2 * Copyright (c) 2014-2015 Hisilicon Limited.
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.
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>
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>
25 #include "hns_dsaf_mac.h"
27 #define NIC_MAX_Q_PER_VF 16
28 #define HNS_NIC_TX_TIMEOUT (5 * HZ)
30 #define SERVICE_TIMER_HZ (1 * HZ)
32 #define NIC_TX_CLEAN_MAX_NUM 256
33 #define NIC_RX_CLEAN_MAX_NUM 64
35 #define RCB_IRQ_NOT_INITED 0
36 #define RCB_IRQ_INITED 1
37 #define HNS_BUFFER_SIZE_2048 2048
39 #define BD_MAX_SEND_SIZE 8191
40 #define SKB_TMP_LEN(SKB) \
41 (((SKB)->transport_header - (SKB)->mac_header) + tcp_hdrlen(SKB))
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)
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];
50 struct ipv6hdr *ipv6hdr;
62 desc_cb->length = size;
66 desc->addr = cpu_to_le64(dma);
67 desc->tx.send_size = cpu_to_le16((u16)size);
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);
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);
77 if (type == DESC_TYPE_SKB) {
78 skb = (struct sk_buff *)priv;
80 if (skb->ip_summed == CHECKSUM_PARTIAL) {
81 skb_reset_mac_len(skb);
82 protocol = skb->protocol;
85 if (protocol == htons(ETH_P_8021Q)) {
86 ip_offset += VLAN_HLEN;
87 protocol = vlan_get_protocol(skb);
88 skb->protocol = protocol;
91 if (skb->protocol == htons(ETH_P_IP)) {
93 hnae_set_bit(rrcfv, HNSV2_TXD_L3CS_B, 1);
94 hnae_set_bit(rrcfv, HNSV2_TXD_L4CS_B, 1);
96 /* check for tcp/udp header */
97 if (iphdr->protocol == IPPROTO_TCP &&
101 l4_len = tcp_hdrlen(skb);
102 mss = skb_shinfo(skb)->gso_size;
103 paylen = skb->len - SKB_TMP_LEN(skb);
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);
110 /* check for tcp/udp header */
111 if (ipv6hdr->nexthdr == IPPROTO_TCP &&
112 skb_is_gso(skb) && skb_is_gso_v6(skb)) {
115 l4_len = tcp_hdrlen(skb);
116 mss = skb_shinfo(skb)->gso_size;
117 paylen = skb->len - SKB_TMP_LEN(skb);
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);
128 hnae_set_bit(rrcfv, HNSV2_TXD_FE_B, frag_end);
130 desc->tx.bn_pid = bn_pid;
131 desc->tx.ra_ri_cs_fe_vld = rrcfv;
133 ring_ptr_move_fw(ring, next_to_use);
136 static const struct acpi_device_id hns_enet_acpi_match[] = {
141 MODULE_DEVICE_TABLE(acpi, hns_enet_acpi_match);
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)
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];
152 u32 asid_bufnum_pid = 0;
153 u32 flag_ipoffset = 0;
155 desc_cb->priv = priv;
156 desc_cb->length = size;
158 desc_cb->type = type;
160 desc->addr = cpu_to_le64(dma);
161 desc->tx.send_size = cpu_to_le16((u16)size);
163 /*config bd buffer end */
164 flag_ipoffset |= 1 << HNS_TXD_VLD_B;
166 asid_bufnum_pid |= buf_num << HNS_TXD_BUFNUM_S;
168 if (type == DESC_TYPE_SKB) {
169 skb = (struct sk_buff *)priv;
171 if (skb->ip_summed == CHECKSUM_PARTIAL) {
172 protocol = skb->protocol;
173 ip_offset = ETH_HLEN;
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;
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;
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;
192 flag_ipoffset |= ip_offset << HNS_TXD_IPOFFSET_S;
196 flag_ipoffset |= frag_end << HNS_TXD_FE_B;
198 desc->tx.asid_bufnum_pid = cpu_to_le16(asid_bufnum_pid);
199 desc->tx.flag_ipoffset = cpu_to_le32(flag_ipoffset);
201 ring_ptr_move_fw(ring, next_to_use);
204 static void unfill_desc(struct hnae_ring *ring)
206 ring_ptr_move_bw(ring, next_to_use);
209 static int hns_nic_maybe_stop_tx(
210 struct sk_buff **out_skb, int *bnum, struct hnae_ring *ring)
212 struct sk_buff *skb = *out_skb;
213 struct sk_buff *new_skb = NULL;
216 /* no. of segments (plus a header) */
217 buf_num = skb_shinfo(skb)->nr_frags + 1;
219 if (unlikely(buf_num > ring->max_desc_num_per_pkt)) {
220 if (ring_space(ring) < 1)
223 new_skb = skb_copy(skb, GFP_ATOMIC);
227 dev_kfree_skb_any(skb);
230 } else if (buf_num > ring_space(ring)) {
238 static int hns_nic_maybe_stop_tso(
239 struct sk_buff **out_skb, int *bnum, struct hnae_ring *ring)
245 struct sk_buff *skb = *out_skb;
246 struct sk_buff *new_skb = NULL;
247 struct skb_frag_struct *frag;
249 size = skb_headlen(skb);
250 buf_num = (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
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;
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)
263 /* manual split the send packet */
264 new_skb = skb_copy(skb, GFP_ATOMIC);
267 dev_kfree_skb_any(skb);
270 } else if (ring_space(ring) < buf_num) {
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)
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;
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,
298 (type == DESC_TYPE_SKB && !k) ?
299 DESC_TYPE_SKB : DESC_TYPE_PAGE,
303 netdev_tx_t hns_nic_net_xmit_hw(struct net_device *ndev,
305 struct hns_nic_ring_data *ring_data)
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;
315 int size, next_to_use;
318 switch (priv->ops.maybe_stop_tx(&skb, &buf_num, ring)) {
320 ring->stats.tx_busy++;
321 goto out_net_tx_busy;
323 ring->stats.sw_err_cnt++;
324 netdev_err(ndev, "no memory to xmit!\n");
330 /* no. of segments (plus a header) */
331 seg_num = skb_shinfo(skb)->nr_frags + 1;
332 next_to_use = ring->next_to_use;
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++;
342 priv->ops.fill_desc(ring, skb, size, dma, seg_num == 1 ? 1 : 0,
343 buf_num, DESC_TYPE_SKB, ndev->mtu);
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;
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);
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);
364 netif_trans_update(ndev);
365 ndev->stats.tx_bytes += skb->len;
366 ndev->stats.tx_packets++;
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;
378 while (ring->next_to_use != next_to_use) {
380 if (ring->next_to_use != next_to_use)
382 ring->desc_cb[ring->next_to_use].dma,
383 ring->desc_cb[ring->next_to_use].length,
386 dma_unmap_single(dev,
387 ring->desc_cb[next_to_use].dma,
388 ring->desc_cb[next_to_use].length,
394 dev_kfree_skb_any(skb);
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.
406 return NETDEV_TX_BUSY;
410 * hns_nic_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
411 * @data: pointer to the start of the headers
412 * @max: total length of section to find headers in
414 * This function is meant to determine the length of headers that will
415 * be recognized by hardware for LRO, GRO, and RSC offloads. The main
416 * motivation of doing this is to only perform one pull for IPv4 TCP
417 * packets so that we can do basic things like calculating the gso_size
418 * based on the average data per packet.
420 static unsigned int hns_nic_get_headlen(unsigned char *data, u32 flag,
421 unsigned int max_size)
423 unsigned char *network;
426 /* this should never happen, but better safe than sorry */
427 if (max_size < ETH_HLEN)
430 /* initialize network frame pointer */
433 /* set first protocol and move network header forward */
436 /* handle any vlan tag if present */
437 if (hnae_get_field(flag, HNS_RXD_VLAN_M, HNS_RXD_VLAN_S)
438 == HNS_RX_FLAG_VLAN_PRESENT) {
439 if ((typeof(max_size))(network - data) > (max_size - VLAN_HLEN))
442 network += VLAN_HLEN;
445 /* handle L3 protocols */
446 if (hnae_get_field(flag, HNS_RXD_L3ID_M, HNS_RXD_L3ID_S)
447 == HNS_RX_FLAG_L3ID_IPV4) {
448 if ((typeof(max_size))(network - data) >
449 (max_size - sizeof(struct iphdr)))
452 /* access ihl as a u8 to avoid unaligned access on ia64 */
453 hlen = (network[0] & 0x0F) << 2;
455 /* verify hlen meets minimum size requirements */
456 if (hlen < sizeof(struct iphdr))
457 return network - data;
459 /* record next protocol if header is present */
460 } else if (hnae_get_field(flag, HNS_RXD_L3ID_M, HNS_RXD_L3ID_S)
461 == HNS_RX_FLAG_L3ID_IPV6) {
462 if ((typeof(max_size))(network - data) >
463 (max_size - sizeof(struct ipv6hdr)))
466 /* record next protocol */
467 hlen = sizeof(struct ipv6hdr);
469 return network - data;
472 /* relocate pointer to start of L4 header */
475 /* finally sort out TCP/UDP */
476 if (hnae_get_field(flag, HNS_RXD_L4ID_M, HNS_RXD_L4ID_S)
477 == HNS_RX_FLAG_L4ID_TCP) {
478 if ((typeof(max_size))(network - data) >
479 (max_size - sizeof(struct tcphdr)))
482 /* access doff as a u8 to avoid unaligned access on ia64 */
483 hlen = (network[12] & 0xF0) >> 2;
485 /* verify hlen meets minimum size requirements */
486 if (hlen < sizeof(struct tcphdr))
487 return network - data;
490 } else if (hnae_get_field(flag, HNS_RXD_L4ID_M, HNS_RXD_L4ID_S)
491 == HNS_RX_FLAG_L4ID_UDP) {
492 if ((typeof(max_size))(network - data) >
493 (max_size - sizeof(struct udphdr)))
496 network += sizeof(struct udphdr);
499 /* If everything has gone correctly network should be the
500 * data section of the packet and will be the end of the header.
501 * If not then it probably represents the end of the last recognized
504 if ((typeof(max_size))(network - data) < max_size)
505 return network - data;
510 static void hns_nic_reuse_page(struct sk_buff *skb, int i,
511 struct hnae_ring *ring, int pull_len,
512 struct hnae_desc_cb *desc_cb)
514 struct hnae_desc *desc;
519 twobufs = ((PAGE_SIZE < 8192) &&
520 hnae_buf_size(ring) == HNS_BUFFER_SIZE_2048);
522 desc = &ring->desc[ring->next_to_clean];
523 size = le16_to_cpu(desc->rx.size);
526 truesize = hnae_buf_size(ring);
528 truesize = ALIGN(size, L1_CACHE_BYTES);
529 last_offset = hnae_page_size(ring) - hnae_buf_size(ring);
532 skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
533 size - pull_len, truesize - pull_len);
535 /* avoid re-using remote pages,flag default unreuse */
536 if (unlikely(page_to_nid(desc_cb->priv) != numa_node_id()))
540 /* if we are only owner of page we can reuse it */
541 if (likely(page_count(desc_cb->priv) == 1)) {
542 /* flip page offset to other buffer */
543 desc_cb->page_offset ^= truesize;
545 desc_cb->reuse_flag = 1;
546 /* bump ref count on page before it is given*/
547 get_page(desc_cb->priv);
552 /* move offset up to the next cache line */
553 desc_cb->page_offset += truesize;
555 if (desc_cb->page_offset <= last_offset) {
556 desc_cb->reuse_flag = 1;
557 /* bump ref count on page before it is given*/
558 get_page(desc_cb->priv);
562 static void get_v2rx_desc_bnum(u32 bnum_flag, int *out_bnum)
564 *out_bnum = hnae_get_field(bnum_flag,
565 HNS_RXD_BUFNUM_M, HNS_RXD_BUFNUM_S) + 1;
568 static void get_rx_desc_bnum(u32 bnum_flag, int *out_bnum)
570 *out_bnum = hnae_get_field(bnum_flag,
571 HNS_RXD_BUFNUM_M, HNS_RXD_BUFNUM_S);
574 static void hns_nic_rx_checksum(struct hns_nic_ring_data *ring_data,
575 struct sk_buff *skb, u32 flag)
577 struct net_device *netdev = ring_data->napi.dev;
581 /* check if RX checksum offload is enabled */
582 if (unlikely(!(netdev->features & NETIF_F_RXCSUM)))
585 /* In hardware, we only support checksum for the following protocols:
587 * 2) TCP(over IPv4 or IPv6),
588 * 3) UDP(over IPv4 or IPv6),
589 * 4) SCTP(over IPv4 or IPv6)
590 * but we support many L3(IPv4, IPv6, MPLS, PPPoE etc) and L4(TCP,
591 * UDP, GRE, SCTP, IGMP, ICMP etc.) protocols.
593 * Hardware limitation:
594 * Our present hardware RX Descriptor lacks L3/L4 checksum "Status &
595 * Error" bit (which usually can be used to indicate whether checksum
596 * was calculated by the hardware and if there was any error encountered
597 * during checksum calculation).
599 * Software workaround:
600 * We do get info within the RX descriptor about the kind of L3/L4
601 * protocol coming in the packet and the error status. These errors
602 * might not just be checksum errors but could be related to version,
603 * length of IPv4, UDP, TCP etc.
604 * Because there is no-way of knowing if it is a L3/L4 error due to bad
605 * checksum or any other L3/L4 error, we will not (cannot) convey
606 * checksum status for such cases to upper stack and will not maintain
607 * the RX L3/L4 checksum counters as well.
610 l3id = hnae_get_field(flag, HNS_RXD_L3ID_M, HNS_RXD_L3ID_S);
611 l4id = hnae_get_field(flag, HNS_RXD_L4ID_M, HNS_RXD_L4ID_S);
613 /* check L3 protocol for which checksum is supported */
614 if ((l3id != HNS_RX_FLAG_L3ID_IPV4) && (l3id != HNS_RX_FLAG_L3ID_IPV6))
617 /* check for any(not just checksum)flagged L3 protocol errors */
618 if (unlikely(hnae_get_bit(flag, HNS_RXD_L3E_B)))
621 /* we do not support checksum of fragmented packets */
622 if (unlikely(hnae_get_bit(flag, HNS_RXD_FRAG_B)))
625 /* check L4 protocol for which checksum is supported */
626 if ((l4id != HNS_RX_FLAG_L4ID_TCP) &&
627 (l4id != HNS_RX_FLAG_L4ID_UDP) &&
628 (l4id != HNS_RX_FLAG_L4ID_SCTP))
631 /* check for any(not just checksum)flagged L4 protocol errors */
632 if (unlikely(hnae_get_bit(flag, HNS_RXD_L4E_B)))
635 /* now, this has to be a packet with valid RX checksum */
636 skb->ip_summed = CHECKSUM_UNNECESSARY;
639 static int hns_nic_poll_rx_skb(struct hns_nic_ring_data *ring_data,
640 struct sk_buff **out_skb, int *out_bnum)
642 struct hnae_ring *ring = ring_data->ring;
643 struct net_device *ndev = ring_data->napi.dev;
644 struct hns_nic_priv *priv = netdev_priv(ndev);
646 struct hnae_desc *desc;
647 struct hnae_desc_cb *desc_cb;
653 desc = &ring->desc[ring->next_to_clean];
654 desc_cb = &ring->desc_cb[ring->next_to_clean];
658 va = (unsigned char *)desc_cb->buf + desc_cb->page_offset;
660 /* prefetch first cache line of first page */
662 #if L1_CACHE_BYTES < 128
663 prefetch(va + L1_CACHE_BYTES);
666 skb = *out_skb = napi_alloc_skb(&ring_data->napi,
668 if (unlikely(!skb)) {
669 netdev_err(ndev, "alloc rx skb fail\n");
670 ring->stats.sw_err_cnt++;
674 prefetchw(skb->data);
675 length = le16_to_cpu(desc->rx.pkt_len);
676 bnum_flag = le32_to_cpu(desc->rx.ipoff_bnum_pid_flag);
677 priv->ops.get_rxd_bnum(bnum_flag, &bnum);
680 if (length <= HNS_RX_HEAD_SIZE) {
681 memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
683 /* we can reuse buffer as-is, just make sure it is local */
684 if (likely(page_to_nid(desc_cb->priv) == numa_node_id()))
685 desc_cb->reuse_flag = 1;
686 else /* this page cannot be reused so discard it */
687 put_page(desc_cb->priv);
689 ring_ptr_move_fw(ring, next_to_clean);
691 if (unlikely(bnum != 1)) { /* check err*/
696 ring->stats.seg_pkt_cnt++;
698 pull_len = hns_nic_get_headlen(va, bnum_flag, HNS_RX_HEAD_SIZE);
699 memcpy(__skb_put(skb, pull_len), va,
700 ALIGN(pull_len, sizeof(long)));
702 hns_nic_reuse_page(skb, 0, ring, pull_len, desc_cb);
703 ring_ptr_move_fw(ring, next_to_clean);
705 if (unlikely(bnum >= (int)MAX_SKB_FRAGS)) { /* check err*/
709 for (i = 1; i < bnum; i++) {
710 desc = &ring->desc[ring->next_to_clean];
711 desc_cb = &ring->desc_cb[ring->next_to_clean];
713 hns_nic_reuse_page(skb, i, ring, 0, desc_cb);
714 ring_ptr_move_fw(ring, next_to_clean);
718 /* check except process, free skb and jump the desc */
719 if (unlikely((!bnum) || (bnum > ring->max_desc_num_per_pkt))) {
721 *out_bnum = *out_bnum ? *out_bnum : 1; /* ntc moved,cannot 0*/
722 netdev_err(ndev, "invalid bnum(%d,%d,%d,%d),%016llx,%016llx\n",
723 bnum, ring->max_desc_num_per_pkt,
724 length, (int)MAX_SKB_FRAGS,
725 ((u64 *)desc)[0], ((u64 *)desc)[1]);
726 ring->stats.err_bd_num++;
727 dev_kfree_skb_any(skb);
731 bnum_flag = le32_to_cpu(desc->rx.ipoff_bnum_pid_flag);
733 if (unlikely(!hnae_get_bit(bnum_flag, HNS_RXD_VLD_B))) {
734 netdev_err(ndev, "no valid bd,%016llx,%016llx\n",
735 ((u64 *)desc)[0], ((u64 *)desc)[1]);
736 ring->stats.non_vld_descs++;
737 dev_kfree_skb_any(skb);
741 if (unlikely((!desc->rx.pkt_len) ||
742 hnae_get_bit(bnum_flag, HNS_RXD_DROP_B))) {
743 ring->stats.err_pkt_len++;
744 dev_kfree_skb_any(skb);
748 if (unlikely(hnae_get_bit(bnum_flag, HNS_RXD_L2E_B))) {
749 ring->stats.l2_err++;
750 dev_kfree_skb_any(skb);
754 ring->stats.rx_pkts++;
755 ring->stats.rx_bytes += skb->len;
757 /* indicate to upper stack if our hardware has already calculated
760 hns_nic_rx_checksum(ring_data, skb, bnum_flag);
766 hns_nic_alloc_rx_buffers(struct hns_nic_ring_data *ring_data, int cleand_count)
769 struct hnae_desc_cb res_cbs;
770 struct hnae_desc_cb *desc_cb;
771 struct hnae_ring *ring = ring_data->ring;
772 struct net_device *ndev = ring_data->napi.dev;
774 for (i = 0; i < cleand_count; i++) {
775 desc_cb = &ring->desc_cb[ring->next_to_use];
776 if (desc_cb->reuse_flag) {
777 ring->stats.reuse_pg_cnt++;
778 hnae_reuse_buffer(ring, ring->next_to_use);
780 ret = hnae_reserve_buffer_map(ring, &res_cbs);
782 ring->stats.sw_err_cnt++;
783 netdev_err(ndev, "hnae reserve buffer map failed.\n");
786 hnae_replace_buffer(ring, ring->next_to_use, &res_cbs);
789 ring_ptr_move_fw(ring, next_to_use);
792 wmb(); /* make all data has been write before submit */
793 writel_relaxed(i, ring->io_base + RCB_REG_HEAD);
796 /* return error number for error or number of desc left to take
798 static void hns_nic_rx_up_pro(struct hns_nic_ring_data *ring_data,
801 struct net_device *ndev = ring_data->napi.dev;
803 skb->protocol = eth_type_trans(skb, ndev);
804 (void)napi_gro_receive(&ring_data->napi, skb);
807 static int hns_desc_unused(struct hnae_ring *ring)
809 int ntc = ring->next_to_clean;
810 int ntu = ring->next_to_use;
812 return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
815 #define HNS_LOWEST_LATENCY_RATE 27 /* 27 MB/s */
816 #define HNS_LOW_LATENCY_RATE 80 /* 80 MB/s */
818 #define HNS_COAL_BDNUM 3
820 static u32 hns_coal_rx_bdnum(struct hnae_ring *ring)
822 bool coal_enable = ring->q->handle->coal_adapt_en;
825 ring->coal_last_rx_bytes > HNS_LOWEST_LATENCY_RATE)
826 return HNS_COAL_BDNUM;
831 static void hns_update_rx_rate(struct hnae_ring *ring)
833 bool coal_enable = ring->q->handle->coal_adapt_en;
838 time_before(jiffies, ring->coal_last_jiffies + (HZ >> 4)))
841 /* ring->stats.rx_bytes overflowed */
842 if (ring->coal_last_rx_bytes > ring->stats.rx_bytes) {
843 ring->coal_last_rx_bytes = ring->stats.rx_bytes;
844 ring->coal_last_jiffies = jiffies;
848 total_bytes = ring->stats.rx_bytes - ring->coal_last_rx_bytes;
849 time_passed_ms = jiffies_to_msecs(jiffies - ring->coal_last_jiffies);
850 do_div(total_bytes, time_passed_ms);
851 ring->coal_rx_rate = total_bytes >> 10;
853 ring->coal_last_rx_bytes = ring->stats.rx_bytes;
854 ring->coal_last_jiffies = jiffies;
858 * smooth_alg - smoothing algrithm for adjusting coalesce parameter
860 static u32 smooth_alg(u32 new_param, u32 old_param)
862 u32 gap = (new_param > old_param) ? new_param - old_param
863 : old_param - new_param;
868 if (new_param > old_param)
869 return old_param + gap;
871 return old_param - gap;
875 * hns_nic_adp_coalesce - self adapte coalesce according to rx rate
876 * @ring_data: pointer to hns_nic_ring_data
878 static void hns_nic_adpt_coalesce(struct hns_nic_ring_data *ring_data)
880 struct hnae_ring *ring = ring_data->ring;
881 struct hnae_handle *handle = ring->q->handle;
882 u32 new_coal_param, old_coal_param = ring->coal_param;
884 if (ring->coal_rx_rate < HNS_LOWEST_LATENCY_RATE)
885 new_coal_param = HNAE_LOWEST_LATENCY_COAL_PARAM;
886 else if (ring->coal_rx_rate < HNS_LOW_LATENCY_RATE)
887 new_coal_param = HNAE_LOW_LATENCY_COAL_PARAM;
889 new_coal_param = HNAE_BULK_LATENCY_COAL_PARAM;
891 if (new_coal_param == old_coal_param &&
892 new_coal_param == handle->coal_param)
895 new_coal_param = smooth_alg(new_coal_param, old_coal_param);
896 ring->coal_param = new_coal_param;
899 * Because all ring in one port has one coalesce param, when one ring
900 * calculate its own coalesce param, it cannot write to hardware at
901 * once. There are three conditions as follows:
902 * 1. current ring's coalesce param is larger than the hardware.
903 * 2. or ring which adapt last time can change again.
906 if (new_coal_param == handle->coal_param) {
907 handle->coal_last_jiffies = jiffies;
908 handle->coal_ring_idx = ring_data->queue_index;
909 } else if (new_coal_param > handle->coal_param ||
910 handle->coal_ring_idx == ring_data->queue_index ||
911 time_after(jiffies, handle->coal_last_jiffies + (HZ >> 4))) {
912 handle->dev->ops->set_coalesce_usecs(handle,
914 handle->dev->ops->set_coalesce_frames(handle,
916 handle->coal_param = new_coal_param;
917 handle->coal_ring_idx = ring_data->queue_index;
918 handle->coal_last_jiffies = jiffies;
922 static int hns_nic_rx_poll_one(struct hns_nic_ring_data *ring_data,
925 struct hnae_ring *ring = ring_data->ring;
928 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
929 int recv_pkts, recv_bds, clean_count, err;
930 int unused_count = hns_desc_unused(ring);
932 num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
933 rmb(); /* make sure num taken effect before the other data is touched */
935 recv_pkts = 0, recv_bds = 0, clean_count = 0;
938 while (recv_pkts < budget && recv_bds < num) {
939 /* reuse or realloc buffers */
940 if (clean_count + unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
941 hns_nic_alloc_rx_buffers(ring_data,
942 clean_count + unused_count);
944 unused_count = hns_desc_unused(ring);
948 err = hns_nic_poll_rx_skb(ring_data, &skb, &bnum);
949 if (unlikely(!skb)) /* this fault cannot be repaired */
954 if (unlikely(err)) { /* do jump the err */
959 /* do update ip stack process*/
960 ((void (*)(struct hns_nic_ring_data *, struct sk_buff *))v)(
966 /* make all data has been write before submit */
967 if (clean_count + unused_count > 0)
968 hns_nic_alloc_rx_buffers(ring_data,
969 clean_count + unused_count);
974 static bool hns_nic_rx_fini_pro(struct hns_nic_ring_data *ring_data)
976 struct hnae_ring *ring = ring_data->ring;
980 hns_update_rx_rate(ring);
982 /* for hardware bug fixed */
983 ring_data->ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
984 num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
986 if (num <= hns_coal_rx_bdnum(ring)) {
987 if (ring->q->handle->coal_adapt_en)
988 hns_nic_adpt_coalesce(ring_data);
992 ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1001 static bool hns_nic_rx_fini_pro_v2(struct hns_nic_ring_data *ring_data)
1003 struct hnae_ring *ring = ring_data->ring;
1006 hns_update_rx_rate(ring);
1007 num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
1009 if (num <= hns_coal_rx_bdnum(ring)) {
1010 if (ring->q->handle->coal_adapt_en)
1011 hns_nic_adpt_coalesce(ring_data);
1019 static inline void hns_nic_reclaim_one_desc(struct hnae_ring *ring,
1020 int *bytes, int *pkts)
1022 struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
1024 (*pkts) += (desc_cb->type == DESC_TYPE_SKB);
1025 (*bytes) += desc_cb->length;
1026 /* desc_cb will be cleaned, after hnae_free_buffer_detach*/
1027 hnae_free_buffer_detach(ring, ring->next_to_clean);
1029 ring_ptr_move_fw(ring, next_to_clean);
1032 static int is_valid_clean_head(struct hnae_ring *ring, int h)
1034 int u = ring->next_to_use;
1035 int c = ring->next_to_clean;
1037 if (unlikely(h > ring->desc_num))
1040 assert(u > 0 && u < ring->desc_num);
1041 assert(c > 0 && c < ring->desc_num);
1042 assert(u != c && h != c); /* must be checked before call this func */
1044 return u > c ? (h > c && h <= u) : (h > c || h <= u);
1047 /* netif_tx_lock will turn down the performance, set only when necessary */
1048 #ifdef CONFIG_NET_POLL_CONTROLLER
1049 #define NETIF_TX_LOCK(ring) spin_lock(&(ring)->lock)
1050 #define NETIF_TX_UNLOCK(ring) spin_unlock(&(ring)->lock)
1052 #define NETIF_TX_LOCK(ring)
1053 #define NETIF_TX_UNLOCK(ring)
1056 /* reclaim all desc in one budget
1057 * return error or number of desc left
1059 static int hns_nic_tx_poll_one(struct hns_nic_ring_data *ring_data,
1060 int budget, void *v)
1062 struct hnae_ring *ring = ring_data->ring;
1063 struct net_device *ndev = ring_data->napi.dev;
1064 struct netdev_queue *dev_queue;
1065 struct hns_nic_priv *priv = netdev_priv(ndev);
1069 NETIF_TX_LOCK(ring);
1071 head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1072 rmb(); /* make sure head is ready before touch any data */
1074 if (is_ring_empty(ring) || head == ring->next_to_clean) {
1075 NETIF_TX_UNLOCK(ring);
1076 return 0; /* no data to poll */
1079 if (!is_valid_clean_head(ring, head)) {
1080 netdev_err(ndev, "wrong head (%d, %d-%d)\n", head,
1081 ring->next_to_use, ring->next_to_clean);
1082 ring->stats.io_err_cnt++;
1083 NETIF_TX_UNLOCK(ring);
1089 while (head != ring->next_to_clean) {
1090 hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
1091 /* issue prefetch for next Tx descriptor */
1092 prefetch(&ring->desc_cb[ring->next_to_clean]);
1095 NETIF_TX_UNLOCK(ring);
1097 dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
1098 netdev_tx_completed_queue(dev_queue, pkts, bytes);
1100 if (unlikely(priv->link && !netif_carrier_ok(ndev)))
1101 netif_carrier_on(ndev);
1103 if (unlikely(pkts && netif_carrier_ok(ndev) &&
1104 (ring_space(ring) >= ring->max_desc_num_per_pkt * 2))) {
1105 /* Make sure that anybody stopping the queue after this
1106 * sees the new next_to_clean.
1109 if (netif_tx_queue_stopped(dev_queue) &&
1110 !test_bit(NIC_STATE_DOWN, &priv->state)) {
1111 netif_tx_wake_queue(dev_queue);
1112 ring->stats.restart_queue++;
1118 static bool hns_nic_tx_fini_pro(struct hns_nic_ring_data *ring_data)
1120 struct hnae_ring *ring = ring_data->ring;
1123 ring_data->ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
1125 head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1127 if (head != ring->next_to_clean) {
1128 ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1129 ring_data->ring, 1);
1137 static bool hns_nic_tx_fini_pro_v2(struct hns_nic_ring_data *ring_data)
1139 struct hnae_ring *ring = ring_data->ring;
1140 int head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1142 if (head == ring->next_to_clean)
1148 static void hns_nic_tx_clr_all_bufs(struct hns_nic_ring_data *ring_data)
1150 struct hnae_ring *ring = ring_data->ring;
1151 struct net_device *ndev = ring_data->napi.dev;
1152 struct netdev_queue *dev_queue;
1156 NETIF_TX_LOCK(ring);
1158 head = ring->next_to_use; /* ntu :soft setted ring position*/
1161 while (head != ring->next_to_clean)
1162 hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
1164 NETIF_TX_UNLOCK(ring);
1166 dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
1167 netdev_tx_reset_queue(dev_queue);
1170 static int hns_nic_common_poll(struct napi_struct *napi, int budget)
1172 int clean_complete = 0;
1173 struct hns_nic_ring_data *ring_data =
1174 container_of(napi, struct hns_nic_ring_data, napi);
1175 struct hnae_ring *ring = ring_data->ring;
1178 clean_complete += ring_data->poll_one(
1179 ring_data, budget - clean_complete,
1180 ring_data->ex_process);
1182 if (clean_complete < budget) {
1183 if (ring_data->fini_process(ring_data)) {
1184 napi_complete(napi);
1185 ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
1191 return clean_complete;
1194 static irqreturn_t hns_irq_handle(int irq, void *dev)
1196 struct hns_nic_ring_data *ring_data = (struct hns_nic_ring_data *)dev;
1198 ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1199 ring_data->ring, 1);
1200 napi_schedule(&ring_data->napi);
1206 *hns_nic_adjust_link - adjust net work mode by the phy stat or new param
1209 static void hns_nic_adjust_link(struct net_device *ndev)
1211 struct hns_nic_priv *priv = netdev_priv(ndev);
1212 struct hnae_handle *h = priv->ae_handle;
1216 h->dev->ops->adjust_link(h, ndev->phydev->speed,
1217 ndev->phydev->duplex);
1218 state = ndev->phydev->link;
1220 state = state && h->dev->ops->get_status(h);
1222 if (state != priv->link) {
1224 netif_carrier_on(ndev);
1225 netif_tx_wake_all_queues(ndev);
1226 netdev_info(ndev, "link up\n");
1228 netif_carrier_off(ndev);
1229 netdev_info(ndev, "link down\n");
1236 *hns_nic_init_phy - init phy
1239 * Return 0 on success, negative on failure
1241 int hns_nic_init_phy(struct net_device *ndev, struct hnae_handle *h)
1243 struct phy_device *phy_dev = h->phy_dev;
1249 if (h->phy_if != PHY_INTERFACE_MODE_XGMII) {
1250 phy_dev->dev_flags = 0;
1252 ret = phy_connect_direct(ndev, phy_dev, hns_nic_adjust_link,
1255 ret = phy_attach_direct(ndev, phy_dev, 0, h->phy_if);
1260 phy_dev->supported &= h->if_support;
1261 phy_dev->advertising = phy_dev->supported;
1263 if (h->phy_if == PHY_INTERFACE_MODE_XGMII)
1264 phy_dev->autoneg = false;
1269 static int hns_nic_ring_open(struct net_device *netdev, int idx)
1271 struct hns_nic_priv *priv = netdev_priv(netdev);
1272 struct hnae_handle *h = priv->ae_handle;
1274 napi_enable(&priv->ring_data[idx].napi);
1276 enable_irq(priv->ring_data[idx].ring->irq);
1277 h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 0);
1282 static int hns_nic_net_set_mac_address(struct net_device *ndev, void *p)
1284 struct hns_nic_priv *priv = netdev_priv(ndev);
1285 struct hnae_handle *h = priv->ae_handle;
1286 struct sockaddr *mac_addr = p;
1289 if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1290 return -EADDRNOTAVAIL;
1292 ret = h->dev->ops->set_mac_addr(h, mac_addr->sa_data);
1294 netdev_err(ndev, "set_mac_address fail, ret=%d!\n", ret);
1298 memcpy(ndev->dev_addr, mac_addr->sa_data, ndev->addr_len);
1303 void hns_nic_update_stats(struct net_device *netdev)
1305 struct hns_nic_priv *priv = netdev_priv(netdev);
1306 struct hnae_handle *h = priv->ae_handle;
1308 h->dev->ops->update_stats(h, &netdev->stats);
1311 /* set mac addr if it is configed. or leave it to the AE driver */
1312 static void hns_init_mac_addr(struct net_device *ndev)
1314 struct hns_nic_priv *priv = netdev_priv(ndev);
1316 if (!device_get_mac_address(priv->dev, ndev->dev_addr, ETH_ALEN)) {
1317 eth_hw_addr_random(ndev);
1318 dev_warn(priv->dev, "No valid mac, use random mac %pM",
1323 static void hns_nic_ring_close(struct net_device *netdev, int idx)
1325 struct hns_nic_priv *priv = netdev_priv(netdev);
1326 struct hnae_handle *h = priv->ae_handle;
1328 h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 1);
1329 disable_irq(priv->ring_data[idx].ring->irq);
1331 napi_disable(&priv->ring_data[idx].napi);
1334 static int hns_nic_init_affinity_mask(int q_num, int ring_idx,
1335 struct hnae_ring *ring, cpumask_t *mask)
1339 /* Diffrent irq banlance between 16core and 32core.
1340 * The cpu mask set by ring index according to the ring flag
1341 * which indicate the ring is tx or rx.
1343 if (q_num == num_possible_cpus()) {
1344 if (is_tx_ring(ring))
1347 cpu = ring_idx - q_num;
1349 if (is_tx_ring(ring))
1352 cpu = (ring_idx - q_num) * 2 + 1;
1355 cpumask_clear(mask);
1356 cpumask_set_cpu(cpu, mask);
1361 static int hns_nic_init_irq(struct hns_nic_priv *priv)
1363 struct hnae_handle *h = priv->ae_handle;
1364 struct hns_nic_ring_data *rd;
1369 for (i = 0; i < h->q_num * 2; i++) {
1370 rd = &priv->ring_data[i];
1372 if (rd->ring->irq_init_flag == RCB_IRQ_INITED)
1375 snprintf(rd->ring->ring_name, RCB_RING_NAME_LEN,
1376 "%s-%s%d", priv->netdev->name,
1377 (is_tx_ring(rd->ring) ? "tx" : "rx"), rd->queue_index);
1379 rd->ring->ring_name[RCB_RING_NAME_LEN - 1] = '\0';
1381 ret = request_irq(rd->ring->irq,
1382 hns_irq_handle, 0, rd->ring->ring_name, rd);
1384 netdev_err(priv->netdev, "request irq(%d) fail\n",
1388 disable_irq(rd->ring->irq);
1390 cpu = hns_nic_init_affinity_mask(h->q_num, i,
1391 rd->ring, &rd->mask);
1393 if (cpu_online(cpu))
1394 irq_set_affinity_hint(rd->ring->irq,
1397 rd->ring->irq_init_flag = RCB_IRQ_INITED;
1403 static int hns_nic_net_up(struct net_device *ndev)
1405 struct hns_nic_priv *priv = netdev_priv(ndev);
1406 struct hnae_handle *h = priv->ae_handle;
1410 ret = hns_nic_init_irq(priv);
1412 netdev_err(ndev, "hns init irq failed! ret=%d\n", ret);
1416 for (i = 0; i < h->q_num * 2; i++) {
1417 ret = hns_nic_ring_open(ndev, i);
1419 goto out_has_some_queues;
1422 ret = h->dev->ops->set_mac_addr(h, ndev->dev_addr);
1424 goto out_set_mac_addr_err;
1426 ret = h->dev->ops->start ? h->dev->ops->start(h) : 0;
1431 phy_start(ndev->phydev);
1433 clear_bit(NIC_STATE_DOWN, &priv->state);
1434 (void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
1439 netif_stop_queue(ndev);
1440 out_set_mac_addr_err:
1441 out_has_some_queues:
1442 for (j = i - 1; j >= 0; j--)
1443 hns_nic_ring_close(ndev, j);
1445 set_bit(NIC_STATE_DOWN, &priv->state);
1450 static void hns_nic_net_down(struct net_device *ndev)
1453 struct hnae_ae_ops *ops;
1454 struct hns_nic_priv *priv = netdev_priv(ndev);
1456 if (test_and_set_bit(NIC_STATE_DOWN, &priv->state))
1459 (void)del_timer_sync(&priv->service_timer);
1460 netif_tx_stop_all_queues(ndev);
1461 netif_carrier_off(ndev);
1462 netif_tx_disable(ndev);
1466 phy_stop(ndev->phydev);
1468 ops = priv->ae_handle->dev->ops;
1471 ops->stop(priv->ae_handle);
1473 netif_tx_stop_all_queues(ndev);
1475 for (i = priv->ae_handle->q_num - 1; i >= 0; i--) {
1476 hns_nic_ring_close(ndev, i);
1477 hns_nic_ring_close(ndev, i + priv->ae_handle->q_num);
1479 /* clean tx buffers*/
1480 hns_nic_tx_clr_all_bufs(priv->ring_data + i);
1484 void hns_nic_net_reset(struct net_device *ndev)
1486 struct hns_nic_priv *priv = netdev_priv(ndev);
1487 struct hnae_handle *handle = priv->ae_handle;
1489 while (test_and_set_bit(NIC_STATE_RESETTING, &priv->state))
1490 usleep_range(1000, 2000);
1492 (void)hnae_reinit_handle(handle);
1494 clear_bit(NIC_STATE_RESETTING, &priv->state);
1497 void hns_nic_net_reinit(struct net_device *netdev)
1499 struct hns_nic_priv *priv = netdev_priv(netdev);
1500 enum hnae_port_type type = priv->ae_handle->port_type;
1502 netif_trans_update(priv->netdev);
1503 while (test_and_set_bit(NIC_STATE_REINITING, &priv->state))
1504 usleep_range(1000, 2000);
1506 hns_nic_net_down(netdev);
1508 /* Only do hns_nic_net_reset in debug mode
1509 * because of hardware limitation.
1511 if (type == HNAE_PORT_DEBUG)
1512 hns_nic_net_reset(netdev);
1514 (void)hns_nic_net_up(netdev);
1515 clear_bit(NIC_STATE_REINITING, &priv->state);
1518 static int hns_nic_net_open(struct net_device *ndev)
1520 struct hns_nic_priv *priv = netdev_priv(ndev);
1521 struct hnae_handle *h = priv->ae_handle;
1524 if (test_bit(NIC_STATE_TESTING, &priv->state))
1528 netif_carrier_off(ndev);
1530 ret = netif_set_real_num_tx_queues(ndev, h->q_num);
1532 netdev_err(ndev, "netif_set_real_num_tx_queues fail, ret=%d!\n",
1537 ret = netif_set_real_num_rx_queues(ndev, h->q_num);
1540 "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
1544 ret = hns_nic_net_up(ndev);
1547 "hns net up fail, ret=%d!\n", ret);
1554 static int hns_nic_net_stop(struct net_device *ndev)
1556 hns_nic_net_down(ndev);
1561 static void hns_tx_timeout_reset(struct hns_nic_priv *priv);
1562 static void hns_nic_net_timeout(struct net_device *ndev)
1564 struct hns_nic_priv *priv = netdev_priv(ndev);
1566 hns_tx_timeout_reset(priv);
1569 static int hns_nic_do_ioctl(struct net_device *netdev, struct ifreq *ifr,
1572 struct phy_device *phy_dev = netdev->phydev;
1574 if (!netif_running(netdev))
1580 return phy_mii_ioctl(phy_dev, ifr, cmd);
1583 /* use only for netconsole to poll with the device without interrupt */
1584 #ifdef CONFIG_NET_POLL_CONTROLLER
1585 void hns_nic_poll_controller(struct net_device *ndev)
1587 struct hns_nic_priv *priv = netdev_priv(ndev);
1588 unsigned long flags;
1591 local_irq_save(flags);
1592 for (i = 0; i < priv->ae_handle->q_num * 2; i++)
1593 napi_schedule(&priv->ring_data[i].napi);
1594 local_irq_restore(flags);
1598 static netdev_tx_t hns_nic_net_xmit(struct sk_buff *skb,
1599 struct net_device *ndev)
1601 struct hns_nic_priv *priv = netdev_priv(ndev);
1603 assert(skb->queue_mapping < ndev->ae_handle->q_num);
1605 return hns_nic_net_xmit_hw(ndev, skb,
1606 &tx_ring_data(priv, skb->queue_mapping));
1609 static void hns_nic_drop_rx_fetch(struct hns_nic_ring_data *ring_data,
1610 struct sk_buff *skb)
1612 dev_kfree_skb_any(skb);
1615 #define HNS_LB_TX_RING 0
1616 static struct sk_buff *hns_assemble_skb(struct net_device *ndev)
1618 struct sk_buff *skb;
1619 struct ethhdr *ethhdr;
1622 /* allocate test skb */
1623 skb = alloc_skb(64, GFP_KERNEL);
1629 memset(skb->data, 0xFF, skb->len);
1631 /* must be tcp/ip package */
1632 ethhdr = (struct ethhdr *)skb->data;
1633 ethhdr->h_proto = htons(ETH_P_IP);
1635 frame_len = skb->len & (~1ul);
1636 memset(&skb->data[frame_len / 2], 0xAA,
1639 skb->queue_mapping = HNS_LB_TX_RING;
1644 static int hns_enable_serdes_lb(struct net_device *ndev)
1646 struct hns_nic_priv *priv = netdev_priv(ndev);
1647 struct hnae_handle *h = priv->ae_handle;
1648 struct hnae_ae_ops *ops = h->dev->ops;
1652 ret = ops->set_loopback(h, MAC_INTERNALLOOP_SERDES, 1);
1656 ret = ops->start ? ops->start(h) : 0;
1660 /* link adjust duplex*/
1661 if (h->phy_if != PHY_INTERFACE_MODE_XGMII)
1667 ops->adjust_link(h, speed, duplex);
1669 /* wait h/w ready */
1675 static void hns_disable_serdes_lb(struct net_device *ndev)
1677 struct hns_nic_priv *priv = netdev_priv(ndev);
1678 struct hnae_handle *h = priv->ae_handle;
1679 struct hnae_ae_ops *ops = h->dev->ops;
1682 ops->set_loopback(h, MAC_INTERNALLOOP_SERDES, 0);
1686 *hns_nic_clear_all_rx_fetch - clear the chip fetched descriptions. The
1687 *function as follows:
1688 * 1. if one rx ring has found the page_offset is not equal 0 between head
1689 * and tail, it means that the chip fetched the wrong descs for the ring
1690 * which buffer size is 4096.
1691 * 2. we set the chip serdes loopback and set rss indirection to the ring.
1692 * 3. construct 64-bytes ip broadcast packages, wait the associated rx ring
1693 * recieving all packages and it will fetch new descriptions.
1694 * 4. recover to the original state.
1698 static int hns_nic_clear_all_rx_fetch(struct net_device *ndev)
1700 struct hns_nic_priv *priv = netdev_priv(ndev);
1701 struct hnae_handle *h = priv->ae_handle;
1702 struct hnae_ae_ops *ops = h->dev->ops;
1703 struct hns_nic_ring_data *rd;
1704 struct hnae_ring *ring;
1705 struct sk_buff *skb;
1716 /* alloc indir memory */
1717 indir_size = ops->get_rss_indir_size(h) * sizeof(*org_indir);
1718 org_indir = kzalloc(indir_size, GFP_KERNEL);
1722 /* store the orginal indirection */
1723 ops->get_rss(h, org_indir, NULL, NULL);
1725 cur_indir = kzalloc(indir_size, GFP_KERNEL);
1728 goto cur_indir_alloc_err;
1732 if (hns_enable_serdes_lb(ndev)) {
1734 goto enable_serdes_lb_err;
1737 /* foreach every rx ring to clear fetch desc */
1738 for (i = 0; i < h->q_num; i++) {
1739 ring = &h->qs[i]->rx_ring;
1740 head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1741 tail = readl_relaxed(ring->io_base + RCB_REG_TAIL);
1743 fetch_num = ring_dist(ring, head, tail);
1745 while (head != tail) {
1746 if (ring->desc_cb[head].page_offset != 0) {
1752 if (head == ring->desc_num)
1757 for (j = 0; j < indir_size / sizeof(*org_indir); j++)
1759 ops->set_rss(h, cur_indir, NULL, 0);
1761 for (j = 0; j < fetch_num; j++) {
1762 /* alloc one skb and init */
1763 skb = hns_assemble_skb(ndev);
1766 rd = &tx_ring_data(priv, skb->queue_mapping);
1767 hns_nic_net_xmit_hw(ndev, skb, rd);
1770 while (retry_times++ < 10) {
1773 rd = &rx_ring_data(priv, i);
1774 if (rd->poll_one(rd, fetch_num,
1775 hns_nic_drop_rx_fetch))
1780 while (retry_times++ < 10) {
1782 /* clean tx ring 0 send package */
1783 rd = &tx_ring_data(priv,
1785 if (rd->poll_one(rd, fetch_num, NULL))
1793 /* restore everything */
1794 ops->set_rss(h, org_indir, NULL, 0);
1795 hns_disable_serdes_lb(ndev);
1796 enable_serdes_lb_err:
1798 cur_indir_alloc_err:
1804 static int hns_nic_change_mtu(struct net_device *ndev, int new_mtu)
1806 struct hns_nic_priv *priv = netdev_priv(ndev);
1807 struct hnae_handle *h = priv->ae_handle;
1808 bool if_running = netif_running(ndev);
1811 /* MTU < 68 is an error and causes problems on some kernels */
1816 if (new_mtu == ndev->mtu)
1819 if (!h->dev->ops->set_mtu)
1823 (void)hns_nic_net_stop(ndev);
1827 if (priv->enet_ver != AE_VERSION_1 &&
1828 ndev->mtu <= BD_SIZE_2048_MAX_MTU &&
1829 new_mtu > BD_SIZE_2048_MAX_MTU) {
1831 hnae_reinit_all_ring_desc(h);
1833 /* clear the package which the chip has fetched */
1834 ret = hns_nic_clear_all_rx_fetch(ndev);
1836 /* the page offset must be consist with desc */
1837 hnae_reinit_all_ring_page_off(h);
1840 netdev_err(ndev, "clear the fetched desc fail\n");
1845 ret = h->dev->ops->set_mtu(h, new_mtu);
1847 netdev_err(ndev, "set mtu fail, return value %d\n",
1852 /* finally, set new mtu to netdevice */
1853 ndev->mtu = new_mtu;
1857 if (hns_nic_net_open(ndev)) {
1858 netdev_err(ndev, "hns net open fail\n");
1866 static int hns_nic_set_features(struct net_device *netdev,
1867 netdev_features_t features)
1869 struct hns_nic_priv *priv = netdev_priv(netdev);
1871 switch (priv->enet_ver) {
1873 if (features & (NETIF_F_TSO | NETIF_F_TSO6))
1874 netdev_info(netdev, "enet v1 do not support tso!\n");
1877 if (features & (NETIF_F_TSO | NETIF_F_TSO6)) {
1878 priv->ops.fill_desc = fill_tso_desc;
1879 priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tso;
1880 /* The chip only support 7*4096 */
1881 netif_set_gso_max_size(netdev, 7 * 4096);
1883 priv->ops.fill_desc = fill_v2_desc;
1884 priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
1888 netdev->features = features;
1892 static netdev_features_t hns_nic_fix_features(
1893 struct net_device *netdev, netdev_features_t features)
1895 struct hns_nic_priv *priv = netdev_priv(netdev);
1897 switch (priv->enet_ver) {
1899 features &= ~(NETIF_F_TSO | NETIF_F_TSO6 |
1900 NETIF_F_HW_VLAN_CTAG_FILTER);
1908 static int hns_nic_uc_sync(struct net_device *netdev, const unsigned char *addr)
1910 struct hns_nic_priv *priv = netdev_priv(netdev);
1911 struct hnae_handle *h = priv->ae_handle;
1913 if (h->dev->ops->add_uc_addr)
1914 return h->dev->ops->add_uc_addr(h, addr);
1919 static int hns_nic_uc_unsync(struct net_device *netdev,
1920 const unsigned char *addr)
1922 struct hns_nic_priv *priv = netdev_priv(netdev);
1923 struct hnae_handle *h = priv->ae_handle;
1925 if (h->dev->ops->rm_uc_addr)
1926 return h->dev->ops->rm_uc_addr(h, addr);
1932 * nic_set_multicast_list - set mutl mac address
1933 * @netdev: net device
1938 void hns_set_multicast_list(struct net_device *ndev)
1940 struct hns_nic_priv *priv = netdev_priv(ndev);
1941 struct hnae_handle *h = priv->ae_handle;
1942 struct netdev_hw_addr *ha = NULL;
1945 netdev_err(ndev, "hnae handle is null\n");
1949 if (h->dev->ops->clr_mc_addr)
1950 if (h->dev->ops->clr_mc_addr(h))
1951 netdev_err(ndev, "clear multicast address fail\n");
1953 if (h->dev->ops->set_mc_addr) {
1954 netdev_for_each_mc_addr(ha, ndev)
1955 if (h->dev->ops->set_mc_addr(h, ha->addr))
1956 netdev_err(ndev, "set multicast fail\n");
1960 void hns_nic_set_rx_mode(struct net_device *ndev)
1962 struct hns_nic_priv *priv = netdev_priv(ndev);
1963 struct hnae_handle *h = priv->ae_handle;
1965 if (h->dev->ops->set_promisc_mode) {
1966 if (ndev->flags & IFF_PROMISC)
1967 h->dev->ops->set_promisc_mode(h, 1);
1969 h->dev->ops->set_promisc_mode(h, 0);
1972 hns_set_multicast_list(ndev);
1974 if (__dev_uc_sync(ndev, hns_nic_uc_sync, hns_nic_uc_unsync))
1975 netdev_err(ndev, "sync uc address fail\n");
1978 static void hns_nic_get_stats64(struct net_device *ndev,
1979 struct rtnl_link_stats64 *stats)
1986 struct hns_nic_priv *priv = netdev_priv(ndev);
1987 struct hnae_handle *h = priv->ae_handle;
1989 for (idx = 0; idx < h->q_num; idx++) {
1990 tx_bytes += h->qs[idx]->tx_ring.stats.tx_bytes;
1991 tx_pkts += h->qs[idx]->tx_ring.stats.tx_pkts;
1992 rx_bytes += h->qs[idx]->rx_ring.stats.rx_bytes;
1993 rx_pkts += h->qs[idx]->rx_ring.stats.rx_pkts;
1996 stats->tx_bytes = tx_bytes;
1997 stats->tx_packets = tx_pkts;
1998 stats->rx_bytes = rx_bytes;
1999 stats->rx_packets = rx_pkts;
2001 stats->rx_errors = ndev->stats.rx_errors;
2002 stats->multicast = ndev->stats.multicast;
2003 stats->rx_length_errors = ndev->stats.rx_length_errors;
2004 stats->rx_crc_errors = ndev->stats.rx_crc_errors;
2005 stats->rx_missed_errors = ndev->stats.rx_missed_errors;
2007 stats->tx_errors = ndev->stats.tx_errors;
2008 stats->rx_dropped = ndev->stats.rx_dropped;
2009 stats->tx_dropped = ndev->stats.tx_dropped;
2010 stats->collisions = ndev->stats.collisions;
2011 stats->rx_over_errors = ndev->stats.rx_over_errors;
2012 stats->rx_frame_errors = ndev->stats.rx_frame_errors;
2013 stats->rx_fifo_errors = ndev->stats.rx_fifo_errors;
2014 stats->tx_aborted_errors = ndev->stats.tx_aborted_errors;
2015 stats->tx_carrier_errors = ndev->stats.tx_carrier_errors;
2016 stats->tx_fifo_errors = ndev->stats.tx_fifo_errors;
2017 stats->tx_heartbeat_errors = ndev->stats.tx_heartbeat_errors;
2018 stats->tx_window_errors = ndev->stats.tx_window_errors;
2019 stats->rx_compressed = ndev->stats.rx_compressed;
2020 stats->tx_compressed = ndev->stats.tx_compressed;
2024 hns_nic_select_queue(struct net_device *ndev, struct sk_buff *skb,
2025 void *accel_priv, select_queue_fallback_t fallback)
2027 struct ethhdr *eth_hdr = (struct ethhdr *)skb->data;
2028 struct hns_nic_priv *priv = netdev_priv(ndev);
2030 /* fix hardware broadcast/multicast packets queue loopback */
2031 if (!AE_IS_VER1(priv->enet_ver) &&
2032 is_multicast_ether_addr(eth_hdr->h_dest))
2035 return fallback(ndev, skb);
2038 static const struct net_device_ops hns_nic_netdev_ops = {
2039 .ndo_open = hns_nic_net_open,
2040 .ndo_stop = hns_nic_net_stop,
2041 .ndo_start_xmit = hns_nic_net_xmit,
2042 .ndo_tx_timeout = hns_nic_net_timeout,
2043 .ndo_set_mac_address = hns_nic_net_set_mac_address,
2044 .ndo_change_mtu = hns_nic_change_mtu,
2045 .ndo_do_ioctl = hns_nic_do_ioctl,
2046 .ndo_set_features = hns_nic_set_features,
2047 .ndo_fix_features = hns_nic_fix_features,
2048 .ndo_get_stats64 = hns_nic_get_stats64,
2049 #ifdef CONFIG_NET_POLL_CONTROLLER
2050 .ndo_poll_controller = hns_nic_poll_controller,
2052 .ndo_set_rx_mode = hns_nic_set_rx_mode,
2053 .ndo_select_queue = hns_nic_select_queue,
2056 static void hns_nic_update_link_status(struct net_device *netdev)
2058 struct hns_nic_priv *priv = netdev_priv(netdev);
2060 struct hnae_handle *h = priv->ae_handle;
2063 if (h->phy_if != PHY_INTERFACE_MODE_XGMII)
2066 (void)genphy_read_status(h->phy_dev);
2068 hns_nic_adjust_link(netdev);
2071 /* for dumping key regs*/
2072 static void hns_nic_dump(struct hns_nic_priv *priv)
2074 struct hnae_handle *h = priv->ae_handle;
2075 struct hnae_ae_ops *ops = h->dev->ops;
2076 u32 *data, reg_num, i;
2078 if (ops->get_regs_len && ops->get_regs) {
2079 reg_num = ops->get_regs_len(priv->ae_handle);
2080 reg_num = (reg_num + 3ul) & ~3ul;
2081 data = kcalloc(reg_num, sizeof(u32), GFP_KERNEL);
2083 ops->get_regs(priv->ae_handle, data);
2084 for (i = 0; i < reg_num; i += 4)
2085 pr_info("0x%08x: 0x%08x 0x%08x 0x%08x 0x%08x\n",
2086 i, data[i], data[i + 1],
2087 data[i + 2], data[i + 3]);
2092 for (i = 0; i < h->q_num; i++) {
2093 pr_info("tx_queue%d_next_to_clean:%d\n",
2094 i, h->qs[i]->tx_ring.next_to_clean);
2095 pr_info("tx_queue%d_next_to_use:%d\n",
2096 i, h->qs[i]->tx_ring.next_to_use);
2097 pr_info("rx_queue%d_next_to_clean:%d\n",
2098 i, h->qs[i]->rx_ring.next_to_clean);
2099 pr_info("rx_queue%d_next_to_use:%d\n",
2100 i, h->qs[i]->rx_ring.next_to_use);
2104 /* for resetting subtask */
2105 static void hns_nic_reset_subtask(struct hns_nic_priv *priv)
2107 enum hnae_port_type type = priv->ae_handle->port_type;
2109 if (!test_bit(NIC_STATE2_RESET_REQUESTED, &priv->state))
2111 clear_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
2113 /* If we're already down, removing or resetting, just bail */
2114 if (test_bit(NIC_STATE_DOWN, &priv->state) ||
2115 test_bit(NIC_STATE_REMOVING, &priv->state) ||
2116 test_bit(NIC_STATE_RESETTING, &priv->state))
2120 netdev_info(priv->netdev, "try to reset %s port!\n",
2121 (type == HNAE_PORT_DEBUG ? "debug" : "service"));
2124 /* put off any impending NetWatchDogTimeout */
2125 netif_trans_update(priv->netdev);
2126 hns_nic_net_reinit(priv->netdev);
2131 /* for doing service complete*/
2132 static void hns_nic_service_event_complete(struct hns_nic_priv *priv)
2134 WARN_ON(!test_bit(NIC_STATE_SERVICE_SCHED, &priv->state));
2135 /* make sure to commit the things */
2136 smp_mb__before_atomic();
2137 clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
2140 static void hns_nic_service_task(struct work_struct *work)
2142 struct hns_nic_priv *priv
2143 = container_of(work, struct hns_nic_priv, service_task);
2144 struct hnae_handle *h = priv->ae_handle;
2146 hns_nic_update_link_status(priv->netdev);
2147 h->dev->ops->update_led_status(h);
2148 hns_nic_update_stats(priv->netdev);
2150 hns_nic_reset_subtask(priv);
2151 hns_nic_service_event_complete(priv);
2154 static void hns_nic_task_schedule(struct hns_nic_priv *priv)
2156 if (!test_bit(NIC_STATE_DOWN, &priv->state) &&
2157 !test_bit(NIC_STATE_REMOVING, &priv->state) &&
2158 !test_and_set_bit(NIC_STATE_SERVICE_SCHED, &priv->state))
2159 (void)schedule_work(&priv->service_task);
2162 static void hns_nic_service_timer(struct timer_list *t)
2164 struct hns_nic_priv *priv = from_timer(priv, t, service_timer);
2166 (void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
2168 hns_nic_task_schedule(priv);
2172 * hns_tx_timeout_reset - initiate reset due to Tx timeout
2173 * @priv: driver private struct
2175 static void hns_tx_timeout_reset(struct hns_nic_priv *priv)
2177 /* Do the reset outside of interrupt context */
2178 if (!test_bit(NIC_STATE_DOWN, &priv->state)) {
2179 set_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
2180 netdev_warn(priv->netdev,
2181 "initiating reset due to tx timeout(%llu,0x%lx)\n",
2182 priv->tx_timeout_count, priv->state);
2183 priv->tx_timeout_count++;
2184 hns_nic_task_schedule(priv);
2188 static int hns_nic_init_ring_data(struct hns_nic_priv *priv)
2190 struct hnae_handle *h = priv->ae_handle;
2191 struct hns_nic_ring_data *rd;
2192 bool is_ver1 = AE_IS_VER1(priv->enet_ver);
2195 if (h->q_num > NIC_MAX_Q_PER_VF) {
2196 netdev_err(priv->netdev, "too much queue (%d)\n", h->q_num);
2200 priv->ring_data = kzalloc(array3_size(h->q_num,
2201 sizeof(*priv->ring_data), 2),
2203 if (!priv->ring_data)
2206 for (i = 0; i < h->q_num; i++) {
2207 rd = &priv->ring_data[i];
2208 rd->queue_index = i;
2209 rd->ring = &h->qs[i]->tx_ring;
2210 rd->poll_one = hns_nic_tx_poll_one;
2211 rd->fini_process = is_ver1 ? hns_nic_tx_fini_pro :
2212 hns_nic_tx_fini_pro_v2;
2214 netif_napi_add(priv->netdev, &rd->napi,
2215 hns_nic_common_poll, NIC_TX_CLEAN_MAX_NUM);
2216 rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
2218 for (i = h->q_num; i < h->q_num * 2; i++) {
2219 rd = &priv->ring_data[i];
2220 rd->queue_index = i - h->q_num;
2221 rd->ring = &h->qs[i - h->q_num]->rx_ring;
2222 rd->poll_one = hns_nic_rx_poll_one;
2223 rd->ex_process = hns_nic_rx_up_pro;
2224 rd->fini_process = is_ver1 ? hns_nic_rx_fini_pro :
2225 hns_nic_rx_fini_pro_v2;
2227 netif_napi_add(priv->netdev, &rd->napi,
2228 hns_nic_common_poll, NIC_RX_CLEAN_MAX_NUM);
2229 rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
2235 static void hns_nic_uninit_ring_data(struct hns_nic_priv *priv)
2237 struct hnae_handle *h = priv->ae_handle;
2240 for (i = 0; i < h->q_num * 2; i++) {
2241 netif_napi_del(&priv->ring_data[i].napi);
2242 if (priv->ring_data[i].ring->irq_init_flag == RCB_IRQ_INITED) {
2243 (void)irq_set_affinity_hint(
2244 priv->ring_data[i].ring->irq,
2246 free_irq(priv->ring_data[i].ring->irq,
2247 &priv->ring_data[i]);
2250 priv->ring_data[i].ring->irq_init_flag = RCB_IRQ_NOT_INITED;
2252 kfree(priv->ring_data);
2255 static void hns_nic_set_priv_ops(struct net_device *netdev)
2257 struct hns_nic_priv *priv = netdev_priv(netdev);
2258 struct hnae_handle *h = priv->ae_handle;
2260 if (AE_IS_VER1(priv->enet_ver)) {
2261 priv->ops.fill_desc = fill_desc;
2262 priv->ops.get_rxd_bnum = get_rx_desc_bnum;
2263 priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
2265 priv->ops.get_rxd_bnum = get_v2rx_desc_bnum;
2266 if ((netdev->features & NETIF_F_TSO) ||
2267 (netdev->features & NETIF_F_TSO6)) {
2268 priv->ops.fill_desc = fill_tso_desc;
2269 priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tso;
2270 /* This chip only support 7*4096 */
2271 netif_set_gso_max_size(netdev, 7 * 4096);
2273 priv->ops.fill_desc = fill_v2_desc;
2274 priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
2276 /* enable tso when init
2277 * control tso on/off through TSE bit in bd
2279 h->dev->ops->set_tso_stats(h, 1);
2283 static int hns_nic_try_get_ae(struct net_device *ndev)
2285 struct hns_nic_priv *priv = netdev_priv(ndev);
2286 struct hnae_handle *h;
2289 h = hnae_get_handle(&priv->netdev->dev,
2290 priv->fwnode, priv->port_id, NULL);
2291 if (IS_ERR_OR_NULL(h)) {
2293 dev_dbg(priv->dev, "has not handle, register notifier!\n");
2296 priv->ae_handle = h;
2298 ret = hns_nic_init_phy(ndev, h);
2300 dev_err(priv->dev, "probe phy device fail!\n");
2304 ret = hns_nic_init_ring_data(priv);
2307 goto out_init_ring_data;
2310 hns_nic_set_priv_ops(ndev);
2312 ret = register_netdev(ndev);
2314 dev_err(priv->dev, "probe register netdev fail!\n");
2315 goto out_reg_ndev_fail;
2320 hns_nic_uninit_ring_data(priv);
2321 priv->ring_data = NULL;
2324 hnae_put_handle(priv->ae_handle);
2325 priv->ae_handle = NULL;
2330 static int hns_nic_notifier_action(struct notifier_block *nb,
2331 unsigned long action, void *data)
2333 struct hns_nic_priv *priv =
2334 container_of(nb, struct hns_nic_priv, notifier_block);
2336 assert(action == HNAE_AE_REGISTER);
2338 if (!hns_nic_try_get_ae(priv->netdev)) {
2339 hnae_unregister_notifier(&priv->notifier_block);
2340 priv->notifier_block.notifier_call = NULL;
2345 static int hns_nic_dev_probe(struct platform_device *pdev)
2347 struct device *dev = &pdev->dev;
2348 struct net_device *ndev;
2349 struct hns_nic_priv *priv;
2353 ndev = alloc_etherdev_mq(sizeof(struct hns_nic_priv), NIC_MAX_Q_PER_VF);
2357 platform_set_drvdata(pdev, ndev);
2359 priv = netdev_priv(ndev);
2361 priv->netdev = ndev;
2363 if (dev_of_node(dev)) {
2364 struct device_node *ae_node;
2366 if (of_device_is_compatible(dev->of_node,
2367 "hisilicon,hns-nic-v1"))
2368 priv->enet_ver = AE_VERSION_1;
2370 priv->enet_ver = AE_VERSION_2;
2372 ae_node = of_parse_phandle(dev->of_node, "ae-handle", 0);
2375 dev_err(dev, "not find ae-handle\n");
2376 goto out_read_prop_fail;
2378 priv->fwnode = &ae_node->fwnode;
2379 } else if (is_acpi_node(dev->fwnode)) {
2380 struct acpi_reference_args args;
2382 if (acpi_dev_found(hns_enet_acpi_match[0].id))
2383 priv->enet_ver = AE_VERSION_1;
2384 else if (acpi_dev_found(hns_enet_acpi_match[1].id))
2385 priv->enet_ver = AE_VERSION_2;
2389 /* try to find port-idx-in-ae first */
2390 ret = acpi_node_get_property_reference(dev->fwnode,
2391 "ae-handle", 0, &args);
2393 dev_err(dev, "not find ae-handle\n");
2394 goto out_read_prop_fail;
2396 priv->fwnode = acpi_fwnode_handle(args.adev);
2398 dev_err(dev, "cannot read cfg data from OF or acpi\n");
2402 ret = device_property_read_u32(dev, "port-idx-in-ae", &port_id);
2404 /* only for old code compatible */
2405 ret = device_property_read_u32(dev, "port-id", &port_id);
2407 goto out_read_prop_fail;
2408 /* for old dts, we need to caculate the port offset */
2409 port_id = port_id < HNS_SRV_OFFSET ? port_id + HNS_DEBUG_OFFSET
2410 : port_id - HNS_SRV_OFFSET;
2412 priv->port_id = port_id;
2414 hns_init_mac_addr(ndev);
2416 ndev->watchdog_timeo = HNS_NIC_TX_TIMEOUT;
2417 ndev->priv_flags |= IFF_UNICAST_FLT;
2418 ndev->netdev_ops = &hns_nic_netdev_ops;
2419 hns_ethtool_set_ops(ndev);
2421 ndev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2422 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2424 ndev->vlan_features |=
2425 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM;
2426 ndev->vlan_features |= NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
2428 /* MTU range: 68 - 9578 (v1) or 9706 (v2) */
2429 ndev->min_mtu = MAC_MIN_MTU;
2430 switch (priv->enet_ver) {
2432 ndev->features |= NETIF_F_TSO | NETIF_F_TSO6;
2433 ndev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2434 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2435 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6;
2436 ndev->max_mtu = MAC_MAX_MTU_V2 -
2437 (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
2440 ndev->max_mtu = MAC_MAX_MTU -
2441 (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
2445 SET_NETDEV_DEV(ndev, dev);
2447 if (!dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64)))
2448 dev_dbg(dev, "set mask to 64bit\n");
2450 dev_err(dev, "set mask to 64bit fail!\n");
2452 /* carrier off reporting is important to ethtool even BEFORE open */
2453 netif_carrier_off(ndev);
2455 timer_setup(&priv->service_timer, hns_nic_service_timer, 0);
2456 INIT_WORK(&priv->service_task, hns_nic_service_task);
2458 set_bit(NIC_STATE_SERVICE_INITED, &priv->state);
2459 clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
2460 set_bit(NIC_STATE_DOWN, &priv->state);
2462 if (hns_nic_try_get_ae(priv->netdev)) {
2463 priv->notifier_block.notifier_call = hns_nic_notifier_action;
2464 ret = hnae_register_notifier(&priv->notifier_block);
2466 dev_err(dev, "register notifier fail!\n");
2467 goto out_notify_fail;
2469 dev_dbg(dev, "has not handle, register notifier!\n");
2475 (void)cancel_work_sync(&priv->service_task);
2481 static int hns_nic_dev_remove(struct platform_device *pdev)
2483 struct net_device *ndev = platform_get_drvdata(pdev);
2484 struct hns_nic_priv *priv = netdev_priv(ndev);
2486 if (ndev->reg_state != NETREG_UNINITIALIZED)
2487 unregister_netdev(ndev);
2489 if (priv->ring_data)
2490 hns_nic_uninit_ring_data(priv);
2491 priv->ring_data = NULL;
2494 phy_disconnect(ndev->phydev);
2496 if (!IS_ERR_OR_NULL(priv->ae_handle))
2497 hnae_put_handle(priv->ae_handle);
2498 priv->ae_handle = NULL;
2499 if (priv->notifier_block.notifier_call)
2500 hnae_unregister_notifier(&priv->notifier_block);
2501 priv->notifier_block.notifier_call = NULL;
2503 set_bit(NIC_STATE_REMOVING, &priv->state);
2504 (void)cancel_work_sync(&priv->service_task);
2510 static const struct of_device_id hns_enet_of_match[] = {
2511 {.compatible = "hisilicon,hns-nic-v1",},
2512 {.compatible = "hisilicon,hns-nic-v2",},
2516 MODULE_DEVICE_TABLE(of, hns_enet_of_match);
2518 static struct platform_driver hns_nic_dev_driver = {
2521 .of_match_table = hns_enet_of_match,
2522 .acpi_match_table = ACPI_PTR(hns_enet_acpi_match),
2524 .probe = hns_nic_dev_probe,
2525 .remove = hns_nic_dev_remove,
2528 module_platform_driver(hns_nic_dev_driver);
2530 MODULE_DESCRIPTION("HISILICON HNS Ethernet driver");
2531 MODULE_AUTHOR("Hisilicon, Inc.");
2532 MODULE_LICENSE("GPL");
2533 MODULE_ALIAS("platform:hns-nic");