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)))
25 #define hns3_tx_bd_count(S) DIV_ROUND_UP(S, HNS3_MAX_BD_SIZE)
27 static void hns3_clear_all_ring(struct hnae3_handle *h);
28 static void hns3_force_clear_all_rx_ring(struct hnae3_handle *h);
29 static void hns3_remove_hw_addr(struct net_device *netdev);
31 static const char hns3_driver_name[] = "hns3";
32 const char hns3_driver_version[] = VERMAGIC_STRING;
33 static const char hns3_driver_string[] =
34 "Hisilicon Ethernet Network Driver for Hip08 Family";
35 static const char hns3_copyright[] = "Copyright (c) 2017 Huawei Corporation.";
36 static struct hnae3_client client;
38 /* hns3_pci_tbl - PCI Device ID Table
40 * Last entry must be all 0s
42 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
43 * Class, Class Mask, private data (not used) }
45 static const struct pci_device_id hns3_pci_tbl[] = {
46 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
47 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
48 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA),
49 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
50 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC),
51 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
52 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA),
53 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
54 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC),
55 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
56 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC),
57 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
58 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_VF), 0},
59 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF),
60 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
61 /* required last entry */
64 MODULE_DEVICE_TABLE(pci, hns3_pci_tbl);
66 static irqreturn_t hns3_irq_handle(int irq, void *vector)
68 struct hns3_enet_tqp_vector *tqp_vector = vector;
70 napi_schedule(&tqp_vector->napi);
75 /* This callback function is used to set affinity changes to the irq affinity
76 * masks when the irq_set_affinity_notifier function is used.
78 static void hns3_nic_irq_affinity_notify(struct irq_affinity_notify *notify,
79 const cpumask_t *mask)
81 struct hns3_enet_tqp_vector *tqp_vectors =
82 container_of(notify, struct hns3_enet_tqp_vector,
85 tqp_vectors->affinity_mask = *mask;
88 static void hns3_nic_irq_affinity_release(struct kref *ref)
92 static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv)
94 struct hns3_enet_tqp_vector *tqp_vectors;
97 for (i = 0; i < priv->vector_num; i++) {
98 tqp_vectors = &priv->tqp_vector[i];
100 if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED)
103 /* clear the affinity notifier and affinity mask */
104 irq_set_affinity_notifier(tqp_vectors->vector_irq, NULL);
105 irq_set_affinity_hint(tqp_vectors->vector_irq, NULL);
107 /* release the irq resource */
108 free_irq(tqp_vectors->vector_irq, tqp_vectors);
109 tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED;
113 static int hns3_nic_init_irq(struct hns3_nic_priv *priv)
115 struct hns3_enet_tqp_vector *tqp_vectors;
116 int txrx_int_idx = 0;
122 for (i = 0; i < priv->vector_num; i++) {
123 tqp_vectors = &priv->tqp_vector[i];
125 if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED)
128 if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) {
129 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
130 "%s-%s-%d", priv->netdev->name, "TxRx",
133 } else if (tqp_vectors->rx_group.ring) {
134 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
135 "%s-%s-%d", priv->netdev->name, "Rx",
137 } else if (tqp_vectors->tx_group.ring) {
138 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
139 "%s-%s-%d", priv->netdev->name, "Tx",
142 /* Skip this unused q_vector */
146 tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0';
148 ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0,
152 netdev_err(priv->netdev, "request irq(%d) fail\n",
153 tqp_vectors->vector_irq);
157 tqp_vectors->affinity_notify.notify =
158 hns3_nic_irq_affinity_notify;
159 tqp_vectors->affinity_notify.release =
160 hns3_nic_irq_affinity_release;
161 irq_set_affinity_notifier(tqp_vectors->vector_irq,
162 &tqp_vectors->affinity_notify);
163 irq_set_affinity_hint(tqp_vectors->vector_irq,
164 &tqp_vectors->affinity_mask);
166 tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED;
172 static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector,
175 writel(mask_en, tqp_vector->mask_addr);
178 static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector)
180 napi_enable(&tqp_vector->napi);
183 hns3_mask_vector_irq(tqp_vector, 1);
186 static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector)
189 hns3_mask_vector_irq(tqp_vector, 0);
191 disable_irq(tqp_vector->vector_irq);
192 napi_disable(&tqp_vector->napi);
195 void hns3_set_vector_coalesce_rl(struct hns3_enet_tqp_vector *tqp_vector,
198 u32 rl_reg = hns3_rl_usec_to_reg(rl_value);
200 /* this defines the configuration for RL (Interrupt Rate Limiter).
201 * Rl defines rate of interrupts i.e. number of interrupts-per-second
202 * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing
205 if (rl_reg > 0 && !tqp_vector->tx_group.coal.gl_adapt_enable &&
206 !tqp_vector->rx_group.coal.gl_adapt_enable)
207 /* According to the hardware, the range of rl_reg is
208 * 0-59 and the unit is 4.
210 rl_reg |= HNS3_INT_RL_ENABLE_MASK;
212 writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
215 void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
218 u32 rx_gl_reg = hns3_gl_usec_to_reg(gl_value);
220 writel(rx_gl_reg, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
223 void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
226 u32 tx_gl_reg = hns3_gl_usec_to_reg(gl_value);
228 writel(tx_gl_reg, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
231 static void hns3_vector_gl_rl_init(struct hns3_enet_tqp_vector *tqp_vector,
232 struct hns3_nic_priv *priv)
234 /* initialize the configuration for interrupt coalescing.
235 * 1. GL (Interrupt Gap Limiter)
236 * 2. RL (Interrupt Rate Limiter)
239 /* Default: enable interrupt coalescing self-adaptive and GL */
240 tqp_vector->tx_group.coal.gl_adapt_enable = 1;
241 tqp_vector->rx_group.coal.gl_adapt_enable = 1;
243 tqp_vector->tx_group.coal.int_gl = HNS3_INT_GL_50K;
244 tqp_vector->rx_group.coal.int_gl = HNS3_INT_GL_50K;
246 tqp_vector->rx_group.coal.flow_level = HNS3_FLOW_LOW;
247 tqp_vector->tx_group.coal.flow_level = HNS3_FLOW_LOW;
250 static void hns3_vector_gl_rl_init_hw(struct hns3_enet_tqp_vector *tqp_vector,
251 struct hns3_nic_priv *priv)
253 struct hnae3_handle *h = priv->ae_handle;
255 hns3_set_vector_coalesce_tx_gl(tqp_vector,
256 tqp_vector->tx_group.coal.int_gl);
257 hns3_set_vector_coalesce_rx_gl(tqp_vector,
258 tqp_vector->rx_group.coal.int_gl);
259 hns3_set_vector_coalesce_rl(tqp_vector, h->kinfo.int_rl_setting);
262 static int hns3_nic_set_real_num_queue(struct net_device *netdev)
264 struct hnae3_handle *h = hns3_get_handle(netdev);
265 struct hnae3_knic_private_info *kinfo = &h->kinfo;
266 unsigned int queue_size = kinfo->rss_size * kinfo->num_tc;
269 if (kinfo->num_tc <= 1) {
270 netdev_reset_tc(netdev);
272 ret = netdev_set_num_tc(netdev, kinfo->num_tc);
275 "netdev_set_num_tc fail, ret=%d!\n", ret);
279 for (i = 0; i < HNAE3_MAX_TC; i++) {
280 if (!kinfo->tc_info[i].enable)
283 netdev_set_tc_queue(netdev,
284 kinfo->tc_info[i].tc,
285 kinfo->tc_info[i].tqp_count,
286 kinfo->tc_info[i].tqp_offset);
290 ret = netif_set_real_num_tx_queues(netdev, queue_size);
293 "netif_set_real_num_tx_queues fail, ret=%d!\n",
298 ret = netif_set_real_num_rx_queues(netdev, queue_size);
301 "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
308 static u16 hns3_get_max_available_channels(struct hnae3_handle *h)
310 u16 alloc_tqps, max_rss_size, rss_size;
312 h->ae_algo->ops->get_tqps_and_rss_info(h, &alloc_tqps, &max_rss_size);
313 rss_size = alloc_tqps / h->kinfo.num_tc;
315 return min_t(u16, rss_size, max_rss_size);
318 static void hns3_tqp_enable(struct hnae3_queue *tqp)
322 rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
323 rcb_reg |= BIT(HNS3_RING_EN_B);
324 hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
327 static void hns3_tqp_disable(struct hnae3_queue *tqp)
331 rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
332 rcb_reg &= ~BIT(HNS3_RING_EN_B);
333 hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
336 static int hns3_nic_net_up(struct net_device *netdev)
338 struct hns3_nic_priv *priv = netdev_priv(netdev);
339 struct hnae3_handle *h = priv->ae_handle;
343 ret = hns3_nic_reset_all_ring(h);
347 /* get irq resource for all vectors */
348 ret = hns3_nic_init_irq(priv);
350 netdev_err(netdev, "hns init irq failed! ret=%d\n", ret);
354 clear_bit(HNS3_NIC_STATE_DOWN, &priv->state);
356 /* enable the vectors */
357 for (i = 0; i < priv->vector_num; i++)
358 hns3_vector_enable(&priv->tqp_vector[i]);
361 for (j = 0; j < h->kinfo.num_tqps; j++)
362 hns3_tqp_enable(h->kinfo.tqp[j]);
364 /* start the ae_dev */
365 ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0;
372 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
374 hns3_tqp_disable(h->kinfo.tqp[j]);
376 for (j = i - 1; j >= 0; j--)
377 hns3_vector_disable(&priv->tqp_vector[j]);
379 hns3_nic_uninit_irq(priv);
384 static void hns3_config_xps(struct hns3_nic_priv *priv)
388 for (i = 0; i < priv->vector_num; i++) {
389 struct hns3_enet_tqp_vector *tqp_vector = &priv->tqp_vector[i];
390 struct hns3_enet_ring *ring = tqp_vector->tx_group.ring;
395 ret = netif_set_xps_queue(priv->netdev,
396 &tqp_vector->affinity_mask,
397 ring->tqp->tqp_index);
399 netdev_warn(priv->netdev,
400 "set xps queue failed: %d", ret);
407 static int hns3_nic_net_open(struct net_device *netdev)
409 struct hns3_nic_priv *priv = netdev_priv(netdev);
410 struct hnae3_handle *h = hns3_get_handle(netdev);
411 struct hnae3_knic_private_info *kinfo;
414 if (hns3_nic_resetting(netdev))
417 netif_carrier_off(netdev);
419 ret = hns3_nic_set_real_num_queue(netdev);
423 ret = hns3_nic_net_up(netdev);
426 "hns net up fail, ret=%d!\n", ret);
431 for (i = 0; i < HNAE3_MAX_USER_PRIO; i++) {
432 netdev_set_prio_tc_map(netdev, i,
436 if (h->ae_algo->ops->set_timer_task)
437 h->ae_algo->ops->set_timer_task(priv->ae_handle, true);
439 hns3_config_xps(priv);
443 static void hns3_nic_net_down(struct net_device *netdev)
445 struct hns3_nic_priv *priv = netdev_priv(netdev);
446 struct hnae3_handle *h = hns3_get_handle(netdev);
447 const struct hnae3_ae_ops *ops;
450 /* disable vectors */
451 for (i = 0; i < priv->vector_num; i++)
452 hns3_vector_disable(&priv->tqp_vector[i]);
455 for (i = 0; i < h->kinfo.num_tqps; i++)
456 hns3_tqp_disable(h->kinfo.tqp[i]);
459 ops = priv->ae_handle->ae_algo->ops;
461 ops->stop(priv->ae_handle);
463 /* free irq resources */
464 hns3_nic_uninit_irq(priv);
466 hns3_clear_all_ring(priv->ae_handle);
469 static int hns3_nic_net_stop(struct net_device *netdev)
471 struct hns3_nic_priv *priv = netdev_priv(netdev);
472 struct hnae3_handle *h = hns3_get_handle(netdev);
474 if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
477 if (h->ae_algo->ops->set_timer_task)
478 h->ae_algo->ops->set_timer_task(priv->ae_handle, false);
480 netif_tx_stop_all_queues(netdev);
481 netif_carrier_off(netdev);
483 hns3_nic_net_down(netdev);
488 static int hns3_nic_uc_sync(struct net_device *netdev,
489 const unsigned char *addr)
491 struct hnae3_handle *h = hns3_get_handle(netdev);
493 if (h->ae_algo->ops->add_uc_addr)
494 return h->ae_algo->ops->add_uc_addr(h, addr);
499 static int hns3_nic_uc_unsync(struct net_device *netdev,
500 const unsigned char *addr)
502 struct hnae3_handle *h = hns3_get_handle(netdev);
504 if (h->ae_algo->ops->rm_uc_addr)
505 return h->ae_algo->ops->rm_uc_addr(h, addr);
510 static int hns3_nic_mc_sync(struct net_device *netdev,
511 const unsigned char *addr)
513 struct hnae3_handle *h = hns3_get_handle(netdev);
515 if (h->ae_algo->ops->add_mc_addr)
516 return h->ae_algo->ops->add_mc_addr(h, addr);
521 static int hns3_nic_mc_unsync(struct net_device *netdev,
522 const unsigned char *addr)
524 struct hnae3_handle *h = hns3_get_handle(netdev);
526 if (h->ae_algo->ops->rm_mc_addr)
527 return h->ae_algo->ops->rm_mc_addr(h, addr);
532 static u8 hns3_get_netdev_flags(struct net_device *netdev)
536 if (netdev->flags & IFF_PROMISC) {
537 flags = HNAE3_USER_UPE | HNAE3_USER_MPE | HNAE3_BPE;
539 flags |= HNAE3_VLAN_FLTR;
540 if (netdev->flags & IFF_ALLMULTI)
541 flags |= HNAE3_USER_MPE;
547 static void hns3_nic_set_rx_mode(struct net_device *netdev)
549 struct hnae3_handle *h = hns3_get_handle(netdev);
553 new_flags = hns3_get_netdev_flags(netdev);
555 ret = __dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync);
557 netdev_err(netdev, "sync uc address fail\n");
559 new_flags |= HNAE3_OVERFLOW_UPE;
562 if (netdev->flags & IFF_MULTICAST) {
563 ret = __dev_mc_sync(netdev, hns3_nic_mc_sync,
566 netdev_err(netdev, "sync mc address fail\n");
568 new_flags |= HNAE3_OVERFLOW_MPE;
572 /* User mode Promisc mode enable and vlan filtering is disabled to
573 * let all packets in. MAC-VLAN Table overflow Promisc enabled and
574 * vlan fitering is enabled
576 hns3_enable_vlan_filter(netdev, new_flags & HNAE3_VLAN_FLTR);
577 h->netdev_flags = new_flags;
578 hns3_update_promisc_mode(netdev, new_flags);
581 int hns3_update_promisc_mode(struct net_device *netdev, u8 promisc_flags)
583 struct hns3_nic_priv *priv = netdev_priv(netdev);
584 struct hnae3_handle *h = priv->ae_handle;
586 if (h->ae_algo->ops->set_promisc_mode) {
587 return h->ae_algo->ops->set_promisc_mode(h,
588 promisc_flags & HNAE3_UPE,
589 promisc_flags & HNAE3_MPE);
595 void hns3_enable_vlan_filter(struct net_device *netdev, bool enable)
597 struct hns3_nic_priv *priv = netdev_priv(netdev);
598 struct hnae3_handle *h = priv->ae_handle;
601 if (h->pdev->revision >= 0x21 && h->ae_algo->ops->enable_vlan_filter) {
602 last_state = h->netdev_flags & HNAE3_VLAN_FLTR ? true : false;
603 if (enable != last_state) {
606 enable ? "enable" : "disable");
607 h->ae_algo->ops->enable_vlan_filter(h, enable);
612 static int hns3_set_tso(struct sk_buff *skb, u32 *paylen,
613 u16 *mss, u32 *type_cs_vlan_tso)
615 u32 l4_offset, hdr_len;
616 union l3_hdr_info l3;
617 union l4_hdr_info l4;
621 if (!skb_is_gso(skb))
624 ret = skb_cow_head(skb, 0);
628 l3.hdr = skb_network_header(skb);
629 l4.hdr = skb_transport_header(skb);
631 /* Software should clear the IPv4's checksum field when tso is
634 if (l3.v4->version == 4)
638 if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
641 SKB_GSO_UDP_TUNNEL_CSUM)) {
642 if ((!(skb_shinfo(skb)->gso_type &
644 (skb_shinfo(skb)->gso_type &
645 SKB_GSO_UDP_TUNNEL_CSUM)) {
646 /* Software should clear the udp's checksum
647 * field when tso is needed.
651 /* reset l3&l4 pointers from outer to inner headers */
652 l3.hdr = skb_inner_network_header(skb);
653 l4.hdr = skb_inner_transport_header(skb);
655 /* Software should clear the IPv4's checksum field when
658 if (l3.v4->version == 4)
662 /* normal or tunnel packet*/
663 l4_offset = l4.hdr - skb->data;
664 hdr_len = (l4.tcp->doff << 2) + l4_offset;
666 /* remove payload length from inner pseudo checksum when tso*/
667 l4_paylen = skb->len - l4_offset;
668 csum_replace_by_diff(&l4.tcp->check,
669 (__force __wsum)htonl(l4_paylen));
671 /* find the txbd field values */
672 *paylen = skb->len - hdr_len;
673 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_TSO_B, 1);
675 /* get MSS for TSO */
676 *mss = skb_shinfo(skb)->gso_size;
681 static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
684 union l3_hdr_info l3;
685 unsigned char *l4_hdr;
686 unsigned char *exthdr;
690 /* find outer header point */
691 l3.hdr = skb_network_header(skb);
692 l4_hdr = skb_transport_header(skb);
694 if (skb->protocol == htons(ETH_P_IPV6)) {
695 exthdr = l3.hdr + sizeof(*l3.v6);
696 l4_proto_tmp = l3.v6->nexthdr;
697 if (l4_hdr != exthdr)
698 ipv6_skip_exthdr(skb, exthdr - skb->data,
699 &l4_proto_tmp, &frag_off);
700 } else if (skb->protocol == htons(ETH_P_IP)) {
701 l4_proto_tmp = l3.v4->protocol;
706 *ol4_proto = l4_proto_tmp;
709 if (!skb->encapsulation) {
714 /* find inner header point */
715 l3.hdr = skb_inner_network_header(skb);
716 l4_hdr = skb_inner_transport_header(skb);
718 if (l3.v6->version == 6) {
719 exthdr = l3.hdr + sizeof(*l3.v6);
720 l4_proto_tmp = l3.v6->nexthdr;
721 if (l4_hdr != exthdr)
722 ipv6_skip_exthdr(skb, exthdr - skb->data,
723 &l4_proto_tmp, &frag_off);
724 } else if (l3.v4->version == 4) {
725 l4_proto_tmp = l3.v4->protocol;
728 *il4_proto = l4_proto_tmp;
733 static void hns3_set_l2l3l4_len(struct sk_buff *skb, u8 ol4_proto,
734 u8 il4_proto, u32 *type_cs_vlan_tso,
735 u32 *ol_type_vlan_len_msec)
737 union l3_hdr_info l3;
738 union l4_hdr_info l4;
739 unsigned char *l2_hdr;
740 u8 l4_proto = ol4_proto;
747 l3.hdr = skb_network_header(skb);
748 l4.hdr = skb_transport_header(skb);
750 /* compute L2 header size for normal packet, defined in 2 Bytes */
751 l2_len = l3.hdr - skb->data;
752 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S, l2_len >> 1);
755 if (skb->encapsulation) {
756 /* compute OL2 header size, defined in 2 Bytes */
758 hns3_set_field(*ol_type_vlan_len_msec,
759 HNS3_TXD_L2LEN_S, ol2_len >> 1);
761 /* compute OL3 header size, defined in 4 Bytes */
762 ol3_len = l4.hdr - l3.hdr;
763 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_S,
766 /* MAC in UDP, MAC in GRE (0x6558)*/
767 if ((ol4_proto == IPPROTO_UDP) || (ol4_proto == IPPROTO_GRE)) {
768 /* switch MAC header ptr from outer to inner header.*/
769 l2_hdr = skb_inner_mac_header(skb);
771 /* compute OL4 header size, defined in 4 Bytes. */
772 ol4_len = l2_hdr - l4.hdr;
773 hns3_set_field(*ol_type_vlan_len_msec,
774 HNS3_TXD_L4LEN_S, ol4_len >> 2);
776 /* switch IP header ptr from outer to inner header */
777 l3.hdr = skb_inner_network_header(skb);
779 /* compute inner l2 header size, defined in 2 Bytes. */
780 l2_len = l3.hdr - l2_hdr;
781 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S,
784 /* skb packet types not supported by hardware,
785 * txbd len fild doesn't be filled.
790 /* switch L4 header pointer from outer to inner */
791 l4.hdr = skb_inner_transport_header(skb);
793 l4_proto = il4_proto;
796 /* compute inner(/normal) L3 header size, defined in 4 Bytes */
797 l3_len = l4.hdr - l3.hdr;
798 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_S, l3_len >> 2);
800 /* compute inner(/normal) L4 header size, defined in 4 Bytes */
803 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
807 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
808 (sizeof(struct sctphdr) >> 2));
811 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
812 (sizeof(struct udphdr) >> 2));
815 /* skb packet types not supported by hardware,
816 * txbd len fild doesn't be filled.
822 /* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
823 * and it is udp packet, which has a dest port as the IANA assigned.
824 * the hardware is expected to do the checksum offload, but the
825 * hardware will not do the checksum offload when udp dest port is
828 static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
830 #define IANA_VXLAN_PORT 4789
831 union l4_hdr_info l4;
833 l4.hdr = skb_transport_header(skb);
835 if (!(!skb->encapsulation && l4.udp->dest == htons(IANA_VXLAN_PORT)))
838 skb_checksum_help(skb);
843 static int hns3_set_l3l4_type_csum(struct sk_buff *skb, u8 ol4_proto,
844 u8 il4_proto, u32 *type_cs_vlan_tso,
845 u32 *ol_type_vlan_len_msec)
847 union l3_hdr_info l3;
848 u32 l4_proto = ol4_proto;
850 l3.hdr = skb_network_header(skb);
852 /* define OL3 type and tunnel type(OL4).*/
853 if (skb->encapsulation) {
854 /* define outer network header type.*/
855 if (skb->protocol == htons(ETH_P_IP)) {
857 hns3_set_field(*ol_type_vlan_len_msec,
859 HNS3_OL3T_IPV4_CSUM);
861 hns3_set_field(*ol_type_vlan_len_msec,
863 HNS3_OL3T_IPV4_NO_CSUM);
865 } else if (skb->protocol == htons(ETH_P_IPV6)) {
866 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_S,
870 /* define tunnel type(OL4).*/
873 hns3_set_field(*ol_type_vlan_len_msec,
875 HNS3_TUN_MAC_IN_UDP);
878 hns3_set_field(*ol_type_vlan_len_msec,
883 /* drop the skb tunnel packet if hardware don't support,
884 * because hardware can't calculate csum when TSO.
889 /* the stack computes the IP header already,
890 * driver calculate l4 checksum when not TSO.
892 skb_checksum_help(skb);
896 l3.hdr = skb_inner_network_header(skb);
897 l4_proto = il4_proto;
900 if (l3.v4->version == 4) {
901 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
904 /* the stack computes the IP header already, the only time we
905 * need the hardware to recompute it is in the case of TSO.
908 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3CS_B, 1);
909 } else if (l3.v6->version == 6) {
910 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
916 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
917 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
921 if (hns3_tunnel_csum_bug(skb))
924 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
925 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
929 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
930 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
934 /* drop the skb tunnel packet if hardware don't support,
935 * because hardware can't calculate csum when TSO.
940 /* the stack computes the IP header already,
941 * driver calculate l4 checksum when not TSO.
943 skb_checksum_help(skb);
950 static void hns3_set_txbd_baseinfo(u16 *bdtp_fe_sc_vld_ra_ri, int frag_end)
952 /* Config bd buffer end */
953 hns3_set_field(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_FE_B, !!frag_end);
954 hns3_set_field(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_VLD_B, 1);
957 static int hns3_fill_desc_vtags(struct sk_buff *skb,
958 struct hns3_enet_ring *tx_ring,
959 u32 *inner_vlan_flag,
964 #define HNS3_TX_VLAN_PRIO_SHIFT 13
966 if (skb->protocol == htons(ETH_P_8021Q) &&
967 !(tx_ring->tqp->handle->kinfo.netdev->features &
968 NETIF_F_HW_VLAN_CTAG_TX)) {
969 /* When HW VLAN acceleration is turned off, and the stack
970 * sets the protocol to 802.1q, the driver just need to
971 * set the protocol to the encapsulated ethertype.
973 skb->protocol = vlan_get_protocol(skb);
977 if (skb_vlan_tag_present(skb)) {
980 vlan_tag = skb_vlan_tag_get(skb);
981 vlan_tag |= (skb->priority & 0x7) << HNS3_TX_VLAN_PRIO_SHIFT;
983 /* Based on hw strategy, use out_vtag in two layer tag case,
984 * and use inner_vtag in one tag case.
986 if (skb->protocol == htons(ETH_P_8021Q)) {
987 hns3_set_field(*out_vlan_flag, HNS3_TXD_OVLAN_B, 1);
988 *out_vtag = vlan_tag;
990 hns3_set_field(*inner_vlan_flag, HNS3_TXD_VLAN_B, 1);
991 *inner_vtag = vlan_tag;
993 } else if (skb->protocol == htons(ETH_P_8021Q)) {
994 struct vlan_ethhdr *vhdr;
997 rc = skb_cow_head(skb, 0);
998 if (unlikely(rc < 0))
1000 vhdr = (struct vlan_ethhdr *)skb->data;
1001 vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority & 0x7)
1002 << HNS3_TX_VLAN_PRIO_SHIFT);
1005 skb->protocol = vlan_get_protocol(skb);
1009 static int hns3_fill_desc(struct hns3_enet_ring *ring, void *priv,
1010 int size, int frag_end, enum hns_desc_type type)
1012 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1013 struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1014 struct device *dev = ring_to_dev(ring);
1015 u16 bdtp_fe_sc_vld_ra_ri = 0;
1016 struct skb_frag_struct *frag;
1017 unsigned int frag_buf_num;
1021 if (type == DESC_TYPE_SKB) {
1022 struct sk_buff *skb = (struct sk_buff *)priv;
1023 u32 ol_type_vlan_len_msec = 0;
1024 u32 type_cs_vlan_tso = 0;
1025 u32 paylen = skb->len;
1031 ret = hns3_fill_desc_vtags(skb, ring, &type_cs_vlan_tso,
1032 &ol_type_vlan_len_msec,
1033 &inner_vtag, &out_vtag);
1037 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1038 u8 ol4_proto, il4_proto;
1040 skb_reset_mac_len(skb);
1042 ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
1045 hns3_set_l2l3l4_len(skb, ol4_proto, il4_proto,
1047 &ol_type_vlan_len_msec);
1048 ret = hns3_set_l3l4_type_csum(skb, ol4_proto, il4_proto,
1050 &ol_type_vlan_len_msec);
1054 ret = hns3_set_tso(skb, &paylen, &mss,
1061 desc->tx.ol_type_vlan_len_msec =
1062 cpu_to_le32(ol_type_vlan_len_msec);
1063 desc->tx.type_cs_vlan_tso_len =
1064 cpu_to_le32(type_cs_vlan_tso);
1065 desc->tx.paylen = cpu_to_le32(paylen);
1066 desc->tx.mss = cpu_to_le16(mss);
1067 desc->tx.vlan_tag = cpu_to_le16(inner_vtag);
1068 desc->tx.outer_vlan_tag = cpu_to_le16(out_vtag);
1070 dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1072 frag = (struct skb_frag_struct *)priv;
1073 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1076 if (unlikely(dma_mapping_error(ring->dev, dma))) {
1077 ring->stats.sw_err_cnt++;
1081 desc_cb->length = size;
1083 frag_buf_num = hns3_tx_bd_count(size);
1084 sizeoflast = size & HNS3_TX_LAST_SIZE_M;
1085 sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1087 /* When frag size is bigger than hardware limit, split this frag */
1088 for (k = 0; k < frag_buf_num; k++) {
1089 /* The txbd's baseinfo of DESC_TYPE_PAGE & DESC_TYPE_SKB */
1090 desc_cb->priv = priv;
1091 desc_cb->dma = dma + HNS3_MAX_BD_SIZE * k;
1092 desc_cb->type = (type == DESC_TYPE_SKB && !k) ?
1093 DESC_TYPE_SKB : DESC_TYPE_PAGE;
1095 /* now, fill the descriptor */
1096 desc->addr = cpu_to_le64(dma + HNS3_MAX_BD_SIZE * k);
1097 desc->tx.send_size = cpu_to_le16((k == frag_buf_num - 1) ?
1098 (u16)sizeoflast : (u16)HNS3_MAX_BD_SIZE);
1099 hns3_set_txbd_baseinfo(&bdtp_fe_sc_vld_ra_ri,
1100 frag_end && (k == frag_buf_num - 1) ?
1102 desc->tx.bdtp_fe_sc_vld_ra_ri =
1103 cpu_to_le16(bdtp_fe_sc_vld_ra_ri);
1105 /* move ring pointer to next.*/
1106 ring_ptr_move_fw(ring, next_to_use);
1108 desc_cb = &ring->desc_cb[ring->next_to_use];
1109 desc = &ring->desc[ring->next_to_use];
1115 static int hns3_nic_maybe_stop_tso(struct sk_buff **out_skb, int *bnum,
1116 struct hns3_enet_ring *ring)
1118 struct sk_buff *skb = *out_skb;
1119 struct sk_buff *new_skb = NULL;
1120 struct skb_frag_struct *frag;
1127 size = skb_headlen(skb);
1128 buf_num = hns3_tx_bd_count(size);
1130 frag_num = skb_shinfo(skb)->nr_frags;
1131 for (i = 0; i < frag_num; i++) {
1132 frag = &skb_shinfo(skb)->frags[i];
1133 size = skb_frag_size(frag);
1134 bdnum_for_frag = hns3_tx_bd_count(size);
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 = hns3_tx_bd_count(skb->len);
1143 if (ring_space(ring) < buf_num)
1145 /* manual split the send packet */
1146 new_skb = skb_copy(skb, GFP_ATOMIC);
1149 dev_kfree_skb_any(skb);
1153 if (unlikely(ring_space(ring) < buf_num))
1160 static int hns3_nic_maybe_stop_tx(struct sk_buff **out_skb, int *bnum,
1161 struct hns3_enet_ring *ring)
1163 struct sk_buff *skb = *out_skb;
1164 struct sk_buff *new_skb = NULL;
1167 /* No. of segments (plus a header) */
1168 buf_num = skb_shinfo(skb)->nr_frags + 1;
1170 if (unlikely(buf_num > HNS3_MAX_BD_PER_FRAG)) {
1171 buf_num = hns3_tx_bd_count(skb->len);
1172 if (ring_space(ring) < buf_num)
1174 /* manual split the send packet */
1175 new_skb = skb_copy(skb, GFP_ATOMIC);
1178 dev_kfree_skb_any(skb);
1182 if (unlikely(ring_space(ring) < buf_num))
1190 static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
1192 struct device *dev = ring_to_dev(ring);
1195 for (i = 0; i < ring->desc_num; i++) {
1196 /* check if this is where we started */
1197 if (ring->next_to_use == next_to_use_orig)
1200 /* unmap the descriptor dma address */
1201 if (ring->desc_cb[ring->next_to_use].type == DESC_TYPE_SKB)
1202 dma_unmap_single(dev,
1203 ring->desc_cb[ring->next_to_use].dma,
1204 ring->desc_cb[ring->next_to_use].length,
1206 else if (ring->desc_cb[ring->next_to_use].length)
1208 ring->desc_cb[ring->next_to_use].dma,
1209 ring->desc_cb[ring->next_to_use].length,
1212 ring->desc_cb[ring->next_to_use].length = 0;
1215 ring_ptr_move_bw(ring, next_to_use);
1219 netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
1221 struct hns3_nic_priv *priv = netdev_priv(netdev);
1222 struct hns3_nic_ring_data *ring_data =
1223 &tx_ring_data(priv, skb->queue_mapping);
1224 struct hns3_enet_ring *ring = ring_data->ring;
1225 struct netdev_queue *dev_queue;
1226 struct skb_frag_struct *frag;
1227 int next_to_use_head;
1228 int next_to_use_frag;
1235 /* Prefetch the data used later */
1236 prefetch(skb->data);
1238 switch (priv->ops.maybe_stop_tx(&skb, &buf_num, ring)) {
1240 u64_stats_update_begin(&ring->syncp);
1241 ring->stats.tx_busy++;
1242 u64_stats_update_end(&ring->syncp);
1244 goto out_net_tx_busy;
1246 u64_stats_update_begin(&ring->syncp);
1247 ring->stats.sw_err_cnt++;
1248 u64_stats_update_end(&ring->syncp);
1249 netdev_err(netdev, "no memory to xmit!\n");
1256 /* No. of segments (plus a header) */
1257 seg_num = skb_shinfo(skb)->nr_frags + 1;
1258 /* Fill the first part */
1259 size = skb_headlen(skb);
1261 next_to_use_head = ring->next_to_use;
1263 ret = hns3_fill_desc(ring, skb, size, seg_num == 1 ? 1 : 0,
1268 next_to_use_frag = ring->next_to_use;
1269 /* Fill the fragments */
1270 for (i = 1; i < seg_num; i++) {
1271 frag = &skb_shinfo(skb)->frags[i - 1];
1272 size = skb_frag_size(frag);
1274 ret = hns3_fill_desc(ring, frag, size,
1275 seg_num - 1 == i ? 1 : 0,
1282 /* Complete translate all packets */
1283 dev_queue = netdev_get_tx_queue(netdev, ring_data->queue_index);
1284 netdev_tx_sent_queue(dev_queue, skb->len);
1286 wmb(); /* Commit all data before submit */
1288 hnae3_queue_xmit(ring->tqp, buf_num);
1290 return NETDEV_TX_OK;
1293 hns3_clear_desc(ring, next_to_use_frag);
1296 hns3_clear_desc(ring, next_to_use_head);
1299 dev_kfree_skb_any(skb);
1300 return NETDEV_TX_OK;
1303 netif_stop_subqueue(netdev, ring_data->queue_index);
1304 smp_mb(); /* Commit all data before submit */
1306 return NETDEV_TX_BUSY;
1309 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
1311 struct hnae3_handle *h = hns3_get_handle(netdev);
1312 struct sockaddr *mac_addr = p;
1315 if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1316 return -EADDRNOTAVAIL;
1318 if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
1319 netdev_info(netdev, "already using mac address %pM\n",
1324 ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
1326 netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
1330 ether_addr_copy(netdev->dev_addr, mac_addr->sa_data);
1335 static int hns3_nic_do_ioctl(struct net_device *netdev,
1336 struct ifreq *ifr, int cmd)
1338 struct hnae3_handle *h = hns3_get_handle(netdev);
1340 if (!netif_running(netdev))
1343 if (!h->ae_algo->ops->do_ioctl)
1346 return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
1349 static int hns3_nic_set_features(struct net_device *netdev,
1350 netdev_features_t features)
1352 netdev_features_t changed = netdev->features ^ features;
1353 struct hns3_nic_priv *priv = netdev_priv(netdev);
1354 struct hnae3_handle *h = priv->ae_handle;
1358 if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) {
1359 if (features & (NETIF_F_TSO | NETIF_F_TSO6))
1360 priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tso;
1362 priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tx;
1365 if (changed & (NETIF_F_GRO_HW) && h->ae_algo->ops->set_gro_en) {
1366 enable = !!(features & NETIF_F_GRO_HW);
1367 ret = h->ae_algo->ops->set_gro_en(h, enable);
1372 if ((changed & NETIF_F_HW_VLAN_CTAG_FILTER) &&
1373 h->ae_algo->ops->enable_vlan_filter) {
1374 enable = !!(features & NETIF_F_HW_VLAN_CTAG_FILTER);
1375 h->ae_algo->ops->enable_vlan_filter(h, enable);
1378 if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
1379 h->ae_algo->ops->enable_hw_strip_rxvtag) {
1380 enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
1381 ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, enable);
1386 if ((changed & NETIF_F_NTUPLE) && h->ae_algo->ops->enable_fd) {
1387 enable = !!(features & NETIF_F_NTUPLE);
1388 h->ae_algo->ops->enable_fd(h, enable);
1391 netdev->features = features;
1395 static void hns3_nic_get_stats64(struct net_device *netdev,
1396 struct rtnl_link_stats64 *stats)
1398 struct hns3_nic_priv *priv = netdev_priv(netdev);
1399 int queue_num = priv->ae_handle->kinfo.num_tqps;
1400 struct hnae3_handle *handle = priv->ae_handle;
1401 struct hns3_enet_ring *ring;
1402 u64 rx_length_errors = 0;
1403 u64 rx_crc_errors = 0;
1404 u64 rx_multicast = 0;
1416 if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
1419 handle->ae_algo->ops->update_stats(handle, &netdev->stats);
1421 for (idx = 0; idx < queue_num; idx++) {
1422 /* fetch the tx stats */
1423 ring = priv->ring_data[idx].ring;
1425 start = u64_stats_fetch_begin_irq(&ring->syncp);
1426 tx_bytes += ring->stats.tx_bytes;
1427 tx_pkts += ring->stats.tx_pkts;
1428 tx_drop += ring->stats.sw_err_cnt;
1429 tx_errors += ring->stats.sw_err_cnt;
1430 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1432 /* fetch the rx stats */
1433 ring = priv->ring_data[idx + queue_num].ring;
1435 start = u64_stats_fetch_begin_irq(&ring->syncp);
1436 rx_bytes += ring->stats.rx_bytes;
1437 rx_pkts += ring->stats.rx_pkts;
1438 rx_drop += ring->stats.non_vld_descs;
1439 rx_drop += ring->stats.l2_err;
1440 rx_errors += ring->stats.non_vld_descs;
1441 rx_errors += ring->stats.l2_err;
1442 rx_crc_errors += ring->stats.l2_err;
1443 rx_crc_errors += ring->stats.l3l4_csum_err;
1444 rx_multicast += ring->stats.rx_multicast;
1445 rx_length_errors += ring->stats.err_pkt_len;
1446 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1449 stats->tx_bytes = tx_bytes;
1450 stats->tx_packets = tx_pkts;
1451 stats->rx_bytes = rx_bytes;
1452 stats->rx_packets = rx_pkts;
1454 stats->rx_errors = rx_errors;
1455 stats->multicast = rx_multicast;
1456 stats->rx_length_errors = rx_length_errors;
1457 stats->rx_crc_errors = rx_crc_errors;
1458 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
1460 stats->tx_errors = tx_errors;
1461 stats->rx_dropped = rx_drop;
1462 stats->tx_dropped = tx_drop;
1463 stats->collisions = netdev->stats.collisions;
1464 stats->rx_over_errors = netdev->stats.rx_over_errors;
1465 stats->rx_frame_errors = netdev->stats.rx_frame_errors;
1466 stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
1467 stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
1468 stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
1469 stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
1470 stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
1471 stats->tx_window_errors = netdev->stats.tx_window_errors;
1472 stats->rx_compressed = netdev->stats.rx_compressed;
1473 stats->tx_compressed = netdev->stats.tx_compressed;
1476 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
1478 struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
1479 struct hnae3_handle *h = hns3_get_handle(netdev);
1480 struct hnae3_knic_private_info *kinfo = &h->kinfo;
1481 u8 *prio_tc = mqprio_qopt->qopt.prio_tc_map;
1482 u8 tc = mqprio_qopt->qopt.num_tc;
1483 u16 mode = mqprio_qopt->mode;
1484 u8 hw = mqprio_qopt->qopt.hw;
1486 if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
1487 mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
1490 if (tc > HNAE3_MAX_TC)
1496 return (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
1497 kinfo->dcb_ops->setup_tc(h, tc, prio_tc) : -EOPNOTSUPP;
1500 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
1503 if (type != TC_SETUP_QDISC_MQPRIO)
1506 return hns3_setup_tc(dev, type_data);
1509 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
1510 __be16 proto, u16 vid)
1512 struct hnae3_handle *h = hns3_get_handle(netdev);
1513 struct hns3_nic_priv *priv = netdev_priv(netdev);
1516 if (h->ae_algo->ops->set_vlan_filter)
1517 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
1520 set_bit(vid, priv->active_vlans);
1525 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
1526 __be16 proto, u16 vid)
1528 struct hnae3_handle *h = hns3_get_handle(netdev);
1529 struct hns3_nic_priv *priv = netdev_priv(netdev);
1532 if (h->ae_algo->ops->set_vlan_filter)
1533 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
1536 clear_bit(vid, priv->active_vlans);
1541 static int hns3_restore_vlan(struct net_device *netdev)
1543 struct hns3_nic_priv *priv = netdev_priv(netdev);
1547 for_each_set_bit(vid, priv->active_vlans, VLAN_N_VID) {
1548 ret = hns3_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
1550 netdev_err(netdev, "Restore vlan: %d filter, ret:%d\n",
1559 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
1560 u8 qos, __be16 vlan_proto)
1562 struct hnae3_handle *h = hns3_get_handle(netdev);
1565 if (h->ae_algo->ops->set_vf_vlan_filter)
1566 ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
1572 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
1574 struct hnae3_handle *h = hns3_get_handle(netdev);
1577 if (!h->ae_algo->ops->set_mtu)
1580 ret = h->ae_algo->ops->set_mtu(h, new_mtu);
1582 netdev_err(netdev, "failed to change MTU in hardware %d\n",
1585 netdev->mtu = new_mtu;
1590 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
1592 struct hns3_nic_priv *priv = netdev_priv(ndev);
1593 struct hns3_enet_ring *tx_ring = NULL;
1594 int timeout_queue = 0;
1595 int hw_head, hw_tail;
1598 /* Find the stopped queue the same way the stack does */
1599 for (i = 0; i < ndev->real_num_tx_queues; i++) {
1600 struct netdev_queue *q;
1601 unsigned long trans_start;
1603 q = netdev_get_tx_queue(ndev, i);
1604 trans_start = q->trans_start;
1605 if (netif_xmit_stopped(q) &&
1607 (trans_start + ndev->watchdog_timeo))) {
1613 if (i == ndev->num_tx_queues) {
1615 "no netdev TX timeout queue found, timeout count: %llu\n",
1616 priv->tx_timeout_count);
1620 tx_ring = priv->ring_data[timeout_queue].ring;
1622 hw_head = readl_relaxed(tx_ring->tqp->io_base +
1623 HNS3_RING_TX_RING_HEAD_REG);
1624 hw_tail = readl_relaxed(tx_ring->tqp->io_base +
1625 HNS3_RING_TX_RING_TAIL_REG);
1627 "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",
1628 priv->tx_timeout_count,
1630 tx_ring->next_to_use,
1631 tx_ring->next_to_clean,
1634 readl(tx_ring->tqp_vector->mask_addr));
1639 static void hns3_nic_net_timeout(struct net_device *ndev)
1641 struct hns3_nic_priv *priv = netdev_priv(ndev);
1642 struct hnae3_handle *h = priv->ae_handle;
1644 if (!hns3_get_tx_timeo_queue_info(ndev))
1647 priv->tx_timeout_count++;
1649 /* request the reset, and let the hclge to determine
1650 * which reset level should be done
1652 if (h->ae_algo->ops->reset_event)
1653 h->ae_algo->ops->reset_event(h->pdev, h);
1656 static const struct net_device_ops hns3_nic_netdev_ops = {
1657 .ndo_open = hns3_nic_net_open,
1658 .ndo_stop = hns3_nic_net_stop,
1659 .ndo_start_xmit = hns3_nic_net_xmit,
1660 .ndo_tx_timeout = hns3_nic_net_timeout,
1661 .ndo_set_mac_address = hns3_nic_net_set_mac_address,
1662 .ndo_do_ioctl = hns3_nic_do_ioctl,
1663 .ndo_change_mtu = hns3_nic_change_mtu,
1664 .ndo_set_features = hns3_nic_set_features,
1665 .ndo_get_stats64 = hns3_nic_get_stats64,
1666 .ndo_setup_tc = hns3_nic_setup_tc,
1667 .ndo_set_rx_mode = hns3_nic_set_rx_mode,
1668 .ndo_vlan_rx_add_vid = hns3_vlan_rx_add_vid,
1669 .ndo_vlan_rx_kill_vid = hns3_vlan_rx_kill_vid,
1670 .ndo_set_vf_vlan = hns3_ndo_set_vf_vlan,
1673 static bool hns3_is_phys_func(struct pci_dev *pdev)
1675 u32 dev_id = pdev->device;
1678 case HNAE3_DEV_ID_GE:
1679 case HNAE3_DEV_ID_25GE:
1680 case HNAE3_DEV_ID_25GE_RDMA:
1681 case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
1682 case HNAE3_DEV_ID_50GE_RDMA:
1683 case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
1684 case HNAE3_DEV_ID_100G_RDMA_MACSEC:
1686 case HNAE3_DEV_ID_100G_VF:
1687 case HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF:
1690 dev_warn(&pdev->dev, "un-recognized pci device-id %d",
1697 static void hns3_disable_sriov(struct pci_dev *pdev)
1699 /* If our VFs are assigned we cannot shut down SR-IOV
1700 * without causing issues, so just leave the hardware
1701 * available but disabled
1703 if (pci_vfs_assigned(pdev)) {
1704 dev_warn(&pdev->dev,
1705 "disabling driver while VFs are assigned\n");
1709 pci_disable_sriov(pdev);
1712 static void hns3_get_dev_capability(struct pci_dev *pdev,
1713 struct hnae3_ae_dev *ae_dev)
1715 if (pdev->revision >= 0x21) {
1716 hnae3_set_bit(ae_dev->flag, HNAE3_DEV_SUPPORT_FD_B, 1);
1717 hnae3_set_bit(ae_dev->flag, HNAE3_DEV_SUPPORT_GRO_B, 1);
1721 /* hns3_probe - Device initialization routine
1722 * @pdev: PCI device information struct
1723 * @ent: entry in hns3_pci_tbl
1725 * hns3_probe initializes a PF identified by a pci_dev structure.
1726 * The OS initialization, configuring of the PF private structure,
1727 * and a hardware reset occur.
1729 * Returns 0 on success, negative on failure
1731 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1733 struct hnae3_ae_dev *ae_dev;
1736 ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev),
1743 ae_dev->pdev = pdev;
1744 ae_dev->flag = ent->driver_data;
1745 ae_dev->dev_type = HNAE3_DEV_KNIC;
1746 ae_dev->reset_type = HNAE3_NONE_RESET;
1747 hns3_get_dev_capability(pdev, ae_dev);
1748 pci_set_drvdata(pdev, ae_dev);
1750 ret = hnae3_register_ae_dev(ae_dev);
1752 devm_kfree(&pdev->dev, ae_dev);
1753 pci_set_drvdata(pdev, NULL);
1759 /* hns3_remove - Device removal routine
1760 * @pdev: PCI device information struct
1762 static void hns3_remove(struct pci_dev *pdev)
1764 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1766 if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
1767 hns3_disable_sriov(pdev);
1769 hnae3_unregister_ae_dev(ae_dev);
1770 pci_set_drvdata(pdev, NULL);
1774 * hns3_pci_sriov_configure
1775 * @pdev: pointer to a pci_dev structure
1776 * @num_vfs: number of VFs to allocate
1778 * Enable or change the number of VFs. Called when the user updates the number
1781 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
1785 if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
1786 dev_warn(&pdev->dev, "Can not config SRIOV\n");
1791 ret = pci_enable_sriov(pdev, num_vfs);
1793 dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
1796 } else if (!pci_vfs_assigned(pdev)) {
1797 pci_disable_sriov(pdev);
1799 dev_warn(&pdev->dev,
1800 "Unable to free VFs because some are assigned to VMs.\n");
1806 static void hns3_shutdown(struct pci_dev *pdev)
1808 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1810 hnae3_unregister_ae_dev(ae_dev);
1811 devm_kfree(&pdev->dev, ae_dev);
1812 pci_set_drvdata(pdev, NULL);
1814 if (system_state == SYSTEM_POWER_OFF)
1815 pci_set_power_state(pdev, PCI_D3hot);
1818 static pci_ers_result_t hns3_error_detected(struct pci_dev *pdev,
1819 pci_channel_state_t state)
1821 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1822 pci_ers_result_t ret;
1824 dev_info(&pdev->dev, "PCI error detected, state(=%d)!!\n", state);
1826 if (state == pci_channel_io_perm_failure)
1827 return PCI_ERS_RESULT_DISCONNECT;
1831 "Can't recover - error happened during device init\n");
1832 return PCI_ERS_RESULT_NONE;
1835 if (ae_dev->ops->handle_hw_ras_error)
1836 ret = ae_dev->ops->handle_hw_ras_error(ae_dev);
1838 return PCI_ERS_RESULT_NONE;
1843 static pci_ers_result_t hns3_slot_reset(struct pci_dev *pdev)
1845 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1846 struct device *dev = &pdev->dev;
1848 dev_info(dev, "requesting reset due to PCI error\n");
1850 /* request the reset */
1851 if (ae_dev->ops->reset_event) {
1852 if (!ae_dev->override_pci_need_reset)
1853 ae_dev->ops->reset_event(pdev, NULL);
1855 return PCI_ERS_RESULT_RECOVERED;
1858 return PCI_ERS_RESULT_DISCONNECT;
1861 static void hns3_reset_prepare(struct pci_dev *pdev)
1863 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1865 dev_info(&pdev->dev, "hns3 flr prepare\n");
1866 if (ae_dev && ae_dev->ops && ae_dev->ops->flr_prepare)
1867 ae_dev->ops->flr_prepare(ae_dev);
1870 static void hns3_reset_done(struct pci_dev *pdev)
1872 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1874 dev_info(&pdev->dev, "hns3 flr done\n");
1875 if (ae_dev && ae_dev->ops && ae_dev->ops->flr_done)
1876 ae_dev->ops->flr_done(ae_dev);
1879 static const struct pci_error_handlers hns3_err_handler = {
1880 .error_detected = hns3_error_detected,
1881 .slot_reset = hns3_slot_reset,
1882 .reset_prepare = hns3_reset_prepare,
1883 .reset_done = hns3_reset_done,
1886 static struct pci_driver hns3_driver = {
1887 .name = hns3_driver_name,
1888 .id_table = hns3_pci_tbl,
1889 .probe = hns3_probe,
1890 .remove = hns3_remove,
1891 .shutdown = hns3_shutdown,
1892 .sriov_configure = hns3_pci_sriov_configure,
1893 .err_handler = &hns3_err_handler,
1896 /* set default feature to hns3 */
1897 static void hns3_set_default_feature(struct net_device *netdev)
1899 struct hnae3_handle *h = hns3_get_handle(netdev);
1900 struct pci_dev *pdev = h->pdev;
1902 netdev->priv_flags |= IFF_UNICAST_FLT;
1904 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1905 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1906 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1907 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1908 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1910 netdev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
1912 netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
1914 netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1915 NETIF_F_HW_VLAN_CTAG_FILTER |
1916 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
1917 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1918 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1919 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1920 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1922 netdev->vlan_features |=
1923 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
1924 NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO |
1925 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1926 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1927 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1929 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1930 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
1931 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1932 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1933 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1934 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1936 if (pdev->revision >= 0x21) {
1937 netdev->hw_features |= NETIF_F_GRO_HW;
1938 netdev->features |= NETIF_F_GRO_HW;
1940 if (!(h->flags & HNAE3_SUPPORT_VF)) {
1941 netdev->hw_features |= NETIF_F_NTUPLE;
1942 netdev->features |= NETIF_F_NTUPLE;
1947 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
1948 struct hns3_desc_cb *cb)
1950 unsigned int order = hnae3_page_order(ring);
1953 p = dev_alloc_pages(order);
1958 cb->page_offset = 0;
1960 cb->buf = page_address(p);
1961 cb->length = hnae3_page_size(ring);
1962 cb->type = DESC_TYPE_PAGE;
1967 static void hns3_free_buffer(struct hns3_enet_ring *ring,
1968 struct hns3_desc_cb *cb)
1970 if (cb->type == DESC_TYPE_SKB)
1971 dev_kfree_skb_any((struct sk_buff *)cb->priv);
1972 else if (!HNAE3_IS_TX_RING(ring))
1973 put_page((struct page *)cb->priv);
1974 memset(cb, 0, sizeof(*cb));
1977 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
1979 cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
1980 cb->length, ring_to_dma_dir(ring));
1982 if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
1988 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
1989 struct hns3_desc_cb *cb)
1991 if (cb->type == DESC_TYPE_SKB)
1992 dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
1993 ring_to_dma_dir(ring));
1994 else if (cb->length)
1995 dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
1996 ring_to_dma_dir(ring));
1999 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
2001 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2002 ring->desc[i].addr = 0;
2005 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i)
2007 struct hns3_desc_cb *cb = &ring->desc_cb[i];
2009 if (!ring->desc_cb[i].dma)
2012 hns3_buffer_detach(ring, i);
2013 hns3_free_buffer(ring, cb);
2016 static void hns3_free_buffers(struct hns3_enet_ring *ring)
2020 for (i = 0; i < ring->desc_num; i++)
2021 hns3_free_buffer_detach(ring, i);
2024 /* free desc along with its attached buffer */
2025 static void hns3_free_desc(struct hns3_enet_ring *ring)
2027 int size = ring->desc_num * sizeof(ring->desc[0]);
2029 hns3_free_buffers(ring);
2032 dma_free_coherent(ring_to_dev(ring), size,
2033 ring->desc, ring->desc_dma_addr);
2038 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
2040 int size = ring->desc_num * sizeof(ring->desc[0]);
2042 ring->desc = dma_alloc_coherent(ring_to_dev(ring), size,
2043 &ring->desc_dma_addr, GFP_KERNEL);
2050 static int hns3_reserve_buffer_map(struct hns3_enet_ring *ring,
2051 struct hns3_desc_cb *cb)
2055 ret = hns3_alloc_buffer(ring, cb);
2059 ret = hns3_map_buffer(ring, cb);
2066 hns3_free_buffer(ring, cb);
2071 static int hns3_alloc_buffer_attach(struct hns3_enet_ring *ring, int i)
2073 int ret = hns3_reserve_buffer_map(ring, &ring->desc_cb[i]);
2078 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2083 /* Allocate memory for raw pkg, and map with dma */
2084 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
2088 for (i = 0; i < ring->desc_num; i++) {
2089 ret = hns3_alloc_buffer_attach(ring, i);
2091 goto out_buffer_fail;
2097 for (j = i - 1; j >= 0; j--)
2098 hns3_free_buffer_detach(ring, j);
2102 /* detach a in-used buffer and replace with a reserved one */
2103 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
2104 struct hns3_desc_cb *res_cb)
2106 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2107 ring->desc_cb[i] = *res_cb;
2108 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2109 ring->desc[i].rx.bd_base_info = 0;
2112 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
2114 ring->desc_cb[i].reuse_flag = 0;
2115 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma
2116 + ring->desc_cb[i].page_offset);
2117 ring->desc[i].rx.bd_base_info = 0;
2120 static void hns3_nic_reclaim_one_desc(struct hns3_enet_ring *ring, int *bytes,
2123 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
2125 (*pkts) += (desc_cb->type == DESC_TYPE_SKB);
2126 (*bytes) += desc_cb->length;
2127 /* desc_cb will be cleaned, after hnae3_free_buffer_detach*/
2128 hns3_free_buffer_detach(ring, ring->next_to_clean);
2130 ring_ptr_move_fw(ring, next_to_clean);
2133 static int is_valid_clean_head(struct hns3_enet_ring *ring, int h)
2135 int u = ring->next_to_use;
2136 int c = ring->next_to_clean;
2138 if (unlikely(h > ring->desc_num))
2141 return u > c ? (h > c && h <= u) : (h > c || h <= u);
2144 void hns3_clean_tx_ring(struct hns3_enet_ring *ring)
2146 struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2147 struct hns3_nic_priv *priv = netdev_priv(netdev);
2148 struct netdev_queue *dev_queue;
2152 head = readl_relaxed(ring->tqp->io_base + HNS3_RING_TX_RING_HEAD_REG);
2153 rmb(); /* Make sure head is ready before touch any data */
2155 if (is_ring_empty(ring) || head == ring->next_to_clean)
2156 return; /* no data to poll */
2158 if (unlikely(!is_valid_clean_head(ring, head))) {
2159 netdev_err(netdev, "wrong head (%d, %d-%d)\n", head,
2160 ring->next_to_use, ring->next_to_clean);
2162 u64_stats_update_begin(&ring->syncp);
2163 ring->stats.io_err_cnt++;
2164 u64_stats_update_end(&ring->syncp);
2170 while (head != ring->next_to_clean) {
2171 hns3_nic_reclaim_one_desc(ring, &bytes, &pkts);
2172 /* Issue prefetch for next Tx descriptor */
2173 prefetch(&ring->desc_cb[ring->next_to_clean]);
2176 ring->tqp_vector->tx_group.total_bytes += bytes;
2177 ring->tqp_vector->tx_group.total_packets += pkts;
2179 u64_stats_update_begin(&ring->syncp);
2180 ring->stats.tx_bytes += bytes;
2181 ring->stats.tx_pkts += pkts;
2182 u64_stats_update_end(&ring->syncp);
2184 dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
2185 netdev_tx_completed_queue(dev_queue, pkts, bytes);
2187 if (unlikely(pkts && netif_carrier_ok(netdev) &&
2188 (ring_space(ring) > HNS3_MAX_BD_PER_PKT))) {
2189 /* Make sure that anybody stopping the queue after this
2190 * sees the new next_to_clean.
2193 if (netif_tx_queue_stopped(dev_queue) &&
2194 !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
2195 netif_tx_wake_queue(dev_queue);
2196 ring->stats.restart_queue++;
2201 static int hns3_desc_unused(struct hns3_enet_ring *ring)
2203 int ntc = ring->next_to_clean;
2204 int ntu = ring->next_to_use;
2206 return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
2210 hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring, int cleand_count)
2212 struct hns3_desc_cb *desc_cb;
2213 struct hns3_desc_cb res_cbs;
2216 for (i = 0; i < cleand_count; i++) {
2217 desc_cb = &ring->desc_cb[ring->next_to_use];
2218 if (desc_cb->reuse_flag) {
2219 u64_stats_update_begin(&ring->syncp);
2220 ring->stats.reuse_pg_cnt++;
2221 u64_stats_update_end(&ring->syncp);
2223 hns3_reuse_buffer(ring, ring->next_to_use);
2225 ret = hns3_reserve_buffer_map(ring, &res_cbs);
2227 u64_stats_update_begin(&ring->syncp);
2228 ring->stats.sw_err_cnt++;
2229 u64_stats_update_end(&ring->syncp);
2231 netdev_err(ring->tqp->handle->kinfo.netdev,
2232 "hnae reserve buffer map failed.\n");
2235 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
2238 ring_ptr_move_fw(ring, next_to_use);
2241 wmb(); /* Make all data has been write before submit */
2242 writel_relaxed(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
2245 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
2246 struct hns3_enet_ring *ring, int pull_len,
2247 struct hns3_desc_cb *desc_cb)
2249 struct hns3_desc *desc;
2255 twobufs = ((PAGE_SIZE < 8192) &&
2256 hnae3_buf_size(ring) == HNS3_BUFFER_SIZE_2048);
2258 desc = &ring->desc[ring->next_to_clean];
2259 size = le16_to_cpu(desc->rx.size);
2261 truesize = hnae3_buf_size(ring);
2264 last_offset = hnae3_page_size(ring) - hnae3_buf_size(ring);
2266 skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
2267 size - pull_len, truesize);
2269 /* Avoid re-using remote pages,flag default unreuse */
2270 if (unlikely(page_to_nid(desc_cb->priv) != numa_node_id()))
2274 /* If we are only owner of page we can reuse it */
2275 if (likely(page_count(desc_cb->priv) == 1)) {
2276 /* Flip page offset to other buffer */
2277 desc_cb->page_offset ^= truesize;
2279 desc_cb->reuse_flag = 1;
2280 /* bump ref count on page before it is given*/
2281 get_page(desc_cb->priv);
2286 /* Move offset up to the next cache line */
2287 desc_cb->page_offset += truesize;
2289 if (desc_cb->page_offset <= last_offset) {
2290 desc_cb->reuse_flag = 1;
2291 /* Bump ref count on page before it is given*/
2292 get_page(desc_cb->priv);
2296 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
2297 struct hns3_desc *desc)
2299 struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2300 int l3_type, l4_type;
2305 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2306 l234info = le32_to_cpu(desc->rx.l234_info);
2308 skb->ip_summed = CHECKSUM_NONE;
2310 skb_checksum_none_assert(skb);
2312 if (!(netdev->features & NETIF_F_RXCSUM))
2315 /* We MUST enable hardware checksum before enabling hardware GRO */
2316 if (skb_shinfo(skb)->gso_size) {
2317 skb->ip_summed = CHECKSUM_UNNECESSARY;
2321 /* check if hardware has done checksum */
2322 if (!(bd_base_info & BIT(HNS3_RXD_L3L4P_B)))
2325 if (unlikely(l234info & (BIT(HNS3_RXD_L3E_B) | BIT(HNS3_RXD_L4E_B) |
2326 BIT(HNS3_RXD_OL3E_B) |
2327 BIT(HNS3_RXD_OL4E_B)))) {
2328 u64_stats_update_begin(&ring->syncp);
2329 ring->stats.l3l4_csum_err++;
2330 u64_stats_update_end(&ring->syncp);
2335 ol4_type = hnae3_get_field(l234info, HNS3_RXD_OL4ID_M,
2338 case HNS3_OL4_TYPE_MAC_IN_UDP:
2339 case HNS3_OL4_TYPE_NVGRE:
2340 skb->csum_level = 1;
2342 case HNS3_OL4_TYPE_NO_TUN:
2343 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
2345 l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
2348 /* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
2349 if ((l3_type == HNS3_L3_TYPE_IPV4 ||
2350 l3_type == HNS3_L3_TYPE_IPV6) &&
2351 (l4_type == HNS3_L4_TYPE_UDP ||
2352 l4_type == HNS3_L4_TYPE_TCP ||
2353 l4_type == HNS3_L4_TYPE_SCTP))
2354 skb->ip_summed = CHECKSUM_UNNECESSARY;
2361 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
2363 if (skb_has_frag_list(skb))
2364 napi_gro_flush(&ring->tqp_vector->napi, false);
2366 napi_gro_receive(&ring->tqp_vector->napi, skb);
2369 static bool hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
2370 struct hns3_desc *desc, u32 l234info,
2373 struct pci_dev *pdev = ring->tqp->handle->pdev;
2375 if (pdev->revision == 0x20) {
2376 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2377 if (!(*vlan_tag & VLAN_VID_MASK))
2378 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2380 return (*vlan_tag != 0);
2383 #define HNS3_STRP_OUTER_VLAN 0x1
2384 #define HNS3_STRP_INNER_VLAN 0x2
2386 switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
2387 HNS3_RXD_STRP_TAGP_S)) {
2388 case HNS3_STRP_OUTER_VLAN:
2389 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2391 case HNS3_STRP_INNER_VLAN:
2392 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2399 static int hns3_alloc_skb(struct hns3_enet_ring *ring, int length,
2402 #define HNS3_NEED_ADD_FRAG 1
2403 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
2404 struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2405 struct sk_buff *skb;
2407 ring->skb = napi_alloc_skb(&ring->tqp_vector->napi, HNS3_RX_HEAD_SIZE);
2409 if (unlikely(!skb)) {
2410 netdev_err(netdev, "alloc rx skb fail\n");
2412 u64_stats_update_begin(&ring->syncp);
2413 ring->stats.sw_err_cnt++;
2414 u64_stats_update_end(&ring->syncp);
2419 prefetchw(skb->data);
2421 ring->pending_buf = 1;
2423 ring->tail_skb = NULL;
2424 if (length <= HNS3_RX_HEAD_SIZE) {
2425 memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
2427 /* We can reuse buffer as-is, just make sure it is local */
2428 if (likely(page_to_nid(desc_cb->priv) == numa_node_id()))
2429 desc_cb->reuse_flag = 1;
2430 else /* This page cannot be reused so discard it */
2431 put_page(desc_cb->priv);
2433 ring_ptr_move_fw(ring, next_to_clean);
2436 u64_stats_update_begin(&ring->syncp);
2437 ring->stats.seg_pkt_cnt++;
2438 u64_stats_update_end(&ring->syncp);
2440 ring->pull_len = eth_get_headlen(va, HNS3_RX_HEAD_SIZE);
2441 __skb_put(skb, ring->pull_len);
2442 hns3_nic_reuse_page(skb, ring->frag_num++, ring, ring->pull_len,
2444 ring_ptr_move_fw(ring, next_to_clean);
2446 return HNS3_NEED_ADD_FRAG;
2449 static int hns3_add_frag(struct hns3_enet_ring *ring, struct hns3_desc *desc,
2450 struct sk_buff **out_skb, bool pending)
2452 struct sk_buff *skb = *out_skb;
2453 struct sk_buff *head_skb = *out_skb;
2454 struct sk_buff *new_skb;
2455 struct hns3_desc_cb *desc_cb;
2456 struct hns3_desc *pre_desc;
2460 /* if there is pending bd, the SW param next_to_clean has moved
2461 * to next and the next is NULL
2464 pre_bd = (ring->next_to_clean - 1 + ring->desc_num) %
2466 pre_desc = &ring->desc[pre_bd];
2467 bd_base_info = le32_to_cpu(pre_desc->rx.bd_base_info);
2469 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2472 while (!(bd_base_info & BIT(HNS3_RXD_FE_B))) {
2473 desc = &ring->desc[ring->next_to_clean];
2474 desc_cb = &ring->desc_cb[ring->next_to_clean];
2475 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2476 /* make sure HW write desc complete */
2478 if (!(bd_base_info & BIT(HNS3_RXD_VLD_B)))
2481 if (unlikely(ring->frag_num >= MAX_SKB_FRAGS)) {
2482 new_skb = napi_alloc_skb(&ring->tqp_vector->napi,
2484 if (unlikely(!new_skb)) {
2485 netdev_err(ring->tqp->handle->kinfo.netdev,
2486 "alloc rx skb frag fail\n");
2491 if (ring->tail_skb) {
2492 ring->tail_skb->next = new_skb;
2493 ring->tail_skb = new_skb;
2495 skb_shinfo(skb)->frag_list = new_skb;
2496 ring->tail_skb = new_skb;
2500 if (ring->tail_skb) {
2501 head_skb->truesize += hnae3_buf_size(ring);
2502 head_skb->data_len += le16_to_cpu(desc->rx.size);
2503 head_skb->len += le16_to_cpu(desc->rx.size);
2504 skb = ring->tail_skb;
2507 hns3_nic_reuse_page(skb, ring->frag_num++, ring, 0, desc_cb);
2508 ring_ptr_move_fw(ring, next_to_clean);
2509 ring->pending_buf++;
2515 static void hns3_set_gro_param(struct sk_buff *skb, u32 l234info,
2521 gro_count = hnae3_get_field(l234info, HNS3_RXD_GRO_COUNT_M,
2522 HNS3_RXD_GRO_COUNT_S);
2523 /* if there is no HW GRO, do not set gro params */
2527 /* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
2528 * to skb_shinfo(skb)->gso_segs
2530 NAPI_GRO_CB(skb)->count = gro_count;
2532 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
2534 if (l3_type == HNS3_L3_TYPE_IPV4)
2535 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2536 else if (l3_type == HNS3_L3_TYPE_IPV6)
2537 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
2541 skb_shinfo(skb)->gso_size = hnae3_get_field(bd_base_info,
2542 HNS3_RXD_GRO_SIZE_M,
2543 HNS3_RXD_GRO_SIZE_S);
2544 if (skb_shinfo(skb)->gso_size)
2545 tcp_gro_complete(skb);
2548 static void hns3_set_rx_skb_rss_type(struct hns3_enet_ring *ring,
2549 struct sk_buff *skb)
2551 struct hnae3_handle *handle = ring->tqp->handle;
2552 enum pkt_hash_types rss_type;
2553 struct hns3_desc *desc;
2556 /* When driver handle the rss type, ring->next_to_clean indicates the
2557 * first descriptor of next packet, need -1 here.
2559 last_bd = (ring->next_to_clean - 1 + ring->desc_num) % ring->desc_num;
2560 desc = &ring->desc[last_bd];
2562 if (le32_to_cpu(desc->rx.rss_hash))
2563 rss_type = handle->kinfo.rss_type;
2565 rss_type = PKT_HASH_TYPE_NONE;
2567 skb_set_hash(skb, le32_to_cpu(desc->rx.rss_hash), rss_type);
2570 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring,
2571 struct sk_buff **out_skb)
2573 struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2574 enum hns3_pkt_l2t_type l2_frame_type;
2575 struct sk_buff *skb = ring->skb;
2576 struct hns3_desc_cb *desc_cb;
2577 struct hns3_desc *desc;
2583 desc = &ring->desc[ring->next_to_clean];
2584 desc_cb = &ring->desc_cb[ring->next_to_clean];
2588 length = le16_to_cpu(desc->rx.size);
2589 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2591 /* Check valid BD */
2592 if (unlikely(!(bd_base_info & BIT(HNS3_RXD_VLD_B))))
2596 ring->va = (unsigned char *)desc_cb->buf + desc_cb->page_offset;
2598 /* Prefetch first cache line of first page
2599 * Idea is to cache few bytes of the header of the packet. Our L1 Cache
2600 * line size is 64B so need to prefetch twice to make it 128B. But in
2601 * actual we can have greater size of caches with 128B Level 1 cache
2602 * lines. In such a case, single fetch would suffice to cache in the
2603 * relevant part of the header.
2606 #if L1_CACHE_BYTES < 128
2607 prefetch(ring->va + L1_CACHE_BYTES);
2611 ret = hns3_alloc_skb(ring, length, ring->va);
2612 *out_skb = skb = ring->skb;
2614 if (ret < 0) /* alloc buffer fail */
2616 if (ret > 0) { /* need add frag */
2617 ret = hns3_add_frag(ring, desc, &skb, false);
2621 /* As the head data may be changed when GRO enable, copy
2622 * the head data in after other data rx completed
2624 memcpy(skb->data, ring->va,
2625 ALIGN(ring->pull_len, sizeof(long)));
2628 ret = hns3_add_frag(ring, desc, &skb, true);
2632 /* As the head data may be changed when GRO enable, copy
2633 * the head data in after other data rx completed
2635 memcpy(skb->data, ring->va,
2636 ALIGN(ring->pull_len, sizeof(long)));
2639 l234info = le32_to_cpu(desc->rx.l234_info);
2640 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2642 /* Based on hw strategy, the tag offloaded will be stored at
2643 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
2644 * in one layer tag case.
2646 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
2649 if (hns3_parse_vlan_tag(ring, desc, l234info, &vlan_tag))
2650 __vlan_hwaccel_put_tag(skb,
2655 if (unlikely(!(bd_base_info & BIT(HNS3_RXD_VLD_B)))) {
2656 u64_stats_update_begin(&ring->syncp);
2657 ring->stats.non_vld_descs++;
2658 u64_stats_update_end(&ring->syncp);
2660 dev_kfree_skb_any(skb);
2664 if (unlikely((!desc->rx.pkt_len) ||
2665 (l234info & (BIT(HNS3_RXD_TRUNCAT_B) |
2666 BIT(HNS3_RXD_L2E_B))))) {
2667 u64_stats_update_begin(&ring->syncp);
2668 if (l234info & BIT(HNS3_RXD_L2E_B))
2669 ring->stats.l2_err++;
2671 ring->stats.err_pkt_len++;
2672 u64_stats_update_end(&ring->syncp);
2674 dev_kfree_skb_any(skb);
2679 l2_frame_type = hnae3_get_field(l234info, HNS3_RXD_DMAC_M,
2681 u64_stats_update_begin(&ring->syncp);
2682 if (l2_frame_type == HNS3_L2_TYPE_MULTICAST)
2683 ring->stats.rx_multicast++;
2685 ring->stats.rx_pkts++;
2686 ring->stats.rx_bytes += skb->len;
2687 u64_stats_update_end(&ring->syncp);
2689 ring->tqp_vector->rx_group.total_bytes += skb->len;
2691 /* This is needed in order to enable forwarding support */
2692 hns3_set_gro_param(skb, l234info, bd_base_info);
2694 hns3_rx_checksum(ring, skb, desc);
2696 hns3_set_rx_skb_rss_type(ring, skb);
2701 int hns3_clean_rx_ring(
2702 struct hns3_enet_ring *ring, int budget,
2703 void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
2705 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
2706 struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2707 int recv_pkts, recv_bds, clean_count, err;
2708 int unused_count = hns3_desc_unused(ring) - ring->pending_buf;
2709 struct sk_buff *skb = ring->skb;
2712 num = readl_relaxed(ring->tqp->io_base + HNS3_RING_RX_RING_FBDNUM_REG);
2713 rmb(); /* Make sure num taken effect before the other data is touched */
2715 recv_pkts = 0, recv_bds = 0, clean_count = 0;
2716 num -= unused_count;
2718 while (recv_pkts < budget && recv_bds < num) {
2719 /* Reuse or realloc buffers */
2720 if (clean_count + unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
2721 hns3_nic_alloc_rx_buffers(ring,
2722 clean_count + unused_count);
2724 unused_count = hns3_desc_unused(ring) -
2729 err = hns3_handle_rx_bd(ring, &skb);
2730 if (unlikely(!skb)) /* This fault cannot be repaired */
2733 if (err == -ENXIO) { /* Do not get FE for the packet */
2735 } else if (unlikely(err)) { /* Do jump the err */
2736 recv_bds += ring->pending_buf;
2737 clean_count += ring->pending_buf;
2739 ring->pending_buf = 0;
2743 /* Do update ip stack process */
2744 skb->protocol = eth_type_trans(skb, netdev);
2746 recv_bds += ring->pending_buf;
2747 clean_count += ring->pending_buf;
2749 ring->pending_buf = 0;
2755 /* Make all data has been write before submit */
2756 if (clean_count + unused_count > 0)
2757 hns3_nic_alloc_rx_buffers(ring,
2758 clean_count + unused_count);
2763 static bool hns3_get_new_int_gl(struct hns3_enet_ring_group *ring_group)
2765 struct hns3_enet_tqp_vector *tqp_vector =
2766 ring_group->ring->tqp_vector;
2767 enum hns3_flow_level_range new_flow_level;
2768 int packets_per_msecs;
2769 int bytes_per_msecs;
2773 if (!tqp_vector->last_jiffies)
2776 if (ring_group->total_packets == 0) {
2777 ring_group->coal.int_gl = HNS3_INT_GL_50K;
2778 ring_group->coal.flow_level = HNS3_FLOW_LOW;
2782 /* Simple throttlerate management
2783 * 0-10MB/s lower (50000 ints/s)
2784 * 10-20MB/s middle (20000 ints/s)
2785 * 20-1249MB/s high (18000 ints/s)
2786 * > 40000pps ultra (8000 ints/s)
2788 new_flow_level = ring_group->coal.flow_level;
2789 new_int_gl = ring_group->coal.int_gl;
2791 jiffies_to_msecs(jiffies - tqp_vector->last_jiffies);
2793 if (!time_passed_ms)
2796 do_div(ring_group->total_packets, time_passed_ms);
2797 packets_per_msecs = ring_group->total_packets;
2799 do_div(ring_group->total_bytes, time_passed_ms);
2800 bytes_per_msecs = ring_group->total_bytes;
2802 #define HNS3_RX_LOW_BYTE_RATE 10000
2803 #define HNS3_RX_MID_BYTE_RATE 20000
2805 switch (new_flow_level) {
2807 if (bytes_per_msecs > HNS3_RX_LOW_BYTE_RATE)
2808 new_flow_level = HNS3_FLOW_MID;
2811 if (bytes_per_msecs > HNS3_RX_MID_BYTE_RATE)
2812 new_flow_level = HNS3_FLOW_HIGH;
2813 else if (bytes_per_msecs <= HNS3_RX_LOW_BYTE_RATE)
2814 new_flow_level = HNS3_FLOW_LOW;
2816 case HNS3_FLOW_HIGH:
2817 case HNS3_FLOW_ULTRA:
2819 if (bytes_per_msecs <= HNS3_RX_MID_BYTE_RATE)
2820 new_flow_level = HNS3_FLOW_MID;
2824 #define HNS3_RX_ULTRA_PACKET_RATE 40
2826 if (packets_per_msecs > HNS3_RX_ULTRA_PACKET_RATE &&
2827 &tqp_vector->rx_group == ring_group)
2828 new_flow_level = HNS3_FLOW_ULTRA;
2830 switch (new_flow_level) {
2832 new_int_gl = HNS3_INT_GL_50K;
2835 new_int_gl = HNS3_INT_GL_20K;
2837 case HNS3_FLOW_HIGH:
2838 new_int_gl = HNS3_INT_GL_18K;
2840 case HNS3_FLOW_ULTRA:
2841 new_int_gl = HNS3_INT_GL_8K;
2847 ring_group->total_bytes = 0;
2848 ring_group->total_packets = 0;
2849 ring_group->coal.flow_level = new_flow_level;
2850 if (new_int_gl != ring_group->coal.int_gl) {
2851 ring_group->coal.int_gl = new_int_gl;
2857 static void hns3_update_new_int_gl(struct hns3_enet_tqp_vector *tqp_vector)
2859 struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
2860 struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
2861 bool rx_update, tx_update;
2863 /* update param every 1000ms */
2864 if (time_before(jiffies,
2865 tqp_vector->last_jiffies + msecs_to_jiffies(1000)))
2868 if (rx_group->coal.gl_adapt_enable) {
2869 rx_update = hns3_get_new_int_gl(rx_group);
2871 hns3_set_vector_coalesce_rx_gl(tqp_vector,
2872 rx_group->coal.int_gl);
2875 if (tx_group->coal.gl_adapt_enable) {
2876 tx_update = hns3_get_new_int_gl(tx_group);
2878 hns3_set_vector_coalesce_tx_gl(tqp_vector,
2879 tx_group->coal.int_gl);
2882 tqp_vector->last_jiffies = jiffies;
2885 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
2887 struct hns3_nic_priv *priv = netdev_priv(napi->dev);
2888 struct hns3_enet_ring *ring;
2889 int rx_pkt_total = 0;
2891 struct hns3_enet_tqp_vector *tqp_vector =
2892 container_of(napi, struct hns3_enet_tqp_vector, napi);
2893 bool clean_complete = true;
2896 if (unlikely(test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
2897 napi_complete(napi);
2901 /* Since the actual Tx work is minimal, we can give the Tx a larger
2902 * budget and be more aggressive about cleaning up the Tx descriptors.
2904 hns3_for_each_ring(ring, tqp_vector->tx_group)
2905 hns3_clean_tx_ring(ring);
2907 /* make sure rx ring budget not smaller than 1 */
2908 rx_budget = max(budget / tqp_vector->num_tqps, 1);
2910 hns3_for_each_ring(ring, tqp_vector->rx_group) {
2911 int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
2914 if (rx_cleaned >= rx_budget)
2915 clean_complete = false;
2917 rx_pkt_total += rx_cleaned;
2920 tqp_vector->rx_group.total_packets += rx_pkt_total;
2922 if (!clean_complete)
2925 if (napi_complete(napi) &&
2926 likely(!test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
2927 hns3_update_new_int_gl(tqp_vector);
2928 hns3_mask_vector_irq(tqp_vector, 1);
2931 return rx_pkt_total;
2934 static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
2935 struct hnae3_ring_chain_node *head)
2937 struct pci_dev *pdev = tqp_vector->handle->pdev;
2938 struct hnae3_ring_chain_node *cur_chain = head;
2939 struct hnae3_ring_chain_node *chain;
2940 struct hns3_enet_ring *tx_ring;
2941 struct hns3_enet_ring *rx_ring;
2943 tx_ring = tqp_vector->tx_group.ring;
2945 cur_chain->tqp_index = tx_ring->tqp->tqp_index;
2946 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
2947 HNAE3_RING_TYPE_TX);
2948 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
2949 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_TX);
2951 cur_chain->next = NULL;
2953 while (tx_ring->next) {
2954 tx_ring = tx_ring->next;
2956 chain = devm_kzalloc(&pdev->dev, sizeof(*chain),
2959 goto err_free_chain;
2961 cur_chain->next = chain;
2962 chain->tqp_index = tx_ring->tqp->tqp_index;
2963 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
2964 HNAE3_RING_TYPE_TX);
2965 hnae3_set_field(chain->int_gl_idx,
2966 HNAE3_RING_GL_IDX_M,
2967 HNAE3_RING_GL_IDX_S,
2974 rx_ring = tqp_vector->rx_group.ring;
2975 if (!tx_ring && rx_ring) {
2976 cur_chain->next = NULL;
2977 cur_chain->tqp_index = rx_ring->tqp->tqp_index;
2978 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
2979 HNAE3_RING_TYPE_RX);
2980 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
2981 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
2983 rx_ring = rx_ring->next;
2987 chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
2989 goto err_free_chain;
2991 cur_chain->next = chain;
2992 chain->tqp_index = rx_ring->tqp->tqp_index;
2993 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
2994 HNAE3_RING_TYPE_RX);
2995 hnae3_set_field(chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
2996 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
3000 rx_ring = rx_ring->next;
3006 cur_chain = head->next;
3008 chain = cur_chain->next;
3009 devm_kfree(&pdev->dev, cur_chain);
3017 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
3018 struct hnae3_ring_chain_node *head)
3020 struct pci_dev *pdev = tqp_vector->handle->pdev;
3021 struct hnae3_ring_chain_node *chain_tmp, *chain;
3026 chain_tmp = chain->next;
3027 devm_kfree(&pdev->dev, chain);
3032 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
3033 struct hns3_enet_ring *ring)
3035 ring->next = group->ring;
3041 static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
3043 struct pci_dev *pdev = priv->ae_handle->pdev;
3044 struct hns3_enet_tqp_vector *tqp_vector;
3045 int num_vectors = priv->vector_num;
3049 numa_node = dev_to_node(&pdev->dev);
3051 for (vector_i = 0; vector_i < num_vectors; vector_i++) {
3052 tqp_vector = &priv->tqp_vector[vector_i];
3053 cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
3054 &tqp_vector->affinity_mask);
3058 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
3060 struct hnae3_ring_chain_node vector_ring_chain;
3061 struct hnae3_handle *h = priv->ae_handle;
3062 struct hns3_enet_tqp_vector *tqp_vector;
3066 hns3_nic_set_cpumask(priv);
3068 for (i = 0; i < priv->vector_num; i++) {
3069 tqp_vector = &priv->tqp_vector[i];
3070 hns3_vector_gl_rl_init_hw(tqp_vector, priv);
3071 tqp_vector->num_tqps = 0;
3074 for (i = 0; i < h->kinfo.num_tqps; i++) {
3075 u16 vector_i = i % priv->vector_num;
3076 u16 tqp_num = h->kinfo.num_tqps;
3078 tqp_vector = &priv->tqp_vector[vector_i];
3080 hns3_add_ring_to_group(&tqp_vector->tx_group,
3081 priv->ring_data[i].ring);
3083 hns3_add_ring_to_group(&tqp_vector->rx_group,
3084 priv->ring_data[i + tqp_num].ring);
3086 priv->ring_data[i].ring->tqp_vector = tqp_vector;
3087 priv->ring_data[i + tqp_num].ring->tqp_vector = tqp_vector;
3088 tqp_vector->num_tqps++;
3091 for (i = 0; i < priv->vector_num; i++) {
3092 tqp_vector = &priv->tqp_vector[i];
3094 tqp_vector->rx_group.total_bytes = 0;
3095 tqp_vector->rx_group.total_packets = 0;
3096 tqp_vector->tx_group.total_bytes = 0;
3097 tqp_vector->tx_group.total_packets = 0;
3098 tqp_vector->handle = h;
3100 ret = hns3_get_vector_ring_chain(tqp_vector,
3101 &vector_ring_chain);
3105 ret = h->ae_algo->ops->map_ring_to_vector(h,
3106 tqp_vector->vector_irq, &vector_ring_chain);
3108 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3113 netif_napi_add(priv->netdev, &tqp_vector->napi,
3114 hns3_nic_common_poll, NAPI_POLL_WEIGHT);
3121 netif_napi_del(&priv->tqp_vector[i].napi);
3126 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
3128 #define HNS3_VECTOR_PF_MAX_NUM 64
3130 struct hnae3_handle *h = priv->ae_handle;
3131 struct hns3_enet_tqp_vector *tqp_vector;
3132 struct hnae3_vector_info *vector;
3133 struct pci_dev *pdev = h->pdev;
3134 u16 tqp_num = h->kinfo.num_tqps;
3139 /* RSS size, cpu online and vector_num should be the same */
3140 /* Should consider 2p/4p later */
3141 vector_num = min_t(u16, num_online_cpus(), tqp_num);
3142 vector_num = min_t(u16, vector_num, HNS3_VECTOR_PF_MAX_NUM);
3144 vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
3149 vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
3151 priv->vector_num = vector_num;
3152 priv->tqp_vector = (struct hns3_enet_tqp_vector *)
3153 devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
3155 if (!priv->tqp_vector) {
3160 for (i = 0; i < priv->vector_num; i++) {
3161 tqp_vector = &priv->tqp_vector[i];
3162 tqp_vector->idx = i;
3163 tqp_vector->mask_addr = vector[i].io_addr;
3164 tqp_vector->vector_irq = vector[i].vector;
3165 hns3_vector_gl_rl_init(tqp_vector, priv);
3169 devm_kfree(&pdev->dev, vector);
3173 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
3179 static void hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
3181 struct hnae3_ring_chain_node vector_ring_chain;
3182 struct hnae3_handle *h = priv->ae_handle;
3183 struct hns3_enet_tqp_vector *tqp_vector;
3186 for (i = 0; i < priv->vector_num; i++) {
3187 tqp_vector = &priv->tqp_vector[i];
3189 if (!tqp_vector->rx_group.ring && !tqp_vector->tx_group.ring)
3192 hns3_get_vector_ring_chain(tqp_vector, &vector_ring_chain);
3194 h->ae_algo->ops->unmap_ring_from_vector(h,
3195 tqp_vector->vector_irq, &vector_ring_chain);
3197 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3199 if (tqp_vector->irq_init_flag == HNS3_VECTOR_INITED) {
3200 irq_set_affinity_notifier(tqp_vector->vector_irq,
3202 irq_set_affinity_hint(tqp_vector->vector_irq, NULL);
3203 free_irq(tqp_vector->vector_irq, tqp_vector);
3204 tqp_vector->irq_init_flag = HNS3_VECTOR_NOT_INITED;
3207 hns3_clear_ring_group(&tqp_vector->rx_group);
3208 hns3_clear_ring_group(&tqp_vector->tx_group);
3209 netif_napi_del(&priv->tqp_vector[i].napi);
3213 static int hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
3215 struct hnae3_handle *h = priv->ae_handle;
3216 struct pci_dev *pdev = h->pdev;
3219 for (i = 0; i < priv->vector_num; i++) {
3220 struct hns3_enet_tqp_vector *tqp_vector;
3222 tqp_vector = &priv->tqp_vector[i];
3223 ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
3228 devm_kfree(&pdev->dev, priv->tqp_vector);
3232 static int hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
3235 struct hns3_nic_ring_data *ring_data = priv->ring_data;
3236 int queue_num = priv->ae_handle->kinfo.num_tqps;
3237 struct pci_dev *pdev = priv->ae_handle->pdev;
3238 struct hns3_enet_ring *ring;
3241 ring = devm_kzalloc(&pdev->dev, sizeof(*ring), GFP_KERNEL);
3245 if (ring_type == HNAE3_RING_TYPE_TX) {
3246 desc_num = priv->ae_handle->kinfo.num_tx_desc;
3247 ring_data[q->tqp_index].ring = ring;
3248 ring_data[q->tqp_index].queue_index = q->tqp_index;
3249 ring->io_base = (u8 __iomem *)q->io_base + HNS3_TX_REG_OFFSET;
3251 desc_num = priv->ae_handle->kinfo.num_rx_desc;
3252 ring_data[q->tqp_index + queue_num].ring = ring;
3253 ring_data[q->tqp_index + queue_num].queue_index = q->tqp_index;
3254 ring->io_base = q->io_base;
3257 hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
3261 ring->desc_cb = NULL;
3262 ring->dev = priv->dev;
3263 ring->desc_dma_addr = 0;
3264 ring->buf_size = q->buf_size;
3265 ring->desc_num = desc_num;
3266 ring->next_to_use = 0;
3267 ring->next_to_clean = 0;
3272 static int hns3_queue_to_ring(struct hnae3_queue *tqp,
3273 struct hns3_nic_priv *priv)
3277 ret = hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
3281 ret = hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
3283 devm_kfree(priv->dev, priv->ring_data[tqp->tqp_index].ring);
3290 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
3292 struct hnae3_handle *h = priv->ae_handle;
3293 struct pci_dev *pdev = h->pdev;
3296 priv->ring_data = devm_kzalloc(&pdev->dev,
3297 array3_size(h->kinfo.num_tqps,
3298 sizeof(*priv->ring_data),
3301 if (!priv->ring_data)
3304 for (i = 0; i < h->kinfo.num_tqps; i++) {
3305 ret = hns3_queue_to_ring(h->kinfo.tqp[i], priv);
3313 devm_kfree(priv->dev, priv->ring_data[i].ring);
3314 devm_kfree(priv->dev,
3315 priv->ring_data[i + h->kinfo.num_tqps].ring);
3318 devm_kfree(&pdev->dev, priv->ring_data);
3322 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
3324 struct hnae3_handle *h = priv->ae_handle;
3327 for (i = 0; i < h->kinfo.num_tqps; i++) {
3328 devm_kfree(priv->dev, priv->ring_data[i].ring);
3329 devm_kfree(priv->dev,
3330 priv->ring_data[i + h->kinfo.num_tqps].ring);
3332 devm_kfree(priv->dev, priv->ring_data);
3335 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
3339 if (ring->desc_num <= 0 || ring->buf_size <= 0)
3342 ring->desc_cb = kcalloc(ring->desc_num, sizeof(ring->desc_cb[0]),
3344 if (!ring->desc_cb) {
3349 ret = hns3_alloc_desc(ring);
3351 goto out_with_desc_cb;
3353 if (!HNAE3_IS_TX_RING(ring)) {
3354 ret = hns3_alloc_ring_buffers(ring);
3362 hns3_free_desc(ring);
3364 kfree(ring->desc_cb);
3365 ring->desc_cb = NULL;
3370 static void hns3_fini_ring(struct hns3_enet_ring *ring)
3372 hns3_free_desc(ring);
3373 kfree(ring->desc_cb);
3374 ring->desc_cb = NULL;
3375 ring->next_to_clean = 0;
3376 ring->next_to_use = 0;
3377 ring->pending_buf = 0;
3379 dev_kfree_skb_any(ring->skb);
3384 static int hns3_buf_size2type(u32 buf_size)
3390 bd_size_type = HNS3_BD_SIZE_512_TYPE;
3393 bd_size_type = HNS3_BD_SIZE_1024_TYPE;
3396 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
3399 bd_size_type = HNS3_BD_SIZE_4096_TYPE;
3402 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
3405 return bd_size_type;
3408 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
3410 dma_addr_t dma = ring->desc_dma_addr;
3411 struct hnae3_queue *q = ring->tqp;
3413 if (!HNAE3_IS_TX_RING(ring)) {
3414 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG,
3416 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
3417 (u32)((dma >> 31) >> 1));
3419 hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
3420 hns3_buf_size2type(ring->buf_size));
3421 hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
3422 ring->desc_num / 8 - 1);
3425 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
3427 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
3428 (u32)((dma >> 31) >> 1));
3430 hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
3431 ring->desc_num / 8 - 1);
3435 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
3437 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
3440 for (i = 0; i < HNAE3_MAX_TC; i++) {
3441 struct hnae3_tc_info *tc_info = &kinfo->tc_info[i];
3444 if (!tc_info->enable)
3447 for (j = 0; j < tc_info->tqp_count; j++) {
3448 struct hnae3_queue *q;
3450 q = priv->ring_data[tc_info->tqp_offset + j].ring->tqp;
3451 hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG,
3457 int hns3_init_all_ring(struct hns3_nic_priv *priv)
3459 struct hnae3_handle *h = priv->ae_handle;
3460 int ring_num = h->kinfo.num_tqps * 2;
3464 for (i = 0; i < ring_num; i++) {
3465 ret = hns3_alloc_ring_memory(priv->ring_data[i].ring);
3468 "Alloc ring memory fail! ret=%d\n", ret);
3469 goto out_when_alloc_ring_memory;
3472 u64_stats_init(&priv->ring_data[i].ring->syncp);
3477 out_when_alloc_ring_memory:
3478 for (j = i - 1; j >= 0; j--)
3479 hns3_fini_ring(priv->ring_data[j].ring);
3484 int hns3_uninit_all_ring(struct hns3_nic_priv *priv)
3486 struct hnae3_handle *h = priv->ae_handle;
3489 for (i = 0; i < h->kinfo.num_tqps; i++) {
3490 hns3_fini_ring(priv->ring_data[i].ring);
3491 hns3_fini_ring(priv->ring_data[i + h->kinfo.num_tqps].ring);
3496 /* Set mac addr if it is configured. or leave it to the AE driver */
3497 static int hns3_init_mac_addr(struct net_device *netdev, bool init)
3499 struct hns3_nic_priv *priv = netdev_priv(netdev);
3500 struct hnae3_handle *h = priv->ae_handle;
3501 u8 mac_addr_temp[ETH_ALEN];
3504 if (h->ae_algo->ops->get_mac_addr && init) {
3505 h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
3506 ether_addr_copy(netdev->dev_addr, mac_addr_temp);
3509 /* Check if the MAC address is valid, if not get a random one */
3510 if (!is_valid_ether_addr(netdev->dev_addr)) {
3511 eth_hw_addr_random(netdev);
3512 dev_warn(priv->dev, "using random MAC address %pM\n",
3516 if (h->ae_algo->ops->set_mac_addr)
3517 ret = h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
3522 static int hns3_init_phy(struct net_device *netdev)
3524 struct hnae3_handle *h = hns3_get_handle(netdev);
3527 if (h->ae_algo->ops->mac_connect_phy)
3528 ret = h->ae_algo->ops->mac_connect_phy(h);
3533 static void hns3_uninit_phy(struct net_device *netdev)
3535 struct hnae3_handle *h = hns3_get_handle(netdev);
3537 if (h->ae_algo->ops->mac_disconnect_phy)
3538 h->ae_algo->ops->mac_disconnect_phy(h);
3541 static int hns3_restore_fd_rules(struct net_device *netdev)
3543 struct hnae3_handle *h = hns3_get_handle(netdev);
3546 if (h->ae_algo->ops->restore_fd_rules)
3547 ret = h->ae_algo->ops->restore_fd_rules(h);
3552 static void hns3_del_all_fd_rules(struct net_device *netdev, bool clear_list)
3554 struct hnae3_handle *h = hns3_get_handle(netdev);
3556 if (h->ae_algo->ops->del_all_fd_entries)
3557 h->ae_algo->ops->del_all_fd_entries(h, clear_list);
3560 static void hns3_nic_set_priv_ops(struct net_device *netdev)
3562 struct hns3_nic_priv *priv = netdev_priv(netdev);
3564 if ((netdev->features & NETIF_F_TSO) ||
3565 (netdev->features & NETIF_F_TSO6))
3566 priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tso;
3568 priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tx;
3571 static int hns3_client_start(struct hnae3_handle *handle)
3573 if (!handle->ae_algo->ops->client_start)
3576 return handle->ae_algo->ops->client_start(handle);
3579 static void hns3_client_stop(struct hnae3_handle *handle)
3581 if (!handle->ae_algo->ops->client_stop)
3584 handle->ae_algo->ops->client_stop(handle);
3587 static int hns3_client_init(struct hnae3_handle *handle)
3589 struct pci_dev *pdev = handle->pdev;
3590 u16 alloc_tqps, max_rss_size;
3591 struct hns3_nic_priv *priv;
3592 struct net_device *netdev;
3595 handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
3597 netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
3601 priv = netdev_priv(netdev);
3602 priv->dev = &pdev->dev;
3603 priv->netdev = netdev;
3604 priv->ae_handle = handle;
3605 priv->tx_timeout_count = 0;
3606 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
3608 handle->kinfo.netdev = netdev;
3609 handle->priv = (void *)priv;
3611 hns3_init_mac_addr(netdev, true);
3613 hns3_set_default_feature(netdev);
3615 netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
3616 netdev->priv_flags |= IFF_UNICAST_FLT;
3617 netdev->netdev_ops = &hns3_nic_netdev_ops;
3618 SET_NETDEV_DEV(netdev, &pdev->dev);
3619 hns3_ethtool_set_ops(netdev);
3620 hns3_nic_set_priv_ops(netdev);
3622 /* Carrier off reporting is important to ethtool even BEFORE open */
3623 netif_carrier_off(netdev);
3625 ret = hns3_get_ring_config(priv);
3628 goto out_get_ring_cfg;
3631 ret = hns3_nic_alloc_vector_data(priv);
3634 goto out_alloc_vector_data;
3637 ret = hns3_nic_init_vector_data(priv);
3640 goto out_init_vector_data;
3643 ret = hns3_init_all_ring(priv);
3646 goto out_init_ring_data;
3649 ret = hns3_init_phy(netdev);
3653 ret = register_netdev(netdev);
3655 dev_err(priv->dev, "probe register netdev fail!\n");
3656 goto out_reg_netdev_fail;
3659 ret = hns3_client_start(handle);
3661 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
3662 goto out_client_start;
3665 hns3_dcbnl_setup(handle);
3667 hns3_dbg_init(handle);
3669 /* MTU range: (ETH_MIN_MTU(kernel default) - 9702) */
3670 netdev->max_mtu = HNS3_MAX_MTU;
3672 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
3677 unregister_netdev(netdev);
3678 out_reg_netdev_fail:
3679 hns3_uninit_phy(netdev);
3681 hns3_uninit_all_ring(priv);
3683 hns3_nic_uninit_vector_data(priv);
3684 out_init_vector_data:
3685 hns3_nic_dealloc_vector_data(priv);
3686 out_alloc_vector_data:
3687 priv->ring_data = NULL;
3689 priv->ae_handle = NULL;
3690 free_netdev(netdev);
3694 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
3696 struct net_device *netdev = handle->kinfo.netdev;
3697 struct hns3_nic_priv *priv = netdev_priv(netdev);
3700 hns3_client_stop(handle);
3702 hns3_remove_hw_addr(netdev);
3704 if (netdev->reg_state != NETREG_UNINITIALIZED)
3705 unregister_netdev(netdev);
3707 if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
3708 netdev_warn(netdev, "already uninitialized\n");
3709 goto out_netdev_free;
3712 hns3_del_all_fd_rules(netdev, true);
3714 hns3_force_clear_all_rx_ring(handle);
3716 hns3_uninit_phy(netdev);
3718 hns3_nic_uninit_vector_data(priv);
3720 ret = hns3_nic_dealloc_vector_data(priv);
3722 netdev_err(netdev, "dealloc vector error\n");
3724 ret = hns3_uninit_all_ring(priv);
3726 netdev_err(netdev, "uninit ring error\n");
3728 hns3_put_ring_config(priv);
3730 hns3_dbg_uninit(handle);
3732 priv->ring_data = NULL;
3735 free_netdev(netdev);
3738 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
3740 struct net_device *netdev = handle->kinfo.netdev;
3746 netif_carrier_on(netdev);
3747 netif_tx_wake_all_queues(netdev);
3748 netdev_info(netdev, "link up\n");
3750 netif_carrier_off(netdev);
3751 netif_tx_stop_all_queues(netdev);
3752 netdev_info(netdev, "link down\n");
3756 static int hns3_client_setup_tc(struct hnae3_handle *handle, u8 tc)
3758 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
3759 struct net_device *ndev = kinfo->netdev;
3761 if (tc > HNAE3_MAX_TC)
3767 return hns3_nic_set_real_num_queue(ndev);
3770 static int hns3_recover_hw_addr(struct net_device *ndev)
3772 struct netdev_hw_addr_list *list;
3773 struct netdev_hw_addr *ha, *tmp;
3776 /* go through and sync uc_addr entries to the device */
3778 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3779 ret = hns3_nic_uc_sync(ndev, ha->addr);
3784 /* go through and sync mc_addr entries to the device */
3786 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3787 ret = hns3_nic_mc_sync(ndev, ha->addr);
3795 static void hns3_remove_hw_addr(struct net_device *netdev)
3797 struct netdev_hw_addr_list *list;
3798 struct netdev_hw_addr *ha, *tmp;
3800 hns3_nic_uc_unsync(netdev, netdev->dev_addr);
3802 /* go through and unsync uc_addr entries to the device */
3804 list_for_each_entry_safe(ha, tmp, &list->list, list)
3805 hns3_nic_uc_unsync(netdev, ha->addr);
3807 /* go through and unsync mc_addr entries to the device */
3809 list_for_each_entry_safe(ha, tmp, &list->list, list)
3810 if (ha->refcount > 1)
3811 hns3_nic_mc_unsync(netdev, ha->addr);
3814 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
3816 while (ring->next_to_clean != ring->next_to_use) {
3817 ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
3818 hns3_free_buffer_detach(ring, ring->next_to_clean);
3819 ring_ptr_move_fw(ring, next_to_clean);
3823 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
3825 struct hns3_desc_cb res_cbs;
3828 while (ring->next_to_use != ring->next_to_clean) {
3829 /* When a buffer is not reused, it's memory has been
3830 * freed in hns3_handle_rx_bd or will be freed by
3831 * stack, so we need to replace the buffer here.
3833 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
3834 ret = hns3_reserve_buffer_map(ring, &res_cbs);
3836 u64_stats_update_begin(&ring->syncp);
3837 ring->stats.sw_err_cnt++;
3838 u64_stats_update_end(&ring->syncp);
3839 /* if alloc new buffer fail, exit directly
3840 * and reclear in up flow.
3842 netdev_warn(ring->tqp->handle->kinfo.netdev,
3843 "reserve buffer map failed, ret = %d\n",
3847 hns3_replace_buffer(ring, ring->next_to_use,
3850 ring_ptr_move_fw(ring, next_to_use);
3856 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
3858 while (ring->next_to_use != ring->next_to_clean) {
3859 /* When a buffer is not reused, it's memory has been
3860 * freed in hns3_handle_rx_bd or will be freed by
3861 * stack, so only need to unmap the buffer here.
3863 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
3864 hns3_unmap_buffer(ring,
3865 &ring->desc_cb[ring->next_to_use]);
3866 ring->desc_cb[ring->next_to_use].dma = 0;
3869 ring_ptr_move_fw(ring, next_to_use);
3873 static void hns3_force_clear_all_rx_ring(struct hnae3_handle *h)
3875 struct net_device *ndev = h->kinfo.netdev;
3876 struct hns3_nic_priv *priv = netdev_priv(ndev);
3877 struct hns3_enet_ring *ring;
3880 for (i = 0; i < h->kinfo.num_tqps; i++) {
3881 ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
3882 hns3_force_clear_rx_ring(ring);
3886 static void hns3_clear_all_ring(struct hnae3_handle *h)
3888 struct net_device *ndev = h->kinfo.netdev;
3889 struct hns3_nic_priv *priv = netdev_priv(ndev);
3892 for (i = 0; i < h->kinfo.num_tqps; i++) {
3893 struct netdev_queue *dev_queue;
3894 struct hns3_enet_ring *ring;
3896 ring = priv->ring_data[i].ring;
3897 hns3_clear_tx_ring(ring);
3898 dev_queue = netdev_get_tx_queue(ndev,
3899 priv->ring_data[i].queue_index);
3900 netdev_tx_reset_queue(dev_queue);
3902 ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
3903 /* Continue to clear other rings even if clearing some
3906 hns3_clear_rx_ring(ring);
3910 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
3912 struct net_device *ndev = h->kinfo.netdev;
3913 struct hns3_nic_priv *priv = netdev_priv(ndev);
3914 struct hns3_enet_ring *rx_ring;
3918 for (i = 0; i < h->kinfo.num_tqps; i++) {
3919 ret = h->ae_algo->ops->reset_queue(h, i);
3923 hns3_init_ring_hw(priv->ring_data[i].ring);
3925 /* We need to clear tx ring here because self test will
3926 * use the ring and will not run down before up
3928 hns3_clear_tx_ring(priv->ring_data[i].ring);
3929 priv->ring_data[i].ring->next_to_clean = 0;
3930 priv->ring_data[i].ring->next_to_use = 0;
3932 rx_ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
3933 hns3_init_ring_hw(rx_ring);
3934 ret = hns3_clear_rx_ring(rx_ring);
3938 /* We can not know the hardware head and tail when this
3939 * function is called in reset flow, so we reuse all desc.
3941 for (j = 0; j < rx_ring->desc_num; j++)
3942 hns3_reuse_buffer(rx_ring, j);
3944 rx_ring->next_to_clean = 0;
3945 rx_ring->next_to_use = 0;
3948 hns3_init_tx_ring_tc(priv);
3953 static void hns3_store_coal(struct hns3_nic_priv *priv)
3955 /* ethtool only support setting and querying one coal
3956 * configuation for now, so save the vector 0' coal
3957 * configuation here in order to restore it.
3959 memcpy(&priv->tx_coal, &priv->tqp_vector[0].tx_group.coal,
3960 sizeof(struct hns3_enet_coalesce));
3961 memcpy(&priv->rx_coal, &priv->tqp_vector[0].rx_group.coal,
3962 sizeof(struct hns3_enet_coalesce));
3965 static void hns3_restore_coal(struct hns3_nic_priv *priv)
3967 u16 vector_num = priv->vector_num;
3970 for (i = 0; i < vector_num; i++) {
3971 memcpy(&priv->tqp_vector[i].tx_group.coal, &priv->tx_coal,
3972 sizeof(struct hns3_enet_coalesce));
3973 memcpy(&priv->tqp_vector[i].rx_group.coal, &priv->rx_coal,
3974 sizeof(struct hns3_enet_coalesce));
3978 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
3980 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(handle->pdev);
3981 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
3982 struct net_device *ndev = kinfo->netdev;
3983 struct hns3_nic_priv *priv = netdev_priv(ndev);
3985 if (test_and_set_bit(HNS3_NIC_STATE_RESETTING, &priv->state))
3988 /* it is cumbersome for hardware to pick-and-choose entries for deletion
3989 * from table space. Hence, for function reset software intervention is
3990 * required to delete the entries
3992 if (hns3_dev_ongoing_func_reset(ae_dev)) {
3993 hns3_remove_hw_addr(ndev);
3994 hns3_del_all_fd_rules(ndev, false);
3997 if (!netif_running(ndev))
4000 return hns3_nic_net_stop(ndev);
4003 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
4005 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4006 struct hns3_nic_priv *priv = netdev_priv(kinfo->netdev);
4009 clear_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4011 if (netif_running(kinfo->netdev)) {
4012 ret = hns3_nic_net_open(kinfo->netdev);
4014 set_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4015 netdev_err(kinfo->netdev,
4016 "hns net up fail, ret=%d!\n", ret);
4024 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
4026 struct net_device *netdev = handle->kinfo.netdev;
4027 struct hns3_nic_priv *priv = netdev_priv(netdev);
4030 /* Carrier off reporting is important to ethtool even BEFORE open */
4031 netif_carrier_off(netdev);
4033 ret = hns3_get_ring_config(priv);
4037 ret = hns3_nic_alloc_vector_data(priv);
4041 hns3_restore_coal(priv);
4043 ret = hns3_nic_init_vector_data(priv);
4045 goto err_dealloc_vector;
4047 ret = hns3_init_all_ring(priv);
4049 goto err_uninit_vector;
4051 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
4056 hns3_nic_uninit_vector_data(priv);
4057 priv->ring_data = NULL;
4059 hns3_nic_dealloc_vector_data(priv);
4061 hns3_put_ring_config(priv);
4062 priv->ring_data = NULL;
4067 static int hns3_reset_notify_restore_enet(struct hnae3_handle *handle)
4069 struct net_device *netdev = handle->kinfo.netdev;
4070 bool vlan_filter_enable;
4073 ret = hns3_init_mac_addr(netdev, false);
4077 ret = hns3_recover_hw_addr(netdev);
4081 ret = hns3_update_promisc_mode(netdev, handle->netdev_flags);
4085 vlan_filter_enable = netdev->flags & IFF_PROMISC ? false : true;
4086 hns3_enable_vlan_filter(netdev, vlan_filter_enable);
4088 /* Hardware table is only clear when pf resets */
4089 if (!(handle->flags & HNAE3_SUPPORT_VF)) {
4090 ret = hns3_restore_vlan(netdev);
4095 return hns3_restore_fd_rules(netdev);
4098 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
4100 struct net_device *netdev = handle->kinfo.netdev;
4101 struct hns3_nic_priv *priv = netdev_priv(netdev);
4104 if (!test_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
4105 netdev_warn(netdev, "already uninitialized\n");
4109 hns3_force_clear_all_rx_ring(handle);
4111 hns3_nic_uninit_vector_data(priv);
4113 hns3_store_coal(priv);
4115 ret = hns3_nic_dealloc_vector_data(priv);
4117 netdev_err(netdev, "dealloc vector error\n");
4119 ret = hns3_uninit_all_ring(priv);
4121 netdev_err(netdev, "uninit ring error\n");
4123 hns3_put_ring_config(priv);
4124 priv->ring_data = NULL;
4126 clear_bit(HNS3_NIC_STATE_INITED, &priv->state);
4131 static int hns3_reset_notify(struct hnae3_handle *handle,
4132 enum hnae3_reset_notify_type type)
4137 case HNAE3_UP_CLIENT:
4138 ret = hns3_reset_notify_up_enet(handle);
4140 case HNAE3_DOWN_CLIENT:
4141 ret = hns3_reset_notify_down_enet(handle);
4143 case HNAE3_INIT_CLIENT:
4144 ret = hns3_reset_notify_init_enet(handle);
4146 case HNAE3_UNINIT_CLIENT:
4147 ret = hns3_reset_notify_uninit_enet(handle);
4149 case HNAE3_RESTORE_CLIENT:
4150 ret = hns3_reset_notify_restore_enet(handle);
4159 int hns3_set_channels(struct net_device *netdev,
4160 struct ethtool_channels *ch)
4162 struct hnae3_handle *h = hns3_get_handle(netdev);
4163 struct hnae3_knic_private_info *kinfo = &h->kinfo;
4164 bool rxfh_configured = netif_is_rxfh_configured(netdev);
4165 u32 new_tqp_num = ch->combined_count;
4169 if (ch->rx_count || ch->tx_count)
4172 if (new_tqp_num > hns3_get_max_available_channels(h) ||
4174 dev_err(&netdev->dev,
4175 "Change tqps fail, the tqp range is from 1 to %d",
4176 hns3_get_max_available_channels(h));
4180 if (kinfo->rss_size == new_tqp_num)
4183 ret = hns3_reset_notify(h, HNAE3_DOWN_CLIENT);
4187 ret = hns3_reset_notify(h, HNAE3_UNINIT_CLIENT);
4191 org_tqp_num = h->kinfo.num_tqps;
4192 ret = h->ae_algo->ops->set_channels(h, new_tqp_num, rxfh_configured);
4194 ret = h->ae_algo->ops->set_channels(h, org_tqp_num,
4197 /* If revert to old tqp failed, fatal error occurred */
4198 dev_err(&netdev->dev,
4199 "Revert to old tqp num fail, ret=%d", ret);
4202 dev_info(&netdev->dev,
4203 "Change tqp num fail, Revert to old tqp num");
4205 ret = hns3_reset_notify(h, HNAE3_INIT_CLIENT);
4209 return hns3_reset_notify(h, HNAE3_UP_CLIENT);
4212 static const struct hnae3_client_ops client_ops = {
4213 .init_instance = hns3_client_init,
4214 .uninit_instance = hns3_client_uninit,
4215 .link_status_change = hns3_link_status_change,
4216 .setup_tc = hns3_client_setup_tc,
4217 .reset_notify = hns3_reset_notify,
4220 /* hns3_init_module - Driver registration routine
4221 * hns3_init_module is the first routine called when the driver is
4222 * loaded. All it does is register with the PCI subsystem.
4224 static int __init hns3_init_module(void)
4228 pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
4229 pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
4231 client.type = HNAE3_CLIENT_KNIC;
4232 snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH - 1, "%s",
4235 client.ops = &client_ops;
4237 INIT_LIST_HEAD(&client.node);
4239 hns3_dbg_register_debugfs(hns3_driver_name);
4241 ret = hnae3_register_client(&client);
4243 goto err_reg_client;
4245 ret = pci_register_driver(&hns3_driver);
4247 goto err_reg_driver;
4252 hnae3_unregister_client(&client);
4254 hns3_dbg_unregister_debugfs();
4257 module_init(hns3_init_module);
4259 /* hns3_exit_module - Driver exit cleanup routine
4260 * hns3_exit_module is called just before the driver is removed
4263 static void __exit hns3_exit_module(void)
4265 pci_unregister_driver(&hns3_driver);
4266 hnae3_unregister_client(&client);
4267 hns3_dbg_unregister_debugfs();
4269 module_exit(hns3_exit_module);
4271 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
4272 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
4273 MODULE_LICENSE("GPL");
4274 MODULE_ALIAS("pci:hns-nic");
4275 MODULE_VERSION(HNS3_MOD_VERSION);