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 #ifdef CONFIG_RFS_ACCEL
8 #include <linux/cpu_rmap.h>
10 #include <linux/if_vlan.h>
12 #include <linux/ipv6.h>
13 #include <linux/module.h>
14 #include <linux/pci.h>
15 #include <linux/aer.h>
16 #include <linux/skbuff.h>
17 #include <linux/sctp.h>
18 #include <linux/vermagic.h>
20 #include <net/ip6_checksum.h>
21 #include <net/pkt_cls.h>
23 #include <net/vxlan.h>
26 #include "hns3_enet.h"
28 #define hns3_set_field(origin, shift, val) ((origin) |= ((val) << (shift)))
29 #define hns3_tx_bd_count(S) DIV_ROUND_UP(S, HNS3_MAX_BD_SIZE)
31 #define hns3_rl_err(fmt, ...) \
33 if (net_ratelimit()) \
34 netdev_err(fmt, ##__VA_ARGS__); \
37 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force);
38 static void hns3_remove_hw_addr(struct net_device *netdev);
40 static const char hns3_driver_name[] = "hns3";
41 const char hns3_driver_version[] = VERMAGIC_STRING;
42 static const char hns3_driver_string[] =
43 "Hisilicon Ethernet Network Driver for Hip08 Family";
44 static const char hns3_copyright[] = "Copyright (c) 2017 Huawei Corporation.";
45 static struct hnae3_client client;
47 static int debug = -1;
48 module_param(debug, int, 0);
49 MODULE_PARM_DESC(debug, " Network interface message level setting");
51 #define DEFAULT_MSG_LEVEL (NETIF_MSG_PROBE | NETIF_MSG_LINK | \
52 NETIF_MSG_IFDOWN | NETIF_MSG_IFUP)
54 #define HNS3_INNER_VLAN_TAG 1
55 #define HNS3_OUTER_VLAN_TAG 2
57 /* hns3_pci_tbl - PCI Device ID Table
59 * Last entry must be all 0s
61 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
62 * Class, Class Mask, private data (not used) }
64 static const struct pci_device_id hns3_pci_tbl[] = {
65 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
66 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
67 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA),
68 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
69 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC),
70 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
71 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA),
72 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
73 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC),
74 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
75 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC),
76 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
77 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_VF), 0},
78 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF),
79 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
80 /* required last entry */
83 MODULE_DEVICE_TABLE(pci, hns3_pci_tbl);
85 static irqreturn_t hns3_irq_handle(int irq, void *vector)
87 struct hns3_enet_tqp_vector *tqp_vector = vector;
89 napi_schedule_irqoff(&tqp_vector->napi);
94 static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv)
96 struct hns3_enet_tqp_vector *tqp_vectors;
99 for (i = 0; i < priv->vector_num; i++) {
100 tqp_vectors = &priv->tqp_vector[i];
102 if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED)
105 /* clear the affinity mask */
106 irq_set_affinity_hint(tqp_vectors->vector_irq, NULL);
108 /* release the irq resource */
109 free_irq(tqp_vectors->vector_irq, tqp_vectors);
110 tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED;
114 static int hns3_nic_init_irq(struct hns3_nic_priv *priv)
116 struct hns3_enet_tqp_vector *tqp_vectors;
117 int txrx_int_idx = 0;
123 for (i = 0; i < priv->vector_num; i++) {
124 tqp_vectors = &priv->tqp_vector[i];
126 if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED)
129 if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) {
130 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
131 "%s-%s-%d", priv->netdev->name, "TxRx",
134 } else if (tqp_vectors->rx_group.ring) {
135 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
136 "%s-%s-%d", priv->netdev->name, "Rx",
138 } else if (tqp_vectors->tx_group.ring) {
139 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
140 "%s-%s-%d", priv->netdev->name, "Tx",
143 /* Skip this unused q_vector */
147 tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0';
149 ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0,
150 tqp_vectors->name, tqp_vectors);
152 netdev_err(priv->netdev, "request irq(%d) fail\n",
153 tqp_vectors->vector_irq);
154 hns3_nic_uninit_irq(priv);
158 irq_set_affinity_hint(tqp_vectors->vector_irq,
159 &tqp_vectors->affinity_mask);
161 tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED;
167 static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector,
170 writel(mask_en, tqp_vector->mask_addr);
173 static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector)
175 napi_enable(&tqp_vector->napi);
178 hns3_mask_vector_irq(tqp_vector, 1);
181 static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector)
184 hns3_mask_vector_irq(tqp_vector, 0);
186 disable_irq(tqp_vector->vector_irq);
187 napi_disable(&tqp_vector->napi);
190 void hns3_set_vector_coalesce_rl(struct hns3_enet_tqp_vector *tqp_vector,
193 u32 rl_reg = hns3_rl_usec_to_reg(rl_value);
195 /* this defines the configuration for RL (Interrupt Rate Limiter).
196 * Rl defines rate of interrupts i.e. number of interrupts-per-second
197 * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing
200 if (rl_reg > 0 && !tqp_vector->tx_group.coal.gl_adapt_enable &&
201 !tqp_vector->rx_group.coal.gl_adapt_enable)
202 /* According to the hardware, the range of rl_reg is
203 * 0-59 and the unit is 4.
205 rl_reg |= HNS3_INT_RL_ENABLE_MASK;
207 writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
210 void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
213 u32 rx_gl_reg = hns3_gl_usec_to_reg(gl_value);
215 writel(rx_gl_reg, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
218 void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
221 u32 tx_gl_reg = hns3_gl_usec_to_reg(gl_value);
223 writel(tx_gl_reg, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
226 static void hns3_vector_gl_rl_init(struct hns3_enet_tqp_vector *tqp_vector,
227 struct hns3_nic_priv *priv)
229 /* initialize the configuration for interrupt coalescing.
230 * 1. GL (Interrupt Gap Limiter)
231 * 2. RL (Interrupt Rate Limiter)
233 * Default: enable interrupt coalescing self-adaptive and GL
235 tqp_vector->tx_group.coal.gl_adapt_enable = 1;
236 tqp_vector->rx_group.coal.gl_adapt_enable = 1;
238 tqp_vector->tx_group.coal.int_gl = HNS3_INT_GL_50K;
239 tqp_vector->rx_group.coal.int_gl = HNS3_INT_GL_50K;
241 tqp_vector->rx_group.coal.flow_level = HNS3_FLOW_LOW;
242 tqp_vector->tx_group.coal.flow_level = HNS3_FLOW_LOW;
245 static void hns3_vector_gl_rl_init_hw(struct hns3_enet_tqp_vector *tqp_vector,
246 struct hns3_nic_priv *priv)
248 struct hnae3_handle *h = priv->ae_handle;
250 hns3_set_vector_coalesce_tx_gl(tqp_vector,
251 tqp_vector->tx_group.coal.int_gl);
252 hns3_set_vector_coalesce_rx_gl(tqp_vector,
253 tqp_vector->rx_group.coal.int_gl);
254 hns3_set_vector_coalesce_rl(tqp_vector, h->kinfo.int_rl_setting);
257 static int hns3_nic_set_real_num_queue(struct net_device *netdev)
259 struct hnae3_handle *h = hns3_get_handle(netdev);
260 struct hnae3_knic_private_info *kinfo = &h->kinfo;
261 unsigned int queue_size = kinfo->rss_size * kinfo->num_tc;
264 if (kinfo->num_tc <= 1) {
265 netdev_reset_tc(netdev);
267 ret = netdev_set_num_tc(netdev, kinfo->num_tc);
270 "netdev_set_num_tc fail, ret=%d!\n", ret);
274 for (i = 0; i < HNAE3_MAX_TC; i++) {
275 if (!kinfo->tc_info[i].enable)
278 netdev_set_tc_queue(netdev,
279 kinfo->tc_info[i].tc,
280 kinfo->tc_info[i].tqp_count,
281 kinfo->tc_info[i].tqp_offset);
285 ret = netif_set_real_num_tx_queues(netdev, queue_size);
288 "netif_set_real_num_tx_queues fail, ret=%d!\n", ret);
292 ret = netif_set_real_num_rx_queues(netdev, queue_size);
295 "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
302 static u16 hns3_get_max_available_channels(struct hnae3_handle *h)
304 u16 alloc_tqps, max_rss_size, rss_size;
306 h->ae_algo->ops->get_tqps_and_rss_info(h, &alloc_tqps, &max_rss_size);
307 rss_size = alloc_tqps / h->kinfo.num_tc;
309 return min_t(u16, rss_size, max_rss_size);
312 static void hns3_tqp_enable(struct hnae3_queue *tqp)
316 rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
317 rcb_reg |= BIT(HNS3_RING_EN_B);
318 hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
321 static void hns3_tqp_disable(struct hnae3_queue *tqp)
325 rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
326 rcb_reg &= ~BIT(HNS3_RING_EN_B);
327 hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
330 static void hns3_free_rx_cpu_rmap(struct net_device *netdev)
332 #ifdef CONFIG_RFS_ACCEL
333 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
334 netdev->rx_cpu_rmap = NULL;
338 static int hns3_set_rx_cpu_rmap(struct net_device *netdev)
340 #ifdef CONFIG_RFS_ACCEL
341 struct hns3_nic_priv *priv = netdev_priv(netdev);
342 struct hns3_enet_tqp_vector *tqp_vector;
345 if (!netdev->rx_cpu_rmap) {
346 netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->vector_num);
347 if (!netdev->rx_cpu_rmap)
351 for (i = 0; i < priv->vector_num; i++) {
352 tqp_vector = &priv->tqp_vector[i];
353 ret = irq_cpu_rmap_add(netdev->rx_cpu_rmap,
354 tqp_vector->vector_irq);
356 hns3_free_rx_cpu_rmap(netdev);
364 static int hns3_nic_net_up(struct net_device *netdev)
366 struct hns3_nic_priv *priv = netdev_priv(netdev);
367 struct hnae3_handle *h = priv->ae_handle;
371 ret = hns3_nic_reset_all_ring(h);
375 /* the device can work without cpu rmap, only aRFS needs it */
376 ret = hns3_set_rx_cpu_rmap(netdev);
378 netdev_warn(netdev, "set rx cpu rmap fail, ret=%d!\n", ret);
380 /* get irq resource for all vectors */
381 ret = hns3_nic_init_irq(priv);
383 netdev_err(netdev, "init irq failed! ret=%d\n", ret);
387 clear_bit(HNS3_NIC_STATE_DOWN, &priv->state);
389 /* enable the vectors */
390 for (i = 0; i < priv->vector_num; i++)
391 hns3_vector_enable(&priv->tqp_vector[i]);
394 for (j = 0; j < h->kinfo.num_tqps; j++)
395 hns3_tqp_enable(h->kinfo.tqp[j]);
397 /* start the ae_dev */
398 ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0;
405 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
407 hns3_tqp_disable(h->kinfo.tqp[j]);
409 for (j = i - 1; j >= 0; j--)
410 hns3_vector_disable(&priv->tqp_vector[j]);
412 hns3_nic_uninit_irq(priv);
414 hns3_free_rx_cpu_rmap(netdev);
418 static void hns3_config_xps(struct hns3_nic_priv *priv)
422 for (i = 0; i < priv->vector_num; i++) {
423 struct hns3_enet_tqp_vector *tqp_vector = &priv->tqp_vector[i];
424 struct hns3_enet_ring *ring = tqp_vector->tx_group.ring;
429 ret = netif_set_xps_queue(priv->netdev,
430 &tqp_vector->affinity_mask,
431 ring->tqp->tqp_index);
433 netdev_warn(priv->netdev,
434 "set xps queue failed: %d", ret);
441 static int hns3_nic_net_open(struct net_device *netdev)
443 struct hns3_nic_priv *priv = netdev_priv(netdev);
444 struct hnae3_handle *h = hns3_get_handle(netdev);
445 struct hnae3_knic_private_info *kinfo;
448 if (hns3_nic_resetting(netdev))
451 netif_carrier_off(netdev);
453 ret = hns3_nic_set_real_num_queue(netdev);
457 ret = hns3_nic_net_up(netdev);
459 netdev_err(netdev, "net up fail, ret=%d!\n", ret);
464 for (i = 0; i < HNAE3_MAX_USER_PRIO; i++)
465 netdev_set_prio_tc_map(netdev, i, kinfo->prio_tc[i]);
467 if (h->ae_algo->ops->set_timer_task)
468 h->ae_algo->ops->set_timer_task(priv->ae_handle, true);
470 hns3_config_xps(priv);
472 netif_dbg(h, drv, netdev, "net open\n");
477 static void hns3_reset_tx_queue(struct hnae3_handle *h)
479 struct net_device *ndev = h->kinfo.netdev;
480 struct hns3_nic_priv *priv = netdev_priv(ndev);
481 struct netdev_queue *dev_queue;
484 for (i = 0; i < h->kinfo.num_tqps; i++) {
485 dev_queue = netdev_get_tx_queue(ndev,
486 priv->ring_data[i].queue_index);
487 netdev_tx_reset_queue(dev_queue);
491 static void hns3_nic_net_down(struct net_device *netdev)
493 struct hns3_nic_priv *priv = netdev_priv(netdev);
494 struct hnae3_handle *h = hns3_get_handle(netdev);
495 const struct hnae3_ae_ops *ops;
498 /* disable vectors */
499 for (i = 0; i < priv->vector_num; i++)
500 hns3_vector_disable(&priv->tqp_vector[i]);
503 for (i = 0; i < h->kinfo.num_tqps; i++)
504 hns3_tqp_disable(h->kinfo.tqp[i]);
507 ops = priv->ae_handle->ae_algo->ops;
509 ops->stop(priv->ae_handle);
511 hns3_free_rx_cpu_rmap(netdev);
513 /* free irq resources */
514 hns3_nic_uninit_irq(priv);
516 /* delay ring buffer clearing to hns3_reset_notify_uninit_enet
517 * during reset process, because driver may not be able
518 * to disable the ring through firmware when downing the netdev.
520 if (!hns3_nic_resetting(netdev))
521 hns3_clear_all_ring(priv->ae_handle, false);
523 hns3_reset_tx_queue(priv->ae_handle);
526 static int hns3_nic_net_stop(struct net_device *netdev)
528 struct hns3_nic_priv *priv = netdev_priv(netdev);
529 struct hnae3_handle *h = hns3_get_handle(netdev);
531 if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
534 netif_dbg(h, drv, netdev, "net stop\n");
536 if (h->ae_algo->ops->set_timer_task)
537 h->ae_algo->ops->set_timer_task(priv->ae_handle, false);
539 netif_tx_stop_all_queues(netdev);
540 netif_carrier_off(netdev);
542 hns3_nic_net_down(netdev);
547 static int hns3_nic_uc_sync(struct net_device *netdev,
548 const unsigned char *addr)
550 struct hnae3_handle *h = hns3_get_handle(netdev);
552 if (h->ae_algo->ops->add_uc_addr)
553 return h->ae_algo->ops->add_uc_addr(h, addr);
558 static int hns3_nic_uc_unsync(struct net_device *netdev,
559 const unsigned char *addr)
561 struct hnae3_handle *h = hns3_get_handle(netdev);
563 if (h->ae_algo->ops->rm_uc_addr)
564 return h->ae_algo->ops->rm_uc_addr(h, addr);
569 static int hns3_nic_mc_sync(struct net_device *netdev,
570 const unsigned char *addr)
572 struct hnae3_handle *h = hns3_get_handle(netdev);
574 if (h->ae_algo->ops->add_mc_addr)
575 return h->ae_algo->ops->add_mc_addr(h, addr);
580 static int hns3_nic_mc_unsync(struct net_device *netdev,
581 const unsigned char *addr)
583 struct hnae3_handle *h = hns3_get_handle(netdev);
585 if (h->ae_algo->ops->rm_mc_addr)
586 return h->ae_algo->ops->rm_mc_addr(h, addr);
591 static u8 hns3_get_netdev_flags(struct net_device *netdev)
595 if (netdev->flags & IFF_PROMISC) {
596 flags = HNAE3_USER_UPE | HNAE3_USER_MPE | HNAE3_BPE;
598 flags |= HNAE3_VLAN_FLTR;
599 if (netdev->flags & IFF_ALLMULTI)
600 flags |= HNAE3_USER_MPE;
606 static void hns3_nic_set_rx_mode(struct net_device *netdev)
608 struct hnae3_handle *h = hns3_get_handle(netdev);
612 new_flags = hns3_get_netdev_flags(netdev);
614 ret = __dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync);
616 netdev_err(netdev, "sync uc address fail\n");
618 new_flags |= HNAE3_OVERFLOW_UPE;
621 if (netdev->flags & IFF_MULTICAST) {
622 ret = __dev_mc_sync(netdev, hns3_nic_mc_sync,
625 netdev_err(netdev, "sync mc address fail\n");
627 new_flags |= HNAE3_OVERFLOW_MPE;
631 /* User mode Promisc mode enable and vlan filtering is disabled to
632 * let all packets in. MAC-VLAN Table overflow Promisc enabled and
633 * vlan fitering is enabled
635 hns3_enable_vlan_filter(netdev, new_flags & HNAE3_VLAN_FLTR);
636 h->netdev_flags = new_flags;
637 hns3_update_promisc_mode(netdev, new_flags);
640 int hns3_update_promisc_mode(struct net_device *netdev, u8 promisc_flags)
642 struct hns3_nic_priv *priv = netdev_priv(netdev);
643 struct hnae3_handle *h = priv->ae_handle;
645 if (h->ae_algo->ops->set_promisc_mode) {
646 return h->ae_algo->ops->set_promisc_mode(h,
647 promisc_flags & HNAE3_UPE,
648 promisc_flags & HNAE3_MPE);
654 void hns3_enable_vlan_filter(struct net_device *netdev, bool enable)
656 struct hns3_nic_priv *priv = netdev_priv(netdev);
657 struct hnae3_handle *h = priv->ae_handle;
660 if (h->pdev->revision >= 0x21 && h->ae_algo->ops->enable_vlan_filter) {
661 last_state = h->netdev_flags & HNAE3_VLAN_FLTR ? true : false;
662 if (enable != last_state) {
665 enable ? "enable" : "disable");
666 h->ae_algo->ops->enable_vlan_filter(h, enable);
671 static int hns3_set_tso(struct sk_buff *skb, u32 *paylen,
672 u16 *mss, u32 *type_cs_vlan_tso)
674 u32 l4_offset, hdr_len;
675 union l3_hdr_info l3;
676 union l4_hdr_info l4;
680 if (!skb_is_gso(skb))
683 ret = skb_cow_head(skb, 0);
687 l3.hdr = skb_network_header(skb);
688 l4.hdr = skb_transport_header(skb);
690 /* Software should clear the IPv4's checksum field when tso is
693 if (l3.v4->version == 4)
697 if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
700 SKB_GSO_UDP_TUNNEL_CSUM)) {
701 if ((!(skb_shinfo(skb)->gso_type &
703 (skb_shinfo(skb)->gso_type &
704 SKB_GSO_UDP_TUNNEL_CSUM)) {
705 /* Software should clear the udp's checksum
706 * field when tso is needed.
710 /* reset l3&l4 pointers from outer to inner headers */
711 l3.hdr = skb_inner_network_header(skb);
712 l4.hdr = skb_inner_transport_header(skb);
714 /* Software should clear the IPv4's checksum field when
717 if (l3.v4->version == 4)
721 /* normal or tunnel packet */
722 l4_offset = l4.hdr - skb->data;
723 hdr_len = (l4.tcp->doff << 2) + l4_offset;
725 /* remove payload length from inner pseudo checksum when tso */
726 l4_paylen = skb->len - l4_offset;
727 csum_replace_by_diff(&l4.tcp->check,
728 (__force __wsum)htonl(l4_paylen));
730 /* find the txbd field values */
731 *paylen = skb->len - hdr_len;
732 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_TSO_B, 1);
734 /* get MSS for TSO */
735 *mss = skb_shinfo(skb)->gso_size;
740 static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
743 union l3_hdr_info l3;
744 unsigned char *l4_hdr;
745 unsigned char *exthdr;
749 /* find outer header point */
750 l3.hdr = skb_network_header(skb);
751 l4_hdr = skb_transport_header(skb);
753 if (skb->protocol == htons(ETH_P_IPV6)) {
754 exthdr = l3.hdr + sizeof(*l3.v6);
755 l4_proto_tmp = l3.v6->nexthdr;
756 if (l4_hdr != exthdr)
757 ipv6_skip_exthdr(skb, exthdr - skb->data,
758 &l4_proto_tmp, &frag_off);
759 } else if (skb->protocol == htons(ETH_P_IP)) {
760 l4_proto_tmp = l3.v4->protocol;
765 *ol4_proto = l4_proto_tmp;
768 if (!skb->encapsulation) {
773 /* find inner header point */
774 l3.hdr = skb_inner_network_header(skb);
775 l4_hdr = skb_inner_transport_header(skb);
777 if (l3.v6->version == 6) {
778 exthdr = l3.hdr + sizeof(*l3.v6);
779 l4_proto_tmp = l3.v6->nexthdr;
780 if (l4_hdr != exthdr)
781 ipv6_skip_exthdr(skb, exthdr - skb->data,
782 &l4_proto_tmp, &frag_off);
783 } else if (l3.v4->version == 4) {
784 l4_proto_tmp = l3.v4->protocol;
787 *il4_proto = l4_proto_tmp;
792 /* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
793 * and it is udp packet, which has a dest port as the IANA assigned.
794 * the hardware is expected to do the checksum offload, but the
795 * hardware will not do the checksum offload when udp dest port is
798 static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
800 union l4_hdr_info l4;
802 l4.hdr = skb_transport_header(skb);
804 if (!(!skb->encapsulation &&
805 l4.udp->dest == htons(IANA_VXLAN_UDP_PORT)))
808 skb_checksum_help(skb);
813 static void hns3_set_outer_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
814 u32 *ol_type_vlan_len_msec)
816 u32 l2_len, l3_len, l4_len;
817 unsigned char *il2_hdr;
818 union l3_hdr_info l3;
819 union l4_hdr_info l4;
821 l3.hdr = skb_network_header(skb);
822 l4.hdr = skb_transport_header(skb);
824 /* compute OL2 header size, defined in 2 Bytes */
825 l2_len = l3.hdr - skb->data;
826 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L2LEN_S, l2_len >> 1);
828 /* compute OL3 header size, defined in 4 Bytes */
829 l3_len = l4.hdr - l3.hdr;
830 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_S, l3_len >> 2);
832 il2_hdr = skb_inner_mac_header(skb);
833 /* compute OL4 header size, defined in 4 Bytes */
834 l4_len = il2_hdr - l4.hdr;
835 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L4LEN_S, l4_len >> 2);
837 /* define outer network header type */
838 if (skb->protocol == htons(ETH_P_IP)) {
840 hns3_set_field(*ol_type_vlan_len_msec,
842 HNS3_OL3T_IPV4_CSUM);
844 hns3_set_field(*ol_type_vlan_len_msec,
846 HNS3_OL3T_IPV4_NO_CSUM);
848 } else if (skb->protocol == htons(ETH_P_IPV6)) {
849 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_S,
853 if (ol4_proto == IPPROTO_UDP)
854 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
855 HNS3_TUN_MAC_IN_UDP);
856 else if (ol4_proto == IPPROTO_GRE)
857 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
861 static int hns3_set_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
862 u8 il4_proto, u32 *type_cs_vlan_tso,
863 u32 *ol_type_vlan_len_msec)
865 unsigned char *l2_hdr = skb->data;
866 u32 l4_proto = ol4_proto;
867 union l4_hdr_info l4;
868 union l3_hdr_info l3;
871 l4.hdr = skb_transport_header(skb);
872 l3.hdr = skb_network_header(skb);
874 /* handle encapsulation skb */
875 if (skb->encapsulation) {
876 /* If this is a not UDP/GRE encapsulation skb */
877 if (!(ol4_proto == IPPROTO_UDP || ol4_proto == IPPROTO_GRE)) {
878 /* drop the skb tunnel packet if hardware don't support,
879 * because hardware can't calculate csum when TSO.
884 /* the stack computes the IP header already,
885 * driver calculate l4 checksum when not TSO.
887 skb_checksum_help(skb);
891 hns3_set_outer_l2l3l4(skb, ol4_proto, ol_type_vlan_len_msec);
893 /* switch to inner header */
894 l2_hdr = skb_inner_mac_header(skb);
895 l3.hdr = skb_inner_network_header(skb);
896 l4.hdr = skb_inner_transport_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,
914 /* compute inner(/normal) L2 header size, defined in 2 Bytes */
915 l2_len = l3.hdr - l2_hdr;
916 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S, l2_len >> 1);
918 /* compute inner(/normal) L3 header size, defined in 4 Bytes */
919 l3_len = l4.hdr - l3.hdr;
920 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_S, l3_len >> 2);
922 /* compute inner(/normal) L4 header size, defined in 4 Bytes */
925 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
926 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
928 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
932 if (hns3_tunnel_csum_bug(skb))
935 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
936 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
938 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
939 (sizeof(struct udphdr) >> 2));
942 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
943 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
945 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
946 (sizeof(struct sctphdr) >> 2));
949 /* drop the skb tunnel packet if hardware don't support,
950 * because hardware can't calculate csum when TSO.
955 /* the stack computes the IP header already,
956 * driver calculate l4 checksum when not TSO.
958 skb_checksum_help(skb);
965 static void hns3_set_txbd_baseinfo(u16 *bdtp_fe_sc_vld_ra_ri, int frag_end)
967 /* Config bd buffer end */
969 hns3_set_field(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_FE_B, 1U);
970 hns3_set_field(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_VLD_B, 1U);
973 static int hns3_handle_vtags(struct hns3_enet_ring *tx_ring,
976 struct hnae3_handle *handle = tx_ring->tqp->handle;
977 struct vlan_ethhdr *vhdr;
980 if (!(skb->protocol == htons(ETH_P_8021Q) ||
981 skb_vlan_tag_present(skb)))
984 /* Since HW limitation, if port based insert VLAN enabled, only one VLAN
985 * header is allowed in skb, otherwise it will cause RAS error.
987 if (unlikely(skb_vlan_tagged_multi(skb) &&
988 handle->port_base_vlan_state ==
989 HNAE3_PORT_BASE_VLAN_ENABLE))
992 if (skb->protocol == htons(ETH_P_8021Q) &&
993 !(handle->kinfo.netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
994 /* When HW VLAN acceleration is turned off, and the stack
995 * sets the protocol to 802.1q, the driver just need to
996 * set the protocol to the encapsulated ethertype.
998 skb->protocol = vlan_get_protocol(skb);
1002 if (skb_vlan_tag_present(skb)) {
1003 /* Based on hw strategy, use out_vtag in two layer tag case,
1004 * and use inner_vtag in one tag case.
1006 if (skb->protocol == htons(ETH_P_8021Q) &&
1007 handle->port_base_vlan_state ==
1008 HNAE3_PORT_BASE_VLAN_DISABLE)
1009 rc = HNS3_OUTER_VLAN_TAG;
1011 rc = HNS3_INNER_VLAN_TAG;
1013 skb->protocol = vlan_get_protocol(skb);
1017 rc = skb_cow_head(skb, 0);
1018 if (unlikely(rc < 0))
1021 vhdr = (struct vlan_ethhdr *)skb->data;
1022 vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority << VLAN_PRIO_SHIFT)
1025 skb->protocol = vlan_get_protocol(skb);
1029 static int hns3_fill_skb_desc(struct hns3_enet_ring *ring,
1030 struct sk_buff *skb, struct hns3_desc *desc)
1032 u32 ol_type_vlan_len_msec = 0;
1033 u32 type_cs_vlan_tso = 0;
1034 u32 paylen = skb->len;
1040 ret = hns3_handle_vtags(ring, skb);
1041 if (unlikely(ret < 0)) {
1042 u64_stats_update_begin(&ring->syncp);
1043 ring->stats.tx_vlan_err++;
1044 u64_stats_update_end(&ring->syncp);
1046 } else if (ret == HNS3_INNER_VLAN_TAG) {
1047 inner_vtag = skb_vlan_tag_get(skb);
1048 inner_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1050 hns3_set_field(type_cs_vlan_tso, HNS3_TXD_VLAN_B, 1);
1051 } else if (ret == HNS3_OUTER_VLAN_TAG) {
1052 out_vtag = skb_vlan_tag_get(skb);
1053 out_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1055 hns3_set_field(ol_type_vlan_len_msec, HNS3_TXD_OVLAN_B,
1059 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1060 u8 ol4_proto, il4_proto;
1062 skb_reset_mac_len(skb);
1064 ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
1065 if (unlikely(ret)) {
1066 u64_stats_update_begin(&ring->syncp);
1067 ring->stats.tx_l4_proto_err++;
1068 u64_stats_update_end(&ring->syncp);
1072 ret = hns3_set_l2l3l4(skb, ol4_proto, il4_proto,
1074 &ol_type_vlan_len_msec);
1075 if (unlikely(ret)) {
1076 u64_stats_update_begin(&ring->syncp);
1077 ring->stats.tx_l2l3l4_err++;
1078 u64_stats_update_end(&ring->syncp);
1082 ret = hns3_set_tso(skb, &paylen, &mss,
1084 if (unlikely(ret)) {
1085 u64_stats_update_begin(&ring->syncp);
1086 ring->stats.tx_tso_err++;
1087 u64_stats_update_end(&ring->syncp);
1093 desc->tx.ol_type_vlan_len_msec =
1094 cpu_to_le32(ol_type_vlan_len_msec);
1095 desc->tx.type_cs_vlan_tso_len = cpu_to_le32(type_cs_vlan_tso);
1096 desc->tx.paylen = cpu_to_le32(paylen);
1097 desc->tx.mss = cpu_to_le16(mss);
1098 desc->tx.vlan_tag = cpu_to_le16(inner_vtag);
1099 desc->tx.outer_vlan_tag = cpu_to_le16(out_vtag);
1104 static int hns3_fill_desc(struct hns3_enet_ring *ring, void *priv,
1105 unsigned int size, int frag_end,
1106 enum hns_desc_type type)
1108 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1109 struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1110 struct device *dev = ring_to_dev(ring);
1112 unsigned int frag_buf_num;
1116 if (type == DESC_TYPE_SKB) {
1117 struct sk_buff *skb = (struct sk_buff *)priv;
1120 ret = hns3_fill_skb_desc(ring, skb, desc);
1124 dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1126 frag = (skb_frag_t *)priv;
1127 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1130 if (unlikely(dma_mapping_error(dev, dma))) {
1131 u64_stats_update_begin(&ring->syncp);
1132 ring->stats.sw_err_cnt++;
1133 u64_stats_update_end(&ring->syncp);
1137 desc_cb->length = size;
1139 if (likely(size <= HNS3_MAX_BD_SIZE)) {
1140 u16 bdtp_fe_sc_vld_ra_ri = 0;
1142 desc_cb->priv = priv;
1144 desc_cb->type = type;
1145 desc->addr = cpu_to_le64(dma);
1146 desc->tx.send_size = cpu_to_le16(size);
1147 hns3_set_txbd_baseinfo(&bdtp_fe_sc_vld_ra_ri, frag_end);
1148 desc->tx.bdtp_fe_sc_vld_ra_ri =
1149 cpu_to_le16(bdtp_fe_sc_vld_ra_ri);
1151 ring_ptr_move_fw(ring, next_to_use);
1155 frag_buf_num = hns3_tx_bd_count(size);
1156 sizeoflast = size & HNS3_TX_LAST_SIZE_M;
1157 sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1159 /* When frag size is bigger than hardware limit, split this frag */
1160 for (k = 0; k < frag_buf_num; k++) {
1161 u16 bdtp_fe_sc_vld_ra_ri = 0;
1163 /* The txbd's baseinfo of DESC_TYPE_PAGE & DESC_TYPE_SKB */
1164 desc_cb->priv = priv;
1165 desc_cb->dma = dma + HNS3_MAX_BD_SIZE * k;
1166 desc_cb->type = (type == DESC_TYPE_SKB && !k) ?
1167 DESC_TYPE_SKB : DESC_TYPE_PAGE;
1169 /* now, fill the descriptor */
1170 desc->addr = cpu_to_le64(dma + HNS3_MAX_BD_SIZE * k);
1171 desc->tx.send_size = cpu_to_le16((k == frag_buf_num - 1) ?
1172 (u16)sizeoflast : (u16)HNS3_MAX_BD_SIZE);
1173 hns3_set_txbd_baseinfo(&bdtp_fe_sc_vld_ra_ri,
1174 frag_end && (k == frag_buf_num - 1) ?
1176 desc->tx.bdtp_fe_sc_vld_ra_ri =
1177 cpu_to_le16(bdtp_fe_sc_vld_ra_ri);
1179 /* move ring pointer to next */
1180 ring_ptr_move_fw(ring, next_to_use);
1182 desc_cb = &ring->desc_cb[ring->next_to_use];
1183 desc = &ring->desc[ring->next_to_use];
1189 static unsigned int hns3_nic_bd_num(struct sk_buff *skb)
1191 unsigned int bd_num;
1194 /* if the total len is within the max bd limit */
1195 if (likely(skb->len <= HNS3_MAX_BD_SIZE))
1196 return skb_shinfo(skb)->nr_frags + 1;
1198 bd_num = hns3_tx_bd_count(skb_headlen(skb));
1200 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1201 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1202 bd_num += hns3_tx_bd_count(skb_frag_size(frag));
1208 static unsigned int hns3_gso_hdr_len(struct sk_buff *skb)
1210 if (!skb->encapsulation)
1211 return skb_transport_offset(skb) + tcp_hdrlen(skb);
1213 return skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb);
1216 /* HW need every continuous 8 buffer data to be larger than MSS,
1217 * we simplify it by ensuring skb_headlen + the first continuous
1218 * 7 frags to to be larger than gso header len + mss, and the remaining
1219 * continuous 7 frags to be larger than MSS except the last 7 frags.
1221 static bool hns3_skb_need_linearized(struct sk_buff *skb)
1223 int bd_limit = HNS3_MAX_BD_NUM_NORMAL - 1;
1224 unsigned int tot_len = 0;
1227 for (i = 0; i < bd_limit; i++)
1228 tot_len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1230 /* ensure headlen + the first 7 frags is greater than mss + header
1231 * and the first 7 frags is greater than mss.
1233 if (((tot_len + skb_headlen(skb)) < (skb_shinfo(skb)->gso_size +
1234 hns3_gso_hdr_len(skb))) || (tot_len < skb_shinfo(skb)->gso_size))
1237 /* ensure the remaining continuous 7 buffer is greater than mss */
1238 for (i = 0; i < (skb_shinfo(skb)->nr_frags - bd_limit - 1); i++) {
1239 tot_len -= skb_frag_size(&skb_shinfo(skb)->frags[i]);
1240 tot_len += skb_frag_size(&skb_shinfo(skb)->frags[i + bd_limit]);
1242 if (tot_len < skb_shinfo(skb)->gso_size)
1249 static int hns3_nic_maybe_stop_tx(struct hns3_enet_ring *ring,
1250 struct sk_buff **out_skb)
1252 struct sk_buff *skb = *out_skb;
1253 unsigned int bd_num;
1255 bd_num = hns3_nic_bd_num(skb);
1256 if (unlikely(bd_num > HNS3_MAX_BD_NUM_NORMAL)) {
1257 struct sk_buff *new_skb;
1259 if (skb_is_gso(skb) && bd_num <= HNS3_MAX_BD_NUM_TSO &&
1260 !hns3_skb_need_linearized(skb))
1263 /* manual split the send packet */
1264 new_skb = skb_copy(skb, GFP_ATOMIC);
1267 dev_kfree_skb_any(skb);
1270 bd_num = hns3_nic_bd_num(new_skb);
1271 if ((skb_is_gso(new_skb) && bd_num > HNS3_MAX_BD_NUM_TSO) ||
1272 (!skb_is_gso(new_skb) && bd_num > HNS3_MAX_BD_NUM_NORMAL))
1275 u64_stats_update_begin(&ring->syncp);
1276 ring->stats.tx_copy++;
1277 u64_stats_update_end(&ring->syncp);
1281 if (unlikely(ring_space(ring) < bd_num))
1287 static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
1289 struct device *dev = ring_to_dev(ring);
1292 for (i = 0; i < ring->desc_num; i++) {
1293 /* check if this is where we started */
1294 if (ring->next_to_use == next_to_use_orig)
1298 ring_ptr_move_bw(ring, next_to_use);
1300 /* unmap the descriptor dma address */
1301 if (ring->desc_cb[ring->next_to_use].type == DESC_TYPE_SKB)
1302 dma_unmap_single(dev,
1303 ring->desc_cb[ring->next_to_use].dma,
1304 ring->desc_cb[ring->next_to_use].length,
1306 else if (ring->desc_cb[ring->next_to_use].length)
1308 ring->desc_cb[ring->next_to_use].dma,
1309 ring->desc_cb[ring->next_to_use].length,
1312 ring->desc_cb[ring->next_to_use].length = 0;
1313 ring->desc_cb[ring->next_to_use].dma = 0;
1317 netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
1319 struct hns3_nic_priv *priv = netdev_priv(netdev);
1320 struct hns3_nic_ring_data *ring_data =
1321 &tx_ring_data(priv, skb->queue_mapping);
1322 struct hns3_enet_ring *ring = ring_data->ring;
1323 struct netdev_queue *dev_queue;
1325 int next_to_use_head;
1332 /* Prefetch the data used later */
1333 prefetch(skb->data);
1335 buf_num = hns3_nic_maybe_stop_tx(ring, &skb);
1336 if (unlikely(buf_num <= 0)) {
1337 if (buf_num == -EBUSY) {
1338 u64_stats_update_begin(&ring->syncp);
1339 ring->stats.tx_busy++;
1340 u64_stats_update_end(&ring->syncp);
1341 goto out_net_tx_busy;
1342 } else if (buf_num == -ENOMEM) {
1343 u64_stats_update_begin(&ring->syncp);
1344 ring->stats.sw_err_cnt++;
1345 u64_stats_update_end(&ring->syncp);
1348 hns3_rl_err(netdev, "xmit error: %d!\n", buf_num);
1352 /* No. of segments (plus a header) */
1353 seg_num = skb_shinfo(skb)->nr_frags + 1;
1354 /* Fill the first part */
1355 size = skb_headlen(skb);
1357 next_to_use_head = ring->next_to_use;
1359 ret = hns3_fill_desc(ring, skb, size, seg_num == 1 ? 1 : 0,
1364 /* Fill the fragments */
1365 for (i = 1; i < seg_num; i++) {
1366 frag = &skb_shinfo(skb)->frags[i - 1];
1367 size = skb_frag_size(frag);
1369 ret = hns3_fill_desc(ring, frag, size,
1370 seg_num - 1 == i ? 1 : 0,
1377 /* Complete translate all packets */
1378 dev_queue = netdev_get_tx_queue(netdev, ring_data->queue_index);
1379 netdev_tx_sent_queue(dev_queue, skb->len);
1381 wmb(); /* Commit all data before submit */
1383 hnae3_queue_xmit(ring->tqp, buf_num);
1385 return NETDEV_TX_OK;
1388 hns3_clear_desc(ring, next_to_use_head);
1391 dev_kfree_skb_any(skb);
1392 return NETDEV_TX_OK;
1395 netif_stop_subqueue(netdev, ring_data->queue_index);
1396 smp_mb(); /* Commit all data before submit */
1398 return NETDEV_TX_BUSY;
1401 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
1403 struct hnae3_handle *h = hns3_get_handle(netdev);
1404 struct sockaddr *mac_addr = p;
1407 if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1408 return -EADDRNOTAVAIL;
1410 if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
1411 netdev_info(netdev, "already using mac address %pM\n",
1416 ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
1418 netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
1422 ether_addr_copy(netdev->dev_addr, mac_addr->sa_data);
1427 static int hns3_nic_do_ioctl(struct net_device *netdev,
1428 struct ifreq *ifr, int cmd)
1430 struct hnae3_handle *h = hns3_get_handle(netdev);
1432 if (!netif_running(netdev))
1435 if (!h->ae_algo->ops->do_ioctl)
1438 return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
1441 static int hns3_nic_set_features(struct net_device *netdev,
1442 netdev_features_t features)
1444 netdev_features_t changed = netdev->features ^ features;
1445 struct hns3_nic_priv *priv = netdev_priv(netdev);
1446 struct hnae3_handle *h = priv->ae_handle;
1450 if (changed & (NETIF_F_GRO_HW) && h->ae_algo->ops->set_gro_en) {
1451 enable = !!(features & NETIF_F_GRO_HW);
1452 ret = h->ae_algo->ops->set_gro_en(h, enable);
1457 if ((changed & NETIF_F_HW_VLAN_CTAG_FILTER) &&
1458 h->ae_algo->ops->enable_vlan_filter) {
1459 enable = !!(features & NETIF_F_HW_VLAN_CTAG_FILTER);
1460 h->ae_algo->ops->enable_vlan_filter(h, enable);
1463 if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
1464 h->ae_algo->ops->enable_hw_strip_rxvtag) {
1465 enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
1466 ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, enable);
1471 if ((changed & NETIF_F_NTUPLE) && h->ae_algo->ops->enable_fd) {
1472 enable = !!(features & NETIF_F_NTUPLE);
1473 h->ae_algo->ops->enable_fd(h, enable);
1476 netdev->features = features;
1480 static void hns3_nic_get_stats64(struct net_device *netdev,
1481 struct rtnl_link_stats64 *stats)
1483 struct hns3_nic_priv *priv = netdev_priv(netdev);
1484 int queue_num = priv->ae_handle->kinfo.num_tqps;
1485 struct hnae3_handle *handle = priv->ae_handle;
1486 struct hns3_enet_ring *ring;
1487 u64 rx_length_errors = 0;
1488 u64 rx_crc_errors = 0;
1489 u64 rx_multicast = 0;
1501 if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
1504 handle->ae_algo->ops->update_stats(handle, &netdev->stats);
1506 for (idx = 0; idx < queue_num; idx++) {
1507 /* fetch the tx stats */
1508 ring = priv->ring_data[idx].ring;
1510 start = u64_stats_fetch_begin_irq(&ring->syncp);
1511 tx_bytes += ring->stats.tx_bytes;
1512 tx_pkts += ring->stats.tx_pkts;
1513 tx_drop += ring->stats.sw_err_cnt;
1514 tx_drop += ring->stats.tx_vlan_err;
1515 tx_drop += ring->stats.tx_l4_proto_err;
1516 tx_drop += ring->stats.tx_l2l3l4_err;
1517 tx_drop += ring->stats.tx_tso_err;
1518 tx_errors += ring->stats.sw_err_cnt;
1519 tx_errors += ring->stats.tx_vlan_err;
1520 tx_errors += ring->stats.tx_l4_proto_err;
1521 tx_errors += ring->stats.tx_l2l3l4_err;
1522 tx_errors += ring->stats.tx_tso_err;
1523 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1525 /* fetch the rx stats */
1526 ring = priv->ring_data[idx + queue_num].ring;
1528 start = u64_stats_fetch_begin_irq(&ring->syncp);
1529 rx_bytes += ring->stats.rx_bytes;
1530 rx_pkts += ring->stats.rx_pkts;
1531 rx_drop += ring->stats.l2_err;
1532 rx_errors += ring->stats.l2_err;
1533 rx_errors += ring->stats.l3l4_csum_err;
1534 rx_crc_errors += ring->stats.l2_err;
1535 rx_multicast += ring->stats.rx_multicast;
1536 rx_length_errors += ring->stats.err_pkt_len;
1537 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1540 stats->tx_bytes = tx_bytes;
1541 stats->tx_packets = tx_pkts;
1542 stats->rx_bytes = rx_bytes;
1543 stats->rx_packets = rx_pkts;
1545 stats->rx_errors = rx_errors;
1546 stats->multicast = rx_multicast;
1547 stats->rx_length_errors = rx_length_errors;
1548 stats->rx_crc_errors = rx_crc_errors;
1549 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
1551 stats->tx_errors = tx_errors;
1552 stats->rx_dropped = rx_drop;
1553 stats->tx_dropped = tx_drop;
1554 stats->collisions = netdev->stats.collisions;
1555 stats->rx_over_errors = netdev->stats.rx_over_errors;
1556 stats->rx_frame_errors = netdev->stats.rx_frame_errors;
1557 stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
1558 stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
1559 stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
1560 stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
1561 stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
1562 stats->tx_window_errors = netdev->stats.tx_window_errors;
1563 stats->rx_compressed = netdev->stats.rx_compressed;
1564 stats->tx_compressed = netdev->stats.tx_compressed;
1567 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
1569 struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
1570 u8 *prio_tc = mqprio_qopt->qopt.prio_tc_map;
1571 struct hnae3_knic_private_info *kinfo;
1572 u8 tc = mqprio_qopt->qopt.num_tc;
1573 u16 mode = mqprio_qopt->mode;
1574 u8 hw = mqprio_qopt->qopt.hw;
1575 struct hnae3_handle *h;
1577 if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
1578 mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
1581 if (tc > HNAE3_MAX_TC)
1587 h = hns3_get_handle(netdev);
1590 netif_dbg(h, drv, netdev, "setup tc: num_tc=%u\n", tc);
1592 return (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
1593 kinfo->dcb_ops->setup_tc(h, tc, prio_tc) : -EOPNOTSUPP;
1596 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
1599 if (type != TC_SETUP_QDISC_MQPRIO)
1602 return hns3_setup_tc(dev, type_data);
1605 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
1606 __be16 proto, u16 vid)
1608 struct hnae3_handle *h = hns3_get_handle(netdev);
1611 if (h->ae_algo->ops->set_vlan_filter)
1612 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
1617 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
1618 __be16 proto, u16 vid)
1620 struct hnae3_handle *h = hns3_get_handle(netdev);
1623 if (h->ae_algo->ops->set_vlan_filter)
1624 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
1629 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
1630 u8 qos, __be16 vlan_proto)
1632 struct hnae3_handle *h = hns3_get_handle(netdev);
1635 netif_dbg(h, drv, netdev,
1636 "set vf vlan: vf=%d, vlan=%u, qos=%u, vlan_proto=%u\n",
1637 vf, vlan, qos, vlan_proto);
1639 if (h->ae_algo->ops->set_vf_vlan_filter)
1640 ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
1646 static int hns3_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
1648 struct hnae3_handle *handle = hns3_get_handle(netdev);
1650 if (hns3_nic_resetting(netdev))
1653 if (!handle->ae_algo->ops->set_vf_spoofchk)
1656 return handle->ae_algo->ops->set_vf_spoofchk(handle, vf, enable);
1659 static int hns3_set_vf_trust(struct net_device *netdev, int vf, bool enable)
1661 struct hnae3_handle *handle = hns3_get_handle(netdev);
1663 if (!handle->ae_algo->ops->set_vf_trust)
1666 return handle->ae_algo->ops->set_vf_trust(handle, vf, enable);
1669 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
1671 struct hnae3_handle *h = hns3_get_handle(netdev);
1674 if (hns3_nic_resetting(netdev))
1677 if (!h->ae_algo->ops->set_mtu)
1680 netif_dbg(h, drv, netdev,
1681 "change mtu from %u to %d\n", netdev->mtu, new_mtu);
1683 ret = h->ae_algo->ops->set_mtu(h, new_mtu);
1685 netdev_err(netdev, "failed to change MTU in hardware %d\n",
1688 netdev->mtu = new_mtu;
1693 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
1695 struct hns3_nic_priv *priv = netdev_priv(ndev);
1696 struct hnae3_handle *h = hns3_get_handle(ndev);
1697 struct hns3_enet_ring *tx_ring = NULL;
1698 struct napi_struct *napi;
1699 int timeout_queue = 0;
1700 int hw_head, hw_tail;
1701 int fbd_num, fbd_oft;
1702 int ebd_num, ebd_oft;
1707 /* Find the stopped queue the same way the stack does */
1708 for (i = 0; i < ndev->num_tx_queues; i++) {
1709 struct netdev_queue *q;
1710 unsigned long trans_start;
1712 q = netdev_get_tx_queue(ndev, i);
1713 trans_start = q->trans_start;
1714 if (netif_xmit_stopped(q) &&
1716 (trans_start + ndev->watchdog_timeo))) {
1722 if (i == ndev->num_tx_queues) {
1724 "no netdev TX timeout queue found, timeout count: %llu\n",
1725 priv->tx_timeout_count);
1729 priv->tx_timeout_count++;
1731 tx_ring = priv->ring_data[timeout_queue].ring;
1732 napi = &tx_ring->tqp_vector->napi;
1735 "tx_timeout count: %llu, queue id: %d, SW_NTU: 0x%x, SW_NTC: 0x%x, napi state: %lu\n",
1736 priv->tx_timeout_count, timeout_queue, tx_ring->next_to_use,
1737 tx_ring->next_to_clean, napi->state);
1740 "tx_pkts: %llu, tx_bytes: %llu, io_err_cnt: %llu, sw_err_cnt: %llu\n",
1741 tx_ring->stats.tx_pkts, tx_ring->stats.tx_bytes,
1742 tx_ring->stats.io_err_cnt, tx_ring->stats.sw_err_cnt);
1745 "seg_pkt_cnt: %llu, tx_err_cnt: %llu, restart_queue: %llu, tx_busy: %llu\n",
1746 tx_ring->stats.seg_pkt_cnt, tx_ring->stats.tx_err_cnt,
1747 tx_ring->stats.restart_queue, tx_ring->stats.tx_busy);
1749 /* When mac received many pause frames continuous, it's unable to send
1750 * packets, which may cause tx timeout
1752 if (h->ae_algo->ops->get_mac_stats) {
1753 struct hns3_mac_stats mac_stats;
1755 h->ae_algo->ops->get_mac_stats(h, &mac_stats);
1756 netdev_info(ndev, "tx_pause_cnt: %llu, rx_pause_cnt: %llu\n",
1757 mac_stats.tx_pause_cnt, mac_stats.rx_pause_cnt);
1760 hw_head = readl_relaxed(tx_ring->tqp->io_base +
1761 HNS3_RING_TX_RING_HEAD_REG);
1762 hw_tail = readl_relaxed(tx_ring->tqp->io_base +
1763 HNS3_RING_TX_RING_TAIL_REG);
1764 fbd_num = readl_relaxed(tx_ring->tqp->io_base +
1765 HNS3_RING_TX_RING_FBDNUM_REG);
1766 fbd_oft = readl_relaxed(tx_ring->tqp->io_base +
1767 HNS3_RING_TX_RING_OFFSET_REG);
1768 ebd_num = readl_relaxed(tx_ring->tqp->io_base +
1769 HNS3_RING_TX_RING_EBDNUM_REG);
1770 ebd_oft = readl_relaxed(tx_ring->tqp->io_base +
1771 HNS3_RING_TX_RING_EBD_OFFSET_REG);
1772 bd_num = readl_relaxed(tx_ring->tqp->io_base +
1773 HNS3_RING_TX_RING_BD_NUM_REG);
1774 bd_err = readl_relaxed(tx_ring->tqp->io_base +
1775 HNS3_RING_TX_RING_BD_ERR_REG);
1776 ring_en = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_EN_REG);
1777 tc = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_TX_RING_TC_REG);
1780 "BD_NUM: 0x%x HW_HEAD: 0x%x, HW_TAIL: 0x%x, BD_ERR: 0x%x, INT: 0x%x\n",
1781 bd_num, hw_head, hw_tail, bd_err,
1782 readl(tx_ring->tqp_vector->mask_addr));
1784 "RING_EN: 0x%x, TC: 0x%x, FBD_NUM: 0x%x FBD_OFT: 0x%x, EBD_NUM: 0x%x, EBD_OFT: 0x%x\n",
1785 ring_en, tc, fbd_num, fbd_oft, ebd_num, ebd_oft);
1790 static void hns3_nic_net_timeout(struct net_device *ndev)
1792 struct hns3_nic_priv *priv = netdev_priv(ndev);
1793 struct hnae3_handle *h = priv->ae_handle;
1795 if (!hns3_get_tx_timeo_queue_info(ndev))
1798 /* request the reset, and let the hclge to determine
1799 * which reset level should be done
1801 if (h->ae_algo->ops->reset_event)
1802 h->ae_algo->ops->reset_event(h->pdev, h);
1805 #ifdef CONFIG_RFS_ACCEL
1806 static int hns3_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
1807 u16 rxq_index, u32 flow_id)
1809 struct hnae3_handle *h = hns3_get_handle(dev);
1810 struct flow_keys fkeys;
1812 if (!h->ae_algo->ops->add_arfs_entry)
1815 if (skb->encapsulation)
1816 return -EPROTONOSUPPORT;
1818 if (!skb_flow_dissect_flow_keys(skb, &fkeys, 0))
1819 return -EPROTONOSUPPORT;
1821 if ((fkeys.basic.n_proto != htons(ETH_P_IP) &&
1822 fkeys.basic.n_proto != htons(ETH_P_IPV6)) ||
1823 (fkeys.basic.ip_proto != IPPROTO_TCP &&
1824 fkeys.basic.ip_proto != IPPROTO_UDP))
1825 return -EPROTONOSUPPORT;
1827 return h->ae_algo->ops->add_arfs_entry(h, rxq_index, flow_id, &fkeys);
1831 static int hns3_nic_get_vf_config(struct net_device *ndev, int vf,
1832 struct ifla_vf_info *ivf)
1834 struct hnae3_handle *h = hns3_get_handle(ndev);
1836 if (!h->ae_algo->ops->get_vf_config)
1839 return h->ae_algo->ops->get_vf_config(h, vf, ivf);
1842 static int hns3_nic_set_vf_link_state(struct net_device *ndev, int vf,
1845 struct hnae3_handle *h = hns3_get_handle(ndev);
1847 if (!h->ae_algo->ops->set_vf_link_state)
1850 return h->ae_algo->ops->set_vf_link_state(h, vf, link_state);
1853 static int hns3_nic_set_vf_rate(struct net_device *ndev, int vf,
1854 int min_tx_rate, int max_tx_rate)
1856 struct hnae3_handle *h = hns3_get_handle(ndev);
1858 if (!h->ae_algo->ops->set_vf_rate)
1861 return h->ae_algo->ops->set_vf_rate(h, vf, min_tx_rate, max_tx_rate,
1865 static const struct net_device_ops hns3_nic_netdev_ops = {
1866 .ndo_open = hns3_nic_net_open,
1867 .ndo_stop = hns3_nic_net_stop,
1868 .ndo_start_xmit = hns3_nic_net_xmit,
1869 .ndo_tx_timeout = hns3_nic_net_timeout,
1870 .ndo_set_mac_address = hns3_nic_net_set_mac_address,
1871 .ndo_do_ioctl = hns3_nic_do_ioctl,
1872 .ndo_change_mtu = hns3_nic_change_mtu,
1873 .ndo_set_features = hns3_nic_set_features,
1874 .ndo_get_stats64 = hns3_nic_get_stats64,
1875 .ndo_setup_tc = hns3_nic_setup_tc,
1876 .ndo_set_rx_mode = hns3_nic_set_rx_mode,
1877 .ndo_vlan_rx_add_vid = hns3_vlan_rx_add_vid,
1878 .ndo_vlan_rx_kill_vid = hns3_vlan_rx_kill_vid,
1879 .ndo_set_vf_vlan = hns3_ndo_set_vf_vlan,
1880 .ndo_set_vf_spoofchk = hns3_set_vf_spoofchk,
1881 .ndo_set_vf_trust = hns3_set_vf_trust,
1882 #ifdef CONFIG_RFS_ACCEL
1883 .ndo_rx_flow_steer = hns3_rx_flow_steer,
1885 .ndo_get_vf_config = hns3_nic_get_vf_config,
1886 .ndo_set_vf_link_state = hns3_nic_set_vf_link_state,
1887 .ndo_set_vf_rate = hns3_nic_set_vf_rate,
1890 bool hns3_is_phys_func(struct pci_dev *pdev)
1892 u32 dev_id = pdev->device;
1895 case HNAE3_DEV_ID_GE:
1896 case HNAE3_DEV_ID_25GE:
1897 case HNAE3_DEV_ID_25GE_RDMA:
1898 case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
1899 case HNAE3_DEV_ID_50GE_RDMA:
1900 case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
1901 case HNAE3_DEV_ID_100G_RDMA_MACSEC:
1903 case HNAE3_DEV_ID_100G_VF:
1904 case HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF:
1907 dev_warn(&pdev->dev, "un-recognized pci device-id %d",
1914 static void hns3_disable_sriov(struct pci_dev *pdev)
1916 /* If our VFs are assigned we cannot shut down SR-IOV
1917 * without causing issues, so just leave the hardware
1918 * available but disabled
1920 if (pci_vfs_assigned(pdev)) {
1921 dev_warn(&pdev->dev,
1922 "disabling driver while VFs are assigned\n");
1926 pci_disable_sriov(pdev);
1929 static void hns3_get_dev_capability(struct pci_dev *pdev,
1930 struct hnae3_ae_dev *ae_dev)
1932 if (pdev->revision >= 0x21) {
1933 hnae3_set_bit(ae_dev->flag, HNAE3_DEV_SUPPORT_FD_B, 1);
1934 hnae3_set_bit(ae_dev->flag, HNAE3_DEV_SUPPORT_GRO_B, 1);
1938 /* hns3_probe - Device initialization routine
1939 * @pdev: PCI device information struct
1940 * @ent: entry in hns3_pci_tbl
1942 * hns3_probe initializes a PF identified by a pci_dev structure.
1943 * The OS initialization, configuring of the PF private structure,
1944 * and a hardware reset occur.
1946 * Returns 0 on success, negative on failure
1948 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1950 struct hnae3_ae_dev *ae_dev;
1953 ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev), GFP_KERNEL);
1959 ae_dev->pdev = pdev;
1960 ae_dev->flag = ent->driver_data;
1961 ae_dev->reset_type = HNAE3_NONE_RESET;
1962 hns3_get_dev_capability(pdev, ae_dev);
1963 pci_set_drvdata(pdev, ae_dev);
1965 ret = hnae3_register_ae_dev(ae_dev);
1967 devm_kfree(&pdev->dev, ae_dev);
1968 pci_set_drvdata(pdev, NULL);
1974 /* hns3_remove - Device removal routine
1975 * @pdev: PCI device information struct
1977 static void hns3_remove(struct pci_dev *pdev)
1979 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1981 if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
1982 hns3_disable_sriov(pdev);
1984 hnae3_unregister_ae_dev(ae_dev);
1985 pci_set_drvdata(pdev, NULL);
1989 * hns3_pci_sriov_configure
1990 * @pdev: pointer to a pci_dev structure
1991 * @num_vfs: number of VFs to allocate
1993 * Enable or change the number of VFs. Called when the user updates the number
1996 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
2000 if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
2001 dev_warn(&pdev->dev, "Can not config SRIOV\n");
2006 ret = pci_enable_sriov(pdev, num_vfs);
2008 dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
2011 } else if (!pci_vfs_assigned(pdev)) {
2012 pci_disable_sriov(pdev);
2014 dev_warn(&pdev->dev,
2015 "Unable to free VFs because some are assigned to VMs.\n");
2021 static void hns3_shutdown(struct pci_dev *pdev)
2023 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2025 hnae3_unregister_ae_dev(ae_dev);
2026 devm_kfree(&pdev->dev, ae_dev);
2027 pci_set_drvdata(pdev, NULL);
2029 if (system_state == SYSTEM_POWER_OFF)
2030 pci_set_power_state(pdev, PCI_D3hot);
2033 static pci_ers_result_t hns3_error_detected(struct pci_dev *pdev,
2034 pci_channel_state_t state)
2036 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2037 pci_ers_result_t ret;
2039 dev_info(&pdev->dev, "PCI error detected, state(=%d)!!\n", state);
2041 if (state == pci_channel_io_perm_failure)
2042 return PCI_ERS_RESULT_DISCONNECT;
2044 if (!ae_dev || !ae_dev->ops) {
2046 "Can't recover - error happened before device initialized\n");
2047 return PCI_ERS_RESULT_NONE;
2050 if (ae_dev->ops->handle_hw_ras_error)
2051 ret = ae_dev->ops->handle_hw_ras_error(ae_dev);
2053 return PCI_ERS_RESULT_NONE;
2058 static pci_ers_result_t hns3_slot_reset(struct pci_dev *pdev)
2060 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2061 const struct hnae3_ae_ops *ops;
2062 enum hnae3_reset_type reset_type;
2063 struct device *dev = &pdev->dev;
2065 if (!ae_dev || !ae_dev->ops)
2066 return PCI_ERS_RESULT_NONE;
2069 /* request the reset */
2070 if (ops->reset_event && ops->get_reset_level &&
2071 ops->set_default_reset_request) {
2072 if (ae_dev->hw_err_reset_req) {
2073 reset_type = ops->get_reset_level(ae_dev,
2074 &ae_dev->hw_err_reset_req);
2075 ops->set_default_reset_request(ae_dev, reset_type);
2076 dev_info(dev, "requesting reset due to PCI error\n");
2077 ops->reset_event(pdev, NULL);
2080 return PCI_ERS_RESULT_RECOVERED;
2083 return PCI_ERS_RESULT_DISCONNECT;
2086 static void hns3_reset_prepare(struct pci_dev *pdev)
2088 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2090 dev_info(&pdev->dev, "hns3 flr prepare\n");
2091 if (ae_dev && ae_dev->ops && ae_dev->ops->flr_prepare)
2092 ae_dev->ops->flr_prepare(ae_dev);
2095 static void hns3_reset_done(struct pci_dev *pdev)
2097 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2099 dev_info(&pdev->dev, "hns3 flr done\n");
2100 if (ae_dev && ae_dev->ops && ae_dev->ops->flr_done)
2101 ae_dev->ops->flr_done(ae_dev);
2104 static const struct pci_error_handlers hns3_err_handler = {
2105 .error_detected = hns3_error_detected,
2106 .slot_reset = hns3_slot_reset,
2107 .reset_prepare = hns3_reset_prepare,
2108 .reset_done = hns3_reset_done,
2111 static struct pci_driver hns3_driver = {
2112 .name = hns3_driver_name,
2113 .id_table = hns3_pci_tbl,
2114 .probe = hns3_probe,
2115 .remove = hns3_remove,
2116 .shutdown = hns3_shutdown,
2117 .sriov_configure = hns3_pci_sriov_configure,
2118 .err_handler = &hns3_err_handler,
2121 /* set default feature to hns3 */
2122 static void hns3_set_default_feature(struct net_device *netdev)
2124 struct hnae3_handle *h = hns3_get_handle(netdev);
2125 struct pci_dev *pdev = h->pdev;
2127 netdev->priv_flags |= IFF_UNICAST_FLT;
2129 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2130 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2131 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2132 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2133 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
2135 netdev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
2137 netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
2139 netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2140 NETIF_F_HW_VLAN_CTAG_FILTER |
2141 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
2142 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2143 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2144 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2145 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
2147 netdev->vlan_features |=
2148 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
2149 NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO |
2150 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2151 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2152 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
2154 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2155 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
2156 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2157 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2158 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2159 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
2161 if (pdev->revision >= 0x21) {
2162 netdev->hw_features |= NETIF_F_GRO_HW;
2163 netdev->features |= NETIF_F_GRO_HW;
2165 if (!(h->flags & HNAE3_SUPPORT_VF)) {
2166 netdev->hw_features |= NETIF_F_NTUPLE;
2167 netdev->features |= NETIF_F_NTUPLE;
2172 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
2173 struct hns3_desc_cb *cb)
2175 unsigned int order = hns3_page_order(ring);
2178 p = dev_alloc_pages(order);
2183 cb->page_offset = 0;
2185 cb->buf = page_address(p);
2186 cb->length = hns3_page_size(ring);
2187 cb->type = DESC_TYPE_PAGE;
2192 static void hns3_free_buffer(struct hns3_enet_ring *ring,
2193 struct hns3_desc_cb *cb)
2195 if (cb->type == DESC_TYPE_SKB)
2196 dev_kfree_skb_any((struct sk_buff *)cb->priv);
2197 else if (!HNAE3_IS_TX_RING(ring))
2198 put_page((struct page *)cb->priv);
2199 memset(cb, 0, sizeof(*cb));
2202 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
2204 cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
2205 cb->length, ring_to_dma_dir(ring));
2207 if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
2213 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
2214 struct hns3_desc_cb *cb)
2216 if (cb->type == DESC_TYPE_SKB)
2217 dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
2218 ring_to_dma_dir(ring));
2219 else if (cb->length)
2220 dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
2221 ring_to_dma_dir(ring));
2224 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
2226 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2227 ring->desc[i].addr = 0;
2230 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i)
2232 struct hns3_desc_cb *cb = &ring->desc_cb[i];
2234 if (!ring->desc_cb[i].dma)
2237 hns3_buffer_detach(ring, i);
2238 hns3_free_buffer(ring, cb);
2241 static void hns3_free_buffers(struct hns3_enet_ring *ring)
2245 for (i = 0; i < ring->desc_num; i++)
2246 hns3_free_buffer_detach(ring, i);
2249 /* free desc along with its attached buffer */
2250 static void hns3_free_desc(struct hns3_enet_ring *ring)
2252 int size = ring->desc_num * sizeof(ring->desc[0]);
2254 hns3_free_buffers(ring);
2257 dma_free_coherent(ring_to_dev(ring), size,
2258 ring->desc, ring->desc_dma_addr);
2263 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
2265 int size = ring->desc_num * sizeof(ring->desc[0]);
2267 ring->desc = dma_alloc_coherent(ring_to_dev(ring), size,
2268 &ring->desc_dma_addr, GFP_KERNEL);
2275 static int hns3_reserve_buffer_map(struct hns3_enet_ring *ring,
2276 struct hns3_desc_cb *cb)
2280 ret = hns3_alloc_buffer(ring, cb);
2284 ret = hns3_map_buffer(ring, cb);
2291 hns3_free_buffer(ring, cb);
2296 static int hns3_alloc_buffer_attach(struct hns3_enet_ring *ring, int i)
2298 int ret = hns3_reserve_buffer_map(ring, &ring->desc_cb[i]);
2303 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2308 /* Allocate memory for raw pkg, and map with dma */
2309 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
2313 for (i = 0; i < ring->desc_num; i++) {
2314 ret = hns3_alloc_buffer_attach(ring, i);
2316 goto out_buffer_fail;
2322 for (j = i - 1; j >= 0; j--)
2323 hns3_free_buffer_detach(ring, j);
2327 /* detach a in-used buffer and replace with a reserved one */
2328 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
2329 struct hns3_desc_cb *res_cb)
2331 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2332 ring->desc_cb[i] = *res_cb;
2333 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2334 ring->desc[i].rx.bd_base_info = 0;
2337 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
2339 ring->desc_cb[i].reuse_flag = 0;
2340 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
2341 ring->desc_cb[i].page_offset);
2342 ring->desc[i].rx.bd_base_info = 0;
2345 static void hns3_nic_reclaim_desc(struct hns3_enet_ring *ring, int head,
2346 int *bytes, int *pkts)
2348 int ntc = ring->next_to_clean;
2349 struct hns3_desc_cb *desc_cb;
2351 while (head != ntc) {
2352 desc_cb = &ring->desc_cb[ntc];
2353 (*pkts) += (desc_cb->type == DESC_TYPE_SKB);
2354 (*bytes) += desc_cb->length;
2355 /* desc_cb will be cleaned, after hnae3_free_buffer_detach */
2356 hns3_free_buffer_detach(ring, ntc);
2358 if (++ntc == ring->desc_num)
2361 /* Issue prefetch for next Tx descriptor */
2362 prefetch(&ring->desc_cb[ntc]);
2365 /* This smp_store_release() pairs with smp_load_acquire() in
2366 * ring_space called by hns3_nic_net_xmit.
2368 smp_store_release(&ring->next_to_clean, ntc);
2371 static int is_valid_clean_head(struct hns3_enet_ring *ring, int h)
2373 int u = ring->next_to_use;
2374 int c = ring->next_to_clean;
2376 if (unlikely(h > ring->desc_num))
2379 return u > c ? (h > c && h <= u) : (h > c || h <= u);
2382 void hns3_clean_tx_ring(struct hns3_enet_ring *ring)
2384 struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2385 struct hns3_nic_priv *priv = netdev_priv(netdev);
2386 struct netdev_queue *dev_queue;
2390 head = readl_relaxed(ring->tqp->io_base + HNS3_RING_TX_RING_HEAD_REG);
2391 rmb(); /* Make sure head is ready before touch any data */
2393 if (is_ring_empty(ring) || head == ring->next_to_clean)
2394 return; /* no data to poll */
2396 if (unlikely(!is_valid_clean_head(ring, head))) {
2397 netdev_err(netdev, "wrong head (%d, %d-%d)\n", head,
2398 ring->next_to_use, ring->next_to_clean);
2400 u64_stats_update_begin(&ring->syncp);
2401 ring->stats.io_err_cnt++;
2402 u64_stats_update_end(&ring->syncp);
2408 hns3_nic_reclaim_desc(ring, head, &bytes, &pkts);
2410 ring->tqp_vector->tx_group.total_bytes += bytes;
2411 ring->tqp_vector->tx_group.total_packets += pkts;
2413 u64_stats_update_begin(&ring->syncp);
2414 ring->stats.tx_bytes += bytes;
2415 ring->stats.tx_pkts += pkts;
2416 u64_stats_update_end(&ring->syncp);
2418 dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
2419 netdev_tx_completed_queue(dev_queue, pkts, bytes);
2421 if (unlikely(pkts && netif_carrier_ok(netdev) &&
2422 (ring_space(ring) > HNS3_MAX_BD_PER_PKT))) {
2423 /* Make sure that anybody stopping the queue after this
2424 * sees the new next_to_clean.
2427 if (netif_tx_queue_stopped(dev_queue) &&
2428 !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
2429 netif_tx_wake_queue(dev_queue);
2430 ring->stats.restart_queue++;
2435 static int hns3_desc_unused(struct hns3_enet_ring *ring)
2437 int ntc = ring->next_to_clean;
2438 int ntu = ring->next_to_use;
2440 return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
2443 static void hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring,
2446 struct hns3_desc_cb *desc_cb;
2447 struct hns3_desc_cb res_cbs;
2450 for (i = 0; i < cleand_count; i++) {
2451 desc_cb = &ring->desc_cb[ring->next_to_use];
2452 if (desc_cb->reuse_flag) {
2453 u64_stats_update_begin(&ring->syncp);
2454 ring->stats.reuse_pg_cnt++;
2455 u64_stats_update_end(&ring->syncp);
2457 hns3_reuse_buffer(ring, ring->next_to_use);
2459 ret = hns3_reserve_buffer_map(ring, &res_cbs);
2461 u64_stats_update_begin(&ring->syncp);
2462 ring->stats.sw_err_cnt++;
2463 u64_stats_update_end(&ring->syncp);
2465 hns3_rl_err(ring->tqp_vector->napi.dev,
2466 "alloc rx buffer failed: %d\n",
2470 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
2472 u64_stats_update_begin(&ring->syncp);
2473 ring->stats.non_reuse_pg++;
2474 u64_stats_update_end(&ring->syncp);
2477 ring_ptr_move_fw(ring, next_to_use);
2480 wmb(); /* Make all data has been write before submit */
2481 writel_relaxed(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
2484 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
2485 struct hns3_enet_ring *ring, int pull_len,
2486 struct hns3_desc_cb *desc_cb)
2488 struct hns3_desc *desc = &ring->desc[ring->next_to_clean];
2489 int size = le16_to_cpu(desc->rx.size);
2490 u32 truesize = hns3_buf_size(ring);
2492 skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
2493 size - pull_len, truesize);
2495 /* Avoid re-using remote pages, or the stack is still using the page
2496 * when page_offset rollback to zero, flag default unreuse
2498 if (unlikely(page_to_nid(desc_cb->priv) != numa_mem_id()) ||
2499 (!desc_cb->page_offset && page_count(desc_cb->priv) > 1))
2502 /* Move offset up to the next cache line */
2503 desc_cb->page_offset += truesize;
2505 if (desc_cb->page_offset + truesize <= hns3_page_size(ring)) {
2506 desc_cb->reuse_flag = 1;
2507 /* Bump ref count on page before it is given */
2508 get_page(desc_cb->priv);
2509 } else if (page_count(desc_cb->priv) == 1) {
2510 desc_cb->reuse_flag = 1;
2511 desc_cb->page_offset = 0;
2512 get_page(desc_cb->priv);
2516 static int hns3_gro_complete(struct sk_buff *skb, u32 l234info)
2518 __be16 type = skb->protocol;
2522 while (eth_type_vlan(type)) {
2523 struct vlan_hdr *vh;
2525 if ((depth + VLAN_HLEN) > skb_headlen(skb))
2528 vh = (struct vlan_hdr *)(skb->data + depth);
2529 type = vh->h_vlan_encapsulated_proto;
2533 skb_set_network_header(skb, depth);
2535 if (type == htons(ETH_P_IP)) {
2536 const struct iphdr *iph = ip_hdr(skb);
2538 depth += sizeof(struct iphdr);
2539 skb_set_transport_header(skb, depth);
2541 th->check = ~tcp_v4_check(skb->len - depth, iph->saddr,
2543 } else if (type == htons(ETH_P_IPV6)) {
2544 const struct ipv6hdr *iph = ipv6_hdr(skb);
2546 depth += sizeof(struct ipv6hdr);
2547 skb_set_transport_header(skb, depth);
2549 th->check = ~tcp_v6_check(skb->len - depth, &iph->saddr,
2552 hns3_rl_err(skb->dev,
2553 "Error: FW GRO supports only IPv4/IPv6, not 0x%04x, depth: %d\n",
2554 be16_to_cpu(type), depth);
2558 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
2560 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
2562 if (l234info & BIT(HNS3_RXD_GRO_FIXID_B))
2563 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_FIXEDID;
2565 skb->csum_start = (unsigned char *)th - skb->head;
2566 skb->csum_offset = offsetof(struct tcphdr, check);
2567 skb->ip_summed = CHECKSUM_PARTIAL;
2571 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
2572 u32 l234info, u32 bd_base_info, u32 ol_info)
2574 struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2575 int l3_type, l4_type;
2578 skb->ip_summed = CHECKSUM_NONE;
2580 skb_checksum_none_assert(skb);
2582 if (!(netdev->features & NETIF_F_RXCSUM))
2585 /* check if hardware has done checksum */
2586 if (!(bd_base_info & BIT(HNS3_RXD_L3L4P_B)))
2589 if (unlikely(l234info & (BIT(HNS3_RXD_L3E_B) | BIT(HNS3_RXD_L4E_B) |
2590 BIT(HNS3_RXD_OL3E_B) |
2591 BIT(HNS3_RXD_OL4E_B)))) {
2592 u64_stats_update_begin(&ring->syncp);
2593 ring->stats.l3l4_csum_err++;
2594 u64_stats_update_end(&ring->syncp);
2599 ol4_type = hnae3_get_field(ol_info, HNS3_RXD_OL4ID_M,
2602 case HNS3_OL4_TYPE_MAC_IN_UDP:
2603 case HNS3_OL4_TYPE_NVGRE:
2604 skb->csum_level = 1;
2606 case HNS3_OL4_TYPE_NO_TUN:
2607 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
2609 l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
2612 /* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
2613 if ((l3_type == HNS3_L3_TYPE_IPV4 ||
2614 l3_type == HNS3_L3_TYPE_IPV6) &&
2615 (l4_type == HNS3_L4_TYPE_UDP ||
2616 l4_type == HNS3_L4_TYPE_TCP ||
2617 l4_type == HNS3_L4_TYPE_SCTP))
2618 skb->ip_summed = CHECKSUM_UNNECESSARY;
2625 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
2627 if (skb_has_frag_list(skb))
2628 napi_gro_flush(&ring->tqp_vector->napi, false);
2630 napi_gro_receive(&ring->tqp_vector->napi, skb);
2633 static bool hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
2634 struct hns3_desc *desc, u32 l234info,
2637 struct hnae3_handle *handle = ring->tqp->handle;
2638 struct pci_dev *pdev = ring->tqp->handle->pdev;
2640 if (pdev->revision == 0x20) {
2641 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2642 if (!(*vlan_tag & VLAN_VID_MASK))
2643 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2645 return (*vlan_tag != 0);
2648 #define HNS3_STRP_OUTER_VLAN 0x1
2649 #define HNS3_STRP_INNER_VLAN 0x2
2650 #define HNS3_STRP_BOTH 0x3
2652 /* Hardware always insert VLAN tag into RX descriptor when
2653 * remove the tag from packet, driver needs to determine
2654 * reporting which tag to stack.
2656 switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
2657 HNS3_RXD_STRP_TAGP_S)) {
2658 case HNS3_STRP_OUTER_VLAN:
2659 if (handle->port_base_vlan_state !=
2660 HNAE3_PORT_BASE_VLAN_DISABLE)
2663 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2665 case HNS3_STRP_INNER_VLAN:
2666 if (handle->port_base_vlan_state !=
2667 HNAE3_PORT_BASE_VLAN_DISABLE)
2670 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2672 case HNS3_STRP_BOTH:
2673 if (handle->port_base_vlan_state ==
2674 HNAE3_PORT_BASE_VLAN_DISABLE)
2675 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2677 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2685 static int hns3_alloc_skb(struct hns3_enet_ring *ring, unsigned int length,
2688 #define HNS3_NEED_ADD_FRAG 1
2689 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
2690 struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2691 struct sk_buff *skb;
2693 ring->skb = napi_alloc_skb(&ring->tqp_vector->napi, HNS3_RX_HEAD_SIZE);
2695 if (unlikely(!skb)) {
2696 hns3_rl_err(netdev, "alloc rx skb fail\n");
2698 u64_stats_update_begin(&ring->syncp);
2699 ring->stats.sw_err_cnt++;
2700 u64_stats_update_end(&ring->syncp);
2705 prefetchw(skb->data);
2707 ring->pending_buf = 1;
2709 ring->tail_skb = NULL;
2710 if (length <= HNS3_RX_HEAD_SIZE) {
2711 memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
2713 /* We can reuse buffer as-is, just make sure it is local */
2714 if (likely(page_to_nid(desc_cb->priv) == numa_mem_id()))
2715 desc_cb->reuse_flag = 1;
2716 else /* This page cannot be reused so discard it */
2717 put_page(desc_cb->priv);
2719 ring_ptr_move_fw(ring, next_to_clean);
2722 u64_stats_update_begin(&ring->syncp);
2723 ring->stats.seg_pkt_cnt++;
2724 u64_stats_update_end(&ring->syncp);
2726 ring->pull_len = eth_get_headlen(netdev, va, HNS3_RX_HEAD_SIZE);
2727 __skb_put(skb, ring->pull_len);
2728 hns3_nic_reuse_page(skb, ring->frag_num++, ring, ring->pull_len,
2730 ring_ptr_move_fw(ring, next_to_clean);
2732 return HNS3_NEED_ADD_FRAG;
2735 static int hns3_add_frag(struct hns3_enet_ring *ring, struct hns3_desc *desc,
2736 struct sk_buff **out_skb, bool pending)
2738 struct sk_buff *skb = *out_skb;
2739 struct sk_buff *head_skb = *out_skb;
2740 struct sk_buff *new_skb;
2741 struct hns3_desc_cb *desc_cb;
2742 struct hns3_desc *pre_desc;
2746 /* if there is pending bd, the SW param next_to_clean has moved
2747 * to next and the next is NULL
2750 pre_bd = (ring->next_to_clean - 1 + ring->desc_num) %
2752 pre_desc = &ring->desc[pre_bd];
2753 bd_base_info = le32_to_cpu(pre_desc->rx.bd_base_info);
2755 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2758 while (!(bd_base_info & BIT(HNS3_RXD_FE_B))) {
2759 desc = &ring->desc[ring->next_to_clean];
2760 desc_cb = &ring->desc_cb[ring->next_to_clean];
2761 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2762 /* make sure HW write desc complete */
2764 if (!(bd_base_info & BIT(HNS3_RXD_VLD_B)))
2767 if (unlikely(ring->frag_num >= MAX_SKB_FRAGS)) {
2768 new_skb = napi_alloc_skb(&ring->tqp_vector->napi,
2770 if (unlikely(!new_skb)) {
2771 hns3_rl_err(ring->tqp_vector->napi.dev,
2772 "alloc rx fraglist skb fail\n");
2777 if (ring->tail_skb) {
2778 ring->tail_skb->next = new_skb;
2779 ring->tail_skb = new_skb;
2781 skb_shinfo(skb)->frag_list = new_skb;
2782 ring->tail_skb = new_skb;
2786 if (ring->tail_skb) {
2787 head_skb->truesize += hns3_buf_size(ring);
2788 head_skb->data_len += le16_to_cpu(desc->rx.size);
2789 head_skb->len += le16_to_cpu(desc->rx.size);
2790 skb = ring->tail_skb;
2793 hns3_nic_reuse_page(skb, ring->frag_num++, ring, 0, desc_cb);
2794 ring_ptr_move_fw(ring, next_to_clean);
2795 ring->pending_buf++;
2801 static int hns3_set_gro_and_checksum(struct hns3_enet_ring *ring,
2802 struct sk_buff *skb, u32 l234info,
2803 u32 bd_base_info, u32 ol_info)
2807 skb_shinfo(skb)->gso_size = hnae3_get_field(bd_base_info,
2808 HNS3_RXD_GRO_SIZE_M,
2809 HNS3_RXD_GRO_SIZE_S);
2810 /* if there is no HW GRO, do not set gro params */
2811 if (!skb_shinfo(skb)->gso_size) {
2812 hns3_rx_checksum(ring, skb, l234info, bd_base_info, ol_info);
2816 NAPI_GRO_CB(skb)->count = hnae3_get_field(l234info,
2817 HNS3_RXD_GRO_COUNT_M,
2818 HNS3_RXD_GRO_COUNT_S);
2820 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M, HNS3_RXD_L3ID_S);
2821 if (l3_type == HNS3_L3_TYPE_IPV4)
2822 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2823 else if (l3_type == HNS3_L3_TYPE_IPV6)
2824 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
2828 return hns3_gro_complete(skb, l234info);
2831 static void hns3_set_rx_skb_rss_type(struct hns3_enet_ring *ring,
2832 struct sk_buff *skb, u32 rss_hash)
2834 struct hnae3_handle *handle = ring->tqp->handle;
2835 enum pkt_hash_types rss_type;
2838 rss_type = handle->kinfo.rss_type;
2840 rss_type = PKT_HASH_TYPE_NONE;
2842 skb_set_hash(skb, rss_hash, rss_type);
2845 static int hns3_handle_bdinfo(struct hns3_enet_ring *ring, struct sk_buff *skb)
2847 struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2848 enum hns3_pkt_l2t_type l2_frame_type;
2849 u32 bd_base_info, l234info, ol_info;
2850 struct hns3_desc *desc;
2854 /* bdinfo handled below is only valid on the last BD of the
2855 * current packet, and ring->next_to_clean indicates the first
2856 * descriptor of next packet, so need - 1 below.
2858 pre_ntc = ring->next_to_clean ? (ring->next_to_clean - 1) :
2859 (ring->desc_num - 1);
2860 desc = &ring->desc[pre_ntc];
2861 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2862 l234info = le32_to_cpu(desc->rx.l234_info);
2863 ol_info = le32_to_cpu(desc->rx.ol_info);
2865 /* Based on hw strategy, the tag offloaded will be stored at
2866 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
2867 * in one layer tag case.
2869 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
2872 if (hns3_parse_vlan_tag(ring, desc, l234info, &vlan_tag))
2873 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2877 if (unlikely(!desc->rx.pkt_len || (l234info & (BIT(HNS3_RXD_TRUNCAT_B) |
2878 BIT(HNS3_RXD_L2E_B))))) {
2879 u64_stats_update_begin(&ring->syncp);
2880 if (l234info & BIT(HNS3_RXD_L2E_B))
2881 ring->stats.l2_err++;
2883 ring->stats.err_pkt_len++;
2884 u64_stats_update_end(&ring->syncp);
2891 /* Do update ip stack process */
2892 skb->protocol = eth_type_trans(skb, netdev);
2894 /* This is needed in order to enable forwarding support */
2895 ret = hns3_set_gro_and_checksum(ring, skb, l234info,
2896 bd_base_info, ol_info);
2897 if (unlikely(ret)) {
2898 u64_stats_update_begin(&ring->syncp);
2899 ring->stats.rx_err_cnt++;
2900 u64_stats_update_end(&ring->syncp);
2904 l2_frame_type = hnae3_get_field(l234info, HNS3_RXD_DMAC_M,
2907 u64_stats_update_begin(&ring->syncp);
2908 ring->stats.rx_pkts++;
2909 ring->stats.rx_bytes += len;
2911 if (l2_frame_type == HNS3_L2_TYPE_MULTICAST)
2912 ring->stats.rx_multicast++;
2914 u64_stats_update_end(&ring->syncp);
2916 ring->tqp_vector->rx_group.total_bytes += len;
2918 hns3_set_rx_skb_rss_type(ring, skb, le32_to_cpu(desc->rx.rss_hash));
2922 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring,
2923 struct sk_buff **out_skb)
2925 struct sk_buff *skb = ring->skb;
2926 struct hns3_desc_cb *desc_cb;
2927 struct hns3_desc *desc;
2928 unsigned int length;
2932 desc = &ring->desc[ring->next_to_clean];
2933 desc_cb = &ring->desc_cb[ring->next_to_clean];
2937 length = le16_to_cpu(desc->rx.size);
2938 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2940 /* Check valid BD */
2941 if (unlikely(!(bd_base_info & BIT(HNS3_RXD_VLD_B))))
2945 ring->va = (unsigned char *)desc_cb->buf + desc_cb->page_offset;
2947 /* Prefetch first cache line of first page
2948 * Idea is to cache few bytes of the header of the packet. Our L1 Cache
2949 * line size is 64B so need to prefetch twice to make it 128B. But in
2950 * actual we can have greater size of caches with 128B Level 1 cache
2951 * lines. In such a case, single fetch would suffice to cache in the
2952 * relevant part of the header.
2955 #if L1_CACHE_BYTES < 128
2956 prefetch(ring->va + L1_CACHE_BYTES);
2960 ret = hns3_alloc_skb(ring, length, ring->va);
2961 *out_skb = skb = ring->skb;
2963 if (ret < 0) /* alloc buffer fail */
2965 if (ret > 0) { /* need add frag */
2966 ret = hns3_add_frag(ring, desc, &skb, false);
2970 /* As the head data may be changed when GRO enable, copy
2971 * the head data in after other data rx completed
2973 memcpy(skb->data, ring->va,
2974 ALIGN(ring->pull_len, sizeof(long)));
2977 ret = hns3_add_frag(ring, desc, &skb, true);
2981 /* As the head data may be changed when GRO enable, copy
2982 * the head data in after other data rx completed
2984 memcpy(skb->data, ring->va,
2985 ALIGN(ring->pull_len, sizeof(long)));
2988 ret = hns3_handle_bdinfo(ring, skb);
2989 if (unlikely(ret)) {
2990 dev_kfree_skb_any(skb);
2994 skb_record_rx_queue(skb, ring->tqp->tqp_index);
3000 int hns3_clean_rx_ring(struct hns3_enet_ring *ring, int budget,
3001 void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
3003 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
3004 int unused_count = hns3_desc_unused(ring);
3005 struct sk_buff *skb = ring->skb;
3010 num = readl_relaxed(ring->tqp->io_base + HNS3_RING_RX_RING_FBDNUM_REG);
3011 rmb(); /* Make sure num taken effect before the other data is touched */
3013 num -= unused_count;
3014 unused_count -= ring->pending_buf;
3016 while (recv_pkts < budget && recv_bds < num) {
3017 /* Reuse or realloc buffers */
3018 if (unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
3019 hns3_nic_alloc_rx_buffers(ring, unused_count);
3020 unused_count = hns3_desc_unused(ring) -
3025 err = hns3_handle_rx_bd(ring, &skb);
3026 if (unlikely(!skb)) /* This fault cannot be repaired */
3029 if (err == -ENXIO) { /* Do not get FE for the packet */
3031 } else if (unlikely(err)) { /* Do jump the err */
3032 recv_bds += ring->pending_buf;
3033 unused_count += ring->pending_buf;
3035 ring->pending_buf = 0;
3040 recv_bds += ring->pending_buf;
3041 unused_count += ring->pending_buf;
3043 ring->pending_buf = 0;
3049 /* Make all data has been write before submit */
3050 if (unused_count > 0)
3051 hns3_nic_alloc_rx_buffers(ring, unused_count);
3056 static bool hns3_get_new_flow_lvl(struct hns3_enet_ring_group *ring_group)
3058 #define HNS3_RX_LOW_BYTE_RATE 10000
3059 #define HNS3_RX_MID_BYTE_RATE 20000
3060 #define HNS3_RX_ULTRA_PACKET_RATE 40
3062 enum hns3_flow_level_range new_flow_level;
3063 struct hns3_enet_tqp_vector *tqp_vector;
3064 int packets_per_msecs, bytes_per_msecs;
3067 tqp_vector = ring_group->ring->tqp_vector;
3069 jiffies_to_msecs(jiffies - tqp_vector->last_jiffies);
3070 if (!time_passed_ms)
3073 do_div(ring_group->total_packets, time_passed_ms);
3074 packets_per_msecs = ring_group->total_packets;
3076 do_div(ring_group->total_bytes, time_passed_ms);
3077 bytes_per_msecs = ring_group->total_bytes;
3079 new_flow_level = ring_group->coal.flow_level;
3081 /* Simple throttlerate management
3082 * 0-10MB/s lower (50000 ints/s)
3083 * 10-20MB/s middle (20000 ints/s)
3084 * 20-1249MB/s high (18000 ints/s)
3085 * > 40000pps ultra (8000 ints/s)
3087 switch (new_flow_level) {
3089 if (bytes_per_msecs > HNS3_RX_LOW_BYTE_RATE)
3090 new_flow_level = HNS3_FLOW_MID;
3093 if (bytes_per_msecs > HNS3_RX_MID_BYTE_RATE)
3094 new_flow_level = HNS3_FLOW_HIGH;
3095 else if (bytes_per_msecs <= HNS3_RX_LOW_BYTE_RATE)
3096 new_flow_level = HNS3_FLOW_LOW;
3098 case HNS3_FLOW_HIGH:
3099 case HNS3_FLOW_ULTRA:
3101 if (bytes_per_msecs <= HNS3_RX_MID_BYTE_RATE)
3102 new_flow_level = HNS3_FLOW_MID;
3106 if (packets_per_msecs > HNS3_RX_ULTRA_PACKET_RATE &&
3107 &tqp_vector->rx_group == ring_group)
3108 new_flow_level = HNS3_FLOW_ULTRA;
3110 ring_group->total_bytes = 0;
3111 ring_group->total_packets = 0;
3112 ring_group->coal.flow_level = new_flow_level;
3117 static bool hns3_get_new_int_gl(struct hns3_enet_ring_group *ring_group)
3119 struct hns3_enet_tqp_vector *tqp_vector;
3122 if (!ring_group->ring)
3125 tqp_vector = ring_group->ring->tqp_vector;
3126 if (!tqp_vector->last_jiffies)
3129 if (ring_group->total_packets == 0) {
3130 ring_group->coal.int_gl = HNS3_INT_GL_50K;
3131 ring_group->coal.flow_level = HNS3_FLOW_LOW;
3135 if (!hns3_get_new_flow_lvl(ring_group))
3138 new_int_gl = ring_group->coal.int_gl;
3139 switch (ring_group->coal.flow_level) {
3141 new_int_gl = HNS3_INT_GL_50K;
3144 new_int_gl = HNS3_INT_GL_20K;
3146 case HNS3_FLOW_HIGH:
3147 new_int_gl = HNS3_INT_GL_18K;
3149 case HNS3_FLOW_ULTRA:
3150 new_int_gl = HNS3_INT_GL_8K;
3156 if (new_int_gl != ring_group->coal.int_gl) {
3157 ring_group->coal.int_gl = new_int_gl;
3163 static void hns3_update_new_int_gl(struct hns3_enet_tqp_vector *tqp_vector)
3165 struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
3166 struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
3167 bool rx_update, tx_update;
3169 /* update param every 1000ms */
3170 if (time_before(jiffies,
3171 tqp_vector->last_jiffies + msecs_to_jiffies(1000)))
3174 if (rx_group->coal.gl_adapt_enable) {
3175 rx_update = hns3_get_new_int_gl(rx_group);
3177 hns3_set_vector_coalesce_rx_gl(tqp_vector,
3178 rx_group->coal.int_gl);
3181 if (tx_group->coal.gl_adapt_enable) {
3182 tx_update = hns3_get_new_int_gl(tx_group);
3184 hns3_set_vector_coalesce_tx_gl(tqp_vector,
3185 tx_group->coal.int_gl);
3188 tqp_vector->last_jiffies = jiffies;
3191 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
3193 struct hns3_nic_priv *priv = netdev_priv(napi->dev);
3194 struct hns3_enet_ring *ring;
3195 int rx_pkt_total = 0;
3197 struct hns3_enet_tqp_vector *tqp_vector =
3198 container_of(napi, struct hns3_enet_tqp_vector, napi);
3199 bool clean_complete = true;
3200 int rx_budget = budget;
3202 if (unlikely(test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
3203 napi_complete(napi);
3207 /* Since the actual Tx work is minimal, we can give the Tx a larger
3208 * budget and be more aggressive about cleaning up the Tx descriptors.
3210 hns3_for_each_ring(ring, tqp_vector->tx_group)
3211 hns3_clean_tx_ring(ring);
3213 /* make sure rx ring budget not smaller than 1 */
3214 if (tqp_vector->num_tqps > 1)
3215 rx_budget = max(budget / tqp_vector->num_tqps, 1);
3217 hns3_for_each_ring(ring, tqp_vector->rx_group) {
3218 int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
3221 if (rx_cleaned >= rx_budget)
3222 clean_complete = false;
3224 rx_pkt_total += rx_cleaned;
3227 tqp_vector->rx_group.total_packets += rx_pkt_total;
3229 if (!clean_complete)
3232 if (napi_complete(napi) &&
3233 likely(!test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
3234 hns3_update_new_int_gl(tqp_vector);
3235 hns3_mask_vector_irq(tqp_vector, 1);
3238 return rx_pkt_total;
3241 static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
3242 struct hnae3_ring_chain_node *head)
3244 struct pci_dev *pdev = tqp_vector->handle->pdev;
3245 struct hnae3_ring_chain_node *cur_chain = head;
3246 struct hnae3_ring_chain_node *chain;
3247 struct hns3_enet_ring *tx_ring;
3248 struct hns3_enet_ring *rx_ring;
3250 tx_ring = tqp_vector->tx_group.ring;
3252 cur_chain->tqp_index = tx_ring->tqp->tqp_index;
3253 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
3254 HNAE3_RING_TYPE_TX);
3255 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3256 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_TX);
3258 cur_chain->next = NULL;
3260 while (tx_ring->next) {
3261 tx_ring = tx_ring->next;
3263 chain = devm_kzalloc(&pdev->dev, sizeof(*chain),
3266 goto err_free_chain;
3268 cur_chain->next = chain;
3269 chain->tqp_index = tx_ring->tqp->tqp_index;
3270 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
3271 HNAE3_RING_TYPE_TX);
3272 hnae3_set_field(chain->int_gl_idx,
3273 HNAE3_RING_GL_IDX_M,
3274 HNAE3_RING_GL_IDX_S,
3281 rx_ring = tqp_vector->rx_group.ring;
3282 if (!tx_ring && rx_ring) {
3283 cur_chain->next = NULL;
3284 cur_chain->tqp_index = rx_ring->tqp->tqp_index;
3285 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
3286 HNAE3_RING_TYPE_RX);
3287 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3288 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
3290 rx_ring = rx_ring->next;
3294 chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
3296 goto err_free_chain;
3298 cur_chain->next = chain;
3299 chain->tqp_index = rx_ring->tqp->tqp_index;
3300 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
3301 HNAE3_RING_TYPE_RX);
3302 hnae3_set_field(chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3303 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
3307 rx_ring = rx_ring->next;
3313 cur_chain = head->next;
3315 chain = cur_chain->next;
3316 devm_kfree(&pdev->dev, cur_chain);
3324 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
3325 struct hnae3_ring_chain_node *head)
3327 struct pci_dev *pdev = tqp_vector->handle->pdev;
3328 struct hnae3_ring_chain_node *chain_tmp, *chain;
3333 chain_tmp = chain->next;
3334 devm_kfree(&pdev->dev, chain);
3339 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
3340 struct hns3_enet_ring *ring)
3342 ring->next = group->ring;
3348 static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
3350 struct pci_dev *pdev = priv->ae_handle->pdev;
3351 struct hns3_enet_tqp_vector *tqp_vector;
3352 int num_vectors = priv->vector_num;
3356 numa_node = dev_to_node(&pdev->dev);
3358 for (vector_i = 0; vector_i < num_vectors; vector_i++) {
3359 tqp_vector = &priv->tqp_vector[vector_i];
3360 cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
3361 &tqp_vector->affinity_mask);
3365 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
3367 struct hnae3_ring_chain_node vector_ring_chain;
3368 struct hnae3_handle *h = priv->ae_handle;
3369 struct hns3_enet_tqp_vector *tqp_vector;
3373 hns3_nic_set_cpumask(priv);
3375 for (i = 0; i < priv->vector_num; i++) {
3376 tqp_vector = &priv->tqp_vector[i];
3377 hns3_vector_gl_rl_init_hw(tqp_vector, priv);
3378 tqp_vector->num_tqps = 0;
3381 for (i = 0; i < h->kinfo.num_tqps; i++) {
3382 u16 vector_i = i % priv->vector_num;
3383 u16 tqp_num = h->kinfo.num_tqps;
3385 tqp_vector = &priv->tqp_vector[vector_i];
3387 hns3_add_ring_to_group(&tqp_vector->tx_group,
3388 priv->ring_data[i].ring);
3390 hns3_add_ring_to_group(&tqp_vector->rx_group,
3391 priv->ring_data[i + tqp_num].ring);
3393 priv->ring_data[i].ring->tqp_vector = tqp_vector;
3394 priv->ring_data[i + tqp_num].ring->tqp_vector = tqp_vector;
3395 tqp_vector->num_tqps++;
3398 for (i = 0; i < priv->vector_num; i++) {
3399 tqp_vector = &priv->tqp_vector[i];
3401 tqp_vector->rx_group.total_bytes = 0;
3402 tqp_vector->rx_group.total_packets = 0;
3403 tqp_vector->tx_group.total_bytes = 0;
3404 tqp_vector->tx_group.total_packets = 0;
3405 tqp_vector->handle = h;
3407 ret = hns3_get_vector_ring_chain(tqp_vector,
3408 &vector_ring_chain);
3412 ret = h->ae_algo->ops->map_ring_to_vector(h,
3413 tqp_vector->vector_irq, &vector_ring_chain);
3415 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3420 netif_napi_add(priv->netdev, &tqp_vector->napi,
3421 hns3_nic_common_poll, NAPI_POLL_WEIGHT);
3428 netif_napi_del(&priv->tqp_vector[i].napi);
3433 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
3435 #define HNS3_VECTOR_PF_MAX_NUM 64
3437 struct hnae3_handle *h = priv->ae_handle;
3438 struct hns3_enet_tqp_vector *tqp_vector;
3439 struct hnae3_vector_info *vector;
3440 struct pci_dev *pdev = h->pdev;
3441 u16 tqp_num = h->kinfo.num_tqps;
3446 /* RSS size, cpu online and vector_num should be the same */
3447 /* Should consider 2p/4p later */
3448 vector_num = min_t(u16, num_online_cpus(), tqp_num);
3449 vector_num = min_t(u16, vector_num, HNS3_VECTOR_PF_MAX_NUM);
3451 vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
3456 /* save the actual available vector number */
3457 vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
3459 priv->vector_num = vector_num;
3460 priv->tqp_vector = (struct hns3_enet_tqp_vector *)
3461 devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
3463 if (!priv->tqp_vector) {
3468 for (i = 0; i < priv->vector_num; i++) {
3469 tqp_vector = &priv->tqp_vector[i];
3470 tqp_vector->idx = i;
3471 tqp_vector->mask_addr = vector[i].io_addr;
3472 tqp_vector->vector_irq = vector[i].vector;
3473 hns3_vector_gl_rl_init(tqp_vector, priv);
3477 devm_kfree(&pdev->dev, vector);
3481 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
3487 static void hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
3489 struct hnae3_ring_chain_node vector_ring_chain;
3490 struct hnae3_handle *h = priv->ae_handle;
3491 struct hns3_enet_tqp_vector *tqp_vector;
3494 for (i = 0; i < priv->vector_num; i++) {
3495 tqp_vector = &priv->tqp_vector[i];
3497 if (!tqp_vector->rx_group.ring && !tqp_vector->tx_group.ring)
3500 hns3_get_vector_ring_chain(tqp_vector, &vector_ring_chain);
3502 h->ae_algo->ops->unmap_ring_from_vector(h,
3503 tqp_vector->vector_irq, &vector_ring_chain);
3505 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3507 if (tqp_vector->irq_init_flag == HNS3_VECTOR_INITED) {
3508 irq_set_affinity_hint(tqp_vector->vector_irq, NULL);
3509 free_irq(tqp_vector->vector_irq, tqp_vector);
3510 tqp_vector->irq_init_flag = HNS3_VECTOR_NOT_INITED;
3513 hns3_clear_ring_group(&tqp_vector->rx_group);
3514 hns3_clear_ring_group(&tqp_vector->tx_group);
3515 netif_napi_del(&priv->tqp_vector[i].napi);
3519 static int hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
3521 struct hnae3_handle *h = priv->ae_handle;
3522 struct pci_dev *pdev = h->pdev;
3525 for (i = 0; i < priv->vector_num; i++) {
3526 struct hns3_enet_tqp_vector *tqp_vector;
3528 tqp_vector = &priv->tqp_vector[i];
3529 ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
3534 devm_kfree(&pdev->dev, priv->tqp_vector);
3538 static int hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
3539 unsigned int ring_type)
3541 struct hns3_nic_ring_data *ring_data = priv->ring_data;
3542 int queue_num = priv->ae_handle->kinfo.num_tqps;
3543 struct pci_dev *pdev = priv->ae_handle->pdev;
3544 struct hns3_enet_ring *ring;
3547 ring = devm_kzalloc(&pdev->dev, sizeof(*ring), GFP_KERNEL);
3551 if (ring_type == HNAE3_RING_TYPE_TX) {
3552 desc_num = priv->ae_handle->kinfo.num_tx_desc;
3553 ring_data[q->tqp_index].ring = ring;
3554 ring_data[q->tqp_index].queue_index = q->tqp_index;
3555 ring->io_base = (u8 __iomem *)q->io_base + HNS3_TX_REG_OFFSET;
3557 desc_num = priv->ae_handle->kinfo.num_rx_desc;
3558 ring_data[q->tqp_index + queue_num].ring = ring;
3559 ring_data[q->tqp_index + queue_num].queue_index = q->tqp_index;
3560 ring->io_base = q->io_base;
3563 hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
3567 ring->desc_cb = NULL;
3568 ring->dev = priv->dev;
3569 ring->desc_dma_addr = 0;
3570 ring->buf_size = q->buf_size;
3571 ring->desc_num = desc_num;
3572 ring->next_to_use = 0;
3573 ring->next_to_clean = 0;
3578 static int hns3_queue_to_ring(struct hnae3_queue *tqp,
3579 struct hns3_nic_priv *priv)
3583 ret = hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
3587 ret = hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
3589 devm_kfree(priv->dev, priv->ring_data[tqp->tqp_index].ring);
3596 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
3598 struct hnae3_handle *h = priv->ae_handle;
3599 struct pci_dev *pdev = h->pdev;
3602 priv->ring_data = devm_kzalloc(&pdev->dev,
3603 array3_size(h->kinfo.num_tqps,
3604 sizeof(*priv->ring_data),
3607 if (!priv->ring_data)
3610 for (i = 0; i < h->kinfo.num_tqps; i++) {
3611 ret = hns3_queue_to_ring(h->kinfo.tqp[i], priv);
3619 devm_kfree(priv->dev, priv->ring_data[i].ring);
3620 devm_kfree(priv->dev,
3621 priv->ring_data[i + h->kinfo.num_tqps].ring);
3624 devm_kfree(&pdev->dev, priv->ring_data);
3625 priv->ring_data = NULL;
3629 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
3631 struct hnae3_handle *h = priv->ae_handle;
3634 if (!priv->ring_data)
3637 for (i = 0; i < h->kinfo.num_tqps; i++) {
3638 devm_kfree(priv->dev, priv->ring_data[i].ring);
3639 devm_kfree(priv->dev,
3640 priv->ring_data[i + h->kinfo.num_tqps].ring);
3642 devm_kfree(priv->dev, priv->ring_data);
3643 priv->ring_data = NULL;
3646 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
3650 if (ring->desc_num <= 0 || ring->buf_size <= 0)
3653 ring->desc_cb = devm_kcalloc(ring_to_dev(ring), ring->desc_num,
3654 sizeof(ring->desc_cb[0]), GFP_KERNEL);
3655 if (!ring->desc_cb) {
3660 ret = hns3_alloc_desc(ring);
3662 goto out_with_desc_cb;
3664 if (!HNAE3_IS_TX_RING(ring)) {
3665 ret = hns3_alloc_ring_buffers(ring);
3673 hns3_free_desc(ring);
3675 devm_kfree(ring_to_dev(ring), ring->desc_cb);
3676 ring->desc_cb = NULL;
3681 void hns3_fini_ring(struct hns3_enet_ring *ring)
3683 hns3_free_desc(ring);
3684 devm_kfree(ring_to_dev(ring), ring->desc_cb);
3685 ring->desc_cb = NULL;
3686 ring->next_to_clean = 0;
3687 ring->next_to_use = 0;
3688 ring->pending_buf = 0;
3690 dev_kfree_skb_any(ring->skb);
3695 static int hns3_buf_size2type(u32 buf_size)
3701 bd_size_type = HNS3_BD_SIZE_512_TYPE;
3704 bd_size_type = HNS3_BD_SIZE_1024_TYPE;
3707 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
3710 bd_size_type = HNS3_BD_SIZE_4096_TYPE;
3713 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
3716 return bd_size_type;
3719 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
3721 dma_addr_t dma = ring->desc_dma_addr;
3722 struct hnae3_queue *q = ring->tqp;
3724 if (!HNAE3_IS_TX_RING(ring)) {
3725 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG, (u32)dma);
3726 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
3727 (u32)((dma >> 31) >> 1));
3729 hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
3730 hns3_buf_size2type(ring->buf_size));
3731 hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
3732 ring->desc_num / 8 - 1);
3735 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
3737 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
3738 (u32)((dma >> 31) >> 1));
3740 hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
3741 ring->desc_num / 8 - 1);
3745 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
3747 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
3750 for (i = 0; i < HNAE3_MAX_TC; i++) {
3751 struct hnae3_tc_info *tc_info = &kinfo->tc_info[i];
3754 if (!tc_info->enable)
3757 for (j = 0; j < tc_info->tqp_count; j++) {
3758 struct hnae3_queue *q;
3760 q = priv->ring_data[tc_info->tqp_offset + j].ring->tqp;
3761 hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG,
3767 int hns3_init_all_ring(struct hns3_nic_priv *priv)
3769 struct hnae3_handle *h = priv->ae_handle;
3770 int ring_num = h->kinfo.num_tqps * 2;
3774 for (i = 0; i < ring_num; i++) {
3775 ret = hns3_alloc_ring_memory(priv->ring_data[i].ring);
3778 "Alloc ring memory fail! ret=%d\n", ret);
3779 goto out_when_alloc_ring_memory;
3782 u64_stats_init(&priv->ring_data[i].ring->syncp);
3787 out_when_alloc_ring_memory:
3788 for (j = i - 1; j >= 0; j--)
3789 hns3_fini_ring(priv->ring_data[j].ring);
3794 int hns3_uninit_all_ring(struct hns3_nic_priv *priv)
3796 struct hnae3_handle *h = priv->ae_handle;
3799 for (i = 0; i < h->kinfo.num_tqps; i++) {
3800 hns3_fini_ring(priv->ring_data[i].ring);
3801 hns3_fini_ring(priv->ring_data[i + h->kinfo.num_tqps].ring);
3806 /* Set mac addr if it is configured. or leave it to the AE driver */
3807 static int hns3_init_mac_addr(struct net_device *netdev, bool init)
3809 struct hns3_nic_priv *priv = netdev_priv(netdev);
3810 struct hnae3_handle *h = priv->ae_handle;
3811 u8 mac_addr_temp[ETH_ALEN];
3814 if (h->ae_algo->ops->get_mac_addr && init) {
3815 h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
3816 ether_addr_copy(netdev->dev_addr, mac_addr_temp);
3819 /* Check if the MAC address is valid, if not get a random one */
3820 if (!is_valid_ether_addr(netdev->dev_addr)) {
3821 eth_hw_addr_random(netdev);
3822 dev_warn(priv->dev, "using random MAC address %pM\n",
3826 if (h->ae_algo->ops->set_mac_addr)
3827 ret = h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
3832 static int hns3_init_phy(struct net_device *netdev)
3834 struct hnae3_handle *h = hns3_get_handle(netdev);
3837 if (h->ae_algo->ops->mac_connect_phy)
3838 ret = h->ae_algo->ops->mac_connect_phy(h);
3843 static void hns3_uninit_phy(struct net_device *netdev)
3845 struct hnae3_handle *h = hns3_get_handle(netdev);
3847 if (h->ae_algo->ops->mac_disconnect_phy)
3848 h->ae_algo->ops->mac_disconnect_phy(h);
3851 static int hns3_restore_fd_rules(struct net_device *netdev)
3853 struct hnae3_handle *h = hns3_get_handle(netdev);
3856 if (h->ae_algo->ops->restore_fd_rules)
3857 ret = h->ae_algo->ops->restore_fd_rules(h);
3862 static void hns3_del_all_fd_rules(struct net_device *netdev, bool clear_list)
3864 struct hnae3_handle *h = hns3_get_handle(netdev);
3866 if (h->ae_algo->ops->del_all_fd_entries)
3867 h->ae_algo->ops->del_all_fd_entries(h, clear_list);
3870 static int hns3_client_start(struct hnae3_handle *handle)
3872 if (!handle->ae_algo->ops->client_start)
3875 return handle->ae_algo->ops->client_start(handle);
3878 static void hns3_client_stop(struct hnae3_handle *handle)
3880 if (!handle->ae_algo->ops->client_stop)
3883 handle->ae_algo->ops->client_stop(handle);
3886 static void hns3_info_show(struct hns3_nic_priv *priv)
3888 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
3890 dev_info(priv->dev, "MAC address: %pM\n", priv->netdev->dev_addr);
3891 dev_info(priv->dev, "Task queue pairs numbers: %d\n", kinfo->num_tqps);
3892 dev_info(priv->dev, "RSS size: %d\n", kinfo->rss_size);
3893 dev_info(priv->dev, "Allocated RSS size: %d\n", kinfo->req_rss_size);
3894 dev_info(priv->dev, "RX buffer length: %d\n", kinfo->rx_buf_len);
3895 dev_info(priv->dev, "Desc num per TX queue: %d\n", kinfo->num_tx_desc);
3896 dev_info(priv->dev, "Desc num per RX queue: %d\n", kinfo->num_rx_desc);
3897 dev_info(priv->dev, "Total number of enabled TCs: %d\n", kinfo->num_tc);
3898 dev_info(priv->dev, "Max mtu size: %d\n", priv->netdev->max_mtu);
3901 static int hns3_client_init(struct hnae3_handle *handle)
3903 struct pci_dev *pdev = handle->pdev;
3904 u16 alloc_tqps, max_rss_size;
3905 struct hns3_nic_priv *priv;
3906 struct net_device *netdev;
3909 handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
3911 netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
3915 priv = netdev_priv(netdev);
3916 priv->dev = &pdev->dev;
3917 priv->netdev = netdev;
3918 priv->ae_handle = handle;
3919 priv->tx_timeout_count = 0;
3920 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
3922 handle->msg_enable = netif_msg_init(debug, DEFAULT_MSG_LEVEL);
3924 handle->kinfo.netdev = netdev;
3925 handle->priv = (void *)priv;
3927 hns3_init_mac_addr(netdev, true);
3929 hns3_set_default_feature(netdev);
3931 netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
3932 netdev->priv_flags |= IFF_UNICAST_FLT;
3933 netdev->netdev_ops = &hns3_nic_netdev_ops;
3934 SET_NETDEV_DEV(netdev, &pdev->dev);
3935 hns3_ethtool_set_ops(netdev);
3937 /* Carrier off reporting is important to ethtool even BEFORE open */
3938 netif_carrier_off(netdev);
3940 ret = hns3_get_ring_config(priv);
3943 goto out_get_ring_cfg;
3946 ret = hns3_nic_alloc_vector_data(priv);
3949 goto out_alloc_vector_data;
3952 ret = hns3_nic_init_vector_data(priv);
3955 goto out_init_vector_data;
3958 ret = hns3_init_all_ring(priv);
3961 goto out_init_ring_data;
3964 ret = hns3_init_phy(netdev);
3968 ret = register_netdev(netdev);
3970 dev_err(priv->dev, "probe register netdev fail!\n");
3971 goto out_reg_netdev_fail;
3974 ret = hns3_client_start(handle);
3976 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
3977 goto out_client_start;
3980 hns3_dcbnl_setup(handle);
3982 hns3_dbg_init(handle);
3984 /* MTU range: (ETH_MIN_MTU(kernel default) - 9702) */
3985 netdev->max_mtu = HNS3_MAX_MTU;
3987 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
3989 if (netif_msg_drv(handle))
3990 hns3_info_show(priv);
3995 unregister_netdev(netdev);
3996 out_reg_netdev_fail:
3997 hns3_uninit_phy(netdev);
3999 hns3_uninit_all_ring(priv);
4001 hns3_nic_uninit_vector_data(priv);
4002 out_init_vector_data:
4003 hns3_nic_dealloc_vector_data(priv);
4004 out_alloc_vector_data:
4005 priv->ring_data = NULL;
4007 priv->ae_handle = NULL;
4008 free_netdev(netdev);
4012 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
4014 struct net_device *netdev = handle->kinfo.netdev;
4015 struct hns3_nic_priv *priv = netdev_priv(netdev);
4018 hns3_remove_hw_addr(netdev);
4020 if (netdev->reg_state != NETREG_UNINITIALIZED)
4021 unregister_netdev(netdev);
4023 hns3_client_stop(handle);
4025 hns3_uninit_phy(netdev);
4027 if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
4028 netdev_warn(netdev, "already uninitialized\n");
4029 goto out_netdev_free;
4032 hns3_del_all_fd_rules(netdev, true);
4034 hns3_clear_all_ring(handle, true);
4036 hns3_nic_uninit_vector_data(priv);
4038 ret = hns3_nic_dealloc_vector_data(priv);
4040 netdev_err(netdev, "dealloc vector error\n");
4042 ret = hns3_uninit_all_ring(priv);
4044 netdev_err(netdev, "uninit ring error\n");
4046 hns3_put_ring_config(priv);
4048 hns3_dbg_uninit(handle);
4051 free_netdev(netdev);
4054 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
4056 struct net_device *netdev = handle->kinfo.netdev;
4062 netif_carrier_on(netdev);
4063 netif_tx_wake_all_queues(netdev);
4064 if (netif_msg_link(handle))
4065 netdev_info(netdev, "link up\n");
4067 netif_carrier_off(netdev);
4068 netif_tx_stop_all_queues(netdev);
4069 if (netif_msg_link(handle))
4070 netdev_info(netdev, "link down\n");
4074 static int hns3_client_setup_tc(struct hnae3_handle *handle, u8 tc)
4076 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4077 struct net_device *ndev = kinfo->netdev;
4079 if (tc > HNAE3_MAX_TC)
4085 return hns3_nic_set_real_num_queue(ndev);
4088 static int hns3_recover_hw_addr(struct net_device *ndev)
4090 struct netdev_hw_addr_list *list;
4091 struct netdev_hw_addr *ha, *tmp;
4094 netif_addr_lock_bh(ndev);
4095 /* go through and sync uc_addr entries to the device */
4097 list_for_each_entry_safe(ha, tmp, &list->list, list) {
4098 ret = hns3_nic_uc_sync(ndev, ha->addr);
4103 /* go through and sync mc_addr entries to the device */
4105 list_for_each_entry_safe(ha, tmp, &list->list, list) {
4106 ret = hns3_nic_mc_sync(ndev, ha->addr);
4112 netif_addr_unlock_bh(ndev);
4116 static void hns3_remove_hw_addr(struct net_device *netdev)
4118 struct netdev_hw_addr_list *list;
4119 struct netdev_hw_addr *ha, *tmp;
4121 hns3_nic_uc_unsync(netdev, netdev->dev_addr);
4123 netif_addr_lock_bh(netdev);
4124 /* go through and unsync uc_addr entries to the device */
4126 list_for_each_entry_safe(ha, tmp, &list->list, list)
4127 hns3_nic_uc_unsync(netdev, ha->addr);
4129 /* go through and unsync mc_addr entries to the device */
4131 list_for_each_entry_safe(ha, tmp, &list->list, list)
4132 if (ha->refcount > 1)
4133 hns3_nic_mc_unsync(netdev, ha->addr);
4135 netif_addr_unlock_bh(netdev);
4138 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
4140 while (ring->next_to_clean != ring->next_to_use) {
4141 ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
4142 hns3_free_buffer_detach(ring, ring->next_to_clean);
4143 ring_ptr_move_fw(ring, next_to_clean);
4147 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
4149 struct hns3_desc_cb res_cbs;
4152 while (ring->next_to_use != ring->next_to_clean) {
4153 /* When a buffer is not reused, it's memory has been
4154 * freed in hns3_handle_rx_bd or will be freed by
4155 * stack, so we need to replace the buffer here.
4157 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
4158 ret = hns3_reserve_buffer_map(ring, &res_cbs);
4160 u64_stats_update_begin(&ring->syncp);
4161 ring->stats.sw_err_cnt++;
4162 u64_stats_update_end(&ring->syncp);
4163 /* if alloc new buffer fail, exit directly
4164 * and reclear in up flow.
4166 netdev_warn(ring->tqp->handle->kinfo.netdev,
4167 "reserve buffer map failed, ret = %d\n",
4171 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
4173 ring_ptr_move_fw(ring, next_to_use);
4176 /* Free the pending skb in rx ring */
4178 dev_kfree_skb_any(ring->skb);
4180 ring->pending_buf = 0;
4186 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
4188 while (ring->next_to_use != ring->next_to_clean) {
4189 /* When a buffer is not reused, it's memory has been
4190 * freed in hns3_handle_rx_bd or will be freed by
4191 * stack, so only need to unmap the buffer here.
4193 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
4194 hns3_unmap_buffer(ring,
4195 &ring->desc_cb[ring->next_to_use]);
4196 ring->desc_cb[ring->next_to_use].dma = 0;
4199 ring_ptr_move_fw(ring, next_to_use);
4203 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force)
4205 struct net_device *ndev = h->kinfo.netdev;
4206 struct hns3_nic_priv *priv = netdev_priv(ndev);
4209 for (i = 0; i < h->kinfo.num_tqps; i++) {
4210 struct hns3_enet_ring *ring;
4212 ring = priv->ring_data[i].ring;
4213 hns3_clear_tx_ring(ring);
4215 ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
4216 /* Continue to clear other rings even if clearing some
4220 hns3_force_clear_rx_ring(ring);
4222 hns3_clear_rx_ring(ring);
4226 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
4228 struct net_device *ndev = h->kinfo.netdev;
4229 struct hns3_nic_priv *priv = netdev_priv(ndev);
4230 struct hns3_enet_ring *rx_ring;
4234 for (i = 0; i < h->kinfo.num_tqps; i++) {
4235 ret = h->ae_algo->ops->reset_queue(h, i);
4239 hns3_init_ring_hw(priv->ring_data[i].ring);
4241 /* We need to clear tx ring here because self test will
4242 * use the ring and will not run down before up
4244 hns3_clear_tx_ring(priv->ring_data[i].ring);
4245 priv->ring_data[i].ring->next_to_clean = 0;
4246 priv->ring_data[i].ring->next_to_use = 0;
4248 rx_ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
4249 hns3_init_ring_hw(rx_ring);
4250 ret = hns3_clear_rx_ring(rx_ring);
4254 /* We can not know the hardware head and tail when this
4255 * function is called in reset flow, so we reuse all desc.
4257 for (j = 0; j < rx_ring->desc_num; j++)
4258 hns3_reuse_buffer(rx_ring, j);
4260 rx_ring->next_to_clean = 0;
4261 rx_ring->next_to_use = 0;
4264 hns3_init_tx_ring_tc(priv);
4269 static void hns3_store_coal(struct hns3_nic_priv *priv)
4271 /* ethtool only support setting and querying one coal
4272 * configuration for now, so save the vector 0' coal
4273 * configuration here in order to restore it.
4275 memcpy(&priv->tx_coal, &priv->tqp_vector[0].tx_group.coal,
4276 sizeof(struct hns3_enet_coalesce));
4277 memcpy(&priv->rx_coal, &priv->tqp_vector[0].rx_group.coal,
4278 sizeof(struct hns3_enet_coalesce));
4281 static void hns3_restore_coal(struct hns3_nic_priv *priv)
4283 u16 vector_num = priv->vector_num;
4286 for (i = 0; i < vector_num; i++) {
4287 memcpy(&priv->tqp_vector[i].tx_group.coal, &priv->tx_coal,
4288 sizeof(struct hns3_enet_coalesce));
4289 memcpy(&priv->tqp_vector[i].rx_group.coal, &priv->rx_coal,
4290 sizeof(struct hns3_enet_coalesce));
4294 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
4296 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(handle->pdev);
4297 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4298 struct net_device *ndev = kinfo->netdev;
4299 struct hns3_nic_priv *priv = netdev_priv(ndev);
4301 if (test_and_set_bit(HNS3_NIC_STATE_RESETTING, &priv->state))
4304 /* it is cumbersome for hardware to pick-and-choose entries for deletion
4305 * from table space. Hence, for function reset software intervention is
4306 * required to delete the entries
4308 if (hns3_dev_ongoing_func_reset(ae_dev)) {
4309 hns3_remove_hw_addr(ndev);
4310 hns3_del_all_fd_rules(ndev, false);
4313 if (!netif_running(ndev))
4316 return hns3_nic_net_stop(ndev);
4319 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
4321 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4322 struct hns3_nic_priv *priv = netdev_priv(kinfo->netdev);
4325 clear_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4327 if (netif_running(kinfo->netdev)) {
4328 ret = hns3_nic_net_open(kinfo->netdev);
4330 set_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4331 netdev_err(kinfo->netdev,
4332 "net up fail, ret=%d!\n", ret);
4340 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
4342 struct net_device *netdev = handle->kinfo.netdev;
4343 struct hns3_nic_priv *priv = netdev_priv(netdev);
4346 /* Carrier off reporting is important to ethtool even BEFORE open */
4347 netif_carrier_off(netdev);
4349 ret = hns3_get_ring_config(priv);
4353 ret = hns3_nic_alloc_vector_data(priv);
4357 hns3_restore_coal(priv);
4359 ret = hns3_nic_init_vector_data(priv);
4361 goto err_dealloc_vector;
4363 ret = hns3_init_all_ring(priv);
4365 goto err_uninit_vector;
4367 ret = hns3_client_start(handle);
4369 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
4370 goto err_uninit_ring;
4373 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
4378 hns3_uninit_all_ring(priv);
4380 hns3_nic_uninit_vector_data(priv);
4382 hns3_nic_dealloc_vector_data(priv);
4384 hns3_put_ring_config(priv);
4389 static int hns3_reset_notify_restore_enet(struct hnae3_handle *handle)
4391 struct net_device *netdev = handle->kinfo.netdev;
4392 bool vlan_filter_enable;
4395 ret = hns3_init_mac_addr(netdev, false);
4399 ret = hns3_recover_hw_addr(netdev);
4403 ret = hns3_update_promisc_mode(netdev, handle->netdev_flags);
4407 vlan_filter_enable = netdev->flags & IFF_PROMISC ? false : true;
4408 hns3_enable_vlan_filter(netdev, vlan_filter_enable);
4410 if (handle->ae_algo->ops->restore_vlan_table)
4411 handle->ae_algo->ops->restore_vlan_table(handle);
4413 return hns3_restore_fd_rules(netdev);
4416 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
4418 struct net_device *netdev = handle->kinfo.netdev;
4419 struct hns3_nic_priv *priv = netdev_priv(netdev);
4422 if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
4423 netdev_warn(netdev, "already uninitialized\n");
4427 hns3_clear_all_ring(handle, true);
4428 hns3_reset_tx_queue(priv->ae_handle);
4430 hns3_nic_uninit_vector_data(priv);
4432 hns3_store_coal(priv);
4434 ret = hns3_nic_dealloc_vector_data(priv);
4436 netdev_err(netdev, "dealloc vector error\n");
4438 ret = hns3_uninit_all_ring(priv);
4440 netdev_err(netdev, "uninit ring error\n");
4442 hns3_put_ring_config(priv);
4447 static int hns3_reset_notify(struct hnae3_handle *handle,
4448 enum hnae3_reset_notify_type type)
4453 case HNAE3_UP_CLIENT:
4454 ret = hns3_reset_notify_up_enet(handle);
4456 case HNAE3_DOWN_CLIENT:
4457 ret = hns3_reset_notify_down_enet(handle);
4459 case HNAE3_INIT_CLIENT:
4460 ret = hns3_reset_notify_init_enet(handle);
4462 case HNAE3_UNINIT_CLIENT:
4463 ret = hns3_reset_notify_uninit_enet(handle);
4465 case HNAE3_RESTORE_CLIENT:
4466 ret = hns3_reset_notify_restore_enet(handle);
4475 static int hns3_change_channels(struct hnae3_handle *handle, u32 new_tqp_num,
4476 bool rxfh_configured)
4480 ret = handle->ae_algo->ops->set_channels(handle, new_tqp_num,
4483 dev_err(&handle->pdev->dev,
4484 "Change tqp num(%u) fail.\n", new_tqp_num);
4488 ret = hns3_reset_notify(handle, HNAE3_INIT_CLIENT);
4492 ret = hns3_reset_notify(handle, HNAE3_UP_CLIENT);
4494 hns3_reset_notify(handle, HNAE3_UNINIT_CLIENT);
4499 int hns3_set_channels(struct net_device *netdev,
4500 struct ethtool_channels *ch)
4502 struct hnae3_handle *h = hns3_get_handle(netdev);
4503 struct hnae3_knic_private_info *kinfo = &h->kinfo;
4504 bool rxfh_configured = netif_is_rxfh_configured(netdev);
4505 u32 new_tqp_num = ch->combined_count;
4509 if (hns3_nic_resetting(netdev))
4512 if (ch->rx_count || ch->tx_count)
4515 if (new_tqp_num > hns3_get_max_available_channels(h) ||
4517 dev_err(&netdev->dev,
4518 "Change tqps fail, the tqp range is from 1 to %d",
4519 hns3_get_max_available_channels(h));
4523 if (kinfo->rss_size == new_tqp_num)
4526 netif_dbg(h, drv, netdev,
4527 "set channels: tqp_num=%u, rxfh=%d\n",
4528 new_tqp_num, rxfh_configured);
4530 ret = hns3_reset_notify(h, HNAE3_DOWN_CLIENT);
4534 ret = hns3_reset_notify(h, HNAE3_UNINIT_CLIENT);
4538 org_tqp_num = h->kinfo.num_tqps;
4539 ret = hns3_change_channels(h, new_tqp_num, rxfh_configured);
4544 "Change channels fail, revert to old value\n");
4545 ret1 = hns3_change_channels(h, org_tqp_num, rxfh_configured);
4548 "revert to old channel fail\n");
4558 static const struct hns3_hw_error_info hns3_hw_err[] = {
4559 { .type = HNAE3_PPU_POISON_ERROR,
4560 .msg = "PPU poison" },
4561 { .type = HNAE3_CMDQ_ECC_ERROR,
4562 .msg = "IMP CMDQ error" },
4563 { .type = HNAE3_IMP_RD_POISON_ERROR,
4564 .msg = "IMP RD poison" },
4567 static void hns3_process_hw_error(struct hnae3_handle *handle,
4568 enum hnae3_hw_error_type type)
4572 for (i = 0; i < ARRAY_SIZE(hns3_hw_err); i++) {
4573 if (hns3_hw_err[i].type == type) {
4574 dev_err(&handle->pdev->dev, "Detected %s!\n",
4575 hns3_hw_err[i].msg);
4581 static const struct hnae3_client_ops client_ops = {
4582 .init_instance = hns3_client_init,
4583 .uninit_instance = hns3_client_uninit,
4584 .link_status_change = hns3_link_status_change,
4585 .setup_tc = hns3_client_setup_tc,
4586 .reset_notify = hns3_reset_notify,
4587 .process_hw_error = hns3_process_hw_error,
4590 /* hns3_init_module - Driver registration routine
4591 * hns3_init_module is the first routine called when the driver is
4592 * loaded. All it does is register with the PCI subsystem.
4594 static int __init hns3_init_module(void)
4598 pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
4599 pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
4601 client.type = HNAE3_CLIENT_KNIC;
4602 snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH - 1, "%s",
4605 client.ops = &client_ops;
4607 INIT_LIST_HEAD(&client.node);
4609 hns3_dbg_register_debugfs(hns3_driver_name);
4611 ret = hnae3_register_client(&client);
4613 goto err_reg_client;
4615 ret = pci_register_driver(&hns3_driver);
4617 goto err_reg_driver;
4622 hnae3_unregister_client(&client);
4624 hns3_dbg_unregister_debugfs();
4627 module_init(hns3_init_module);
4629 /* hns3_exit_module - Driver exit cleanup routine
4630 * hns3_exit_module is called just before the driver is removed
4633 static void __exit hns3_exit_module(void)
4635 pci_unregister_driver(&hns3_driver);
4636 hnae3_unregister_client(&client);
4637 hns3_dbg_unregister_debugfs();
4639 module_exit(hns3_exit_module);
4641 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
4642 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
4643 MODULE_LICENSE("GPL");
4644 MODULE_ALIAS("pci:hns-nic");
4645 MODULE_VERSION(HNS3_MOD_VERSION);