GNU Linux-libre 6.8.7-gnu
[releases.git] / drivers / net / ethernet / hisilicon / hns / hns_enet.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (c) 2014-2015 Hisilicon Limited.
4  */
5
6 #include <linux/clk.h>
7 #include <linux/cpumask.h>
8 #include <linux/etherdevice.h>
9 #include <linux/if_vlan.h>
10 #include <linux/interrupt.h>
11 #include <linux/io.h>
12 #include <linux/ip.h>
13 #include <linux/ipv6.h>
14 #include <linux/irq.h>
15 #include <linux/module.h>
16 #include <linux/phy.h>
17 #include <linux/platform_device.h>
18 #include <linux/skbuff.h>
19
20 #include "hnae.h"
21 #include "hns_enet.h"
22 #include "hns_dsaf_mac.h"
23
24 #define NIC_MAX_Q_PER_VF 16
25 #define HNS_NIC_TX_TIMEOUT (5 * HZ)
26
27 #define SERVICE_TIMER_HZ (1 * HZ)
28
29 #define RCB_IRQ_NOT_INITED 0
30 #define RCB_IRQ_INITED 1
31 #define HNS_BUFFER_SIZE_2048 2048
32
33 #define BD_MAX_SEND_SIZE 8191
34
35 static void fill_v2_desc_hw(struct hnae_ring *ring, void *priv, int size,
36                             int send_sz, dma_addr_t dma, int frag_end,
37                             int buf_num, enum hns_desc_type type, int mtu)
38 {
39         struct hnae_desc *desc = &ring->desc[ring->next_to_use];
40         struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
41         struct iphdr *iphdr;
42         struct ipv6hdr *ipv6hdr;
43         struct sk_buff *skb;
44         __be16 protocol;
45         u8 bn_pid = 0;
46         u8 rrcfv = 0;
47         u8 ip_offset = 0;
48         u8 tvsvsn = 0;
49         u16 mss = 0;
50         u8 l4_len = 0;
51         u16 paylen = 0;
52
53         desc_cb->priv = priv;
54         desc_cb->length = size;
55         desc_cb->dma = dma;
56         desc_cb->type = type;
57
58         desc->addr = cpu_to_le64(dma);
59         desc->tx.send_size = cpu_to_le16((u16)send_sz);
60
61         /* config bd buffer end */
62         hnae_set_bit(rrcfv, HNSV2_TXD_VLD_B, 1);
63         hnae_set_field(bn_pid, HNSV2_TXD_BUFNUM_M, 0, buf_num - 1);
64
65         /* fill port_id in the tx bd for sending management pkts */
66         hnae_set_field(bn_pid, HNSV2_TXD_PORTID_M,
67                        HNSV2_TXD_PORTID_S, ring->q->handle->dport_id);
68
69         if (type == DESC_TYPE_SKB) {
70                 skb = (struct sk_buff *)priv;
71
72                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
73                         skb_reset_mac_len(skb);
74                         protocol = skb->protocol;
75                         ip_offset = ETH_HLEN;
76
77                         if (protocol == htons(ETH_P_8021Q)) {
78                                 ip_offset += VLAN_HLEN;
79                                 protocol = vlan_get_protocol(skb);
80                                 skb->protocol = protocol;
81                         }
82
83                         if (skb->protocol == htons(ETH_P_IP)) {
84                                 iphdr = ip_hdr(skb);
85                                 hnae_set_bit(rrcfv, HNSV2_TXD_L3CS_B, 1);
86                                 hnae_set_bit(rrcfv, HNSV2_TXD_L4CS_B, 1);
87
88                                 /* check for tcp/udp header */
89                                 if (iphdr->protocol == IPPROTO_TCP &&
90                                     skb_is_gso(skb)) {
91                                         hnae_set_bit(tvsvsn,
92                                                      HNSV2_TXD_TSE_B, 1);
93                                         l4_len = tcp_hdrlen(skb);
94                                         mss = skb_shinfo(skb)->gso_size;
95                                         paylen = skb->len - skb_tcp_all_headers(skb);
96                                 }
97                         } else if (skb->protocol == htons(ETH_P_IPV6)) {
98                                 hnae_set_bit(tvsvsn, HNSV2_TXD_IPV6_B, 1);
99                                 ipv6hdr = ipv6_hdr(skb);
100                                 hnae_set_bit(rrcfv, HNSV2_TXD_L4CS_B, 1);
101
102                                 /* check for tcp/udp header */
103                                 if (ipv6hdr->nexthdr == IPPROTO_TCP &&
104                                     skb_is_gso(skb) && skb_is_gso_v6(skb)) {
105                                         hnae_set_bit(tvsvsn,
106                                                      HNSV2_TXD_TSE_B, 1);
107                                         l4_len = tcp_hdrlen(skb);
108                                         mss = skb_shinfo(skb)->gso_size;
109                                         paylen = skb->len - skb_tcp_all_headers(skb);
110                                 }
111                         }
112                         desc->tx.ip_offset = ip_offset;
113                         desc->tx.tse_vlan_snap_v6_sctp_nth = tvsvsn;
114                         desc->tx.mss = cpu_to_le16(mss);
115                         desc->tx.l4_len = l4_len;
116                         desc->tx.paylen = cpu_to_le16(paylen);
117                 }
118         }
119
120         hnae_set_bit(rrcfv, HNSV2_TXD_FE_B, frag_end);
121
122         desc->tx.bn_pid = bn_pid;
123         desc->tx.ra_ri_cs_fe_vld = rrcfv;
124
125         ring_ptr_move_fw(ring, next_to_use);
126 }
127
128 static void fill_v2_desc(struct hnae_ring *ring, void *priv,
129                          int size, dma_addr_t dma, int frag_end,
130                          int buf_num, enum hns_desc_type type, int mtu)
131 {
132         fill_v2_desc_hw(ring, priv, size, size, dma, frag_end,
133                         buf_num, type, mtu);
134 }
135
136 static const struct acpi_device_id hns_enet_acpi_match[] = {
137         { "HISI00C1", 0 },
138         { "HISI00C2", 0 },
139         { },
140 };
141 MODULE_DEVICE_TABLE(acpi, hns_enet_acpi_match);
142
143 static void fill_desc(struct hnae_ring *ring, void *priv,
144                       int size, dma_addr_t dma, int frag_end,
145                       int buf_num, enum hns_desc_type type, int mtu,
146                       bool is_gso)
147 {
148         struct hnae_desc *desc = &ring->desc[ring->next_to_use];
149         struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
150         struct sk_buff *skb;
151         __be16 protocol;
152         u32 ip_offset;
153         u32 asid_bufnum_pid = 0;
154         u32 flag_ipoffset = 0;
155
156         desc_cb->priv = priv;
157         desc_cb->length = size;
158         desc_cb->dma = dma;
159         desc_cb->type = type;
160
161         desc->addr = cpu_to_le64(dma);
162         desc->tx.send_size = cpu_to_le16((u16)size);
163
164         /*config bd buffer end */
165         flag_ipoffset |= 1 << HNS_TXD_VLD_B;
166
167         asid_bufnum_pid |= buf_num << HNS_TXD_BUFNUM_S;
168
169         if (type == DESC_TYPE_SKB) {
170                 skb = (struct sk_buff *)priv;
171
172                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
173                         protocol = skb->protocol;
174                         ip_offset = ETH_HLEN;
175
176                         /*if it is a SW VLAN check the next protocol*/
177                         if (protocol == htons(ETH_P_8021Q)) {
178                                 ip_offset += VLAN_HLEN;
179                                 protocol = vlan_get_protocol(skb);
180                                 skb->protocol = protocol;
181                         }
182
183                         if (skb->protocol == htons(ETH_P_IP)) {
184                                 flag_ipoffset |= 1 << HNS_TXD_L3CS_B;
185                                 /* check for tcp/udp header */
186                                 flag_ipoffset |= 1 << HNS_TXD_L4CS_B;
187
188                         } else if (skb->protocol == htons(ETH_P_IPV6)) {
189                                 /* ipv6 has not l3 cs, check for L4 header */
190                                 flag_ipoffset |= 1 << HNS_TXD_L4CS_B;
191                         }
192
193                         flag_ipoffset |= ip_offset << HNS_TXD_IPOFFSET_S;
194                 }
195         }
196
197         flag_ipoffset |= frag_end << HNS_TXD_FE_B;
198
199         desc->tx.asid_bufnum_pid = cpu_to_le16(asid_bufnum_pid);
200         desc->tx.flag_ipoffset = cpu_to_le32(flag_ipoffset);
201
202         ring_ptr_move_fw(ring, next_to_use);
203 }
204
205 static void unfill_desc(struct hnae_ring *ring)
206 {
207         ring_ptr_move_bw(ring, next_to_use);
208 }
209
210 static int hns_nic_maybe_stop_tx(
211         struct sk_buff **out_skb, int *bnum, struct hnae_ring *ring)
212 {
213         struct sk_buff *skb = *out_skb;
214         struct sk_buff *new_skb = NULL;
215         int buf_num;
216
217         /* no. of segments (plus a header) */
218         buf_num = skb_shinfo(skb)->nr_frags + 1;
219
220         if (unlikely(buf_num > ring->max_desc_num_per_pkt)) {
221                 if (ring_space(ring) < 1)
222                         return -EBUSY;
223
224                 new_skb = skb_copy(skb, GFP_ATOMIC);
225                 if (!new_skb)
226                         return -ENOMEM;
227
228                 dev_kfree_skb_any(skb);
229                 *out_skb = new_skb;
230                 buf_num = 1;
231         } else if (buf_num > ring_space(ring)) {
232                 return -EBUSY;
233         }
234
235         *bnum = buf_num;
236         return 0;
237 }
238
239 static int hns_nic_maybe_stop_tso(
240         struct sk_buff **out_skb, int *bnum, struct hnae_ring *ring)
241 {
242         int i;
243         int size;
244         int buf_num;
245         int frag_num;
246         struct sk_buff *skb = *out_skb;
247         struct sk_buff *new_skb = NULL;
248         skb_frag_t *frag;
249
250         size = skb_headlen(skb);
251         buf_num = (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
252
253         frag_num = skb_shinfo(skb)->nr_frags;
254         for (i = 0; i < frag_num; i++) {
255                 frag = &skb_shinfo(skb)->frags[i];
256                 size = skb_frag_size(frag);
257                 buf_num += (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
258         }
259
260         if (unlikely(buf_num > ring->max_desc_num_per_pkt)) {
261                 buf_num = (skb->len + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
262                 if (ring_space(ring) < buf_num)
263                         return -EBUSY;
264                 /* manual split the send packet */
265                 new_skb = skb_copy(skb, GFP_ATOMIC);
266                 if (!new_skb)
267                         return -ENOMEM;
268                 dev_kfree_skb_any(skb);
269                 *out_skb = new_skb;
270
271         } else if (ring_space(ring) < buf_num) {
272                 return -EBUSY;
273         }
274
275         *bnum = buf_num;
276         return 0;
277 }
278
279 static int hns_nic_maybe_stop_tx_v2(struct sk_buff **out_skb, int *bnum,
280                                     struct hnae_ring *ring)
281 {
282         if (skb_is_gso(*out_skb))
283                 return hns_nic_maybe_stop_tso(out_skb, bnum, ring);
284         else
285                 return hns_nic_maybe_stop_tx(out_skb, bnum, ring);
286 }
287
288 static void fill_tso_desc(struct hnae_ring *ring, void *priv,
289                           int size, dma_addr_t dma, int frag_end,
290                           int buf_num, enum hns_desc_type type, int mtu)
291 {
292         int frag_buf_num;
293         int sizeoflast;
294         int k;
295
296         frag_buf_num = (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
297         sizeoflast = size % BD_MAX_SEND_SIZE;
298         sizeoflast = sizeoflast ? sizeoflast : BD_MAX_SEND_SIZE;
299
300         /* when the frag size is bigger than hardware, split this frag */
301         for (k = 0; k < frag_buf_num; k++)
302                 fill_v2_desc_hw(ring, priv, k == 0 ? size : 0,
303                                 (k == frag_buf_num - 1) ?
304                                         sizeoflast : BD_MAX_SEND_SIZE,
305                                 dma + BD_MAX_SEND_SIZE * k,
306                                 frag_end && (k == frag_buf_num - 1) ? 1 : 0,
307                                 buf_num,
308                                 (type == DESC_TYPE_SKB && !k) ?
309                                         DESC_TYPE_SKB : DESC_TYPE_PAGE,
310                                 mtu);
311 }
312
313 static void fill_desc_v2(struct hnae_ring *ring, void *priv,
314                          int size, dma_addr_t dma, int frag_end,
315                          int buf_num, enum hns_desc_type type, int mtu,
316                          bool is_gso)
317 {
318         if (is_gso)
319                 fill_tso_desc(ring, priv, size, dma, frag_end, buf_num, type,
320                               mtu);
321         else
322                 fill_v2_desc(ring, priv, size, dma, frag_end, buf_num, type,
323                              mtu);
324 }
325
326 netdev_tx_t hns_nic_net_xmit_hw(struct net_device *ndev,
327                                 struct sk_buff *skb,
328                                 struct hns_nic_ring_data *ring_data)
329 {
330         struct hns_nic_priv *priv = netdev_priv(ndev);
331         struct hnae_ring *ring = ring_data->ring;
332         struct device *dev = ring_to_dev(ring);
333         struct netdev_queue *dev_queue;
334         skb_frag_t *frag;
335         int buf_num;
336         int seg_num;
337         dma_addr_t dma;
338         int size, next_to_use;
339         bool is_gso;
340         int i;
341
342         switch (priv->ops.maybe_stop_tx(&skb, &buf_num, ring)) {
343         case -EBUSY:
344                 ring->stats.tx_busy++;
345                 goto out_net_tx_busy;
346         case -ENOMEM:
347                 ring->stats.sw_err_cnt++;
348                 netdev_err(ndev, "no memory to xmit!\n");
349                 goto out_err_tx_ok;
350         default:
351                 break;
352         }
353
354         /* no. of segments (plus a header) */
355         seg_num = skb_shinfo(skb)->nr_frags + 1;
356         next_to_use = ring->next_to_use;
357
358         /* fill the first part */
359         size = skb_headlen(skb);
360         dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
361         if (dma_mapping_error(dev, dma)) {
362                 netdev_err(ndev, "TX head DMA map failed\n");
363                 ring->stats.sw_err_cnt++;
364                 goto out_err_tx_ok;
365         }
366         is_gso = skb_is_gso(skb);
367         priv->ops.fill_desc(ring, skb, size, dma, seg_num == 1 ? 1 : 0,
368                             buf_num, DESC_TYPE_SKB, ndev->mtu, is_gso);
369
370         /* fill the fragments */
371         for (i = 1; i < seg_num; i++) {
372                 frag = &skb_shinfo(skb)->frags[i - 1];
373                 size = skb_frag_size(frag);
374                 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
375                 if (dma_mapping_error(dev, dma)) {
376                         netdev_err(ndev, "TX frag(%d) DMA map failed\n", i);
377                         ring->stats.sw_err_cnt++;
378                         goto out_map_frag_fail;
379                 }
380                 priv->ops.fill_desc(ring, skb_frag_page(frag), size, dma,
381                                     seg_num - 1 == i ? 1 : 0, buf_num,
382                                     DESC_TYPE_PAGE, ndev->mtu, is_gso);
383         }
384
385         /*complete translate all packets*/
386         dev_queue = netdev_get_tx_queue(ndev, skb->queue_mapping);
387         netdev_tx_sent_queue(dev_queue, skb->len);
388
389         netif_trans_update(ndev);
390         ndev->stats.tx_bytes += skb->len;
391         ndev->stats.tx_packets++;
392
393         wmb(); /* commit all data before submit */
394         assert(skb->queue_mapping < priv->ae_handle->q_num);
395         hnae_queue_xmit(priv->ae_handle->qs[skb->queue_mapping], buf_num);
396
397         return NETDEV_TX_OK;
398
399 out_map_frag_fail:
400
401         while (ring->next_to_use != next_to_use) {
402                 unfill_desc(ring);
403                 if (ring->next_to_use != next_to_use)
404                         dma_unmap_page(dev,
405                                        ring->desc_cb[ring->next_to_use].dma,
406                                        ring->desc_cb[ring->next_to_use].length,
407                                        DMA_TO_DEVICE);
408                 else
409                         dma_unmap_single(dev,
410                                          ring->desc_cb[next_to_use].dma,
411                                          ring->desc_cb[next_to_use].length,
412                                          DMA_TO_DEVICE);
413         }
414
415 out_err_tx_ok:
416
417         dev_kfree_skb_any(skb);
418         return NETDEV_TX_OK;
419
420 out_net_tx_busy:
421
422         netif_stop_subqueue(ndev, skb->queue_mapping);
423
424         /* Herbert's original patch had:
425          *  smp_mb__after_netif_stop_queue();
426          * but since that doesn't exist yet, just open code it.
427          */
428         smp_mb();
429         return NETDEV_TX_BUSY;
430 }
431
432 static void hns_nic_reuse_page(struct sk_buff *skb, int i,
433                                struct hnae_ring *ring, int pull_len,
434                                struct hnae_desc_cb *desc_cb)
435 {
436         struct hnae_desc *desc;
437         u32 truesize;
438         int size;
439         int last_offset;
440         bool twobufs;
441
442         twobufs = ((PAGE_SIZE < 8192) &&
443                 hnae_buf_size(ring) == HNS_BUFFER_SIZE_2048);
444
445         desc = &ring->desc[ring->next_to_clean];
446         size = le16_to_cpu(desc->rx.size);
447
448         if (twobufs) {
449                 truesize = hnae_buf_size(ring);
450         } else {
451                 truesize = ALIGN(size, L1_CACHE_BYTES);
452                 last_offset = hnae_page_size(ring) - hnae_buf_size(ring);
453         }
454
455         skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
456                         size - pull_len, truesize);
457
458          /* avoid re-using remote pages,flag default unreuse */
459         if (unlikely(page_to_nid(desc_cb->priv) != numa_node_id()))
460                 return;
461
462         if (twobufs) {
463                 /* if we are only owner of page we can reuse it */
464                 if (likely(page_count(desc_cb->priv) == 1)) {
465                         /* flip page offset to other buffer */
466                         desc_cb->page_offset ^= truesize;
467
468                         desc_cb->reuse_flag = 1;
469                         /* bump ref count on page before it is given*/
470                         get_page(desc_cb->priv);
471                 }
472                 return;
473         }
474
475         /* move offset up to the next cache line */
476         desc_cb->page_offset += truesize;
477
478         if (desc_cb->page_offset <= last_offset) {
479                 desc_cb->reuse_flag = 1;
480                 /* bump ref count on page before it is given*/
481                 get_page(desc_cb->priv);
482         }
483 }
484
485 static void get_v2rx_desc_bnum(u32 bnum_flag, int *out_bnum)
486 {
487         *out_bnum = hnae_get_field(bnum_flag,
488                                    HNS_RXD_BUFNUM_M, HNS_RXD_BUFNUM_S) + 1;
489 }
490
491 static void get_rx_desc_bnum(u32 bnum_flag, int *out_bnum)
492 {
493         *out_bnum = hnae_get_field(bnum_flag,
494                                    HNS_RXD_BUFNUM_M, HNS_RXD_BUFNUM_S);
495 }
496
497 static void hns_nic_rx_checksum(struct hns_nic_ring_data *ring_data,
498                                 struct sk_buff *skb, u32 flag)
499 {
500         struct net_device *netdev = ring_data->napi.dev;
501         u32 l3id;
502         u32 l4id;
503
504         /* check if RX checksum offload is enabled */
505         if (unlikely(!(netdev->features & NETIF_F_RXCSUM)))
506                 return;
507
508         /* In hardware, we only support checksum for the following protocols:
509          * 1) IPv4,
510          * 2) TCP(over IPv4 or IPv6),
511          * 3) UDP(over IPv4 or IPv6),
512          * 4) SCTP(over IPv4 or IPv6)
513          * but we support many L3(IPv4, IPv6, MPLS, PPPoE etc) and L4(TCP,
514          * UDP, GRE, SCTP, IGMP, ICMP etc.) protocols.
515          *
516          * Hardware limitation:
517          * Our present hardware RX Descriptor lacks L3/L4 checksum "Status &
518          * Error" bit (which usually can be used to indicate whether checksum
519          * was calculated by the hardware and if there was any error encountered
520          * during checksum calculation).
521          *
522          * Software workaround:
523          * We do get info within the RX descriptor about the kind of L3/L4
524          * protocol coming in the packet and the error status. These errors
525          * might not just be checksum errors but could be related to version,
526          * length of IPv4, UDP, TCP etc.
527          * Because there is no-way of knowing if it is a L3/L4 error due to bad
528          * checksum or any other L3/L4 error, we will not (cannot) convey
529          * checksum status for such cases to upper stack and will not maintain
530          * the RX L3/L4 checksum counters as well.
531          */
532
533         l3id = hnae_get_field(flag, HNS_RXD_L3ID_M, HNS_RXD_L3ID_S);
534         l4id = hnae_get_field(flag, HNS_RXD_L4ID_M, HNS_RXD_L4ID_S);
535
536         /*  check L3 protocol for which checksum is supported */
537         if ((l3id != HNS_RX_FLAG_L3ID_IPV4) && (l3id != HNS_RX_FLAG_L3ID_IPV6))
538                 return;
539
540         /* check for any(not just checksum)flagged L3 protocol errors */
541         if (unlikely(hnae_get_bit(flag, HNS_RXD_L3E_B)))
542                 return;
543
544         /* we do not support checksum of fragmented packets */
545         if (unlikely(hnae_get_bit(flag, HNS_RXD_FRAG_B)))
546                 return;
547
548         /*  check L4 protocol for which checksum is supported */
549         if ((l4id != HNS_RX_FLAG_L4ID_TCP) &&
550             (l4id != HNS_RX_FLAG_L4ID_UDP) &&
551             (l4id != HNS_RX_FLAG_L4ID_SCTP))
552                 return;
553
554         /* check for any(not just checksum)flagged L4 protocol errors */
555         if (unlikely(hnae_get_bit(flag, HNS_RXD_L4E_B)))
556                 return;
557
558         /* now, this has to be a packet with valid RX checksum */
559         skb->ip_summed = CHECKSUM_UNNECESSARY;
560 }
561
562 static int hns_nic_poll_rx_skb(struct hns_nic_ring_data *ring_data,
563                                struct sk_buff **out_skb, int *out_bnum)
564 {
565         struct hnae_ring *ring = ring_data->ring;
566         struct net_device *ndev = ring_data->napi.dev;
567         struct hns_nic_priv *priv = netdev_priv(ndev);
568         struct sk_buff *skb;
569         struct hnae_desc *desc;
570         struct hnae_desc_cb *desc_cb;
571         unsigned char *va;
572         int bnum, length, i;
573         int pull_len;
574         u32 bnum_flag;
575
576         desc = &ring->desc[ring->next_to_clean];
577         desc_cb = &ring->desc_cb[ring->next_to_clean];
578
579         prefetch(desc);
580
581         va = (unsigned char *)desc_cb->buf + desc_cb->page_offset;
582
583         /* prefetch first cache line of first page */
584         net_prefetch(va);
585
586         skb = *out_skb = napi_alloc_skb(&ring_data->napi,
587                                         HNS_RX_HEAD_SIZE);
588         if (unlikely(!skb)) {
589                 ring->stats.sw_err_cnt++;
590                 return -ENOMEM;
591         }
592
593         prefetchw(skb->data);
594         length = le16_to_cpu(desc->rx.pkt_len);
595         bnum_flag = le32_to_cpu(desc->rx.ipoff_bnum_pid_flag);
596         priv->ops.get_rxd_bnum(bnum_flag, &bnum);
597         *out_bnum = bnum;
598
599         if (length <= HNS_RX_HEAD_SIZE) {
600                 memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
601
602                 /* we can reuse buffer as-is, just make sure it is local */
603                 if (likely(page_to_nid(desc_cb->priv) == numa_node_id()))
604                         desc_cb->reuse_flag = 1;
605                 else /* this page cannot be reused so discard it */
606                         put_page(desc_cb->priv);
607
608                 ring_ptr_move_fw(ring, next_to_clean);
609
610                 if (unlikely(bnum != 1)) { /* check err*/
611                         *out_bnum = 1;
612                         goto out_bnum_err;
613                 }
614         } else {
615                 ring->stats.seg_pkt_cnt++;
616
617                 pull_len = eth_get_headlen(ndev, va, HNS_RX_HEAD_SIZE);
618                 memcpy(__skb_put(skb, pull_len), va,
619                        ALIGN(pull_len, sizeof(long)));
620
621                 hns_nic_reuse_page(skb, 0, ring, pull_len, desc_cb);
622                 ring_ptr_move_fw(ring, next_to_clean);
623
624                 if (unlikely(bnum >= (int)MAX_SKB_FRAGS)) { /* check err*/
625                         *out_bnum = 1;
626                         goto out_bnum_err;
627                 }
628                 for (i = 1; i < bnum; i++) {
629                         desc = &ring->desc[ring->next_to_clean];
630                         desc_cb = &ring->desc_cb[ring->next_to_clean];
631
632                         hns_nic_reuse_page(skb, i, ring, 0, desc_cb);
633                         ring_ptr_move_fw(ring, next_to_clean);
634                 }
635         }
636
637         /* check except process, free skb and jump the desc */
638         if (unlikely((!bnum) || (bnum > ring->max_desc_num_per_pkt))) {
639 out_bnum_err:
640                 *out_bnum = *out_bnum ? *out_bnum : 1; /* ntc moved,cannot 0*/
641                 netdev_err(ndev, "invalid bnum(%d,%d,%d,%d),%016llx,%016llx\n",
642                            bnum, ring->max_desc_num_per_pkt,
643                            length, (int)MAX_SKB_FRAGS,
644                            ((u64 *)desc)[0], ((u64 *)desc)[1]);
645                 ring->stats.err_bd_num++;
646                 dev_kfree_skb_any(skb);
647                 return -EDOM;
648         }
649
650         bnum_flag = le32_to_cpu(desc->rx.ipoff_bnum_pid_flag);
651
652         if (unlikely(!hnae_get_bit(bnum_flag, HNS_RXD_VLD_B))) {
653                 netdev_err(ndev, "no valid bd,%016llx,%016llx\n",
654                            ((u64 *)desc)[0], ((u64 *)desc)[1]);
655                 ring->stats.non_vld_descs++;
656                 dev_kfree_skb_any(skb);
657                 return -EINVAL;
658         }
659
660         if (unlikely((!desc->rx.pkt_len) ||
661                      hnae_get_bit(bnum_flag, HNS_RXD_DROP_B))) {
662                 ring->stats.err_pkt_len++;
663                 dev_kfree_skb_any(skb);
664                 return -EFAULT;
665         }
666
667         if (unlikely(hnae_get_bit(bnum_flag, HNS_RXD_L2E_B))) {
668                 ring->stats.l2_err++;
669                 dev_kfree_skb_any(skb);
670                 return -EFAULT;
671         }
672
673         ring->stats.rx_pkts++;
674         ring->stats.rx_bytes += skb->len;
675
676         /* indicate to upper stack if our hardware has already calculated
677          * the RX checksum
678          */
679         hns_nic_rx_checksum(ring_data, skb, bnum_flag);
680
681         return 0;
682 }
683
684 static void
685 hns_nic_alloc_rx_buffers(struct hns_nic_ring_data *ring_data, int cleand_count)
686 {
687         int i, ret;
688         struct hnae_desc_cb res_cbs;
689         struct hnae_desc_cb *desc_cb;
690         struct hnae_ring *ring = ring_data->ring;
691         struct net_device *ndev = ring_data->napi.dev;
692
693         for (i = 0; i < cleand_count; i++) {
694                 desc_cb = &ring->desc_cb[ring->next_to_use];
695                 if (desc_cb->reuse_flag) {
696                         ring->stats.reuse_pg_cnt++;
697                         hnae_reuse_buffer(ring, ring->next_to_use);
698                 } else {
699                         ret = hnae_reserve_buffer_map(ring, &res_cbs);
700                         if (ret) {
701                                 ring->stats.sw_err_cnt++;
702                                 netdev_err(ndev, "hnae reserve buffer map failed.\n");
703                                 break;
704                         }
705                         hnae_replace_buffer(ring, ring->next_to_use, &res_cbs);
706                 }
707
708                 ring_ptr_move_fw(ring, next_to_use);
709         }
710
711         wmb(); /* make all data has been write before submit */
712         writel_relaxed(i, ring->io_base + RCB_REG_HEAD);
713 }
714
715 /* return error number for error or number of desc left to take
716  */
717 static void hns_nic_rx_up_pro(struct hns_nic_ring_data *ring_data,
718                               struct sk_buff *skb)
719 {
720         struct net_device *ndev = ring_data->napi.dev;
721
722         skb->protocol = eth_type_trans(skb, ndev);
723         napi_gro_receive(&ring_data->napi, skb);
724 }
725
726 static int hns_desc_unused(struct hnae_ring *ring)
727 {
728         int ntc = ring->next_to_clean;
729         int ntu = ring->next_to_use;
730
731         return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
732 }
733
734 #define HNS_LOWEST_LATENCY_RATE         27      /* 27 MB/s */
735 #define HNS_LOW_LATENCY_RATE                    80      /* 80 MB/s */
736
737 #define HNS_COAL_BDNUM                  3
738
739 static u32 hns_coal_rx_bdnum(struct hnae_ring *ring)
740 {
741         bool coal_enable = ring->q->handle->coal_adapt_en;
742
743         if (coal_enable &&
744             ring->coal_last_rx_bytes > HNS_LOWEST_LATENCY_RATE)
745                 return HNS_COAL_BDNUM;
746         else
747                 return 0;
748 }
749
750 static void hns_update_rx_rate(struct hnae_ring *ring)
751 {
752         bool coal_enable = ring->q->handle->coal_adapt_en;
753         u32 time_passed_ms;
754         u64 total_bytes;
755
756         if (!coal_enable ||
757             time_before(jiffies, ring->coal_last_jiffies + (HZ >> 4)))
758                 return;
759
760         /* ring->stats.rx_bytes overflowed */
761         if (ring->coal_last_rx_bytes > ring->stats.rx_bytes) {
762                 ring->coal_last_rx_bytes = ring->stats.rx_bytes;
763                 ring->coal_last_jiffies = jiffies;
764                 return;
765         }
766
767         total_bytes = ring->stats.rx_bytes - ring->coal_last_rx_bytes;
768         time_passed_ms = jiffies_to_msecs(jiffies - ring->coal_last_jiffies);
769         do_div(total_bytes, time_passed_ms);
770         ring->coal_rx_rate = total_bytes >> 10;
771
772         ring->coal_last_rx_bytes = ring->stats.rx_bytes;
773         ring->coal_last_jiffies = jiffies;
774 }
775
776 /**
777  * smooth_alg - smoothing algrithm for adjusting coalesce parameter
778  * @new_param: new value
779  * @old_param: old value
780  **/
781 static u32 smooth_alg(u32 new_param, u32 old_param)
782 {
783         u32 gap = (new_param > old_param) ? new_param - old_param
784                                           : old_param - new_param;
785
786         if (gap > 8)
787                 gap >>= 3;
788
789         if (new_param > old_param)
790                 return old_param + gap;
791         else
792                 return old_param - gap;
793 }
794
795 /**
796  * hns_nic_adpt_coalesce - self adapte coalesce according to rx rate
797  * @ring_data: pointer to hns_nic_ring_data
798  **/
799 static void hns_nic_adpt_coalesce(struct hns_nic_ring_data *ring_data)
800 {
801         struct hnae_ring *ring = ring_data->ring;
802         struct hnae_handle *handle = ring->q->handle;
803         u32 new_coal_param, old_coal_param = ring->coal_param;
804
805         if (ring->coal_rx_rate < HNS_LOWEST_LATENCY_RATE)
806                 new_coal_param = HNAE_LOWEST_LATENCY_COAL_PARAM;
807         else if (ring->coal_rx_rate < HNS_LOW_LATENCY_RATE)
808                 new_coal_param = HNAE_LOW_LATENCY_COAL_PARAM;
809         else
810                 new_coal_param = HNAE_BULK_LATENCY_COAL_PARAM;
811
812         if (new_coal_param == old_coal_param &&
813             new_coal_param == handle->coal_param)
814                 return;
815
816         new_coal_param = smooth_alg(new_coal_param, old_coal_param);
817         ring->coal_param = new_coal_param;
818
819         /**
820          * Because all ring in one port has one coalesce param, when one ring
821          * calculate its own coalesce param, it cannot write to hardware at
822          * once. There are three conditions as follows:
823          *       1. current ring's coalesce param is larger than the hardware.
824          *       2. or ring which adapt last time can change again.
825          *       3. timeout.
826          */
827         if (new_coal_param == handle->coal_param) {
828                 handle->coal_last_jiffies = jiffies;
829                 handle->coal_ring_idx = ring_data->queue_index;
830         } else if (new_coal_param > handle->coal_param ||
831                    handle->coal_ring_idx == ring_data->queue_index ||
832                    time_after(jiffies, handle->coal_last_jiffies + (HZ >> 4))) {
833                 handle->dev->ops->set_coalesce_usecs(handle,
834                                         new_coal_param);
835                 handle->dev->ops->set_coalesce_frames(handle,
836                                         1, new_coal_param);
837                 handle->coal_param = new_coal_param;
838                 handle->coal_ring_idx = ring_data->queue_index;
839                 handle->coal_last_jiffies = jiffies;
840         }
841 }
842
843 static int hns_nic_rx_poll_one(struct hns_nic_ring_data *ring_data,
844                                int budget, void *v)
845 {
846         struct hnae_ring *ring = ring_data->ring;
847         struct sk_buff *skb;
848         int num, bnum;
849 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
850         int recv_pkts, recv_bds, clean_count, err;
851         int unused_count = hns_desc_unused(ring);
852
853         num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
854         rmb(); /* make sure num taken effect before the other data is touched */
855
856         recv_pkts = 0, recv_bds = 0, clean_count = 0;
857         num -= unused_count;
858
859         while (recv_pkts < budget && recv_bds < num) {
860                 /* reuse or realloc buffers */
861                 if (clean_count + unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
862                         hns_nic_alloc_rx_buffers(ring_data,
863                                                  clean_count + unused_count);
864                         clean_count = 0;
865                         unused_count = hns_desc_unused(ring);
866                 }
867
868                 /* poll one pkt */
869                 err = hns_nic_poll_rx_skb(ring_data, &skb, &bnum);
870                 if (unlikely(!skb)) /* this fault cannot be repaired */
871                         goto out;
872
873                 recv_bds += bnum;
874                 clean_count += bnum;
875                 if (unlikely(err)) {  /* do jump the err */
876                         recv_pkts++;
877                         continue;
878                 }
879
880                 /* do update ip stack process*/
881                 ((void (*)(struct hns_nic_ring_data *, struct sk_buff *))v)(
882                                                         ring_data, skb);
883                 recv_pkts++;
884         }
885
886 out:
887         /* make all data has been write before submit */
888         if (clean_count + unused_count > 0)
889                 hns_nic_alloc_rx_buffers(ring_data,
890                                          clean_count + unused_count);
891
892         return recv_pkts;
893 }
894
895 static bool hns_nic_rx_fini_pro(struct hns_nic_ring_data *ring_data)
896 {
897         struct hnae_ring *ring = ring_data->ring;
898         int num;
899         bool rx_stopped;
900
901         hns_update_rx_rate(ring);
902
903         /* for hardware bug fixed */
904         ring_data->ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
905         num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
906
907         if (num <= hns_coal_rx_bdnum(ring)) {
908                 if (ring->q->handle->coal_adapt_en)
909                         hns_nic_adpt_coalesce(ring_data);
910
911                 rx_stopped = true;
912         } else {
913                 ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
914                         ring_data->ring, 1);
915
916                 rx_stopped = false;
917         }
918
919         return rx_stopped;
920 }
921
922 static bool hns_nic_rx_fini_pro_v2(struct hns_nic_ring_data *ring_data)
923 {
924         struct hnae_ring *ring = ring_data->ring;
925         int num;
926
927         hns_update_rx_rate(ring);
928         num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
929
930         if (num <= hns_coal_rx_bdnum(ring)) {
931                 if (ring->q->handle->coal_adapt_en)
932                         hns_nic_adpt_coalesce(ring_data);
933
934                 return true;
935         }
936
937         return false;
938 }
939
940 static inline void hns_nic_reclaim_one_desc(struct hnae_ring *ring,
941                                             int *bytes, int *pkts)
942 {
943         struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
944
945         (*pkts) += (desc_cb->type == DESC_TYPE_SKB);
946         (*bytes) += desc_cb->length;
947         /* desc_cb will be cleaned, after hnae_free_buffer_detach*/
948         hnae_free_buffer_detach(ring, ring->next_to_clean);
949
950         ring_ptr_move_fw(ring, next_to_clean);
951 }
952
953 static int is_valid_clean_head(struct hnae_ring *ring, int h)
954 {
955         int u = ring->next_to_use;
956         int c = ring->next_to_clean;
957
958         if (unlikely(h > ring->desc_num))
959                 return 0;
960
961         assert(u > 0 && u < ring->desc_num);
962         assert(c > 0 && c < ring->desc_num);
963         assert(u != c && h != c); /* must be checked before call this func */
964
965         return u > c ? (h > c && h <= u) : (h > c || h <= u);
966 }
967
968 /* reclaim all desc in one budget
969  * return error or number of desc left
970  */
971 static int hns_nic_tx_poll_one(struct hns_nic_ring_data *ring_data,
972                                int budget, void *v)
973 {
974         struct hnae_ring *ring = ring_data->ring;
975         struct net_device *ndev = ring_data->napi.dev;
976         struct netdev_queue *dev_queue;
977         struct hns_nic_priv *priv = netdev_priv(ndev);
978         int head;
979         int bytes, pkts;
980
981         head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
982         rmb(); /* make sure head is ready before touch any data */
983
984         if (is_ring_empty(ring) || head == ring->next_to_clean)
985                 return 0; /* no data to poll */
986
987         if (!is_valid_clean_head(ring, head)) {
988                 netdev_err(ndev, "wrong head (%d, %d-%d)\n", head,
989                            ring->next_to_use, ring->next_to_clean);
990                 ring->stats.io_err_cnt++;
991                 return -EIO;
992         }
993
994         bytes = 0;
995         pkts = 0;
996         while (head != ring->next_to_clean) {
997                 hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
998                 /* issue prefetch for next Tx descriptor */
999                 prefetch(&ring->desc_cb[ring->next_to_clean]);
1000         }
1001         /* update tx ring statistics. */
1002         ring->stats.tx_pkts += pkts;
1003         ring->stats.tx_bytes += bytes;
1004
1005         dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
1006         netdev_tx_completed_queue(dev_queue, pkts, bytes);
1007
1008         if (unlikely(priv->link && !netif_carrier_ok(ndev)))
1009                 netif_carrier_on(ndev);
1010
1011         if (unlikely(pkts && netif_carrier_ok(ndev) &&
1012                      (ring_space(ring) >= ring->max_desc_num_per_pkt * 2))) {
1013                 /* Make sure that anybody stopping the queue after this
1014                  * sees the new next_to_clean.
1015                  */
1016                 smp_mb();
1017                 if (netif_tx_queue_stopped(dev_queue) &&
1018                     !test_bit(NIC_STATE_DOWN, &priv->state)) {
1019                         netif_tx_wake_queue(dev_queue);
1020                         ring->stats.restart_queue++;
1021                 }
1022         }
1023         return 0;
1024 }
1025
1026 static bool hns_nic_tx_fini_pro(struct hns_nic_ring_data *ring_data)
1027 {
1028         struct hnae_ring *ring = ring_data->ring;
1029         int head;
1030
1031         ring_data->ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
1032
1033         head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1034
1035         if (head != ring->next_to_clean) {
1036                 ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1037                         ring_data->ring, 1);
1038
1039                 return false;
1040         } else {
1041                 return true;
1042         }
1043 }
1044
1045 static bool hns_nic_tx_fini_pro_v2(struct hns_nic_ring_data *ring_data)
1046 {
1047         struct hnae_ring *ring = ring_data->ring;
1048         int head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1049
1050         if (head == ring->next_to_clean)
1051                 return true;
1052         else
1053                 return false;
1054 }
1055
1056 static void hns_nic_tx_clr_all_bufs(struct hns_nic_ring_data *ring_data)
1057 {
1058         struct hnae_ring *ring = ring_data->ring;
1059         struct net_device *ndev = ring_data->napi.dev;
1060         struct netdev_queue *dev_queue;
1061         int head;
1062         int bytes, pkts;
1063
1064         head = ring->next_to_use; /* ntu :soft setted ring position*/
1065         bytes = 0;
1066         pkts = 0;
1067         while (head != ring->next_to_clean)
1068                 hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
1069
1070         dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
1071         netdev_tx_reset_queue(dev_queue);
1072 }
1073
1074 static int hns_nic_common_poll(struct napi_struct *napi, int budget)
1075 {
1076         int clean_complete = 0;
1077         struct hns_nic_ring_data *ring_data =
1078                 container_of(napi, struct hns_nic_ring_data, napi);
1079         struct hnae_ring *ring = ring_data->ring;
1080
1081         clean_complete += ring_data->poll_one(
1082                                 ring_data, budget - clean_complete,
1083                                 ring_data->ex_process);
1084
1085         if (clean_complete < budget) {
1086                 if (ring_data->fini_process(ring_data)) {
1087                         napi_complete(napi);
1088                         ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
1089                 } else {
1090                         return budget;
1091                 }
1092         }
1093
1094         return clean_complete;
1095 }
1096
1097 static irqreturn_t hns_irq_handle(int irq, void *dev)
1098 {
1099         struct hns_nic_ring_data *ring_data = (struct hns_nic_ring_data *)dev;
1100
1101         ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1102                 ring_data->ring, 1);
1103         napi_schedule(&ring_data->napi);
1104
1105         return IRQ_HANDLED;
1106 }
1107
1108 /**
1109  *hns_nic_adjust_link - adjust net work mode by the phy stat or new param
1110  *@ndev: net device
1111  */
1112 static void hns_nic_adjust_link(struct net_device *ndev)
1113 {
1114         struct hns_nic_priv *priv = netdev_priv(ndev);
1115         struct hnae_handle *h = priv->ae_handle;
1116         int state = 1;
1117
1118         /* If there is no phy, do not need adjust link */
1119         if (ndev->phydev) {
1120                 /* When phy link down, do nothing */
1121                 if (ndev->phydev->link == 0)
1122                         return;
1123
1124                 if (h->dev->ops->need_adjust_link(h, ndev->phydev->speed,
1125                                                   ndev->phydev->duplex)) {
1126                         /* because Hi161X chip don't support to change gmac
1127                          * speed and duplex with traffic. Delay 200ms to
1128                          * make sure there is no more data in chip FIFO.
1129                          */
1130                         netif_carrier_off(ndev);
1131                         msleep(200);
1132                         h->dev->ops->adjust_link(h, ndev->phydev->speed,
1133                                                  ndev->phydev->duplex);
1134                         netif_carrier_on(ndev);
1135                 }
1136         }
1137
1138         state = state && h->dev->ops->get_status(h);
1139
1140         if (state != priv->link) {
1141                 if (state) {
1142                         netif_carrier_on(ndev);
1143                         netif_tx_wake_all_queues(ndev);
1144                         netdev_info(ndev, "link up\n");
1145                 } else {
1146                         netif_carrier_off(ndev);
1147                         netdev_info(ndev, "link down\n");
1148                 }
1149                 priv->link = state;
1150         }
1151 }
1152
1153 /**
1154  *hns_nic_init_phy - init phy
1155  *@ndev: net device
1156  *@h: ae handle
1157  * Return 0 on success, negative on failure
1158  */
1159 int hns_nic_init_phy(struct net_device *ndev, struct hnae_handle *h)
1160 {
1161         __ETHTOOL_DECLARE_LINK_MODE_MASK(supported) = { 0, };
1162         struct phy_device *phy_dev = h->phy_dev;
1163         int ret;
1164
1165         if (!h->phy_dev)
1166                 return 0;
1167
1168         ethtool_convert_legacy_u32_to_link_mode(supported, h->if_support);
1169         linkmode_and(phy_dev->supported, phy_dev->supported, supported);
1170         linkmode_copy(phy_dev->advertising, phy_dev->supported);
1171
1172         if (h->phy_if == PHY_INTERFACE_MODE_XGMII)
1173                 phy_dev->autoneg = false;
1174
1175         if (h->phy_if != PHY_INTERFACE_MODE_XGMII) {
1176                 phy_dev->dev_flags = 0;
1177
1178                 ret = phy_connect_direct(ndev, phy_dev, hns_nic_adjust_link,
1179                                          h->phy_if);
1180         } else {
1181                 ret = phy_attach_direct(ndev, phy_dev, 0, h->phy_if);
1182         }
1183         if (unlikely(ret))
1184                 return -ENODEV;
1185
1186         phy_attached_info(phy_dev);
1187
1188         return 0;
1189 }
1190
1191 static int hns_nic_ring_open(struct net_device *netdev, int idx)
1192 {
1193         struct hns_nic_priv *priv = netdev_priv(netdev);
1194         struct hnae_handle *h = priv->ae_handle;
1195
1196         napi_enable(&priv->ring_data[idx].napi);
1197
1198         enable_irq(priv->ring_data[idx].ring->irq);
1199         h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 0);
1200
1201         return 0;
1202 }
1203
1204 static int hns_nic_net_set_mac_address(struct net_device *ndev, void *p)
1205 {
1206         struct hns_nic_priv *priv = netdev_priv(ndev);
1207         struct hnae_handle *h = priv->ae_handle;
1208         struct sockaddr *mac_addr = p;
1209         int ret;
1210
1211         if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1212                 return -EADDRNOTAVAIL;
1213
1214         ret = h->dev->ops->set_mac_addr(h, mac_addr->sa_data);
1215         if (ret) {
1216                 netdev_err(ndev, "set_mac_address fail, ret=%d!\n", ret);
1217                 return ret;
1218         }
1219
1220         eth_hw_addr_set(ndev, mac_addr->sa_data);
1221
1222         return 0;
1223 }
1224
1225 static void hns_nic_update_stats(struct net_device *netdev)
1226 {
1227         struct hns_nic_priv *priv = netdev_priv(netdev);
1228         struct hnae_handle *h = priv->ae_handle;
1229
1230         h->dev->ops->update_stats(h, &netdev->stats);
1231 }
1232
1233 /* set mac addr if it is configed. or leave it to the AE driver */
1234 static void hns_init_mac_addr(struct net_device *ndev)
1235 {
1236         struct hns_nic_priv *priv = netdev_priv(ndev);
1237
1238         if (device_get_ethdev_address(priv->dev, ndev)) {
1239                 eth_hw_addr_random(ndev);
1240                 dev_warn(priv->dev, "No valid mac, use random mac %pM",
1241                          ndev->dev_addr);
1242         }
1243 }
1244
1245 static void hns_nic_ring_close(struct net_device *netdev, int idx)
1246 {
1247         struct hns_nic_priv *priv = netdev_priv(netdev);
1248         struct hnae_handle *h = priv->ae_handle;
1249
1250         h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 1);
1251         disable_irq(priv->ring_data[idx].ring->irq);
1252
1253         napi_disable(&priv->ring_data[idx].napi);
1254 }
1255
1256 static int hns_nic_init_affinity_mask(int q_num, int ring_idx,
1257                                       struct hnae_ring *ring, cpumask_t *mask)
1258 {
1259         int cpu;
1260
1261         /* Different irq balance between 16core and 32core.
1262          * The cpu mask set by ring index according to the ring flag
1263          * which indicate the ring is tx or rx.
1264          */
1265         if (q_num == num_possible_cpus()) {
1266                 if (is_tx_ring(ring))
1267                         cpu = ring_idx;
1268                 else
1269                         cpu = ring_idx - q_num;
1270         } else {
1271                 if (is_tx_ring(ring))
1272                         cpu = ring_idx * 2;
1273                 else
1274                         cpu = (ring_idx - q_num) * 2 + 1;
1275         }
1276
1277         cpumask_clear(mask);
1278         cpumask_set_cpu(cpu, mask);
1279
1280         return cpu;
1281 }
1282
1283 static void hns_nic_free_irq(int q_num, struct hns_nic_priv *priv)
1284 {
1285         int i;
1286
1287         for (i = 0; i < q_num * 2; i++) {
1288                 if (priv->ring_data[i].ring->irq_init_flag == RCB_IRQ_INITED) {
1289                         irq_set_affinity_hint(priv->ring_data[i].ring->irq,
1290                                               NULL);
1291                         free_irq(priv->ring_data[i].ring->irq,
1292                                  &priv->ring_data[i]);
1293                         priv->ring_data[i].ring->irq_init_flag =
1294                                 RCB_IRQ_NOT_INITED;
1295                 }
1296         }
1297 }
1298
1299 static int hns_nic_init_irq(struct hns_nic_priv *priv)
1300 {
1301         struct hnae_handle *h = priv->ae_handle;
1302         struct hns_nic_ring_data *rd;
1303         int i;
1304         int ret;
1305         int cpu;
1306
1307         for (i = 0; i < h->q_num * 2; i++) {
1308                 rd = &priv->ring_data[i];
1309
1310                 if (rd->ring->irq_init_flag == RCB_IRQ_INITED)
1311                         break;
1312
1313                 snprintf(rd->ring->ring_name, RCB_RING_NAME_LEN,
1314                          "%s-%s%d", priv->netdev->name,
1315                          (is_tx_ring(rd->ring) ? "tx" : "rx"), rd->queue_index);
1316
1317                 rd->ring->ring_name[RCB_RING_NAME_LEN - 1] = '\0';
1318
1319                 irq_set_status_flags(rd->ring->irq, IRQ_NOAUTOEN);
1320                 ret = request_irq(rd->ring->irq,
1321                                   hns_irq_handle, 0, rd->ring->ring_name, rd);
1322                 if (ret) {
1323                         netdev_err(priv->netdev, "request irq(%d) fail\n",
1324                                    rd->ring->irq);
1325                         goto out_free_irq;
1326                 }
1327
1328                 cpu = hns_nic_init_affinity_mask(h->q_num, i,
1329                                                  rd->ring, &rd->mask);
1330
1331                 if (cpu_online(cpu))
1332                         irq_set_affinity_hint(rd->ring->irq,
1333                                               &rd->mask);
1334
1335                 rd->ring->irq_init_flag = RCB_IRQ_INITED;
1336         }
1337
1338         return 0;
1339
1340 out_free_irq:
1341         hns_nic_free_irq(h->q_num, priv);
1342         return ret;
1343 }
1344
1345 static int hns_nic_net_up(struct net_device *ndev)
1346 {
1347         struct hns_nic_priv *priv = netdev_priv(ndev);
1348         struct hnae_handle *h = priv->ae_handle;
1349         int i, j;
1350         int ret;
1351
1352         if (!test_bit(NIC_STATE_DOWN, &priv->state))
1353                 return 0;
1354
1355         ret = hns_nic_init_irq(priv);
1356         if (ret != 0) {
1357                 netdev_err(ndev, "hns init irq failed! ret=%d\n", ret);
1358                 return ret;
1359         }
1360
1361         for (i = 0; i < h->q_num * 2; i++) {
1362                 ret = hns_nic_ring_open(ndev, i);
1363                 if (ret)
1364                         goto out_has_some_queues;
1365         }
1366
1367         ret = h->dev->ops->set_mac_addr(h, ndev->dev_addr);
1368         if (ret)
1369                 goto out_set_mac_addr_err;
1370
1371         ret = h->dev->ops->start ? h->dev->ops->start(h) : 0;
1372         if (ret)
1373                 goto out_start_err;
1374
1375         if (ndev->phydev)
1376                 phy_start(ndev->phydev);
1377
1378         clear_bit(NIC_STATE_DOWN, &priv->state);
1379         (void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
1380
1381         return 0;
1382
1383 out_start_err:
1384         netif_stop_queue(ndev);
1385 out_set_mac_addr_err:
1386 out_has_some_queues:
1387         for (j = i - 1; j >= 0; j--)
1388                 hns_nic_ring_close(ndev, j);
1389
1390         hns_nic_free_irq(h->q_num, priv);
1391         set_bit(NIC_STATE_DOWN, &priv->state);
1392
1393         return ret;
1394 }
1395
1396 static void hns_nic_net_down(struct net_device *ndev)
1397 {
1398         int i;
1399         struct hnae_ae_ops *ops;
1400         struct hns_nic_priv *priv = netdev_priv(ndev);
1401
1402         if (test_and_set_bit(NIC_STATE_DOWN, &priv->state))
1403                 return;
1404
1405         (void)del_timer_sync(&priv->service_timer);
1406         netif_tx_stop_all_queues(ndev);
1407         netif_carrier_off(ndev);
1408         netif_tx_disable(ndev);
1409         priv->link = 0;
1410
1411         if (ndev->phydev)
1412                 phy_stop(ndev->phydev);
1413
1414         ops = priv->ae_handle->dev->ops;
1415
1416         if (ops->stop)
1417                 ops->stop(priv->ae_handle);
1418
1419         netif_tx_stop_all_queues(ndev);
1420
1421         for (i = priv->ae_handle->q_num - 1; i >= 0; i--) {
1422                 hns_nic_ring_close(ndev, i);
1423                 hns_nic_ring_close(ndev, i + priv->ae_handle->q_num);
1424
1425                 /* clean tx buffers*/
1426                 hns_nic_tx_clr_all_bufs(priv->ring_data + i);
1427         }
1428 }
1429
1430 void hns_nic_net_reset(struct net_device *ndev)
1431 {
1432         struct hns_nic_priv *priv = netdev_priv(ndev);
1433         struct hnae_handle *handle = priv->ae_handle;
1434
1435         while (test_and_set_bit(NIC_STATE_RESETTING, &priv->state))
1436                 usleep_range(1000, 2000);
1437
1438         (void)hnae_reinit_handle(handle);
1439
1440         clear_bit(NIC_STATE_RESETTING, &priv->state);
1441 }
1442
1443 void hns_nic_net_reinit(struct net_device *netdev)
1444 {
1445         struct hns_nic_priv *priv = netdev_priv(netdev);
1446         enum hnae_port_type type = priv->ae_handle->port_type;
1447
1448         netif_trans_update(priv->netdev);
1449         while (test_and_set_bit(NIC_STATE_REINITING, &priv->state))
1450                 usleep_range(1000, 2000);
1451
1452         hns_nic_net_down(netdev);
1453
1454         /* Only do hns_nic_net_reset in debug mode
1455          * because of hardware limitation.
1456          */
1457         if (type == HNAE_PORT_DEBUG)
1458                 hns_nic_net_reset(netdev);
1459
1460         (void)hns_nic_net_up(netdev);
1461         clear_bit(NIC_STATE_REINITING, &priv->state);
1462 }
1463
1464 static int hns_nic_net_open(struct net_device *ndev)
1465 {
1466         struct hns_nic_priv *priv = netdev_priv(ndev);
1467         struct hnae_handle *h = priv->ae_handle;
1468         int ret;
1469
1470         if (test_bit(NIC_STATE_TESTING, &priv->state))
1471                 return -EBUSY;
1472
1473         priv->link = 0;
1474         netif_carrier_off(ndev);
1475
1476         ret = netif_set_real_num_tx_queues(ndev, h->q_num);
1477         if (ret < 0) {
1478                 netdev_err(ndev, "netif_set_real_num_tx_queues fail, ret=%d!\n",
1479                            ret);
1480                 return ret;
1481         }
1482
1483         ret = netif_set_real_num_rx_queues(ndev, h->q_num);
1484         if (ret < 0) {
1485                 netdev_err(ndev,
1486                            "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
1487                 return ret;
1488         }
1489
1490         ret = hns_nic_net_up(ndev);
1491         if (ret) {
1492                 netdev_err(ndev,
1493                            "hns net up fail, ret=%d!\n", ret);
1494                 return ret;
1495         }
1496
1497         return 0;
1498 }
1499
1500 static int hns_nic_net_stop(struct net_device *ndev)
1501 {
1502         hns_nic_net_down(ndev);
1503
1504         return 0;
1505 }
1506
1507 static void hns_tx_timeout_reset(struct hns_nic_priv *priv);
1508 #define HNS_TX_TIMEO_LIMIT (40 * HZ)
1509 static void hns_nic_net_timeout(struct net_device *ndev, unsigned int txqueue)
1510 {
1511         struct hns_nic_priv *priv = netdev_priv(ndev);
1512
1513         if (ndev->watchdog_timeo < HNS_TX_TIMEO_LIMIT) {
1514                 ndev->watchdog_timeo *= 2;
1515                 netdev_info(ndev, "watchdog_timo changed to %d.\n",
1516                             ndev->watchdog_timeo);
1517         } else {
1518                 ndev->watchdog_timeo = HNS_NIC_TX_TIMEOUT;
1519                 hns_tx_timeout_reset(priv);
1520         }
1521 }
1522
1523 static netdev_tx_t hns_nic_net_xmit(struct sk_buff *skb,
1524                                     struct net_device *ndev)
1525 {
1526         struct hns_nic_priv *priv = netdev_priv(ndev);
1527
1528         assert(skb->queue_mapping < priv->ae_handle->q_num);
1529
1530         return hns_nic_net_xmit_hw(ndev, skb,
1531                                    &tx_ring_data(priv, skb->queue_mapping));
1532 }
1533
1534 static void hns_nic_drop_rx_fetch(struct hns_nic_ring_data *ring_data,
1535                                   struct sk_buff *skb)
1536 {
1537         dev_kfree_skb_any(skb);
1538 }
1539
1540 #define HNS_LB_TX_RING  0
1541 static struct sk_buff *hns_assemble_skb(struct net_device *ndev)
1542 {
1543         struct sk_buff *skb;
1544         struct ethhdr *ethhdr;
1545         int frame_len;
1546
1547         /* allocate test skb */
1548         skb = alloc_skb(64, GFP_KERNEL);
1549         if (!skb)
1550                 return NULL;
1551
1552         skb_put(skb, 64);
1553         skb->dev = ndev;
1554         memset(skb->data, 0xFF, skb->len);
1555
1556         /* must be tcp/ip package */
1557         ethhdr = (struct ethhdr *)skb->data;
1558         ethhdr->h_proto = htons(ETH_P_IP);
1559
1560         frame_len = skb->len & (~1ul);
1561         memset(&skb->data[frame_len / 2], 0xAA,
1562                frame_len / 2 - 1);
1563
1564         skb->queue_mapping = HNS_LB_TX_RING;
1565
1566         return skb;
1567 }
1568
1569 static int hns_enable_serdes_lb(struct net_device *ndev)
1570 {
1571         struct hns_nic_priv *priv = netdev_priv(ndev);
1572         struct hnae_handle *h = priv->ae_handle;
1573         struct hnae_ae_ops *ops = h->dev->ops;
1574         int speed, duplex;
1575         int ret;
1576
1577         ret = ops->set_loopback(h, MAC_INTERNALLOOP_SERDES, 1);
1578         if (ret)
1579                 return ret;
1580
1581         ret = ops->start ? ops->start(h) : 0;
1582         if (ret)
1583                 return ret;
1584
1585         /* link adjust duplex*/
1586         if (h->phy_if != PHY_INTERFACE_MODE_XGMII)
1587                 speed = 1000;
1588         else
1589                 speed = 10000;
1590         duplex = 1;
1591
1592         ops->adjust_link(h, speed, duplex);
1593
1594         /* wait h/w ready */
1595         mdelay(300);
1596
1597         return 0;
1598 }
1599
1600 static void hns_disable_serdes_lb(struct net_device *ndev)
1601 {
1602         struct hns_nic_priv *priv = netdev_priv(ndev);
1603         struct hnae_handle *h = priv->ae_handle;
1604         struct hnae_ae_ops *ops = h->dev->ops;
1605
1606         ops->stop(h);
1607         ops->set_loopback(h, MAC_INTERNALLOOP_SERDES, 0);
1608 }
1609
1610 /**
1611  *hns_nic_clear_all_rx_fetch - clear the chip fetched descriptions. The
1612  *function as follows:
1613  *    1. if one rx ring has found the page_offset is not equal 0 between head
1614  *       and tail, it means that the chip fetched the wrong descs for the ring
1615  *       which buffer size is 4096.
1616  *    2. we set the chip serdes loopback and set rss indirection to the ring.
1617  *    3. construct 64-bytes ip broadcast packages, wait the associated rx ring
1618  *       receiving all packages and it will fetch new descriptions.
1619  *    4. recover to the original state.
1620  *
1621  *@ndev: net device
1622  */
1623 static int hns_nic_clear_all_rx_fetch(struct net_device *ndev)
1624 {
1625         struct hns_nic_priv *priv = netdev_priv(ndev);
1626         struct hnae_handle *h = priv->ae_handle;
1627         struct hnae_ae_ops *ops = h->dev->ops;
1628         struct hns_nic_ring_data *rd;
1629         struct hnae_ring *ring;
1630         struct sk_buff *skb;
1631         u32 *org_indir;
1632         u32 *cur_indir;
1633         int indir_size;
1634         int head, tail;
1635         int fetch_num;
1636         int i, j;
1637         bool found;
1638         int retry_times;
1639         int ret = 0;
1640
1641         /* alloc indir memory */
1642         indir_size = ops->get_rss_indir_size(h) * sizeof(*org_indir);
1643         org_indir = kzalloc(indir_size, GFP_KERNEL);
1644         if (!org_indir)
1645                 return -ENOMEM;
1646
1647         /* store the original indirection */
1648         ops->get_rss(h, org_indir, NULL, NULL);
1649
1650         cur_indir = kzalloc(indir_size, GFP_KERNEL);
1651         if (!cur_indir) {
1652                 ret = -ENOMEM;
1653                 goto cur_indir_alloc_err;
1654         }
1655
1656         /* set loopback */
1657         if (hns_enable_serdes_lb(ndev)) {
1658                 ret = -EINVAL;
1659                 goto enable_serdes_lb_err;
1660         }
1661
1662         /* foreach every rx ring to clear fetch desc */
1663         for (i = 0; i < h->q_num; i++) {
1664                 ring = &h->qs[i]->rx_ring;
1665                 head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1666                 tail = readl_relaxed(ring->io_base + RCB_REG_TAIL);
1667                 found = false;
1668                 fetch_num = ring_dist(ring, head, tail);
1669
1670                 while (head != tail) {
1671                         if (ring->desc_cb[head].page_offset != 0) {
1672                                 found = true;
1673                                 break;
1674                         }
1675
1676                         head++;
1677                         if (head == ring->desc_num)
1678                                 head = 0;
1679                 }
1680
1681                 if (found) {
1682                         for (j = 0; j < indir_size / sizeof(*org_indir); j++)
1683                                 cur_indir[j] = i;
1684                         ops->set_rss(h, cur_indir, NULL, 0);
1685
1686                         for (j = 0; j < fetch_num; j++) {
1687                                 /* alloc one skb and init */
1688                                 skb = hns_assemble_skb(ndev);
1689                                 if (!skb) {
1690                                         ret = -ENOMEM;
1691                                         goto out;
1692                                 }
1693                                 rd = &tx_ring_data(priv, skb->queue_mapping);
1694                                 hns_nic_net_xmit_hw(ndev, skb, rd);
1695
1696                                 retry_times = 0;
1697                                 while (retry_times++ < 10) {
1698                                         mdelay(10);
1699                                         /* clean rx */
1700                                         rd = &rx_ring_data(priv, i);
1701                                         if (rd->poll_one(rd, fetch_num,
1702                                                          hns_nic_drop_rx_fetch))
1703                                                 break;
1704                                 }
1705
1706                                 retry_times = 0;
1707                                 while (retry_times++ < 10) {
1708                                         mdelay(10);
1709                                         /* clean tx ring 0 send package */
1710                                         rd = &tx_ring_data(priv,
1711                                                            HNS_LB_TX_RING);
1712                                         if (rd->poll_one(rd, fetch_num, NULL))
1713                                                 break;
1714                                 }
1715                         }
1716                 }
1717         }
1718
1719 out:
1720         /* restore everything */
1721         ops->set_rss(h, org_indir, NULL, 0);
1722         hns_disable_serdes_lb(ndev);
1723 enable_serdes_lb_err:
1724         kfree(cur_indir);
1725 cur_indir_alloc_err:
1726         kfree(org_indir);
1727
1728         return ret;
1729 }
1730
1731 static int hns_nic_change_mtu(struct net_device *ndev, int new_mtu)
1732 {
1733         struct hns_nic_priv *priv = netdev_priv(ndev);
1734         struct hnae_handle *h = priv->ae_handle;
1735         bool if_running = netif_running(ndev);
1736         int ret;
1737
1738         /* MTU < 68 is an error and causes problems on some kernels */
1739         if (new_mtu < 68)
1740                 return -EINVAL;
1741
1742         /* MTU no change */
1743         if (new_mtu == ndev->mtu)
1744                 return 0;
1745
1746         if (!h->dev->ops->set_mtu)
1747                 return -ENOTSUPP;
1748
1749         if (if_running) {
1750                 (void)hns_nic_net_stop(ndev);
1751                 msleep(100);
1752         }
1753
1754         if (priv->enet_ver != AE_VERSION_1 &&
1755             ndev->mtu <= BD_SIZE_2048_MAX_MTU &&
1756             new_mtu > BD_SIZE_2048_MAX_MTU) {
1757                 /* update desc */
1758                 hnae_reinit_all_ring_desc(h);
1759
1760                 /* clear the package which the chip has fetched */
1761                 ret = hns_nic_clear_all_rx_fetch(ndev);
1762
1763                 /* the page offset must be consist with desc */
1764                 hnae_reinit_all_ring_page_off(h);
1765
1766                 if (ret) {
1767                         netdev_err(ndev, "clear the fetched desc fail\n");
1768                         goto out;
1769                 }
1770         }
1771
1772         ret = h->dev->ops->set_mtu(h, new_mtu);
1773         if (ret) {
1774                 netdev_err(ndev, "set mtu fail, return value %d\n",
1775                            ret);
1776                 goto out;
1777         }
1778
1779         /* finally, set new mtu to netdevice */
1780         ndev->mtu = new_mtu;
1781
1782 out:
1783         if (if_running) {
1784                 if (hns_nic_net_open(ndev)) {
1785                         netdev_err(ndev, "hns net open fail\n");
1786                         ret = -EINVAL;
1787                 }
1788         }
1789
1790         return ret;
1791 }
1792
1793 static int hns_nic_set_features(struct net_device *netdev,
1794                                 netdev_features_t features)
1795 {
1796         struct hns_nic_priv *priv = netdev_priv(netdev);
1797
1798         switch (priv->enet_ver) {
1799         case AE_VERSION_1:
1800                 if (features & (NETIF_F_TSO | NETIF_F_TSO6))
1801                         netdev_info(netdev, "enet v1 do not support tso!\n");
1802                 break;
1803         default:
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  * hns_set_multicast_list - set mutl mac address
1851  * @ndev: net device
1852  *
1853  * return void
1854  */
1855 static void hns_set_multicast_list(struct net_device *ndev)
1856 {
1857         struct hns_nic_priv *priv = netdev_priv(ndev);
1858         struct hnae_handle *h = priv->ae_handle;
1859         struct netdev_hw_addr *ha = NULL;
1860
1861         if (!h) {
1862                 netdev_err(ndev, "hnae handle is null\n");
1863                 return;
1864         }
1865
1866         if (h->dev->ops->clr_mc_addr)
1867                 if (h->dev->ops->clr_mc_addr(h))
1868                         netdev_err(ndev, "clear multicast address fail\n");
1869
1870         if (h->dev->ops->set_mc_addr) {
1871                 netdev_for_each_mc_addr(ha, ndev)
1872                         if (h->dev->ops->set_mc_addr(h, ha->addr))
1873                                 netdev_err(ndev, "set multicast fail\n");
1874         }
1875 }
1876
1877 static void hns_nic_set_rx_mode(struct net_device *ndev)
1878 {
1879         struct hns_nic_priv *priv = netdev_priv(ndev);
1880         struct hnae_handle *h = priv->ae_handle;
1881
1882         if (h->dev->ops->set_promisc_mode) {
1883                 if (ndev->flags & IFF_PROMISC)
1884                         h->dev->ops->set_promisc_mode(h, 1);
1885                 else
1886                         h->dev->ops->set_promisc_mode(h, 0);
1887         }
1888
1889         hns_set_multicast_list(ndev);
1890
1891         if (__dev_uc_sync(ndev, hns_nic_uc_sync, hns_nic_uc_unsync))
1892                 netdev_err(ndev, "sync uc address fail\n");
1893 }
1894
1895 static void hns_nic_get_stats64(struct net_device *ndev,
1896                                 struct rtnl_link_stats64 *stats)
1897 {
1898         int idx;
1899         u64 tx_bytes = 0;
1900         u64 rx_bytes = 0;
1901         u64 tx_pkts = 0;
1902         u64 rx_pkts = 0;
1903         struct hns_nic_priv *priv = netdev_priv(ndev);
1904         struct hnae_handle *h = priv->ae_handle;
1905
1906         for (idx = 0; idx < h->q_num; idx++) {
1907                 tx_bytes += h->qs[idx]->tx_ring.stats.tx_bytes;
1908                 tx_pkts += h->qs[idx]->tx_ring.stats.tx_pkts;
1909                 rx_bytes += h->qs[idx]->rx_ring.stats.rx_bytes;
1910                 rx_pkts += h->qs[idx]->rx_ring.stats.rx_pkts;
1911         }
1912
1913         stats->tx_bytes = tx_bytes;
1914         stats->tx_packets = tx_pkts;
1915         stats->rx_bytes = rx_bytes;
1916         stats->rx_packets = rx_pkts;
1917
1918         stats->rx_errors = ndev->stats.rx_errors;
1919         stats->multicast = ndev->stats.multicast;
1920         stats->rx_length_errors = ndev->stats.rx_length_errors;
1921         stats->rx_crc_errors = ndev->stats.rx_crc_errors;
1922         stats->rx_missed_errors = ndev->stats.rx_missed_errors;
1923
1924         stats->tx_errors = ndev->stats.tx_errors;
1925         stats->rx_dropped = ndev->stats.rx_dropped;
1926         stats->tx_dropped = ndev->stats.tx_dropped;
1927         stats->collisions = ndev->stats.collisions;
1928         stats->rx_over_errors = ndev->stats.rx_over_errors;
1929         stats->rx_frame_errors = ndev->stats.rx_frame_errors;
1930         stats->rx_fifo_errors = ndev->stats.rx_fifo_errors;
1931         stats->tx_aborted_errors = ndev->stats.tx_aborted_errors;
1932         stats->tx_carrier_errors = ndev->stats.tx_carrier_errors;
1933         stats->tx_fifo_errors = ndev->stats.tx_fifo_errors;
1934         stats->tx_heartbeat_errors = ndev->stats.tx_heartbeat_errors;
1935         stats->tx_window_errors = ndev->stats.tx_window_errors;
1936         stats->rx_compressed = ndev->stats.rx_compressed;
1937         stats->tx_compressed = ndev->stats.tx_compressed;
1938 }
1939
1940 static u16
1941 hns_nic_select_queue(struct net_device *ndev, struct sk_buff *skb,
1942                      struct net_device *sb_dev)
1943 {
1944         struct ethhdr *eth_hdr = (struct ethhdr *)skb->data;
1945         struct hns_nic_priv *priv = netdev_priv(ndev);
1946
1947         /* fix hardware broadcast/multicast packets queue loopback */
1948         if (!AE_IS_VER1(priv->enet_ver) &&
1949             is_multicast_ether_addr(eth_hdr->h_dest))
1950                 return 0;
1951         else
1952                 return netdev_pick_tx(ndev, skb, NULL);
1953 }
1954
1955 static const struct net_device_ops hns_nic_netdev_ops = {
1956         .ndo_open = hns_nic_net_open,
1957         .ndo_stop = hns_nic_net_stop,
1958         .ndo_start_xmit = hns_nic_net_xmit,
1959         .ndo_tx_timeout = hns_nic_net_timeout,
1960         .ndo_set_mac_address = hns_nic_net_set_mac_address,
1961         .ndo_change_mtu = hns_nic_change_mtu,
1962         .ndo_eth_ioctl = phy_do_ioctl_running,
1963         .ndo_set_features = hns_nic_set_features,
1964         .ndo_fix_features = hns_nic_fix_features,
1965         .ndo_get_stats64 = hns_nic_get_stats64,
1966         .ndo_set_rx_mode = hns_nic_set_rx_mode,
1967         .ndo_select_queue = hns_nic_select_queue,
1968 };
1969
1970 static void hns_nic_update_link_status(struct net_device *netdev)
1971 {
1972         struct hns_nic_priv *priv = netdev_priv(netdev);
1973
1974         struct hnae_handle *h = priv->ae_handle;
1975
1976         if (h->phy_dev) {
1977                 if (h->phy_if != PHY_INTERFACE_MODE_XGMII)
1978                         return;
1979
1980                 (void)genphy_read_status(h->phy_dev);
1981         }
1982         hns_nic_adjust_link(netdev);
1983 }
1984
1985 /* for dumping key regs*/
1986 static void hns_nic_dump(struct hns_nic_priv *priv)
1987 {
1988         struct hnae_handle *h = priv->ae_handle;
1989         struct hnae_ae_ops *ops = h->dev->ops;
1990         u32 *data, reg_num, i;
1991
1992         if (ops->get_regs_len && ops->get_regs) {
1993                 reg_num = ops->get_regs_len(priv->ae_handle);
1994                 reg_num = (reg_num + 3ul) & ~3ul;
1995                 data = kcalloc(reg_num, sizeof(u32), GFP_KERNEL);
1996                 if (data) {
1997                         ops->get_regs(priv->ae_handle, data);
1998                         for (i = 0; i < reg_num; i += 4)
1999                                 pr_info("0x%08x: 0x%08x 0x%08x 0x%08x 0x%08x\n",
2000                                         i, data[i], data[i + 1],
2001                                         data[i + 2], data[i + 3]);
2002                         kfree(data);
2003                 }
2004         }
2005
2006         for (i = 0; i < h->q_num; i++) {
2007                 pr_info("tx_queue%d_next_to_clean:%d\n",
2008                         i, h->qs[i]->tx_ring.next_to_clean);
2009                 pr_info("tx_queue%d_next_to_use:%d\n",
2010                         i, h->qs[i]->tx_ring.next_to_use);
2011                 pr_info("rx_queue%d_next_to_clean:%d\n",
2012                         i, h->qs[i]->rx_ring.next_to_clean);
2013                 pr_info("rx_queue%d_next_to_use:%d\n",
2014                         i, h->qs[i]->rx_ring.next_to_use);
2015         }
2016 }
2017
2018 /* for resetting subtask */
2019 static void hns_nic_reset_subtask(struct hns_nic_priv *priv)
2020 {
2021         enum hnae_port_type type = priv->ae_handle->port_type;
2022
2023         if (!test_bit(NIC_STATE2_RESET_REQUESTED, &priv->state))
2024                 return;
2025         clear_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
2026
2027         /* If we're already down, removing or resetting, just bail */
2028         if (test_bit(NIC_STATE_DOWN, &priv->state) ||
2029             test_bit(NIC_STATE_REMOVING, &priv->state) ||
2030             test_bit(NIC_STATE_RESETTING, &priv->state))
2031                 return;
2032
2033         hns_nic_dump(priv);
2034         netdev_info(priv->netdev, "try to reset %s port!\n",
2035                     (type == HNAE_PORT_DEBUG ? "debug" : "service"));
2036
2037         rtnl_lock();
2038         /* put off any impending NetWatchDogTimeout */
2039         netif_trans_update(priv->netdev);
2040         hns_nic_net_reinit(priv->netdev);
2041
2042         rtnl_unlock();
2043 }
2044
2045 /* for doing service complete*/
2046 static void hns_nic_service_event_complete(struct hns_nic_priv *priv)
2047 {
2048         WARN_ON(!test_bit(NIC_STATE_SERVICE_SCHED, &priv->state));
2049         /* make sure to commit the things */
2050         smp_mb__before_atomic();
2051         clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
2052 }
2053
2054 static void hns_nic_service_task(struct work_struct *work)
2055 {
2056         struct hns_nic_priv *priv
2057                 = container_of(work, struct hns_nic_priv, service_task);
2058         struct hnae_handle *h = priv->ae_handle;
2059
2060         hns_nic_reset_subtask(priv);
2061         hns_nic_update_link_status(priv->netdev);
2062         h->dev->ops->update_led_status(h);
2063         hns_nic_update_stats(priv->netdev);
2064
2065         hns_nic_service_event_complete(priv);
2066 }
2067
2068 static void hns_nic_task_schedule(struct hns_nic_priv *priv)
2069 {
2070         if (!test_bit(NIC_STATE_DOWN, &priv->state) &&
2071             !test_bit(NIC_STATE_REMOVING, &priv->state) &&
2072             !test_and_set_bit(NIC_STATE_SERVICE_SCHED, &priv->state))
2073                 (void)schedule_work(&priv->service_task);
2074 }
2075
2076 static void hns_nic_service_timer(struct timer_list *t)
2077 {
2078         struct hns_nic_priv *priv = from_timer(priv, t, service_timer);
2079
2080         (void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
2081
2082         hns_nic_task_schedule(priv);
2083 }
2084
2085 /**
2086  * hns_tx_timeout_reset - initiate reset due to Tx timeout
2087  * @priv: driver private struct
2088  **/
2089 static void hns_tx_timeout_reset(struct hns_nic_priv *priv)
2090 {
2091         /* Do the reset outside of interrupt context */
2092         if (!test_bit(NIC_STATE_DOWN, &priv->state)) {
2093                 set_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
2094                 netdev_warn(priv->netdev,
2095                             "initiating reset due to tx timeout(%llu,0x%lx)\n",
2096                             priv->tx_timeout_count, priv->state);
2097                 priv->tx_timeout_count++;
2098                 hns_nic_task_schedule(priv);
2099         }
2100 }
2101
2102 static int hns_nic_init_ring_data(struct hns_nic_priv *priv)
2103 {
2104         struct hnae_handle *h = priv->ae_handle;
2105         struct hns_nic_ring_data *rd;
2106         bool is_ver1 = AE_IS_VER1(priv->enet_ver);
2107         int i;
2108
2109         if (h->q_num > NIC_MAX_Q_PER_VF) {
2110                 netdev_err(priv->netdev, "too much queue (%d)\n", h->q_num);
2111                 return -EINVAL;
2112         }
2113
2114         priv->ring_data = kzalloc(array3_size(h->q_num,
2115                                               sizeof(*priv->ring_data), 2),
2116                                   GFP_KERNEL);
2117         if (!priv->ring_data)
2118                 return -ENOMEM;
2119
2120         for (i = 0; i < h->q_num; i++) {
2121                 rd = &priv->ring_data[i];
2122                 rd->queue_index = i;
2123                 rd->ring = &h->qs[i]->tx_ring;
2124                 rd->poll_one = hns_nic_tx_poll_one;
2125                 rd->fini_process = is_ver1 ? hns_nic_tx_fini_pro :
2126                         hns_nic_tx_fini_pro_v2;
2127
2128                 netif_napi_add(priv->netdev, &rd->napi, hns_nic_common_poll);
2129                 rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
2130         }
2131         for (i = h->q_num; i < h->q_num * 2; i++) {
2132                 rd = &priv->ring_data[i];
2133                 rd->queue_index = i - h->q_num;
2134                 rd->ring = &h->qs[i - h->q_num]->rx_ring;
2135                 rd->poll_one = hns_nic_rx_poll_one;
2136                 rd->ex_process = hns_nic_rx_up_pro;
2137                 rd->fini_process = is_ver1 ? hns_nic_rx_fini_pro :
2138                         hns_nic_rx_fini_pro_v2;
2139
2140                 netif_napi_add(priv->netdev, &rd->napi, hns_nic_common_poll);
2141                 rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
2142         }
2143
2144         return 0;
2145 }
2146
2147 static void hns_nic_uninit_ring_data(struct hns_nic_priv *priv)
2148 {
2149         struct hnae_handle *h = priv->ae_handle;
2150         int i;
2151
2152         for (i = 0; i < h->q_num * 2; i++) {
2153                 netif_napi_del(&priv->ring_data[i].napi);
2154                 if (priv->ring_data[i].ring->irq_init_flag == RCB_IRQ_INITED) {
2155                         (void)irq_set_affinity_hint(
2156                                 priv->ring_data[i].ring->irq,
2157                                 NULL);
2158                         free_irq(priv->ring_data[i].ring->irq,
2159                                  &priv->ring_data[i]);
2160                 }
2161
2162                 priv->ring_data[i].ring->irq_init_flag = RCB_IRQ_NOT_INITED;
2163         }
2164         kfree(priv->ring_data);
2165 }
2166
2167 static void hns_nic_set_priv_ops(struct net_device *netdev)
2168 {
2169         struct hns_nic_priv *priv = netdev_priv(netdev);
2170         struct hnae_handle *h = priv->ae_handle;
2171
2172         if (AE_IS_VER1(priv->enet_ver)) {
2173                 priv->ops.fill_desc = fill_desc;
2174                 priv->ops.get_rxd_bnum = get_rx_desc_bnum;
2175                 priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
2176         } else {
2177                 priv->ops.get_rxd_bnum = get_v2rx_desc_bnum;
2178                 priv->ops.fill_desc = fill_desc_v2;
2179                 priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx_v2;
2180                 netif_set_tso_max_size(netdev, 7 * 4096);
2181                 /* enable tso when init
2182                  * control tso on/off through TSE bit in bd
2183                  */
2184                 h->dev->ops->set_tso_stats(h, 1);
2185         }
2186 }
2187
2188 static int hns_nic_try_get_ae(struct net_device *ndev)
2189 {
2190         struct hns_nic_priv *priv = netdev_priv(ndev);
2191         struct hnae_handle *h;
2192         int ret;
2193
2194         h = hnae_get_handle(&priv->netdev->dev,
2195                             priv->fwnode, priv->port_id, NULL);
2196         if (IS_ERR_OR_NULL(h)) {
2197                 ret = -ENODEV;
2198                 dev_dbg(priv->dev, "has not handle, register notifier!\n");
2199                 goto out;
2200         }
2201         priv->ae_handle = h;
2202
2203         ret = hns_nic_init_phy(ndev, h);
2204         if (ret) {
2205                 dev_err(priv->dev, "probe phy device fail!\n");
2206                 goto out_init_phy;
2207         }
2208
2209         ret = hns_nic_init_ring_data(priv);
2210         if (ret) {
2211                 ret = -ENOMEM;
2212                 goto out_init_ring_data;
2213         }
2214
2215         hns_nic_set_priv_ops(ndev);
2216
2217         ret = register_netdev(ndev);
2218         if (ret) {
2219                 dev_err(priv->dev, "probe register netdev fail!\n");
2220                 goto out_reg_ndev_fail;
2221         }
2222         return 0;
2223
2224 out_reg_ndev_fail:
2225         hns_nic_uninit_ring_data(priv);
2226         priv->ring_data = NULL;
2227 out_init_phy:
2228 out_init_ring_data:
2229         hnae_put_handle(priv->ae_handle);
2230         priv->ae_handle = NULL;
2231 out:
2232         return ret;
2233 }
2234
2235 static int hns_nic_notifier_action(struct notifier_block *nb,
2236                                    unsigned long action, void *data)
2237 {
2238         struct hns_nic_priv *priv =
2239                 container_of(nb, struct hns_nic_priv, notifier_block);
2240
2241         assert(action == HNAE_AE_REGISTER);
2242
2243         if (!hns_nic_try_get_ae(priv->netdev)) {
2244                 hnae_unregister_notifier(&priv->notifier_block);
2245                 priv->notifier_block.notifier_call = NULL;
2246         }
2247         return 0;
2248 }
2249
2250 static int hns_nic_dev_probe(struct platform_device *pdev)
2251 {
2252         struct device *dev = &pdev->dev;
2253         struct net_device *ndev;
2254         struct hns_nic_priv *priv;
2255         u32 port_id;
2256         int ret;
2257
2258         ndev = alloc_etherdev_mq(sizeof(struct hns_nic_priv), NIC_MAX_Q_PER_VF);
2259         if (!ndev)
2260                 return -ENOMEM;
2261
2262         platform_set_drvdata(pdev, ndev);
2263
2264         priv = netdev_priv(ndev);
2265         priv->dev = dev;
2266         priv->netdev = ndev;
2267
2268         if (dev_of_node(dev)) {
2269                 struct device_node *ae_node;
2270
2271                 if (of_device_is_compatible(dev->of_node,
2272                                             "hisilicon,hns-nic-v1"))
2273                         priv->enet_ver = AE_VERSION_1;
2274                 else
2275                         priv->enet_ver = AE_VERSION_2;
2276
2277                 ae_node = of_parse_phandle(dev->of_node, "ae-handle", 0);
2278                 if (!ae_node) {
2279                         ret = -ENODEV;
2280                         dev_err(dev, "not find ae-handle\n");
2281                         goto out_read_prop_fail;
2282                 }
2283                 priv->fwnode = &ae_node->fwnode;
2284         } else if (is_acpi_node(dev->fwnode)) {
2285                 struct fwnode_reference_args args;
2286
2287                 if (acpi_dev_found(hns_enet_acpi_match[0].id))
2288                         priv->enet_ver = AE_VERSION_1;
2289                 else if (acpi_dev_found(hns_enet_acpi_match[1].id))
2290                         priv->enet_ver = AE_VERSION_2;
2291                 else {
2292                         ret = -ENXIO;
2293                         goto out_read_prop_fail;
2294                 }
2295
2296                 /* try to find port-idx-in-ae first */
2297                 ret = acpi_node_get_property_reference(dev->fwnode,
2298                                                        "ae-handle", 0, &args);
2299                 if (ret) {
2300                         dev_err(dev, "not find ae-handle\n");
2301                         goto out_read_prop_fail;
2302                 }
2303                 if (!is_acpi_device_node(args.fwnode)) {
2304                         ret = -EINVAL;
2305                         goto out_read_prop_fail;
2306                 }
2307                 priv->fwnode = args.fwnode;
2308         } else {
2309                 dev_err(dev, "cannot read cfg data from OF or acpi\n");
2310                 ret = -ENXIO;
2311                 goto out_read_prop_fail;
2312         }
2313
2314         ret = device_property_read_u32(dev, "port-idx-in-ae", &port_id);
2315         if (ret) {
2316                 /* only for old code compatible */
2317                 ret = device_property_read_u32(dev, "port-id", &port_id);
2318                 if (ret)
2319                         goto out_read_prop_fail;
2320                 /* for old dts, we need to caculate the port offset */
2321                 port_id = port_id < HNS_SRV_OFFSET ? port_id + HNS_DEBUG_OFFSET
2322                         : port_id - HNS_SRV_OFFSET;
2323         }
2324         priv->port_id = port_id;
2325
2326         hns_init_mac_addr(ndev);
2327
2328         ndev->watchdog_timeo = HNS_NIC_TX_TIMEOUT;
2329         ndev->priv_flags |= IFF_UNICAST_FLT;
2330         ndev->netdev_ops = &hns_nic_netdev_ops;
2331         hns_ethtool_set_ops(ndev);
2332
2333         ndev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2334                 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2335                 NETIF_F_GRO;
2336         ndev->vlan_features |=
2337                 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM;
2338         ndev->vlan_features |= NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
2339
2340         /* MTU range: 68 - 9578 (v1) or 9706 (v2) */
2341         ndev->min_mtu = MAC_MIN_MTU;
2342         switch (priv->enet_ver) {
2343         case AE_VERSION_2:
2344                 ndev->features |= NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_NTUPLE;
2345                 ndev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2346                         NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2347                         NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6;
2348                 ndev->vlan_features |= NETIF_F_TSO | NETIF_F_TSO6;
2349                 ndev->max_mtu = MAC_MAX_MTU_V2 -
2350                                 (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
2351                 break;
2352         default:
2353                 ndev->max_mtu = MAC_MAX_MTU -
2354                                 (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
2355                 break;
2356         }
2357
2358         SET_NETDEV_DEV(ndev, dev);
2359
2360         if (!dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64)))
2361                 dev_dbg(dev, "set mask to 64bit\n");
2362         else
2363                 dev_err(dev, "set mask to 64bit fail!\n");
2364
2365         /* carrier off reporting is important to ethtool even BEFORE open */
2366         netif_carrier_off(ndev);
2367
2368         timer_setup(&priv->service_timer, hns_nic_service_timer, 0);
2369         INIT_WORK(&priv->service_task, hns_nic_service_task);
2370
2371         set_bit(NIC_STATE_SERVICE_INITED, &priv->state);
2372         clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
2373         set_bit(NIC_STATE_DOWN, &priv->state);
2374
2375         if (hns_nic_try_get_ae(priv->netdev)) {
2376                 priv->notifier_block.notifier_call = hns_nic_notifier_action;
2377                 ret = hnae_register_notifier(&priv->notifier_block);
2378                 if (ret) {
2379                         dev_err(dev, "register notifier fail!\n");
2380                         goto out_notify_fail;
2381                 }
2382                 dev_dbg(dev, "has not handle, register notifier!\n");
2383         }
2384
2385         return 0;
2386
2387 out_notify_fail:
2388         (void)cancel_work_sync(&priv->service_task);
2389 out_read_prop_fail:
2390         /* safe for ACPI FW */
2391         of_node_put(to_of_node(priv->fwnode));
2392         free_netdev(ndev);
2393         return ret;
2394 }
2395
2396 static void hns_nic_dev_remove(struct platform_device *pdev)
2397 {
2398         struct net_device *ndev = platform_get_drvdata(pdev);
2399         struct hns_nic_priv *priv = netdev_priv(ndev);
2400
2401         if (ndev->reg_state != NETREG_UNINITIALIZED)
2402                 unregister_netdev(ndev);
2403
2404         if (priv->ring_data)
2405                 hns_nic_uninit_ring_data(priv);
2406         priv->ring_data = NULL;
2407
2408         if (ndev->phydev)
2409                 phy_disconnect(ndev->phydev);
2410
2411         if (!IS_ERR_OR_NULL(priv->ae_handle))
2412                 hnae_put_handle(priv->ae_handle);
2413         priv->ae_handle = NULL;
2414         if (priv->notifier_block.notifier_call)
2415                 hnae_unregister_notifier(&priv->notifier_block);
2416         priv->notifier_block.notifier_call = NULL;
2417
2418         set_bit(NIC_STATE_REMOVING, &priv->state);
2419         (void)cancel_work_sync(&priv->service_task);
2420
2421         /* safe for ACPI FW */
2422         of_node_put(to_of_node(priv->fwnode));
2423
2424         free_netdev(ndev);
2425 }
2426
2427 static const struct of_device_id hns_enet_of_match[] = {
2428         {.compatible = "hisilicon,hns-nic-v1",},
2429         {.compatible = "hisilicon,hns-nic-v2",},
2430         {},
2431 };
2432
2433 MODULE_DEVICE_TABLE(of, hns_enet_of_match);
2434
2435 static struct platform_driver hns_nic_dev_driver = {
2436         .driver = {
2437                 .name = "hns-nic",
2438                 .of_match_table = hns_enet_of_match,
2439                 .acpi_match_table = ACPI_PTR(hns_enet_acpi_match),
2440         },
2441         .probe = hns_nic_dev_probe,
2442         .remove_new = hns_nic_dev_remove,
2443 };
2444
2445 module_platform_driver(hns_nic_dev_driver);
2446
2447 MODULE_DESCRIPTION("HISILICON HNS Ethernet driver");
2448 MODULE_AUTHOR("Hisilicon, Inc.");
2449 MODULE_LICENSE("GPL");
2450 MODULE_ALIAS("platform:hns-nic");