Merge tag 'kvm-ppc-fixes-4.19-2' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6-microblaze.git] / drivers / net / ethernet / hisilicon / hns / hns_enet.c
1 /*
2  * Copyright (c) 2014-2015 Hisilicon Limited.
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License as published by
6  * the Free Software Foundation; either version 2 of the License, or
7  * (at your option) any later version.
8  */
9
10 #include <linux/clk.h>
11 #include <linux/cpumask.h>
12 #include <linux/etherdevice.h>
13 #include <linux/if_vlan.h>
14 #include <linux/interrupt.h>
15 #include <linux/io.h>
16 #include <linux/ip.h>
17 #include <linux/ipv6.h>
18 #include <linux/module.h>
19 #include <linux/phy.h>
20 #include <linux/platform_device.h>
21 #include <linux/skbuff.h>
22
23 #include "hnae.h"
24 #include "hns_enet.h"
25 #include "hns_dsaf_mac.h"
26
27 #define NIC_MAX_Q_PER_VF 16
28 #define HNS_NIC_TX_TIMEOUT (5 * HZ)
29
30 #define SERVICE_TIMER_HZ (1 * HZ)
31
32 #define NIC_TX_CLEAN_MAX_NUM 256
33 #define NIC_RX_CLEAN_MAX_NUM 64
34
35 #define RCB_IRQ_NOT_INITED 0
36 #define RCB_IRQ_INITED 1
37 #define HNS_BUFFER_SIZE_2048 2048
38
39 #define BD_MAX_SEND_SIZE 8191
40 #define SKB_TMP_LEN(SKB) \
41         (((SKB)->transport_header - (SKB)->mac_header) + tcp_hdrlen(SKB))
42
43 static void fill_v2_desc(struct hnae_ring *ring, void *priv,
44                          int size, dma_addr_t dma, int frag_end,
45                          int buf_num, enum hns_desc_type type, int mtu)
46 {
47         struct hnae_desc *desc = &ring->desc[ring->next_to_use];
48         struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
49         struct iphdr *iphdr;
50         struct ipv6hdr *ipv6hdr;
51         struct sk_buff *skb;
52         __be16 protocol;
53         u8 bn_pid = 0;
54         u8 rrcfv = 0;
55         u8 ip_offset = 0;
56         u8 tvsvsn = 0;
57         u16 mss = 0;
58         u8 l4_len = 0;
59         u16 paylen = 0;
60
61         desc_cb->priv = priv;
62         desc_cb->length = size;
63         desc_cb->dma = dma;
64         desc_cb->type = type;
65
66         desc->addr = cpu_to_le64(dma);
67         desc->tx.send_size = cpu_to_le16((u16)size);
68
69         /* config bd buffer end */
70         hnae_set_bit(rrcfv, HNSV2_TXD_VLD_B, 1);
71         hnae_set_field(bn_pid, HNSV2_TXD_BUFNUM_M, 0, buf_num - 1);
72
73         /* fill port_id in the tx bd for sending management pkts */
74         hnae_set_field(bn_pid, HNSV2_TXD_PORTID_M,
75                        HNSV2_TXD_PORTID_S, ring->q->handle->dport_id);
76
77         if (type == DESC_TYPE_SKB) {
78                 skb = (struct sk_buff *)priv;
79
80                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
81                         skb_reset_mac_len(skb);
82                         protocol = skb->protocol;
83                         ip_offset = ETH_HLEN;
84
85                         if (protocol == htons(ETH_P_8021Q)) {
86                                 ip_offset += VLAN_HLEN;
87                                 protocol = vlan_get_protocol(skb);
88                                 skb->protocol = protocol;
89                         }
90
91                         if (skb->protocol == htons(ETH_P_IP)) {
92                                 iphdr = ip_hdr(skb);
93                                 hnae_set_bit(rrcfv, HNSV2_TXD_L3CS_B, 1);
94                                 hnae_set_bit(rrcfv, HNSV2_TXD_L4CS_B, 1);
95
96                                 /* check for tcp/udp header */
97                                 if (iphdr->protocol == IPPROTO_TCP &&
98                                     skb_is_gso(skb)) {
99                                         hnae_set_bit(tvsvsn,
100                                                      HNSV2_TXD_TSE_B, 1);
101                                         l4_len = tcp_hdrlen(skb);
102                                         mss = skb_shinfo(skb)->gso_size;
103                                         paylen = skb->len - SKB_TMP_LEN(skb);
104                                 }
105                         } else if (skb->protocol == htons(ETH_P_IPV6)) {
106                                 hnae_set_bit(tvsvsn, HNSV2_TXD_IPV6_B, 1);
107                                 ipv6hdr = ipv6_hdr(skb);
108                                 hnae_set_bit(rrcfv, HNSV2_TXD_L4CS_B, 1);
109
110                                 /* check for tcp/udp header */
111                                 if (ipv6hdr->nexthdr == IPPROTO_TCP &&
112                                     skb_is_gso(skb) && skb_is_gso_v6(skb)) {
113                                         hnae_set_bit(tvsvsn,
114                                                      HNSV2_TXD_TSE_B, 1);
115                                         l4_len = tcp_hdrlen(skb);
116                                         mss = skb_shinfo(skb)->gso_size;
117                                         paylen = skb->len - SKB_TMP_LEN(skb);
118                                 }
119                         }
120                         desc->tx.ip_offset = ip_offset;
121                         desc->tx.tse_vlan_snap_v6_sctp_nth = tvsvsn;
122                         desc->tx.mss = cpu_to_le16(mss);
123                         desc->tx.l4_len = l4_len;
124                         desc->tx.paylen = cpu_to_le16(paylen);
125                 }
126         }
127
128         hnae_set_bit(rrcfv, HNSV2_TXD_FE_B, frag_end);
129
130         desc->tx.bn_pid = bn_pid;
131         desc->tx.ra_ri_cs_fe_vld = rrcfv;
132
133         ring_ptr_move_fw(ring, next_to_use);
134 }
135
136 static const struct acpi_device_id hns_enet_acpi_match[] = {
137         { "HISI00C1", 0 },
138         { "HISI00C2", 0 },
139         { },
140 };
141 MODULE_DEVICE_TABLE(acpi, hns_enet_acpi_match);
142
143 static void fill_desc(struct hnae_ring *ring, void *priv,
144                       int size, dma_addr_t dma, int frag_end,
145                       int buf_num, enum hns_desc_type type, int mtu)
146 {
147         struct hnae_desc *desc = &ring->desc[ring->next_to_use];
148         struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
149         struct sk_buff *skb;
150         __be16 protocol;
151         u32 ip_offset;
152         u32 asid_bufnum_pid = 0;
153         u32 flag_ipoffset = 0;
154
155         desc_cb->priv = priv;
156         desc_cb->length = size;
157         desc_cb->dma = dma;
158         desc_cb->type = type;
159
160         desc->addr = cpu_to_le64(dma);
161         desc->tx.send_size = cpu_to_le16((u16)size);
162
163         /*config bd buffer end */
164         flag_ipoffset |= 1 << HNS_TXD_VLD_B;
165
166         asid_bufnum_pid |= buf_num << HNS_TXD_BUFNUM_S;
167
168         if (type == DESC_TYPE_SKB) {
169                 skb = (struct sk_buff *)priv;
170
171                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
172                         protocol = skb->protocol;
173                         ip_offset = ETH_HLEN;
174
175                         /*if it is a SW VLAN check the next protocol*/
176                         if (protocol == htons(ETH_P_8021Q)) {
177                                 ip_offset += VLAN_HLEN;
178                                 protocol = vlan_get_protocol(skb);
179                                 skb->protocol = protocol;
180                         }
181
182                         if (skb->protocol == htons(ETH_P_IP)) {
183                                 flag_ipoffset |= 1 << HNS_TXD_L3CS_B;
184                                 /* check for tcp/udp header */
185                                 flag_ipoffset |= 1 << HNS_TXD_L4CS_B;
186
187                         } else if (skb->protocol == htons(ETH_P_IPV6)) {
188                                 /* ipv6 has not l3 cs, check for L4 header */
189                                 flag_ipoffset |= 1 << HNS_TXD_L4CS_B;
190                         }
191
192                         flag_ipoffset |= ip_offset << HNS_TXD_IPOFFSET_S;
193                 }
194         }
195
196         flag_ipoffset |= frag_end << HNS_TXD_FE_B;
197
198         desc->tx.asid_bufnum_pid = cpu_to_le16(asid_bufnum_pid);
199         desc->tx.flag_ipoffset = cpu_to_le32(flag_ipoffset);
200
201         ring_ptr_move_fw(ring, next_to_use);
202 }
203
204 static void unfill_desc(struct hnae_ring *ring)
205 {
206         ring_ptr_move_bw(ring, next_to_use);
207 }
208
209 static int hns_nic_maybe_stop_tx(
210         struct sk_buff **out_skb, int *bnum, struct hnae_ring *ring)
211 {
212         struct sk_buff *skb = *out_skb;
213         struct sk_buff *new_skb = NULL;
214         int buf_num;
215
216         /* no. of segments (plus a header) */
217         buf_num = skb_shinfo(skb)->nr_frags + 1;
218
219         if (unlikely(buf_num > ring->max_desc_num_per_pkt)) {
220                 if (ring_space(ring) < 1)
221                         return -EBUSY;
222
223                 new_skb = skb_copy(skb, GFP_ATOMIC);
224                 if (!new_skb)
225                         return -ENOMEM;
226
227                 dev_kfree_skb_any(skb);
228                 *out_skb = new_skb;
229                 buf_num = 1;
230         } else if (buf_num > ring_space(ring)) {
231                 return -EBUSY;
232         }
233
234         *bnum = buf_num;
235         return 0;
236 }
237
238 static int hns_nic_maybe_stop_tso(
239         struct sk_buff **out_skb, int *bnum, struct hnae_ring *ring)
240 {
241         int i;
242         int size;
243         int buf_num;
244         int frag_num;
245         struct sk_buff *skb = *out_skb;
246         struct sk_buff *new_skb = NULL;
247         struct skb_frag_struct *frag;
248
249         size = skb_headlen(skb);
250         buf_num = (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
251
252         frag_num = skb_shinfo(skb)->nr_frags;
253         for (i = 0; i < frag_num; i++) {
254                 frag = &skb_shinfo(skb)->frags[i];
255                 size = skb_frag_size(frag);
256                 buf_num += (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
257         }
258
259         if (unlikely(buf_num > ring->max_desc_num_per_pkt)) {
260                 buf_num = (skb->len + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
261                 if (ring_space(ring) < buf_num)
262                         return -EBUSY;
263                 /* manual split the send packet */
264                 new_skb = skb_copy(skb, GFP_ATOMIC);
265                 if (!new_skb)
266                         return -ENOMEM;
267                 dev_kfree_skb_any(skb);
268                 *out_skb = new_skb;
269
270         } else if (ring_space(ring) < buf_num) {
271                 return -EBUSY;
272         }
273
274         *bnum = buf_num;
275         return 0;
276 }
277
278 static void fill_tso_desc(struct hnae_ring *ring, void *priv,
279                           int size, dma_addr_t dma, int frag_end,
280                           int buf_num, enum hns_desc_type type, int mtu)
281 {
282         int frag_buf_num;
283         int sizeoflast;
284         int k;
285
286         frag_buf_num = (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
287         sizeoflast = size % BD_MAX_SEND_SIZE;
288         sizeoflast = sizeoflast ? sizeoflast : BD_MAX_SEND_SIZE;
289
290         /* when the frag size is bigger than hardware, split this frag */
291         for (k = 0; k < frag_buf_num; k++)
292                 fill_v2_desc(ring, priv,
293                              (k == frag_buf_num - 1) ?
294                                         sizeoflast : BD_MAX_SEND_SIZE,
295                              dma + BD_MAX_SEND_SIZE * k,
296                              frag_end && (k == frag_buf_num - 1) ? 1 : 0,
297                              buf_num,
298                              (type == DESC_TYPE_SKB && !k) ?
299                                         DESC_TYPE_SKB : DESC_TYPE_PAGE,
300                              mtu);
301 }
302
303 netdev_tx_t hns_nic_net_xmit_hw(struct net_device *ndev,
304                                 struct sk_buff *skb,
305                                 struct hns_nic_ring_data *ring_data)
306 {
307         struct hns_nic_priv *priv = netdev_priv(ndev);
308         struct hnae_ring *ring = ring_data->ring;
309         struct device *dev = ring_to_dev(ring);
310         struct netdev_queue *dev_queue;
311         struct skb_frag_struct *frag;
312         int buf_num;
313         int seg_num;
314         dma_addr_t dma;
315         int size, next_to_use;
316         int i;
317
318         switch (priv->ops.maybe_stop_tx(&skb, &buf_num, ring)) {
319         case -EBUSY:
320                 ring->stats.tx_busy++;
321                 goto out_net_tx_busy;
322         case -ENOMEM:
323                 ring->stats.sw_err_cnt++;
324                 netdev_err(ndev, "no memory to xmit!\n");
325                 goto out_err_tx_ok;
326         default:
327                 break;
328         }
329
330         /* no. of segments (plus a header) */
331         seg_num = skb_shinfo(skb)->nr_frags + 1;
332         next_to_use = ring->next_to_use;
333
334         /* fill the first part */
335         size = skb_headlen(skb);
336         dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
337         if (dma_mapping_error(dev, dma)) {
338                 netdev_err(ndev, "TX head DMA map failed\n");
339                 ring->stats.sw_err_cnt++;
340                 goto out_err_tx_ok;
341         }
342         priv->ops.fill_desc(ring, skb, size, dma, seg_num == 1 ? 1 : 0,
343                             buf_num, DESC_TYPE_SKB, ndev->mtu);
344
345         /* fill the fragments */
346         for (i = 1; i < seg_num; i++) {
347                 frag = &skb_shinfo(skb)->frags[i - 1];
348                 size = skb_frag_size(frag);
349                 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
350                 if (dma_mapping_error(dev, dma)) {
351                         netdev_err(ndev, "TX frag(%d) DMA map failed\n", i);
352                         ring->stats.sw_err_cnt++;
353                         goto out_map_frag_fail;
354                 }
355                 priv->ops.fill_desc(ring, skb_frag_page(frag), size, dma,
356                                     seg_num - 1 == i ? 1 : 0, buf_num,
357                                     DESC_TYPE_PAGE, ndev->mtu);
358         }
359
360         /*complete translate all packets*/
361         dev_queue = netdev_get_tx_queue(ndev, skb->queue_mapping);
362         netdev_tx_sent_queue(dev_queue, skb->len);
363
364         netif_trans_update(ndev);
365         ndev->stats.tx_bytes += skb->len;
366         ndev->stats.tx_packets++;
367
368         wmb(); /* commit all data before submit */
369         assert(skb->queue_mapping < priv->ae_handle->q_num);
370         hnae_queue_xmit(priv->ae_handle->qs[skb->queue_mapping], buf_num);
371         ring->stats.tx_pkts++;
372         ring->stats.tx_bytes += skb->len;
373
374         return NETDEV_TX_OK;
375
376 out_map_frag_fail:
377
378         while (ring->next_to_use != next_to_use) {
379                 unfill_desc(ring);
380                 if (ring->next_to_use != next_to_use)
381                         dma_unmap_page(dev,
382                                        ring->desc_cb[ring->next_to_use].dma,
383                                        ring->desc_cb[ring->next_to_use].length,
384                                        DMA_TO_DEVICE);
385                 else
386                         dma_unmap_single(dev,
387                                          ring->desc_cb[next_to_use].dma,
388                                          ring->desc_cb[next_to_use].length,
389                                          DMA_TO_DEVICE);
390         }
391
392 out_err_tx_ok:
393
394         dev_kfree_skb_any(skb);
395         return NETDEV_TX_OK;
396
397 out_net_tx_busy:
398
399         netif_stop_subqueue(ndev, skb->queue_mapping);
400
401         /* Herbert's original patch had:
402          *  smp_mb__after_netif_stop_queue();
403          * but since that doesn't exist yet, just open code it.
404          */
405         smp_mb();
406         return NETDEV_TX_BUSY;
407 }
408
409 static void hns_nic_reuse_page(struct sk_buff *skb, int i,
410                                struct hnae_ring *ring, int pull_len,
411                                struct hnae_desc_cb *desc_cb)
412 {
413         struct hnae_desc *desc;
414         u32 truesize;
415         int size;
416         int last_offset;
417         bool twobufs;
418
419         twobufs = ((PAGE_SIZE < 8192) &&
420                 hnae_buf_size(ring) == HNS_BUFFER_SIZE_2048);
421
422         desc = &ring->desc[ring->next_to_clean];
423         size = le16_to_cpu(desc->rx.size);
424
425         if (twobufs) {
426                 truesize = hnae_buf_size(ring);
427         } else {
428                 truesize = ALIGN(size, L1_CACHE_BYTES);
429                 last_offset = hnae_page_size(ring) - hnae_buf_size(ring);
430         }
431
432         skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
433                         size - pull_len, truesize);
434
435          /* avoid re-using remote pages,flag default unreuse */
436         if (unlikely(page_to_nid(desc_cb->priv) != numa_node_id()))
437                 return;
438
439         if (twobufs) {
440                 /* if we are only owner of page we can reuse it */
441                 if (likely(page_count(desc_cb->priv) == 1)) {
442                         /* flip page offset to other buffer */
443                         desc_cb->page_offset ^= truesize;
444
445                         desc_cb->reuse_flag = 1;
446                         /* bump ref count on page before it is given*/
447                         get_page(desc_cb->priv);
448                 }
449                 return;
450         }
451
452         /* move offset up to the next cache line */
453         desc_cb->page_offset += truesize;
454
455         if (desc_cb->page_offset <= last_offset) {
456                 desc_cb->reuse_flag = 1;
457                 /* bump ref count on page before it is given*/
458                 get_page(desc_cb->priv);
459         }
460 }
461
462 static void get_v2rx_desc_bnum(u32 bnum_flag, int *out_bnum)
463 {
464         *out_bnum = hnae_get_field(bnum_flag,
465                                    HNS_RXD_BUFNUM_M, HNS_RXD_BUFNUM_S) + 1;
466 }
467
468 static void get_rx_desc_bnum(u32 bnum_flag, int *out_bnum)
469 {
470         *out_bnum = hnae_get_field(bnum_flag,
471                                    HNS_RXD_BUFNUM_M, HNS_RXD_BUFNUM_S);
472 }
473
474 static void hns_nic_rx_checksum(struct hns_nic_ring_data *ring_data,
475                                 struct sk_buff *skb, u32 flag)
476 {
477         struct net_device *netdev = ring_data->napi.dev;
478         u32 l3id;
479         u32 l4id;
480
481         /* check if RX checksum offload is enabled */
482         if (unlikely(!(netdev->features & NETIF_F_RXCSUM)))
483                 return;
484
485         /* In hardware, we only support checksum for the following protocols:
486          * 1) IPv4,
487          * 2) TCP(over IPv4 or IPv6),
488          * 3) UDP(over IPv4 or IPv6),
489          * 4) SCTP(over IPv4 or IPv6)
490          * but we support many L3(IPv4, IPv6, MPLS, PPPoE etc) and L4(TCP,
491          * UDP, GRE, SCTP, IGMP, ICMP etc.) protocols.
492          *
493          * Hardware limitation:
494          * Our present hardware RX Descriptor lacks L3/L4 checksum "Status &
495          * Error" bit (which usually can be used to indicate whether checksum
496          * was calculated by the hardware and if there was any error encountered
497          * during checksum calculation).
498          *
499          * Software workaround:
500          * We do get info within the RX descriptor about the kind of L3/L4
501          * protocol coming in the packet and the error status. These errors
502          * might not just be checksum errors but could be related to version,
503          * length of IPv4, UDP, TCP etc.
504          * Because there is no-way of knowing if it is a L3/L4 error due to bad
505          * checksum or any other L3/L4 error, we will not (cannot) convey
506          * checksum status for such cases to upper stack and will not maintain
507          * the RX L3/L4 checksum counters as well.
508          */
509
510         l3id = hnae_get_field(flag, HNS_RXD_L3ID_M, HNS_RXD_L3ID_S);
511         l4id = hnae_get_field(flag, HNS_RXD_L4ID_M, HNS_RXD_L4ID_S);
512
513         /*  check L3 protocol for which checksum is supported */
514         if ((l3id != HNS_RX_FLAG_L3ID_IPV4) && (l3id != HNS_RX_FLAG_L3ID_IPV6))
515                 return;
516
517         /* check for any(not just checksum)flagged L3 protocol errors */
518         if (unlikely(hnae_get_bit(flag, HNS_RXD_L3E_B)))
519                 return;
520
521         /* we do not support checksum of fragmented packets */
522         if (unlikely(hnae_get_bit(flag, HNS_RXD_FRAG_B)))
523                 return;
524
525         /*  check L4 protocol for which checksum is supported */
526         if ((l4id != HNS_RX_FLAG_L4ID_TCP) &&
527             (l4id != HNS_RX_FLAG_L4ID_UDP) &&
528             (l4id != HNS_RX_FLAG_L4ID_SCTP))
529                 return;
530
531         /* check for any(not just checksum)flagged L4 protocol errors */
532         if (unlikely(hnae_get_bit(flag, HNS_RXD_L4E_B)))
533                 return;
534
535         /* now, this has to be a packet with valid RX checksum */
536         skb->ip_summed = CHECKSUM_UNNECESSARY;
537 }
538
539 static int hns_nic_poll_rx_skb(struct hns_nic_ring_data *ring_data,
540                                struct sk_buff **out_skb, int *out_bnum)
541 {
542         struct hnae_ring *ring = ring_data->ring;
543         struct net_device *ndev = ring_data->napi.dev;
544         struct hns_nic_priv *priv = netdev_priv(ndev);
545         struct sk_buff *skb;
546         struct hnae_desc *desc;
547         struct hnae_desc_cb *desc_cb;
548         unsigned char *va;
549         int bnum, length, i;
550         int pull_len;
551         u32 bnum_flag;
552
553         desc = &ring->desc[ring->next_to_clean];
554         desc_cb = &ring->desc_cb[ring->next_to_clean];
555
556         prefetch(desc);
557
558         va = (unsigned char *)desc_cb->buf + desc_cb->page_offset;
559
560         /* prefetch first cache line of first page */
561         prefetch(va);
562 #if L1_CACHE_BYTES < 128
563         prefetch(va + L1_CACHE_BYTES);
564 #endif
565
566         skb = *out_skb = napi_alloc_skb(&ring_data->napi,
567                                         HNS_RX_HEAD_SIZE);
568         if (unlikely(!skb)) {
569                 netdev_err(ndev, "alloc rx skb fail\n");
570                 ring->stats.sw_err_cnt++;
571                 return -ENOMEM;
572         }
573
574         prefetchw(skb->data);
575         length = le16_to_cpu(desc->rx.pkt_len);
576         bnum_flag = le32_to_cpu(desc->rx.ipoff_bnum_pid_flag);
577         priv->ops.get_rxd_bnum(bnum_flag, &bnum);
578         *out_bnum = bnum;
579
580         if (length <= HNS_RX_HEAD_SIZE) {
581                 memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
582
583                 /* we can reuse buffer as-is, just make sure it is local */
584                 if (likely(page_to_nid(desc_cb->priv) == numa_node_id()))
585                         desc_cb->reuse_flag = 1;
586                 else /* this page cannot be reused so discard it */
587                         put_page(desc_cb->priv);
588
589                 ring_ptr_move_fw(ring, next_to_clean);
590
591                 if (unlikely(bnum != 1)) { /* check err*/
592                         *out_bnum = 1;
593                         goto out_bnum_err;
594                 }
595         } else {
596                 ring->stats.seg_pkt_cnt++;
597
598                 pull_len = eth_get_headlen(va, HNS_RX_HEAD_SIZE);
599                 memcpy(__skb_put(skb, pull_len), va,
600                        ALIGN(pull_len, sizeof(long)));
601
602                 hns_nic_reuse_page(skb, 0, ring, pull_len, desc_cb);
603                 ring_ptr_move_fw(ring, next_to_clean);
604
605                 if (unlikely(bnum >= (int)MAX_SKB_FRAGS)) { /* check err*/
606                         *out_bnum = 1;
607                         goto out_bnum_err;
608                 }
609                 for (i = 1; i < bnum; i++) {
610                         desc = &ring->desc[ring->next_to_clean];
611                         desc_cb = &ring->desc_cb[ring->next_to_clean];
612
613                         hns_nic_reuse_page(skb, i, ring, 0, desc_cb);
614                         ring_ptr_move_fw(ring, next_to_clean);
615                 }
616         }
617
618         /* check except process, free skb and jump the desc */
619         if (unlikely((!bnum) || (bnum > ring->max_desc_num_per_pkt))) {
620 out_bnum_err:
621                 *out_bnum = *out_bnum ? *out_bnum : 1; /* ntc moved,cannot 0*/
622                 netdev_err(ndev, "invalid bnum(%d,%d,%d,%d),%016llx,%016llx\n",
623                            bnum, ring->max_desc_num_per_pkt,
624                            length, (int)MAX_SKB_FRAGS,
625                            ((u64 *)desc)[0], ((u64 *)desc)[1]);
626                 ring->stats.err_bd_num++;
627                 dev_kfree_skb_any(skb);
628                 return -EDOM;
629         }
630
631         bnum_flag = le32_to_cpu(desc->rx.ipoff_bnum_pid_flag);
632
633         if (unlikely(!hnae_get_bit(bnum_flag, HNS_RXD_VLD_B))) {
634                 netdev_err(ndev, "no valid bd,%016llx,%016llx\n",
635                            ((u64 *)desc)[0], ((u64 *)desc)[1]);
636                 ring->stats.non_vld_descs++;
637                 dev_kfree_skb_any(skb);
638                 return -EINVAL;
639         }
640
641         if (unlikely((!desc->rx.pkt_len) ||
642                      hnae_get_bit(bnum_flag, HNS_RXD_DROP_B))) {
643                 ring->stats.err_pkt_len++;
644                 dev_kfree_skb_any(skb);
645                 return -EFAULT;
646         }
647
648         if (unlikely(hnae_get_bit(bnum_flag, HNS_RXD_L2E_B))) {
649                 ring->stats.l2_err++;
650                 dev_kfree_skb_any(skb);
651                 return -EFAULT;
652         }
653
654         ring->stats.rx_pkts++;
655         ring->stats.rx_bytes += skb->len;
656
657         /* indicate to upper stack if our hardware has already calculated
658          * the RX checksum
659          */
660         hns_nic_rx_checksum(ring_data, skb, bnum_flag);
661
662         return 0;
663 }
664
665 static void
666 hns_nic_alloc_rx_buffers(struct hns_nic_ring_data *ring_data, int cleand_count)
667 {
668         int i, ret;
669         struct hnae_desc_cb res_cbs;
670         struct hnae_desc_cb *desc_cb;
671         struct hnae_ring *ring = ring_data->ring;
672         struct net_device *ndev = ring_data->napi.dev;
673
674         for (i = 0; i < cleand_count; i++) {
675                 desc_cb = &ring->desc_cb[ring->next_to_use];
676                 if (desc_cb->reuse_flag) {
677                         ring->stats.reuse_pg_cnt++;
678                         hnae_reuse_buffer(ring, ring->next_to_use);
679                 } else {
680                         ret = hnae_reserve_buffer_map(ring, &res_cbs);
681                         if (ret) {
682                                 ring->stats.sw_err_cnt++;
683                                 netdev_err(ndev, "hnae reserve buffer map failed.\n");
684                                 break;
685                         }
686                         hnae_replace_buffer(ring, ring->next_to_use, &res_cbs);
687                 }
688
689                 ring_ptr_move_fw(ring, next_to_use);
690         }
691
692         wmb(); /* make all data has been write before submit */
693         writel_relaxed(i, ring->io_base + RCB_REG_HEAD);
694 }
695
696 /* return error number for error or number of desc left to take
697  */
698 static void hns_nic_rx_up_pro(struct hns_nic_ring_data *ring_data,
699                               struct sk_buff *skb)
700 {
701         struct net_device *ndev = ring_data->napi.dev;
702
703         skb->protocol = eth_type_trans(skb, ndev);
704         (void)napi_gro_receive(&ring_data->napi, skb);
705 }
706
707 static int hns_desc_unused(struct hnae_ring *ring)
708 {
709         int ntc = ring->next_to_clean;
710         int ntu = ring->next_to_use;
711
712         return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
713 }
714
715 #define HNS_LOWEST_LATENCY_RATE         27      /* 27 MB/s */
716 #define HNS_LOW_LATENCY_RATE                    80      /* 80 MB/s */
717
718 #define HNS_COAL_BDNUM                  3
719
720 static u32 hns_coal_rx_bdnum(struct hnae_ring *ring)
721 {
722         bool coal_enable = ring->q->handle->coal_adapt_en;
723
724         if (coal_enable &&
725             ring->coal_last_rx_bytes > HNS_LOWEST_LATENCY_RATE)
726                 return HNS_COAL_BDNUM;
727         else
728                 return 0;
729 }
730
731 static void hns_update_rx_rate(struct hnae_ring *ring)
732 {
733         bool coal_enable = ring->q->handle->coal_adapt_en;
734         u32 time_passed_ms;
735         u64 total_bytes;
736
737         if (!coal_enable ||
738             time_before(jiffies, ring->coal_last_jiffies + (HZ >> 4)))
739                 return;
740
741         /* ring->stats.rx_bytes overflowed */
742         if (ring->coal_last_rx_bytes > ring->stats.rx_bytes) {
743                 ring->coal_last_rx_bytes = ring->stats.rx_bytes;
744                 ring->coal_last_jiffies = jiffies;
745                 return;
746         }
747
748         total_bytes = ring->stats.rx_bytes - ring->coal_last_rx_bytes;
749         time_passed_ms = jiffies_to_msecs(jiffies - ring->coal_last_jiffies);
750         do_div(total_bytes, time_passed_ms);
751         ring->coal_rx_rate = total_bytes >> 10;
752
753         ring->coal_last_rx_bytes = ring->stats.rx_bytes;
754         ring->coal_last_jiffies = jiffies;
755 }
756
757 /**
758  * smooth_alg - smoothing algrithm for adjusting coalesce parameter
759  **/
760 static u32 smooth_alg(u32 new_param, u32 old_param)
761 {
762         u32 gap = (new_param > old_param) ? new_param - old_param
763                                           : old_param - new_param;
764
765         if (gap > 8)
766                 gap >>= 3;
767
768         if (new_param > old_param)
769                 return old_param + gap;
770         else
771                 return old_param - gap;
772 }
773
774 /**
775  * hns_nic_adp_coalesce - self adapte coalesce according to rx rate
776  * @ring_data: pointer to hns_nic_ring_data
777  **/
778 static void hns_nic_adpt_coalesce(struct hns_nic_ring_data *ring_data)
779 {
780         struct hnae_ring *ring = ring_data->ring;
781         struct hnae_handle *handle = ring->q->handle;
782         u32 new_coal_param, old_coal_param = ring->coal_param;
783
784         if (ring->coal_rx_rate < HNS_LOWEST_LATENCY_RATE)
785                 new_coal_param = HNAE_LOWEST_LATENCY_COAL_PARAM;
786         else if (ring->coal_rx_rate < HNS_LOW_LATENCY_RATE)
787                 new_coal_param = HNAE_LOW_LATENCY_COAL_PARAM;
788         else
789                 new_coal_param = HNAE_BULK_LATENCY_COAL_PARAM;
790
791         if (new_coal_param == old_coal_param &&
792             new_coal_param == handle->coal_param)
793                 return;
794
795         new_coal_param = smooth_alg(new_coal_param, old_coal_param);
796         ring->coal_param = new_coal_param;
797
798         /**
799          * Because all ring in one port has one coalesce param, when one ring
800          * calculate its own coalesce param, it cannot write to hardware at
801          * once. There are three conditions as follows:
802          *       1. current ring's coalesce param is larger than the hardware.
803          *       2. or ring which adapt last time can change again.
804          *       3. timeout.
805          */
806         if (new_coal_param == handle->coal_param) {
807                 handle->coal_last_jiffies = jiffies;
808                 handle->coal_ring_idx = ring_data->queue_index;
809         } else if (new_coal_param > handle->coal_param ||
810                    handle->coal_ring_idx == ring_data->queue_index ||
811                    time_after(jiffies, handle->coal_last_jiffies + (HZ >> 4))) {
812                 handle->dev->ops->set_coalesce_usecs(handle,
813                                         new_coal_param);
814                 handle->dev->ops->set_coalesce_frames(handle,
815                                         1, new_coal_param);
816                 handle->coal_param = new_coal_param;
817                 handle->coal_ring_idx = ring_data->queue_index;
818                 handle->coal_last_jiffies = jiffies;
819         }
820 }
821
822 static int hns_nic_rx_poll_one(struct hns_nic_ring_data *ring_data,
823                                int budget, void *v)
824 {
825         struct hnae_ring *ring = ring_data->ring;
826         struct sk_buff *skb;
827         int num, bnum;
828 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
829         int recv_pkts, recv_bds, clean_count, err;
830         int unused_count = hns_desc_unused(ring);
831
832         num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
833         rmb(); /* make sure num taken effect before the other data is touched */
834
835         recv_pkts = 0, recv_bds = 0, clean_count = 0;
836         num -= unused_count;
837
838         while (recv_pkts < budget && recv_bds < num) {
839                 /* reuse or realloc buffers */
840                 if (clean_count + unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
841                         hns_nic_alloc_rx_buffers(ring_data,
842                                                  clean_count + unused_count);
843                         clean_count = 0;
844                         unused_count = hns_desc_unused(ring);
845                 }
846
847                 /* poll one pkt */
848                 err = hns_nic_poll_rx_skb(ring_data, &skb, &bnum);
849                 if (unlikely(!skb)) /* this fault cannot be repaired */
850                         goto out;
851
852                 recv_bds += bnum;
853                 clean_count += bnum;
854                 if (unlikely(err)) {  /* do jump the err */
855                         recv_pkts++;
856                         continue;
857                 }
858
859                 /* do update ip stack process*/
860                 ((void (*)(struct hns_nic_ring_data *, struct sk_buff *))v)(
861                                                         ring_data, skb);
862                 recv_pkts++;
863         }
864
865 out:
866         /* make all data has been write before submit */
867         if (clean_count + unused_count > 0)
868                 hns_nic_alloc_rx_buffers(ring_data,
869                                          clean_count + unused_count);
870
871         return recv_pkts;
872 }
873
874 static bool hns_nic_rx_fini_pro(struct hns_nic_ring_data *ring_data)
875 {
876         struct hnae_ring *ring = ring_data->ring;
877         int num = 0;
878         bool rx_stopped;
879
880         hns_update_rx_rate(ring);
881
882         /* for hardware bug fixed */
883         ring_data->ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
884         num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
885
886         if (num <= hns_coal_rx_bdnum(ring)) {
887                 if (ring->q->handle->coal_adapt_en)
888                         hns_nic_adpt_coalesce(ring_data);
889
890                 rx_stopped = true;
891         } else {
892                 ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
893                         ring_data->ring, 1);
894
895                 rx_stopped = false;
896         }
897
898         return rx_stopped;
899 }
900
901 static bool hns_nic_rx_fini_pro_v2(struct hns_nic_ring_data *ring_data)
902 {
903         struct hnae_ring *ring = ring_data->ring;
904         int num;
905
906         hns_update_rx_rate(ring);
907         num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
908
909         if (num <= hns_coal_rx_bdnum(ring)) {
910                 if (ring->q->handle->coal_adapt_en)
911                         hns_nic_adpt_coalesce(ring_data);
912
913                 return true;
914         }
915
916         return false;
917 }
918
919 static inline void hns_nic_reclaim_one_desc(struct hnae_ring *ring,
920                                             int *bytes, int *pkts)
921 {
922         struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
923
924         (*pkts) += (desc_cb->type == DESC_TYPE_SKB);
925         (*bytes) += desc_cb->length;
926         /* desc_cb will be cleaned, after hnae_free_buffer_detach*/
927         hnae_free_buffer_detach(ring, ring->next_to_clean);
928
929         ring_ptr_move_fw(ring, next_to_clean);
930 }
931
932 static int is_valid_clean_head(struct hnae_ring *ring, int h)
933 {
934         int u = ring->next_to_use;
935         int c = ring->next_to_clean;
936
937         if (unlikely(h > ring->desc_num))
938                 return 0;
939
940         assert(u > 0 && u < ring->desc_num);
941         assert(c > 0 && c < ring->desc_num);
942         assert(u != c && h != c); /* must be checked before call this func */
943
944         return u > c ? (h > c && h <= u) : (h > c || h <= u);
945 }
946
947 /* netif_tx_lock will turn down the performance, set only when necessary */
948 #ifdef CONFIG_NET_POLL_CONTROLLER
949 #define NETIF_TX_LOCK(ring) spin_lock(&(ring)->lock)
950 #define NETIF_TX_UNLOCK(ring) spin_unlock(&(ring)->lock)
951 #else
952 #define NETIF_TX_LOCK(ring)
953 #define NETIF_TX_UNLOCK(ring)
954 #endif
955
956 /* reclaim all desc in one budget
957  * return error or number of desc left
958  */
959 static int hns_nic_tx_poll_one(struct hns_nic_ring_data *ring_data,
960                                int budget, void *v)
961 {
962         struct hnae_ring *ring = ring_data->ring;
963         struct net_device *ndev = ring_data->napi.dev;
964         struct netdev_queue *dev_queue;
965         struct hns_nic_priv *priv = netdev_priv(ndev);
966         int head;
967         int bytes, pkts;
968
969         NETIF_TX_LOCK(ring);
970
971         head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
972         rmb(); /* make sure head is ready before touch any data */
973
974         if (is_ring_empty(ring) || head == ring->next_to_clean) {
975                 NETIF_TX_UNLOCK(ring);
976                 return 0; /* no data to poll */
977         }
978
979         if (!is_valid_clean_head(ring, head)) {
980                 netdev_err(ndev, "wrong head (%d, %d-%d)\n", head,
981                            ring->next_to_use, ring->next_to_clean);
982                 ring->stats.io_err_cnt++;
983                 NETIF_TX_UNLOCK(ring);
984                 return -EIO;
985         }
986
987         bytes = 0;
988         pkts = 0;
989         while (head != ring->next_to_clean) {
990                 hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
991                 /* issue prefetch for next Tx descriptor */
992                 prefetch(&ring->desc_cb[ring->next_to_clean]);
993         }
994
995         NETIF_TX_UNLOCK(ring);
996
997         dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
998         netdev_tx_completed_queue(dev_queue, pkts, bytes);
999
1000         if (unlikely(priv->link && !netif_carrier_ok(ndev)))
1001                 netif_carrier_on(ndev);
1002
1003         if (unlikely(pkts && netif_carrier_ok(ndev) &&
1004                      (ring_space(ring) >= ring->max_desc_num_per_pkt * 2))) {
1005                 /* Make sure that anybody stopping the queue after this
1006                  * sees the new next_to_clean.
1007                  */
1008                 smp_mb();
1009                 if (netif_tx_queue_stopped(dev_queue) &&
1010                     !test_bit(NIC_STATE_DOWN, &priv->state)) {
1011                         netif_tx_wake_queue(dev_queue);
1012                         ring->stats.restart_queue++;
1013                 }
1014         }
1015         return 0;
1016 }
1017
1018 static bool hns_nic_tx_fini_pro(struct hns_nic_ring_data *ring_data)
1019 {
1020         struct hnae_ring *ring = ring_data->ring;
1021         int head;
1022
1023         ring_data->ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
1024
1025         head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1026
1027         if (head != ring->next_to_clean) {
1028                 ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1029                         ring_data->ring, 1);
1030
1031                 return false;
1032         } else {
1033                 return true;
1034         }
1035 }
1036
1037 static bool hns_nic_tx_fini_pro_v2(struct hns_nic_ring_data *ring_data)
1038 {
1039         struct hnae_ring *ring = ring_data->ring;
1040         int head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1041
1042         if (head == ring->next_to_clean)
1043                 return true;
1044         else
1045                 return false;
1046 }
1047
1048 static void hns_nic_tx_clr_all_bufs(struct hns_nic_ring_data *ring_data)
1049 {
1050         struct hnae_ring *ring = ring_data->ring;
1051         struct net_device *ndev = ring_data->napi.dev;
1052         struct netdev_queue *dev_queue;
1053         int head;
1054         int bytes, pkts;
1055
1056         NETIF_TX_LOCK(ring);
1057
1058         head = ring->next_to_use; /* ntu :soft setted ring position*/
1059         bytes = 0;
1060         pkts = 0;
1061         while (head != ring->next_to_clean)
1062                 hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
1063
1064         NETIF_TX_UNLOCK(ring);
1065
1066         dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
1067         netdev_tx_reset_queue(dev_queue);
1068 }
1069
1070 static int hns_nic_common_poll(struct napi_struct *napi, int budget)
1071 {
1072         int clean_complete = 0;
1073         struct hns_nic_ring_data *ring_data =
1074                 container_of(napi, struct hns_nic_ring_data, napi);
1075         struct hnae_ring *ring = ring_data->ring;
1076
1077 try_again:
1078         clean_complete += ring_data->poll_one(
1079                                 ring_data, budget - clean_complete,
1080                                 ring_data->ex_process);
1081
1082         if (clean_complete < budget) {
1083                 if (ring_data->fini_process(ring_data)) {
1084                         napi_complete(napi);
1085                         ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
1086                 } else {
1087                         goto try_again;
1088                 }
1089         }
1090
1091         return clean_complete;
1092 }
1093
1094 static irqreturn_t hns_irq_handle(int irq, void *dev)
1095 {
1096         struct hns_nic_ring_data *ring_data = (struct hns_nic_ring_data *)dev;
1097
1098         ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1099                 ring_data->ring, 1);
1100         napi_schedule(&ring_data->napi);
1101
1102         return IRQ_HANDLED;
1103 }
1104
1105 /**
1106  *hns_nic_adjust_link - adjust net work mode by the phy stat or new param
1107  *@ndev: net device
1108  */
1109 static void hns_nic_adjust_link(struct net_device *ndev)
1110 {
1111         struct hns_nic_priv *priv = netdev_priv(ndev);
1112         struct hnae_handle *h = priv->ae_handle;
1113         int state = 1;
1114
1115         /* If there is no phy, do not need adjust link */
1116         if (ndev->phydev) {
1117                 /* When phy link down, do nothing */
1118                 if (ndev->phydev->link == 0)
1119                         return;
1120
1121                 if (h->dev->ops->need_adjust_link(h, ndev->phydev->speed,
1122                                                   ndev->phydev->duplex)) {
1123                         /* because Hi161X chip don't support to change gmac
1124                          * speed and duplex with traffic. Delay 200ms to
1125                          * make sure there is no more data in chip FIFO.
1126                          */
1127                         netif_carrier_off(ndev);
1128                         msleep(200);
1129                         h->dev->ops->adjust_link(h, ndev->phydev->speed,
1130                                                  ndev->phydev->duplex);
1131                         netif_carrier_on(ndev);
1132                 }
1133         }
1134
1135         state = state && h->dev->ops->get_status(h);
1136
1137         if (state != priv->link) {
1138                 if (state) {
1139                         netif_carrier_on(ndev);
1140                         netif_tx_wake_all_queues(ndev);
1141                         netdev_info(ndev, "link up\n");
1142                 } else {
1143                         netif_carrier_off(ndev);
1144                         netdev_info(ndev, "link down\n");
1145                 }
1146                 priv->link = state;
1147         }
1148 }
1149
1150 /**
1151  *hns_nic_init_phy - init phy
1152  *@ndev: net device
1153  *@h: ae handle
1154  * Return 0 on success, negative on failure
1155  */
1156 int hns_nic_init_phy(struct net_device *ndev, struct hnae_handle *h)
1157 {
1158         struct phy_device *phy_dev = h->phy_dev;
1159         int ret;
1160
1161         if (!h->phy_dev)
1162                 return 0;
1163
1164         if (h->phy_if != PHY_INTERFACE_MODE_XGMII) {
1165                 phy_dev->dev_flags = 0;
1166
1167                 ret = phy_connect_direct(ndev, phy_dev, hns_nic_adjust_link,
1168                                          h->phy_if);
1169         } else {
1170                 ret = phy_attach_direct(ndev, phy_dev, 0, h->phy_if);
1171         }
1172         if (unlikely(ret))
1173                 return -ENODEV;
1174
1175         phy_dev->supported &= h->if_support;
1176         phy_dev->advertising = phy_dev->supported;
1177
1178         if (h->phy_if == PHY_INTERFACE_MODE_XGMII)
1179                 phy_dev->autoneg = false;
1180
1181         return 0;
1182 }
1183
1184 static int hns_nic_ring_open(struct net_device *netdev, int idx)
1185 {
1186         struct hns_nic_priv *priv = netdev_priv(netdev);
1187         struct hnae_handle *h = priv->ae_handle;
1188
1189         napi_enable(&priv->ring_data[idx].napi);
1190
1191         enable_irq(priv->ring_data[idx].ring->irq);
1192         h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 0);
1193
1194         return 0;
1195 }
1196
1197 static int hns_nic_net_set_mac_address(struct net_device *ndev, void *p)
1198 {
1199         struct hns_nic_priv *priv = netdev_priv(ndev);
1200         struct hnae_handle *h = priv->ae_handle;
1201         struct sockaddr *mac_addr = p;
1202         int ret;
1203
1204         if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1205                 return -EADDRNOTAVAIL;
1206
1207         ret = h->dev->ops->set_mac_addr(h, mac_addr->sa_data);
1208         if (ret) {
1209                 netdev_err(ndev, "set_mac_address fail, ret=%d!\n", ret);
1210                 return ret;
1211         }
1212
1213         memcpy(ndev->dev_addr, mac_addr->sa_data, ndev->addr_len);
1214
1215         return 0;
1216 }
1217
1218 static void hns_nic_update_stats(struct net_device *netdev)
1219 {
1220         struct hns_nic_priv *priv = netdev_priv(netdev);
1221         struct hnae_handle *h = priv->ae_handle;
1222
1223         h->dev->ops->update_stats(h, &netdev->stats);
1224 }
1225
1226 /* set mac addr if it is configed. or leave it to the AE driver */
1227 static void hns_init_mac_addr(struct net_device *ndev)
1228 {
1229         struct hns_nic_priv *priv = netdev_priv(ndev);
1230
1231         if (!device_get_mac_address(priv->dev, ndev->dev_addr, ETH_ALEN)) {
1232                 eth_hw_addr_random(ndev);
1233                 dev_warn(priv->dev, "No valid mac, use random mac %pM",
1234                          ndev->dev_addr);
1235         }
1236 }
1237
1238 static void hns_nic_ring_close(struct net_device *netdev, int idx)
1239 {
1240         struct hns_nic_priv *priv = netdev_priv(netdev);
1241         struct hnae_handle *h = priv->ae_handle;
1242
1243         h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 1);
1244         disable_irq(priv->ring_data[idx].ring->irq);
1245
1246         napi_disable(&priv->ring_data[idx].napi);
1247 }
1248
1249 static int hns_nic_init_affinity_mask(int q_num, int ring_idx,
1250                                       struct hnae_ring *ring, cpumask_t *mask)
1251 {
1252         int cpu;
1253
1254         /* Diffrent irq banlance between 16core and 32core.
1255          * The cpu mask set by ring index according to the ring flag
1256          * which indicate the ring is tx or rx.
1257          */
1258         if (q_num == num_possible_cpus()) {
1259                 if (is_tx_ring(ring))
1260                         cpu = ring_idx;
1261                 else
1262                         cpu = ring_idx - q_num;
1263         } else {
1264                 if (is_tx_ring(ring))
1265                         cpu = ring_idx * 2;
1266                 else
1267                         cpu = (ring_idx - q_num) * 2 + 1;
1268         }
1269
1270         cpumask_clear(mask);
1271         cpumask_set_cpu(cpu, mask);
1272
1273         return cpu;
1274 }
1275
1276 static int hns_nic_init_irq(struct hns_nic_priv *priv)
1277 {
1278         struct hnae_handle *h = priv->ae_handle;
1279         struct hns_nic_ring_data *rd;
1280         int i;
1281         int ret;
1282         int cpu;
1283
1284         for (i = 0; i < h->q_num * 2; i++) {
1285                 rd = &priv->ring_data[i];
1286
1287                 if (rd->ring->irq_init_flag == RCB_IRQ_INITED)
1288                         break;
1289
1290                 snprintf(rd->ring->ring_name, RCB_RING_NAME_LEN,
1291                          "%s-%s%d", priv->netdev->name,
1292                          (is_tx_ring(rd->ring) ? "tx" : "rx"), rd->queue_index);
1293
1294                 rd->ring->ring_name[RCB_RING_NAME_LEN - 1] = '\0';
1295
1296                 ret = request_irq(rd->ring->irq,
1297                                   hns_irq_handle, 0, rd->ring->ring_name, rd);
1298                 if (ret) {
1299                         netdev_err(priv->netdev, "request irq(%d) fail\n",
1300                                    rd->ring->irq);
1301                         return ret;
1302                 }
1303                 disable_irq(rd->ring->irq);
1304
1305                 cpu = hns_nic_init_affinity_mask(h->q_num, i,
1306                                                  rd->ring, &rd->mask);
1307
1308                 if (cpu_online(cpu))
1309                         irq_set_affinity_hint(rd->ring->irq,
1310                                               &rd->mask);
1311
1312                 rd->ring->irq_init_flag = RCB_IRQ_INITED;
1313         }
1314
1315         return 0;
1316 }
1317
1318 static int hns_nic_net_up(struct net_device *ndev)
1319 {
1320         struct hns_nic_priv *priv = netdev_priv(ndev);
1321         struct hnae_handle *h = priv->ae_handle;
1322         int i, j;
1323         int ret;
1324
1325         ret = hns_nic_init_irq(priv);
1326         if (ret != 0) {
1327                 netdev_err(ndev, "hns init irq failed! ret=%d\n", ret);
1328                 return ret;
1329         }
1330
1331         for (i = 0; i < h->q_num * 2; i++) {
1332                 ret = hns_nic_ring_open(ndev, i);
1333                 if (ret)
1334                         goto out_has_some_queues;
1335         }
1336
1337         ret = h->dev->ops->set_mac_addr(h, ndev->dev_addr);
1338         if (ret)
1339                 goto out_set_mac_addr_err;
1340
1341         ret = h->dev->ops->start ? h->dev->ops->start(h) : 0;
1342         if (ret)
1343                 goto out_start_err;
1344
1345         if (ndev->phydev)
1346                 phy_start(ndev->phydev);
1347
1348         clear_bit(NIC_STATE_DOWN, &priv->state);
1349         (void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
1350
1351         return 0;
1352
1353 out_start_err:
1354         netif_stop_queue(ndev);
1355 out_set_mac_addr_err:
1356 out_has_some_queues:
1357         for (j = i - 1; j >= 0; j--)
1358                 hns_nic_ring_close(ndev, j);
1359
1360         set_bit(NIC_STATE_DOWN, &priv->state);
1361
1362         return ret;
1363 }
1364
1365 static void hns_nic_net_down(struct net_device *ndev)
1366 {
1367         int i;
1368         struct hnae_ae_ops *ops;
1369         struct hns_nic_priv *priv = netdev_priv(ndev);
1370
1371         if (test_and_set_bit(NIC_STATE_DOWN, &priv->state))
1372                 return;
1373
1374         (void)del_timer_sync(&priv->service_timer);
1375         netif_tx_stop_all_queues(ndev);
1376         netif_carrier_off(ndev);
1377         netif_tx_disable(ndev);
1378         priv->link = 0;
1379
1380         if (ndev->phydev)
1381                 phy_stop(ndev->phydev);
1382
1383         ops = priv->ae_handle->dev->ops;
1384
1385         if (ops->stop)
1386                 ops->stop(priv->ae_handle);
1387
1388         netif_tx_stop_all_queues(ndev);
1389
1390         for (i = priv->ae_handle->q_num - 1; i >= 0; i--) {
1391                 hns_nic_ring_close(ndev, i);
1392                 hns_nic_ring_close(ndev, i + priv->ae_handle->q_num);
1393
1394                 /* clean tx buffers*/
1395                 hns_nic_tx_clr_all_bufs(priv->ring_data + i);
1396         }
1397 }
1398
1399 void hns_nic_net_reset(struct net_device *ndev)
1400 {
1401         struct hns_nic_priv *priv = netdev_priv(ndev);
1402         struct hnae_handle *handle = priv->ae_handle;
1403
1404         while (test_and_set_bit(NIC_STATE_RESETTING, &priv->state))
1405                 usleep_range(1000, 2000);
1406
1407         (void)hnae_reinit_handle(handle);
1408
1409         clear_bit(NIC_STATE_RESETTING, &priv->state);
1410 }
1411
1412 void hns_nic_net_reinit(struct net_device *netdev)
1413 {
1414         struct hns_nic_priv *priv = netdev_priv(netdev);
1415         enum hnae_port_type type = priv->ae_handle->port_type;
1416
1417         netif_trans_update(priv->netdev);
1418         while (test_and_set_bit(NIC_STATE_REINITING, &priv->state))
1419                 usleep_range(1000, 2000);
1420
1421         hns_nic_net_down(netdev);
1422
1423         /* Only do hns_nic_net_reset in debug mode
1424          * because of hardware limitation.
1425          */
1426         if (type == HNAE_PORT_DEBUG)
1427                 hns_nic_net_reset(netdev);
1428
1429         (void)hns_nic_net_up(netdev);
1430         clear_bit(NIC_STATE_REINITING, &priv->state);
1431 }
1432
1433 static int hns_nic_net_open(struct net_device *ndev)
1434 {
1435         struct hns_nic_priv *priv = netdev_priv(ndev);
1436         struct hnae_handle *h = priv->ae_handle;
1437         int ret;
1438
1439         if (test_bit(NIC_STATE_TESTING, &priv->state))
1440                 return -EBUSY;
1441
1442         priv->link = 0;
1443         netif_carrier_off(ndev);
1444
1445         ret = netif_set_real_num_tx_queues(ndev, h->q_num);
1446         if (ret < 0) {
1447                 netdev_err(ndev, "netif_set_real_num_tx_queues fail, ret=%d!\n",
1448                            ret);
1449                 return ret;
1450         }
1451
1452         ret = netif_set_real_num_rx_queues(ndev, h->q_num);
1453         if (ret < 0) {
1454                 netdev_err(ndev,
1455                            "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
1456                 return ret;
1457         }
1458
1459         ret = hns_nic_net_up(ndev);
1460         if (ret) {
1461                 netdev_err(ndev,
1462                            "hns net up fail, ret=%d!\n", ret);
1463                 return ret;
1464         }
1465
1466         return 0;
1467 }
1468
1469 static int hns_nic_net_stop(struct net_device *ndev)
1470 {
1471         hns_nic_net_down(ndev);
1472
1473         return 0;
1474 }
1475
1476 static void hns_tx_timeout_reset(struct hns_nic_priv *priv);
1477 static void hns_nic_net_timeout(struct net_device *ndev)
1478 {
1479         struct hns_nic_priv *priv = netdev_priv(ndev);
1480
1481         hns_tx_timeout_reset(priv);
1482 }
1483
1484 static int hns_nic_do_ioctl(struct net_device *netdev, struct ifreq *ifr,
1485                             int cmd)
1486 {
1487         struct phy_device *phy_dev = netdev->phydev;
1488
1489         if (!netif_running(netdev))
1490                 return -EINVAL;
1491
1492         if (!phy_dev)
1493                 return -ENOTSUPP;
1494
1495         return phy_mii_ioctl(phy_dev, ifr, cmd);
1496 }
1497
1498 /* use only for netconsole to poll with the device without interrupt */
1499 #ifdef CONFIG_NET_POLL_CONTROLLER
1500 static void hns_nic_poll_controller(struct net_device *ndev)
1501 {
1502         struct hns_nic_priv *priv = netdev_priv(ndev);
1503         unsigned long flags;
1504         int i;
1505
1506         local_irq_save(flags);
1507         for (i = 0; i < priv->ae_handle->q_num * 2; i++)
1508                 napi_schedule(&priv->ring_data[i].napi);
1509         local_irq_restore(flags);
1510 }
1511 #endif
1512
1513 static netdev_tx_t hns_nic_net_xmit(struct sk_buff *skb,
1514                                     struct net_device *ndev)
1515 {
1516         struct hns_nic_priv *priv = netdev_priv(ndev);
1517
1518         assert(skb->queue_mapping < ndev->ae_handle->q_num);
1519
1520         return hns_nic_net_xmit_hw(ndev, skb,
1521                                    &tx_ring_data(priv, skb->queue_mapping));
1522 }
1523
1524 static void hns_nic_drop_rx_fetch(struct hns_nic_ring_data *ring_data,
1525                                   struct sk_buff *skb)
1526 {
1527         dev_kfree_skb_any(skb);
1528 }
1529
1530 #define HNS_LB_TX_RING  0
1531 static struct sk_buff *hns_assemble_skb(struct net_device *ndev)
1532 {
1533         struct sk_buff *skb;
1534         struct ethhdr *ethhdr;
1535         int frame_len;
1536
1537         /* allocate test skb */
1538         skb = alloc_skb(64, GFP_KERNEL);
1539         if (!skb)
1540                 return NULL;
1541
1542         skb_put(skb, 64);
1543         skb->dev = ndev;
1544         memset(skb->data, 0xFF, skb->len);
1545
1546         /* must be tcp/ip package */
1547         ethhdr = (struct ethhdr *)skb->data;
1548         ethhdr->h_proto = htons(ETH_P_IP);
1549
1550         frame_len = skb->len & (~1ul);
1551         memset(&skb->data[frame_len / 2], 0xAA,
1552                frame_len / 2 - 1);
1553
1554         skb->queue_mapping = HNS_LB_TX_RING;
1555
1556         return skb;
1557 }
1558
1559 static int hns_enable_serdes_lb(struct net_device *ndev)
1560 {
1561         struct hns_nic_priv *priv = netdev_priv(ndev);
1562         struct hnae_handle *h = priv->ae_handle;
1563         struct hnae_ae_ops *ops = h->dev->ops;
1564         int speed, duplex;
1565         int ret;
1566
1567         ret = ops->set_loopback(h, MAC_INTERNALLOOP_SERDES, 1);
1568         if (ret)
1569                 return ret;
1570
1571         ret = ops->start ? ops->start(h) : 0;
1572         if (ret)
1573                 return ret;
1574
1575         /* link adjust duplex*/
1576         if (h->phy_if != PHY_INTERFACE_MODE_XGMII)
1577                 speed = 1000;
1578         else
1579                 speed = 10000;
1580         duplex = 1;
1581
1582         ops->adjust_link(h, speed, duplex);
1583
1584         /* wait h/w ready */
1585         mdelay(300);
1586
1587         return 0;
1588 }
1589
1590 static void hns_disable_serdes_lb(struct net_device *ndev)
1591 {
1592         struct hns_nic_priv *priv = netdev_priv(ndev);
1593         struct hnae_handle *h = priv->ae_handle;
1594         struct hnae_ae_ops *ops = h->dev->ops;
1595
1596         ops->stop(h);
1597         ops->set_loopback(h, MAC_INTERNALLOOP_SERDES, 0);
1598 }
1599
1600 /**
1601  *hns_nic_clear_all_rx_fetch - clear the chip fetched descriptions. The
1602  *function as follows:
1603  *    1. if one rx ring has found the page_offset is not equal 0 between head
1604  *       and tail, it means that the chip fetched the wrong descs for the ring
1605  *       which buffer size is 4096.
1606  *    2. we set the chip serdes loopback and set rss indirection to the ring.
1607  *    3. construct 64-bytes ip broadcast packages, wait the associated rx ring
1608  *       recieving all packages and it will fetch new descriptions.
1609  *    4. recover to the original state.
1610  *
1611  *@ndev: net device
1612  */
1613 static int hns_nic_clear_all_rx_fetch(struct net_device *ndev)
1614 {
1615         struct hns_nic_priv *priv = netdev_priv(ndev);
1616         struct hnae_handle *h = priv->ae_handle;
1617         struct hnae_ae_ops *ops = h->dev->ops;
1618         struct hns_nic_ring_data *rd;
1619         struct hnae_ring *ring;
1620         struct sk_buff *skb;
1621         u32 *org_indir;
1622         u32 *cur_indir;
1623         int indir_size;
1624         int head, tail;
1625         int fetch_num;
1626         int i, j;
1627         bool found;
1628         int retry_times;
1629         int ret = 0;
1630
1631         /* alloc indir memory */
1632         indir_size = ops->get_rss_indir_size(h) * sizeof(*org_indir);
1633         org_indir = kzalloc(indir_size, GFP_KERNEL);
1634         if (!org_indir)
1635                 return -ENOMEM;
1636
1637         /* store the orginal indirection */
1638         ops->get_rss(h, org_indir, NULL, NULL);
1639
1640         cur_indir = kzalloc(indir_size, GFP_KERNEL);
1641         if (!cur_indir) {
1642                 ret = -ENOMEM;
1643                 goto cur_indir_alloc_err;
1644         }
1645
1646         /* set loopback */
1647         if (hns_enable_serdes_lb(ndev)) {
1648                 ret = -EINVAL;
1649                 goto enable_serdes_lb_err;
1650         }
1651
1652         /* foreach every rx ring to clear fetch desc */
1653         for (i = 0; i < h->q_num; i++) {
1654                 ring = &h->qs[i]->rx_ring;
1655                 head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1656                 tail = readl_relaxed(ring->io_base + RCB_REG_TAIL);
1657                 found = false;
1658                 fetch_num = ring_dist(ring, head, tail);
1659
1660                 while (head != tail) {
1661                         if (ring->desc_cb[head].page_offset != 0) {
1662                                 found = true;
1663                                 break;
1664                         }
1665
1666                         head++;
1667                         if (head == ring->desc_num)
1668                                 head = 0;
1669                 }
1670
1671                 if (found) {
1672                         for (j = 0; j < indir_size / sizeof(*org_indir); j++)
1673                                 cur_indir[j] = i;
1674                         ops->set_rss(h, cur_indir, NULL, 0);
1675
1676                         for (j = 0; j < fetch_num; j++) {
1677                                 /* alloc one skb and init */
1678                                 skb = hns_assemble_skb(ndev);
1679                                 if (!skb)
1680                                         goto out;
1681                                 rd = &tx_ring_data(priv, skb->queue_mapping);
1682                                 hns_nic_net_xmit_hw(ndev, skb, rd);
1683
1684                                 retry_times = 0;
1685                                 while (retry_times++ < 10) {
1686                                         mdelay(10);
1687                                         /* clean rx */
1688                                         rd = &rx_ring_data(priv, i);
1689                                         if (rd->poll_one(rd, fetch_num,
1690                                                          hns_nic_drop_rx_fetch))
1691                                                 break;
1692                                 }
1693
1694                                 retry_times = 0;
1695                                 while (retry_times++ < 10) {
1696                                         mdelay(10);
1697                                         /* clean tx ring 0 send package */
1698                                         rd = &tx_ring_data(priv,
1699                                                            HNS_LB_TX_RING);
1700                                         if (rd->poll_one(rd, fetch_num, NULL))
1701                                                 break;
1702                                 }
1703                         }
1704                 }
1705         }
1706
1707 out:
1708         /* restore everything */
1709         ops->set_rss(h, org_indir, NULL, 0);
1710         hns_disable_serdes_lb(ndev);
1711 enable_serdes_lb_err:
1712         kfree(cur_indir);
1713 cur_indir_alloc_err:
1714         kfree(org_indir);
1715
1716         return ret;
1717 }
1718
1719 static int hns_nic_change_mtu(struct net_device *ndev, int new_mtu)
1720 {
1721         struct hns_nic_priv *priv = netdev_priv(ndev);
1722         struct hnae_handle *h = priv->ae_handle;
1723         bool if_running = netif_running(ndev);
1724         int ret;
1725
1726         /* MTU < 68 is an error and causes problems on some kernels */
1727         if (new_mtu < 68)
1728                 return -EINVAL;
1729
1730         /* MTU no change */
1731         if (new_mtu == ndev->mtu)
1732                 return 0;
1733
1734         if (!h->dev->ops->set_mtu)
1735                 return -ENOTSUPP;
1736
1737         if (if_running) {
1738                 (void)hns_nic_net_stop(ndev);
1739                 msleep(100);
1740         }
1741
1742         if (priv->enet_ver != AE_VERSION_1 &&
1743             ndev->mtu <= BD_SIZE_2048_MAX_MTU &&
1744             new_mtu > BD_SIZE_2048_MAX_MTU) {
1745                 /* update desc */
1746                 hnae_reinit_all_ring_desc(h);
1747
1748                 /* clear the package which the chip has fetched */
1749                 ret = hns_nic_clear_all_rx_fetch(ndev);
1750
1751                 /* the page offset must be consist with desc */
1752                 hnae_reinit_all_ring_page_off(h);
1753
1754                 if (ret) {
1755                         netdev_err(ndev, "clear the fetched desc fail\n");
1756                         goto out;
1757                 }
1758         }
1759
1760         ret = h->dev->ops->set_mtu(h, new_mtu);
1761         if (ret) {
1762                 netdev_err(ndev, "set mtu fail, return value %d\n",
1763                            ret);
1764                 goto out;
1765         }
1766
1767         /* finally, set new mtu to netdevice */
1768         ndev->mtu = new_mtu;
1769
1770 out:
1771         if (if_running) {
1772                 if (hns_nic_net_open(ndev)) {
1773                         netdev_err(ndev, "hns net open fail\n");
1774                         ret = -EINVAL;
1775                 }
1776         }
1777
1778         return ret;
1779 }
1780
1781 static int hns_nic_set_features(struct net_device *netdev,
1782                                 netdev_features_t features)
1783 {
1784         struct hns_nic_priv *priv = netdev_priv(netdev);
1785
1786         switch (priv->enet_ver) {
1787         case AE_VERSION_1:
1788                 if (features & (NETIF_F_TSO | NETIF_F_TSO6))
1789                         netdev_info(netdev, "enet v1 do not support tso!\n");
1790                 break;
1791         default:
1792                 if (features & (NETIF_F_TSO | NETIF_F_TSO6)) {
1793                         priv->ops.fill_desc = fill_tso_desc;
1794                         priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tso;
1795                         /* The chip only support 7*4096 */
1796                         netif_set_gso_max_size(netdev, 7 * 4096);
1797                 } else {
1798                         priv->ops.fill_desc = fill_v2_desc;
1799                         priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
1800                 }
1801                 break;
1802         }
1803         netdev->features = features;
1804         return 0;
1805 }
1806
1807 static netdev_features_t hns_nic_fix_features(
1808                 struct net_device *netdev, netdev_features_t features)
1809 {
1810         struct hns_nic_priv *priv = netdev_priv(netdev);
1811
1812         switch (priv->enet_ver) {
1813         case AE_VERSION_1:
1814                 features &= ~(NETIF_F_TSO | NETIF_F_TSO6 |
1815                                 NETIF_F_HW_VLAN_CTAG_FILTER);
1816                 break;
1817         default:
1818                 break;
1819         }
1820         return features;
1821 }
1822
1823 static int hns_nic_uc_sync(struct net_device *netdev, const unsigned char *addr)
1824 {
1825         struct hns_nic_priv *priv = netdev_priv(netdev);
1826         struct hnae_handle *h = priv->ae_handle;
1827
1828         if (h->dev->ops->add_uc_addr)
1829                 return h->dev->ops->add_uc_addr(h, addr);
1830
1831         return 0;
1832 }
1833
1834 static int hns_nic_uc_unsync(struct net_device *netdev,
1835                              const unsigned char *addr)
1836 {
1837         struct hns_nic_priv *priv = netdev_priv(netdev);
1838         struct hnae_handle *h = priv->ae_handle;
1839
1840         if (h->dev->ops->rm_uc_addr)
1841                 return h->dev->ops->rm_uc_addr(h, addr);
1842
1843         return 0;
1844 }
1845
1846 /**
1847  * nic_set_multicast_list - set mutl mac address
1848  * @netdev: net device
1849  * @p: mac address
1850  *
1851  * return void
1852  */
1853 static void hns_set_multicast_list(struct net_device *ndev)
1854 {
1855         struct hns_nic_priv *priv = netdev_priv(ndev);
1856         struct hnae_handle *h = priv->ae_handle;
1857         struct netdev_hw_addr *ha = NULL;
1858
1859         if (!h) {
1860                 netdev_err(ndev, "hnae handle is null\n");
1861                 return;
1862         }
1863
1864         if (h->dev->ops->clr_mc_addr)
1865                 if (h->dev->ops->clr_mc_addr(h))
1866                         netdev_err(ndev, "clear multicast address fail\n");
1867
1868         if (h->dev->ops->set_mc_addr) {
1869                 netdev_for_each_mc_addr(ha, ndev)
1870                         if (h->dev->ops->set_mc_addr(h, ha->addr))
1871                                 netdev_err(ndev, "set multicast fail\n");
1872         }
1873 }
1874
1875 static void hns_nic_set_rx_mode(struct net_device *ndev)
1876 {
1877         struct hns_nic_priv *priv = netdev_priv(ndev);
1878         struct hnae_handle *h = priv->ae_handle;
1879
1880         if (h->dev->ops->set_promisc_mode) {
1881                 if (ndev->flags & IFF_PROMISC)
1882                         h->dev->ops->set_promisc_mode(h, 1);
1883                 else
1884                         h->dev->ops->set_promisc_mode(h, 0);
1885         }
1886
1887         hns_set_multicast_list(ndev);
1888
1889         if (__dev_uc_sync(ndev, hns_nic_uc_sync, hns_nic_uc_unsync))
1890                 netdev_err(ndev, "sync uc address fail\n");
1891 }
1892
1893 static void hns_nic_get_stats64(struct net_device *ndev,
1894                                 struct rtnl_link_stats64 *stats)
1895 {
1896         int idx = 0;
1897         u64 tx_bytes = 0;
1898         u64 rx_bytes = 0;
1899         u64 tx_pkts = 0;
1900         u64 rx_pkts = 0;
1901         struct hns_nic_priv *priv = netdev_priv(ndev);
1902         struct hnae_handle *h = priv->ae_handle;
1903
1904         for (idx = 0; idx < h->q_num; idx++) {
1905                 tx_bytes += h->qs[idx]->tx_ring.stats.tx_bytes;
1906                 tx_pkts += h->qs[idx]->tx_ring.stats.tx_pkts;
1907                 rx_bytes += h->qs[idx]->rx_ring.stats.rx_bytes;
1908                 rx_pkts += h->qs[idx]->rx_ring.stats.rx_pkts;
1909         }
1910
1911         stats->tx_bytes = tx_bytes;
1912         stats->tx_packets = tx_pkts;
1913         stats->rx_bytes = rx_bytes;
1914         stats->rx_packets = rx_pkts;
1915
1916         stats->rx_errors = ndev->stats.rx_errors;
1917         stats->multicast = ndev->stats.multicast;
1918         stats->rx_length_errors = ndev->stats.rx_length_errors;
1919         stats->rx_crc_errors = ndev->stats.rx_crc_errors;
1920         stats->rx_missed_errors = ndev->stats.rx_missed_errors;
1921
1922         stats->tx_errors = ndev->stats.tx_errors;
1923         stats->rx_dropped = ndev->stats.rx_dropped;
1924         stats->tx_dropped = ndev->stats.tx_dropped;
1925         stats->collisions = ndev->stats.collisions;
1926         stats->rx_over_errors = ndev->stats.rx_over_errors;
1927         stats->rx_frame_errors = ndev->stats.rx_frame_errors;
1928         stats->rx_fifo_errors = ndev->stats.rx_fifo_errors;
1929         stats->tx_aborted_errors = ndev->stats.tx_aborted_errors;
1930         stats->tx_carrier_errors = ndev->stats.tx_carrier_errors;
1931         stats->tx_fifo_errors = ndev->stats.tx_fifo_errors;
1932         stats->tx_heartbeat_errors = ndev->stats.tx_heartbeat_errors;
1933         stats->tx_window_errors = ndev->stats.tx_window_errors;
1934         stats->rx_compressed = ndev->stats.rx_compressed;
1935         stats->tx_compressed = ndev->stats.tx_compressed;
1936 }
1937
1938 static u16
1939 hns_nic_select_queue(struct net_device *ndev, struct sk_buff *skb,
1940                      struct net_device *sb_dev,
1941                      select_queue_fallback_t fallback)
1942 {
1943         struct ethhdr *eth_hdr = (struct ethhdr *)skb->data;
1944         struct hns_nic_priv *priv = netdev_priv(ndev);
1945
1946         /* fix hardware broadcast/multicast packets queue loopback */
1947         if (!AE_IS_VER1(priv->enet_ver) &&
1948             is_multicast_ether_addr(eth_hdr->h_dest))
1949                 return 0;
1950         else
1951                 return fallback(ndev, skb, NULL);
1952 }
1953
1954 static const struct net_device_ops hns_nic_netdev_ops = {
1955         .ndo_open = hns_nic_net_open,
1956         .ndo_stop = hns_nic_net_stop,
1957         .ndo_start_xmit = hns_nic_net_xmit,
1958         .ndo_tx_timeout = hns_nic_net_timeout,
1959         .ndo_set_mac_address = hns_nic_net_set_mac_address,
1960         .ndo_change_mtu = hns_nic_change_mtu,
1961         .ndo_do_ioctl = hns_nic_do_ioctl,
1962         .ndo_set_features = hns_nic_set_features,
1963         .ndo_fix_features = hns_nic_fix_features,
1964         .ndo_get_stats64 = hns_nic_get_stats64,
1965 #ifdef CONFIG_NET_POLL_CONTROLLER
1966         .ndo_poll_controller = hns_nic_poll_controller,
1967 #endif
1968         .ndo_set_rx_mode = hns_nic_set_rx_mode,
1969         .ndo_select_queue = hns_nic_select_queue,
1970 };
1971
1972 static void hns_nic_update_link_status(struct net_device *netdev)
1973 {
1974         struct hns_nic_priv *priv = netdev_priv(netdev);
1975
1976         struct hnae_handle *h = priv->ae_handle;
1977
1978         if (h->phy_dev) {
1979                 if (h->phy_if != PHY_INTERFACE_MODE_XGMII)
1980                         return;
1981
1982                 (void)genphy_read_status(h->phy_dev);
1983         }
1984         hns_nic_adjust_link(netdev);
1985 }
1986
1987 /* for dumping key regs*/
1988 static void hns_nic_dump(struct hns_nic_priv *priv)
1989 {
1990         struct hnae_handle *h = priv->ae_handle;
1991         struct hnae_ae_ops *ops = h->dev->ops;
1992         u32 *data, reg_num, i;
1993
1994         if (ops->get_regs_len && ops->get_regs) {
1995                 reg_num = ops->get_regs_len(priv->ae_handle);
1996                 reg_num = (reg_num + 3ul) & ~3ul;
1997                 data = kcalloc(reg_num, sizeof(u32), GFP_KERNEL);
1998                 if (data) {
1999                         ops->get_regs(priv->ae_handle, data);
2000                         for (i = 0; i < reg_num; i += 4)
2001                                 pr_info("0x%08x: 0x%08x 0x%08x 0x%08x 0x%08x\n",
2002                                         i, data[i], data[i + 1],
2003                                         data[i + 2], data[i + 3]);
2004                         kfree(data);
2005                 }
2006         }
2007
2008         for (i = 0; i < h->q_num; i++) {
2009                 pr_info("tx_queue%d_next_to_clean:%d\n",
2010                         i, h->qs[i]->tx_ring.next_to_clean);
2011                 pr_info("tx_queue%d_next_to_use:%d\n",
2012                         i, h->qs[i]->tx_ring.next_to_use);
2013                 pr_info("rx_queue%d_next_to_clean:%d\n",
2014                         i, h->qs[i]->rx_ring.next_to_clean);
2015                 pr_info("rx_queue%d_next_to_use:%d\n",
2016                         i, h->qs[i]->rx_ring.next_to_use);
2017         }
2018 }
2019
2020 /* for resetting subtask */
2021 static void hns_nic_reset_subtask(struct hns_nic_priv *priv)
2022 {
2023         enum hnae_port_type type = priv->ae_handle->port_type;
2024
2025         if (!test_bit(NIC_STATE2_RESET_REQUESTED, &priv->state))
2026                 return;
2027         clear_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
2028
2029         /* If we're already down, removing or resetting, just bail */
2030         if (test_bit(NIC_STATE_DOWN, &priv->state) ||
2031             test_bit(NIC_STATE_REMOVING, &priv->state) ||
2032             test_bit(NIC_STATE_RESETTING, &priv->state))
2033                 return;
2034
2035         hns_nic_dump(priv);
2036         netdev_info(priv->netdev, "try to reset %s port!\n",
2037                     (type == HNAE_PORT_DEBUG ? "debug" : "service"));
2038
2039         rtnl_lock();
2040         /* put off any impending NetWatchDogTimeout */
2041         netif_trans_update(priv->netdev);
2042         hns_nic_net_reinit(priv->netdev);
2043
2044         rtnl_unlock();
2045 }
2046
2047 /* for doing service complete*/
2048 static void hns_nic_service_event_complete(struct hns_nic_priv *priv)
2049 {
2050         WARN_ON(!test_bit(NIC_STATE_SERVICE_SCHED, &priv->state));
2051         /* make sure to commit the things */
2052         smp_mb__before_atomic();
2053         clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
2054 }
2055
2056 static void hns_nic_service_task(struct work_struct *work)
2057 {
2058         struct hns_nic_priv *priv
2059                 = container_of(work, struct hns_nic_priv, service_task);
2060         struct hnae_handle *h = priv->ae_handle;
2061
2062         hns_nic_update_link_status(priv->netdev);
2063         h->dev->ops->update_led_status(h);
2064         hns_nic_update_stats(priv->netdev);
2065
2066         hns_nic_reset_subtask(priv);
2067         hns_nic_service_event_complete(priv);
2068 }
2069
2070 static void hns_nic_task_schedule(struct hns_nic_priv *priv)
2071 {
2072         if (!test_bit(NIC_STATE_DOWN, &priv->state) &&
2073             !test_bit(NIC_STATE_REMOVING, &priv->state) &&
2074             !test_and_set_bit(NIC_STATE_SERVICE_SCHED, &priv->state))
2075                 (void)schedule_work(&priv->service_task);
2076 }
2077
2078 static void hns_nic_service_timer(struct timer_list *t)
2079 {
2080         struct hns_nic_priv *priv = from_timer(priv, t, service_timer);
2081
2082         (void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
2083
2084         hns_nic_task_schedule(priv);
2085 }
2086
2087 /**
2088  * hns_tx_timeout_reset - initiate reset due to Tx timeout
2089  * @priv: driver private struct
2090  **/
2091 static void hns_tx_timeout_reset(struct hns_nic_priv *priv)
2092 {
2093         /* Do the reset outside of interrupt context */
2094         if (!test_bit(NIC_STATE_DOWN, &priv->state)) {
2095                 set_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
2096                 netdev_warn(priv->netdev,
2097                             "initiating reset due to tx timeout(%llu,0x%lx)\n",
2098                             priv->tx_timeout_count, priv->state);
2099                 priv->tx_timeout_count++;
2100                 hns_nic_task_schedule(priv);
2101         }
2102 }
2103
2104 static int hns_nic_init_ring_data(struct hns_nic_priv *priv)
2105 {
2106         struct hnae_handle *h = priv->ae_handle;
2107         struct hns_nic_ring_data *rd;
2108         bool is_ver1 = AE_IS_VER1(priv->enet_ver);
2109         int i;
2110
2111         if (h->q_num > NIC_MAX_Q_PER_VF) {
2112                 netdev_err(priv->netdev, "too much queue (%d)\n", h->q_num);
2113                 return -EINVAL;
2114         }
2115
2116         priv->ring_data = kzalloc(array3_size(h->q_num,
2117                                               sizeof(*priv->ring_data), 2),
2118                                   GFP_KERNEL);
2119         if (!priv->ring_data)
2120                 return -ENOMEM;
2121
2122         for (i = 0; i < h->q_num; i++) {
2123                 rd = &priv->ring_data[i];
2124                 rd->queue_index = i;
2125                 rd->ring = &h->qs[i]->tx_ring;
2126                 rd->poll_one = hns_nic_tx_poll_one;
2127                 rd->fini_process = is_ver1 ? hns_nic_tx_fini_pro :
2128                         hns_nic_tx_fini_pro_v2;
2129
2130                 netif_napi_add(priv->netdev, &rd->napi,
2131                                hns_nic_common_poll, NIC_TX_CLEAN_MAX_NUM);
2132                 rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
2133         }
2134         for (i = h->q_num; i < h->q_num * 2; i++) {
2135                 rd = &priv->ring_data[i];
2136                 rd->queue_index = i - h->q_num;
2137                 rd->ring = &h->qs[i - h->q_num]->rx_ring;
2138                 rd->poll_one = hns_nic_rx_poll_one;
2139                 rd->ex_process = hns_nic_rx_up_pro;
2140                 rd->fini_process = is_ver1 ? hns_nic_rx_fini_pro :
2141                         hns_nic_rx_fini_pro_v2;
2142
2143                 netif_napi_add(priv->netdev, &rd->napi,
2144                                hns_nic_common_poll, NIC_RX_CLEAN_MAX_NUM);
2145                 rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
2146         }
2147
2148         return 0;
2149 }
2150
2151 static void hns_nic_uninit_ring_data(struct hns_nic_priv *priv)
2152 {
2153         struct hnae_handle *h = priv->ae_handle;
2154         int i;
2155
2156         for (i = 0; i < h->q_num * 2; i++) {
2157                 netif_napi_del(&priv->ring_data[i].napi);
2158                 if (priv->ring_data[i].ring->irq_init_flag == RCB_IRQ_INITED) {
2159                         (void)irq_set_affinity_hint(
2160                                 priv->ring_data[i].ring->irq,
2161                                 NULL);
2162                         free_irq(priv->ring_data[i].ring->irq,
2163                                  &priv->ring_data[i]);
2164                 }
2165
2166                 priv->ring_data[i].ring->irq_init_flag = RCB_IRQ_NOT_INITED;
2167         }
2168         kfree(priv->ring_data);
2169 }
2170
2171 static void hns_nic_set_priv_ops(struct net_device *netdev)
2172 {
2173         struct hns_nic_priv *priv = netdev_priv(netdev);
2174         struct hnae_handle *h = priv->ae_handle;
2175
2176         if (AE_IS_VER1(priv->enet_ver)) {
2177                 priv->ops.fill_desc = fill_desc;
2178                 priv->ops.get_rxd_bnum = get_rx_desc_bnum;
2179                 priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
2180         } else {
2181                 priv->ops.get_rxd_bnum = get_v2rx_desc_bnum;
2182                 if ((netdev->features & NETIF_F_TSO) ||
2183                     (netdev->features & NETIF_F_TSO6)) {
2184                         priv->ops.fill_desc = fill_tso_desc;
2185                         priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tso;
2186                         /* This chip only support 7*4096 */
2187                         netif_set_gso_max_size(netdev, 7 * 4096);
2188                 } else {
2189                         priv->ops.fill_desc = fill_v2_desc;
2190                         priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
2191                 }
2192                 /* enable tso when init
2193                  * control tso on/off through TSE bit in bd
2194                  */
2195                 h->dev->ops->set_tso_stats(h, 1);
2196         }
2197 }
2198
2199 static int hns_nic_try_get_ae(struct net_device *ndev)
2200 {
2201         struct hns_nic_priv *priv = netdev_priv(ndev);
2202         struct hnae_handle *h;
2203         int ret;
2204
2205         h = hnae_get_handle(&priv->netdev->dev,
2206                             priv->fwnode, priv->port_id, NULL);
2207         if (IS_ERR_OR_NULL(h)) {
2208                 ret = -ENODEV;
2209                 dev_dbg(priv->dev, "has not handle, register notifier!\n");
2210                 goto out;
2211         }
2212         priv->ae_handle = h;
2213
2214         ret = hns_nic_init_phy(ndev, h);
2215         if (ret) {
2216                 dev_err(priv->dev, "probe phy device fail!\n");
2217                 goto out_init_phy;
2218         }
2219
2220         ret = hns_nic_init_ring_data(priv);
2221         if (ret) {
2222                 ret = -ENOMEM;
2223                 goto out_init_ring_data;
2224         }
2225
2226         hns_nic_set_priv_ops(ndev);
2227
2228         ret = register_netdev(ndev);
2229         if (ret) {
2230                 dev_err(priv->dev, "probe register netdev fail!\n");
2231                 goto out_reg_ndev_fail;
2232         }
2233         return 0;
2234
2235 out_reg_ndev_fail:
2236         hns_nic_uninit_ring_data(priv);
2237         priv->ring_data = NULL;
2238 out_init_phy:
2239 out_init_ring_data:
2240         hnae_put_handle(priv->ae_handle);
2241         priv->ae_handle = NULL;
2242 out:
2243         return ret;
2244 }
2245
2246 static int hns_nic_notifier_action(struct notifier_block *nb,
2247                                    unsigned long action, void *data)
2248 {
2249         struct hns_nic_priv *priv =
2250                 container_of(nb, struct hns_nic_priv, notifier_block);
2251
2252         assert(action == HNAE_AE_REGISTER);
2253
2254         if (!hns_nic_try_get_ae(priv->netdev)) {
2255                 hnae_unregister_notifier(&priv->notifier_block);
2256                 priv->notifier_block.notifier_call = NULL;
2257         }
2258         return 0;
2259 }
2260
2261 static int hns_nic_dev_probe(struct platform_device *pdev)
2262 {
2263         struct device *dev = &pdev->dev;
2264         struct net_device *ndev;
2265         struct hns_nic_priv *priv;
2266         u32 port_id;
2267         int ret;
2268
2269         ndev = alloc_etherdev_mq(sizeof(struct hns_nic_priv), NIC_MAX_Q_PER_VF);
2270         if (!ndev)
2271                 return -ENOMEM;
2272
2273         platform_set_drvdata(pdev, ndev);
2274
2275         priv = netdev_priv(ndev);
2276         priv->dev = dev;
2277         priv->netdev = ndev;
2278
2279         if (dev_of_node(dev)) {
2280                 struct device_node *ae_node;
2281
2282                 if (of_device_is_compatible(dev->of_node,
2283                                             "hisilicon,hns-nic-v1"))
2284                         priv->enet_ver = AE_VERSION_1;
2285                 else
2286                         priv->enet_ver = AE_VERSION_2;
2287
2288                 ae_node = of_parse_phandle(dev->of_node, "ae-handle", 0);
2289                 if (!ae_node) {
2290                         ret = -ENODEV;
2291                         dev_err(dev, "not find ae-handle\n");
2292                         goto out_read_prop_fail;
2293                 }
2294                 priv->fwnode = &ae_node->fwnode;
2295         } else if (is_acpi_node(dev->fwnode)) {
2296                 struct fwnode_reference_args args;
2297
2298                 if (acpi_dev_found(hns_enet_acpi_match[0].id))
2299                         priv->enet_ver = AE_VERSION_1;
2300                 else if (acpi_dev_found(hns_enet_acpi_match[1].id))
2301                         priv->enet_ver = AE_VERSION_2;
2302                 else
2303                         return -ENXIO;
2304
2305                 /* try to find port-idx-in-ae first */
2306                 ret = acpi_node_get_property_reference(dev->fwnode,
2307                                                        "ae-handle", 0, &args);
2308                 if (ret) {
2309                         dev_err(dev, "not find ae-handle\n");
2310                         goto out_read_prop_fail;
2311                 }
2312                 if (!is_acpi_device_node(args.fwnode)) {
2313                         ret = -EINVAL;
2314                         goto out_read_prop_fail;
2315                 }
2316                 priv->fwnode = args.fwnode;
2317         } else {
2318                 dev_err(dev, "cannot read cfg data from OF or acpi\n");
2319                 return -ENXIO;
2320         }
2321
2322         ret = device_property_read_u32(dev, "port-idx-in-ae", &port_id);
2323         if (ret) {
2324                 /* only for old code compatible */
2325                 ret = device_property_read_u32(dev, "port-id", &port_id);
2326                 if (ret)
2327                         goto out_read_prop_fail;
2328                 /* for old dts, we need to caculate the port offset */
2329                 port_id = port_id < HNS_SRV_OFFSET ? port_id + HNS_DEBUG_OFFSET
2330                         : port_id - HNS_SRV_OFFSET;
2331         }
2332         priv->port_id = port_id;
2333
2334         hns_init_mac_addr(ndev);
2335
2336         ndev->watchdog_timeo = HNS_NIC_TX_TIMEOUT;
2337         ndev->priv_flags |= IFF_UNICAST_FLT;
2338         ndev->netdev_ops = &hns_nic_netdev_ops;
2339         hns_ethtool_set_ops(ndev);
2340
2341         ndev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2342                 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2343                 NETIF_F_GRO;
2344         ndev->vlan_features |=
2345                 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM;
2346         ndev->vlan_features |= NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
2347
2348         /* MTU range: 68 - 9578 (v1) or 9706 (v2) */
2349         ndev->min_mtu = MAC_MIN_MTU;
2350         switch (priv->enet_ver) {
2351         case AE_VERSION_2:
2352                 ndev->features |= NETIF_F_TSO | NETIF_F_TSO6;
2353                 ndev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2354                         NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2355                         NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6;
2356                 ndev->max_mtu = MAC_MAX_MTU_V2 -
2357                                 (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
2358                 break;
2359         default:
2360                 ndev->max_mtu = MAC_MAX_MTU -
2361                                 (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
2362                 break;
2363         }
2364
2365         SET_NETDEV_DEV(ndev, dev);
2366
2367         if (!dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64)))
2368                 dev_dbg(dev, "set mask to 64bit\n");
2369         else
2370                 dev_err(dev, "set mask to 64bit fail!\n");
2371
2372         /* carrier off reporting is important to ethtool even BEFORE open */
2373         netif_carrier_off(ndev);
2374
2375         timer_setup(&priv->service_timer, hns_nic_service_timer, 0);
2376         INIT_WORK(&priv->service_task, hns_nic_service_task);
2377
2378         set_bit(NIC_STATE_SERVICE_INITED, &priv->state);
2379         clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
2380         set_bit(NIC_STATE_DOWN, &priv->state);
2381
2382         if (hns_nic_try_get_ae(priv->netdev)) {
2383                 priv->notifier_block.notifier_call = hns_nic_notifier_action;
2384                 ret = hnae_register_notifier(&priv->notifier_block);
2385                 if (ret) {
2386                         dev_err(dev, "register notifier fail!\n");
2387                         goto out_notify_fail;
2388                 }
2389                 dev_dbg(dev, "has not handle, register notifier!\n");
2390         }
2391
2392         return 0;
2393
2394 out_notify_fail:
2395         (void)cancel_work_sync(&priv->service_task);
2396 out_read_prop_fail:
2397         free_netdev(ndev);
2398         return ret;
2399 }
2400
2401 static int hns_nic_dev_remove(struct platform_device *pdev)
2402 {
2403         struct net_device *ndev = platform_get_drvdata(pdev);
2404         struct hns_nic_priv *priv = netdev_priv(ndev);
2405
2406         if (ndev->reg_state != NETREG_UNINITIALIZED)
2407                 unregister_netdev(ndev);
2408
2409         if (priv->ring_data)
2410                 hns_nic_uninit_ring_data(priv);
2411         priv->ring_data = NULL;
2412
2413         if (ndev->phydev)
2414                 phy_disconnect(ndev->phydev);
2415
2416         if (!IS_ERR_OR_NULL(priv->ae_handle))
2417                 hnae_put_handle(priv->ae_handle);
2418         priv->ae_handle = NULL;
2419         if (priv->notifier_block.notifier_call)
2420                 hnae_unregister_notifier(&priv->notifier_block);
2421         priv->notifier_block.notifier_call = NULL;
2422
2423         set_bit(NIC_STATE_REMOVING, &priv->state);
2424         (void)cancel_work_sync(&priv->service_task);
2425
2426         free_netdev(ndev);
2427         return 0;
2428 }
2429
2430 static const struct of_device_id hns_enet_of_match[] = {
2431         {.compatible = "hisilicon,hns-nic-v1",},
2432         {.compatible = "hisilicon,hns-nic-v2",},
2433         {},
2434 };
2435
2436 MODULE_DEVICE_TABLE(of, hns_enet_of_match);
2437
2438 static struct platform_driver hns_nic_dev_driver = {
2439         .driver = {
2440                 .name = "hns-nic",
2441                 .of_match_table = hns_enet_of_match,
2442                 .acpi_match_table = ACPI_PTR(hns_enet_acpi_match),
2443         },
2444         .probe = hns_nic_dev_probe,
2445         .remove = hns_nic_dev_remove,
2446 };
2447
2448 module_platform_driver(hns_nic_dev_driver);
2449
2450 MODULE_DESCRIPTION("HISILICON HNS Ethernet driver");
2451 MODULE_AUTHOR("Hisilicon, Inc.");
2452 MODULE_LICENSE("GPL");
2453 MODULE_ALIAS("platform:hns-nic");