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>
11 #include <linux/irq.h>
13 #include <linux/ipv6.h>
14 #include <linux/module.h>
15 #include <linux/pci.h>
16 #include <linux/aer.h>
17 #include <linux/skbuff.h>
18 #include <linux/sctp.h>
20 #include <net/ip6_checksum.h>
21 #include <net/pkt_cls.h>
23 #include <net/vxlan.h>
24 #include <net/geneve.h>
27 #include "hns3_enet.h"
28 /* All hns3 tracepoints are defined by the include below, which
29 * must be included exactly once across the whole kernel with
30 * CREATE_TRACE_POINTS defined
32 #define CREATE_TRACE_POINTS
33 #include "hns3_trace.h"
35 #define hns3_set_field(origin, shift, val) ((origin) |= ((val) << (shift)))
36 #define hns3_tx_bd_count(S) DIV_ROUND_UP(S, HNS3_MAX_BD_SIZE)
38 #define hns3_rl_err(fmt, ...) \
40 if (net_ratelimit()) \
41 netdev_err(fmt, ##__VA_ARGS__); \
44 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force);
46 static const char hns3_driver_name[] = "hns3";
47 static const char hns3_driver_string[] =
48 "Hisilicon Ethernet Network Driver for Hip08 Family";
49 static const char hns3_copyright[] = "Copyright (c) 2017 Huawei Corporation.";
50 static struct hnae3_client client;
52 static int debug = -1;
53 module_param(debug, int, 0);
54 MODULE_PARM_DESC(debug, " Network interface message level setting");
56 #define DEFAULT_MSG_LEVEL (NETIF_MSG_PROBE | NETIF_MSG_LINK | \
57 NETIF_MSG_IFDOWN | NETIF_MSG_IFUP)
59 #define HNS3_INNER_VLAN_TAG 1
60 #define HNS3_OUTER_VLAN_TAG 2
62 #define HNS3_MIN_TX_LEN 33U
64 /* hns3_pci_tbl - PCI Device ID Table
66 * Last entry must be all 0s
68 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
69 * Class, Class Mask, private data (not used) }
71 static const struct pci_device_id hns3_pci_tbl[] = {
72 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
73 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
74 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA),
75 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
76 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC),
77 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
78 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA),
79 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
80 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC),
81 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
82 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC),
83 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
84 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_200G_RDMA),
85 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
86 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_VF), 0},
87 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_RDMA_DCB_PFC_VF),
88 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
89 /* required last entry */
92 MODULE_DEVICE_TABLE(pci, hns3_pci_tbl);
94 static irqreturn_t hns3_irq_handle(int irq, void *vector)
96 struct hns3_enet_tqp_vector *tqp_vector = vector;
98 napi_schedule_irqoff(&tqp_vector->napi);
103 static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv)
105 struct hns3_enet_tqp_vector *tqp_vectors;
108 for (i = 0; i < priv->vector_num; i++) {
109 tqp_vectors = &priv->tqp_vector[i];
111 if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED)
114 /* clear the affinity mask */
115 irq_set_affinity_hint(tqp_vectors->vector_irq, NULL);
117 /* release the irq resource */
118 free_irq(tqp_vectors->vector_irq, tqp_vectors);
119 tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED;
123 static int hns3_nic_init_irq(struct hns3_nic_priv *priv)
125 struct hns3_enet_tqp_vector *tqp_vectors;
126 int txrx_int_idx = 0;
132 for (i = 0; i < priv->vector_num; i++) {
133 tqp_vectors = &priv->tqp_vector[i];
135 if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED)
138 if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) {
139 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
140 "%s-%s-%s-%d", hns3_driver_name,
141 pci_name(priv->ae_handle->pdev),
142 "TxRx", txrx_int_idx++);
144 } else if (tqp_vectors->rx_group.ring) {
145 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
146 "%s-%s-%s-%d", hns3_driver_name,
147 pci_name(priv->ae_handle->pdev),
149 } else if (tqp_vectors->tx_group.ring) {
150 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
151 "%s-%s-%s-%d", hns3_driver_name,
152 pci_name(priv->ae_handle->pdev),
155 /* Skip this unused q_vector */
159 tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0';
161 irq_set_status_flags(tqp_vectors->vector_irq, IRQ_NOAUTOEN);
162 ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0,
163 tqp_vectors->name, tqp_vectors);
165 netdev_err(priv->netdev, "request irq(%d) fail\n",
166 tqp_vectors->vector_irq);
167 hns3_nic_uninit_irq(priv);
171 irq_set_affinity_hint(tqp_vectors->vector_irq,
172 &tqp_vectors->affinity_mask);
174 tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED;
180 static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector,
183 writel(mask_en, tqp_vector->mask_addr);
186 static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector)
188 napi_enable(&tqp_vector->napi);
189 enable_irq(tqp_vector->vector_irq);
192 hns3_mask_vector_irq(tqp_vector, 1);
195 static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector)
198 hns3_mask_vector_irq(tqp_vector, 0);
200 disable_irq(tqp_vector->vector_irq);
201 napi_disable(&tqp_vector->napi);
204 void hns3_set_vector_coalesce_rl(struct hns3_enet_tqp_vector *tqp_vector,
207 u32 rl_reg = hns3_rl_usec_to_reg(rl_value);
209 /* this defines the configuration for RL (Interrupt Rate Limiter).
210 * Rl defines rate of interrupts i.e. number of interrupts-per-second
211 * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing
214 if (rl_reg > 0 && !tqp_vector->tx_group.coal.adapt_enable &&
215 !tqp_vector->rx_group.coal.adapt_enable)
216 /* According to the hardware, the range of rl_reg is
217 * 0-59 and the unit is 4.
219 rl_reg |= HNS3_INT_RL_ENABLE_MASK;
221 writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
224 void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
229 if (tqp_vector->rx_group.coal.unit_1us)
230 new_val = gl_value | HNS3_INT_GL_1US;
232 new_val = hns3_gl_usec_to_reg(gl_value);
234 writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
237 void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
242 if (tqp_vector->tx_group.coal.unit_1us)
243 new_val = gl_value | HNS3_INT_GL_1US;
245 new_val = hns3_gl_usec_to_reg(gl_value);
247 writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
250 void hns3_set_vector_coalesce_tx_ql(struct hns3_enet_tqp_vector *tqp_vector,
253 writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_TX_QL_OFFSET);
256 void hns3_set_vector_coalesce_rx_ql(struct hns3_enet_tqp_vector *tqp_vector,
259 writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_RX_QL_OFFSET);
262 static void hns3_vector_coalesce_init(struct hns3_enet_tqp_vector *tqp_vector,
263 struct hns3_nic_priv *priv)
265 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
266 struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
267 struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
269 /* initialize the configuration for interrupt coalescing.
270 * 1. GL (Interrupt Gap Limiter)
271 * 2. RL (Interrupt Rate Limiter)
272 * 3. QL (Interrupt Quantity Limiter)
274 * Default: enable interrupt coalescing self-adaptive and GL
276 tx_coal->adapt_enable = 1;
277 rx_coal->adapt_enable = 1;
279 tx_coal->int_gl = HNS3_INT_GL_50K;
280 rx_coal->int_gl = HNS3_INT_GL_50K;
282 rx_coal->flow_level = HNS3_FLOW_LOW;
283 tx_coal->flow_level = HNS3_FLOW_LOW;
285 /* device version above V3(include V3), GL can configure 1us
286 * unit, so uses 1us unit.
288 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3) {
289 tx_coal->unit_1us = 1;
290 rx_coal->unit_1us = 1;
293 if (ae_dev->dev_specs.int_ql_max) {
294 tx_coal->ql_enable = 1;
295 rx_coal->ql_enable = 1;
296 tx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
297 rx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
298 tx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
299 rx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
304 hns3_vector_coalesce_init_hw(struct hns3_enet_tqp_vector *tqp_vector,
305 struct hns3_nic_priv *priv)
307 struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
308 struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
309 struct hnae3_handle *h = priv->ae_handle;
311 hns3_set_vector_coalesce_tx_gl(tqp_vector, tx_coal->int_gl);
312 hns3_set_vector_coalesce_rx_gl(tqp_vector, rx_coal->int_gl);
313 hns3_set_vector_coalesce_rl(tqp_vector, h->kinfo.int_rl_setting);
315 if (tx_coal->ql_enable)
316 hns3_set_vector_coalesce_tx_ql(tqp_vector, tx_coal->int_ql);
318 if (rx_coal->ql_enable)
319 hns3_set_vector_coalesce_rx_ql(tqp_vector, rx_coal->int_ql);
322 static int hns3_nic_set_real_num_queue(struct net_device *netdev)
324 struct hnae3_handle *h = hns3_get_handle(netdev);
325 struct hnae3_knic_private_info *kinfo = &h->kinfo;
326 unsigned int queue_size = kinfo->rss_size * kinfo->num_tc;
329 if (kinfo->num_tc <= 1) {
330 netdev_reset_tc(netdev);
332 ret = netdev_set_num_tc(netdev, kinfo->num_tc);
335 "netdev_set_num_tc fail, ret=%d!\n", ret);
339 for (i = 0; i < HNAE3_MAX_TC; i++) {
340 if (!kinfo->tc_info[i].enable)
343 netdev_set_tc_queue(netdev,
344 kinfo->tc_info[i].tc,
345 kinfo->tc_info[i].tqp_count,
346 kinfo->tc_info[i].tqp_offset);
350 ret = netif_set_real_num_tx_queues(netdev, queue_size);
353 "netif_set_real_num_tx_queues fail, ret=%d!\n", ret);
357 ret = netif_set_real_num_rx_queues(netdev, queue_size);
360 "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
367 static u16 hns3_get_max_available_channels(struct hnae3_handle *h)
369 u16 alloc_tqps, max_rss_size, rss_size;
371 h->ae_algo->ops->get_tqps_and_rss_info(h, &alloc_tqps, &max_rss_size);
372 rss_size = alloc_tqps / h->kinfo.num_tc;
374 return min_t(u16, rss_size, max_rss_size);
377 static void hns3_tqp_enable(struct hnae3_queue *tqp)
381 rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
382 rcb_reg |= BIT(HNS3_RING_EN_B);
383 hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
386 static void hns3_tqp_disable(struct hnae3_queue *tqp)
390 rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
391 rcb_reg &= ~BIT(HNS3_RING_EN_B);
392 hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
395 static void hns3_free_rx_cpu_rmap(struct net_device *netdev)
397 #ifdef CONFIG_RFS_ACCEL
398 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
399 netdev->rx_cpu_rmap = NULL;
403 static int hns3_set_rx_cpu_rmap(struct net_device *netdev)
405 #ifdef CONFIG_RFS_ACCEL
406 struct hns3_nic_priv *priv = netdev_priv(netdev);
407 struct hns3_enet_tqp_vector *tqp_vector;
410 if (!netdev->rx_cpu_rmap) {
411 netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->vector_num);
412 if (!netdev->rx_cpu_rmap)
416 for (i = 0; i < priv->vector_num; i++) {
417 tqp_vector = &priv->tqp_vector[i];
418 ret = irq_cpu_rmap_add(netdev->rx_cpu_rmap,
419 tqp_vector->vector_irq);
421 hns3_free_rx_cpu_rmap(netdev);
429 static int hns3_nic_net_up(struct net_device *netdev)
431 struct hns3_nic_priv *priv = netdev_priv(netdev);
432 struct hnae3_handle *h = priv->ae_handle;
436 ret = hns3_nic_reset_all_ring(h);
440 clear_bit(HNS3_NIC_STATE_DOWN, &priv->state);
442 /* enable the vectors */
443 for (i = 0; i < priv->vector_num; i++)
444 hns3_vector_enable(&priv->tqp_vector[i]);
447 for (j = 0; j < h->kinfo.num_tqps; j++)
448 hns3_tqp_enable(h->kinfo.tqp[j]);
450 /* start the ae_dev */
451 ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0;
453 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
455 hns3_tqp_disable(h->kinfo.tqp[j]);
457 for (j = i - 1; j >= 0; j--)
458 hns3_vector_disable(&priv->tqp_vector[j]);
464 static void hns3_config_xps(struct hns3_nic_priv *priv)
468 for (i = 0; i < priv->vector_num; i++) {
469 struct hns3_enet_tqp_vector *tqp_vector = &priv->tqp_vector[i];
470 struct hns3_enet_ring *ring = tqp_vector->tx_group.ring;
475 ret = netif_set_xps_queue(priv->netdev,
476 &tqp_vector->affinity_mask,
477 ring->tqp->tqp_index);
479 netdev_warn(priv->netdev,
480 "set xps queue failed: %d", ret);
487 static int hns3_nic_net_open(struct net_device *netdev)
489 struct hns3_nic_priv *priv = netdev_priv(netdev);
490 struct hnae3_handle *h = hns3_get_handle(netdev);
491 struct hnae3_knic_private_info *kinfo;
494 if (hns3_nic_resetting(netdev))
497 netif_carrier_off(netdev);
499 ret = hns3_nic_set_real_num_queue(netdev);
503 ret = hns3_nic_net_up(netdev);
505 netdev_err(netdev, "net up fail, ret=%d!\n", ret);
510 for (i = 0; i < HNAE3_MAX_USER_PRIO; i++)
511 netdev_set_prio_tc_map(netdev, i, kinfo->prio_tc[i]);
513 if (h->ae_algo->ops->set_timer_task)
514 h->ae_algo->ops->set_timer_task(priv->ae_handle, true);
516 hns3_config_xps(priv);
518 netif_dbg(h, drv, netdev, "net open\n");
523 static void hns3_reset_tx_queue(struct hnae3_handle *h)
525 struct net_device *ndev = h->kinfo.netdev;
526 struct hns3_nic_priv *priv = netdev_priv(ndev);
527 struct netdev_queue *dev_queue;
530 for (i = 0; i < h->kinfo.num_tqps; i++) {
531 dev_queue = netdev_get_tx_queue(ndev,
532 priv->ring[i].queue_index);
533 netdev_tx_reset_queue(dev_queue);
537 static void hns3_nic_net_down(struct net_device *netdev)
539 struct hns3_nic_priv *priv = netdev_priv(netdev);
540 struct hnae3_handle *h = hns3_get_handle(netdev);
541 const struct hnae3_ae_ops *ops;
544 /* disable vectors */
545 for (i = 0; i < priv->vector_num; i++)
546 hns3_vector_disable(&priv->tqp_vector[i]);
549 for (i = 0; i < h->kinfo.num_tqps; i++)
550 hns3_tqp_disable(h->kinfo.tqp[i]);
553 ops = priv->ae_handle->ae_algo->ops;
555 ops->stop(priv->ae_handle);
557 /* delay ring buffer clearing to hns3_reset_notify_uninit_enet
558 * during reset process, because driver may not be able
559 * to disable the ring through firmware when downing the netdev.
561 if (!hns3_nic_resetting(netdev))
562 hns3_clear_all_ring(priv->ae_handle, false);
564 hns3_reset_tx_queue(priv->ae_handle);
567 static int hns3_nic_net_stop(struct net_device *netdev)
569 struct hns3_nic_priv *priv = netdev_priv(netdev);
570 struct hnae3_handle *h = hns3_get_handle(netdev);
572 if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
575 netif_dbg(h, drv, netdev, "net stop\n");
577 if (h->ae_algo->ops->set_timer_task)
578 h->ae_algo->ops->set_timer_task(priv->ae_handle, false);
580 netif_tx_stop_all_queues(netdev);
581 netif_carrier_off(netdev);
583 hns3_nic_net_down(netdev);
588 static int hns3_nic_uc_sync(struct net_device *netdev,
589 const unsigned char *addr)
591 struct hnae3_handle *h = hns3_get_handle(netdev);
593 if (h->ae_algo->ops->add_uc_addr)
594 return h->ae_algo->ops->add_uc_addr(h, addr);
599 static int hns3_nic_uc_unsync(struct net_device *netdev,
600 const unsigned char *addr)
602 struct hnae3_handle *h = hns3_get_handle(netdev);
604 /* need ignore the request of removing device address, because
605 * we store the device address and other addresses of uc list
606 * in the function's mac filter list.
608 if (ether_addr_equal(addr, netdev->dev_addr))
611 if (h->ae_algo->ops->rm_uc_addr)
612 return h->ae_algo->ops->rm_uc_addr(h, addr);
617 static int hns3_nic_mc_sync(struct net_device *netdev,
618 const unsigned char *addr)
620 struct hnae3_handle *h = hns3_get_handle(netdev);
622 if (h->ae_algo->ops->add_mc_addr)
623 return h->ae_algo->ops->add_mc_addr(h, addr);
628 static int hns3_nic_mc_unsync(struct net_device *netdev,
629 const unsigned char *addr)
631 struct hnae3_handle *h = hns3_get_handle(netdev);
633 if (h->ae_algo->ops->rm_mc_addr)
634 return h->ae_algo->ops->rm_mc_addr(h, addr);
639 static u8 hns3_get_netdev_flags(struct net_device *netdev)
643 if (netdev->flags & IFF_PROMISC) {
644 flags = HNAE3_USER_UPE | HNAE3_USER_MPE | HNAE3_BPE;
646 flags |= HNAE3_VLAN_FLTR;
647 if (netdev->flags & IFF_ALLMULTI)
648 flags |= HNAE3_USER_MPE;
654 static void hns3_nic_set_rx_mode(struct net_device *netdev)
656 struct hnae3_handle *h = hns3_get_handle(netdev);
659 new_flags = hns3_get_netdev_flags(netdev);
661 __dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync);
662 __dev_mc_sync(netdev, hns3_nic_mc_sync, hns3_nic_mc_unsync);
664 /* User mode Promisc mode enable and vlan filtering is disabled to
665 * let all packets in.
667 h->netdev_flags = new_flags;
668 hns3_request_update_promisc_mode(h);
671 void hns3_request_update_promisc_mode(struct hnae3_handle *handle)
673 const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
675 if (ops->request_update_promisc_mode)
676 ops->request_update_promisc_mode(handle);
679 void hns3_enable_vlan_filter(struct net_device *netdev, bool enable)
681 struct hns3_nic_priv *priv = netdev_priv(netdev);
682 struct hnae3_handle *h = priv->ae_handle;
683 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(h->pdev);
686 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V2 &&
687 h->ae_algo->ops->enable_vlan_filter) {
688 last_state = h->netdev_flags & HNAE3_VLAN_FLTR ? true : false;
689 if (enable != last_state) {
692 enable ? "enable" : "disable");
693 h->ae_algo->ops->enable_vlan_filter(h, enable);
698 static int hns3_set_tso(struct sk_buff *skb, u32 *paylen_fdop_ol4cs,
699 u16 *mss, u32 *type_cs_vlan_tso)
701 u32 l4_offset, hdr_len;
702 union l3_hdr_info l3;
703 union l4_hdr_info l4;
707 if (!skb_is_gso(skb))
710 ret = skb_cow_head(skb, 0);
711 if (unlikely(ret < 0))
714 l3.hdr = skb_network_header(skb);
715 l4.hdr = skb_transport_header(skb);
717 /* Software should clear the IPv4's checksum field when tso is
720 if (l3.v4->version == 4)
724 if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
727 SKB_GSO_UDP_TUNNEL_CSUM)) {
728 /* reset l3&l4 pointers from outer to inner headers */
729 l3.hdr = skb_inner_network_header(skb);
730 l4.hdr = skb_inner_transport_header(skb);
732 /* Software should clear the IPv4's checksum field when
735 if (l3.v4->version == 4)
739 /* normal or tunnel packet */
740 l4_offset = l4.hdr - skb->data;
742 /* remove payload length from inner pseudo checksum when tso */
743 l4_paylen = skb->len - l4_offset;
745 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) {
746 hdr_len = sizeof(*l4.udp) + l4_offset;
747 csum_replace_by_diff(&l4.udp->check,
748 (__force __wsum)htonl(l4_paylen));
750 hdr_len = (l4.tcp->doff << 2) + l4_offset;
751 csum_replace_by_diff(&l4.tcp->check,
752 (__force __wsum)htonl(l4_paylen));
755 /* find the txbd field values */
756 *paylen_fdop_ol4cs = skb->len - hdr_len;
757 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_TSO_B, 1);
759 /* offload outer UDP header checksum */
760 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)
761 hns3_set_field(*paylen_fdop_ol4cs, HNS3_TXD_OL4CS_B, 1);
763 /* get MSS for TSO */
764 *mss = skb_shinfo(skb)->gso_size;
771 static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
774 union l3_hdr_info l3;
775 unsigned char *l4_hdr;
776 unsigned char *exthdr;
780 /* find outer header point */
781 l3.hdr = skb_network_header(skb);
782 l4_hdr = skb_transport_header(skb);
784 if (skb->protocol == htons(ETH_P_IPV6)) {
785 exthdr = l3.hdr + sizeof(*l3.v6);
786 l4_proto_tmp = l3.v6->nexthdr;
787 if (l4_hdr != exthdr)
788 ipv6_skip_exthdr(skb, exthdr - skb->data,
789 &l4_proto_tmp, &frag_off);
790 } else if (skb->protocol == htons(ETH_P_IP)) {
791 l4_proto_tmp = l3.v4->protocol;
796 *ol4_proto = l4_proto_tmp;
799 if (!skb->encapsulation) {
804 /* find inner header point */
805 l3.hdr = skb_inner_network_header(skb);
806 l4_hdr = skb_inner_transport_header(skb);
808 if (l3.v6->version == 6) {
809 exthdr = l3.hdr + sizeof(*l3.v6);
810 l4_proto_tmp = l3.v6->nexthdr;
811 if (l4_hdr != exthdr)
812 ipv6_skip_exthdr(skb, exthdr - skb->data,
813 &l4_proto_tmp, &frag_off);
814 } else if (l3.v4->version == 4) {
815 l4_proto_tmp = l3.v4->protocol;
818 *il4_proto = l4_proto_tmp;
823 /* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
824 * and it is udp packet, which has a dest port as the IANA assigned.
825 * the hardware is expected to do the checksum offload, but the
826 * hardware will not do the checksum offload when udp dest port is
829 static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
831 struct hns3_nic_priv *priv = netdev_priv(skb->dev);
832 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
833 union l4_hdr_info l4;
835 /* device version above V3(include V3), the hardware can
836 * do this checksum offload.
838 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
841 l4.hdr = skb_transport_header(skb);
843 if (!(!skb->encapsulation &&
844 (l4.udp->dest == htons(IANA_VXLAN_UDP_PORT) ||
845 l4.udp->dest == htons(GENEVE_UDP_PORT))))
848 skb_checksum_help(skb);
853 static void hns3_set_outer_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
854 u32 *ol_type_vlan_len_msec)
856 u32 l2_len, l3_len, l4_len;
857 unsigned char *il2_hdr;
858 union l3_hdr_info l3;
859 union l4_hdr_info l4;
861 l3.hdr = skb_network_header(skb);
862 l4.hdr = skb_transport_header(skb);
864 /* compute OL2 header size, defined in 2 Bytes */
865 l2_len = l3.hdr - skb->data;
866 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L2LEN_S, l2_len >> 1);
868 /* compute OL3 header size, defined in 4 Bytes */
869 l3_len = l4.hdr - l3.hdr;
870 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_S, l3_len >> 2);
872 il2_hdr = skb_inner_mac_header(skb);
873 /* compute OL4 header size, defined in 4 Bytes */
874 l4_len = il2_hdr - l4.hdr;
875 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L4LEN_S, l4_len >> 2);
877 /* define outer network header type */
878 if (skb->protocol == htons(ETH_P_IP)) {
880 hns3_set_field(*ol_type_vlan_len_msec,
882 HNS3_OL3T_IPV4_CSUM);
884 hns3_set_field(*ol_type_vlan_len_msec,
886 HNS3_OL3T_IPV4_NO_CSUM);
888 } else if (skb->protocol == htons(ETH_P_IPV6)) {
889 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_S,
893 if (ol4_proto == IPPROTO_UDP)
894 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
895 HNS3_TUN_MAC_IN_UDP);
896 else if (ol4_proto == IPPROTO_GRE)
897 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
901 static int hns3_set_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
902 u8 il4_proto, u32 *type_cs_vlan_tso,
903 u32 *ol_type_vlan_len_msec)
905 unsigned char *l2_hdr = skb->data;
906 u32 l4_proto = ol4_proto;
907 union l4_hdr_info l4;
908 union l3_hdr_info l3;
911 l4.hdr = skb_transport_header(skb);
912 l3.hdr = skb_network_header(skb);
914 /* handle encapsulation skb */
915 if (skb->encapsulation) {
916 /* If this is a not UDP/GRE encapsulation skb */
917 if (!(ol4_proto == IPPROTO_UDP || ol4_proto == IPPROTO_GRE)) {
918 /* drop the skb tunnel packet if hardware don't support,
919 * because hardware can't calculate csum when TSO.
924 /* the stack computes the IP header already,
925 * driver calculate l4 checksum when not TSO.
927 skb_checksum_help(skb);
931 hns3_set_outer_l2l3l4(skb, ol4_proto, ol_type_vlan_len_msec);
933 /* switch to inner header */
934 l2_hdr = skb_inner_mac_header(skb);
935 l3.hdr = skb_inner_network_header(skb);
936 l4.hdr = skb_inner_transport_header(skb);
937 l4_proto = il4_proto;
940 if (l3.v4->version == 4) {
941 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
944 /* the stack computes the IP header already, the only time we
945 * need the hardware to recompute it is in the case of TSO.
948 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3CS_B, 1);
949 } else if (l3.v6->version == 6) {
950 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
954 /* compute inner(/normal) L2 header size, defined in 2 Bytes */
955 l2_len = l3.hdr - l2_hdr;
956 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S, l2_len >> 1);
958 /* compute inner(/normal) L3 header size, defined in 4 Bytes */
959 l3_len = l4.hdr - l3.hdr;
960 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_S, l3_len >> 2);
962 /* compute inner(/normal) L4 header size, defined in 4 Bytes */
965 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
966 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
968 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
972 if (hns3_tunnel_csum_bug(skb))
975 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
976 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
978 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
979 (sizeof(struct udphdr) >> 2));
982 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
983 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
985 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
986 (sizeof(struct sctphdr) >> 2));
989 /* drop the skb tunnel packet if hardware don't support,
990 * because hardware can't calculate csum when TSO.
995 /* the stack computes the IP header already,
996 * driver calculate l4 checksum when not TSO.
998 skb_checksum_help(skb);
1005 static int hns3_handle_vtags(struct hns3_enet_ring *tx_ring,
1006 struct sk_buff *skb)
1008 struct hnae3_handle *handle = tx_ring->tqp->handle;
1009 struct hnae3_ae_dev *ae_dev;
1010 struct vlan_ethhdr *vhdr;
1013 if (!(skb->protocol == htons(ETH_P_8021Q) ||
1014 skb_vlan_tag_present(skb)))
1017 /* For HW limitation on HNAE3_DEVICE_VERSION_V2, if port based insert
1018 * VLAN enabled, only one VLAN header is allowed in skb, otherwise it
1019 * will cause RAS error.
1021 ae_dev = pci_get_drvdata(handle->pdev);
1022 if (unlikely(skb_vlan_tagged_multi(skb) &&
1023 ae_dev->dev_version <= HNAE3_DEVICE_VERSION_V2 &&
1024 handle->port_base_vlan_state ==
1025 HNAE3_PORT_BASE_VLAN_ENABLE))
1028 if (skb->protocol == htons(ETH_P_8021Q) &&
1029 !(handle->kinfo.netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
1030 /* When HW VLAN acceleration is turned off, and the stack
1031 * sets the protocol to 802.1q, the driver just need to
1032 * set the protocol to the encapsulated ethertype.
1034 skb->protocol = vlan_get_protocol(skb);
1038 if (skb_vlan_tag_present(skb)) {
1039 /* Based on hw strategy, use out_vtag in two layer tag case,
1040 * and use inner_vtag in one tag case.
1042 if (skb->protocol == htons(ETH_P_8021Q) &&
1043 handle->port_base_vlan_state ==
1044 HNAE3_PORT_BASE_VLAN_DISABLE)
1045 rc = HNS3_OUTER_VLAN_TAG;
1047 rc = HNS3_INNER_VLAN_TAG;
1049 skb->protocol = vlan_get_protocol(skb);
1053 rc = skb_cow_head(skb, 0);
1054 if (unlikely(rc < 0))
1057 vhdr = (struct vlan_ethhdr *)skb->data;
1058 vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority << VLAN_PRIO_SHIFT)
1061 skb->protocol = vlan_get_protocol(skb);
1065 /* check if the hardware is capable of checksum offloading */
1066 static bool hns3_check_hw_tx_csum(struct sk_buff *skb)
1068 struct hns3_nic_priv *priv = netdev_priv(skb->dev);
1070 /* Kindly note, due to backward compatibility of the TX descriptor,
1071 * HW checksum of the non-IP packets and GSO packets is handled at
1072 * different place in the following code
1074 if (skb->csum_not_inet || skb_is_gso(skb) ||
1075 !test_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state))
1081 static int hns3_fill_skb_desc(struct hns3_enet_ring *ring,
1082 struct sk_buff *skb, struct hns3_desc *desc)
1084 u32 ol_type_vlan_len_msec = 0;
1085 u32 paylen_ol4cs = skb->len;
1086 u32 type_cs_vlan_tso = 0;
1087 u16 mss_hw_csum = 0;
1092 ret = hns3_handle_vtags(ring, skb);
1093 if (unlikely(ret < 0)) {
1094 u64_stats_update_begin(&ring->syncp);
1095 ring->stats.tx_vlan_err++;
1096 u64_stats_update_end(&ring->syncp);
1098 } else if (ret == HNS3_INNER_VLAN_TAG) {
1099 inner_vtag = skb_vlan_tag_get(skb);
1100 inner_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1102 hns3_set_field(type_cs_vlan_tso, HNS3_TXD_VLAN_B, 1);
1103 } else if (ret == HNS3_OUTER_VLAN_TAG) {
1104 out_vtag = skb_vlan_tag_get(skb);
1105 out_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1107 hns3_set_field(ol_type_vlan_len_msec, HNS3_TXD_OVLAN_B,
1111 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1112 u8 ol4_proto, il4_proto;
1114 if (hns3_check_hw_tx_csum(skb)) {
1115 /* set checksum start and offset, defined in 2 Bytes */
1116 hns3_set_field(type_cs_vlan_tso, HNS3_TXD_CSUM_START_S,
1117 skb_checksum_start_offset(skb) >> 1);
1118 hns3_set_field(ol_type_vlan_len_msec,
1119 HNS3_TXD_CSUM_OFFSET_S,
1120 skb->csum_offset >> 1);
1121 mss_hw_csum |= BIT(HNS3_TXD_HW_CS_B);
1122 goto out_hw_tx_csum;
1125 skb_reset_mac_len(skb);
1127 ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
1128 if (unlikely(ret < 0)) {
1129 u64_stats_update_begin(&ring->syncp);
1130 ring->stats.tx_l4_proto_err++;
1131 u64_stats_update_end(&ring->syncp);
1135 ret = hns3_set_l2l3l4(skb, ol4_proto, il4_proto,
1137 &ol_type_vlan_len_msec);
1138 if (unlikely(ret < 0)) {
1139 u64_stats_update_begin(&ring->syncp);
1140 ring->stats.tx_l2l3l4_err++;
1141 u64_stats_update_end(&ring->syncp);
1145 ret = hns3_set_tso(skb, &paylen_ol4cs, &mss_hw_csum,
1147 if (unlikely(ret < 0)) {
1148 u64_stats_update_begin(&ring->syncp);
1149 ring->stats.tx_tso_err++;
1150 u64_stats_update_end(&ring->syncp);
1157 desc->tx.ol_type_vlan_len_msec =
1158 cpu_to_le32(ol_type_vlan_len_msec);
1159 desc->tx.type_cs_vlan_tso_len = cpu_to_le32(type_cs_vlan_tso);
1160 desc->tx.paylen_ol4cs = cpu_to_le32(paylen_ol4cs);
1161 desc->tx.mss_hw_csum = cpu_to_le16(mss_hw_csum);
1162 desc->tx.vlan_tag = cpu_to_le16(inner_vtag);
1163 desc->tx.outer_vlan_tag = cpu_to_le16(out_vtag);
1168 static int hns3_fill_desc(struct hns3_enet_ring *ring, void *priv,
1169 unsigned int size, enum hns_desc_type type)
1171 #define HNS3_LIKELY_BD_NUM 1
1173 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1174 struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1175 struct device *dev = ring_to_dev(ring);
1177 unsigned int frag_buf_num;
1181 if (type == DESC_TYPE_FRAGLIST_SKB ||
1182 type == DESC_TYPE_SKB) {
1183 struct sk_buff *skb = (struct sk_buff *)priv;
1185 dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1187 frag = (skb_frag_t *)priv;
1188 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1191 if (unlikely(dma_mapping_error(dev, dma))) {
1192 u64_stats_update_begin(&ring->syncp);
1193 ring->stats.sw_err_cnt++;
1194 u64_stats_update_end(&ring->syncp);
1198 desc_cb->priv = priv;
1199 desc_cb->length = size;
1201 desc_cb->type = type;
1203 if (likely(size <= HNS3_MAX_BD_SIZE)) {
1204 desc->addr = cpu_to_le64(dma);
1205 desc->tx.send_size = cpu_to_le16(size);
1206 desc->tx.bdtp_fe_sc_vld_ra_ri =
1207 cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1209 trace_hns3_tx_desc(ring, ring->next_to_use);
1210 ring_ptr_move_fw(ring, next_to_use);
1211 return HNS3_LIKELY_BD_NUM;
1214 frag_buf_num = hns3_tx_bd_count(size);
1215 sizeoflast = size % HNS3_MAX_BD_SIZE;
1216 sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1218 /* When frag size is bigger than hardware limit, split this frag */
1219 for (k = 0; k < frag_buf_num; k++) {
1220 /* now, fill the descriptor */
1221 desc->addr = cpu_to_le64(dma + HNS3_MAX_BD_SIZE * k);
1222 desc->tx.send_size = cpu_to_le16((k == frag_buf_num - 1) ?
1223 (u16)sizeoflast : (u16)HNS3_MAX_BD_SIZE);
1224 desc->tx.bdtp_fe_sc_vld_ra_ri =
1225 cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1227 trace_hns3_tx_desc(ring, ring->next_to_use);
1228 /* move ring pointer to next */
1229 ring_ptr_move_fw(ring, next_to_use);
1231 desc = &ring->desc[ring->next_to_use];
1234 return frag_buf_num;
1237 static unsigned int hns3_skb_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1238 unsigned int bd_num)
1243 size = skb_headlen(skb);
1244 while (size > HNS3_MAX_BD_SIZE) {
1245 bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1246 size -= HNS3_MAX_BD_SIZE;
1248 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1253 bd_size[bd_num++] = size;
1254 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1258 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1259 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1260 size = skb_frag_size(frag);
1264 while (size > HNS3_MAX_BD_SIZE) {
1265 bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1266 size -= HNS3_MAX_BD_SIZE;
1268 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1272 bd_size[bd_num++] = size;
1273 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1280 static unsigned int hns3_tx_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1281 u8 max_non_tso_bd_num)
1283 struct sk_buff *frag_skb;
1284 unsigned int bd_num = 0;
1286 /* If the total len is within the max bd limit */
1287 if (likely(skb->len <= HNS3_MAX_BD_SIZE && !skb_has_frag_list(skb) &&
1288 skb_shinfo(skb)->nr_frags < max_non_tso_bd_num))
1289 return skb_shinfo(skb)->nr_frags + 1U;
1291 /* The below case will always be linearized, return
1292 * HNS3_MAX_BD_NUM_TSO + 1U to make sure it is linearized.
1294 if (unlikely(skb->len > HNS3_MAX_TSO_SIZE ||
1295 (!skb_is_gso(skb) && skb->len >
1296 HNS3_MAX_NON_TSO_SIZE(max_non_tso_bd_num))))
1297 return HNS3_MAX_TSO_BD_NUM + 1U;
1299 bd_num = hns3_skb_bd_num(skb, bd_size, bd_num);
1301 if (!skb_has_frag_list(skb) || bd_num > HNS3_MAX_TSO_BD_NUM)
1304 skb_walk_frags(skb, frag_skb) {
1305 bd_num = hns3_skb_bd_num(frag_skb, bd_size, bd_num);
1306 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1313 static unsigned int hns3_gso_hdr_len(struct sk_buff *skb)
1315 if (!skb->encapsulation)
1316 return skb_transport_offset(skb) + tcp_hdrlen(skb);
1318 return skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb);
1321 /* HW need every continuous max_non_tso_bd_num buffer data to be larger
1322 * than MSS, we simplify it by ensuring skb_headlen + the first continuous
1323 * max_non_tso_bd_num - 1 frags to be larger than gso header len + mss,
1324 * and the remaining continuous max_non_tso_bd_num - 1 frags to be larger
1325 * than MSS except the last max_non_tso_bd_num - 1 frags.
1327 static bool hns3_skb_need_linearized(struct sk_buff *skb, unsigned int *bd_size,
1328 unsigned int bd_num, u8 max_non_tso_bd_num)
1330 unsigned int tot_len = 0;
1333 for (i = 0; i < max_non_tso_bd_num - 1U; i++)
1334 tot_len += bd_size[i];
1336 /* ensure the first max_non_tso_bd_num frags is greater than
1339 if (tot_len + bd_size[max_non_tso_bd_num - 1U] <
1340 skb_shinfo(skb)->gso_size + hns3_gso_hdr_len(skb))
1343 /* ensure every continuous max_non_tso_bd_num - 1 buffer is greater
1344 * than mss except the last one.
1346 for (i = 0; i < bd_num - max_non_tso_bd_num; i++) {
1347 tot_len -= bd_size[i];
1348 tot_len += bd_size[i + max_non_tso_bd_num - 1U];
1350 if (tot_len < skb_shinfo(skb)->gso_size)
1357 void hns3_shinfo_pack(struct skb_shared_info *shinfo, __u32 *size)
1361 for (i = 0; i < MAX_SKB_FRAGS; i++)
1362 size[i] = skb_frag_size(&shinfo->frags[i]);
1365 static int hns3_nic_maybe_stop_tx(struct hns3_enet_ring *ring,
1366 struct net_device *netdev,
1367 struct sk_buff *skb)
1369 struct hns3_nic_priv *priv = netdev_priv(netdev);
1370 u8 max_non_tso_bd_num = priv->max_non_tso_bd_num;
1371 unsigned int bd_size[HNS3_MAX_TSO_BD_NUM + 1U];
1372 unsigned int bd_num;
1374 bd_num = hns3_tx_bd_num(skb, bd_size, max_non_tso_bd_num);
1375 if (unlikely(bd_num > max_non_tso_bd_num)) {
1376 if (bd_num <= HNS3_MAX_TSO_BD_NUM && skb_is_gso(skb) &&
1377 !hns3_skb_need_linearized(skb, bd_size, bd_num,
1378 max_non_tso_bd_num)) {
1379 trace_hns3_over_max_bd(skb);
1383 if (__skb_linearize(skb))
1386 bd_num = hns3_tx_bd_count(skb->len);
1387 if ((skb_is_gso(skb) && bd_num > HNS3_MAX_TSO_BD_NUM) ||
1388 (!skb_is_gso(skb) &&
1389 bd_num > max_non_tso_bd_num)) {
1390 trace_hns3_over_max_bd(skb);
1394 u64_stats_update_begin(&ring->syncp);
1395 ring->stats.tx_copy++;
1396 u64_stats_update_end(&ring->syncp);
1400 if (likely(ring_space(ring) >= bd_num))
1403 netif_stop_subqueue(netdev, ring->queue_index);
1404 smp_mb(); /* Memory barrier before checking ring_space */
1406 /* Start queue in case hns3_clean_tx_ring has just made room
1407 * available and has not seen the queue stopped state performed
1408 * by netif_stop_subqueue above.
1410 if (ring_space(ring) >= bd_num && netif_carrier_ok(netdev) &&
1411 !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
1412 netif_start_subqueue(netdev, ring->queue_index);
1419 static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
1421 struct device *dev = ring_to_dev(ring);
1424 for (i = 0; i < ring->desc_num; i++) {
1425 struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1427 memset(desc, 0, sizeof(*desc));
1429 /* check if this is where we started */
1430 if (ring->next_to_use == next_to_use_orig)
1434 ring_ptr_move_bw(ring, next_to_use);
1436 if (!ring->desc_cb[ring->next_to_use].dma)
1439 /* unmap the descriptor dma address */
1440 if (ring->desc_cb[ring->next_to_use].type == DESC_TYPE_SKB ||
1441 ring->desc_cb[ring->next_to_use].type ==
1442 DESC_TYPE_FRAGLIST_SKB)
1443 dma_unmap_single(dev,
1444 ring->desc_cb[ring->next_to_use].dma,
1445 ring->desc_cb[ring->next_to_use].length,
1447 else if (ring->desc_cb[ring->next_to_use].length)
1449 ring->desc_cb[ring->next_to_use].dma,
1450 ring->desc_cb[ring->next_to_use].length,
1453 ring->desc_cb[ring->next_to_use].length = 0;
1454 ring->desc_cb[ring->next_to_use].dma = 0;
1455 ring->desc_cb[ring->next_to_use].type = DESC_TYPE_UNKNOWN;
1459 static int hns3_fill_skb_to_desc(struct hns3_enet_ring *ring,
1460 struct sk_buff *skb, enum hns_desc_type type)
1462 unsigned int size = skb_headlen(skb);
1463 int i, ret, bd_num = 0;
1466 ret = hns3_fill_desc(ring, skb, size, type);
1467 if (unlikely(ret < 0))
1473 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1474 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1476 size = skb_frag_size(frag);
1480 ret = hns3_fill_desc(ring, frag, size, DESC_TYPE_PAGE);
1481 if (unlikely(ret < 0))
1490 static void hns3_tx_doorbell(struct hns3_enet_ring *ring, int num,
1493 ring->pending_buf += num;
1496 u64_stats_update_begin(&ring->syncp);
1497 ring->stats.tx_more++;
1498 u64_stats_update_end(&ring->syncp);
1502 if (!ring->pending_buf)
1505 writel(ring->pending_buf,
1506 ring->tqp->io_base + HNS3_RING_TX_RING_TAIL_REG);
1507 ring->pending_buf = 0;
1508 WRITE_ONCE(ring->last_to_use, ring->next_to_use);
1511 netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
1513 struct hns3_nic_priv *priv = netdev_priv(netdev);
1514 struct hns3_enet_ring *ring = &priv->ring[skb->queue_mapping];
1515 struct netdev_queue *dev_queue;
1516 int pre_ntu, next_to_use_head;
1517 struct sk_buff *frag_skb;
1522 /* Hardware can only handle short frames above 32 bytes */
1523 if (skb_put_padto(skb, HNS3_MIN_TX_LEN)) {
1524 hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
1525 return NETDEV_TX_OK;
1528 /* Prefetch the data used later */
1529 prefetch(skb->data);
1531 ret = hns3_nic_maybe_stop_tx(ring, netdev, skb);
1532 if (unlikely(ret <= 0)) {
1533 if (ret == -EBUSY) {
1534 u64_stats_update_begin(&ring->syncp);
1535 ring->stats.tx_busy++;
1536 u64_stats_update_end(&ring->syncp);
1537 hns3_tx_doorbell(ring, 0, true);
1538 return NETDEV_TX_BUSY;
1539 } else if (ret == -ENOMEM) {
1540 u64_stats_update_begin(&ring->syncp);
1541 ring->stats.sw_err_cnt++;
1542 u64_stats_update_end(&ring->syncp);
1545 hns3_rl_err(netdev, "xmit error: %d!\n", ret);
1549 next_to_use_head = ring->next_to_use;
1551 ret = hns3_fill_skb_desc(ring, skb, &ring->desc[ring->next_to_use]);
1552 if (unlikely(ret < 0))
1555 ret = hns3_fill_skb_to_desc(ring, skb, DESC_TYPE_SKB);
1556 if (unlikely(ret < 0))
1561 skb_walk_frags(skb, frag_skb) {
1562 ret = hns3_fill_skb_to_desc(ring, frag_skb,
1563 DESC_TYPE_FRAGLIST_SKB);
1564 if (unlikely(ret < 0))
1570 pre_ntu = ring->next_to_use ? (ring->next_to_use - 1) :
1571 (ring->desc_num - 1);
1572 ring->desc[pre_ntu].tx.bdtp_fe_sc_vld_ra_ri |=
1573 cpu_to_le16(BIT(HNS3_TXD_FE_B));
1574 trace_hns3_tx_desc(ring, pre_ntu);
1576 /* Complete translate all packets */
1577 dev_queue = netdev_get_tx_queue(netdev, ring->queue_index);
1578 doorbell = __netdev_tx_sent_queue(dev_queue, skb->len,
1579 netdev_xmit_more());
1580 hns3_tx_doorbell(ring, bd_num, doorbell);
1582 return NETDEV_TX_OK;
1585 hns3_clear_desc(ring, next_to_use_head);
1588 dev_kfree_skb_any(skb);
1589 hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
1590 return NETDEV_TX_OK;
1593 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
1595 struct hnae3_handle *h = hns3_get_handle(netdev);
1596 struct sockaddr *mac_addr = p;
1599 if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1600 return -EADDRNOTAVAIL;
1602 if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
1603 netdev_info(netdev, "already using mac address %pM\n",
1608 /* For VF device, if there is a perm_addr, then the user will not
1609 * be allowed to change the address.
1611 if (!hns3_is_phys_func(h->pdev) &&
1612 !is_zero_ether_addr(netdev->perm_addr)) {
1613 netdev_err(netdev, "has permanent MAC %pM, user MAC %pM not allow\n",
1614 netdev->perm_addr, mac_addr->sa_data);
1618 ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
1620 netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
1624 ether_addr_copy(netdev->dev_addr, mac_addr->sa_data);
1629 static int hns3_nic_do_ioctl(struct net_device *netdev,
1630 struct ifreq *ifr, int cmd)
1632 struct hnae3_handle *h = hns3_get_handle(netdev);
1634 if (!netif_running(netdev))
1637 if (!h->ae_algo->ops->do_ioctl)
1640 return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
1643 static int hns3_nic_set_features(struct net_device *netdev,
1644 netdev_features_t features)
1646 netdev_features_t changed = netdev->features ^ features;
1647 struct hns3_nic_priv *priv = netdev_priv(netdev);
1648 struct hnae3_handle *h = priv->ae_handle;
1652 if (changed & (NETIF_F_GRO_HW) && h->ae_algo->ops->set_gro_en) {
1653 enable = !!(features & NETIF_F_GRO_HW);
1654 ret = h->ae_algo->ops->set_gro_en(h, enable);
1659 if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
1660 h->ae_algo->ops->enable_hw_strip_rxvtag) {
1661 enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
1662 ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, enable);
1667 if ((changed & NETIF_F_NTUPLE) && h->ae_algo->ops->enable_fd) {
1668 enable = !!(features & NETIF_F_NTUPLE);
1669 h->ae_algo->ops->enable_fd(h, enable);
1672 netdev->features = features;
1676 static netdev_features_t hns3_features_check(struct sk_buff *skb,
1677 struct net_device *dev,
1678 netdev_features_t features)
1680 #define HNS3_MAX_HDR_LEN 480U
1681 #define HNS3_MAX_L4_HDR_LEN 60U
1685 if (skb->ip_summed != CHECKSUM_PARTIAL)
1688 if (skb->encapsulation)
1689 len = skb_inner_transport_header(skb) - skb->data;
1691 len = skb_transport_header(skb) - skb->data;
1693 /* Assume L4 is 60 byte as TCP is the only protocol with a
1694 * a flexible value, and it's max len is 60 bytes.
1696 len += HNS3_MAX_L4_HDR_LEN;
1698 /* Hardware only supports checksum on the skb with a max header
1701 if (len > HNS3_MAX_HDR_LEN)
1702 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
1707 static void hns3_nic_get_stats64(struct net_device *netdev,
1708 struct rtnl_link_stats64 *stats)
1710 struct hns3_nic_priv *priv = netdev_priv(netdev);
1711 int queue_num = priv->ae_handle->kinfo.num_tqps;
1712 struct hnae3_handle *handle = priv->ae_handle;
1713 struct hns3_enet_ring *ring;
1714 u64 rx_length_errors = 0;
1715 u64 rx_crc_errors = 0;
1716 u64 rx_multicast = 0;
1728 if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
1731 handle->ae_algo->ops->update_stats(handle, &netdev->stats);
1733 for (idx = 0; idx < queue_num; idx++) {
1734 /* fetch the tx stats */
1735 ring = &priv->ring[idx];
1737 start = u64_stats_fetch_begin_irq(&ring->syncp);
1738 tx_bytes += ring->stats.tx_bytes;
1739 tx_pkts += ring->stats.tx_pkts;
1740 tx_drop += ring->stats.sw_err_cnt;
1741 tx_drop += ring->stats.tx_vlan_err;
1742 tx_drop += ring->stats.tx_l4_proto_err;
1743 tx_drop += ring->stats.tx_l2l3l4_err;
1744 tx_drop += ring->stats.tx_tso_err;
1745 tx_errors += ring->stats.sw_err_cnt;
1746 tx_errors += ring->stats.tx_vlan_err;
1747 tx_errors += ring->stats.tx_l4_proto_err;
1748 tx_errors += ring->stats.tx_l2l3l4_err;
1749 tx_errors += ring->stats.tx_tso_err;
1750 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1752 /* fetch the rx stats */
1753 ring = &priv->ring[idx + queue_num];
1755 start = u64_stats_fetch_begin_irq(&ring->syncp);
1756 rx_bytes += ring->stats.rx_bytes;
1757 rx_pkts += ring->stats.rx_pkts;
1758 rx_drop += ring->stats.l2_err;
1759 rx_errors += ring->stats.l2_err;
1760 rx_errors += ring->stats.l3l4_csum_err;
1761 rx_crc_errors += ring->stats.l2_err;
1762 rx_multicast += ring->stats.rx_multicast;
1763 rx_length_errors += ring->stats.err_pkt_len;
1764 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1767 stats->tx_bytes = tx_bytes;
1768 stats->tx_packets = tx_pkts;
1769 stats->rx_bytes = rx_bytes;
1770 stats->rx_packets = rx_pkts;
1772 stats->rx_errors = rx_errors;
1773 stats->multicast = rx_multicast;
1774 stats->rx_length_errors = rx_length_errors;
1775 stats->rx_crc_errors = rx_crc_errors;
1776 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
1778 stats->tx_errors = tx_errors;
1779 stats->rx_dropped = rx_drop;
1780 stats->tx_dropped = tx_drop;
1781 stats->collisions = netdev->stats.collisions;
1782 stats->rx_over_errors = netdev->stats.rx_over_errors;
1783 stats->rx_frame_errors = netdev->stats.rx_frame_errors;
1784 stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
1785 stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
1786 stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
1787 stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
1788 stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
1789 stats->tx_window_errors = netdev->stats.tx_window_errors;
1790 stats->rx_compressed = netdev->stats.rx_compressed;
1791 stats->tx_compressed = netdev->stats.tx_compressed;
1794 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
1796 struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
1797 u8 *prio_tc = mqprio_qopt->qopt.prio_tc_map;
1798 struct hnae3_knic_private_info *kinfo;
1799 u8 tc = mqprio_qopt->qopt.num_tc;
1800 u16 mode = mqprio_qopt->mode;
1801 u8 hw = mqprio_qopt->qopt.hw;
1802 struct hnae3_handle *h;
1804 if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
1805 mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
1808 if (tc > HNAE3_MAX_TC)
1814 h = hns3_get_handle(netdev);
1817 netif_dbg(h, drv, netdev, "setup tc: num_tc=%u\n", tc);
1819 return (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
1820 kinfo->dcb_ops->setup_tc(h, tc ? tc : 1, prio_tc) : -EOPNOTSUPP;
1823 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
1826 if (type != TC_SETUP_QDISC_MQPRIO)
1829 return hns3_setup_tc(dev, type_data);
1832 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
1833 __be16 proto, u16 vid)
1835 struct hnae3_handle *h = hns3_get_handle(netdev);
1838 if (h->ae_algo->ops->set_vlan_filter)
1839 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
1844 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
1845 __be16 proto, u16 vid)
1847 struct hnae3_handle *h = hns3_get_handle(netdev);
1850 if (h->ae_algo->ops->set_vlan_filter)
1851 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
1856 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
1857 u8 qos, __be16 vlan_proto)
1859 struct hnae3_handle *h = hns3_get_handle(netdev);
1862 netif_dbg(h, drv, netdev,
1863 "set vf vlan: vf=%d, vlan=%u, qos=%u, vlan_proto=0x%x\n",
1864 vf, vlan, qos, ntohs(vlan_proto));
1866 if (h->ae_algo->ops->set_vf_vlan_filter)
1867 ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
1873 static int hns3_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
1875 struct hnae3_handle *handle = hns3_get_handle(netdev);
1877 if (hns3_nic_resetting(netdev))
1880 if (!handle->ae_algo->ops->set_vf_spoofchk)
1883 return handle->ae_algo->ops->set_vf_spoofchk(handle, vf, enable);
1886 static int hns3_set_vf_trust(struct net_device *netdev, int vf, bool enable)
1888 struct hnae3_handle *handle = hns3_get_handle(netdev);
1890 if (!handle->ae_algo->ops->set_vf_trust)
1893 return handle->ae_algo->ops->set_vf_trust(handle, vf, enable);
1896 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
1898 struct hnae3_handle *h = hns3_get_handle(netdev);
1901 if (hns3_nic_resetting(netdev))
1904 if (!h->ae_algo->ops->set_mtu)
1907 netif_dbg(h, drv, netdev,
1908 "change mtu from %u to %d\n", netdev->mtu, new_mtu);
1910 ret = h->ae_algo->ops->set_mtu(h, new_mtu);
1912 netdev_err(netdev, "failed to change MTU in hardware %d\n",
1915 netdev->mtu = new_mtu;
1920 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
1922 struct hns3_nic_priv *priv = netdev_priv(ndev);
1923 struct hnae3_handle *h = hns3_get_handle(ndev);
1924 struct hns3_enet_ring *tx_ring;
1925 struct napi_struct *napi;
1926 int timeout_queue = 0;
1927 int hw_head, hw_tail;
1928 int fbd_num, fbd_oft;
1929 int ebd_num, ebd_oft;
1934 /* Find the stopped queue the same way the stack does */
1935 for (i = 0; i < ndev->num_tx_queues; i++) {
1936 struct netdev_queue *q;
1937 unsigned long trans_start;
1939 q = netdev_get_tx_queue(ndev, i);
1940 trans_start = q->trans_start;
1941 if (netif_xmit_stopped(q) &&
1943 (trans_start + ndev->watchdog_timeo))) {
1945 netdev_info(ndev, "queue state: 0x%lx, delta msecs: %u\n",
1947 jiffies_to_msecs(jiffies - trans_start));
1952 if (i == ndev->num_tx_queues) {
1954 "no netdev TX timeout queue found, timeout count: %llu\n",
1955 priv->tx_timeout_count);
1959 priv->tx_timeout_count++;
1961 tx_ring = &priv->ring[timeout_queue];
1962 napi = &tx_ring->tqp_vector->napi;
1965 "tx_timeout count: %llu, queue id: %d, SW_NTU: 0x%x, SW_NTC: 0x%x, napi state: %lu\n",
1966 priv->tx_timeout_count, timeout_queue, tx_ring->next_to_use,
1967 tx_ring->next_to_clean, napi->state);
1970 "tx_pkts: %llu, tx_bytes: %llu, sw_err_cnt: %llu, tx_pending: %d\n",
1971 tx_ring->stats.tx_pkts, tx_ring->stats.tx_bytes,
1972 tx_ring->stats.sw_err_cnt, tx_ring->pending_buf);
1975 "seg_pkt_cnt: %llu, tx_more: %llu, restart_queue: %llu, tx_busy: %llu\n",
1976 tx_ring->stats.seg_pkt_cnt, tx_ring->stats.tx_more,
1977 tx_ring->stats.restart_queue, tx_ring->stats.tx_busy);
1979 /* When mac received many pause frames continuous, it's unable to send
1980 * packets, which may cause tx timeout
1982 if (h->ae_algo->ops->get_mac_stats) {
1983 struct hns3_mac_stats mac_stats;
1985 h->ae_algo->ops->get_mac_stats(h, &mac_stats);
1986 netdev_info(ndev, "tx_pause_cnt: %llu, rx_pause_cnt: %llu\n",
1987 mac_stats.tx_pause_cnt, mac_stats.rx_pause_cnt);
1990 hw_head = readl_relaxed(tx_ring->tqp->io_base +
1991 HNS3_RING_TX_RING_HEAD_REG);
1992 hw_tail = readl_relaxed(tx_ring->tqp->io_base +
1993 HNS3_RING_TX_RING_TAIL_REG);
1994 fbd_num = readl_relaxed(tx_ring->tqp->io_base +
1995 HNS3_RING_TX_RING_FBDNUM_REG);
1996 fbd_oft = readl_relaxed(tx_ring->tqp->io_base +
1997 HNS3_RING_TX_RING_OFFSET_REG);
1998 ebd_num = readl_relaxed(tx_ring->tqp->io_base +
1999 HNS3_RING_TX_RING_EBDNUM_REG);
2000 ebd_oft = readl_relaxed(tx_ring->tqp->io_base +
2001 HNS3_RING_TX_RING_EBD_OFFSET_REG);
2002 bd_num = readl_relaxed(tx_ring->tqp->io_base +
2003 HNS3_RING_TX_RING_BD_NUM_REG);
2004 bd_err = readl_relaxed(tx_ring->tqp->io_base +
2005 HNS3_RING_TX_RING_BD_ERR_REG);
2006 ring_en = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_EN_REG);
2007 tc = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_TX_RING_TC_REG);
2010 "BD_NUM: 0x%x HW_HEAD: 0x%x, HW_TAIL: 0x%x, BD_ERR: 0x%x, INT: 0x%x\n",
2011 bd_num, hw_head, hw_tail, bd_err,
2012 readl(tx_ring->tqp_vector->mask_addr));
2014 "RING_EN: 0x%x, TC: 0x%x, FBD_NUM: 0x%x FBD_OFT: 0x%x, EBD_NUM: 0x%x, EBD_OFT: 0x%x\n",
2015 ring_en, tc, fbd_num, fbd_oft, ebd_num, ebd_oft);
2020 static void hns3_nic_net_timeout(struct net_device *ndev, unsigned int txqueue)
2022 struct hns3_nic_priv *priv = netdev_priv(ndev);
2023 struct hnae3_handle *h = priv->ae_handle;
2025 if (!hns3_get_tx_timeo_queue_info(ndev))
2028 /* request the reset, and let the hclge to determine
2029 * which reset level should be done
2031 if (h->ae_algo->ops->reset_event)
2032 h->ae_algo->ops->reset_event(h->pdev, h);
2035 #ifdef CONFIG_RFS_ACCEL
2036 static int hns3_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
2037 u16 rxq_index, u32 flow_id)
2039 struct hnae3_handle *h = hns3_get_handle(dev);
2040 struct flow_keys fkeys;
2042 if (!h->ae_algo->ops->add_arfs_entry)
2045 if (skb->encapsulation)
2046 return -EPROTONOSUPPORT;
2048 if (!skb_flow_dissect_flow_keys(skb, &fkeys, 0))
2049 return -EPROTONOSUPPORT;
2051 if ((fkeys.basic.n_proto != htons(ETH_P_IP) &&
2052 fkeys.basic.n_proto != htons(ETH_P_IPV6)) ||
2053 (fkeys.basic.ip_proto != IPPROTO_TCP &&
2054 fkeys.basic.ip_proto != IPPROTO_UDP))
2055 return -EPROTONOSUPPORT;
2057 return h->ae_algo->ops->add_arfs_entry(h, rxq_index, flow_id, &fkeys);
2061 static int hns3_nic_get_vf_config(struct net_device *ndev, int vf,
2062 struct ifla_vf_info *ivf)
2064 struct hnae3_handle *h = hns3_get_handle(ndev);
2066 if (!h->ae_algo->ops->get_vf_config)
2069 return h->ae_algo->ops->get_vf_config(h, vf, ivf);
2072 static int hns3_nic_set_vf_link_state(struct net_device *ndev, int vf,
2075 struct hnae3_handle *h = hns3_get_handle(ndev);
2077 if (!h->ae_algo->ops->set_vf_link_state)
2080 return h->ae_algo->ops->set_vf_link_state(h, vf, link_state);
2083 static int hns3_nic_set_vf_rate(struct net_device *ndev, int vf,
2084 int min_tx_rate, int max_tx_rate)
2086 struct hnae3_handle *h = hns3_get_handle(ndev);
2088 if (!h->ae_algo->ops->set_vf_rate)
2091 return h->ae_algo->ops->set_vf_rate(h, vf, min_tx_rate, max_tx_rate,
2095 static int hns3_nic_set_vf_mac(struct net_device *netdev, int vf_id, u8 *mac)
2097 struct hnae3_handle *h = hns3_get_handle(netdev);
2099 if (!h->ae_algo->ops->set_vf_mac)
2102 if (is_multicast_ether_addr(mac)) {
2104 "Invalid MAC:%pM specified. Could not set MAC\n",
2109 return h->ae_algo->ops->set_vf_mac(h, vf_id, mac);
2112 static const struct net_device_ops hns3_nic_netdev_ops = {
2113 .ndo_open = hns3_nic_net_open,
2114 .ndo_stop = hns3_nic_net_stop,
2115 .ndo_start_xmit = hns3_nic_net_xmit,
2116 .ndo_tx_timeout = hns3_nic_net_timeout,
2117 .ndo_set_mac_address = hns3_nic_net_set_mac_address,
2118 .ndo_do_ioctl = hns3_nic_do_ioctl,
2119 .ndo_change_mtu = hns3_nic_change_mtu,
2120 .ndo_set_features = hns3_nic_set_features,
2121 .ndo_features_check = hns3_features_check,
2122 .ndo_get_stats64 = hns3_nic_get_stats64,
2123 .ndo_setup_tc = hns3_nic_setup_tc,
2124 .ndo_set_rx_mode = hns3_nic_set_rx_mode,
2125 .ndo_vlan_rx_add_vid = hns3_vlan_rx_add_vid,
2126 .ndo_vlan_rx_kill_vid = hns3_vlan_rx_kill_vid,
2127 .ndo_set_vf_vlan = hns3_ndo_set_vf_vlan,
2128 .ndo_set_vf_spoofchk = hns3_set_vf_spoofchk,
2129 .ndo_set_vf_trust = hns3_set_vf_trust,
2130 #ifdef CONFIG_RFS_ACCEL
2131 .ndo_rx_flow_steer = hns3_rx_flow_steer,
2133 .ndo_get_vf_config = hns3_nic_get_vf_config,
2134 .ndo_set_vf_link_state = hns3_nic_set_vf_link_state,
2135 .ndo_set_vf_rate = hns3_nic_set_vf_rate,
2136 .ndo_set_vf_mac = hns3_nic_set_vf_mac,
2139 bool hns3_is_phys_func(struct pci_dev *pdev)
2141 u32 dev_id = pdev->device;
2144 case HNAE3_DEV_ID_GE:
2145 case HNAE3_DEV_ID_25GE:
2146 case HNAE3_DEV_ID_25GE_RDMA:
2147 case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
2148 case HNAE3_DEV_ID_50GE_RDMA:
2149 case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
2150 case HNAE3_DEV_ID_100G_RDMA_MACSEC:
2151 case HNAE3_DEV_ID_200G_RDMA:
2153 case HNAE3_DEV_ID_VF:
2154 case HNAE3_DEV_ID_RDMA_DCB_PFC_VF:
2157 dev_warn(&pdev->dev, "un-recognized pci device-id %u",
2164 static void hns3_disable_sriov(struct pci_dev *pdev)
2166 /* If our VFs are assigned we cannot shut down SR-IOV
2167 * without causing issues, so just leave the hardware
2168 * available but disabled
2170 if (pci_vfs_assigned(pdev)) {
2171 dev_warn(&pdev->dev,
2172 "disabling driver while VFs are assigned\n");
2176 pci_disable_sriov(pdev);
2179 /* hns3_probe - Device initialization routine
2180 * @pdev: PCI device information struct
2181 * @ent: entry in hns3_pci_tbl
2183 * hns3_probe initializes a PF identified by a pci_dev structure.
2184 * The OS initialization, configuring of the PF private structure,
2185 * and a hardware reset occur.
2187 * Returns 0 on success, negative on failure
2189 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2191 struct hnae3_ae_dev *ae_dev;
2194 ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev), GFP_KERNEL);
2198 ae_dev->pdev = pdev;
2199 ae_dev->flag = ent->driver_data;
2200 pci_set_drvdata(pdev, ae_dev);
2202 ret = hnae3_register_ae_dev(ae_dev);
2204 pci_set_drvdata(pdev, NULL);
2209 /* hns3_remove - Device removal routine
2210 * @pdev: PCI device information struct
2212 static void hns3_remove(struct pci_dev *pdev)
2214 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2216 if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
2217 hns3_disable_sriov(pdev);
2219 hnae3_unregister_ae_dev(ae_dev);
2220 pci_set_drvdata(pdev, NULL);
2224 * hns3_pci_sriov_configure
2225 * @pdev: pointer to a pci_dev structure
2226 * @num_vfs: number of VFs to allocate
2228 * Enable or change the number of VFs. Called when the user updates the number
2231 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
2235 if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
2236 dev_warn(&pdev->dev, "Can not config SRIOV\n");
2241 ret = pci_enable_sriov(pdev, num_vfs);
2243 dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
2246 } else if (!pci_vfs_assigned(pdev)) {
2247 pci_disable_sriov(pdev);
2249 dev_warn(&pdev->dev,
2250 "Unable to free VFs because some are assigned to VMs.\n");
2256 static void hns3_shutdown(struct pci_dev *pdev)
2258 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2260 hnae3_unregister_ae_dev(ae_dev);
2261 pci_set_drvdata(pdev, NULL);
2263 if (system_state == SYSTEM_POWER_OFF)
2264 pci_set_power_state(pdev, PCI_D3hot);
2267 static pci_ers_result_t hns3_error_detected(struct pci_dev *pdev,
2268 pci_channel_state_t state)
2270 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2271 pci_ers_result_t ret;
2273 dev_info(&pdev->dev, "PCI error detected, state(=%d)!!\n", state);
2275 if (state == pci_channel_io_perm_failure)
2276 return PCI_ERS_RESULT_DISCONNECT;
2278 if (!ae_dev || !ae_dev->ops) {
2280 "Can't recover - error happened before device initialized\n");
2281 return PCI_ERS_RESULT_NONE;
2284 if (ae_dev->ops->handle_hw_ras_error)
2285 ret = ae_dev->ops->handle_hw_ras_error(ae_dev);
2287 return PCI_ERS_RESULT_NONE;
2292 static pci_ers_result_t hns3_slot_reset(struct pci_dev *pdev)
2294 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2295 const struct hnae3_ae_ops *ops;
2296 enum hnae3_reset_type reset_type;
2297 struct device *dev = &pdev->dev;
2299 if (!ae_dev || !ae_dev->ops)
2300 return PCI_ERS_RESULT_NONE;
2303 /* request the reset */
2304 if (ops->reset_event && ops->get_reset_level &&
2305 ops->set_default_reset_request) {
2306 if (ae_dev->hw_err_reset_req) {
2307 reset_type = ops->get_reset_level(ae_dev,
2308 &ae_dev->hw_err_reset_req);
2309 ops->set_default_reset_request(ae_dev, reset_type);
2310 dev_info(dev, "requesting reset due to PCI error\n");
2311 ops->reset_event(pdev, NULL);
2314 return PCI_ERS_RESULT_RECOVERED;
2317 return PCI_ERS_RESULT_DISCONNECT;
2320 static void hns3_reset_prepare(struct pci_dev *pdev)
2322 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2324 dev_info(&pdev->dev, "FLR prepare\n");
2325 if (ae_dev && ae_dev->ops && ae_dev->ops->flr_prepare)
2326 ae_dev->ops->flr_prepare(ae_dev);
2329 static void hns3_reset_done(struct pci_dev *pdev)
2331 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2333 dev_info(&pdev->dev, "FLR done\n");
2334 if (ae_dev && ae_dev->ops && ae_dev->ops->flr_done)
2335 ae_dev->ops->flr_done(ae_dev);
2338 static const struct pci_error_handlers hns3_err_handler = {
2339 .error_detected = hns3_error_detected,
2340 .slot_reset = hns3_slot_reset,
2341 .reset_prepare = hns3_reset_prepare,
2342 .reset_done = hns3_reset_done,
2345 static struct pci_driver hns3_driver = {
2346 .name = hns3_driver_name,
2347 .id_table = hns3_pci_tbl,
2348 .probe = hns3_probe,
2349 .remove = hns3_remove,
2350 .shutdown = hns3_shutdown,
2351 .sriov_configure = hns3_pci_sriov_configure,
2352 .err_handler = &hns3_err_handler,
2355 /* set default feature to hns3 */
2356 static void hns3_set_default_feature(struct net_device *netdev)
2358 struct hnae3_handle *h = hns3_get_handle(netdev);
2359 struct pci_dev *pdev = h->pdev;
2360 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2362 netdev->priv_flags |= IFF_UNICAST_FLT;
2364 netdev->hw_enc_features |= NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2365 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2366 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2367 NETIF_F_SCTP_CRC | NETIF_F_TSO_MANGLEID | NETIF_F_FRAGLIST;
2369 netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
2371 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
2372 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
2373 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2374 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2375 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2376 NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
2378 netdev->vlan_features |= NETIF_F_RXCSUM |
2379 NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO |
2380 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2381 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2382 NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
2384 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX |
2385 NETIF_F_HW_VLAN_CTAG_RX |
2386 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2387 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2388 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2389 NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
2391 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V2) {
2392 netdev->hw_features |= NETIF_F_GRO_HW;
2393 netdev->features |= NETIF_F_GRO_HW;
2395 if (!(h->flags & HNAE3_SUPPORT_VF)) {
2396 netdev->hw_features |= NETIF_F_NTUPLE;
2397 netdev->features |= NETIF_F_NTUPLE;
2401 if (test_bit(HNAE3_DEV_SUPPORT_UDP_GSO_B, ae_dev->caps)) {
2402 netdev->hw_features |= NETIF_F_GSO_UDP_L4;
2403 netdev->features |= NETIF_F_GSO_UDP_L4;
2404 netdev->vlan_features |= NETIF_F_GSO_UDP_L4;
2405 netdev->hw_enc_features |= NETIF_F_GSO_UDP_L4;
2408 if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps)) {
2409 netdev->hw_features |= NETIF_F_HW_CSUM;
2410 netdev->features |= NETIF_F_HW_CSUM;
2411 netdev->vlan_features |= NETIF_F_HW_CSUM;
2412 netdev->hw_enc_features |= NETIF_F_HW_CSUM;
2414 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2415 netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2416 netdev->vlan_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2417 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2420 if (test_bit(HNAE3_DEV_SUPPORT_UDP_TUNNEL_CSUM_B, ae_dev->caps)) {
2421 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
2422 netdev->features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
2423 netdev->vlan_features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
2424 netdev->hw_enc_features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
2428 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
2429 struct hns3_desc_cb *cb)
2431 unsigned int order = hns3_page_order(ring);
2434 p = dev_alloc_pages(order);
2439 cb->page_offset = 0;
2441 cb->buf = page_address(p);
2442 cb->length = hns3_page_size(ring);
2443 cb->type = DESC_TYPE_PAGE;
2444 page_ref_add(p, USHRT_MAX - 1);
2445 cb->pagecnt_bias = USHRT_MAX;
2450 static void hns3_free_buffer(struct hns3_enet_ring *ring,
2451 struct hns3_desc_cb *cb, int budget)
2453 if (cb->type == DESC_TYPE_SKB)
2454 napi_consume_skb(cb->priv, budget);
2455 else if (!HNAE3_IS_TX_RING(ring) && cb->pagecnt_bias)
2456 __page_frag_cache_drain(cb->priv, cb->pagecnt_bias);
2457 memset(cb, 0, sizeof(*cb));
2460 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
2462 cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
2463 cb->length, ring_to_dma_dir(ring));
2465 if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
2471 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
2472 struct hns3_desc_cb *cb)
2474 if (cb->type == DESC_TYPE_SKB || cb->type == DESC_TYPE_FRAGLIST_SKB)
2475 dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
2476 ring_to_dma_dir(ring));
2477 else if (cb->length)
2478 dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
2479 ring_to_dma_dir(ring));
2482 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
2484 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2485 ring->desc[i].addr = 0;
2488 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i,
2491 struct hns3_desc_cb *cb = &ring->desc_cb[i];
2493 if (!ring->desc_cb[i].dma)
2496 hns3_buffer_detach(ring, i);
2497 hns3_free_buffer(ring, cb, budget);
2500 static void hns3_free_buffers(struct hns3_enet_ring *ring)
2504 for (i = 0; i < ring->desc_num; i++)
2505 hns3_free_buffer_detach(ring, i, 0);
2508 /* free desc along with its attached buffer */
2509 static void hns3_free_desc(struct hns3_enet_ring *ring)
2511 int size = ring->desc_num * sizeof(ring->desc[0]);
2513 hns3_free_buffers(ring);
2516 dma_free_coherent(ring_to_dev(ring), size,
2517 ring->desc, ring->desc_dma_addr);
2522 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
2524 int size = ring->desc_num * sizeof(ring->desc[0]);
2526 ring->desc = dma_alloc_coherent(ring_to_dev(ring), size,
2527 &ring->desc_dma_addr, GFP_KERNEL);
2534 static int hns3_alloc_and_map_buffer(struct hns3_enet_ring *ring,
2535 struct hns3_desc_cb *cb)
2539 ret = hns3_alloc_buffer(ring, cb);
2543 ret = hns3_map_buffer(ring, cb);
2550 hns3_free_buffer(ring, cb, 0);
2555 static int hns3_alloc_and_attach_buffer(struct hns3_enet_ring *ring, int i)
2557 int ret = hns3_alloc_and_map_buffer(ring, &ring->desc_cb[i]);
2562 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2567 /* Allocate memory for raw pkg, and map with dma */
2568 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
2572 for (i = 0; i < ring->desc_num; i++) {
2573 ret = hns3_alloc_and_attach_buffer(ring, i);
2575 goto out_buffer_fail;
2581 for (j = i - 1; j >= 0; j--)
2582 hns3_free_buffer_detach(ring, j, 0);
2586 /* detach a in-used buffer and replace with a reserved one */
2587 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
2588 struct hns3_desc_cb *res_cb)
2590 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2591 ring->desc_cb[i] = *res_cb;
2592 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2593 ring->desc[i].rx.bd_base_info = 0;
2596 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
2598 ring->desc_cb[i].reuse_flag = 0;
2599 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
2600 ring->desc_cb[i].page_offset);
2601 ring->desc[i].rx.bd_base_info = 0;
2603 dma_sync_single_for_device(ring_to_dev(ring),
2604 ring->desc_cb[i].dma + ring->desc_cb[i].page_offset,
2605 hns3_buf_size(ring),
2609 static bool hns3_nic_reclaim_desc(struct hns3_enet_ring *ring,
2610 int *bytes, int *pkts, int budget)
2612 /* pair with ring->last_to_use update in hns3_tx_doorbell(),
2613 * smp_store_release() is not used in hns3_tx_doorbell() because
2614 * the doorbell operation already have the needed barrier operation.
2616 int ltu = smp_load_acquire(&ring->last_to_use);
2617 int ntc = ring->next_to_clean;
2618 struct hns3_desc_cb *desc_cb;
2619 bool reclaimed = false;
2620 struct hns3_desc *desc;
2622 while (ltu != ntc) {
2623 desc = &ring->desc[ntc];
2625 if (le16_to_cpu(desc->tx.bdtp_fe_sc_vld_ra_ri) &
2626 BIT(HNS3_TXD_VLD_B))
2629 desc_cb = &ring->desc_cb[ntc];
2630 (*pkts) += (desc_cb->type == DESC_TYPE_SKB);
2631 (*bytes) += desc_cb->length;
2632 /* desc_cb will be cleaned, after hnae3_free_buffer_detach */
2633 hns3_free_buffer_detach(ring, ntc, budget);
2635 if (++ntc == ring->desc_num)
2638 /* Issue prefetch for next Tx descriptor */
2639 prefetch(&ring->desc_cb[ntc]);
2643 if (unlikely(!reclaimed))
2646 /* This smp_store_release() pairs with smp_load_acquire() in
2647 * ring_space called by hns3_nic_net_xmit.
2649 smp_store_release(&ring->next_to_clean, ntc);
2653 void hns3_clean_tx_ring(struct hns3_enet_ring *ring, int budget)
2655 struct net_device *netdev = ring_to_netdev(ring);
2656 struct hns3_nic_priv *priv = netdev_priv(netdev);
2657 struct netdev_queue *dev_queue;
2663 if (unlikely(!hns3_nic_reclaim_desc(ring, &bytes, &pkts, budget)))
2666 ring->tqp_vector->tx_group.total_bytes += bytes;
2667 ring->tqp_vector->tx_group.total_packets += pkts;
2669 u64_stats_update_begin(&ring->syncp);
2670 ring->stats.tx_bytes += bytes;
2671 ring->stats.tx_pkts += pkts;
2672 u64_stats_update_end(&ring->syncp);
2674 dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
2675 netdev_tx_completed_queue(dev_queue, pkts, bytes);
2677 if (unlikely(netif_carrier_ok(netdev) &&
2678 ring_space(ring) > HNS3_MAX_TSO_BD_NUM)) {
2679 /* Make sure that anybody stopping the queue after this
2680 * sees the new next_to_clean.
2683 if (netif_tx_queue_stopped(dev_queue) &&
2684 !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
2685 netif_tx_wake_queue(dev_queue);
2686 ring->stats.restart_queue++;
2691 static int hns3_desc_unused(struct hns3_enet_ring *ring)
2693 int ntc = ring->next_to_clean;
2694 int ntu = ring->next_to_use;
2696 return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
2699 static void hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring,
2702 struct hns3_desc_cb *desc_cb;
2703 struct hns3_desc_cb res_cbs;
2706 for (i = 0; i < cleand_count; i++) {
2707 desc_cb = &ring->desc_cb[ring->next_to_use];
2708 if (desc_cb->reuse_flag) {
2709 u64_stats_update_begin(&ring->syncp);
2710 ring->stats.reuse_pg_cnt++;
2711 u64_stats_update_end(&ring->syncp);
2713 hns3_reuse_buffer(ring, ring->next_to_use);
2715 ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
2717 u64_stats_update_begin(&ring->syncp);
2718 ring->stats.sw_err_cnt++;
2719 u64_stats_update_end(&ring->syncp);
2721 hns3_rl_err(ring_to_netdev(ring),
2722 "alloc rx buffer failed: %d\n",
2726 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
2728 u64_stats_update_begin(&ring->syncp);
2729 ring->stats.non_reuse_pg++;
2730 u64_stats_update_end(&ring->syncp);
2733 ring_ptr_move_fw(ring, next_to_use);
2736 writel(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
2739 static bool hns3_page_is_reusable(struct page *page)
2741 return page_to_nid(page) == numa_mem_id() &&
2742 !page_is_pfmemalloc(page);
2745 static bool hns3_can_reuse_page(struct hns3_desc_cb *cb)
2747 return (page_count(cb->priv) - cb->pagecnt_bias) == 1;
2750 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
2751 struct hns3_enet_ring *ring, int pull_len,
2752 struct hns3_desc_cb *desc_cb)
2754 struct hns3_desc *desc = &ring->desc[ring->next_to_clean];
2755 int size = le16_to_cpu(desc->rx.size);
2756 u32 truesize = hns3_buf_size(ring);
2758 desc_cb->pagecnt_bias--;
2759 skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
2760 size - pull_len, truesize);
2762 /* Avoid re-using remote pages, or the stack is still using the page
2763 * when page_offset rollback to zero, flag default unreuse
2765 if (unlikely(!hns3_page_is_reusable(desc_cb->priv)) ||
2766 (!desc_cb->page_offset && !hns3_can_reuse_page(desc_cb))) {
2767 __page_frag_cache_drain(desc_cb->priv, desc_cb->pagecnt_bias);
2771 /* Move offset up to the next cache line */
2772 desc_cb->page_offset += truesize;
2774 if (desc_cb->page_offset + truesize <= hns3_page_size(ring)) {
2775 desc_cb->reuse_flag = 1;
2776 } else if (hns3_can_reuse_page(desc_cb)) {
2777 desc_cb->reuse_flag = 1;
2778 desc_cb->page_offset = 0;
2779 } else if (desc_cb->pagecnt_bias) {
2780 __page_frag_cache_drain(desc_cb->priv, desc_cb->pagecnt_bias);
2784 if (unlikely(!desc_cb->pagecnt_bias)) {
2785 page_ref_add(desc_cb->priv, USHRT_MAX);
2786 desc_cb->pagecnt_bias = USHRT_MAX;
2790 static int hns3_gro_complete(struct sk_buff *skb, u32 l234info)
2792 __be16 type = skb->protocol;
2796 while (eth_type_vlan(type)) {
2797 struct vlan_hdr *vh;
2799 if ((depth + VLAN_HLEN) > skb_headlen(skb))
2802 vh = (struct vlan_hdr *)(skb->data + depth);
2803 type = vh->h_vlan_encapsulated_proto;
2807 skb_set_network_header(skb, depth);
2809 if (type == htons(ETH_P_IP)) {
2810 const struct iphdr *iph = ip_hdr(skb);
2812 depth += sizeof(struct iphdr);
2813 skb_set_transport_header(skb, depth);
2815 th->check = ~tcp_v4_check(skb->len - depth, iph->saddr,
2817 } else if (type == htons(ETH_P_IPV6)) {
2818 const struct ipv6hdr *iph = ipv6_hdr(skb);
2820 depth += sizeof(struct ipv6hdr);
2821 skb_set_transport_header(skb, depth);
2823 th->check = ~tcp_v6_check(skb->len - depth, &iph->saddr,
2826 hns3_rl_err(skb->dev,
2827 "Error: FW GRO supports only IPv4/IPv6, not 0x%04x, depth: %d\n",
2828 be16_to_cpu(type), depth);
2832 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
2834 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
2836 if (l234info & BIT(HNS3_RXD_GRO_FIXID_B))
2837 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_FIXEDID;
2839 skb->csum_start = (unsigned char *)th - skb->head;
2840 skb->csum_offset = offsetof(struct tcphdr, check);
2841 skb->ip_summed = CHECKSUM_PARTIAL;
2843 trace_hns3_gro(skb);
2848 static void hns3_checksum_complete(struct hns3_enet_ring *ring,
2849 struct sk_buff *skb, u32 l234info)
2853 u64_stats_update_begin(&ring->syncp);
2854 ring->stats.csum_complete++;
2855 u64_stats_update_end(&ring->syncp);
2856 skb->ip_summed = CHECKSUM_COMPLETE;
2857 lo = hnae3_get_field(l234info, HNS3_RXD_L2_CSUM_L_M,
2858 HNS3_RXD_L2_CSUM_L_S);
2859 hi = hnae3_get_field(l234info, HNS3_RXD_L2_CSUM_H_M,
2860 HNS3_RXD_L2_CSUM_H_S);
2861 skb->csum = csum_unfold((__force __sum16)(lo | hi << 8));
2864 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
2865 u32 l234info, u32 bd_base_info, u32 ol_info)
2867 struct net_device *netdev = ring_to_netdev(ring);
2868 int l3_type, l4_type;
2871 skb->ip_summed = CHECKSUM_NONE;
2873 skb_checksum_none_assert(skb);
2875 if (!(netdev->features & NETIF_F_RXCSUM))
2878 if (l234info & BIT(HNS3_RXD_L2_CSUM_B)) {
2879 hns3_checksum_complete(ring, skb, l234info);
2883 /* check if hardware has done checksum */
2884 if (!(bd_base_info & BIT(HNS3_RXD_L3L4P_B)))
2887 if (unlikely(l234info & (BIT(HNS3_RXD_L3E_B) | BIT(HNS3_RXD_L4E_B) |
2888 BIT(HNS3_RXD_OL3E_B) |
2889 BIT(HNS3_RXD_OL4E_B)))) {
2890 u64_stats_update_begin(&ring->syncp);
2891 ring->stats.l3l4_csum_err++;
2892 u64_stats_update_end(&ring->syncp);
2897 ol4_type = hnae3_get_field(ol_info, HNS3_RXD_OL4ID_M,
2900 case HNS3_OL4_TYPE_MAC_IN_UDP:
2901 case HNS3_OL4_TYPE_NVGRE:
2902 skb->csum_level = 1;
2904 case HNS3_OL4_TYPE_NO_TUN:
2905 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
2907 l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
2910 /* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
2911 if ((l3_type == HNS3_L3_TYPE_IPV4 ||
2912 l3_type == HNS3_L3_TYPE_IPV6) &&
2913 (l4_type == HNS3_L4_TYPE_UDP ||
2914 l4_type == HNS3_L4_TYPE_TCP ||
2915 l4_type == HNS3_L4_TYPE_SCTP))
2916 skb->ip_summed = CHECKSUM_UNNECESSARY;
2923 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
2925 if (skb_has_frag_list(skb))
2926 napi_gro_flush(&ring->tqp_vector->napi, false);
2928 napi_gro_receive(&ring->tqp_vector->napi, skb);
2931 static bool hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
2932 struct hns3_desc *desc, u32 l234info,
2935 struct hnae3_handle *handle = ring->tqp->handle;
2936 struct pci_dev *pdev = ring->tqp->handle->pdev;
2937 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2939 if (unlikely(ae_dev->dev_version < HNAE3_DEVICE_VERSION_V2)) {
2940 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2941 if (!(*vlan_tag & VLAN_VID_MASK))
2942 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2944 return (*vlan_tag != 0);
2947 #define HNS3_STRP_OUTER_VLAN 0x1
2948 #define HNS3_STRP_INNER_VLAN 0x2
2949 #define HNS3_STRP_BOTH 0x3
2951 /* Hardware always insert VLAN tag into RX descriptor when
2952 * remove the tag from packet, driver needs to determine
2953 * reporting which tag to stack.
2955 switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
2956 HNS3_RXD_STRP_TAGP_S)) {
2957 case HNS3_STRP_OUTER_VLAN:
2958 if (handle->port_base_vlan_state !=
2959 HNAE3_PORT_BASE_VLAN_DISABLE)
2962 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2964 case HNS3_STRP_INNER_VLAN:
2965 if (handle->port_base_vlan_state !=
2966 HNAE3_PORT_BASE_VLAN_DISABLE)
2969 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2971 case HNS3_STRP_BOTH:
2972 if (handle->port_base_vlan_state ==
2973 HNAE3_PORT_BASE_VLAN_DISABLE)
2974 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2976 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2984 static void hns3_rx_ring_move_fw(struct hns3_enet_ring *ring)
2986 ring->desc[ring->next_to_clean].rx.bd_base_info &=
2987 cpu_to_le32(~BIT(HNS3_RXD_VLD_B));
2988 ring->next_to_clean += 1;
2990 if (unlikely(ring->next_to_clean == ring->desc_num))
2991 ring->next_to_clean = 0;
2994 static int hns3_alloc_skb(struct hns3_enet_ring *ring, unsigned int length,
2997 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
2998 struct net_device *netdev = ring_to_netdev(ring);
2999 struct sk_buff *skb;
3001 ring->skb = napi_alloc_skb(&ring->tqp_vector->napi, HNS3_RX_HEAD_SIZE);
3003 if (unlikely(!skb)) {
3004 hns3_rl_err(netdev, "alloc rx skb fail\n");
3006 u64_stats_update_begin(&ring->syncp);
3007 ring->stats.sw_err_cnt++;
3008 u64_stats_update_end(&ring->syncp);
3013 trace_hns3_rx_desc(ring);
3014 prefetchw(skb->data);
3016 ring->pending_buf = 1;
3018 ring->tail_skb = NULL;
3019 if (length <= HNS3_RX_HEAD_SIZE) {
3020 memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
3022 /* We can reuse buffer as-is, just make sure it is local */
3023 if (likely(hns3_page_is_reusable(desc_cb->priv)))
3024 desc_cb->reuse_flag = 1;
3025 else /* This page cannot be reused so discard it */
3026 __page_frag_cache_drain(desc_cb->priv,
3027 desc_cb->pagecnt_bias);
3029 hns3_rx_ring_move_fw(ring);
3032 u64_stats_update_begin(&ring->syncp);
3033 ring->stats.seg_pkt_cnt++;
3034 u64_stats_update_end(&ring->syncp);
3036 ring->pull_len = eth_get_headlen(netdev, va, HNS3_RX_HEAD_SIZE);
3037 __skb_put(skb, ring->pull_len);
3038 hns3_nic_reuse_page(skb, ring->frag_num++, ring, ring->pull_len,
3040 hns3_rx_ring_move_fw(ring);
3045 static int hns3_add_frag(struct hns3_enet_ring *ring)
3047 struct sk_buff *skb = ring->skb;
3048 struct sk_buff *head_skb = skb;
3049 struct sk_buff *new_skb;
3050 struct hns3_desc_cb *desc_cb;
3051 struct hns3_desc *desc;
3055 desc = &ring->desc[ring->next_to_clean];
3056 desc_cb = &ring->desc_cb[ring->next_to_clean];
3057 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
3058 /* make sure HW write desc complete */
3060 if (!(bd_base_info & BIT(HNS3_RXD_VLD_B)))
3063 if (unlikely(ring->frag_num >= MAX_SKB_FRAGS)) {
3064 new_skb = napi_alloc_skb(&ring->tqp_vector->napi, 0);
3065 if (unlikely(!new_skb)) {
3066 hns3_rl_err(ring_to_netdev(ring),
3067 "alloc rx fraglist skb fail\n");
3072 if (ring->tail_skb) {
3073 ring->tail_skb->next = new_skb;
3074 ring->tail_skb = new_skb;
3076 skb_shinfo(skb)->frag_list = new_skb;
3077 ring->tail_skb = new_skb;
3081 if (ring->tail_skb) {
3082 head_skb->truesize += hns3_buf_size(ring);
3083 head_skb->data_len += le16_to_cpu(desc->rx.size);
3084 head_skb->len += le16_to_cpu(desc->rx.size);
3085 skb = ring->tail_skb;
3088 dma_sync_single_for_cpu(ring_to_dev(ring),
3089 desc_cb->dma + desc_cb->page_offset,
3090 hns3_buf_size(ring),
3093 hns3_nic_reuse_page(skb, ring->frag_num++, ring, 0, desc_cb);
3094 trace_hns3_rx_desc(ring);
3095 hns3_rx_ring_move_fw(ring);
3096 ring->pending_buf++;
3097 } while (!(bd_base_info & BIT(HNS3_RXD_FE_B)));
3102 static int hns3_set_gro_and_checksum(struct hns3_enet_ring *ring,
3103 struct sk_buff *skb, u32 l234info,
3104 u32 bd_base_info, u32 ol_info)
3108 skb_shinfo(skb)->gso_size = hnae3_get_field(bd_base_info,
3109 HNS3_RXD_GRO_SIZE_M,
3110 HNS3_RXD_GRO_SIZE_S);
3111 /* if there is no HW GRO, do not set gro params */
3112 if (!skb_shinfo(skb)->gso_size) {
3113 hns3_rx_checksum(ring, skb, l234info, bd_base_info, ol_info);
3117 NAPI_GRO_CB(skb)->count = hnae3_get_field(l234info,
3118 HNS3_RXD_GRO_COUNT_M,
3119 HNS3_RXD_GRO_COUNT_S);
3121 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M, HNS3_RXD_L3ID_S);
3122 if (l3_type == HNS3_L3_TYPE_IPV4)
3123 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
3124 else if (l3_type == HNS3_L3_TYPE_IPV6)
3125 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
3129 return hns3_gro_complete(skb, l234info);
3132 static void hns3_set_rx_skb_rss_type(struct hns3_enet_ring *ring,
3133 struct sk_buff *skb, u32 rss_hash)
3135 struct hnae3_handle *handle = ring->tqp->handle;
3136 enum pkt_hash_types rss_type;
3139 rss_type = handle->kinfo.rss_type;
3141 rss_type = PKT_HASH_TYPE_NONE;
3143 skb_set_hash(skb, rss_hash, rss_type);
3146 static int hns3_handle_bdinfo(struct hns3_enet_ring *ring, struct sk_buff *skb)
3148 struct net_device *netdev = ring_to_netdev(ring);
3149 enum hns3_pkt_l2t_type l2_frame_type;
3150 u32 bd_base_info, l234info, ol_info;
3151 struct hns3_desc *desc;
3155 /* bdinfo handled below is only valid on the last BD of the
3156 * current packet, and ring->next_to_clean indicates the first
3157 * descriptor of next packet, so need - 1 below.
3159 pre_ntc = ring->next_to_clean ? (ring->next_to_clean - 1) :
3160 (ring->desc_num - 1);
3161 desc = &ring->desc[pre_ntc];
3162 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
3163 l234info = le32_to_cpu(desc->rx.l234_info);
3164 ol_info = le32_to_cpu(desc->rx.ol_info);
3166 /* Based on hw strategy, the tag offloaded will be stored at
3167 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
3168 * in one layer tag case.
3170 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
3173 if (hns3_parse_vlan_tag(ring, desc, l234info, &vlan_tag))
3174 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
3178 if (unlikely(!desc->rx.pkt_len || (l234info & (BIT(HNS3_RXD_TRUNCAT_B) |
3179 BIT(HNS3_RXD_L2E_B))))) {
3180 u64_stats_update_begin(&ring->syncp);
3181 if (l234info & BIT(HNS3_RXD_L2E_B))
3182 ring->stats.l2_err++;
3184 ring->stats.err_pkt_len++;
3185 u64_stats_update_end(&ring->syncp);
3192 /* Do update ip stack process */
3193 skb->protocol = eth_type_trans(skb, netdev);
3195 /* This is needed in order to enable forwarding support */
3196 ret = hns3_set_gro_and_checksum(ring, skb, l234info,
3197 bd_base_info, ol_info);
3198 if (unlikely(ret)) {
3199 u64_stats_update_begin(&ring->syncp);
3200 ring->stats.rx_err_cnt++;
3201 u64_stats_update_end(&ring->syncp);
3205 l2_frame_type = hnae3_get_field(l234info, HNS3_RXD_DMAC_M,
3208 u64_stats_update_begin(&ring->syncp);
3209 ring->stats.rx_pkts++;
3210 ring->stats.rx_bytes += len;
3212 if (l2_frame_type == HNS3_L2_TYPE_MULTICAST)
3213 ring->stats.rx_multicast++;
3215 u64_stats_update_end(&ring->syncp);
3217 ring->tqp_vector->rx_group.total_bytes += len;
3219 hns3_set_rx_skb_rss_type(ring, skb, le32_to_cpu(desc->rx.rss_hash));
3223 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring)
3225 struct sk_buff *skb = ring->skb;
3226 struct hns3_desc_cb *desc_cb;
3227 struct hns3_desc *desc;
3228 unsigned int length;
3232 desc = &ring->desc[ring->next_to_clean];
3233 desc_cb = &ring->desc_cb[ring->next_to_clean];
3238 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
3240 /* Check valid BD */
3241 if (unlikely(!(bd_base_info & BIT(HNS3_RXD_VLD_B))))
3245 length = le16_to_cpu(desc->rx.size);
3247 ring->va = desc_cb->buf + desc_cb->page_offset;
3249 dma_sync_single_for_cpu(ring_to_dev(ring),
3250 desc_cb->dma + desc_cb->page_offset,
3251 hns3_buf_size(ring),
3254 /* Prefetch first cache line of first page.
3255 * Idea is to cache few bytes of the header of the packet.
3256 * Our L1 Cache line size is 64B so need to prefetch twice to make
3257 * it 128B. But in actual we can have greater size of caches with
3258 * 128B Level 1 cache lines. In such a case, single fetch would
3259 * suffice to cache in the relevant part of the header.
3261 net_prefetch(ring->va);
3263 ret = hns3_alloc_skb(ring, length, ring->va);
3266 if (ret < 0) /* alloc buffer fail */
3268 if (!(bd_base_info & BIT(HNS3_RXD_FE_B))) { /* need add frag */
3269 ret = hns3_add_frag(ring);
3274 ret = hns3_add_frag(ring);
3279 /* As the head data may be changed when GRO enable, copy
3280 * the head data in after other data rx completed
3282 if (skb->len > HNS3_RX_HEAD_SIZE)
3283 memcpy(skb->data, ring->va,
3284 ALIGN(ring->pull_len, sizeof(long)));
3286 ret = hns3_handle_bdinfo(ring, skb);
3287 if (unlikely(ret)) {
3288 dev_kfree_skb_any(skb);
3292 skb_record_rx_queue(skb, ring->tqp->tqp_index);
3296 int hns3_clean_rx_ring(struct hns3_enet_ring *ring, int budget,
3297 void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
3299 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
3300 int unused_count = hns3_desc_unused(ring);
3304 unused_count -= ring->pending_buf;
3306 while (recv_pkts < budget) {
3307 /* Reuse or realloc buffers */
3308 if (unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
3309 hns3_nic_alloc_rx_buffers(ring, unused_count);
3310 unused_count = hns3_desc_unused(ring) -
3315 err = hns3_handle_rx_bd(ring);
3316 /* Do not get FE for the packet or failed to alloc skb */
3317 if (unlikely(!ring->skb || err == -ENXIO)) {
3319 } else if (likely(!err)) {
3320 rx_fn(ring, ring->skb);
3324 unused_count += ring->pending_buf;
3326 ring->pending_buf = 0;
3330 /* Make all data has been write before submit */
3331 if (unused_count > 0)
3332 hns3_nic_alloc_rx_buffers(ring, unused_count);
3337 static bool hns3_get_new_flow_lvl(struct hns3_enet_ring_group *ring_group)
3339 #define HNS3_RX_LOW_BYTE_RATE 10000
3340 #define HNS3_RX_MID_BYTE_RATE 20000
3341 #define HNS3_RX_ULTRA_PACKET_RATE 40
3343 enum hns3_flow_level_range new_flow_level;
3344 struct hns3_enet_tqp_vector *tqp_vector;
3345 int packets_per_msecs, bytes_per_msecs;
3348 tqp_vector = ring_group->ring->tqp_vector;
3350 jiffies_to_msecs(jiffies - tqp_vector->last_jiffies);
3351 if (!time_passed_ms)
3354 do_div(ring_group->total_packets, time_passed_ms);
3355 packets_per_msecs = ring_group->total_packets;
3357 do_div(ring_group->total_bytes, time_passed_ms);
3358 bytes_per_msecs = ring_group->total_bytes;
3360 new_flow_level = ring_group->coal.flow_level;
3362 /* Simple throttlerate management
3363 * 0-10MB/s lower (50000 ints/s)
3364 * 10-20MB/s middle (20000 ints/s)
3365 * 20-1249MB/s high (18000 ints/s)
3366 * > 40000pps ultra (8000 ints/s)
3368 switch (new_flow_level) {
3370 if (bytes_per_msecs > HNS3_RX_LOW_BYTE_RATE)
3371 new_flow_level = HNS3_FLOW_MID;
3374 if (bytes_per_msecs > HNS3_RX_MID_BYTE_RATE)
3375 new_flow_level = HNS3_FLOW_HIGH;
3376 else if (bytes_per_msecs <= HNS3_RX_LOW_BYTE_RATE)
3377 new_flow_level = HNS3_FLOW_LOW;
3379 case HNS3_FLOW_HIGH:
3380 case HNS3_FLOW_ULTRA:
3382 if (bytes_per_msecs <= HNS3_RX_MID_BYTE_RATE)
3383 new_flow_level = HNS3_FLOW_MID;
3387 if (packets_per_msecs > HNS3_RX_ULTRA_PACKET_RATE &&
3388 &tqp_vector->rx_group == ring_group)
3389 new_flow_level = HNS3_FLOW_ULTRA;
3391 ring_group->total_bytes = 0;
3392 ring_group->total_packets = 0;
3393 ring_group->coal.flow_level = new_flow_level;
3398 static bool hns3_get_new_int_gl(struct hns3_enet_ring_group *ring_group)
3400 struct hns3_enet_tqp_vector *tqp_vector;
3403 if (!ring_group->ring)
3406 tqp_vector = ring_group->ring->tqp_vector;
3407 if (!tqp_vector->last_jiffies)
3410 if (ring_group->total_packets == 0) {
3411 ring_group->coal.int_gl = HNS3_INT_GL_50K;
3412 ring_group->coal.flow_level = HNS3_FLOW_LOW;
3416 if (!hns3_get_new_flow_lvl(ring_group))
3419 new_int_gl = ring_group->coal.int_gl;
3420 switch (ring_group->coal.flow_level) {
3422 new_int_gl = HNS3_INT_GL_50K;
3425 new_int_gl = HNS3_INT_GL_20K;
3427 case HNS3_FLOW_HIGH:
3428 new_int_gl = HNS3_INT_GL_18K;
3430 case HNS3_FLOW_ULTRA:
3431 new_int_gl = HNS3_INT_GL_8K;
3437 if (new_int_gl != ring_group->coal.int_gl) {
3438 ring_group->coal.int_gl = new_int_gl;
3444 static void hns3_update_new_int_gl(struct hns3_enet_tqp_vector *tqp_vector)
3446 struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
3447 struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
3448 bool rx_update, tx_update;
3450 /* update param every 1000ms */
3451 if (time_before(jiffies,
3452 tqp_vector->last_jiffies + msecs_to_jiffies(1000)))
3455 if (rx_group->coal.adapt_enable) {
3456 rx_update = hns3_get_new_int_gl(rx_group);
3458 hns3_set_vector_coalesce_rx_gl(tqp_vector,
3459 rx_group->coal.int_gl);
3462 if (tx_group->coal.adapt_enable) {
3463 tx_update = hns3_get_new_int_gl(tx_group);
3465 hns3_set_vector_coalesce_tx_gl(tqp_vector,
3466 tx_group->coal.int_gl);
3469 tqp_vector->last_jiffies = jiffies;
3472 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
3474 struct hns3_nic_priv *priv = netdev_priv(napi->dev);
3475 struct hns3_enet_ring *ring;
3476 int rx_pkt_total = 0;
3478 struct hns3_enet_tqp_vector *tqp_vector =
3479 container_of(napi, struct hns3_enet_tqp_vector, napi);
3480 bool clean_complete = true;
3481 int rx_budget = budget;
3483 if (unlikely(test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
3484 napi_complete(napi);
3488 /* Since the actual Tx work is minimal, we can give the Tx a larger
3489 * budget and be more aggressive about cleaning up the Tx descriptors.
3491 hns3_for_each_ring(ring, tqp_vector->tx_group)
3492 hns3_clean_tx_ring(ring, budget);
3494 /* make sure rx ring budget not smaller than 1 */
3495 if (tqp_vector->num_tqps > 1)
3496 rx_budget = max(budget / tqp_vector->num_tqps, 1);
3498 hns3_for_each_ring(ring, tqp_vector->rx_group) {
3499 int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
3502 if (rx_cleaned >= rx_budget)
3503 clean_complete = false;
3505 rx_pkt_total += rx_cleaned;
3508 tqp_vector->rx_group.total_packets += rx_pkt_total;
3510 if (!clean_complete)
3513 if (napi_complete(napi) &&
3514 likely(!test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
3515 hns3_update_new_int_gl(tqp_vector);
3516 hns3_mask_vector_irq(tqp_vector, 1);
3519 return rx_pkt_total;
3522 static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
3523 struct hnae3_ring_chain_node *head)
3525 struct pci_dev *pdev = tqp_vector->handle->pdev;
3526 struct hnae3_ring_chain_node *cur_chain = head;
3527 struct hnae3_ring_chain_node *chain;
3528 struct hns3_enet_ring *tx_ring;
3529 struct hns3_enet_ring *rx_ring;
3531 tx_ring = tqp_vector->tx_group.ring;
3533 cur_chain->tqp_index = tx_ring->tqp->tqp_index;
3534 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
3535 HNAE3_RING_TYPE_TX);
3536 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3537 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_TX);
3539 cur_chain->next = NULL;
3541 while (tx_ring->next) {
3542 tx_ring = tx_ring->next;
3544 chain = devm_kzalloc(&pdev->dev, sizeof(*chain),
3547 goto err_free_chain;
3549 cur_chain->next = chain;
3550 chain->tqp_index = tx_ring->tqp->tqp_index;
3551 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
3552 HNAE3_RING_TYPE_TX);
3553 hnae3_set_field(chain->int_gl_idx,
3554 HNAE3_RING_GL_IDX_M,
3555 HNAE3_RING_GL_IDX_S,
3562 rx_ring = tqp_vector->rx_group.ring;
3563 if (!tx_ring && rx_ring) {
3564 cur_chain->next = NULL;
3565 cur_chain->tqp_index = rx_ring->tqp->tqp_index;
3566 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
3567 HNAE3_RING_TYPE_RX);
3568 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3569 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
3571 rx_ring = rx_ring->next;
3575 chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
3577 goto err_free_chain;
3579 cur_chain->next = chain;
3580 chain->tqp_index = rx_ring->tqp->tqp_index;
3581 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
3582 HNAE3_RING_TYPE_RX);
3583 hnae3_set_field(chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3584 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
3588 rx_ring = rx_ring->next;
3594 cur_chain = head->next;
3596 chain = cur_chain->next;
3597 devm_kfree(&pdev->dev, cur_chain);
3605 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
3606 struct hnae3_ring_chain_node *head)
3608 struct pci_dev *pdev = tqp_vector->handle->pdev;
3609 struct hnae3_ring_chain_node *chain_tmp, *chain;
3614 chain_tmp = chain->next;
3615 devm_kfree(&pdev->dev, chain);
3620 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
3621 struct hns3_enet_ring *ring)
3623 ring->next = group->ring;
3629 static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
3631 struct pci_dev *pdev = priv->ae_handle->pdev;
3632 struct hns3_enet_tqp_vector *tqp_vector;
3633 int num_vectors = priv->vector_num;
3637 numa_node = dev_to_node(&pdev->dev);
3639 for (vector_i = 0; vector_i < num_vectors; vector_i++) {
3640 tqp_vector = &priv->tqp_vector[vector_i];
3641 cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
3642 &tqp_vector->affinity_mask);
3646 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
3648 struct hnae3_ring_chain_node vector_ring_chain;
3649 struct hnae3_handle *h = priv->ae_handle;
3650 struct hns3_enet_tqp_vector *tqp_vector;
3654 hns3_nic_set_cpumask(priv);
3656 for (i = 0; i < priv->vector_num; i++) {
3657 tqp_vector = &priv->tqp_vector[i];
3658 hns3_vector_coalesce_init_hw(tqp_vector, priv);
3659 tqp_vector->num_tqps = 0;
3662 for (i = 0; i < h->kinfo.num_tqps; i++) {
3663 u16 vector_i = i % priv->vector_num;
3664 u16 tqp_num = h->kinfo.num_tqps;
3666 tqp_vector = &priv->tqp_vector[vector_i];
3668 hns3_add_ring_to_group(&tqp_vector->tx_group,
3671 hns3_add_ring_to_group(&tqp_vector->rx_group,
3672 &priv->ring[i + tqp_num]);
3674 priv->ring[i].tqp_vector = tqp_vector;
3675 priv->ring[i + tqp_num].tqp_vector = tqp_vector;
3676 tqp_vector->num_tqps++;
3679 for (i = 0; i < priv->vector_num; i++) {
3680 tqp_vector = &priv->tqp_vector[i];
3682 tqp_vector->rx_group.total_bytes = 0;
3683 tqp_vector->rx_group.total_packets = 0;
3684 tqp_vector->tx_group.total_bytes = 0;
3685 tqp_vector->tx_group.total_packets = 0;
3686 tqp_vector->handle = h;
3688 ret = hns3_get_vector_ring_chain(tqp_vector,
3689 &vector_ring_chain);
3693 ret = h->ae_algo->ops->map_ring_to_vector(h,
3694 tqp_vector->vector_irq, &vector_ring_chain);
3696 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3701 netif_napi_add(priv->netdev, &tqp_vector->napi,
3702 hns3_nic_common_poll, NAPI_POLL_WEIGHT);
3709 netif_napi_del(&priv->tqp_vector[i].napi);
3714 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
3716 struct hnae3_handle *h = priv->ae_handle;
3717 struct hns3_enet_tqp_vector *tqp_vector;
3718 struct hnae3_vector_info *vector;
3719 struct pci_dev *pdev = h->pdev;
3720 u16 tqp_num = h->kinfo.num_tqps;
3725 /* RSS size, cpu online and vector_num should be the same */
3726 /* Should consider 2p/4p later */
3727 vector_num = min_t(u16, num_online_cpus(), tqp_num);
3729 vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
3734 /* save the actual available vector number */
3735 vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
3737 priv->vector_num = vector_num;
3738 priv->tqp_vector = (struct hns3_enet_tqp_vector *)
3739 devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
3741 if (!priv->tqp_vector) {
3746 for (i = 0; i < priv->vector_num; i++) {
3747 tqp_vector = &priv->tqp_vector[i];
3748 tqp_vector->idx = i;
3749 tqp_vector->mask_addr = vector[i].io_addr;
3750 tqp_vector->vector_irq = vector[i].vector;
3751 hns3_vector_coalesce_init(tqp_vector, priv);
3755 devm_kfree(&pdev->dev, vector);
3759 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
3765 static void hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
3767 struct hnae3_ring_chain_node vector_ring_chain;
3768 struct hnae3_handle *h = priv->ae_handle;
3769 struct hns3_enet_tqp_vector *tqp_vector;
3772 for (i = 0; i < priv->vector_num; i++) {
3773 tqp_vector = &priv->tqp_vector[i];
3775 if (!tqp_vector->rx_group.ring && !tqp_vector->tx_group.ring)
3778 /* Since the mapping can be overwritten, when fail to get the
3779 * chain between vector and ring, we should go on to deal with
3780 * the remaining options.
3782 if (hns3_get_vector_ring_chain(tqp_vector, &vector_ring_chain))
3783 dev_warn(priv->dev, "failed to get ring chain\n");
3785 h->ae_algo->ops->unmap_ring_from_vector(h,
3786 tqp_vector->vector_irq, &vector_ring_chain);
3788 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3790 hns3_clear_ring_group(&tqp_vector->rx_group);
3791 hns3_clear_ring_group(&tqp_vector->tx_group);
3792 netif_napi_del(&priv->tqp_vector[i].napi);
3796 static void hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
3798 struct hnae3_handle *h = priv->ae_handle;
3799 struct pci_dev *pdev = h->pdev;
3802 for (i = 0; i < priv->vector_num; i++) {
3803 struct hns3_enet_tqp_vector *tqp_vector;
3805 tqp_vector = &priv->tqp_vector[i];
3806 ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
3811 devm_kfree(&pdev->dev, priv->tqp_vector);
3814 static void hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
3815 unsigned int ring_type)
3817 int queue_num = priv->ae_handle->kinfo.num_tqps;
3818 struct hns3_enet_ring *ring;
3821 if (ring_type == HNAE3_RING_TYPE_TX) {
3822 ring = &priv->ring[q->tqp_index];
3823 desc_num = priv->ae_handle->kinfo.num_tx_desc;
3824 ring->queue_index = q->tqp_index;
3826 ring = &priv->ring[q->tqp_index + queue_num];
3827 desc_num = priv->ae_handle->kinfo.num_rx_desc;
3828 ring->queue_index = q->tqp_index;
3831 hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
3835 ring->desc_cb = NULL;
3836 ring->dev = priv->dev;
3837 ring->desc_dma_addr = 0;
3838 ring->buf_size = q->buf_size;
3839 ring->desc_num = desc_num;
3840 ring->next_to_use = 0;
3841 ring->next_to_clean = 0;
3842 ring->last_to_use = 0;
3845 static void hns3_queue_to_ring(struct hnae3_queue *tqp,
3846 struct hns3_nic_priv *priv)
3848 hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
3849 hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
3852 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
3854 struct hnae3_handle *h = priv->ae_handle;
3855 struct pci_dev *pdev = h->pdev;
3858 priv->ring = devm_kzalloc(&pdev->dev,
3859 array3_size(h->kinfo.num_tqps,
3860 sizeof(*priv->ring), 2),
3865 for (i = 0; i < h->kinfo.num_tqps; i++)
3866 hns3_queue_to_ring(h->kinfo.tqp[i], priv);
3871 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
3876 devm_kfree(priv->dev, priv->ring);
3880 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
3884 if (ring->desc_num <= 0 || ring->buf_size <= 0)
3887 ring->desc_cb = devm_kcalloc(ring_to_dev(ring), ring->desc_num,
3888 sizeof(ring->desc_cb[0]), GFP_KERNEL);
3889 if (!ring->desc_cb) {
3894 ret = hns3_alloc_desc(ring);
3896 goto out_with_desc_cb;
3898 if (!HNAE3_IS_TX_RING(ring)) {
3899 ret = hns3_alloc_ring_buffers(ring);
3907 hns3_free_desc(ring);
3909 devm_kfree(ring_to_dev(ring), ring->desc_cb);
3910 ring->desc_cb = NULL;
3915 void hns3_fini_ring(struct hns3_enet_ring *ring)
3917 hns3_free_desc(ring);
3918 devm_kfree(ring_to_dev(ring), ring->desc_cb);
3919 ring->desc_cb = NULL;
3920 ring->next_to_clean = 0;
3921 ring->next_to_use = 0;
3922 ring->last_to_use = 0;
3923 ring->pending_buf = 0;
3925 dev_kfree_skb_any(ring->skb);
3930 static int hns3_buf_size2type(u32 buf_size)
3936 bd_size_type = HNS3_BD_SIZE_512_TYPE;
3939 bd_size_type = HNS3_BD_SIZE_1024_TYPE;
3942 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
3945 bd_size_type = HNS3_BD_SIZE_4096_TYPE;
3948 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
3951 return bd_size_type;
3954 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
3956 dma_addr_t dma = ring->desc_dma_addr;
3957 struct hnae3_queue *q = ring->tqp;
3959 if (!HNAE3_IS_TX_RING(ring)) {
3960 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG, (u32)dma);
3961 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
3962 (u32)((dma >> 31) >> 1));
3964 hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
3965 hns3_buf_size2type(ring->buf_size));
3966 hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
3967 ring->desc_num / 8 - 1);
3970 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
3972 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
3973 (u32)((dma >> 31) >> 1));
3975 hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
3976 ring->desc_num / 8 - 1);
3980 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
3982 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
3985 for (i = 0; i < HNAE3_MAX_TC; i++) {
3986 struct hnae3_tc_info *tc_info = &kinfo->tc_info[i];
3989 if (!tc_info->enable)
3992 for (j = 0; j < tc_info->tqp_count; j++) {
3993 struct hnae3_queue *q;
3995 q = priv->ring[tc_info->tqp_offset + j].tqp;
3996 hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG,
4002 int hns3_init_all_ring(struct hns3_nic_priv *priv)
4004 struct hnae3_handle *h = priv->ae_handle;
4005 int ring_num = h->kinfo.num_tqps * 2;
4009 for (i = 0; i < ring_num; i++) {
4010 ret = hns3_alloc_ring_memory(&priv->ring[i]);
4013 "Alloc ring memory fail! ret=%d\n", ret);
4014 goto out_when_alloc_ring_memory;
4017 u64_stats_init(&priv->ring[i].syncp);
4022 out_when_alloc_ring_memory:
4023 for (j = i - 1; j >= 0; j--)
4024 hns3_fini_ring(&priv->ring[j]);
4029 int hns3_uninit_all_ring(struct hns3_nic_priv *priv)
4031 struct hnae3_handle *h = priv->ae_handle;
4034 for (i = 0; i < h->kinfo.num_tqps; i++) {
4035 hns3_fini_ring(&priv->ring[i]);
4036 hns3_fini_ring(&priv->ring[i + h->kinfo.num_tqps]);
4041 /* Set mac addr if it is configured. or leave it to the AE driver */
4042 static int hns3_init_mac_addr(struct net_device *netdev)
4044 struct hns3_nic_priv *priv = netdev_priv(netdev);
4045 struct hnae3_handle *h = priv->ae_handle;
4046 u8 mac_addr_temp[ETH_ALEN];
4049 if (h->ae_algo->ops->get_mac_addr)
4050 h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
4052 /* Check if the MAC address is valid, if not get a random one */
4053 if (!is_valid_ether_addr(mac_addr_temp)) {
4054 eth_hw_addr_random(netdev);
4055 dev_warn(priv->dev, "using random MAC address %pM\n",
4057 } else if (!ether_addr_equal(netdev->dev_addr, mac_addr_temp)) {
4058 ether_addr_copy(netdev->dev_addr, mac_addr_temp);
4059 ether_addr_copy(netdev->perm_addr, mac_addr_temp);
4064 if (h->ae_algo->ops->set_mac_addr)
4065 ret = h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
4070 static int hns3_init_phy(struct net_device *netdev)
4072 struct hnae3_handle *h = hns3_get_handle(netdev);
4075 if (h->ae_algo->ops->mac_connect_phy)
4076 ret = h->ae_algo->ops->mac_connect_phy(h);
4081 static void hns3_uninit_phy(struct net_device *netdev)
4083 struct hnae3_handle *h = hns3_get_handle(netdev);
4085 if (h->ae_algo->ops->mac_disconnect_phy)
4086 h->ae_algo->ops->mac_disconnect_phy(h);
4089 static void hns3_del_all_fd_rules(struct net_device *netdev, bool clear_list)
4091 struct hnae3_handle *h = hns3_get_handle(netdev);
4093 if (h->ae_algo->ops->del_all_fd_entries)
4094 h->ae_algo->ops->del_all_fd_entries(h, clear_list);
4097 static int hns3_client_start(struct hnae3_handle *handle)
4099 if (!handle->ae_algo->ops->client_start)
4102 return handle->ae_algo->ops->client_start(handle);
4105 static void hns3_client_stop(struct hnae3_handle *handle)
4107 if (!handle->ae_algo->ops->client_stop)
4110 handle->ae_algo->ops->client_stop(handle);
4113 static void hns3_info_show(struct hns3_nic_priv *priv)
4115 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
4117 dev_info(priv->dev, "MAC address: %pM\n", priv->netdev->dev_addr);
4118 dev_info(priv->dev, "Task queue pairs numbers: %u\n", kinfo->num_tqps);
4119 dev_info(priv->dev, "RSS size: %u\n", kinfo->rss_size);
4120 dev_info(priv->dev, "Allocated RSS size: %u\n", kinfo->req_rss_size);
4121 dev_info(priv->dev, "RX buffer length: %u\n", kinfo->rx_buf_len);
4122 dev_info(priv->dev, "Desc num per TX queue: %u\n", kinfo->num_tx_desc);
4123 dev_info(priv->dev, "Desc num per RX queue: %u\n", kinfo->num_rx_desc);
4124 dev_info(priv->dev, "Total number of enabled TCs: %u\n", kinfo->num_tc);
4125 dev_info(priv->dev, "Max mtu size: %u\n", priv->netdev->max_mtu);
4128 static int hns3_client_init(struct hnae3_handle *handle)
4130 struct pci_dev *pdev = handle->pdev;
4131 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
4132 u16 alloc_tqps, max_rss_size;
4133 struct hns3_nic_priv *priv;
4134 struct net_device *netdev;
4137 handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
4139 netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
4143 priv = netdev_priv(netdev);
4144 priv->dev = &pdev->dev;
4145 priv->netdev = netdev;
4146 priv->ae_handle = handle;
4147 priv->tx_timeout_count = 0;
4148 priv->max_non_tso_bd_num = ae_dev->dev_specs.max_non_tso_bd_num;
4149 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
4151 handle->msg_enable = netif_msg_init(debug, DEFAULT_MSG_LEVEL);
4153 handle->kinfo.netdev = netdev;
4154 handle->priv = (void *)priv;
4156 hns3_init_mac_addr(netdev);
4158 hns3_set_default_feature(netdev);
4160 netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
4161 netdev->priv_flags |= IFF_UNICAST_FLT;
4162 netdev->netdev_ops = &hns3_nic_netdev_ops;
4163 SET_NETDEV_DEV(netdev, &pdev->dev);
4164 hns3_ethtool_set_ops(netdev);
4166 /* Carrier off reporting is important to ethtool even BEFORE open */
4167 netif_carrier_off(netdev);
4169 ret = hns3_get_ring_config(priv);
4172 goto out_get_ring_cfg;
4175 ret = hns3_nic_alloc_vector_data(priv);
4178 goto out_alloc_vector_data;
4181 ret = hns3_nic_init_vector_data(priv);
4184 goto out_init_vector_data;
4187 ret = hns3_init_all_ring(priv);
4193 ret = hns3_init_phy(netdev);
4197 ret = register_netdev(netdev);
4199 dev_err(priv->dev, "probe register netdev fail!\n");
4200 goto out_reg_netdev_fail;
4203 /* the device can work without cpu rmap, only aRFS needs it */
4204 ret = hns3_set_rx_cpu_rmap(netdev);
4206 dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
4208 ret = hns3_nic_init_irq(priv);
4210 dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
4211 hns3_free_rx_cpu_rmap(netdev);
4212 goto out_init_irq_fail;
4215 ret = hns3_client_start(handle);
4217 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
4218 goto out_client_start;
4221 hns3_dcbnl_setup(handle);
4223 hns3_dbg_init(handle);
4225 /* MTU range: (ETH_MIN_MTU(kernel default) - 9702) */
4226 netdev->max_mtu = HNS3_MAX_MTU;
4228 if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps))
4229 set_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state);
4231 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
4233 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
4234 set_bit(HNAE3_PFLAG_LIMIT_PROMISC, &handle->supported_pflags);
4236 if (netif_msg_drv(handle))
4237 hns3_info_show(priv);
4242 hns3_free_rx_cpu_rmap(netdev);
4243 hns3_nic_uninit_irq(priv);
4245 unregister_netdev(netdev);
4246 out_reg_netdev_fail:
4247 hns3_uninit_phy(netdev);
4249 hns3_uninit_all_ring(priv);
4251 hns3_nic_uninit_vector_data(priv);
4252 out_init_vector_data:
4253 hns3_nic_dealloc_vector_data(priv);
4254 out_alloc_vector_data:
4257 priv->ae_handle = NULL;
4258 free_netdev(netdev);
4262 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
4264 struct net_device *netdev = handle->kinfo.netdev;
4265 struct hns3_nic_priv *priv = netdev_priv(netdev);
4268 if (netdev->reg_state != NETREG_UNINITIALIZED)
4269 unregister_netdev(netdev);
4271 hns3_client_stop(handle);
4273 hns3_uninit_phy(netdev);
4275 if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
4276 netdev_warn(netdev, "already uninitialized\n");
4277 goto out_netdev_free;
4280 hns3_free_rx_cpu_rmap(netdev);
4282 hns3_nic_uninit_irq(priv);
4284 hns3_del_all_fd_rules(netdev, true);
4286 hns3_clear_all_ring(handle, true);
4288 hns3_nic_uninit_vector_data(priv);
4290 hns3_nic_dealloc_vector_data(priv);
4292 ret = hns3_uninit_all_ring(priv);
4294 netdev_err(netdev, "uninit ring error\n");
4296 hns3_put_ring_config(priv);
4299 hns3_dbg_uninit(handle);
4300 free_netdev(netdev);
4303 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
4305 struct net_device *netdev = handle->kinfo.netdev;
4311 netif_tx_wake_all_queues(netdev);
4312 netif_carrier_on(netdev);
4313 if (netif_msg_link(handle))
4314 netdev_info(netdev, "link up\n");
4316 netif_carrier_off(netdev);
4317 netif_tx_stop_all_queues(netdev);
4318 if (netif_msg_link(handle))
4319 netdev_info(netdev, "link down\n");
4323 static int hns3_client_setup_tc(struct hnae3_handle *handle, u8 tc)
4325 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4326 struct net_device *ndev = kinfo->netdev;
4328 if (tc > HNAE3_MAX_TC)
4334 return hns3_nic_set_real_num_queue(ndev);
4337 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
4339 while (ring->next_to_clean != ring->next_to_use) {
4340 ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
4341 hns3_free_buffer_detach(ring, ring->next_to_clean, 0);
4342 ring_ptr_move_fw(ring, next_to_clean);
4345 ring->pending_buf = 0;
4348 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
4350 struct hns3_desc_cb res_cbs;
4353 while (ring->next_to_use != ring->next_to_clean) {
4354 /* When a buffer is not reused, it's memory has been
4355 * freed in hns3_handle_rx_bd or will be freed by
4356 * stack, so we need to replace the buffer here.
4358 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
4359 ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
4361 u64_stats_update_begin(&ring->syncp);
4362 ring->stats.sw_err_cnt++;
4363 u64_stats_update_end(&ring->syncp);
4364 /* if alloc new buffer fail, exit directly
4365 * and reclear in up flow.
4367 netdev_warn(ring_to_netdev(ring),
4368 "reserve buffer map failed, ret = %d\n",
4372 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
4374 ring_ptr_move_fw(ring, next_to_use);
4377 /* Free the pending skb in rx ring */
4379 dev_kfree_skb_any(ring->skb);
4381 ring->pending_buf = 0;
4387 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
4389 while (ring->next_to_use != ring->next_to_clean) {
4390 /* When a buffer is not reused, it's memory has been
4391 * freed in hns3_handle_rx_bd or will be freed by
4392 * stack, so only need to unmap the buffer here.
4394 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
4395 hns3_unmap_buffer(ring,
4396 &ring->desc_cb[ring->next_to_use]);
4397 ring->desc_cb[ring->next_to_use].dma = 0;
4400 ring_ptr_move_fw(ring, next_to_use);
4404 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force)
4406 struct net_device *ndev = h->kinfo.netdev;
4407 struct hns3_nic_priv *priv = netdev_priv(ndev);
4410 for (i = 0; i < h->kinfo.num_tqps; i++) {
4411 struct hns3_enet_ring *ring;
4413 ring = &priv->ring[i];
4414 hns3_clear_tx_ring(ring);
4416 ring = &priv->ring[i + h->kinfo.num_tqps];
4417 /* Continue to clear other rings even if clearing some
4421 hns3_force_clear_rx_ring(ring);
4423 hns3_clear_rx_ring(ring);
4427 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
4429 struct net_device *ndev = h->kinfo.netdev;
4430 struct hns3_nic_priv *priv = netdev_priv(ndev);
4431 struct hns3_enet_ring *rx_ring;
4435 for (i = 0; i < h->kinfo.num_tqps; i++) {
4436 ret = h->ae_algo->ops->reset_queue(h, i);
4440 hns3_init_ring_hw(&priv->ring[i]);
4442 /* We need to clear tx ring here because self test will
4443 * use the ring and will not run down before up
4445 hns3_clear_tx_ring(&priv->ring[i]);
4446 priv->ring[i].next_to_clean = 0;
4447 priv->ring[i].next_to_use = 0;
4448 priv->ring[i].last_to_use = 0;
4450 rx_ring = &priv->ring[i + h->kinfo.num_tqps];
4451 hns3_init_ring_hw(rx_ring);
4452 ret = hns3_clear_rx_ring(rx_ring);
4456 /* We can not know the hardware head and tail when this
4457 * function is called in reset flow, so we reuse all desc.
4459 for (j = 0; j < rx_ring->desc_num; j++)
4460 hns3_reuse_buffer(rx_ring, j);
4462 rx_ring->next_to_clean = 0;
4463 rx_ring->next_to_use = 0;
4466 hns3_init_tx_ring_tc(priv);
4471 static void hns3_store_coal(struct hns3_nic_priv *priv)
4473 /* ethtool only support setting and querying one coal
4474 * configuration for now, so save the vector 0' coal
4475 * configuration here in order to restore it.
4477 memcpy(&priv->tx_coal, &priv->tqp_vector[0].tx_group.coal,
4478 sizeof(struct hns3_enet_coalesce));
4479 memcpy(&priv->rx_coal, &priv->tqp_vector[0].rx_group.coal,
4480 sizeof(struct hns3_enet_coalesce));
4483 static void hns3_restore_coal(struct hns3_nic_priv *priv)
4485 u16 vector_num = priv->vector_num;
4488 for (i = 0; i < vector_num; i++) {
4489 memcpy(&priv->tqp_vector[i].tx_group.coal, &priv->tx_coal,
4490 sizeof(struct hns3_enet_coalesce));
4491 memcpy(&priv->tqp_vector[i].rx_group.coal, &priv->rx_coal,
4492 sizeof(struct hns3_enet_coalesce));
4496 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
4498 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4499 struct net_device *ndev = kinfo->netdev;
4500 struct hns3_nic_priv *priv = netdev_priv(ndev);
4502 if (test_and_set_bit(HNS3_NIC_STATE_RESETTING, &priv->state))
4505 if (!netif_running(ndev))
4508 return hns3_nic_net_stop(ndev);
4511 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
4513 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4514 struct hns3_nic_priv *priv = netdev_priv(kinfo->netdev);
4517 clear_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4519 if (netif_running(kinfo->netdev)) {
4520 ret = hns3_nic_net_open(kinfo->netdev);
4522 set_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4523 netdev_err(kinfo->netdev,
4524 "net up fail, ret=%d!\n", ret);
4532 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
4534 struct net_device *netdev = handle->kinfo.netdev;
4535 struct hns3_nic_priv *priv = netdev_priv(netdev);
4538 /* Carrier off reporting is important to ethtool even BEFORE open */
4539 netif_carrier_off(netdev);
4541 ret = hns3_get_ring_config(priv);
4545 ret = hns3_nic_alloc_vector_data(priv);
4549 hns3_restore_coal(priv);
4551 ret = hns3_nic_init_vector_data(priv);
4553 goto err_dealloc_vector;
4555 ret = hns3_init_all_ring(priv);
4557 goto err_uninit_vector;
4559 /* the device can work without cpu rmap, only aRFS needs it */
4560 ret = hns3_set_rx_cpu_rmap(netdev);
4562 dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
4564 ret = hns3_nic_init_irq(priv);
4566 dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
4567 hns3_free_rx_cpu_rmap(netdev);
4568 goto err_init_irq_fail;
4571 if (!hns3_is_phys_func(handle->pdev))
4572 hns3_init_mac_addr(netdev);
4574 ret = hns3_client_start(handle);
4576 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
4577 goto err_client_start_fail;
4580 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
4584 err_client_start_fail:
4585 hns3_free_rx_cpu_rmap(netdev);
4586 hns3_nic_uninit_irq(priv);
4588 hns3_uninit_all_ring(priv);
4590 hns3_nic_uninit_vector_data(priv);
4592 hns3_nic_dealloc_vector_data(priv);
4594 hns3_put_ring_config(priv);
4599 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
4601 struct net_device *netdev = handle->kinfo.netdev;
4602 struct hns3_nic_priv *priv = netdev_priv(netdev);
4605 if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
4606 netdev_warn(netdev, "already uninitialized\n");
4610 hns3_free_rx_cpu_rmap(netdev);
4611 hns3_nic_uninit_irq(priv);
4612 hns3_clear_all_ring(handle, true);
4613 hns3_reset_tx_queue(priv->ae_handle);
4615 hns3_nic_uninit_vector_data(priv);
4617 hns3_store_coal(priv);
4619 hns3_nic_dealloc_vector_data(priv);
4621 ret = hns3_uninit_all_ring(priv);
4623 netdev_err(netdev, "uninit ring error\n");
4625 hns3_put_ring_config(priv);
4630 static int hns3_reset_notify(struct hnae3_handle *handle,
4631 enum hnae3_reset_notify_type type)
4636 case HNAE3_UP_CLIENT:
4637 ret = hns3_reset_notify_up_enet(handle);
4639 case HNAE3_DOWN_CLIENT:
4640 ret = hns3_reset_notify_down_enet(handle);
4642 case HNAE3_INIT_CLIENT:
4643 ret = hns3_reset_notify_init_enet(handle);
4645 case HNAE3_UNINIT_CLIENT:
4646 ret = hns3_reset_notify_uninit_enet(handle);
4655 static int hns3_change_channels(struct hnae3_handle *handle, u32 new_tqp_num,
4656 bool rxfh_configured)
4660 ret = handle->ae_algo->ops->set_channels(handle, new_tqp_num,
4663 dev_err(&handle->pdev->dev,
4664 "Change tqp num(%u) fail.\n", new_tqp_num);
4668 ret = hns3_reset_notify(handle, HNAE3_INIT_CLIENT);
4672 ret = hns3_reset_notify(handle, HNAE3_UP_CLIENT);
4674 hns3_reset_notify(handle, HNAE3_UNINIT_CLIENT);
4679 int hns3_set_channels(struct net_device *netdev,
4680 struct ethtool_channels *ch)
4682 struct hnae3_handle *h = hns3_get_handle(netdev);
4683 struct hnae3_knic_private_info *kinfo = &h->kinfo;
4684 bool rxfh_configured = netif_is_rxfh_configured(netdev);
4685 u32 new_tqp_num = ch->combined_count;
4689 if (hns3_nic_resetting(netdev))
4692 if (ch->rx_count || ch->tx_count)
4695 if (new_tqp_num > hns3_get_max_available_channels(h) ||
4697 dev_err(&netdev->dev,
4698 "Change tqps fail, the tqp range is from 1 to %u",
4699 hns3_get_max_available_channels(h));
4703 if (kinfo->rss_size == new_tqp_num)
4706 netif_dbg(h, drv, netdev,
4707 "set channels: tqp_num=%u, rxfh=%d\n",
4708 new_tqp_num, rxfh_configured);
4710 ret = hns3_reset_notify(h, HNAE3_DOWN_CLIENT);
4714 ret = hns3_reset_notify(h, HNAE3_UNINIT_CLIENT);
4718 org_tqp_num = h->kinfo.num_tqps;
4719 ret = hns3_change_channels(h, new_tqp_num, rxfh_configured);
4724 "Change channels fail, revert to old value\n");
4725 ret1 = hns3_change_channels(h, org_tqp_num, rxfh_configured);
4728 "revert to old channel fail\n");
4738 static const struct hns3_hw_error_info hns3_hw_err[] = {
4739 { .type = HNAE3_PPU_POISON_ERROR,
4740 .msg = "PPU poison" },
4741 { .type = HNAE3_CMDQ_ECC_ERROR,
4742 .msg = "IMP CMDQ error" },
4743 { .type = HNAE3_IMP_RD_POISON_ERROR,
4744 .msg = "IMP RD poison" },
4745 { .type = HNAE3_ROCEE_AXI_RESP_ERROR,
4746 .msg = "ROCEE AXI RESP error" },
4749 static void hns3_process_hw_error(struct hnae3_handle *handle,
4750 enum hnae3_hw_error_type type)
4754 for (i = 0; i < ARRAY_SIZE(hns3_hw_err); i++) {
4755 if (hns3_hw_err[i].type == type) {
4756 dev_err(&handle->pdev->dev, "Detected %s!\n",
4757 hns3_hw_err[i].msg);
4763 static const struct hnae3_client_ops client_ops = {
4764 .init_instance = hns3_client_init,
4765 .uninit_instance = hns3_client_uninit,
4766 .link_status_change = hns3_link_status_change,
4767 .setup_tc = hns3_client_setup_tc,
4768 .reset_notify = hns3_reset_notify,
4769 .process_hw_error = hns3_process_hw_error,
4772 /* hns3_init_module - Driver registration routine
4773 * hns3_init_module is the first routine called when the driver is
4774 * loaded. All it does is register with the PCI subsystem.
4776 static int __init hns3_init_module(void)
4780 pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
4781 pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
4783 client.type = HNAE3_CLIENT_KNIC;
4784 snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH, "%s",
4787 client.ops = &client_ops;
4789 INIT_LIST_HEAD(&client.node);
4791 hns3_dbg_register_debugfs(hns3_driver_name);
4793 ret = hnae3_register_client(&client);
4795 goto err_reg_client;
4797 ret = pci_register_driver(&hns3_driver);
4799 goto err_reg_driver;
4804 hnae3_unregister_client(&client);
4806 hns3_dbg_unregister_debugfs();
4809 module_init(hns3_init_module);
4811 /* hns3_exit_module - Driver exit cleanup routine
4812 * hns3_exit_module is called just before the driver is removed
4815 static void __exit hns3_exit_module(void)
4817 pci_unregister_driver(&hns3_driver);
4818 hnae3_unregister_client(&client);
4819 hns3_dbg_unregister_debugfs();
4821 module_exit(hns3_exit_module);
4823 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
4824 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
4825 MODULE_LICENSE("GPL");
4826 MODULE_ALIAS("pci:hns-nic");