1 // SPDX-License-Identifier: GPL-2.0+
2 // Copyright (c) 2016-2017 Hisilicon Limited.
4 #include <linux/dma-mapping.h>
5 #include <linux/etherdevice.h>
6 #include <linux/interrupt.h>
7 #ifdef CONFIG_RFS_ACCEL
8 #include <linux/cpu_rmap.h>
10 #include <linux/if_vlan.h>
11 #include <linux/irq.h>
13 #include <linux/ipv6.h>
14 #include <linux/module.h>
15 #include <linux/pci.h>
16 #include <linux/aer.h>
17 #include <linux/skbuff.h>
18 #include <linux/sctp.h>
20 #include <net/ip6_checksum.h>
21 #include <net/pkt_cls.h>
23 #include <net/vxlan.h>
24 #include <net/geneve.h>
27 #include "hns3_enet.h"
28 /* All hns3 tracepoints are defined by the include below, which
29 * must be included exactly once across the whole kernel with
30 * CREATE_TRACE_POINTS defined
32 #define CREATE_TRACE_POINTS
33 #include "hns3_trace.h"
35 #define hns3_set_field(origin, shift, val) ((origin) |= (val) << (shift))
36 #define hns3_tx_bd_count(S) DIV_ROUND_UP(S, HNS3_MAX_BD_SIZE)
38 #define hns3_rl_err(fmt, ...) \
40 if (net_ratelimit()) \
41 netdev_err(fmt, ##__VA_ARGS__); \
44 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force);
46 static const char hns3_driver_name[] = "hns3";
47 static const char hns3_driver_string[] =
48 "Hisilicon Ethernet Network Driver for Hip08 Family";
49 static const char hns3_copyright[] = "Copyright (c) 2017 Huawei Corporation.";
50 static struct hnae3_client client;
52 static int debug = -1;
53 module_param(debug, int, 0);
54 MODULE_PARM_DESC(debug, " Network interface message level setting");
56 #define DEFAULT_MSG_LEVEL (NETIF_MSG_PROBE | NETIF_MSG_LINK | \
57 NETIF_MSG_IFDOWN | NETIF_MSG_IFUP)
59 #define HNS3_INNER_VLAN_TAG 1
60 #define HNS3_OUTER_VLAN_TAG 2
62 #define HNS3_MIN_TX_LEN 33U
64 /* hns3_pci_tbl - PCI Device ID Table
66 * Last entry must be all 0s
68 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
69 * Class, Class Mask, private data (not used) }
71 static const struct pci_device_id hns3_pci_tbl[] = {
72 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
73 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
74 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA),
75 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
76 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC),
77 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
78 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA),
79 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
80 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC),
81 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
82 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC),
83 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
84 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_200G_RDMA),
85 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
86 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_VF), 0},
87 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_RDMA_DCB_PFC_VF),
88 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
89 /* required last entry */
92 MODULE_DEVICE_TABLE(pci, hns3_pci_tbl);
94 static irqreturn_t hns3_irq_handle(int irq, void *vector)
96 struct hns3_enet_tqp_vector *tqp_vector = vector;
98 napi_schedule_irqoff(&tqp_vector->napi);
103 static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv)
105 struct hns3_enet_tqp_vector *tqp_vectors;
108 for (i = 0; i < priv->vector_num; i++) {
109 tqp_vectors = &priv->tqp_vector[i];
111 if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED)
114 /* clear the affinity mask */
115 irq_set_affinity_hint(tqp_vectors->vector_irq, NULL);
117 /* release the irq resource */
118 free_irq(tqp_vectors->vector_irq, tqp_vectors);
119 tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED;
123 static int hns3_nic_init_irq(struct hns3_nic_priv *priv)
125 struct hns3_enet_tqp_vector *tqp_vectors;
126 int txrx_int_idx = 0;
132 for (i = 0; i < priv->vector_num; i++) {
133 tqp_vectors = &priv->tqp_vector[i];
135 if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED)
138 if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) {
139 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
140 "%s-%s-%s-%d", hns3_driver_name,
141 pci_name(priv->ae_handle->pdev),
142 "TxRx", txrx_int_idx++);
144 } else if (tqp_vectors->rx_group.ring) {
145 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
146 "%s-%s-%s-%d", hns3_driver_name,
147 pci_name(priv->ae_handle->pdev),
149 } else if (tqp_vectors->tx_group.ring) {
150 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
151 "%s-%s-%s-%d", hns3_driver_name,
152 pci_name(priv->ae_handle->pdev),
155 /* Skip this unused q_vector */
159 tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0';
161 irq_set_status_flags(tqp_vectors->vector_irq, IRQ_NOAUTOEN);
162 ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0,
163 tqp_vectors->name, tqp_vectors);
165 netdev_err(priv->netdev, "request irq(%d) fail\n",
166 tqp_vectors->vector_irq);
167 hns3_nic_uninit_irq(priv);
171 irq_set_affinity_hint(tqp_vectors->vector_irq,
172 &tqp_vectors->affinity_mask);
174 tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED;
180 static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector,
183 writel(mask_en, tqp_vector->mask_addr);
186 static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector)
188 napi_enable(&tqp_vector->napi);
189 enable_irq(tqp_vector->vector_irq);
192 hns3_mask_vector_irq(tqp_vector, 1);
195 static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector)
198 hns3_mask_vector_irq(tqp_vector, 0);
200 disable_irq(tqp_vector->vector_irq);
201 napi_disable(&tqp_vector->napi);
204 void hns3_set_vector_coalesce_rl(struct hns3_enet_tqp_vector *tqp_vector,
207 u32 rl_reg = hns3_rl_usec_to_reg(rl_value);
209 /* this defines the configuration for RL (Interrupt Rate Limiter).
210 * Rl defines rate of interrupts i.e. number of interrupts-per-second
211 * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing
214 if (rl_reg > 0 && !tqp_vector->tx_group.coal.adapt_enable &&
215 !tqp_vector->rx_group.coal.adapt_enable)
216 /* According to the hardware, the range of rl_reg is
217 * 0-59 and the unit is 4.
219 rl_reg |= HNS3_INT_RL_ENABLE_MASK;
221 writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
224 void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
229 if (tqp_vector->rx_group.coal.unit_1us)
230 new_val = gl_value | HNS3_INT_GL_1US;
232 new_val = hns3_gl_usec_to_reg(gl_value);
234 writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
237 void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
242 if (tqp_vector->tx_group.coal.unit_1us)
243 new_val = gl_value | HNS3_INT_GL_1US;
245 new_val = hns3_gl_usec_to_reg(gl_value);
247 writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
250 void hns3_set_vector_coalesce_tx_ql(struct hns3_enet_tqp_vector *tqp_vector,
253 writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_TX_QL_OFFSET);
256 void hns3_set_vector_coalesce_rx_ql(struct hns3_enet_tqp_vector *tqp_vector,
259 writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_RX_QL_OFFSET);
262 static void hns3_vector_coalesce_init(struct hns3_enet_tqp_vector *tqp_vector,
263 struct hns3_nic_priv *priv)
265 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
266 struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
267 struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
269 /* initialize the configuration for interrupt coalescing.
270 * 1. GL (Interrupt Gap Limiter)
271 * 2. RL (Interrupt Rate Limiter)
272 * 3. QL (Interrupt Quantity Limiter)
274 * Default: enable interrupt coalescing self-adaptive and GL
276 tx_coal->adapt_enable = 1;
277 rx_coal->adapt_enable = 1;
279 tx_coal->int_gl = HNS3_INT_GL_50K;
280 rx_coal->int_gl = HNS3_INT_GL_50K;
282 rx_coal->flow_level = HNS3_FLOW_LOW;
283 tx_coal->flow_level = HNS3_FLOW_LOW;
285 /* device version above V3(include V3), GL can configure 1us
286 * unit, so uses 1us unit.
288 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3) {
289 tx_coal->unit_1us = 1;
290 rx_coal->unit_1us = 1;
293 if (ae_dev->dev_specs.int_ql_max) {
294 tx_coal->ql_enable = 1;
295 rx_coal->ql_enable = 1;
296 tx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
297 rx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
298 tx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
299 rx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
304 hns3_vector_coalesce_init_hw(struct hns3_enet_tqp_vector *tqp_vector,
305 struct hns3_nic_priv *priv)
307 struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
308 struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
309 struct hnae3_handle *h = priv->ae_handle;
311 hns3_set_vector_coalesce_tx_gl(tqp_vector, tx_coal->int_gl);
312 hns3_set_vector_coalesce_rx_gl(tqp_vector, rx_coal->int_gl);
313 hns3_set_vector_coalesce_rl(tqp_vector, h->kinfo.int_rl_setting);
315 if (tx_coal->ql_enable)
316 hns3_set_vector_coalesce_tx_ql(tqp_vector, tx_coal->int_ql);
318 if (rx_coal->ql_enable)
319 hns3_set_vector_coalesce_rx_ql(tqp_vector, rx_coal->int_ql);
322 static int hns3_nic_set_real_num_queue(struct net_device *netdev)
324 struct hnae3_handle *h = hns3_get_handle(netdev);
325 struct hnae3_knic_private_info *kinfo = &h->kinfo;
326 struct hnae3_tc_info *tc_info = &kinfo->tc_info;
327 unsigned int queue_size = kinfo->num_tqps;
330 if (tc_info->num_tc <= 1 && !tc_info->mqprio_active) {
331 netdev_reset_tc(netdev);
333 ret = netdev_set_num_tc(netdev, tc_info->num_tc);
336 "netdev_set_num_tc fail, ret=%d!\n", ret);
340 for (i = 0; i < HNAE3_MAX_TC; i++) {
341 if (!test_bit(i, &tc_info->tc_en))
344 netdev_set_tc_queue(netdev, i, tc_info->tqp_count[i],
345 tc_info->tqp_offset[i]);
349 ret = netif_set_real_num_tx_queues(netdev, queue_size);
352 "netif_set_real_num_tx_queues fail, ret=%d!\n", ret);
356 ret = netif_set_real_num_rx_queues(netdev, queue_size);
359 "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
366 static u16 hns3_get_max_available_channels(struct hnae3_handle *h)
368 u16 alloc_tqps, max_rss_size, rss_size;
370 h->ae_algo->ops->get_tqps_and_rss_info(h, &alloc_tqps, &max_rss_size);
371 rss_size = alloc_tqps / h->kinfo.tc_info.num_tc;
373 return min_t(u16, rss_size, max_rss_size);
376 static void hns3_tqp_enable(struct hnae3_queue *tqp)
380 rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
381 rcb_reg |= BIT(HNS3_RING_EN_B);
382 hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
385 static void hns3_tqp_disable(struct hnae3_queue *tqp)
389 rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
390 rcb_reg &= ~BIT(HNS3_RING_EN_B);
391 hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
394 static void hns3_free_rx_cpu_rmap(struct net_device *netdev)
396 #ifdef CONFIG_RFS_ACCEL
397 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
398 netdev->rx_cpu_rmap = NULL;
402 static int hns3_set_rx_cpu_rmap(struct net_device *netdev)
404 #ifdef CONFIG_RFS_ACCEL
405 struct hns3_nic_priv *priv = netdev_priv(netdev);
406 struct hns3_enet_tqp_vector *tqp_vector;
409 if (!netdev->rx_cpu_rmap) {
410 netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->vector_num);
411 if (!netdev->rx_cpu_rmap)
415 for (i = 0; i < priv->vector_num; i++) {
416 tqp_vector = &priv->tqp_vector[i];
417 ret = irq_cpu_rmap_add(netdev->rx_cpu_rmap,
418 tqp_vector->vector_irq);
420 hns3_free_rx_cpu_rmap(netdev);
428 static int hns3_nic_net_up(struct net_device *netdev)
430 struct hns3_nic_priv *priv = netdev_priv(netdev);
431 struct hnae3_handle *h = priv->ae_handle;
435 ret = hns3_nic_reset_all_ring(h);
439 clear_bit(HNS3_NIC_STATE_DOWN, &priv->state);
441 /* enable the vectors */
442 for (i = 0; i < priv->vector_num; i++)
443 hns3_vector_enable(&priv->tqp_vector[i]);
446 for (j = 0; j < h->kinfo.num_tqps; j++)
447 hns3_tqp_enable(h->kinfo.tqp[j]);
449 /* start the ae_dev */
450 ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0;
452 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
454 hns3_tqp_disable(h->kinfo.tqp[j]);
456 for (j = i - 1; j >= 0; j--)
457 hns3_vector_disable(&priv->tqp_vector[j]);
463 static void hns3_config_xps(struct hns3_nic_priv *priv)
467 for (i = 0; i < priv->vector_num; i++) {
468 struct hns3_enet_tqp_vector *tqp_vector = &priv->tqp_vector[i];
469 struct hns3_enet_ring *ring = tqp_vector->tx_group.ring;
474 ret = netif_set_xps_queue(priv->netdev,
475 &tqp_vector->affinity_mask,
476 ring->tqp->tqp_index);
478 netdev_warn(priv->netdev,
479 "set xps queue failed: %d", ret);
486 static int hns3_nic_net_open(struct net_device *netdev)
488 struct hns3_nic_priv *priv = netdev_priv(netdev);
489 struct hnae3_handle *h = hns3_get_handle(netdev);
490 struct hnae3_knic_private_info *kinfo;
493 if (hns3_nic_resetting(netdev))
496 netif_carrier_off(netdev);
498 ret = hns3_nic_set_real_num_queue(netdev);
502 ret = hns3_nic_net_up(netdev);
504 netdev_err(netdev, "net up fail, ret=%d!\n", ret);
509 for (i = 0; i < HNAE3_MAX_USER_PRIO; i++)
510 netdev_set_prio_tc_map(netdev, i, kinfo->tc_info.prio_tc[i]);
512 if (h->ae_algo->ops->set_timer_task)
513 h->ae_algo->ops->set_timer_task(priv->ae_handle, true);
515 hns3_config_xps(priv);
517 netif_dbg(h, drv, netdev, "net open\n");
522 static void hns3_reset_tx_queue(struct hnae3_handle *h)
524 struct net_device *ndev = h->kinfo.netdev;
525 struct hns3_nic_priv *priv = netdev_priv(ndev);
526 struct netdev_queue *dev_queue;
529 for (i = 0; i < h->kinfo.num_tqps; i++) {
530 dev_queue = netdev_get_tx_queue(ndev,
531 priv->ring[i].queue_index);
532 netdev_tx_reset_queue(dev_queue);
536 static void hns3_nic_net_down(struct net_device *netdev)
538 struct hns3_nic_priv *priv = netdev_priv(netdev);
539 struct hnae3_handle *h = hns3_get_handle(netdev);
540 const struct hnae3_ae_ops *ops;
543 /* disable vectors */
544 for (i = 0; i < priv->vector_num; i++)
545 hns3_vector_disable(&priv->tqp_vector[i]);
548 for (i = 0; i < h->kinfo.num_tqps; i++)
549 hns3_tqp_disable(h->kinfo.tqp[i]);
552 ops = priv->ae_handle->ae_algo->ops;
554 ops->stop(priv->ae_handle);
556 /* delay ring buffer clearing to hns3_reset_notify_uninit_enet
557 * during reset process, because driver may not be able
558 * to disable the ring through firmware when downing the netdev.
560 if (!hns3_nic_resetting(netdev))
561 hns3_clear_all_ring(priv->ae_handle, false);
563 hns3_reset_tx_queue(priv->ae_handle);
566 static int hns3_nic_net_stop(struct net_device *netdev)
568 struct hns3_nic_priv *priv = netdev_priv(netdev);
569 struct hnae3_handle *h = hns3_get_handle(netdev);
571 if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
574 netif_dbg(h, drv, netdev, "net stop\n");
576 if (h->ae_algo->ops->set_timer_task)
577 h->ae_algo->ops->set_timer_task(priv->ae_handle, false);
579 netif_tx_stop_all_queues(netdev);
580 netif_carrier_off(netdev);
582 hns3_nic_net_down(netdev);
587 static int hns3_nic_uc_sync(struct net_device *netdev,
588 const unsigned char *addr)
590 struct hnae3_handle *h = hns3_get_handle(netdev);
592 if (h->ae_algo->ops->add_uc_addr)
593 return h->ae_algo->ops->add_uc_addr(h, addr);
598 static int hns3_nic_uc_unsync(struct net_device *netdev,
599 const unsigned char *addr)
601 struct hnae3_handle *h = hns3_get_handle(netdev);
603 /* need ignore the request of removing device address, because
604 * we store the device address and other addresses of uc list
605 * in the function's mac filter list.
607 if (ether_addr_equal(addr, netdev->dev_addr))
610 if (h->ae_algo->ops->rm_uc_addr)
611 return h->ae_algo->ops->rm_uc_addr(h, addr);
616 static int hns3_nic_mc_sync(struct net_device *netdev,
617 const unsigned char *addr)
619 struct hnae3_handle *h = hns3_get_handle(netdev);
621 if (h->ae_algo->ops->add_mc_addr)
622 return h->ae_algo->ops->add_mc_addr(h, addr);
627 static int hns3_nic_mc_unsync(struct net_device *netdev,
628 const unsigned char *addr)
630 struct hnae3_handle *h = hns3_get_handle(netdev);
632 if (h->ae_algo->ops->rm_mc_addr)
633 return h->ae_algo->ops->rm_mc_addr(h, addr);
638 static u8 hns3_get_netdev_flags(struct net_device *netdev)
642 if (netdev->flags & IFF_PROMISC) {
643 flags = HNAE3_USER_UPE | HNAE3_USER_MPE | HNAE3_BPE;
645 flags |= HNAE3_VLAN_FLTR;
646 if (netdev->flags & IFF_ALLMULTI)
647 flags |= HNAE3_USER_MPE;
653 static void hns3_nic_set_rx_mode(struct net_device *netdev)
655 struct hnae3_handle *h = hns3_get_handle(netdev);
658 new_flags = hns3_get_netdev_flags(netdev);
660 __dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync);
661 __dev_mc_sync(netdev, hns3_nic_mc_sync, hns3_nic_mc_unsync);
663 /* User mode Promisc mode enable and vlan filtering is disabled to
664 * let all packets in.
666 h->netdev_flags = new_flags;
667 hns3_request_update_promisc_mode(h);
670 void hns3_request_update_promisc_mode(struct hnae3_handle *handle)
672 const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
674 if (ops->request_update_promisc_mode)
675 ops->request_update_promisc_mode(handle);
678 void hns3_enable_vlan_filter(struct net_device *netdev, bool enable)
680 struct hns3_nic_priv *priv = netdev_priv(netdev);
681 struct hnae3_handle *h = priv->ae_handle;
682 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(h->pdev);
685 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V2 &&
686 h->ae_algo->ops->enable_vlan_filter) {
687 last_state = h->netdev_flags & HNAE3_VLAN_FLTR ? true : false;
688 if (enable != last_state) {
691 enable ? "enable" : "disable");
692 h->ae_algo->ops->enable_vlan_filter(h, enable);
697 static int hns3_set_tso(struct sk_buff *skb, u32 *paylen_fdop_ol4cs,
698 u16 *mss, u32 *type_cs_vlan_tso)
700 u32 l4_offset, hdr_len;
701 union l3_hdr_info l3;
702 union l4_hdr_info l4;
706 if (!skb_is_gso(skb))
709 ret = skb_cow_head(skb, 0);
710 if (unlikely(ret < 0))
713 l3.hdr = skb_network_header(skb);
714 l4.hdr = skb_transport_header(skb);
716 /* Software should clear the IPv4's checksum field when tso is
719 if (l3.v4->version == 4)
723 if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
726 SKB_GSO_UDP_TUNNEL_CSUM)) {
727 /* reset l3&l4 pointers from outer to inner headers */
728 l3.hdr = skb_inner_network_header(skb);
729 l4.hdr = skb_inner_transport_header(skb);
731 /* Software should clear the IPv4's checksum field when
734 if (l3.v4->version == 4)
738 /* normal or tunnel packet */
739 l4_offset = l4.hdr - skb->data;
741 /* remove payload length from inner pseudo checksum when tso */
742 l4_paylen = skb->len - l4_offset;
744 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) {
745 hdr_len = sizeof(*l4.udp) + l4_offset;
746 csum_replace_by_diff(&l4.udp->check,
747 (__force __wsum)htonl(l4_paylen));
749 hdr_len = (l4.tcp->doff << 2) + l4_offset;
750 csum_replace_by_diff(&l4.tcp->check,
751 (__force __wsum)htonl(l4_paylen));
754 /* find the txbd field values */
755 *paylen_fdop_ol4cs = skb->len - hdr_len;
756 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_TSO_B, 1);
758 /* offload outer UDP header checksum */
759 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)
760 hns3_set_field(*paylen_fdop_ol4cs, HNS3_TXD_OL4CS_B, 1);
762 /* get MSS for TSO */
763 *mss = skb_shinfo(skb)->gso_size;
770 static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
773 union l3_hdr_info l3;
774 unsigned char *l4_hdr;
775 unsigned char *exthdr;
779 /* find outer header point */
780 l3.hdr = skb_network_header(skb);
781 l4_hdr = skb_transport_header(skb);
783 if (skb->protocol == htons(ETH_P_IPV6)) {
784 exthdr = l3.hdr + sizeof(*l3.v6);
785 l4_proto_tmp = l3.v6->nexthdr;
786 if (l4_hdr != exthdr)
787 ipv6_skip_exthdr(skb, exthdr - skb->data,
788 &l4_proto_tmp, &frag_off);
789 } else if (skb->protocol == htons(ETH_P_IP)) {
790 l4_proto_tmp = l3.v4->protocol;
795 *ol4_proto = l4_proto_tmp;
798 if (!skb->encapsulation) {
803 /* find inner header point */
804 l3.hdr = skb_inner_network_header(skb);
805 l4_hdr = skb_inner_transport_header(skb);
807 if (l3.v6->version == 6) {
808 exthdr = l3.hdr + sizeof(*l3.v6);
809 l4_proto_tmp = l3.v6->nexthdr;
810 if (l4_hdr != exthdr)
811 ipv6_skip_exthdr(skb, exthdr - skb->data,
812 &l4_proto_tmp, &frag_off);
813 } else if (l3.v4->version == 4) {
814 l4_proto_tmp = l3.v4->protocol;
817 *il4_proto = l4_proto_tmp;
822 /* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
823 * and it is udp packet, which has a dest port as the IANA assigned.
824 * the hardware is expected to do the checksum offload, but the
825 * hardware will not do the checksum offload when udp dest port is
828 static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
830 struct hns3_nic_priv *priv = netdev_priv(skb->dev);
831 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
832 union l4_hdr_info l4;
834 /* device version above V3(include V3), the hardware can
835 * do this checksum offload.
837 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
840 l4.hdr = skb_transport_header(skb);
842 if (!(!skb->encapsulation &&
843 (l4.udp->dest == htons(IANA_VXLAN_UDP_PORT) ||
844 l4.udp->dest == htons(GENEVE_UDP_PORT))))
847 skb_checksum_help(skb);
852 static void hns3_set_outer_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
853 u32 *ol_type_vlan_len_msec)
855 u32 l2_len, l3_len, l4_len;
856 unsigned char *il2_hdr;
857 union l3_hdr_info l3;
858 union l4_hdr_info l4;
860 l3.hdr = skb_network_header(skb);
861 l4.hdr = skb_transport_header(skb);
863 /* compute OL2 header size, defined in 2 Bytes */
864 l2_len = l3.hdr - skb->data;
865 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L2LEN_S, l2_len >> 1);
867 /* compute OL3 header size, defined in 4 Bytes */
868 l3_len = l4.hdr - l3.hdr;
869 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_S, l3_len >> 2);
871 il2_hdr = skb_inner_mac_header(skb);
872 /* compute OL4 header size, defined in 4 Bytes */
873 l4_len = il2_hdr - l4.hdr;
874 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L4LEN_S, l4_len >> 2);
876 /* define outer network header type */
877 if (skb->protocol == htons(ETH_P_IP)) {
879 hns3_set_field(*ol_type_vlan_len_msec,
881 HNS3_OL3T_IPV4_CSUM);
883 hns3_set_field(*ol_type_vlan_len_msec,
885 HNS3_OL3T_IPV4_NO_CSUM);
887 } else if (skb->protocol == htons(ETH_P_IPV6)) {
888 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_S,
892 if (ol4_proto == IPPROTO_UDP)
893 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
894 HNS3_TUN_MAC_IN_UDP);
895 else if (ol4_proto == IPPROTO_GRE)
896 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
900 static int hns3_set_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
901 u8 il4_proto, u32 *type_cs_vlan_tso,
902 u32 *ol_type_vlan_len_msec)
904 unsigned char *l2_hdr = skb->data;
905 u32 l4_proto = ol4_proto;
906 union l4_hdr_info l4;
907 union l3_hdr_info l3;
910 l4.hdr = skb_transport_header(skb);
911 l3.hdr = skb_network_header(skb);
913 /* handle encapsulation skb */
914 if (skb->encapsulation) {
915 /* If this is a not UDP/GRE encapsulation skb */
916 if (!(ol4_proto == IPPROTO_UDP || ol4_proto == IPPROTO_GRE)) {
917 /* drop the skb tunnel packet if hardware don't support,
918 * because hardware can't calculate csum when TSO.
923 /* the stack computes the IP header already,
924 * driver calculate l4 checksum when not TSO.
926 skb_checksum_help(skb);
930 hns3_set_outer_l2l3l4(skb, ol4_proto, ol_type_vlan_len_msec);
932 /* switch to inner header */
933 l2_hdr = skb_inner_mac_header(skb);
934 l3.hdr = skb_inner_network_header(skb);
935 l4.hdr = skb_inner_transport_header(skb);
936 l4_proto = il4_proto;
939 if (l3.v4->version == 4) {
940 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
943 /* the stack computes the IP header already, the only time we
944 * need the hardware to recompute it is in the case of TSO.
947 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3CS_B, 1);
948 } else if (l3.v6->version == 6) {
949 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
953 /* compute inner(/normal) L2 header size, defined in 2 Bytes */
954 l2_len = l3.hdr - l2_hdr;
955 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S, l2_len >> 1);
957 /* compute inner(/normal) L3 header size, defined in 4 Bytes */
958 l3_len = l4.hdr - l3.hdr;
959 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_S, l3_len >> 2);
961 /* compute inner(/normal) L4 header size, defined in 4 Bytes */
964 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
965 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
967 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
971 if (hns3_tunnel_csum_bug(skb))
974 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
975 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
977 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
978 (sizeof(struct udphdr) >> 2));
981 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
982 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
984 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
985 (sizeof(struct sctphdr) >> 2));
988 /* drop the skb tunnel packet if hardware don't support,
989 * because hardware can't calculate csum when TSO.
994 /* the stack computes the IP header already,
995 * driver calculate l4 checksum when not TSO.
997 skb_checksum_help(skb);
1004 static int hns3_handle_vtags(struct hns3_enet_ring *tx_ring,
1005 struct sk_buff *skb)
1007 struct hnae3_handle *handle = tx_ring->tqp->handle;
1008 struct hnae3_ae_dev *ae_dev;
1009 struct vlan_ethhdr *vhdr;
1012 if (!(skb->protocol == htons(ETH_P_8021Q) ||
1013 skb_vlan_tag_present(skb)))
1016 /* For HW limitation on HNAE3_DEVICE_VERSION_V2, if port based insert
1017 * VLAN enabled, only one VLAN header is allowed in skb, otherwise it
1018 * will cause RAS error.
1020 ae_dev = pci_get_drvdata(handle->pdev);
1021 if (unlikely(skb_vlan_tagged_multi(skb) &&
1022 ae_dev->dev_version <= HNAE3_DEVICE_VERSION_V2 &&
1023 handle->port_base_vlan_state ==
1024 HNAE3_PORT_BASE_VLAN_ENABLE))
1027 if (skb->protocol == htons(ETH_P_8021Q) &&
1028 !(handle->kinfo.netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
1029 /* When HW VLAN acceleration is turned off, and the stack
1030 * sets the protocol to 802.1q, the driver just need to
1031 * set the protocol to the encapsulated ethertype.
1033 skb->protocol = vlan_get_protocol(skb);
1037 if (skb_vlan_tag_present(skb)) {
1038 /* Based on hw strategy, use out_vtag in two layer tag case,
1039 * and use inner_vtag in one tag case.
1041 if (skb->protocol == htons(ETH_P_8021Q) &&
1042 handle->port_base_vlan_state ==
1043 HNAE3_PORT_BASE_VLAN_DISABLE)
1044 rc = HNS3_OUTER_VLAN_TAG;
1046 rc = HNS3_INNER_VLAN_TAG;
1048 skb->protocol = vlan_get_protocol(skb);
1052 rc = skb_cow_head(skb, 0);
1053 if (unlikely(rc < 0))
1056 vhdr = (struct vlan_ethhdr *)skb->data;
1057 vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority << VLAN_PRIO_SHIFT)
1060 skb->protocol = vlan_get_protocol(skb);
1064 /* check if the hardware is capable of checksum offloading */
1065 static bool hns3_check_hw_tx_csum(struct sk_buff *skb)
1067 struct hns3_nic_priv *priv = netdev_priv(skb->dev);
1069 /* Kindly note, due to backward compatibility of the TX descriptor,
1070 * HW checksum of the non-IP packets and GSO packets is handled at
1071 * different place in the following code
1073 if (skb_csum_is_sctp(skb) || skb_is_gso(skb) ||
1074 !test_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state))
1080 static int hns3_fill_skb_desc(struct hns3_enet_ring *ring,
1081 struct sk_buff *skb, struct hns3_desc *desc)
1083 u32 ol_type_vlan_len_msec = 0;
1084 u32 paylen_ol4cs = skb->len;
1085 u32 type_cs_vlan_tso = 0;
1086 u16 mss_hw_csum = 0;
1091 ret = hns3_handle_vtags(ring, skb);
1092 if (unlikely(ret < 0)) {
1093 u64_stats_update_begin(&ring->syncp);
1094 ring->stats.tx_vlan_err++;
1095 u64_stats_update_end(&ring->syncp);
1097 } else if (ret == HNS3_INNER_VLAN_TAG) {
1098 inner_vtag = skb_vlan_tag_get(skb);
1099 inner_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1101 hns3_set_field(type_cs_vlan_tso, HNS3_TXD_VLAN_B, 1);
1102 } else if (ret == HNS3_OUTER_VLAN_TAG) {
1103 out_vtag = skb_vlan_tag_get(skb);
1104 out_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1106 hns3_set_field(ol_type_vlan_len_msec, HNS3_TXD_OVLAN_B,
1110 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1111 u8 ol4_proto, il4_proto;
1113 if (hns3_check_hw_tx_csum(skb)) {
1114 /* set checksum start and offset, defined in 2 Bytes */
1115 hns3_set_field(type_cs_vlan_tso, HNS3_TXD_CSUM_START_S,
1116 skb_checksum_start_offset(skb) >> 1);
1117 hns3_set_field(ol_type_vlan_len_msec,
1118 HNS3_TXD_CSUM_OFFSET_S,
1119 skb->csum_offset >> 1);
1120 mss_hw_csum |= BIT(HNS3_TXD_HW_CS_B);
1121 goto out_hw_tx_csum;
1124 skb_reset_mac_len(skb);
1126 ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
1127 if (unlikely(ret < 0)) {
1128 u64_stats_update_begin(&ring->syncp);
1129 ring->stats.tx_l4_proto_err++;
1130 u64_stats_update_end(&ring->syncp);
1134 ret = hns3_set_l2l3l4(skb, ol4_proto, il4_proto,
1136 &ol_type_vlan_len_msec);
1137 if (unlikely(ret < 0)) {
1138 u64_stats_update_begin(&ring->syncp);
1139 ring->stats.tx_l2l3l4_err++;
1140 u64_stats_update_end(&ring->syncp);
1144 ret = hns3_set_tso(skb, &paylen_ol4cs, &mss_hw_csum,
1146 if (unlikely(ret < 0)) {
1147 u64_stats_update_begin(&ring->syncp);
1148 ring->stats.tx_tso_err++;
1149 u64_stats_update_end(&ring->syncp);
1156 desc->tx.ol_type_vlan_len_msec =
1157 cpu_to_le32(ol_type_vlan_len_msec);
1158 desc->tx.type_cs_vlan_tso_len = cpu_to_le32(type_cs_vlan_tso);
1159 desc->tx.paylen_ol4cs = cpu_to_le32(paylen_ol4cs);
1160 desc->tx.mss_hw_csum = cpu_to_le16(mss_hw_csum);
1161 desc->tx.vlan_tag = cpu_to_le16(inner_vtag);
1162 desc->tx.outer_vlan_tag = cpu_to_le16(out_vtag);
1167 static int hns3_fill_desc(struct hns3_enet_ring *ring, void *priv,
1168 unsigned int size, enum hns_desc_type type)
1170 #define HNS3_LIKELY_BD_NUM 1
1172 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1173 struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1174 struct device *dev = ring_to_dev(ring);
1176 unsigned int frag_buf_num;
1180 if (type == DESC_TYPE_FRAGLIST_SKB ||
1181 type == DESC_TYPE_SKB) {
1182 struct sk_buff *skb = (struct sk_buff *)priv;
1184 dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1186 frag = (skb_frag_t *)priv;
1187 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1190 if (unlikely(dma_mapping_error(dev, dma))) {
1191 u64_stats_update_begin(&ring->syncp);
1192 ring->stats.sw_err_cnt++;
1193 u64_stats_update_end(&ring->syncp);
1197 desc_cb->priv = priv;
1198 desc_cb->length = size;
1200 desc_cb->type = type;
1202 if (likely(size <= HNS3_MAX_BD_SIZE)) {
1203 desc->addr = cpu_to_le64(dma);
1204 desc->tx.send_size = cpu_to_le16(size);
1205 desc->tx.bdtp_fe_sc_vld_ra_ri =
1206 cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1208 trace_hns3_tx_desc(ring, ring->next_to_use);
1209 ring_ptr_move_fw(ring, next_to_use);
1210 return HNS3_LIKELY_BD_NUM;
1213 frag_buf_num = hns3_tx_bd_count(size);
1214 sizeoflast = size % HNS3_MAX_BD_SIZE;
1215 sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1217 /* When frag size is bigger than hardware limit, split this frag */
1218 for (k = 0; k < frag_buf_num; k++) {
1219 /* now, fill the descriptor */
1220 desc->addr = cpu_to_le64(dma + HNS3_MAX_BD_SIZE * k);
1221 desc->tx.send_size = cpu_to_le16((k == frag_buf_num - 1) ?
1222 (u16)sizeoflast : (u16)HNS3_MAX_BD_SIZE);
1223 desc->tx.bdtp_fe_sc_vld_ra_ri =
1224 cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1226 trace_hns3_tx_desc(ring, ring->next_to_use);
1227 /* move ring pointer to next */
1228 ring_ptr_move_fw(ring, next_to_use);
1230 desc = &ring->desc[ring->next_to_use];
1233 return frag_buf_num;
1236 static unsigned int hns3_skb_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1237 unsigned int bd_num)
1242 size = skb_headlen(skb);
1243 while (size > HNS3_MAX_BD_SIZE) {
1244 bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1245 size -= HNS3_MAX_BD_SIZE;
1247 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1252 bd_size[bd_num++] = size;
1253 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1257 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1258 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1259 size = skb_frag_size(frag);
1263 while (size > HNS3_MAX_BD_SIZE) {
1264 bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1265 size -= HNS3_MAX_BD_SIZE;
1267 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1271 bd_size[bd_num++] = size;
1272 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1279 static unsigned int hns3_tx_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1280 u8 max_non_tso_bd_num)
1282 struct sk_buff *frag_skb;
1283 unsigned int bd_num = 0;
1285 /* If the total len is within the max bd limit */
1286 if (likely(skb->len <= HNS3_MAX_BD_SIZE && !skb_has_frag_list(skb) &&
1287 skb_shinfo(skb)->nr_frags < max_non_tso_bd_num))
1288 return skb_shinfo(skb)->nr_frags + 1U;
1290 /* The below case will always be linearized, return
1291 * HNS3_MAX_BD_NUM_TSO + 1U to make sure it is linearized.
1293 if (unlikely(skb->len > HNS3_MAX_TSO_SIZE ||
1294 (!skb_is_gso(skb) && skb->len >
1295 HNS3_MAX_NON_TSO_SIZE(max_non_tso_bd_num))))
1296 return HNS3_MAX_TSO_BD_NUM + 1U;
1298 bd_num = hns3_skb_bd_num(skb, bd_size, bd_num);
1300 if (!skb_has_frag_list(skb) || bd_num > HNS3_MAX_TSO_BD_NUM)
1303 skb_walk_frags(skb, frag_skb) {
1304 bd_num = hns3_skb_bd_num(frag_skb, bd_size, bd_num);
1305 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1312 static unsigned int hns3_gso_hdr_len(struct sk_buff *skb)
1314 if (!skb->encapsulation)
1315 return skb_transport_offset(skb) + tcp_hdrlen(skb);
1317 return skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb);
1320 /* HW need every continuous max_non_tso_bd_num buffer data to be larger
1321 * than MSS, we simplify it by ensuring skb_headlen + the first continuous
1322 * max_non_tso_bd_num - 1 frags to be larger than gso header len + mss,
1323 * and the remaining continuous max_non_tso_bd_num - 1 frags to be larger
1324 * than MSS except the last max_non_tso_bd_num - 1 frags.
1326 static bool hns3_skb_need_linearized(struct sk_buff *skb, unsigned int *bd_size,
1327 unsigned int bd_num, u8 max_non_tso_bd_num)
1329 unsigned int tot_len = 0;
1332 for (i = 0; i < max_non_tso_bd_num - 1U; i++)
1333 tot_len += bd_size[i];
1335 /* ensure the first max_non_tso_bd_num frags is greater than
1338 if (tot_len + bd_size[max_non_tso_bd_num - 1U] <
1339 skb_shinfo(skb)->gso_size + hns3_gso_hdr_len(skb))
1342 /* ensure every continuous max_non_tso_bd_num - 1 buffer is greater
1343 * than mss except the last one.
1345 for (i = 0; i < bd_num - max_non_tso_bd_num; i++) {
1346 tot_len -= bd_size[i];
1347 tot_len += bd_size[i + max_non_tso_bd_num - 1U];
1349 if (tot_len < skb_shinfo(skb)->gso_size)
1356 void hns3_shinfo_pack(struct skb_shared_info *shinfo, __u32 *size)
1360 for (i = 0; i < MAX_SKB_FRAGS; i++)
1361 size[i] = skb_frag_size(&shinfo->frags[i]);
1364 static int hns3_nic_maybe_stop_tx(struct hns3_enet_ring *ring,
1365 struct net_device *netdev,
1366 struct sk_buff *skb)
1368 struct hns3_nic_priv *priv = netdev_priv(netdev);
1369 u8 max_non_tso_bd_num = priv->max_non_tso_bd_num;
1370 unsigned int bd_size[HNS3_MAX_TSO_BD_NUM + 1U];
1371 unsigned int bd_num;
1373 bd_num = hns3_tx_bd_num(skb, bd_size, max_non_tso_bd_num);
1374 if (unlikely(bd_num > max_non_tso_bd_num)) {
1375 if (bd_num <= HNS3_MAX_TSO_BD_NUM && skb_is_gso(skb) &&
1376 !hns3_skb_need_linearized(skb, bd_size, bd_num,
1377 max_non_tso_bd_num)) {
1378 trace_hns3_over_max_bd(skb);
1382 if (__skb_linearize(skb))
1385 bd_num = hns3_tx_bd_count(skb->len);
1386 if ((skb_is_gso(skb) && bd_num > HNS3_MAX_TSO_BD_NUM) ||
1387 (!skb_is_gso(skb) &&
1388 bd_num > max_non_tso_bd_num)) {
1389 trace_hns3_over_max_bd(skb);
1393 u64_stats_update_begin(&ring->syncp);
1394 ring->stats.tx_copy++;
1395 u64_stats_update_end(&ring->syncp);
1399 if (likely(ring_space(ring) >= bd_num))
1402 netif_stop_subqueue(netdev, ring->queue_index);
1403 smp_mb(); /* Memory barrier before checking ring_space */
1405 /* Start queue in case hns3_clean_tx_ring has just made room
1406 * available and has not seen the queue stopped state performed
1407 * by netif_stop_subqueue above.
1409 if (ring_space(ring) >= bd_num && netif_carrier_ok(netdev) &&
1410 !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
1411 netif_start_subqueue(netdev, ring->queue_index);
1418 static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
1420 struct device *dev = ring_to_dev(ring);
1423 for (i = 0; i < ring->desc_num; i++) {
1424 struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1426 memset(desc, 0, sizeof(*desc));
1428 /* check if this is where we started */
1429 if (ring->next_to_use == next_to_use_orig)
1433 ring_ptr_move_bw(ring, next_to_use);
1435 if (!ring->desc_cb[ring->next_to_use].dma)
1438 /* unmap the descriptor dma address */
1439 if (ring->desc_cb[ring->next_to_use].type == DESC_TYPE_SKB ||
1440 ring->desc_cb[ring->next_to_use].type ==
1441 DESC_TYPE_FRAGLIST_SKB)
1442 dma_unmap_single(dev,
1443 ring->desc_cb[ring->next_to_use].dma,
1444 ring->desc_cb[ring->next_to_use].length,
1446 else if (ring->desc_cb[ring->next_to_use].length)
1448 ring->desc_cb[ring->next_to_use].dma,
1449 ring->desc_cb[ring->next_to_use].length,
1452 ring->desc_cb[ring->next_to_use].length = 0;
1453 ring->desc_cb[ring->next_to_use].dma = 0;
1454 ring->desc_cb[ring->next_to_use].type = DESC_TYPE_UNKNOWN;
1458 static int hns3_fill_skb_to_desc(struct hns3_enet_ring *ring,
1459 struct sk_buff *skb, enum hns_desc_type type)
1461 unsigned int size = skb_headlen(skb);
1462 int i, ret, bd_num = 0;
1465 ret = hns3_fill_desc(ring, skb, size, type);
1466 if (unlikely(ret < 0))
1472 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1473 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1475 size = skb_frag_size(frag);
1479 ret = hns3_fill_desc(ring, frag, size, DESC_TYPE_PAGE);
1480 if (unlikely(ret < 0))
1489 static void hns3_tx_doorbell(struct hns3_enet_ring *ring, int num,
1492 ring->pending_buf += num;
1495 u64_stats_update_begin(&ring->syncp);
1496 ring->stats.tx_more++;
1497 u64_stats_update_end(&ring->syncp);
1501 if (!ring->pending_buf)
1504 writel(ring->pending_buf,
1505 ring->tqp->io_base + HNS3_RING_TX_RING_TAIL_REG);
1506 ring->pending_buf = 0;
1507 WRITE_ONCE(ring->last_to_use, ring->next_to_use);
1510 netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
1512 struct hns3_nic_priv *priv = netdev_priv(netdev);
1513 struct hns3_enet_ring *ring = &priv->ring[skb->queue_mapping];
1514 struct netdev_queue *dev_queue;
1515 int pre_ntu, next_to_use_head;
1516 struct sk_buff *frag_skb;
1521 /* Hardware can only handle short frames above 32 bytes */
1522 if (skb_put_padto(skb, HNS3_MIN_TX_LEN)) {
1523 hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
1524 return NETDEV_TX_OK;
1527 /* Prefetch the data used later */
1528 prefetch(skb->data);
1530 ret = hns3_nic_maybe_stop_tx(ring, netdev, skb);
1531 if (unlikely(ret <= 0)) {
1532 if (ret == -EBUSY) {
1533 u64_stats_update_begin(&ring->syncp);
1534 ring->stats.tx_busy++;
1535 u64_stats_update_end(&ring->syncp);
1536 hns3_tx_doorbell(ring, 0, true);
1537 return NETDEV_TX_BUSY;
1538 } else if (ret == -ENOMEM) {
1539 u64_stats_update_begin(&ring->syncp);
1540 ring->stats.sw_err_cnt++;
1541 u64_stats_update_end(&ring->syncp);
1544 hns3_rl_err(netdev, "xmit error: %d!\n", ret);
1548 next_to_use_head = ring->next_to_use;
1550 ret = hns3_fill_skb_desc(ring, skb, &ring->desc[ring->next_to_use]);
1551 if (unlikely(ret < 0))
1554 ret = hns3_fill_skb_to_desc(ring, skb, DESC_TYPE_SKB);
1555 if (unlikely(ret < 0))
1560 skb_walk_frags(skb, frag_skb) {
1561 ret = hns3_fill_skb_to_desc(ring, frag_skb,
1562 DESC_TYPE_FRAGLIST_SKB);
1563 if (unlikely(ret < 0))
1569 pre_ntu = ring->next_to_use ? (ring->next_to_use - 1) :
1570 (ring->desc_num - 1);
1571 ring->desc[pre_ntu].tx.bdtp_fe_sc_vld_ra_ri |=
1572 cpu_to_le16(BIT(HNS3_TXD_FE_B));
1573 trace_hns3_tx_desc(ring, pre_ntu);
1575 /* Complete translate all packets */
1576 dev_queue = netdev_get_tx_queue(netdev, ring->queue_index);
1577 doorbell = __netdev_tx_sent_queue(dev_queue, skb->len,
1578 netdev_xmit_more());
1579 hns3_tx_doorbell(ring, bd_num, doorbell);
1581 return NETDEV_TX_OK;
1584 hns3_clear_desc(ring, next_to_use_head);
1587 dev_kfree_skb_any(skb);
1588 hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
1589 return NETDEV_TX_OK;
1592 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
1594 struct hnae3_handle *h = hns3_get_handle(netdev);
1595 struct sockaddr *mac_addr = p;
1598 if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1599 return -EADDRNOTAVAIL;
1601 if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
1602 netdev_info(netdev, "already using mac address %pM\n",
1607 /* For VF device, if there is a perm_addr, then the user will not
1608 * be allowed to change the address.
1610 if (!hns3_is_phys_func(h->pdev) &&
1611 !is_zero_ether_addr(netdev->perm_addr)) {
1612 netdev_err(netdev, "has permanent MAC %pM, user MAC %pM not allow\n",
1613 netdev->perm_addr, mac_addr->sa_data);
1617 ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
1619 netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
1623 ether_addr_copy(netdev->dev_addr, mac_addr->sa_data);
1628 static int hns3_nic_do_ioctl(struct net_device *netdev,
1629 struct ifreq *ifr, int cmd)
1631 struct hnae3_handle *h = hns3_get_handle(netdev);
1633 if (!netif_running(netdev))
1636 if (!h->ae_algo->ops->do_ioctl)
1639 return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
1642 static int hns3_nic_set_features(struct net_device *netdev,
1643 netdev_features_t features)
1645 netdev_features_t changed = netdev->features ^ features;
1646 struct hns3_nic_priv *priv = netdev_priv(netdev);
1647 struct hnae3_handle *h = priv->ae_handle;
1651 if (changed & (NETIF_F_GRO_HW) && h->ae_algo->ops->set_gro_en) {
1652 enable = !!(features & NETIF_F_GRO_HW);
1653 ret = h->ae_algo->ops->set_gro_en(h, enable);
1658 if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
1659 h->ae_algo->ops->enable_hw_strip_rxvtag) {
1660 enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
1661 ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, enable);
1666 if ((changed & NETIF_F_NTUPLE) && h->ae_algo->ops->enable_fd) {
1667 enable = !!(features & NETIF_F_NTUPLE);
1668 h->ae_algo->ops->enable_fd(h, enable);
1671 if ((netdev->features & NETIF_F_HW_TC) > (features & NETIF_F_HW_TC) &&
1672 h->ae_algo->ops->cls_flower_active(h)) {
1674 "there are offloaded TC filters active, cannot disable HW TC offload");
1678 netdev->features = features;
1682 static netdev_features_t hns3_features_check(struct sk_buff *skb,
1683 struct net_device *dev,
1684 netdev_features_t features)
1686 #define HNS3_MAX_HDR_LEN 480U
1687 #define HNS3_MAX_L4_HDR_LEN 60U
1691 if (skb->ip_summed != CHECKSUM_PARTIAL)
1694 if (skb->encapsulation)
1695 len = skb_inner_transport_header(skb) - skb->data;
1697 len = skb_transport_header(skb) - skb->data;
1699 /* Assume L4 is 60 byte as TCP is the only protocol with a
1700 * a flexible value, and it's max len is 60 bytes.
1702 len += HNS3_MAX_L4_HDR_LEN;
1704 /* Hardware only supports checksum on the skb with a max header
1707 if (len > HNS3_MAX_HDR_LEN)
1708 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
1713 static void hns3_nic_get_stats64(struct net_device *netdev,
1714 struct rtnl_link_stats64 *stats)
1716 struct hns3_nic_priv *priv = netdev_priv(netdev);
1717 int queue_num = priv->ae_handle->kinfo.num_tqps;
1718 struct hnae3_handle *handle = priv->ae_handle;
1719 struct hns3_enet_ring *ring;
1720 u64 rx_length_errors = 0;
1721 u64 rx_crc_errors = 0;
1722 u64 rx_multicast = 0;
1734 if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
1737 handle->ae_algo->ops->update_stats(handle, &netdev->stats);
1739 for (idx = 0; idx < queue_num; idx++) {
1740 /* fetch the tx stats */
1741 ring = &priv->ring[idx];
1743 start = u64_stats_fetch_begin_irq(&ring->syncp);
1744 tx_bytes += ring->stats.tx_bytes;
1745 tx_pkts += ring->stats.tx_pkts;
1746 tx_drop += ring->stats.sw_err_cnt;
1747 tx_drop += ring->stats.tx_vlan_err;
1748 tx_drop += ring->stats.tx_l4_proto_err;
1749 tx_drop += ring->stats.tx_l2l3l4_err;
1750 tx_drop += ring->stats.tx_tso_err;
1751 tx_errors += ring->stats.sw_err_cnt;
1752 tx_errors += ring->stats.tx_vlan_err;
1753 tx_errors += ring->stats.tx_l4_proto_err;
1754 tx_errors += ring->stats.tx_l2l3l4_err;
1755 tx_errors += ring->stats.tx_tso_err;
1756 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1758 /* fetch the rx stats */
1759 ring = &priv->ring[idx + queue_num];
1761 start = u64_stats_fetch_begin_irq(&ring->syncp);
1762 rx_bytes += ring->stats.rx_bytes;
1763 rx_pkts += ring->stats.rx_pkts;
1764 rx_drop += ring->stats.l2_err;
1765 rx_errors += ring->stats.l2_err;
1766 rx_errors += ring->stats.l3l4_csum_err;
1767 rx_crc_errors += ring->stats.l2_err;
1768 rx_multicast += ring->stats.rx_multicast;
1769 rx_length_errors += ring->stats.err_pkt_len;
1770 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1773 stats->tx_bytes = tx_bytes;
1774 stats->tx_packets = tx_pkts;
1775 stats->rx_bytes = rx_bytes;
1776 stats->rx_packets = rx_pkts;
1778 stats->rx_errors = rx_errors;
1779 stats->multicast = rx_multicast;
1780 stats->rx_length_errors = rx_length_errors;
1781 stats->rx_crc_errors = rx_crc_errors;
1782 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
1784 stats->tx_errors = tx_errors;
1785 stats->rx_dropped = rx_drop;
1786 stats->tx_dropped = tx_drop;
1787 stats->collisions = netdev->stats.collisions;
1788 stats->rx_over_errors = netdev->stats.rx_over_errors;
1789 stats->rx_frame_errors = netdev->stats.rx_frame_errors;
1790 stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
1791 stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
1792 stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
1793 stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
1794 stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
1795 stats->tx_window_errors = netdev->stats.tx_window_errors;
1796 stats->rx_compressed = netdev->stats.rx_compressed;
1797 stats->tx_compressed = netdev->stats.tx_compressed;
1800 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
1802 struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
1803 struct hnae3_knic_private_info *kinfo;
1804 u8 tc = mqprio_qopt->qopt.num_tc;
1805 u16 mode = mqprio_qopt->mode;
1806 u8 hw = mqprio_qopt->qopt.hw;
1807 struct hnae3_handle *h;
1809 if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
1810 mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
1813 if (tc > HNAE3_MAX_TC)
1819 h = hns3_get_handle(netdev);
1822 netif_dbg(h, drv, netdev, "setup tc: num_tc=%u\n", tc);
1824 return (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
1825 kinfo->dcb_ops->setup_tc(h, mqprio_qopt) : -EOPNOTSUPP;
1828 static int hns3_setup_tc_cls_flower(struct hns3_nic_priv *priv,
1829 struct flow_cls_offload *flow)
1831 int tc = tc_classid_to_hwtc(priv->netdev, flow->classid);
1832 struct hnae3_handle *h = hns3_get_handle(priv->netdev);
1834 switch (flow->command) {
1835 case FLOW_CLS_REPLACE:
1836 if (h->ae_algo->ops->add_cls_flower)
1837 return h->ae_algo->ops->add_cls_flower(h, flow, tc);
1839 case FLOW_CLS_DESTROY:
1840 if (h->ae_algo->ops->del_cls_flower)
1841 return h->ae_algo->ops->del_cls_flower(h, flow);
1850 static int hns3_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
1853 struct hns3_nic_priv *priv = cb_priv;
1855 if (!tc_cls_can_offload_and_chain0(priv->netdev, type_data))
1859 case TC_SETUP_CLSFLOWER:
1860 return hns3_setup_tc_cls_flower(priv, type_data);
1866 static LIST_HEAD(hns3_block_cb_list);
1868 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
1871 struct hns3_nic_priv *priv = netdev_priv(dev);
1875 case TC_SETUP_QDISC_MQPRIO:
1876 ret = hns3_setup_tc(dev, type_data);
1878 case TC_SETUP_BLOCK:
1879 ret = flow_block_cb_setup_simple(type_data,
1880 &hns3_block_cb_list,
1881 hns3_setup_tc_block_cb,
1891 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
1892 __be16 proto, u16 vid)
1894 struct hnae3_handle *h = hns3_get_handle(netdev);
1897 if (h->ae_algo->ops->set_vlan_filter)
1898 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
1903 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
1904 __be16 proto, u16 vid)
1906 struct hnae3_handle *h = hns3_get_handle(netdev);
1909 if (h->ae_algo->ops->set_vlan_filter)
1910 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
1915 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
1916 u8 qos, __be16 vlan_proto)
1918 struct hnae3_handle *h = hns3_get_handle(netdev);
1921 netif_dbg(h, drv, netdev,
1922 "set vf vlan: vf=%d, vlan=%u, qos=%u, vlan_proto=0x%x\n",
1923 vf, vlan, qos, ntohs(vlan_proto));
1925 if (h->ae_algo->ops->set_vf_vlan_filter)
1926 ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
1932 static int hns3_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
1934 struct hnae3_handle *handle = hns3_get_handle(netdev);
1936 if (hns3_nic_resetting(netdev))
1939 if (!handle->ae_algo->ops->set_vf_spoofchk)
1942 return handle->ae_algo->ops->set_vf_spoofchk(handle, vf, enable);
1945 static int hns3_set_vf_trust(struct net_device *netdev, int vf, bool enable)
1947 struct hnae3_handle *handle = hns3_get_handle(netdev);
1949 if (!handle->ae_algo->ops->set_vf_trust)
1952 return handle->ae_algo->ops->set_vf_trust(handle, vf, enable);
1955 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
1957 struct hnae3_handle *h = hns3_get_handle(netdev);
1960 if (hns3_nic_resetting(netdev))
1963 if (!h->ae_algo->ops->set_mtu)
1966 netif_dbg(h, drv, netdev,
1967 "change mtu from %u to %d\n", netdev->mtu, new_mtu);
1969 ret = h->ae_algo->ops->set_mtu(h, new_mtu);
1971 netdev_err(netdev, "failed to change MTU in hardware %d\n",
1974 netdev->mtu = new_mtu;
1979 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
1981 struct hns3_nic_priv *priv = netdev_priv(ndev);
1982 struct hnae3_handle *h = hns3_get_handle(ndev);
1983 struct hns3_enet_ring *tx_ring;
1984 struct napi_struct *napi;
1985 int timeout_queue = 0;
1986 int hw_head, hw_tail;
1987 int fbd_num, fbd_oft;
1988 int ebd_num, ebd_oft;
1993 /* Find the stopped queue the same way the stack does */
1994 for (i = 0; i < ndev->num_tx_queues; i++) {
1995 struct netdev_queue *q;
1996 unsigned long trans_start;
1998 q = netdev_get_tx_queue(ndev, i);
1999 trans_start = q->trans_start;
2000 if (netif_xmit_stopped(q) &&
2002 (trans_start + ndev->watchdog_timeo))) {
2004 netdev_info(ndev, "queue state: 0x%lx, delta msecs: %u\n",
2006 jiffies_to_msecs(jiffies - trans_start));
2011 if (i == ndev->num_tx_queues) {
2013 "no netdev TX timeout queue found, timeout count: %llu\n",
2014 priv->tx_timeout_count);
2018 priv->tx_timeout_count++;
2020 tx_ring = &priv->ring[timeout_queue];
2021 napi = &tx_ring->tqp_vector->napi;
2024 "tx_timeout count: %llu, queue id: %d, SW_NTU: 0x%x, SW_NTC: 0x%x, napi state: %lu\n",
2025 priv->tx_timeout_count, timeout_queue, tx_ring->next_to_use,
2026 tx_ring->next_to_clean, napi->state);
2029 "tx_pkts: %llu, tx_bytes: %llu, sw_err_cnt: %llu, tx_pending: %d\n",
2030 tx_ring->stats.tx_pkts, tx_ring->stats.tx_bytes,
2031 tx_ring->stats.sw_err_cnt, tx_ring->pending_buf);
2034 "seg_pkt_cnt: %llu, tx_more: %llu, restart_queue: %llu, tx_busy: %llu\n",
2035 tx_ring->stats.seg_pkt_cnt, tx_ring->stats.tx_more,
2036 tx_ring->stats.restart_queue, tx_ring->stats.tx_busy);
2038 /* When mac received many pause frames continuous, it's unable to send
2039 * packets, which may cause tx timeout
2041 if (h->ae_algo->ops->get_mac_stats) {
2042 struct hns3_mac_stats mac_stats;
2044 h->ae_algo->ops->get_mac_stats(h, &mac_stats);
2045 netdev_info(ndev, "tx_pause_cnt: %llu, rx_pause_cnt: %llu\n",
2046 mac_stats.tx_pause_cnt, mac_stats.rx_pause_cnt);
2049 hw_head = readl_relaxed(tx_ring->tqp->io_base +
2050 HNS3_RING_TX_RING_HEAD_REG);
2051 hw_tail = readl_relaxed(tx_ring->tqp->io_base +
2052 HNS3_RING_TX_RING_TAIL_REG);
2053 fbd_num = readl_relaxed(tx_ring->tqp->io_base +
2054 HNS3_RING_TX_RING_FBDNUM_REG);
2055 fbd_oft = readl_relaxed(tx_ring->tqp->io_base +
2056 HNS3_RING_TX_RING_OFFSET_REG);
2057 ebd_num = readl_relaxed(tx_ring->tqp->io_base +
2058 HNS3_RING_TX_RING_EBDNUM_REG);
2059 ebd_oft = readl_relaxed(tx_ring->tqp->io_base +
2060 HNS3_RING_TX_RING_EBD_OFFSET_REG);
2061 bd_num = readl_relaxed(tx_ring->tqp->io_base +
2062 HNS3_RING_TX_RING_BD_NUM_REG);
2063 bd_err = readl_relaxed(tx_ring->tqp->io_base +
2064 HNS3_RING_TX_RING_BD_ERR_REG);
2065 ring_en = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_EN_REG);
2066 tc = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_TX_RING_TC_REG);
2069 "BD_NUM: 0x%x HW_HEAD: 0x%x, HW_TAIL: 0x%x, BD_ERR: 0x%x, INT: 0x%x\n",
2070 bd_num, hw_head, hw_tail, bd_err,
2071 readl(tx_ring->tqp_vector->mask_addr));
2073 "RING_EN: 0x%x, TC: 0x%x, FBD_NUM: 0x%x FBD_OFT: 0x%x, EBD_NUM: 0x%x, EBD_OFT: 0x%x\n",
2074 ring_en, tc, fbd_num, fbd_oft, ebd_num, ebd_oft);
2079 static void hns3_nic_net_timeout(struct net_device *ndev, unsigned int txqueue)
2081 struct hns3_nic_priv *priv = netdev_priv(ndev);
2082 struct hnae3_handle *h = priv->ae_handle;
2084 if (!hns3_get_tx_timeo_queue_info(ndev))
2087 /* request the reset, and let the hclge to determine
2088 * which reset level should be done
2090 if (h->ae_algo->ops->reset_event)
2091 h->ae_algo->ops->reset_event(h->pdev, h);
2094 #ifdef CONFIG_RFS_ACCEL
2095 static int hns3_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
2096 u16 rxq_index, u32 flow_id)
2098 struct hnae3_handle *h = hns3_get_handle(dev);
2099 struct flow_keys fkeys;
2101 if (!h->ae_algo->ops->add_arfs_entry)
2104 if (skb->encapsulation)
2105 return -EPROTONOSUPPORT;
2107 if (!skb_flow_dissect_flow_keys(skb, &fkeys, 0))
2108 return -EPROTONOSUPPORT;
2110 if ((fkeys.basic.n_proto != htons(ETH_P_IP) &&
2111 fkeys.basic.n_proto != htons(ETH_P_IPV6)) ||
2112 (fkeys.basic.ip_proto != IPPROTO_TCP &&
2113 fkeys.basic.ip_proto != IPPROTO_UDP))
2114 return -EPROTONOSUPPORT;
2116 return h->ae_algo->ops->add_arfs_entry(h, rxq_index, flow_id, &fkeys);
2120 static int hns3_nic_get_vf_config(struct net_device *ndev, int vf,
2121 struct ifla_vf_info *ivf)
2123 struct hnae3_handle *h = hns3_get_handle(ndev);
2125 if (!h->ae_algo->ops->get_vf_config)
2128 return h->ae_algo->ops->get_vf_config(h, vf, ivf);
2131 static int hns3_nic_set_vf_link_state(struct net_device *ndev, int vf,
2134 struct hnae3_handle *h = hns3_get_handle(ndev);
2136 if (!h->ae_algo->ops->set_vf_link_state)
2139 return h->ae_algo->ops->set_vf_link_state(h, vf, link_state);
2142 static int hns3_nic_set_vf_rate(struct net_device *ndev, int vf,
2143 int min_tx_rate, int max_tx_rate)
2145 struct hnae3_handle *h = hns3_get_handle(ndev);
2147 if (!h->ae_algo->ops->set_vf_rate)
2150 return h->ae_algo->ops->set_vf_rate(h, vf, min_tx_rate, max_tx_rate,
2154 static int hns3_nic_set_vf_mac(struct net_device *netdev, int vf_id, u8 *mac)
2156 struct hnae3_handle *h = hns3_get_handle(netdev);
2158 if (!h->ae_algo->ops->set_vf_mac)
2161 if (is_multicast_ether_addr(mac)) {
2163 "Invalid MAC:%pM specified. Could not set MAC\n",
2168 return h->ae_algo->ops->set_vf_mac(h, vf_id, mac);
2171 static const struct net_device_ops hns3_nic_netdev_ops = {
2172 .ndo_open = hns3_nic_net_open,
2173 .ndo_stop = hns3_nic_net_stop,
2174 .ndo_start_xmit = hns3_nic_net_xmit,
2175 .ndo_tx_timeout = hns3_nic_net_timeout,
2176 .ndo_set_mac_address = hns3_nic_net_set_mac_address,
2177 .ndo_do_ioctl = hns3_nic_do_ioctl,
2178 .ndo_change_mtu = hns3_nic_change_mtu,
2179 .ndo_set_features = hns3_nic_set_features,
2180 .ndo_features_check = hns3_features_check,
2181 .ndo_get_stats64 = hns3_nic_get_stats64,
2182 .ndo_setup_tc = hns3_nic_setup_tc,
2183 .ndo_set_rx_mode = hns3_nic_set_rx_mode,
2184 .ndo_vlan_rx_add_vid = hns3_vlan_rx_add_vid,
2185 .ndo_vlan_rx_kill_vid = hns3_vlan_rx_kill_vid,
2186 .ndo_set_vf_vlan = hns3_ndo_set_vf_vlan,
2187 .ndo_set_vf_spoofchk = hns3_set_vf_spoofchk,
2188 .ndo_set_vf_trust = hns3_set_vf_trust,
2189 #ifdef CONFIG_RFS_ACCEL
2190 .ndo_rx_flow_steer = hns3_rx_flow_steer,
2192 .ndo_get_vf_config = hns3_nic_get_vf_config,
2193 .ndo_set_vf_link_state = hns3_nic_set_vf_link_state,
2194 .ndo_set_vf_rate = hns3_nic_set_vf_rate,
2195 .ndo_set_vf_mac = hns3_nic_set_vf_mac,
2198 bool hns3_is_phys_func(struct pci_dev *pdev)
2200 u32 dev_id = pdev->device;
2203 case HNAE3_DEV_ID_GE:
2204 case HNAE3_DEV_ID_25GE:
2205 case HNAE3_DEV_ID_25GE_RDMA:
2206 case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
2207 case HNAE3_DEV_ID_50GE_RDMA:
2208 case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
2209 case HNAE3_DEV_ID_100G_RDMA_MACSEC:
2210 case HNAE3_DEV_ID_200G_RDMA:
2212 case HNAE3_DEV_ID_VF:
2213 case HNAE3_DEV_ID_RDMA_DCB_PFC_VF:
2216 dev_warn(&pdev->dev, "un-recognized pci device-id %u",
2223 static void hns3_disable_sriov(struct pci_dev *pdev)
2225 /* If our VFs are assigned we cannot shut down SR-IOV
2226 * without causing issues, so just leave the hardware
2227 * available but disabled
2229 if (pci_vfs_assigned(pdev)) {
2230 dev_warn(&pdev->dev,
2231 "disabling driver while VFs are assigned\n");
2235 pci_disable_sriov(pdev);
2238 /* hns3_probe - Device initialization routine
2239 * @pdev: PCI device information struct
2240 * @ent: entry in hns3_pci_tbl
2242 * hns3_probe initializes a PF identified by a pci_dev structure.
2243 * The OS initialization, configuring of the PF private structure,
2244 * and a hardware reset occur.
2246 * Returns 0 on success, negative on failure
2248 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2250 struct hnae3_ae_dev *ae_dev;
2253 ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev), GFP_KERNEL);
2257 ae_dev->pdev = pdev;
2258 ae_dev->flag = ent->driver_data;
2259 pci_set_drvdata(pdev, ae_dev);
2261 ret = hnae3_register_ae_dev(ae_dev);
2263 pci_set_drvdata(pdev, NULL);
2268 /* hns3_remove - Device removal routine
2269 * @pdev: PCI device information struct
2271 static void hns3_remove(struct pci_dev *pdev)
2273 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2275 if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
2276 hns3_disable_sriov(pdev);
2278 hnae3_unregister_ae_dev(ae_dev);
2279 pci_set_drvdata(pdev, NULL);
2283 * hns3_pci_sriov_configure
2284 * @pdev: pointer to a pci_dev structure
2285 * @num_vfs: number of VFs to allocate
2287 * Enable or change the number of VFs. Called when the user updates the number
2290 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
2294 if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
2295 dev_warn(&pdev->dev, "Can not config SRIOV\n");
2300 ret = pci_enable_sriov(pdev, num_vfs);
2302 dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
2305 } else if (!pci_vfs_assigned(pdev)) {
2306 pci_disable_sriov(pdev);
2308 dev_warn(&pdev->dev,
2309 "Unable to free VFs because some are assigned to VMs.\n");
2315 static void hns3_shutdown(struct pci_dev *pdev)
2317 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2319 hnae3_unregister_ae_dev(ae_dev);
2320 pci_set_drvdata(pdev, NULL);
2322 if (system_state == SYSTEM_POWER_OFF)
2323 pci_set_power_state(pdev, PCI_D3hot);
2326 static pci_ers_result_t hns3_error_detected(struct pci_dev *pdev,
2327 pci_channel_state_t state)
2329 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2330 pci_ers_result_t ret;
2332 dev_info(&pdev->dev, "PCI error detected, state(=%u)!!\n", state);
2334 if (state == pci_channel_io_perm_failure)
2335 return PCI_ERS_RESULT_DISCONNECT;
2337 if (!ae_dev || !ae_dev->ops) {
2339 "Can't recover - error happened before device initialized\n");
2340 return PCI_ERS_RESULT_NONE;
2343 if (ae_dev->ops->handle_hw_ras_error)
2344 ret = ae_dev->ops->handle_hw_ras_error(ae_dev);
2346 return PCI_ERS_RESULT_NONE;
2351 static pci_ers_result_t hns3_slot_reset(struct pci_dev *pdev)
2353 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2354 const struct hnae3_ae_ops *ops;
2355 enum hnae3_reset_type reset_type;
2356 struct device *dev = &pdev->dev;
2358 if (!ae_dev || !ae_dev->ops)
2359 return PCI_ERS_RESULT_NONE;
2362 /* request the reset */
2363 if (ops->reset_event && ops->get_reset_level &&
2364 ops->set_default_reset_request) {
2365 if (ae_dev->hw_err_reset_req) {
2366 reset_type = ops->get_reset_level(ae_dev,
2367 &ae_dev->hw_err_reset_req);
2368 ops->set_default_reset_request(ae_dev, reset_type);
2369 dev_info(dev, "requesting reset due to PCI error\n");
2370 ops->reset_event(pdev, NULL);
2373 return PCI_ERS_RESULT_RECOVERED;
2376 return PCI_ERS_RESULT_DISCONNECT;
2379 static void hns3_reset_prepare(struct pci_dev *pdev)
2381 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2383 dev_info(&pdev->dev, "FLR prepare\n");
2384 if (ae_dev && ae_dev->ops && ae_dev->ops->flr_prepare)
2385 ae_dev->ops->flr_prepare(ae_dev);
2388 static void hns3_reset_done(struct pci_dev *pdev)
2390 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2392 dev_info(&pdev->dev, "FLR done\n");
2393 if (ae_dev && ae_dev->ops && ae_dev->ops->flr_done)
2394 ae_dev->ops->flr_done(ae_dev);
2397 static const struct pci_error_handlers hns3_err_handler = {
2398 .error_detected = hns3_error_detected,
2399 .slot_reset = hns3_slot_reset,
2400 .reset_prepare = hns3_reset_prepare,
2401 .reset_done = hns3_reset_done,
2404 static struct pci_driver hns3_driver = {
2405 .name = hns3_driver_name,
2406 .id_table = hns3_pci_tbl,
2407 .probe = hns3_probe,
2408 .remove = hns3_remove,
2409 .shutdown = hns3_shutdown,
2410 .sriov_configure = hns3_pci_sriov_configure,
2411 .err_handler = &hns3_err_handler,
2414 /* set default feature to hns3 */
2415 static void hns3_set_default_feature(struct net_device *netdev)
2417 struct hnae3_handle *h = hns3_get_handle(netdev);
2418 struct pci_dev *pdev = h->pdev;
2419 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2421 netdev->priv_flags |= IFF_UNICAST_FLT;
2423 netdev->hw_enc_features |= NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2424 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2425 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2426 NETIF_F_SCTP_CRC | NETIF_F_TSO_MANGLEID | NETIF_F_FRAGLIST;
2428 netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
2430 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
2431 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
2432 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2433 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2434 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2435 NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
2437 netdev->vlan_features |= NETIF_F_RXCSUM |
2438 NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO |
2439 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2440 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2441 NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
2443 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX |
2444 NETIF_F_HW_VLAN_CTAG_RX |
2445 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2446 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2447 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2448 NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
2450 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V2) {
2451 netdev->hw_features |= NETIF_F_GRO_HW;
2452 netdev->features |= NETIF_F_GRO_HW;
2454 if (!(h->flags & HNAE3_SUPPORT_VF)) {
2455 netdev->hw_features |= NETIF_F_NTUPLE;
2456 netdev->features |= NETIF_F_NTUPLE;
2460 if (test_bit(HNAE3_DEV_SUPPORT_UDP_GSO_B, ae_dev->caps)) {
2461 netdev->hw_features |= NETIF_F_GSO_UDP_L4;
2462 netdev->features |= NETIF_F_GSO_UDP_L4;
2463 netdev->vlan_features |= NETIF_F_GSO_UDP_L4;
2464 netdev->hw_enc_features |= NETIF_F_GSO_UDP_L4;
2467 if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps)) {
2468 netdev->hw_features |= NETIF_F_HW_CSUM;
2469 netdev->features |= NETIF_F_HW_CSUM;
2470 netdev->vlan_features |= NETIF_F_HW_CSUM;
2471 netdev->hw_enc_features |= NETIF_F_HW_CSUM;
2473 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2474 netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2475 netdev->vlan_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2476 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2479 if (test_bit(HNAE3_DEV_SUPPORT_UDP_TUNNEL_CSUM_B, ae_dev->caps)) {
2480 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
2481 netdev->features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
2482 netdev->vlan_features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
2483 netdev->hw_enc_features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
2486 if (test_bit(HNAE3_DEV_SUPPORT_FD_FORWARD_TC_B, ae_dev->caps)) {
2487 netdev->hw_features |= NETIF_F_HW_TC;
2488 netdev->features |= NETIF_F_HW_TC;
2492 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
2493 struct hns3_desc_cb *cb)
2495 unsigned int order = hns3_page_order(ring);
2498 p = dev_alloc_pages(order);
2503 cb->page_offset = 0;
2505 cb->buf = page_address(p);
2506 cb->length = hns3_page_size(ring);
2507 cb->type = DESC_TYPE_PAGE;
2508 page_ref_add(p, USHRT_MAX - 1);
2509 cb->pagecnt_bias = USHRT_MAX;
2514 static void hns3_free_buffer(struct hns3_enet_ring *ring,
2515 struct hns3_desc_cb *cb, int budget)
2517 if (cb->type == DESC_TYPE_SKB)
2518 napi_consume_skb(cb->priv, budget);
2519 else if (!HNAE3_IS_TX_RING(ring) && cb->pagecnt_bias)
2520 __page_frag_cache_drain(cb->priv, cb->pagecnt_bias);
2521 memset(cb, 0, sizeof(*cb));
2524 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
2526 cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
2527 cb->length, ring_to_dma_dir(ring));
2529 if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
2535 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
2536 struct hns3_desc_cb *cb)
2538 if (cb->type == DESC_TYPE_SKB || cb->type == DESC_TYPE_FRAGLIST_SKB)
2539 dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
2540 ring_to_dma_dir(ring));
2541 else if (cb->length)
2542 dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
2543 ring_to_dma_dir(ring));
2546 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
2548 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2549 ring->desc[i].addr = 0;
2552 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i,
2555 struct hns3_desc_cb *cb = &ring->desc_cb[i];
2557 if (!ring->desc_cb[i].dma)
2560 hns3_buffer_detach(ring, i);
2561 hns3_free_buffer(ring, cb, budget);
2564 static void hns3_free_buffers(struct hns3_enet_ring *ring)
2568 for (i = 0; i < ring->desc_num; i++)
2569 hns3_free_buffer_detach(ring, i, 0);
2572 /* free desc along with its attached buffer */
2573 static void hns3_free_desc(struct hns3_enet_ring *ring)
2575 int size = ring->desc_num * sizeof(ring->desc[0]);
2577 hns3_free_buffers(ring);
2580 dma_free_coherent(ring_to_dev(ring), size,
2581 ring->desc, ring->desc_dma_addr);
2586 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
2588 int size = ring->desc_num * sizeof(ring->desc[0]);
2590 ring->desc = dma_alloc_coherent(ring_to_dev(ring), size,
2591 &ring->desc_dma_addr, GFP_KERNEL);
2598 static int hns3_alloc_and_map_buffer(struct hns3_enet_ring *ring,
2599 struct hns3_desc_cb *cb)
2603 ret = hns3_alloc_buffer(ring, cb);
2607 ret = hns3_map_buffer(ring, cb);
2614 hns3_free_buffer(ring, cb, 0);
2619 static int hns3_alloc_and_attach_buffer(struct hns3_enet_ring *ring, int i)
2621 int ret = hns3_alloc_and_map_buffer(ring, &ring->desc_cb[i]);
2626 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2631 /* Allocate memory for raw pkg, and map with dma */
2632 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
2636 for (i = 0; i < ring->desc_num; i++) {
2637 ret = hns3_alloc_and_attach_buffer(ring, i);
2639 goto out_buffer_fail;
2645 for (j = i - 1; j >= 0; j--)
2646 hns3_free_buffer_detach(ring, j, 0);
2650 /* detach a in-used buffer and replace with a reserved one */
2651 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
2652 struct hns3_desc_cb *res_cb)
2654 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2655 ring->desc_cb[i] = *res_cb;
2656 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2657 ring->desc[i].rx.bd_base_info = 0;
2660 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
2662 ring->desc_cb[i].reuse_flag = 0;
2663 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
2664 ring->desc_cb[i].page_offset);
2665 ring->desc[i].rx.bd_base_info = 0;
2667 dma_sync_single_for_device(ring_to_dev(ring),
2668 ring->desc_cb[i].dma + ring->desc_cb[i].page_offset,
2669 hns3_buf_size(ring),
2673 static bool hns3_nic_reclaim_desc(struct hns3_enet_ring *ring,
2674 int *bytes, int *pkts, int budget)
2676 /* pair with ring->last_to_use update in hns3_tx_doorbell(),
2677 * smp_store_release() is not used in hns3_tx_doorbell() because
2678 * the doorbell operation already have the needed barrier operation.
2680 int ltu = smp_load_acquire(&ring->last_to_use);
2681 int ntc = ring->next_to_clean;
2682 struct hns3_desc_cb *desc_cb;
2683 bool reclaimed = false;
2684 struct hns3_desc *desc;
2686 while (ltu != ntc) {
2687 desc = &ring->desc[ntc];
2689 if (le16_to_cpu(desc->tx.bdtp_fe_sc_vld_ra_ri) &
2690 BIT(HNS3_TXD_VLD_B))
2693 desc_cb = &ring->desc_cb[ntc];
2694 (*pkts) += (desc_cb->type == DESC_TYPE_SKB);
2695 (*bytes) += desc_cb->length;
2696 /* desc_cb will be cleaned, after hnae3_free_buffer_detach */
2697 hns3_free_buffer_detach(ring, ntc, budget);
2699 if (++ntc == ring->desc_num)
2702 /* Issue prefetch for next Tx descriptor */
2703 prefetch(&ring->desc_cb[ntc]);
2707 if (unlikely(!reclaimed))
2710 /* This smp_store_release() pairs with smp_load_acquire() in
2711 * ring_space called by hns3_nic_net_xmit.
2713 smp_store_release(&ring->next_to_clean, ntc);
2717 void hns3_clean_tx_ring(struct hns3_enet_ring *ring, int budget)
2719 struct net_device *netdev = ring_to_netdev(ring);
2720 struct hns3_nic_priv *priv = netdev_priv(netdev);
2721 struct netdev_queue *dev_queue;
2727 if (unlikely(!hns3_nic_reclaim_desc(ring, &bytes, &pkts, budget)))
2730 ring->tqp_vector->tx_group.total_bytes += bytes;
2731 ring->tqp_vector->tx_group.total_packets += pkts;
2733 u64_stats_update_begin(&ring->syncp);
2734 ring->stats.tx_bytes += bytes;
2735 ring->stats.tx_pkts += pkts;
2736 u64_stats_update_end(&ring->syncp);
2738 dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
2739 netdev_tx_completed_queue(dev_queue, pkts, bytes);
2741 if (unlikely(netif_carrier_ok(netdev) &&
2742 ring_space(ring) > HNS3_MAX_TSO_BD_NUM)) {
2743 /* Make sure that anybody stopping the queue after this
2744 * sees the new next_to_clean.
2747 if (netif_tx_queue_stopped(dev_queue) &&
2748 !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
2749 netif_tx_wake_queue(dev_queue);
2750 ring->stats.restart_queue++;
2755 static int hns3_desc_unused(struct hns3_enet_ring *ring)
2757 int ntc = ring->next_to_clean;
2758 int ntu = ring->next_to_use;
2760 return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
2763 static void hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring,
2766 struct hns3_desc_cb *desc_cb;
2767 struct hns3_desc_cb res_cbs;
2770 for (i = 0; i < cleand_count; i++) {
2771 desc_cb = &ring->desc_cb[ring->next_to_use];
2772 if (desc_cb->reuse_flag) {
2773 u64_stats_update_begin(&ring->syncp);
2774 ring->stats.reuse_pg_cnt++;
2775 u64_stats_update_end(&ring->syncp);
2777 hns3_reuse_buffer(ring, ring->next_to_use);
2779 ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
2781 u64_stats_update_begin(&ring->syncp);
2782 ring->stats.sw_err_cnt++;
2783 u64_stats_update_end(&ring->syncp);
2785 hns3_rl_err(ring_to_netdev(ring),
2786 "alloc rx buffer failed: %d\n",
2790 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
2792 u64_stats_update_begin(&ring->syncp);
2793 ring->stats.non_reuse_pg++;
2794 u64_stats_update_end(&ring->syncp);
2797 ring_ptr_move_fw(ring, next_to_use);
2800 writel(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
2803 static bool hns3_can_reuse_page(struct hns3_desc_cb *cb)
2805 return (page_count(cb->priv) - cb->pagecnt_bias) == 1;
2808 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
2809 struct hns3_enet_ring *ring, int pull_len,
2810 struct hns3_desc_cb *desc_cb)
2812 struct hns3_desc *desc = &ring->desc[ring->next_to_clean];
2813 int size = le16_to_cpu(desc->rx.size);
2814 u32 truesize = hns3_buf_size(ring);
2816 desc_cb->pagecnt_bias--;
2817 skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
2818 size - pull_len, truesize);
2820 /* Avoid re-using remote and pfmemalloc pages, or the stack is still
2821 * using the page when page_offset rollback to zero, flag default
2824 if (!dev_page_is_reusable(desc_cb->priv) ||
2825 (!desc_cb->page_offset && !hns3_can_reuse_page(desc_cb))) {
2826 __page_frag_cache_drain(desc_cb->priv, desc_cb->pagecnt_bias);
2830 /* Move offset up to the next cache line */
2831 desc_cb->page_offset += truesize;
2833 if (desc_cb->page_offset + truesize <= hns3_page_size(ring)) {
2834 desc_cb->reuse_flag = 1;
2835 } else if (hns3_can_reuse_page(desc_cb)) {
2836 desc_cb->reuse_flag = 1;
2837 desc_cb->page_offset = 0;
2838 } else if (desc_cb->pagecnt_bias) {
2839 __page_frag_cache_drain(desc_cb->priv, desc_cb->pagecnt_bias);
2843 if (unlikely(!desc_cb->pagecnt_bias)) {
2844 page_ref_add(desc_cb->priv, USHRT_MAX);
2845 desc_cb->pagecnt_bias = USHRT_MAX;
2849 static int hns3_gro_complete(struct sk_buff *skb, u32 l234info)
2851 __be16 type = skb->protocol;
2855 while (eth_type_vlan(type)) {
2856 struct vlan_hdr *vh;
2858 if ((depth + VLAN_HLEN) > skb_headlen(skb))
2861 vh = (struct vlan_hdr *)(skb->data + depth);
2862 type = vh->h_vlan_encapsulated_proto;
2866 skb_set_network_header(skb, depth);
2868 if (type == htons(ETH_P_IP)) {
2869 const struct iphdr *iph = ip_hdr(skb);
2871 depth += sizeof(struct iphdr);
2872 skb_set_transport_header(skb, depth);
2874 th->check = ~tcp_v4_check(skb->len - depth, iph->saddr,
2876 } else if (type == htons(ETH_P_IPV6)) {
2877 const struct ipv6hdr *iph = ipv6_hdr(skb);
2879 depth += sizeof(struct ipv6hdr);
2880 skb_set_transport_header(skb, depth);
2882 th->check = ~tcp_v6_check(skb->len - depth, &iph->saddr,
2885 hns3_rl_err(skb->dev,
2886 "Error: FW GRO supports only IPv4/IPv6, not 0x%04x, depth: %d\n",
2887 be16_to_cpu(type), depth);
2891 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
2893 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
2895 if (l234info & BIT(HNS3_RXD_GRO_FIXID_B))
2896 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_FIXEDID;
2898 skb->csum_start = (unsigned char *)th - skb->head;
2899 skb->csum_offset = offsetof(struct tcphdr, check);
2900 skb->ip_summed = CHECKSUM_PARTIAL;
2902 trace_hns3_gro(skb);
2907 static void hns3_checksum_complete(struct hns3_enet_ring *ring,
2908 struct sk_buff *skb, u32 l234info)
2912 u64_stats_update_begin(&ring->syncp);
2913 ring->stats.csum_complete++;
2914 u64_stats_update_end(&ring->syncp);
2915 skb->ip_summed = CHECKSUM_COMPLETE;
2916 lo = hnae3_get_field(l234info, HNS3_RXD_L2_CSUM_L_M,
2917 HNS3_RXD_L2_CSUM_L_S);
2918 hi = hnae3_get_field(l234info, HNS3_RXD_L2_CSUM_H_M,
2919 HNS3_RXD_L2_CSUM_H_S);
2920 skb->csum = csum_unfold((__force __sum16)(lo | hi << 8));
2923 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
2924 u32 l234info, u32 bd_base_info, u32 ol_info)
2926 struct net_device *netdev = ring_to_netdev(ring);
2927 int l3_type, l4_type;
2930 skb->ip_summed = CHECKSUM_NONE;
2932 skb_checksum_none_assert(skb);
2934 if (!(netdev->features & NETIF_F_RXCSUM))
2937 if (l234info & BIT(HNS3_RXD_L2_CSUM_B)) {
2938 hns3_checksum_complete(ring, skb, l234info);
2942 /* check if hardware has done checksum */
2943 if (!(bd_base_info & BIT(HNS3_RXD_L3L4P_B)))
2946 if (unlikely(l234info & (BIT(HNS3_RXD_L3E_B) | BIT(HNS3_RXD_L4E_B) |
2947 BIT(HNS3_RXD_OL3E_B) |
2948 BIT(HNS3_RXD_OL4E_B)))) {
2949 u64_stats_update_begin(&ring->syncp);
2950 ring->stats.l3l4_csum_err++;
2951 u64_stats_update_end(&ring->syncp);
2956 ol4_type = hnae3_get_field(ol_info, HNS3_RXD_OL4ID_M,
2959 case HNS3_OL4_TYPE_MAC_IN_UDP:
2960 case HNS3_OL4_TYPE_NVGRE:
2961 skb->csum_level = 1;
2963 case HNS3_OL4_TYPE_NO_TUN:
2964 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
2966 l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
2969 /* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
2970 if ((l3_type == HNS3_L3_TYPE_IPV4 ||
2971 l3_type == HNS3_L3_TYPE_IPV6) &&
2972 (l4_type == HNS3_L4_TYPE_UDP ||
2973 l4_type == HNS3_L4_TYPE_TCP ||
2974 l4_type == HNS3_L4_TYPE_SCTP))
2975 skb->ip_summed = CHECKSUM_UNNECESSARY;
2982 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
2984 if (skb_has_frag_list(skb))
2985 napi_gro_flush(&ring->tqp_vector->napi, false);
2987 napi_gro_receive(&ring->tqp_vector->napi, skb);
2990 static bool hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
2991 struct hns3_desc *desc, u32 l234info,
2994 struct hnae3_handle *handle = ring->tqp->handle;
2995 struct pci_dev *pdev = ring->tqp->handle->pdev;
2996 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2998 if (unlikely(ae_dev->dev_version < HNAE3_DEVICE_VERSION_V2)) {
2999 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3000 if (!(*vlan_tag & VLAN_VID_MASK))
3001 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3003 return (*vlan_tag != 0);
3006 #define HNS3_STRP_OUTER_VLAN 0x1
3007 #define HNS3_STRP_INNER_VLAN 0x2
3008 #define HNS3_STRP_BOTH 0x3
3010 /* Hardware always insert VLAN tag into RX descriptor when
3011 * remove the tag from packet, driver needs to determine
3012 * reporting which tag to stack.
3014 switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
3015 HNS3_RXD_STRP_TAGP_S)) {
3016 case HNS3_STRP_OUTER_VLAN:
3017 if (handle->port_base_vlan_state !=
3018 HNAE3_PORT_BASE_VLAN_DISABLE)
3021 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3023 case HNS3_STRP_INNER_VLAN:
3024 if (handle->port_base_vlan_state !=
3025 HNAE3_PORT_BASE_VLAN_DISABLE)
3028 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3030 case HNS3_STRP_BOTH:
3031 if (handle->port_base_vlan_state ==
3032 HNAE3_PORT_BASE_VLAN_DISABLE)
3033 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3035 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3043 static void hns3_rx_ring_move_fw(struct hns3_enet_ring *ring)
3045 ring->desc[ring->next_to_clean].rx.bd_base_info &=
3046 cpu_to_le32(~BIT(HNS3_RXD_VLD_B));
3047 ring->next_to_clean += 1;
3049 if (unlikely(ring->next_to_clean == ring->desc_num))
3050 ring->next_to_clean = 0;
3053 static int hns3_alloc_skb(struct hns3_enet_ring *ring, unsigned int length,
3056 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
3057 struct net_device *netdev = ring_to_netdev(ring);
3058 struct sk_buff *skb;
3060 ring->skb = napi_alloc_skb(&ring->tqp_vector->napi, HNS3_RX_HEAD_SIZE);
3062 if (unlikely(!skb)) {
3063 hns3_rl_err(netdev, "alloc rx skb fail\n");
3065 u64_stats_update_begin(&ring->syncp);
3066 ring->stats.sw_err_cnt++;
3067 u64_stats_update_end(&ring->syncp);
3072 trace_hns3_rx_desc(ring);
3073 prefetchw(skb->data);
3075 ring->pending_buf = 1;
3077 ring->tail_skb = NULL;
3078 if (length <= HNS3_RX_HEAD_SIZE) {
3079 memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
3081 /* We can reuse buffer as-is, just make sure it is reusable */
3082 if (dev_page_is_reusable(desc_cb->priv))
3083 desc_cb->reuse_flag = 1;
3084 else /* This page cannot be reused so discard it */
3085 __page_frag_cache_drain(desc_cb->priv,
3086 desc_cb->pagecnt_bias);
3088 hns3_rx_ring_move_fw(ring);
3091 u64_stats_update_begin(&ring->syncp);
3092 ring->stats.seg_pkt_cnt++;
3093 u64_stats_update_end(&ring->syncp);
3095 ring->pull_len = eth_get_headlen(netdev, va, HNS3_RX_HEAD_SIZE);
3096 __skb_put(skb, ring->pull_len);
3097 hns3_nic_reuse_page(skb, ring->frag_num++, ring, ring->pull_len,
3099 hns3_rx_ring_move_fw(ring);
3104 static int hns3_add_frag(struct hns3_enet_ring *ring)
3106 struct sk_buff *skb = ring->skb;
3107 struct sk_buff *head_skb = skb;
3108 struct sk_buff *new_skb;
3109 struct hns3_desc_cb *desc_cb;
3110 struct hns3_desc *desc;
3114 desc = &ring->desc[ring->next_to_clean];
3115 desc_cb = &ring->desc_cb[ring->next_to_clean];
3116 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
3117 /* make sure HW write desc complete */
3119 if (!(bd_base_info & BIT(HNS3_RXD_VLD_B)))
3122 if (unlikely(ring->frag_num >= MAX_SKB_FRAGS)) {
3123 new_skb = napi_alloc_skb(&ring->tqp_vector->napi, 0);
3124 if (unlikely(!new_skb)) {
3125 hns3_rl_err(ring_to_netdev(ring),
3126 "alloc rx fraglist skb fail\n");
3131 if (ring->tail_skb) {
3132 ring->tail_skb->next = new_skb;
3133 ring->tail_skb = new_skb;
3135 skb_shinfo(skb)->frag_list = new_skb;
3136 ring->tail_skb = new_skb;
3140 if (ring->tail_skb) {
3141 head_skb->truesize += hns3_buf_size(ring);
3142 head_skb->data_len += le16_to_cpu(desc->rx.size);
3143 head_skb->len += le16_to_cpu(desc->rx.size);
3144 skb = ring->tail_skb;
3147 dma_sync_single_for_cpu(ring_to_dev(ring),
3148 desc_cb->dma + desc_cb->page_offset,
3149 hns3_buf_size(ring),
3152 hns3_nic_reuse_page(skb, ring->frag_num++, ring, 0, desc_cb);
3153 trace_hns3_rx_desc(ring);
3154 hns3_rx_ring_move_fw(ring);
3155 ring->pending_buf++;
3156 } while (!(bd_base_info & BIT(HNS3_RXD_FE_B)));
3161 static int hns3_set_gro_and_checksum(struct hns3_enet_ring *ring,
3162 struct sk_buff *skb, u32 l234info,
3163 u32 bd_base_info, u32 ol_info)
3167 skb_shinfo(skb)->gso_size = hnae3_get_field(bd_base_info,
3168 HNS3_RXD_GRO_SIZE_M,
3169 HNS3_RXD_GRO_SIZE_S);
3170 /* if there is no HW GRO, do not set gro params */
3171 if (!skb_shinfo(skb)->gso_size) {
3172 hns3_rx_checksum(ring, skb, l234info, bd_base_info, ol_info);
3176 NAPI_GRO_CB(skb)->count = hnae3_get_field(l234info,
3177 HNS3_RXD_GRO_COUNT_M,
3178 HNS3_RXD_GRO_COUNT_S);
3180 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M, HNS3_RXD_L3ID_S);
3181 if (l3_type == HNS3_L3_TYPE_IPV4)
3182 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
3183 else if (l3_type == HNS3_L3_TYPE_IPV6)
3184 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
3188 return hns3_gro_complete(skb, l234info);
3191 static void hns3_set_rx_skb_rss_type(struct hns3_enet_ring *ring,
3192 struct sk_buff *skb, u32 rss_hash)
3194 struct hnae3_handle *handle = ring->tqp->handle;
3195 enum pkt_hash_types rss_type;
3198 rss_type = handle->kinfo.rss_type;
3200 rss_type = PKT_HASH_TYPE_NONE;
3202 skb_set_hash(skb, rss_hash, rss_type);
3205 static int hns3_handle_bdinfo(struct hns3_enet_ring *ring, struct sk_buff *skb)
3207 struct net_device *netdev = ring_to_netdev(ring);
3208 enum hns3_pkt_l2t_type l2_frame_type;
3209 u32 bd_base_info, l234info, ol_info;
3210 struct hns3_desc *desc;
3214 /* bdinfo handled below is only valid on the last BD of the
3215 * current packet, and ring->next_to_clean indicates the first
3216 * descriptor of next packet, so need - 1 below.
3218 pre_ntc = ring->next_to_clean ? (ring->next_to_clean - 1) :
3219 (ring->desc_num - 1);
3220 desc = &ring->desc[pre_ntc];
3221 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
3222 l234info = le32_to_cpu(desc->rx.l234_info);
3223 ol_info = le32_to_cpu(desc->rx.ol_info);
3225 /* Based on hw strategy, the tag offloaded will be stored at
3226 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
3227 * in one layer tag case.
3229 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
3232 if (hns3_parse_vlan_tag(ring, desc, l234info, &vlan_tag))
3233 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
3237 if (unlikely(!desc->rx.pkt_len || (l234info & (BIT(HNS3_RXD_TRUNCAT_B) |
3238 BIT(HNS3_RXD_L2E_B))))) {
3239 u64_stats_update_begin(&ring->syncp);
3240 if (l234info & BIT(HNS3_RXD_L2E_B))
3241 ring->stats.l2_err++;
3243 ring->stats.err_pkt_len++;
3244 u64_stats_update_end(&ring->syncp);
3251 /* Do update ip stack process */
3252 skb->protocol = eth_type_trans(skb, netdev);
3254 /* This is needed in order to enable forwarding support */
3255 ret = hns3_set_gro_and_checksum(ring, skb, l234info,
3256 bd_base_info, ol_info);
3257 if (unlikely(ret)) {
3258 u64_stats_update_begin(&ring->syncp);
3259 ring->stats.rx_err_cnt++;
3260 u64_stats_update_end(&ring->syncp);
3264 l2_frame_type = hnae3_get_field(l234info, HNS3_RXD_DMAC_M,
3267 u64_stats_update_begin(&ring->syncp);
3268 ring->stats.rx_pkts++;
3269 ring->stats.rx_bytes += len;
3271 if (l2_frame_type == HNS3_L2_TYPE_MULTICAST)
3272 ring->stats.rx_multicast++;
3274 u64_stats_update_end(&ring->syncp);
3276 ring->tqp_vector->rx_group.total_bytes += len;
3278 hns3_set_rx_skb_rss_type(ring, skb, le32_to_cpu(desc->rx.rss_hash));
3282 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring)
3284 struct sk_buff *skb = ring->skb;
3285 struct hns3_desc_cb *desc_cb;
3286 struct hns3_desc *desc;
3287 unsigned int length;
3291 desc = &ring->desc[ring->next_to_clean];
3292 desc_cb = &ring->desc_cb[ring->next_to_clean];
3297 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
3299 /* Check valid BD */
3300 if (unlikely(!(bd_base_info & BIT(HNS3_RXD_VLD_B))))
3304 length = le16_to_cpu(desc->rx.size);
3306 ring->va = desc_cb->buf + desc_cb->page_offset;
3308 dma_sync_single_for_cpu(ring_to_dev(ring),
3309 desc_cb->dma + desc_cb->page_offset,
3310 hns3_buf_size(ring),
3313 /* Prefetch first cache line of first page.
3314 * Idea is to cache few bytes of the header of the packet.
3315 * Our L1 Cache line size is 64B so need to prefetch twice to make
3316 * it 128B. But in actual we can have greater size of caches with
3317 * 128B Level 1 cache lines. In such a case, single fetch would
3318 * suffice to cache in the relevant part of the header.
3320 net_prefetch(ring->va);
3322 ret = hns3_alloc_skb(ring, length, ring->va);
3325 if (ret < 0) /* alloc buffer fail */
3327 if (!(bd_base_info & BIT(HNS3_RXD_FE_B))) { /* need add frag */
3328 ret = hns3_add_frag(ring);
3333 ret = hns3_add_frag(ring);
3338 /* As the head data may be changed when GRO enable, copy
3339 * the head data in after other data rx completed
3341 if (skb->len > HNS3_RX_HEAD_SIZE)
3342 memcpy(skb->data, ring->va,
3343 ALIGN(ring->pull_len, sizeof(long)));
3345 ret = hns3_handle_bdinfo(ring, skb);
3346 if (unlikely(ret)) {
3347 dev_kfree_skb_any(skb);
3351 skb_record_rx_queue(skb, ring->tqp->tqp_index);
3355 int hns3_clean_rx_ring(struct hns3_enet_ring *ring, int budget,
3356 void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
3358 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
3359 int unused_count = hns3_desc_unused(ring);
3363 unused_count -= ring->pending_buf;
3365 while (recv_pkts < budget) {
3366 /* Reuse or realloc buffers */
3367 if (unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
3368 hns3_nic_alloc_rx_buffers(ring, unused_count);
3369 unused_count = hns3_desc_unused(ring) -
3374 err = hns3_handle_rx_bd(ring);
3375 /* Do not get FE for the packet or failed to alloc skb */
3376 if (unlikely(!ring->skb || err == -ENXIO)) {
3378 } else if (likely(!err)) {
3379 rx_fn(ring, ring->skb);
3383 unused_count += ring->pending_buf;
3385 ring->pending_buf = 0;
3389 /* Make all data has been write before submit */
3390 if (unused_count > 0)
3391 hns3_nic_alloc_rx_buffers(ring, unused_count);
3396 static bool hns3_get_new_flow_lvl(struct hns3_enet_ring_group *ring_group)
3398 #define HNS3_RX_LOW_BYTE_RATE 10000
3399 #define HNS3_RX_MID_BYTE_RATE 20000
3400 #define HNS3_RX_ULTRA_PACKET_RATE 40
3402 enum hns3_flow_level_range new_flow_level;
3403 struct hns3_enet_tqp_vector *tqp_vector;
3404 int packets_per_msecs, bytes_per_msecs;
3407 tqp_vector = ring_group->ring->tqp_vector;
3409 jiffies_to_msecs(jiffies - tqp_vector->last_jiffies);
3410 if (!time_passed_ms)
3413 do_div(ring_group->total_packets, time_passed_ms);
3414 packets_per_msecs = ring_group->total_packets;
3416 do_div(ring_group->total_bytes, time_passed_ms);
3417 bytes_per_msecs = ring_group->total_bytes;
3419 new_flow_level = ring_group->coal.flow_level;
3421 /* Simple throttlerate management
3422 * 0-10MB/s lower (50000 ints/s)
3423 * 10-20MB/s middle (20000 ints/s)
3424 * 20-1249MB/s high (18000 ints/s)
3425 * > 40000pps ultra (8000 ints/s)
3427 switch (new_flow_level) {
3429 if (bytes_per_msecs > HNS3_RX_LOW_BYTE_RATE)
3430 new_flow_level = HNS3_FLOW_MID;
3433 if (bytes_per_msecs > HNS3_RX_MID_BYTE_RATE)
3434 new_flow_level = HNS3_FLOW_HIGH;
3435 else if (bytes_per_msecs <= HNS3_RX_LOW_BYTE_RATE)
3436 new_flow_level = HNS3_FLOW_LOW;
3438 case HNS3_FLOW_HIGH:
3439 case HNS3_FLOW_ULTRA:
3441 if (bytes_per_msecs <= HNS3_RX_MID_BYTE_RATE)
3442 new_flow_level = HNS3_FLOW_MID;
3446 if (packets_per_msecs > HNS3_RX_ULTRA_PACKET_RATE &&
3447 &tqp_vector->rx_group == ring_group)
3448 new_flow_level = HNS3_FLOW_ULTRA;
3450 ring_group->total_bytes = 0;
3451 ring_group->total_packets = 0;
3452 ring_group->coal.flow_level = new_flow_level;
3457 static bool hns3_get_new_int_gl(struct hns3_enet_ring_group *ring_group)
3459 struct hns3_enet_tqp_vector *tqp_vector;
3462 if (!ring_group->ring)
3465 tqp_vector = ring_group->ring->tqp_vector;
3466 if (!tqp_vector->last_jiffies)
3469 if (ring_group->total_packets == 0) {
3470 ring_group->coal.int_gl = HNS3_INT_GL_50K;
3471 ring_group->coal.flow_level = HNS3_FLOW_LOW;
3475 if (!hns3_get_new_flow_lvl(ring_group))
3478 new_int_gl = ring_group->coal.int_gl;
3479 switch (ring_group->coal.flow_level) {
3481 new_int_gl = HNS3_INT_GL_50K;
3484 new_int_gl = HNS3_INT_GL_20K;
3486 case HNS3_FLOW_HIGH:
3487 new_int_gl = HNS3_INT_GL_18K;
3489 case HNS3_FLOW_ULTRA:
3490 new_int_gl = HNS3_INT_GL_8K;
3496 if (new_int_gl != ring_group->coal.int_gl) {
3497 ring_group->coal.int_gl = new_int_gl;
3503 static void hns3_update_new_int_gl(struct hns3_enet_tqp_vector *tqp_vector)
3505 struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
3506 struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
3507 bool rx_update, tx_update;
3509 /* update param every 1000ms */
3510 if (time_before(jiffies,
3511 tqp_vector->last_jiffies + msecs_to_jiffies(1000)))
3514 if (rx_group->coal.adapt_enable) {
3515 rx_update = hns3_get_new_int_gl(rx_group);
3517 hns3_set_vector_coalesce_rx_gl(tqp_vector,
3518 rx_group->coal.int_gl);
3521 if (tx_group->coal.adapt_enable) {
3522 tx_update = hns3_get_new_int_gl(tx_group);
3524 hns3_set_vector_coalesce_tx_gl(tqp_vector,
3525 tx_group->coal.int_gl);
3528 tqp_vector->last_jiffies = jiffies;
3531 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
3533 struct hns3_nic_priv *priv = netdev_priv(napi->dev);
3534 struct hns3_enet_ring *ring;
3535 int rx_pkt_total = 0;
3537 struct hns3_enet_tqp_vector *tqp_vector =
3538 container_of(napi, struct hns3_enet_tqp_vector, napi);
3539 bool clean_complete = true;
3540 int rx_budget = budget;
3542 if (unlikely(test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
3543 napi_complete(napi);
3547 /* Since the actual Tx work is minimal, we can give the Tx a larger
3548 * budget and be more aggressive about cleaning up the Tx descriptors.
3550 hns3_for_each_ring(ring, tqp_vector->tx_group)
3551 hns3_clean_tx_ring(ring, budget);
3553 /* make sure rx ring budget not smaller than 1 */
3554 if (tqp_vector->num_tqps > 1)
3555 rx_budget = max(budget / tqp_vector->num_tqps, 1);
3557 hns3_for_each_ring(ring, tqp_vector->rx_group) {
3558 int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
3561 if (rx_cleaned >= rx_budget)
3562 clean_complete = false;
3564 rx_pkt_total += rx_cleaned;
3567 tqp_vector->rx_group.total_packets += rx_pkt_total;
3569 if (!clean_complete)
3572 if (napi_complete(napi) &&
3573 likely(!test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
3574 hns3_update_new_int_gl(tqp_vector);
3575 hns3_mask_vector_irq(tqp_vector, 1);
3578 return rx_pkt_total;
3581 static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
3582 struct hnae3_ring_chain_node *head)
3584 struct pci_dev *pdev = tqp_vector->handle->pdev;
3585 struct hnae3_ring_chain_node *cur_chain = head;
3586 struct hnae3_ring_chain_node *chain;
3587 struct hns3_enet_ring *tx_ring;
3588 struct hns3_enet_ring *rx_ring;
3590 tx_ring = tqp_vector->tx_group.ring;
3592 cur_chain->tqp_index = tx_ring->tqp->tqp_index;
3593 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
3594 HNAE3_RING_TYPE_TX);
3595 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3596 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_TX);
3598 cur_chain->next = NULL;
3600 while (tx_ring->next) {
3601 tx_ring = tx_ring->next;
3603 chain = devm_kzalloc(&pdev->dev, sizeof(*chain),
3606 goto err_free_chain;
3608 cur_chain->next = chain;
3609 chain->tqp_index = tx_ring->tqp->tqp_index;
3610 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
3611 HNAE3_RING_TYPE_TX);
3612 hnae3_set_field(chain->int_gl_idx,
3613 HNAE3_RING_GL_IDX_M,
3614 HNAE3_RING_GL_IDX_S,
3621 rx_ring = tqp_vector->rx_group.ring;
3622 if (!tx_ring && rx_ring) {
3623 cur_chain->next = NULL;
3624 cur_chain->tqp_index = rx_ring->tqp->tqp_index;
3625 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
3626 HNAE3_RING_TYPE_RX);
3627 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3628 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
3630 rx_ring = rx_ring->next;
3634 chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
3636 goto err_free_chain;
3638 cur_chain->next = chain;
3639 chain->tqp_index = rx_ring->tqp->tqp_index;
3640 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
3641 HNAE3_RING_TYPE_RX);
3642 hnae3_set_field(chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3643 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
3647 rx_ring = rx_ring->next;
3653 cur_chain = head->next;
3655 chain = cur_chain->next;
3656 devm_kfree(&pdev->dev, cur_chain);
3664 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
3665 struct hnae3_ring_chain_node *head)
3667 struct pci_dev *pdev = tqp_vector->handle->pdev;
3668 struct hnae3_ring_chain_node *chain_tmp, *chain;
3673 chain_tmp = chain->next;
3674 devm_kfree(&pdev->dev, chain);
3679 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
3680 struct hns3_enet_ring *ring)
3682 ring->next = group->ring;
3688 static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
3690 struct pci_dev *pdev = priv->ae_handle->pdev;
3691 struct hns3_enet_tqp_vector *tqp_vector;
3692 int num_vectors = priv->vector_num;
3696 numa_node = dev_to_node(&pdev->dev);
3698 for (vector_i = 0; vector_i < num_vectors; vector_i++) {
3699 tqp_vector = &priv->tqp_vector[vector_i];
3700 cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
3701 &tqp_vector->affinity_mask);
3705 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
3707 struct hnae3_ring_chain_node vector_ring_chain;
3708 struct hnae3_handle *h = priv->ae_handle;
3709 struct hns3_enet_tqp_vector *tqp_vector;
3713 hns3_nic_set_cpumask(priv);
3715 for (i = 0; i < priv->vector_num; i++) {
3716 tqp_vector = &priv->tqp_vector[i];
3717 hns3_vector_coalesce_init_hw(tqp_vector, priv);
3718 tqp_vector->num_tqps = 0;
3721 for (i = 0; i < h->kinfo.num_tqps; i++) {
3722 u16 vector_i = i % priv->vector_num;
3723 u16 tqp_num = h->kinfo.num_tqps;
3725 tqp_vector = &priv->tqp_vector[vector_i];
3727 hns3_add_ring_to_group(&tqp_vector->tx_group,
3730 hns3_add_ring_to_group(&tqp_vector->rx_group,
3731 &priv->ring[i + tqp_num]);
3733 priv->ring[i].tqp_vector = tqp_vector;
3734 priv->ring[i + tqp_num].tqp_vector = tqp_vector;
3735 tqp_vector->num_tqps++;
3738 for (i = 0; i < priv->vector_num; i++) {
3739 tqp_vector = &priv->tqp_vector[i];
3741 tqp_vector->rx_group.total_bytes = 0;
3742 tqp_vector->rx_group.total_packets = 0;
3743 tqp_vector->tx_group.total_bytes = 0;
3744 tqp_vector->tx_group.total_packets = 0;
3745 tqp_vector->handle = h;
3747 ret = hns3_get_vector_ring_chain(tqp_vector,
3748 &vector_ring_chain);
3752 ret = h->ae_algo->ops->map_ring_to_vector(h,
3753 tqp_vector->vector_irq, &vector_ring_chain);
3755 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3760 netif_napi_add(priv->netdev, &tqp_vector->napi,
3761 hns3_nic_common_poll, NAPI_POLL_WEIGHT);
3768 netif_napi_del(&priv->tqp_vector[i].napi);
3773 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
3775 struct hnae3_handle *h = priv->ae_handle;
3776 struct hns3_enet_tqp_vector *tqp_vector;
3777 struct hnae3_vector_info *vector;
3778 struct pci_dev *pdev = h->pdev;
3779 u16 tqp_num = h->kinfo.num_tqps;
3784 /* RSS size, cpu online and vector_num should be the same */
3785 /* Should consider 2p/4p later */
3786 vector_num = min_t(u16, num_online_cpus(), tqp_num);
3788 vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
3793 /* save the actual available vector number */
3794 vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
3796 priv->vector_num = vector_num;
3797 priv->tqp_vector = (struct hns3_enet_tqp_vector *)
3798 devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
3800 if (!priv->tqp_vector) {
3805 for (i = 0; i < priv->vector_num; i++) {
3806 tqp_vector = &priv->tqp_vector[i];
3807 tqp_vector->idx = i;
3808 tqp_vector->mask_addr = vector[i].io_addr;
3809 tqp_vector->vector_irq = vector[i].vector;
3810 hns3_vector_coalesce_init(tqp_vector, priv);
3814 devm_kfree(&pdev->dev, vector);
3818 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
3824 static void hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
3826 struct hnae3_ring_chain_node vector_ring_chain;
3827 struct hnae3_handle *h = priv->ae_handle;
3828 struct hns3_enet_tqp_vector *tqp_vector;
3831 for (i = 0; i < priv->vector_num; i++) {
3832 tqp_vector = &priv->tqp_vector[i];
3834 if (!tqp_vector->rx_group.ring && !tqp_vector->tx_group.ring)
3837 /* Since the mapping can be overwritten, when fail to get the
3838 * chain between vector and ring, we should go on to deal with
3839 * the remaining options.
3841 if (hns3_get_vector_ring_chain(tqp_vector, &vector_ring_chain))
3842 dev_warn(priv->dev, "failed to get ring chain\n");
3844 h->ae_algo->ops->unmap_ring_from_vector(h,
3845 tqp_vector->vector_irq, &vector_ring_chain);
3847 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3849 hns3_clear_ring_group(&tqp_vector->rx_group);
3850 hns3_clear_ring_group(&tqp_vector->tx_group);
3851 netif_napi_del(&priv->tqp_vector[i].napi);
3855 static void hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
3857 struct hnae3_handle *h = priv->ae_handle;
3858 struct pci_dev *pdev = h->pdev;
3861 for (i = 0; i < priv->vector_num; i++) {
3862 struct hns3_enet_tqp_vector *tqp_vector;
3864 tqp_vector = &priv->tqp_vector[i];
3865 ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
3870 devm_kfree(&pdev->dev, priv->tqp_vector);
3873 static void hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
3874 unsigned int ring_type)
3876 int queue_num = priv->ae_handle->kinfo.num_tqps;
3877 struct hns3_enet_ring *ring;
3880 if (ring_type == HNAE3_RING_TYPE_TX) {
3881 ring = &priv->ring[q->tqp_index];
3882 desc_num = priv->ae_handle->kinfo.num_tx_desc;
3883 ring->queue_index = q->tqp_index;
3885 ring = &priv->ring[q->tqp_index + queue_num];
3886 desc_num = priv->ae_handle->kinfo.num_rx_desc;
3887 ring->queue_index = q->tqp_index;
3890 hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
3894 ring->desc_cb = NULL;
3895 ring->dev = priv->dev;
3896 ring->desc_dma_addr = 0;
3897 ring->buf_size = q->buf_size;
3898 ring->desc_num = desc_num;
3899 ring->next_to_use = 0;
3900 ring->next_to_clean = 0;
3901 ring->last_to_use = 0;
3904 static void hns3_queue_to_ring(struct hnae3_queue *tqp,
3905 struct hns3_nic_priv *priv)
3907 hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
3908 hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
3911 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
3913 struct hnae3_handle *h = priv->ae_handle;
3914 struct pci_dev *pdev = h->pdev;
3917 priv->ring = devm_kzalloc(&pdev->dev,
3918 array3_size(h->kinfo.num_tqps,
3919 sizeof(*priv->ring), 2),
3924 for (i = 0; i < h->kinfo.num_tqps; i++)
3925 hns3_queue_to_ring(h->kinfo.tqp[i], priv);
3930 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
3935 devm_kfree(priv->dev, priv->ring);
3939 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
3943 if (ring->desc_num <= 0 || ring->buf_size <= 0)
3946 ring->desc_cb = devm_kcalloc(ring_to_dev(ring), ring->desc_num,
3947 sizeof(ring->desc_cb[0]), GFP_KERNEL);
3948 if (!ring->desc_cb) {
3953 ret = hns3_alloc_desc(ring);
3955 goto out_with_desc_cb;
3957 if (!HNAE3_IS_TX_RING(ring)) {
3958 ret = hns3_alloc_ring_buffers(ring);
3966 hns3_free_desc(ring);
3968 devm_kfree(ring_to_dev(ring), ring->desc_cb);
3969 ring->desc_cb = NULL;
3974 void hns3_fini_ring(struct hns3_enet_ring *ring)
3976 hns3_free_desc(ring);
3977 devm_kfree(ring_to_dev(ring), ring->desc_cb);
3978 ring->desc_cb = NULL;
3979 ring->next_to_clean = 0;
3980 ring->next_to_use = 0;
3981 ring->last_to_use = 0;
3982 ring->pending_buf = 0;
3984 dev_kfree_skb_any(ring->skb);
3989 static int hns3_buf_size2type(u32 buf_size)
3995 bd_size_type = HNS3_BD_SIZE_512_TYPE;
3998 bd_size_type = HNS3_BD_SIZE_1024_TYPE;
4001 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4004 bd_size_type = HNS3_BD_SIZE_4096_TYPE;
4007 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4010 return bd_size_type;
4013 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
4015 dma_addr_t dma = ring->desc_dma_addr;
4016 struct hnae3_queue *q = ring->tqp;
4018 if (!HNAE3_IS_TX_RING(ring)) {
4019 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG, (u32)dma);
4020 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
4021 (u32)((dma >> 31) >> 1));
4023 hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
4024 hns3_buf_size2type(ring->buf_size));
4025 hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
4026 ring->desc_num / 8 - 1);
4029 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
4031 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
4032 (u32)((dma >> 31) >> 1));
4034 hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
4035 ring->desc_num / 8 - 1);
4039 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
4041 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
4042 struct hnae3_tc_info *tc_info = &kinfo->tc_info;
4045 for (i = 0; i < HNAE3_MAX_TC; i++) {
4048 if (!test_bit(i, &tc_info->tc_en))
4051 for (j = 0; j < tc_info->tqp_count[i]; j++) {
4052 struct hnae3_queue *q;
4054 q = priv->ring[tc_info->tqp_offset[i] + j].tqp;
4055 hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG, i);
4060 int hns3_init_all_ring(struct hns3_nic_priv *priv)
4062 struct hnae3_handle *h = priv->ae_handle;
4063 int ring_num = h->kinfo.num_tqps * 2;
4067 for (i = 0; i < ring_num; i++) {
4068 ret = hns3_alloc_ring_memory(&priv->ring[i]);
4071 "Alloc ring memory fail! ret=%d\n", ret);
4072 goto out_when_alloc_ring_memory;
4075 u64_stats_init(&priv->ring[i].syncp);
4080 out_when_alloc_ring_memory:
4081 for (j = i - 1; j >= 0; j--)
4082 hns3_fini_ring(&priv->ring[j]);
4087 static void hns3_uninit_all_ring(struct hns3_nic_priv *priv)
4089 struct hnae3_handle *h = priv->ae_handle;
4092 for (i = 0; i < h->kinfo.num_tqps; i++) {
4093 hns3_fini_ring(&priv->ring[i]);
4094 hns3_fini_ring(&priv->ring[i + h->kinfo.num_tqps]);
4098 /* Set mac addr if it is configured. or leave it to the AE driver */
4099 static int hns3_init_mac_addr(struct net_device *netdev)
4101 struct hns3_nic_priv *priv = netdev_priv(netdev);
4102 struct hnae3_handle *h = priv->ae_handle;
4103 u8 mac_addr_temp[ETH_ALEN];
4106 if (h->ae_algo->ops->get_mac_addr)
4107 h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
4109 /* Check if the MAC address is valid, if not get a random one */
4110 if (!is_valid_ether_addr(mac_addr_temp)) {
4111 eth_hw_addr_random(netdev);
4112 dev_warn(priv->dev, "using random MAC address %pM\n",
4114 } else if (!ether_addr_equal(netdev->dev_addr, mac_addr_temp)) {
4115 ether_addr_copy(netdev->dev_addr, mac_addr_temp);
4116 ether_addr_copy(netdev->perm_addr, mac_addr_temp);
4121 if (h->ae_algo->ops->set_mac_addr)
4122 ret = h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
4127 static int hns3_init_phy(struct net_device *netdev)
4129 struct hnae3_handle *h = hns3_get_handle(netdev);
4132 if (h->ae_algo->ops->mac_connect_phy)
4133 ret = h->ae_algo->ops->mac_connect_phy(h);
4138 static void hns3_uninit_phy(struct net_device *netdev)
4140 struct hnae3_handle *h = hns3_get_handle(netdev);
4142 if (h->ae_algo->ops->mac_disconnect_phy)
4143 h->ae_algo->ops->mac_disconnect_phy(h);
4146 static void hns3_del_all_fd_rules(struct net_device *netdev, bool clear_list)
4148 struct hnae3_handle *h = hns3_get_handle(netdev);
4150 if (h->ae_algo->ops->del_all_fd_entries)
4151 h->ae_algo->ops->del_all_fd_entries(h, clear_list);
4154 static int hns3_client_start(struct hnae3_handle *handle)
4156 if (!handle->ae_algo->ops->client_start)
4159 return handle->ae_algo->ops->client_start(handle);
4162 static void hns3_client_stop(struct hnae3_handle *handle)
4164 if (!handle->ae_algo->ops->client_stop)
4167 handle->ae_algo->ops->client_stop(handle);
4170 static void hns3_info_show(struct hns3_nic_priv *priv)
4172 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
4174 dev_info(priv->dev, "MAC address: %pM\n", priv->netdev->dev_addr);
4175 dev_info(priv->dev, "Task queue pairs numbers: %u\n", kinfo->num_tqps);
4176 dev_info(priv->dev, "RSS size: %u\n", kinfo->rss_size);
4177 dev_info(priv->dev, "Allocated RSS size: %u\n", kinfo->req_rss_size);
4178 dev_info(priv->dev, "RX buffer length: %u\n", kinfo->rx_buf_len);
4179 dev_info(priv->dev, "Desc num per TX queue: %u\n", kinfo->num_tx_desc);
4180 dev_info(priv->dev, "Desc num per RX queue: %u\n", kinfo->num_rx_desc);
4181 dev_info(priv->dev, "Total number of enabled TCs: %u\n",
4182 kinfo->tc_info.num_tc);
4183 dev_info(priv->dev, "Max mtu size: %u\n", priv->netdev->max_mtu);
4186 static int hns3_client_init(struct hnae3_handle *handle)
4188 struct pci_dev *pdev = handle->pdev;
4189 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
4190 u16 alloc_tqps, max_rss_size;
4191 struct hns3_nic_priv *priv;
4192 struct net_device *netdev;
4195 handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
4197 netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
4201 priv = netdev_priv(netdev);
4202 priv->dev = &pdev->dev;
4203 priv->netdev = netdev;
4204 priv->ae_handle = handle;
4205 priv->tx_timeout_count = 0;
4206 priv->max_non_tso_bd_num = ae_dev->dev_specs.max_non_tso_bd_num;
4207 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
4209 handle->msg_enable = netif_msg_init(debug, DEFAULT_MSG_LEVEL);
4211 handle->kinfo.netdev = netdev;
4212 handle->priv = (void *)priv;
4214 hns3_init_mac_addr(netdev);
4216 hns3_set_default_feature(netdev);
4218 netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
4219 netdev->priv_flags |= IFF_UNICAST_FLT;
4220 netdev->netdev_ops = &hns3_nic_netdev_ops;
4221 SET_NETDEV_DEV(netdev, &pdev->dev);
4222 hns3_ethtool_set_ops(netdev);
4224 /* Carrier off reporting is important to ethtool even BEFORE open */
4225 netif_carrier_off(netdev);
4227 ret = hns3_get_ring_config(priv);
4230 goto out_get_ring_cfg;
4233 ret = hns3_nic_alloc_vector_data(priv);
4236 goto out_alloc_vector_data;
4239 ret = hns3_nic_init_vector_data(priv);
4242 goto out_init_vector_data;
4245 ret = hns3_init_all_ring(priv);
4251 ret = hns3_init_phy(netdev);
4255 ret = register_netdev(netdev);
4257 dev_err(priv->dev, "probe register netdev fail!\n");
4258 goto out_reg_netdev_fail;
4261 /* the device can work without cpu rmap, only aRFS needs it */
4262 ret = hns3_set_rx_cpu_rmap(netdev);
4264 dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
4266 ret = hns3_nic_init_irq(priv);
4268 dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
4269 hns3_free_rx_cpu_rmap(netdev);
4270 goto out_init_irq_fail;
4273 ret = hns3_client_start(handle);
4275 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
4276 goto out_client_start;
4279 hns3_dcbnl_setup(handle);
4281 hns3_dbg_init(handle);
4283 netdev->max_mtu = HNS3_MAX_MTU(ae_dev->dev_specs.max_frm_size);
4285 if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps))
4286 set_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state);
4288 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
4290 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
4291 set_bit(HNAE3_PFLAG_LIMIT_PROMISC, &handle->supported_pflags);
4293 if (netif_msg_drv(handle))
4294 hns3_info_show(priv);
4299 hns3_free_rx_cpu_rmap(netdev);
4300 hns3_nic_uninit_irq(priv);
4302 unregister_netdev(netdev);
4303 out_reg_netdev_fail:
4304 hns3_uninit_phy(netdev);
4306 hns3_uninit_all_ring(priv);
4308 hns3_nic_uninit_vector_data(priv);
4309 out_init_vector_data:
4310 hns3_nic_dealloc_vector_data(priv);
4311 out_alloc_vector_data:
4314 priv->ae_handle = NULL;
4315 free_netdev(netdev);
4319 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
4321 struct net_device *netdev = handle->kinfo.netdev;
4322 struct hns3_nic_priv *priv = netdev_priv(netdev);
4324 if (netdev->reg_state != NETREG_UNINITIALIZED)
4325 unregister_netdev(netdev);
4327 hns3_client_stop(handle);
4329 hns3_uninit_phy(netdev);
4331 if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
4332 netdev_warn(netdev, "already uninitialized\n");
4333 goto out_netdev_free;
4336 hns3_free_rx_cpu_rmap(netdev);
4338 hns3_nic_uninit_irq(priv);
4340 hns3_del_all_fd_rules(netdev, true);
4342 hns3_clear_all_ring(handle, true);
4344 hns3_nic_uninit_vector_data(priv);
4346 hns3_nic_dealloc_vector_data(priv);
4348 hns3_uninit_all_ring(priv);
4350 hns3_put_ring_config(priv);
4353 hns3_dbg_uninit(handle);
4354 free_netdev(netdev);
4357 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
4359 struct net_device *netdev = handle->kinfo.netdev;
4365 netif_tx_wake_all_queues(netdev);
4366 netif_carrier_on(netdev);
4367 if (netif_msg_link(handle))
4368 netdev_info(netdev, "link up\n");
4370 netif_carrier_off(netdev);
4371 netif_tx_stop_all_queues(netdev);
4372 if (netif_msg_link(handle))
4373 netdev_info(netdev, "link down\n");
4377 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
4379 while (ring->next_to_clean != ring->next_to_use) {
4380 ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
4381 hns3_free_buffer_detach(ring, ring->next_to_clean, 0);
4382 ring_ptr_move_fw(ring, next_to_clean);
4385 ring->pending_buf = 0;
4388 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
4390 struct hns3_desc_cb res_cbs;
4393 while (ring->next_to_use != ring->next_to_clean) {
4394 /* When a buffer is not reused, it's memory has been
4395 * freed in hns3_handle_rx_bd or will be freed by
4396 * stack, so we need to replace the buffer here.
4398 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
4399 ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
4401 u64_stats_update_begin(&ring->syncp);
4402 ring->stats.sw_err_cnt++;
4403 u64_stats_update_end(&ring->syncp);
4404 /* if alloc new buffer fail, exit directly
4405 * and reclear in up flow.
4407 netdev_warn(ring_to_netdev(ring),
4408 "reserve buffer map failed, ret = %d\n",
4412 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
4414 ring_ptr_move_fw(ring, next_to_use);
4417 /* Free the pending skb in rx ring */
4419 dev_kfree_skb_any(ring->skb);
4421 ring->pending_buf = 0;
4427 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
4429 while (ring->next_to_use != ring->next_to_clean) {
4430 /* When a buffer is not reused, it's memory has been
4431 * freed in hns3_handle_rx_bd or will be freed by
4432 * stack, so only need to unmap the buffer here.
4434 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
4435 hns3_unmap_buffer(ring,
4436 &ring->desc_cb[ring->next_to_use]);
4437 ring->desc_cb[ring->next_to_use].dma = 0;
4440 ring_ptr_move_fw(ring, next_to_use);
4444 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force)
4446 struct net_device *ndev = h->kinfo.netdev;
4447 struct hns3_nic_priv *priv = netdev_priv(ndev);
4450 for (i = 0; i < h->kinfo.num_tqps; i++) {
4451 struct hns3_enet_ring *ring;
4453 ring = &priv->ring[i];
4454 hns3_clear_tx_ring(ring);
4456 ring = &priv->ring[i + h->kinfo.num_tqps];
4457 /* Continue to clear other rings even if clearing some
4461 hns3_force_clear_rx_ring(ring);
4463 hns3_clear_rx_ring(ring);
4467 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
4469 struct net_device *ndev = h->kinfo.netdev;
4470 struct hns3_nic_priv *priv = netdev_priv(ndev);
4471 struct hns3_enet_ring *rx_ring;
4475 for (i = 0; i < h->kinfo.num_tqps; i++) {
4476 ret = h->ae_algo->ops->reset_queue(h, i);
4480 hns3_init_ring_hw(&priv->ring[i]);
4482 /* We need to clear tx ring here because self test will
4483 * use the ring and will not run down before up
4485 hns3_clear_tx_ring(&priv->ring[i]);
4486 priv->ring[i].next_to_clean = 0;
4487 priv->ring[i].next_to_use = 0;
4488 priv->ring[i].last_to_use = 0;
4490 rx_ring = &priv->ring[i + h->kinfo.num_tqps];
4491 hns3_init_ring_hw(rx_ring);
4492 ret = hns3_clear_rx_ring(rx_ring);
4496 /* We can not know the hardware head and tail when this
4497 * function is called in reset flow, so we reuse all desc.
4499 for (j = 0; j < rx_ring->desc_num; j++)
4500 hns3_reuse_buffer(rx_ring, j);
4502 rx_ring->next_to_clean = 0;
4503 rx_ring->next_to_use = 0;
4506 hns3_init_tx_ring_tc(priv);
4511 static void hns3_store_coal(struct hns3_nic_priv *priv)
4513 /* ethtool only support setting and querying one coal
4514 * configuration for now, so save the vector 0' coal
4515 * configuration here in order to restore it.
4517 memcpy(&priv->tx_coal, &priv->tqp_vector[0].tx_group.coal,
4518 sizeof(struct hns3_enet_coalesce));
4519 memcpy(&priv->rx_coal, &priv->tqp_vector[0].rx_group.coal,
4520 sizeof(struct hns3_enet_coalesce));
4523 static void hns3_restore_coal(struct hns3_nic_priv *priv)
4525 u16 vector_num = priv->vector_num;
4528 for (i = 0; i < vector_num; i++) {
4529 memcpy(&priv->tqp_vector[i].tx_group.coal, &priv->tx_coal,
4530 sizeof(struct hns3_enet_coalesce));
4531 memcpy(&priv->tqp_vector[i].rx_group.coal, &priv->rx_coal,
4532 sizeof(struct hns3_enet_coalesce));
4536 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
4538 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4539 struct net_device *ndev = kinfo->netdev;
4540 struct hns3_nic_priv *priv = netdev_priv(ndev);
4542 if (test_and_set_bit(HNS3_NIC_STATE_RESETTING, &priv->state))
4545 if (!netif_running(ndev))
4548 return hns3_nic_net_stop(ndev);
4551 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
4553 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4554 struct hns3_nic_priv *priv = netdev_priv(kinfo->netdev);
4557 clear_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4559 if (netif_running(kinfo->netdev)) {
4560 ret = hns3_nic_net_open(kinfo->netdev);
4562 set_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4563 netdev_err(kinfo->netdev,
4564 "net up fail, ret=%d!\n", ret);
4572 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
4574 struct net_device *netdev = handle->kinfo.netdev;
4575 struct hns3_nic_priv *priv = netdev_priv(netdev);
4578 /* Carrier off reporting is important to ethtool even BEFORE open */
4579 netif_carrier_off(netdev);
4581 ret = hns3_get_ring_config(priv);
4585 ret = hns3_nic_alloc_vector_data(priv);
4589 hns3_restore_coal(priv);
4591 ret = hns3_nic_init_vector_data(priv);
4593 goto err_dealloc_vector;
4595 ret = hns3_init_all_ring(priv);
4597 goto err_uninit_vector;
4599 /* the device can work without cpu rmap, only aRFS needs it */
4600 ret = hns3_set_rx_cpu_rmap(netdev);
4602 dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
4604 ret = hns3_nic_init_irq(priv);
4606 dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
4607 hns3_free_rx_cpu_rmap(netdev);
4608 goto err_init_irq_fail;
4611 if (!hns3_is_phys_func(handle->pdev))
4612 hns3_init_mac_addr(netdev);
4614 ret = hns3_client_start(handle);
4616 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
4617 goto err_client_start_fail;
4620 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
4624 err_client_start_fail:
4625 hns3_free_rx_cpu_rmap(netdev);
4626 hns3_nic_uninit_irq(priv);
4628 hns3_uninit_all_ring(priv);
4630 hns3_nic_uninit_vector_data(priv);
4632 hns3_nic_dealloc_vector_data(priv);
4634 hns3_put_ring_config(priv);
4639 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
4641 struct net_device *netdev = handle->kinfo.netdev;
4642 struct hns3_nic_priv *priv = netdev_priv(netdev);
4644 if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
4645 netdev_warn(netdev, "already uninitialized\n");
4649 hns3_free_rx_cpu_rmap(netdev);
4650 hns3_nic_uninit_irq(priv);
4651 hns3_clear_all_ring(handle, true);
4652 hns3_reset_tx_queue(priv->ae_handle);
4654 hns3_nic_uninit_vector_data(priv);
4656 hns3_store_coal(priv);
4658 hns3_nic_dealloc_vector_data(priv);
4660 hns3_uninit_all_ring(priv);
4662 hns3_put_ring_config(priv);
4667 static int hns3_reset_notify(struct hnae3_handle *handle,
4668 enum hnae3_reset_notify_type type)
4673 case HNAE3_UP_CLIENT:
4674 ret = hns3_reset_notify_up_enet(handle);
4676 case HNAE3_DOWN_CLIENT:
4677 ret = hns3_reset_notify_down_enet(handle);
4679 case HNAE3_INIT_CLIENT:
4680 ret = hns3_reset_notify_init_enet(handle);
4682 case HNAE3_UNINIT_CLIENT:
4683 ret = hns3_reset_notify_uninit_enet(handle);
4692 static int hns3_change_channels(struct hnae3_handle *handle, u32 new_tqp_num,
4693 bool rxfh_configured)
4697 ret = handle->ae_algo->ops->set_channels(handle, new_tqp_num,
4700 dev_err(&handle->pdev->dev,
4701 "Change tqp num(%u) fail.\n", new_tqp_num);
4705 ret = hns3_reset_notify(handle, HNAE3_INIT_CLIENT);
4709 ret = hns3_reset_notify(handle, HNAE3_UP_CLIENT);
4711 hns3_reset_notify(handle, HNAE3_UNINIT_CLIENT);
4716 int hns3_set_channels(struct net_device *netdev,
4717 struct ethtool_channels *ch)
4719 struct hnae3_handle *h = hns3_get_handle(netdev);
4720 struct hnae3_knic_private_info *kinfo = &h->kinfo;
4721 bool rxfh_configured = netif_is_rxfh_configured(netdev);
4722 u32 new_tqp_num = ch->combined_count;
4726 if (hns3_nic_resetting(netdev))
4729 if (ch->rx_count || ch->tx_count)
4732 if (kinfo->tc_info.mqprio_active) {
4733 dev_err(&netdev->dev,
4734 "it's not allowed to set channels via ethtool when MQPRIO mode is on\n");
4738 if (new_tqp_num > hns3_get_max_available_channels(h) ||
4740 dev_err(&netdev->dev,
4741 "Change tqps fail, the tqp range is from 1 to %u",
4742 hns3_get_max_available_channels(h));
4746 if (kinfo->rss_size == new_tqp_num)
4749 netif_dbg(h, drv, netdev,
4750 "set channels: tqp_num=%u, rxfh=%d\n",
4751 new_tqp_num, rxfh_configured);
4753 ret = hns3_reset_notify(h, HNAE3_DOWN_CLIENT);
4757 ret = hns3_reset_notify(h, HNAE3_UNINIT_CLIENT);
4761 org_tqp_num = h->kinfo.num_tqps;
4762 ret = hns3_change_channels(h, new_tqp_num, rxfh_configured);
4767 "Change channels fail, revert to old value\n");
4768 ret1 = hns3_change_channels(h, org_tqp_num, rxfh_configured);
4771 "revert to old channel fail\n");
4781 static const struct hns3_hw_error_info hns3_hw_err[] = {
4782 { .type = HNAE3_PPU_POISON_ERROR,
4783 .msg = "PPU poison" },
4784 { .type = HNAE3_CMDQ_ECC_ERROR,
4785 .msg = "IMP CMDQ error" },
4786 { .type = HNAE3_IMP_RD_POISON_ERROR,
4787 .msg = "IMP RD poison" },
4788 { .type = HNAE3_ROCEE_AXI_RESP_ERROR,
4789 .msg = "ROCEE AXI RESP error" },
4792 static void hns3_process_hw_error(struct hnae3_handle *handle,
4793 enum hnae3_hw_error_type type)
4797 for (i = 0; i < ARRAY_SIZE(hns3_hw_err); i++) {
4798 if (hns3_hw_err[i].type == type) {
4799 dev_err(&handle->pdev->dev, "Detected %s!\n",
4800 hns3_hw_err[i].msg);
4806 static const struct hnae3_client_ops client_ops = {
4807 .init_instance = hns3_client_init,
4808 .uninit_instance = hns3_client_uninit,
4809 .link_status_change = hns3_link_status_change,
4810 .reset_notify = hns3_reset_notify,
4811 .process_hw_error = hns3_process_hw_error,
4814 /* hns3_init_module - Driver registration routine
4815 * hns3_init_module is the first routine called when the driver is
4816 * loaded. All it does is register with the PCI subsystem.
4818 static int __init hns3_init_module(void)
4822 pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
4823 pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
4825 client.type = HNAE3_CLIENT_KNIC;
4826 snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH, "%s",
4829 client.ops = &client_ops;
4831 INIT_LIST_HEAD(&client.node);
4833 hns3_dbg_register_debugfs(hns3_driver_name);
4835 ret = hnae3_register_client(&client);
4837 goto err_reg_client;
4839 ret = pci_register_driver(&hns3_driver);
4841 goto err_reg_driver;
4846 hnae3_unregister_client(&client);
4848 hns3_dbg_unregister_debugfs();
4851 module_init(hns3_init_module);
4853 /* hns3_exit_module - Driver exit cleanup routine
4854 * hns3_exit_module is called just before the driver is removed
4857 static void __exit hns3_exit_module(void)
4859 pci_unregister_driver(&hns3_driver);
4860 hnae3_unregister_client(&client);
4861 hns3_dbg_unregister_debugfs();
4863 module_exit(hns3_exit_module);
4865 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
4866 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
4867 MODULE_LICENSE("GPL");
4868 MODULE_ALIAS("pci:hns-nic");