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
213 if (rl_reg > 0 && !tqp_vector->tx_group.coal.adapt_enable &&
214 !tqp_vector->rx_group.coal.adapt_enable)
215 /* According to the hardware, the range of rl_reg is
216 * 0-59 and the unit is 4.
218 rl_reg |= HNS3_INT_RL_ENABLE_MASK;
220 writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
223 void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
228 if (tqp_vector->rx_group.coal.unit_1us)
229 new_val = gl_value | HNS3_INT_GL_1US;
231 new_val = hns3_gl_usec_to_reg(gl_value);
233 writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
236 void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
241 if (tqp_vector->tx_group.coal.unit_1us)
242 new_val = gl_value | HNS3_INT_GL_1US;
244 new_val = hns3_gl_usec_to_reg(gl_value);
246 writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
249 void hns3_set_vector_coalesce_tx_ql(struct hns3_enet_tqp_vector *tqp_vector,
252 writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_TX_QL_OFFSET);
255 void hns3_set_vector_coalesce_rx_ql(struct hns3_enet_tqp_vector *tqp_vector,
258 writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_RX_QL_OFFSET);
261 static void hns3_vector_coalesce_init(struct hns3_enet_tqp_vector *tqp_vector,
262 struct hns3_nic_priv *priv)
264 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
265 struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
266 struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
268 /* initialize the configuration for interrupt coalescing.
269 * 1. GL (Interrupt Gap Limiter)
270 * 2. RL (Interrupt Rate Limiter)
271 * 3. QL (Interrupt Quantity Limiter)
273 * Default: enable interrupt coalescing self-adaptive and GL
275 tx_coal->adapt_enable = 1;
276 rx_coal->adapt_enable = 1;
278 tx_coal->int_gl = HNS3_INT_GL_50K;
279 rx_coal->int_gl = HNS3_INT_GL_50K;
281 rx_coal->flow_level = HNS3_FLOW_LOW;
282 tx_coal->flow_level = HNS3_FLOW_LOW;
284 /* device version above V3(include V3), GL can configure 1us
285 * unit, so uses 1us unit.
287 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3) {
288 tx_coal->unit_1us = 1;
289 rx_coal->unit_1us = 1;
292 if (ae_dev->dev_specs.int_ql_max) {
293 tx_coal->ql_enable = 1;
294 rx_coal->ql_enable = 1;
295 tx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
296 rx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
297 tx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
298 rx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
303 hns3_vector_coalesce_init_hw(struct hns3_enet_tqp_vector *tqp_vector,
304 struct hns3_nic_priv *priv)
306 struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
307 struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
308 struct hnae3_handle *h = priv->ae_handle;
310 hns3_set_vector_coalesce_tx_gl(tqp_vector, tx_coal->int_gl);
311 hns3_set_vector_coalesce_rx_gl(tqp_vector, rx_coal->int_gl);
312 hns3_set_vector_coalesce_rl(tqp_vector, h->kinfo.int_rl_setting);
314 if (tx_coal->ql_enable)
315 hns3_set_vector_coalesce_tx_ql(tqp_vector, tx_coal->int_ql);
317 if (rx_coal->ql_enable)
318 hns3_set_vector_coalesce_rx_ql(tqp_vector, rx_coal->int_ql);
321 static int hns3_nic_set_real_num_queue(struct net_device *netdev)
323 struct hnae3_handle *h = hns3_get_handle(netdev);
324 struct hnae3_knic_private_info *kinfo = &h->kinfo;
325 struct hnae3_tc_info *tc_info = &kinfo->tc_info;
326 unsigned int queue_size = kinfo->num_tqps;
329 if (tc_info->num_tc <= 1 && !tc_info->mqprio_active) {
330 netdev_reset_tc(netdev);
332 ret = netdev_set_num_tc(netdev, tc_info->num_tc);
335 "netdev_set_num_tc fail, ret=%d!\n", ret);
339 for (i = 0; i < HNAE3_MAX_TC; i++) {
340 if (!test_bit(i, &tc_info->tc_en))
343 netdev_set_tc_queue(netdev, i, tc_info->tqp_count[i],
344 tc_info->tqp_offset[i]);
348 ret = netif_set_real_num_tx_queues(netdev, queue_size);
351 "netif_set_real_num_tx_queues fail, ret=%d!\n", ret);
355 ret = netif_set_real_num_rx_queues(netdev, queue_size);
358 "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
365 static u16 hns3_get_max_available_channels(struct hnae3_handle *h)
367 u16 alloc_tqps, max_rss_size, rss_size;
369 h->ae_algo->ops->get_tqps_and_rss_info(h, &alloc_tqps, &max_rss_size);
370 rss_size = alloc_tqps / h->kinfo.tc_info.num_tc;
372 return min_t(u16, rss_size, max_rss_size);
375 static void hns3_tqp_enable(struct hnae3_queue *tqp)
379 rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
380 rcb_reg |= BIT(HNS3_RING_EN_B);
381 hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
384 static void hns3_tqp_disable(struct hnae3_queue *tqp)
388 rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
389 rcb_reg &= ~BIT(HNS3_RING_EN_B);
390 hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
393 static void hns3_free_rx_cpu_rmap(struct net_device *netdev)
395 #ifdef CONFIG_RFS_ACCEL
396 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
397 netdev->rx_cpu_rmap = NULL;
401 static int hns3_set_rx_cpu_rmap(struct net_device *netdev)
403 #ifdef CONFIG_RFS_ACCEL
404 struct hns3_nic_priv *priv = netdev_priv(netdev);
405 struct hns3_enet_tqp_vector *tqp_vector;
408 if (!netdev->rx_cpu_rmap) {
409 netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->vector_num);
410 if (!netdev->rx_cpu_rmap)
414 for (i = 0; i < priv->vector_num; i++) {
415 tqp_vector = &priv->tqp_vector[i];
416 ret = irq_cpu_rmap_add(netdev->rx_cpu_rmap,
417 tqp_vector->vector_irq);
419 hns3_free_rx_cpu_rmap(netdev);
427 static int hns3_nic_net_up(struct net_device *netdev)
429 struct hns3_nic_priv *priv = netdev_priv(netdev);
430 struct hnae3_handle *h = priv->ae_handle;
434 ret = hns3_nic_reset_all_ring(h);
438 clear_bit(HNS3_NIC_STATE_DOWN, &priv->state);
440 /* enable the vectors */
441 for (i = 0; i < priv->vector_num; i++)
442 hns3_vector_enable(&priv->tqp_vector[i]);
445 for (j = 0; j < h->kinfo.num_tqps; j++)
446 hns3_tqp_enable(h->kinfo.tqp[j]);
448 /* start the ae_dev */
449 ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0;
451 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
453 hns3_tqp_disable(h->kinfo.tqp[j]);
455 for (j = i - 1; j >= 0; j--)
456 hns3_vector_disable(&priv->tqp_vector[j]);
462 static void hns3_config_xps(struct hns3_nic_priv *priv)
466 for (i = 0; i < priv->vector_num; i++) {
467 struct hns3_enet_tqp_vector *tqp_vector = &priv->tqp_vector[i];
468 struct hns3_enet_ring *ring = tqp_vector->tx_group.ring;
473 ret = netif_set_xps_queue(priv->netdev,
474 &tqp_vector->affinity_mask,
475 ring->tqp->tqp_index);
477 netdev_warn(priv->netdev,
478 "set xps queue failed: %d", ret);
485 static int hns3_nic_net_open(struct net_device *netdev)
487 struct hns3_nic_priv *priv = netdev_priv(netdev);
488 struct hnae3_handle *h = hns3_get_handle(netdev);
489 struct hnae3_knic_private_info *kinfo;
492 if (hns3_nic_resetting(netdev))
495 netif_carrier_off(netdev);
497 ret = hns3_nic_set_real_num_queue(netdev);
501 ret = hns3_nic_net_up(netdev);
503 netdev_err(netdev, "net up fail, ret=%d!\n", ret);
508 for (i = 0; i < HNAE3_MAX_USER_PRIO; i++)
509 netdev_set_prio_tc_map(netdev, i, kinfo->tc_info.prio_tc[i]);
511 if (h->ae_algo->ops->set_timer_task)
512 h->ae_algo->ops->set_timer_task(priv->ae_handle, true);
514 hns3_config_xps(priv);
516 netif_dbg(h, drv, netdev, "net open\n");
521 static void hns3_reset_tx_queue(struct hnae3_handle *h)
523 struct net_device *ndev = h->kinfo.netdev;
524 struct hns3_nic_priv *priv = netdev_priv(ndev);
525 struct netdev_queue *dev_queue;
528 for (i = 0; i < h->kinfo.num_tqps; i++) {
529 dev_queue = netdev_get_tx_queue(ndev,
530 priv->ring[i].queue_index);
531 netdev_tx_reset_queue(dev_queue);
535 static void hns3_nic_net_down(struct net_device *netdev)
537 struct hns3_nic_priv *priv = netdev_priv(netdev);
538 struct hnae3_handle *h = hns3_get_handle(netdev);
539 const struct hnae3_ae_ops *ops;
542 /* disable vectors */
543 for (i = 0; i < priv->vector_num; i++)
544 hns3_vector_disable(&priv->tqp_vector[i]);
547 for (i = 0; i < h->kinfo.num_tqps; i++)
548 hns3_tqp_disable(h->kinfo.tqp[i]);
551 ops = priv->ae_handle->ae_algo->ops;
553 ops->stop(priv->ae_handle);
555 /* delay ring buffer clearing to hns3_reset_notify_uninit_enet
556 * during reset process, because driver may not be able
557 * to disable the ring through firmware when downing the netdev.
559 if (!hns3_nic_resetting(netdev))
560 hns3_clear_all_ring(priv->ae_handle, false);
562 hns3_reset_tx_queue(priv->ae_handle);
565 static int hns3_nic_net_stop(struct net_device *netdev)
567 struct hns3_nic_priv *priv = netdev_priv(netdev);
568 struct hnae3_handle *h = hns3_get_handle(netdev);
570 if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
573 netif_dbg(h, drv, netdev, "net stop\n");
575 if (h->ae_algo->ops->set_timer_task)
576 h->ae_algo->ops->set_timer_task(priv->ae_handle, false);
578 netif_carrier_off(netdev);
579 netif_tx_disable(netdev);
581 hns3_nic_net_down(netdev);
586 static int hns3_nic_uc_sync(struct net_device *netdev,
587 const unsigned char *addr)
589 struct hnae3_handle *h = hns3_get_handle(netdev);
591 if (h->ae_algo->ops->add_uc_addr)
592 return h->ae_algo->ops->add_uc_addr(h, addr);
597 static int hns3_nic_uc_unsync(struct net_device *netdev,
598 const unsigned char *addr)
600 struct hnae3_handle *h = hns3_get_handle(netdev);
602 /* need ignore the request of removing device address, because
603 * we store the device address and other addresses of uc list
604 * in the function's mac filter list.
606 if (ether_addr_equal(addr, netdev->dev_addr))
609 if (h->ae_algo->ops->rm_uc_addr)
610 return h->ae_algo->ops->rm_uc_addr(h, addr);
615 static int hns3_nic_mc_sync(struct net_device *netdev,
616 const unsigned char *addr)
618 struct hnae3_handle *h = hns3_get_handle(netdev);
620 if (h->ae_algo->ops->add_mc_addr)
621 return h->ae_algo->ops->add_mc_addr(h, addr);
626 static int hns3_nic_mc_unsync(struct net_device *netdev,
627 const unsigned char *addr)
629 struct hnae3_handle *h = hns3_get_handle(netdev);
631 if (h->ae_algo->ops->rm_mc_addr)
632 return h->ae_algo->ops->rm_mc_addr(h, addr);
637 static u8 hns3_get_netdev_flags(struct net_device *netdev)
641 if (netdev->flags & IFF_PROMISC) {
642 flags = HNAE3_USER_UPE | HNAE3_USER_MPE | HNAE3_BPE;
644 flags |= HNAE3_VLAN_FLTR;
645 if (netdev->flags & IFF_ALLMULTI)
646 flags |= HNAE3_USER_MPE;
652 static void hns3_nic_set_rx_mode(struct net_device *netdev)
654 struct hnae3_handle *h = hns3_get_handle(netdev);
657 new_flags = hns3_get_netdev_flags(netdev);
659 __dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync);
660 __dev_mc_sync(netdev, hns3_nic_mc_sync, hns3_nic_mc_unsync);
662 /* User mode Promisc mode enable and vlan filtering is disabled to
663 * let all packets in.
665 h->netdev_flags = new_flags;
666 hns3_request_update_promisc_mode(h);
669 void hns3_request_update_promisc_mode(struct hnae3_handle *handle)
671 const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
673 if (ops->request_update_promisc_mode)
674 ops->request_update_promisc_mode(handle);
677 void hns3_enable_vlan_filter(struct net_device *netdev, bool enable)
679 struct hns3_nic_priv *priv = netdev_priv(netdev);
680 struct hnae3_handle *h = priv->ae_handle;
681 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(h->pdev);
684 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V2 &&
685 h->ae_algo->ops->enable_vlan_filter) {
686 last_state = h->netdev_flags & HNAE3_VLAN_FLTR ? true : false;
687 if (enable != last_state) {
690 enable ? "enable" : "disable");
691 h->ae_algo->ops->enable_vlan_filter(h, enable);
696 static int hns3_set_tso(struct sk_buff *skb, u32 *paylen_fdop_ol4cs,
697 u16 *mss, u32 *type_cs_vlan_tso, u32 *send_bytes)
699 u32 l4_offset, hdr_len;
700 union l3_hdr_info l3;
701 union l4_hdr_info l4;
705 if (!skb_is_gso(skb))
708 ret = skb_cow_head(skb, 0);
709 if (unlikely(ret < 0))
712 l3.hdr = skb_network_header(skb);
713 l4.hdr = skb_transport_header(skb);
715 /* Software should clear the IPv4's checksum field when tso is
718 if (l3.v4->version == 4)
722 if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
725 SKB_GSO_UDP_TUNNEL_CSUM)) {
726 /* reset l3&l4 pointers from outer to inner headers */
727 l3.hdr = skb_inner_network_header(skb);
728 l4.hdr = skb_inner_transport_header(skb);
730 /* Software should clear the IPv4's checksum field when
733 if (l3.v4->version == 4)
737 /* normal or tunnel packet */
738 l4_offset = l4.hdr - skb->data;
740 /* remove payload length from inner pseudo checksum when tso */
741 l4_paylen = skb->len - l4_offset;
743 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) {
744 hdr_len = sizeof(*l4.udp) + l4_offset;
745 csum_replace_by_diff(&l4.udp->check,
746 (__force __wsum)htonl(l4_paylen));
748 hdr_len = (l4.tcp->doff << 2) + l4_offset;
749 csum_replace_by_diff(&l4.tcp->check,
750 (__force __wsum)htonl(l4_paylen));
753 *send_bytes = (skb_shinfo(skb)->gso_segs - 1) * hdr_len + skb->len;
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
827 * 4789, 4790 or 6081.
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) ||
846 l4.udp->dest == htons(4790))))
849 skb_checksum_help(skb);
854 static void hns3_set_outer_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
855 u32 *ol_type_vlan_len_msec)
857 u32 l2_len, l3_len, l4_len;
858 unsigned char *il2_hdr;
859 union l3_hdr_info l3;
860 union l4_hdr_info l4;
862 l3.hdr = skb_network_header(skb);
863 l4.hdr = skb_transport_header(skb);
865 /* compute OL2 header size, defined in 2 Bytes */
866 l2_len = l3.hdr - skb->data;
867 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L2LEN_S, l2_len >> 1);
869 /* compute OL3 header size, defined in 4 Bytes */
870 l3_len = l4.hdr - l3.hdr;
871 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_S, l3_len >> 2);
873 il2_hdr = skb_inner_mac_header(skb);
874 /* compute OL4 header size, defined in 4 Bytes */
875 l4_len = il2_hdr - l4.hdr;
876 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L4LEN_S, l4_len >> 2);
878 /* define outer network header type */
879 if (skb->protocol == htons(ETH_P_IP)) {
881 hns3_set_field(*ol_type_vlan_len_msec,
883 HNS3_OL3T_IPV4_CSUM);
885 hns3_set_field(*ol_type_vlan_len_msec,
887 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_is_sctp(skb) || 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,
1083 struct hns3_desc_cb *desc_cb)
1085 u32 ol_type_vlan_len_msec = 0;
1086 u32 paylen_ol4cs = skb->len;
1087 u32 type_cs_vlan_tso = 0;
1088 u16 mss_hw_csum = 0;
1093 ret = hns3_handle_vtags(ring, skb);
1094 if (unlikely(ret < 0)) {
1095 u64_stats_update_begin(&ring->syncp);
1096 ring->stats.tx_vlan_err++;
1097 u64_stats_update_end(&ring->syncp);
1099 } else if (ret == HNS3_INNER_VLAN_TAG) {
1100 inner_vtag = skb_vlan_tag_get(skb);
1101 inner_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1103 hns3_set_field(type_cs_vlan_tso, HNS3_TXD_VLAN_B, 1);
1104 } else if (ret == HNS3_OUTER_VLAN_TAG) {
1105 out_vtag = skb_vlan_tag_get(skb);
1106 out_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1108 hns3_set_field(ol_type_vlan_len_msec, HNS3_TXD_OVLAN_B,
1112 desc_cb->send_bytes = skb->len;
1114 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1115 u8 ol4_proto, il4_proto;
1117 if (hns3_check_hw_tx_csum(skb)) {
1118 /* set checksum start and offset, defined in 2 Bytes */
1119 hns3_set_field(type_cs_vlan_tso, HNS3_TXD_CSUM_START_S,
1120 skb_checksum_start_offset(skb) >> 1);
1121 hns3_set_field(ol_type_vlan_len_msec,
1122 HNS3_TXD_CSUM_OFFSET_S,
1123 skb->csum_offset >> 1);
1124 mss_hw_csum |= BIT(HNS3_TXD_HW_CS_B);
1125 goto out_hw_tx_csum;
1128 skb_reset_mac_len(skb);
1130 ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
1131 if (unlikely(ret < 0)) {
1132 u64_stats_update_begin(&ring->syncp);
1133 ring->stats.tx_l4_proto_err++;
1134 u64_stats_update_end(&ring->syncp);
1138 ret = hns3_set_l2l3l4(skb, ol4_proto, il4_proto,
1140 &ol_type_vlan_len_msec);
1141 if (unlikely(ret < 0)) {
1142 u64_stats_update_begin(&ring->syncp);
1143 ring->stats.tx_l2l3l4_err++;
1144 u64_stats_update_end(&ring->syncp);
1148 ret = hns3_set_tso(skb, &paylen_ol4cs, &mss_hw_csum,
1149 &type_cs_vlan_tso, &desc_cb->send_bytes);
1150 if (unlikely(ret < 0)) {
1151 u64_stats_update_begin(&ring->syncp);
1152 ring->stats.tx_tso_err++;
1153 u64_stats_update_end(&ring->syncp);
1160 desc->tx.ol_type_vlan_len_msec =
1161 cpu_to_le32(ol_type_vlan_len_msec);
1162 desc->tx.type_cs_vlan_tso_len = cpu_to_le32(type_cs_vlan_tso);
1163 desc->tx.paylen_ol4cs = cpu_to_le32(paylen_ol4cs);
1164 desc->tx.mss_hw_csum = cpu_to_le16(mss_hw_csum);
1165 desc->tx.vlan_tag = cpu_to_le16(inner_vtag);
1166 desc->tx.outer_vlan_tag = cpu_to_le16(out_vtag);
1171 static int hns3_fill_desc(struct hns3_enet_ring *ring, void *priv,
1172 unsigned int size, enum hns_desc_type type)
1174 #define HNS3_LIKELY_BD_NUM 1
1176 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1177 struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1178 struct device *dev = ring_to_dev(ring);
1180 unsigned int frag_buf_num;
1184 if (type == DESC_TYPE_FRAGLIST_SKB ||
1185 type == DESC_TYPE_SKB) {
1186 struct sk_buff *skb = (struct sk_buff *)priv;
1188 dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1190 frag = (skb_frag_t *)priv;
1191 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1194 if (unlikely(dma_mapping_error(dev, dma))) {
1195 u64_stats_update_begin(&ring->syncp);
1196 ring->stats.sw_err_cnt++;
1197 u64_stats_update_end(&ring->syncp);
1201 desc_cb->priv = priv;
1202 desc_cb->length = size;
1204 desc_cb->type = type;
1206 if (likely(size <= HNS3_MAX_BD_SIZE)) {
1207 desc->addr = cpu_to_le64(dma);
1208 desc->tx.send_size = cpu_to_le16(size);
1209 desc->tx.bdtp_fe_sc_vld_ra_ri =
1210 cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1212 trace_hns3_tx_desc(ring, ring->next_to_use);
1213 ring_ptr_move_fw(ring, next_to_use);
1214 return HNS3_LIKELY_BD_NUM;
1217 frag_buf_num = hns3_tx_bd_count(size);
1218 sizeoflast = size % HNS3_MAX_BD_SIZE;
1219 sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1221 /* When frag size is bigger than hardware limit, split this frag */
1222 for (k = 0; k < frag_buf_num; k++) {
1223 /* now, fill the descriptor */
1224 desc->addr = cpu_to_le64(dma + HNS3_MAX_BD_SIZE * k);
1225 desc->tx.send_size = cpu_to_le16((k == frag_buf_num - 1) ?
1226 (u16)sizeoflast : (u16)HNS3_MAX_BD_SIZE);
1227 desc->tx.bdtp_fe_sc_vld_ra_ri =
1228 cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1230 trace_hns3_tx_desc(ring, ring->next_to_use);
1231 /* move ring pointer to next */
1232 ring_ptr_move_fw(ring, next_to_use);
1234 desc = &ring->desc[ring->next_to_use];
1237 return frag_buf_num;
1240 static unsigned int hns3_skb_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1241 unsigned int bd_num)
1246 size = skb_headlen(skb);
1247 while (size > HNS3_MAX_BD_SIZE) {
1248 bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1249 size -= HNS3_MAX_BD_SIZE;
1251 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1256 bd_size[bd_num++] = size;
1257 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1261 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1262 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1263 size = skb_frag_size(frag);
1267 while (size > HNS3_MAX_BD_SIZE) {
1268 bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1269 size -= HNS3_MAX_BD_SIZE;
1271 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1275 bd_size[bd_num++] = size;
1276 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1283 static unsigned int hns3_tx_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1284 u8 max_non_tso_bd_num, unsigned int bd_num,
1285 unsigned int recursion_level)
1287 #define HNS3_MAX_RECURSION_LEVEL 24
1289 struct sk_buff *frag_skb;
1291 /* If the total len is within the max bd limit */
1292 if (likely(skb->len <= HNS3_MAX_BD_SIZE && !recursion_level &&
1293 !skb_has_frag_list(skb) &&
1294 skb_shinfo(skb)->nr_frags < max_non_tso_bd_num))
1295 return skb_shinfo(skb)->nr_frags + 1U;
1297 if (unlikely(recursion_level >= HNS3_MAX_RECURSION_LEVEL))
1300 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_tx_bd_num(frag_skb, bd_size, max_non_tso_bd_num,
1306 bd_num, recursion_level + 1);
1307 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1314 static unsigned int hns3_gso_hdr_len(struct sk_buff *skb)
1316 if (!skb->encapsulation)
1317 return skb_transport_offset(skb) + tcp_hdrlen(skb);
1319 return skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb);
1322 /* HW need every continuous max_non_tso_bd_num buffer data to be larger
1323 * than MSS, we simplify it by ensuring skb_headlen + the first continuous
1324 * max_non_tso_bd_num - 1 frags to be larger than gso header len + mss,
1325 * and the remaining continuous max_non_tso_bd_num - 1 frags to be larger
1326 * than MSS except the last max_non_tso_bd_num - 1 frags.
1328 static bool hns3_skb_need_linearized(struct sk_buff *skb, unsigned int *bd_size,
1329 unsigned int bd_num, u8 max_non_tso_bd_num)
1331 unsigned int tot_len = 0;
1334 for (i = 0; i < max_non_tso_bd_num - 1U; i++)
1335 tot_len += bd_size[i];
1337 /* ensure the first max_non_tso_bd_num frags is greater than
1340 if (tot_len + bd_size[max_non_tso_bd_num - 1U] <
1341 skb_shinfo(skb)->gso_size + hns3_gso_hdr_len(skb))
1344 /* ensure every continuous max_non_tso_bd_num - 1 buffer is greater
1345 * than mss except the last one.
1347 for (i = 0; i < bd_num - max_non_tso_bd_num; i++) {
1348 tot_len -= bd_size[i];
1349 tot_len += bd_size[i + max_non_tso_bd_num - 1U];
1351 if (tot_len < skb_shinfo(skb)->gso_size)
1358 void hns3_shinfo_pack(struct skb_shared_info *shinfo, __u32 *size)
1362 for (i = 0; i < MAX_SKB_FRAGS; i++)
1363 size[i] = skb_frag_size(&shinfo->frags[i]);
1366 static int hns3_skb_linearize(struct hns3_enet_ring *ring,
1367 struct sk_buff *skb,
1368 u8 max_non_tso_bd_num,
1369 unsigned int bd_num)
1371 /* 'bd_num == UINT_MAX' means the skb' fraglist has a
1372 * recursion level of over HNS3_MAX_RECURSION_LEVEL.
1374 if (bd_num == UINT_MAX) {
1375 u64_stats_update_begin(&ring->syncp);
1376 ring->stats.over_max_recursion++;
1377 u64_stats_update_end(&ring->syncp);
1381 /* The skb->len has exceeded the hw limitation, linearization
1384 if (skb->len > HNS3_MAX_TSO_SIZE ||
1385 (!skb_is_gso(skb) && skb->len >
1386 HNS3_MAX_NON_TSO_SIZE(max_non_tso_bd_num))) {
1387 u64_stats_update_begin(&ring->syncp);
1388 ring->stats.hw_limitation++;
1389 u64_stats_update_end(&ring->syncp);
1393 if (__skb_linearize(skb)) {
1394 u64_stats_update_begin(&ring->syncp);
1395 ring->stats.sw_err_cnt++;
1396 u64_stats_update_end(&ring->syncp);
1403 static int hns3_nic_maybe_stop_tx(struct hns3_enet_ring *ring,
1404 struct net_device *netdev,
1405 struct sk_buff *skb)
1407 struct hns3_nic_priv *priv = netdev_priv(netdev);
1408 u8 max_non_tso_bd_num = priv->max_non_tso_bd_num;
1409 unsigned int bd_size[HNS3_MAX_TSO_BD_NUM + 1U];
1410 unsigned int bd_num;
1412 bd_num = hns3_tx_bd_num(skb, bd_size, max_non_tso_bd_num, 0, 0);
1413 if (unlikely(bd_num > max_non_tso_bd_num)) {
1414 if (bd_num <= HNS3_MAX_TSO_BD_NUM && skb_is_gso(skb) &&
1415 !hns3_skb_need_linearized(skb, bd_size, bd_num,
1416 max_non_tso_bd_num)) {
1417 trace_hns3_over_max_bd(skb);
1421 if (hns3_skb_linearize(ring, skb, max_non_tso_bd_num,
1425 bd_num = hns3_tx_bd_count(skb->len);
1427 u64_stats_update_begin(&ring->syncp);
1428 ring->stats.tx_copy++;
1429 u64_stats_update_end(&ring->syncp);
1433 if (likely(ring_space(ring) >= bd_num))
1436 netif_stop_subqueue(netdev, ring->queue_index);
1437 smp_mb(); /* Memory barrier before checking ring_space */
1439 /* Start queue in case hns3_clean_tx_ring has just made room
1440 * available and has not seen the queue stopped state performed
1441 * by netif_stop_subqueue above.
1443 if (ring_space(ring) >= bd_num && netif_carrier_ok(netdev) &&
1444 !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
1445 netif_start_subqueue(netdev, ring->queue_index);
1449 u64_stats_update_begin(&ring->syncp);
1450 ring->stats.tx_busy++;
1451 u64_stats_update_end(&ring->syncp);
1456 static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
1458 struct device *dev = ring_to_dev(ring);
1461 for (i = 0; i < ring->desc_num; i++) {
1462 struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1464 memset(desc, 0, sizeof(*desc));
1466 /* check if this is where we started */
1467 if (ring->next_to_use == next_to_use_orig)
1471 ring_ptr_move_bw(ring, next_to_use);
1473 if (!ring->desc_cb[ring->next_to_use].dma)
1476 /* unmap the descriptor dma address */
1477 if (ring->desc_cb[ring->next_to_use].type == DESC_TYPE_SKB ||
1478 ring->desc_cb[ring->next_to_use].type ==
1479 DESC_TYPE_FRAGLIST_SKB)
1480 dma_unmap_single(dev,
1481 ring->desc_cb[ring->next_to_use].dma,
1482 ring->desc_cb[ring->next_to_use].length,
1484 else if (ring->desc_cb[ring->next_to_use].length)
1486 ring->desc_cb[ring->next_to_use].dma,
1487 ring->desc_cb[ring->next_to_use].length,
1490 ring->desc_cb[ring->next_to_use].length = 0;
1491 ring->desc_cb[ring->next_to_use].dma = 0;
1492 ring->desc_cb[ring->next_to_use].type = DESC_TYPE_UNKNOWN;
1496 static int hns3_fill_skb_to_desc(struct hns3_enet_ring *ring,
1497 struct sk_buff *skb, enum hns_desc_type type)
1499 unsigned int size = skb_headlen(skb);
1500 struct sk_buff *frag_skb;
1501 int i, ret, bd_num = 0;
1504 ret = hns3_fill_desc(ring, skb, size, type);
1505 if (unlikely(ret < 0))
1511 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1512 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1514 size = skb_frag_size(frag);
1518 ret = hns3_fill_desc(ring, frag, size, DESC_TYPE_PAGE);
1519 if (unlikely(ret < 0))
1525 skb_walk_frags(skb, frag_skb) {
1526 ret = hns3_fill_skb_to_desc(ring, frag_skb,
1527 DESC_TYPE_FRAGLIST_SKB);
1528 if (unlikely(ret < 0))
1537 static void hns3_tx_doorbell(struct hns3_enet_ring *ring, int num,
1540 ring->pending_buf += num;
1543 u64_stats_update_begin(&ring->syncp);
1544 ring->stats.tx_more++;
1545 u64_stats_update_end(&ring->syncp);
1549 if (!ring->pending_buf)
1552 writel(ring->pending_buf,
1553 ring->tqp->io_base + HNS3_RING_TX_RING_TAIL_REG);
1554 ring->pending_buf = 0;
1555 WRITE_ONCE(ring->last_to_use, ring->next_to_use);
1558 netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
1560 struct hns3_nic_priv *priv = netdev_priv(netdev);
1561 struct hns3_enet_ring *ring = &priv->ring[skb->queue_mapping];
1562 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1563 struct netdev_queue *dev_queue;
1564 int pre_ntu, next_to_use_head;
1568 /* Hardware can only handle short frames above 32 bytes */
1569 if (skb_put_padto(skb, HNS3_MIN_TX_LEN)) {
1570 hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
1572 u64_stats_update_begin(&ring->syncp);
1573 ring->stats.sw_err_cnt++;
1574 u64_stats_update_end(&ring->syncp);
1576 return NETDEV_TX_OK;
1579 /* Prefetch the data used later */
1580 prefetch(skb->data);
1582 ret = hns3_nic_maybe_stop_tx(ring, netdev, skb);
1583 if (unlikely(ret <= 0)) {
1584 if (ret == -EBUSY) {
1585 hns3_tx_doorbell(ring, 0, true);
1586 return NETDEV_TX_BUSY;
1589 hns3_rl_err(netdev, "xmit error: %d!\n", ret);
1593 next_to_use_head = ring->next_to_use;
1595 ret = hns3_fill_skb_desc(ring, skb, &ring->desc[ring->next_to_use],
1597 if (unlikely(ret < 0))
1600 /* 'ret < 0' means filling error, 'ret == 0' means skb->len is
1601 * zero, which is unlikely, and 'ret > 0' means how many tx desc
1602 * need to be notified to the hw.
1604 ret = hns3_fill_skb_to_desc(ring, skb, DESC_TYPE_SKB);
1605 if (unlikely(ret <= 0))
1608 pre_ntu = ring->next_to_use ? (ring->next_to_use - 1) :
1609 (ring->desc_num - 1);
1610 ring->desc[pre_ntu].tx.bdtp_fe_sc_vld_ra_ri |=
1611 cpu_to_le16(BIT(HNS3_TXD_FE_B));
1612 trace_hns3_tx_desc(ring, pre_ntu);
1614 /* Complete translate all packets */
1615 dev_queue = netdev_get_tx_queue(netdev, ring->queue_index);
1616 doorbell = __netdev_tx_sent_queue(dev_queue, desc_cb->send_bytes,
1617 netdev_xmit_more());
1618 hns3_tx_doorbell(ring, ret, doorbell);
1620 return NETDEV_TX_OK;
1623 hns3_clear_desc(ring, next_to_use_head);
1626 dev_kfree_skb_any(skb);
1627 hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
1628 return NETDEV_TX_OK;
1631 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
1633 struct hnae3_handle *h = hns3_get_handle(netdev);
1634 struct sockaddr *mac_addr = p;
1637 if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1638 return -EADDRNOTAVAIL;
1640 if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
1641 netdev_info(netdev, "already using mac address %pM\n",
1646 /* For VF device, if there is a perm_addr, then the user will not
1647 * be allowed to change the address.
1649 if (!hns3_is_phys_func(h->pdev) &&
1650 !is_zero_ether_addr(netdev->perm_addr)) {
1651 netdev_err(netdev, "has permanent MAC %pM, user MAC %pM not allow\n",
1652 netdev->perm_addr, mac_addr->sa_data);
1656 ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
1658 netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
1662 ether_addr_copy(netdev->dev_addr, mac_addr->sa_data);
1667 static int hns3_nic_do_ioctl(struct net_device *netdev,
1668 struct ifreq *ifr, int cmd)
1670 struct hnae3_handle *h = hns3_get_handle(netdev);
1672 if (!netif_running(netdev))
1675 if (!h->ae_algo->ops->do_ioctl)
1678 return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
1681 static int hns3_nic_set_features(struct net_device *netdev,
1682 netdev_features_t features)
1684 netdev_features_t changed = netdev->features ^ features;
1685 struct hns3_nic_priv *priv = netdev_priv(netdev);
1686 struct hnae3_handle *h = priv->ae_handle;
1690 if (changed & (NETIF_F_GRO_HW) && h->ae_algo->ops->set_gro_en) {
1691 enable = !!(features & NETIF_F_GRO_HW);
1692 ret = h->ae_algo->ops->set_gro_en(h, enable);
1697 if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
1698 h->ae_algo->ops->enable_hw_strip_rxvtag) {
1699 enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
1700 ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, enable);
1705 if ((changed & NETIF_F_NTUPLE) && h->ae_algo->ops->enable_fd) {
1706 enable = !!(features & NETIF_F_NTUPLE);
1707 h->ae_algo->ops->enable_fd(h, enable);
1710 if ((netdev->features & NETIF_F_HW_TC) > (features & NETIF_F_HW_TC) &&
1711 h->ae_algo->ops->cls_flower_active(h)) {
1713 "there are offloaded TC filters active, cannot disable HW TC offload");
1717 netdev->features = features;
1721 static netdev_features_t hns3_features_check(struct sk_buff *skb,
1722 struct net_device *dev,
1723 netdev_features_t features)
1725 #define HNS3_MAX_HDR_LEN 480U
1726 #define HNS3_MAX_L4_HDR_LEN 60U
1730 if (skb->ip_summed != CHECKSUM_PARTIAL)
1733 if (skb->encapsulation)
1734 len = skb_inner_transport_header(skb) - skb->data;
1736 len = skb_transport_header(skb) - skb->data;
1738 /* Assume L4 is 60 byte as TCP is the only protocol with a
1739 * a flexible value, and it's max len is 60 bytes.
1741 len += HNS3_MAX_L4_HDR_LEN;
1743 /* Hardware only supports checksum on the skb with a max header
1746 if (len > HNS3_MAX_HDR_LEN)
1747 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
1752 static void hns3_nic_get_stats64(struct net_device *netdev,
1753 struct rtnl_link_stats64 *stats)
1755 struct hns3_nic_priv *priv = netdev_priv(netdev);
1756 int queue_num = priv->ae_handle->kinfo.num_tqps;
1757 struct hnae3_handle *handle = priv->ae_handle;
1758 struct hns3_enet_ring *ring;
1759 u64 rx_length_errors = 0;
1760 u64 rx_crc_errors = 0;
1761 u64 rx_multicast = 0;
1773 if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
1776 handle->ae_algo->ops->update_stats(handle, &netdev->stats);
1778 for (idx = 0; idx < queue_num; idx++) {
1779 /* fetch the tx stats */
1780 ring = &priv->ring[idx];
1782 start = u64_stats_fetch_begin_irq(&ring->syncp);
1783 tx_bytes += ring->stats.tx_bytes;
1784 tx_pkts += ring->stats.tx_pkts;
1785 tx_drop += ring->stats.sw_err_cnt;
1786 tx_drop += ring->stats.tx_vlan_err;
1787 tx_drop += ring->stats.tx_l4_proto_err;
1788 tx_drop += ring->stats.tx_l2l3l4_err;
1789 tx_drop += ring->stats.tx_tso_err;
1790 tx_drop += ring->stats.over_max_recursion;
1791 tx_drop += ring->stats.hw_limitation;
1792 tx_errors += ring->stats.sw_err_cnt;
1793 tx_errors += ring->stats.tx_vlan_err;
1794 tx_errors += ring->stats.tx_l4_proto_err;
1795 tx_errors += ring->stats.tx_l2l3l4_err;
1796 tx_errors += ring->stats.tx_tso_err;
1797 tx_errors += ring->stats.over_max_recursion;
1798 tx_errors += ring->stats.hw_limitation;
1799 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1801 /* fetch the rx stats */
1802 ring = &priv->ring[idx + queue_num];
1804 start = u64_stats_fetch_begin_irq(&ring->syncp);
1805 rx_bytes += ring->stats.rx_bytes;
1806 rx_pkts += ring->stats.rx_pkts;
1807 rx_drop += ring->stats.l2_err;
1808 rx_errors += ring->stats.l2_err;
1809 rx_errors += ring->stats.l3l4_csum_err;
1810 rx_crc_errors += ring->stats.l2_err;
1811 rx_multicast += ring->stats.rx_multicast;
1812 rx_length_errors += ring->stats.err_pkt_len;
1813 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1816 stats->tx_bytes = tx_bytes;
1817 stats->tx_packets = tx_pkts;
1818 stats->rx_bytes = rx_bytes;
1819 stats->rx_packets = rx_pkts;
1821 stats->rx_errors = rx_errors;
1822 stats->multicast = rx_multicast;
1823 stats->rx_length_errors = rx_length_errors;
1824 stats->rx_crc_errors = rx_crc_errors;
1825 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
1827 stats->tx_errors = tx_errors;
1828 stats->rx_dropped = rx_drop;
1829 stats->tx_dropped = tx_drop;
1830 stats->collisions = netdev->stats.collisions;
1831 stats->rx_over_errors = netdev->stats.rx_over_errors;
1832 stats->rx_frame_errors = netdev->stats.rx_frame_errors;
1833 stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
1834 stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
1835 stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
1836 stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
1837 stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
1838 stats->tx_window_errors = netdev->stats.tx_window_errors;
1839 stats->rx_compressed = netdev->stats.rx_compressed;
1840 stats->tx_compressed = netdev->stats.tx_compressed;
1843 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
1845 struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
1846 struct hnae3_knic_private_info *kinfo;
1847 u8 tc = mqprio_qopt->qopt.num_tc;
1848 u16 mode = mqprio_qopt->mode;
1849 u8 hw = mqprio_qopt->qopt.hw;
1850 struct hnae3_handle *h;
1852 if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
1853 mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
1856 if (tc > HNAE3_MAX_TC)
1862 h = hns3_get_handle(netdev);
1865 netif_dbg(h, drv, netdev, "setup tc: num_tc=%u\n", tc);
1867 return (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
1868 kinfo->dcb_ops->setup_tc(h, mqprio_qopt) : -EOPNOTSUPP;
1871 static int hns3_setup_tc_cls_flower(struct hns3_nic_priv *priv,
1872 struct flow_cls_offload *flow)
1874 int tc = tc_classid_to_hwtc(priv->netdev, flow->classid);
1875 struct hnae3_handle *h = hns3_get_handle(priv->netdev);
1877 switch (flow->command) {
1878 case FLOW_CLS_REPLACE:
1879 if (h->ae_algo->ops->add_cls_flower)
1880 return h->ae_algo->ops->add_cls_flower(h, flow, tc);
1882 case FLOW_CLS_DESTROY:
1883 if (h->ae_algo->ops->del_cls_flower)
1884 return h->ae_algo->ops->del_cls_flower(h, flow);
1893 static int hns3_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
1896 struct hns3_nic_priv *priv = cb_priv;
1898 if (!tc_cls_can_offload_and_chain0(priv->netdev, type_data))
1902 case TC_SETUP_CLSFLOWER:
1903 return hns3_setup_tc_cls_flower(priv, type_data);
1909 static LIST_HEAD(hns3_block_cb_list);
1911 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
1914 struct hns3_nic_priv *priv = netdev_priv(dev);
1918 case TC_SETUP_QDISC_MQPRIO:
1919 ret = hns3_setup_tc(dev, type_data);
1921 case TC_SETUP_BLOCK:
1922 ret = flow_block_cb_setup_simple(type_data,
1923 &hns3_block_cb_list,
1924 hns3_setup_tc_block_cb,
1934 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
1935 __be16 proto, u16 vid)
1937 struct hnae3_handle *h = hns3_get_handle(netdev);
1940 if (h->ae_algo->ops->set_vlan_filter)
1941 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
1946 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
1947 __be16 proto, u16 vid)
1949 struct hnae3_handle *h = hns3_get_handle(netdev);
1952 if (h->ae_algo->ops->set_vlan_filter)
1953 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
1958 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
1959 u8 qos, __be16 vlan_proto)
1961 struct hnae3_handle *h = hns3_get_handle(netdev);
1964 netif_dbg(h, drv, netdev,
1965 "set vf vlan: vf=%d, vlan=%u, qos=%u, vlan_proto=0x%x\n",
1966 vf, vlan, qos, ntohs(vlan_proto));
1968 if (h->ae_algo->ops->set_vf_vlan_filter)
1969 ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
1975 static int hns3_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
1977 struct hnae3_handle *handle = hns3_get_handle(netdev);
1979 if (hns3_nic_resetting(netdev))
1982 if (!handle->ae_algo->ops->set_vf_spoofchk)
1985 return handle->ae_algo->ops->set_vf_spoofchk(handle, vf, enable);
1988 static int hns3_set_vf_trust(struct net_device *netdev, int vf, bool enable)
1990 struct hnae3_handle *handle = hns3_get_handle(netdev);
1992 if (!handle->ae_algo->ops->set_vf_trust)
1995 return handle->ae_algo->ops->set_vf_trust(handle, vf, enable);
1998 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
2000 struct hnae3_handle *h = hns3_get_handle(netdev);
2003 if (hns3_nic_resetting(netdev))
2006 if (!h->ae_algo->ops->set_mtu)
2009 netif_dbg(h, drv, netdev,
2010 "change mtu from %u to %d\n", netdev->mtu, new_mtu);
2012 ret = h->ae_algo->ops->set_mtu(h, new_mtu);
2014 netdev_err(netdev, "failed to change MTU in hardware %d\n",
2017 netdev->mtu = new_mtu;
2022 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
2024 struct hns3_nic_priv *priv = netdev_priv(ndev);
2025 struct hnae3_handle *h = hns3_get_handle(ndev);
2026 struct hns3_enet_ring *tx_ring;
2027 struct napi_struct *napi;
2028 int timeout_queue = 0;
2029 int hw_head, hw_tail;
2030 int fbd_num, fbd_oft;
2031 int ebd_num, ebd_oft;
2036 /* Find the stopped queue the same way the stack does */
2037 for (i = 0; i < ndev->num_tx_queues; i++) {
2038 struct netdev_queue *q;
2039 unsigned long trans_start;
2041 q = netdev_get_tx_queue(ndev, i);
2042 trans_start = q->trans_start;
2043 if (netif_xmit_stopped(q) &&
2045 (trans_start + ndev->watchdog_timeo))) {
2047 netdev_info(ndev, "queue state: 0x%lx, delta msecs: %u\n",
2049 jiffies_to_msecs(jiffies - trans_start));
2054 if (i == ndev->num_tx_queues) {
2056 "no netdev TX timeout queue found, timeout count: %llu\n",
2057 priv->tx_timeout_count);
2061 priv->tx_timeout_count++;
2063 tx_ring = &priv->ring[timeout_queue];
2064 napi = &tx_ring->tqp_vector->napi;
2067 "tx_timeout count: %llu, queue id: %d, SW_NTU: 0x%x, SW_NTC: 0x%x, napi state: %lu\n",
2068 priv->tx_timeout_count, timeout_queue, tx_ring->next_to_use,
2069 tx_ring->next_to_clean, napi->state);
2072 "tx_pkts: %llu, tx_bytes: %llu, sw_err_cnt: %llu, tx_pending: %d\n",
2073 tx_ring->stats.tx_pkts, tx_ring->stats.tx_bytes,
2074 tx_ring->stats.sw_err_cnt, tx_ring->pending_buf);
2077 "seg_pkt_cnt: %llu, tx_more: %llu, restart_queue: %llu, tx_busy: %llu\n",
2078 tx_ring->stats.seg_pkt_cnt, tx_ring->stats.tx_more,
2079 tx_ring->stats.restart_queue, tx_ring->stats.tx_busy);
2081 /* When mac received many pause frames continuous, it's unable to send
2082 * packets, which may cause tx timeout
2084 if (h->ae_algo->ops->get_mac_stats) {
2085 struct hns3_mac_stats mac_stats;
2087 h->ae_algo->ops->get_mac_stats(h, &mac_stats);
2088 netdev_info(ndev, "tx_pause_cnt: %llu, rx_pause_cnt: %llu\n",
2089 mac_stats.tx_pause_cnt, mac_stats.rx_pause_cnt);
2092 hw_head = readl_relaxed(tx_ring->tqp->io_base +
2093 HNS3_RING_TX_RING_HEAD_REG);
2094 hw_tail = readl_relaxed(tx_ring->tqp->io_base +
2095 HNS3_RING_TX_RING_TAIL_REG);
2096 fbd_num = readl_relaxed(tx_ring->tqp->io_base +
2097 HNS3_RING_TX_RING_FBDNUM_REG);
2098 fbd_oft = readl_relaxed(tx_ring->tqp->io_base +
2099 HNS3_RING_TX_RING_OFFSET_REG);
2100 ebd_num = readl_relaxed(tx_ring->tqp->io_base +
2101 HNS3_RING_TX_RING_EBDNUM_REG);
2102 ebd_oft = readl_relaxed(tx_ring->tqp->io_base +
2103 HNS3_RING_TX_RING_EBD_OFFSET_REG);
2104 bd_num = readl_relaxed(tx_ring->tqp->io_base +
2105 HNS3_RING_TX_RING_BD_NUM_REG);
2106 bd_err = readl_relaxed(tx_ring->tqp->io_base +
2107 HNS3_RING_TX_RING_BD_ERR_REG);
2108 ring_en = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_EN_REG);
2109 tc = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_TX_RING_TC_REG);
2112 "BD_NUM: 0x%x HW_HEAD: 0x%x, HW_TAIL: 0x%x, BD_ERR: 0x%x, INT: 0x%x\n",
2113 bd_num, hw_head, hw_tail, bd_err,
2114 readl(tx_ring->tqp_vector->mask_addr));
2116 "RING_EN: 0x%x, TC: 0x%x, FBD_NUM: 0x%x FBD_OFT: 0x%x, EBD_NUM: 0x%x, EBD_OFT: 0x%x\n",
2117 ring_en, tc, fbd_num, fbd_oft, ebd_num, ebd_oft);
2122 static void hns3_nic_net_timeout(struct net_device *ndev, unsigned int txqueue)
2124 struct hns3_nic_priv *priv = netdev_priv(ndev);
2125 struct hnae3_handle *h = priv->ae_handle;
2127 if (!hns3_get_tx_timeo_queue_info(ndev))
2130 /* request the reset, and let the hclge to determine
2131 * which reset level should be done
2133 if (h->ae_algo->ops->reset_event)
2134 h->ae_algo->ops->reset_event(h->pdev, h);
2137 #ifdef CONFIG_RFS_ACCEL
2138 static int hns3_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
2139 u16 rxq_index, u32 flow_id)
2141 struct hnae3_handle *h = hns3_get_handle(dev);
2142 struct flow_keys fkeys;
2144 if (!h->ae_algo->ops->add_arfs_entry)
2147 if (skb->encapsulation)
2148 return -EPROTONOSUPPORT;
2150 if (!skb_flow_dissect_flow_keys(skb, &fkeys, 0))
2151 return -EPROTONOSUPPORT;
2153 if ((fkeys.basic.n_proto != htons(ETH_P_IP) &&
2154 fkeys.basic.n_proto != htons(ETH_P_IPV6)) ||
2155 (fkeys.basic.ip_proto != IPPROTO_TCP &&
2156 fkeys.basic.ip_proto != IPPROTO_UDP))
2157 return -EPROTONOSUPPORT;
2159 return h->ae_algo->ops->add_arfs_entry(h, rxq_index, flow_id, &fkeys);
2163 static int hns3_nic_get_vf_config(struct net_device *ndev, int vf,
2164 struct ifla_vf_info *ivf)
2166 struct hnae3_handle *h = hns3_get_handle(ndev);
2168 if (!h->ae_algo->ops->get_vf_config)
2171 return h->ae_algo->ops->get_vf_config(h, vf, ivf);
2174 static int hns3_nic_set_vf_link_state(struct net_device *ndev, int vf,
2177 struct hnae3_handle *h = hns3_get_handle(ndev);
2179 if (!h->ae_algo->ops->set_vf_link_state)
2182 return h->ae_algo->ops->set_vf_link_state(h, vf, link_state);
2185 static int hns3_nic_set_vf_rate(struct net_device *ndev, int vf,
2186 int min_tx_rate, int max_tx_rate)
2188 struct hnae3_handle *h = hns3_get_handle(ndev);
2190 if (!h->ae_algo->ops->set_vf_rate)
2193 return h->ae_algo->ops->set_vf_rate(h, vf, min_tx_rate, max_tx_rate,
2197 static int hns3_nic_set_vf_mac(struct net_device *netdev, int vf_id, u8 *mac)
2199 struct hnae3_handle *h = hns3_get_handle(netdev);
2201 if (!h->ae_algo->ops->set_vf_mac)
2204 if (is_multicast_ether_addr(mac)) {
2206 "Invalid MAC:%pM specified. Could not set MAC\n",
2211 return h->ae_algo->ops->set_vf_mac(h, vf_id, mac);
2214 static const struct net_device_ops hns3_nic_netdev_ops = {
2215 .ndo_open = hns3_nic_net_open,
2216 .ndo_stop = hns3_nic_net_stop,
2217 .ndo_start_xmit = hns3_nic_net_xmit,
2218 .ndo_tx_timeout = hns3_nic_net_timeout,
2219 .ndo_set_mac_address = hns3_nic_net_set_mac_address,
2220 .ndo_do_ioctl = hns3_nic_do_ioctl,
2221 .ndo_change_mtu = hns3_nic_change_mtu,
2222 .ndo_set_features = hns3_nic_set_features,
2223 .ndo_features_check = hns3_features_check,
2224 .ndo_get_stats64 = hns3_nic_get_stats64,
2225 .ndo_setup_tc = hns3_nic_setup_tc,
2226 .ndo_set_rx_mode = hns3_nic_set_rx_mode,
2227 .ndo_vlan_rx_add_vid = hns3_vlan_rx_add_vid,
2228 .ndo_vlan_rx_kill_vid = hns3_vlan_rx_kill_vid,
2229 .ndo_set_vf_vlan = hns3_ndo_set_vf_vlan,
2230 .ndo_set_vf_spoofchk = hns3_set_vf_spoofchk,
2231 .ndo_set_vf_trust = hns3_set_vf_trust,
2232 #ifdef CONFIG_RFS_ACCEL
2233 .ndo_rx_flow_steer = hns3_rx_flow_steer,
2235 .ndo_get_vf_config = hns3_nic_get_vf_config,
2236 .ndo_set_vf_link_state = hns3_nic_set_vf_link_state,
2237 .ndo_set_vf_rate = hns3_nic_set_vf_rate,
2238 .ndo_set_vf_mac = hns3_nic_set_vf_mac,
2241 bool hns3_is_phys_func(struct pci_dev *pdev)
2243 u32 dev_id = pdev->device;
2246 case HNAE3_DEV_ID_GE:
2247 case HNAE3_DEV_ID_25GE:
2248 case HNAE3_DEV_ID_25GE_RDMA:
2249 case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
2250 case HNAE3_DEV_ID_50GE_RDMA:
2251 case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
2252 case HNAE3_DEV_ID_100G_RDMA_MACSEC:
2253 case HNAE3_DEV_ID_200G_RDMA:
2255 case HNAE3_DEV_ID_VF:
2256 case HNAE3_DEV_ID_RDMA_DCB_PFC_VF:
2259 dev_warn(&pdev->dev, "un-recognized pci device-id %u",
2266 static void hns3_disable_sriov(struct pci_dev *pdev)
2268 /* If our VFs are assigned we cannot shut down SR-IOV
2269 * without causing issues, so just leave the hardware
2270 * available but disabled
2272 if (pci_vfs_assigned(pdev)) {
2273 dev_warn(&pdev->dev,
2274 "disabling driver while VFs are assigned\n");
2278 pci_disable_sriov(pdev);
2281 /* hns3_probe - Device initialization routine
2282 * @pdev: PCI device information struct
2283 * @ent: entry in hns3_pci_tbl
2285 * hns3_probe initializes a PF identified by a pci_dev structure.
2286 * The OS initialization, configuring of the PF private structure,
2287 * and a hardware reset occur.
2289 * Returns 0 on success, negative on failure
2291 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2293 struct hnae3_ae_dev *ae_dev;
2296 ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev), GFP_KERNEL);
2300 ae_dev->pdev = pdev;
2301 ae_dev->flag = ent->driver_data;
2302 pci_set_drvdata(pdev, ae_dev);
2304 ret = hnae3_register_ae_dev(ae_dev);
2306 pci_set_drvdata(pdev, NULL);
2311 /* hns3_remove - Device removal routine
2312 * @pdev: PCI device information struct
2314 static void hns3_remove(struct pci_dev *pdev)
2316 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2318 if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
2319 hns3_disable_sriov(pdev);
2321 hnae3_unregister_ae_dev(ae_dev);
2322 pci_set_drvdata(pdev, NULL);
2326 * hns3_pci_sriov_configure
2327 * @pdev: pointer to a pci_dev structure
2328 * @num_vfs: number of VFs to allocate
2330 * Enable or change the number of VFs. Called when the user updates the number
2333 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
2337 if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
2338 dev_warn(&pdev->dev, "Can not config SRIOV\n");
2343 ret = pci_enable_sriov(pdev, num_vfs);
2345 dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
2348 } else if (!pci_vfs_assigned(pdev)) {
2349 pci_disable_sriov(pdev);
2351 dev_warn(&pdev->dev,
2352 "Unable to free VFs because some are assigned to VMs.\n");
2358 static void hns3_shutdown(struct pci_dev *pdev)
2360 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2362 hnae3_unregister_ae_dev(ae_dev);
2363 pci_set_drvdata(pdev, NULL);
2365 if (system_state == SYSTEM_POWER_OFF)
2366 pci_set_power_state(pdev, PCI_D3hot);
2369 static int __maybe_unused hns3_suspend(struct device *dev)
2371 struct hnae3_ae_dev *ae_dev = dev_get_drvdata(dev);
2373 if (ae_dev && hns3_is_phys_func(ae_dev->pdev)) {
2374 dev_info(dev, "Begin to suspend.\n");
2375 if (ae_dev->ops && ae_dev->ops->reset_prepare)
2376 ae_dev->ops->reset_prepare(ae_dev, HNAE3_FUNC_RESET);
2382 static int __maybe_unused hns3_resume(struct device *dev)
2384 struct hnae3_ae_dev *ae_dev = dev_get_drvdata(dev);
2386 if (ae_dev && hns3_is_phys_func(ae_dev->pdev)) {
2387 dev_info(dev, "Begin to resume.\n");
2388 if (ae_dev->ops && ae_dev->ops->reset_done)
2389 ae_dev->ops->reset_done(ae_dev);
2395 static pci_ers_result_t hns3_error_detected(struct pci_dev *pdev,
2396 pci_channel_state_t state)
2398 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2399 pci_ers_result_t ret;
2401 dev_info(&pdev->dev, "PCI error detected, state(=%u)!!\n", state);
2403 if (state == pci_channel_io_perm_failure)
2404 return PCI_ERS_RESULT_DISCONNECT;
2406 if (!ae_dev || !ae_dev->ops) {
2408 "Can't recover - error happened before device initialized\n");
2409 return PCI_ERS_RESULT_NONE;
2412 if (ae_dev->ops->handle_hw_ras_error)
2413 ret = ae_dev->ops->handle_hw_ras_error(ae_dev);
2415 return PCI_ERS_RESULT_NONE;
2420 static pci_ers_result_t hns3_slot_reset(struct pci_dev *pdev)
2422 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2423 const struct hnae3_ae_ops *ops;
2424 enum hnae3_reset_type reset_type;
2425 struct device *dev = &pdev->dev;
2427 if (!ae_dev || !ae_dev->ops)
2428 return PCI_ERS_RESULT_NONE;
2431 /* request the reset */
2432 if (ops->reset_event && ops->get_reset_level &&
2433 ops->set_default_reset_request) {
2434 if (ae_dev->hw_err_reset_req) {
2435 reset_type = ops->get_reset_level(ae_dev,
2436 &ae_dev->hw_err_reset_req);
2437 ops->set_default_reset_request(ae_dev, reset_type);
2438 dev_info(dev, "requesting reset due to PCI error\n");
2439 ops->reset_event(pdev, NULL);
2442 return PCI_ERS_RESULT_RECOVERED;
2445 return PCI_ERS_RESULT_DISCONNECT;
2448 static void hns3_reset_prepare(struct pci_dev *pdev)
2450 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2452 dev_info(&pdev->dev, "FLR prepare\n");
2453 if (ae_dev && ae_dev->ops && ae_dev->ops->reset_prepare)
2454 ae_dev->ops->reset_prepare(ae_dev, HNAE3_FLR_RESET);
2457 static void hns3_reset_done(struct pci_dev *pdev)
2459 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2461 dev_info(&pdev->dev, "FLR done\n");
2462 if (ae_dev && ae_dev->ops && ae_dev->ops->reset_done)
2463 ae_dev->ops->reset_done(ae_dev);
2466 static const struct pci_error_handlers hns3_err_handler = {
2467 .error_detected = hns3_error_detected,
2468 .slot_reset = hns3_slot_reset,
2469 .reset_prepare = hns3_reset_prepare,
2470 .reset_done = hns3_reset_done,
2473 static SIMPLE_DEV_PM_OPS(hns3_pm_ops, hns3_suspend, hns3_resume);
2475 static struct pci_driver hns3_driver = {
2476 .name = hns3_driver_name,
2477 .id_table = hns3_pci_tbl,
2478 .probe = hns3_probe,
2479 .remove = hns3_remove,
2480 .shutdown = hns3_shutdown,
2481 .driver.pm = &hns3_pm_ops,
2482 .sriov_configure = hns3_pci_sriov_configure,
2483 .err_handler = &hns3_err_handler,
2486 /* set default feature to hns3 */
2487 static void hns3_set_default_feature(struct net_device *netdev)
2489 struct hnae3_handle *h = hns3_get_handle(netdev);
2490 struct pci_dev *pdev = h->pdev;
2491 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2493 netdev->priv_flags |= IFF_UNICAST_FLT;
2495 netdev->hw_enc_features |= NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2496 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2497 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2498 NETIF_F_SCTP_CRC | NETIF_F_TSO_MANGLEID | NETIF_F_FRAGLIST;
2500 netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
2502 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
2503 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
2504 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2505 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2506 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2507 NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
2509 netdev->vlan_features |= NETIF_F_RXCSUM |
2510 NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO |
2511 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2512 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2513 NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
2515 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX |
2516 NETIF_F_HW_VLAN_CTAG_RX |
2517 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2518 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2519 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2520 NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
2522 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V2) {
2523 netdev->hw_features |= NETIF_F_GRO_HW;
2524 netdev->features |= NETIF_F_GRO_HW;
2526 if (!(h->flags & HNAE3_SUPPORT_VF)) {
2527 netdev->hw_features |= NETIF_F_NTUPLE;
2528 netdev->features |= NETIF_F_NTUPLE;
2532 if (test_bit(HNAE3_DEV_SUPPORT_UDP_GSO_B, ae_dev->caps)) {
2533 netdev->hw_features |= NETIF_F_GSO_UDP_L4;
2534 netdev->features |= NETIF_F_GSO_UDP_L4;
2535 netdev->vlan_features |= NETIF_F_GSO_UDP_L4;
2536 netdev->hw_enc_features |= NETIF_F_GSO_UDP_L4;
2539 if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps)) {
2540 netdev->hw_features |= NETIF_F_HW_CSUM;
2541 netdev->features |= NETIF_F_HW_CSUM;
2542 netdev->vlan_features |= NETIF_F_HW_CSUM;
2543 netdev->hw_enc_features |= NETIF_F_HW_CSUM;
2545 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2546 netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2547 netdev->vlan_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2548 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2551 if (test_bit(HNAE3_DEV_SUPPORT_UDP_TUNNEL_CSUM_B, ae_dev->caps)) {
2552 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
2553 netdev->features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
2554 netdev->vlan_features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
2555 netdev->hw_enc_features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
2558 if (test_bit(HNAE3_DEV_SUPPORT_FD_FORWARD_TC_B, ae_dev->caps)) {
2559 netdev->hw_features |= NETIF_F_HW_TC;
2560 netdev->features |= NETIF_F_HW_TC;
2564 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
2565 struct hns3_desc_cb *cb)
2567 unsigned int order = hns3_page_order(ring);
2570 p = dev_alloc_pages(order);
2575 cb->page_offset = 0;
2577 cb->buf = page_address(p);
2578 cb->length = hns3_page_size(ring);
2579 cb->type = DESC_TYPE_PAGE;
2580 page_ref_add(p, USHRT_MAX - 1);
2581 cb->pagecnt_bias = USHRT_MAX;
2586 static void hns3_free_buffer(struct hns3_enet_ring *ring,
2587 struct hns3_desc_cb *cb, int budget)
2589 if (cb->type == DESC_TYPE_SKB)
2590 napi_consume_skb(cb->priv, budget);
2591 else if (!HNAE3_IS_TX_RING(ring) && cb->pagecnt_bias)
2592 __page_frag_cache_drain(cb->priv, cb->pagecnt_bias);
2593 memset(cb, 0, sizeof(*cb));
2596 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
2598 cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
2599 cb->length, ring_to_dma_dir(ring));
2601 if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
2607 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
2608 struct hns3_desc_cb *cb)
2610 if (cb->type == DESC_TYPE_SKB || cb->type == DESC_TYPE_FRAGLIST_SKB)
2611 dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
2612 ring_to_dma_dir(ring));
2613 else if (cb->length)
2614 dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
2615 ring_to_dma_dir(ring));
2618 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
2620 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2621 ring->desc[i].addr = 0;
2624 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i,
2627 struct hns3_desc_cb *cb = &ring->desc_cb[i];
2629 if (!ring->desc_cb[i].dma)
2632 hns3_buffer_detach(ring, i);
2633 hns3_free_buffer(ring, cb, budget);
2636 static void hns3_free_buffers(struct hns3_enet_ring *ring)
2640 for (i = 0; i < ring->desc_num; i++)
2641 hns3_free_buffer_detach(ring, i, 0);
2644 /* free desc along with its attached buffer */
2645 static void hns3_free_desc(struct hns3_enet_ring *ring)
2647 int size = ring->desc_num * sizeof(ring->desc[0]);
2649 hns3_free_buffers(ring);
2652 dma_free_coherent(ring_to_dev(ring), size,
2653 ring->desc, ring->desc_dma_addr);
2658 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
2660 int size = ring->desc_num * sizeof(ring->desc[0]);
2662 ring->desc = dma_alloc_coherent(ring_to_dev(ring), size,
2663 &ring->desc_dma_addr, GFP_KERNEL);
2670 static int hns3_alloc_and_map_buffer(struct hns3_enet_ring *ring,
2671 struct hns3_desc_cb *cb)
2675 ret = hns3_alloc_buffer(ring, cb);
2679 ret = hns3_map_buffer(ring, cb);
2686 hns3_free_buffer(ring, cb, 0);
2691 static int hns3_alloc_and_attach_buffer(struct hns3_enet_ring *ring, int i)
2693 int ret = hns3_alloc_and_map_buffer(ring, &ring->desc_cb[i]);
2698 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2703 /* Allocate memory for raw pkg, and map with dma */
2704 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
2708 for (i = 0; i < ring->desc_num; i++) {
2709 ret = hns3_alloc_and_attach_buffer(ring, i);
2711 goto out_buffer_fail;
2717 for (j = i - 1; j >= 0; j--)
2718 hns3_free_buffer_detach(ring, j, 0);
2722 /* detach a in-used buffer and replace with a reserved one */
2723 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
2724 struct hns3_desc_cb *res_cb)
2726 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2727 ring->desc_cb[i] = *res_cb;
2728 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2729 ring->desc[i].rx.bd_base_info = 0;
2732 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
2734 ring->desc_cb[i].reuse_flag = 0;
2735 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
2736 ring->desc_cb[i].page_offset);
2737 ring->desc[i].rx.bd_base_info = 0;
2739 dma_sync_single_for_device(ring_to_dev(ring),
2740 ring->desc_cb[i].dma + ring->desc_cb[i].page_offset,
2741 hns3_buf_size(ring),
2745 static bool hns3_nic_reclaim_desc(struct hns3_enet_ring *ring,
2746 int *bytes, int *pkts, int budget)
2748 /* pair with ring->last_to_use update in hns3_tx_doorbell(),
2749 * smp_store_release() is not used in hns3_tx_doorbell() because
2750 * the doorbell operation already have the needed barrier operation.
2752 int ltu = smp_load_acquire(&ring->last_to_use);
2753 int ntc = ring->next_to_clean;
2754 struct hns3_desc_cb *desc_cb;
2755 bool reclaimed = false;
2756 struct hns3_desc *desc;
2758 while (ltu != ntc) {
2759 desc = &ring->desc[ntc];
2761 if (le16_to_cpu(desc->tx.bdtp_fe_sc_vld_ra_ri) &
2762 BIT(HNS3_TXD_VLD_B))
2765 desc_cb = &ring->desc_cb[ntc];
2767 if (desc_cb->type == DESC_TYPE_SKB) {
2769 (*bytes) += desc_cb->send_bytes;
2772 /* desc_cb will be cleaned, after hnae3_free_buffer_detach */
2773 hns3_free_buffer_detach(ring, ntc, budget);
2775 if (++ntc == ring->desc_num)
2778 /* Issue prefetch for next Tx descriptor */
2779 prefetch(&ring->desc_cb[ntc]);
2783 if (unlikely(!reclaimed))
2786 /* This smp_store_release() pairs with smp_load_acquire() in
2787 * ring_space called by hns3_nic_net_xmit.
2789 smp_store_release(&ring->next_to_clean, ntc);
2793 void hns3_clean_tx_ring(struct hns3_enet_ring *ring, int budget)
2795 struct net_device *netdev = ring_to_netdev(ring);
2796 struct hns3_nic_priv *priv = netdev_priv(netdev);
2797 struct netdev_queue *dev_queue;
2803 if (unlikely(!hns3_nic_reclaim_desc(ring, &bytes, &pkts, budget)))
2806 ring->tqp_vector->tx_group.total_bytes += bytes;
2807 ring->tqp_vector->tx_group.total_packets += pkts;
2809 u64_stats_update_begin(&ring->syncp);
2810 ring->stats.tx_bytes += bytes;
2811 ring->stats.tx_pkts += pkts;
2812 u64_stats_update_end(&ring->syncp);
2814 dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
2815 netdev_tx_completed_queue(dev_queue, pkts, bytes);
2817 if (unlikely(netif_carrier_ok(netdev) &&
2818 ring_space(ring) > HNS3_MAX_TSO_BD_NUM)) {
2819 /* Make sure that anybody stopping the queue after this
2820 * sees the new next_to_clean.
2823 if (netif_tx_queue_stopped(dev_queue) &&
2824 !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
2825 netif_tx_wake_queue(dev_queue);
2826 ring->stats.restart_queue++;
2831 static int hns3_desc_unused(struct hns3_enet_ring *ring)
2833 int ntc = ring->next_to_clean;
2834 int ntu = ring->next_to_use;
2836 return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
2839 static void hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring,
2842 struct hns3_desc_cb *desc_cb;
2843 struct hns3_desc_cb res_cbs;
2846 for (i = 0; i < cleand_count; i++) {
2847 desc_cb = &ring->desc_cb[ring->next_to_use];
2848 if (desc_cb->reuse_flag) {
2849 u64_stats_update_begin(&ring->syncp);
2850 ring->stats.reuse_pg_cnt++;
2851 u64_stats_update_end(&ring->syncp);
2853 hns3_reuse_buffer(ring, ring->next_to_use);
2855 ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
2857 u64_stats_update_begin(&ring->syncp);
2858 ring->stats.sw_err_cnt++;
2859 u64_stats_update_end(&ring->syncp);
2861 hns3_rl_err(ring_to_netdev(ring),
2862 "alloc rx buffer failed: %d\n",
2866 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
2868 u64_stats_update_begin(&ring->syncp);
2869 ring->stats.non_reuse_pg++;
2870 u64_stats_update_end(&ring->syncp);
2873 ring_ptr_move_fw(ring, next_to_use);
2876 writel(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
2879 static bool hns3_can_reuse_page(struct hns3_desc_cb *cb)
2881 return (page_count(cb->priv) - cb->pagecnt_bias) == 1;
2884 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
2885 struct hns3_enet_ring *ring, int pull_len,
2886 struct hns3_desc_cb *desc_cb)
2888 struct hns3_desc *desc = &ring->desc[ring->next_to_clean];
2889 int size = le16_to_cpu(desc->rx.size);
2890 u32 truesize = hns3_buf_size(ring);
2892 desc_cb->pagecnt_bias--;
2893 skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
2894 size - pull_len, truesize);
2896 /* Avoid re-using remote and pfmemalloc pages, or the stack is still
2897 * using the page when page_offset rollback to zero, flag default
2900 if (!dev_page_is_reusable(desc_cb->priv) ||
2901 (!desc_cb->page_offset && !hns3_can_reuse_page(desc_cb))) {
2902 __page_frag_cache_drain(desc_cb->priv, desc_cb->pagecnt_bias);
2906 /* Move offset up to the next cache line */
2907 desc_cb->page_offset += truesize;
2909 if (desc_cb->page_offset + truesize <= hns3_page_size(ring)) {
2910 desc_cb->reuse_flag = 1;
2911 } else if (hns3_can_reuse_page(desc_cb)) {
2912 desc_cb->reuse_flag = 1;
2913 desc_cb->page_offset = 0;
2914 } else if (desc_cb->pagecnt_bias) {
2915 __page_frag_cache_drain(desc_cb->priv, desc_cb->pagecnt_bias);
2919 if (unlikely(!desc_cb->pagecnt_bias)) {
2920 page_ref_add(desc_cb->priv, USHRT_MAX);
2921 desc_cb->pagecnt_bias = USHRT_MAX;
2925 static int hns3_gro_complete(struct sk_buff *skb, u32 l234info)
2927 __be16 type = skb->protocol;
2931 while (eth_type_vlan(type)) {
2932 struct vlan_hdr *vh;
2934 if ((depth + VLAN_HLEN) > skb_headlen(skb))
2937 vh = (struct vlan_hdr *)(skb->data + depth);
2938 type = vh->h_vlan_encapsulated_proto;
2942 skb_set_network_header(skb, depth);
2944 if (type == htons(ETH_P_IP)) {
2945 const struct iphdr *iph = ip_hdr(skb);
2947 depth += sizeof(struct iphdr);
2948 skb_set_transport_header(skb, depth);
2950 th->check = ~tcp_v4_check(skb->len - depth, iph->saddr,
2952 } else if (type == htons(ETH_P_IPV6)) {
2953 const struct ipv6hdr *iph = ipv6_hdr(skb);
2955 depth += sizeof(struct ipv6hdr);
2956 skb_set_transport_header(skb, depth);
2958 th->check = ~tcp_v6_check(skb->len - depth, &iph->saddr,
2961 hns3_rl_err(skb->dev,
2962 "Error: FW GRO supports only IPv4/IPv6, not 0x%04x, depth: %d\n",
2963 be16_to_cpu(type), depth);
2967 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
2969 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
2971 if (l234info & BIT(HNS3_RXD_GRO_FIXID_B))
2972 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_FIXEDID;
2974 skb->csum_start = (unsigned char *)th - skb->head;
2975 skb->csum_offset = offsetof(struct tcphdr, check);
2976 skb->ip_summed = CHECKSUM_PARTIAL;
2978 trace_hns3_gro(skb);
2983 static void hns3_checksum_complete(struct hns3_enet_ring *ring,
2984 struct sk_buff *skb, u32 l234info)
2988 u64_stats_update_begin(&ring->syncp);
2989 ring->stats.csum_complete++;
2990 u64_stats_update_end(&ring->syncp);
2991 skb->ip_summed = CHECKSUM_COMPLETE;
2992 lo = hnae3_get_field(l234info, HNS3_RXD_L2_CSUM_L_M,
2993 HNS3_RXD_L2_CSUM_L_S);
2994 hi = hnae3_get_field(l234info, HNS3_RXD_L2_CSUM_H_M,
2995 HNS3_RXD_L2_CSUM_H_S);
2996 skb->csum = csum_unfold((__force __sum16)(lo | hi << 8));
2999 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
3000 u32 l234info, u32 bd_base_info, u32 ol_info)
3002 struct net_device *netdev = ring_to_netdev(ring);
3003 int l3_type, l4_type;
3006 skb->ip_summed = CHECKSUM_NONE;
3008 skb_checksum_none_assert(skb);
3010 if (!(netdev->features & NETIF_F_RXCSUM))
3013 if (l234info & BIT(HNS3_RXD_L2_CSUM_B)) {
3014 hns3_checksum_complete(ring, skb, l234info);
3018 /* check if hardware has done checksum */
3019 if (!(bd_base_info & BIT(HNS3_RXD_L3L4P_B)))
3022 if (unlikely(l234info & (BIT(HNS3_RXD_L3E_B) | BIT(HNS3_RXD_L4E_B) |
3023 BIT(HNS3_RXD_OL3E_B) |
3024 BIT(HNS3_RXD_OL4E_B)))) {
3025 u64_stats_update_begin(&ring->syncp);
3026 ring->stats.l3l4_csum_err++;
3027 u64_stats_update_end(&ring->syncp);
3032 ol4_type = hnae3_get_field(ol_info, HNS3_RXD_OL4ID_M,
3035 case HNS3_OL4_TYPE_MAC_IN_UDP:
3036 case HNS3_OL4_TYPE_NVGRE:
3037 skb->csum_level = 1;
3039 case HNS3_OL4_TYPE_NO_TUN:
3040 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
3042 l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
3044 /* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
3045 if ((l3_type == HNS3_L3_TYPE_IPV4 ||
3046 l3_type == HNS3_L3_TYPE_IPV6) &&
3047 (l4_type == HNS3_L4_TYPE_UDP ||
3048 l4_type == HNS3_L4_TYPE_TCP ||
3049 l4_type == HNS3_L4_TYPE_SCTP))
3050 skb->ip_summed = CHECKSUM_UNNECESSARY;
3057 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
3059 if (skb_has_frag_list(skb))
3060 napi_gro_flush(&ring->tqp_vector->napi, false);
3062 napi_gro_receive(&ring->tqp_vector->napi, skb);
3065 static bool hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
3066 struct hns3_desc *desc, u32 l234info,
3069 struct hnae3_handle *handle = ring->tqp->handle;
3070 struct pci_dev *pdev = ring->tqp->handle->pdev;
3071 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3073 if (unlikely(ae_dev->dev_version < HNAE3_DEVICE_VERSION_V2)) {
3074 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3075 if (!(*vlan_tag & VLAN_VID_MASK))
3076 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3078 return (*vlan_tag != 0);
3081 #define HNS3_STRP_OUTER_VLAN 0x1
3082 #define HNS3_STRP_INNER_VLAN 0x2
3083 #define HNS3_STRP_BOTH 0x3
3085 /* Hardware always insert VLAN tag into RX descriptor when
3086 * remove the tag from packet, driver needs to determine
3087 * reporting which tag to stack.
3089 switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
3090 HNS3_RXD_STRP_TAGP_S)) {
3091 case HNS3_STRP_OUTER_VLAN:
3092 if (handle->port_base_vlan_state !=
3093 HNAE3_PORT_BASE_VLAN_DISABLE)
3096 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3098 case HNS3_STRP_INNER_VLAN:
3099 if (handle->port_base_vlan_state !=
3100 HNAE3_PORT_BASE_VLAN_DISABLE)
3103 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3105 case HNS3_STRP_BOTH:
3106 if (handle->port_base_vlan_state ==
3107 HNAE3_PORT_BASE_VLAN_DISABLE)
3108 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3110 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3118 static void hns3_rx_ring_move_fw(struct hns3_enet_ring *ring)
3120 ring->desc[ring->next_to_clean].rx.bd_base_info &=
3121 cpu_to_le32(~BIT(HNS3_RXD_VLD_B));
3122 ring->next_to_clean += 1;
3124 if (unlikely(ring->next_to_clean == ring->desc_num))
3125 ring->next_to_clean = 0;
3128 static int hns3_alloc_skb(struct hns3_enet_ring *ring, unsigned int length,
3131 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
3132 struct net_device *netdev = ring_to_netdev(ring);
3133 struct sk_buff *skb;
3135 ring->skb = napi_alloc_skb(&ring->tqp_vector->napi, HNS3_RX_HEAD_SIZE);
3137 if (unlikely(!skb)) {
3138 hns3_rl_err(netdev, "alloc rx skb fail\n");
3140 u64_stats_update_begin(&ring->syncp);
3141 ring->stats.sw_err_cnt++;
3142 u64_stats_update_end(&ring->syncp);
3147 trace_hns3_rx_desc(ring);
3148 prefetchw(skb->data);
3150 ring->pending_buf = 1;
3152 ring->tail_skb = NULL;
3153 if (length <= HNS3_RX_HEAD_SIZE) {
3154 memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
3156 /* We can reuse buffer as-is, just make sure it is reusable */
3157 if (dev_page_is_reusable(desc_cb->priv))
3158 desc_cb->reuse_flag = 1;
3159 else /* This page cannot be reused so discard it */
3160 __page_frag_cache_drain(desc_cb->priv,
3161 desc_cb->pagecnt_bias);
3163 hns3_rx_ring_move_fw(ring);
3166 u64_stats_update_begin(&ring->syncp);
3167 ring->stats.seg_pkt_cnt++;
3168 u64_stats_update_end(&ring->syncp);
3170 ring->pull_len = eth_get_headlen(netdev, va, HNS3_RX_HEAD_SIZE);
3171 __skb_put(skb, ring->pull_len);
3172 hns3_nic_reuse_page(skb, ring->frag_num++, ring, ring->pull_len,
3174 hns3_rx_ring_move_fw(ring);
3179 static int hns3_add_frag(struct hns3_enet_ring *ring)
3181 struct sk_buff *skb = ring->skb;
3182 struct sk_buff *head_skb = skb;
3183 struct sk_buff *new_skb;
3184 struct hns3_desc_cb *desc_cb;
3185 struct hns3_desc *desc;
3189 desc = &ring->desc[ring->next_to_clean];
3190 desc_cb = &ring->desc_cb[ring->next_to_clean];
3191 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
3192 /* make sure HW write desc complete */
3194 if (!(bd_base_info & BIT(HNS3_RXD_VLD_B)))
3197 if (unlikely(ring->frag_num >= MAX_SKB_FRAGS)) {
3198 new_skb = napi_alloc_skb(&ring->tqp_vector->napi, 0);
3199 if (unlikely(!new_skb)) {
3200 hns3_rl_err(ring_to_netdev(ring),
3201 "alloc rx fraglist skb fail\n");
3206 if (ring->tail_skb) {
3207 ring->tail_skb->next = new_skb;
3208 ring->tail_skb = new_skb;
3210 skb_shinfo(skb)->frag_list = new_skb;
3211 ring->tail_skb = new_skb;
3215 if (ring->tail_skb) {
3216 head_skb->truesize += hns3_buf_size(ring);
3217 head_skb->data_len += le16_to_cpu(desc->rx.size);
3218 head_skb->len += le16_to_cpu(desc->rx.size);
3219 skb = ring->tail_skb;
3222 dma_sync_single_for_cpu(ring_to_dev(ring),
3223 desc_cb->dma + desc_cb->page_offset,
3224 hns3_buf_size(ring),
3227 hns3_nic_reuse_page(skb, ring->frag_num++, ring, 0, desc_cb);
3228 trace_hns3_rx_desc(ring);
3229 hns3_rx_ring_move_fw(ring);
3230 ring->pending_buf++;
3231 } while (!(bd_base_info & BIT(HNS3_RXD_FE_B)));
3236 static int hns3_set_gro_and_checksum(struct hns3_enet_ring *ring,
3237 struct sk_buff *skb, u32 l234info,
3238 u32 bd_base_info, u32 ol_info)
3242 skb_shinfo(skb)->gso_size = hnae3_get_field(bd_base_info,
3243 HNS3_RXD_GRO_SIZE_M,
3244 HNS3_RXD_GRO_SIZE_S);
3245 /* if there is no HW GRO, do not set gro params */
3246 if (!skb_shinfo(skb)->gso_size) {
3247 hns3_rx_checksum(ring, skb, l234info, bd_base_info, ol_info);
3251 NAPI_GRO_CB(skb)->count = hnae3_get_field(l234info,
3252 HNS3_RXD_GRO_COUNT_M,
3253 HNS3_RXD_GRO_COUNT_S);
3255 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M, HNS3_RXD_L3ID_S);
3256 if (l3_type == HNS3_L3_TYPE_IPV4)
3257 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
3258 else if (l3_type == HNS3_L3_TYPE_IPV6)
3259 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
3263 return hns3_gro_complete(skb, l234info);
3266 static void hns3_set_rx_skb_rss_type(struct hns3_enet_ring *ring,
3267 struct sk_buff *skb, u32 rss_hash)
3269 struct hnae3_handle *handle = ring->tqp->handle;
3270 enum pkt_hash_types rss_type;
3273 rss_type = handle->kinfo.rss_type;
3275 rss_type = PKT_HASH_TYPE_NONE;
3277 skb_set_hash(skb, rss_hash, rss_type);
3280 static int hns3_handle_bdinfo(struct hns3_enet_ring *ring, struct sk_buff *skb)
3282 struct net_device *netdev = ring_to_netdev(ring);
3283 enum hns3_pkt_l2t_type l2_frame_type;
3284 u32 bd_base_info, l234info, ol_info;
3285 struct hns3_desc *desc;
3289 /* bdinfo handled below is only valid on the last BD of the
3290 * current packet, and ring->next_to_clean indicates the first
3291 * descriptor of next packet, so need - 1 below.
3293 pre_ntc = ring->next_to_clean ? (ring->next_to_clean - 1) :
3294 (ring->desc_num - 1);
3295 desc = &ring->desc[pre_ntc];
3296 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
3297 l234info = le32_to_cpu(desc->rx.l234_info);
3298 ol_info = le32_to_cpu(desc->rx.ol_info);
3300 /* Based on hw strategy, the tag offloaded will be stored at
3301 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
3302 * in one layer tag case.
3304 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
3307 if (hns3_parse_vlan_tag(ring, desc, l234info, &vlan_tag))
3308 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
3312 if (unlikely(!desc->rx.pkt_len || (l234info & (BIT(HNS3_RXD_TRUNCAT_B) |
3313 BIT(HNS3_RXD_L2E_B))))) {
3314 u64_stats_update_begin(&ring->syncp);
3315 if (l234info & BIT(HNS3_RXD_L2E_B))
3316 ring->stats.l2_err++;
3318 ring->stats.err_pkt_len++;
3319 u64_stats_update_end(&ring->syncp);
3326 /* Do update ip stack process */
3327 skb->protocol = eth_type_trans(skb, netdev);
3329 /* This is needed in order to enable forwarding support */
3330 ret = hns3_set_gro_and_checksum(ring, skb, l234info,
3331 bd_base_info, ol_info);
3332 if (unlikely(ret)) {
3333 u64_stats_update_begin(&ring->syncp);
3334 ring->stats.rx_err_cnt++;
3335 u64_stats_update_end(&ring->syncp);
3339 l2_frame_type = hnae3_get_field(l234info, HNS3_RXD_DMAC_M,
3342 u64_stats_update_begin(&ring->syncp);
3343 ring->stats.rx_pkts++;
3344 ring->stats.rx_bytes += len;
3346 if (l2_frame_type == HNS3_L2_TYPE_MULTICAST)
3347 ring->stats.rx_multicast++;
3349 u64_stats_update_end(&ring->syncp);
3351 ring->tqp_vector->rx_group.total_bytes += len;
3353 hns3_set_rx_skb_rss_type(ring, skb, le32_to_cpu(desc->rx.rss_hash));
3357 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring)
3359 struct sk_buff *skb = ring->skb;
3360 struct hns3_desc_cb *desc_cb;
3361 struct hns3_desc *desc;
3362 unsigned int length;
3366 desc = &ring->desc[ring->next_to_clean];
3367 desc_cb = &ring->desc_cb[ring->next_to_clean];
3372 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
3373 /* Check valid BD */
3374 if (unlikely(!(bd_base_info & BIT(HNS3_RXD_VLD_B))))
3378 length = le16_to_cpu(desc->rx.size);
3380 ring->va = desc_cb->buf + desc_cb->page_offset;
3382 dma_sync_single_for_cpu(ring_to_dev(ring),
3383 desc_cb->dma + desc_cb->page_offset,
3384 hns3_buf_size(ring),
3387 /* Prefetch first cache line of first page.
3388 * Idea is to cache few bytes of the header of the packet.
3389 * Our L1 Cache line size is 64B so need to prefetch twice to make
3390 * it 128B. But in actual we can have greater size of caches with
3391 * 128B Level 1 cache lines. In such a case, single fetch would
3392 * suffice to cache in the relevant part of the header.
3394 net_prefetch(ring->va);
3396 ret = hns3_alloc_skb(ring, length, ring->va);
3399 if (ret < 0) /* alloc buffer fail */
3401 if (!(bd_base_info & BIT(HNS3_RXD_FE_B))) { /* need add frag */
3402 ret = hns3_add_frag(ring);
3407 ret = hns3_add_frag(ring);
3412 /* As the head data may be changed when GRO enable, copy
3413 * the head data in after other data rx completed
3415 if (skb->len > HNS3_RX_HEAD_SIZE)
3416 memcpy(skb->data, ring->va,
3417 ALIGN(ring->pull_len, sizeof(long)));
3419 ret = hns3_handle_bdinfo(ring, skb);
3420 if (unlikely(ret)) {
3421 dev_kfree_skb_any(skb);
3425 skb_record_rx_queue(skb, ring->tqp->tqp_index);
3429 int hns3_clean_rx_ring(struct hns3_enet_ring *ring, int budget,
3430 void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
3432 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
3433 int unused_count = hns3_desc_unused(ring);
3437 unused_count -= ring->pending_buf;
3439 while (recv_pkts < budget) {
3440 /* Reuse or realloc buffers */
3441 if (unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
3442 hns3_nic_alloc_rx_buffers(ring, unused_count);
3443 unused_count = hns3_desc_unused(ring) -
3448 err = hns3_handle_rx_bd(ring);
3449 /* Do not get FE for the packet or failed to alloc skb */
3450 if (unlikely(!ring->skb || err == -ENXIO)) {
3452 } else if (likely(!err)) {
3453 rx_fn(ring, ring->skb);
3457 unused_count += ring->pending_buf;
3459 ring->pending_buf = 0;
3463 /* Make all data has been write before submit */
3464 if (unused_count > 0)
3465 hns3_nic_alloc_rx_buffers(ring, unused_count);
3470 static bool hns3_get_new_flow_lvl(struct hns3_enet_ring_group *ring_group)
3472 #define HNS3_RX_LOW_BYTE_RATE 10000
3473 #define HNS3_RX_MID_BYTE_RATE 20000
3474 #define HNS3_RX_ULTRA_PACKET_RATE 40
3476 enum hns3_flow_level_range new_flow_level;
3477 struct hns3_enet_tqp_vector *tqp_vector;
3478 int packets_per_msecs, bytes_per_msecs;
3481 tqp_vector = ring_group->ring->tqp_vector;
3483 jiffies_to_msecs(jiffies - tqp_vector->last_jiffies);
3484 if (!time_passed_ms)
3487 do_div(ring_group->total_packets, time_passed_ms);
3488 packets_per_msecs = ring_group->total_packets;
3490 do_div(ring_group->total_bytes, time_passed_ms);
3491 bytes_per_msecs = ring_group->total_bytes;
3493 new_flow_level = ring_group->coal.flow_level;
3495 /* Simple throttlerate management
3496 * 0-10MB/s lower (50000 ints/s)
3497 * 10-20MB/s middle (20000 ints/s)
3498 * 20-1249MB/s high (18000 ints/s)
3499 * > 40000pps ultra (8000 ints/s)
3501 switch (new_flow_level) {
3503 if (bytes_per_msecs > HNS3_RX_LOW_BYTE_RATE)
3504 new_flow_level = HNS3_FLOW_MID;
3507 if (bytes_per_msecs > HNS3_RX_MID_BYTE_RATE)
3508 new_flow_level = HNS3_FLOW_HIGH;
3509 else if (bytes_per_msecs <= HNS3_RX_LOW_BYTE_RATE)
3510 new_flow_level = HNS3_FLOW_LOW;
3512 case HNS3_FLOW_HIGH:
3513 case HNS3_FLOW_ULTRA:
3515 if (bytes_per_msecs <= HNS3_RX_MID_BYTE_RATE)
3516 new_flow_level = HNS3_FLOW_MID;
3520 if (packets_per_msecs > HNS3_RX_ULTRA_PACKET_RATE &&
3521 &tqp_vector->rx_group == ring_group)
3522 new_flow_level = HNS3_FLOW_ULTRA;
3524 ring_group->total_bytes = 0;
3525 ring_group->total_packets = 0;
3526 ring_group->coal.flow_level = new_flow_level;
3531 static bool hns3_get_new_int_gl(struct hns3_enet_ring_group *ring_group)
3533 struct hns3_enet_tqp_vector *tqp_vector;
3536 if (!ring_group->ring)
3539 tqp_vector = ring_group->ring->tqp_vector;
3540 if (!tqp_vector->last_jiffies)
3543 if (ring_group->total_packets == 0) {
3544 ring_group->coal.int_gl = HNS3_INT_GL_50K;
3545 ring_group->coal.flow_level = HNS3_FLOW_LOW;
3549 if (!hns3_get_new_flow_lvl(ring_group))
3552 new_int_gl = ring_group->coal.int_gl;
3553 switch (ring_group->coal.flow_level) {
3555 new_int_gl = HNS3_INT_GL_50K;
3558 new_int_gl = HNS3_INT_GL_20K;
3560 case HNS3_FLOW_HIGH:
3561 new_int_gl = HNS3_INT_GL_18K;
3563 case HNS3_FLOW_ULTRA:
3564 new_int_gl = HNS3_INT_GL_8K;
3570 if (new_int_gl != ring_group->coal.int_gl) {
3571 ring_group->coal.int_gl = new_int_gl;
3577 static void hns3_update_new_int_gl(struct hns3_enet_tqp_vector *tqp_vector)
3579 struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
3580 struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
3581 bool rx_update, tx_update;
3583 /* update param every 1000ms */
3584 if (time_before(jiffies,
3585 tqp_vector->last_jiffies + msecs_to_jiffies(1000)))
3588 if (rx_group->coal.adapt_enable) {
3589 rx_update = hns3_get_new_int_gl(rx_group);
3591 hns3_set_vector_coalesce_rx_gl(tqp_vector,
3592 rx_group->coal.int_gl);
3595 if (tx_group->coal.adapt_enable) {
3596 tx_update = hns3_get_new_int_gl(tx_group);
3598 hns3_set_vector_coalesce_tx_gl(tqp_vector,
3599 tx_group->coal.int_gl);
3602 tqp_vector->last_jiffies = jiffies;
3605 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
3607 struct hns3_nic_priv *priv = netdev_priv(napi->dev);
3608 struct hns3_enet_ring *ring;
3609 int rx_pkt_total = 0;
3611 struct hns3_enet_tqp_vector *tqp_vector =
3612 container_of(napi, struct hns3_enet_tqp_vector, napi);
3613 bool clean_complete = true;
3614 int rx_budget = budget;
3616 if (unlikely(test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
3617 napi_complete(napi);
3621 /* Since the actual Tx work is minimal, we can give the Tx a larger
3622 * budget and be more aggressive about cleaning up the Tx descriptors.
3624 hns3_for_each_ring(ring, tqp_vector->tx_group)
3625 hns3_clean_tx_ring(ring, budget);
3627 /* make sure rx ring budget not smaller than 1 */
3628 if (tqp_vector->num_tqps > 1)
3629 rx_budget = max(budget / tqp_vector->num_tqps, 1);
3631 hns3_for_each_ring(ring, tqp_vector->rx_group) {
3632 int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
3634 if (rx_cleaned >= rx_budget)
3635 clean_complete = false;
3637 rx_pkt_total += rx_cleaned;
3640 tqp_vector->rx_group.total_packets += rx_pkt_total;
3642 if (!clean_complete)
3645 if (napi_complete(napi) &&
3646 likely(!test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
3647 hns3_update_new_int_gl(tqp_vector);
3648 hns3_mask_vector_irq(tqp_vector, 1);
3651 return rx_pkt_total;
3654 static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
3655 struct hnae3_ring_chain_node *head)
3657 struct pci_dev *pdev = tqp_vector->handle->pdev;
3658 struct hnae3_ring_chain_node *cur_chain = head;
3659 struct hnae3_ring_chain_node *chain;
3660 struct hns3_enet_ring *tx_ring;
3661 struct hns3_enet_ring *rx_ring;
3663 tx_ring = tqp_vector->tx_group.ring;
3665 cur_chain->tqp_index = tx_ring->tqp->tqp_index;
3666 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
3667 HNAE3_RING_TYPE_TX);
3668 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3669 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_TX);
3671 cur_chain->next = NULL;
3673 while (tx_ring->next) {
3674 tx_ring = tx_ring->next;
3676 chain = devm_kzalloc(&pdev->dev, sizeof(*chain),
3679 goto err_free_chain;
3681 cur_chain->next = chain;
3682 chain->tqp_index = tx_ring->tqp->tqp_index;
3683 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
3684 HNAE3_RING_TYPE_TX);
3685 hnae3_set_field(chain->int_gl_idx,
3686 HNAE3_RING_GL_IDX_M,
3687 HNAE3_RING_GL_IDX_S,
3694 rx_ring = tqp_vector->rx_group.ring;
3695 if (!tx_ring && rx_ring) {
3696 cur_chain->next = NULL;
3697 cur_chain->tqp_index = rx_ring->tqp->tqp_index;
3698 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
3699 HNAE3_RING_TYPE_RX);
3700 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3701 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
3703 rx_ring = rx_ring->next;
3707 chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
3709 goto err_free_chain;
3711 cur_chain->next = chain;
3712 chain->tqp_index = rx_ring->tqp->tqp_index;
3713 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
3714 HNAE3_RING_TYPE_RX);
3715 hnae3_set_field(chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3716 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
3720 rx_ring = rx_ring->next;
3726 cur_chain = head->next;
3728 chain = cur_chain->next;
3729 devm_kfree(&pdev->dev, cur_chain);
3737 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
3738 struct hnae3_ring_chain_node *head)
3740 struct pci_dev *pdev = tqp_vector->handle->pdev;
3741 struct hnae3_ring_chain_node *chain_tmp, *chain;
3746 chain_tmp = chain->next;
3747 devm_kfree(&pdev->dev, chain);
3752 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
3753 struct hns3_enet_ring *ring)
3755 ring->next = group->ring;
3761 static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
3763 struct pci_dev *pdev = priv->ae_handle->pdev;
3764 struct hns3_enet_tqp_vector *tqp_vector;
3765 int num_vectors = priv->vector_num;
3769 numa_node = dev_to_node(&pdev->dev);
3771 for (vector_i = 0; vector_i < num_vectors; vector_i++) {
3772 tqp_vector = &priv->tqp_vector[vector_i];
3773 cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
3774 &tqp_vector->affinity_mask);
3778 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
3780 struct hnae3_handle *h = priv->ae_handle;
3781 struct hns3_enet_tqp_vector *tqp_vector;
3785 hns3_nic_set_cpumask(priv);
3787 for (i = 0; i < priv->vector_num; i++) {
3788 tqp_vector = &priv->tqp_vector[i];
3789 hns3_vector_coalesce_init_hw(tqp_vector, priv);
3790 tqp_vector->num_tqps = 0;
3793 for (i = 0; i < h->kinfo.num_tqps; i++) {
3794 u16 vector_i = i % priv->vector_num;
3795 u16 tqp_num = h->kinfo.num_tqps;
3797 tqp_vector = &priv->tqp_vector[vector_i];
3799 hns3_add_ring_to_group(&tqp_vector->tx_group,
3802 hns3_add_ring_to_group(&tqp_vector->rx_group,
3803 &priv->ring[i + tqp_num]);
3805 priv->ring[i].tqp_vector = tqp_vector;
3806 priv->ring[i + tqp_num].tqp_vector = tqp_vector;
3807 tqp_vector->num_tqps++;
3810 for (i = 0; i < priv->vector_num; i++) {
3811 struct hnae3_ring_chain_node vector_ring_chain;
3813 tqp_vector = &priv->tqp_vector[i];
3815 tqp_vector->rx_group.total_bytes = 0;
3816 tqp_vector->rx_group.total_packets = 0;
3817 tqp_vector->tx_group.total_bytes = 0;
3818 tqp_vector->tx_group.total_packets = 0;
3819 tqp_vector->handle = h;
3821 ret = hns3_get_vector_ring_chain(tqp_vector,
3822 &vector_ring_chain);
3826 ret = h->ae_algo->ops->map_ring_to_vector(h,
3827 tqp_vector->vector_irq, &vector_ring_chain);
3829 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3834 netif_napi_add(priv->netdev, &tqp_vector->napi,
3835 hns3_nic_common_poll, NAPI_POLL_WEIGHT);
3842 netif_napi_del(&priv->tqp_vector[i].napi);
3847 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
3849 struct hnae3_handle *h = priv->ae_handle;
3850 struct hns3_enet_tqp_vector *tqp_vector;
3851 struct hnae3_vector_info *vector;
3852 struct pci_dev *pdev = h->pdev;
3853 u16 tqp_num = h->kinfo.num_tqps;
3858 /* RSS size, cpu online and vector_num should be the same */
3859 /* Should consider 2p/4p later */
3860 vector_num = min_t(u16, num_online_cpus(), tqp_num);
3862 vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
3867 /* save the actual available vector number */
3868 vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
3870 priv->vector_num = vector_num;
3871 priv->tqp_vector = (struct hns3_enet_tqp_vector *)
3872 devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
3874 if (!priv->tqp_vector) {
3879 for (i = 0; i < priv->vector_num; i++) {
3880 tqp_vector = &priv->tqp_vector[i];
3881 tqp_vector->idx = i;
3882 tqp_vector->mask_addr = vector[i].io_addr;
3883 tqp_vector->vector_irq = vector[i].vector;
3884 hns3_vector_coalesce_init(tqp_vector, priv);
3888 devm_kfree(&pdev->dev, vector);
3892 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
3898 static void hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
3900 struct hnae3_ring_chain_node vector_ring_chain;
3901 struct hnae3_handle *h = priv->ae_handle;
3902 struct hns3_enet_tqp_vector *tqp_vector;
3905 for (i = 0; i < priv->vector_num; i++) {
3906 tqp_vector = &priv->tqp_vector[i];
3908 if (!tqp_vector->rx_group.ring && !tqp_vector->tx_group.ring)
3911 /* Since the mapping can be overwritten, when fail to get the
3912 * chain between vector and ring, we should go on to deal with
3913 * the remaining options.
3915 if (hns3_get_vector_ring_chain(tqp_vector, &vector_ring_chain))
3916 dev_warn(priv->dev, "failed to get ring chain\n");
3918 h->ae_algo->ops->unmap_ring_from_vector(h,
3919 tqp_vector->vector_irq, &vector_ring_chain);
3921 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3923 hns3_clear_ring_group(&tqp_vector->rx_group);
3924 hns3_clear_ring_group(&tqp_vector->tx_group);
3925 netif_napi_del(&priv->tqp_vector[i].napi);
3929 static void hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
3931 struct hnae3_handle *h = priv->ae_handle;
3932 struct pci_dev *pdev = h->pdev;
3935 for (i = 0; i < priv->vector_num; i++) {
3936 struct hns3_enet_tqp_vector *tqp_vector;
3938 tqp_vector = &priv->tqp_vector[i];
3939 ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
3944 devm_kfree(&pdev->dev, priv->tqp_vector);
3947 static void hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
3948 unsigned int ring_type)
3950 int queue_num = priv->ae_handle->kinfo.num_tqps;
3951 struct hns3_enet_ring *ring;
3954 if (ring_type == HNAE3_RING_TYPE_TX) {
3955 ring = &priv->ring[q->tqp_index];
3956 desc_num = priv->ae_handle->kinfo.num_tx_desc;
3957 ring->queue_index = q->tqp_index;
3959 ring = &priv->ring[q->tqp_index + queue_num];
3960 desc_num = priv->ae_handle->kinfo.num_rx_desc;
3961 ring->queue_index = q->tqp_index;
3964 hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
3968 ring->desc_cb = NULL;
3969 ring->dev = priv->dev;
3970 ring->desc_dma_addr = 0;
3971 ring->buf_size = q->buf_size;
3972 ring->desc_num = desc_num;
3973 ring->next_to_use = 0;
3974 ring->next_to_clean = 0;
3975 ring->last_to_use = 0;
3978 static void hns3_queue_to_ring(struct hnae3_queue *tqp,
3979 struct hns3_nic_priv *priv)
3981 hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
3982 hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
3985 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
3987 struct hnae3_handle *h = priv->ae_handle;
3988 struct pci_dev *pdev = h->pdev;
3991 priv->ring = devm_kzalloc(&pdev->dev,
3992 array3_size(h->kinfo.num_tqps,
3993 sizeof(*priv->ring), 2),
3998 for (i = 0; i < h->kinfo.num_tqps; i++)
3999 hns3_queue_to_ring(h->kinfo.tqp[i], priv);
4004 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
4009 devm_kfree(priv->dev, priv->ring);
4013 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
4017 if (ring->desc_num <= 0 || ring->buf_size <= 0)
4020 ring->desc_cb = devm_kcalloc(ring_to_dev(ring), ring->desc_num,
4021 sizeof(ring->desc_cb[0]), GFP_KERNEL);
4022 if (!ring->desc_cb) {
4027 ret = hns3_alloc_desc(ring);
4029 goto out_with_desc_cb;
4031 if (!HNAE3_IS_TX_RING(ring)) {
4032 ret = hns3_alloc_ring_buffers(ring);
4040 hns3_free_desc(ring);
4042 devm_kfree(ring_to_dev(ring), ring->desc_cb);
4043 ring->desc_cb = NULL;
4048 void hns3_fini_ring(struct hns3_enet_ring *ring)
4050 hns3_free_desc(ring);
4051 devm_kfree(ring_to_dev(ring), ring->desc_cb);
4052 ring->desc_cb = NULL;
4053 ring->next_to_clean = 0;
4054 ring->next_to_use = 0;
4055 ring->last_to_use = 0;
4056 ring->pending_buf = 0;
4058 dev_kfree_skb_any(ring->skb);
4063 static int hns3_buf_size2type(u32 buf_size)
4069 bd_size_type = HNS3_BD_SIZE_512_TYPE;
4072 bd_size_type = HNS3_BD_SIZE_1024_TYPE;
4075 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4078 bd_size_type = HNS3_BD_SIZE_4096_TYPE;
4081 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4084 return bd_size_type;
4087 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
4089 dma_addr_t dma = ring->desc_dma_addr;
4090 struct hnae3_queue *q = ring->tqp;
4092 if (!HNAE3_IS_TX_RING(ring)) {
4093 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG, (u32)dma);
4094 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
4095 (u32)((dma >> 31) >> 1));
4097 hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
4098 hns3_buf_size2type(ring->buf_size));
4099 hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
4100 ring->desc_num / 8 - 1);
4102 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
4104 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
4105 (u32)((dma >> 31) >> 1));
4107 hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
4108 ring->desc_num / 8 - 1);
4112 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
4114 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
4115 struct hnae3_tc_info *tc_info = &kinfo->tc_info;
4118 for (i = 0; i < HNAE3_MAX_TC; i++) {
4121 if (!test_bit(i, &tc_info->tc_en))
4124 for (j = 0; j < tc_info->tqp_count[i]; j++) {
4125 struct hnae3_queue *q;
4127 q = priv->ring[tc_info->tqp_offset[i] + j].tqp;
4128 hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG, i);
4133 int hns3_init_all_ring(struct hns3_nic_priv *priv)
4135 struct hnae3_handle *h = priv->ae_handle;
4136 int ring_num = h->kinfo.num_tqps * 2;
4140 for (i = 0; i < ring_num; i++) {
4141 ret = hns3_alloc_ring_memory(&priv->ring[i]);
4144 "Alloc ring memory fail! ret=%d\n", ret);
4145 goto out_when_alloc_ring_memory;
4148 u64_stats_init(&priv->ring[i].syncp);
4153 out_when_alloc_ring_memory:
4154 for (j = i - 1; j >= 0; j--)
4155 hns3_fini_ring(&priv->ring[j]);
4160 static void hns3_uninit_all_ring(struct hns3_nic_priv *priv)
4162 struct hnae3_handle *h = priv->ae_handle;
4165 for (i = 0; i < h->kinfo.num_tqps; i++) {
4166 hns3_fini_ring(&priv->ring[i]);
4167 hns3_fini_ring(&priv->ring[i + h->kinfo.num_tqps]);
4171 /* Set mac addr if it is configured. or leave it to the AE driver */
4172 static int hns3_init_mac_addr(struct net_device *netdev)
4174 struct hns3_nic_priv *priv = netdev_priv(netdev);
4175 struct hnae3_handle *h = priv->ae_handle;
4176 u8 mac_addr_temp[ETH_ALEN];
4179 if (h->ae_algo->ops->get_mac_addr)
4180 h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
4182 /* Check if the MAC address is valid, if not get a random one */
4183 if (!is_valid_ether_addr(mac_addr_temp)) {
4184 eth_hw_addr_random(netdev);
4185 dev_warn(priv->dev, "using random MAC address %pM\n",
4187 } else if (!ether_addr_equal(netdev->dev_addr, mac_addr_temp)) {
4188 ether_addr_copy(netdev->dev_addr, mac_addr_temp);
4189 ether_addr_copy(netdev->perm_addr, mac_addr_temp);
4194 if (h->ae_algo->ops->set_mac_addr)
4195 ret = h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
4200 static int hns3_init_phy(struct net_device *netdev)
4202 struct hnae3_handle *h = hns3_get_handle(netdev);
4205 if (h->ae_algo->ops->mac_connect_phy)
4206 ret = h->ae_algo->ops->mac_connect_phy(h);
4211 static void hns3_uninit_phy(struct net_device *netdev)
4213 struct hnae3_handle *h = hns3_get_handle(netdev);
4215 if (h->ae_algo->ops->mac_disconnect_phy)
4216 h->ae_algo->ops->mac_disconnect_phy(h);
4219 static int hns3_client_start(struct hnae3_handle *handle)
4221 if (!handle->ae_algo->ops->client_start)
4224 return handle->ae_algo->ops->client_start(handle);
4227 static void hns3_client_stop(struct hnae3_handle *handle)
4229 if (!handle->ae_algo->ops->client_stop)
4232 handle->ae_algo->ops->client_stop(handle);
4235 static void hns3_info_show(struct hns3_nic_priv *priv)
4237 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
4239 dev_info(priv->dev, "MAC address: %pM\n", priv->netdev->dev_addr);
4240 dev_info(priv->dev, "Task queue pairs numbers: %u\n", kinfo->num_tqps);
4241 dev_info(priv->dev, "RSS size: %u\n", kinfo->rss_size);
4242 dev_info(priv->dev, "Allocated RSS size: %u\n", kinfo->req_rss_size);
4243 dev_info(priv->dev, "RX buffer length: %u\n", kinfo->rx_buf_len);
4244 dev_info(priv->dev, "Desc num per TX queue: %u\n", kinfo->num_tx_desc);
4245 dev_info(priv->dev, "Desc num per RX queue: %u\n", kinfo->num_rx_desc);
4246 dev_info(priv->dev, "Total number of enabled TCs: %u\n",
4247 kinfo->tc_info.num_tc);
4248 dev_info(priv->dev, "Max mtu size: %u\n", priv->netdev->max_mtu);
4251 static int hns3_client_init(struct hnae3_handle *handle)
4253 struct pci_dev *pdev = handle->pdev;
4254 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
4255 u16 alloc_tqps, max_rss_size;
4256 struct hns3_nic_priv *priv;
4257 struct net_device *netdev;
4260 handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
4262 netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
4266 priv = netdev_priv(netdev);
4267 priv->dev = &pdev->dev;
4268 priv->netdev = netdev;
4269 priv->ae_handle = handle;
4270 priv->tx_timeout_count = 0;
4271 priv->max_non_tso_bd_num = ae_dev->dev_specs.max_non_tso_bd_num;
4272 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
4274 handle->msg_enable = netif_msg_init(debug, DEFAULT_MSG_LEVEL);
4276 handle->kinfo.netdev = netdev;
4277 handle->priv = (void *)priv;
4279 hns3_init_mac_addr(netdev);
4281 hns3_set_default_feature(netdev);
4283 netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
4284 netdev->priv_flags |= IFF_UNICAST_FLT;
4285 netdev->netdev_ops = &hns3_nic_netdev_ops;
4286 SET_NETDEV_DEV(netdev, &pdev->dev);
4287 hns3_ethtool_set_ops(netdev);
4289 /* Carrier off reporting is important to ethtool even BEFORE open */
4290 netif_carrier_off(netdev);
4292 ret = hns3_get_ring_config(priv);
4295 goto out_get_ring_cfg;
4298 ret = hns3_nic_alloc_vector_data(priv);
4301 goto out_alloc_vector_data;
4304 ret = hns3_nic_init_vector_data(priv);
4307 goto out_init_vector_data;
4310 ret = hns3_init_all_ring(priv);
4316 ret = hns3_init_phy(netdev);
4320 ret = register_netdev(netdev);
4322 dev_err(priv->dev, "probe register netdev fail!\n");
4323 goto out_reg_netdev_fail;
4326 /* the device can work without cpu rmap, only aRFS needs it */
4327 ret = hns3_set_rx_cpu_rmap(netdev);
4329 dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
4331 ret = hns3_nic_init_irq(priv);
4333 dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
4334 hns3_free_rx_cpu_rmap(netdev);
4335 goto out_init_irq_fail;
4338 ret = hns3_client_start(handle);
4340 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
4341 goto out_client_start;
4344 hns3_dcbnl_setup(handle);
4346 hns3_dbg_init(handle);
4348 netdev->max_mtu = HNS3_MAX_MTU(ae_dev->dev_specs.max_frm_size);
4350 if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps))
4351 set_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state);
4353 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
4355 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
4356 set_bit(HNAE3_PFLAG_LIMIT_PROMISC, &handle->supported_pflags);
4358 if (netif_msg_drv(handle))
4359 hns3_info_show(priv);
4364 hns3_free_rx_cpu_rmap(netdev);
4365 hns3_nic_uninit_irq(priv);
4367 unregister_netdev(netdev);
4368 out_reg_netdev_fail:
4369 hns3_uninit_phy(netdev);
4371 hns3_uninit_all_ring(priv);
4373 hns3_nic_uninit_vector_data(priv);
4374 out_init_vector_data:
4375 hns3_nic_dealloc_vector_data(priv);
4376 out_alloc_vector_data:
4379 priv->ae_handle = NULL;
4380 free_netdev(netdev);
4384 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
4386 struct net_device *netdev = handle->kinfo.netdev;
4387 struct hns3_nic_priv *priv = netdev_priv(netdev);
4389 if (netdev->reg_state != NETREG_UNINITIALIZED)
4390 unregister_netdev(netdev);
4392 hns3_client_stop(handle);
4394 hns3_uninit_phy(netdev);
4396 if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
4397 netdev_warn(netdev, "already uninitialized\n");
4398 goto out_netdev_free;
4401 hns3_free_rx_cpu_rmap(netdev);
4403 hns3_nic_uninit_irq(priv);
4405 hns3_clear_all_ring(handle, true);
4407 hns3_nic_uninit_vector_data(priv);
4409 hns3_nic_dealloc_vector_data(priv);
4411 hns3_uninit_all_ring(priv);
4413 hns3_put_ring_config(priv);
4416 hns3_dbg_uninit(handle);
4417 free_netdev(netdev);
4420 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
4422 struct net_device *netdev = handle->kinfo.netdev;
4428 netif_tx_wake_all_queues(netdev);
4429 netif_carrier_on(netdev);
4430 if (netif_msg_link(handle))
4431 netdev_info(netdev, "link up\n");
4433 netif_carrier_off(netdev);
4434 netif_tx_stop_all_queues(netdev);
4435 if (netif_msg_link(handle))
4436 netdev_info(netdev, "link down\n");
4440 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
4442 while (ring->next_to_clean != ring->next_to_use) {
4443 ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
4444 hns3_free_buffer_detach(ring, ring->next_to_clean, 0);
4445 ring_ptr_move_fw(ring, next_to_clean);
4448 ring->pending_buf = 0;
4451 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
4453 struct hns3_desc_cb res_cbs;
4456 while (ring->next_to_use != ring->next_to_clean) {
4457 /* When a buffer is not reused, it's memory has been
4458 * freed in hns3_handle_rx_bd or will be freed by
4459 * stack, so we need to replace the buffer here.
4461 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
4462 ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
4464 u64_stats_update_begin(&ring->syncp);
4465 ring->stats.sw_err_cnt++;
4466 u64_stats_update_end(&ring->syncp);
4467 /* if alloc new buffer fail, exit directly
4468 * and reclear in up flow.
4470 netdev_warn(ring_to_netdev(ring),
4471 "reserve buffer map failed, ret = %d\n",
4475 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
4477 ring_ptr_move_fw(ring, next_to_use);
4480 /* Free the pending skb in rx ring */
4482 dev_kfree_skb_any(ring->skb);
4484 ring->pending_buf = 0;
4490 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
4492 while (ring->next_to_use != ring->next_to_clean) {
4493 /* When a buffer is not reused, it's memory has been
4494 * freed in hns3_handle_rx_bd or will be freed by
4495 * stack, so only need to unmap the buffer here.
4497 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
4498 hns3_unmap_buffer(ring,
4499 &ring->desc_cb[ring->next_to_use]);
4500 ring->desc_cb[ring->next_to_use].dma = 0;
4503 ring_ptr_move_fw(ring, next_to_use);
4507 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force)
4509 struct net_device *ndev = h->kinfo.netdev;
4510 struct hns3_nic_priv *priv = netdev_priv(ndev);
4513 for (i = 0; i < h->kinfo.num_tqps; i++) {
4514 struct hns3_enet_ring *ring;
4516 ring = &priv->ring[i];
4517 hns3_clear_tx_ring(ring);
4519 ring = &priv->ring[i + h->kinfo.num_tqps];
4520 /* Continue to clear other rings even if clearing some
4524 hns3_force_clear_rx_ring(ring);
4526 hns3_clear_rx_ring(ring);
4530 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
4532 struct net_device *ndev = h->kinfo.netdev;
4533 struct hns3_nic_priv *priv = netdev_priv(ndev);
4534 struct hns3_enet_ring *rx_ring;
4538 ret = h->ae_algo->ops->reset_queue(h);
4542 for (i = 0; i < h->kinfo.num_tqps; i++) {
4543 hns3_init_ring_hw(&priv->ring[i]);
4545 /* We need to clear tx ring here because self test will
4546 * use the ring and will not run down before up
4548 hns3_clear_tx_ring(&priv->ring[i]);
4549 priv->ring[i].next_to_clean = 0;
4550 priv->ring[i].next_to_use = 0;
4551 priv->ring[i].last_to_use = 0;
4553 rx_ring = &priv->ring[i + h->kinfo.num_tqps];
4554 hns3_init_ring_hw(rx_ring);
4555 ret = hns3_clear_rx_ring(rx_ring);
4559 /* We can not know the hardware head and tail when this
4560 * function is called in reset flow, so we reuse all desc.
4562 for (j = 0; j < rx_ring->desc_num; j++)
4563 hns3_reuse_buffer(rx_ring, j);
4565 rx_ring->next_to_clean = 0;
4566 rx_ring->next_to_use = 0;
4569 hns3_init_tx_ring_tc(priv);
4574 static void hns3_store_coal(struct hns3_nic_priv *priv)
4576 /* ethtool only support setting and querying one coal
4577 * configuration for now, so save the vector 0' coal
4578 * configuration here in order to restore it.
4580 memcpy(&priv->tx_coal, &priv->tqp_vector[0].tx_group.coal,
4581 sizeof(struct hns3_enet_coalesce));
4582 memcpy(&priv->rx_coal, &priv->tqp_vector[0].rx_group.coal,
4583 sizeof(struct hns3_enet_coalesce));
4586 static void hns3_restore_coal(struct hns3_nic_priv *priv)
4588 u16 vector_num = priv->vector_num;
4591 for (i = 0; i < vector_num; i++) {
4592 memcpy(&priv->tqp_vector[i].tx_group.coal, &priv->tx_coal,
4593 sizeof(struct hns3_enet_coalesce));
4594 memcpy(&priv->tqp_vector[i].rx_group.coal, &priv->rx_coal,
4595 sizeof(struct hns3_enet_coalesce));
4599 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
4601 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4602 struct net_device *ndev = kinfo->netdev;
4603 struct hns3_nic_priv *priv = netdev_priv(ndev);
4605 if (test_and_set_bit(HNS3_NIC_STATE_RESETTING, &priv->state))
4608 if (!netif_running(ndev))
4611 return hns3_nic_net_stop(ndev);
4614 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
4616 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4617 struct hns3_nic_priv *priv = netdev_priv(kinfo->netdev);
4620 if (!test_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
4621 netdev_err(kinfo->netdev, "device is not initialized yet\n");
4625 clear_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4627 if (netif_running(kinfo->netdev)) {
4628 ret = hns3_nic_net_open(kinfo->netdev);
4630 set_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4631 netdev_err(kinfo->netdev,
4632 "net up fail, ret=%d!\n", ret);
4640 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
4642 struct net_device *netdev = handle->kinfo.netdev;
4643 struct hns3_nic_priv *priv = netdev_priv(netdev);
4646 /* Carrier off reporting is important to ethtool even BEFORE open */
4647 netif_carrier_off(netdev);
4649 ret = hns3_get_ring_config(priv);
4653 ret = hns3_nic_alloc_vector_data(priv);
4657 hns3_restore_coal(priv);
4659 ret = hns3_nic_init_vector_data(priv);
4661 goto err_dealloc_vector;
4663 ret = hns3_init_all_ring(priv);
4665 goto err_uninit_vector;
4667 /* the device can work without cpu rmap, only aRFS needs it */
4668 ret = hns3_set_rx_cpu_rmap(netdev);
4670 dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
4672 ret = hns3_nic_init_irq(priv);
4674 dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
4675 hns3_free_rx_cpu_rmap(netdev);
4676 goto err_init_irq_fail;
4679 if (!hns3_is_phys_func(handle->pdev))
4680 hns3_init_mac_addr(netdev);
4682 ret = hns3_client_start(handle);
4684 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
4685 goto err_client_start_fail;
4688 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
4692 err_client_start_fail:
4693 hns3_free_rx_cpu_rmap(netdev);
4694 hns3_nic_uninit_irq(priv);
4696 hns3_uninit_all_ring(priv);
4698 hns3_nic_uninit_vector_data(priv);
4700 hns3_nic_dealloc_vector_data(priv);
4702 hns3_put_ring_config(priv);
4707 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
4709 struct net_device *netdev = handle->kinfo.netdev;
4710 struct hns3_nic_priv *priv = netdev_priv(netdev);
4712 if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
4713 netdev_warn(netdev, "already uninitialized\n");
4717 hns3_free_rx_cpu_rmap(netdev);
4718 hns3_nic_uninit_irq(priv);
4719 hns3_clear_all_ring(handle, true);
4720 hns3_reset_tx_queue(priv->ae_handle);
4722 hns3_nic_uninit_vector_data(priv);
4724 hns3_store_coal(priv);
4726 hns3_nic_dealloc_vector_data(priv);
4728 hns3_uninit_all_ring(priv);
4730 hns3_put_ring_config(priv);
4735 static int hns3_reset_notify(struct hnae3_handle *handle,
4736 enum hnae3_reset_notify_type type)
4741 case HNAE3_UP_CLIENT:
4742 ret = hns3_reset_notify_up_enet(handle);
4744 case HNAE3_DOWN_CLIENT:
4745 ret = hns3_reset_notify_down_enet(handle);
4747 case HNAE3_INIT_CLIENT:
4748 ret = hns3_reset_notify_init_enet(handle);
4750 case HNAE3_UNINIT_CLIENT:
4751 ret = hns3_reset_notify_uninit_enet(handle);
4760 static int hns3_change_channels(struct hnae3_handle *handle, u32 new_tqp_num,
4761 bool rxfh_configured)
4765 ret = handle->ae_algo->ops->set_channels(handle, new_tqp_num,
4768 dev_err(&handle->pdev->dev,
4769 "Change tqp num(%u) fail.\n", new_tqp_num);
4773 ret = hns3_reset_notify(handle, HNAE3_INIT_CLIENT);
4777 ret = hns3_reset_notify(handle, HNAE3_UP_CLIENT);
4779 hns3_reset_notify(handle, HNAE3_UNINIT_CLIENT);
4784 int hns3_set_channels(struct net_device *netdev,
4785 struct ethtool_channels *ch)
4787 struct hnae3_handle *h = hns3_get_handle(netdev);
4788 struct hnae3_knic_private_info *kinfo = &h->kinfo;
4789 bool rxfh_configured = netif_is_rxfh_configured(netdev);
4790 u32 new_tqp_num = ch->combined_count;
4794 if (hns3_nic_resetting(netdev))
4797 if (ch->rx_count || ch->tx_count)
4800 if (kinfo->tc_info.mqprio_active) {
4801 dev_err(&netdev->dev,
4802 "it's not allowed to set channels via ethtool when MQPRIO mode is on\n");
4806 if (new_tqp_num > hns3_get_max_available_channels(h) ||
4808 dev_err(&netdev->dev,
4809 "Change tqps fail, the tqp range is from 1 to %u",
4810 hns3_get_max_available_channels(h));
4814 if (kinfo->rss_size == new_tqp_num)
4817 netif_dbg(h, drv, netdev,
4818 "set channels: tqp_num=%u, rxfh=%d\n",
4819 new_tqp_num, rxfh_configured);
4821 ret = hns3_reset_notify(h, HNAE3_DOWN_CLIENT);
4825 ret = hns3_reset_notify(h, HNAE3_UNINIT_CLIENT);
4829 org_tqp_num = h->kinfo.num_tqps;
4830 ret = hns3_change_channels(h, new_tqp_num, rxfh_configured);
4835 "Change channels fail, revert to old value\n");
4836 ret1 = hns3_change_channels(h, org_tqp_num, rxfh_configured);
4839 "revert to old channel fail\n");
4849 static const struct hns3_hw_error_info hns3_hw_err[] = {
4850 { .type = HNAE3_PPU_POISON_ERROR,
4851 .msg = "PPU poison" },
4852 { .type = HNAE3_CMDQ_ECC_ERROR,
4853 .msg = "IMP CMDQ error" },
4854 { .type = HNAE3_IMP_RD_POISON_ERROR,
4855 .msg = "IMP RD poison" },
4856 { .type = HNAE3_ROCEE_AXI_RESP_ERROR,
4857 .msg = "ROCEE AXI RESP error" },
4860 static void hns3_process_hw_error(struct hnae3_handle *handle,
4861 enum hnae3_hw_error_type type)
4865 for (i = 0; i < ARRAY_SIZE(hns3_hw_err); i++) {
4866 if (hns3_hw_err[i].type == type) {
4867 dev_err(&handle->pdev->dev, "Detected %s!\n",
4868 hns3_hw_err[i].msg);
4874 static const struct hnae3_client_ops client_ops = {
4875 .init_instance = hns3_client_init,
4876 .uninit_instance = hns3_client_uninit,
4877 .link_status_change = hns3_link_status_change,
4878 .reset_notify = hns3_reset_notify,
4879 .process_hw_error = hns3_process_hw_error,
4882 /* hns3_init_module - Driver registration routine
4883 * hns3_init_module is the first routine called when the driver is
4884 * loaded. All it does is register with the PCI subsystem.
4886 static int __init hns3_init_module(void)
4890 pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
4891 pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
4893 client.type = HNAE3_CLIENT_KNIC;
4894 snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH, "%s",
4897 client.ops = &client_ops;
4899 INIT_LIST_HEAD(&client.node);
4901 hns3_dbg_register_debugfs(hns3_driver_name);
4903 ret = hnae3_register_client(&client);
4905 goto err_reg_client;
4907 ret = pci_register_driver(&hns3_driver);
4909 goto err_reg_driver;
4914 hnae3_unregister_client(&client);
4916 hns3_dbg_unregister_debugfs();
4919 module_init(hns3_init_module);
4921 /* hns3_exit_module - Driver exit cleanup routine
4922 * hns3_exit_module is called just before the driver is removed
4925 static void __exit hns3_exit_module(void)
4927 pci_unregister_driver(&hns3_driver);
4928 hnae3_unregister_client(&client);
4929 hns3_dbg_unregister_debugfs();
4931 module_exit(hns3_exit_module);
4933 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
4934 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
4935 MODULE_LICENSE("GPL");
4936 MODULE_ALIAS("pci:hns-nic");