1 // SPDX-License-Identifier: GPL-2.0+
2 // Copyright (c) 2016-2017 Hisilicon Limited.
4 #include <linux/dma-mapping.h>
5 #include <linux/etherdevice.h>
6 #include <linux/interrupt.h>
7 #include <linux/if_vlan.h>
9 #include <linux/ipv6.h>
10 #include <linux/module.h>
11 #include <linux/pci.h>
12 #include <linux/aer.h>
13 #include <linux/skbuff.h>
14 #include <linux/sctp.h>
15 #include <linux/vermagic.h>
17 #include <net/pkt_cls.h>
19 #include <net/vxlan.h>
22 #include "hns3_enet.h"
24 #define hns3_set_field(origin, shift, val) ((origin) |= ((val) << (shift)))
26 static void hns3_clear_all_ring(struct hnae3_handle *h);
27 static void hns3_force_clear_all_rx_ring(struct hnae3_handle *h);
28 static void hns3_remove_hw_addr(struct net_device *netdev);
30 static const char hns3_driver_name[] = "hns3";
31 const char hns3_driver_version[] = VERMAGIC_STRING;
32 static const char hns3_driver_string[] =
33 "Hisilicon Ethernet Network Driver for Hip08 Family";
34 static const char hns3_copyright[] = "Copyright (c) 2017 Huawei Corporation.";
35 static struct hnae3_client client;
37 /* hns3_pci_tbl - PCI Device ID Table
39 * Last entry must be all 0s
41 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
42 * Class, Class Mask, private data (not used) }
44 static const struct pci_device_id hns3_pci_tbl[] = {
45 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
46 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
47 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA),
48 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
49 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC),
50 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
51 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA),
52 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
53 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC),
54 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
55 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC),
56 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
57 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_VF), 0},
58 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF),
59 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
60 /* required last entry */
63 MODULE_DEVICE_TABLE(pci, hns3_pci_tbl);
65 static irqreturn_t hns3_irq_handle(int irq, void *vector)
67 struct hns3_enet_tqp_vector *tqp_vector = vector;
69 napi_schedule(&tqp_vector->napi);
74 /* This callback function is used to set affinity changes to the irq affinity
75 * masks when the irq_set_affinity_notifier function is used.
77 static void hns3_nic_irq_affinity_notify(struct irq_affinity_notify *notify,
78 const cpumask_t *mask)
80 struct hns3_enet_tqp_vector *tqp_vectors =
81 container_of(notify, struct hns3_enet_tqp_vector,
84 tqp_vectors->affinity_mask = *mask;
87 static void hns3_nic_irq_affinity_release(struct kref *ref)
91 static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv)
93 struct hns3_enet_tqp_vector *tqp_vectors;
96 for (i = 0; i < priv->vector_num; i++) {
97 tqp_vectors = &priv->tqp_vector[i];
99 if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED)
102 /* clear the affinity notifier and affinity mask */
103 irq_set_affinity_notifier(tqp_vectors->vector_irq, NULL);
104 irq_set_affinity_hint(tqp_vectors->vector_irq, NULL);
106 /* release the irq resource */
107 free_irq(tqp_vectors->vector_irq, tqp_vectors);
108 tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED;
112 static int hns3_nic_init_irq(struct hns3_nic_priv *priv)
114 struct hns3_enet_tqp_vector *tqp_vectors;
115 int txrx_int_idx = 0;
121 for (i = 0; i < priv->vector_num; i++) {
122 tqp_vectors = &priv->tqp_vector[i];
124 if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED)
127 if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) {
128 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
129 "%s-%s-%d", priv->netdev->name, "TxRx",
132 } else if (tqp_vectors->rx_group.ring) {
133 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
134 "%s-%s-%d", priv->netdev->name, "Rx",
136 } else if (tqp_vectors->tx_group.ring) {
137 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
138 "%s-%s-%d", priv->netdev->name, "Tx",
141 /* Skip this unused q_vector */
145 tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0';
147 ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0,
151 netdev_err(priv->netdev, "request irq(%d) fail\n",
152 tqp_vectors->vector_irq);
156 tqp_vectors->affinity_notify.notify =
157 hns3_nic_irq_affinity_notify;
158 tqp_vectors->affinity_notify.release =
159 hns3_nic_irq_affinity_release;
160 irq_set_affinity_notifier(tqp_vectors->vector_irq,
161 &tqp_vectors->affinity_notify);
162 irq_set_affinity_hint(tqp_vectors->vector_irq,
163 &tqp_vectors->affinity_mask);
165 tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED;
171 static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector,
174 writel(mask_en, tqp_vector->mask_addr);
177 static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector)
179 napi_enable(&tqp_vector->napi);
182 hns3_mask_vector_irq(tqp_vector, 1);
185 static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector)
188 hns3_mask_vector_irq(tqp_vector, 0);
190 disable_irq(tqp_vector->vector_irq);
191 napi_disable(&tqp_vector->napi);
194 void hns3_set_vector_coalesce_rl(struct hns3_enet_tqp_vector *tqp_vector,
197 u32 rl_reg = hns3_rl_usec_to_reg(rl_value);
199 /* this defines the configuration for RL (Interrupt Rate Limiter).
200 * Rl defines rate of interrupts i.e. number of interrupts-per-second
201 * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing
204 if (rl_reg > 0 && !tqp_vector->tx_group.coal.gl_adapt_enable &&
205 !tqp_vector->rx_group.coal.gl_adapt_enable)
206 /* According to the hardware, the range of rl_reg is
207 * 0-59 and the unit is 4.
209 rl_reg |= HNS3_INT_RL_ENABLE_MASK;
211 writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
214 void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
217 u32 rx_gl_reg = hns3_gl_usec_to_reg(gl_value);
219 writel(rx_gl_reg, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
222 void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
225 u32 tx_gl_reg = hns3_gl_usec_to_reg(gl_value);
227 writel(tx_gl_reg, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
230 static void hns3_vector_gl_rl_init(struct hns3_enet_tqp_vector *tqp_vector,
231 struct hns3_nic_priv *priv)
233 /* initialize the configuration for interrupt coalescing.
234 * 1. GL (Interrupt Gap Limiter)
235 * 2. RL (Interrupt Rate Limiter)
238 /* Default: enable interrupt coalescing self-adaptive and GL */
239 tqp_vector->tx_group.coal.gl_adapt_enable = 1;
240 tqp_vector->rx_group.coal.gl_adapt_enable = 1;
242 tqp_vector->tx_group.coal.int_gl = HNS3_INT_GL_50K;
243 tqp_vector->rx_group.coal.int_gl = HNS3_INT_GL_50K;
245 tqp_vector->rx_group.coal.flow_level = HNS3_FLOW_LOW;
246 tqp_vector->tx_group.coal.flow_level = HNS3_FLOW_LOW;
249 static void hns3_vector_gl_rl_init_hw(struct hns3_enet_tqp_vector *tqp_vector,
250 struct hns3_nic_priv *priv)
252 struct hnae3_handle *h = priv->ae_handle;
254 hns3_set_vector_coalesce_tx_gl(tqp_vector,
255 tqp_vector->tx_group.coal.int_gl);
256 hns3_set_vector_coalesce_rx_gl(tqp_vector,
257 tqp_vector->rx_group.coal.int_gl);
258 hns3_set_vector_coalesce_rl(tqp_vector, h->kinfo.int_rl_setting);
261 static int hns3_nic_set_real_num_queue(struct net_device *netdev)
263 struct hnae3_handle *h = hns3_get_handle(netdev);
264 struct hnae3_knic_private_info *kinfo = &h->kinfo;
265 unsigned int queue_size = kinfo->rss_size * kinfo->num_tc;
268 if (kinfo->num_tc <= 1) {
269 netdev_reset_tc(netdev);
271 ret = netdev_set_num_tc(netdev, kinfo->num_tc);
274 "netdev_set_num_tc fail, ret=%d!\n", ret);
278 for (i = 0; i < HNAE3_MAX_TC; i++) {
279 if (!kinfo->tc_info[i].enable)
282 netdev_set_tc_queue(netdev,
283 kinfo->tc_info[i].tc,
284 kinfo->tc_info[i].tqp_count,
285 kinfo->tc_info[i].tqp_offset);
289 ret = netif_set_real_num_tx_queues(netdev, queue_size);
292 "netif_set_real_num_tx_queues fail, ret=%d!\n",
297 ret = netif_set_real_num_rx_queues(netdev, queue_size);
300 "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
307 static u16 hns3_get_max_available_channels(struct hnae3_handle *h)
309 u16 alloc_tqps, max_rss_size, rss_size;
311 h->ae_algo->ops->get_tqps_and_rss_info(h, &alloc_tqps, &max_rss_size);
312 rss_size = alloc_tqps / h->kinfo.num_tc;
314 return min_t(u16, rss_size, max_rss_size);
317 static void hns3_tqp_enable(struct hnae3_queue *tqp)
321 rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
322 rcb_reg |= BIT(HNS3_RING_EN_B);
323 hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
326 static void hns3_tqp_disable(struct hnae3_queue *tqp)
330 rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
331 rcb_reg &= ~BIT(HNS3_RING_EN_B);
332 hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
335 static int hns3_nic_net_up(struct net_device *netdev)
337 struct hns3_nic_priv *priv = netdev_priv(netdev);
338 struct hnae3_handle *h = priv->ae_handle;
342 ret = hns3_nic_reset_all_ring(h);
346 /* get irq resource for all vectors */
347 ret = hns3_nic_init_irq(priv);
349 netdev_err(netdev, "hns init irq failed! ret=%d\n", ret);
353 clear_bit(HNS3_NIC_STATE_DOWN, &priv->state);
355 /* enable the vectors */
356 for (i = 0; i < priv->vector_num; i++)
357 hns3_vector_enable(&priv->tqp_vector[i]);
360 for (j = 0; j < h->kinfo.num_tqps; j++)
361 hns3_tqp_enable(h->kinfo.tqp[j]);
363 /* start the ae_dev */
364 ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0;
371 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
373 hns3_tqp_disable(h->kinfo.tqp[j]);
375 for (j = i - 1; j >= 0; j--)
376 hns3_vector_disable(&priv->tqp_vector[j]);
378 hns3_nic_uninit_irq(priv);
383 static void hns3_config_xps(struct hns3_nic_priv *priv)
387 for (i = 0; i < priv->vector_num; i++) {
388 struct hns3_enet_tqp_vector *tqp_vector = &priv->tqp_vector[i];
389 struct hns3_enet_ring *ring = tqp_vector->tx_group.ring;
394 ret = netif_set_xps_queue(priv->netdev,
395 &tqp_vector->affinity_mask,
396 ring->tqp->tqp_index);
398 netdev_warn(priv->netdev,
399 "set xps queue failed: %d", ret);
406 static int hns3_nic_net_open(struct net_device *netdev)
408 struct hns3_nic_priv *priv = netdev_priv(netdev);
409 struct hnae3_handle *h = hns3_get_handle(netdev);
410 struct hnae3_knic_private_info *kinfo;
413 if (hns3_nic_resetting(netdev))
416 netif_carrier_off(netdev);
418 ret = hns3_nic_set_real_num_queue(netdev);
422 ret = hns3_nic_net_up(netdev);
425 "hns net up fail, ret=%d!\n", ret);
430 for (i = 0; i < HNAE3_MAX_USER_PRIO; i++) {
431 netdev_set_prio_tc_map(netdev, i,
435 if (h->ae_algo->ops->set_timer_task)
436 h->ae_algo->ops->set_timer_task(priv->ae_handle, true);
438 hns3_config_xps(priv);
442 static void hns3_nic_net_down(struct net_device *netdev)
444 struct hns3_nic_priv *priv = netdev_priv(netdev);
445 struct hnae3_handle *h = hns3_get_handle(netdev);
446 const struct hnae3_ae_ops *ops;
449 /* disable vectors */
450 for (i = 0; i < priv->vector_num; i++)
451 hns3_vector_disable(&priv->tqp_vector[i]);
454 for (i = 0; i < h->kinfo.num_tqps; i++)
455 hns3_tqp_disable(h->kinfo.tqp[i]);
458 ops = priv->ae_handle->ae_algo->ops;
460 ops->stop(priv->ae_handle);
462 /* free irq resources */
463 hns3_nic_uninit_irq(priv);
465 hns3_clear_all_ring(priv->ae_handle);
468 static int hns3_nic_net_stop(struct net_device *netdev)
470 struct hns3_nic_priv *priv = netdev_priv(netdev);
471 struct hnae3_handle *h = hns3_get_handle(netdev);
473 if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
476 if (h->ae_algo->ops->set_timer_task)
477 h->ae_algo->ops->set_timer_task(priv->ae_handle, false);
479 netif_tx_stop_all_queues(netdev);
480 netif_carrier_off(netdev);
482 hns3_nic_net_down(netdev);
487 static int hns3_nic_uc_sync(struct net_device *netdev,
488 const unsigned char *addr)
490 struct hnae3_handle *h = hns3_get_handle(netdev);
492 if (h->ae_algo->ops->add_uc_addr)
493 return h->ae_algo->ops->add_uc_addr(h, addr);
498 static int hns3_nic_uc_unsync(struct net_device *netdev,
499 const unsigned char *addr)
501 struct hnae3_handle *h = hns3_get_handle(netdev);
503 if (h->ae_algo->ops->rm_uc_addr)
504 return h->ae_algo->ops->rm_uc_addr(h, addr);
509 static int hns3_nic_mc_sync(struct net_device *netdev,
510 const unsigned char *addr)
512 struct hnae3_handle *h = hns3_get_handle(netdev);
514 if (h->ae_algo->ops->add_mc_addr)
515 return h->ae_algo->ops->add_mc_addr(h, addr);
520 static int hns3_nic_mc_unsync(struct net_device *netdev,
521 const unsigned char *addr)
523 struct hnae3_handle *h = hns3_get_handle(netdev);
525 if (h->ae_algo->ops->rm_mc_addr)
526 return h->ae_algo->ops->rm_mc_addr(h, addr);
531 static u8 hns3_get_netdev_flags(struct net_device *netdev)
535 if (netdev->flags & IFF_PROMISC) {
536 flags = HNAE3_USER_UPE | HNAE3_USER_MPE | HNAE3_BPE;
538 flags |= HNAE3_VLAN_FLTR;
539 if (netdev->flags & IFF_ALLMULTI)
540 flags |= HNAE3_USER_MPE;
546 static void hns3_nic_set_rx_mode(struct net_device *netdev)
548 struct hnae3_handle *h = hns3_get_handle(netdev);
552 new_flags = hns3_get_netdev_flags(netdev);
554 ret = __dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync);
556 netdev_err(netdev, "sync uc address fail\n");
558 new_flags |= HNAE3_OVERFLOW_UPE;
561 if (netdev->flags & IFF_MULTICAST) {
562 ret = __dev_mc_sync(netdev, hns3_nic_mc_sync,
565 netdev_err(netdev, "sync mc address fail\n");
567 new_flags |= HNAE3_OVERFLOW_MPE;
571 /* User mode Promisc mode enable and vlan filtering is disabled to
572 * let all packets in. MAC-VLAN Table overflow Promisc enabled and
573 * vlan fitering is enabled
575 hns3_enable_vlan_filter(netdev, new_flags & HNAE3_VLAN_FLTR);
576 h->netdev_flags = new_flags;
577 hns3_update_promisc_mode(netdev, new_flags);
580 int hns3_update_promisc_mode(struct net_device *netdev, u8 promisc_flags)
582 struct hns3_nic_priv *priv = netdev_priv(netdev);
583 struct hnae3_handle *h = priv->ae_handle;
585 if (h->ae_algo->ops->set_promisc_mode) {
586 return h->ae_algo->ops->set_promisc_mode(h,
587 promisc_flags & HNAE3_UPE,
588 promisc_flags & HNAE3_MPE);
594 void hns3_enable_vlan_filter(struct net_device *netdev, bool enable)
596 struct hns3_nic_priv *priv = netdev_priv(netdev);
597 struct hnae3_handle *h = priv->ae_handle;
600 if (h->pdev->revision >= 0x21 && h->ae_algo->ops->enable_vlan_filter) {
601 last_state = h->netdev_flags & HNAE3_VLAN_FLTR ? true : false;
602 if (enable != last_state) {
605 enable ? "enable" : "disable");
606 h->ae_algo->ops->enable_vlan_filter(h, enable);
611 static int hns3_set_tso(struct sk_buff *skb, u32 *paylen,
612 u16 *mss, u32 *type_cs_vlan_tso)
614 u32 l4_offset, hdr_len;
615 union l3_hdr_info l3;
616 union l4_hdr_info l4;
620 if (!skb_is_gso(skb))
623 ret = skb_cow_head(skb, 0);
627 l3.hdr = skb_network_header(skb);
628 l4.hdr = skb_transport_header(skb);
630 /* Software should clear the IPv4's checksum field when tso is
633 if (l3.v4->version == 4)
637 if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
640 SKB_GSO_UDP_TUNNEL_CSUM)) {
641 if ((!(skb_shinfo(skb)->gso_type &
643 (skb_shinfo(skb)->gso_type &
644 SKB_GSO_UDP_TUNNEL_CSUM)) {
645 /* Software should clear the udp's checksum
646 * field when tso is needed.
650 /* reset l3&l4 pointers from outer to inner headers */
651 l3.hdr = skb_inner_network_header(skb);
652 l4.hdr = skb_inner_transport_header(skb);
654 /* Software should clear the IPv4's checksum field when
657 if (l3.v4->version == 4)
661 /* normal or tunnel packet*/
662 l4_offset = l4.hdr - skb->data;
663 hdr_len = (l4.tcp->doff << 2) + l4_offset;
665 /* remove payload length from inner pseudo checksum when tso*/
666 l4_paylen = skb->len - l4_offset;
667 csum_replace_by_diff(&l4.tcp->check,
668 (__force __wsum)htonl(l4_paylen));
670 /* find the txbd field values */
671 *paylen = skb->len - hdr_len;
672 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_TSO_B, 1);
674 /* get MSS for TSO */
675 *mss = skb_shinfo(skb)->gso_size;
680 static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
683 union l3_hdr_info l3;
684 unsigned char *l4_hdr;
685 unsigned char *exthdr;
689 /* find outer header point */
690 l3.hdr = skb_network_header(skb);
691 l4_hdr = skb_transport_header(skb);
693 if (skb->protocol == htons(ETH_P_IPV6)) {
694 exthdr = l3.hdr + sizeof(*l3.v6);
695 l4_proto_tmp = l3.v6->nexthdr;
696 if (l4_hdr != exthdr)
697 ipv6_skip_exthdr(skb, exthdr - skb->data,
698 &l4_proto_tmp, &frag_off);
699 } else if (skb->protocol == htons(ETH_P_IP)) {
700 l4_proto_tmp = l3.v4->protocol;
705 *ol4_proto = l4_proto_tmp;
708 if (!skb->encapsulation) {
713 /* find inner header point */
714 l3.hdr = skb_inner_network_header(skb);
715 l4_hdr = skb_inner_transport_header(skb);
717 if (l3.v6->version == 6) {
718 exthdr = l3.hdr + sizeof(*l3.v6);
719 l4_proto_tmp = l3.v6->nexthdr;
720 if (l4_hdr != exthdr)
721 ipv6_skip_exthdr(skb, exthdr - skb->data,
722 &l4_proto_tmp, &frag_off);
723 } else if (l3.v4->version == 4) {
724 l4_proto_tmp = l3.v4->protocol;
727 *il4_proto = l4_proto_tmp;
732 static void hns3_set_l2l3l4_len(struct sk_buff *skb, u8 ol4_proto,
733 u8 il4_proto, u32 *type_cs_vlan_tso,
734 u32 *ol_type_vlan_len_msec)
736 union l3_hdr_info l3;
737 union l4_hdr_info l4;
738 unsigned char *l2_hdr;
739 u8 l4_proto = ol4_proto;
746 l3.hdr = skb_network_header(skb);
747 l4.hdr = skb_transport_header(skb);
749 /* compute L2 header size for normal packet, defined in 2 Bytes */
750 l2_len = l3.hdr - skb->data;
751 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S, l2_len >> 1);
754 if (skb->encapsulation) {
755 /* compute OL2 header size, defined in 2 Bytes */
757 hns3_set_field(*ol_type_vlan_len_msec,
758 HNS3_TXD_L2LEN_S, ol2_len >> 1);
760 /* compute OL3 header size, defined in 4 Bytes */
761 ol3_len = l4.hdr - l3.hdr;
762 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_S,
765 /* MAC in UDP, MAC in GRE (0x6558)*/
766 if ((ol4_proto == IPPROTO_UDP) || (ol4_proto == IPPROTO_GRE)) {
767 /* switch MAC header ptr from outer to inner header.*/
768 l2_hdr = skb_inner_mac_header(skb);
770 /* compute OL4 header size, defined in 4 Bytes. */
771 ol4_len = l2_hdr - l4.hdr;
772 hns3_set_field(*ol_type_vlan_len_msec,
773 HNS3_TXD_L4LEN_S, ol4_len >> 2);
775 /* switch IP header ptr from outer to inner header */
776 l3.hdr = skb_inner_network_header(skb);
778 /* compute inner l2 header size, defined in 2 Bytes. */
779 l2_len = l3.hdr - l2_hdr;
780 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S,
783 /* skb packet types not supported by hardware,
784 * txbd len fild doesn't be filled.
789 /* switch L4 header pointer from outer to inner */
790 l4.hdr = skb_inner_transport_header(skb);
792 l4_proto = il4_proto;
795 /* compute inner(/normal) L3 header size, defined in 4 Bytes */
796 l3_len = l4.hdr - l3.hdr;
797 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_S, l3_len >> 2);
799 /* compute inner(/normal) L4 header size, defined in 4 Bytes */
802 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
806 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
807 (sizeof(struct sctphdr) >> 2));
810 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
811 (sizeof(struct udphdr) >> 2));
814 /* skb packet types not supported by hardware,
815 * txbd len fild doesn't be filled.
821 /* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
822 * and it is udp packet, which has a dest port as the IANA assigned.
823 * the hardware is expected to do the checksum offload, but the
824 * hardware will not do the checksum offload when udp dest port is
827 static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
829 #define IANA_VXLAN_PORT 4789
830 union l4_hdr_info l4;
832 l4.hdr = skb_transport_header(skb);
834 if (!(!skb->encapsulation && l4.udp->dest == htons(IANA_VXLAN_PORT)))
837 skb_checksum_help(skb);
842 static int hns3_set_l3l4_type_csum(struct sk_buff *skb, u8 ol4_proto,
843 u8 il4_proto, u32 *type_cs_vlan_tso,
844 u32 *ol_type_vlan_len_msec)
846 union l3_hdr_info l3;
847 u32 l4_proto = ol4_proto;
849 l3.hdr = skb_network_header(skb);
851 /* define OL3 type and tunnel type(OL4).*/
852 if (skb->encapsulation) {
853 /* define outer network header type.*/
854 if (skb->protocol == htons(ETH_P_IP)) {
856 hns3_set_field(*ol_type_vlan_len_msec,
858 HNS3_OL3T_IPV4_CSUM);
860 hns3_set_field(*ol_type_vlan_len_msec,
862 HNS3_OL3T_IPV4_NO_CSUM);
864 } else if (skb->protocol == htons(ETH_P_IPV6)) {
865 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_S,
869 /* define tunnel type(OL4).*/
872 hns3_set_field(*ol_type_vlan_len_msec,
874 HNS3_TUN_MAC_IN_UDP);
877 hns3_set_field(*ol_type_vlan_len_msec,
882 /* drop the skb tunnel packet if hardware don't support,
883 * because hardware can't calculate csum when TSO.
888 /* the stack computes the IP header already,
889 * driver calculate l4 checksum when not TSO.
891 skb_checksum_help(skb);
895 l3.hdr = skb_inner_network_header(skb);
896 l4_proto = il4_proto;
899 if (l3.v4->version == 4) {
900 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
903 /* the stack computes the IP header already, the only time we
904 * need the hardware to recompute it is in the case of TSO.
907 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3CS_B, 1);
908 } else if (l3.v6->version == 6) {
909 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
915 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
916 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
920 if (hns3_tunnel_csum_bug(skb))
923 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
924 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
928 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
929 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
933 /* drop the skb tunnel packet if hardware don't support,
934 * because hardware can't calculate csum when TSO.
939 /* the stack computes the IP header already,
940 * driver calculate l4 checksum when not TSO.
942 skb_checksum_help(skb);
949 static void hns3_set_txbd_baseinfo(u16 *bdtp_fe_sc_vld_ra_ri, int frag_end)
951 /* Config bd buffer end */
952 hns3_set_field(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_FE_B, !!frag_end);
953 hns3_set_field(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_VLD_B, 1);
956 static int hns3_fill_desc_vtags(struct sk_buff *skb,
957 struct hns3_enet_ring *tx_ring,
958 u32 *inner_vlan_flag,
963 #define HNS3_TX_VLAN_PRIO_SHIFT 13
965 if (skb->protocol == htons(ETH_P_8021Q) &&
966 !(tx_ring->tqp->handle->kinfo.netdev->features &
967 NETIF_F_HW_VLAN_CTAG_TX)) {
968 /* When HW VLAN acceleration is turned off, and the stack
969 * sets the protocol to 802.1q, the driver just need to
970 * set the protocol to the encapsulated ethertype.
972 skb->protocol = vlan_get_protocol(skb);
976 if (skb_vlan_tag_present(skb)) {
979 vlan_tag = skb_vlan_tag_get(skb);
980 vlan_tag |= (skb->priority & 0x7) << HNS3_TX_VLAN_PRIO_SHIFT;
982 /* Based on hw strategy, use out_vtag in two layer tag case,
983 * and use inner_vtag in one tag case.
985 if (skb->protocol == htons(ETH_P_8021Q)) {
986 hns3_set_field(*out_vlan_flag, HNS3_TXD_OVLAN_B, 1);
987 *out_vtag = vlan_tag;
989 hns3_set_field(*inner_vlan_flag, HNS3_TXD_VLAN_B, 1);
990 *inner_vtag = vlan_tag;
992 } else if (skb->protocol == htons(ETH_P_8021Q)) {
993 struct vlan_ethhdr *vhdr;
996 rc = skb_cow_head(skb, 0);
997 if (unlikely(rc < 0))
999 vhdr = (struct vlan_ethhdr *)skb->data;
1000 vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority & 0x7)
1001 << HNS3_TX_VLAN_PRIO_SHIFT);
1004 skb->protocol = vlan_get_protocol(skb);
1008 static int hns3_fill_desc(struct hns3_enet_ring *ring, void *priv,
1009 int size, int frag_end, enum hns_desc_type type)
1011 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1012 struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1013 struct device *dev = ring_to_dev(ring);
1014 u16 bdtp_fe_sc_vld_ra_ri = 0;
1015 struct skb_frag_struct *frag;
1016 unsigned int frag_buf_num;
1020 if (type == DESC_TYPE_SKB) {
1021 struct sk_buff *skb = (struct sk_buff *)priv;
1022 u32 ol_type_vlan_len_msec = 0;
1023 u32 type_cs_vlan_tso = 0;
1024 u32 paylen = skb->len;
1030 ret = hns3_fill_desc_vtags(skb, ring, &type_cs_vlan_tso,
1031 &ol_type_vlan_len_msec,
1032 &inner_vtag, &out_vtag);
1036 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1037 u8 ol4_proto, il4_proto;
1039 skb_reset_mac_len(skb);
1041 ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
1044 hns3_set_l2l3l4_len(skb, ol4_proto, il4_proto,
1046 &ol_type_vlan_len_msec);
1047 ret = hns3_set_l3l4_type_csum(skb, ol4_proto, il4_proto,
1049 &ol_type_vlan_len_msec);
1053 ret = hns3_set_tso(skb, &paylen, &mss,
1060 desc->tx.ol_type_vlan_len_msec =
1061 cpu_to_le32(ol_type_vlan_len_msec);
1062 desc->tx.type_cs_vlan_tso_len =
1063 cpu_to_le32(type_cs_vlan_tso);
1064 desc->tx.paylen = cpu_to_le32(paylen);
1065 desc->tx.mss = cpu_to_le16(mss);
1066 desc->tx.vlan_tag = cpu_to_le16(inner_vtag);
1067 desc->tx.outer_vlan_tag = cpu_to_le16(out_vtag);
1069 dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1071 frag = (struct skb_frag_struct *)priv;
1072 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1075 if (unlikely(dma_mapping_error(ring->dev, dma))) {
1076 ring->stats.sw_err_cnt++;
1080 desc_cb->length = size;
1082 frag_buf_num = (size + HNS3_MAX_BD_SIZE - 1) >> HNS3_MAX_BD_SIZE_OFFSET;
1083 sizeoflast = size & HNS3_TX_LAST_SIZE_M;
1084 sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1086 /* When frag size is bigger than hardware limit, split this frag */
1087 for (k = 0; k < frag_buf_num; k++) {
1088 /* The txbd's baseinfo of DESC_TYPE_PAGE & DESC_TYPE_SKB */
1089 desc_cb->priv = priv;
1090 desc_cb->dma = dma + HNS3_MAX_BD_SIZE * k;
1091 desc_cb->type = (type == DESC_TYPE_SKB && !k) ?
1092 DESC_TYPE_SKB : DESC_TYPE_PAGE;
1094 /* now, fill the descriptor */
1095 desc->addr = cpu_to_le64(dma + HNS3_MAX_BD_SIZE * k);
1096 desc->tx.send_size = cpu_to_le16((k == frag_buf_num - 1) ?
1097 (u16)sizeoflast : (u16)HNS3_MAX_BD_SIZE);
1098 hns3_set_txbd_baseinfo(&bdtp_fe_sc_vld_ra_ri,
1099 frag_end && (k == frag_buf_num - 1) ?
1101 desc->tx.bdtp_fe_sc_vld_ra_ri =
1102 cpu_to_le16(bdtp_fe_sc_vld_ra_ri);
1104 /* move ring pointer to next.*/
1105 ring_ptr_move_fw(ring, next_to_use);
1107 desc_cb = &ring->desc_cb[ring->next_to_use];
1108 desc = &ring->desc[ring->next_to_use];
1114 static int hns3_nic_maybe_stop_tso(struct sk_buff **out_skb, int *bnum,
1115 struct hns3_enet_ring *ring)
1117 struct sk_buff *skb = *out_skb;
1118 struct sk_buff *new_skb = NULL;
1119 struct skb_frag_struct *frag;
1126 size = skb_headlen(skb);
1127 buf_num = (size + HNS3_MAX_BD_SIZE - 1) >> HNS3_MAX_BD_SIZE_OFFSET;
1129 frag_num = skb_shinfo(skb)->nr_frags;
1130 for (i = 0; i < frag_num; i++) {
1131 frag = &skb_shinfo(skb)->frags[i];
1132 size = skb_frag_size(frag);
1133 bdnum_for_frag = (size + HNS3_MAX_BD_SIZE - 1) >>
1134 HNS3_MAX_BD_SIZE_OFFSET;
1135 if (unlikely(bdnum_for_frag > HNS3_MAX_BD_PER_FRAG))
1138 buf_num += bdnum_for_frag;
1141 if (unlikely(buf_num > HNS3_MAX_BD_PER_FRAG)) {
1142 buf_num = (skb->len + HNS3_MAX_BD_SIZE - 1) >>
1143 HNS3_MAX_BD_SIZE_OFFSET;
1144 if (ring_space(ring) < buf_num)
1146 /* manual split the send packet */
1147 new_skb = skb_copy(skb, GFP_ATOMIC);
1150 dev_kfree_skb_any(skb);
1154 if (unlikely(ring_space(ring) < buf_num))
1161 static int hns3_nic_maybe_stop_tx(struct sk_buff **out_skb, int *bnum,
1162 struct hns3_enet_ring *ring)
1164 struct sk_buff *skb = *out_skb;
1165 struct sk_buff *new_skb = NULL;
1168 /* No. of segments (plus a header) */
1169 buf_num = skb_shinfo(skb)->nr_frags + 1;
1171 if (unlikely(buf_num > HNS3_MAX_BD_PER_FRAG)) {
1172 buf_num = (skb->len + HNS3_MAX_BD_SIZE - 1) / HNS3_MAX_BD_SIZE;
1173 if (ring_space(ring) < buf_num)
1175 /* manual split the send packet */
1176 new_skb = skb_copy(skb, GFP_ATOMIC);
1179 dev_kfree_skb_any(skb);
1183 if (unlikely(ring_space(ring) < buf_num))
1191 static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
1193 struct device *dev = ring_to_dev(ring);
1196 for (i = 0; i < ring->desc_num; i++) {
1197 /* check if this is where we started */
1198 if (ring->next_to_use == next_to_use_orig)
1201 /* unmap the descriptor dma address */
1202 if (ring->desc_cb[ring->next_to_use].type == DESC_TYPE_SKB)
1203 dma_unmap_single(dev,
1204 ring->desc_cb[ring->next_to_use].dma,
1205 ring->desc_cb[ring->next_to_use].length,
1207 else if (ring->desc_cb[ring->next_to_use].length)
1209 ring->desc_cb[ring->next_to_use].dma,
1210 ring->desc_cb[ring->next_to_use].length,
1213 ring->desc_cb[ring->next_to_use].length = 0;
1216 ring_ptr_move_bw(ring, next_to_use);
1220 netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
1222 struct hns3_nic_priv *priv = netdev_priv(netdev);
1223 struct hns3_nic_ring_data *ring_data =
1224 &tx_ring_data(priv, skb->queue_mapping);
1225 struct hns3_enet_ring *ring = ring_data->ring;
1226 struct netdev_queue *dev_queue;
1227 struct skb_frag_struct *frag;
1228 int next_to_use_head;
1229 int next_to_use_frag;
1236 /* Prefetch the data used later */
1237 prefetch(skb->data);
1239 switch (priv->ops.maybe_stop_tx(&skb, &buf_num, ring)) {
1241 u64_stats_update_begin(&ring->syncp);
1242 ring->stats.tx_busy++;
1243 u64_stats_update_end(&ring->syncp);
1245 goto out_net_tx_busy;
1247 u64_stats_update_begin(&ring->syncp);
1248 ring->stats.sw_err_cnt++;
1249 u64_stats_update_end(&ring->syncp);
1250 netdev_err(netdev, "no memory to xmit!\n");
1257 /* No. of segments (plus a header) */
1258 seg_num = skb_shinfo(skb)->nr_frags + 1;
1259 /* Fill the first part */
1260 size = skb_headlen(skb);
1262 next_to_use_head = ring->next_to_use;
1264 ret = hns3_fill_desc(ring, skb, size, seg_num == 1 ? 1 : 0,
1269 next_to_use_frag = ring->next_to_use;
1270 /* Fill the fragments */
1271 for (i = 1; i < seg_num; i++) {
1272 frag = &skb_shinfo(skb)->frags[i - 1];
1273 size = skb_frag_size(frag);
1275 ret = hns3_fill_desc(ring, frag, size,
1276 seg_num - 1 == i ? 1 : 0,
1283 /* Complete translate all packets */
1284 dev_queue = netdev_get_tx_queue(netdev, ring_data->queue_index);
1285 netdev_tx_sent_queue(dev_queue, skb->len);
1287 wmb(); /* Commit all data before submit */
1289 hnae3_queue_xmit(ring->tqp, buf_num);
1291 return NETDEV_TX_OK;
1294 hns3_clear_desc(ring, next_to_use_frag);
1297 hns3_clear_desc(ring, next_to_use_head);
1300 dev_kfree_skb_any(skb);
1301 return NETDEV_TX_OK;
1304 netif_stop_subqueue(netdev, ring_data->queue_index);
1305 smp_mb(); /* Commit all data before submit */
1307 return NETDEV_TX_BUSY;
1310 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
1312 struct hnae3_handle *h = hns3_get_handle(netdev);
1313 struct sockaddr *mac_addr = p;
1316 if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1317 return -EADDRNOTAVAIL;
1319 if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
1320 netdev_info(netdev, "already using mac address %pM\n",
1325 ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
1327 netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
1331 ether_addr_copy(netdev->dev_addr, mac_addr->sa_data);
1336 static int hns3_nic_do_ioctl(struct net_device *netdev,
1337 struct ifreq *ifr, int cmd)
1339 struct hnae3_handle *h = hns3_get_handle(netdev);
1341 if (!netif_running(netdev))
1344 if (!h->ae_algo->ops->do_ioctl)
1347 return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
1350 static int hns3_nic_set_features(struct net_device *netdev,
1351 netdev_features_t features)
1353 netdev_features_t changed = netdev->features ^ features;
1354 struct hns3_nic_priv *priv = netdev_priv(netdev);
1355 struct hnae3_handle *h = priv->ae_handle;
1359 if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) {
1360 if (features & (NETIF_F_TSO | NETIF_F_TSO6))
1361 priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tso;
1363 priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tx;
1366 if (changed & (NETIF_F_GRO_HW) && h->ae_algo->ops->set_gro_en) {
1367 enable = !!(features & NETIF_F_GRO_HW);
1368 ret = h->ae_algo->ops->set_gro_en(h, enable);
1373 if ((changed & NETIF_F_HW_VLAN_CTAG_FILTER) &&
1374 h->ae_algo->ops->enable_vlan_filter) {
1375 enable = !!(features & NETIF_F_HW_VLAN_CTAG_FILTER);
1376 h->ae_algo->ops->enable_vlan_filter(h, enable);
1379 if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
1380 h->ae_algo->ops->enable_hw_strip_rxvtag) {
1381 enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
1382 ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, enable);
1387 if ((changed & NETIF_F_NTUPLE) && h->ae_algo->ops->enable_fd) {
1388 enable = !!(features & NETIF_F_NTUPLE);
1389 h->ae_algo->ops->enable_fd(h, enable);
1392 netdev->features = features;
1396 static void hns3_nic_get_stats64(struct net_device *netdev,
1397 struct rtnl_link_stats64 *stats)
1399 struct hns3_nic_priv *priv = netdev_priv(netdev);
1400 int queue_num = priv->ae_handle->kinfo.num_tqps;
1401 struct hnae3_handle *handle = priv->ae_handle;
1402 struct hns3_enet_ring *ring;
1403 u64 rx_length_errors = 0;
1404 u64 rx_crc_errors = 0;
1405 u64 rx_multicast = 0;
1417 if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
1420 handle->ae_algo->ops->update_stats(handle, &netdev->stats);
1422 for (idx = 0; idx < queue_num; idx++) {
1423 /* fetch the tx stats */
1424 ring = priv->ring_data[idx].ring;
1426 start = u64_stats_fetch_begin_irq(&ring->syncp);
1427 tx_bytes += ring->stats.tx_bytes;
1428 tx_pkts += ring->stats.tx_pkts;
1429 tx_drop += ring->stats.sw_err_cnt;
1430 tx_errors += ring->stats.sw_err_cnt;
1431 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1433 /* fetch the rx stats */
1434 ring = priv->ring_data[idx + queue_num].ring;
1436 start = u64_stats_fetch_begin_irq(&ring->syncp);
1437 rx_bytes += ring->stats.rx_bytes;
1438 rx_pkts += ring->stats.rx_pkts;
1439 rx_drop += ring->stats.non_vld_descs;
1440 rx_drop += ring->stats.l2_err;
1441 rx_errors += ring->stats.non_vld_descs;
1442 rx_errors += ring->stats.l2_err;
1443 rx_crc_errors += ring->stats.l2_err;
1444 rx_crc_errors += ring->stats.l3l4_csum_err;
1445 rx_multicast += ring->stats.rx_multicast;
1446 rx_length_errors += ring->stats.err_pkt_len;
1447 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1450 stats->tx_bytes = tx_bytes;
1451 stats->tx_packets = tx_pkts;
1452 stats->rx_bytes = rx_bytes;
1453 stats->rx_packets = rx_pkts;
1455 stats->rx_errors = rx_errors;
1456 stats->multicast = rx_multicast;
1457 stats->rx_length_errors = rx_length_errors;
1458 stats->rx_crc_errors = rx_crc_errors;
1459 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
1461 stats->tx_errors = tx_errors;
1462 stats->rx_dropped = rx_drop;
1463 stats->tx_dropped = tx_drop;
1464 stats->collisions = netdev->stats.collisions;
1465 stats->rx_over_errors = netdev->stats.rx_over_errors;
1466 stats->rx_frame_errors = netdev->stats.rx_frame_errors;
1467 stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
1468 stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
1469 stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
1470 stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
1471 stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
1472 stats->tx_window_errors = netdev->stats.tx_window_errors;
1473 stats->rx_compressed = netdev->stats.rx_compressed;
1474 stats->tx_compressed = netdev->stats.tx_compressed;
1477 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
1479 struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
1480 struct hnae3_handle *h = hns3_get_handle(netdev);
1481 struct hnae3_knic_private_info *kinfo = &h->kinfo;
1482 u8 *prio_tc = mqprio_qopt->qopt.prio_tc_map;
1483 u8 tc = mqprio_qopt->qopt.num_tc;
1484 u16 mode = mqprio_qopt->mode;
1485 u8 hw = mqprio_qopt->qopt.hw;
1487 if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
1488 mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
1491 if (tc > HNAE3_MAX_TC)
1497 return (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
1498 kinfo->dcb_ops->setup_tc(h, tc, prio_tc) : -EOPNOTSUPP;
1501 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
1504 if (type != TC_SETUP_QDISC_MQPRIO)
1507 return hns3_setup_tc(dev, type_data);
1510 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
1511 __be16 proto, u16 vid)
1513 struct hnae3_handle *h = hns3_get_handle(netdev);
1514 struct hns3_nic_priv *priv = netdev_priv(netdev);
1517 if (h->ae_algo->ops->set_vlan_filter)
1518 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
1521 set_bit(vid, priv->active_vlans);
1526 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
1527 __be16 proto, u16 vid)
1529 struct hnae3_handle *h = hns3_get_handle(netdev);
1530 struct hns3_nic_priv *priv = netdev_priv(netdev);
1533 if (h->ae_algo->ops->set_vlan_filter)
1534 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
1537 clear_bit(vid, priv->active_vlans);
1542 static int hns3_restore_vlan(struct net_device *netdev)
1544 struct hns3_nic_priv *priv = netdev_priv(netdev);
1548 for_each_set_bit(vid, priv->active_vlans, VLAN_N_VID) {
1549 ret = hns3_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
1551 netdev_err(netdev, "Restore vlan: %d filter, ret:%d\n",
1560 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
1561 u8 qos, __be16 vlan_proto)
1563 struct hnae3_handle *h = hns3_get_handle(netdev);
1566 if (h->ae_algo->ops->set_vf_vlan_filter)
1567 ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
1573 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
1575 struct hnae3_handle *h = hns3_get_handle(netdev);
1578 if (!h->ae_algo->ops->set_mtu)
1581 ret = h->ae_algo->ops->set_mtu(h, new_mtu);
1583 netdev_err(netdev, "failed to change MTU in hardware %d\n",
1586 netdev->mtu = new_mtu;
1591 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
1593 struct hns3_nic_priv *priv = netdev_priv(ndev);
1594 struct hns3_enet_ring *tx_ring = NULL;
1595 int timeout_queue = 0;
1596 int hw_head, hw_tail;
1599 /* Find the stopped queue the same way the stack does */
1600 for (i = 0; i < ndev->real_num_tx_queues; i++) {
1601 struct netdev_queue *q;
1602 unsigned long trans_start;
1604 q = netdev_get_tx_queue(ndev, i);
1605 trans_start = q->trans_start;
1606 if (netif_xmit_stopped(q) &&
1608 (trans_start + ndev->watchdog_timeo))) {
1614 if (i == ndev->num_tx_queues) {
1616 "no netdev TX timeout queue found, timeout count: %llu\n",
1617 priv->tx_timeout_count);
1621 tx_ring = priv->ring_data[timeout_queue].ring;
1623 hw_head = readl_relaxed(tx_ring->tqp->io_base +
1624 HNS3_RING_TX_RING_HEAD_REG);
1625 hw_tail = readl_relaxed(tx_ring->tqp->io_base +
1626 HNS3_RING_TX_RING_TAIL_REG);
1628 "tx_timeout count: %llu, queue id: %d, SW_NTU: 0x%x, SW_NTC: 0x%x, HW_HEAD: 0x%x, HW_TAIL: 0x%x, INT: 0x%x\n",
1629 priv->tx_timeout_count,
1631 tx_ring->next_to_use,
1632 tx_ring->next_to_clean,
1635 readl(tx_ring->tqp_vector->mask_addr));
1640 static void hns3_nic_net_timeout(struct net_device *ndev)
1642 struct hns3_nic_priv *priv = netdev_priv(ndev);
1643 struct hnae3_handle *h = priv->ae_handle;
1645 if (!hns3_get_tx_timeo_queue_info(ndev))
1648 priv->tx_timeout_count++;
1650 /* request the reset, and let the hclge to determine
1651 * which reset level should be done
1653 if (h->ae_algo->ops->reset_event)
1654 h->ae_algo->ops->reset_event(h->pdev, h);
1657 static const struct net_device_ops hns3_nic_netdev_ops = {
1658 .ndo_open = hns3_nic_net_open,
1659 .ndo_stop = hns3_nic_net_stop,
1660 .ndo_start_xmit = hns3_nic_net_xmit,
1661 .ndo_tx_timeout = hns3_nic_net_timeout,
1662 .ndo_set_mac_address = hns3_nic_net_set_mac_address,
1663 .ndo_do_ioctl = hns3_nic_do_ioctl,
1664 .ndo_change_mtu = hns3_nic_change_mtu,
1665 .ndo_set_features = hns3_nic_set_features,
1666 .ndo_get_stats64 = hns3_nic_get_stats64,
1667 .ndo_setup_tc = hns3_nic_setup_tc,
1668 .ndo_set_rx_mode = hns3_nic_set_rx_mode,
1669 .ndo_vlan_rx_add_vid = hns3_vlan_rx_add_vid,
1670 .ndo_vlan_rx_kill_vid = hns3_vlan_rx_kill_vid,
1671 .ndo_set_vf_vlan = hns3_ndo_set_vf_vlan,
1674 static bool hns3_is_phys_func(struct pci_dev *pdev)
1676 u32 dev_id = pdev->device;
1679 case HNAE3_DEV_ID_GE:
1680 case HNAE3_DEV_ID_25GE:
1681 case HNAE3_DEV_ID_25GE_RDMA:
1682 case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
1683 case HNAE3_DEV_ID_50GE_RDMA:
1684 case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
1685 case HNAE3_DEV_ID_100G_RDMA_MACSEC:
1687 case HNAE3_DEV_ID_100G_VF:
1688 case HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF:
1691 dev_warn(&pdev->dev, "un-recognized pci device-id %d",
1698 static void hns3_disable_sriov(struct pci_dev *pdev)
1700 /* If our VFs are assigned we cannot shut down SR-IOV
1701 * without causing issues, so just leave the hardware
1702 * available but disabled
1704 if (pci_vfs_assigned(pdev)) {
1705 dev_warn(&pdev->dev,
1706 "disabling driver while VFs are assigned\n");
1710 pci_disable_sriov(pdev);
1713 static void hns3_get_dev_capability(struct pci_dev *pdev,
1714 struct hnae3_ae_dev *ae_dev)
1716 if (pdev->revision >= 0x21) {
1717 hnae3_set_bit(ae_dev->flag, HNAE3_DEV_SUPPORT_FD_B, 1);
1718 hnae3_set_bit(ae_dev->flag, HNAE3_DEV_SUPPORT_GRO_B, 1);
1722 /* hns3_probe - Device initialization routine
1723 * @pdev: PCI device information struct
1724 * @ent: entry in hns3_pci_tbl
1726 * hns3_probe initializes a PF identified by a pci_dev structure.
1727 * The OS initialization, configuring of the PF private structure,
1728 * and a hardware reset occur.
1730 * Returns 0 on success, negative on failure
1732 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1734 struct hnae3_ae_dev *ae_dev;
1737 ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev),
1744 ae_dev->pdev = pdev;
1745 ae_dev->flag = ent->driver_data;
1746 ae_dev->dev_type = HNAE3_DEV_KNIC;
1747 ae_dev->reset_type = HNAE3_NONE_RESET;
1748 hns3_get_dev_capability(pdev, ae_dev);
1749 pci_set_drvdata(pdev, ae_dev);
1751 ret = hnae3_register_ae_dev(ae_dev);
1753 devm_kfree(&pdev->dev, ae_dev);
1754 pci_set_drvdata(pdev, NULL);
1760 /* hns3_remove - Device removal routine
1761 * @pdev: PCI device information struct
1763 static void hns3_remove(struct pci_dev *pdev)
1765 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1767 if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
1768 hns3_disable_sriov(pdev);
1770 hnae3_unregister_ae_dev(ae_dev);
1771 pci_set_drvdata(pdev, NULL);
1775 * hns3_pci_sriov_configure
1776 * @pdev: pointer to a pci_dev structure
1777 * @num_vfs: number of VFs to allocate
1779 * Enable or change the number of VFs. Called when the user updates the number
1782 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
1786 if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
1787 dev_warn(&pdev->dev, "Can not config SRIOV\n");
1792 ret = pci_enable_sriov(pdev, num_vfs);
1794 dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
1797 } else if (!pci_vfs_assigned(pdev)) {
1798 pci_disable_sriov(pdev);
1800 dev_warn(&pdev->dev,
1801 "Unable to free VFs because some are assigned to VMs.\n");
1807 static void hns3_shutdown(struct pci_dev *pdev)
1809 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1811 hnae3_unregister_ae_dev(ae_dev);
1812 devm_kfree(&pdev->dev, ae_dev);
1813 pci_set_drvdata(pdev, NULL);
1815 if (system_state == SYSTEM_POWER_OFF)
1816 pci_set_power_state(pdev, PCI_D3hot);
1819 static pci_ers_result_t hns3_error_detected(struct pci_dev *pdev,
1820 pci_channel_state_t state)
1822 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1823 pci_ers_result_t ret;
1825 dev_info(&pdev->dev, "PCI error detected, state(=%d)!!\n", state);
1827 if (state == pci_channel_io_perm_failure)
1828 return PCI_ERS_RESULT_DISCONNECT;
1832 "Can't recover - error happened during device init\n");
1833 return PCI_ERS_RESULT_NONE;
1836 if (ae_dev->ops->handle_hw_ras_error)
1837 ret = ae_dev->ops->handle_hw_ras_error(ae_dev);
1839 return PCI_ERS_RESULT_NONE;
1844 static pci_ers_result_t hns3_slot_reset(struct pci_dev *pdev)
1846 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1847 struct device *dev = &pdev->dev;
1849 dev_info(dev, "requesting reset due to PCI error\n");
1851 /* request the reset */
1852 if (ae_dev->ops->reset_event) {
1853 ae_dev->ops->reset_event(pdev, NULL);
1854 return PCI_ERS_RESULT_RECOVERED;
1857 return PCI_ERS_RESULT_DISCONNECT;
1860 static void hns3_reset_prepare(struct pci_dev *pdev)
1862 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1864 dev_info(&pdev->dev, "hns3 flr prepare\n");
1865 if (ae_dev && ae_dev->ops && ae_dev->ops->flr_prepare)
1866 ae_dev->ops->flr_prepare(ae_dev);
1869 static void hns3_reset_done(struct pci_dev *pdev)
1871 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1873 dev_info(&pdev->dev, "hns3 flr done\n");
1874 if (ae_dev && ae_dev->ops && ae_dev->ops->flr_done)
1875 ae_dev->ops->flr_done(ae_dev);
1878 static const struct pci_error_handlers hns3_err_handler = {
1879 .error_detected = hns3_error_detected,
1880 .slot_reset = hns3_slot_reset,
1881 .reset_prepare = hns3_reset_prepare,
1882 .reset_done = hns3_reset_done,
1885 static struct pci_driver hns3_driver = {
1886 .name = hns3_driver_name,
1887 .id_table = hns3_pci_tbl,
1888 .probe = hns3_probe,
1889 .remove = hns3_remove,
1890 .shutdown = hns3_shutdown,
1891 .sriov_configure = hns3_pci_sriov_configure,
1892 .err_handler = &hns3_err_handler,
1895 /* set default feature to hns3 */
1896 static void hns3_set_default_feature(struct net_device *netdev)
1898 struct hnae3_handle *h = hns3_get_handle(netdev);
1899 struct pci_dev *pdev = h->pdev;
1901 netdev->priv_flags |= IFF_UNICAST_FLT;
1903 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1904 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1905 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1906 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1907 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1909 netdev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
1911 netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
1913 netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1914 NETIF_F_HW_VLAN_CTAG_FILTER |
1915 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
1916 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1917 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1918 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1919 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1921 netdev->vlan_features |=
1922 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
1923 NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO |
1924 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1925 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1926 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1928 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1929 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
1930 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1931 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1932 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1933 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1935 if (pdev->revision >= 0x21) {
1936 netdev->hw_features |= NETIF_F_GRO_HW;
1937 netdev->features |= NETIF_F_GRO_HW;
1939 if (!(h->flags & HNAE3_SUPPORT_VF)) {
1940 netdev->hw_features |= NETIF_F_NTUPLE;
1941 netdev->features |= NETIF_F_NTUPLE;
1946 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
1947 struct hns3_desc_cb *cb)
1949 unsigned int order = hnae3_page_order(ring);
1952 p = dev_alloc_pages(order);
1957 cb->page_offset = 0;
1959 cb->buf = page_address(p);
1960 cb->length = hnae3_page_size(ring);
1961 cb->type = DESC_TYPE_PAGE;
1966 static void hns3_free_buffer(struct hns3_enet_ring *ring,
1967 struct hns3_desc_cb *cb)
1969 if (cb->type == DESC_TYPE_SKB)
1970 dev_kfree_skb_any((struct sk_buff *)cb->priv);
1971 else if (!HNAE3_IS_TX_RING(ring))
1972 put_page((struct page *)cb->priv);
1973 memset(cb, 0, sizeof(*cb));
1976 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
1978 cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
1979 cb->length, ring_to_dma_dir(ring));
1981 if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
1987 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
1988 struct hns3_desc_cb *cb)
1990 if (cb->type == DESC_TYPE_SKB)
1991 dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
1992 ring_to_dma_dir(ring));
1993 else if (cb->length)
1994 dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
1995 ring_to_dma_dir(ring));
1998 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
2000 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2001 ring->desc[i].addr = 0;
2004 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i)
2006 struct hns3_desc_cb *cb = &ring->desc_cb[i];
2008 if (!ring->desc_cb[i].dma)
2011 hns3_buffer_detach(ring, i);
2012 hns3_free_buffer(ring, cb);
2015 static void hns3_free_buffers(struct hns3_enet_ring *ring)
2019 for (i = 0; i < ring->desc_num; i++)
2020 hns3_free_buffer_detach(ring, i);
2023 /* free desc along with its attached buffer */
2024 static void hns3_free_desc(struct hns3_enet_ring *ring)
2026 int size = ring->desc_num * sizeof(ring->desc[0]);
2028 hns3_free_buffers(ring);
2031 dma_free_coherent(ring_to_dev(ring), size,
2032 ring->desc, ring->desc_dma_addr);
2037 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
2039 int size = ring->desc_num * sizeof(ring->desc[0]);
2041 ring->desc = dma_alloc_coherent(ring_to_dev(ring), size,
2042 &ring->desc_dma_addr, GFP_KERNEL);
2049 static int hns3_reserve_buffer_map(struct hns3_enet_ring *ring,
2050 struct hns3_desc_cb *cb)
2054 ret = hns3_alloc_buffer(ring, cb);
2058 ret = hns3_map_buffer(ring, cb);
2065 hns3_free_buffer(ring, cb);
2070 static int hns3_alloc_buffer_attach(struct hns3_enet_ring *ring, int i)
2072 int ret = hns3_reserve_buffer_map(ring, &ring->desc_cb[i]);
2077 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2082 /* Allocate memory for raw pkg, and map with dma */
2083 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
2087 for (i = 0; i < ring->desc_num; i++) {
2088 ret = hns3_alloc_buffer_attach(ring, i);
2090 goto out_buffer_fail;
2096 for (j = i - 1; j >= 0; j--)
2097 hns3_free_buffer_detach(ring, j);
2101 /* detach a in-used buffer and replace with a reserved one */
2102 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
2103 struct hns3_desc_cb *res_cb)
2105 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2106 ring->desc_cb[i] = *res_cb;
2107 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2108 ring->desc[i].rx.bd_base_info = 0;
2111 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
2113 ring->desc_cb[i].reuse_flag = 0;
2114 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma
2115 + ring->desc_cb[i].page_offset);
2116 ring->desc[i].rx.bd_base_info = 0;
2119 static void hns3_nic_reclaim_one_desc(struct hns3_enet_ring *ring, int *bytes,
2122 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
2124 (*pkts) += (desc_cb->type == DESC_TYPE_SKB);
2125 (*bytes) += desc_cb->length;
2126 /* desc_cb will be cleaned, after hnae3_free_buffer_detach*/
2127 hns3_free_buffer_detach(ring, ring->next_to_clean);
2129 ring_ptr_move_fw(ring, next_to_clean);
2132 static int is_valid_clean_head(struct hns3_enet_ring *ring, int h)
2134 int u = ring->next_to_use;
2135 int c = ring->next_to_clean;
2137 if (unlikely(h > ring->desc_num))
2140 return u > c ? (h > c && h <= u) : (h > c || h <= u);
2143 void hns3_clean_tx_ring(struct hns3_enet_ring *ring)
2145 struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2146 struct hns3_nic_priv *priv = netdev_priv(netdev);
2147 struct netdev_queue *dev_queue;
2151 head = readl_relaxed(ring->tqp->io_base + HNS3_RING_TX_RING_HEAD_REG);
2152 rmb(); /* Make sure head is ready before touch any data */
2154 if (is_ring_empty(ring) || head == ring->next_to_clean)
2155 return; /* no data to poll */
2157 if (unlikely(!is_valid_clean_head(ring, head))) {
2158 netdev_err(netdev, "wrong head (%d, %d-%d)\n", head,
2159 ring->next_to_use, ring->next_to_clean);
2161 u64_stats_update_begin(&ring->syncp);
2162 ring->stats.io_err_cnt++;
2163 u64_stats_update_end(&ring->syncp);
2169 while (head != ring->next_to_clean) {
2170 hns3_nic_reclaim_one_desc(ring, &bytes, &pkts);
2171 /* Issue prefetch for next Tx descriptor */
2172 prefetch(&ring->desc_cb[ring->next_to_clean]);
2175 ring->tqp_vector->tx_group.total_bytes += bytes;
2176 ring->tqp_vector->tx_group.total_packets += pkts;
2178 u64_stats_update_begin(&ring->syncp);
2179 ring->stats.tx_bytes += bytes;
2180 ring->stats.tx_pkts += pkts;
2181 u64_stats_update_end(&ring->syncp);
2183 dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
2184 netdev_tx_completed_queue(dev_queue, pkts, bytes);
2186 if (unlikely(pkts && netif_carrier_ok(netdev) &&
2187 (ring_space(ring) > HNS3_MAX_BD_PER_PKT))) {
2188 /* Make sure that anybody stopping the queue after this
2189 * sees the new next_to_clean.
2192 if (netif_tx_queue_stopped(dev_queue) &&
2193 !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
2194 netif_tx_wake_queue(dev_queue);
2195 ring->stats.restart_queue++;
2200 static int hns3_desc_unused(struct hns3_enet_ring *ring)
2202 int ntc = ring->next_to_clean;
2203 int ntu = ring->next_to_use;
2205 return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
2209 hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring, int cleand_count)
2211 struct hns3_desc_cb *desc_cb;
2212 struct hns3_desc_cb res_cbs;
2215 for (i = 0; i < cleand_count; i++) {
2216 desc_cb = &ring->desc_cb[ring->next_to_use];
2217 if (desc_cb->reuse_flag) {
2218 u64_stats_update_begin(&ring->syncp);
2219 ring->stats.reuse_pg_cnt++;
2220 u64_stats_update_end(&ring->syncp);
2222 hns3_reuse_buffer(ring, ring->next_to_use);
2224 ret = hns3_reserve_buffer_map(ring, &res_cbs);
2226 u64_stats_update_begin(&ring->syncp);
2227 ring->stats.sw_err_cnt++;
2228 u64_stats_update_end(&ring->syncp);
2230 netdev_err(ring->tqp->handle->kinfo.netdev,
2231 "hnae reserve buffer map failed.\n");
2234 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
2237 ring_ptr_move_fw(ring, next_to_use);
2240 wmb(); /* Make all data has been write before submit */
2241 writel_relaxed(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
2244 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
2245 struct hns3_enet_ring *ring, int pull_len,
2246 struct hns3_desc_cb *desc_cb)
2248 struct hns3_desc *desc;
2254 twobufs = ((PAGE_SIZE < 8192) &&
2255 hnae3_buf_size(ring) == HNS3_BUFFER_SIZE_2048);
2257 desc = &ring->desc[ring->next_to_clean];
2258 size = le16_to_cpu(desc->rx.size);
2260 truesize = hnae3_buf_size(ring);
2263 last_offset = hnae3_page_size(ring) - hnae3_buf_size(ring);
2265 skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
2266 size - pull_len, truesize);
2268 /* Avoid re-using remote pages,flag default unreuse */
2269 if (unlikely(page_to_nid(desc_cb->priv) != numa_node_id()))
2273 /* If we are only owner of page we can reuse it */
2274 if (likely(page_count(desc_cb->priv) == 1)) {
2275 /* Flip page offset to other buffer */
2276 desc_cb->page_offset ^= truesize;
2278 desc_cb->reuse_flag = 1;
2279 /* bump ref count on page before it is given*/
2280 get_page(desc_cb->priv);
2285 /* Move offset up to the next cache line */
2286 desc_cb->page_offset += truesize;
2288 if (desc_cb->page_offset <= last_offset) {
2289 desc_cb->reuse_flag = 1;
2290 /* Bump ref count on page before it is given*/
2291 get_page(desc_cb->priv);
2295 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
2296 struct hns3_desc *desc)
2298 struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2299 int l3_type, l4_type;
2304 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2305 l234info = le32_to_cpu(desc->rx.l234_info);
2307 skb->ip_summed = CHECKSUM_NONE;
2309 skb_checksum_none_assert(skb);
2311 if (!(netdev->features & NETIF_F_RXCSUM))
2314 /* We MUST enable hardware checksum before enabling hardware GRO */
2315 if (skb_shinfo(skb)->gso_size) {
2316 skb->ip_summed = CHECKSUM_UNNECESSARY;
2320 /* check if hardware has done checksum */
2321 if (!(bd_base_info & BIT(HNS3_RXD_L3L4P_B)))
2324 if (unlikely(l234info & (BIT(HNS3_RXD_L3E_B) | BIT(HNS3_RXD_L4E_B) ||
2325 BIT(HNS3_RXD_OL3E_B) ||
2326 BIT(HNS3_RXD_OL4E_B)))) {
2327 u64_stats_update_begin(&ring->syncp);
2328 ring->stats.l3l4_csum_err++;
2329 u64_stats_update_end(&ring->syncp);
2334 ol4_type = hnae3_get_field(l234info, HNS3_RXD_OL4ID_M,
2337 case HNS3_OL4_TYPE_MAC_IN_UDP:
2338 case HNS3_OL4_TYPE_NVGRE:
2339 skb->csum_level = 1;
2341 case HNS3_OL4_TYPE_NO_TUN:
2342 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
2344 l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
2347 /* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
2348 if ((l3_type == HNS3_L3_TYPE_IPV4 ||
2349 l3_type == HNS3_L3_TYPE_IPV6) &&
2350 (l4_type == HNS3_L4_TYPE_UDP ||
2351 l4_type == HNS3_L4_TYPE_TCP ||
2352 l4_type == HNS3_L4_TYPE_SCTP))
2353 skb->ip_summed = CHECKSUM_UNNECESSARY;
2360 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
2362 if (skb_has_frag_list(skb))
2363 napi_gro_flush(&ring->tqp_vector->napi, false);
2365 napi_gro_receive(&ring->tqp_vector->napi, skb);
2368 static bool hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
2369 struct hns3_desc *desc, u32 l234info,
2372 struct pci_dev *pdev = ring->tqp->handle->pdev;
2374 if (pdev->revision == 0x20) {
2375 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2376 if (!(*vlan_tag & VLAN_VID_MASK))
2377 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2379 return (*vlan_tag != 0);
2382 #define HNS3_STRP_OUTER_VLAN 0x1
2383 #define HNS3_STRP_INNER_VLAN 0x2
2385 switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
2386 HNS3_RXD_STRP_TAGP_S)) {
2387 case HNS3_STRP_OUTER_VLAN:
2388 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2390 case HNS3_STRP_INNER_VLAN:
2391 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2398 static int hns3_alloc_skb(struct hns3_enet_ring *ring, int length,
2401 #define HNS3_NEED_ADD_FRAG 1
2402 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
2403 struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2404 struct sk_buff *skb;
2406 ring->skb = napi_alloc_skb(&ring->tqp_vector->napi, HNS3_RX_HEAD_SIZE);
2408 if (unlikely(!skb)) {
2409 netdev_err(netdev, "alloc rx skb fail\n");
2411 u64_stats_update_begin(&ring->syncp);
2412 ring->stats.sw_err_cnt++;
2413 u64_stats_update_end(&ring->syncp);
2418 prefetchw(skb->data);
2420 ring->pending_buf = 1;
2422 ring->tail_skb = NULL;
2423 if (length <= HNS3_RX_HEAD_SIZE) {
2424 memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
2426 /* We can reuse buffer as-is, just make sure it is local */
2427 if (likely(page_to_nid(desc_cb->priv) == numa_node_id()))
2428 desc_cb->reuse_flag = 1;
2429 else /* This page cannot be reused so discard it */
2430 put_page(desc_cb->priv);
2432 ring_ptr_move_fw(ring, next_to_clean);
2435 u64_stats_update_begin(&ring->syncp);
2436 ring->stats.seg_pkt_cnt++;
2437 u64_stats_update_end(&ring->syncp);
2439 ring->pull_len = eth_get_headlen(va, HNS3_RX_HEAD_SIZE);
2440 __skb_put(skb, ring->pull_len);
2441 hns3_nic_reuse_page(skb, ring->frag_num++, ring, ring->pull_len,
2443 ring_ptr_move_fw(ring, next_to_clean);
2445 return HNS3_NEED_ADD_FRAG;
2448 static int hns3_add_frag(struct hns3_enet_ring *ring, struct hns3_desc *desc,
2449 struct sk_buff **out_skb, bool pending)
2451 struct sk_buff *skb = *out_skb;
2452 struct sk_buff *head_skb = *out_skb;
2453 struct sk_buff *new_skb;
2454 struct hns3_desc_cb *desc_cb;
2455 struct hns3_desc *pre_desc;
2459 /* if there is pending bd, the SW param next_to_clean has moved
2460 * to next and the next is NULL
2463 pre_bd = (ring->next_to_clean - 1 + ring->desc_num) %
2465 pre_desc = &ring->desc[pre_bd];
2466 bd_base_info = le32_to_cpu(pre_desc->rx.bd_base_info);
2468 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2471 while (!(bd_base_info & BIT(HNS3_RXD_FE_B))) {
2472 desc = &ring->desc[ring->next_to_clean];
2473 desc_cb = &ring->desc_cb[ring->next_to_clean];
2474 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2475 if (!(bd_base_info & BIT(HNS3_RXD_VLD_B)))
2478 if (unlikely(ring->frag_num >= MAX_SKB_FRAGS)) {
2479 new_skb = napi_alloc_skb(&ring->tqp_vector->napi,
2481 if (unlikely(!new_skb)) {
2482 netdev_err(ring->tqp->handle->kinfo.netdev,
2483 "alloc rx skb frag fail\n");
2488 if (ring->tail_skb) {
2489 ring->tail_skb->next = new_skb;
2490 ring->tail_skb = new_skb;
2492 skb_shinfo(skb)->frag_list = new_skb;
2493 ring->tail_skb = new_skb;
2497 if (ring->tail_skb) {
2498 head_skb->truesize += hnae3_buf_size(ring);
2499 head_skb->data_len += le16_to_cpu(desc->rx.size);
2500 head_skb->len += le16_to_cpu(desc->rx.size);
2501 skb = ring->tail_skb;
2504 hns3_nic_reuse_page(skb, ring->frag_num++, ring, 0, desc_cb);
2505 ring_ptr_move_fw(ring, next_to_clean);
2506 ring->pending_buf++;
2512 static void hns3_set_gro_param(struct sk_buff *skb, u32 l234info,
2518 gro_count = hnae3_get_field(l234info, HNS3_RXD_GRO_COUNT_M,
2519 HNS3_RXD_GRO_COUNT_S);
2520 /* if there is no HW GRO, do not set gro params */
2524 /* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
2525 * to skb_shinfo(skb)->gso_segs
2527 NAPI_GRO_CB(skb)->count = gro_count;
2529 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
2531 if (l3_type == HNS3_L3_TYPE_IPV4)
2532 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2533 else if (l3_type == HNS3_L3_TYPE_IPV6)
2534 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
2538 skb_shinfo(skb)->gso_size = hnae3_get_field(bd_base_info,
2539 HNS3_RXD_GRO_SIZE_M,
2540 HNS3_RXD_GRO_SIZE_S);
2541 if (skb_shinfo(skb)->gso_size)
2542 tcp_gro_complete(skb);
2545 static void hns3_set_rx_skb_rss_type(struct hns3_enet_ring *ring,
2546 struct sk_buff *skb)
2548 struct hnae3_handle *handle = ring->tqp->handle;
2549 enum pkt_hash_types rss_type;
2550 struct hns3_desc *desc;
2553 /* When driver handle the rss type, ring->next_to_clean indicates the
2554 * first descriptor of next packet, need -1 here.
2556 last_bd = (ring->next_to_clean - 1 + ring->desc_num) % ring->desc_num;
2557 desc = &ring->desc[last_bd];
2559 if (le32_to_cpu(desc->rx.rss_hash))
2560 rss_type = handle->kinfo.rss_type;
2562 rss_type = PKT_HASH_TYPE_NONE;
2564 skb_set_hash(skb, le32_to_cpu(desc->rx.rss_hash), rss_type);
2567 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring,
2568 struct sk_buff **out_skb)
2570 struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2571 enum hns3_pkt_l2t_type l2_frame_type;
2572 struct sk_buff *skb = ring->skb;
2573 struct hns3_desc_cb *desc_cb;
2574 struct hns3_desc *desc;
2580 desc = &ring->desc[ring->next_to_clean];
2581 desc_cb = &ring->desc_cb[ring->next_to_clean];
2585 length = le16_to_cpu(desc->rx.size);
2586 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2588 /* Check valid BD */
2589 if (unlikely(!(bd_base_info & BIT(HNS3_RXD_VLD_B))))
2593 ring->va = (unsigned char *)desc_cb->buf + desc_cb->page_offset;
2595 /* Prefetch first cache line of first page
2596 * Idea is to cache few bytes of the header of the packet. Our L1 Cache
2597 * line size is 64B so need to prefetch twice to make it 128B. But in
2598 * actual we can have greater size of caches with 128B Level 1 cache
2599 * lines. In such a case, single fetch would suffice to cache in the
2600 * relevant part of the header.
2603 #if L1_CACHE_BYTES < 128
2604 prefetch(ring->va + L1_CACHE_BYTES);
2608 ret = hns3_alloc_skb(ring, length, ring->va);
2609 *out_skb = skb = ring->skb;
2611 if (ret < 0) /* alloc buffer fail */
2613 if (ret > 0) { /* need add frag */
2614 ret = hns3_add_frag(ring, desc, &skb, false);
2618 /* As the head data may be changed when GRO enable, copy
2619 * the head data in after other data rx completed
2621 memcpy(skb->data, ring->va,
2622 ALIGN(ring->pull_len, sizeof(long)));
2625 ret = hns3_add_frag(ring, desc, &skb, true);
2629 /* As the head data may be changed when GRO enable, copy
2630 * the head data in after other data rx completed
2632 memcpy(skb->data, ring->va,
2633 ALIGN(ring->pull_len, sizeof(long)));
2636 l234info = le32_to_cpu(desc->rx.l234_info);
2637 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2639 /* Based on hw strategy, the tag offloaded will be stored at
2640 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
2641 * in one layer tag case.
2643 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
2646 if (hns3_parse_vlan_tag(ring, desc, l234info, &vlan_tag))
2647 __vlan_hwaccel_put_tag(skb,
2652 if (unlikely(!(bd_base_info & BIT(HNS3_RXD_VLD_B)))) {
2653 u64_stats_update_begin(&ring->syncp);
2654 ring->stats.non_vld_descs++;
2655 u64_stats_update_end(&ring->syncp);
2657 dev_kfree_skb_any(skb);
2661 if (unlikely((!desc->rx.pkt_len) ||
2662 (l234info & (BIT(HNS3_RXD_TRUNCAT_B) |
2663 BIT(HNS3_RXD_L2E_B))))) {
2664 u64_stats_update_begin(&ring->syncp);
2665 if (l234info & BIT(HNS3_RXD_L2E_B))
2666 ring->stats.l2_err++;
2668 ring->stats.err_pkt_len++;
2669 u64_stats_update_end(&ring->syncp);
2671 dev_kfree_skb_any(skb);
2676 l2_frame_type = hnae3_get_field(l234info, HNS3_RXD_DMAC_M,
2678 u64_stats_update_begin(&ring->syncp);
2679 if (l2_frame_type == HNS3_L2_TYPE_MULTICAST)
2680 ring->stats.rx_multicast++;
2682 ring->stats.rx_pkts++;
2683 ring->stats.rx_bytes += skb->len;
2684 u64_stats_update_end(&ring->syncp);
2686 ring->tqp_vector->rx_group.total_bytes += skb->len;
2688 /* This is needed in order to enable forwarding support */
2689 hns3_set_gro_param(skb, l234info, bd_base_info);
2691 hns3_rx_checksum(ring, skb, desc);
2693 hns3_set_rx_skb_rss_type(ring, skb);
2698 int hns3_clean_rx_ring(
2699 struct hns3_enet_ring *ring, int budget,
2700 void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
2702 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
2703 struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2704 int recv_pkts, recv_bds, clean_count, err;
2705 int unused_count = hns3_desc_unused(ring) - ring->pending_buf;
2706 struct sk_buff *skb = ring->skb;
2709 num = readl_relaxed(ring->tqp->io_base + HNS3_RING_RX_RING_FBDNUM_REG);
2710 rmb(); /* Make sure num taken effect before the other data is touched */
2712 recv_pkts = 0, recv_bds = 0, clean_count = 0;
2713 num -= unused_count;
2715 while (recv_pkts < budget && recv_bds < num) {
2716 /* Reuse or realloc buffers */
2717 if (clean_count + unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
2718 hns3_nic_alloc_rx_buffers(ring,
2719 clean_count + unused_count);
2721 unused_count = hns3_desc_unused(ring) -
2726 err = hns3_handle_rx_bd(ring, &skb);
2727 if (unlikely(!skb)) /* This fault cannot be repaired */
2730 if (err == -ENXIO) { /* Do not get FE for the packet */
2732 } else if (unlikely(err)) { /* Do jump the err */
2733 recv_bds += ring->pending_buf;
2734 clean_count += ring->pending_buf;
2736 ring->pending_buf = 0;
2740 /* Do update ip stack process */
2741 skb->protocol = eth_type_trans(skb, netdev);
2743 recv_bds += ring->pending_buf;
2744 clean_count += ring->pending_buf;
2746 ring->pending_buf = 0;
2752 /* Make all data has been write before submit */
2753 if (clean_count + unused_count > 0)
2754 hns3_nic_alloc_rx_buffers(ring,
2755 clean_count + unused_count);
2760 static bool hns3_get_new_int_gl(struct hns3_enet_ring_group *ring_group)
2762 struct hns3_enet_tqp_vector *tqp_vector =
2763 ring_group->ring->tqp_vector;
2764 enum hns3_flow_level_range new_flow_level;
2765 int packets_per_msecs;
2766 int bytes_per_msecs;
2770 if (!tqp_vector->last_jiffies)
2773 if (ring_group->total_packets == 0) {
2774 ring_group->coal.int_gl = HNS3_INT_GL_50K;
2775 ring_group->coal.flow_level = HNS3_FLOW_LOW;
2779 /* Simple throttlerate management
2780 * 0-10MB/s lower (50000 ints/s)
2781 * 10-20MB/s middle (20000 ints/s)
2782 * 20-1249MB/s high (18000 ints/s)
2783 * > 40000pps ultra (8000 ints/s)
2785 new_flow_level = ring_group->coal.flow_level;
2786 new_int_gl = ring_group->coal.int_gl;
2788 jiffies_to_msecs(jiffies - tqp_vector->last_jiffies);
2790 if (!time_passed_ms)
2793 do_div(ring_group->total_packets, time_passed_ms);
2794 packets_per_msecs = ring_group->total_packets;
2796 do_div(ring_group->total_bytes, time_passed_ms);
2797 bytes_per_msecs = ring_group->total_bytes;
2799 #define HNS3_RX_LOW_BYTE_RATE 10000
2800 #define HNS3_RX_MID_BYTE_RATE 20000
2802 switch (new_flow_level) {
2804 if (bytes_per_msecs > HNS3_RX_LOW_BYTE_RATE)
2805 new_flow_level = HNS3_FLOW_MID;
2808 if (bytes_per_msecs > HNS3_RX_MID_BYTE_RATE)
2809 new_flow_level = HNS3_FLOW_HIGH;
2810 else if (bytes_per_msecs <= HNS3_RX_LOW_BYTE_RATE)
2811 new_flow_level = HNS3_FLOW_LOW;
2813 case HNS3_FLOW_HIGH:
2814 case HNS3_FLOW_ULTRA:
2816 if (bytes_per_msecs <= HNS3_RX_MID_BYTE_RATE)
2817 new_flow_level = HNS3_FLOW_MID;
2821 #define HNS3_RX_ULTRA_PACKET_RATE 40
2823 if (packets_per_msecs > HNS3_RX_ULTRA_PACKET_RATE &&
2824 &tqp_vector->rx_group == ring_group)
2825 new_flow_level = HNS3_FLOW_ULTRA;
2827 switch (new_flow_level) {
2829 new_int_gl = HNS3_INT_GL_50K;
2832 new_int_gl = HNS3_INT_GL_20K;
2834 case HNS3_FLOW_HIGH:
2835 new_int_gl = HNS3_INT_GL_18K;
2837 case HNS3_FLOW_ULTRA:
2838 new_int_gl = HNS3_INT_GL_8K;
2844 ring_group->total_bytes = 0;
2845 ring_group->total_packets = 0;
2846 ring_group->coal.flow_level = new_flow_level;
2847 if (new_int_gl != ring_group->coal.int_gl) {
2848 ring_group->coal.int_gl = new_int_gl;
2854 static void hns3_update_new_int_gl(struct hns3_enet_tqp_vector *tqp_vector)
2856 struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
2857 struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
2858 bool rx_update, tx_update;
2860 /* update param every 1000ms */
2861 if (time_before(jiffies,
2862 tqp_vector->last_jiffies + msecs_to_jiffies(1000)))
2865 if (rx_group->coal.gl_adapt_enable) {
2866 rx_update = hns3_get_new_int_gl(rx_group);
2868 hns3_set_vector_coalesce_rx_gl(tqp_vector,
2869 rx_group->coal.int_gl);
2872 if (tx_group->coal.gl_adapt_enable) {
2873 tx_update = hns3_get_new_int_gl(tx_group);
2875 hns3_set_vector_coalesce_tx_gl(tqp_vector,
2876 tx_group->coal.int_gl);
2879 tqp_vector->last_jiffies = jiffies;
2882 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
2884 struct hns3_nic_priv *priv = netdev_priv(napi->dev);
2885 struct hns3_enet_ring *ring;
2886 int rx_pkt_total = 0;
2888 struct hns3_enet_tqp_vector *tqp_vector =
2889 container_of(napi, struct hns3_enet_tqp_vector, napi);
2890 bool clean_complete = true;
2893 if (unlikely(test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
2894 napi_complete(napi);
2898 /* Since the actual Tx work is minimal, we can give the Tx a larger
2899 * budget and be more aggressive about cleaning up the Tx descriptors.
2901 hns3_for_each_ring(ring, tqp_vector->tx_group)
2902 hns3_clean_tx_ring(ring);
2904 /* make sure rx ring budget not smaller than 1 */
2905 rx_budget = max(budget / tqp_vector->num_tqps, 1);
2907 hns3_for_each_ring(ring, tqp_vector->rx_group) {
2908 int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
2911 if (rx_cleaned >= rx_budget)
2912 clean_complete = false;
2914 rx_pkt_total += rx_cleaned;
2917 tqp_vector->rx_group.total_packets += rx_pkt_total;
2919 if (!clean_complete)
2922 if (napi_complete(napi) &&
2923 likely(!test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
2924 hns3_update_new_int_gl(tqp_vector);
2925 hns3_mask_vector_irq(tqp_vector, 1);
2928 return rx_pkt_total;
2931 static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
2932 struct hnae3_ring_chain_node *head)
2934 struct pci_dev *pdev = tqp_vector->handle->pdev;
2935 struct hnae3_ring_chain_node *cur_chain = head;
2936 struct hnae3_ring_chain_node *chain;
2937 struct hns3_enet_ring *tx_ring;
2938 struct hns3_enet_ring *rx_ring;
2940 tx_ring = tqp_vector->tx_group.ring;
2942 cur_chain->tqp_index = tx_ring->tqp->tqp_index;
2943 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
2944 HNAE3_RING_TYPE_TX);
2945 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
2946 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_TX);
2948 cur_chain->next = NULL;
2950 while (tx_ring->next) {
2951 tx_ring = tx_ring->next;
2953 chain = devm_kzalloc(&pdev->dev, sizeof(*chain),
2956 goto err_free_chain;
2958 cur_chain->next = chain;
2959 chain->tqp_index = tx_ring->tqp->tqp_index;
2960 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
2961 HNAE3_RING_TYPE_TX);
2962 hnae3_set_field(chain->int_gl_idx,
2963 HNAE3_RING_GL_IDX_M,
2964 HNAE3_RING_GL_IDX_S,
2971 rx_ring = tqp_vector->rx_group.ring;
2972 if (!tx_ring && rx_ring) {
2973 cur_chain->next = NULL;
2974 cur_chain->tqp_index = rx_ring->tqp->tqp_index;
2975 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
2976 HNAE3_RING_TYPE_RX);
2977 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
2978 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
2980 rx_ring = rx_ring->next;
2984 chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
2986 goto err_free_chain;
2988 cur_chain->next = chain;
2989 chain->tqp_index = rx_ring->tqp->tqp_index;
2990 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
2991 HNAE3_RING_TYPE_RX);
2992 hnae3_set_field(chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
2993 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
2997 rx_ring = rx_ring->next;
3003 cur_chain = head->next;
3005 chain = cur_chain->next;
3006 devm_kfree(&pdev->dev, cur_chain);
3014 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
3015 struct hnae3_ring_chain_node *head)
3017 struct pci_dev *pdev = tqp_vector->handle->pdev;
3018 struct hnae3_ring_chain_node *chain_tmp, *chain;
3023 chain_tmp = chain->next;
3024 devm_kfree(&pdev->dev, chain);
3029 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
3030 struct hns3_enet_ring *ring)
3032 ring->next = group->ring;
3038 static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
3040 struct pci_dev *pdev = priv->ae_handle->pdev;
3041 struct hns3_enet_tqp_vector *tqp_vector;
3042 int num_vectors = priv->vector_num;
3046 numa_node = dev_to_node(&pdev->dev);
3048 for (vector_i = 0; vector_i < num_vectors; vector_i++) {
3049 tqp_vector = &priv->tqp_vector[vector_i];
3050 cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
3051 &tqp_vector->affinity_mask);
3055 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
3057 struct hnae3_ring_chain_node vector_ring_chain;
3058 struct hnae3_handle *h = priv->ae_handle;
3059 struct hns3_enet_tqp_vector *tqp_vector;
3063 hns3_nic_set_cpumask(priv);
3065 for (i = 0; i < priv->vector_num; i++) {
3066 tqp_vector = &priv->tqp_vector[i];
3067 hns3_vector_gl_rl_init_hw(tqp_vector, priv);
3068 tqp_vector->num_tqps = 0;
3071 for (i = 0; i < h->kinfo.num_tqps; i++) {
3072 u16 vector_i = i % priv->vector_num;
3073 u16 tqp_num = h->kinfo.num_tqps;
3075 tqp_vector = &priv->tqp_vector[vector_i];
3077 hns3_add_ring_to_group(&tqp_vector->tx_group,
3078 priv->ring_data[i].ring);
3080 hns3_add_ring_to_group(&tqp_vector->rx_group,
3081 priv->ring_data[i + tqp_num].ring);
3083 priv->ring_data[i].ring->tqp_vector = tqp_vector;
3084 priv->ring_data[i + tqp_num].ring->tqp_vector = tqp_vector;
3085 tqp_vector->num_tqps++;
3088 for (i = 0; i < priv->vector_num; i++) {
3089 tqp_vector = &priv->tqp_vector[i];
3091 tqp_vector->rx_group.total_bytes = 0;
3092 tqp_vector->rx_group.total_packets = 0;
3093 tqp_vector->tx_group.total_bytes = 0;
3094 tqp_vector->tx_group.total_packets = 0;
3095 tqp_vector->handle = h;
3097 ret = hns3_get_vector_ring_chain(tqp_vector,
3098 &vector_ring_chain);
3102 ret = h->ae_algo->ops->map_ring_to_vector(h,
3103 tqp_vector->vector_irq, &vector_ring_chain);
3105 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3110 netif_napi_add(priv->netdev, &tqp_vector->napi,
3111 hns3_nic_common_poll, NAPI_POLL_WEIGHT);
3118 netif_napi_del(&priv->tqp_vector[i].napi);
3123 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
3125 #define HNS3_VECTOR_PF_MAX_NUM 64
3127 struct hnae3_handle *h = priv->ae_handle;
3128 struct hns3_enet_tqp_vector *tqp_vector;
3129 struct hnae3_vector_info *vector;
3130 struct pci_dev *pdev = h->pdev;
3131 u16 tqp_num = h->kinfo.num_tqps;
3136 /* RSS size, cpu online and vector_num should be the same */
3137 /* Should consider 2p/4p later */
3138 vector_num = min_t(u16, num_online_cpus(), tqp_num);
3139 vector_num = min_t(u16, vector_num, HNS3_VECTOR_PF_MAX_NUM);
3141 vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
3146 vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
3148 priv->vector_num = vector_num;
3149 priv->tqp_vector = (struct hns3_enet_tqp_vector *)
3150 devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
3152 if (!priv->tqp_vector) {
3157 for (i = 0; i < priv->vector_num; i++) {
3158 tqp_vector = &priv->tqp_vector[i];
3159 tqp_vector->idx = i;
3160 tqp_vector->mask_addr = vector[i].io_addr;
3161 tqp_vector->vector_irq = vector[i].vector;
3162 hns3_vector_gl_rl_init(tqp_vector, priv);
3166 devm_kfree(&pdev->dev, vector);
3170 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
3176 static void hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
3178 struct hnae3_ring_chain_node vector_ring_chain;
3179 struct hnae3_handle *h = priv->ae_handle;
3180 struct hns3_enet_tqp_vector *tqp_vector;
3183 for (i = 0; i < priv->vector_num; i++) {
3184 tqp_vector = &priv->tqp_vector[i];
3186 if (!tqp_vector->rx_group.ring && !tqp_vector->tx_group.ring)
3189 hns3_get_vector_ring_chain(tqp_vector, &vector_ring_chain);
3191 h->ae_algo->ops->unmap_ring_from_vector(h,
3192 tqp_vector->vector_irq, &vector_ring_chain);
3194 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3196 if (tqp_vector->irq_init_flag == HNS3_VECTOR_INITED) {
3197 irq_set_affinity_notifier(tqp_vector->vector_irq,
3199 irq_set_affinity_hint(tqp_vector->vector_irq, NULL);
3200 free_irq(tqp_vector->vector_irq, tqp_vector);
3201 tqp_vector->irq_init_flag = HNS3_VECTOR_NOT_INITED;
3204 hns3_clear_ring_group(&tqp_vector->rx_group);
3205 hns3_clear_ring_group(&tqp_vector->tx_group);
3206 netif_napi_del(&priv->tqp_vector[i].napi);
3210 static int hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
3212 struct hnae3_handle *h = priv->ae_handle;
3213 struct pci_dev *pdev = h->pdev;
3216 for (i = 0; i < priv->vector_num; i++) {
3217 struct hns3_enet_tqp_vector *tqp_vector;
3219 tqp_vector = &priv->tqp_vector[i];
3220 ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
3225 devm_kfree(&pdev->dev, priv->tqp_vector);
3229 static int hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
3232 struct hns3_nic_ring_data *ring_data = priv->ring_data;
3233 int queue_num = priv->ae_handle->kinfo.num_tqps;
3234 struct pci_dev *pdev = priv->ae_handle->pdev;
3235 struct hns3_enet_ring *ring;
3238 ring = devm_kzalloc(&pdev->dev, sizeof(*ring), GFP_KERNEL);
3242 if (ring_type == HNAE3_RING_TYPE_TX) {
3243 desc_num = priv->ae_handle->kinfo.num_tx_desc;
3244 ring_data[q->tqp_index].ring = ring;
3245 ring_data[q->tqp_index].queue_index = q->tqp_index;
3246 ring->io_base = (u8 __iomem *)q->io_base + HNS3_TX_REG_OFFSET;
3248 desc_num = priv->ae_handle->kinfo.num_rx_desc;
3249 ring_data[q->tqp_index + queue_num].ring = ring;
3250 ring_data[q->tqp_index + queue_num].queue_index = q->tqp_index;
3251 ring->io_base = q->io_base;
3254 hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
3258 ring->desc_cb = NULL;
3259 ring->dev = priv->dev;
3260 ring->desc_dma_addr = 0;
3261 ring->buf_size = q->buf_size;
3262 ring->desc_num = desc_num;
3263 ring->next_to_use = 0;
3264 ring->next_to_clean = 0;
3269 static int hns3_queue_to_ring(struct hnae3_queue *tqp,
3270 struct hns3_nic_priv *priv)
3274 ret = hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
3278 ret = hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
3280 devm_kfree(priv->dev, priv->ring_data[tqp->tqp_index].ring);
3287 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
3289 struct hnae3_handle *h = priv->ae_handle;
3290 struct pci_dev *pdev = h->pdev;
3293 priv->ring_data = devm_kzalloc(&pdev->dev,
3294 array3_size(h->kinfo.num_tqps,
3295 sizeof(*priv->ring_data),
3298 if (!priv->ring_data)
3301 for (i = 0; i < h->kinfo.num_tqps; i++) {
3302 ret = hns3_queue_to_ring(h->kinfo.tqp[i], priv);
3310 devm_kfree(priv->dev, priv->ring_data[i].ring);
3311 devm_kfree(priv->dev,
3312 priv->ring_data[i + h->kinfo.num_tqps].ring);
3315 devm_kfree(&pdev->dev, priv->ring_data);
3319 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
3321 struct hnae3_handle *h = priv->ae_handle;
3324 for (i = 0; i < h->kinfo.num_tqps; i++) {
3325 devm_kfree(priv->dev, priv->ring_data[i].ring);
3326 devm_kfree(priv->dev,
3327 priv->ring_data[i + h->kinfo.num_tqps].ring);
3329 devm_kfree(priv->dev, priv->ring_data);
3332 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
3336 if (ring->desc_num <= 0 || ring->buf_size <= 0)
3339 ring->desc_cb = kcalloc(ring->desc_num, sizeof(ring->desc_cb[0]),
3341 if (!ring->desc_cb) {
3346 ret = hns3_alloc_desc(ring);
3348 goto out_with_desc_cb;
3350 if (!HNAE3_IS_TX_RING(ring)) {
3351 ret = hns3_alloc_ring_buffers(ring);
3359 hns3_free_desc(ring);
3361 kfree(ring->desc_cb);
3362 ring->desc_cb = NULL;
3367 static void hns3_fini_ring(struct hns3_enet_ring *ring)
3369 hns3_free_desc(ring);
3370 kfree(ring->desc_cb);
3371 ring->desc_cb = NULL;
3372 ring->next_to_clean = 0;
3373 ring->next_to_use = 0;
3374 ring->pending_buf = 0;
3376 dev_kfree_skb_any(ring->skb);
3381 static int hns3_buf_size2type(u32 buf_size)
3387 bd_size_type = HNS3_BD_SIZE_512_TYPE;
3390 bd_size_type = HNS3_BD_SIZE_1024_TYPE;
3393 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
3396 bd_size_type = HNS3_BD_SIZE_4096_TYPE;
3399 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
3402 return bd_size_type;
3405 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
3407 dma_addr_t dma = ring->desc_dma_addr;
3408 struct hnae3_queue *q = ring->tqp;
3410 if (!HNAE3_IS_TX_RING(ring)) {
3411 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG,
3413 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
3414 (u32)((dma >> 31) >> 1));
3416 hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
3417 hns3_buf_size2type(ring->buf_size));
3418 hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
3419 ring->desc_num / 8 - 1);
3422 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
3424 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
3425 (u32)((dma >> 31) >> 1));
3427 hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
3428 ring->desc_num / 8 - 1);
3432 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
3434 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
3437 for (i = 0; i < HNAE3_MAX_TC; i++) {
3438 struct hnae3_tc_info *tc_info = &kinfo->tc_info[i];
3441 if (!tc_info->enable)
3444 for (j = 0; j < tc_info->tqp_count; j++) {
3445 struct hnae3_queue *q;
3447 q = priv->ring_data[tc_info->tqp_offset + j].ring->tqp;
3448 hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG,
3454 int hns3_init_all_ring(struct hns3_nic_priv *priv)
3456 struct hnae3_handle *h = priv->ae_handle;
3457 int ring_num = h->kinfo.num_tqps * 2;
3461 for (i = 0; i < ring_num; i++) {
3462 ret = hns3_alloc_ring_memory(priv->ring_data[i].ring);
3465 "Alloc ring memory fail! ret=%d\n", ret);
3466 goto out_when_alloc_ring_memory;
3469 u64_stats_init(&priv->ring_data[i].ring->syncp);
3474 out_when_alloc_ring_memory:
3475 for (j = i - 1; j >= 0; j--)
3476 hns3_fini_ring(priv->ring_data[j].ring);
3481 int hns3_uninit_all_ring(struct hns3_nic_priv *priv)
3483 struct hnae3_handle *h = priv->ae_handle;
3486 for (i = 0; i < h->kinfo.num_tqps; i++) {
3487 hns3_fini_ring(priv->ring_data[i].ring);
3488 hns3_fini_ring(priv->ring_data[i + h->kinfo.num_tqps].ring);
3493 /* Set mac addr if it is configured. or leave it to the AE driver */
3494 static int hns3_init_mac_addr(struct net_device *netdev, bool init)
3496 struct hns3_nic_priv *priv = netdev_priv(netdev);
3497 struct hnae3_handle *h = priv->ae_handle;
3498 u8 mac_addr_temp[ETH_ALEN];
3501 if (h->ae_algo->ops->get_mac_addr && init) {
3502 h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
3503 ether_addr_copy(netdev->dev_addr, mac_addr_temp);
3506 /* Check if the MAC address is valid, if not get a random one */
3507 if (!is_valid_ether_addr(netdev->dev_addr)) {
3508 eth_hw_addr_random(netdev);
3509 dev_warn(priv->dev, "using random MAC address %pM\n",
3513 if (h->ae_algo->ops->set_mac_addr)
3514 ret = h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
3519 static int hns3_init_phy(struct net_device *netdev)
3521 struct hnae3_handle *h = hns3_get_handle(netdev);
3524 if (h->ae_algo->ops->mac_connect_phy)
3525 ret = h->ae_algo->ops->mac_connect_phy(h);
3530 static void hns3_uninit_phy(struct net_device *netdev)
3532 struct hnae3_handle *h = hns3_get_handle(netdev);
3534 if (h->ae_algo->ops->mac_disconnect_phy)
3535 h->ae_algo->ops->mac_disconnect_phy(h);
3538 static int hns3_restore_fd_rules(struct net_device *netdev)
3540 struct hnae3_handle *h = hns3_get_handle(netdev);
3543 if (h->ae_algo->ops->restore_fd_rules)
3544 ret = h->ae_algo->ops->restore_fd_rules(h);
3549 static void hns3_del_all_fd_rules(struct net_device *netdev, bool clear_list)
3551 struct hnae3_handle *h = hns3_get_handle(netdev);
3553 if (h->ae_algo->ops->del_all_fd_entries)
3554 h->ae_algo->ops->del_all_fd_entries(h, clear_list);
3557 static void hns3_nic_set_priv_ops(struct net_device *netdev)
3559 struct hns3_nic_priv *priv = netdev_priv(netdev);
3561 if ((netdev->features & NETIF_F_TSO) ||
3562 (netdev->features & NETIF_F_TSO6))
3563 priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tso;
3565 priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tx;
3568 static int hns3_client_start(struct hnae3_handle *handle)
3570 if (!handle->ae_algo->ops->client_start)
3573 return handle->ae_algo->ops->client_start(handle);
3576 static void hns3_client_stop(struct hnae3_handle *handle)
3578 if (!handle->ae_algo->ops->client_stop)
3581 handle->ae_algo->ops->client_stop(handle);
3584 static int hns3_client_init(struct hnae3_handle *handle)
3586 struct pci_dev *pdev = handle->pdev;
3587 u16 alloc_tqps, max_rss_size;
3588 struct hns3_nic_priv *priv;
3589 struct net_device *netdev;
3592 handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
3594 netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
3598 priv = netdev_priv(netdev);
3599 priv->dev = &pdev->dev;
3600 priv->netdev = netdev;
3601 priv->ae_handle = handle;
3602 priv->tx_timeout_count = 0;
3603 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
3605 handle->kinfo.netdev = netdev;
3606 handle->priv = (void *)priv;
3608 hns3_init_mac_addr(netdev, true);
3610 hns3_set_default_feature(netdev);
3612 netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
3613 netdev->priv_flags |= IFF_UNICAST_FLT;
3614 netdev->netdev_ops = &hns3_nic_netdev_ops;
3615 SET_NETDEV_DEV(netdev, &pdev->dev);
3616 hns3_ethtool_set_ops(netdev);
3617 hns3_nic_set_priv_ops(netdev);
3619 /* Carrier off reporting is important to ethtool even BEFORE open */
3620 netif_carrier_off(netdev);
3622 ret = hns3_get_ring_config(priv);
3625 goto out_get_ring_cfg;
3628 ret = hns3_nic_alloc_vector_data(priv);
3631 goto out_alloc_vector_data;
3634 ret = hns3_nic_init_vector_data(priv);
3637 goto out_init_vector_data;
3640 ret = hns3_init_all_ring(priv);
3643 goto out_init_ring_data;
3646 ret = hns3_init_phy(netdev);
3650 ret = register_netdev(netdev);
3652 dev_err(priv->dev, "probe register netdev fail!\n");
3653 goto out_reg_netdev_fail;
3656 ret = hns3_client_start(handle);
3658 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
3659 goto out_client_start;
3662 hns3_dcbnl_setup(handle);
3664 hns3_dbg_init(handle);
3666 /* MTU range: (ETH_MIN_MTU(kernel default) - 9702) */
3667 netdev->max_mtu = HNS3_MAX_MTU;
3669 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
3674 unregister_netdev(netdev);
3675 out_reg_netdev_fail:
3676 hns3_uninit_phy(netdev);
3678 hns3_uninit_all_ring(priv);
3680 hns3_nic_uninit_vector_data(priv);
3681 out_init_vector_data:
3682 hns3_nic_dealloc_vector_data(priv);
3683 out_alloc_vector_data:
3684 priv->ring_data = NULL;
3686 priv->ae_handle = NULL;
3687 free_netdev(netdev);
3691 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
3693 struct net_device *netdev = handle->kinfo.netdev;
3694 struct hns3_nic_priv *priv = netdev_priv(netdev);
3697 hns3_client_stop(handle);
3699 hns3_remove_hw_addr(netdev);
3701 if (netdev->reg_state != NETREG_UNINITIALIZED)
3702 unregister_netdev(netdev);
3704 if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
3705 netdev_warn(netdev, "already uninitialized\n");
3706 goto out_netdev_free;
3709 hns3_del_all_fd_rules(netdev, true);
3711 hns3_force_clear_all_rx_ring(handle);
3713 hns3_uninit_phy(netdev);
3715 hns3_nic_uninit_vector_data(priv);
3717 ret = hns3_nic_dealloc_vector_data(priv);
3719 netdev_err(netdev, "dealloc vector error\n");
3721 ret = hns3_uninit_all_ring(priv);
3723 netdev_err(netdev, "uninit ring error\n");
3725 hns3_put_ring_config(priv);
3727 hns3_dbg_uninit(handle);
3729 priv->ring_data = NULL;
3732 free_netdev(netdev);
3735 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
3737 struct net_device *netdev = handle->kinfo.netdev;
3743 netif_carrier_on(netdev);
3744 netif_tx_wake_all_queues(netdev);
3745 netdev_info(netdev, "link up\n");
3747 netif_carrier_off(netdev);
3748 netif_tx_stop_all_queues(netdev);
3749 netdev_info(netdev, "link down\n");
3753 static int hns3_client_setup_tc(struct hnae3_handle *handle, u8 tc)
3755 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
3756 struct net_device *ndev = kinfo->netdev;
3758 if (tc > HNAE3_MAX_TC)
3764 return hns3_nic_set_real_num_queue(ndev);
3767 static int hns3_recover_hw_addr(struct net_device *ndev)
3769 struct netdev_hw_addr_list *list;
3770 struct netdev_hw_addr *ha, *tmp;
3773 /* go through and sync uc_addr entries to the device */
3775 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3776 ret = hns3_nic_uc_sync(ndev, ha->addr);
3781 /* go through and sync mc_addr entries to the device */
3783 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3784 ret = hns3_nic_mc_sync(ndev, ha->addr);
3792 static void hns3_remove_hw_addr(struct net_device *netdev)
3794 struct netdev_hw_addr_list *list;
3795 struct netdev_hw_addr *ha, *tmp;
3797 hns3_nic_uc_unsync(netdev, netdev->dev_addr);
3799 /* go through and unsync uc_addr entries to the device */
3801 list_for_each_entry_safe(ha, tmp, &list->list, list)
3802 hns3_nic_uc_unsync(netdev, ha->addr);
3804 /* go through and unsync mc_addr entries to the device */
3806 list_for_each_entry_safe(ha, tmp, &list->list, list)
3807 if (ha->refcount > 1)
3808 hns3_nic_mc_unsync(netdev, ha->addr);
3811 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
3813 while (ring->next_to_clean != ring->next_to_use) {
3814 ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
3815 hns3_free_buffer_detach(ring, ring->next_to_clean);
3816 ring_ptr_move_fw(ring, next_to_clean);
3820 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
3822 struct hns3_desc_cb res_cbs;
3825 while (ring->next_to_use != ring->next_to_clean) {
3826 /* When a buffer is not reused, it's memory has been
3827 * freed in hns3_handle_rx_bd or will be freed by
3828 * stack, so we need to replace the buffer here.
3830 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
3831 ret = hns3_reserve_buffer_map(ring, &res_cbs);
3833 u64_stats_update_begin(&ring->syncp);
3834 ring->stats.sw_err_cnt++;
3835 u64_stats_update_end(&ring->syncp);
3836 /* if alloc new buffer fail, exit directly
3837 * and reclear in up flow.
3839 netdev_warn(ring->tqp->handle->kinfo.netdev,
3840 "reserve buffer map failed, ret = %d\n",
3844 hns3_replace_buffer(ring, ring->next_to_use,
3847 ring_ptr_move_fw(ring, next_to_use);
3853 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
3855 while (ring->next_to_use != ring->next_to_clean) {
3856 /* When a buffer is not reused, it's memory has been
3857 * freed in hns3_handle_rx_bd or will be freed by
3858 * stack, so only need to unmap the buffer here.
3860 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
3861 hns3_unmap_buffer(ring,
3862 &ring->desc_cb[ring->next_to_use]);
3863 ring->desc_cb[ring->next_to_use].dma = 0;
3866 ring_ptr_move_fw(ring, next_to_use);
3870 static void hns3_force_clear_all_rx_ring(struct hnae3_handle *h)
3872 struct net_device *ndev = h->kinfo.netdev;
3873 struct hns3_nic_priv *priv = netdev_priv(ndev);
3874 struct hns3_enet_ring *ring;
3877 for (i = 0; i < h->kinfo.num_tqps; i++) {
3878 ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
3879 hns3_force_clear_rx_ring(ring);
3883 static void hns3_clear_all_ring(struct hnae3_handle *h)
3885 struct net_device *ndev = h->kinfo.netdev;
3886 struct hns3_nic_priv *priv = netdev_priv(ndev);
3889 for (i = 0; i < h->kinfo.num_tqps; i++) {
3890 struct netdev_queue *dev_queue;
3891 struct hns3_enet_ring *ring;
3893 ring = priv->ring_data[i].ring;
3894 hns3_clear_tx_ring(ring);
3895 dev_queue = netdev_get_tx_queue(ndev,
3896 priv->ring_data[i].queue_index);
3897 netdev_tx_reset_queue(dev_queue);
3899 ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
3900 /* Continue to clear other rings even if clearing some
3903 hns3_clear_rx_ring(ring);
3907 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
3909 struct net_device *ndev = h->kinfo.netdev;
3910 struct hns3_nic_priv *priv = netdev_priv(ndev);
3911 struct hns3_enet_ring *rx_ring;
3915 for (i = 0; i < h->kinfo.num_tqps; i++) {
3916 ret = h->ae_algo->ops->reset_queue(h, i);
3920 hns3_init_ring_hw(priv->ring_data[i].ring);
3922 /* We need to clear tx ring here because self test will
3923 * use the ring and will not run down before up
3925 hns3_clear_tx_ring(priv->ring_data[i].ring);
3926 priv->ring_data[i].ring->next_to_clean = 0;
3927 priv->ring_data[i].ring->next_to_use = 0;
3929 rx_ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
3930 hns3_init_ring_hw(rx_ring);
3931 ret = hns3_clear_rx_ring(rx_ring);
3935 /* We can not know the hardware head and tail when this
3936 * function is called in reset flow, so we reuse all desc.
3938 for (j = 0; j < rx_ring->desc_num; j++)
3939 hns3_reuse_buffer(rx_ring, j);
3941 rx_ring->next_to_clean = 0;
3942 rx_ring->next_to_use = 0;
3945 hns3_init_tx_ring_tc(priv);
3950 static void hns3_store_coal(struct hns3_nic_priv *priv)
3952 /* ethtool only support setting and querying one coal
3953 * configuation for now, so save the vector 0' coal
3954 * configuation here in order to restore it.
3956 memcpy(&priv->tx_coal, &priv->tqp_vector[0].tx_group.coal,
3957 sizeof(struct hns3_enet_coalesce));
3958 memcpy(&priv->rx_coal, &priv->tqp_vector[0].rx_group.coal,
3959 sizeof(struct hns3_enet_coalesce));
3962 static void hns3_restore_coal(struct hns3_nic_priv *priv)
3964 u16 vector_num = priv->vector_num;
3967 for (i = 0; i < vector_num; i++) {
3968 memcpy(&priv->tqp_vector[i].tx_group.coal, &priv->tx_coal,
3969 sizeof(struct hns3_enet_coalesce));
3970 memcpy(&priv->tqp_vector[i].rx_group.coal, &priv->rx_coal,
3971 sizeof(struct hns3_enet_coalesce));
3975 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
3977 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(handle->pdev);
3978 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
3979 struct net_device *ndev = kinfo->netdev;
3980 struct hns3_nic_priv *priv = netdev_priv(ndev);
3982 if (test_and_set_bit(HNS3_NIC_STATE_RESETTING, &priv->state))
3985 /* it is cumbersome for hardware to pick-and-choose entries for deletion
3986 * from table space. Hence, for function reset software intervention is
3987 * required to delete the entries
3989 if (hns3_dev_ongoing_func_reset(ae_dev)) {
3990 hns3_remove_hw_addr(ndev);
3991 hns3_del_all_fd_rules(ndev, false);
3994 if (!netif_running(ndev))
3997 return hns3_nic_net_stop(ndev);
4000 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
4002 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4003 struct hns3_nic_priv *priv = netdev_priv(kinfo->netdev);
4006 clear_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4008 if (netif_running(kinfo->netdev)) {
4009 ret = hns3_nic_net_open(kinfo->netdev);
4011 set_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4012 netdev_err(kinfo->netdev,
4013 "hns net up fail, ret=%d!\n", ret);
4021 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
4023 struct net_device *netdev = handle->kinfo.netdev;
4024 struct hns3_nic_priv *priv = netdev_priv(netdev);
4027 /* Carrier off reporting is important to ethtool even BEFORE open */
4028 netif_carrier_off(netdev);
4030 ret = hns3_get_ring_config(priv);
4034 ret = hns3_nic_alloc_vector_data(priv);
4038 hns3_restore_coal(priv);
4040 ret = hns3_nic_init_vector_data(priv);
4042 goto err_dealloc_vector;
4044 ret = hns3_init_all_ring(priv);
4046 goto err_uninit_vector;
4048 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
4053 hns3_nic_uninit_vector_data(priv);
4054 priv->ring_data = NULL;
4056 hns3_nic_dealloc_vector_data(priv);
4058 hns3_put_ring_config(priv);
4059 priv->ring_data = NULL;
4064 static int hns3_reset_notify_restore_enet(struct hnae3_handle *handle)
4066 struct net_device *netdev = handle->kinfo.netdev;
4067 bool vlan_filter_enable;
4070 ret = hns3_init_mac_addr(netdev, false);
4074 ret = hns3_recover_hw_addr(netdev);
4078 ret = hns3_update_promisc_mode(netdev, handle->netdev_flags);
4082 vlan_filter_enable = netdev->flags & IFF_PROMISC ? false : true;
4083 hns3_enable_vlan_filter(netdev, vlan_filter_enable);
4085 /* Hardware table is only clear when pf resets */
4086 if (!(handle->flags & HNAE3_SUPPORT_VF)) {
4087 ret = hns3_restore_vlan(netdev);
4092 return hns3_restore_fd_rules(netdev);
4095 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
4097 struct net_device *netdev = handle->kinfo.netdev;
4098 struct hns3_nic_priv *priv = netdev_priv(netdev);
4101 if (!test_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
4102 netdev_warn(netdev, "already uninitialized\n");
4106 hns3_force_clear_all_rx_ring(handle);
4108 hns3_nic_uninit_vector_data(priv);
4110 hns3_store_coal(priv);
4112 ret = hns3_nic_dealloc_vector_data(priv);
4114 netdev_err(netdev, "dealloc vector error\n");
4116 ret = hns3_uninit_all_ring(priv);
4118 netdev_err(netdev, "uninit ring error\n");
4120 hns3_put_ring_config(priv);
4121 priv->ring_data = NULL;
4123 clear_bit(HNS3_NIC_STATE_INITED, &priv->state);
4128 static int hns3_reset_notify(struct hnae3_handle *handle,
4129 enum hnae3_reset_notify_type type)
4134 case HNAE3_UP_CLIENT:
4135 ret = hns3_reset_notify_up_enet(handle);
4137 case HNAE3_DOWN_CLIENT:
4138 ret = hns3_reset_notify_down_enet(handle);
4140 case HNAE3_INIT_CLIENT:
4141 ret = hns3_reset_notify_init_enet(handle);
4143 case HNAE3_UNINIT_CLIENT:
4144 ret = hns3_reset_notify_uninit_enet(handle);
4146 case HNAE3_RESTORE_CLIENT:
4147 ret = hns3_reset_notify_restore_enet(handle);
4156 int hns3_set_channels(struct net_device *netdev,
4157 struct ethtool_channels *ch)
4159 struct hnae3_handle *h = hns3_get_handle(netdev);
4160 struct hnae3_knic_private_info *kinfo = &h->kinfo;
4161 bool rxfh_configured = netif_is_rxfh_configured(netdev);
4162 u32 new_tqp_num = ch->combined_count;
4166 if (ch->rx_count || ch->tx_count)
4169 if (new_tqp_num > hns3_get_max_available_channels(h) ||
4171 dev_err(&netdev->dev,
4172 "Change tqps fail, the tqp range is from 1 to %d",
4173 hns3_get_max_available_channels(h));
4177 if (kinfo->rss_size == new_tqp_num)
4180 ret = hns3_reset_notify(h, HNAE3_DOWN_CLIENT);
4184 ret = hns3_reset_notify(h, HNAE3_UNINIT_CLIENT);
4188 org_tqp_num = h->kinfo.num_tqps;
4189 ret = h->ae_algo->ops->set_channels(h, new_tqp_num, rxfh_configured);
4191 ret = h->ae_algo->ops->set_channels(h, org_tqp_num,
4194 /* If revert to old tqp failed, fatal error occurred */
4195 dev_err(&netdev->dev,
4196 "Revert to old tqp num fail, ret=%d", ret);
4199 dev_info(&netdev->dev,
4200 "Change tqp num fail, Revert to old tqp num");
4202 ret = hns3_reset_notify(h, HNAE3_INIT_CLIENT);
4206 return hns3_reset_notify(h, HNAE3_UP_CLIENT);
4209 static const struct hnae3_client_ops client_ops = {
4210 .init_instance = hns3_client_init,
4211 .uninit_instance = hns3_client_uninit,
4212 .link_status_change = hns3_link_status_change,
4213 .setup_tc = hns3_client_setup_tc,
4214 .reset_notify = hns3_reset_notify,
4217 /* hns3_init_module - Driver registration routine
4218 * hns3_init_module is the first routine called when the driver is
4219 * loaded. All it does is register with the PCI subsystem.
4221 static int __init hns3_init_module(void)
4225 pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
4226 pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
4228 client.type = HNAE3_CLIENT_KNIC;
4229 snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH - 1, "%s",
4232 client.ops = &client_ops;
4234 INIT_LIST_HEAD(&client.node);
4236 hns3_dbg_register_debugfs(hns3_driver_name);
4238 ret = hnae3_register_client(&client);
4240 goto err_reg_client;
4242 ret = pci_register_driver(&hns3_driver);
4244 goto err_reg_driver;
4249 hnae3_unregister_client(&client);
4251 hns3_dbg_unregister_debugfs();
4254 module_init(hns3_init_module);
4256 /* hns3_exit_module - Driver exit cleanup routine
4257 * hns3_exit_module is called just before the driver is removed
4260 static void __exit hns3_exit_module(void)
4262 pci_unregister_driver(&hns3_driver);
4263 hnae3_unregister_client(&client);
4264 hns3_dbg_unregister_debugfs();
4266 module_exit(hns3_exit_module);
4268 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
4269 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
4270 MODULE_LICENSE("GPL");
4271 MODULE_ALIAS("pci:hns-nic");
4272 MODULE_VERSION(HNS3_MOD_VERSION);