GNU Linux-libre 4.19.286-gnu1
[releases.git] / drivers / net / ethernet / hisilicon / hns / hns_enet.c
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 RCB_IRQ_NOT_INITED 0
33 #define RCB_IRQ_INITED 1
34 #define HNS_BUFFER_SIZE_2048 2048
35
36 #define BD_MAX_SEND_SIZE 8191
37 #define SKB_TMP_LEN(SKB) \
38         (((SKB)->transport_header - (SKB)->mac_header) + tcp_hdrlen(SKB))
39
40 static void fill_v2_desc_hw(struct hnae_ring *ring, void *priv, int size,
41                             int send_sz, dma_addr_t dma, int frag_end,
42                             int buf_num, enum hns_desc_type type, int mtu)
43 {
44         struct hnae_desc *desc = &ring->desc[ring->next_to_use];
45         struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
46         struct iphdr *iphdr;
47         struct ipv6hdr *ipv6hdr;
48         struct sk_buff *skb;
49         __be16 protocol;
50         u8 bn_pid = 0;
51         u8 rrcfv = 0;
52         u8 ip_offset = 0;
53         u8 tvsvsn = 0;
54         u16 mss = 0;
55         u8 l4_len = 0;
56         u16 paylen = 0;
57
58         desc_cb->priv = priv;
59         desc_cb->length = size;
60         desc_cb->dma = dma;
61         desc_cb->type = type;
62
63         desc->addr = cpu_to_le64(dma);
64         desc->tx.send_size = cpu_to_le16((u16)send_sz);
65
66         /* config bd buffer end */
67         hnae_set_bit(rrcfv, HNSV2_TXD_VLD_B, 1);
68         hnae_set_field(bn_pid, HNSV2_TXD_BUFNUM_M, 0, buf_num - 1);
69
70         /* fill port_id in the tx bd for sending management pkts */
71         hnae_set_field(bn_pid, HNSV2_TXD_PORTID_M,
72                        HNSV2_TXD_PORTID_S, ring->q->handle->dport_id);
73
74         if (type == DESC_TYPE_SKB) {
75                 skb = (struct sk_buff *)priv;
76
77                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
78                         skb_reset_mac_len(skb);
79                         protocol = skb->protocol;
80                         ip_offset = ETH_HLEN;
81
82                         if (protocol == htons(ETH_P_8021Q)) {
83                                 ip_offset += VLAN_HLEN;
84                                 protocol = vlan_get_protocol(skb);
85                                 skb->protocol = protocol;
86                         }
87
88                         if (skb->protocol == htons(ETH_P_IP)) {
89                                 iphdr = ip_hdr(skb);
90                                 hnae_set_bit(rrcfv, HNSV2_TXD_L3CS_B, 1);
91                                 hnae_set_bit(rrcfv, HNSV2_TXD_L4CS_B, 1);
92
93                                 /* check for tcp/udp header */
94                                 if (iphdr->protocol == IPPROTO_TCP &&
95                                     skb_is_gso(skb)) {
96                                         hnae_set_bit(tvsvsn,
97                                                      HNSV2_TXD_TSE_B, 1);
98                                         l4_len = tcp_hdrlen(skb);
99                                         mss = skb_shinfo(skb)->gso_size;
100                                         paylen = skb->len - SKB_TMP_LEN(skb);
101                                 }
102                         } else if (skb->protocol == htons(ETH_P_IPV6)) {
103                                 hnae_set_bit(tvsvsn, HNSV2_TXD_IPV6_B, 1);
104                                 ipv6hdr = ipv6_hdr(skb);
105                                 hnae_set_bit(rrcfv, HNSV2_TXD_L4CS_B, 1);
106
107                                 /* check for tcp/udp header */
108                                 if (ipv6hdr->nexthdr == IPPROTO_TCP &&
109                                     skb_is_gso(skb) && skb_is_gso_v6(skb)) {
110                                         hnae_set_bit(tvsvsn,
111                                                      HNSV2_TXD_TSE_B, 1);
112                                         l4_len = tcp_hdrlen(skb);
113                                         mss = skb_shinfo(skb)->gso_size;
114                                         paylen = skb->len - SKB_TMP_LEN(skb);
115                                 }
116                         }
117                         desc->tx.ip_offset = ip_offset;
118                         desc->tx.tse_vlan_snap_v6_sctp_nth = tvsvsn;
119                         desc->tx.mss = cpu_to_le16(mss);
120                         desc->tx.l4_len = l4_len;
121                         desc->tx.paylen = cpu_to_le16(paylen);
122                 }
123         }
124
125         hnae_set_bit(rrcfv, HNSV2_TXD_FE_B, frag_end);
126
127         desc->tx.bn_pid = bn_pid;
128         desc->tx.ra_ri_cs_fe_vld = rrcfv;
129
130         ring_ptr_move_fw(ring, next_to_use);
131 }
132
133 static void fill_v2_desc(struct hnae_ring *ring, void *priv,
134                          int size, dma_addr_t dma, int frag_end,
135                          int buf_num, enum hns_desc_type type, int mtu)
136 {
137         fill_v2_desc_hw(ring, priv, size, size, dma, frag_end,
138                         buf_num, type, mtu);
139 }
140
141 static const struct acpi_device_id hns_enet_acpi_match[] = {
142         { "HISI00C1", 0 },
143         { "HISI00C2", 0 },
144         { },
145 };
146 MODULE_DEVICE_TABLE(acpi, hns_enet_acpi_match);
147
148 static void fill_desc(struct hnae_ring *ring, void *priv,
149                       int size, dma_addr_t dma, int frag_end,
150                       int buf_num, enum hns_desc_type type, int mtu)
151 {
152         struct hnae_desc *desc = &ring->desc[ring->next_to_use];
153         struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
154         struct sk_buff *skb;
155         __be16 protocol;
156         u32 ip_offset;
157         u32 asid_bufnum_pid = 0;
158         u32 flag_ipoffset = 0;
159
160         desc_cb->priv = priv;
161         desc_cb->length = size;
162         desc_cb->dma = dma;
163         desc_cb->type = type;
164
165         desc->addr = cpu_to_le64(dma);
166         desc->tx.send_size = cpu_to_le16((u16)size);
167
168         /*config bd buffer end */
169         flag_ipoffset |= 1 << HNS_TXD_VLD_B;
170
171         asid_bufnum_pid |= buf_num << HNS_TXD_BUFNUM_S;
172
173         if (type == DESC_TYPE_SKB) {
174                 skb = (struct sk_buff *)priv;
175
176                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
177                         protocol = skb->protocol;
178                         ip_offset = ETH_HLEN;
179
180                         /*if it is a SW VLAN check the next protocol*/
181                         if (protocol == htons(ETH_P_8021Q)) {
182                                 ip_offset += VLAN_HLEN;
183                                 protocol = vlan_get_protocol(skb);
184                                 skb->protocol = protocol;
185                         }
186
187                         if (skb->protocol == htons(ETH_P_IP)) {
188                                 flag_ipoffset |= 1 << HNS_TXD_L3CS_B;
189                                 /* check for tcp/udp header */
190                                 flag_ipoffset |= 1 << HNS_TXD_L4CS_B;
191
192                         } else if (skb->protocol == htons(ETH_P_IPV6)) {
193                                 /* ipv6 has not l3 cs, check for L4 header */
194                                 flag_ipoffset |= 1 << HNS_TXD_L4CS_B;
195                         }
196
197                         flag_ipoffset |= ip_offset << HNS_TXD_IPOFFSET_S;
198                 }
199         }
200
201         flag_ipoffset |= frag_end << HNS_TXD_FE_B;
202
203         desc->tx.asid_bufnum_pid = cpu_to_le16(asid_bufnum_pid);
204         desc->tx.flag_ipoffset = cpu_to_le32(flag_ipoffset);
205
206         ring_ptr_move_fw(ring, next_to_use);
207 }
208
209 static void unfill_desc(struct hnae_ring *ring)
210 {
211         ring_ptr_move_bw(ring, next_to_use);
212 }
213
214 static int hns_nic_maybe_stop_tx(
215         struct sk_buff **out_skb, int *bnum, struct hnae_ring *ring)
216 {
217         struct sk_buff *skb = *out_skb;
218         struct sk_buff *new_skb = NULL;
219         int buf_num;
220
221         /* no. of segments (plus a header) */
222         buf_num = skb_shinfo(skb)->nr_frags + 1;
223
224         if (unlikely(buf_num > ring->max_desc_num_per_pkt)) {
225                 if (ring_space(ring) < 1)
226                         return -EBUSY;
227
228                 new_skb = skb_copy(skb, GFP_ATOMIC);
229                 if (!new_skb)
230                         return -ENOMEM;
231
232                 dev_kfree_skb_any(skb);
233                 *out_skb = new_skb;
234                 buf_num = 1;
235         } else if (buf_num > ring_space(ring)) {
236                 return -EBUSY;
237         }
238
239         *bnum = buf_num;
240         return 0;
241 }
242
243 static int hns_nic_maybe_stop_tso(
244         struct sk_buff **out_skb, int *bnum, struct hnae_ring *ring)
245 {
246         int i;
247         int size;
248         int buf_num;
249         int frag_num;
250         struct sk_buff *skb = *out_skb;
251         struct sk_buff *new_skb = NULL;
252         struct skb_frag_struct *frag;
253
254         size = skb_headlen(skb);
255         buf_num = (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
256
257         frag_num = skb_shinfo(skb)->nr_frags;
258         for (i = 0; i < frag_num; i++) {
259                 frag = &skb_shinfo(skb)->frags[i];
260                 size = skb_frag_size(frag);
261                 buf_num += (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
262         }
263
264         if (unlikely(buf_num > ring->max_desc_num_per_pkt)) {
265                 buf_num = (skb->len + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
266                 if (ring_space(ring) < buf_num)
267                         return -EBUSY;
268                 /* manual split the send packet */
269                 new_skb = skb_copy(skb, GFP_ATOMIC);
270                 if (!new_skb)
271                         return -ENOMEM;
272                 dev_kfree_skb_any(skb);
273                 *out_skb = new_skb;
274
275         } else if (ring_space(ring) < buf_num) {
276                 return -EBUSY;
277         }
278
279         *bnum = buf_num;
280         return 0;
281 }
282
283 static void fill_tso_desc(struct hnae_ring *ring, void *priv,
284                           int size, dma_addr_t dma, int frag_end,
285                           int buf_num, enum hns_desc_type type, int mtu)
286 {
287         int frag_buf_num;
288         int sizeoflast;
289         int k;
290
291         frag_buf_num = (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
292         sizeoflast = size % BD_MAX_SEND_SIZE;
293         sizeoflast = sizeoflast ? sizeoflast : BD_MAX_SEND_SIZE;
294
295         /* when the frag size is bigger than hardware, split this frag */
296         for (k = 0; k < frag_buf_num; k++)
297                 fill_v2_desc_hw(ring, priv, k == 0 ? size : 0,
298                                 (k == frag_buf_num - 1) ?
299                                         sizeoflast : BD_MAX_SEND_SIZE,
300                                 dma + BD_MAX_SEND_SIZE * k,
301                                 frag_end && (k == frag_buf_num - 1) ? 1 : 0,
302                                 buf_num,
303                                 (type == DESC_TYPE_SKB && !k) ?
304                                         DESC_TYPE_SKB : DESC_TYPE_PAGE,
305                                 mtu);
306 }
307
308 netdev_tx_t hns_nic_net_xmit_hw(struct net_device *ndev,
309                                 struct sk_buff *skb,
310                                 struct hns_nic_ring_data *ring_data)
311 {
312         struct hns_nic_priv *priv = netdev_priv(ndev);
313         struct hnae_ring *ring = ring_data->ring;
314         struct device *dev = ring_to_dev(ring);
315         struct netdev_queue *dev_queue;
316         struct skb_frag_struct *frag;
317         int buf_num;
318         int seg_num;
319         dma_addr_t dma;
320         int size, next_to_use;
321         int i;
322
323         switch (priv->ops.maybe_stop_tx(&skb, &buf_num, ring)) {
324         case -EBUSY:
325                 ring->stats.tx_busy++;
326                 goto out_net_tx_busy;
327         case -ENOMEM:
328                 ring->stats.sw_err_cnt++;
329                 netdev_err(ndev, "no memory to xmit!\n");
330                 goto out_err_tx_ok;
331         default:
332                 break;
333         }
334
335         /* no. of segments (plus a header) */
336         seg_num = skb_shinfo(skb)->nr_frags + 1;
337         next_to_use = ring->next_to_use;
338
339         /* fill the first part */
340         size = skb_headlen(skb);
341         dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
342         if (dma_mapping_error(dev, dma)) {
343                 netdev_err(ndev, "TX head DMA map failed\n");
344                 ring->stats.sw_err_cnt++;
345                 goto out_err_tx_ok;
346         }
347         priv->ops.fill_desc(ring, skb, size, dma, seg_num == 1 ? 1 : 0,
348                             buf_num, DESC_TYPE_SKB, ndev->mtu);
349
350         /* fill the fragments */
351         for (i = 1; i < seg_num; i++) {
352                 frag = &skb_shinfo(skb)->frags[i - 1];
353                 size = skb_frag_size(frag);
354                 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
355                 if (dma_mapping_error(dev, dma)) {
356                         netdev_err(ndev, "TX frag(%d) DMA map failed\n", i);
357                         ring->stats.sw_err_cnt++;
358                         goto out_map_frag_fail;
359                 }
360                 priv->ops.fill_desc(ring, skb_frag_page(frag), size, dma,
361                                     seg_num - 1 == i ? 1 : 0, buf_num,
362                                     DESC_TYPE_PAGE, ndev->mtu);
363         }
364
365         /*complete translate all packets*/
366         dev_queue = netdev_get_tx_queue(ndev, skb->queue_mapping);
367         netdev_tx_sent_queue(dev_queue, skb->len);
368
369         netif_trans_update(ndev);
370         ndev->stats.tx_bytes += skb->len;
371         ndev->stats.tx_packets++;
372
373         wmb(); /* commit all data before submit */
374         assert(skb->queue_mapping < priv->ae_handle->q_num);
375         hnae_queue_xmit(priv->ae_handle->qs[skb->queue_mapping], buf_num);
376
377         return NETDEV_TX_OK;
378
379 out_map_frag_fail:
380
381         while (ring->next_to_use != next_to_use) {
382                 unfill_desc(ring);
383                 if (ring->next_to_use != next_to_use)
384                         dma_unmap_page(dev,
385                                        ring->desc_cb[ring->next_to_use].dma,
386                                        ring->desc_cb[ring->next_to_use].length,
387                                        DMA_TO_DEVICE);
388                 else
389                         dma_unmap_single(dev,
390                                          ring->desc_cb[next_to_use].dma,
391                                          ring->desc_cb[next_to_use].length,
392                                          DMA_TO_DEVICE);
393         }
394
395 out_err_tx_ok:
396
397         dev_kfree_skb_any(skb);
398         return NETDEV_TX_OK;
399
400 out_net_tx_busy:
401
402         netif_stop_subqueue(ndev, skb->queue_mapping);
403
404         /* Herbert's original patch had:
405          *  smp_mb__after_netif_stop_queue();
406          * but since that doesn't exist yet, just open code it.
407          */
408         smp_mb();
409         return NETDEV_TX_BUSY;
410 }
411
412 static void hns_nic_reuse_page(struct sk_buff *skb, int i,
413                                struct hnae_ring *ring, int pull_len,
414                                struct hnae_desc_cb *desc_cb)
415 {
416         struct hnae_desc *desc;
417         u32 truesize;
418         int size;
419         int last_offset;
420         bool twobufs;
421
422         twobufs = ((PAGE_SIZE < 8192) &&
423                 hnae_buf_size(ring) == HNS_BUFFER_SIZE_2048);
424
425         desc = &ring->desc[ring->next_to_clean];
426         size = le16_to_cpu(desc->rx.size);
427
428         if (twobufs) {
429                 truesize = hnae_buf_size(ring);
430         } else {
431                 truesize = ALIGN(size, L1_CACHE_BYTES);
432                 last_offset = hnae_page_size(ring) - hnae_buf_size(ring);
433         }
434
435         skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
436                         size - pull_len, truesize);
437
438          /* avoid re-using remote pages,flag default unreuse */
439         if (unlikely(page_to_nid(desc_cb->priv) != numa_node_id()))
440                 return;
441
442         if (twobufs) {
443                 /* if we are only owner of page we can reuse it */
444                 if (likely(page_count(desc_cb->priv) == 1)) {
445                         /* flip page offset to other buffer */
446                         desc_cb->page_offset ^= truesize;
447
448                         desc_cb->reuse_flag = 1;
449                         /* bump ref count on page before it is given*/
450                         get_page(desc_cb->priv);
451                 }
452                 return;
453         }
454
455         /* move offset up to the next cache line */
456         desc_cb->page_offset += truesize;
457
458         if (desc_cb->page_offset <= last_offset) {
459                 desc_cb->reuse_flag = 1;
460                 /* bump ref count on page before it is given*/
461                 get_page(desc_cb->priv);
462         }
463 }
464
465 static void get_v2rx_desc_bnum(u32 bnum_flag, int *out_bnum)
466 {
467         *out_bnum = hnae_get_field(bnum_flag,
468                                    HNS_RXD_BUFNUM_M, HNS_RXD_BUFNUM_S) + 1;
469 }
470
471 static void get_rx_desc_bnum(u32 bnum_flag, int *out_bnum)
472 {
473         *out_bnum = hnae_get_field(bnum_flag,
474                                    HNS_RXD_BUFNUM_M, HNS_RXD_BUFNUM_S);
475 }
476
477 static void hns_nic_rx_checksum(struct hns_nic_ring_data *ring_data,
478                                 struct sk_buff *skb, u32 flag)
479 {
480         struct net_device *netdev = ring_data->napi.dev;
481         u32 l3id;
482         u32 l4id;
483
484         /* check if RX checksum offload is enabled */
485         if (unlikely(!(netdev->features & NETIF_F_RXCSUM)))
486                 return;
487
488         /* In hardware, we only support checksum for the following protocols:
489          * 1) IPv4,
490          * 2) TCP(over IPv4 or IPv6),
491          * 3) UDP(over IPv4 or IPv6),
492          * 4) SCTP(over IPv4 or IPv6)
493          * but we support many L3(IPv4, IPv6, MPLS, PPPoE etc) and L4(TCP,
494          * UDP, GRE, SCTP, IGMP, ICMP etc.) protocols.
495          *
496          * Hardware limitation:
497          * Our present hardware RX Descriptor lacks L3/L4 checksum "Status &
498          * Error" bit (which usually can be used to indicate whether checksum
499          * was calculated by the hardware and if there was any error encountered
500          * during checksum calculation).
501          *
502          * Software workaround:
503          * We do get info within the RX descriptor about the kind of L3/L4
504          * protocol coming in the packet and the error status. These errors
505          * might not just be checksum errors but could be related to version,
506          * length of IPv4, UDP, TCP etc.
507          * Because there is no-way of knowing if it is a L3/L4 error due to bad
508          * checksum or any other L3/L4 error, we will not (cannot) convey
509          * checksum status for such cases to upper stack and will not maintain
510          * the RX L3/L4 checksum counters as well.
511          */
512
513         l3id = hnae_get_field(flag, HNS_RXD_L3ID_M, HNS_RXD_L3ID_S);
514         l4id = hnae_get_field(flag, HNS_RXD_L4ID_M, HNS_RXD_L4ID_S);
515
516         /*  check L3 protocol for which checksum is supported */
517         if ((l3id != HNS_RX_FLAG_L3ID_IPV4) && (l3id != HNS_RX_FLAG_L3ID_IPV6))
518                 return;
519
520         /* check for any(not just checksum)flagged L3 protocol errors */
521         if (unlikely(hnae_get_bit(flag, HNS_RXD_L3E_B)))
522                 return;
523
524         /* we do not support checksum of fragmented packets */
525         if (unlikely(hnae_get_bit(flag, HNS_RXD_FRAG_B)))
526                 return;
527
528         /*  check L4 protocol for which checksum is supported */
529         if ((l4id != HNS_RX_FLAG_L4ID_TCP) &&
530             (l4id != HNS_RX_FLAG_L4ID_UDP) &&
531             (l4id != HNS_RX_FLAG_L4ID_SCTP))
532                 return;
533
534         /* check for any(not just checksum)flagged L4 protocol errors */
535         if (unlikely(hnae_get_bit(flag, HNS_RXD_L4E_B)))
536                 return;
537
538         /* now, this has to be a packet with valid RX checksum */
539         skb->ip_summed = CHECKSUM_UNNECESSARY;
540 }
541
542 static int hns_nic_poll_rx_skb(struct hns_nic_ring_data *ring_data,
543                                struct sk_buff **out_skb, int *out_bnum)
544 {
545         struct hnae_ring *ring = ring_data->ring;
546         struct net_device *ndev = ring_data->napi.dev;
547         struct hns_nic_priv *priv = netdev_priv(ndev);
548         struct sk_buff *skb;
549         struct hnae_desc *desc;
550         struct hnae_desc_cb *desc_cb;
551         unsigned char *va;
552         int bnum, length, i;
553         int pull_len;
554         u32 bnum_flag;
555
556         desc = &ring->desc[ring->next_to_clean];
557         desc_cb = &ring->desc_cb[ring->next_to_clean];
558
559         prefetch(desc);
560
561         va = (unsigned char *)desc_cb->buf + desc_cb->page_offset;
562
563         /* prefetch first cache line of first page */
564         prefetch(va);
565 #if L1_CACHE_BYTES < 128
566         prefetch(va + L1_CACHE_BYTES);
567 #endif
568
569         skb = *out_skb = napi_alloc_skb(&ring_data->napi,
570                                         HNS_RX_HEAD_SIZE);
571         if (unlikely(!skb)) {
572                 ring->stats.sw_err_cnt++;
573                 return -ENOMEM;
574         }
575
576         prefetchw(skb->data);
577         length = le16_to_cpu(desc->rx.pkt_len);
578         bnum_flag = le32_to_cpu(desc->rx.ipoff_bnum_pid_flag);
579         priv->ops.get_rxd_bnum(bnum_flag, &bnum);
580         *out_bnum = bnum;
581
582         if (length <= HNS_RX_HEAD_SIZE) {
583                 memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
584
585                 /* we can reuse buffer as-is, just make sure it is local */
586                 if (likely(page_to_nid(desc_cb->priv) == numa_node_id()))
587                         desc_cb->reuse_flag = 1;
588                 else /* this page cannot be reused so discard it */
589                         put_page(desc_cb->priv);
590
591                 ring_ptr_move_fw(ring, next_to_clean);
592
593                 if (unlikely(bnum != 1)) { /* check err*/
594                         *out_bnum = 1;
595                         goto out_bnum_err;
596                 }
597         } else {
598                 ring->stats.seg_pkt_cnt++;
599
600                 pull_len = eth_get_headlen(va, HNS_RX_HEAD_SIZE);
601                 memcpy(__skb_put(skb, pull_len), va,
602                        ALIGN(pull_len, sizeof(long)));
603
604                 hns_nic_reuse_page(skb, 0, ring, pull_len, desc_cb);
605                 ring_ptr_move_fw(ring, next_to_clean);
606
607                 if (unlikely(bnum >= (int)MAX_SKB_FRAGS)) { /* check err*/
608                         *out_bnum = 1;
609                         goto out_bnum_err;
610                 }
611                 for (i = 1; i < bnum; i++) {
612                         desc = &ring->desc[ring->next_to_clean];
613                         desc_cb = &ring->desc_cb[ring->next_to_clean];
614
615                         hns_nic_reuse_page(skb, i, ring, 0, desc_cb);
616                         ring_ptr_move_fw(ring, next_to_clean);
617                 }
618         }
619
620         /* check except process, free skb and jump the desc */
621         if (unlikely((!bnum) || (bnum > ring->max_desc_num_per_pkt))) {
622 out_bnum_err:
623                 *out_bnum = *out_bnum ? *out_bnum : 1; /* ntc moved,cannot 0*/
624                 netdev_err(ndev, "invalid bnum(%d,%d,%d,%d),%016llx,%016llx\n",
625                            bnum, ring->max_desc_num_per_pkt,
626                            length, (int)MAX_SKB_FRAGS,
627                            ((u64 *)desc)[0], ((u64 *)desc)[1]);
628                 ring->stats.err_bd_num++;
629                 dev_kfree_skb_any(skb);
630                 return -EDOM;
631         }
632
633         bnum_flag = le32_to_cpu(desc->rx.ipoff_bnum_pid_flag);
634
635         if (unlikely(!hnae_get_bit(bnum_flag, HNS_RXD_VLD_B))) {
636                 netdev_err(ndev, "no valid bd,%016llx,%016llx\n",
637                            ((u64 *)desc)[0], ((u64 *)desc)[1]);
638                 ring->stats.non_vld_descs++;
639                 dev_kfree_skb_any(skb);
640                 return -EINVAL;
641         }
642
643         if (unlikely((!desc->rx.pkt_len) ||
644                      hnae_get_bit(bnum_flag, HNS_RXD_DROP_B))) {
645                 ring->stats.err_pkt_len++;
646                 dev_kfree_skb_any(skb);
647                 return -EFAULT;
648         }
649
650         if (unlikely(hnae_get_bit(bnum_flag, HNS_RXD_L2E_B))) {
651                 ring->stats.l2_err++;
652                 dev_kfree_skb_any(skb);
653                 return -EFAULT;
654         }
655
656         ring->stats.rx_pkts++;
657         ring->stats.rx_bytes += skb->len;
658
659         /* indicate to upper stack if our hardware has already calculated
660          * the RX checksum
661          */
662         hns_nic_rx_checksum(ring_data, skb, bnum_flag);
663
664         return 0;
665 }
666
667 static void
668 hns_nic_alloc_rx_buffers(struct hns_nic_ring_data *ring_data, int cleand_count)
669 {
670         int i, ret;
671         struct hnae_desc_cb res_cbs;
672         struct hnae_desc_cb *desc_cb;
673         struct hnae_ring *ring = ring_data->ring;
674         struct net_device *ndev = ring_data->napi.dev;
675
676         for (i = 0; i < cleand_count; i++) {
677                 desc_cb = &ring->desc_cb[ring->next_to_use];
678                 if (desc_cb->reuse_flag) {
679                         ring->stats.reuse_pg_cnt++;
680                         hnae_reuse_buffer(ring, ring->next_to_use);
681                 } else {
682                         ret = hnae_reserve_buffer_map(ring, &res_cbs);
683                         if (ret) {
684                                 ring->stats.sw_err_cnt++;
685                                 netdev_err(ndev, "hnae reserve buffer map failed.\n");
686                                 break;
687                         }
688                         hnae_replace_buffer(ring, ring->next_to_use, &res_cbs);
689                 }
690
691                 ring_ptr_move_fw(ring, next_to_use);
692         }
693
694         wmb(); /* make all data has been write before submit */
695         writel_relaxed(i, ring->io_base + RCB_REG_HEAD);
696 }
697
698 /* return error number for error or number of desc left to take
699  */
700 static void hns_nic_rx_up_pro(struct hns_nic_ring_data *ring_data,
701                               struct sk_buff *skb)
702 {
703         struct net_device *ndev = ring_data->napi.dev;
704
705         skb->protocol = eth_type_trans(skb, ndev);
706         (void)napi_gro_receive(&ring_data->napi, skb);
707 }
708
709 static int hns_desc_unused(struct hnae_ring *ring)
710 {
711         int ntc = ring->next_to_clean;
712         int ntu = ring->next_to_use;
713
714         return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
715 }
716
717 #define HNS_LOWEST_LATENCY_RATE         27      /* 27 MB/s */
718 #define HNS_LOW_LATENCY_RATE                    80      /* 80 MB/s */
719
720 #define HNS_COAL_BDNUM                  3
721
722 static u32 hns_coal_rx_bdnum(struct hnae_ring *ring)
723 {
724         bool coal_enable = ring->q->handle->coal_adapt_en;
725
726         if (coal_enable &&
727             ring->coal_last_rx_bytes > HNS_LOWEST_LATENCY_RATE)
728                 return HNS_COAL_BDNUM;
729         else
730                 return 0;
731 }
732
733 static void hns_update_rx_rate(struct hnae_ring *ring)
734 {
735         bool coal_enable = ring->q->handle->coal_adapt_en;
736         u32 time_passed_ms;
737         u64 total_bytes;
738
739         if (!coal_enable ||
740             time_before(jiffies, ring->coal_last_jiffies + (HZ >> 4)))
741                 return;
742
743         /* ring->stats.rx_bytes overflowed */
744         if (ring->coal_last_rx_bytes > ring->stats.rx_bytes) {
745                 ring->coal_last_rx_bytes = ring->stats.rx_bytes;
746                 ring->coal_last_jiffies = jiffies;
747                 return;
748         }
749
750         total_bytes = ring->stats.rx_bytes - ring->coal_last_rx_bytes;
751         time_passed_ms = jiffies_to_msecs(jiffies - ring->coal_last_jiffies);
752         do_div(total_bytes, time_passed_ms);
753         ring->coal_rx_rate = total_bytes >> 10;
754
755         ring->coal_last_rx_bytes = ring->stats.rx_bytes;
756         ring->coal_last_jiffies = jiffies;
757 }
758
759 /**
760  * smooth_alg - smoothing algrithm for adjusting coalesce parameter
761  **/
762 static u32 smooth_alg(u32 new_param, u32 old_param)
763 {
764         u32 gap = (new_param > old_param) ? new_param - old_param
765                                           : old_param - new_param;
766
767         if (gap > 8)
768                 gap >>= 3;
769
770         if (new_param > old_param)
771                 return old_param + gap;
772         else
773                 return old_param - gap;
774 }
775
776 /**
777  * hns_nic_adp_coalesce - self adapte coalesce according to rx rate
778  * @ring_data: pointer to hns_nic_ring_data
779  **/
780 static void hns_nic_adpt_coalesce(struct hns_nic_ring_data *ring_data)
781 {
782         struct hnae_ring *ring = ring_data->ring;
783         struct hnae_handle *handle = ring->q->handle;
784         u32 new_coal_param, old_coal_param = ring->coal_param;
785
786         if (ring->coal_rx_rate < HNS_LOWEST_LATENCY_RATE)
787                 new_coal_param = HNAE_LOWEST_LATENCY_COAL_PARAM;
788         else if (ring->coal_rx_rate < HNS_LOW_LATENCY_RATE)
789                 new_coal_param = HNAE_LOW_LATENCY_COAL_PARAM;
790         else
791                 new_coal_param = HNAE_BULK_LATENCY_COAL_PARAM;
792
793         if (new_coal_param == old_coal_param &&
794             new_coal_param == handle->coal_param)
795                 return;
796
797         new_coal_param = smooth_alg(new_coal_param, old_coal_param);
798         ring->coal_param = new_coal_param;
799
800         /**
801          * Because all ring in one port has one coalesce param, when one ring
802          * calculate its own coalesce param, it cannot write to hardware at
803          * once. There are three conditions as follows:
804          *       1. current ring's coalesce param is larger than the hardware.
805          *       2. or ring which adapt last time can change again.
806          *       3. timeout.
807          */
808         if (new_coal_param == handle->coal_param) {
809                 handle->coal_last_jiffies = jiffies;
810                 handle->coal_ring_idx = ring_data->queue_index;
811         } else if (new_coal_param > handle->coal_param ||
812                    handle->coal_ring_idx == ring_data->queue_index ||
813                    time_after(jiffies, handle->coal_last_jiffies + (HZ >> 4))) {
814                 handle->dev->ops->set_coalesce_usecs(handle,
815                                         new_coal_param);
816                 handle->dev->ops->set_coalesce_frames(handle,
817                                         1, new_coal_param);
818                 handle->coal_param = new_coal_param;
819                 handle->coal_ring_idx = ring_data->queue_index;
820                 handle->coal_last_jiffies = jiffies;
821         }
822 }
823
824 static int hns_nic_rx_poll_one(struct hns_nic_ring_data *ring_data,
825                                int budget, void *v)
826 {
827         struct hnae_ring *ring = ring_data->ring;
828         struct sk_buff *skb;
829         int num, bnum;
830 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
831         int recv_pkts, recv_bds, clean_count, err;
832         int unused_count = hns_desc_unused(ring);
833
834         num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
835         rmb(); /* make sure num taken effect before the other data is touched */
836
837         recv_pkts = 0, recv_bds = 0, clean_count = 0;
838         num -= unused_count;
839
840         while (recv_pkts < budget && recv_bds < num) {
841                 /* reuse or realloc buffers */
842                 if (clean_count + unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
843                         hns_nic_alloc_rx_buffers(ring_data,
844                                                  clean_count + unused_count);
845                         clean_count = 0;
846                         unused_count = hns_desc_unused(ring);
847                 }
848
849                 /* poll one pkt */
850                 err = hns_nic_poll_rx_skb(ring_data, &skb, &bnum);
851                 if (unlikely(!skb)) /* this fault cannot be repaired */
852                         goto out;
853
854                 recv_bds += bnum;
855                 clean_count += bnum;
856                 if (unlikely(err)) {  /* do jump the err */
857                         recv_pkts++;
858                         continue;
859                 }
860
861                 /* do update ip stack process*/
862                 ((void (*)(struct hns_nic_ring_data *, struct sk_buff *))v)(
863                                                         ring_data, skb);
864                 recv_pkts++;
865         }
866
867 out:
868         /* make all data has been write before submit */
869         if (clean_count + unused_count > 0)
870                 hns_nic_alloc_rx_buffers(ring_data,
871                                          clean_count + unused_count);
872
873         return recv_pkts;
874 }
875
876 static bool hns_nic_rx_fini_pro(struct hns_nic_ring_data *ring_data)
877 {
878         struct hnae_ring *ring = ring_data->ring;
879         int num = 0;
880         bool rx_stopped;
881
882         hns_update_rx_rate(ring);
883
884         /* for hardware bug fixed */
885         ring_data->ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
886         num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
887
888         if (num <= hns_coal_rx_bdnum(ring)) {
889                 if (ring->q->handle->coal_adapt_en)
890                         hns_nic_adpt_coalesce(ring_data);
891
892                 rx_stopped = true;
893         } else {
894                 ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
895                         ring_data->ring, 1);
896
897                 rx_stopped = false;
898         }
899
900         return rx_stopped;
901 }
902
903 static bool hns_nic_rx_fini_pro_v2(struct hns_nic_ring_data *ring_data)
904 {
905         struct hnae_ring *ring = ring_data->ring;
906         int num;
907
908         hns_update_rx_rate(ring);
909         num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
910
911         if (num <= hns_coal_rx_bdnum(ring)) {
912                 if (ring->q->handle->coal_adapt_en)
913                         hns_nic_adpt_coalesce(ring_data);
914
915                 return true;
916         }
917
918         return false;
919 }
920
921 static inline void hns_nic_reclaim_one_desc(struct hnae_ring *ring,
922                                             int *bytes, int *pkts)
923 {
924         struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
925
926         (*pkts) += (desc_cb->type == DESC_TYPE_SKB);
927         (*bytes) += desc_cb->length;
928         /* desc_cb will be cleaned, after hnae_free_buffer_detach*/
929         hnae_free_buffer_detach(ring, ring->next_to_clean);
930
931         ring_ptr_move_fw(ring, next_to_clean);
932 }
933
934 static int is_valid_clean_head(struct hnae_ring *ring, int h)
935 {
936         int u = ring->next_to_use;
937         int c = ring->next_to_clean;
938
939         if (unlikely(h > ring->desc_num))
940                 return 0;
941
942         assert(u > 0 && u < ring->desc_num);
943         assert(c > 0 && c < ring->desc_num);
944         assert(u != c && h != c); /* must be checked before call this func */
945
946         return u > c ? (h > c && h <= u) : (h > c || h <= u);
947 }
948
949 /* reclaim all desc in one budget
950  * return error or number of desc left
951  */
952 static int hns_nic_tx_poll_one(struct hns_nic_ring_data *ring_data,
953                                int budget, void *v)
954 {
955         struct hnae_ring *ring = ring_data->ring;
956         struct net_device *ndev = ring_data->napi.dev;
957         struct netdev_queue *dev_queue;
958         struct hns_nic_priv *priv = netdev_priv(ndev);
959         int head;
960         int bytes, pkts;
961
962         head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
963         rmb(); /* make sure head is ready before touch any data */
964
965         if (is_ring_empty(ring) || head == ring->next_to_clean)
966                 return 0; /* no data to poll */
967
968         if (!is_valid_clean_head(ring, head)) {
969                 netdev_err(ndev, "wrong head (%d, %d-%d)\n", head,
970                            ring->next_to_use, ring->next_to_clean);
971                 ring->stats.io_err_cnt++;
972                 return -EIO;
973         }
974
975         bytes = 0;
976         pkts = 0;
977         while (head != ring->next_to_clean) {
978                 hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
979                 /* issue prefetch for next Tx descriptor */
980                 prefetch(&ring->desc_cb[ring->next_to_clean]);
981         }
982         /* update tx ring statistics. */
983         ring->stats.tx_pkts += pkts;
984         ring->stats.tx_bytes += bytes;
985
986         dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
987         netdev_tx_completed_queue(dev_queue, pkts, bytes);
988
989         if (unlikely(priv->link && !netif_carrier_ok(ndev)))
990                 netif_carrier_on(ndev);
991
992         if (unlikely(pkts && netif_carrier_ok(ndev) &&
993                      (ring_space(ring) >= ring->max_desc_num_per_pkt * 2))) {
994                 /* Make sure that anybody stopping the queue after this
995                  * sees the new next_to_clean.
996                  */
997                 smp_mb();
998                 if (netif_tx_queue_stopped(dev_queue) &&
999                     !test_bit(NIC_STATE_DOWN, &priv->state)) {
1000                         netif_tx_wake_queue(dev_queue);
1001                         ring->stats.restart_queue++;
1002                 }
1003         }
1004         return 0;
1005 }
1006
1007 static bool hns_nic_tx_fini_pro(struct hns_nic_ring_data *ring_data)
1008 {
1009         struct hnae_ring *ring = ring_data->ring;
1010         int head;
1011
1012         ring_data->ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
1013
1014         head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1015
1016         if (head != ring->next_to_clean) {
1017                 ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1018                         ring_data->ring, 1);
1019
1020                 return false;
1021         } else {
1022                 return true;
1023         }
1024 }
1025
1026 static bool hns_nic_tx_fini_pro_v2(struct hns_nic_ring_data *ring_data)
1027 {
1028         struct hnae_ring *ring = ring_data->ring;
1029         int head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1030
1031         if (head == ring->next_to_clean)
1032                 return true;
1033         else
1034                 return false;
1035 }
1036
1037 static void hns_nic_tx_clr_all_bufs(struct hns_nic_ring_data *ring_data)
1038 {
1039         struct hnae_ring *ring = ring_data->ring;
1040         struct net_device *ndev = ring_data->napi.dev;
1041         struct netdev_queue *dev_queue;
1042         int head;
1043         int bytes, pkts;
1044
1045         head = ring->next_to_use; /* ntu :soft setted ring position*/
1046         bytes = 0;
1047         pkts = 0;
1048         while (head != ring->next_to_clean)
1049                 hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
1050
1051         dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
1052         netdev_tx_reset_queue(dev_queue);
1053 }
1054
1055 static int hns_nic_common_poll(struct napi_struct *napi, int budget)
1056 {
1057         int clean_complete = 0;
1058         struct hns_nic_ring_data *ring_data =
1059                 container_of(napi, struct hns_nic_ring_data, napi);
1060         struct hnae_ring *ring = ring_data->ring;
1061
1062         clean_complete += ring_data->poll_one(
1063                                 ring_data, budget - clean_complete,
1064                                 ring_data->ex_process);
1065
1066         if (clean_complete < budget) {
1067                 if (ring_data->fini_process(ring_data)) {
1068                         napi_complete(napi);
1069                         ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
1070                 } else {
1071                         return budget;
1072                 }
1073         }
1074
1075         return clean_complete;
1076 }
1077
1078 static irqreturn_t hns_irq_handle(int irq, void *dev)
1079 {
1080         struct hns_nic_ring_data *ring_data = (struct hns_nic_ring_data *)dev;
1081
1082         ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1083                 ring_data->ring, 1);
1084         napi_schedule(&ring_data->napi);
1085
1086         return IRQ_HANDLED;
1087 }
1088
1089 /**
1090  *hns_nic_adjust_link - adjust net work mode by the phy stat or new param
1091  *@ndev: net device
1092  */
1093 static void hns_nic_adjust_link(struct net_device *ndev)
1094 {
1095         struct hns_nic_priv *priv = netdev_priv(ndev);
1096         struct hnae_handle *h = priv->ae_handle;
1097         int state = 1;
1098
1099         /* If there is no phy, do not need adjust link */
1100         if (ndev->phydev) {
1101                 /* When phy link down, do nothing */
1102                 if (ndev->phydev->link == 0)
1103                         return;
1104
1105                 if (h->dev->ops->need_adjust_link(h, ndev->phydev->speed,
1106                                                   ndev->phydev->duplex)) {
1107                         /* because Hi161X chip don't support to change gmac
1108                          * speed and duplex with traffic. Delay 200ms to
1109                          * make sure there is no more data in chip FIFO.
1110                          */
1111                         netif_carrier_off(ndev);
1112                         msleep(200);
1113                         h->dev->ops->adjust_link(h, ndev->phydev->speed,
1114                                                  ndev->phydev->duplex);
1115                         netif_carrier_on(ndev);
1116                 }
1117         }
1118
1119         state = state && h->dev->ops->get_status(h);
1120
1121         if (state != priv->link) {
1122                 if (state) {
1123                         netif_carrier_on(ndev);
1124                         netif_tx_wake_all_queues(ndev);
1125                         netdev_info(ndev, "link up\n");
1126                 } else {
1127                         netif_carrier_off(ndev);
1128                         netdev_info(ndev, "link down\n");
1129                 }
1130                 priv->link = state;
1131         }
1132 }
1133
1134 /**
1135  *hns_nic_init_phy - init phy
1136  *@ndev: net device
1137  *@h: ae handle
1138  * Return 0 on success, negative on failure
1139  */
1140 int hns_nic_init_phy(struct net_device *ndev, struct hnae_handle *h)
1141 {
1142         struct phy_device *phy_dev = h->phy_dev;
1143         int ret;
1144
1145         if (!h->phy_dev)
1146                 return 0;
1147
1148         phy_dev->supported &= h->if_support;
1149         phy_dev->advertising = phy_dev->supported;
1150
1151         if (h->phy_if == PHY_INTERFACE_MODE_XGMII)
1152                 phy_dev->autoneg = false;
1153
1154         if (h->phy_if != PHY_INTERFACE_MODE_XGMII) {
1155                 phy_dev->dev_flags = 0;
1156
1157                 ret = phy_connect_direct(ndev, phy_dev, hns_nic_adjust_link,
1158                                          h->phy_if);
1159         } else {
1160                 ret = phy_attach_direct(ndev, phy_dev, 0, h->phy_if);
1161         }
1162         if (unlikely(ret))
1163                 return -ENODEV;
1164
1165         return 0;
1166 }
1167
1168 static int hns_nic_ring_open(struct net_device *netdev, int idx)
1169 {
1170         struct hns_nic_priv *priv = netdev_priv(netdev);
1171         struct hnae_handle *h = priv->ae_handle;
1172
1173         napi_enable(&priv->ring_data[idx].napi);
1174
1175         enable_irq(priv->ring_data[idx].ring->irq);
1176         h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 0);
1177
1178         return 0;
1179 }
1180
1181 static int hns_nic_net_set_mac_address(struct net_device *ndev, void *p)
1182 {
1183         struct hns_nic_priv *priv = netdev_priv(ndev);
1184         struct hnae_handle *h = priv->ae_handle;
1185         struct sockaddr *mac_addr = p;
1186         int ret;
1187
1188         if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1189                 return -EADDRNOTAVAIL;
1190
1191         ret = h->dev->ops->set_mac_addr(h, mac_addr->sa_data);
1192         if (ret) {
1193                 netdev_err(ndev, "set_mac_address fail, ret=%d!\n", ret);
1194                 return ret;
1195         }
1196
1197         memcpy(ndev->dev_addr, mac_addr->sa_data, ndev->addr_len);
1198
1199         return 0;
1200 }
1201
1202 static void hns_nic_update_stats(struct net_device *netdev)
1203 {
1204         struct hns_nic_priv *priv = netdev_priv(netdev);
1205         struct hnae_handle *h = priv->ae_handle;
1206
1207         h->dev->ops->update_stats(h, &netdev->stats);
1208 }
1209
1210 /* set mac addr if it is configed. or leave it to the AE driver */
1211 static void hns_init_mac_addr(struct net_device *ndev)
1212 {
1213         struct hns_nic_priv *priv = netdev_priv(ndev);
1214
1215         if (!device_get_mac_address(priv->dev, ndev->dev_addr, ETH_ALEN)) {
1216                 eth_hw_addr_random(ndev);
1217                 dev_warn(priv->dev, "No valid mac, use random mac %pM",
1218                          ndev->dev_addr);
1219         }
1220 }
1221
1222 static void hns_nic_ring_close(struct net_device *netdev, int idx)
1223 {
1224         struct hns_nic_priv *priv = netdev_priv(netdev);
1225         struct hnae_handle *h = priv->ae_handle;
1226
1227         h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 1);
1228         disable_irq(priv->ring_data[idx].ring->irq);
1229
1230         napi_disable(&priv->ring_data[idx].napi);
1231 }
1232
1233 static int hns_nic_init_affinity_mask(int q_num, int ring_idx,
1234                                       struct hnae_ring *ring, cpumask_t *mask)
1235 {
1236         int cpu;
1237
1238         /* Diffrent irq banlance between 16core and 32core.
1239          * The cpu mask set by ring index according to the ring flag
1240          * which indicate the ring is tx or rx.
1241          */
1242         if (q_num == num_possible_cpus()) {
1243                 if (is_tx_ring(ring))
1244                         cpu = ring_idx;
1245                 else
1246                         cpu = ring_idx - q_num;
1247         } else {
1248                 if (is_tx_ring(ring))
1249                         cpu = ring_idx * 2;
1250                 else
1251                         cpu = (ring_idx - q_num) * 2 + 1;
1252         }
1253
1254         cpumask_clear(mask);
1255         cpumask_set_cpu(cpu, mask);
1256
1257         return cpu;
1258 }
1259
1260 static void hns_nic_free_irq(int q_num, struct hns_nic_priv *priv)
1261 {
1262         int i;
1263
1264         for (i = 0; i < q_num * 2; i++) {
1265                 if (priv->ring_data[i].ring->irq_init_flag == RCB_IRQ_INITED) {
1266                         irq_set_affinity_hint(priv->ring_data[i].ring->irq,
1267                                               NULL);
1268                         free_irq(priv->ring_data[i].ring->irq,
1269                                  &priv->ring_data[i]);
1270                         priv->ring_data[i].ring->irq_init_flag =
1271                                 RCB_IRQ_NOT_INITED;
1272                 }
1273         }
1274 }
1275
1276 static int hns_nic_init_irq(struct hns_nic_priv *priv)
1277 {
1278         struct hnae_handle *h = priv->ae_handle;
1279         struct hns_nic_ring_data *rd;
1280         int i;
1281         int ret;
1282         int cpu;
1283
1284         for (i = 0; i < h->q_num * 2; i++) {
1285                 rd = &priv->ring_data[i];
1286
1287                 if (rd->ring->irq_init_flag == RCB_IRQ_INITED)
1288                         break;
1289
1290                 snprintf(rd->ring->ring_name, RCB_RING_NAME_LEN,
1291                          "%s-%s%d", priv->netdev->name,
1292                          (is_tx_ring(rd->ring) ? "tx" : "rx"), rd->queue_index);
1293
1294                 rd->ring->ring_name[RCB_RING_NAME_LEN - 1] = '\0';
1295
1296                 ret = request_irq(rd->ring->irq,
1297                                   hns_irq_handle, 0, rd->ring->ring_name, rd);
1298                 if (ret) {
1299                         netdev_err(priv->netdev, "request irq(%d) fail\n",
1300                                    rd->ring->irq);
1301                         goto out_free_irq;
1302                 }
1303                 disable_irq(rd->ring->irq);
1304
1305                 cpu = hns_nic_init_affinity_mask(h->q_num, i,
1306                                                  rd->ring, &rd->mask);
1307
1308                 if (cpu_online(cpu))
1309                         irq_set_affinity_hint(rd->ring->irq,
1310                                               &rd->mask);
1311
1312                 rd->ring->irq_init_flag = RCB_IRQ_INITED;
1313         }
1314
1315         return 0;
1316
1317 out_free_irq:
1318         hns_nic_free_irq(h->q_num, priv);
1319         return ret;
1320 }
1321
1322 static int hns_nic_net_up(struct net_device *ndev)
1323 {
1324         struct hns_nic_priv *priv = netdev_priv(ndev);
1325         struct hnae_handle *h = priv->ae_handle;
1326         int i, j;
1327         int ret;
1328
1329         if (!test_bit(NIC_STATE_DOWN, &priv->state))
1330                 return 0;
1331
1332         ret = hns_nic_init_irq(priv);
1333         if (ret != 0) {
1334                 netdev_err(ndev, "hns init irq failed! ret=%d\n", ret);
1335                 return ret;
1336         }
1337
1338         for (i = 0; i < h->q_num * 2; i++) {
1339                 ret = hns_nic_ring_open(ndev, i);
1340                 if (ret)
1341                         goto out_has_some_queues;
1342         }
1343
1344         ret = h->dev->ops->set_mac_addr(h, ndev->dev_addr);
1345         if (ret)
1346                 goto out_set_mac_addr_err;
1347
1348         ret = h->dev->ops->start ? h->dev->ops->start(h) : 0;
1349         if (ret)
1350                 goto out_start_err;
1351
1352         if (ndev->phydev)
1353                 phy_start(ndev->phydev);
1354
1355         clear_bit(NIC_STATE_DOWN, &priv->state);
1356         (void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
1357
1358         return 0;
1359
1360 out_start_err:
1361         netif_stop_queue(ndev);
1362 out_set_mac_addr_err:
1363 out_has_some_queues:
1364         for (j = i - 1; j >= 0; j--)
1365                 hns_nic_ring_close(ndev, j);
1366
1367         hns_nic_free_irq(h->q_num, priv);
1368         set_bit(NIC_STATE_DOWN, &priv->state);
1369
1370         return ret;
1371 }
1372
1373 static void hns_nic_net_down(struct net_device *ndev)
1374 {
1375         int i;
1376         struct hnae_ae_ops *ops;
1377         struct hns_nic_priv *priv = netdev_priv(ndev);
1378
1379         if (test_and_set_bit(NIC_STATE_DOWN, &priv->state))
1380                 return;
1381
1382         (void)del_timer_sync(&priv->service_timer);
1383         netif_tx_stop_all_queues(ndev);
1384         netif_carrier_off(ndev);
1385         netif_tx_disable(ndev);
1386         priv->link = 0;
1387
1388         if (ndev->phydev)
1389                 phy_stop(ndev->phydev);
1390
1391         ops = priv->ae_handle->dev->ops;
1392
1393         if (ops->stop)
1394                 ops->stop(priv->ae_handle);
1395
1396         netif_tx_stop_all_queues(ndev);
1397
1398         for (i = priv->ae_handle->q_num - 1; i >= 0; i--) {
1399                 hns_nic_ring_close(ndev, i);
1400                 hns_nic_ring_close(ndev, i + priv->ae_handle->q_num);
1401
1402                 /* clean tx buffers*/
1403                 hns_nic_tx_clr_all_bufs(priv->ring_data + i);
1404         }
1405 }
1406
1407 void hns_nic_net_reset(struct net_device *ndev)
1408 {
1409         struct hns_nic_priv *priv = netdev_priv(ndev);
1410         struct hnae_handle *handle = priv->ae_handle;
1411
1412         while (test_and_set_bit(NIC_STATE_RESETTING, &priv->state))
1413                 usleep_range(1000, 2000);
1414
1415         (void)hnae_reinit_handle(handle);
1416
1417         clear_bit(NIC_STATE_RESETTING, &priv->state);
1418 }
1419
1420 void hns_nic_net_reinit(struct net_device *netdev)
1421 {
1422         struct hns_nic_priv *priv = netdev_priv(netdev);
1423         enum hnae_port_type type = priv->ae_handle->port_type;
1424
1425         netif_trans_update(priv->netdev);
1426         while (test_and_set_bit(NIC_STATE_REINITING, &priv->state))
1427                 usleep_range(1000, 2000);
1428
1429         hns_nic_net_down(netdev);
1430
1431         /* Only do hns_nic_net_reset in debug mode
1432          * because of hardware limitation.
1433          */
1434         if (type == HNAE_PORT_DEBUG)
1435                 hns_nic_net_reset(netdev);
1436
1437         (void)hns_nic_net_up(netdev);
1438         clear_bit(NIC_STATE_REINITING, &priv->state);
1439 }
1440
1441 static int hns_nic_net_open(struct net_device *ndev)
1442 {
1443         struct hns_nic_priv *priv = netdev_priv(ndev);
1444         struct hnae_handle *h = priv->ae_handle;
1445         int ret;
1446
1447         if (test_bit(NIC_STATE_TESTING, &priv->state))
1448                 return -EBUSY;
1449
1450         priv->link = 0;
1451         netif_carrier_off(ndev);
1452
1453         ret = netif_set_real_num_tx_queues(ndev, h->q_num);
1454         if (ret < 0) {
1455                 netdev_err(ndev, "netif_set_real_num_tx_queues fail, ret=%d!\n",
1456                            ret);
1457                 return ret;
1458         }
1459
1460         ret = netif_set_real_num_rx_queues(ndev, h->q_num);
1461         if (ret < 0) {
1462                 netdev_err(ndev,
1463                            "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
1464                 return ret;
1465         }
1466
1467         ret = hns_nic_net_up(ndev);
1468         if (ret) {
1469                 netdev_err(ndev,
1470                            "hns net up fail, ret=%d!\n", ret);
1471                 return ret;
1472         }
1473
1474         return 0;
1475 }
1476
1477 static int hns_nic_net_stop(struct net_device *ndev)
1478 {
1479         hns_nic_net_down(ndev);
1480
1481         return 0;
1482 }
1483
1484 static void hns_tx_timeout_reset(struct hns_nic_priv *priv);
1485 #define HNS_TX_TIMEO_LIMIT (40 * HZ)
1486 static void hns_nic_net_timeout(struct net_device *ndev)
1487 {
1488         struct hns_nic_priv *priv = netdev_priv(ndev);
1489
1490         if (ndev->watchdog_timeo < HNS_TX_TIMEO_LIMIT) {
1491                 ndev->watchdog_timeo *= 2;
1492                 netdev_info(ndev, "watchdog_timo changed to %d.\n",
1493                             ndev->watchdog_timeo);
1494         } else {
1495                 ndev->watchdog_timeo = HNS_NIC_TX_TIMEOUT;
1496                 hns_tx_timeout_reset(priv);
1497         }
1498 }
1499
1500 static int hns_nic_do_ioctl(struct net_device *netdev, struct ifreq *ifr,
1501                             int cmd)
1502 {
1503         struct phy_device *phy_dev = netdev->phydev;
1504
1505         if (!netif_running(netdev))
1506                 return -EINVAL;
1507
1508         if (!phy_dev)
1509                 return -ENOTSUPP;
1510
1511         return phy_mii_ioctl(phy_dev, ifr, cmd);
1512 }
1513
1514 static netdev_tx_t hns_nic_net_xmit(struct sk_buff *skb,
1515                                     struct net_device *ndev)
1516 {
1517         struct hns_nic_priv *priv = netdev_priv(ndev);
1518
1519         assert(skb->queue_mapping < ndev->ae_handle->q_num);
1520
1521         return hns_nic_net_xmit_hw(ndev, skb,
1522                                    &tx_ring_data(priv, skb->queue_mapping));
1523 }
1524
1525 static void hns_nic_drop_rx_fetch(struct hns_nic_ring_data *ring_data,
1526                                   struct sk_buff *skb)
1527 {
1528         dev_kfree_skb_any(skb);
1529 }
1530
1531 #define HNS_LB_TX_RING  0
1532 static struct sk_buff *hns_assemble_skb(struct net_device *ndev)
1533 {
1534         struct sk_buff *skb;
1535         struct ethhdr *ethhdr;
1536         int frame_len;
1537
1538         /* allocate test skb */
1539         skb = alloc_skb(64, GFP_KERNEL);
1540         if (!skb)
1541                 return NULL;
1542
1543         skb_put(skb, 64);
1544         skb->dev = ndev;
1545         memset(skb->data, 0xFF, skb->len);
1546
1547         /* must be tcp/ip package */
1548         ethhdr = (struct ethhdr *)skb->data;
1549         ethhdr->h_proto = htons(ETH_P_IP);
1550
1551         frame_len = skb->len & (~1ul);
1552         memset(&skb->data[frame_len / 2], 0xAA,
1553                frame_len / 2 - 1);
1554
1555         skb->queue_mapping = HNS_LB_TX_RING;
1556
1557         return skb;
1558 }
1559
1560 static int hns_enable_serdes_lb(struct net_device *ndev)
1561 {
1562         struct hns_nic_priv *priv = netdev_priv(ndev);
1563         struct hnae_handle *h = priv->ae_handle;
1564         struct hnae_ae_ops *ops = h->dev->ops;
1565         int speed, duplex;
1566         int ret;
1567
1568         ret = ops->set_loopback(h, MAC_INTERNALLOOP_SERDES, 1);
1569         if (ret)
1570                 return ret;
1571
1572         ret = ops->start ? ops->start(h) : 0;
1573         if (ret)
1574                 return ret;
1575
1576         /* link adjust duplex*/
1577         if (h->phy_if != PHY_INTERFACE_MODE_XGMII)
1578                 speed = 1000;
1579         else
1580                 speed = 10000;
1581         duplex = 1;
1582
1583         ops->adjust_link(h, speed, duplex);
1584
1585         /* wait h/w ready */
1586         mdelay(300);
1587
1588         return 0;
1589 }
1590
1591 static void hns_disable_serdes_lb(struct net_device *ndev)
1592 {
1593         struct hns_nic_priv *priv = netdev_priv(ndev);
1594         struct hnae_handle *h = priv->ae_handle;
1595         struct hnae_ae_ops *ops = h->dev->ops;
1596
1597         ops->stop(h);
1598         ops->set_loopback(h, MAC_INTERNALLOOP_SERDES, 0);
1599 }
1600
1601 /**
1602  *hns_nic_clear_all_rx_fetch - clear the chip fetched descriptions. The
1603  *function as follows:
1604  *    1. if one rx ring has found the page_offset is not equal 0 between head
1605  *       and tail, it means that the chip fetched the wrong descs for the ring
1606  *       which buffer size is 4096.
1607  *    2. we set the chip serdes loopback and set rss indirection to the ring.
1608  *    3. construct 64-bytes ip broadcast packages, wait the associated rx ring
1609  *       recieving all packages and it will fetch new descriptions.
1610  *    4. recover to the original state.
1611  *
1612  *@ndev: net device
1613  */
1614 static int hns_nic_clear_all_rx_fetch(struct net_device *ndev)
1615 {
1616         struct hns_nic_priv *priv = netdev_priv(ndev);
1617         struct hnae_handle *h = priv->ae_handle;
1618         struct hnae_ae_ops *ops = h->dev->ops;
1619         struct hns_nic_ring_data *rd;
1620         struct hnae_ring *ring;
1621         struct sk_buff *skb;
1622         u32 *org_indir;
1623         u32 *cur_indir;
1624         int indir_size;
1625         int head, tail;
1626         int fetch_num;
1627         int i, j;
1628         bool found;
1629         int retry_times;
1630         int ret = 0;
1631
1632         /* alloc indir memory */
1633         indir_size = ops->get_rss_indir_size(h) * sizeof(*org_indir);
1634         org_indir = kzalloc(indir_size, GFP_KERNEL);
1635         if (!org_indir)
1636                 return -ENOMEM;
1637
1638         /* store the orginal indirection */
1639         ops->get_rss(h, org_indir, NULL, NULL);
1640
1641         cur_indir = kzalloc(indir_size, GFP_KERNEL);
1642         if (!cur_indir) {
1643                 ret = -ENOMEM;
1644                 goto cur_indir_alloc_err;
1645         }
1646
1647         /* set loopback */
1648         if (hns_enable_serdes_lb(ndev)) {
1649                 ret = -EINVAL;
1650                 goto enable_serdes_lb_err;
1651         }
1652
1653         /* foreach every rx ring to clear fetch desc */
1654         for (i = 0; i < h->q_num; i++) {
1655                 ring = &h->qs[i]->rx_ring;
1656                 head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1657                 tail = readl_relaxed(ring->io_base + RCB_REG_TAIL);
1658                 found = false;
1659                 fetch_num = ring_dist(ring, head, tail);
1660
1661                 while (head != tail) {
1662                         if (ring->desc_cb[head].page_offset != 0) {
1663                                 found = true;
1664                                 break;
1665                         }
1666
1667                         head++;
1668                         if (head == ring->desc_num)
1669                                 head = 0;
1670                 }
1671
1672                 if (found) {
1673                         for (j = 0; j < indir_size / sizeof(*org_indir); j++)
1674                                 cur_indir[j] = i;
1675                         ops->set_rss(h, cur_indir, NULL, 0);
1676
1677                         for (j = 0; j < fetch_num; j++) {
1678                                 /* alloc one skb and init */
1679                                 skb = hns_assemble_skb(ndev);
1680                                 if (!skb) {
1681                                         ret = -ENOMEM;
1682                                         goto out;
1683                                 }
1684                                 rd = &tx_ring_data(priv, skb->queue_mapping);
1685                                 hns_nic_net_xmit_hw(ndev, skb, rd);
1686
1687                                 retry_times = 0;
1688                                 while (retry_times++ < 10) {
1689                                         mdelay(10);
1690                                         /* clean rx */
1691                                         rd = &rx_ring_data(priv, i);
1692                                         if (rd->poll_one(rd, fetch_num,
1693                                                          hns_nic_drop_rx_fetch))
1694                                                 break;
1695                                 }
1696
1697                                 retry_times = 0;
1698                                 while (retry_times++ < 10) {
1699                                         mdelay(10);
1700                                         /* clean tx ring 0 send package */
1701                                         rd = &tx_ring_data(priv,
1702                                                            HNS_LB_TX_RING);
1703                                         if (rd->poll_one(rd, fetch_num, NULL))
1704                                                 break;
1705                                 }
1706                         }
1707                 }
1708         }
1709
1710 out:
1711         /* restore everything */
1712         ops->set_rss(h, org_indir, NULL, 0);
1713         hns_disable_serdes_lb(ndev);
1714 enable_serdes_lb_err:
1715         kfree(cur_indir);
1716 cur_indir_alloc_err:
1717         kfree(org_indir);
1718
1719         return ret;
1720 }
1721
1722 static int hns_nic_change_mtu(struct net_device *ndev, int new_mtu)
1723 {
1724         struct hns_nic_priv *priv = netdev_priv(ndev);
1725         struct hnae_handle *h = priv->ae_handle;
1726         bool if_running = netif_running(ndev);
1727         int ret;
1728
1729         /* MTU < 68 is an error and causes problems on some kernels */
1730         if (new_mtu < 68)
1731                 return -EINVAL;
1732
1733         /* MTU no change */
1734         if (new_mtu == ndev->mtu)
1735                 return 0;
1736
1737         if (!h->dev->ops->set_mtu)
1738                 return -ENOTSUPP;
1739
1740         if (if_running) {
1741                 (void)hns_nic_net_stop(ndev);
1742                 msleep(100);
1743         }
1744
1745         if (priv->enet_ver != AE_VERSION_1 &&
1746             ndev->mtu <= BD_SIZE_2048_MAX_MTU &&
1747             new_mtu > BD_SIZE_2048_MAX_MTU) {
1748                 /* update desc */
1749                 hnae_reinit_all_ring_desc(h);
1750
1751                 /* clear the package which the chip has fetched */
1752                 ret = hns_nic_clear_all_rx_fetch(ndev);
1753
1754                 /* the page offset must be consist with desc */
1755                 hnae_reinit_all_ring_page_off(h);
1756
1757                 if (ret) {
1758                         netdev_err(ndev, "clear the fetched desc fail\n");
1759                         goto out;
1760                 }
1761         }
1762
1763         ret = h->dev->ops->set_mtu(h, new_mtu);
1764         if (ret) {
1765                 netdev_err(ndev, "set mtu fail, return value %d\n",
1766                            ret);
1767                 goto out;
1768         }
1769
1770         /* finally, set new mtu to netdevice */
1771         ndev->mtu = new_mtu;
1772
1773 out:
1774         if (if_running) {
1775                 if (hns_nic_net_open(ndev)) {
1776                         netdev_err(ndev, "hns net open fail\n");
1777                         ret = -EINVAL;
1778                 }
1779         }
1780
1781         return ret;
1782 }
1783
1784 static int hns_nic_set_features(struct net_device *netdev,
1785                                 netdev_features_t features)
1786 {
1787         struct hns_nic_priv *priv = netdev_priv(netdev);
1788
1789         switch (priv->enet_ver) {
1790         case AE_VERSION_1:
1791                 if (features & (NETIF_F_TSO | NETIF_F_TSO6))
1792                         netdev_info(netdev, "enet v1 do not support tso!\n");
1793                 break;
1794         default:
1795                 if (features & (NETIF_F_TSO | NETIF_F_TSO6)) {
1796                         priv->ops.fill_desc = fill_tso_desc;
1797                         priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tso;
1798                         /* The chip only support 7*4096 */
1799                         netif_set_gso_max_size(netdev, 7 * 4096);
1800                 } else {
1801                         priv->ops.fill_desc = fill_v2_desc;
1802                         priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
1803                 }
1804                 break;
1805         }
1806         netdev->features = features;
1807         return 0;
1808 }
1809
1810 static netdev_features_t hns_nic_fix_features(
1811                 struct net_device *netdev, netdev_features_t features)
1812 {
1813         struct hns_nic_priv *priv = netdev_priv(netdev);
1814
1815         switch (priv->enet_ver) {
1816         case AE_VERSION_1:
1817                 features &= ~(NETIF_F_TSO | NETIF_F_TSO6 |
1818                                 NETIF_F_HW_VLAN_CTAG_FILTER);
1819                 break;
1820         default:
1821                 break;
1822         }
1823         return features;
1824 }
1825
1826 static int hns_nic_uc_sync(struct net_device *netdev, const unsigned char *addr)
1827 {
1828         struct hns_nic_priv *priv = netdev_priv(netdev);
1829         struct hnae_handle *h = priv->ae_handle;
1830
1831         if (h->dev->ops->add_uc_addr)
1832                 return h->dev->ops->add_uc_addr(h, addr);
1833
1834         return 0;
1835 }
1836
1837 static int hns_nic_uc_unsync(struct net_device *netdev,
1838                              const unsigned char *addr)
1839 {
1840         struct hns_nic_priv *priv = netdev_priv(netdev);
1841         struct hnae_handle *h = priv->ae_handle;
1842
1843         if (h->dev->ops->rm_uc_addr)
1844                 return h->dev->ops->rm_uc_addr(h, addr);
1845
1846         return 0;
1847 }
1848
1849 /**
1850  * nic_set_multicast_list - set mutl mac address
1851  * @netdev: net device
1852  * @p: mac address
1853  *
1854  * return void
1855  */
1856 static void hns_set_multicast_list(struct net_device *ndev)
1857 {
1858         struct hns_nic_priv *priv = netdev_priv(ndev);
1859         struct hnae_handle *h = priv->ae_handle;
1860         struct netdev_hw_addr *ha = NULL;
1861
1862         if (!h) {
1863                 netdev_err(ndev, "hnae handle is null\n");
1864                 return;
1865         }
1866
1867         if (h->dev->ops->clr_mc_addr)
1868                 if (h->dev->ops->clr_mc_addr(h))
1869                         netdev_err(ndev, "clear multicast address fail\n");
1870
1871         if (h->dev->ops->set_mc_addr) {
1872                 netdev_for_each_mc_addr(ha, ndev)
1873                         if (h->dev->ops->set_mc_addr(h, ha->addr))
1874                                 netdev_err(ndev, "set multicast fail\n");
1875         }
1876 }
1877
1878 static void hns_nic_set_rx_mode(struct net_device *ndev)
1879 {
1880         struct hns_nic_priv *priv = netdev_priv(ndev);
1881         struct hnae_handle *h = priv->ae_handle;
1882
1883         if (h->dev->ops->set_promisc_mode) {
1884                 if (ndev->flags & IFF_PROMISC)
1885                         h->dev->ops->set_promisc_mode(h, 1);
1886                 else
1887                         h->dev->ops->set_promisc_mode(h, 0);
1888         }
1889
1890         hns_set_multicast_list(ndev);
1891
1892         if (__dev_uc_sync(ndev, hns_nic_uc_sync, hns_nic_uc_unsync))
1893                 netdev_err(ndev, "sync uc address fail\n");
1894 }
1895
1896 static void hns_nic_get_stats64(struct net_device *ndev,
1897                                 struct rtnl_link_stats64 *stats)
1898 {
1899         int idx = 0;
1900         u64 tx_bytes = 0;
1901         u64 rx_bytes = 0;
1902         u64 tx_pkts = 0;
1903         u64 rx_pkts = 0;
1904         struct hns_nic_priv *priv = netdev_priv(ndev);
1905         struct hnae_handle *h = priv->ae_handle;
1906
1907         for (idx = 0; idx < h->q_num; idx++) {
1908                 tx_bytes += h->qs[idx]->tx_ring.stats.tx_bytes;
1909                 tx_pkts += h->qs[idx]->tx_ring.stats.tx_pkts;
1910                 rx_bytes += h->qs[idx]->rx_ring.stats.rx_bytes;
1911                 rx_pkts += h->qs[idx]->rx_ring.stats.rx_pkts;
1912         }
1913
1914         stats->tx_bytes = tx_bytes;
1915         stats->tx_packets = tx_pkts;
1916         stats->rx_bytes = rx_bytes;
1917         stats->rx_packets = rx_pkts;
1918
1919         stats->rx_errors = ndev->stats.rx_errors;
1920         stats->multicast = ndev->stats.multicast;
1921         stats->rx_length_errors = ndev->stats.rx_length_errors;
1922         stats->rx_crc_errors = ndev->stats.rx_crc_errors;
1923         stats->rx_missed_errors = ndev->stats.rx_missed_errors;
1924
1925         stats->tx_errors = ndev->stats.tx_errors;
1926         stats->rx_dropped = ndev->stats.rx_dropped;
1927         stats->tx_dropped = ndev->stats.tx_dropped;
1928         stats->collisions = ndev->stats.collisions;
1929         stats->rx_over_errors = ndev->stats.rx_over_errors;
1930         stats->rx_frame_errors = ndev->stats.rx_frame_errors;
1931         stats->rx_fifo_errors = ndev->stats.rx_fifo_errors;
1932         stats->tx_aborted_errors = ndev->stats.tx_aborted_errors;
1933         stats->tx_carrier_errors = ndev->stats.tx_carrier_errors;
1934         stats->tx_fifo_errors = ndev->stats.tx_fifo_errors;
1935         stats->tx_heartbeat_errors = ndev->stats.tx_heartbeat_errors;
1936         stats->tx_window_errors = ndev->stats.tx_window_errors;
1937         stats->rx_compressed = ndev->stats.rx_compressed;
1938         stats->tx_compressed = ndev->stats.tx_compressed;
1939 }
1940
1941 static u16
1942 hns_nic_select_queue(struct net_device *ndev, struct sk_buff *skb,
1943                      struct net_device *sb_dev,
1944                      select_queue_fallback_t fallback)
1945 {
1946         struct ethhdr *eth_hdr = (struct ethhdr *)skb->data;
1947         struct hns_nic_priv *priv = netdev_priv(ndev);
1948
1949         /* fix hardware broadcast/multicast packets queue loopback */
1950         if (!AE_IS_VER1(priv->enet_ver) &&
1951             is_multicast_ether_addr(eth_hdr->h_dest))
1952                 return 0;
1953         else
1954                 return fallback(ndev, skb, NULL);
1955 }
1956
1957 static const struct net_device_ops hns_nic_netdev_ops = {
1958         .ndo_open = hns_nic_net_open,
1959         .ndo_stop = hns_nic_net_stop,
1960         .ndo_start_xmit = hns_nic_net_xmit,
1961         .ndo_tx_timeout = hns_nic_net_timeout,
1962         .ndo_set_mac_address = hns_nic_net_set_mac_address,
1963         .ndo_change_mtu = hns_nic_change_mtu,
1964         .ndo_do_ioctl = hns_nic_do_ioctl,
1965         .ndo_set_features = hns_nic_set_features,
1966         .ndo_fix_features = hns_nic_fix_features,
1967         .ndo_get_stats64 = hns_nic_get_stats64,
1968         .ndo_set_rx_mode = hns_nic_set_rx_mode,
1969         .ndo_select_queue = hns_nic_select_queue,
1970 };
1971
1972 static void hns_nic_update_link_status(struct net_device *netdev)
1973 {
1974         struct hns_nic_priv *priv = netdev_priv(netdev);
1975
1976         struct hnae_handle *h = priv->ae_handle;
1977
1978         if (h->phy_dev) {
1979                 if (h->phy_if != PHY_INTERFACE_MODE_XGMII)
1980                         return;
1981
1982                 (void)genphy_read_status(h->phy_dev);
1983         }
1984         hns_nic_adjust_link(netdev);
1985 }
1986
1987 /* for dumping key regs*/
1988 static void hns_nic_dump(struct hns_nic_priv *priv)
1989 {
1990         struct hnae_handle *h = priv->ae_handle;
1991         struct hnae_ae_ops *ops = h->dev->ops;
1992         u32 *data, reg_num, i;
1993
1994         if (ops->get_regs_len && ops->get_regs) {
1995                 reg_num = ops->get_regs_len(priv->ae_handle);
1996                 reg_num = (reg_num + 3ul) & ~3ul;
1997                 data = kcalloc(reg_num, sizeof(u32), GFP_KERNEL);
1998                 if (data) {
1999                         ops->get_regs(priv->ae_handle, data);
2000                         for (i = 0; i < reg_num; i += 4)
2001                                 pr_info("0x%08x: 0x%08x 0x%08x 0x%08x 0x%08x\n",
2002                                         i, data[i], data[i + 1],
2003                                         data[i + 2], data[i + 3]);
2004                         kfree(data);
2005                 }
2006         }
2007
2008         for (i = 0; i < h->q_num; i++) {
2009                 pr_info("tx_queue%d_next_to_clean:%d\n",
2010                         i, h->qs[i]->tx_ring.next_to_clean);
2011                 pr_info("tx_queue%d_next_to_use:%d\n",
2012                         i, h->qs[i]->tx_ring.next_to_use);
2013                 pr_info("rx_queue%d_next_to_clean:%d\n",
2014                         i, h->qs[i]->rx_ring.next_to_clean);
2015                 pr_info("rx_queue%d_next_to_use:%d\n",
2016                         i, h->qs[i]->rx_ring.next_to_use);
2017         }
2018 }
2019
2020 /* for resetting subtask */
2021 static void hns_nic_reset_subtask(struct hns_nic_priv *priv)
2022 {
2023         enum hnae_port_type type = priv->ae_handle->port_type;
2024
2025         if (!test_bit(NIC_STATE2_RESET_REQUESTED, &priv->state))
2026                 return;
2027         clear_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
2028
2029         /* If we're already down, removing or resetting, just bail */
2030         if (test_bit(NIC_STATE_DOWN, &priv->state) ||
2031             test_bit(NIC_STATE_REMOVING, &priv->state) ||
2032             test_bit(NIC_STATE_RESETTING, &priv->state))
2033                 return;
2034
2035         hns_nic_dump(priv);
2036         netdev_info(priv->netdev, "try to reset %s port!\n",
2037                     (type == HNAE_PORT_DEBUG ? "debug" : "service"));
2038
2039         rtnl_lock();
2040         /* put off any impending NetWatchDogTimeout */
2041         netif_trans_update(priv->netdev);
2042         hns_nic_net_reinit(priv->netdev);
2043
2044         rtnl_unlock();
2045 }
2046
2047 /* for doing service complete*/
2048 static void hns_nic_service_event_complete(struct hns_nic_priv *priv)
2049 {
2050         WARN_ON(!test_bit(NIC_STATE_SERVICE_SCHED, &priv->state));
2051         /* make sure to commit the things */
2052         smp_mb__before_atomic();
2053         clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
2054 }
2055
2056 static void hns_nic_service_task(struct work_struct *work)
2057 {
2058         struct hns_nic_priv *priv
2059                 = container_of(work, struct hns_nic_priv, service_task);
2060         struct hnae_handle *h = priv->ae_handle;
2061
2062         hns_nic_reset_subtask(priv);
2063         hns_nic_update_link_status(priv->netdev);
2064         h->dev->ops->update_led_status(h);
2065         hns_nic_update_stats(priv->netdev);
2066
2067         hns_nic_service_event_complete(priv);
2068 }
2069
2070 static void hns_nic_task_schedule(struct hns_nic_priv *priv)
2071 {
2072         if (!test_bit(NIC_STATE_DOWN, &priv->state) &&
2073             !test_bit(NIC_STATE_REMOVING, &priv->state) &&
2074             !test_and_set_bit(NIC_STATE_SERVICE_SCHED, &priv->state))
2075                 (void)schedule_work(&priv->service_task);
2076 }
2077
2078 static void hns_nic_service_timer(struct timer_list *t)
2079 {
2080         struct hns_nic_priv *priv = from_timer(priv, t, service_timer);
2081
2082         (void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
2083
2084         hns_nic_task_schedule(priv);
2085 }
2086
2087 /**
2088  * hns_tx_timeout_reset - initiate reset due to Tx timeout
2089  * @priv: driver private struct
2090  **/
2091 static void hns_tx_timeout_reset(struct hns_nic_priv *priv)
2092 {
2093         /* Do the reset outside of interrupt context */
2094         if (!test_bit(NIC_STATE_DOWN, &priv->state)) {
2095                 set_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
2096                 netdev_warn(priv->netdev,
2097                             "initiating reset due to tx timeout(%llu,0x%lx)\n",
2098                             priv->tx_timeout_count, priv->state);
2099                 priv->tx_timeout_count++;
2100                 hns_nic_task_schedule(priv);
2101         }
2102 }
2103
2104 static int hns_nic_init_ring_data(struct hns_nic_priv *priv)
2105 {
2106         struct hnae_handle *h = priv->ae_handle;
2107         struct hns_nic_ring_data *rd;
2108         bool is_ver1 = AE_IS_VER1(priv->enet_ver);
2109         int i;
2110
2111         if (h->q_num > NIC_MAX_Q_PER_VF) {
2112                 netdev_err(priv->netdev, "too much queue (%d)\n", h->q_num);
2113                 return -EINVAL;
2114         }
2115
2116         priv->ring_data = kzalloc(array3_size(h->q_num,
2117                                               sizeof(*priv->ring_data), 2),
2118                                   GFP_KERNEL);
2119         if (!priv->ring_data)
2120                 return -ENOMEM;
2121
2122         for (i = 0; i < h->q_num; i++) {
2123                 rd = &priv->ring_data[i];
2124                 rd->queue_index = i;
2125                 rd->ring = &h->qs[i]->tx_ring;
2126                 rd->poll_one = hns_nic_tx_poll_one;
2127                 rd->fini_process = is_ver1 ? hns_nic_tx_fini_pro :
2128                         hns_nic_tx_fini_pro_v2;
2129
2130                 netif_napi_add(priv->netdev, &rd->napi,
2131                                hns_nic_common_poll, NAPI_POLL_WEIGHT);
2132                 rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
2133         }
2134         for (i = h->q_num; i < h->q_num * 2; i++) {
2135                 rd = &priv->ring_data[i];
2136                 rd->queue_index = i - h->q_num;
2137                 rd->ring = &h->qs[i - h->q_num]->rx_ring;
2138                 rd->poll_one = hns_nic_rx_poll_one;
2139                 rd->ex_process = hns_nic_rx_up_pro;
2140                 rd->fini_process = is_ver1 ? hns_nic_rx_fini_pro :
2141                         hns_nic_rx_fini_pro_v2;
2142
2143                 netif_napi_add(priv->netdev, &rd->napi,
2144                                hns_nic_common_poll, NAPI_POLL_WEIGHT);
2145                 rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
2146         }
2147
2148         return 0;
2149 }
2150
2151 static void hns_nic_uninit_ring_data(struct hns_nic_priv *priv)
2152 {
2153         struct hnae_handle *h = priv->ae_handle;
2154         int i;
2155
2156         for (i = 0; i < h->q_num * 2; i++) {
2157                 netif_napi_del(&priv->ring_data[i].napi);
2158                 if (priv->ring_data[i].ring->irq_init_flag == RCB_IRQ_INITED) {
2159                         (void)irq_set_affinity_hint(
2160                                 priv->ring_data[i].ring->irq,
2161                                 NULL);
2162                         free_irq(priv->ring_data[i].ring->irq,
2163                                  &priv->ring_data[i]);
2164                 }
2165
2166                 priv->ring_data[i].ring->irq_init_flag = RCB_IRQ_NOT_INITED;
2167         }
2168         kfree(priv->ring_data);
2169 }
2170
2171 static void hns_nic_set_priv_ops(struct net_device *netdev)
2172 {
2173         struct hns_nic_priv *priv = netdev_priv(netdev);
2174         struct hnae_handle *h = priv->ae_handle;
2175
2176         if (AE_IS_VER1(priv->enet_ver)) {
2177                 priv->ops.fill_desc = fill_desc;
2178                 priv->ops.get_rxd_bnum = get_rx_desc_bnum;
2179                 priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
2180         } else {
2181                 priv->ops.get_rxd_bnum = get_v2rx_desc_bnum;
2182                 if ((netdev->features & NETIF_F_TSO) ||
2183                     (netdev->features & NETIF_F_TSO6)) {
2184                         priv->ops.fill_desc = fill_tso_desc;
2185                         priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tso;
2186                         /* This chip only support 7*4096 */
2187                         netif_set_gso_max_size(netdev, 7 * 4096);
2188                 } else {
2189                         priv->ops.fill_desc = fill_v2_desc;
2190                         priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
2191                 }
2192                 /* enable tso when init
2193                  * control tso on/off through TSE bit in bd
2194                  */
2195                 h->dev->ops->set_tso_stats(h, 1);
2196         }
2197 }
2198
2199 static int hns_nic_try_get_ae(struct net_device *ndev)
2200 {
2201         struct hns_nic_priv *priv = netdev_priv(ndev);
2202         struct hnae_handle *h;
2203         int ret;
2204
2205         h = hnae_get_handle(&priv->netdev->dev,
2206                             priv->fwnode, priv->port_id, NULL);
2207         if (IS_ERR_OR_NULL(h)) {
2208                 ret = -ENODEV;
2209                 dev_dbg(priv->dev, "has not handle, register notifier!\n");
2210                 goto out;
2211         }
2212         priv->ae_handle = h;
2213
2214         ret = hns_nic_init_phy(ndev, h);
2215         if (ret) {
2216                 dev_err(priv->dev, "probe phy device fail!\n");
2217                 goto out_init_phy;
2218         }
2219
2220         ret = hns_nic_init_ring_data(priv);
2221         if (ret) {
2222                 ret = -ENOMEM;
2223                 goto out_init_ring_data;
2224         }
2225
2226         hns_nic_set_priv_ops(ndev);
2227
2228         ret = register_netdev(ndev);
2229         if (ret) {
2230                 dev_err(priv->dev, "probe register netdev fail!\n");
2231                 goto out_reg_ndev_fail;
2232         }
2233         return 0;
2234
2235 out_reg_ndev_fail:
2236         hns_nic_uninit_ring_data(priv);
2237         priv->ring_data = NULL;
2238 out_init_phy:
2239 out_init_ring_data:
2240         hnae_put_handle(priv->ae_handle);
2241         priv->ae_handle = NULL;
2242 out:
2243         return ret;
2244 }
2245
2246 static int hns_nic_notifier_action(struct notifier_block *nb,
2247                                    unsigned long action, void *data)
2248 {
2249         struct hns_nic_priv *priv =
2250                 container_of(nb, struct hns_nic_priv, notifier_block);
2251
2252         assert(action == HNAE_AE_REGISTER);
2253
2254         if (!hns_nic_try_get_ae(priv->netdev)) {
2255                 hnae_unregister_notifier(&priv->notifier_block);
2256                 priv->notifier_block.notifier_call = NULL;
2257         }
2258         return 0;
2259 }
2260
2261 static int hns_nic_dev_probe(struct platform_device *pdev)
2262 {
2263         struct device *dev = &pdev->dev;
2264         struct net_device *ndev;
2265         struct hns_nic_priv *priv;
2266         u32 port_id;
2267         int ret;
2268
2269         ndev = alloc_etherdev_mq(sizeof(struct hns_nic_priv), NIC_MAX_Q_PER_VF);
2270         if (!ndev)
2271                 return -ENOMEM;
2272
2273         platform_set_drvdata(pdev, ndev);
2274
2275         priv = netdev_priv(ndev);
2276         priv->dev = dev;
2277         priv->netdev = ndev;
2278
2279         if (dev_of_node(dev)) {
2280                 struct device_node *ae_node;
2281
2282                 if (of_device_is_compatible(dev->of_node,
2283                                             "hisilicon,hns-nic-v1"))
2284                         priv->enet_ver = AE_VERSION_1;
2285                 else
2286                         priv->enet_ver = AE_VERSION_2;
2287
2288                 ae_node = of_parse_phandle(dev->of_node, "ae-handle", 0);
2289                 if (!ae_node) {
2290                         ret = -ENODEV;
2291                         dev_err(dev, "not find ae-handle\n");
2292                         goto out_read_prop_fail;
2293                 }
2294                 priv->fwnode = &ae_node->fwnode;
2295         } else if (is_acpi_node(dev->fwnode)) {
2296                 struct fwnode_reference_args args;
2297
2298                 if (acpi_dev_found(hns_enet_acpi_match[0].id))
2299                         priv->enet_ver = AE_VERSION_1;
2300                 else if (acpi_dev_found(hns_enet_acpi_match[1].id))
2301                         priv->enet_ver = AE_VERSION_2;
2302                 else {
2303                         ret = -ENXIO;
2304                         goto out_read_prop_fail;
2305                 }
2306
2307                 /* try to find port-idx-in-ae first */
2308                 ret = acpi_node_get_property_reference(dev->fwnode,
2309                                                        "ae-handle", 0, &args);
2310                 if (ret) {
2311                         dev_err(dev, "not find ae-handle\n");
2312                         goto out_read_prop_fail;
2313                 }
2314                 if (!is_acpi_device_node(args.fwnode)) {
2315                         ret = -EINVAL;
2316                         goto out_read_prop_fail;
2317                 }
2318                 priv->fwnode = args.fwnode;
2319         } else {
2320                 dev_err(dev, "cannot read cfg data from OF or acpi\n");
2321                 ret = -ENXIO;
2322                 goto out_read_prop_fail;
2323         }
2324
2325         ret = device_property_read_u32(dev, "port-idx-in-ae", &port_id);
2326         if (ret) {
2327                 /* only for old code compatible */
2328                 ret = device_property_read_u32(dev, "port-id", &port_id);
2329                 if (ret)
2330                         goto out_read_prop_fail;
2331                 /* for old dts, we need to caculate the port offset */
2332                 port_id = port_id < HNS_SRV_OFFSET ? port_id + HNS_DEBUG_OFFSET
2333                         : port_id - HNS_SRV_OFFSET;
2334         }
2335         priv->port_id = port_id;
2336
2337         hns_init_mac_addr(ndev);
2338
2339         ndev->watchdog_timeo = HNS_NIC_TX_TIMEOUT;
2340         ndev->priv_flags |= IFF_UNICAST_FLT;
2341         ndev->netdev_ops = &hns_nic_netdev_ops;
2342         hns_ethtool_set_ops(ndev);
2343
2344         ndev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2345                 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2346                 NETIF_F_GRO;
2347         ndev->vlan_features |=
2348                 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM;
2349         ndev->vlan_features |= NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
2350
2351         /* MTU range: 68 - 9578 (v1) or 9706 (v2) */
2352         ndev->min_mtu = MAC_MIN_MTU;
2353         switch (priv->enet_ver) {
2354         case AE_VERSION_2:
2355                 ndev->features |= NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_NTUPLE;
2356                 ndev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2357                         NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2358                         NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6;
2359                 ndev->max_mtu = MAC_MAX_MTU_V2 -
2360                                 (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
2361                 break;
2362         default:
2363                 ndev->max_mtu = MAC_MAX_MTU -
2364                                 (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
2365                 break;
2366         }
2367
2368         SET_NETDEV_DEV(ndev, dev);
2369
2370         if (!dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64)))
2371                 dev_dbg(dev, "set mask to 64bit\n");
2372         else
2373                 dev_err(dev, "set mask to 64bit fail!\n");
2374
2375         /* carrier off reporting is important to ethtool even BEFORE open */
2376         netif_carrier_off(ndev);
2377
2378         timer_setup(&priv->service_timer, hns_nic_service_timer, 0);
2379         INIT_WORK(&priv->service_task, hns_nic_service_task);
2380
2381         set_bit(NIC_STATE_SERVICE_INITED, &priv->state);
2382         clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
2383         set_bit(NIC_STATE_DOWN, &priv->state);
2384
2385         if (hns_nic_try_get_ae(priv->netdev)) {
2386                 priv->notifier_block.notifier_call = hns_nic_notifier_action;
2387                 ret = hnae_register_notifier(&priv->notifier_block);
2388                 if (ret) {
2389                         dev_err(dev, "register notifier fail!\n");
2390                         goto out_notify_fail;
2391                 }
2392                 dev_dbg(dev, "has not handle, register notifier!\n");
2393         }
2394
2395         return 0;
2396
2397 out_notify_fail:
2398         (void)cancel_work_sync(&priv->service_task);
2399 out_read_prop_fail:
2400         /* safe for ACPI FW */
2401         of_node_put(to_of_node(priv->fwnode));
2402         free_netdev(ndev);
2403         return ret;
2404 }
2405
2406 static int hns_nic_dev_remove(struct platform_device *pdev)
2407 {
2408         struct net_device *ndev = platform_get_drvdata(pdev);
2409         struct hns_nic_priv *priv = netdev_priv(ndev);
2410
2411         if (ndev->reg_state != NETREG_UNINITIALIZED)
2412                 unregister_netdev(ndev);
2413
2414         if (priv->ring_data)
2415                 hns_nic_uninit_ring_data(priv);
2416         priv->ring_data = NULL;
2417
2418         if (ndev->phydev)
2419                 phy_disconnect(ndev->phydev);
2420
2421         if (!IS_ERR_OR_NULL(priv->ae_handle))
2422                 hnae_put_handle(priv->ae_handle);
2423         priv->ae_handle = NULL;
2424         if (priv->notifier_block.notifier_call)
2425                 hnae_unregister_notifier(&priv->notifier_block);
2426         priv->notifier_block.notifier_call = NULL;
2427
2428         set_bit(NIC_STATE_REMOVING, &priv->state);
2429         (void)cancel_work_sync(&priv->service_task);
2430
2431         /* safe for ACPI FW */
2432         of_node_put(to_of_node(priv->fwnode));
2433
2434         free_netdev(ndev);
2435         return 0;
2436 }
2437
2438 static const struct of_device_id hns_enet_of_match[] = {
2439         {.compatible = "hisilicon,hns-nic-v1",},
2440         {.compatible = "hisilicon,hns-nic-v2",},
2441         {},
2442 };
2443
2444 MODULE_DEVICE_TABLE(of, hns_enet_of_match);
2445
2446 static struct platform_driver hns_nic_dev_driver = {
2447         .driver = {
2448                 .name = "hns-nic",
2449                 .of_match_table = hns_enet_of_match,
2450                 .acpi_match_table = ACPI_PTR(hns_enet_acpi_match),
2451         },
2452         .probe = hns_nic_dev_probe,
2453         .remove = hns_nic_dev_remove,
2454 };
2455
2456 module_platform_driver(hns_nic_dev_driver);
2457
2458 MODULE_DESCRIPTION("HISILICON HNS Ethernet driver");
2459 MODULE_AUTHOR("Hisilicon, Inc.");
2460 MODULE_LICENSE("GPL");
2461 MODULE_ALIAS("platform:hns-nic");