WIP FPC-III support
[linux/fpc-iii.git] / drivers / net / ethernet / hisilicon / hns / hns_enet.c
blob858cb293152a9a465ed77939431995853e107ed7
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright (c) 2014-2015 Hisilicon Limited.
4 */
6 #include <linux/clk.h>
7 #include <linux/cpumask.h>
8 #include <linux/etherdevice.h>
9 #include <linux/if_vlan.h>
10 #include <linux/interrupt.h>
11 #include <linux/io.h>
12 #include <linux/ip.h>
13 #include <linux/ipv6.h>
14 #include <linux/irq.h>
15 #include <linux/module.h>
16 #include <linux/phy.h>
17 #include <linux/platform_device.h>
18 #include <linux/skbuff.h>
20 #include "hnae.h"
21 #include "hns_enet.h"
22 #include "hns_dsaf_mac.h"
24 #define NIC_MAX_Q_PER_VF 16
25 #define HNS_NIC_TX_TIMEOUT (5 * HZ)
27 #define SERVICE_TIMER_HZ (1 * HZ)
29 #define RCB_IRQ_NOT_INITED 0
30 #define RCB_IRQ_INITED 1
31 #define HNS_BUFFER_SIZE_2048 2048
33 #define BD_MAX_SEND_SIZE 8191
34 #define SKB_TMP_LEN(SKB) \
35 (((SKB)->transport_header - (SKB)->mac_header) + tcp_hdrlen(SKB))
37 static void fill_v2_desc_hw(struct hnae_ring *ring, void *priv, int size,
38 int send_sz, dma_addr_t dma, int frag_end,
39 int buf_num, enum hns_desc_type type, int mtu)
41 struct hnae_desc *desc = &ring->desc[ring->next_to_use];
42 struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
43 struct iphdr *iphdr;
44 struct ipv6hdr *ipv6hdr;
45 struct sk_buff *skb;
46 __be16 protocol;
47 u8 bn_pid = 0;
48 u8 rrcfv = 0;
49 u8 ip_offset = 0;
50 u8 tvsvsn = 0;
51 u16 mss = 0;
52 u8 l4_len = 0;
53 u16 paylen = 0;
55 desc_cb->priv = priv;
56 desc_cb->length = size;
57 desc_cb->dma = dma;
58 desc_cb->type = type;
60 desc->addr = cpu_to_le64(dma);
61 desc->tx.send_size = cpu_to_le16((u16)send_sz);
63 /* config bd buffer end */
64 hnae_set_bit(rrcfv, HNSV2_TXD_VLD_B, 1);
65 hnae_set_field(bn_pid, HNSV2_TXD_BUFNUM_M, 0, buf_num - 1);
67 /* fill port_id in the tx bd for sending management pkts */
68 hnae_set_field(bn_pid, HNSV2_TXD_PORTID_M,
69 HNSV2_TXD_PORTID_S, ring->q->handle->dport_id);
71 if (type == DESC_TYPE_SKB) {
72 skb = (struct sk_buff *)priv;
74 if (skb->ip_summed == CHECKSUM_PARTIAL) {
75 skb_reset_mac_len(skb);
76 protocol = skb->protocol;
77 ip_offset = ETH_HLEN;
79 if (protocol == htons(ETH_P_8021Q)) {
80 ip_offset += VLAN_HLEN;
81 protocol = vlan_get_protocol(skb);
82 skb->protocol = protocol;
85 if (skb->protocol == htons(ETH_P_IP)) {
86 iphdr = ip_hdr(skb);
87 hnae_set_bit(rrcfv, HNSV2_TXD_L3CS_B, 1);
88 hnae_set_bit(rrcfv, HNSV2_TXD_L4CS_B, 1);
90 /* check for tcp/udp header */
91 if (iphdr->protocol == IPPROTO_TCP &&
92 skb_is_gso(skb)) {
93 hnae_set_bit(tvsvsn,
94 HNSV2_TXD_TSE_B, 1);
95 l4_len = tcp_hdrlen(skb);
96 mss = skb_shinfo(skb)->gso_size;
97 paylen = skb->len - SKB_TMP_LEN(skb);
99 } else if (skb->protocol == htons(ETH_P_IPV6)) {
100 hnae_set_bit(tvsvsn, HNSV2_TXD_IPV6_B, 1);
101 ipv6hdr = ipv6_hdr(skb);
102 hnae_set_bit(rrcfv, HNSV2_TXD_L4CS_B, 1);
104 /* check for tcp/udp header */
105 if (ipv6hdr->nexthdr == IPPROTO_TCP &&
106 skb_is_gso(skb) && skb_is_gso_v6(skb)) {
107 hnae_set_bit(tvsvsn,
108 HNSV2_TXD_TSE_B, 1);
109 l4_len = tcp_hdrlen(skb);
110 mss = skb_shinfo(skb)->gso_size;
111 paylen = skb->len - SKB_TMP_LEN(skb);
114 desc->tx.ip_offset = ip_offset;
115 desc->tx.tse_vlan_snap_v6_sctp_nth = tvsvsn;
116 desc->tx.mss = cpu_to_le16(mss);
117 desc->tx.l4_len = l4_len;
118 desc->tx.paylen = cpu_to_le16(paylen);
122 hnae_set_bit(rrcfv, HNSV2_TXD_FE_B, frag_end);
124 desc->tx.bn_pid = bn_pid;
125 desc->tx.ra_ri_cs_fe_vld = rrcfv;
127 ring_ptr_move_fw(ring, next_to_use);
130 static void fill_v2_desc(struct hnae_ring *ring, void *priv,
131 int size, dma_addr_t dma, int frag_end,
132 int buf_num, enum hns_desc_type type, int mtu)
134 fill_v2_desc_hw(ring, priv, size, size, dma, frag_end,
135 buf_num, type, mtu);
138 static const struct acpi_device_id hns_enet_acpi_match[] = {
139 { "HISI00C1", 0 },
140 { "HISI00C2", 0 },
141 { },
143 MODULE_DEVICE_TABLE(acpi, hns_enet_acpi_match);
145 static void fill_desc(struct hnae_ring *ring, void *priv,
146 int size, dma_addr_t dma, int frag_end,
147 int buf_num, enum hns_desc_type type, int mtu)
149 struct hnae_desc *desc = &ring->desc[ring->next_to_use];
150 struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
151 struct sk_buff *skb;
152 __be16 protocol;
153 u32 ip_offset;
154 u32 asid_bufnum_pid = 0;
155 u32 flag_ipoffset = 0;
157 desc_cb->priv = priv;
158 desc_cb->length = size;
159 desc_cb->dma = dma;
160 desc_cb->type = type;
162 desc->addr = cpu_to_le64(dma);
163 desc->tx.send_size = cpu_to_le16((u16)size);
165 /*config bd buffer end */
166 flag_ipoffset |= 1 << HNS_TXD_VLD_B;
168 asid_bufnum_pid |= buf_num << HNS_TXD_BUFNUM_S;
170 if (type == DESC_TYPE_SKB) {
171 skb = (struct sk_buff *)priv;
173 if (skb->ip_summed == CHECKSUM_PARTIAL) {
174 protocol = skb->protocol;
175 ip_offset = ETH_HLEN;
177 /*if it is a SW VLAN check the next protocol*/
178 if (protocol == htons(ETH_P_8021Q)) {
179 ip_offset += VLAN_HLEN;
180 protocol = vlan_get_protocol(skb);
181 skb->protocol = protocol;
184 if (skb->protocol == htons(ETH_P_IP)) {
185 flag_ipoffset |= 1 << HNS_TXD_L3CS_B;
186 /* check for tcp/udp header */
187 flag_ipoffset |= 1 << HNS_TXD_L4CS_B;
189 } else if (skb->protocol == htons(ETH_P_IPV6)) {
190 /* ipv6 has not l3 cs, check for L4 header */
191 flag_ipoffset |= 1 << HNS_TXD_L4CS_B;
194 flag_ipoffset |= ip_offset << HNS_TXD_IPOFFSET_S;
198 flag_ipoffset |= frag_end << HNS_TXD_FE_B;
200 desc->tx.asid_bufnum_pid = cpu_to_le16(asid_bufnum_pid);
201 desc->tx.flag_ipoffset = cpu_to_le32(flag_ipoffset);
203 ring_ptr_move_fw(ring, next_to_use);
206 static void unfill_desc(struct hnae_ring *ring)
208 ring_ptr_move_bw(ring, next_to_use);
211 static int hns_nic_maybe_stop_tx(
212 struct sk_buff **out_skb, int *bnum, struct hnae_ring *ring)
214 struct sk_buff *skb = *out_skb;
215 struct sk_buff *new_skb = NULL;
216 int buf_num;
218 /* no. of segments (plus a header) */
219 buf_num = skb_shinfo(skb)->nr_frags + 1;
221 if (unlikely(buf_num > ring->max_desc_num_per_pkt)) {
222 if (ring_space(ring) < 1)
223 return -EBUSY;
225 new_skb = skb_copy(skb, GFP_ATOMIC);
226 if (!new_skb)
227 return -ENOMEM;
229 dev_kfree_skb_any(skb);
230 *out_skb = new_skb;
231 buf_num = 1;
232 } else if (buf_num > ring_space(ring)) {
233 return -EBUSY;
236 *bnum = buf_num;
237 return 0;
240 static int hns_nic_maybe_stop_tso(
241 struct sk_buff **out_skb, int *bnum, struct hnae_ring *ring)
243 int i;
244 int size;
245 int buf_num;
246 int frag_num;
247 struct sk_buff *skb = *out_skb;
248 struct sk_buff *new_skb = NULL;
249 skb_frag_t *frag;
251 size = skb_headlen(skb);
252 buf_num = (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
254 frag_num = skb_shinfo(skb)->nr_frags;
255 for (i = 0; i < frag_num; i++) {
256 frag = &skb_shinfo(skb)->frags[i];
257 size = skb_frag_size(frag);
258 buf_num += (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
261 if (unlikely(buf_num > ring->max_desc_num_per_pkt)) {
262 buf_num = (skb->len + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
263 if (ring_space(ring) < buf_num)
264 return -EBUSY;
265 /* manual split the send packet */
266 new_skb = skb_copy(skb, GFP_ATOMIC);
267 if (!new_skb)
268 return -ENOMEM;
269 dev_kfree_skb_any(skb);
270 *out_skb = new_skb;
272 } else if (ring_space(ring) < buf_num) {
273 return -EBUSY;
276 *bnum = buf_num;
277 return 0;
280 static void fill_tso_desc(struct hnae_ring *ring, void *priv,
281 int size, dma_addr_t dma, int frag_end,
282 int buf_num, enum hns_desc_type type, int mtu)
284 int frag_buf_num;
285 int sizeoflast;
286 int k;
288 frag_buf_num = (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
289 sizeoflast = size % BD_MAX_SEND_SIZE;
290 sizeoflast = sizeoflast ? sizeoflast : BD_MAX_SEND_SIZE;
292 /* when the frag size is bigger than hardware, split this frag */
293 for (k = 0; k < frag_buf_num; k++)
294 fill_v2_desc_hw(ring, priv, k == 0 ? size : 0,
295 (k == frag_buf_num - 1) ?
296 sizeoflast : BD_MAX_SEND_SIZE,
297 dma + BD_MAX_SEND_SIZE * k,
298 frag_end && (k == frag_buf_num - 1) ? 1 : 0,
299 buf_num,
300 (type == DESC_TYPE_SKB && !k) ?
301 DESC_TYPE_SKB : DESC_TYPE_PAGE,
302 mtu);
305 netdev_tx_t hns_nic_net_xmit_hw(struct net_device *ndev,
306 struct sk_buff *skb,
307 struct hns_nic_ring_data *ring_data)
309 struct hns_nic_priv *priv = netdev_priv(ndev);
310 struct hnae_ring *ring = ring_data->ring;
311 struct device *dev = ring_to_dev(ring);
312 struct netdev_queue *dev_queue;
313 skb_frag_t *frag;
314 int buf_num;
315 int seg_num;
316 dma_addr_t dma;
317 int size, next_to_use;
318 int i;
320 switch (priv->ops.maybe_stop_tx(&skb, &buf_num, ring)) {
321 case -EBUSY:
322 ring->stats.tx_busy++;
323 goto out_net_tx_busy;
324 case -ENOMEM:
325 ring->stats.sw_err_cnt++;
326 netdev_err(ndev, "no memory to xmit!\n");
327 goto out_err_tx_ok;
328 default:
329 break;
332 /* no. of segments (plus a header) */
333 seg_num = skb_shinfo(skb)->nr_frags + 1;
334 next_to_use = ring->next_to_use;
336 /* fill the first part */
337 size = skb_headlen(skb);
338 dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
339 if (dma_mapping_error(dev, dma)) {
340 netdev_err(ndev, "TX head DMA map failed\n");
341 ring->stats.sw_err_cnt++;
342 goto out_err_tx_ok;
344 priv->ops.fill_desc(ring, skb, size, dma, seg_num == 1 ? 1 : 0,
345 buf_num, DESC_TYPE_SKB, ndev->mtu);
347 /* fill the fragments */
348 for (i = 1; i < seg_num; i++) {
349 frag = &skb_shinfo(skb)->frags[i - 1];
350 size = skb_frag_size(frag);
351 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
352 if (dma_mapping_error(dev, dma)) {
353 netdev_err(ndev, "TX frag(%d) DMA map failed\n", i);
354 ring->stats.sw_err_cnt++;
355 goto out_map_frag_fail;
357 priv->ops.fill_desc(ring, skb_frag_page(frag), size, dma,
358 seg_num - 1 == i ? 1 : 0, buf_num,
359 DESC_TYPE_PAGE, ndev->mtu);
362 /*complete translate all packets*/
363 dev_queue = netdev_get_tx_queue(ndev, skb->queue_mapping);
364 netdev_tx_sent_queue(dev_queue, skb->len);
366 netif_trans_update(ndev);
367 ndev->stats.tx_bytes += skb->len;
368 ndev->stats.tx_packets++;
370 wmb(); /* commit all data before submit */
371 assert(skb->queue_mapping < priv->ae_handle->q_num);
372 hnae_queue_xmit(priv->ae_handle->qs[skb->queue_mapping], buf_num);
374 return NETDEV_TX_OK;
376 out_map_frag_fail:
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);
392 out_err_tx_ok:
394 dev_kfree_skb_any(skb);
395 return NETDEV_TX_OK;
397 out_net_tx_busy:
399 netif_stop_subqueue(ndev, skb->queue_mapping);
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.
405 smp_mb();
406 return NETDEV_TX_BUSY;
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)
413 struct hnae_desc *desc;
414 u32 truesize;
415 int size;
416 int last_offset;
417 bool twobufs;
419 twobufs = ((PAGE_SIZE < 8192) &&
420 hnae_buf_size(ring) == HNS_BUFFER_SIZE_2048);
422 desc = &ring->desc[ring->next_to_clean];
423 size = le16_to_cpu(desc->rx.size);
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);
432 skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
433 size - pull_len, truesize);
435 /* avoid re-using remote pages,flag default unreuse */
436 if (unlikely(page_to_nid(desc_cb->priv) != numa_node_id()))
437 return;
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;
445 desc_cb->reuse_flag = 1;
446 /* bump ref count on page before it is given*/
447 get_page(desc_cb->priv);
449 return;
452 /* move offset up to the next cache line */
453 desc_cb->page_offset += truesize;
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);
462 static void get_v2rx_desc_bnum(u32 bnum_flag, int *out_bnum)
464 *out_bnum = hnae_get_field(bnum_flag,
465 HNS_RXD_BUFNUM_M, HNS_RXD_BUFNUM_S) + 1;
468 static void get_rx_desc_bnum(u32 bnum_flag, int *out_bnum)
470 *out_bnum = hnae_get_field(bnum_flag,
471 HNS_RXD_BUFNUM_M, HNS_RXD_BUFNUM_S);
474 static void hns_nic_rx_checksum(struct hns_nic_ring_data *ring_data,
475 struct sk_buff *skb, u32 flag)
477 struct net_device *netdev = ring_data->napi.dev;
478 u32 l3id;
479 u32 l4id;
481 /* check if RX checksum offload is enabled */
482 if (unlikely(!(netdev->features & NETIF_F_RXCSUM)))
483 return;
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.
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).
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.
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);
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;
517 /* check for any(not just checksum)flagged L3 protocol errors */
518 if (unlikely(hnae_get_bit(flag, HNS_RXD_L3E_B)))
519 return;
521 /* we do not support checksum of fragmented packets */
522 if (unlikely(hnae_get_bit(flag, HNS_RXD_FRAG_B)))
523 return;
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;
531 /* check for any(not just checksum)flagged L4 protocol errors */
532 if (unlikely(hnae_get_bit(flag, HNS_RXD_L4E_B)))
533 return;
535 /* now, this has to be a packet with valid RX checksum */
536 skb->ip_summed = CHECKSUM_UNNECESSARY;
539 static int hns_nic_poll_rx_skb(struct hns_nic_ring_data *ring_data,
540 struct sk_buff **out_skb, int *out_bnum)
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;
553 desc = &ring->desc[ring->next_to_clean];
554 desc_cb = &ring->desc_cb[ring->next_to_clean];
556 prefetch(desc);
558 va = (unsigned char *)desc_cb->buf + desc_cb->page_offset;
560 /* prefetch first cache line of first page */
561 net_prefetch(va);
563 skb = *out_skb = napi_alloc_skb(&ring_data->napi,
564 HNS_RX_HEAD_SIZE);
565 if (unlikely(!skb)) {
566 ring->stats.sw_err_cnt++;
567 return -ENOMEM;
570 prefetchw(skb->data);
571 length = le16_to_cpu(desc->rx.pkt_len);
572 bnum_flag = le32_to_cpu(desc->rx.ipoff_bnum_pid_flag);
573 priv->ops.get_rxd_bnum(bnum_flag, &bnum);
574 *out_bnum = bnum;
576 if (length <= HNS_RX_HEAD_SIZE) {
577 memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
579 /* we can reuse buffer as-is, just make sure it is local */
580 if (likely(page_to_nid(desc_cb->priv) == numa_node_id()))
581 desc_cb->reuse_flag = 1;
582 else /* this page cannot be reused so discard it */
583 put_page(desc_cb->priv);
585 ring_ptr_move_fw(ring, next_to_clean);
587 if (unlikely(bnum != 1)) { /* check err*/
588 *out_bnum = 1;
589 goto out_bnum_err;
591 } else {
592 ring->stats.seg_pkt_cnt++;
594 pull_len = eth_get_headlen(ndev, va, HNS_RX_HEAD_SIZE);
595 memcpy(__skb_put(skb, pull_len), va,
596 ALIGN(pull_len, sizeof(long)));
598 hns_nic_reuse_page(skb, 0, ring, pull_len, desc_cb);
599 ring_ptr_move_fw(ring, next_to_clean);
601 if (unlikely(bnum >= (int)MAX_SKB_FRAGS)) { /* check err*/
602 *out_bnum = 1;
603 goto out_bnum_err;
605 for (i = 1; i < bnum; i++) {
606 desc = &ring->desc[ring->next_to_clean];
607 desc_cb = &ring->desc_cb[ring->next_to_clean];
609 hns_nic_reuse_page(skb, i, ring, 0, desc_cb);
610 ring_ptr_move_fw(ring, next_to_clean);
614 /* check except process, free skb and jump the desc */
615 if (unlikely((!bnum) || (bnum > ring->max_desc_num_per_pkt))) {
616 out_bnum_err:
617 *out_bnum = *out_bnum ? *out_bnum : 1; /* ntc moved,cannot 0*/
618 netdev_err(ndev, "invalid bnum(%d,%d,%d,%d),%016llx,%016llx\n",
619 bnum, ring->max_desc_num_per_pkt,
620 length, (int)MAX_SKB_FRAGS,
621 ((u64 *)desc)[0], ((u64 *)desc)[1]);
622 ring->stats.err_bd_num++;
623 dev_kfree_skb_any(skb);
624 return -EDOM;
627 bnum_flag = le32_to_cpu(desc->rx.ipoff_bnum_pid_flag);
629 if (unlikely(!hnae_get_bit(bnum_flag, HNS_RXD_VLD_B))) {
630 netdev_err(ndev, "no valid bd,%016llx,%016llx\n",
631 ((u64 *)desc)[0], ((u64 *)desc)[1]);
632 ring->stats.non_vld_descs++;
633 dev_kfree_skb_any(skb);
634 return -EINVAL;
637 if (unlikely((!desc->rx.pkt_len) ||
638 hnae_get_bit(bnum_flag, HNS_RXD_DROP_B))) {
639 ring->stats.err_pkt_len++;
640 dev_kfree_skb_any(skb);
641 return -EFAULT;
644 if (unlikely(hnae_get_bit(bnum_flag, HNS_RXD_L2E_B))) {
645 ring->stats.l2_err++;
646 dev_kfree_skb_any(skb);
647 return -EFAULT;
650 ring->stats.rx_pkts++;
651 ring->stats.rx_bytes += skb->len;
653 /* indicate to upper stack if our hardware has already calculated
654 * the RX checksum
656 hns_nic_rx_checksum(ring_data, skb, bnum_flag);
658 return 0;
661 static void
662 hns_nic_alloc_rx_buffers(struct hns_nic_ring_data *ring_data, int cleand_count)
664 int i, ret;
665 struct hnae_desc_cb res_cbs;
666 struct hnae_desc_cb *desc_cb;
667 struct hnae_ring *ring = ring_data->ring;
668 struct net_device *ndev = ring_data->napi.dev;
670 for (i = 0; i < cleand_count; i++) {
671 desc_cb = &ring->desc_cb[ring->next_to_use];
672 if (desc_cb->reuse_flag) {
673 ring->stats.reuse_pg_cnt++;
674 hnae_reuse_buffer(ring, ring->next_to_use);
675 } else {
676 ret = hnae_reserve_buffer_map(ring, &res_cbs);
677 if (ret) {
678 ring->stats.sw_err_cnt++;
679 netdev_err(ndev, "hnae reserve buffer map failed.\n");
680 break;
682 hnae_replace_buffer(ring, ring->next_to_use, &res_cbs);
685 ring_ptr_move_fw(ring, next_to_use);
688 wmb(); /* make all data has been write before submit */
689 writel_relaxed(i, ring->io_base + RCB_REG_HEAD);
692 /* return error number for error or number of desc left to take
694 static void hns_nic_rx_up_pro(struct hns_nic_ring_data *ring_data,
695 struct sk_buff *skb)
697 struct net_device *ndev = ring_data->napi.dev;
699 skb->protocol = eth_type_trans(skb, ndev);
700 napi_gro_receive(&ring_data->napi, skb);
703 static int hns_desc_unused(struct hnae_ring *ring)
705 int ntc = ring->next_to_clean;
706 int ntu = ring->next_to_use;
708 return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
711 #define HNS_LOWEST_LATENCY_RATE 27 /* 27 MB/s */
712 #define HNS_LOW_LATENCY_RATE 80 /* 80 MB/s */
714 #define HNS_COAL_BDNUM 3
716 static u32 hns_coal_rx_bdnum(struct hnae_ring *ring)
718 bool coal_enable = ring->q->handle->coal_adapt_en;
720 if (coal_enable &&
721 ring->coal_last_rx_bytes > HNS_LOWEST_LATENCY_RATE)
722 return HNS_COAL_BDNUM;
723 else
724 return 0;
727 static void hns_update_rx_rate(struct hnae_ring *ring)
729 bool coal_enable = ring->q->handle->coal_adapt_en;
730 u32 time_passed_ms;
731 u64 total_bytes;
733 if (!coal_enable ||
734 time_before(jiffies, ring->coal_last_jiffies + (HZ >> 4)))
735 return;
737 /* ring->stats.rx_bytes overflowed */
738 if (ring->coal_last_rx_bytes > ring->stats.rx_bytes) {
739 ring->coal_last_rx_bytes = ring->stats.rx_bytes;
740 ring->coal_last_jiffies = jiffies;
741 return;
744 total_bytes = ring->stats.rx_bytes - ring->coal_last_rx_bytes;
745 time_passed_ms = jiffies_to_msecs(jiffies - ring->coal_last_jiffies);
746 do_div(total_bytes, time_passed_ms);
747 ring->coal_rx_rate = total_bytes >> 10;
749 ring->coal_last_rx_bytes = ring->stats.rx_bytes;
750 ring->coal_last_jiffies = jiffies;
754 * smooth_alg - smoothing algrithm for adjusting coalesce parameter
755 * @new_param: new value
756 * @old_param: old value
758 static u32 smooth_alg(u32 new_param, u32 old_param)
760 u32 gap = (new_param > old_param) ? new_param - old_param
761 : old_param - new_param;
763 if (gap > 8)
764 gap >>= 3;
766 if (new_param > old_param)
767 return old_param + gap;
768 else
769 return old_param - gap;
773 * hns_nic_adp_coalesce - self adapte coalesce according to rx rate
774 * @ring_data: pointer to hns_nic_ring_data
776 static void hns_nic_adpt_coalesce(struct hns_nic_ring_data *ring_data)
778 struct hnae_ring *ring = ring_data->ring;
779 struct hnae_handle *handle = ring->q->handle;
780 u32 new_coal_param, old_coal_param = ring->coal_param;
782 if (ring->coal_rx_rate < HNS_LOWEST_LATENCY_RATE)
783 new_coal_param = HNAE_LOWEST_LATENCY_COAL_PARAM;
784 else if (ring->coal_rx_rate < HNS_LOW_LATENCY_RATE)
785 new_coal_param = HNAE_LOW_LATENCY_COAL_PARAM;
786 else
787 new_coal_param = HNAE_BULK_LATENCY_COAL_PARAM;
789 if (new_coal_param == old_coal_param &&
790 new_coal_param == handle->coal_param)
791 return;
793 new_coal_param = smooth_alg(new_coal_param, old_coal_param);
794 ring->coal_param = new_coal_param;
797 * Because all ring in one port has one coalesce param, when one ring
798 * calculate its own coalesce param, it cannot write to hardware at
799 * once. There are three conditions as follows:
800 * 1. current ring's coalesce param is larger than the hardware.
801 * 2. or ring which adapt last time can change again.
802 * 3. timeout.
804 if (new_coal_param == handle->coal_param) {
805 handle->coal_last_jiffies = jiffies;
806 handle->coal_ring_idx = ring_data->queue_index;
807 } else if (new_coal_param > handle->coal_param ||
808 handle->coal_ring_idx == ring_data->queue_index ||
809 time_after(jiffies, handle->coal_last_jiffies + (HZ >> 4))) {
810 handle->dev->ops->set_coalesce_usecs(handle,
811 new_coal_param);
812 handle->dev->ops->set_coalesce_frames(handle,
813 1, new_coal_param);
814 handle->coal_param = new_coal_param;
815 handle->coal_ring_idx = ring_data->queue_index;
816 handle->coal_last_jiffies = jiffies;
820 static int hns_nic_rx_poll_one(struct hns_nic_ring_data *ring_data,
821 int budget, void *v)
823 struct hnae_ring *ring = ring_data->ring;
824 struct sk_buff *skb;
825 int num, bnum;
826 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
827 int recv_pkts, recv_bds, clean_count, err;
828 int unused_count = hns_desc_unused(ring);
830 num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
831 rmb(); /* make sure num taken effect before the other data is touched */
833 recv_pkts = 0, recv_bds = 0, clean_count = 0;
834 num -= unused_count;
836 while (recv_pkts < budget && recv_bds < num) {
837 /* reuse or realloc buffers */
838 if (clean_count + unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
839 hns_nic_alloc_rx_buffers(ring_data,
840 clean_count + unused_count);
841 clean_count = 0;
842 unused_count = hns_desc_unused(ring);
845 /* poll one pkt */
846 err = hns_nic_poll_rx_skb(ring_data, &skb, &bnum);
847 if (unlikely(!skb)) /* this fault cannot be repaired */
848 goto out;
850 recv_bds += bnum;
851 clean_count += bnum;
852 if (unlikely(err)) { /* do jump the err */
853 recv_pkts++;
854 continue;
857 /* do update ip stack process*/
858 ((void (*)(struct hns_nic_ring_data *, struct sk_buff *))v)(
859 ring_data, skb);
860 recv_pkts++;
863 out:
864 /* make all data has been write before submit */
865 if (clean_count + unused_count > 0)
866 hns_nic_alloc_rx_buffers(ring_data,
867 clean_count + unused_count);
869 return recv_pkts;
872 static bool hns_nic_rx_fini_pro(struct hns_nic_ring_data *ring_data)
874 struct hnae_ring *ring = ring_data->ring;
875 int num = 0;
876 bool rx_stopped;
878 hns_update_rx_rate(ring);
880 /* for hardware bug fixed */
881 ring_data->ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
882 num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
884 if (num <= hns_coal_rx_bdnum(ring)) {
885 if (ring->q->handle->coal_adapt_en)
886 hns_nic_adpt_coalesce(ring_data);
888 rx_stopped = true;
889 } else {
890 ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
891 ring_data->ring, 1);
893 rx_stopped = false;
896 return rx_stopped;
899 static bool hns_nic_rx_fini_pro_v2(struct hns_nic_ring_data *ring_data)
901 struct hnae_ring *ring = ring_data->ring;
902 int num;
904 hns_update_rx_rate(ring);
905 num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
907 if (num <= hns_coal_rx_bdnum(ring)) {
908 if (ring->q->handle->coal_adapt_en)
909 hns_nic_adpt_coalesce(ring_data);
911 return true;
914 return false;
917 static inline void hns_nic_reclaim_one_desc(struct hnae_ring *ring,
918 int *bytes, int *pkts)
920 struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
922 (*pkts) += (desc_cb->type == DESC_TYPE_SKB);
923 (*bytes) += desc_cb->length;
924 /* desc_cb will be cleaned, after hnae_free_buffer_detach*/
925 hnae_free_buffer_detach(ring, ring->next_to_clean);
927 ring_ptr_move_fw(ring, next_to_clean);
930 static int is_valid_clean_head(struct hnae_ring *ring, int h)
932 int u = ring->next_to_use;
933 int c = ring->next_to_clean;
935 if (unlikely(h > ring->desc_num))
936 return 0;
938 assert(u > 0 && u < ring->desc_num);
939 assert(c > 0 && c < ring->desc_num);
940 assert(u != c && h != c); /* must be checked before call this func */
942 return u > c ? (h > c && h <= u) : (h > c || h <= u);
945 /* reclaim all desc in one budget
946 * return error or number of desc left
948 static int hns_nic_tx_poll_one(struct hns_nic_ring_data *ring_data,
949 int budget, void *v)
951 struct hnae_ring *ring = ring_data->ring;
952 struct net_device *ndev = ring_data->napi.dev;
953 struct netdev_queue *dev_queue;
954 struct hns_nic_priv *priv = netdev_priv(ndev);
955 int head;
956 int bytes, pkts;
958 head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
959 rmb(); /* make sure head is ready before touch any data */
961 if (is_ring_empty(ring) || head == ring->next_to_clean)
962 return 0; /* no data to poll */
964 if (!is_valid_clean_head(ring, head)) {
965 netdev_err(ndev, "wrong head (%d, %d-%d)\n", head,
966 ring->next_to_use, ring->next_to_clean);
967 ring->stats.io_err_cnt++;
968 return -EIO;
971 bytes = 0;
972 pkts = 0;
973 while (head != ring->next_to_clean) {
974 hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
975 /* issue prefetch for next Tx descriptor */
976 prefetch(&ring->desc_cb[ring->next_to_clean]);
978 /* update tx ring statistics. */
979 ring->stats.tx_pkts += pkts;
980 ring->stats.tx_bytes += bytes;
982 dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
983 netdev_tx_completed_queue(dev_queue, pkts, bytes);
985 if (unlikely(priv->link && !netif_carrier_ok(ndev)))
986 netif_carrier_on(ndev);
988 if (unlikely(pkts && netif_carrier_ok(ndev) &&
989 (ring_space(ring) >= ring->max_desc_num_per_pkt * 2))) {
990 /* Make sure that anybody stopping the queue after this
991 * sees the new next_to_clean.
993 smp_mb();
994 if (netif_tx_queue_stopped(dev_queue) &&
995 !test_bit(NIC_STATE_DOWN, &priv->state)) {
996 netif_tx_wake_queue(dev_queue);
997 ring->stats.restart_queue++;
1000 return 0;
1003 static bool hns_nic_tx_fini_pro(struct hns_nic_ring_data *ring_data)
1005 struct hnae_ring *ring = ring_data->ring;
1006 int head;
1008 ring_data->ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
1010 head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1012 if (head != ring->next_to_clean) {
1013 ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1014 ring_data->ring, 1);
1016 return false;
1017 } else {
1018 return true;
1022 static bool hns_nic_tx_fini_pro_v2(struct hns_nic_ring_data *ring_data)
1024 struct hnae_ring *ring = ring_data->ring;
1025 int head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1027 if (head == ring->next_to_clean)
1028 return true;
1029 else
1030 return false;
1033 static void hns_nic_tx_clr_all_bufs(struct hns_nic_ring_data *ring_data)
1035 struct hnae_ring *ring = ring_data->ring;
1036 struct net_device *ndev = ring_data->napi.dev;
1037 struct netdev_queue *dev_queue;
1038 int head;
1039 int bytes, pkts;
1041 head = ring->next_to_use; /* ntu :soft setted ring position*/
1042 bytes = 0;
1043 pkts = 0;
1044 while (head != ring->next_to_clean)
1045 hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
1047 dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
1048 netdev_tx_reset_queue(dev_queue);
1051 static int hns_nic_common_poll(struct napi_struct *napi, int budget)
1053 int clean_complete = 0;
1054 struct hns_nic_ring_data *ring_data =
1055 container_of(napi, struct hns_nic_ring_data, napi);
1056 struct hnae_ring *ring = ring_data->ring;
1058 clean_complete += ring_data->poll_one(
1059 ring_data, budget - clean_complete,
1060 ring_data->ex_process);
1062 if (clean_complete < budget) {
1063 if (ring_data->fini_process(ring_data)) {
1064 napi_complete(napi);
1065 ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
1066 } else {
1067 return budget;
1071 return clean_complete;
1074 static irqreturn_t hns_irq_handle(int irq, void *dev)
1076 struct hns_nic_ring_data *ring_data = (struct hns_nic_ring_data *)dev;
1078 ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1079 ring_data->ring, 1);
1080 napi_schedule(&ring_data->napi);
1082 return IRQ_HANDLED;
1086 *hns_nic_adjust_link - adjust net work mode by the phy stat or new param
1087 *@ndev: net device
1089 static void hns_nic_adjust_link(struct net_device *ndev)
1091 struct hns_nic_priv *priv = netdev_priv(ndev);
1092 struct hnae_handle *h = priv->ae_handle;
1093 int state = 1;
1095 /* If there is no phy, do not need adjust link */
1096 if (ndev->phydev) {
1097 /* When phy link down, do nothing */
1098 if (ndev->phydev->link == 0)
1099 return;
1101 if (h->dev->ops->need_adjust_link(h, ndev->phydev->speed,
1102 ndev->phydev->duplex)) {
1103 /* because Hi161X chip don't support to change gmac
1104 * speed and duplex with traffic. Delay 200ms to
1105 * make sure there is no more data in chip FIFO.
1107 netif_carrier_off(ndev);
1108 msleep(200);
1109 h->dev->ops->adjust_link(h, ndev->phydev->speed,
1110 ndev->phydev->duplex);
1111 netif_carrier_on(ndev);
1115 state = state && h->dev->ops->get_status(h);
1117 if (state != priv->link) {
1118 if (state) {
1119 netif_carrier_on(ndev);
1120 netif_tx_wake_all_queues(ndev);
1121 netdev_info(ndev, "link up\n");
1122 } else {
1123 netif_carrier_off(ndev);
1124 netdev_info(ndev, "link down\n");
1126 priv->link = state;
1131 *hns_nic_init_phy - init phy
1132 *@ndev: net device
1133 *@h: ae handle
1134 * Return 0 on success, negative on failure
1136 int hns_nic_init_phy(struct net_device *ndev, struct hnae_handle *h)
1138 __ETHTOOL_DECLARE_LINK_MODE_MASK(supported) = { 0, };
1139 struct phy_device *phy_dev = h->phy_dev;
1140 int ret;
1142 if (!h->phy_dev)
1143 return 0;
1145 ethtool_convert_legacy_u32_to_link_mode(supported, h->if_support);
1146 linkmode_and(phy_dev->supported, phy_dev->supported, supported);
1147 linkmode_copy(phy_dev->advertising, phy_dev->supported);
1149 if (h->phy_if == PHY_INTERFACE_MODE_XGMII)
1150 phy_dev->autoneg = false;
1152 if (h->phy_if != PHY_INTERFACE_MODE_XGMII) {
1153 phy_dev->dev_flags = 0;
1155 ret = phy_connect_direct(ndev, phy_dev, hns_nic_adjust_link,
1156 h->phy_if);
1157 } else {
1158 ret = phy_attach_direct(ndev, phy_dev, 0, h->phy_if);
1160 if (unlikely(ret))
1161 return -ENODEV;
1163 phy_attached_info(phy_dev);
1165 return 0;
1168 static int hns_nic_ring_open(struct net_device *netdev, int idx)
1170 struct hns_nic_priv *priv = netdev_priv(netdev);
1171 struct hnae_handle *h = priv->ae_handle;
1173 napi_enable(&priv->ring_data[idx].napi);
1175 enable_irq(priv->ring_data[idx].ring->irq);
1176 h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 0);
1178 return 0;
1181 static int hns_nic_net_set_mac_address(struct net_device *ndev, void *p)
1183 struct hns_nic_priv *priv = netdev_priv(ndev);
1184 struct hnae_handle *h = priv->ae_handle;
1185 struct sockaddr *mac_addr = p;
1186 int ret;
1188 if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1189 return -EADDRNOTAVAIL;
1191 ret = h->dev->ops->set_mac_addr(h, mac_addr->sa_data);
1192 if (ret) {
1193 netdev_err(ndev, "set_mac_address fail, ret=%d!\n", ret);
1194 return ret;
1197 memcpy(ndev->dev_addr, mac_addr->sa_data, ndev->addr_len);
1199 return 0;
1202 static void hns_nic_update_stats(struct net_device *netdev)
1204 struct hns_nic_priv *priv = netdev_priv(netdev);
1205 struct hnae_handle *h = priv->ae_handle;
1207 h->dev->ops->update_stats(h, &netdev->stats);
1210 /* set mac addr if it is configed. or leave it to the AE driver */
1211 static void hns_init_mac_addr(struct net_device *ndev)
1213 struct hns_nic_priv *priv = netdev_priv(ndev);
1215 if (!device_get_mac_address(priv->dev, ndev->dev_addr, ETH_ALEN)) {
1216 eth_hw_addr_random(ndev);
1217 dev_warn(priv->dev, "No valid mac, use random mac %pM",
1218 ndev->dev_addr);
1222 static void hns_nic_ring_close(struct net_device *netdev, int idx)
1224 struct hns_nic_priv *priv = netdev_priv(netdev);
1225 struct hnae_handle *h = priv->ae_handle;
1227 h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 1);
1228 disable_irq(priv->ring_data[idx].ring->irq);
1230 napi_disable(&priv->ring_data[idx].napi);
1233 static int hns_nic_init_affinity_mask(int q_num, int ring_idx,
1234 struct hnae_ring *ring, cpumask_t *mask)
1236 int cpu;
1238 /* Diffrent irq banlance between 16core and 32core.
1239 * The cpu mask set by ring index according to the ring flag
1240 * which indicate the ring is tx or rx.
1242 if (q_num == num_possible_cpus()) {
1243 if (is_tx_ring(ring))
1244 cpu = ring_idx;
1245 else
1246 cpu = ring_idx - q_num;
1247 } else {
1248 if (is_tx_ring(ring))
1249 cpu = ring_idx * 2;
1250 else
1251 cpu = (ring_idx - q_num) * 2 + 1;
1254 cpumask_clear(mask);
1255 cpumask_set_cpu(cpu, mask);
1257 return cpu;
1260 static void hns_nic_free_irq(int q_num, struct hns_nic_priv *priv)
1262 int i;
1264 for (i = 0; i < q_num * 2; i++) {
1265 if (priv->ring_data[i].ring->irq_init_flag == RCB_IRQ_INITED) {
1266 irq_set_affinity_hint(priv->ring_data[i].ring->irq,
1267 NULL);
1268 free_irq(priv->ring_data[i].ring->irq,
1269 &priv->ring_data[i]);
1270 priv->ring_data[i].ring->irq_init_flag =
1271 RCB_IRQ_NOT_INITED;
1276 static int hns_nic_init_irq(struct hns_nic_priv *priv)
1278 struct hnae_handle *h = priv->ae_handle;
1279 struct hns_nic_ring_data *rd;
1280 int i;
1281 int ret;
1282 int cpu;
1284 for (i = 0; i < h->q_num * 2; i++) {
1285 rd = &priv->ring_data[i];
1287 if (rd->ring->irq_init_flag == RCB_IRQ_INITED)
1288 break;
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);
1294 rd->ring->ring_name[RCB_RING_NAME_LEN - 1] = '\0';
1296 irq_set_status_flags(rd->ring->irq, IRQ_NOAUTOEN);
1297 ret = request_irq(rd->ring->irq,
1298 hns_irq_handle, 0, rd->ring->ring_name, rd);
1299 if (ret) {
1300 netdev_err(priv->netdev, "request irq(%d) fail\n",
1301 rd->ring->irq);
1302 goto out_free_irq;
1305 cpu = hns_nic_init_affinity_mask(h->q_num, i,
1306 rd->ring, &rd->mask);
1308 if (cpu_online(cpu))
1309 irq_set_affinity_hint(rd->ring->irq,
1310 &rd->mask);
1312 rd->ring->irq_init_flag = RCB_IRQ_INITED;
1315 return 0;
1317 out_free_irq:
1318 hns_nic_free_irq(h->q_num, priv);
1319 return ret;
1322 static int hns_nic_net_up(struct net_device *ndev)
1324 struct hns_nic_priv *priv = netdev_priv(ndev);
1325 struct hnae_handle *h = priv->ae_handle;
1326 int i, j;
1327 int ret;
1329 if (!test_bit(NIC_STATE_DOWN, &priv->state))
1330 return 0;
1332 ret = hns_nic_init_irq(priv);
1333 if (ret != 0) {
1334 netdev_err(ndev, "hns init irq failed! ret=%d\n", ret);
1335 return ret;
1338 for (i = 0; i < h->q_num * 2; i++) {
1339 ret = hns_nic_ring_open(ndev, i);
1340 if (ret)
1341 goto out_has_some_queues;
1344 ret = h->dev->ops->set_mac_addr(h, ndev->dev_addr);
1345 if (ret)
1346 goto out_set_mac_addr_err;
1348 ret = h->dev->ops->start ? h->dev->ops->start(h) : 0;
1349 if (ret)
1350 goto out_start_err;
1352 if (ndev->phydev)
1353 phy_start(ndev->phydev);
1355 clear_bit(NIC_STATE_DOWN, &priv->state);
1356 (void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
1358 return 0;
1360 out_start_err:
1361 netif_stop_queue(ndev);
1362 out_set_mac_addr_err:
1363 out_has_some_queues:
1364 for (j = i - 1; j >= 0; j--)
1365 hns_nic_ring_close(ndev, j);
1367 hns_nic_free_irq(h->q_num, priv);
1368 set_bit(NIC_STATE_DOWN, &priv->state);
1370 return ret;
1373 static void hns_nic_net_down(struct net_device *ndev)
1375 int i;
1376 struct hnae_ae_ops *ops;
1377 struct hns_nic_priv *priv = netdev_priv(ndev);
1379 if (test_and_set_bit(NIC_STATE_DOWN, &priv->state))
1380 return;
1382 (void)del_timer_sync(&priv->service_timer);
1383 netif_tx_stop_all_queues(ndev);
1384 netif_carrier_off(ndev);
1385 netif_tx_disable(ndev);
1386 priv->link = 0;
1388 if (ndev->phydev)
1389 phy_stop(ndev->phydev);
1391 ops = priv->ae_handle->dev->ops;
1393 if (ops->stop)
1394 ops->stop(priv->ae_handle);
1396 netif_tx_stop_all_queues(ndev);
1398 for (i = priv->ae_handle->q_num - 1; i >= 0; i--) {
1399 hns_nic_ring_close(ndev, i);
1400 hns_nic_ring_close(ndev, i + priv->ae_handle->q_num);
1402 /* clean tx buffers*/
1403 hns_nic_tx_clr_all_bufs(priv->ring_data + i);
1407 void hns_nic_net_reset(struct net_device *ndev)
1409 struct hns_nic_priv *priv = netdev_priv(ndev);
1410 struct hnae_handle *handle = priv->ae_handle;
1412 while (test_and_set_bit(NIC_STATE_RESETTING, &priv->state))
1413 usleep_range(1000, 2000);
1415 (void)hnae_reinit_handle(handle);
1417 clear_bit(NIC_STATE_RESETTING, &priv->state);
1420 void hns_nic_net_reinit(struct net_device *netdev)
1422 struct hns_nic_priv *priv = netdev_priv(netdev);
1423 enum hnae_port_type type = priv->ae_handle->port_type;
1425 netif_trans_update(priv->netdev);
1426 while (test_and_set_bit(NIC_STATE_REINITING, &priv->state))
1427 usleep_range(1000, 2000);
1429 hns_nic_net_down(netdev);
1431 /* Only do hns_nic_net_reset in debug mode
1432 * because of hardware limitation.
1434 if (type == HNAE_PORT_DEBUG)
1435 hns_nic_net_reset(netdev);
1437 (void)hns_nic_net_up(netdev);
1438 clear_bit(NIC_STATE_REINITING, &priv->state);
1441 static int hns_nic_net_open(struct net_device *ndev)
1443 struct hns_nic_priv *priv = netdev_priv(ndev);
1444 struct hnae_handle *h = priv->ae_handle;
1445 int ret;
1447 if (test_bit(NIC_STATE_TESTING, &priv->state))
1448 return -EBUSY;
1450 priv->link = 0;
1451 netif_carrier_off(ndev);
1453 ret = netif_set_real_num_tx_queues(ndev, h->q_num);
1454 if (ret < 0) {
1455 netdev_err(ndev, "netif_set_real_num_tx_queues fail, ret=%d!\n",
1456 ret);
1457 return ret;
1460 ret = netif_set_real_num_rx_queues(ndev, h->q_num);
1461 if (ret < 0) {
1462 netdev_err(ndev,
1463 "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
1464 return ret;
1467 ret = hns_nic_net_up(ndev);
1468 if (ret) {
1469 netdev_err(ndev,
1470 "hns net up fail, ret=%d!\n", ret);
1471 return ret;
1474 return 0;
1477 static int hns_nic_net_stop(struct net_device *ndev)
1479 hns_nic_net_down(ndev);
1481 return 0;
1484 static void hns_tx_timeout_reset(struct hns_nic_priv *priv);
1485 #define HNS_TX_TIMEO_LIMIT (40 * HZ)
1486 static void hns_nic_net_timeout(struct net_device *ndev, unsigned int txqueue)
1488 struct hns_nic_priv *priv = netdev_priv(ndev);
1490 if (ndev->watchdog_timeo < HNS_TX_TIMEO_LIMIT) {
1491 ndev->watchdog_timeo *= 2;
1492 netdev_info(ndev, "watchdog_timo changed to %d.\n",
1493 ndev->watchdog_timeo);
1494 } else {
1495 ndev->watchdog_timeo = HNS_NIC_TX_TIMEOUT;
1496 hns_tx_timeout_reset(priv);
1500 static netdev_tx_t hns_nic_net_xmit(struct sk_buff *skb,
1501 struct net_device *ndev)
1503 struct hns_nic_priv *priv = netdev_priv(ndev);
1505 assert(skb->queue_mapping < ndev->ae_handle->q_num);
1507 return hns_nic_net_xmit_hw(ndev, skb,
1508 &tx_ring_data(priv, skb->queue_mapping));
1511 static void hns_nic_drop_rx_fetch(struct hns_nic_ring_data *ring_data,
1512 struct sk_buff *skb)
1514 dev_kfree_skb_any(skb);
1517 #define HNS_LB_TX_RING 0
1518 static struct sk_buff *hns_assemble_skb(struct net_device *ndev)
1520 struct sk_buff *skb;
1521 struct ethhdr *ethhdr;
1522 int frame_len;
1524 /* allocate test skb */
1525 skb = alloc_skb(64, GFP_KERNEL);
1526 if (!skb)
1527 return NULL;
1529 skb_put(skb, 64);
1530 skb->dev = ndev;
1531 memset(skb->data, 0xFF, skb->len);
1533 /* must be tcp/ip package */
1534 ethhdr = (struct ethhdr *)skb->data;
1535 ethhdr->h_proto = htons(ETH_P_IP);
1537 frame_len = skb->len & (~1ul);
1538 memset(&skb->data[frame_len / 2], 0xAA,
1539 frame_len / 2 - 1);
1541 skb->queue_mapping = HNS_LB_TX_RING;
1543 return skb;
1546 static int hns_enable_serdes_lb(struct net_device *ndev)
1548 struct hns_nic_priv *priv = netdev_priv(ndev);
1549 struct hnae_handle *h = priv->ae_handle;
1550 struct hnae_ae_ops *ops = h->dev->ops;
1551 int speed, duplex;
1552 int ret;
1554 ret = ops->set_loopback(h, MAC_INTERNALLOOP_SERDES, 1);
1555 if (ret)
1556 return ret;
1558 ret = ops->start ? ops->start(h) : 0;
1559 if (ret)
1560 return ret;
1562 /* link adjust duplex*/
1563 if (h->phy_if != PHY_INTERFACE_MODE_XGMII)
1564 speed = 1000;
1565 else
1566 speed = 10000;
1567 duplex = 1;
1569 ops->adjust_link(h, speed, duplex);
1571 /* wait h/w ready */
1572 mdelay(300);
1574 return 0;
1577 static void hns_disable_serdes_lb(struct net_device *ndev)
1579 struct hns_nic_priv *priv = netdev_priv(ndev);
1580 struct hnae_handle *h = priv->ae_handle;
1581 struct hnae_ae_ops *ops = h->dev->ops;
1583 ops->stop(h);
1584 ops->set_loopback(h, MAC_INTERNALLOOP_SERDES, 0);
1588 *hns_nic_clear_all_rx_fetch - clear the chip fetched descriptions. The
1589 *function as follows:
1590 * 1. if one rx ring has found the page_offset is not equal 0 between head
1591 * and tail, it means that the chip fetched the wrong descs for the ring
1592 * which buffer size is 4096.
1593 * 2. we set the chip serdes loopback and set rss indirection to the ring.
1594 * 3. construct 64-bytes ip broadcast packages, wait the associated rx ring
1595 * recieving all packages and it will fetch new descriptions.
1596 * 4. recover to the original state.
1598 *@ndev: net device
1600 static int hns_nic_clear_all_rx_fetch(struct net_device *ndev)
1602 struct hns_nic_priv *priv = netdev_priv(ndev);
1603 struct hnae_handle *h = priv->ae_handle;
1604 struct hnae_ae_ops *ops = h->dev->ops;
1605 struct hns_nic_ring_data *rd;
1606 struct hnae_ring *ring;
1607 struct sk_buff *skb;
1608 u32 *org_indir;
1609 u32 *cur_indir;
1610 int indir_size;
1611 int head, tail;
1612 int fetch_num;
1613 int i, j;
1614 bool found;
1615 int retry_times;
1616 int ret = 0;
1618 /* alloc indir memory */
1619 indir_size = ops->get_rss_indir_size(h) * sizeof(*org_indir);
1620 org_indir = kzalloc(indir_size, GFP_KERNEL);
1621 if (!org_indir)
1622 return -ENOMEM;
1624 /* store the orginal indirection */
1625 ops->get_rss(h, org_indir, NULL, NULL);
1627 cur_indir = kzalloc(indir_size, GFP_KERNEL);
1628 if (!cur_indir) {
1629 ret = -ENOMEM;
1630 goto cur_indir_alloc_err;
1633 /* set loopback */
1634 if (hns_enable_serdes_lb(ndev)) {
1635 ret = -EINVAL;
1636 goto enable_serdes_lb_err;
1639 /* foreach every rx ring to clear fetch desc */
1640 for (i = 0; i < h->q_num; i++) {
1641 ring = &h->qs[i]->rx_ring;
1642 head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1643 tail = readl_relaxed(ring->io_base + RCB_REG_TAIL);
1644 found = false;
1645 fetch_num = ring_dist(ring, head, tail);
1647 while (head != tail) {
1648 if (ring->desc_cb[head].page_offset != 0) {
1649 found = true;
1650 break;
1653 head++;
1654 if (head == ring->desc_num)
1655 head = 0;
1658 if (found) {
1659 for (j = 0; j < indir_size / sizeof(*org_indir); j++)
1660 cur_indir[j] = i;
1661 ops->set_rss(h, cur_indir, NULL, 0);
1663 for (j = 0; j < fetch_num; j++) {
1664 /* alloc one skb and init */
1665 skb = hns_assemble_skb(ndev);
1666 if (!skb)
1667 goto out;
1668 rd = &tx_ring_data(priv, skb->queue_mapping);
1669 hns_nic_net_xmit_hw(ndev, skb, rd);
1671 retry_times = 0;
1672 while (retry_times++ < 10) {
1673 mdelay(10);
1674 /* clean rx */
1675 rd = &rx_ring_data(priv, i);
1676 if (rd->poll_one(rd, fetch_num,
1677 hns_nic_drop_rx_fetch))
1678 break;
1681 retry_times = 0;
1682 while (retry_times++ < 10) {
1683 mdelay(10);
1684 /* clean tx ring 0 send package */
1685 rd = &tx_ring_data(priv,
1686 HNS_LB_TX_RING);
1687 if (rd->poll_one(rd, fetch_num, NULL))
1688 break;
1694 out:
1695 /* restore everything */
1696 ops->set_rss(h, org_indir, NULL, 0);
1697 hns_disable_serdes_lb(ndev);
1698 enable_serdes_lb_err:
1699 kfree(cur_indir);
1700 cur_indir_alloc_err:
1701 kfree(org_indir);
1703 return ret;
1706 static int hns_nic_change_mtu(struct net_device *ndev, int new_mtu)
1708 struct hns_nic_priv *priv = netdev_priv(ndev);
1709 struct hnae_handle *h = priv->ae_handle;
1710 bool if_running = netif_running(ndev);
1711 int ret;
1713 /* MTU < 68 is an error and causes problems on some kernels */
1714 if (new_mtu < 68)
1715 return -EINVAL;
1717 /* MTU no change */
1718 if (new_mtu == ndev->mtu)
1719 return 0;
1721 if (!h->dev->ops->set_mtu)
1722 return -ENOTSUPP;
1724 if (if_running) {
1725 (void)hns_nic_net_stop(ndev);
1726 msleep(100);
1729 if (priv->enet_ver != AE_VERSION_1 &&
1730 ndev->mtu <= BD_SIZE_2048_MAX_MTU &&
1731 new_mtu > BD_SIZE_2048_MAX_MTU) {
1732 /* update desc */
1733 hnae_reinit_all_ring_desc(h);
1735 /* clear the package which the chip has fetched */
1736 ret = hns_nic_clear_all_rx_fetch(ndev);
1738 /* the page offset must be consist with desc */
1739 hnae_reinit_all_ring_page_off(h);
1741 if (ret) {
1742 netdev_err(ndev, "clear the fetched desc fail\n");
1743 goto out;
1747 ret = h->dev->ops->set_mtu(h, new_mtu);
1748 if (ret) {
1749 netdev_err(ndev, "set mtu fail, return value %d\n",
1750 ret);
1751 goto out;
1754 /* finally, set new mtu to netdevice */
1755 ndev->mtu = new_mtu;
1757 out:
1758 if (if_running) {
1759 if (hns_nic_net_open(ndev)) {
1760 netdev_err(ndev, "hns net open fail\n");
1761 ret = -EINVAL;
1765 return ret;
1768 static int hns_nic_set_features(struct net_device *netdev,
1769 netdev_features_t features)
1771 struct hns_nic_priv *priv = netdev_priv(netdev);
1773 switch (priv->enet_ver) {
1774 case AE_VERSION_1:
1775 if (features & (NETIF_F_TSO | NETIF_F_TSO6))
1776 netdev_info(netdev, "enet v1 do not support tso!\n");
1777 break;
1778 default:
1779 if (features & (NETIF_F_TSO | NETIF_F_TSO6)) {
1780 priv->ops.fill_desc = fill_tso_desc;
1781 priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tso;
1782 /* The chip only support 7*4096 */
1783 netif_set_gso_max_size(netdev, 7 * 4096);
1784 } else {
1785 priv->ops.fill_desc = fill_v2_desc;
1786 priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
1788 break;
1790 netdev->features = features;
1791 return 0;
1794 static netdev_features_t hns_nic_fix_features(
1795 struct net_device *netdev, netdev_features_t features)
1797 struct hns_nic_priv *priv = netdev_priv(netdev);
1799 switch (priv->enet_ver) {
1800 case AE_VERSION_1:
1801 features &= ~(NETIF_F_TSO | NETIF_F_TSO6 |
1802 NETIF_F_HW_VLAN_CTAG_FILTER);
1803 break;
1804 default:
1805 break;
1807 return features;
1810 static int hns_nic_uc_sync(struct net_device *netdev, const unsigned char *addr)
1812 struct hns_nic_priv *priv = netdev_priv(netdev);
1813 struct hnae_handle *h = priv->ae_handle;
1815 if (h->dev->ops->add_uc_addr)
1816 return h->dev->ops->add_uc_addr(h, addr);
1818 return 0;
1821 static int hns_nic_uc_unsync(struct net_device *netdev,
1822 const unsigned char *addr)
1824 struct hns_nic_priv *priv = netdev_priv(netdev);
1825 struct hnae_handle *h = priv->ae_handle;
1827 if (h->dev->ops->rm_uc_addr)
1828 return h->dev->ops->rm_uc_addr(h, addr);
1830 return 0;
1834 * hns_set_multicast_list - set mutl mac address
1835 * @ndev: net device
1837 * return void
1839 static void hns_set_multicast_list(struct net_device *ndev)
1841 struct hns_nic_priv *priv = netdev_priv(ndev);
1842 struct hnae_handle *h = priv->ae_handle;
1843 struct netdev_hw_addr *ha = NULL;
1845 if (!h) {
1846 netdev_err(ndev, "hnae handle is null\n");
1847 return;
1850 if (h->dev->ops->clr_mc_addr)
1851 if (h->dev->ops->clr_mc_addr(h))
1852 netdev_err(ndev, "clear multicast address fail\n");
1854 if (h->dev->ops->set_mc_addr) {
1855 netdev_for_each_mc_addr(ha, ndev)
1856 if (h->dev->ops->set_mc_addr(h, ha->addr))
1857 netdev_err(ndev, "set multicast fail\n");
1861 static void hns_nic_set_rx_mode(struct net_device *ndev)
1863 struct hns_nic_priv *priv = netdev_priv(ndev);
1864 struct hnae_handle *h = priv->ae_handle;
1866 if (h->dev->ops->set_promisc_mode) {
1867 if (ndev->flags & IFF_PROMISC)
1868 h->dev->ops->set_promisc_mode(h, 1);
1869 else
1870 h->dev->ops->set_promisc_mode(h, 0);
1873 hns_set_multicast_list(ndev);
1875 if (__dev_uc_sync(ndev, hns_nic_uc_sync, hns_nic_uc_unsync))
1876 netdev_err(ndev, "sync uc address fail\n");
1879 static void hns_nic_get_stats64(struct net_device *ndev,
1880 struct rtnl_link_stats64 *stats)
1882 int idx = 0;
1883 u64 tx_bytes = 0;
1884 u64 rx_bytes = 0;
1885 u64 tx_pkts = 0;
1886 u64 rx_pkts = 0;
1887 struct hns_nic_priv *priv = netdev_priv(ndev);
1888 struct hnae_handle *h = priv->ae_handle;
1890 for (idx = 0; idx < h->q_num; idx++) {
1891 tx_bytes += h->qs[idx]->tx_ring.stats.tx_bytes;
1892 tx_pkts += h->qs[idx]->tx_ring.stats.tx_pkts;
1893 rx_bytes += h->qs[idx]->rx_ring.stats.rx_bytes;
1894 rx_pkts += h->qs[idx]->rx_ring.stats.rx_pkts;
1897 stats->tx_bytes = tx_bytes;
1898 stats->tx_packets = tx_pkts;
1899 stats->rx_bytes = rx_bytes;
1900 stats->rx_packets = rx_pkts;
1902 stats->rx_errors = ndev->stats.rx_errors;
1903 stats->multicast = ndev->stats.multicast;
1904 stats->rx_length_errors = ndev->stats.rx_length_errors;
1905 stats->rx_crc_errors = ndev->stats.rx_crc_errors;
1906 stats->rx_missed_errors = ndev->stats.rx_missed_errors;
1908 stats->tx_errors = ndev->stats.tx_errors;
1909 stats->rx_dropped = ndev->stats.rx_dropped;
1910 stats->tx_dropped = ndev->stats.tx_dropped;
1911 stats->collisions = ndev->stats.collisions;
1912 stats->rx_over_errors = ndev->stats.rx_over_errors;
1913 stats->rx_frame_errors = ndev->stats.rx_frame_errors;
1914 stats->rx_fifo_errors = ndev->stats.rx_fifo_errors;
1915 stats->tx_aborted_errors = ndev->stats.tx_aborted_errors;
1916 stats->tx_carrier_errors = ndev->stats.tx_carrier_errors;
1917 stats->tx_fifo_errors = ndev->stats.tx_fifo_errors;
1918 stats->tx_heartbeat_errors = ndev->stats.tx_heartbeat_errors;
1919 stats->tx_window_errors = ndev->stats.tx_window_errors;
1920 stats->rx_compressed = ndev->stats.rx_compressed;
1921 stats->tx_compressed = ndev->stats.tx_compressed;
1924 static u16
1925 hns_nic_select_queue(struct net_device *ndev, struct sk_buff *skb,
1926 struct net_device *sb_dev)
1928 struct ethhdr *eth_hdr = (struct ethhdr *)skb->data;
1929 struct hns_nic_priv *priv = netdev_priv(ndev);
1931 /* fix hardware broadcast/multicast packets queue loopback */
1932 if (!AE_IS_VER1(priv->enet_ver) &&
1933 is_multicast_ether_addr(eth_hdr->h_dest))
1934 return 0;
1935 else
1936 return netdev_pick_tx(ndev, skb, NULL);
1939 static const struct net_device_ops hns_nic_netdev_ops = {
1940 .ndo_open = hns_nic_net_open,
1941 .ndo_stop = hns_nic_net_stop,
1942 .ndo_start_xmit = hns_nic_net_xmit,
1943 .ndo_tx_timeout = hns_nic_net_timeout,
1944 .ndo_set_mac_address = hns_nic_net_set_mac_address,
1945 .ndo_change_mtu = hns_nic_change_mtu,
1946 .ndo_do_ioctl = phy_do_ioctl_running,
1947 .ndo_set_features = hns_nic_set_features,
1948 .ndo_fix_features = hns_nic_fix_features,
1949 .ndo_get_stats64 = hns_nic_get_stats64,
1950 .ndo_set_rx_mode = hns_nic_set_rx_mode,
1951 .ndo_select_queue = hns_nic_select_queue,
1954 static void hns_nic_update_link_status(struct net_device *netdev)
1956 struct hns_nic_priv *priv = netdev_priv(netdev);
1958 struct hnae_handle *h = priv->ae_handle;
1960 if (h->phy_dev) {
1961 if (h->phy_if != PHY_INTERFACE_MODE_XGMII)
1962 return;
1964 (void)genphy_read_status(h->phy_dev);
1966 hns_nic_adjust_link(netdev);
1969 /* for dumping key regs*/
1970 static void hns_nic_dump(struct hns_nic_priv *priv)
1972 struct hnae_handle *h = priv->ae_handle;
1973 struct hnae_ae_ops *ops = h->dev->ops;
1974 u32 *data, reg_num, i;
1976 if (ops->get_regs_len && ops->get_regs) {
1977 reg_num = ops->get_regs_len(priv->ae_handle);
1978 reg_num = (reg_num + 3ul) & ~3ul;
1979 data = kcalloc(reg_num, sizeof(u32), GFP_KERNEL);
1980 if (data) {
1981 ops->get_regs(priv->ae_handle, data);
1982 for (i = 0; i < reg_num; i += 4)
1983 pr_info("0x%08x: 0x%08x 0x%08x 0x%08x 0x%08x\n",
1984 i, data[i], data[i + 1],
1985 data[i + 2], data[i + 3]);
1986 kfree(data);
1990 for (i = 0; i < h->q_num; i++) {
1991 pr_info("tx_queue%d_next_to_clean:%d\n",
1992 i, h->qs[i]->tx_ring.next_to_clean);
1993 pr_info("tx_queue%d_next_to_use:%d\n",
1994 i, h->qs[i]->tx_ring.next_to_use);
1995 pr_info("rx_queue%d_next_to_clean:%d\n",
1996 i, h->qs[i]->rx_ring.next_to_clean);
1997 pr_info("rx_queue%d_next_to_use:%d\n",
1998 i, h->qs[i]->rx_ring.next_to_use);
2002 /* for resetting subtask */
2003 static void hns_nic_reset_subtask(struct hns_nic_priv *priv)
2005 enum hnae_port_type type = priv->ae_handle->port_type;
2007 if (!test_bit(NIC_STATE2_RESET_REQUESTED, &priv->state))
2008 return;
2009 clear_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
2011 /* If we're already down, removing or resetting, just bail */
2012 if (test_bit(NIC_STATE_DOWN, &priv->state) ||
2013 test_bit(NIC_STATE_REMOVING, &priv->state) ||
2014 test_bit(NIC_STATE_RESETTING, &priv->state))
2015 return;
2017 hns_nic_dump(priv);
2018 netdev_info(priv->netdev, "try to reset %s port!\n",
2019 (type == HNAE_PORT_DEBUG ? "debug" : "service"));
2021 rtnl_lock();
2022 /* put off any impending NetWatchDogTimeout */
2023 netif_trans_update(priv->netdev);
2024 hns_nic_net_reinit(priv->netdev);
2026 rtnl_unlock();
2029 /* for doing service complete*/
2030 static void hns_nic_service_event_complete(struct hns_nic_priv *priv)
2032 WARN_ON(!test_bit(NIC_STATE_SERVICE_SCHED, &priv->state));
2033 /* make sure to commit the things */
2034 smp_mb__before_atomic();
2035 clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
2038 static void hns_nic_service_task(struct work_struct *work)
2040 struct hns_nic_priv *priv
2041 = container_of(work, struct hns_nic_priv, service_task);
2042 struct hnae_handle *h = priv->ae_handle;
2044 hns_nic_reset_subtask(priv);
2045 hns_nic_update_link_status(priv->netdev);
2046 h->dev->ops->update_led_status(h);
2047 hns_nic_update_stats(priv->netdev);
2049 hns_nic_service_event_complete(priv);
2052 static void hns_nic_task_schedule(struct hns_nic_priv *priv)
2054 if (!test_bit(NIC_STATE_DOWN, &priv->state) &&
2055 !test_bit(NIC_STATE_REMOVING, &priv->state) &&
2056 !test_and_set_bit(NIC_STATE_SERVICE_SCHED, &priv->state))
2057 (void)schedule_work(&priv->service_task);
2060 static void hns_nic_service_timer(struct timer_list *t)
2062 struct hns_nic_priv *priv = from_timer(priv, t, service_timer);
2064 (void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
2066 hns_nic_task_schedule(priv);
2070 * hns_tx_timeout_reset - initiate reset due to Tx timeout
2071 * @priv: driver private struct
2073 static void hns_tx_timeout_reset(struct hns_nic_priv *priv)
2075 /* Do the reset outside of interrupt context */
2076 if (!test_bit(NIC_STATE_DOWN, &priv->state)) {
2077 set_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
2078 netdev_warn(priv->netdev,
2079 "initiating reset due to tx timeout(%llu,0x%lx)\n",
2080 priv->tx_timeout_count, priv->state);
2081 priv->tx_timeout_count++;
2082 hns_nic_task_schedule(priv);
2086 static int hns_nic_init_ring_data(struct hns_nic_priv *priv)
2088 struct hnae_handle *h = priv->ae_handle;
2089 struct hns_nic_ring_data *rd;
2090 bool is_ver1 = AE_IS_VER1(priv->enet_ver);
2091 int i;
2093 if (h->q_num > NIC_MAX_Q_PER_VF) {
2094 netdev_err(priv->netdev, "too much queue (%d)\n", h->q_num);
2095 return -EINVAL;
2098 priv->ring_data = kzalloc(array3_size(h->q_num,
2099 sizeof(*priv->ring_data), 2),
2100 GFP_KERNEL);
2101 if (!priv->ring_data)
2102 return -ENOMEM;
2104 for (i = 0; i < h->q_num; i++) {
2105 rd = &priv->ring_data[i];
2106 rd->queue_index = i;
2107 rd->ring = &h->qs[i]->tx_ring;
2108 rd->poll_one = hns_nic_tx_poll_one;
2109 rd->fini_process = is_ver1 ? hns_nic_tx_fini_pro :
2110 hns_nic_tx_fini_pro_v2;
2112 netif_napi_add(priv->netdev, &rd->napi,
2113 hns_nic_common_poll, NAPI_POLL_WEIGHT);
2114 rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
2116 for (i = h->q_num; i < h->q_num * 2; i++) {
2117 rd = &priv->ring_data[i];
2118 rd->queue_index = i - h->q_num;
2119 rd->ring = &h->qs[i - h->q_num]->rx_ring;
2120 rd->poll_one = hns_nic_rx_poll_one;
2121 rd->ex_process = hns_nic_rx_up_pro;
2122 rd->fini_process = is_ver1 ? hns_nic_rx_fini_pro :
2123 hns_nic_rx_fini_pro_v2;
2125 netif_napi_add(priv->netdev, &rd->napi,
2126 hns_nic_common_poll, NAPI_POLL_WEIGHT);
2127 rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
2130 return 0;
2133 static void hns_nic_uninit_ring_data(struct hns_nic_priv *priv)
2135 struct hnae_handle *h = priv->ae_handle;
2136 int i;
2138 for (i = 0; i < h->q_num * 2; i++) {
2139 netif_napi_del(&priv->ring_data[i].napi);
2140 if (priv->ring_data[i].ring->irq_init_flag == RCB_IRQ_INITED) {
2141 (void)irq_set_affinity_hint(
2142 priv->ring_data[i].ring->irq,
2143 NULL);
2144 free_irq(priv->ring_data[i].ring->irq,
2145 &priv->ring_data[i]);
2148 priv->ring_data[i].ring->irq_init_flag = RCB_IRQ_NOT_INITED;
2150 kfree(priv->ring_data);
2153 static void hns_nic_set_priv_ops(struct net_device *netdev)
2155 struct hns_nic_priv *priv = netdev_priv(netdev);
2156 struct hnae_handle *h = priv->ae_handle;
2158 if (AE_IS_VER1(priv->enet_ver)) {
2159 priv->ops.fill_desc = fill_desc;
2160 priv->ops.get_rxd_bnum = get_rx_desc_bnum;
2161 priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
2162 } else {
2163 priv->ops.get_rxd_bnum = get_v2rx_desc_bnum;
2164 if ((netdev->features & NETIF_F_TSO) ||
2165 (netdev->features & NETIF_F_TSO6)) {
2166 priv->ops.fill_desc = fill_tso_desc;
2167 priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tso;
2168 /* This chip only support 7*4096 */
2169 netif_set_gso_max_size(netdev, 7 * 4096);
2170 } else {
2171 priv->ops.fill_desc = fill_v2_desc;
2172 priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
2174 /* enable tso when init
2175 * control tso on/off through TSE bit in bd
2177 h->dev->ops->set_tso_stats(h, 1);
2181 static int hns_nic_try_get_ae(struct net_device *ndev)
2183 struct hns_nic_priv *priv = netdev_priv(ndev);
2184 struct hnae_handle *h;
2185 int ret;
2187 h = hnae_get_handle(&priv->netdev->dev,
2188 priv->fwnode, priv->port_id, NULL);
2189 if (IS_ERR_OR_NULL(h)) {
2190 ret = -ENODEV;
2191 dev_dbg(priv->dev, "has not handle, register notifier!\n");
2192 goto out;
2194 priv->ae_handle = h;
2196 ret = hns_nic_init_phy(ndev, h);
2197 if (ret) {
2198 dev_err(priv->dev, "probe phy device fail!\n");
2199 goto out_init_phy;
2202 ret = hns_nic_init_ring_data(priv);
2203 if (ret) {
2204 ret = -ENOMEM;
2205 goto out_init_ring_data;
2208 hns_nic_set_priv_ops(ndev);
2210 ret = register_netdev(ndev);
2211 if (ret) {
2212 dev_err(priv->dev, "probe register netdev fail!\n");
2213 goto out_reg_ndev_fail;
2215 return 0;
2217 out_reg_ndev_fail:
2218 hns_nic_uninit_ring_data(priv);
2219 priv->ring_data = NULL;
2220 out_init_phy:
2221 out_init_ring_data:
2222 hnae_put_handle(priv->ae_handle);
2223 priv->ae_handle = NULL;
2224 out:
2225 return ret;
2228 static int hns_nic_notifier_action(struct notifier_block *nb,
2229 unsigned long action, void *data)
2231 struct hns_nic_priv *priv =
2232 container_of(nb, struct hns_nic_priv, notifier_block);
2234 assert(action == HNAE_AE_REGISTER);
2236 if (!hns_nic_try_get_ae(priv->netdev)) {
2237 hnae_unregister_notifier(&priv->notifier_block);
2238 priv->notifier_block.notifier_call = NULL;
2240 return 0;
2243 static int hns_nic_dev_probe(struct platform_device *pdev)
2245 struct device *dev = &pdev->dev;
2246 struct net_device *ndev;
2247 struct hns_nic_priv *priv;
2248 u32 port_id;
2249 int ret;
2251 ndev = alloc_etherdev_mq(sizeof(struct hns_nic_priv), NIC_MAX_Q_PER_VF);
2252 if (!ndev)
2253 return -ENOMEM;
2255 platform_set_drvdata(pdev, ndev);
2257 priv = netdev_priv(ndev);
2258 priv->dev = dev;
2259 priv->netdev = ndev;
2261 if (dev_of_node(dev)) {
2262 struct device_node *ae_node;
2264 if (of_device_is_compatible(dev->of_node,
2265 "hisilicon,hns-nic-v1"))
2266 priv->enet_ver = AE_VERSION_1;
2267 else
2268 priv->enet_ver = AE_VERSION_2;
2270 ae_node = of_parse_phandle(dev->of_node, "ae-handle", 0);
2271 if (!ae_node) {
2272 ret = -ENODEV;
2273 dev_err(dev, "not find ae-handle\n");
2274 goto out_read_prop_fail;
2276 priv->fwnode = &ae_node->fwnode;
2277 } else if (is_acpi_node(dev->fwnode)) {
2278 struct fwnode_reference_args args;
2280 if (acpi_dev_found(hns_enet_acpi_match[0].id))
2281 priv->enet_ver = AE_VERSION_1;
2282 else if (acpi_dev_found(hns_enet_acpi_match[1].id))
2283 priv->enet_ver = AE_VERSION_2;
2284 else {
2285 ret = -ENXIO;
2286 goto out_read_prop_fail;
2289 /* try to find port-idx-in-ae first */
2290 ret = acpi_node_get_property_reference(dev->fwnode,
2291 "ae-handle", 0, &args);
2292 if (ret) {
2293 dev_err(dev, "not find ae-handle\n");
2294 goto out_read_prop_fail;
2296 if (!is_acpi_device_node(args.fwnode)) {
2297 ret = -EINVAL;
2298 goto out_read_prop_fail;
2300 priv->fwnode = args.fwnode;
2301 } else {
2302 dev_err(dev, "cannot read cfg data from OF or acpi\n");
2303 ret = -ENXIO;
2304 goto out_read_prop_fail;
2307 ret = device_property_read_u32(dev, "port-idx-in-ae", &port_id);
2308 if (ret) {
2309 /* only for old code compatible */
2310 ret = device_property_read_u32(dev, "port-id", &port_id);
2311 if (ret)
2312 goto out_read_prop_fail;
2313 /* for old dts, we need to caculate the port offset */
2314 port_id = port_id < HNS_SRV_OFFSET ? port_id + HNS_DEBUG_OFFSET
2315 : port_id - HNS_SRV_OFFSET;
2317 priv->port_id = port_id;
2319 hns_init_mac_addr(ndev);
2321 ndev->watchdog_timeo = HNS_NIC_TX_TIMEOUT;
2322 ndev->priv_flags |= IFF_UNICAST_FLT;
2323 ndev->netdev_ops = &hns_nic_netdev_ops;
2324 hns_ethtool_set_ops(ndev);
2326 ndev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2327 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2328 NETIF_F_GRO;
2329 ndev->vlan_features |=
2330 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM;
2331 ndev->vlan_features |= NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
2333 /* MTU range: 68 - 9578 (v1) or 9706 (v2) */
2334 ndev->min_mtu = MAC_MIN_MTU;
2335 switch (priv->enet_ver) {
2336 case AE_VERSION_2:
2337 ndev->features |= NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_NTUPLE;
2338 ndev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2339 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2340 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6;
2341 ndev->vlan_features |= NETIF_F_TSO | NETIF_F_TSO6;
2342 ndev->max_mtu = MAC_MAX_MTU_V2 -
2343 (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
2344 break;
2345 default:
2346 ndev->max_mtu = MAC_MAX_MTU -
2347 (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
2348 break;
2351 SET_NETDEV_DEV(ndev, dev);
2353 if (!dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64)))
2354 dev_dbg(dev, "set mask to 64bit\n");
2355 else
2356 dev_err(dev, "set mask to 64bit fail!\n");
2358 /* carrier off reporting is important to ethtool even BEFORE open */
2359 netif_carrier_off(ndev);
2361 timer_setup(&priv->service_timer, hns_nic_service_timer, 0);
2362 INIT_WORK(&priv->service_task, hns_nic_service_task);
2364 set_bit(NIC_STATE_SERVICE_INITED, &priv->state);
2365 clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
2366 set_bit(NIC_STATE_DOWN, &priv->state);
2368 if (hns_nic_try_get_ae(priv->netdev)) {
2369 priv->notifier_block.notifier_call = hns_nic_notifier_action;
2370 ret = hnae_register_notifier(&priv->notifier_block);
2371 if (ret) {
2372 dev_err(dev, "register notifier fail!\n");
2373 goto out_notify_fail;
2375 dev_dbg(dev, "has not handle, register notifier!\n");
2378 return 0;
2380 out_notify_fail:
2381 (void)cancel_work_sync(&priv->service_task);
2382 out_read_prop_fail:
2383 /* safe for ACPI FW */
2384 of_node_put(to_of_node(priv->fwnode));
2385 free_netdev(ndev);
2386 return ret;
2389 static int hns_nic_dev_remove(struct platform_device *pdev)
2391 struct net_device *ndev = platform_get_drvdata(pdev);
2392 struct hns_nic_priv *priv = netdev_priv(ndev);
2394 if (ndev->reg_state != NETREG_UNINITIALIZED)
2395 unregister_netdev(ndev);
2397 if (priv->ring_data)
2398 hns_nic_uninit_ring_data(priv);
2399 priv->ring_data = NULL;
2401 if (ndev->phydev)
2402 phy_disconnect(ndev->phydev);
2404 if (!IS_ERR_OR_NULL(priv->ae_handle))
2405 hnae_put_handle(priv->ae_handle);
2406 priv->ae_handle = NULL;
2407 if (priv->notifier_block.notifier_call)
2408 hnae_unregister_notifier(&priv->notifier_block);
2409 priv->notifier_block.notifier_call = NULL;
2411 set_bit(NIC_STATE_REMOVING, &priv->state);
2412 (void)cancel_work_sync(&priv->service_task);
2414 /* safe for ACPI FW */
2415 of_node_put(to_of_node(priv->fwnode));
2417 free_netdev(ndev);
2418 return 0;
2421 static const struct of_device_id hns_enet_of_match[] = {
2422 {.compatible = "hisilicon,hns-nic-v1",},
2423 {.compatible = "hisilicon,hns-nic-v2",},
2427 MODULE_DEVICE_TABLE(of, hns_enet_of_match);
2429 static struct platform_driver hns_nic_dev_driver = {
2430 .driver = {
2431 .name = "hns-nic",
2432 .of_match_table = hns_enet_of_match,
2433 .acpi_match_table = ACPI_PTR(hns_enet_acpi_match),
2435 .probe = hns_nic_dev_probe,
2436 .remove = hns_nic_dev_remove,
2439 module_platform_driver(hns_nic_dev_driver);
2441 MODULE_DESCRIPTION("HISILICON HNS Ethernet driver");
2442 MODULE_AUTHOR("Hisilicon, Inc.");
2443 MODULE_LICENSE("GPL");
2444 MODULE_ALIAS("platform:hns-nic");