1 // SPDX-License-Identifier: GPL-2.0+
2 // Copyright (c) 2016-2017 Hisilicon Limited.
4 #include <linux/dma-mapping.h>
5 #include <linux/etherdevice.h>
6 #include <linux/interrupt.h>
7 #include <linux/if_vlan.h>
9 #include <linux/ipv6.h>
10 #include <linux/module.h>
11 #include <linux/pci.h>
12 #include <linux/skbuff.h>
13 #include <linux/sctp.h>
14 #include <linux/vermagic.h>
16 #include <net/pkt_cls.h>
17 #include <net/vxlan.h>
20 #include "hns3_enet.h"
22 static void hns3_clear_all_ring(struct hnae3_handle *h);
23 static void hns3_force_clear_all_rx_ring(struct hnae3_handle *h);
25 static const char hns3_driver_name[] = "hns3";
26 const char hns3_driver_version[] = VERMAGIC_STRING;
27 static const char hns3_driver_string[] =
28 "Hisilicon Ethernet Network Driver for Hip08 Family";
29 static const char hns3_copyright[] = "Copyright (c) 2017 Huawei Corporation.";
30 static struct hnae3_client client;
32 /* hns3_pci_tbl - PCI Device ID Table
34 * Last entry must be all 0s
36 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
37 * Class, Class Mask, private data (not used) }
39 static const struct pci_device_id hns3_pci_tbl[] = {
40 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
41 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
42 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA),
43 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
44 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC),
45 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
46 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA),
47 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
48 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC),
49 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
50 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC),
51 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
52 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_VF), 0},
53 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF),
54 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
55 /* required last entry */
58 MODULE_DEVICE_TABLE(pci, hns3_pci_tbl);
60 static irqreturn_t hns3_irq_handle(int irq, void *vector)
62 struct hns3_enet_tqp_vector *tqp_vector = vector;
64 napi_schedule(&tqp_vector->napi);
69 /* This callback function is used to set affinity changes to the irq affinity
70 * masks when the irq_set_affinity_notifier function is used.
72 static void hns3_nic_irq_affinity_notify(struct irq_affinity_notify *notify,
73 const cpumask_t *mask)
75 struct hns3_enet_tqp_vector *tqp_vectors =
76 container_of(notify, struct hns3_enet_tqp_vector,
79 tqp_vectors->affinity_mask = *mask;
82 static void hns3_nic_irq_affinity_release(struct kref *ref)
86 static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv)
88 struct hns3_enet_tqp_vector *tqp_vectors;
91 for (i = 0; i < priv->vector_num; i++) {
92 tqp_vectors = &priv->tqp_vector[i];
94 if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED)
97 /* clear the affinity notifier and affinity mask */
98 irq_set_affinity_notifier(tqp_vectors->vector_irq, NULL);
99 irq_set_affinity_hint(tqp_vectors->vector_irq, NULL);
101 /* release the irq resource */
102 free_irq(tqp_vectors->vector_irq, tqp_vectors);
103 tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED;
107 static int hns3_nic_init_irq(struct hns3_nic_priv *priv)
109 struct hns3_enet_tqp_vector *tqp_vectors;
110 int txrx_int_idx = 0;
116 for (i = 0; i < priv->vector_num; i++) {
117 tqp_vectors = &priv->tqp_vector[i];
119 if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED)
122 if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) {
123 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
124 "%s-%s-%d", priv->netdev->name, "TxRx",
127 } else if (tqp_vectors->rx_group.ring) {
128 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
129 "%s-%s-%d", priv->netdev->name, "Rx",
131 } else if (tqp_vectors->tx_group.ring) {
132 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
133 "%s-%s-%d", priv->netdev->name, "Tx",
136 /* Skip this unused q_vector */
140 tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0';
142 ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0,
146 netdev_err(priv->netdev, "request irq(%d) fail\n",
147 tqp_vectors->vector_irq);
151 tqp_vectors->affinity_notify.notify =
152 hns3_nic_irq_affinity_notify;
153 tqp_vectors->affinity_notify.release =
154 hns3_nic_irq_affinity_release;
155 irq_set_affinity_notifier(tqp_vectors->vector_irq,
156 &tqp_vectors->affinity_notify);
157 irq_set_affinity_hint(tqp_vectors->vector_irq,
158 &tqp_vectors->affinity_mask);
160 tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED;
166 static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector,
169 writel(mask_en, tqp_vector->mask_addr);
172 static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector)
174 napi_enable(&tqp_vector->napi);
177 hns3_mask_vector_irq(tqp_vector, 1);
180 static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector)
183 hns3_mask_vector_irq(tqp_vector, 0);
185 disable_irq(tqp_vector->vector_irq);
186 napi_disable(&tqp_vector->napi);
189 void hns3_set_vector_coalesce_rl(struct hns3_enet_tqp_vector *tqp_vector,
192 u32 rl_reg = hns3_rl_usec_to_reg(rl_value);
194 /* this defines the configuration for RL (Interrupt Rate Limiter).
195 * Rl defines rate of interrupts i.e. number of interrupts-per-second
196 * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing
199 if (rl_reg > 0 && !tqp_vector->tx_group.coal.gl_adapt_enable &&
200 !tqp_vector->rx_group.coal.gl_adapt_enable)
201 /* According to the hardware, the range of rl_reg is
202 * 0-59 and the unit is 4.
204 rl_reg |= HNS3_INT_RL_ENABLE_MASK;
206 writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
209 void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
212 u32 rx_gl_reg = hns3_gl_usec_to_reg(gl_value);
214 writel(rx_gl_reg, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
217 void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
220 u32 tx_gl_reg = hns3_gl_usec_to_reg(gl_value);
222 writel(tx_gl_reg, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
225 static void hns3_vector_gl_rl_init(struct hns3_enet_tqp_vector *tqp_vector,
226 struct hns3_nic_priv *priv)
228 /* initialize the configuration for interrupt coalescing.
229 * 1. GL (Interrupt Gap Limiter)
230 * 2. RL (Interrupt Rate Limiter)
233 /* Default: enable interrupt coalescing self-adaptive and GL */
234 tqp_vector->tx_group.coal.gl_adapt_enable = 1;
235 tqp_vector->rx_group.coal.gl_adapt_enable = 1;
237 tqp_vector->tx_group.coal.int_gl = HNS3_INT_GL_50K;
238 tqp_vector->rx_group.coal.int_gl = HNS3_INT_GL_50K;
240 tqp_vector->int_adapt_down = HNS3_INT_ADAPT_DOWN_START;
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",
293 ret = netif_set_real_num_rx_queues(netdev, queue_size);
296 "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
303 static u16 hns3_get_max_available_channels(struct hnae3_handle *h)
305 u16 alloc_tqps, max_rss_size, rss_size;
307 h->ae_algo->ops->get_tqps_and_rss_info(h, &alloc_tqps, &max_rss_size);
308 rss_size = alloc_tqps / h->kinfo.num_tc;
310 return min_t(u16, rss_size, max_rss_size);
313 static int hns3_nic_net_up(struct net_device *netdev)
315 struct hns3_nic_priv *priv = netdev_priv(netdev);
316 struct hnae3_handle *h = priv->ae_handle;
320 ret = hns3_nic_reset_all_ring(h);
324 /* get irq resource for all vectors */
325 ret = hns3_nic_init_irq(priv);
327 netdev_err(netdev, "hns init irq failed! ret=%d\n", ret);
331 /* enable the vectors */
332 for (i = 0; i < priv->vector_num; i++)
333 hns3_vector_enable(&priv->tqp_vector[i]);
335 /* start the ae_dev */
336 ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0;
340 clear_bit(HNS3_NIC_STATE_DOWN, &priv->state);
345 for (j = i - 1; j >= 0; j--)
346 hns3_vector_disable(&priv->tqp_vector[j]);
348 hns3_nic_uninit_irq(priv);
353 static int hns3_nic_net_open(struct net_device *netdev)
355 struct hns3_nic_priv *priv = netdev_priv(netdev);
356 struct hnae3_handle *h = hns3_get_handle(netdev);
357 struct hnae3_knic_private_info *kinfo;
360 netif_carrier_off(netdev);
362 ret = hns3_nic_set_real_num_queue(netdev);
366 ret = hns3_nic_net_up(netdev);
369 "hns net up fail, ret=%d!\n", ret);
374 for (i = 0; i < HNAE3_MAX_USER_PRIO; i++) {
375 netdev_set_prio_tc_map(netdev, i,
379 priv->ae_handle->last_reset_time = jiffies;
383 static void hns3_nic_net_down(struct net_device *netdev)
385 struct hns3_nic_priv *priv = netdev_priv(netdev);
386 const struct hnae3_ae_ops *ops;
389 if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
392 /* disable vectors */
393 for (i = 0; i < priv->vector_num; i++)
394 hns3_vector_disable(&priv->tqp_vector[i]);
397 ops = priv->ae_handle->ae_algo->ops;
399 ops->stop(priv->ae_handle);
401 /* free irq resources */
402 hns3_nic_uninit_irq(priv);
404 hns3_clear_all_ring(priv->ae_handle);
407 static int hns3_nic_net_stop(struct net_device *netdev)
409 netif_tx_stop_all_queues(netdev);
410 netif_carrier_off(netdev);
412 hns3_nic_net_down(netdev);
417 static int hns3_nic_uc_sync(struct net_device *netdev,
418 const unsigned char *addr)
420 struct hnae3_handle *h = hns3_get_handle(netdev);
422 if (h->ae_algo->ops->add_uc_addr)
423 return h->ae_algo->ops->add_uc_addr(h, addr);
428 static int hns3_nic_uc_unsync(struct net_device *netdev,
429 const unsigned char *addr)
431 struct hnae3_handle *h = hns3_get_handle(netdev);
433 if (h->ae_algo->ops->rm_uc_addr)
434 return h->ae_algo->ops->rm_uc_addr(h, addr);
439 static int hns3_nic_mc_sync(struct net_device *netdev,
440 const unsigned char *addr)
442 struct hnae3_handle *h = hns3_get_handle(netdev);
444 if (h->ae_algo->ops->add_mc_addr)
445 return h->ae_algo->ops->add_mc_addr(h, addr);
450 static int hns3_nic_mc_unsync(struct net_device *netdev,
451 const unsigned char *addr)
453 struct hnae3_handle *h = hns3_get_handle(netdev);
455 if (h->ae_algo->ops->rm_mc_addr)
456 return h->ae_algo->ops->rm_mc_addr(h, addr);
461 static void hns3_nic_set_rx_mode(struct net_device *netdev)
463 struct hnae3_handle *h = hns3_get_handle(netdev);
465 if (h->ae_algo->ops->set_promisc_mode) {
466 if (netdev->flags & IFF_PROMISC)
467 h->ae_algo->ops->set_promisc_mode(h, true, true);
468 else if (netdev->flags & IFF_ALLMULTI)
469 h->ae_algo->ops->set_promisc_mode(h, false, true);
471 h->ae_algo->ops->set_promisc_mode(h, false, false);
473 if (__dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync))
474 netdev_err(netdev, "sync uc address fail\n");
475 if (netdev->flags & IFF_MULTICAST) {
476 if (__dev_mc_sync(netdev, hns3_nic_mc_sync, hns3_nic_mc_unsync))
477 netdev_err(netdev, "sync mc address fail\n");
479 if (h->ae_algo->ops->update_mta_status)
480 h->ae_algo->ops->update_mta_status(h);
484 static int hns3_set_tso(struct sk_buff *skb, u32 *paylen,
485 u16 *mss, u32 *type_cs_vlan_tso)
487 u32 l4_offset, hdr_len;
488 union l3_hdr_info l3;
489 union l4_hdr_info l4;
493 if (!skb_is_gso(skb))
496 ret = skb_cow_head(skb, 0);
500 l3.hdr = skb_network_header(skb);
501 l4.hdr = skb_transport_header(skb);
503 /* Software should clear the IPv4's checksum field when tso is
506 if (l3.v4->version == 4)
510 if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
513 SKB_GSO_UDP_TUNNEL_CSUM)) {
514 if ((!(skb_shinfo(skb)->gso_type &
516 (skb_shinfo(skb)->gso_type &
517 SKB_GSO_UDP_TUNNEL_CSUM)) {
518 /* Software should clear the udp's checksum
519 * field when tso is needed.
523 /* reset l3&l4 pointers from outer to inner headers */
524 l3.hdr = skb_inner_network_header(skb);
525 l4.hdr = skb_inner_transport_header(skb);
527 /* Software should clear the IPv4's checksum field when
530 if (l3.v4->version == 4)
534 /* normal or tunnel packet*/
535 l4_offset = l4.hdr - skb->data;
536 hdr_len = (l4.tcp->doff * 4) + l4_offset;
538 /* remove payload length from inner pseudo checksum when tso*/
539 l4_paylen = skb->len - l4_offset;
540 csum_replace_by_diff(&l4.tcp->check,
541 (__force __wsum)htonl(l4_paylen));
543 /* find the txbd field values */
544 *paylen = skb->len - hdr_len;
545 hnae3_set_bit(*type_cs_vlan_tso,
548 /* get MSS for TSO */
549 *mss = skb_shinfo(skb)->gso_size;
554 static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
562 unsigned char *l4_hdr;
563 unsigned char *exthdr;
567 /* find outer header point */
568 l3.hdr = skb_network_header(skb);
569 l4_hdr = skb_transport_header(skb);
571 if (skb->protocol == htons(ETH_P_IPV6)) {
572 exthdr = l3.hdr + sizeof(*l3.v6);
573 l4_proto_tmp = l3.v6->nexthdr;
574 if (l4_hdr != exthdr)
575 ipv6_skip_exthdr(skb, exthdr - skb->data,
576 &l4_proto_tmp, &frag_off);
577 } else if (skb->protocol == htons(ETH_P_IP)) {
578 l4_proto_tmp = l3.v4->protocol;
583 *ol4_proto = l4_proto_tmp;
586 if (!skb->encapsulation) {
591 /* find inner header point */
592 l3.hdr = skb_inner_network_header(skb);
593 l4_hdr = skb_inner_transport_header(skb);
595 if (l3.v6->version == 6) {
596 exthdr = l3.hdr + sizeof(*l3.v6);
597 l4_proto_tmp = l3.v6->nexthdr;
598 if (l4_hdr != exthdr)
599 ipv6_skip_exthdr(skb, exthdr - skb->data,
600 &l4_proto_tmp, &frag_off);
601 } else if (l3.v4->version == 4) {
602 l4_proto_tmp = l3.v4->protocol;
605 *il4_proto = l4_proto_tmp;
610 static void hns3_set_l2l3l4_len(struct sk_buff *skb, u8 ol4_proto,
611 u8 il4_proto, u32 *type_cs_vlan_tso,
612 u32 *ol_type_vlan_len_msec)
622 struct gre_base_hdr *gre;
625 unsigned char *l2_hdr;
626 u8 l4_proto = ol4_proto;
633 l3.hdr = skb_network_header(skb);
634 l4.hdr = skb_transport_header(skb);
636 /* compute L2 header size for normal packet, defined in 2 Bytes */
637 l2_len = l3.hdr - skb->data;
638 hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_M,
639 HNS3_TXD_L2LEN_S, l2_len >> 1);
642 if (skb->encapsulation) {
643 /* compute OL2 header size, defined in 2 Bytes */
645 hnae3_set_field(*ol_type_vlan_len_msec,
647 HNS3_TXD_L2LEN_S, ol2_len >> 1);
649 /* compute OL3 header size, defined in 4 Bytes */
650 ol3_len = l4.hdr - l3.hdr;
651 hnae3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_M,
652 HNS3_TXD_L3LEN_S, ol3_len >> 2);
654 /* MAC in UDP, MAC in GRE (0x6558)*/
655 if ((ol4_proto == IPPROTO_UDP) || (ol4_proto == IPPROTO_GRE)) {
656 /* switch MAC header ptr from outer to inner header.*/
657 l2_hdr = skb_inner_mac_header(skb);
659 /* compute OL4 header size, defined in 4 Bytes. */
660 ol4_len = l2_hdr - l4.hdr;
661 hnae3_set_field(*ol_type_vlan_len_msec,
662 HNS3_TXD_L4LEN_M, HNS3_TXD_L4LEN_S,
665 /* switch IP header ptr from outer to inner header */
666 l3.hdr = skb_inner_network_header(skb);
668 /* compute inner l2 header size, defined in 2 Bytes. */
669 l2_len = l3.hdr - l2_hdr;
670 hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_M,
671 HNS3_TXD_L2LEN_S, l2_len >> 1);
673 /* skb packet types not supported by hardware,
674 * txbd len fild doesn't be filled.
679 /* switch L4 header pointer from outer to inner */
680 l4.hdr = skb_inner_transport_header(skb);
682 l4_proto = il4_proto;
685 /* compute inner(/normal) L3 header size, defined in 4 Bytes */
686 l3_len = l4.hdr - l3.hdr;
687 hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_M,
688 HNS3_TXD_L3LEN_S, l3_len >> 2);
690 /* compute inner(/normal) L4 header size, defined in 4 Bytes */
693 hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_M,
694 HNS3_TXD_L4LEN_S, l4.tcp->doff);
697 hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_M,
699 (sizeof(struct sctphdr) >> 2));
702 hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_M,
704 (sizeof(struct udphdr) >> 2));
707 /* skb packet types not supported by hardware,
708 * txbd len fild doesn't be filled.
714 /* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
715 * and it is udp packet, which has a dest port as the IANA assigned.
716 * the hardware is expected to do the checksum offload, but the
717 * hardware will not do the checksum offload when udp dest port is
720 static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
722 #define IANA_VXLAN_PORT 4789
726 struct gre_base_hdr *gre;
730 l4.hdr = skb_transport_header(skb);
732 if (!(!skb->encapsulation && l4.udp->dest == htons(IANA_VXLAN_PORT)))
735 skb_checksum_help(skb);
740 static int hns3_set_l3l4_type_csum(struct sk_buff *skb, u8 ol4_proto,
741 u8 il4_proto, u32 *type_cs_vlan_tso,
742 u32 *ol_type_vlan_len_msec)
749 u32 l4_proto = ol4_proto;
751 l3.hdr = skb_network_header(skb);
753 /* define OL3 type and tunnel type(OL4).*/
754 if (skb->encapsulation) {
755 /* define outer network header type.*/
756 if (skb->protocol == htons(ETH_P_IP)) {
758 hnae3_set_field(*ol_type_vlan_len_msec,
761 HNS3_OL3T_IPV4_CSUM);
763 hnae3_set_field(*ol_type_vlan_len_msec,
766 HNS3_OL3T_IPV4_NO_CSUM);
768 } else if (skb->protocol == htons(ETH_P_IPV6)) {
769 hnae3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_M,
770 HNS3_TXD_OL3T_S, HNS3_OL3T_IPV6);
773 /* define tunnel type(OL4).*/
776 hnae3_set_field(*ol_type_vlan_len_msec,
779 HNS3_TUN_MAC_IN_UDP);
782 hnae3_set_field(*ol_type_vlan_len_msec,
788 /* drop the skb tunnel packet if hardware don't support,
789 * because hardware can't calculate csum when TSO.
794 /* the stack computes the IP header already,
795 * driver calculate l4 checksum when not TSO.
797 skb_checksum_help(skb);
801 l3.hdr = skb_inner_network_header(skb);
802 l4_proto = il4_proto;
805 if (l3.v4->version == 4) {
806 hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_M,
807 HNS3_TXD_L3T_S, HNS3_L3T_IPV4);
809 /* the stack computes the IP header already, the only time we
810 * need the hardware to recompute it is in the case of TSO.
813 hnae3_set_bit(*type_cs_vlan_tso, HNS3_TXD_L3CS_B, 1);
814 } else if (l3.v6->version == 6) {
815 hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_M,
816 HNS3_TXD_L3T_S, HNS3_L3T_IPV6);
821 hnae3_set_bit(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
822 hnae3_set_field(*type_cs_vlan_tso,
828 if (hns3_tunnel_csum_bug(skb))
831 hnae3_set_bit(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
832 hnae3_set_field(*type_cs_vlan_tso,
838 hnae3_set_bit(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
839 hnae3_set_field(*type_cs_vlan_tso,
845 /* drop the skb tunnel packet if hardware don't support,
846 * because hardware can't calculate csum when TSO.
851 /* the stack computes the IP header already,
852 * driver calculate l4 checksum when not TSO.
854 skb_checksum_help(skb);
861 static void hns3_set_txbd_baseinfo(u16 *bdtp_fe_sc_vld_ra_ri, int frag_end)
863 /* Config bd buffer end */
864 hnae3_set_field(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_BDTYPE_M,
865 HNS3_TXD_BDTYPE_S, 0);
866 hnae3_set_bit(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_FE_B, !!frag_end);
867 hnae3_set_bit(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_VLD_B, 1);
868 hnae3_set_field(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_SC_M, HNS3_TXD_SC_S, 0);
871 static int hns3_fill_desc_vtags(struct sk_buff *skb,
872 struct hns3_enet_ring *tx_ring,
873 u32 *inner_vlan_flag,
878 #define HNS3_TX_VLAN_PRIO_SHIFT 13
880 if (skb->protocol == htons(ETH_P_8021Q) &&
881 !(tx_ring->tqp->handle->kinfo.netdev->features &
882 NETIF_F_HW_VLAN_CTAG_TX)) {
883 /* When HW VLAN acceleration is turned off, and the stack
884 * sets the protocol to 802.1q, the driver just need to
885 * set the protocol to the encapsulated ethertype.
887 skb->protocol = vlan_get_protocol(skb);
891 if (skb_vlan_tag_present(skb)) {
894 vlan_tag = skb_vlan_tag_get(skb);
895 vlan_tag |= (skb->priority & 0x7) << HNS3_TX_VLAN_PRIO_SHIFT;
897 /* Based on hw strategy, use out_vtag in two layer tag case,
898 * and use inner_vtag in one tag case.
900 if (skb->protocol == htons(ETH_P_8021Q)) {
901 hnae3_set_bit(*out_vlan_flag, HNS3_TXD_OVLAN_B, 1);
902 *out_vtag = vlan_tag;
904 hnae3_set_bit(*inner_vlan_flag, HNS3_TXD_VLAN_B, 1);
905 *inner_vtag = vlan_tag;
907 } else if (skb->protocol == htons(ETH_P_8021Q)) {
908 struct vlan_ethhdr *vhdr;
911 rc = skb_cow_head(skb, 0);
914 vhdr = (struct vlan_ethhdr *)skb->data;
915 vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority & 0x7)
916 << HNS3_TX_VLAN_PRIO_SHIFT);
919 skb->protocol = vlan_get_protocol(skb);
923 static int hns3_fill_desc(struct hns3_enet_ring *ring, void *priv,
924 int size, dma_addr_t dma, int frag_end,
925 enum hns_desc_type type)
927 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
928 struct hns3_desc *desc = &ring->desc[ring->next_to_use];
929 u32 ol_type_vlan_len_msec = 0;
930 u16 bdtp_fe_sc_vld_ra_ri = 0;
931 u32 type_cs_vlan_tso = 0;
941 /* The txbd's baseinfo of DESC_TYPE_PAGE & DESC_TYPE_SKB */
942 desc_cb->priv = priv;
943 desc_cb->length = size;
945 desc_cb->type = type;
947 /* now, fill the descriptor */
948 desc->addr = cpu_to_le64(dma);
949 desc->tx.send_size = cpu_to_le16((u16)size);
950 hns3_set_txbd_baseinfo(&bdtp_fe_sc_vld_ra_ri, frag_end);
951 desc->tx.bdtp_fe_sc_vld_ra_ri = cpu_to_le16(bdtp_fe_sc_vld_ra_ri);
953 if (type == DESC_TYPE_SKB) {
954 skb = (struct sk_buff *)priv;
957 ret = hns3_fill_desc_vtags(skb, ring, &type_cs_vlan_tso,
958 &ol_type_vlan_len_msec,
959 &inner_vtag, &out_vtag);
963 if (skb->ip_summed == CHECKSUM_PARTIAL) {
964 skb_reset_mac_len(skb);
966 ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
969 hns3_set_l2l3l4_len(skb, ol4_proto, il4_proto,
971 &ol_type_vlan_len_msec);
972 ret = hns3_set_l3l4_type_csum(skb, ol4_proto, il4_proto,
974 &ol_type_vlan_len_msec);
978 ret = hns3_set_tso(skb, &paylen, &mss,
985 desc->tx.ol_type_vlan_len_msec =
986 cpu_to_le32(ol_type_vlan_len_msec);
987 desc->tx.type_cs_vlan_tso_len =
988 cpu_to_le32(type_cs_vlan_tso);
989 desc->tx.paylen = cpu_to_le32(paylen);
990 desc->tx.mss = cpu_to_le16(mss);
991 desc->tx.vlan_tag = cpu_to_le16(inner_vtag);
992 desc->tx.outer_vlan_tag = cpu_to_le16(out_vtag);
995 /* move ring pointer to next.*/
996 ring_ptr_move_fw(ring, next_to_use);
1001 static int hns3_fill_desc_tso(struct hns3_enet_ring *ring, void *priv,
1002 int size, dma_addr_t dma, int frag_end,
1003 enum hns_desc_type type)
1005 unsigned int frag_buf_num;
1010 frag_buf_num = (size + HNS3_MAX_BD_SIZE - 1) / HNS3_MAX_BD_SIZE;
1011 sizeoflast = size % HNS3_MAX_BD_SIZE;
1012 sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1014 /* When the frag size is bigger than hardware, split this frag */
1015 for (k = 0; k < frag_buf_num; k++) {
1016 ret = hns3_fill_desc(ring, priv,
1017 (k == frag_buf_num - 1) ?
1018 sizeoflast : HNS3_MAX_BD_SIZE,
1019 dma + HNS3_MAX_BD_SIZE * k,
1020 frag_end && (k == frag_buf_num - 1) ? 1 : 0,
1021 (type == DESC_TYPE_SKB && !k) ?
1022 DESC_TYPE_SKB : DESC_TYPE_PAGE);
1030 static int hns3_nic_maybe_stop_tso(struct sk_buff **out_skb, int *bnum,
1031 struct hns3_enet_ring *ring)
1033 struct sk_buff *skb = *out_skb;
1034 struct skb_frag_struct *frag;
1041 size = skb_headlen(skb);
1042 buf_num = (size + HNS3_MAX_BD_SIZE - 1) / HNS3_MAX_BD_SIZE;
1044 frag_num = skb_shinfo(skb)->nr_frags;
1045 for (i = 0; i < frag_num; i++) {
1046 frag = &skb_shinfo(skb)->frags[i];
1047 size = skb_frag_size(frag);
1049 (size + HNS3_MAX_BD_SIZE - 1) / HNS3_MAX_BD_SIZE;
1050 if (bdnum_for_frag > HNS3_MAX_BD_PER_FRAG)
1053 buf_num += bdnum_for_frag;
1056 if (buf_num > ring_space(ring))
1063 static int hns3_nic_maybe_stop_tx(struct sk_buff **out_skb, int *bnum,
1064 struct hns3_enet_ring *ring)
1066 struct sk_buff *skb = *out_skb;
1069 /* No. of segments (plus a header) */
1070 buf_num = skb_shinfo(skb)->nr_frags + 1;
1072 if (unlikely(ring_space(ring) < buf_num))
1080 static void hns_nic_dma_unmap(struct hns3_enet_ring *ring, int next_to_use_orig)
1082 struct device *dev = ring_to_dev(ring);
1085 for (i = 0; i < ring->desc_num; i++) {
1086 /* check if this is where we started */
1087 if (ring->next_to_use == next_to_use_orig)
1090 /* unmap the descriptor dma address */
1091 if (ring->desc_cb[ring->next_to_use].type == DESC_TYPE_SKB)
1092 dma_unmap_single(dev,
1093 ring->desc_cb[ring->next_to_use].dma,
1094 ring->desc_cb[ring->next_to_use].length,
1098 ring->desc_cb[ring->next_to_use].dma,
1099 ring->desc_cb[ring->next_to_use].length,
1103 ring_ptr_move_bw(ring, next_to_use);
1107 netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
1109 struct hns3_nic_priv *priv = netdev_priv(netdev);
1110 struct hns3_nic_ring_data *ring_data =
1111 &tx_ring_data(priv, skb->queue_mapping);
1112 struct hns3_enet_ring *ring = ring_data->ring;
1113 struct device *dev = priv->dev;
1114 struct netdev_queue *dev_queue;
1115 struct skb_frag_struct *frag;
1116 int next_to_use_head;
1117 int next_to_use_frag;
1125 /* Prefetch the data used later */
1126 prefetch(skb->data);
1128 switch (priv->ops.maybe_stop_tx(&skb, &buf_num, ring)) {
1130 u64_stats_update_begin(&ring->syncp);
1131 ring->stats.tx_busy++;
1132 u64_stats_update_end(&ring->syncp);
1134 goto out_net_tx_busy;
1136 u64_stats_update_begin(&ring->syncp);
1137 ring->stats.sw_err_cnt++;
1138 u64_stats_update_end(&ring->syncp);
1139 netdev_err(netdev, "no memory to xmit!\n");
1146 /* No. of segments (plus a header) */
1147 seg_num = skb_shinfo(skb)->nr_frags + 1;
1148 /* Fill the first part */
1149 size = skb_headlen(skb);
1151 next_to_use_head = ring->next_to_use;
1153 dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1154 if (dma_mapping_error(dev, dma)) {
1155 netdev_err(netdev, "TX head DMA map failed\n");
1156 ring->stats.sw_err_cnt++;
1160 ret = priv->ops.fill_desc(ring, skb, size, dma, seg_num == 1 ? 1 : 0,
1163 goto head_dma_map_err;
1165 next_to_use_frag = ring->next_to_use;
1166 /* Fill the fragments */
1167 for (i = 1; i < seg_num; i++) {
1168 frag = &skb_shinfo(skb)->frags[i - 1];
1169 size = skb_frag_size(frag);
1170 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1171 if (dma_mapping_error(dev, dma)) {
1172 netdev_err(netdev, "TX frag(%d) DMA map failed\n", i);
1173 ring->stats.sw_err_cnt++;
1174 goto frag_dma_map_err;
1176 ret = priv->ops.fill_desc(ring, skb_frag_page(frag), size, dma,
1177 seg_num - 1 == i ? 1 : 0,
1181 goto frag_dma_map_err;
1184 /* Complete translate all packets */
1185 dev_queue = netdev_get_tx_queue(netdev, ring_data->queue_index);
1186 netdev_tx_sent_queue(dev_queue, skb->len);
1188 wmb(); /* Commit all data before submit */
1190 hnae3_queue_xmit(ring->tqp, buf_num);
1192 return NETDEV_TX_OK;
1195 hns_nic_dma_unmap(ring, next_to_use_frag);
1198 hns_nic_dma_unmap(ring, next_to_use_head);
1201 dev_kfree_skb_any(skb);
1202 return NETDEV_TX_OK;
1205 netif_stop_subqueue(netdev, ring_data->queue_index);
1206 smp_mb(); /* Commit all data before submit */
1208 return NETDEV_TX_BUSY;
1211 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
1213 struct hnae3_handle *h = hns3_get_handle(netdev);
1214 struct sockaddr *mac_addr = p;
1217 if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1218 return -EADDRNOTAVAIL;
1220 if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
1221 netdev_info(netdev, "already using mac address %pM\n",
1226 ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
1228 netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
1232 ether_addr_copy(netdev->dev_addr, mac_addr->sa_data);
1237 static int hns3_nic_do_ioctl(struct net_device *netdev,
1238 struct ifreq *ifr, int cmd)
1240 struct hnae3_handle *h = hns3_get_handle(netdev);
1242 if (!netif_running(netdev))
1245 if (!h->ae_algo->ops->do_ioctl)
1248 return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
1251 static int hns3_nic_set_features(struct net_device *netdev,
1252 netdev_features_t features)
1254 netdev_features_t changed = netdev->features ^ features;
1255 struct hns3_nic_priv *priv = netdev_priv(netdev);
1256 struct hnae3_handle *h = priv->ae_handle;
1259 if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) {
1260 if (features & (NETIF_F_TSO | NETIF_F_TSO6)) {
1261 priv->ops.fill_desc = hns3_fill_desc_tso;
1262 priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tso;
1264 priv->ops.fill_desc = hns3_fill_desc;
1265 priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tx;
1269 if ((changed & NETIF_F_HW_VLAN_CTAG_FILTER) &&
1270 h->ae_algo->ops->enable_vlan_filter) {
1271 if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
1272 h->ae_algo->ops->enable_vlan_filter(h, true);
1274 h->ae_algo->ops->enable_vlan_filter(h, false);
1277 if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
1278 h->ae_algo->ops->enable_hw_strip_rxvtag) {
1279 if (features & NETIF_F_HW_VLAN_CTAG_RX)
1280 ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, true);
1282 ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, false);
1288 netdev->features = features;
1292 static void hns3_nic_get_stats64(struct net_device *netdev,
1293 struct rtnl_link_stats64 *stats)
1295 struct hns3_nic_priv *priv = netdev_priv(netdev);
1296 int queue_num = priv->ae_handle->kinfo.num_tqps;
1297 struct hnae3_handle *handle = priv->ae_handle;
1298 struct hns3_enet_ring *ring;
1308 if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
1311 handle->ae_algo->ops->update_stats(handle, &netdev->stats);
1313 for (idx = 0; idx < queue_num; idx++) {
1314 /* fetch the tx stats */
1315 ring = priv->ring_data[idx].ring;
1317 start = u64_stats_fetch_begin_irq(&ring->syncp);
1318 tx_bytes += ring->stats.tx_bytes;
1319 tx_pkts += ring->stats.tx_pkts;
1320 tx_drop += ring->stats.tx_busy;
1321 tx_drop += ring->stats.sw_err_cnt;
1322 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1324 /* fetch the rx stats */
1325 ring = priv->ring_data[idx + queue_num].ring;
1327 start = u64_stats_fetch_begin_irq(&ring->syncp);
1328 rx_bytes += ring->stats.rx_bytes;
1329 rx_pkts += ring->stats.rx_pkts;
1330 rx_drop += ring->stats.non_vld_descs;
1331 rx_drop += ring->stats.err_pkt_len;
1332 rx_drop += ring->stats.l2_err;
1333 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1336 stats->tx_bytes = tx_bytes;
1337 stats->tx_packets = tx_pkts;
1338 stats->rx_bytes = rx_bytes;
1339 stats->rx_packets = rx_pkts;
1341 stats->rx_errors = netdev->stats.rx_errors;
1342 stats->multicast = netdev->stats.multicast;
1343 stats->rx_length_errors = netdev->stats.rx_length_errors;
1344 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
1345 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
1347 stats->tx_errors = netdev->stats.tx_errors;
1348 stats->rx_dropped = rx_drop + netdev->stats.rx_dropped;
1349 stats->tx_dropped = tx_drop + netdev->stats.tx_dropped;
1350 stats->collisions = netdev->stats.collisions;
1351 stats->rx_over_errors = netdev->stats.rx_over_errors;
1352 stats->rx_frame_errors = netdev->stats.rx_frame_errors;
1353 stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
1354 stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
1355 stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
1356 stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
1357 stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
1358 stats->tx_window_errors = netdev->stats.tx_window_errors;
1359 stats->rx_compressed = netdev->stats.rx_compressed;
1360 stats->tx_compressed = netdev->stats.tx_compressed;
1363 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
1365 struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
1366 struct hnae3_handle *h = hns3_get_handle(netdev);
1367 struct hnae3_knic_private_info *kinfo = &h->kinfo;
1368 u8 *prio_tc = mqprio_qopt->qopt.prio_tc_map;
1369 u8 tc = mqprio_qopt->qopt.num_tc;
1370 u16 mode = mqprio_qopt->mode;
1371 u8 hw = mqprio_qopt->qopt.hw;
1375 if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
1376 mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
1379 if (tc > HNAE3_MAX_TC)
1385 if_running = netif_running(netdev);
1387 hns3_nic_net_stop(netdev);
1391 ret = (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
1392 kinfo->dcb_ops->setup_tc(h, tc, prio_tc) : -EOPNOTSUPP;
1396 ret = hns3_nic_set_real_num_queue(netdev);
1400 hns3_nic_net_open(netdev);
1405 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
1408 if (type != TC_SETUP_QDISC_MQPRIO)
1411 return hns3_setup_tc(dev, type_data);
1414 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
1415 __be16 proto, u16 vid)
1417 struct hnae3_handle *h = hns3_get_handle(netdev);
1418 struct hns3_nic_priv *priv = netdev_priv(netdev);
1421 if (h->ae_algo->ops->set_vlan_filter)
1422 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
1425 set_bit(vid, priv->active_vlans);
1430 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
1431 __be16 proto, u16 vid)
1433 struct hnae3_handle *h = hns3_get_handle(netdev);
1434 struct hns3_nic_priv *priv = netdev_priv(netdev);
1437 if (h->ae_algo->ops->set_vlan_filter)
1438 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
1441 clear_bit(vid, priv->active_vlans);
1446 static void hns3_restore_vlan(struct net_device *netdev)
1448 struct hns3_nic_priv *priv = netdev_priv(netdev);
1452 for_each_set_bit(vid, priv->active_vlans, VLAN_N_VID) {
1453 ret = hns3_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
1455 netdev_warn(netdev, "Restore vlan: %d filter, ret:%d\n",
1460 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
1461 u8 qos, __be16 vlan_proto)
1463 struct hnae3_handle *h = hns3_get_handle(netdev);
1466 if (h->ae_algo->ops->set_vf_vlan_filter)
1467 ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
1473 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
1475 struct hnae3_handle *h = hns3_get_handle(netdev);
1476 bool if_running = netif_running(netdev);
1479 if (!h->ae_algo->ops->set_mtu)
1482 /* if this was called with netdev up then bring netdevice down */
1484 (void)hns3_nic_net_stop(netdev);
1488 ret = h->ae_algo->ops->set_mtu(h, new_mtu);
1490 netdev_err(netdev, "failed to change MTU in hardware %d\n",
1493 netdev->mtu = new_mtu;
1495 /* if the netdev was running earlier, bring it up again */
1496 if (if_running && hns3_nic_net_open(netdev))
1502 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
1504 struct hns3_nic_priv *priv = netdev_priv(ndev);
1505 struct hns3_enet_ring *tx_ring = NULL;
1506 int timeout_queue = 0;
1507 int hw_head, hw_tail;
1510 /* Find the stopped queue the same way the stack does */
1511 for (i = 0; i < ndev->real_num_tx_queues; i++) {
1512 struct netdev_queue *q;
1513 unsigned long trans_start;
1515 q = netdev_get_tx_queue(ndev, i);
1516 trans_start = q->trans_start;
1517 if (netif_xmit_stopped(q) &&
1519 (trans_start + ndev->watchdog_timeo))) {
1525 if (i == ndev->num_tx_queues) {
1527 "no netdev TX timeout queue found, timeout count: %llu\n",
1528 priv->tx_timeout_count);
1532 tx_ring = priv->ring_data[timeout_queue].ring;
1534 hw_head = readl_relaxed(tx_ring->tqp->io_base +
1535 HNS3_RING_TX_RING_HEAD_REG);
1536 hw_tail = readl_relaxed(tx_ring->tqp->io_base +
1537 HNS3_RING_TX_RING_TAIL_REG);
1539 "tx_timeout count: %llu, queue id: %d, SW_NTU: 0x%x, SW_NTC: 0x%x, HW_HEAD: 0x%x, HW_TAIL: 0x%x, INT: 0x%x\n",
1540 priv->tx_timeout_count,
1542 tx_ring->next_to_use,
1543 tx_ring->next_to_clean,
1546 readl(tx_ring->tqp_vector->mask_addr));
1551 static void hns3_nic_net_timeout(struct net_device *ndev)
1553 struct hns3_nic_priv *priv = netdev_priv(ndev);
1554 struct hnae3_handle *h = priv->ae_handle;
1556 if (!hns3_get_tx_timeo_queue_info(ndev))
1559 priv->tx_timeout_count++;
1561 if (time_before(jiffies, (h->last_reset_time + ndev->watchdog_timeo)))
1564 /* request the reset */
1565 if (h->ae_algo->ops->reset_event)
1566 h->ae_algo->ops->reset_event(h);
1569 static const struct net_device_ops hns3_nic_netdev_ops = {
1570 .ndo_open = hns3_nic_net_open,
1571 .ndo_stop = hns3_nic_net_stop,
1572 .ndo_start_xmit = hns3_nic_net_xmit,
1573 .ndo_tx_timeout = hns3_nic_net_timeout,
1574 .ndo_set_mac_address = hns3_nic_net_set_mac_address,
1575 .ndo_do_ioctl = hns3_nic_do_ioctl,
1576 .ndo_change_mtu = hns3_nic_change_mtu,
1577 .ndo_set_features = hns3_nic_set_features,
1578 .ndo_get_stats64 = hns3_nic_get_stats64,
1579 .ndo_setup_tc = hns3_nic_setup_tc,
1580 .ndo_set_rx_mode = hns3_nic_set_rx_mode,
1581 .ndo_vlan_rx_add_vid = hns3_vlan_rx_add_vid,
1582 .ndo_vlan_rx_kill_vid = hns3_vlan_rx_kill_vid,
1583 .ndo_set_vf_vlan = hns3_ndo_set_vf_vlan,
1586 static bool hns3_is_phys_func(struct pci_dev *pdev)
1588 u32 dev_id = pdev->device;
1591 case HNAE3_DEV_ID_GE:
1592 case HNAE3_DEV_ID_25GE:
1593 case HNAE3_DEV_ID_25GE_RDMA:
1594 case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
1595 case HNAE3_DEV_ID_50GE_RDMA:
1596 case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
1597 case HNAE3_DEV_ID_100G_RDMA_MACSEC:
1599 case HNAE3_DEV_ID_100G_VF:
1600 case HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF:
1603 dev_warn(&pdev->dev, "un-recognized pci device-id %d",
1610 static void hns3_disable_sriov(struct pci_dev *pdev)
1612 /* If our VFs are assigned we cannot shut down SR-IOV
1613 * without causing issues, so just leave the hardware
1614 * available but disabled
1616 if (pci_vfs_assigned(pdev)) {
1617 dev_warn(&pdev->dev,
1618 "disabling driver while VFs are assigned\n");
1622 pci_disable_sriov(pdev);
1625 static void hns3_get_dev_capability(struct pci_dev *pdev,
1626 struct hnae3_ae_dev *ae_dev)
1628 if (pdev->revision >= 0x21)
1629 hnae3_set_bit(ae_dev->flag, HNAE3_DEV_SUPPORT_FD_B, 1);
1632 /* hns3_probe - Device initialization routine
1633 * @pdev: PCI device information struct
1634 * @ent: entry in hns3_pci_tbl
1636 * hns3_probe initializes a PF identified by a pci_dev structure.
1637 * The OS initialization, configuring of the PF private structure,
1638 * and a hardware reset occur.
1640 * Returns 0 on success, negative on failure
1642 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1644 struct hnae3_ae_dev *ae_dev;
1647 ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev),
1654 ae_dev->pdev = pdev;
1655 ae_dev->flag = ent->driver_data;
1656 ae_dev->dev_type = HNAE3_DEV_KNIC;
1657 ae_dev->reset_type = HNAE3_NONE_RESET;
1658 hns3_get_dev_capability(pdev, ae_dev);
1659 pci_set_drvdata(pdev, ae_dev);
1661 hnae3_register_ae_dev(ae_dev);
1666 /* hns3_remove - Device removal routine
1667 * @pdev: PCI device information struct
1669 static void hns3_remove(struct pci_dev *pdev)
1671 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1673 if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
1674 hns3_disable_sriov(pdev);
1676 hnae3_unregister_ae_dev(ae_dev);
1680 * hns3_pci_sriov_configure
1681 * @pdev: pointer to a pci_dev structure
1682 * @num_vfs: number of VFs to allocate
1684 * Enable or change the number of VFs. Called when the user updates the number
1687 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
1691 if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
1692 dev_warn(&pdev->dev, "Can not config SRIOV\n");
1697 ret = pci_enable_sriov(pdev, num_vfs);
1699 dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
1702 } else if (!pci_vfs_assigned(pdev)) {
1703 pci_disable_sriov(pdev);
1705 dev_warn(&pdev->dev,
1706 "Unable to free VFs because some are assigned to VMs.\n");
1712 static void hns3_shutdown(struct pci_dev *pdev)
1714 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1716 hnae3_unregister_ae_dev(ae_dev);
1717 devm_kfree(&pdev->dev, ae_dev);
1718 pci_set_drvdata(pdev, NULL);
1720 if (system_state == SYSTEM_POWER_OFF)
1721 pci_set_power_state(pdev, PCI_D3hot);
1724 static struct pci_driver hns3_driver = {
1725 .name = hns3_driver_name,
1726 .id_table = hns3_pci_tbl,
1727 .probe = hns3_probe,
1728 .remove = hns3_remove,
1729 .shutdown = hns3_shutdown,
1730 .sriov_configure = hns3_pci_sriov_configure,
1733 /* set default feature to hns3 */
1734 static void hns3_set_default_feature(struct net_device *netdev)
1736 struct hnae3_handle *h = hns3_get_handle(netdev);
1737 struct pci_dev *pdev = h->pdev;
1739 netdev->priv_flags |= IFF_UNICAST_FLT;
1741 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1742 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1743 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1744 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1745 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1747 netdev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
1749 netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
1751 netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1752 NETIF_F_HW_VLAN_CTAG_FILTER |
1753 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
1754 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1755 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1756 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1757 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1759 netdev->vlan_features |=
1760 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
1761 NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO |
1762 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1763 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1764 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1766 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1767 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
1768 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1769 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1770 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1771 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1773 if (pdev->revision != 0x20)
1774 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
1777 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
1778 struct hns3_desc_cb *cb)
1780 unsigned int order = hnae3_page_order(ring);
1783 p = dev_alloc_pages(order);
1788 cb->page_offset = 0;
1790 cb->buf = page_address(p);
1791 cb->length = hnae3_page_size(ring);
1792 cb->type = DESC_TYPE_PAGE;
1797 static void hns3_free_buffer(struct hns3_enet_ring *ring,
1798 struct hns3_desc_cb *cb)
1800 if (cb->type == DESC_TYPE_SKB)
1801 dev_kfree_skb_any((struct sk_buff *)cb->priv);
1802 else if (!HNAE3_IS_TX_RING(ring))
1803 put_page((struct page *)cb->priv);
1804 memset(cb, 0, sizeof(*cb));
1807 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
1809 cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
1810 cb->length, ring_to_dma_dir(ring));
1812 if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
1818 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
1819 struct hns3_desc_cb *cb)
1821 if (cb->type == DESC_TYPE_SKB)
1822 dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
1823 ring_to_dma_dir(ring));
1825 dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
1826 ring_to_dma_dir(ring));
1829 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
1831 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
1832 ring->desc[i].addr = 0;
1835 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i)
1837 struct hns3_desc_cb *cb = &ring->desc_cb[i];
1839 if (!ring->desc_cb[i].dma)
1842 hns3_buffer_detach(ring, i);
1843 hns3_free_buffer(ring, cb);
1846 static void hns3_free_buffers(struct hns3_enet_ring *ring)
1850 for (i = 0; i < ring->desc_num; i++)
1851 hns3_free_buffer_detach(ring, i);
1854 /* free desc along with its attached buffer */
1855 static void hns3_free_desc(struct hns3_enet_ring *ring)
1857 int size = ring->desc_num * sizeof(ring->desc[0]);
1859 hns3_free_buffers(ring);
1862 dma_free_coherent(ring_to_dev(ring), size,
1863 ring->desc, ring->desc_dma_addr);
1868 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
1870 int size = ring->desc_num * sizeof(ring->desc[0]);
1872 ring->desc = dma_zalloc_coherent(ring_to_dev(ring), size,
1873 &ring->desc_dma_addr,
1881 static int hns3_reserve_buffer_map(struct hns3_enet_ring *ring,
1882 struct hns3_desc_cb *cb)
1886 ret = hns3_alloc_buffer(ring, cb);
1890 ret = hns3_map_buffer(ring, cb);
1897 hns3_free_buffer(ring, cb);
1902 static int hns3_alloc_buffer_attach(struct hns3_enet_ring *ring, int i)
1904 int ret = hns3_reserve_buffer_map(ring, &ring->desc_cb[i]);
1909 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
1914 /* Allocate memory for raw pkg, and map with dma */
1915 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
1919 for (i = 0; i < ring->desc_num; i++) {
1920 ret = hns3_alloc_buffer_attach(ring, i);
1922 goto out_buffer_fail;
1928 for (j = i - 1; j >= 0; j--)
1929 hns3_free_buffer_detach(ring, j);
1933 /* detach a in-used buffer and replace with a reserved one */
1934 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
1935 struct hns3_desc_cb *res_cb)
1937 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
1938 ring->desc_cb[i] = *res_cb;
1939 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
1940 ring->desc[i].rx.bd_base_info = 0;
1943 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
1945 ring->desc_cb[i].reuse_flag = 0;
1946 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma
1947 + ring->desc_cb[i].page_offset);
1948 ring->desc[i].rx.bd_base_info = 0;
1951 static void hns3_nic_reclaim_one_desc(struct hns3_enet_ring *ring, int *bytes,
1954 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
1956 (*pkts) += (desc_cb->type == DESC_TYPE_SKB);
1957 (*bytes) += desc_cb->length;
1958 /* desc_cb will be cleaned, after hnae3_free_buffer_detach*/
1959 hns3_free_buffer_detach(ring, ring->next_to_clean);
1961 ring_ptr_move_fw(ring, next_to_clean);
1964 static int is_valid_clean_head(struct hns3_enet_ring *ring, int h)
1966 int u = ring->next_to_use;
1967 int c = ring->next_to_clean;
1969 if (unlikely(h > ring->desc_num))
1972 return u > c ? (h > c && h <= u) : (h > c || h <= u);
1975 void hns3_clean_tx_ring(struct hns3_enet_ring *ring)
1977 struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
1978 struct hns3_nic_priv *priv = netdev_priv(netdev);
1979 struct netdev_queue *dev_queue;
1983 head = readl_relaxed(ring->tqp->io_base + HNS3_RING_TX_RING_HEAD_REG);
1984 rmb(); /* Make sure head is ready before touch any data */
1986 if (is_ring_empty(ring) || head == ring->next_to_clean)
1987 return; /* no data to poll */
1989 if (unlikely(!is_valid_clean_head(ring, head))) {
1990 netdev_err(netdev, "wrong head (%d, %d-%d)\n", head,
1991 ring->next_to_use, ring->next_to_clean);
1993 u64_stats_update_begin(&ring->syncp);
1994 ring->stats.io_err_cnt++;
1995 u64_stats_update_end(&ring->syncp);
2001 while (head != ring->next_to_clean) {
2002 hns3_nic_reclaim_one_desc(ring, &bytes, &pkts);
2003 /* Issue prefetch for next Tx descriptor */
2004 prefetch(&ring->desc_cb[ring->next_to_clean]);
2007 ring->tqp_vector->tx_group.total_bytes += bytes;
2008 ring->tqp_vector->tx_group.total_packets += pkts;
2010 u64_stats_update_begin(&ring->syncp);
2011 ring->stats.tx_bytes += bytes;
2012 ring->stats.tx_pkts += pkts;
2013 u64_stats_update_end(&ring->syncp);
2015 dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
2016 netdev_tx_completed_queue(dev_queue, pkts, bytes);
2018 if (unlikely(pkts && netif_carrier_ok(netdev) &&
2019 (ring_space(ring) > HNS3_MAX_BD_PER_PKT))) {
2020 /* Make sure that anybody stopping the queue after this
2021 * sees the new next_to_clean.
2024 if (netif_tx_queue_stopped(dev_queue) &&
2025 !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
2026 netif_tx_wake_queue(dev_queue);
2027 ring->stats.restart_queue++;
2032 static int hns3_desc_unused(struct hns3_enet_ring *ring)
2034 int ntc = ring->next_to_clean;
2035 int ntu = ring->next_to_use;
2037 return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
2041 hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring, int cleand_count)
2043 struct hns3_desc_cb *desc_cb;
2044 struct hns3_desc_cb res_cbs;
2047 for (i = 0; i < cleand_count; i++) {
2048 desc_cb = &ring->desc_cb[ring->next_to_use];
2049 if (desc_cb->reuse_flag) {
2050 u64_stats_update_begin(&ring->syncp);
2051 ring->stats.reuse_pg_cnt++;
2052 u64_stats_update_end(&ring->syncp);
2054 hns3_reuse_buffer(ring, ring->next_to_use);
2056 ret = hns3_reserve_buffer_map(ring, &res_cbs);
2058 u64_stats_update_begin(&ring->syncp);
2059 ring->stats.sw_err_cnt++;
2060 u64_stats_update_end(&ring->syncp);
2062 netdev_err(ring->tqp->handle->kinfo.netdev,
2063 "hnae reserve buffer map failed.\n");
2066 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
2069 ring_ptr_move_fw(ring, next_to_use);
2072 wmb(); /* Make all data has been write before submit */
2073 writel_relaxed(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
2076 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
2077 struct hns3_enet_ring *ring, int pull_len,
2078 struct hns3_desc_cb *desc_cb)
2080 struct hns3_desc *desc;
2086 twobufs = ((PAGE_SIZE < 8192) &&
2087 hnae3_buf_size(ring) == HNS3_BUFFER_SIZE_2048);
2089 desc = &ring->desc[ring->next_to_clean];
2090 size = le16_to_cpu(desc->rx.size);
2092 truesize = hnae3_buf_size(ring);
2095 last_offset = hnae3_page_size(ring) - hnae3_buf_size(ring);
2097 skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
2098 size - pull_len, truesize);
2100 /* Avoid re-using remote pages,flag default unreuse */
2101 if (unlikely(page_to_nid(desc_cb->priv) != numa_node_id()))
2105 /* If we are only owner of page we can reuse it */
2106 if (likely(page_count(desc_cb->priv) == 1)) {
2107 /* Flip page offset to other buffer */
2108 desc_cb->page_offset ^= truesize;
2110 desc_cb->reuse_flag = 1;
2111 /* bump ref count on page before it is given*/
2112 get_page(desc_cb->priv);
2117 /* Move offset up to the next cache line */
2118 desc_cb->page_offset += truesize;
2120 if (desc_cb->page_offset <= last_offset) {
2121 desc_cb->reuse_flag = 1;
2122 /* Bump ref count on page before it is given*/
2123 get_page(desc_cb->priv);
2127 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
2128 struct hns3_desc *desc)
2130 struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2131 int l3_type, l4_type;
2136 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2137 l234info = le32_to_cpu(desc->rx.l234_info);
2139 skb->ip_summed = CHECKSUM_NONE;
2141 skb_checksum_none_assert(skb);
2143 if (!(netdev->features & NETIF_F_RXCSUM))
2146 /* check if hardware has done checksum */
2147 if (!hnae3_get_bit(bd_base_info, HNS3_RXD_L3L4P_B))
2150 if (unlikely(hnae3_get_bit(l234info, HNS3_RXD_L3E_B) ||
2151 hnae3_get_bit(l234info, HNS3_RXD_L4E_B) ||
2152 hnae3_get_bit(l234info, HNS3_RXD_OL3E_B) ||
2153 hnae3_get_bit(l234info, HNS3_RXD_OL4E_B))) {
2154 u64_stats_update_begin(&ring->syncp);
2155 ring->stats.l3l4_csum_err++;
2156 u64_stats_update_end(&ring->syncp);
2161 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
2163 l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
2166 ol4_type = hnae3_get_field(l234info, HNS3_RXD_OL4ID_M,
2169 case HNS3_OL4_TYPE_MAC_IN_UDP:
2170 case HNS3_OL4_TYPE_NVGRE:
2171 skb->csum_level = 1;
2173 case HNS3_OL4_TYPE_NO_TUN:
2174 /* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
2175 if ((l3_type == HNS3_L3_TYPE_IPV4 ||
2176 l3_type == HNS3_L3_TYPE_IPV6) &&
2177 (l4_type == HNS3_L4_TYPE_UDP ||
2178 l4_type == HNS3_L4_TYPE_TCP ||
2179 l4_type == HNS3_L4_TYPE_SCTP))
2180 skb->ip_summed = CHECKSUM_UNNECESSARY;
2187 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
2189 napi_gro_receive(&ring->tqp_vector->napi, skb);
2192 static u16 hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
2193 struct hns3_desc *desc, u32 l234info)
2195 struct pci_dev *pdev = ring->tqp->handle->pdev;
2198 if (pdev->revision == 0x20) {
2199 vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2200 if (!(vlan_tag & VLAN_VID_MASK))
2201 vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2206 #define HNS3_STRP_OUTER_VLAN 0x1
2207 #define HNS3_STRP_INNER_VLAN 0x2
2209 switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
2210 HNS3_RXD_STRP_TAGP_S)) {
2211 case HNS3_STRP_OUTER_VLAN:
2212 vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2214 case HNS3_STRP_INNER_VLAN:
2215 vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2225 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring,
2226 struct sk_buff **out_skb, int *out_bnum)
2228 struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2229 struct hns3_desc_cb *desc_cb;
2230 struct hns3_desc *desc;
2231 struct sk_buff *skb;
2239 desc = &ring->desc[ring->next_to_clean];
2240 desc_cb = &ring->desc_cb[ring->next_to_clean];
2244 length = le16_to_cpu(desc->rx.size);
2245 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2247 /* Check valid BD */
2248 if (unlikely(!hnae3_get_bit(bd_base_info, HNS3_RXD_VLD_B)))
2251 va = (unsigned char *)desc_cb->buf + desc_cb->page_offset;
2253 /* Prefetch first cache line of first page
2254 * Idea is to cache few bytes of the header of the packet. Our L1 Cache
2255 * line size is 64B so need to prefetch twice to make it 128B. But in
2256 * actual we can have greater size of caches with 128B Level 1 cache
2257 * lines. In such a case, single fetch would suffice to cache in the
2258 * relevant part of the header.
2261 #if L1_CACHE_BYTES < 128
2262 prefetch(va + L1_CACHE_BYTES);
2265 skb = *out_skb = napi_alloc_skb(&ring->tqp_vector->napi,
2267 if (unlikely(!skb)) {
2268 netdev_err(netdev, "alloc rx skb fail\n");
2270 u64_stats_update_begin(&ring->syncp);
2271 ring->stats.sw_err_cnt++;
2272 u64_stats_update_end(&ring->syncp);
2277 prefetchw(skb->data);
2280 if (length <= HNS3_RX_HEAD_SIZE) {
2281 memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
2283 /* We can reuse buffer as-is, just make sure it is local */
2284 if (likely(page_to_nid(desc_cb->priv) == numa_node_id()))
2285 desc_cb->reuse_flag = 1;
2286 else /* This page cannot be reused so discard it */
2287 put_page(desc_cb->priv);
2289 ring_ptr_move_fw(ring, next_to_clean);
2291 u64_stats_update_begin(&ring->syncp);
2292 ring->stats.seg_pkt_cnt++;
2293 u64_stats_update_end(&ring->syncp);
2295 pull_len = eth_get_headlen(va, HNS3_RX_HEAD_SIZE);
2297 memcpy(__skb_put(skb, pull_len), va,
2298 ALIGN(pull_len, sizeof(long)));
2300 hns3_nic_reuse_page(skb, 0, ring, pull_len, desc_cb);
2301 ring_ptr_move_fw(ring, next_to_clean);
2303 while (!hnae3_get_bit(bd_base_info, HNS3_RXD_FE_B)) {
2304 desc = &ring->desc[ring->next_to_clean];
2305 desc_cb = &ring->desc_cb[ring->next_to_clean];
2306 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2307 hns3_nic_reuse_page(skb, bnum, ring, 0, desc_cb);
2308 ring_ptr_move_fw(ring, next_to_clean);
2315 l234info = le32_to_cpu(desc->rx.l234_info);
2317 /* Based on hw strategy, the tag offloaded will be stored at
2318 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
2319 * in one layer tag case.
2321 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
2324 vlan_tag = hns3_parse_vlan_tag(ring, desc, l234info);
2325 if (vlan_tag & VLAN_VID_MASK)
2326 __vlan_hwaccel_put_tag(skb,
2331 if (unlikely(!hnae3_get_bit(bd_base_info, HNS3_RXD_VLD_B))) {
2332 u64_stats_update_begin(&ring->syncp);
2333 ring->stats.non_vld_descs++;
2334 u64_stats_update_end(&ring->syncp);
2336 dev_kfree_skb_any(skb);
2340 if (unlikely((!desc->rx.pkt_len) ||
2341 hnae3_get_bit(l234info, HNS3_RXD_TRUNCAT_B))) {
2342 u64_stats_update_begin(&ring->syncp);
2343 ring->stats.err_pkt_len++;
2344 u64_stats_update_end(&ring->syncp);
2346 dev_kfree_skb_any(skb);
2350 if (unlikely(hnae3_get_bit(l234info, HNS3_RXD_L2E_B))) {
2351 u64_stats_update_begin(&ring->syncp);
2352 ring->stats.l2_err++;
2353 u64_stats_update_end(&ring->syncp);
2355 dev_kfree_skb_any(skb);
2359 u64_stats_update_begin(&ring->syncp);
2360 ring->stats.rx_pkts++;
2361 ring->stats.rx_bytes += skb->len;
2362 u64_stats_update_end(&ring->syncp);
2364 ring->tqp_vector->rx_group.total_bytes += skb->len;
2366 hns3_rx_checksum(ring, skb, desc);
2370 int hns3_clean_rx_ring(
2371 struct hns3_enet_ring *ring, int budget,
2372 void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
2374 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
2375 struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2376 int recv_pkts, recv_bds, clean_count, err;
2377 int unused_count = hns3_desc_unused(ring);
2378 struct sk_buff *skb = NULL;
2381 num = readl_relaxed(ring->tqp->io_base + HNS3_RING_RX_RING_FBDNUM_REG);
2382 rmb(); /* Make sure num taken effect before the other data is touched */
2384 recv_pkts = 0, recv_bds = 0, clean_count = 0;
2385 num -= unused_count;
2387 while (recv_pkts < budget && recv_bds < num) {
2388 /* Reuse or realloc buffers */
2389 if (clean_count + unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
2390 hns3_nic_alloc_rx_buffers(ring,
2391 clean_count + unused_count);
2393 unused_count = hns3_desc_unused(ring);
2397 err = hns3_handle_rx_bd(ring, &skb, &bnum);
2398 if (unlikely(!skb)) /* This fault cannot be repaired */
2402 clean_count += bnum;
2403 if (unlikely(err)) { /* Do jump the err */
2408 /* Do update ip stack process */
2409 skb->protocol = eth_type_trans(skb, netdev);
2416 /* Make all data has been write before submit */
2417 if (clean_count + unused_count > 0)
2418 hns3_nic_alloc_rx_buffers(ring,
2419 clean_count + unused_count);
2424 static bool hns3_get_new_int_gl(struct hns3_enet_ring_group *ring_group)
2426 struct hns3_enet_tqp_vector *tqp_vector =
2427 ring_group->ring->tqp_vector;
2428 enum hns3_flow_level_range new_flow_level;
2429 int packets_per_msecs;
2430 int bytes_per_msecs;
2434 if (!ring_group->coal.int_gl || !tqp_vector->last_jiffies)
2437 if (ring_group->total_packets == 0) {
2438 ring_group->coal.int_gl = HNS3_INT_GL_50K;
2439 ring_group->coal.flow_level = HNS3_FLOW_LOW;
2443 /* Simple throttlerate management
2444 * 0-10MB/s lower (50000 ints/s)
2445 * 10-20MB/s middle (20000 ints/s)
2446 * 20-1249MB/s high (18000 ints/s)
2447 * > 40000pps ultra (8000 ints/s)
2449 new_flow_level = ring_group->coal.flow_level;
2450 new_int_gl = ring_group->coal.int_gl;
2452 jiffies_to_msecs(jiffies - tqp_vector->last_jiffies);
2454 if (!time_passed_ms)
2457 do_div(ring_group->total_packets, time_passed_ms);
2458 packets_per_msecs = ring_group->total_packets;
2460 do_div(ring_group->total_bytes, time_passed_ms);
2461 bytes_per_msecs = ring_group->total_bytes;
2463 #define HNS3_RX_LOW_BYTE_RATE 10000
2464 #define HNS3_RX_MID_BYTE_RATE 20000
2466 switch (new_flow_level) {
2468 if (bytes_per_msecs > HNS3_RX_LOW_BYTE_RATE)
2469 new_flow_level = HNS3_FLOW_MID;
2472 if (bytes_per_msecs > HNS3_RX_MID_BYTE_RATE)
2473 new_flow_level = HNS3_FLOW_HIGH;
2474 else if (bytes_per_msecs <= HNS3_RX_LOW_BYTE_RATE)
2475 new_flow_level = HNS3_FLOW_LOW;
2477 case HNS3_FLOW_HIGH:
2478 case HNS3_FLOW_ULTRA:
2480 if (bytes_per_msecs <= HNS3_RX_MID_BYTE_RATE)
2481 new_flow_level = HNS3_FLOW_MID;
2485 #define HNS3_RX_ULTRA_PACKET_RATE 40
2487 if (packets_per_msecs > HNS3_RX_ULTRA_PACKET_RATE &&
2488 &tqp_vector->rx_group == ring_group)
2489 new_flow_level = HNS3_FLOW_ULTRA;
2491 switch (new_flow_level) {
2493 new_int_gl = HNS3_INT_GL_50K;
2496 new_int_gl = HNS3_INT_GL_20K;
2498 case HNS3_FLOW_HIGH:
2499 new_int_gl = HNS3_INT_GL_18K;
2501 case HNS3_FLOW_ULTRA:
2502 new_int_gl = HNS3_INT_GL_8K;
2508 ring_group->total_bytes = 0;
2509 ring_group->total_packets = 0;
2510 ring_group->coal.flow_level = new_flow_level;
2511 if (new_int_gl != ring_group->coal.int_gl) {
2512 ring_group->coal.int_gl = new_int_gl;
2518 static void hns3_update_new_int_gl(struct hns3_enet_tqp_vector *tqp_vector)
2520 struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
2521 struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
2522 bool rx_update, tx_update;
2524 if (tqp_vector->int_adapt_down > 0) {
2525 tqp_vector->int_adapt_down--;
2529 if (rx_group->coal.gl_adapt_enable) {
2530 rx_update = hns3_get_new_int_gl(rx_group);
2532 hns3_set_vector_coalesce_rx_gl(tqp_vector,
2533 rx_group->coal.int_gl);
2536 if (tx_group->coal.gl_adapt_enable) {
2537 tx_update = hns3_get_new_int_gl(&tqp_vector->tx_group);
2539 hns3_set_vector_coalesce_tx_gl(tqp_vector,
2540 tx_group->coal.int_gl);
2543 tqp_vector->last_jiffies = jiffies;
2544 tqp_vector->int_adapt_down = HNS3_INT_ADAPT_DOWN_START;
2547 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
2549 struct hns3_enet_ring *ring;
2550 int rx_pkt_total = 0;
2552 struct hns3_enet_tqp_vector *tqp_vector =
2553 container_of(napi, struct hns3_enet_tqp_vector, napi);
2554 bool clean_complete = true;
2557 /* Since the actual Tx work is minimal, we can give the Tx a larger
2558 * budget and be more aggressive about cleaning up the Tx descriptors.
2560 hns3_for_each_ring(ring, tqp_vector->tx_group)
2561 hns3_clean_tx_ring(ring);
2563 /* make sure rx ring budget not smaller than 1 */
2564 rx_budget = max(budget / tqp_vector->num_tqps, 1);
2566 hns3_for_each_ring(ring, tqp_vector->rx_group) {
2567 int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
2570 if (rx_cleaned >= rx_budget)
2571 clean_complete = false;
2573 rx_pkt_total += rx_cleaned;
2576 tqp_vector->rx_group.total_packets += rx_pkt_total;
2578 if (!clean_complete)
2581 napi_complete(napi);
2582 hns3_update_new_int_gl(tqp_vector);
2583 hns3_mask_vector_irq(tqp_vector, 1);
2585 return rx_pkt_total;
2588 static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
2589 struct hnae3_ring_chain_node *head)
2591 struct pci_dev *pdev = tqp_vector->handle->pdev;
2592 struct hnae3_ring_chain_node *cur_chain = head;
2593 struct hnae3_ring_chain_node *chain;
2594 struct hns3_enet_ring *tx_ring;
2595 struct hns3_enet_ring *rx_ring;
2597 tx_ring = tqp_vector->tx_group.ring;
2599 cur_chain->tqp_index = tx_ring->tqp->tqp_index;
2600 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
2601 HNAE3_RING_TYPE_TX);
2602 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
2603 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_TX);
2605 cur_chain->next = NULL;
2607 while (tx_ring->next) {
2608 tx_ring = tx_ring->next;
2610 chain = devm_kzalloc(&pdev->dev, sizeof(*chain),
2615 cur_chain->next = chain;
2616 chain->tqp_index = tx_ring->tqp->tqp_index;
2617 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
2618 HNAE3_RING_TYPE_TX);
2619 hnae3_set_field(chain->int_gl_idx,
2620 HNAE3_RING_GL_IDX_M,
2621 HNAE3_RING_GL_IDX_S,
2628 rx_ring = tqp_vector->rx_group.ring;
2629 if (!tx_ring && rx_ring) {
2630 cur_chain->next = NULL;
2631 cur_chain->tqp_index = rx_ring->tqp->tqp_index;
2632 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
2633 HNAE3_RING_TYPE_RX);
2634 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
2635 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
2637 rx_ring = rx_ring->next;
2641 chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
2645 cur_chain->next = chain;
2646 chain->tqp_index = rx_ring->tqp->tqp_index;
2647 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
2648 HNAE3_RING_TYPE_RX);
2649 hnae3_set_field(chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
2650 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
2654 rx_ring = rx_ring->next;
2660 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
2661 struct hnae3_ring_chain_node *head)
2663 struct pci_dev *pdev = tqp_vector->handle->pdev;
2664 struct hnae3_ring_chain_node *chain_tmp, *chain;
2669 chain_tmp = chain->next;
2670 devm_kfree(&pdev->dev, chain);
2675 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
2676 struct hns3_enet_ring *ring)
2678 ring->next = group->ring;
2684 static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
2686 struct pci_dev *pdev = priv->ae_handle->pdev;
2687 struct hns3_enet_tqp_vector *tqp_vector;
2688 int num_vectors = priv->vector_num;
2692 numa_node = dev_to_node(&pdev->dev);
2694 for (vector_i = 0; vector_i < num_vectors; vector_i++) {
2695 tqp_vector = &priv->tqp_vector[vector_i];
2696 cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
2697 &tqp_vector->affinity_mask);
2701 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
2703 struct hnae3_ring_chain_node vector_ring_chain;
2704 struct hnae3_handle *h = priv->ae_handle;
2705 struct hns3_enet_tqp_vector *tqp_vector;
2709 hns3_nic_set_cpumask(priv);
2711 for (i = 0; i < priv->vector_num; i++) {
2712 tqp_vector = &priv->tqp_vector[i];
2713 hns3_vector_gl_rl_init_hw(tqp_vector, priv);
2714 tqp_vector->num_tqps = 0;
2717 for (i = 0; i < h->kinfo.num_tqps; i++) {
2718 u16 vector_i = i % priv->vector_num;
2719 u16 tqp_num = h->kinfo.num_tqps;
2721 tqp_vector = &priv->tqp_vector[vector_i];
2723 hns3_add_ring_to_group(&tqp_vector->tx_group,
2724 priv->ring_data[i].ring);
2726 hns3_add_ring_to_group(&tqp_vector->rx_group,
2727 priv->ring_data[i + tqp_num].ring);
2729 priv->ring_data[i].ring->tqp_vector = tqp_vector;
2730 priv->ring_data[i + tqp_num].ring->tqp_vector = tqp_vector;
2731 tqp_vector->num_tqps++;
2734 for (i = 0; i < priv->vector_num; i++) {
2735 tqp_vector = &priv->tqp_vector[i];
2737 tqp_vector->rx_group.total_bytes = 0;
2738 tqp_vector->rx_group.total_packets = 0;
2739 tqp_vector->tx_group.total_bytes = 0;
2740 tqp_vector->tx_group.total_packets = 0;
2741 tqp_vector->handle = h;
2743 ret = hns3_get_vector_ring_chain(tqp_vector,
2744 &vector_ring_chain);
2748 ret = h->ae_algo->ops->map_ring_to_vector(h,
2749 tqp_vector->vector_irq, &vector_ring_chain);
2751 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
2756 netif_napi_add(priv->netdev, &tqp_vector->napi,
2757 hns3_nic_common_poll, NAPI_POLL_WEIGHT);
2763 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
2765 struct hnae3_handle *h = priv->ae_handle;
2766 struct hns3_enet_tqp_vector *tqp_vector;
2767 struct hnae3_vector_info *vector;
2768 struct pci_dev *pdev = h->pdev;
2769 u16 tqp_num = h->kinfo.num_tqps;
2774 /* RSS size, cpu online and vector_num should be the same */
2775 /* Should consider 2p/4p later */
2776 vector_num = min_t(u16, num_online_cpus(), tqp_num);
2777 vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
2782 vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
2784 priv->vector_num = vector_num;
2785 priv->tqp_vector = (struct hns3_enet_tqp_vector *)
2786 devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
2788 if (!priv->tqp_vector) {
2793 for (i = 0; i < priv->vector_num; i++) {
2794 tqp_vector = &priv->tqp_vector[i];
2795 tqp_vector->idx = i;
2796 tqp_vector->mask_addr = vector[i].io_addr;
2797 tqp_vector->vector_irq = vector[i].vector;
2798 hns3_vector_gl_rl_init(tqp_vector, priv);
2802 devm_kfree(&pdev->dev, vector);
2806 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
2812 static int hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
2814 struct hnae3_ring_chain_node vector_ring_chain;
2815 struct hnae3_handle *h = priv->ae_handle;
2816 struct hns3_enet_tqp_vector *tqp_vector;
2819 for (i = 0; i < priv->vector_num; i++) {
2820 tqp_vector = &priv->tqp_vector[i];
2822 ret = hns3_get_vector_ring_chain(tqp_vector,
2823 &vector_ring_chain);
2827 ret = h->ae_algo->ops->unmap_ring_from_vector(h,
2828 tqp_vector->vector_irq, &vector_ring_chain);
2832 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
2834 if (priv->tqp_vector[i].irq_init_flag == HNS3_VECTOR_INITED) {
2835 (void)irq_set_affinity_hint(
2836 priv->tqp_vector[i].vector_irq,
2838 free_irq(priv->tqp_vector[i].vector_irq,
2839 &priv->tqp_vector[i]);
2842 priv->ring_data[i].ring->irq_init_flag = HNS3_VECTOR_NOT_INITED;
2843 hns3_clear_ring_group(&tqp_vector->rx_group);
2844 hns3_clear_ring_group(&tqp_vector->tx_group);
2845 netif_napi_del(&priv->tqp_vector[i].napi);
2851 static int hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
2853 struct hnae3_handle *h = priv->ae_handle;
2854 struct pci_dev *pdev = h->pdev;
2857 for (i = 0; i < priv->vector_num; i++) {
2858 struct hns3_enet_tqp_vector *tqp_vector;
2860 tqp_vector = &priv->tqp_vector[i];
2861 ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
2866 devm_kfree(&pdev->dev, priv->tqp_vector);
2870 static int hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
2873 struct hns3_nic_ring_data *ring_data = priv->ring_data;
2874 int queue_num = priv->ae_handle->kinfo.num_tqps;
2875 struct pci_dev *pdev = priv->ae_handle->pdev;
2876 struct hns3_enet_ring *ring;
2878 ring = devm_kzalloc(&pdev->dev, sizeof(*ring), GFP_KERNEL);
2882 if (ring_type == HNAE3_RING_TYPE_TX) {
2883 ring_data[q->tqp_index].ring = ring;
2884 ring_data[q->tqp_index].queue_index = q->tqp_index;
2885 ring->io_base = (u8 __iomem *)q->io_base + HNS3_TX_REG_OFFSET;
2887 ring_data[q->tqp_index + queue_num].ring = ring;
2888 ring_data[q->tqp_index + queue_num].queue_index = q->tqp_index;
2889 ring->io_base = q->io_base;
2892 hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
2896 ring->desc_cb = NULL;
2897 ring->dev = priv->dev;
2898 ring->desc_dma_addr = 0;
2899 ring->buf_size = q->buf_size;
2900 ring->desc_num = q->desc_num;
2901 ring->next_to_use = 0;
2902 ring->next_to_clean = 0;
2907 static int hns3_queue_to_ring(struct hnae3_queue *tqp,
2908 struct hns3_nic_priv *priv)
2912 ret = hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
2916 ret = hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
2923 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
2925 struct hnae3_handle *h = priv->ae_handle;
2926 struct pci_dev *pdev = h->pdev;
2929 priv->ring_data = devm_kzalloc(&pdev->dev,
2930 array3_size(h->kinfo.num_tqps,
2931 sizeof(*priv->ring_data),
2934 if (!priv->ring_data)
2937 for (i = 0; i < h->kinfo.num_tqps; i++) {
2938 ret = hns3_queue_to_ring(h->kinfo.tqp[i], priv);
2945 devm_kfree(&pdev->dev, priv->ring_data);
2949 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
2951 struct hnae3_handle *h = priv->ae_handle;
2954 for (i = 0; i < h->kinfo.num_tqps; i++) {
2955 devm_kfree(priv->dev, priv->ring_data[i].ring);
2956 devm_kfree(priv->dev,
2957 priv->ring_data[i + h->kinfo.num_tqps].ring);
2959 devm_kfree(priv->dev, priv->ring_data);
2962 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
2966 if (ring->desc_num <= 0 || ring->buf_size <= 0)
2969 ring->desc_cb = kcalloc(ring->desc_num, sizeof(ring->desc_cb[0]),
2971 if (!ring->desc_cb) {
2976 ret = hns3_alloc_desc(ring);
2978 goto out_with_desc_cb;
2980 if (!HNAE3_IS_TX_RING(ring)) {
2981 ret = hns3_alloc_ring_buffers(ring);
2989 hns3_free_desc(ring);
2991 kfree(ring->desc_cb);
2992 ring->desc_cb = NULL;
2997 static void hns3_fini_ring(struct hns3_enet_ring *ring)
2999 hns3_free_desc(ring);
3000 kfree(ring->desc_cb);
3001 ring->desc_cb = NULL;
3002 ring->next_to_clean = 0;
3003 ring->next_to_use = 0;
3006 static int hns3_buf_size2type(u32 buf_size)
3012 bd_size_type = HNS3_BD_SIZE_512_TYPE;
3015 bd_size_type = HNS3_BD_SIZE_1024_TYPE;
3018 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
3021 bd_size_type = HNS3_BD_SIZE_4096_TYPE;
3024 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
3027 return bd_size_type;
3030 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
3032 dma_addr_t dma = ring->desc_dma_addr;
3033 struct hnae3_queue *q = ring->tqp;
3035 if (!HNAE3_IS_TX_RING(ring)) {
3036 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG,
3038 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
3039 (u32)((dma >> 31) >> 1));
3041 hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
3042 hns3_buf_size2type(ring->buf_size));
3043 hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
3044 ring->desc_num / 8 - 1);
3047 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
3049 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
3050 (u32)((dma >> 31) >> 1));
3052 hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
3053 ring->desc_num / 8 - 1);
3057 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
3059 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
3062 for (i = 0; i < HNAE3_MAX_TC; i++) {
3063 struct hnae3_tc_info *tc_info = &kinfo->tc_info[i];
3066 if (!tc_info->enable)
3069 for (j = 0; j < tc_info->tqp_count; j++) {
3070 struct hnae3_queue *q;
3072 q = priv->ring_data[tc_info->tqp_offset + j].ring->tqp;
3073 hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG,
3079 int hns3_init_all_ring(struct hns3_nic_priv *priv)
3081 struct hnae3_handle *h = priv->ae_handle;
3082 int ring_num = h->kinfo.num_tqps * 2;
3086 for (i = 0; i < ring_num; i++) {
3087 ret = hns3_alloc_ring_memory(priv->ring_data[i].ring);
3090 "Alloc ring memory fail! ret=%d\n", ret);
3091 goto out_when_alloc_ring_memory;
3094 u64_stats_init(&priv->ring_data[i].ring->syncp);
3099 out_when_alloc_ring_memory:
3100 for (j = i - 1; j >= 0; j--)
3101 hns3_fini_ring(priv->ring_data[j].ring);
3106 int hns3_uninit_all_ring(struct hns3_nic_priv *priv)
3108 struct hnae3_handle *h = priv->ae_handle;
3111 for (i = 0; i < h->kinfo.num_tqps; i++) {
3112 if (h->ae_algo->ops->reset_queue)
3113 h->ae_algo->ops->reset_queue(h, i);
3115 hns3_fini_ring(priv->ring_data[i].ring);
3116 hns3_fini_ring(priv->ring_data[i + h->kinfo.num_tqps].ring);
3121 /* Set mac addr if it is configured. or leave it to the AE driver */
3122 static void hns3_init_mac_addr(struct net_device *netdev, bool init)
3124 struct hns3_nic_priv *priv = netdev_priv(netdev);
3125 struct hnae3_handle *h = priv->ae_handle;
3126 u8 mac_addr_temp[ETH_ALEN];
3128 if (h->ae_algo->ops->get_mac_addr && init) {
3129 h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
3130 ether_addr_copy(netdev->dev_addr, mac_addr_temp);
3133 /* Check if the MAC address is valid, if not get a random one */
3134 if (!is_valid_ether_addr(netdev->dev_addr)) {
3135 eth_hw_addr_random(netdev);
3136 dev_warn(priv->dev, "using random MAC address %pM\n",
3140 if (h->ae_algo->ops->set_mac_addr)
3141 h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
3145 static void hns3_uninit_mac_addr(struct net_device *netdev)
3147 struct hns3_nic_priv *priv = netdev_priv(netdev);
3148 struct hnae3_handle *h = priv->ae_handle;
3150 if (h->ae_algo->ops->rm_uc_addr)
3151 h->ae_algo->ops->rm_uc_addr(h, netdev->dev_addr);
3154 static int hns3_restore_fd_rules(struct net_device *netdev)
3156 struct hnae3_handle *h = hns3_get_handle(netdev);
3159 if (h->ae_algo->ops->restore_fd_rules)
3160 ret = h->ae_algo->ops->restore_fd_rules(h);
3165 static void hns3_del_all_fd_rules(struct net_device *netdev, bool clear_list)
3167 struct hnae3_handle *h = hns3_get_handle(netdev);
3169 if (h->ae_algo->ops->del_all_fd_entries)
3170 h->ae_algo->ops->del_all_fd_entries(h, clear_list);
3173 static void hns3_nic_set_priv_ops(struct net_device *netdev)
3175 struct hns3_nic_priv *priv = netdev_priv(netdev);
3177 if ((netdev->features & NETIF_F_TSO) ||
3178 (netdev->features & NETIF_F_TSO6)) {
3179 priv->ops.fill_desc = hns3_fill_desc_tso;
3180 priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tso;
3182 priv->ops.fill_desc = hns3_fill_desc;
3183 priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tx;
3187 static int hns3_client_init(struct hnae3_handle *handle)
3189 struct pci_dev *pdev = handle->pdev;
3190 u16 alloc_tqps, max_rss_size;
3191 struct hns3_nic_priv *priv;
3192 struct net_device *netdev;
3195 handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
3197 netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
3201 priv = netdev_priv(netdev);
3202 priv->dev = &pdev->dev;
3203 priv->netdev = netdev;
3204 priv->ae_handle = handle;
3205 priv->ae_handle->last_reset_time = jiffies;
3206 priv->tx_timeout_count = 0;
3208 handle->kinfo.netdev = netdev;
3209 handle->priv = (void *)priv;
3211 hns3_init_mac_addr(netdev, true);
3213 hns3_set_default_feature(netdev);
3215 netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
3216 netdev->priv_flags |= IFF_UNICAST_FLT;
3217 netdev->netdev_ops = &hns3_nic_netdev_ops;
3218 SET_NETDEV_DEV(netdev, &pdev->dev);
3219 hns3_ethtool_set_ops(netdev);
3220 hns3_nic_set_priv_ops(netdev);
3222 /* Carrier off reporting is important to ethtool even BEFORE open */
3223 netif_carrier_off(netdev);
3225 if (handle->flags & HNAE3_SUPPORT_VF)
3226 handle->reset_level = HNAE3_VF_RESET;
3228 handle->reset_level = HNAE3_FUNC_RESET;
3230 ret = hns3_get_ring_config(priv);
3233 goto out_get_ring_cfg;
3236 ret = hns3_nic_alloc_vector_data(priv);
3239 goto out_alloc_vector_data;
3242 ret = hns3_nic_init_vector_data(priv);
3245 goto out_init_vector_data;
3248 ret = hns3_init_all_ring(priv);
3251 goto out_init_ring_data;
3254 ret = register_netdev(netdev);
3256 dev_err(priv->dev, "probe register netdev fail!\n");
3257 goto out_reg_netdev_fail;
3260 hns3_dcbnl_setup(handle);
3262 /* MTU range: (ETH_MIN_MTU(kernel default) - 9706) */
3263 netdev->max_mtu = HNS3_MAX_MTU - (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
3267 out_reg_netdev_fail:
3269 (void)hns3_nic_uninit_vector_data(priv);
3270 out_init_vector_data:
3271 hns3_nic_dealloc_vector_data(priv);
3272 out_alloc_vector_data:
3273 priv->ring_data = NULL;
3275 priv->ae_handle = NULL;
3276 free_netdev(netdev);
3280 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
3282 struct net_device *netdev = handle->kinfo.netdev;
3283 struct hns3_nic_priv *priv = netdev_priv(netdev);
3286 if (netdev->reg_state != NETREG_UNINITIALIZED)
3287 unregister_netdev(netdev);
3289 hns3_force_clear_all_rx_ring(handle);
3291 ret = hns3_nic_uninit_vector_data(priv);
3293 netdev_err(netdev, "uninit vector error\n");
3295 ret = hns3_nic_dealloc_vector_data(priv);
3297 netdev_err(netdev, "dealloc vector error\n");
3299 ret = hns3_uninit_all_ring(priv);
3301 netdev_err(netdev, "uninit ring error\n");
3303 hns3_put_ring_config(priv);
3305 priv->ring_data = NULL;
3307 hns3_uninit_mac_addr(netdev);
3309 free_netdev(netdev);
3312 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
3314 struct net_device *netdev = handle->kinfo.netdev;
3320 netif_carrier_on(netdev);
3321 netif_tx_wake_all_queues(netdev);
3322 netdev_info(netdev, "link up\n");
3324 netif_carrier_off(netdev);
3325 netif_tx_stop_all_queues(netdev);
3326 netdev_info(netdev, "link down\n");
3330 static int hns3_client_setup_tc(struct hnae3_handle *handle, u8 tc)
3332 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
3333 struct net_device *ndev = kinfo->netdev;
3337 if (tc > HNAE3_MAX_TC)
3343 if_running = netif_running(ndev);
3346 (void)hns3_nic_net_stop(ndev);
3350 ret = (kinfo->dcb_ops && kinfo->dcb_ops->map_update) ?
3351 kinfo->dcb_ops->map_update(handle) : -EOPNOTSUPP;
3355 ret = hns3_nic_set_real_num_queue(ndev);
3359 (void)hns3_nic_net_open(ndev);
3364 static void hns3_recover_hw_addr(struct net_device *ndev)
3366 struct netdev_hw_addr_list *list;
3367 struct netdev_hw_addr *ha, *tmp;
3369 /* go through and sync uc_addr entries to the device */
3371 list_for_each_entry_safe(ha, tmp, &list->list, list)
3372 hns3_nic_uc_sync(ndev, ha->addr);
3374 /* go through and sync mc_addr entries to the device */
3376 list_for_each_entry_safe(ha, tmp, &list->list, list)
3377 hns3_nic_mc_sync(ndev, ha->addr);
3380 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
3382 while (ring->next_to_clean != ring->next_to_use) {
3383 ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
3384 hns3_free_buffer_detach(ring, ring->next_to_clean);
3385 ring_ptr_move_fw(ring, next_to_clean);
3389 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
3391 struct hns3_desc_cb res_cbs;
3394 while (ring->next_to_use != ring->next_to_clean) {
3395 /* When a buffer is not reused, it's memory has been
3396 * freed in hns3_handle_rx_bd or will be freed by
3397 * stack, so we need to replace the buffer here.
3399 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
3400 ret = hns3_reserve_buffer_map(ring, &res_cbs);
3402 u64_stats_update_begin(&ring->syncp);
3403 ring->stats.sw_err_cnt++;
3404 u64_stats_update_end(&ring->syncp);
3405 /* if alloc new buffer fail, exit directly
3406 * and reclear in up flow.
3408 netdev_warn(ring->tqp->handle->kinfo.netdev,
3409 "reserve buffer map failed, ret = %d\n",
3413 hns3_replace_buffer(ring, ring->next_to_use,
3416 ring_ptr_move_fw(ring, next_to_use);
3422 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
3424 while (ring->next_to_use != ring->next_to_clean) {
3425 /* When a buffer is not reused, it's memory has been
3426 * freed in hns3_handle_rx_bd or will be freed by
3427 * stack, so only need to unmap the buffer here.
3429 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
3430 hns3_unmap_buffer(ring,
3431 &ring->desc_cb[ring->next_to_use]);
3432 ring->desc_cb[ring->next_to_use].dma = 0;
3435 ring_ptr_move_fw(ring, next_to_use);
3439 static void hns3_force_clear_all_rx_ring(struct hnae3_handle *h)
3441 struct net_device *ndev = h->kinfo.netdev;
3442 struct hns3_nic_priv *priv = netdev_priv(ndev);
3443 struct hns3_enet_ring *ring;
3446 for (i = 0; i < h->kinfo.num_tqps; i++) {
3447 ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
3448 hns3_force_clear_rx_ring(ring);
3452 static void hns3_clear_all_ring(struct hnae3_handle *h)
3454 struct net_device *ndev = h->kinfo.netdev;
3455 struct hns3_nic_priv *priv = netdev_priv(ndev);
3458 for (i = 0; i < h->kinfo.num_tqps; i++) {
3459 struct netdev_queue *dev_queue;
3460 struct hns3_enet_ring *ring;
3462 ring = priv->ring_data[i].ring;
3463 hns3_clear_tx_ring(ring);
3464 dev_queue = netdev_get_tx_queue(ndev,
3465 priv->ring_data[i].queue_index);
3466 netdev_tx_reset_queue(dev_queue);
3468 ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
3469 /* Continue to clear other rings even if clearing some
3472 hns3_clear_rx_ring(ring);
3476 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
3478 struct net_device *ndev = h->kinfo.netdev;
3479 struct hns3_nic_priv *priv = netdev_priv(ndev);
3480 struct hns3_enet_ring *rx_ring;
3484 for (i = 0; i < h->kinfo.num_tqps; i++) {
3485 h->ae_algo->ops->reset_queue(h, i);
3486 hns3_init_ring_hw(priv->ring_data[i].ring);
3488 /* We need to clear tx ring here because self test will
3489 * use the ring and will not run down before up
3491 hns3_clear_tx_ring(priv->ring_data[i].ring);
3492 priv->ring_data[i].ring->next_to_clean = 0;
3493 priv->ring_data[i].ring->next_to_use = 0;
3495 rx_ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
3496 hns3_init_ring_hw(rx_ring);
3497 ret = hns3_clear_rx_ring(rx_ring);
3501 /* We can not know the hardware head and tail when this
3502 * function is called in reset flow, so we reuse all desc.
3504 for (j = 0; j < rx_ring->desc_num; j++)
3505 hns3_reuse_buffer(rx_ring, j);
3507 rx_ring->next_to_clean = 0;
3508 rx_ring->next_to_use = 0;
3511 hns3_init_tx_ring_tc(priv);
3516 static void hns3_store_coal(struct hns3_nic_priv *priv)
3518 /* ethtool only support setting and querying one coal
3519 * configuation for now, so save the vector 0' coal
3520 * configuation here in order to restore it.
3522 memcpy(&priv->tx_coal, &priv->tqp_vector[0].tx_group.coal,
3523 sizeof(struct hns3_enet_coalesce));
3524 memcpy(&priv->rx_coal, &priv->tqp_vector[0].rx_group.coal,
3525 sizeof(struct hns3_enet_coalesce));
3528 static void hns3_restore_coal(struct hns3_nic_priv *priv)
3530 u16 vector_num = priv->vector_num;
3533 for (i = 0; i < vector_num; i++) {
3534 memcpy(&priv->tqp_vector[i].tx_group.coal, &priv->tx_coal,
3535 sizeof(struct hns3_enet_coalesce));
3536 memcpy(&priv->tqp_vector[i].rx_group.coal, &priv->rx_coal,
3537 sizeof(struct hns3_enet_coalesce));
3541 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
3543 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
3544 struct net_device *ndev = kinfo->netdev;
3546 if (!netif_running(ndev))
3549 return hns3_nic_net_stop(ndev);
3552 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
3554 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
3557 if (netif_running(kinfo->netdev)) {
3558 ret = hns3_nic_net_up(kinfo->netdev);
3560 netdev_err(kinfo->netdev,
3561 "hns net up fail, ret=%d!\n", ret);
3564 handle->last_reset_time = jiffies;
3570 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
3572 struct net_device *netdev = handle->kinfo.netdev;
3573 struct hns3_nic_priv *priv = netdev_priv(netdev);
3576 hns3_init_mac_addr(netdev, false);
3577 hns3_nic_set_rx_mode(netdev);
3578 hns3_recover_hw_addr(netdev);
3580 /* Hardware table is only clear when pf resets */
3581 if (!(handle->flags & HNAE3_SUPPORT_VF))
3582 hns3_restore_vlan(netdev);
3584 hns3_restore_fd_rules(netdev);
3586 /* Carrier off reporting is important to ethtool even BEFORE open */
3587 netif_carrier_off(netdev);
3589 hns3_restore_coal(priv);
3591 ret = hns3_nic_init_vector_data(priv);
3595 ret = hns3_init_all_ring(priv);
3597 hns3_nic_uninit_vector_data(priv);
3598 priv->ring_data = NULL;
3604 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
3606 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(handle->pdev);
3607 struct net_device *netdev = handle->kinfo.netdev;
3608 struct hns3_nic_priv *priv = netdev_priv(netdev);
3611 hns3_force_clear_all_rx_ring(handle);
3613 ret = hns3_nic_uninit_vector_data(priv);
3615 netdev_err(netdev, "uninit vector error\n");
3619 hns3_store_coal(priv);
3621 ret = hns3_uninit_all_ring(priv);
3623 netdev_err(netdev, "uninit ring error\n");
3625 hns3_uninit_mac_addr(netdev);
3627 /* it is cumbersome for hardware to pick-and-choose rules for deletion
3628 * from TCAM. Hence, for function reset software intervention is
3629 * required to delete the rules
3631 if (hns3_dev_ongoing_func_reset(ae_dev))
3632 hns3_del_all_fd_rules(netdev, false);
3637 static int hns3_reset_notify(struct hnae3_handle *handle,
3638 enum hnae3_reset_notify_type type)
3643 case HNAE3_UP_CLIENT:
3644 ret = hns3_reset_notify_up_enet(handle);
3646 case HNAE3_DOWN_CLIENT:
3647 ret = hns3_reset_notify_down_enet(handle);
3649 case HNAE3_INIT_CLIENT:
3650 ret = hns3_reset_notify_init_enet(handle);
3652 case HNAE3_UNINIT_CLIENT:
3653 ret = hns3_reset_notify_uninit_enet(handle);
3662 static int hns3_modify_tqp_num(struct net_device *netdev, u16 new_tqp_num)
3664 struct hns3_nic_priv *priv = netdev_priv(netdev);
3665 struct hnae3_handle *h = hns3_get_handle(netdev);
3668 ret = h->ae_algo->ops->set_channels(h, new_tqp_num);
3672 ret = hns3_get_ring_config(priv);
3676 ret = hns3_nic_alloc_vector_data(priv);
3678 goto err_alloc_vector;
3680 hns3_restore_coal(priv);
3682 ret = hns3_nic_init_vector_data(priv);
3684 goto err_uninit_vector;
3686 ret = hns3_init_all_ring(priv);
3693 hns3_put_ring_config(priv);
3695 hns3_nic_uninit_vector_data(priv);
3697 hns3_nic_dealloc_vector_data(priv);
3701 static int hns3_adjust_tqps_num(u8 num_tc, u32 new_tqp_num)
3703 return (new_tqp_num / num_tc) * num_tc;
3706 int hns3_set_channels(struct net_device *netdev,
3707 struct ethtool_channels *ch)
3709 struct hns3_nic_priv *priv = netdev_priv(netdev);
3710 struct hnae3_handle *h = hns3_get_handle(netdev);
3711 struct hnae3_knic_private_info *kinfo = &h->kinfo;
3712 bool if_running = netif_running(netdev);
3713 u32 new_tqp_num = ch->combined_count;
3717 if (ch->rx_count || ch->tx_count)
3720 if (new_tqp_num > hns3_get_max_available_channels(h) ||
3721 new_tqp_num < kinfo->num_tc) {
3722 dev_err(&netdev->dev,
3723 "Change tqps fail, the tqp range is from %d to %d",
3725 hns3_get_max_available_channels(h));
3729 new_tqp_num = hns3_adjust_tqps_num(kinfo->num_tc, new_tqp_num);
3730 if (kinfo->num_tqps == new_tqp_num)
3734 hns3_nic_net_stop(netdev);
3736 ret = hns3_nic_uninit_vector_data(priv);
3738 dev_err(&netdev->dev,
3739 "Unbind vector with tqp fail, nothing is changed");
3743 hns3_store_coal(priv);
3745 hns3_nic_dealloc_vector_data(priv);
3747 hns3_uninit_all_ring(priv);
3748 hns3_put_ring_config(priv);
3750 org_tqp_num = h->kinfo.num_tqps;
3751 ret = hns3_modify_tqp_num(netdev, new_tqp_num);
3753 ret = hns3_modify_tqp_num(netdev, org_tqp_num);
3755 /* If revert to old tqp failed, fatal error occurred */
3756 dev_err(&netdev->dev,
3757 "Revert to old tqp num fail, ret=%d", ret);
3760 dev_info(&netdev->dev,
3761 "Change tqp num fail, Revert to old tqp num");
3766 hns3_nic_net_open(netdev);
3771 static const struct hnae3_client_ops client_ops = {
3772 .init_instance = hns3_client_init,
3773 .uninit_instance = hns3_client_uninit,
3774 .link_status_change = hns3_link_status_change,
3775 .setup_tc = hns3_client_setup_tc,
3776 .reset_notify = hns3_reset_notify,
3779 /* hns3_init_module - Driver registration routine
3780 * hns3_init_module is the first routine called when the driver is
3781 * loaded. All it does is register with the PCI subsystem.
3783 static int __init hns3_init_module(void)
3787 pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
3788 pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
3790 client.type = HNAE3_CLIENT_KNIC;
3791 snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH - 1, "%s",
3794 client.ops = &client_ops;
3796 INIT_LIST_HEAD(&client.node);
3798 ret = hnae3_register_client(&client);
3802 ret = pci_register_driver(&hns3_driver);
3804 hnae3_unregister_client(&client);
3808 module_init(hns3_init_module);
3810 /* hns3_exit_module - Driver exit cleanup routine
3811 * hns3_exit_module is called just before the driver is removed
3814 static void __exit hns3_exit_module(void)
3816 pci_unregister_driver(&hns3_driver);
3817 hnae3_unregister_client(&client);
3819 module_exit(hns3_exit_module);
3821 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
3822 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
3823 MODULE_LICENSE("GPL");
3824 MODULE_ALIAS("pci:hns-nic");
3825 MODULE_VERSION(HNS3_MOD_VERSION);