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 hnae_ring *ring = ring_data->ring;
309 	struct device *dev = ring_to_dev(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 }
801 
802 static int hns_desc_unused(struct hnae_ring *ring)
803 {
804 	int ntc = ring->next_to_clean;
805 	int ntu = ring->next_to_use;
806 
807 	return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
808 }
809 
810 static int hns_nic_rx_poll_one(struct hns_nic_ring_data *ring_data,
811 			       int budget, void *v)
812 {
813 	struct hnae_ring *ring = ring_data->ring;
814 	struct sk_buff *skb;
815 	int num, bnum;
816 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
817 	int recv_pkts, recv_bds, clean_count, err;
818 	int unused_count = hns_desc_unused(ring);
819 
820 	num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
821 	rmb(); /* make sure num taken effect before the other data is touched */
822 
823 	recv_pkts = 0, recv_bds = 0, clean_count = 0;
824 	num -= unused_count;
825 
826 	while (recv_pkts < budget && recv_bds < num) {
827 		/* reuse or realloc buffers */
828 		if (clean_count + unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
829 			hns_nic_alloc_rx_buffers(ring_data,
830 						 clean_count + unused_count);
831 			clean_count = 0;
832 			unused_count = hns_desc_unused(ring);
833 		}
834 
835 		/* poll one pkt */
836 		err = hns_nic_poll_rx_skb(ring_data, &skb, &bnum);
837 		if (unlikely(!skb)) /* this fault cannot be repaired */
838 			goto out;
839 
840 		recv_bds += bnum;
841 		clean_count += bnum;
842 		if (unlikely(err)) {  /* do jump the err */
843 			recv_pkts++;
844 			continue;
845 		}
846 
847 		/* do update ip stack process*/
848 		((void (*)(struct hns_nic_ring_data *, struct sk_buff *))v)(
849 							ring_data, skb);
850 		recv_pkts++;
851 	}
852 
853 out:
854 	/* make all data has been write before submit */
855 	if (clean_count + unused_count > 0)
856 		hns_nic_alloc_rx_buffers(ring_data,
857 					 clean_count + unused_count);
858 
859 	return recv_pkts;
860 }
861 
862 static void hns_nic_rx_fini_pro(struct hns_nic_ring_data *ring_data)
863 {
864 	struct hnae_ring *ring = ring_data->ring;
865 	int num = 0;
866 
867 	ring_data->ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
868 
869 	/* for hardware bug fixed */
870 	num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
871 
872 	if (num > 0) {
873 		ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
874 			ring_data->ring, 1);
875 
876 		napi_schedule(&ring_data->napi);
877 	}
878 }
879 
880 static void hns_nic_rx_fini_pro_v2(struct hns_nic_ring_data *ring_data)
881 {
882 	struct hnae_ring *ring = ring_data->ring;
883 	int num = 0;
884 
885 	num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
886 
887 	if (num == 0)
888 		ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
889 			ring, 0);
890 	else
891 		napi_schedule(&ring_data->napi);
892 }
893 
894 static inline void hns_nic_reclaim_one_desc(struct hnae_ring *ring,
895 					    int *bytes, int *pkts)
896 {
897 	struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
898 
899 	(*pkts) += (desc_cb->type == DESC_TYPE_SKB);
900 	(*bytes) += desc_cb->length;
901 	/* desc_cb will be cleaned, after hnae_free_buffer_detach*/
902 	hnae_free_buffer_detach(ring, ring->next_to_clean);
903 
904 	ring_ptr_move_fw(ring, next_to_clean);
905 }
906 
907 static int is_valid_clean_head(struct hnae_ring *ring, int h)
908 {
909 	int u = ring->next_to_use;
910 	int c = ring->next_to_clean;
911 
912 	if (unlikely(h > ring->desc_num))
913 		return 0;
914 
915 	assert(u > 0 && u < ring->desc_num);
916 	assert(c > 0 && c < ring->desc_num);
917 	assert(u != c && h != c); /* must be checked before call this func */
918 
919 	return u > c ? (h > c && h <= u) : (h > c || h <= u);
920 }
921 
922 /* netif_tx_lock will turn down the performance, set only when necessary */
923 #ifdef CONFIG_NET_POLL_CONTROLLER
924 #define NETIF_TX_LOCK(ndev) netif_tx_lock(ndev)
925 #define NETIF_TX_UNLOCK(ndev) netif_tx_unlock(ndev)
926 #else
927 #define NETIF_TX_LOCK(ndev)
928 #define NETIF_TX_UNLOCK(ndev)
929 #endif
930 /* reclaim all desc in one budget
931  * return error or number of desc left
932  */
933 static int hns_nic_tx_poll_one(struct hns_nic_ring_data *ring_data,
934 			       int budget, void *v)
935 {
936 	struct hnae_ring *ring = ring_data->ring;
937 	struct net_device *ndev = ring_data->napi.dev;
938 	struct netdev_queue *dev_queue;
939 	struct hns_nic_priv *priv = netdev_priv(ndev);
940 	int head;
941 	int bytes, pkts;
942 
943 	NETIF_TX_LOCK(ndev);
944 
945 	head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
946 	rmb(); /* make sure head is ready before touch any data */
947 
948 	if (is_ring_empty(ring) || head == ring->next_to_clean) {
949 		NETIF_TX_UNLOCK(ndev);
950 		return 0; /* no data to poll */
951 	}
952 
953 	if (!is_valid_clean_head(ring, head)) {
954 		netdev_err(ndev, "wrong head (%d, %d-%d)\n", head,
955 			   ring->next_to_use, ring->next_to_clean);
956 		ring->stats.io_err_cnt++;
957 		NETIF_TX_UNLOCK(ndev);
958 		return -EIO;
959 	}
960 
961 	bytes = 0;
962 	pkts = 0;
963 	while (head != ring->next_to_clean) {
964 		hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
965 		/* issue prefetch for next Tx descriptor */
966 		prefetch(&ring->desc_cb[ring->next_to_clean]);
967 	}
968 
969 	NETIF_TX_UNLOCK(ndev);
970 
971 	dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
972 	netdev_tx_completed_queue(dev_queue, pkts, bytes);
973 
974 	if (unlikely(priv->link && !netif_carrier_ok(ndev)))
975 		netif_carrier_on(ndev);
976 
977 	if (unlikely(pkts && netif_carrier_ok(ndev) &&
978 		     (ring_space(ring) >= ring->max_desc_num_per_pkt * 2))) {
979 		/* Make sure that anybody stopping the queue after this
980 		 * sees the new next_to_clean.
981 		 */
982 		smp_mb();
983 		if (netif_tx_queue_stopped(dev_queue) &&
984 		    !test_bit(NIC_STATE_DOWN, &priv->state)) {
985 			netif_tx_wake_queue(dev_queue);
986 			ring->stats.restart_queue++;
987 		}
988 	}
989 	return 0;
990 }
991 
992 static void hns_nic_tx_fini_pro(struct hns_nic_ring_data *ring_data)
993 {
994 	struct hnae_ring *ring = ring_data->ring;
995 	int head;
996 
997 	ring_data->ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
998 
999 	head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1000 
1001 	if (head != ring->next_to_clean) {
1002 		ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1003 			ring_data->ring, 1);
1004 
1005 		napi_schedule(&ring_data->napi);
1006 	}
1007 }
1008 
1009 static void hns_nic_tx_fini_pro_v2(struct hns_nic_ring_data *ring_data)
1010 {
1011 	struct hnae_ring *ring = ring_data->ring;
1012 	int head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1013 
1014 	if (head == ring->next_to_clean)
1015 		ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1016 			ring, 0);
1017 	else
1018 		napi_schedule(&ring_data->napi);
1019 }
1020 
1021 static void hns_nic_tx_clr_all_bufs(struct hns_nic_ring_data *ring_data)
1022 {
1023 	struct hnae_ring *ring = ring_data->ring;
1024 	struct net_device *ndev = ring_data->napi.dev;
1025 	struct netdev_queue *dev_queue;
1026 	int head;
1027 	int bytes, pkts;
1028 
1029 	NETIF_TX_LOCK(ndev);
1030 
1031 	head = ring->next_to_use; /* ntu :soft setted ring position*/
1032 	bytes = 0;
1033 	pkts = 0;
1034 	while (head != ring->next_to_clean)
1035 		hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
1036 
1037 	NETIF_TX_UNLOCK(ndev);
1038 
1039 	dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
1040 	netdev_tx_reset_queue(dev_queue);
1041 }
1042 
1043 static int hns_nic_common_poll(struct napi_struct *napi, int budget)
1044 {
1045 	struct hns_nic_ring_data *ring_data =
1046 		container_of(napi, struct hns_nic_ring_data, napi);
1047 	int clean_complete = ring_data->poll_one(
1048 				ring_data, budget, ring_data->ex_process);
1049 
1050 	if (clean_complete >= 0 && clean_complete < budget) {
1051 		napi_complete(napi);
1052 		ring_data->fini_process(ring_data);
1053 		return 0;
1054 	}
1055 
1056 	return clean_complete;
1057 }
1058 
1059 static irqreturn_t hns_irq_handle(int irq, void *dev)
1060 {
1061 	struct hns_nic_ring_data *ring_data = (struct hns_nic_ring_data *)dev;
1062 
1063 	ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1064 		ring_data->ring, 1);
1065 	napi_schedule(&ring_data->napi);
1066 
1067 	return IRQ_HANDLED;
1068 }
1069 
1070 /**
1071  *hns_nic_adjust_link - adjust net work mode by the phy stat or new param
1072  *@ndev: net device
1073  */
1074 static void hns_nic_adjust_link(struct net_device *ndev)
1075 {
1076 	struct hns_nic_priv *priv = netdev_priv(ndev);
1077 	struct hnae_handle *h = priv->ae_handle;
1078 	int state = 1;
1079 
1080 	if (ndev->phydev) {
1081 		h->dev->ops->adjust_link(h, ndev->phydev->speed,
1082 					 ndev->phydev->duplex);
1083 		state = ndev->phydev->link;
1084 	}
1085 	state = state && h->dev->ops->get_status(h);
1086 
1087 	if (state != priv->link) {
1088 		if (state) {
1089 			netif_carrier_on(ndev);
1090 			netif_tx_wake_all_queues(ndev);
1091 			netdev_info(ndev, "link up\n");
1092 		} else {
1093 			netif_carrier_off(ndev);
1094 			netdev_info(ndev, "link down\n");
1095 		}
1096 		priv->link = state;
1097 	}
1098 }
1099 
1100 /**
1101  *hns_nic_init_phy - init phy
1102  *@ndev: net device
1103  *@h: ae handle
1104  * Return 0 on success, negative on failure
1105  */
1106 int hns_nic_init_phy(struct net_device *ndev, struct hnae_handle *h)
1107 {
1108 	struct phy_device *phy_dev = h->phy_dev;
1109 	int ret;
1110 
1111 	if (!h->phy_dev)
1112 		return 0;
1113 
1114 	if (h->phy_if != PHY_INTERFACE_MODE_XGMII) {
1115 		phy_dev->dev_flags = 0;
1116 
1117 		ret = phy_connect_direct(ndev, phy_dev, hns_nic_adjust_link,
1118 					 h->phy_if);
1119 	} else {
1120 		ret = phy_attach_direct(ndev, phy_dev, 0, h->phy_if);
1121 	}
1122 	if (unlikely(ret))
1123 		return -ENODEV;
1124 
1125 	phy_dev->supported &= h->if_support;
1126 	phy_dev->advertising = phy_dev->supported;
1127 
1128 	if (h->phy_if == PHY_INTERFACE_MODE_XGMII)
1129 		phy_dev->autoneg = false;
1130 
1131 	return 0;
1132 }
1133 
1134 static int hns_nic_ring_open(struct net_device *netdev, int idx)
1135 {
1136 	struct hns_nic_priv *priv = netdev_priv(netdev);
1137 	struct hnae_handle *h = priv->ae_handle;
1138 
1139 	napi_enable(&priv->ring_data[idx].napi);
1140 
1141 	enable_irq(priv->ring_data[idx].ring->irq);
1142 	h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 0);
1143 
1144 	return 0;
1145 }
1146 
1147 static int hns_nic_net_set_mac_address(struct net_device *ndev, void *p)
1148 {
1149 	struct hns_nic_priv *priv = netdev_priv(ndev);
1150 	struct hnae_handle *h = priv->ae_handle;
1151 	struct sockaddr *mac_addr = p;
1152 	int ret;
1153 
1154 	if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1155 		return -EADDRNOTAVAIL;
1156 
1157 	ret = h->dev->ops->set_mac_addr(h, mac_addr->sa_data);
1158 	if (ret) {
1159 		netdev_err(ndev, "set_mac_address fail, ret=%d!\n", ret);
1160 		return ret;
1161 	}
1162 
1163 	memcpy(ndev->dev_addr, mac_addr->sa_data, ndev->addr_len);
1164 
1165 	return 0;
1166 }
1167 
1168 void hns_nic_update_stats(struct net_device *netdev)
1169 {
1170 	struct hns_nic_priv *priv = netdev_priv(netdev);
1171 	struct hnae_handle *h = priv->ae_handle;
1172 
1173 	h->dev->ops->update_stats(h, &netdev->stats);
1174 }
1175 
1176 /* set mac addr if it is configed. or leave it to the AE driver */
1177 static void hns_init_mac_addr(struct net_device *ndev)
1178 {
1179 	struct hns_nic_priv *priv = netdev_priv(ndev);
1180 
1181 	if (!device_get_mac_address(priv->dev, ndev->dev_addr, ETH_ALEN)) {
1182 		eth_hw_addr_random(ndev);
1183 		dev_warn(priv->dev, "No valid mac, use random mac %pM",
1184 			 ndev->dev_addr);
1185 	}
1186 }
1187 
1188 static void hns_nic_ring_close(struct net_device *netdev, int idx)
1189 {
1190 	struct hns_nic_priv *priv = netdev_priv(netdev);
1191 	struct hnae_handle *h = priv->ae_handle;
1192 
1193 	h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 1);
1194 	disable_irq(priv->ring_data[idx].ring->irq);
1195 
1196 	napi_disable(&priv->ring_data[idx].napi);
1197 }
1198 
1199 static void hns_set_irq_affinity(struct hns_nic_priv *priv)
1200 {
1201 	struct hnae_handle *h = priv->ae_handle;
1202 	struct hns_nic_ring_data *rd;
1203 	int i;
1204 	int cpu;
1205 	cpumask_var_t mask;
1206 
1207 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1208 		return;
1209 
1210 	/*diffrent irq banlance for 16core and 32core*/
1211 	if (h->q_num == num_possible_cpus()) {
1212 		for (i = 0; i < h->q_num * 2; i++) {
1213 			rd = &priv->ring_data[i];
1214 			if (cpu_online(rd->queue_index)) {
1215 				cpumask_clear(mask);
1216 				cpu = rd->queue_index;
1217 				cpumask_set_cpu(cpu, mask);
1218 				(void)irq_set_affinity_hint(rd->ring->irq,
1219 							    mask);
1220 			}
1221 		}
1222 	} else {
1223 		for (i = 0; i < h->q_num; i++) {
1224 			rd = &priv->ring_data[i];
1225 			if (cpu_online(rd->queue_index * 2)) {
1226 				cpumask_clear(mask);
1227 				cpu = rd->queue_index * 2;
1228 				cpumask_set_cpu(cpu, mask);
1229 				(void)irq_set_affinity_hint(rd->ring->irq,
1230 							    mask);
1231 			}
1232 		}
1233 
1234 		for (i = h->q_num; i < h->q_num * 2; i++) {
1235 			rd = &priv->ring_data[i];
1236 			if (cpu_online(rd->queue_index * 2 + 1)) {
1237 				cpumask_clear(mask);
1238 				cpu = rd->queue_index * 2 + 1;
1239 				cpumask_set_cpu(cpu, mask);
1240 				(void)irq_set_affinity_hint(rd->ring->irq,
1241 							    mask);
1242 			}
1243 		}
1244 	}
1245 
1246 	free_cpumask_var(mask);
1247 }
1248 
1249 static int hns_nic_init_irq(struct hns_nic_priv *priv)
1250 {
1251 	struct hnae_handle *h = priv->ae_handle;
1252 	struct hns_nic_ring_data *rd;
1253 	int i;
1254 	int ret;
1255 
1256 	for (i = 0; i < h->q_num * 2; i++) {
1257 		rd = &priv->ring_data[i];
1258 
1259 		if (rd->ring->irq_init_flag == RCB_IRQ_INITED)
1260 			break;
1261 
1262 		snprintf(rd->ring->ring_name, RCB_RING_NAME_LEN,
1263 			 "%s-%s%d", priv->netdev->name,
1264 			 (i < h->q_num ? "tx" : "rx"), rd->queue_index);
1265 
1266 		rd->ring->ring_name[RCB_RING_NAME_LEN - 1] = '\0';
1267 
1268 		ret = request_irq(rd->ring->irq,
1269 				  hns_irq_handle, 0, rd->ring->ring_name, rd);
1270 		if (ret) {
1271 			netdev_err(priv->netdev, "request irq(%d) fail\n",
1272 				   rd->ring->irq);
1273 			return ret;
1274 		}
1275 		disable_irq(rd->ring->irq);
1276 		rd->ring->irq_init_flag = RCB_IRQ_INITED;
1277 	}
1278 
1279 	/*set cpu affinity*/
1280 	hns_set_irq_affinity(priv);
1281 
1282 	return 0;
1283 }
1284 
1285 static int hns_nic_net_up(struct net_device *ndev)
1286 {
1287 	struct hns_nic_priv *priv = netdev_priv(ndev);
1288 	struct hnae_handle *h = priv->ae_handle;
1289 	int i, j;
1290 	int ret;
1291 
1292 	ret = hns_nic_init_irq(priv);
1293 	if (ret != 0) {
1294 		netdev_err(ndev, "hns init irq failed! ret=%d\n", ret);
1295 		return ret;
1296 	}
1297 
1298 	for (i = 0; i < h->q_num * 2; i++) {
1299 		ret = hns_nic_ring_open(ndev, i);
1300 		if (ret)
1301 			goto out_has_some_queues;
1302 	}
1303 
1304 	ret = h->dev->ops->set_mac_addr(h, ndev->dev_addr);
1305 	if (ret)
1306 		goto out_set_mac_addr_err;
1307 
1308 	ret = h->dev->ops->start ? h->dev->ops->start(h) : 0;
1309 	if (ret)
1310 		goto out_start_err;
1311 
1312 	if (ndev->phydev)
1313 		phy_start(ndev->phydev);
1314 
1315 	clear_bit(NIC_STATE_DOWN, &priv->state);
1316 	(void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
1317 
1318 	return 0;
1319 
1320 out_start_err:
1321 	netif_stop_queue(ndev);
1322 out_set_mac_addr_err:
1323 out_has_some_queues:
1324 	for (j = i - 1; j >= 0; j--)
1325 		hns_nic_ring_close(ndev, j);
1326 
1327 	set_bit(NIC_STATE_DOWN, &priv->state);
1328 
1329 	return ret;
1330 }
1331 
1332 static void hns_nic_net_down(struct net_device *ndev)
1333 {
1334 	int i;
1335 	struct hnae_ae_ops *ops;
1336 	struct hns_nic_priv *priv = netdev_priv(ndev);
1337 
1338 	if (test_and_set_bit(NIC_STATE_DOWN, &priv->state))
1339 		return;
1340 
1341 	(void)del_timer_sync(&priv->service_timer);
1342 	netif_tx_stop_all_queues(ndev);
1343 	netif_carrier_off(ndev);
1344 	netif_tx_disable(ndev);
1345 	priv->link = 0;
1346 
1347 	if (ndev->phydev)
1348 		phy_stop(ndev->phydev);
1349 
1350 	ops = priv->ae_handle->dev->ops;
1351 
1352 	if (ops->stop)
1353 		ops->stop(priv->ae_handle);
1354 
1355 	netif_tx_stop_all_queues(ndev);
1356 
1357 	for (i = priv->ae_handle->q_num - 1; i >= 0; i--) {
1358 		hns_nic_ring_close(ndev, i);
1359 		hns_nic_ring_close(ndev, i + priv->ae_handle->q_num);
1360 
1361 		/* clean tx buffers*/
1362 		hns_nic_tx_clr_all_bufs(priv->ring_data + i);
1363 	}
1364 }
1365 
1366 void hns_nic_net_reset(struct net_device *ndev)
1367 {
1368 	struct hns_nic_priv *priv = netdev_priv(ndev);
1369 	struct hnae_handle *handle = priv->ae_handle;
1370 
1371 	while (test_and_set_bit(NIC_STATE_RESETTING, &priv->state))
1372 		usleep_range(1000, 2000);
1373 
1374 	(void)hnae_reinit_handle(handle);
1375 
1376 	clear_bit(NIC_STATE_RESETTING, &priv->state);
1377 }
1378 
1379 void hns_nic_net_reinit(struct net_device *netdev)
1380 {
1381 	struct hns_nic_priv *priv = netdev_priv(netdev);
1382 
1383 	netif_trans_update(priv->netdev);
1384 	while (test_and_set_bit(NIC_STATE_REINITING, &priv->state))
1385 		usleep_range(1000, 2000);
1386 
1387 	hns_nic_net_down(netdev);
1388 	hns_nic_net_reset(netdev);
1389 	(void)hns_nic_net_up(netdev);
1390 	clear_bit(NIC_STATE_REINITING, &priv->state);
1391 }
1392 
1393 static int hns_nic_net_open(struct net_device *ndev)
1394 {
1395 	struct hns_nic_priv *priv = netdev_priv(ndev);
1396 	struct hnae_handle *h = priv->ae_handle;
1397 	int ret;
1398 
1399 	if (test_bit(NIC_STATE_TESTING, &priv->state))
1400 		return -EBUSY;
1401 
1402 	priv->link = 0;
1403 	netif_carrier_off(ndev);
1404 
1405 	ret = netif_set_real_num_tx_queues(ndev, h->q_num);
1406 	if (ret < 0) {
1407 		netdev_err(ndev, "netif_set_real_num_tx_queues fail, ret=%d!\n",
1408 			   ret);
1409 		return ret;
1410 	}
1411 
1412 	ret = netif_set_real_num_rx_queues(ndev, h->q_num);
1413 	if (ret < 0) {
1414 		netdev_err(ndev,
1415 			   "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
1416 		return ret;
1417 	}
1418 
1419 	ret = hns_nic_net_up(ndev);
1420 	if (ret) {
1421 		netdev_err(ndev,
1422 			   "hns net up fail, ret=%d!\n", ret);
1423 		return ret;
1424 	}
1425 
1426 	return 0;
1427 }
1428 
1429 static int hns_nic_net_stop(struct net_device *ndev)
1430 {
1431 	hns_nic_net_down(ndev);
1432 
1433 	return 0;
1434 }
1435 
1436 static void hns_tx_timeout_reset(struct hns_nic_priv *priv);
1437 static void hns_nic_net_timeout(struct net_device *ndev)
1438 {
1439 	struct hns_nic_priv *priv = netdev_priv(ndev);
1440 
1441 	hns_tx_timeout_reset(priv);
1442 }
1443 
1444 static int hns_nic_do_ioctl(struct net_device *netdev, struct ifreq *ifr,
1445 			    int cmd)
1446 {
1447 	struct phy_device *phy_dev = netdev->phydev;
1448 
1449 	if (!netif_running(netdev))
1450 		return -EINVAL;
1451 
1452 	if (!phy_dev)
1453 		return -ENOTSUPP;
1454 
1455 	return phy_mii_ioctl(phy_dev, ifr, cmd);
1456 }
1457 
1458 /* use only for netconsole to poll with the device without interrupt */
1459 #ifdef CONFIG_NET_POLL_CONTROLLER
1460 void hns_nic_poll_controller(struct net_device *ndev)
1461 {
1462 	struct hns_nic_priv *priv = netdev_priv(ndev);
1463 	unsigned long flags;
1464 	int i;
1465 
1466 	local_irq_save(flags);
1467 	for (i = 0; i < priv->ae_handle->q_num * 2; i++)
1468 		napi_schedule(&priv->ring_data[i].napi);
1469 	local_irq_restore(flags);
1470 }
1471 #endif
1472 
1473 static netdev_tx_t hns_nic_net_xmit(struct sk_buff *skb,
1474 				    struct net_device *ndev)
1475 {
1476 	struct hns_nic_priv *priv = netdev_priv(ndev);
1477 	int ret;
1478 
1479 	assert(skb->queue_mapping < ndev->ae_handle->q_num);
1480 	ret = hns_nic_net_xmit_hw(ndev, skb,
1481 				  &tx_ring_data(priv, skb->queue_mapping));
1482 	if (ret == NETDEV_TX_OK) {
1483 		netif_trans_update(ndev);
1484 		ndev->stats.tx_bytes += skb->len;
1485 		ndev->stats.tx_packets++;
1486 	}
1487 	return (netdev_tx_t)ret;
1488 }
1489 
1490 static int hns_nic_change_mtu(struct net_device *ndev, int new_mtu)
1491 {
1492 	struct hns_nic_priv *priv = netdev_priv(ndev);
1493 	struct hnae_handle *h = priv->ae_handle;
1494 	int ret;
1495 
1496 	if (!h->dev->ops->set_mtu)
1497 		return -ENOTSUPP;
1498 
1499 	if (netif_running(ndev)) {
1500 		(void)hns_nic_net_stop(ndev);
1501 		msleep(100);
1502 
1503 		ret = h->dev->ops->set_mtu(h, new_mtu);
1504 		if (ret)
1505 			netdev_err(ndev, "set mtu fail, return value %d\n",
1506 				   ret);
1507 
1508 		if (hns_nic_net_open(ndev))
1509 			netdev_err(ndev, "hns net open fail\n");
1510 	} else {
1511 		ret = h->dev->ops->set_mtu(h, new_mtu);
1512 	}
1513 
1514 	if (!ret)
1515 		ndev->mtu = new_mtu;
1516 
1517 	return ret;
1518 }
1519 
1520 static int hns_nic_set_features(struct net_device *netdev,
1521 				netdev_features_t features)
1522 {
1523 	struct hns_nic_priv *priv = netdev_priv(netdev);
1524 
1525 	switch (priv->enet_ver) {
1526 	case AE_VERSION_1:
1527 		if (features & (NETIF_F_TSO | NETIF_F_TSO6))
1528 			netdev_info(netdev, "enet v1 do not support tso!\n");
1529 		break;
1530 	default:
1531 		if (features & (NETIF_F_TSO | NETIF_F_TSO6)) {
1532 			priv->ops.fill_desc = fill_tso_desc;
1533 			priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tso;
1534 			/* The chip only support 7*4096 */
1535 			netif_set_gso_max_size(netdev, 7 * 4096);
1536 		} else {
1537 			priv->ops.fill_desc = fill_v2_desc;
1538 			priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
1539 		}
1540 		break;
1541 	}
1542 	netdev->features = features;
1543 	return 0;
1544 }
1545 
1546 static netdev_features_t hns_nic_fix_features(
1547 		struct net_device *netdev, netdev_features_t features)
1548 {
1549 	struct hns_nic_priv *priv = netdev_priv(netdev);
1550 
1551 	switch (priv->enet_ver) {
1552 	case AE_VERSION_1:
1553 		features &= ~(NETIF_F_TSO | NETIF_F_TSO6 |
1554 				NETIF_F_HW_VLAN_CTAG_FILTER);
1555 		break;
1556 	default:
1557 		break;
1558 	}
1559 	return features;
1560 }
1561 
1562 static int hns_nic_uc_sync(struct net_device *netdev, const unsigned char *addr)
1563 {
1564 	struct hns_nic_priv *priv = netdev_priv(netdev);
1565 	struct hnae_handle *h = priv->ae_handle;
1566 
1567 	if (h->dev->ops->add_uc_addr)
1568 		return h->dev->ops->add_uc_addr(h, addr);
1569 
1570 	return 0;
1571 }
1572 
1573 static int hns_nic_uc_unsync(struct net_device *netdev,
1574 			     const unsigned char *addr)
1575 {
1576 	struct hns_nic_priv *priv = netdev_priv(netdev);
1577 	struct hnae_handle *h = priv->ae_handle;
1578 
1579 	if (h->dev->ops->rm_uc_addr)
1580 		return h->dev->ops->rm_uc_addr(h, addr);
1581 
1582 	return 0;
1583 }
1584 
1585 /**
1586  * nic_set_multicast_list - set mutl mac address
1587  * @netdev: net device
1588  * @p: mac address
1589  *
1590  * return void
1591  */
1592 void hns_set_multicast_list(struct net_device *ndev)
1593 {
1594 	struct hns_nic_priv *priv = netdev_priv(ndev);
1595 	struct hnae_handle *h = priv->ae_handle;
1596 	struct netdev_hw_addr *ha = NULL;
1597 
1598 	if (!h)	{
1599 		netdev_err(ndev, "hnae handle is null\n");
1600 		return;
1601 	}
1602 
1603 	if (h->dev->ops->clr_mc_addr)
1604 		if (h->dev->ops->clr_mc_addr(h))
1605 			netdev_err(ndev, "clear multicast address fail\n");
1606 
1607 	if (h->dev->ops->set_mc_addr) {
1608 		netdev_for_each_mc_addr(ha, ndev)
1609 			if (h->dev->ops->set_mc_addr(h, ha->addr))
1610 				netdev_err(ndev, "set multicast fail\n");
1611 	}
1612 }
1613 
1614 void hns_nic_set_rx_mode(struct net_device *ndev)
1615 {
1616 	struct hns_nic_priv *priv = netdev_priv(ndev);
1617 	struct hnae_handle *h = priv->ae_handle;
1618 
1619 	if (h->dev->ops->set_promisc_mode) {
1620 		if (ndev->flags & IFF_PROMISC)
1621 			h->dev->ops->set_promisc_mode(h, 1);
1622 		else
1623 			h->dev->ops->set_promisc_mode(h, 0);
1624 	}
1625 
1626 	hns_set_multicast_list(ndev);
1627 
1628 	if (__dev_uc_sync(ndev, hns_nic_uc_sync, hns_nic_uc_unsync))
1629 		netdev_err(ndev, "sync uc address fail\n");
1630 }
1631 
1632 static void hns_nic_get_stats64(struct net_device *ndev,
1633 				struct rtnl_link_stats64 *stats)
1634 {
1635 	int idx = 0;
1636 	u64 tx_bytes = 0;
1637 	u64 rx_bytes = 0;
1638 	u64 tx_pkts = 0;
1639 	u64 rx_pkts = 0;
1640 	struct hns_nic_priv *priv = netdev_priv(ndev);
1641 	struct hnae_handle *h = priv->ae_handle;
1642 
1643 	for (idx = 0; idx < h->q_num; idx++) {
1644 		tx_bytes += h->qs[idx]->tx_ring.stats.tx_bytes;
1645 		tx_pkts += h->qs[idx]->tx_ring.stats.tx_pkts;
1646 		rx_bytes += h->qs[idx]->rx_ring.stats.rx_bytes;
1647 		rx_pkts += h->qs[idx]->rx_ring.stats.rx_pkts;
1648 	}
1649 
1650 	stats->tx_bytes = tx_bytes;
1651 	stats->tx_packets = tx_pkts;
1652 	stats->rx_bytes = rx_bytes;
1653 	stats->rx_packets = rx_pkts;
1654 
1655 	stats->rx_errors = ndev->stats.rx_errors;
1656 	stats->multicast = ndev->stats.multicast;
1657 	stats->rx_length_errors = ndev->stats.rx_length_errors;
1658 	stats->rx_crc_errors = ndev->stats.rx_crc_errors;
1659 	stats->rx_missed_errors = ndev->stats.rx_missed_errors;
1660 
1661 	stats->tx_errors = ndev->stats.tx_errors;
1662 	stats->rx_dropped = ndev->stats.rx_dropped;
1663 	stats->tx_dropped = ndev->stats.tx_dropped;
1664 	stats->collisions = ndev->stats.collisions;
1665 	stats->rx_over_errors = ndev->stats.rx_over_errors;
1666 	stats->rx_frame_errors = ndev->stats.rx_frame_errors;
1667 	stats->rx_fifo_errors = ndev->stats.rx_fifo_errors;
1668 	stats->tx_aborted_errors = ndev->stats.tx_aborted_errors;
1669 	stats->tx_carrier_errors = ndev->stats.tx_carrier_errors;
1670 	stats->tx_fifo_errors = ndev->stats.tx_fifo_errors;
1671 	stats->tx_heartbeat_errors = ndev->stats.tx_heartbeat_errors;
1672 	stats->tx_window_errors = ndev->stats.tx_window_errors;
1673 	stats->rx_compressed = ndev->stats.rx_compressed;
1674 	stats->tx_compressed = ndev->stats.tx_compressed;
1675 }
1676 
1677 static u16
1678 hns_nic_select_queue(struct net_device *ndev, struct sk_buff *skb,
1679 		     void *accel_priv, select_queue_fallback_t fallback)
1680 {
1681 	struct ethhdr *eth_hdr = (struct ethhdr *)skb->data;
1682 	struct hns_nic_priv *priv = netdev_priv(ndev);
1683 
1684 	/* fix hardware broadcast/multicast packets queue loopback */
1685 	if (!AE_IS_VER1(priv->enet_ver) &&
1686 	    is_multicast_ether_addr(eth_hdr->h_dest))
1687 		return 0;
1688 	else
1689 		return fallback(ndev, skb);
1690 }
1691 
1692 static const struct net_device_ops hns_nic_netdev_ops = {
1693 	.ndo_open = hns_nic_net_open,
1694 	.ndo_stop = hns_nic_net_stop,
1695 	.ndo_start_xmit = hns_nic_net_xmit,
1696 	.ndo_tx_timeout = hns_nic_net_timeout,
1697 	.ndo_set_mac_address = hns_nic_net_set_mac_address,
1698 	.ndo_change_mtu = hns_nic_change_mtu,
1699 	.ndo_do_ioctl = hns_nic_do_ioctl,
1700 	.ndo_set_features = hns_nic_set_features,
1701 	.ndo_fix_features = hns_nic_fix_features,
1702 	.ndo_get_stats64 = hns_nic_get_stats64,
1703 #ifdef CONFIG_NET_POLL_CONTROLLER
1704 	.ndo_poll_controller = hns_nic_poll_controller,
1705 #endif
1706 	.ndo_set_rx_mode = hns_nic_set_rx_mode,
1707 	.ndo_select_queue = hns_nic_select_queue,
1708 };
1709 
1710 static void hns_nic_update_link_status(struct net_device *netdev)
1711 {
1712 	struct hns_nic_priv *priv = netdev_priv(netdev);
1713 
1714 	struct hnae_handle *h = priv->ae_handle;
1715 
1716 	if (h->phy_dev) {
1717 		if (h->phy_if != PHY_INTERFACE_MODE_XGMII)
1718 			return;
1719 
1720 		(void)genphy_read_status(h->phy_dev);
1721 	}
1722 	hns_nic_adjust_link(netdev);
1723 }
1724 
1725 /* for dumping key regs*/
1726 static void hns_nic_dump(struct hns_nic_priv *priv)
1727 {
1728 	struct hnae_handle *h = priv->ae_handle;
1729 	struct hnae_ae_ops *ops = h->dev->ops;
1730 	u32 *data, reg_num, i;
1731 
1732 	if (ops->get_regs_len && ops->get_regs) {
1733 		reg_num = ops->get_regs_len(priv->ae_handle);
1734 		reg_num = (reg_num + 3ul) & ~3ul;
1735 		data = kcalloc(reg_num, sizeof(u32), GFP_KERNEL);
1736 		if (data) {
1737 			ops->get_regs(priv->ae_handle, data);
1738 			for (i = 0; i < reg_num; i += 4)
1739 				pr_info("0x%08x: 0x%08x 0x%08x 0x%08x 0x%08x\n",
1740 					i, data[i], data[i + 1],
1741 					data[i + 2], data[i + 3]);
1742 			kfree(data);
1743 		}
1744 	}
1745 
1746 	for (i = 0; i < h->q_num; i++) {
1747 		pr_info("tx_queue%d_next_to_clean:%d\n",
1748 			i, h->qs[i]->tx_ring.next_to_clean);
1749 		pr_info("tx_queue%d_next_to_use:%d\n",
1750 			i, h->qs[i]->tx_ring.next_to_use);
1751 		pr_info("rx_queue%d_next_to_clean:%d\n",
1752 			i, h->qs[i]->rx_ring.next_to_clean);
1753 		pr_info("rx_queue%d_next_to_use:%d\n",
1754 			i, h->qs[i]->rx_ring.next_to_use);
1755 	}
1756 }
1757 
1758 /* for resetting subtask */
1759 static void hns_nic_reset_subtask(struct hns_nic_priv *priv)
1760 {
1761 	enum hnae_port_type type = priv->ae_handle->port_type;
1762 
1763 	if (!test_bit(NIC_STATE2_RESET_REQUESTED, &priv->state))
1764 		return;
1765 	clear_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
1766 
1767 	/* If we're already down, removing or resetting, just bail */
1768 	if (test_bit(NIC_STATE_DOWN, &priv->state) ||
1769 	    test_bit(NIC_STATE_REMOVING, &priv->state) ||
1770 	    test_bit(NIC_STATE_RESETTING, &priv->state))
1771 		return;
1772 
1773 	hns_nic_dump(priv);
1774 	netdev_info(priv->netdev, "try to reset %s port!\n",
1775 		    (type == HNAE_PORT_DEBUG ? "debug" : "service"));
1776 
1777 	rtnl_lock();
1778 	/* put off any impending NetWatchDogTimeout */
1779 	netif_trans_update(priv->netdev);
1780 
1781 	if (type == HNAE_PORT_DEBUG) {
1782 		hns_nic_net_reinit(priv->netdev);
1783 	} else {
1784 		netif_carrier_off(priv->netdev);
1785 		netif_tx_disable(priv->netdev);
1786 	}
1787 	rtnl_unlock();
1788 }
1789 
1790 /* for doing service complete*/
1791 static void hns_nic_service_event_complete(struct hns_nic_priv *priv)
1792 {
1793 	WARN_ON(!test_bit(NIC_STATE_SERVICE_SCHED, &priv->state));
1794 
1795 	smp_mb__before_atomic();
1796 	clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
1797 }
1798 
1799 static void hns_nic_service_task(struct work_struct *work)
1800 {
1801 	struct hns_nic_priv *priv
1802 		= container_of(work, struct hns_nic_priv, service_task);
1803 	struct hnae_handle *h = priv->ae_handle;
1804 
1805 	hns_nic_update_link_status(priv->netdev);
1806 	h->dev->ops->update_led_status(h);
1807 	hns_nic_update_stats(priv->netdev);
1808 
1809 	hns_nic_reset_subtask(priv);
1810 	hns_nic_service_event_complete(priv);
1811 }
1812 
1813 static void hns_nic_task_schedule(struct hns_nic_priv *priv)
1814 {
1815 	if (!test_bit(NIC_STATE_DOWN, &priv->state) &&
1816 	    !test_bit(NIC_STATE_REMOVING, &priv->state) &&
1817 	    !test_and_set_bit(NIC_STATE_SERVICE_SCHED, &priv->state))
1818 		(void)schedule_work(&priv->service_task);
1819 }
1820 
1821 static void hns_nic_service_timer(unsigned long data)
1822 {
1823 	struct hns_nic_priv *priv = (struct hns_nic_priv *)data;
1824 
1825 	(void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
1826 
1827 	hns_nic_task_schedule(priv);
1828 }
1829 
1830 /**
1831  * hns_tx_timeout_reset - initiate reset due to Tx timeout
1832  * @priv: driver private struct
1833  **/
1834 static void hns_tx_timeout_reset(struct hns_nic_priv *priv)
1835 {
1836 	/* Do the reset outside of interrupt context */
1837 	if (!test_bit(NIC_STATE_DOWN, &priv->state)) {
1838 		set_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
1839 		netdev_warn(priv->netdev,
1840 			    "initiating reset due to tx timeout(%llu,0x%lx)\n",
1841 			    priv->tx_timeout_count, priv->state);
1842 		priv->tx_timeout_count++;
1843 		hns_nic_task_schedule(priv);
1844 	}
1845 }
1846 
1847 static int hns_nic_init_ring_data(struct hns_nic_priv *priv)
1848 {
1849 	struct hnae_handle *h = priv->ae_handle;
1850 	struct hns_nic_ring_data *rd;
1851 	bool is_ver1 = AE_IS_VER1(priv->enet_ver);
1852 	int i;
1853 
1854 	if (h->q_num > NIC_MAX_Q_PER_VF) {
1855 		netdev_err(priv->netdev, "too much queue (%d)\n", h->q_num);
1856 		return -EINVAL;
1857 	}
1858 
1859 	priv->ring_data = kzalloc(h->q_num * sizeof(*priv->ring_data) * 2,
1860 				  GFP_KERNEL);
1861 	if (!priv->ring_data)
1862 		return -ENOMEM;
1863 
1864 	for (i = 0; i < h->q_num; i++) {
1865 		rd = &priv->ring_data[i];
1866 		rd->queue_index = i;
1867 		rd->ring = &h->qs[i]->tx_ring;
1868 		rd->poll_one = hns_nic_tx_poll_one;
1869 		rd->fini_process = is_ver1 ? hns_nic_tx_fini_pro :
1870 			hns_nic_tx_fini_pro_v2;
1871 
1872 		netif_napi_add(priv->netdev, &rd->napi,
1873 			       hns_nic_common_poll, NIC_TX_CLEAN_MAX_NUM);
1874 		rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
1875 	}
1876 	for (i = h->q_num; i < h->q_num * 2; i++) {
1877 		rd = &priv->ring_data[i];
1878 		rd->queue_index = i - h->q_num;
1879 		rd->ring = &h->qs[i - h->q_num]->rx_ring;
1880 		rd->poll_one = hns_nic_rx_poll_one;
1881 		rd->ex_process = hns_nic_rx_up_pro;
1882 		rd->fini_process = is_ver1 ? hns_nic_rx_fini_pro :
1883 			hns_nic_rx_fini_pro_v2;
1884 
1885 		netif_napi_add(priv->netdev, &rd->napi,
1886 			       hns_nic_common_poll, NIC_RX_CLEAN_MAX_NUM);
1887 		rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
1888 	}
1889 
1890 	return 0;
1891 }
1892 
1893 static void hns_nic_uninit_ring_data(struct hns_nic_priv *priv)
1894 {
1895 	struct hnae_handle *h = priv->ae_handle;
1896 	int i;
1897 
1898 	for (i = 0; i < h->q_num * 2; i++) {
1899 		netif_napi_del(&priv->ring_data[i].napi);
1900 		if (priv->ring_data[i].ring->irq_init_flag == RCB_IRQ_INITED) {
1901 			(void)irq_set_affinity_hint(
1902 				priv->ring_data[i].ring->irq,
1903 				NULL);
1904 			free_irq(priv->ring_data[i].ring->irq,
1905 				 &priv->ring_data[i]);
1906 		}
1907 
1908 		priv->ring_data[i].ring->irq_init_flag = RCB_IRQ_NOT_INITED;
1909 	}
1910 	kfree(priv->ring_data);
1911 }
1912 
1913 static void hns_nic_set_priv_ops(struct net_device *netdev)
1914 {
1915 	struct hns_nic_priv *priv = netdev_priv(netdev);
1916 	struct hnae_handle *h = priv->ae_handle;
1917 
1918 	if (AE_IS_VER1(priv->enet_ver)) {
1919 		priv->ops.fill_desc = fill_desc;
1920 		priv->ops.get_rxd_bnum = get_rx_desc_bnum;
1921 		priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
1922 	} else {
1923 		priv->ops.get_rxd_bnum = get_v2rx_desc_bnum;
1924 		if ((netdev->features & NETIF_F_TSO) ||
1925 		    (netdev->features & NETIF_F_TSO6)) {
1926 			priv->ops.fill_desc = fill_tso_desc;
1927 			priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tso;
1928 			/* This chip only support 7*4096 */
1929 			netif_set_gso_max_size(netdev, 7 * 4096);
1930 		} else {
1931 			priv->ops.fill_desc = fill_v2_desc;
1932 			priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
1933 		}
1934 		/* enable tso when init
1935 		 * control tso on/off through TSE bit in bd
1936 		 */
1937 		h->dev->ops->set_tso_stats(h, 1);
1938 	}
1939 }
1940 
1941 static int hns_nic_try_get_ae(struct net_device *ndev)
1942 {
1943 	struct hns_nic_priv *priv = netdev_priv(ndev);
1944 	struct hnae_handle *h;
1945 	int ret;
1946 
1947 	h = hnae_get_handle(&priv->netdev->dev,
1948 			    priv->fwnode, priv->port_id, NULL);
1949 	if (IS_ERR_OR_NULL(h)) {
1950 		ret = -ENODEV;
1951 		dev_dbg(priv->dev, "has not handle, register notifier!\n");
1952 		goto out;
1953 	}
1954 	priv->ae_handle = h;
1955 
1956 	ret = hns_nic_init_phy(ndev, h);
1957 	if (ret) {
1958 		dev_err(priv->dev, "probe phy device fail!\n");
1959 		goto out_init_phy;
1960 	}
1961 
1962 	ret = hns_nic_init_ring_data(priv);
1963 	if (ret) {
1964 		ret = -ENOMEM;
1965 		goto out_init_ring_data;
1966 	}
1967 
1968 	hns_nic_set_priv_ops(ndev);
1969 
1970 	ret = register_netdev(ndev);
1971 	if (ret) {
1972 		dev_err(priv->dev, "probe register netdev fail!\n");
1973 		goto out_reg_ndev_fail;
1974 	}
1975 	return 0;
1976 
1977 out_reg_ndev_fail:
1978 	hns_nic_uninit_ring_data(priv);
1979 	priv->ring_data = NULL;
1980 out_init_phy:
1981 out_init_ring_data:
1982 	hnae_put_handle(priv->ae_handle);
1983 	priv->ae_handle = NULL;
1984 out:
1985 	return ret;
1986 }
1987 
1988 static int hns_nic_notifier_action(struct notifier_block *nb,
1989 				   unsigned long action, void *data)
1990 {
1991 	struct hns_nic_priv *priv =
1992 		container_of(nb, struct hns_nic_priv, notifier_block);
1993 
1994 	assert(action == HNAE_AE_REGISTER);
1995 
1996 	if (!hns_nic_try_get_ae(priv->netdev)) {
1997 		hnae_unregister_notifier(&priv->notifier_block);
1998 		priv->notifier_block.notifier_call = NULL;
1999 	}
2000 	return 0;
2001 }
2002 
2003 static int hns_nic_dev_probe(struct platform_device *pdev)
2004 {
2005 	struct device *dev = &pdev->dev;
2006 	struct net_device *ndev;
2007 	struct hns_nic_priv *priv;
2008 	u32 port_id;
2009 	int ret;
2010 
2011 	ndev = alloc_etherdev_mq(sizeof(struct hns_nic_priv), NIC_MAX_Q_PER_VF);
2012 	if (!ndev)
2013 		return -ENOMEM;
2014 
2015 	platform_set_drvdata(pdev, ndev);
2016 
2017 	priv = netdev_priv(ndev);
2018 	priv->dev = dev;
2019 	priv->netdev = ndev;
2020 
2021 	if (dev_of_node(dev)) {
2022 		struct device_node *ae_node;
2023 
2024 		if (of_device_is_compatible(dev->of_node,
2025 					    "hisilicon,hns-nic-v1"))
2026 			priv->enet_ver = AE_VERSION_1;
2027 		else
2028 			priv->enet_ver = AE_VERSION_2;
2029 
2030 		ae_node = of_parse_phandle(dev->of_node, "ae-handle", 0);
2031 		if (IS_ERR_OR_NULL(ae_node)) {
2032 			ret = PTR_ERR(ae_node);
2033 			dev_err(dev, "not find ae-handle\n");
2034 			goto out_read_prop_fail;
2035 		}
2036 		priv->fwnode = &ae_node->fwnode;
2037 	} else if (is_acpi_node(dev->fwnode)) {
2038 		struct acpi_reference_args args;
2039 
2040 		if (acpi_dev_found(hns_enet_acpi_match[0].id))
2041 			priv->enet_ver = AE_VERSION_1;
2042 		else if (acpi_dev_found(hns_enet_acpi_match[1].id))
2043 			priv->enet_ver = AE_VERSION_2;
2044 		else
2045 			return -ENXIO;
2046 
2047 		/* try to find port-idx-in-ae first */
2048 		ret = acpi_node_get_property_reference(dev->fwnode,
2049 						       "ae-handle", 0, &args);
2050 		if (ret) {
2051 			dev_err(dev, "not find ae-handle\n");
2052 			goto out_read_prop_fail;
2053 		}
2054 		priv->fwnode = acpi_fwnode_handle(args.adev);
2055 	} else {
2056 		dev_err(dev, "cannot read cfg data from OF or acpi\n");
2057 		return -ENXIO;
2058 	}
2059 
2060 	ret = device_property_read_u32(dev, "port-idx-in-ae", &port_id);
2061 	if (ret) {
2062 		/* only for old code compatible */
2063 		ret = device_property_read_u32(dev, "port-id", &port_id);
2064 		if (ret)
2065 			goto out_read_prop_fail;
2066 		/* for old dts, we need to caculate the port offset */
2067 		port_id = port_id < HNS_SRV_OFFSET ? port_id + HNS_DEBUG_OFFSET
2068 			: port_id - HNS_SRV_OFFSET;
2069 	}
2070 	priv->port_id = port_id;
2071 
2072 	hns_init_mac_addr(ndev);
2073 
2074 	ndev->watchdog_timeo = HNS_NIC_TX_TIMEOUT;
2075 	ndev->priv_flags |= IFF_UNICAST_FLT;
2076 	ndev->netdev_ops = &hns_nic_netdev_ops;
2077 	hns_ethtool_set_ops(ndev);
2078 
2079 	ndev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2080 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2081 		NETIF_F_GRO;
2082 	ndev->vlan_features |=
2083 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM;
2084 	ndev->vlan_features |= NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
2085 
2086 	/* MTU range: 68 - 9578 (v1) or 9706 (v2) */
2087 	ndev->min_mtu = MAC_MIN_MTU;
2088 	switch (priv->enet_ver) {
2089 	case AE_VERSION_2:
2090 		ndev->features |= NETIF_F_TSO | NETIF_F_TSO6;
2091 		ndev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2092 			NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2093 			NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6;
2094 		ndev->max_mtu = MAC_MAX_MTU_V2 -
2095 				(ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
2096 		break;
2097 	default:
2098 		ndev->max_mtu = MAC_MAX_MTU -
2099 				(ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
2100 		break;
2101 	}
2102 
2103 	SET_NETDEV_DEV(ndev, dev);
2104 
2105 	if (!dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64)))
2106 		dev_dbg(dev, "set mask to 64bit\n");
2107 	else
2108 		dev_err(dev, "set mask to 64bit fail!\n");
2109 
2110 	/* carrier off reporting is important to ethtool even BEFORE open */
2111 	netif_carrier_off(ndev);
2112 
2113 	setup_timer(&priv->service_timer, hns_nic_service_timer,
2114 		    (unsigned long)priv);
2115 	INIT_WORK(&priv->service_task, hns_nic_service_task);
2116 
2117 	set_bit(NIC_STATE_SERVICE_INITED, &priv->state);
2118 	clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
2119 	set_bit(NIC_STATE_DOWN, &priv->state);
2120 
2121 	if (hns_nic_try_get_ae(priv->netdev)) {
2122 		priv->notifier_block.notifier_call = hns_nic_notifier_action;
2123 		ret = hnae_register_notifier(&priv->notifier_block);
2124 		if (ret) {
2125 			dev_err(dev, "register notifier fail!\n");
2126 			goto out_notify_fail;
2127 		}
2128 		dev_dbg(dev, "has not handle, register notifier!\n");
2129 	}
2130 
2131 	return 0;
2132 
2133 out_notify_fail:
2134 	(void)cancel_work_sync(&priv->service_task);
2135 out_read_prop_fail:
2136 	free_netdev(ndev);
2137 	return ret;
2138 }
2139 
2140 static int hns_nic_dev_remove(struct platform_device *pdev)
2141 {
2142 	struct net_device *ndev = platform_get_drvdata(pdev);
2143 	struct hns_nic_priv *priv = netdev_priv(ndev);
2144 
2145 	if (ndev->reg_state != NETREG_UNINITIALIZED)
2146 		unregister_netdev(ndev);
2147 
2148 	if (priv->ring_data)
2149 		hns_nic_uninit_ring_data(priv);
2150 	priv->ring_data = NULL;
2151 
2152 	if (ndev->phydev)
2153 		phy_disconnect(ndev->phydev);
2154 
2155 	if (!IS_ERR_OR_NULL(priv->ae_handle))
2156 		hnae_put_handle(priv->ae_handle);
2157 	priv->ae_handle = NULL;
2158 	if (priv->notifier_block.notifier_call)
2159 		hnae_unregister_notifier(&priv->notifier_block);
2160 	priv->notifier_block.notifier_call = NULL;
2161 
2162 	set_bit(NIC_STATE_REMOVING, &priv->state);
2163 	(void)cancel_work_sync(&priv->service_task);
2164 
2165 	free_netdev(ndev);
2166 	return 0;
2167 }
2168 
2169 static const struct of_device_id hns_enet_of_match[] = {
2170 	{.compatible = "hisilicon,hns-nic-v1",},
2171 	{.compatible = "hisilicon,hns-nic-v2",},
2172 	{},
2173 };
2174 
2175 MODULE_DEVICE_TABLE(of, hns_enet_of_match);
2176 
2177 static struct platform_driver hns_nic_dev_driver = {
2178 	.driver = {
2179 		.name = "hns-nic",
2180 		.of_match_table = hns_enet_of_match,
2181 		.acpi_match_table = ACPI_PTR(hns_enet_acpi_match),
2182 	},
2183 	.probe = hns_nic_dev_probe,
2184 	.remove = hns_nic_dev_remove,
2185 };
2186 
2187 module_platform_driver(hns_nic_dev_driver);
2188 
2189 MODULE_DESCRIPTION("HISILICON HNS Ethernet driver");
2190 MODULE_AUTHOR("Hisilicon, Inc.");
2191 MODULE_LICENSE("GPL");
2192 MODULE_ALIAS("platform:hns-nic");
2193