netdev: pass the stuck queue to the timeout handler
[linux-block.git] / drivers / net / ethernet / hisilicon / hns3 / hns3_enet.c
1 // SPDX-License-Identifier: GPL-2.0+
2 // Copyright (c) 2016-2017 Hisilicon Limited.
3
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>
9 #endif
10 #include <linux/if_vlan.h>
11 #include <linux/ip.h>
12 #include <linux/ipv6.h>
13 #include <linux/module.h>
14 #include <linux/pci.h>
15 #include <linux/aer.h>
16 #include <linux/skbuff.h>
17 #include <linux/sctp.h>
18 #include <linux/vermagic.h>
19 #include <net/gre.h>
20 #include <net/ip6_checksum.h>
21 #include <net/pkt_cls.h>
22 #include <net/tcp.h>
23 #include <net/vxlan.h>
24
25 #include "hnae3.h"
26 #include "hns3_enet.h"
27
28 #define hns3_set_field(origin, shift, val)      ((origin) |= ((val) << (shift)))
29 #define hns3_tx_bd_count(S)     DIV_ROUND_UP(S, HNS3_MAX_BD_SIZE)
30
31 #define hns3_rl_err(fmt, ...)                                           \
32         do {                                                            \
33                 if (net_ratelimit())                                    \
34                         netdev_err(fmt, ##__VA_ARGS__);                 \
35         } while (0)
36
37 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force);
38 static void hns3_remove_hw_addr(struct net_device *netdev);
39
40 static const char hns3_driver_name[] = "hns3";
41 const char hns3_driver_version[] = VERMAGIC_STRING;
42 static const char hns3_driver_string[] =
43                         "Hisilicon Ethernet Network Driver for Hip08 Family";
44 static const char hns3_copyright[] = "Copyright (c) 2017 Huawei Corporation.";
45 static struct hnae3_client client;
46
47 static int debug = -1;
48 module_param(debug, int, 0);
49 MODULE_PARM_DESC(debug, " Network interface message level setting");
50
51 #define DEFAULT_MSG_LEVEL (NETIF_MSG_PROBE | NETIF_MSG_LINK | \
52                            NETIF_MSG_IFDOWN | NETIF_MSG_IFUP)
53
54 #define HNS3_INNER_VLAN_TAG     1
55 #define HNS3_OUTER_VLAN_TAG     2
56
57 /* hns3_pci_tbl - PCI Device ID Table
58  *
59  * Last entry must be all 0s
60  *
61  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
62  *   Class, Class Mask, private data (not used) }
63  */
64 static const struct pci_device_id hns3_pci_tbl[] = {
65         {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
66         {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
67         {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA),
68          HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
69         {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC),
70          HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
71         {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA),
72          HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
73         {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC),
74          HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
75         {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC),
76          HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
77         {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_VF), 0},
78         {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF),
79          HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
80         /* required last entry */
81         {0, }
82 };
83 MODULE_DEVICE_TABLE(pci, hns3_pci_tbl);
84
85 static irqreturn_t hns3_irq_handle(int irq, void *vector)
86 {
87         struct hns3_enet_tqp_vector *tqp_vector = vector;
88
89         napi_schedule_irqoff(&tqp_vector->napi);
90
91         return IRQ_HANDLED;
92 }
93
94 static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv)
95 {
96         struct hns3_enet_tqp_vector *tqp_vectors;
97         unsigned int i;
98
99         for (i = 0; i < priv->vector_num; i++) {
100                 tqp_vectors = &priv->tqp_vector[i];
101
102                 if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED)
103                         continue;
104
105                 /* clear the affinity mask */
106                 irq_set_affinity_hint(tqp_vectors->vector_irq, NULL);
107
108                 /* release the irq resource */
109                 free_irq(tqp_vectors->vector_irq, tqp_vectors);
110                 tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED;
111         }
112 }
113
114 static int hns3_nic_init_irq(struct hns3_nic_priv *priv)
115 {
116         struct hns3_enet_tqp_vector *tqp_vectors;
117         int txrx_int_idx = 0;
118         int rx_int_idx = 0;
119         int tx_int_idx = 0;
120         unsigned int i;
121         int ret;
122
123         for (i = 0; i < priv->vector_num; i++) {
124                 tqp_vectors = &priv->tqp_vector[i];
125
126                 if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED)
127                         continue;
128
129                 if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) {
130                         snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
131                                  "%s-%s-%d", priv->netdev->name, "TxRx",
132                                  txrx_int_idx++);
133                         txrx_int_idx++;
134                 } else if (tqp_vectors->rx_group.ring) {
135                         snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
136                                  "%s-%s-%d", priv->netdev->name, "Rx",
137                                  rx_int_idx++);
138                 } else if (tqp_vectors->tx_group.ring) {
139                         snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
140                                  "%s-%s-%d", priv->netdev->name, "Tx",
141                                  tx_int_idx++);
142                 } else {
143                         /* Skip this unused q_vector */
144                         continue;
145                 }
146
147                 tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0';
148
149                 ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0,
150                                   tqp_vectors->name, tqp_vectors);
151                 if (ret) {
152                         netdev_err(priv->netdev, "request irq(%d) fail\n",
153                                    tqp_vectors->vector_irq);
154                         hns3_nic_uninit_irq(priv);
155                         return ret;
156                 }
157
158                 irq_set_affinity_hint(tqp_vectors->vector_irq,
159                                       &tqp_vectors->affinity_mask);
160
161                 tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED;
162         }
163
164         return 0;
165 }
166
167 static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector,
168                                  u32 mask_en)
169 {
170         writel(mask_en, tqp_vector->mask_addr);
171 }
172
173 static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector)
174 {
175         napi_enable(&tqp_vector->napi);
176
177         /* enable vector */
178         hns3_mask_vector_irq(tqp_vector, 1);
179 }
180
181 static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector)
182 {
183         /* disable vector */
184         hns3_mask_vector_irq(tqp_vector, 0);
185
186         disable_irq(tqp_vector->vector_irq);
187         napi_disable(&tqp_vector->napi);
188 }
189
190 void hns3_set_vector_coalesce_rl(struct hns3_enet_tqp_vector *tqp_vector,
191                                  u32 rl_value)
192 {
193         u32 rl_reg = hns3_rl_usec_to_reg(rl_value);
194
195         /* this defines the configuration for RL (Interrupt Rate Limiter).
196          * Rl defines rate of interrupts i.e. number of interrupts-per-second
197          * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing
198          */
199
200         if (rl_reg > 0 && !tqp_vector->tx_group.coal.gl_adapt_enable &&
201             !tqp_vector->rx_group.coal.gl_adapt_enable)
202                 /* According to the hardware, the range of rl_reg is
203                  * 0-59 and the unit is 4.
204                  */
205                 rl_reg |=  HNS3_INT_RL_ENABLE_MASK;
206
207         writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
208 }
209
210 void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
211                                     u32 gl_value)
212 {
213         u32 rx_gl_reg = hns3_gl_usec_to_reg(gl_value);
214
215         writel(rx_gl_reg, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
216 }
217
218 void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
219                                     u32 gl_value)
220 {
221         u32 tx_gl_reg = hns3_gl_usec_to_reg(gl_value);
222
223         writel(tx_gl_reg, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
224 }
225
226 static void hns3_vector_gl_rl_init(struct hns3_enet_tqp_vector *tqp_vector,
227                                    struct hns3_nic_priv *priv)
228 {
229         /* initialize the configuration for interrupt coalescing.
230          * 1. GL (Interrupt Gap Limiter)
231          * 2. RL (Interrupt Rate Limiter)
232          *
233          * Default: enable interrupt coalescing self-adaptive and GL
234          */
235         tqp_vector->tx_group.coal.gl_adapt_enable = 1;
236         tqp_vector->rx_group.coal.gl_adapt_enable = 1;
237
238         tqp_vector->tx_group.coal.int_gl = HNS3_INT_GL_50K;
239         tqp_vector->rx_group.coal.int_gl = HNS3_INT_GL_50K;
240
241         tqp_vector->rx_group.coal.flow_level = HNS3_FLOW_LOW;
242         tqp_vector->tx_group.coal.flow_level = HNS3_FLOW_LOW;
243 }
244
245 static void hns3_vector_gl_rl_init_hw(struct hns3_enet_tqp_vector *tqp_vector,
246                                       struct hns3_nic_priv *priv)
247 {
248         struct hnae3_handle *h = priv->ae_handle;
249
250         hns3_set_vector_coalesce_tx_gl(tqp_vector,
251                                        tqp_vector->tx_group.coal.int_gl);
252         hns3_set_vector_coalesce_rx_gl(tqp_vector,
253                                        tqp_vector->rx_group.coal.int_gl);
254         hns3_set_vector_coalesce_rl(tqp_vector, h->kinfo.int_rl_setting);
255 }
256
257 static int hns3_nic_set_real_num_queue(struct net_device *netdev)
258 {
259         struct hnae3_handle *h = hns3_get_handle(netdev);
260         struct hnae3_knic_private_info *kinfo = &h->kinfo;
261         unsigned int queue_size = kinfo->rss_size * kinfo->num_tc;
262         int i, ret;
263
264         if (kinfo->num_tc <= 1) {
265                 netdev_reset_tc(netdev);
266         } else {
267                 ret = netdev_set_num_tc(netdev, kinfo->num_tc);
268                 if (ret) {
269                         netdev_err(netdev,
270                                    "netdev_set_num_tc fail, ret=%d!\n", ret);
271                         return ret;
272                 }
273
274                 for (i = 0; i < HNAE3_MAX_TC; i++) {
275                         if (!kinfo->tc_info[i].enable)
276                                 continue;
277
278                         netdev_set_tc_queue(netdev,
279                                             kinfo->tc_info[i].tc,
280                                             kinfo->tc_info[i].tqp_count,
281                                             kinfo->tc_info[i].tqp_offset);
282                 }
283         }
284
285         ret = netif_set_real_num_tx_queues(netdev, queue_size);
286         if (ret) {
287                 netdev_err(netdev,
288                            "netif_set_real_num_tx_queues fail, ret=%d!\n", ret);
289                 return ret;
290         }
291
292         ret = netif_set_real_num_rx_queues(netdev, queue_size);
293         if (ret) {
294                 netdev_err(netdev,
295                            "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
296                 return ret;
297         }
298
299         return 0;
300 }
301
302 static u16 hns3_get_max_available_channels(struct hnae3_handle *h)
303 {
304         u16 alloc_tqps, max_rss_size, rss_size;
305
306         h->ae_algo->ops->get_tqps_and_rss_info(h, &alloc_tqps, &max_rss_size);
307         rss_size = alloc_tqps / h->kinfo.num_tc;
308
309         return min_t(u16, rss_size, max_rss_size);
310 }
311
312 static void hns3_tqp_enable(struct hnae3_queue *tqp)
313 {
314         u32 rcb_reg;
315
316         rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
317         rcb_reg |= BIT(HNS3_RING_EN_B);
318         hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
319 }
320
321 static void hns3_tqp_disable(struct hnae3_queue *tqp)
322 {
323         u32 rcb_reg;
324
325         rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
326         rcb_reg &= ~BIT(HNS3_RING_EN_B);
327         hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
328 }
329
330 static void hns3_free_rx_cpu_rmap(struct net_device *netdev)
331 {
332 #ifdef CONFIG_RFS_ACCEL
333         free_irq_cpu_rmap(netdev->rx_cpu_rmap);
334         netdev->rx_cpu_rmap = NULL;
335 #endif
336 }
337
338 static int hns3_set_rx_cpu_rmap(struct net_device *netdev)
339 {
340 #ifdef CONFIG_RFS_ACCEL
341         struct hns3_nic_priv *priv = netdev_priv(netdev);
342         struct hns3_enet_tqp_vector *tqp_vector;
343         int i, ret;
344
345         if (!netdev->rx_cpu_rmap) {
346                 netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->vector_num);
347                 if (!netdev->rx_cpu_rmap)
348                         return -ENOMEM;
349         }
350
351         for (i = 0; i < priv->vector_num; i++) {
352                 tqp_vector = &priv->tqp_vector[i];
353                 ret = irq_cpu_rmap_add(netdev->rx_cpu_rmap,
354                                        tqp_vector->vector_irq);
355                 if (ret) {
356                         hns3_free_rx_cpu_rmap(netdev);
357                         return ret;
358                 }
359         }
360 #endif
361         return 0;
362 }
363
364 static int hns3_nic_net_up(struct net_device *netdev)
365 {
366         struct hns3_nic_priv *priv = netdev_priv(netdev);
367         struct hnae3_handle *h = priv->ae_handle;
368         int i, j;
369         int ret;
370
371         ret = hns3_nic_reset_all_ring(h);
372         if (ret)
373                 return ret;
374
375         /* the device can work without cpu rmap, only aRFS needs it */
376         ret = hns3_set_rx_cpu_rmap(netdev);
377         if (ret)
378                 netdev_warn(netdev, "set rx cpu rmap fail, ret=%d!\n", ret);
379
380         /* get irq resource for all vectors */
381         ret = hns3_nic_init_irq(priv);
382         if (ret) {
383                 netdev_err(netdev, "init irq failed! ret=%d\n", ret);
384                 goto free_rmap;
385         }
386
387         clear_bit(HNS3_NIC_STATE_DOWN, &priv->state);
388
389         /* enable the vectors */
390         for (i = 0; i < priv->vector_num; i++)
391                 hns3_vector_enable(&priv->tqp_vector[i]);
392
393         /* enable rcb */
394         for (j = 0; j < h->kinfo.num_tqps; j++)
395                 hns3_tqp_enable(h->kinfo.tqp[j]);
396
397         /* start the ae_dev */
398         ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0;
399         if (ret)
400                 goto out_start_err;
401
402         return 0;
403
404 out_start_err:
405         set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
406         while (j--)
407                 hns3_tqp_disable(h->kinfo.tqp[j]);
408
409         for (j = i - 1; j >= 0; j--)
410                 hns3_vector_disable(&priv->tqp_vector[j]);
411
412         hns3_nic_uninit_irq(priv);
413 free_rmap:
414         hns3_free_rx_cpu_rmap(netdev);
415         return ret;
416 }
417
418 static void hns3_config_xps(struct hns3_nic_priv *priv)
419 {
420         int i;
421
422         for (i = 0; i < priv->vector_num; i++) {
423                 struct hns3_enet_tqp_vector *tqp_vector = &priv->tqp_vector[i];
424                 struct hns3_enet_ring *ring = tqp_vector->tx_group.ring;
425
426                 while (ring) {
427                         int ret;
428
429                         ret = netif_set_xps_queue(priv->netdev,
430                                                   &tqp_vector->affinity_mask,
431                                                   ring->tqp->tqp_index);
432                         if (ret)
433                                 netdev_warn(priv->netdev,
434                                             "set xps queue failed: %d", ret);
435
436                         ring = ring->next;
437                 }
438         }
439 }
440
441 static int hns3_nic_net_open(struct net_device *netdev)
442 {
443         struct hns3_nic_priv *priv = netdev_priv(netdev);
444         struct hnae3_handle *h = hns3_get_handle(netdev);
445         struct hnae3_knic_private_info *kinfo;
446         int i, ret;
447
448         if (hns3_nic_resetting(netdev))
449                 return -EBUSY;
450
451         netif_carrier_off(netdev);
452
453         ret = hns3_nic_set_real_num_queue(netdev);
454         if (ret)
455                 return ret;
456
457         ret = hns3_nic_net_up(netdev);
458         if (ret) {
459                 netdev_err(netdev, "net up fail, ret=%d!\n", ret);
460                 return ret;
461         }
462
463         kinfo = &h->kinfo;
464         for (i = 0; i < HNAE3_MAX_USER_PRIO; i++)
465                 netdev_set_prio_tc_map(netdev, i, kinfo->prio_tc[i]);
466
467         if (h->ae_algo->ops->set_timer_task)
468                 h->ae_algo->ops->set_timer_task(priv->ae_handle, true);
469
470         hns3_config_xps(priv);
471
472         netif_dbg(h, drv, netdev, "net open\n");
473
474         return 0;
475 }
476
477 static void hns3_reset_tx_queue(struct hnae3_handle *h)
478 {
479         struct net_device *ndev = h->kinfo.netdev;
480         struct hns3_nic_priv *priv = netdev_priv(ndev);
481         struct netdev_queue *dev_queue;
482         u32 i;
483
484         for (i = 0; i < h->kinfo.num_tqps; i++) {
485                 dev_queue = netdev_get_tx_queue(ndev,
486                                                 priv->ring[i].queue_index);
487                 netdev_tx_reset_queue(dev_queue);
488         }
489 }
490
491 static void hns3_nic_net_down(struct net_device *netdev)
492 {
493         struct hns3_nic_priv *priv = netdev_priv(netdev);
494         struct hnae3_handle *h = hns3_get_handle(netdev);
495         const struct hnae3_ae_ops *ops;
496         int i;
497
498         /* disable vectors */
499         for (i = 0; i < priv->vector_num; i++)
500                 hns3_vector_disable(&priv->tqp_vector[i]);
501
502         /* disable rcb */
503         for (i = 0; i < h->kinfo.num_tqps; i++)
504                 hns3_tqp_disable(h->kinfo.tqp[i]);
505
506         /* stop ae_dev */
507         ops = priv->ae_handle->ae_algo->ops;
508         if (ops->stop)
509                 ops->stop(priv->ae_handle);
510
511         hns3_free_rx_cpu_rmap(netdev);
512
513         /* free irq resources */
514         hns3_nic_uninit_irq(priv);
515
516         /* delay ring buffer clearing to hns3_reset_notify_uninit_enet
517          * during reset process, because driver may not be able
518          * to disable the ring through firmware when downing the netdev.
519          */
520         if (!hns3_nic_resetting(netdev))
521                 hns3_clear_all_ring(priv->ae_handle, false);
522
523         hns3_reset_tx_queue(priv->ae_handle);
524 }
525
526 static int hns3_nic_net_stop(struct net_device *netdev)
527 {
528         struct hns3_nic_priv *priv = netdev_priv(netdev);
529         struct hnae3_handle *h = hns3_get_handle(netdev);
530
531         if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
532                 return 0;
533
534         netif_dbg(h, drv, netdev, "net stop\n");
535
536         if (h->ae_algo->ops->set_timer_task)
537                 h->ae_algo->ops->set_timer_task(priv->ae_handle, false);
538
539         netif_tx_stop_all_queues(netdev);
540         netif_carrier_off(netdev);
541
542         hns3_nic_net_down(netdev);
543
544         return 0;
545 }
546
547 static int hns3_nic_uc_sync(struct net_device *netdev,
548                             const unsigned char *addr)
549 {
550         struct hnae3_handle *h = hns3_get_handle(netdev);
551
552         if (h->ae_algo->ops->add_uc_addr)
553                 return h->ae_algo->ops->add_uc_addr(h, addr);
554
555         return 0;
556 }
557
558 static int hns3_nic_uc_unsync(struct net_device *netdev,
559                               const unsigned char *addr)
560 {
561         struct hnae3_handle *h = hns3_get_handle(netdev);
562
563         if (h->ae_algo->ops->rm_uc_addr)
564                 return h->ae_algo->ops->rm_uc_addr(h, addr);
565
566         return 0;
567 }
568
569 static int hns3_nic_mc_sync(struct net_device *netdev,
570                             const unsigned char *addr)
571 {
572         struct hnae3_handle *h = hns3_get_handle(netdev);
573
574         if (h->ae_algo->ops->add_mc_addr)
575                 return h->ae_algo->ops->add_mc_addr(h, addr);
576
577         return 0;
578 }
579
580 static int hns3_nic_mc_unsync(struct net_device *netdev,
581                               const unsigned char *addr)
582 {
583         struct hnae3_handle *h = hns3_get_handle(netdev);
584
585         if (h->ae_algo->ops->rm_mc_addr)
586                 return h->ae_algo->ops->rm_mc_addr(h, addr);
587
588         return 0;
589 }
590
591 static u8 hns3_get_netdev_flags(struct net_device *netdev)
592 {
593         u8 flags = 0;
594
595         if (netdev->flags & IFF_PROMISC) {
596                 flags = HNAE3_USER_UPE | HNAE3_USER_MPE | HNAE3_BPE;
597         } else {
598                 flags |= HNAE3_VLAN_FLTR;
599                 if (netdev->flags & IFF_ALLMULTI)
600                         flags |= HNAE3_USER_MPE;
601         }
602
603         return flags;
604 }
605
606 static void hns3_nic_set_rx_mode(struct net_device *netdev)
607 {
608         struct hnae3_handle *h = hns3_get_handle(netdev);
609         u8 new_flags;
610         int ret;
611
612         new_flags = hns3_get_netdev_flags(netdev);
613
614         ret = __dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync);
615         if (ret) {
616                 netdev_err(netdev, "sync uc address fail\n");
617                 if (ret == -ENOSPC)
618                         new_flags |= HNAE3_OVERFLOW_UPE;
619         }
620
621         if (netdev->flags & IFF_MULTICAST) {
622                 ret = __dev_mc_sync(netdev, hns3_nic_mc_sync,
623                                     hns3_nic_mc_unsync);
624                 if (ret) {
625                         netdev_err(netdev, "sync mc address fail\n");
626                         if (ret == -ENOSPC)
627                                 new_flags |= HNAE3_OVERFLOW_MPE;
628                 }
629         }
630
631         /* User mode Promisc mode enable and vlan filtering is disabled to
632          * let all packets in. MAC-VLAN Table overflow Promisc enabled and
633          * vlan fitering is enabled
634          */
635         hns3_enable_vlan_filter(netdev, new_flags & HNAE3_VLAN_FLTR);
636         h->netdev_flags = new_flags;
637         hns3_update_promisc_mode(netdev, new_flags);
638 }
639
640 int hns3_update_promisc_mode(struct net_device *netdev, u8 promisc_flags)
641 {
642         struct hns3_nic_priv *priv = netdev_priv(netdev);
643         struct hnae3_handle *h = priv->ae_handle;
644
645         if (h->ae_algo->ops->set_promisc_mode) {
646                 return h->ae_algo->ops->set_promisc_mode(h,
647                                                 promisc_flags & HNAE3_UPE,
648                                                 promisc_flags & HNAE3_MPE);
649         }
650
651         return 0;
652 }
653
654 void hns3_enable_vlan_filter(struct net_device *netdev, bool enable)
655 {
656         struct hns3_nic_priv *priv = netdev_priv(netdev);
657         struct hnae3_handle *h = priv->ae_handle;
658         bool last_state;
659
660         if (h->pdev->revision >= 0x21 && h->ae_algo->ops->enable_vlan_filter) {
661                 last_state = h->netdev_flags & HNAE3_VLAN_FLTR ? true : false;
662                 if (enable != last_state) {
663                         netdev_info(netdev,
664                                     "%s vlan filter\n",
665                                     enable ? "enable" : "disable");
666                         h->ae_algo->ops->enable_vlan_filter(h, enable);
667                 }
668         }
669 }
670
671 static int hns3_set_tso(struct sk_buff *skb, u32 *paylen,
672                         u16 *mss, u32 *type_cs_vlan_tso)
673 {
674         u32 l4_offset, hdr_len;
675         union l3_hdr_info l3;
676         union l4_hdr_info l4;
677         u32 l4_paylen;
678         int ret;
679
680         if (!skb_is_gso(skb))
681                 return 0;
682
683         ret = skb_cow_head(skb, 0);
684         if (unlikely(ret < 0))
685                 return ret;
686
687         l3.hdr = skb_network_header(skb);
688         l4.hdr = skb_transport_header(skb);
689
690         /* Software should clear the IPv4's checksum field when tso is
691          * needed.
692          */
693         if (l3.v4->version == 4)
694                 l3.v4->check = 0;
695
696         /* tunnel packet */
697         if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
698                                          SKB_GSO_GRE_CSUM |
699                                          SKB_GSO_UDP_TUNNEL |
700                                          SKB_GSO_UDP_TUNNEL_CSUM)) {
701                 if ((!(skb_shinfo(skb)->gso_type &
702                     SKB_GSO_PARTIAL)) &&
703                     (skb_shinfo(skb)->gso_type &
704                     SKB_GSO_UDP_TUNNEL_CSUM)) {
705                         /* Software should clear the udp's checksum
706                          * field when tso is needed.
707                          */
708                         l4.udp->check = 0;
709                 }
710                 /* reset l3&l4 pointers from outer to inner headers */
711                 l3.hdr = skb_inner_network_header(skb);
712                 l4.hdr = skb_inner_transport_header(skb);
713
714                 /* Software should clear the IPv4's checksum field when
715                  * tso is needed.
716                  */
717                 if (l3.v4->version == 4)
718                         l3.v4->check = 0;
719         }
720
721         /* normal or tunnel packet */
722         l4_offset = l4.hdr - skb->data;
723         hdr_len = (l4.tcp->doff << 2) + l4_offset;
724
725         /* remove payload length from inner pseudo checksum when tso */
726         l4_paylen = skb->len - l4_offset;
727         csum_replace_by_diff(&l4.tcp->check,
728                              (__force __wsum)htonl(l4_paylen));
729
730         /* find the txbd field values */
731         *paylen = skb->len - hdr_len;
732         hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_TSO_B, 1);
733
734         /* get MSS for TSO */
735         *mss = skb_shinfo(skb)->gso_size;
736
737         return 0;
738 }
739
740 static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
741                                 u8 *il4_proto)
742 {
743         union l3_hdr_info l3;
744         unsigned char *l4_hdr;
745         unsigned char *exthdr;
746         u8 l4_proto_tmp;
747         __be16 frag_off;
748
749         /* find outer header point */
750         l3.hdr = skb_network_header(skb);
751         l4_hdr = skb_transport_header(skb);
752
753         if (skb->protocol == htons(ETH_P_IPV6)) {
754                 exthdr = l3.hdr + sizeof(*l3.v6);
755                 l4_proto_tmp = l3.v6->nexthdr;
756                 if (l4_hdr != exthdr)
757                         ipv6_skip_exthdr(skb, exthdr - skb->data,
758                                          &l4_proto_tmp, &frag_off);
759         } else if (skb->protocol == htons(ETH_P_IP)) {
760                 l4_proto_tmp = l3.v4->protocol;
761         } else {
762                 return -EINVAL;
763         }
764
765         *ol4_proto = l4_proto_tmp;
766
767         /* tunnel packet */
768         if (!skb->encapsulation) {
769                 *il4_proto = 0;
770                 return 0;
771         }
772
773         /* find inner header point */
774         l3.hdr = skb_inner_network_header(skb);
775         l4_hdr = skb_inner_transport_header(skb);
776
777         if (l3.v6->version == 6) {
778                 exthdr = l3.hdr + sizeof(*l3.v6);
779                 l4_proto_tmp = l3.v6->nexthdr;
780                 if (l4_hdr != exthdr)
781                         ipv6_skip_exthdr(skb, exthdr - skb->data,
782                                          &l4_proto_tmp, &frag_off);
783         } else if (l3.v4->version == 4) {
784                 l4_proto_tmp = l3.v4->protocol;
785         }
786
787         *il4_proto = l4_proto_tmp;
788
789         return 0;
790 }
791
792 /* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
793  * and it is udp packet, which has a dest port as the IANA assigned.
794  * the hardware is expected to do the checksum offload, but the
795  * hardware will not do the checksum offload when udp dest port is
796  * 4789.
797  */
798 static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
799 {
800         union l4_hdr_info l4;
801
802         l4.hdr = skb_transport_header(skb);
803
804         if (!(!skb->encapsulation &&
805               l4.udp->dest == htons(IANA_VXLAN_UDP_PORT)))
806                 return false;
807
808         skb_checksum_help(skb);
809
810         return true;
811 }
812
813 static void hns3_set_outer_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
814                                   u32 *ol_type_vlan_len_msec)
815 {
816         u32 l2_len, l3_len, l4_len;
817         unsigned char *il2_hdr;
818         union l3_hdr_info l3;
819         union l4_hdr_info l4;
820
821         l3.hdr = skb_network_header(skb);
822         l4.hdr = skb_transport_header(skb);
823
824         /* compute OL2 header size, defined in 2 Bytes */
825         l2_len = l3.hdr - skb->data;
826         hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L2LEN_S, l2_len >> 1);
827
828         /* compute OL3 header size, defined in 4 Bytes */
829         l3_len = l4.hdr - l3.hdr;
830         hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_S, l3_len >> 2);
831
832         il2_hdr = skb_inner_mac_header(skb);
833         /* compute OL4 header size, defined in 4 Bytes */
834         l4_len = il2_hdr - l4.hdr;
835         hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L4LEN_S, l4_len >> 2);
836
837         /* define outer network header type */
838         if (skb->protocol == htons(ETH_P_IP)) {
839                 if (skb_is_gso(skb))
840                         hns3_set_field(*ol_type_vlan_len_msec,
841                                        HNS3_TXD_OL3T_S,
842                                        HNS3_OL3T_IPV4_CSUM);
843                 else
844                         hns3_set_field(*ol_type_vlan_len_msec,
845                                        HNS3_TXD_OL3T_S,
846                                        HNS3_OL3T_IPV4_NO_CSUM);
847
848         } else if (skb->protocol == htons(ETH_P_IPV6)) {
849                 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_S,
850                                HNS3_OL3T_IPV6);
851         }
852
853         if (ol4_proto == IPPROTO_UDP)
854                 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
855                                HNS3_TUN_MAC_IN_UDP);
856         else if (ol4_proto == IPPROTO_GRE)
857                 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
858                                HNS3_TUN_NVGRE);
859 }
860
861 static int hns3_set_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
862                            u8 il4_proto, u32 *type_cs_vlan_tso,
863                            u32 *ol_type_vlan_len_msec)
864 {
865         unsigned char *l2_hdr = skb->data;
866         u32 l4_proto = ol4_proto;
867         union l4_hdr_info l4;
868         union l3_hdr_info l3;
869         u32 l2_len, l3_len;
870
871         l4.hdr = skb_transport_header(skb);
872         l3.hdr = skb_network_header(skb);
873
874         /* handle encapsulation skb */
875         if (skb->encapsulation) {
876                 /* If this is a not UDP/GRE encapsulation skb */
877                 if (!(ol4_proto == IPPROTO_UDP || ol4_proto == IPPROTO_GRE)) {
878                         /* drop the skb tunnel packet if hardware don't support,
879                          * because hardware can't calculate csum when TSO.
880                          */
881                         if (skb_is_gso(skb))
882                                 return -EDOM;
883
884                         /* the stack computes the IP header already,
885                          * driver calculate l4 checksum when not TSO.
886                          */
887                         skb_checksum_help(skb);
888                         return 0;
889                 }
890
891                 hns3_set_outer_l2l3l4(skb, ol4_proto, ol_type_vlan_len_msec);
892
893                 /* switch to inner header */
894                 l2_hdr = skb_inner_mac_header(skb);
895                 l3.hdr = skb_inner_network_header(skb);
896                 l4.hdr = skb_inner_transport_header(skb);
897                 l4_proto = il4_proto;
898         }
899
900         if (l3.v4->version == 4) {
901                 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
902                                HNS3_L3T_IPV4);
903
904                 /* the stack computes the IP header already, the only time we
905                  * need the hardware to recompute it is in the case of TSO.
906                  */
907                 if (skb_is_gso(skb))
908                         hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3CS_B, 1);
909         } else if (l3.v6->version == 6) {
910                 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
911                                HNS3_L3T_IPV6);
912         }
913
914         /* compute inner(/normal) L2 header size, defined in 2 Bytes */
915         l2_len = l3.hdr - l2_hdr;
916         hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S, l2_len >> 1);
917
918         /* compute inner(/normal) L3 header size, defined in 4 Bytes */
919         l3_len = l4.hdr - l3.hdr;
920         hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_S, l3_len >> 2);
921
922         /* compute inner(/normal) L4 header size, defined in 4 Bytes */
923         switch (l4_proto) {
924         case IPPROTO_TCP:
925                 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
926                 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
927                                HNS3_L4T_TCP);
928                 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
929                                l4.tcp->doff);
930                 break;
931         case IPPROTO_UDP:
932                 if (hns3_tunnel_csum_bug(skb))
933                         break;
934
935                 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
936                 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
937                                HNS3_L4T_UDP);
938                 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
939                                (sizeof(struct udphdr) >> 2));
940                 break;
941         case IPPROTO_SCTP:
942                 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
943                 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
944                                HNS3_L4T_SCTP);
945                 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
946                                (sizeof(struct sctphdr) >> 2));
947                 break;
948         default:
949                 /* drop the skb tunnel packet if hardware don't support,
950                  * because hardware can't calculate csum when TSO.
951                  */
952                 if (skb_is_gso(skb))
953                         return -EDOM;
954
955                 /* the stack computes the IP header already,
956                  * driver calculate l4 checksum when not TSO.
957                  */
958                 skb_checksum_help(skb);
959                 return 0;
960         }
961
962         return 0;
963 }
964
965 static int hns3_handle_vtags(struct hns3_enet_ring *tx_ring,
966                              struct sk_buff *skb)
967 {
968         struct hnae3_handle *handle = tx_ring->tqp->handle;
969         struct vlan_ethhdr *vhdr;
970         int rc;
971
972         if (!(skb->protocol == htons(ETH_P_8021Q) ||
973               skb_vlan_tag_present(skb)))
974                 return 0;
975
976         /* Since HW limitation, if port based insert VLAN enabled, only one VLAN
977          * header is allowed in skb, otherwise it will cause RAS error.
978          */
979         if (unlikely(skb_vlan_tagged_multi(skb) &&
980                      handle->port_base_vlan_state ==
981                      HNAE3_PORT_BASE_VLAN_ENABLE))
982                 return -EINVAL;
983
984         if (skb->protocol == htons(ETH_P_8021Q) &&
985             !(handle->kinfo.netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
986                 /* When HW VLAN acceleration is turned off, and the stack
987                  * sets the protocol to 802.1q, the driver just need to
988                  * set the protocol to the encapsulated ethertype.
989                  */
990                 skb->protocol = vlan_get_protocol(skb);
991                 return 0;
992         }
993
994         if (skb_vlan_tag_present(skb)) {
995                 /* Based on hw strategy, use out_vtag in two layer tag case,
996                  * and use inner_vtag in one tag case.
997                  */
998                 if (skb->protocol == htons(ETH_P_8021Q) &&
999                     handle->port_base_vlan_state ==
1000                     HNAE3_PORT_BASE_VLAN_DISABLE)
1001                         rc = HNS3_OUTER_VLAN_TAG;
1002                 else
1003                         rc = HNS3_INNER_VLAN_TAG;
1004
1005                 skb->protocol = vlan_get_protocol(skb);
1006                 return rc;
1007         }
1008
1009         rc = skb_cow_head(skb, 0);
1010         if (unlikely(rc < 0))
1011                 return rc;
1012
1013         vhdr = (struct vlan_ethhdr *)skb->data;
1014         vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority << VLAN_PRIO_SHIFT)
1015                                          & VLAN_PRIO_MASK);
1016
1017         skb->protocol = vlan_get_protocol(skb);
1018         return 0;
1019 }
1020
1021 static int hns3_fill_skb_desc(struct hns3_enet_ring *ring,
1022                               struct sk_buff *skb, struct hns3_desc *desc)
1023 {
1024         u32 ol_type_vlan_len_msec = 0;
1025         u32 type_cs_vlan_tso = 0;
1026         u32 paylen = skb->len;
1027         u16 inner_vtag = 0;
1028         u16 out_vtag = 0;
1029         u16 mss = 0;
1030         int ret;
1031
1032         ret = hns3_handle_vtags(ring, skb);
1033         if (unlikely(ret < 0)) {
1034                 u64_stats_update_begin(&ring->syncp);
1035                 ring->stats.tx_vlan_err++;
1036                 u64_stats_update_end(&ring->syncp);
1037                 return ret;
1038         } else if (ret == HNS3_INNER_VLAN_TAG) {
1039                 inner_vtag = skb_vlan_tag_get(skb);
1040                 inner_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1041                                 VLAN_PRIO_MASK;
1042                 hns3_set_field(type_cs_vlan_tso, HNS3_TXD_VLAN_B, 1);
1043         } else if (ret == HNS3_OUTER_VLAN_TAG) {
1044                 out_vtag = skb_vlan_tag_get(skb);
1045                 out_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1046                                 VLAN_PRIO_MASK;
1047                 hns3_set_field(ol_type_vlan_len_msec, HNS3_TXD_OVLAN_B,
1048                                1);
1049         }
1050
1051         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1052                 u8 ol4_proto, il4_proto;
1053
1054                 skb_reset_mac_len(skb);
1055
1056                 ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
1057                 if (unlikely(ret < 0)) {
1058                         u64_stats_update_begin(&ring->syncp);
1059                         ring->stats.tx_l4_proto_err++;
1060                         u64_stats_update_end(&ring->syncp);
1061                         return ret;
1062                 }
1063
1064                 ret = hns3_set_l2l3l4(skb, ol4_proto, il4_proto,
1065                                       &type_cs_vlan_tso,
1066                                       &ol_type_vlan_len_msec);
1067                 if (unlikely(ret < 0)) {
1068                         u64_stats_update_begin(&ring->syncp);
1069                         ring->stats.tx_l2l3l4_err++;
1070                         u64_stats_update_end(&ring->syncp);
1071                         return ret;
1072                 }
1073
1074                 ret = hns3_set_tso(skb, &paylen, &mss,
1075                                    &type_cs_vlan_tso);
1076                 if (unlikely(ret < 0)) {
1077                         u64_stats_update_begin(&ring->syncp);
1078                         ring->stats.tx_tso_err++;
1079                         u64_stats_update_end(&ring->syncp);
1080                         return ret;
1081                 }
1082         }
1083
1084         /* Set txbd */
1085         desc->tx.ol_type_vlan_len_msec =
1086                 cpu_to_le32(ol_type_vlan_len_msec);
1087         desc->tx.type_cs_vlan_tso_len = cpu_to_le32(type_cs_vlan_tso);
1088         desc->tx.paylen = cpu_to_le32(paylen);
1089         desc->tx.mss = cpu_to_le16(mss);
1090         desc->tx.vlan_tag = cpu_to_le16(inner_vtag);
1091         desc->tx.outer_vlan_tag = cpu_to_le16(out_vtag);
1092
1093         return 0;
1094 }
1095
1096 static int hns3_fill_desc(struct hns3_enet_ring *ring, void *priv,
1097                           unsigned int size, enum hns_desc_type type)
1098 {
1099 #define HNS3_LIKELY_BD_NUM      1
1100
1101         struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1102         struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1103         struct device *dev = ring_to_dev(ring);
1104         skb_frag_t *frag;
1105         unsigned int frag_buf_num;
1106         int k, sizeoflast;
1107         dma_addr_t dma;
1108
1109         if (type == DESC_TYPE_SKB) {
1110                 struct sk_buff *skb = (struct sk_buff *)priv;
1111                 int ret;
1112
1113                 ret = hns3_fill_skb_desc(ring, skb, desc);
1114                 if (unlikely(ret < 0))
1115                         return ret;
1116
1117                 dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1118         } else {
1119                 frag = (skb_frag_t *)priv;
1120                 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1121         }
1122
1123         if (unlikely(dma_mapping_error(dev, dma))) {
1124                 u64_stats_update_begin(&ring->syncp);
1125                 ring->stats.sw_err_cnt++;
1126                 u64_stats_update_end(&ring->syncp);
1127                 return -ENOMEM;
1128         }
1129
1130         desc_cb->length = size;
1131
1132         if (likely(size <= HNS3_MAX_BD_SIZE)) {
1133                 desc_cb->priv = priv;
1134                 desc_cb->dma = dma;
1135                 desc_cb->type = type;
1136                 desc->addr = cpu_to_le64(dma);
1137                 desc->tx.send_size = cpu_to_le16(size);
1138                 desc->tx.bdtp_fe_sc_vld_ra_ri =
1139                         cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1140
1141                 ring_ptr_move_fw(ring, next_to_use);
1142                 return HNS3_LIKELY_BD_NUM;
1143         }
1144
1145         frag_buf_num = hns3_tx_bd_count(size);
1146         sizeoflast = size & HNS3_TX_LAST_SIZE_M;
1147         sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1148
1149         /* When frag size is bigger than hardware limit, split this frag */
1150         for (k = 0; k < frag_buf_num; k++) {
1151                 /* The txbd's baseinfo of DESC_TYPE_PAGE & DESC_TYPE_SKB */
1152                 desc_cb->priv = priv;
1153                 desc_cb->dma = dma + HNS3_MAX_BD_SIZE * k;
1154                 desc_cb->type = (type == DESC_TYPE_SKB && !k) ?
1155                                 DESC_TYPE_SKB : DESC_TYPE_PAGE;
1156
1157                 /* now, fill the descriptor */
1158                 desc->addr = cpu_to_le64(dma + HNS3_MAX_BD_SIZE * k);
1159                 desc->tx.send_size = cpu_to_le16((k == frag_buf_num - 1) ?
1160                                      (u16)sizeoflast : (u16)HNS3_MAX_BD_SIZE);
1161                 desc->tx.bdtp_fe_sc_vld_ra_ri =
1162                                 cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1163
1164                 /* move ring pointer to next */
1165                 ring_ptr_move_fw(ring, next_to_use);
1166
1167                 desc_cb = &ring->desc_cb[ring->next_to_use];
1168                 desc = &ring->desc[ring->next_to_use];
1169         }
1170
1171         return frag_buf_num;
1172 }
1173
1174 static unsigned int hns3_skb_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1175                                     unsigned int bd_num)
1176 {
1177         unsigned int size;
1178         int i;
1179
1180         size = skb_headlen(skb);
1181         while (size > HNS3_MAX_BD_SIZE) {
1182                 bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1183                 size -= HNS3_MAX_BD_SIZE;
1184
1185                 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1186                         return bd_num;
1187         }
1188
1189         if (size) {
1190                 bd_size[bd_num++] = size;
1191                 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1192                         return bd_num;
1193         }
1194
1195         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1196                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1197                 size = skb_frag_size(frag);
1198                 if (!size)
1199                         continue;
1200
1201                 while (size > HNS3_MAX_BD_SIZE) {
1202                         bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1203                         size -= HNS3_MAX_BD_SIZE;
1204
1205                         if (bd_num > HNS3_MAX_TSO_BD_NUM)
1206                                 return bd_num;
1207                 }
1208
1209                 bd_size[bd_num++] = size;
1210                 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1211                         return bd_num;
1212         }
1213
1214         return bd_num;
1215 }
1216
1217 static unsigned int hns3_tx_bd_num(struct sk_buff *skb, unsigned int *bd_size)
1218 {
1219         struct sk_buff *frag_skb;
1220         unsigned int bd_num = 0;
1221
1222         /* If the total len is within the max bd limit */
1223         if (likely(skb->len <= HNS3_MAX_BD_SIZE && !skb_has_frag_list(skb) &&
1224                    skb_shinfo(skb)->nr_frags < HNS3_MAX_NON_TSO_BD_NUM))
1225                 return skb_shinfo(skb)->nr_frags + 1U;
1226
1227         /* The below case will always be linearized, return
1228          * HNS3_MAX_BD_NUM_TSO + 1U to make sure it is linearized.
1229          */
1230         if (unlikely(skb->len > HNS3_MAX_TSO_SIZE ||
1231                      (!skb_is_gso(skb) && skb->len > HNS3_MAX_NON_TSO_SIZE)))
1232                 return HNS3_MAX_TSO_BD_NUM + 1U;
1233
1234         bd_num = hns3_skb_bd_num(skb, bd_size, bd_num);
1235
1236         if (!skb_has_frag_list(skb) || bd_num > HNS3_MAX_TSO_BD_NUM)
1237                 return bd_num;
1238
1239         skb_walk_frags(skb, frag_skb) {
1240                 bd_num = hns3_skb_bd_num(frag_skb, bd_size, bd_num);
1241                 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1242                         return bd_num;
1243         }
1244
1245         return bd_num;
1246 }
1247
1248 static unsigned int hns3_gso_hdr_len(struct sk_buff *skb)
1249 {
1250         if (!skb->encapsulation)
1251                 return skb_transport_offset(skb) + tcp_hdrlen(skb);
1252
1253         return skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb);
1254 }
1255
1256 /* HW need every continuous 8 buffer data to be larger than MSS,
1257  * we simplify it by ensuring skb_headlen + the first continuous
1258  * 7 frags to to be larger than gso header len + mss, and the remaining
1259  * continuous 7 frags to be larger than MSS except the last 7 frags.
1260  */
1261 static bool hns3_skb_need_linearized(struct sk_buff *skb, unsigned int *bd_size,
1262                                      unsigned int bd_num)
1263 {
1264         unsigned int tot_len = 0;
1265         int i;
1266
1267         for (i = 0; i < HNS3_MAX_NON_TSO_BD_NUM - 1U; i++)
1268                 tot_len += bd_size[i];
1269
1270         /* ensure the first 8 frags is greater than mss + header */
1271         if (tot_len + bd_size[HNS3_MAX_NON_TSO_BD_NUM - 1U] <
1272             skb_shinfo(skb)->gso_size + hns3_gso_hdr_len(skb))
1273                 return true;
1274
1275         /* ensure every continuous 7 buffer is greater than mss
1276          * except the last one.
1277          */
1278         for (i = 0; i < bd_num - HNS3_MAX_NON_TSO_BD_NUM; i++) {
1279                 tot_len -= bd_size[i];
1280                 tot_len += bd_size[i + HNS3_MAX_NON_TSO_BD_NUM - 1U];
1281
1282                 if (tot_len < skb_shinfo(skb)->gso_size)
1283                         return true;
1284         }
1285
1286         return false;
1287 }
1288
1289 static int hns3_nic_maybe_stop_tx(struct hns3_enet_ring *ring,
1290                                   struct net_device *netdev,
1291                                   struct sk_buff *skb)
1292 {
1293         struct hns3_nic_priv *priv = netdev_priv(netdev);
1294         unsigned int bd_size[HNS3_MAX_TSO_BD_NUM + 1U];
1295         unsigned int bd_num;
1296
1297         bd_num = hns3_tx_bd_num(skb, bd_size);
1298         if (unlikely(bd_num > HNS3_MAX_NON_TSO_BD_NUM)) {
1299                 if (bd_num <= HNS3_MAX_TSO_BD_NUM && skb_is_gso(skb) &&
1300                     !hns3_skb_need_linearized(skb, bd_size, bd_num))
1301                         goto out;
1302
1303                 if (__skb_linearize(skb))
1304                         return -ENOMEM;
1305
1306                 bd_num = hns3_tx_bd_count(skb->len);
1307                 if ((skb_is_gso(skb) && bd_num > HNS3_MAX_TSO_BD_NUM) ||
1308                     (!skb_is_gso(skb) &&
1309                      bd_num > HNS3_MAX_NON_TSO_BD_NUM))
1310                         return -ENOMEM;
1311
1312                 u64_stats_update_begin(&ring->syncp);
1313                 ring->stats.tx_copy++;
1314                 u64_stats_update_end(&ring->syncp);
1315         }
1316
1317 out:
1318         if (likely(ring_space(ring) >= bd_num))
1319                 return bd_num;
1320
1321         netif_stop_subqueue(netdev, ring->queue_index);
1322         smp_mb(); /* Memory barrier before checking ring_space */
1323
1324         /* Start queue in case hns3_clean_tx_ring has just made room
1325          * available and has not seen the queue stopped state performed
1326          * by netif_stop_subqueue above.
1327          */
1328         if (ring_space(ring) >= bd_num && netif_carrier_ok(netdev) &&
1329             !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
1330                 netif_start_subqueue(netdev, ring->queue_index);
1331                 return bd_num;
1332         }
1333
1334         return -EBUSY;
1335 }
1336
1337 static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
1338 {
1339         struct device *dev = ring_to_dev(ring);
1340         unsigned int i;
1341
1342         for (i = 0; i < ring->desc_num; i++) {
1343                 /* check if this is where we started */
1344                 if (ring->next_to_use == next_to_use_orig)
1345                         break;
1346
1347                 /* rollback one */
1348                 ring_ptr_move_bw(ring, next_to_use);
1349
1350                 /* unmap the descriptor dma address */
1351                 if (ring->desc_cb[ring->next_to_use].type == DESC_TYPE_SKB)
1352                         dma_unmap_single(dev,
1353                                          ring->desc_cb[ring->next_to_use].dma,
1354                                         ring->desc_cb[ring->next_to_use].length,
1355                                         DMA_TO_DEVICE);
1356                 else if (ring->desc_cb[ring->next_to_use].length)
1357                         dma_unmap_page(dev,
1358                                        ring->desc_cb[ring->next_to_use].dma,
1359                                        ring->desc_cb[ring->next_to_use].length,
1360                                        DMA_TO_DEVICE);
1361
1362                 ring->desc_cb[ring->next_to_use].length = 0;
1363                 ring->desc_cb[ring->next_to_use].dma = 0;
1364         }
1365 }
1366
1367 static int hns3_fill_skb_to_desc(struct hns3_enet_ring *ring,
1368                                  struct sk_buff *skb, enum hns_desc_type type)
1369 {
1370         unsigned int size = skb_headlen(skb);
1371         int i, ret, bd_num = 0;
1372
1373         if (size) {
1374                 ret = hns3_fill_desc(ring, skb, size, type);
1375                 if (unlikely(ret < 0))
1376                         return ret;
1377
1378                 bd_num += ret;
1379         }
1380
1381         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1382                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1383
1384                 size = skb_frag_size(frag);
1385                 if (!size)
1386                         continue;
1387
1388                 ret = hns3_fill_desc(ring, frag, size, DESC_TYPE_PAGE);
1389                 if (unlikely(ret < 0))
1390                         return ret;
1391
1392                 bd_num += ret;
1393         }
1394
1395         return bd_num;
1396 }
1397
1398 netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
1399 {
1400         struct hns3_nic_priv *priv = netdev_priv(netdev);
1401         struct hns3_enet_ring *ring = &priv->ring[skb->queue_mapping];
1402         struct netdev_queue *dev_queue;
1403         int pre_ntu, next_to_use_head;
1404         struct sk_buff *frag_skb;
1405         int bd_num = 0;
1406         int ret;
1407
1408         /* Prefetch the data used later */
1409         prefetch(skb->data);
1410
1411         ret = hns3_nic_maybe_stop_tx(ring, netdev, skb);
1412         if (unlikely(ret <= 0)) {
1413                 if (ret == -EBUSY) {
1414                         u64_stats_update_begin(&ring->syncp);
1415                         ring->stats.tx_busy++;
1416                         u64_stats_update_end(&ring->syncp);
1417                         return NETDEV_TX_BUSY;
1418                 } else if (ret == -ENOMEM) {
1419                         u64_stats_update_begin(&ring->syncp);
1420                         ring->stats.sw_err_cnt++;
1421                         u64_stats_update_end(&ring->syncp);
1422                 }
1423
1424                 hns3_rl_err(netdev, "xmit error: %d!\n", ret);
1425                 goto out_err_tx_ok;
1426         }
1427
1428         next_to_use_head = ring->next_to_use;
1429
1430         ret = hns3_fill_skb_to_desc(ring, skb, DESC_TYPE_SKB);
1431         if (unlikely(ret < 0))
1432                 goto fill_err;
1433
1434         bd_num += ret;
1435
1436         if (!skb_has_frag_list(skb))
1437                 goto out;
1438
1439         skb_walk_frags(skb, frag_skb) {
1440                 ret = hns3_fill_skb_to_desc(ring, frag_skb, DESC_TYPE_PAGE);
1441                 if (unlikely(ret < 0))
1442                         goto fill_err;
1443
1444                 bd_num += ret;
1445         }
1446 out:
1447         pre_ntu = ring->next_to_use ? (ring->next_to_use - 1) :
1448                                         (ring->desc_num - 1);
1449         ring->desc[pre_ntu].tx.bdtp_fe_sc_vld_ra_ri |=
1450                                 cpu_to_le16(BIT(HNS3_TXD_FE_B));
1451
1452         /* Complete translate all packets */
1453         dev_queue = netdev_get_tx_queue(netdev, ring->queue_index);
1454         netdev_tx_sent_queue(dev_queue, skb->len);
1455
1456         wmb(); /* Commit all data before submit */
1457
1458         hnae3_queue_xmit(ring->tqp, bd_num);
1459
1460         return NETDEV_TX_OK;
1461
1462 fill_err:
1463         hns3_clear_desc(ring, next_to_use_head);
1464
1465 out_err_tx_ok:
1466         dev_kfree_skb_any(skb);
1467         return NETDEV_TX_OK;
1468 }
1469
1470 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
1471 {
1472         struct hnae3_handle *h = hns3_get_handle(netdev);
1473         struct sockaddr *mac_addr = p;
1474         int ret;
1475
1476         if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1477                 return -EADDRNOTAVAIL;
1478
1479         if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
1480                 netdev_info(netdev, "already using mac address %pM\n",
1481                             mac_addr->sa_data);
1482                 return 0;
1483         }
1484
1485         /* For VF device, if there is a perm_addr, then the user will not
1486          * be allowed to change the address.
1487          */
1488         if (!hns3_is_phys_func(h->pdev) &&
1489             !is_zero_ether_addr(netdev->perm_addr)) {
1490                 netdev_err(netdev, "has permanent MAC %pM, user MAC %pM not allow\n",
1491                            netdev->perm_addr, mac_addr->sa_data);
1492                 return -EPERM;
1493         }
1494
1495         ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
1496         if (ret) {
1497                 netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
1498                 return ret;
1499         }
1500
1501         ether_addr_copy(netdev->dev_addr, mac_addr->sa_data);
1502
1503         return 0;
1504 }
1505
1506 static int hns3_nic_do_ioctl(struct net_device *netdev,
1507                              struct ifreq *ifr, int cmd)
1508 {
1509         struct hnae3_handle *h = hns3_get_handle(netdev);
1510
1511         if (!netif_running(netdev))
1512                 return -EINVAL;
1513
1514         if (!h->ae_algo->ops->do_ioctl)
1515                 return -EOPNOTSUPP;
1516
1517         return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
1518 }
1519
1520 static int hns3_nic_set_features(struct net_device *netdev,
1521                                  netdev_features_t features)
1522 {
1523         netdev_features_t changed = netdev->features ^ features;
1524         struct hns3_nic_priv *priv = netdev_priv(netdev);
1525         struct hnae3_handle *h = priv->ae_handle;
1526         bool enable;
1527         int ret;
1528
1529         if (changed & (NETIF_F_GRO_HW) && h->ae_algo->ops->set_gro_en) {
1530                 enable = !!(features & NETIF_F_GRO_HW);
1531                 ret = h->ae_algo->ops->set_gro_en(h, enable);
1532                 if (ret)
1533                         return ret;
1534         }
1535
1536         if ((changed & NETIF_F_HW_VLAN_CTAG_FILTER) &&
1537             h->ae_algo->ops->enable_vlan_filter) {
1538                 enable = !!(features & NETIF_F_HW_VLAN_CTAG_FILTER);
1539                 h->ae_algo->ops->enable_vlan_filter(h, enable);
1540         }
1541
1542         if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
1543             h->ae_algo->ops->enable_hw_strip_rxvtag) {
1544                 enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
1545                 ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, enable);
1546                 if (ret)
1547                         return ret;
1548         }
1549
1550         if ((changed & NETIF_F_NTUPLE) && h->ae_algo->ops->enable_fd) {
1551                 enable = !!(features & NETIF_F_NTUPLE);
1552                 h->ae_algo->ops->enable_fd(h, enable);
1553         }
1554
1555         netdev->features = features;
1556         return 0;
1557 }
1558
1559 static void hns3_nic_get_stats64(struct net_device *netdev,
1560                                  struct rtnl_link_stats64 *stats)
1561 {
1562         struct hns3_nic_priv *priv = netdev_priv(netdev);
1563         int queue_num = priv->ae_handle->kinfo.num_tqps;
1564         struct hnae3_handle *handle = priv->ae_handle;
1565         struct hns3_enet_ring *ring;
1566         u64 rx_length_errors = 0;
1567         u64 rx_crc_errors = 0;
1568         u64 rx_multicast = 0;
1569         unsigned int start;
1570         u64 tx_errors = 0;
1571         u64 rx_errors = 0;
1572         unsigned int idx;
1573         u64 tx_bytes = 0;
1574         u64 rx_bytes = 0;
1575         u64 tx_pkts = 0;
1576         u64 rx_pkts = 0;
1577         u64 tx_drop = 0;
1578         u64 rx_drop = 0;
1579
1580         if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
1581                 return;
1582
1583         handle->ae_algo->ops->update_stats(handle, &netdev->stats);
1584
1585         for (idx = 0; idx < queue_num; idx++) {
1586                 /* fetch the tx stats */
1587                 ring = &priv->ring[idx];
1588                 do {
1589                         start = u64_stats_fetch_begin_irq(&ring->syncp);
1590                         tx_bytes += ring->stats.tx_bytes;
1591                         tx_pkts += ring->stats.tx_pkts;
1592                         tx_drop += ring->stats.sw_err_cnt;
1593                         tx_drop += ring->stats.tx_vlan_err;
1594                         tx_drop += ring->stats.tx_l4_proto_err;
1595                         tx_drop += ring->stats.tx_l2l3l4_err;
1596                         tx_drop += ring->stats.tx_tso_err;
1597                         tx_errors += ring->stats.sw_err_cnt;
1598                         tx_errors += ring->stats.tx_vlan_err;
1599                         tx_errors += ring->stats.tx_l4_proto_err;
1600                         tx_errors += ring->stats.tx_l2l3l4_err;
1601                         tx_errors += ring->stats.tx_tso_err;
1602                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1603
1604                 /* fetch the rx stats */
1605                 ring = &priv->ring[idx + queue_num];
1606                 do {
1607                         start = u64_stats_fetch_begin_irq(&ring->syncp);
1608                         rx_bytes += ring->stats.rx_bytes;
1609                         rx_pkts += ring->stats.rx_pkts;
1610                         rx_drop += ring->stats.l2_err;
1611                         rx_errors += ring->stats.l2_err;
1612                         rx_errors += ring->stats.l3l4_csum_err;
1613                         rx_crc_errors += ring->stats.l2_err;
1614                         rx_multicast += ring->stats.rx_multicast;
1615                         rx_length_errors += ring->stats.err_pkt_len;
1616                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1617         }
1618
1619         stats->tx_bytes = tx_bytes;
1620         stats->tx_packets = tx_pkts;
1621         stats->rx_bytes = rx_bytes;
1622         stats->rx_packets = rx_pkts;
1623
1624         stats->rx_errors = rx_errors;
1625         stats->multicast = rx_multicast;
1626         stats->rx_length_errors = rx_length_errors;
1627         stats->rx_crc_errors = rx_crc_errors;
1628         stats->rx_missed_errors = netdev->stats.rx_missed_errors;
1629
1630         stats->tx_errors = tx_errors;
1631         stats->rx_dropped = rx_drop;
1632         stats->tx_dropped = tx_drop;
1633         stats->collisions = netdev->stats.collisions;
1634         stats->rx_over_errors = netdev->stats.rx_over_errors;
1635         stats->rx_frame_errors = netdev->stats.rx_frame_errors;
1636         stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
1637         stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
1638         stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
1639         stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
1640         stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
1641         stats->tx_window_errors = netdev->stats.tx_window_errors;
1642         stats->rx_compressed = netdev->stats.rx_compressed;
1643         stats->tx_compressed = netdev->stats.tx_compressed;
1644 }
1645
1646 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
1647 {
1648         struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
1649         u8 *prio_tc = mqprio_qopt->qopt.prio_tc_map;
1650         struct hnae3_knic_private_info *kinfo;
1651         u8 tc = mqprio_qopt->qopt.num_tc;
1652         u16 mode = mqprio_qopt->mode;
1653         u8 hw = mqprio_qopt->qopt.hw;
1654         struct hnae3_handle *h;
1655
1656         if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
1657                mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
1658                 return -EOPNOTSUPP;
1659
1660         if (tc > HNAE3_MAX_TC)
1661                 return -EINVAL;
1662
1663         if (!netdev)
1664                 return -EINVAL;
1665
1666         h = hns3_get_handle(netdev);
1667         kinfo = &h->kinfo;
1668
1669         netif_dbg(h, drv, netdev, "setup tc: num_tc=%u\n", tc);
1670
1671         return (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
1672                 kinfo->dcb_ops->setup_tc(h, tc, prio_tc) : -EOPNOTSUPP;
1673 }
1674
1675 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
1676                              void *type_data)
1677 {
1678         if (type != TC_SETUP_QDISC_MQPRIO)
1679                 return -EOPNOTSUPP;
1680
1681         return hns3_setup_tc(dev, type_data);
1682 }
1683
1684 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
1685                                 __be16 proto, u16 vid)
1686 {
1687         struct hnae3_handle *h = hns3_get_handle(netdev);
1688         int ret = -EIO;
1689
1690         if (h->ae_algo->ops->set_vlan_filter)
1691                 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
1692
1693         return ret;
1694 }
1695
1696 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
1697                                  __be16 proto, u16 vid)
1698 {
1699         struct hnae3_handle *h = hns3_get_handle(netdev);
1700         int ret = -EIO;
1701
1702         if (h->ae_algo->ops->set_vlan_filter)
1703                 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
1704
1705         return ret;
1706 }
1707
1708 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
1709                                 u8 qos, __be16 vlan_proto)
1710 {
1711         struct hnae3_handle *h = hns3_get_handle(netdev);
1712         int ret = -EIO;
1713
1714         netif_dbg(h, drv, netdev,
1715                   "set vf vlan: vf=%d, vlan=%u, qos=%u, vlan_proto=0x%x\n",
1716                   vf, vlan, qos, ntohs(vlan_proto));
1717
1718         if (h->ae_algo->ops->set_vf_vlan_filter)
1719                 ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
1720                                                           qos, vlan_proto);
1721
1722         return ret;
1723 }
1724
1725 static int hns3_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
1726 {
1727         struct hnae3_handle *handle = hns3_get_handle(netdev);
1728
1729         if (hns3_nic_resetting(netdev))
1730                 return -EBUSY;
1731
1732         if (!handle->ae_algo->ops->set_vf_spoofchk)
1733                 return -EOPNOTSUPP;
1734
1735         return handle->ae_algo->ops->set_vf_spoofchk(handle, vf, enable);
1736 }
1737
1738 static int hns3_set_vf_trust(struct net_device *netdev, int vf, bool enable)
1739 {
1740         struct hnae3_handle *handle = hns3_get_handle(netdev);
1741
1742         if (!handle->ae_algo->ops->set_vf_trust)
1743                 return -EOPNOTSUPP;
1744
1745         return handle->ae_algo->ops->set_vf_trust(handle, vf, enable);
1746 }
1747
1748 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
1749 {
1750         struct hnae3_handle *h = hns3_get_handle(netdev);
1751         int ret;
1752
1753         if (hns3_nic_resetting(netdev))
1754                 return -EBUSY;
1755
1756         if (!h->ae_algo->ops->set_mtu)
1757                 return -EOPNOTSUPP;
1758
1759         netif_dbg(h, drv, netdev,
1760                   "change mtu from %u to %d\n", netdev->mtu, new_mtu);
1761
1762         ret = h->ae_algo->ops->set_mtu(h, new_mtu);
1763         if (ret)
1764                 netdev_err(netdev, "failed to change MTU in hardware %d\n",
1765                            ret);
1766         else
1767                 netdev->mtu = new_mtu;
1768
1769         return ret;
1770 }
1771
1772 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
1773 {
1774         struct hns3_nic_priv *priv = netdev_priv(ndev);
1775         struct hnae3_handle *h = hns3_get_handle(ndev);
1776         struct hns3_enet_ring *tx_ring;
1777         struct napi_struct *napi;
1778         int timeout_queue = 0;
1779         int hw_head, hw_tail;
1780         int fbd_num, fbd_oft;
1781         int ebd_num, ebd_oft;
1782         int bd_num, bd_err;
1783         int ring_en, tc;
1784         int i;
1785
1786         /* Find the stopped queue the same way the stack does */
1787         for (i = 0; i < ndev->num_tx_queues; i++) {
1788                 struct netdev_queue *q;
1789                 unsigned long trans_start;
1790
1791                 q = netdev_get_tx_queue(ndev, i);
1792                 trans_start = q->trans_start;
1793                 if (netif_xmit_stopped(q) &&
1794                     time_after(jiffies,
1795                                (trans_start + ndev->watchdog_timeo))) {
1796                         timeout_queue = i;
1797                         netdev_info(ndev, "queue state: 0x%lx, delta msecs: %u\n",
1798                                     q->state,
1799                                     jiffies_to_msecs(jiffies - trans_start));
1800                         break;
1801                 }
1802         }
1803
1804         if (i == ndev->num_tx_queues) {
1805                 netdev_info(ndev,
1806                             "no netdev TX timeout queue found, timeout count: %llu\n",
1807                             priv->tx_timeout_count);
1808                 return false;
1809         }
1810
1811         priv->tx_timeout_count++;
1812
1813         tx_ring = &priv->ring[timeout_queue];
1814         napi = &tx_ring->tqp_vector->napi;
1815
1816         netdev_info(ndev,
1817                     "tx_timeout count: %llu, queue id: %d, SW_NTU: 0x%x, SW_NTC: 0x%x, napi state: %lu\n",
1818                     priv->tx_timeout_count, timeout_queue, tx_ring->next_to_use,
1819                     tx_ring->next_to_clean, napi->state);
1820
1821         netdev_info(ndev,
1822                     "tx_pkts: %llu, tx_bytes: %llu, io_err_cnt: %llu, sw_err_cnt: %llu\n",
1823                     tx_ring->stats.tx_pkts, tx_ring->stats.tx_bytes,
1824                     tx_ring->stats.io_err_cnt, tx_ring->stats.sw_err_cnt);
1825
1826         netdev_info(ndev,
1827                     "seg_pkt_cnt: %llu, tx_err_cnt: %llu, restart_queue: %llu, tx_busy: %llu\n",
1828                     tx_ring->stats.seg_pkt_cnt, tx_ring->stats.tx_err_cnt,
1829                     tx_ring->stats.restart_queue, tx_ring->stats.tx_busy);
1830
1831         /* When mac received many pause frames continuous, it's unable to send
1832          * packets, which may cause tx timeout
1833          */
1834         if (h->ae_algo->ops->get_mac_stats) {
1835                 struct hns3_mac_stats mac_stats;
1836
1837                 h->ae_algo->ops->get_mac_stats(h, &mac_stats);
1838                 netdev_info(ndev, "tx_pause_cnt: %llu, rx_pause_cnt: %llu\n",
1839                             mac_stats.tx_pause_cnt, mac_stats.rx_pause_cnt);
1840         }
1841
1842         hw_head = readl_relaxed(tx_ring->tqp->io_base +
1843                                 HNS3_RING_TX_RING_HEAD_REG);
1844         hw_tail = readl_relaxed(tx_ring->tqp->io_base +
1845                                 HNS3_RING_TX_RING_TAIL_REG);
1846         fbd_num = readl_relaxed(tx_ring->tqp->io_base +
1847                                 HNS3_RING_TX_RING_FBDNUM_REG);
1848         fbd_oft = readl_relaxed(tx_ring->tqp->io_base +
1849                                 HNS3_RING_TX_RING_OFFSET_REG);
1850         ebd_num = readl_relaxed(tx_ring->tqp->io_base +
1851                                 HNS3_RING_TX_RING_EBDNUM_REG);
1852         ebd_oft = readl_relaxed(tx_ring->tqp->io_base +
1853                                 HNS3_RING_TX_RING_EBD_OFFSET_REG);
1854         bd_num = readl_relaxed(tx_ring->tqp->io_base +
1855                                HNS3_RING_TX_RING_BD_NUM_REG);
1856         bd_err = readl_relaxed(tx_ring->tqp->io_base +
1857                                HNS3_RING_TX_RING_BD_ERR_REG);
1858         ring_en = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_EN_REG);
1859         tc = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_TX_RING_TC_REG);
1860
1861         netdev_info(ndev,
1862                     "BD_NUM: 0x%x HW_HEAD: 0x%x, HW_TAIL: 0x%x, BD_ERR: 0x%x, INT: 0x%x\n",
1863                     bd_num, hw_head, hw_tail, bd_err,
1864                     readl(tx_ring->tqp_vector->mask_addr));
1865         netdev_info(ndev,
1866                     "RING_EN: 0x%x, TC: 0x%x, FBD_NUM: 0x%x FBD_OFT: 0x%x, EBD_NUM: 0x%x, EBD_OFT: 0x%x\n",
1867                     ring_en, tc, fbd_num, fbd_oft, ebd_num, ebd_oft);
1868
1869         return true;
1870 }
1871
1872 static void hns3_nic_net_timeout(struct net_device *ndev, unsigned int txqueue)
1873 {
1874         struct hns3_nic_priv *priv = netdev_priv(ndev);
1875         struct hnae3_handle *h = priv->ae_handle;
1876
1877         if (!hns3_get_tx_timeo_queue_info(ndev))
1878                 return;
1879
1880         /* request the reset, and let the hclge to determine
1881          * which reset level should be done
1882          */
1883         if (h->ae_algo->ops->reset_event)
1884                 h->ae_algo->ops->reset_event(h->pdev, h);
1885 }
1886
1887 #ifdef CONFIG_RFS_ACCEL
1888 static int hns3_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
1889                               u16 rxq_index, u32 flow_id)
1890 {
1891         struct hnae3_handle *h = hns3_get_handle(dev);
1892         struct flow_keys fkeys;
1893
1894         if (!h->ae_algo->ops->add_arfs_entry)
1895                 return -EOPNOTSUPP;
1896
1897         if (skb->encapsulation)
1898                 return -EPROTONOSUPPORT;
1899
1900         if (!skb_flow_dissect_flow_keys(skb, &fkeys, 0))
1901                 return -EPROTONOSUPPORT;
1902
1903         if ((fkeys.basic.n_proto != htons(ETH_P_IP) &&
1904              fkeys.basic.n_proto != htons(ETH_P_IPV6)) ||
1905             (fkeys.basic.ip_proto != IPPROTO_TCP &&
1906              fkeys.basic.ip_proto != IPPROTO_UDP))
1907                 return -EPROTONOSUPPORT;
1908
1909         return h->ae_algo->ops->add_arfs_entry(h, rxq_index, flow_id, &fkeys);
1910 }
1911 #endif
1912
1913 static int hns3_nic_get_vf_config(struct net_device *ndev, int vf,
1914                                   struct ifla_vf_info *ivf)
1915 {
1916         struct hnae3_handle *h = hns3_get_handle(ndev);
1917
1918         if (!h->ae_algo->ops->get_vf_config)
1919                 return -EOPNOTSUPP;
1920
1921         return h->ae_algo->ops->get_vf_config(h, vf, ivf);
1922 }
1923
1924 static int hns3_nic_set_vf_link_state(struct net_device *ndev, int vf,
1925                                       int link_state)
1926 {
1927         struct hnae3_handle *h = hns3_get_handle(ndev);
1928
1929         if (!h->ae_algo->ops->set_vf_link_state)
1930                 return -EOPNOTSUPP;
1931
1932         return h->ae_algo->ops->set_vf_link_state(h, vf, link_state);
1933 }
1934
1935 static int hns3_nic_set_vf_rate(struct net_device *ndev, int vf,
1936                                 int min_tx_rate, int max_tx_rate)
1937 {
1938         struct hnae3_handle *h = hns3_get_handle(ndev);
1939
1940         if (!h->ae_algo->ops->set_vf_rate)
1941                 return -EOPNOTSUPP;
1942
1943         return h->ae_algo->ops->set_vf_rate(h, vf, min_tx_rate, max_tx_rate,
1944                                             false);
1945 }
1946
1947 static int hns3_nic_set_vf_mac(struct net_device *netdev, int vf_id, u8 *mac)
1948 {
1949         struct hnae3_handle *h = hns3_get_handle(netdev);
1950
1951         if (!h->ae_algo->ops->set_vf_mac)
1952                 return -EOPNOTSUPP;
1953
1954         if (is_multicast_ether_addr(mac)) {
1955                 netdev_err(netdev,
1956                            "Invalid MAC:%pM specified. Could not set MAC\n",
1957                            mac);
1958                 return -EINVAL;
1959         }
1960
1961         return h->ae_algo->ops->set_vf_mac(h, vf_id, mac);
1962 }
1963
1964 static const struct net_device_ops hns3_nic_netdev_ops = {
1965         .ndo_open               = hns3_nic_net_open,
1966         .ndo_stop               = hns3_nic_net_stop,
1967         .ndo_start_xmit         = hns3_nic_net_xmit,
1968         .ndo_tx_timeout         = hns3_nic_net_timeout,
1969         .ndo_set_mac_address    = hns3_nic_net_set_mac_address,
1970         .ndo_do_ioctl           = hns3_nic_do_ioctl,
1971         .ndo_change_mtu         = hns3_nic_change_mtu,
1972         .ndo_set_features       = hns3_nic_set_features,
1973         .ndo_get_stats64        = hns3_nic_get_stats64,
1974         .ndo_setup_tc           = hns3_nic_setup_tc,
1975         .ndo_set_rx_mode        = hns3_nic_set_rx_mode,
1976         .ndo_vlan_rx_add_vid    = hns3_vlan_rx_add_vid,
1977         .ndo_vlan_rx_kill_vid   = hns3_vlan_rx_kill_vid,
1978         .ndo_set_vf_vlan        = hns3_ndo_set_vf_vlan,
1979         .ndo_set_vf_spoofchk    = hns3_set_vf_spoofchk,
1980         .ndo_set_vf_trust       = hns3_set_vf_trust,
1981 #ifdef CONFIG_RFS_ACCEL
1982         .ndo_rx_flow_steer      = hns3_rx_flow_steer,
1983 #endif
1984         .ndo_get_vf_config      = hns3_nic_get_vf_config,
1985         .ndo_set_vf_link_state  = hns3_nic_set_vf_link_state,
1986         .ndo_set_vf_rate        = hns3_nic_set_vf_rate,
1987         .ndo_set_vf_mac         = hns3_nic_set_vf_mac,
1988 };
1989
1990 bool hns3_is_phys_func(struct pci_dev *pdev)
1991 {
1992         u32 dev_id = pdev->device;
1993
1994         switch (dev_id) {
1995         case HNAE3_DEV_ID_GE:
1996         case HNAE3_DEV_ID_25GE:
1997         case HNAE3_DEV_ID_25GE_RDMA:
1998         case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
1999         case HNAE3_DEV_ID_50GE_RDMA:
2000         case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
2001         case HNAE3_DEV_ID_100G_RDMA_MACSEC:
2002                 return true;
2003         case HNAE3_DEV_ID_100G_VF:
2004         case HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF:
2005                 return false;
2006         default:
2007                 dev_warn(&pdev->dev, "un-recognized pci device-id %u",
2008                          dev_id);
2009         }
2010
2011         return false;
2012 }
2013
2014 static void hns3_disable_sriov(struct pci_dev *pdev)
2015 {
2016         /* If our VFs are assigned we cannot shut down SR-IOV
2017          * without causing issues, so just leave the hardware
2018          * available but disabled
2019          */
2020         if (pci_vfs_assigned(pdev)) {
2021                 dev_warn(&pdev->dev,
2022                          "disabling driver while VFs are assigned\n");
2023                 return;
2024         }
2025
2026         pci_disable_sriov(pdev);
2027 }
2028
2029 static void hns3_get_dev_capability(struct pci_dev *pdev,
2030                                     struct hnae3_ae_dev *ae_dev)
2031 {
2032         if (pdev->revision >= 0x21) {
2033                 hnae3_set_bit(ae_dev->flag, HNAE3_DEV_SUPPORT_FD_B, 1);
2034                 hnae3_set_bit(ae_dev->flag, HNAE3_DEV_SUPPORT_GRO_B, 1);
2035         }
2036 }
2037
2038 /* hns3_probe - Device initialization routine
2039  * @pdev: PCI device information struct
2040  * @ent: entry in hns3_pci_tbl
2041  *
2042  * hns3_probe initializes a PF identified by a pci_dev structure.
2043  * The OS initialization, configuring of the PF private structure,
2044  * and a hardware reset occur.
2045  *
2046  * Returns 0 on success, negative on failure
2047  */
2048 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2049 {
2050         struct hnae3_ae_dev *ae_dev;
2051         int ret;
2052
2053         ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev), GFP_KERNEL);
2054         if (!ae_dev) {
2055                 ret = -ENOMEM;
2056                 return ret;
2057         }
2058
2059         ae_dev->pdev = pdev;
2060         ae_dev->flag = ent->driver_data;
2061         ae_dev->reset_type = HNAE3_NONE_RESET;
2062         hns3_get_dev_capability(pdev, ae_dev);
2063         pci_set_drvdata(pdev, ae_dev);
2064
2065         ret = hnae3_register_ae_dev(ae_dev);
2066         if (ret) {
2067                 devm_kfree(&pdev->dev, ae_dev);
2068                 pci_set_drvdata(pdev, NULL);
2069         }
2070
2071         return ret;
2072 }
2073
2074 /* hns3_remove - Device removal routine
2075  * @pdev: PCI device information struct
2076  */
2077 static void hns3_remove(struct pci_dev *pdev)
2078 {
2079         struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2080
2081         if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
2082                 hns3_disable_sriov(pdev);
2083
2084         hnae3_unregister_ae_dev(ae_dev);
2085         pci_set_drvdata(pdev, NULL);
2086 }
2087
2088 /**
2089  * hns3_pci_sriov_configure
2090  * @pdev: pointer to a pci_dev structure
2091  * @num_vfs: number of VFs to allocate
2092  *
2093  * Enable or change the number of VFs. Called when the user updates the number
2094  * of VFs in sysfs.
2095  **/
2096 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
2097 {
2098         int ret;
2099
2100         if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
2101                 dev_warn(&pdev->dev, "Can not config SRIOV\n");
2102                 return -EINVAL;
2103         }
2104
2105         if (num_vfs) {
2106                 ret = pci_enable_sriov(pdev, num_vfs);
2107                 if (ret)
2108                         dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
2109                 else
2110                         return num_vfs;
2111         } else if (!pci_vfs_assigned(pdev)) {
2112                 pci_disable_sriov(pdev);
2113         } else {
2114                 dev_warn(&pdev->dev,
2115                          "Unable to free VFs because some are assigned to VMs.\n");
2116         }
2117
2118         return 0;
2119 }
2120
2121 static void hns3_shutdown(struct pci_dev *pdev)
2122 {
2123         struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2124
2125         hnae3_unregister_ae_dev(ae_dev);
2126         devm_kfree(&pdev->dev, ae_dev);
2127         pci_set_drvdata(pdev, NULL);
2128
2129         if (system_state == SYSTEM_POWER_OFF)
2130                 pci_set_power_state(pdev, PCI_D3hot);
2131 }
2132
2133 static pci_ers_result_t hns3_error_detected(struct pci_dev *pdev,
2134                                             pci_channel_state_t state)
2135 {
2136         struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2137         pci_ers_result_t ret;
2138
2139         dev_info(&pdev->dev, "PCI error detected, state(=%d)!!\n", state);
2140
2141         if (state == pci_channel_io_perm_failure)
2142                 return PCI_ERS_RESULT_DISCONNECT;
2143
2144         if (!ae_dev || !ae_dev->ops) {
2145                 dev_err(&pdev->dev,
2146                         "Can't recover - error happened before device initialized\n");
2147                 return PCI_ERS_RESULT_NONE;
2148         }
2149
2150         if (ae_dev->ops->handle_hw_ras_error)
2151                 ret = ae_dev->ops->handle_hw_ras_error(ae_dev);
2152         else
2153                 return PCI_ERS_RESULT_NONE;
2154
2155         return ret;
2156 }
2157
2158 static pci_ers_result_t hns3_slot_reset(struct pci_dev *pdev)
2159 {
2160         struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2161         const struct hnae3_ae_ops *ops;
2162         enum hnae3_reset_type reset_type;
2163         struct device *dev = &pdev->dev;
2164
2165         if (!ae_dev || !ae_dev->ops)
2166                 return PCI_ERS_RESULT_NONE;
2167
2168         ops = ae_dev->ops;
2169         /* request the reset */
2170         if (ops->reset_event && ops->get_reset_level &&
2171             ops->set_default_reset_request) {
2172                 if (ae_dev->hw_err_reset_req) {
2173                         reset_type = ops->get_reset_level(ae_dev,
2174                                                 &ae_dev->hw_err_reset_req);
2175                         ops->set_default_reset_request(ae_dev, reset_type);
2176                         dev_info(dev, "requesting reset due to PCI error\n");
2177                         ops->reset_event(pdev, NULL);
2178                 }
2179
2180                 return PCI_ERS_RESULT_RECOVERED;
2181         }
2182
2183         return PCI_ERS_RESULT_DISCONNECT;
2184 }
2185
2186 static void hns3_reset_prepare(struct pci_dev *pdev)
2187 {
2188         struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2189
2190         dev_info(&pdev->dev, "hns3 flr prepare\n");
2191         if (ae_dev && ae_dev->ops && ae_dev->ops->flr_prepare)
2192                 ae_dev->ops->flr_prepare(ae_dev);
2193 }
2194
2195 static void hns3_reset_done(struct pci_dev *pdev)
2196 {
2197         struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2198
2199         dev_info(&pdev->dev, "hns3 flr done\n");
2200         if (ae_dev && ae_dev->ops && ae_dev->ops->flr_done)
2201                 ae_dev->ops->flr_done(ae_dev);
2202 }
2203
2204 static const struct pci_error_handlers hns3_err_handler = {
2205         .error_detected = hns3_error_detected,
2206         .slot_reset     = hns3_slot_reset,
2207         .reset_prepare  = hns3_reset_prepare,
2208         .reset_done     = hns3_reset_done,
2209 };
2210
2211 static struct pci_driver hns3_driver = {
2212         .name     = hns3_driver_name,
2213         .id_table = hns3_pci_tbl,
2214         .probe    = hns3_probe,
2215         .remove   = hns3_remove,
2216         .shutdown = hns3_shutdown,
2217         .sriov_configure = hns3_pci_sriov_configure,
2218         .err_handler    = &hns3_err_handler,
2219 };
2220
2221 /* set default feature to hns3 */
2222 static void hns3_set_default_feature(struct net_device *netdev)
2223 {
2224         struct hnae3_handle *h = hns3_get_handle(netdev);
2225         struct pci_dev *pdev = h->pdev;
2226
2227         netdev->priv_flags |= IFF_UNICAST_FLT;
2228
2229         netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2230                 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2231                 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2232                 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2233                 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC |
2234                 NETIF_F_TSO_MANGLEID | NETIF_F_FRAGLIST;
2235
2236         netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
2237
2238         netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2239                 NETIF_F_HW_VLAN_CTAG_FILTER |
2240                 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
2241                 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2242                 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2243                 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2244                 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC |
2245                 NETIF_F_FRAGLIST;
2246
2247         netdev->vlan_features |=
2248                 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
2249                 NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO |
2250                 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2251                 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2252                 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC |
2253                 NETIF_F_FRAGLIST;
2254
2255         netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2256                 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
2257                 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2258                 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2259                 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2260                 NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC |
2261                 NETIF_F_FRAGLIST;
2262
2263         if (pdev->revision >= 0x21) {
2264                 netdev->hw_features |= NETIF_F_GRO_HW;
2265                 netdev->features |= NETIF_F_GRO_HW;
2266
2267                 if (!(h->flags & HNAE3_SUPPORT_VF)) {
2268                         netdev->hw_features |= NETIF_F_NTUPLE;
2269                         netdev->features |= NETIF_F_NTUPLE;
2270                 }
2271         }
2272 }
2273
2274 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
2275                              struct hns3_desc_cb *cb)
2276 {
2277         unsigned int order = hns3_page_order(ring);
2278         struct page *p;
2279
2280         p = dev_alloc_pages(order);
2281         if (!p)
2282                 return -ENOMEM;
2283
2284         cb->priv = p;
2285         cb->page_offset = 0;
2286         cb->reuse_flag = 0;
2287         cb->buf  = page_address(p);
2288         cb->length = hns3_page_size(ring);
2289         cb->type = DESC_TYPE_PAGE;
2290
2291         return 0;
2292 }
2293
2294 static void hns3_free_buffer(struct hns3_enet_ring *ring,
2295                              struct hns3_desc_cb *cb)
2296 {
2297         if (cb->type == DESC_TYPE_SKB)
2298                 dev_kfree_skb_any((struct sk_buff *)cb->priv);
2299         else if (!HNAE3_IS_TX_RING(ring))
2300                 put_page((struct page *)cb->priv);
2301         memset(cb, 0, sizeof(*cb));
2302 }
2303
2304 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
2305 {
2306         cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
2307                                cb->length, ring_to_dma_dir(ring));
2308
2309         if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
2310                 return -EIO;
2311
2312         return 0;
2313 }
2314
2315 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
2316                               struct hns3_desc_cb *cb)
2317 {
2318         if (cb->type == DESC_TYPE_SKB)
2319                 dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
2320                                  ring_to_dma_dir(ring));
2321         else if (cb->length)
2322                 dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
2323                                ring_to_dma_dir(ring));
2324 }
2325
2326 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
2327 {
2328         hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2329         ring->desc[i].addr = 0;
2330 }
2331
2332 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i)
2333 {
2334         struct hns3_desc_cb *cb = &ring->desc_cb[i];
2335
2336         if (!ring->desc_cb[i].dma)
2337                 return;
2338
2339         hns3_buffer_detach(ring, i);
2340         hns3_free_buffer(ring, cb);
2341 }
2342
2343 static void hns3_free_buffers(struct hns3_enet_ring *ring)
2344 {
2345         int i;
2346
2347         for (i = 0; i < ring->desc_num; i++)
2348                 hns3_free_buffer_detach(ring, i);
2349 }
2350
2351 /* free desc along with its attached buffer */
2352 static void hns3_free_desc(struct hns3_enet_ring *ring)
2353 {
2354         int size = ring->desc_num * sizeof(ring->desc[0]);
2355
2356         hns3_free_buffers(ring);
2357
2358         if (ring->desc) {
2359                 dma_free_coherent(ring_to_dev(ring), size,
2360                                   ring->desc, ring->desc_dma_addr);
2361                 ring->desc = NULL;
2362         }
2363 }
2364
2365 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
2366 {
2367         int size = ring->desc_num * sizeof(ring->desc[0]);
2368
2369         ring->desc = dma_alloc_coherent(ring_to_dev(ring), size,
2370                                         &ring->desc_dma_addr, GFP_KERNEL);
2371         if (!ring->desc)
2372                 return -ENOMEM;
2373
2374         return 0;
2375 }
2376
2377 static int hns3_reserve_buffer_map(struct hns3_enet_ring *ring,
2378                                    struct hns3_desc_cb *cb)
2379 {
2380         int ret;
2381
2382         ret = hns3_alloc_buffer(ring, cb);
2383         if (ret)
2384                 goto out;
2385
2386         ret = hns3_map_buffer(ring, cb);
2387         if (ret)
2388                 goto out_with_buf;
2389
2390         return 0;
2391
2392 out_with_buf:
2393         hns3_free_buffer(ring, cb);
2394 out:
2395         return ret;
2396 }
2397
2398 static int hns3_alloc_buffer_attach(struct hns3_enet_ring *ring, int i)
2399 {
2400         int ret = hns3_reserve_buffer_map(ring, &ring->desc_cb[i]);
2401
2402         if (ret)
2403                 return ret;
2404
2405         ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2406
2407         return 0;
2408 }
2409
2410 /* Allocate memory for raw pkg, and map with dma */
2411 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
2412 {
2413         int i, j, ret;
2414
2415         for (i = 0; i < ring->desc_num; i++) {
2416                 ret = hns3_alloc_buffer_attach(ring, i);
2417                 if (ret)
2418                         goto out_buffer_fail;
2419         }
2420
2421         return 0;
2422
2423 out_buffer_fail:
2424         for (j = i - 1; j >= 0; j--)
2425                 hns3_free_buffer_detach(ring, j);
2426         return ret;
2427 }
2428
2429 /* detach a in-used buffer and replace with a reserved one */
2430 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
2431                                 struct hns3_desc_cb *res_cb)
2432 {
2433         hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2434         ring->desc_cb[i] = *res_cb;
2435         ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2436         ring->desc[i].rx.bd_base_info = 0;
2437 }
2438
2439 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
2440 {
2441         ring->desc_cb[i].reuse_flag = 0;
2442         ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
2443                                          ring->desc_cb[i].page_offset);
2444         ring->desc[i].rx.bd_base_info = 0;
2445 }
2446
2447 static void hns3_nic_reclaim_desc(struct hns3_enet_ring *ring, int head,
2448                                   int *bytes, int *pkts)
2449 {
2450         int ntc = ring->next_to_clean;
2451         struct hns3_desc_cb *desc_cb;
2452
2453         while (head != ntc) {
2454                 desc_cb = &ring->desc_cb[ntc];
2455                 (*pkts) += (desc_cb->type == DESC_TYPE_SKB);
2456                 (*bytes) += desc_cb->length;
2457                 /* desc_cb will be cleaned, after hnae3_free_buffer_detach */
2458                 hns3_free_buffer_detach(ring, ntc);
2459
2460                 if (++ntc == ring->desc_num)
2461                         ntc = 0;
2462
2463                 /* Issue prefetch for next Tx descriptor */
2464                 prefetch(&ring->desc_cb[ntc]);
2465         }
2466
2467         /* This smp_store_release() pairs with smp_load_acquire() in
2468          * ring_space called by hns3_nic_net_xmit.
2469          */
2470         smp_store_release(&ring->next_to_clean, ntc);
2471 }
2472
2473 static int is_valid_clean_head(struct hns3_enet_ring *ring, int h)
2474 {
2475         int u = ring->next_to_use;
2476         int c = ring->next_to_clean;
2477
2478         if (unlikely(h > ring->desc_num))
2479                 return 0;
2480
2481         return u > c ? (h > c && h <= u) : (h > c || h <= u);
2482 }
2483
2484 void hns3_clean_tx_ring(struct hns3_enet_ring *ring)
2485 {
2486         struct net_device *netdev = ring_to_netdev(ring);
2487         struct hns3_nic_priv *priv = netdev_priv(netdev);
2488         struct netdev_queue *dev_queue;
2489         int bytes, pkts;
2490         int head;
2491
2492         head = readl_relaxed(ring->tqp->io_base + HNS3_RING_TX_RING_HEAD_REG);
2493
2494         if (is_ring_empty(ring) || head == ring->next_to_clean)
2495                 return; /* no data to poll */
2496
2497         rmb(); /* Make sure head is ready before touch any data */
2498
2499         if (unlikely(!is_valid_clean_head(ring, head))) {
2500                 netdev_err(netdev, "wrong head (%d, %d-%d)\n", head,
2501                            ring->next_to_use, ring->next_to_clean);
2502
2503                 u64_stats_update_begin(&ring->syncp);
2504                 ring->stats.io_err_cnt++;
2505                 u64_stats_update_end(&ring->syncp);
2506                 return;
2507         }
2508
2509         bytes = 0;
2510         pkts = 0;
2511         hns3_nic_reclaim_desc(ring, head, &bytes, &pkts);
2512
2513         ring->tqp_vector->tx_group.total_bytes += bytes;
2514         ring->tqp_vector->tx_group.total_packets += pkts;
2515
2516         u64_stats_update_begin(&ring->syncp);
2517         ring->stats.tx_bytes += bytes;
2518         ring->stats.tx_pkts += pkts;
2519         u64_stats_update_end(&ring->syncp);
2520
2521         dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
2522         netdev_tx_completed_queue(dev_queue, pkts, bytes);
2523
2524         if (unlikely(netif_carrier_ok(netdev) &&
2525                      ring_space(ring) > HNS3_MAX_TSO_BD_NUM)) {
2526                 /* Make sure that anybody stopping the queue after this
2527                  * sees the new next_to_clean.
2528                  */
2529                 smp_mb();
2530                 if (netif_tx_queue_stopped(dev_queue) &&
2531                     !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
2532                         netif_tx_wake_queue(dev_queue);
2533                         ring->stats.restart_queue++;
2534                 }
2535         }
2536 }
2537
2538 static int hns3_desc_unused(struct hns3_enet_ring *ring)
2539 {
2540         int ntc = ring->next_to_clean;
2541         int ntu = ring->next_to_use;
2542
2543         return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
2544 }
2545
2546 static void hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring,
2547                                       int cleand_count)
2548 {
2549         struct hns3_desc_cb *desc_cb;
2550         struct hns3_desc_cb res_cbs;
2551         int i, ret;
2552
2553         for (i = 0; i < cleand_count; i++) {
2554                 desc_cb = &ring->desc_cb[ring->next_to_use];
2555                 if (desc_cb->reuse_flag) {
2556                         u64_stats_update_begin(&ring->syncp);
2557                         ring->stats.reuse_pg_cnt++;
2558                         u64_stats_update_end(&ring->syncp);
2559
2560                         hns3_reuse_buffer(ring, ring->next_to_use);
2561                 } else {
2562                         ret = hns3_reserve_buffer_map(ring, &res_cbs);
2563                         if (ret) {
2564                                 u64_stats_update_begin(&ring->syncp);
2565                                 ring->stats.sw_err_cnt++;
2566                                 u64_stats_update_end(&ring->syncp);
2567
2568                                 hns3_rl_err(ring_to_netdev(ring),
2569                                             "alloc rx buffer failed: %d\n",
2570                                             ret);
2571                                 break;
2572                         }
2573                         hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
2574
2575                         u64_stats_update_begin(&ring->syncp);
2576                         ring->stats.non_reuse_pg++;
2577                         u64_stats_update_end(&ring->syncp);
2578                 }
2579
2580                 ring_ptr_move_fw(ring, next_to_use);
2581         }
2582
2583         wmb(); /* Make all data has been write before submit */
2584         writel_relaxed(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
2585 }
2586
2587 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
2588                                 struct hns3_enet_ring *ring, int pull_len,
2589                                 struct hns3_desc_cb *desc_cb)
2590 {
2591         struct hns3_desc *desc = &ring->desc[ring->next_to_clean];
2592         int size = le16_to_cpu(desc->rx.size);
2593         u32 truesize = hns3_buf_size(ring);
2594
2595         skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
2596                         size - pull_len, truesize);
2597
2598         /* Avoid re-using remote pages, or the stack is still using the page
2599          * when page_offset rollback to zero, flag default unreuse
2600          */
2601         if (unlikely(page_to_nid(desc_cb->priv) != numa_mem_id()) ||
2602             (!desc_cb->page_offset && page_count(desc_cb->priv) > 1))
2603                 return;
2604
2605         /* Move offset up to the next cache line */
2606         desc_cb->page_offset += truesize;
2607
2608         if (desc_cb->page_offset + truesize <= hns3_page_size(ring)) {
2609                 desc_cb->reuse_flag = 1;
2610                 /* Bump ref count on page before it is given */
2611                 get_page(desc_cb->priv);
2612         } else if (page_count(desc_cb->priv) == 1) {
2613                 desc_cb->reuse_flag = 1;
2614                 desc_cb->page_offset = 0;
2615                 get_page(desc_cb->priv);
2616         }
2617 }
2618
2619 static int hns3_gro_complete(struct sk_buff *skb, u32 l234info)
2620 {
2621         __be16 type = skb->protocol;
2622         struct tcphdr *th;
2623         int depth = 0;
2624
2625         while (eth_type_vlan(type)) {
2626                 struct vlan_hdr *vh;
2627
2628                 if ((depth + VLAN_HLEN) > skb_headlen(skb))
2629                         return -EFAULT;
2630
2631                 vh = (struct vlan_hdr *)(skb->data + depth);
2632                 type = vh->h_vlan_encapsulated_proto;
2633                 depth += VLAN_HLEN;
2634         }
2635
2636         skb_set_network_header(skb, depth);
2637
2638         if (type == htons(ETH_P_IP)) {
2639                 const struct iphdr *iph = ip_hdr(skb);
2640
2641                 depth += sizeof(struct iphdr);
2642                 skb_set_transport_header(skb, depth);
2643                 th = tcp_hdr(skb);
2644                 th->check = ~tcp_v4_check(skb->len - depth, iph->saddr,
2645                                           iph->daddr, 0);
2646         } else if (type == htons(ETH_P_IPV6)) {
2647                 const struct ipv6hdr *iph = ipv6_hdr(skb);
2648
2649                 depth += sizeof(struct ipv6hdr);
2650                 skb_set_transport_header(skb, depth);
2651                 th = tcp_hdr(skb);
2652                 th->check = ~tcp_v6_check(skb->len - depth, &iph->saddr,
2653                                           &iph->daddr, 0);
2654         } else {
2655                 hns3_rl_err(skb->dev,
2656                             "Error: FW GRO supports only IPv4/IPv6, not 0x%04x, depth: %d\n",
2657                             be16_to_cpu(type), depth);
2658                 return -EFAULT;
2659         }
2660
2661         skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
2662         if (th->cwr)
2663                 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
2664
2665         if (l234info & BIT(HNS3_RXD_GRO_FIXID_B))
2666                 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_FIXEDID;
2667
2668         skb->csum_start = (unsigned char *)th - skb->head;
2669         skb->csum_offset = offsetof(struct tcphdr, check);
2670         skb->ip_summed = CHECKSUM_PARTIAL;
2671         return 0;
2672 }
2673
2674 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
2675                              u32 l234info, u32 bd_base_info, u32 ol_info)
2676 {
2677         struct net_device *netdev = ring_to_netdev(ring);
2678         int l3_type, l4_type;
2679         int ol4_type;
2680
2681         skb->ip_summed = CHECKSUM_NONE;
2682
2683         skb_checksum_none_assert(skb);
2684
2685         if (!(netdev->features & NETIF_F_RXCSUM))
2686                 return;
2687
2688         /* check if hardware has done checksum */
2689         if (!(bd_base_info & BIT(HNS3_RXD_L3L4P_B)))
2690                 return;
2691
2692         if (unlikely(l234info & (BIT(HNS3_RXD_L3E_B) | BIT(HNS3_RXD_L4E_B) |
2693                                  BIT(HNS3_RXD_OL3E_B) |
2694                                  BIT(HNS3_RXD_OL4E_B)))) {
2695                 u64_stats_update_begin(&ring->syncp);
2696                 ring->stats.l3l4_csum_err++;
2697                 u64_stats_update_end(&ring->syncp);
2698
2699                 return;
2700         }
2701
2702         ol4_type = hnae3_get_field(ol_info, HNS3_RXD_OL4ID_M,
2703                                    HNS3_RXD_OL4ID_S);
2704         switch (ol4_type) {
2705         case HNS3_OL4_TYPE_MAC_IN_UDP:
2706         case HNS3_OL4_TYPE_NVGRE:
2707                 skb->csum_level = 1;
2708                 /* fall through */
2709         case HNS3_OL4_TYPE_NO_TUN:
2710                 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
2711                                           HNS3_RXD_L3ID_S);
2712                 l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
2713                                           HNS3_RXD_L4ID_S);
2714
2715                 /* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
2716                 if ((l3_type == HNS3_L3_TYPE_IPV4 ||
2717                      l3_type == HNS3_L3_TYPE_IPV6) &&
2718                     (l4_type == HNS3_L4_TYPE_UDP ||
2719                      l4_type == HNS3_L4_TYPE_TCP ||
2720                      l4_type == HNS3_L4_TYPE_SCTP))
2721                         skb->ip_summed = CHECKSUM_UNNECESSARY;
2722                 break;
2723         default:
2724                 break;
2725         }
2726 }
2727
2728 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
2729 {
2730         if (skb_has_frag_list(skb))
2731                 napi_gro_flush(&ring->tqp_vector->napi, false);
2732
2733         napi_gro_receive(&ring->tqp_vector->napi, skb);
2734 }
2735
2736 static bool hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
2737                                 struct hns3_desc *desc, u32 l234info,
2738                                 u16 *vlan_tag)
2739 {
2740         struct hnae3_handle *handle = ring->tqp->handle;
2741         struct pci_dev *pdev = ring->tqp->handle->pdev;
2742
2743         if (pdev->revision == 0x20) {
2744                 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2745                 if (!(*vlan_tag & VLAN_VID_MASK))
2746                         *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2747
2748                 return (*vlan_tag != 0);
2749         }
2750
2751 #define HNS3_STRP_OUTER_VLAN    0x1
2752 #define HNS3_STRP_INNER_VLAN    0x2
2753 #define HNS3_STRP_BOTH          0x3
2754
2755         /* Hardware always insert VLAN tag into RX descriptor when
2756          * remove the tag from packet, driver needs to determine
2757          * reporting which tag to stack.
2758          */
2759         switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
2760                                 HNS3_RXD_STRP_TAGP_S)) {
2761         case HNS3_STRP_OUTER_VLAN:
2762                 if (handle->port_base_vlan_state !=
2763                                 HNAE3_PORT_BASE_VLAN_DISABLE)
2764                         return false;
2765
2766                 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2767                 return true;
2768         case HNS3_STRP_INNER_VLAN:
2769                 if (handle->port_base_vlan_state !=
2770                                 HNAE3_PORT_BASE_VLAN_DISABLE)
2771                         return false;
2772
2773                 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2774                 return true;
2775         case HNS3_STRP_BOTH:
2776                 if (handle->port_base_vlan_state ==
2777                                 HNAE3_PORT_BASE_VLAN_DISABLE)
2778                         *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2779                 else
2780                         *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2781
2782                 return true;
2783         default:
2784                 return false;
2785         }
2786 }
2787
2788 static int hns3_alloc_skb(struct hns3_enet_ring *ring, unsigned int length,
2789                           unsigned char *va)
2790 {
2791 #define HNS3_NEED_ADD_FRAG      1
2792         struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
2793         struct net_device *netdev = ring_to_netdev(ring);
2794         struct sk_buff *skb;
2795
2796         ring->skb = napi_alloc_skb(&ring->tqp_vector->napi, HNS3_RX_HEAD_SIZE);
2797         skb = ring->skb;
2798         if (unlikely(!skb)) {
2799                 hns3_rl_err(netdev, "alloc rx skb fail\n");
2800
2801                 u64_stats_update_begin(&ring->syncp);
2802                 ring->stats.sw_err_cnt++;
2803                 u64_stats_update_end(&ring->syncp);
2804
2805                 return -ENOMEM;
2806         }
2807
2808         prefetchw(skb->data);
2809
2810         ring->pending_buf = 1;
2811         ring->frag_num = 0;
2812         ring->tail_skb = NULL;
2813         if (length <= HNS3_RX_HEAD_SIZE) {
2814                 memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
2815
2816                 /* We can reuse buffer as-is, just make sure it is local */
2817                 if (likely(page_to_nid(desc_cb->priv) == numa_mem_id()))
2818                         desc_cb->reuse_flag = 1;
2819                 else /* This page cannot be reused so discard it */
2820                         put_page(desc_cb->priv);
2821
2822                 ring_ptr_move_fw(ring, next_to_clean);
2823                 return 0;
2824         }
2825         u64_stats_update_begin(&ring->syncp);
2826         ring->stats.seg_pkt_cnt++;
2827         u64_stats_update_end(&ring->syncp);
2828
2829         ring->pull_len = eth_get_headlen(netdev, va, HNS3_RX_HEAD_SIZE);
2830         __skb_put(skb, ring->pull_len);
2831         hns3_nic_reuse_page(skb, ring->frag_num++, ring, ring->pull_len,
2832                             desc_cb);
2833         ring_ptr_move_fw(ring, next_to_clean);
2834
2835         return HNS3_NEED_ADD_FRAG;
2836 }
2837
2838 static int hns3_add_frag(struct hns3_enet_ring *ring, struct hns3_desc *desc,
2839                          bool pending)
2840 {
2841         struct sk_buff *skb = ring->skb;
2842         struct sk_buff *head_skb = skb;
2843         struct sk_buff *new_skb;
2844         struct hns3_desc_cb *desc_cb;
2845         struct hns3_desc *pre_desc;
2846         u32 bd_base_info;
2847         int pre_bd;
2848
2849         /* if there is pending bd, the SW param next_to_clean has moved
2850          * to next and the next is NULL
2851          */
2852         if (pending) {
2853                 pre_bd = (ring->next_to_clean - 1 + ring->desc_num) %
2854                          ring->desc_num;
2855                 pre_desc = &ring->desc[pre_bd];
2856                 bd_base_info = le32_to_cpu(pre_desc->rx.bd_base_info);
2857         } else {
2858                 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2859         }
2860
2861         while (!(bd_base_info & BIT(HNS3_RXD_FE_B))) {
2862                 desc = &ring->desc[ring->next_to_clean];
2863                 desc_cb = &ring->desc_cb[ring->next_to_clean];
2864                 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2865                 /* make sure HW write desc complete */
2866                 dma_rmb();
2867                 if (!(bd_base_info & BIT(HNS3_RXD_VLD_B)))
2868                         return -ENXIO;
2869
2870                 if (unlikely(ring->frag_num >= MAX_SKB_FRAGS)) {
2871                         new_skb = napi_alloc_skb(&ring->tqp_vector->napi, 0);
2872                         if (unlikely(!new_skb)) {
2873                                 hns3_rl_err(ring_to_netdev(ring),
2874                                             "alloc rx fraglist skb fail\n");
2875                                 return -ENXIO;
2876                         }
2877                         ring->frag_num = 0;
2878
2879                         if (ring->tail_skb) {
2880                                 ring->tail_skb->next = new_skb;
2881                                 ring->tail_skb = new_skb;
2882                         } else {
2883                                 skb_shinfo(skb)->frag_list = new_skb;
2884                                 ring->tail_skb = new_skb;
2885                         }
2886                 }
2887
2888                 if (ring->tail_skb) {
2889                         head_skb->truesize += hns3_buf_size(ring);
2890                         head_skb->data_len += le16_to_cpu(desc->rx.size);
2891                         head_skb->len += le16_to_cpu(desc->rx.size);
2892                         skb = ring->tail_skb;
2893                 }
2894
2895                 hns3_nic_reuse_page(skb, ring->frag_num++, ring, 0, desc_cb);
2896                 ring_ptr_move_fw(ring, next_to_clean);
2897                 ring->pending_buf++;
2898         }
2899
2900         return 0;
2901 }
2902
2903 static int hns3_set_gro_and_checksum(struct hns3_enet_ring *ring,
2904                                      struct sk_buff *skb, u32 l234info,
2905                                      u32 bd_base_info, u32 ol_info)
2906 {
2907         u32 l3_type;
2908
2909         skb_shinfo(skb)->gso_size = hnae3_get_field(bd_base_info,
2910                                                     HNS3_RXD_GRO_SIZE_M,
2911                                                     HNS3_RXD_GRO_SIZE_S);
2912         /* if there is no HW GRO, do not set gro params */
2913         if (!skb_shinfo(skb)->gso_size) {
2914                 hns3_rx_checksum(ring, skb, l234info, bd_base_info, ol_info);
2915                 return 0;
2916         }
2917
2918         NAPI_GRO_CB(skb)->count = hnae3_get_field(l234info,
2919                                                   HNS3_RXD_GRO_COUNT_M,
2920                                                   HNS3_RXD_GRO_COUNT_S);
2921
2922         l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M, HNS3_RXD_L3ID_S);
2923         if (l3_type == HNS3_L3_TYPE_IPV4)
2924                 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2925         else if (l3_type == HNS3_L3_TYPE_IPV6)
2926                 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
2927         else
2928                 return -EFAULT;
2929
2930         return  hns3_gro_complete(skb, l234info);
2931 }
2932
2933 static void hns3_set_rx_skb_rss_type(struct hns3_enet_ring *ring,
2934                                      struct sk_buff *skb, u32 rss_hash)
2935 {
2936         struct hnae3_handle *handle = ring->tqp->handle;
2937         enum pkt_hash_types rss_type;
2938
2939         if (rss_hash)
2940                 rss_type = handle->kinfo.rss_type;
2941         else
2942                 rss_type = PKT_HASH_TYPE_NONE;
2943
2944         skb_set_hash(skb, rss_hash, rss_type);
2945 }
2946
2947 static int hns3_handle_bdinfo(struct hns3_enet_ring *ring, struct sk_buff *skb)
2948 {
2949         struct net_device *netdev = ring_to_netdev(ring);
2950         enum hns3_pkt_l2t_type l2_frame_type;
2951         u32 bd_base_info, l234info, ol_info;
2952         struct hns3_desc *desc;
2953         unsigned int len;
2954         int pre_ntc, ret;
2955
2956         /* bdinfo handled below is only valid on the last BD of the
2957          * current packet, and ring->next_to_clean indicates the first
2958          * descriptor of next packet, so need - 1 below.
2959          */
2960         pre_ntc = ring->next_to_clean ? (ring->next_to_clean - 1) :
2961                                         (ring->desc_num - 1);
2962         desc = &ring->desc[pre_ntc];
2963         bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2964         l234info = le32_to_cpu(desc->rx.l234_info);
2965         ol_info = le32_to_cpu(desc->rx.ol_info);
2966
2967         /* Based on hw strategy, the tag offloaded will be stored at
2968          * ot_vlan_tag in two layer tag case, and stored at vlan_tag
2969          * in one layer tag case.
2970          */
2971         if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
2972                 u16 vlan_tag;
2973
2974                 if (hns3_parse_vlan_tag(ring, desc, l234info, &vlan_tag))
2975                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2976                                                vlan_tag);
2977         }
2978
2979         if (unlikely(!desc->rx.pkt_len || (l234info & (BIT(HNS3_RXD_TRUNCAT_B) |
2980                                   BIT(HNS3_RXD_L2E_B))))) {
2981                 u64_stats_update_begin(&ring->syncp);
2982                 if (l234info & BIT(HNS3_RXD_L2E_B))
2983                         ring->stats.l2_err++;
2984                 else
2985                         ring->stats.err_pkt_len++;
2986                 u64_stats_update_end(&ring->syncp);
2987
2988                 return -EFAULT;
2989         }
2990
2991         len = skb->len;
2992
2993         /* Do update ip stack process */
2994         skb->protocol = eth_type_trans(skb, netdev);
2995
2996         /* This is needed in order to enable forwarding support */
2997         ret = hns3_set_gro_and_checksum(ring, skb, l234info,
2998                                         bd_base_info, ol_info);
2999         if (unlikely(ret)) {
3000                 u64_stats_update_begin(&ring->syncp);
3001                 ring->stats.rx_err_cnt++;
3002                 u64_stats_update_end(&ring->syncp);
3003                 return ret;
3004         }
3005
3006         l2_frame_type = hnae3_get_field(l234info, HNS3_RXD_DMAC_M,
3007                                         HNS3_RXD_DMAC_S);
3008
3009         u64_stats_update_begin(&ring->syncp);
3010         ring->stats.rx_pkts++;
3011         ring->stats.rx_bytes += len;
3012
3013         if (l2_frame_type == HNS3_L2_TYPE_MULTICAST)
3014                 ring->stats.rx_multicast++;
3015
3016         u64_stats_update_end(&ring->syncp);
3017
3018         ring->tqp_vector->rx_group.total_bytes += len;
3019
3020         hns3_set_rx_skb_rss_type(ring, skb, le32_to_cpu(desc->rx.rss_hash));
3021         return 0;
3022 }
3023
3024 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring)
3025 {
3026         struct sk_buff *skb = ring->skb;
3027         struct hns3_desc_cb *desc_cb;
3028         struct hns3_desc *desc;
3029         unsigned int length;
3030         u32 bd_base_info;
3031         int ret;
3032
3033         desc = &ring->desc[ring->next_to_clean];
3034         desc_cb = &ring->desc_cb[ring->next_to_clean];
3035
3036         prefetch(desc);
3037
3038         length = le16_to_cpu(desc->rx.size);
3039         bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
3040
3041         /* Check valid BD */
3042         if (unlikely(!(bd_base_info & BIT(HNS3_RXD_VLD_B))))
3043                 return -ENXIO;
3044
3045         if (!skb)
3046                 ring->va = (unsigned char *)desc_cb->buf + desc_cb->page_offset;
3047
3048         /* Prefetch first cache line of first page
3049          * Idea is to cache few bytes of the header of the packet. Our L1 Cache
3050          * line size is 64B so need to prefetch twice to make it 128B. But in
3051          * actual we can have greater size of caches with 128B Level 1 cache
3052          * lines. In such a case, single fetch would suffice to cache in the
3053          * relevant part of the header.
3054          */
3055         prefetch(ring->va);
3056 #if L1_CACHE_BYTES < 128
3057         prefetch(ring->va + L1_CACHE_BYTES);
3058 #endif
3059
3060         if (!skb) {
3061                 ret = hns3_alloc_skb(ring, length, ring->va);
3062                 skb = ring->skb;
3063
3064                 if (ret < 0) /* alloc buffer fail */
3065                         return ret;
3066                 if (ret > 0) { /* need add frag */
3067                         ret = hns3_add_frag(ring, desc, false);
3068                         if (ret)
3069                                 return ret;
3070
3071                         /* As the head data may be changed when GRO enable, copy
3072                          * the head data in after other data rx completed
3073                          */
3074                         memcpy(skb->data, ring->va,
3075                                ALIGN(ring->pull_len, sizeof(long)));
3076                 }
3077         } else {
3078                 ret = hns3_add_frag(ring, desc, true);
3079                 if (ret)
3080                         return ret;
3081
3082                 /* As the head data may be changed when GRO enable, copy
3083                  * the head data in after other data rx completed
3084                  */
3085                 memcpy(skb->data, ring->va,
3086                        ALIGN(ring->pull_len, sizeof(long)));
3087         }
3088
3089         ret = hns3_handle_bdinfo(ring, skb);
3090         if (unlikely(ret)) {
3091                 dev_kfree_skb_any(skb);
3092                 return ret;
3093         }
3094
3095         skb_record_rx_queue(skb, ring->tqp->tqp_index);
3096         return 0;
3097 }
3098
3099 int hns3_clean_rx_ring(struct hns3_enet_ring *ring, int budget,
3100                        void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
3101 {
3102 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
3103         int unused_count = hns3_desc_unused(ring);
3104         int recv_pkts = 0;
3105         int recv_bds = 0;
3106         int err, num;
3107
3108         num = readl_relaxed(ring->tqp->io_base + HNS3_RING_RX_RING_FBDNUM_REG);
3109         num -= unused_count;
3110         unused_count -= ring->pending_buf;
3111
3112         if (num <= 0)
3113                 goto out;
3114
3115         rmb(); /* Make sure num taken effect before the other data is touched */
3116
3117         while (recv_pkts < budget && recv_bds < num) {
3118                 /* Reuse or realloc buffers */
3119                 if (unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
3120                         hns3_nic_alloc_rx_buffers(ring, unused_count);
3121                         unused_count = hns3_desc_unused(ring) -
3122                                         ring->pending_buf;
3123                 }
3124
3125                 /* Poll one pkt */
3126                 err = hns3_handle_rx_bd(ring);
3127                 /* Do not get FE for the packet or failed to alloc skb */
3128                 if (unlikely(!ring->skb || err == -ENXIO)) {
3129                         goto out;
3130                 } else if (likely(!err)) {
3131                         rx_fn(ring, ring->skb);
3132                         recv_pkts++;
3133                 }
3134
3135                 recv_bds += ring->pending_buf;
3136                 unused_count += ring->pending_buf;
3137                 ring->skb = NULL;
3138                 ring->pending_buf = 0;
3139         }
3140
3141 out:
3142         /* Make all data has been write before submit */
3143         if (unused_count > 0)
3144                 hns3_nic_alloc_rx_buffers(ring, unused_count);
3145
3146         return recv_pkts;
3147 }
3148
3149 static bool hns3_get_new_flow_lvl(struct hns3_enet_ring_group *ring_group)
3150 {
3151 #define HNS3_RX_LOW_BYTE_RATE 10000
3152 #define HNS3_RX_MID_BYTE_RATE 20000
3153 #define HNS3_RX_ULTRA_PACKET_RATE 40
3154
3155         enum hns3_flow_level_range new_flow_level;
3156         struct hns3_enet_tqp_vector *tqp_vector;
3157         int packets_per_msecs, bytes_per_msecs;
3158         u32 time_passed_ms;
3159
3160         tqp_vector = ring_group->ring->tqp_vector;
3161         time_passed_ms =
3162                 jiffies_to_msecs(jiffies - tqp_vector->last_jiffies);
3163         if (!time_passed_ms)
3164                 return false;
3165
3166         do_div(ring_group->total_packets, time_passed_ms);
3167         packets_per_msecs = ring_group->total_packets;
3168
3169         do_div(ring_group->total_bytes, time_passed_ms);
3170         bytes_per_msecs = ring_group->total_bytes;
3171
3172         new_flow_level = ring_group->coal.flow_level;
3173
3174         /* Simple throttlerate management
3175          * 0-10MB/s   lower     (50000 ints/s)
3176          * 10-20MB/s   middle    (20000 ints/s)
3177          * 20-1249MB/s high      (18000 ints/s)
3178          * > 40000pps  ultra     (8000 ints/s)
3179          */
3180         switch (new_flow_level) {
3181         case HNS3_FLOW_LOW:
3182                 if (bytes_per_msecs > HNS3_RX_LOW_BYTE_RATE)
3183                         new_flow_level = HNS3_FLOW_MID;
3184                 break;
3185         case HNS3_FLOW_MID:
3186                 if (bytes_per_msecs > HNS3_RX_MID_BYTE_RATE)
3187                         new_flow_level = HNS3_FLOW_HIGH;
3188                 else if (bytes_per_msecs <= HNS3_RX_LOW_BYTE_RATE)
3189                         new_flow_level = HNS3_FLOW_LOW;
3190                 break;
3191         case HNS3_FLOW_HIGH:
3192         case HNS3_FLOW_ULTRA:
3193         default:
3194                 if (bytes_per_msecs <= HNS3_RX_MID_BYTE_RATE)
3195                         new_flow_level = HNS3_FLOW_MID;
3196                 break;
3197         }
3198
3199         if (packets_per_msecs > HNS3_RX_ULTRA_PACKET_RATE &&
3200             &tqp_vector->rx_group == ring_group)
3201                 new_flow_level = HNS3_FLOW_ULTRA;
3202
3203         ring_group->total_bytes = 0;
3204         ring_group->total_packets = 0;
3205         ring_group->coal.flow_level = new_flow_level;
3206
3207         return true;
3208 }
3209
3210 static bool hns3_get_new_int_gl(struct hns3_enet_ring_group *ring_group)
3211 {
3212         struct hns3_enet_tqp_vector *tqp_vector;
3213         u16 new_int_gl;
3214
3215         if (!ring_group->ring)
3216                 return false;
3217
3218         tqp_vector = ring_group->ring->tqp_vector;
3219         if (!tqp_vector->last_jiffies)
3220                 return false;
3221
3222         if (ring_group->total_packets == 0) {
3223                 ring_group->coal.int_gl = HNS3_INT_GL_50K;
3224                 ring_group->coal.flow_level = HNS3_FLOW_LOW;
3225                 return true;
3226         }
3227
3228         if (!hns3_get_new_flow_lvl(ring_group))
3229                 return false;
3230
3231         new_int_gl = ring_group->coal.int_gl;
3232         switch (ring_group->coal.flow_level) {
3233         case HNS3_FLOW_LOW:
3234                 new_int_gl = HNS3_INT_GL_50K;
3235                 break;
3236         case HNS3_FLOW_MID:
3237                 new_int_gl = HNS3_INT_GL_20K;
3238                 break;
3239         case HNS3_FLOW_HIGH:
3240                 new_int_gl = HNS3_INT_GL_18K;
3241                 break;
3242         case HNS3_FLOW_ULTRA:
3243                 new_int_gl = HNS3_INT_GL_8K;
3244                 break;
3245         default:
3246                 break;
3247         }
3248
3249         if (new_int_gl != ring_group->coal.int_gl) {
3250                 ring_group->coal.int_gl = new_int_gl;
3251                 return true;
3252         }
3253         return false;
3254 }
3255
3256 static void hns3_update_new_int_gl(struct hns3_enet_tqp_vector *tqp_vector)
3257 {
3258         struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
3259         struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
3260         bool rx_update, tx_update;
3261
3262         /* update param every 1000ms */
3263         if (time_before(jiffies,
3264                         tqp_vector->last_jiffies + msecs_to_jiffies(1000)))
3265                 return;
3266
3267         if (rx_group->coal.gl_adapt_enable) {
3268                 rx_update = hns3_get_new_int_gl(rx_group);
3269                 if (rx_update)
3270                         hns3_set_vector_coalesce_rx_gl(tqp_vector,
3271                                                        rx_group->coal.int_gl);
3272         }
3273
3274         if (tx_group->coal.gl_adapt_enable) {
3275                 tx_update = hns3_get_new_int_gl(tx_group);
3276                 if (tx_update)
3277                         hns3_set_vector_coalesce_tx_gl(tqp_vector,
3278                                                        tx_group->coal.int_gl);
3279         }
3280
3281         tqp_vector->last_jiffies = jiffies;
3282 }
3283
3284 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
3285 {
3286         struct hns3_nic_priv *priv = netdev_priv(napi->dev);
3287         struct hns3_enet_ring *ring;
3288         int rx_pkt_total = 0;
3289
3290         struct hns3_enet_tqp_vector *tqp_vector =
3291                 container_of(napi, struct hns3_enet_tqp_vector, napi);
3292         bool clean_complete = true;
3293         int rx_budget = budget;
3294
3295         if (unlikely(test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
3296                 napi_complete(napi);
3297                 return 0;
3298         }
3299
3300         /* Since the actual Tx work is minimal, we can give the Tx a larger
3301          * budget and be more aggressive about cleaning up the Tx descriptors.
3302          */
3303         hns3_for_each_ring(ring, tqp_vector->tx_group)
3304                 hns3_clean_tx_ring(ring);
3305
3306         /* make sure rx ring budget not smaller than 1 */
3307         if (tqp_vector->num_tqps > 1)
3308                 rx_budget = max(budget / tqp_vector->num_tqps, 1);
3309
3310         hns3_for_each_ring(ring, tqp_vector->rx_group) {
3311                 int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
3312                                                     hns3_rx_skb);
3313
3314                 if (rx_cleaned >= rx_budget)
3315                         clean_complete = false;
3316
3317                 rx_pkt_total += rx_cleaned;
3318         }
3319
3320         tqp_vector->rx_group.total_packets += rx_pkt_total;
3321
3322         if (!clean_complete)
3323                 return budget;
3324
3325         if (napi_complete(napi) &&
3326             likely(!test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
3327                 hns3_update_new_int_gl(tqp_vector);
3328                 hns3_mask_vector_irq(tqp_vector, 1);
3329         }
3330
3331         return rx_pkt_total;
3332 }
3333
3334 static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
3335                                       struct hnae3_ring_chain_node *head)
3336 {
3337         struct pci_dev *pdev = tqp_vector->handle->pdev;
3338         struct hnae3_ring_chain_node *cur_chain = head;
3339         struct hnae3_ring_chain_node *chain;
3340         struct hns3_enet_ring *tx_ring;
3341         struct hns3_enet_ring *rx_ring;
3342
3343         tx_ring = tqp_vector->tx_group.ring;
3344         if (tx_ring) {
3345                 cur_chain->tqp_index = tx_ring->tqp->tqp_index;
3346                 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
3347                               HNAE3_RING_TYPE_TX);
3348                 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3349                                 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_TX);
3350
3351                 cur_chain->next = NULL;
3352
3353                 while (tx_ring->next) {
3354                         tx_ring = tx_ring->next;
3355
3356                         chain = devm_kzalloc(&pdev->dev, sizeof(*chain),
3357                                              GFP_KERNEL);
3358                         if (!chain)
3359                                 goto err_free_chain;
3360
3361                         cur_chain->next = chain;
3362                         chain->tqp_index = tx_ring->tqp->tqp_index;
3363                         hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
3364                                       HNAE3_RING_TYPE_TX);
3365                         hnae3_set_field(chain->int_gl_idx,
3366                                         HNAE3_RING_GL_IDX_M,
3367                                         HNAE3_RING_GL_IDX_S,
3368                                         HNAE3_RING_GL_TX);
3369
3370                         cur_chain = chain;
3371                 }
3372         }
3373
3374         rx_ring = tqp_vector->rx_group.ring;
3375         if (!tx_ring && rx_ring) {
3376                 cur_chain->next = NULL;
3377                 cur_chain->tqp_index = rx_ring->tqp->tqp_index;
3378                 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
3379                               HNAE3_RING_TYPE_RX);
3380                 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3381                                 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
3382
3383                 rx_ring = rx_ring->next;
3384         }
3385
3386         while (rx_ring) {
3387                 chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
3388                 if (!chain)
3389                         goto err_free_chain;
3390
3391                 cur_chain->next = chain;
3392                 chain->tqp_index = rx_ring->tqp->tqp_index;
3393                 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
3394                               HNAE3_RING_TYPE_RX);
3395                 hnae3_set_field(chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3396                                 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
3397
3398                 cur_chain = chain;
3399
3400                 rx_ring = rx_ring->next;
3401         }
3402
3403         return 0;
3404
3405 err_free_chain:
3406         cur_chain = head->next;
3407         while (cur_chain) {
3408                 chain = cur_chain->next;
3409                 devm_kfree(&pdev->dev, cur_chain);
3410                 cur_chain = chain;
3411         }
3412         head->next = NULL;
3413
3414         return -ENOMEM;
3415 }
3416
3417 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
3418                                         struct hnae3_ring_chain_node *head)
3419 {
3420         struct pci_dev *pdev = tqp_vector->handle->pdev;
3421         struct hnae3_ring_chain_node *chain_tmp, *chain;
3422
3423         chain = head->next;
3424
3425         while (chain) {
3426                 chain_tmp = chain->next;
3427                 devm_kfree(&pdev->dev, chain);
3428                 chain = chain_tmp;
3429         }
3430 }
3431
3432 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
3433                                    struct hns3_enet_ring *ring)
3434 {
3435         ring->next = group->ring;
3436         group->ring = ring;
3437
3438         group->count++;
3439 }
3440
3441 static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
3442 {
3443         struct pci_dev *pdev = priv->ae_handle->pdev;
3444         struct hns3_enet_tqp_vector *tqp_vector;
3445         int num_vectors = priv->vector_num;
3446         int numa_node;
3447         int vector_i;
3448
3449         numa_node = dev_to_node(&pdev->dev);
3450
3451         for (vector_i = 0; vector_i < num_vectors; vector_i++) {
3452                 tqp_vector = &priv->tqp_vector[vector_i];
3453                 cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
3454                                 &tqp_vector->affinity_mask);
3455         }
3456 }
3457
3458 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
3459 {
3460         struct hnae3_ring_chain_node vector_ring_chain;
3461         struct hnae3_handle *h = priv->ae_handle;
3462         struct hns3_enet_tqp_vector *tqp_vector;
3463         int ret = 0;
3464         int i;
3465
3466         hns3_nic_set_cpumask(priv);
3467
3468         for (i = 0; i < priv->vector_num; i++) {
3469                 tqp_vector = &priv->tqp_vector[i];
3470                 hns3_vector_gl_rl_init_hw(tqp_vector, priv);
3471                 tqp_vector->num_tqps = 0;
3472         }
3473
3474         for (i = 0; i < h->kinfo.num_tqps; i++) {
3475                 u16 vector_i = i % priv->vector_num;
3476                 u16 tqp_num = h->kinfo.num_tqps;
3477
3478                 tqp_vector = &priv->tqp_vector[vector_i];
3479
3480                 hns3_add_ring_to_group(&tqp_vector->tx_group,
3481                                        &priv->ring[i]);
3482
3483                 hns3_add_ring_to_group(&tqp_vector->rx_group,
3484                                        &priv->ring[i + tqp_num]);
3485
3486                 priv->ring[i].tqp_vector = tqp_vector;
3487                 priv->ring[i + tqp_num].tqp_vector = tqp_vector;
3488                 tqp_vector->num_tqps++;
3489         }
3490
3491         for (i = 0; i < priv->vector_num; i++) {
3492                 tqp_vector = &priv->tqp_vector[i];
3493
3494                 tqp_vector->rx_group.total_bytes = 0;
3495                 tqp_vector->rx_group.total_packets = 0;
3496                 tqp_vector->tx_group.total_bytes = 0;
3497                 tqp_vector->tx_group.total_packets = 0;
3498                 tqp_vector->handle = h;
3499
3500                 ret = hns3_get_vector_ring_chain(tqp_vector,
3501                                                  &vector_ring_chain);
3502                 if (ret)
3503                         goto map_ring_fail;
3504
3505                 ret = h->ae_algo->ops->map_ring_to_vector(h,
3506                         tqp_vector->vector_irq, &vector_ring_chain);
3507
3508                 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3509
3510                 if (ret)
3511                         goto map_ring_fail;
3512
3513                 netif_napi_add(priv->netdev, &tqp_vector->napi,
3514                                hns3_nic_common_poll, NAPI_POLL_WEIGHT);
3515         }
3516
3517         return 0;
3518
3519 map_ring_fail:
3520         while (i--)
3521                 netif_napi_del(&priv->tqp_vector[i].napi);
3522
3523         return ret;
3524 }
3525
3526 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
3527 {
3528 #define HNS3_VECTOR_PF_MAX_NUM          64
3529
3530         struct hnae3_handle *h = priv->ae_handle;
3531         struct hns3_enet_tqp_vector *tqp_vector;
3532         struct hnae3_vector_info *vector;
3533         struct pci_dev *pdev = h->pdev;
3534         u16 tqp_num = h->kinfo.num_tqps;
3535         u16 vector_num;
3536         int ret = 0;
3537         u16 i;
3538
3539         /* RSS size, cpu online and vector_num should be the same */
3540         /* Should consider 2p/4p later */
3541         vector_num = min_t(u16, num_online_cpus(), tqp_num);
3542         vector_num = min_t(u16, vector_num, HNS3_VECTOR_PF_MAX_NUM);
3543
3544         vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
3545                               GFP_KERNEL);
3546         if (!vector)
3547                 return -ENOMEM;
3548
3549         /* save the actual available vector number */
3550         vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
3551
3552         priv->vector_num = vector_num;
3553         priv->tqp_vector = (struct hns3_enet_tqp_vector *)
3554                 devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
3555                              GFP_KERNEL);
3556         if (!priv->tqp_vector) {
3557                 ret = -ENOMEM;
3558                 goto out;
3559         }
3560
3561         for (i = 0; i < priv->vector_num; i++) {
3562                 tqp_vector = &priv->tqp_vector[i];
3563                 tqp_vector->idx = i;
3564                 tqp_vector->mask_addr = vector[i].io_addr;
3565                 tqp_vector->vector_irq = vector[i].vector;
3566                 hns3_vector_gl_rl_init(tqp_vector, priv);
3567         }
3568
3569 out:
3570         devm_kfree(&pdev->dev, vector);
3571         return ret;
3572 }
3573
3574 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
3575 {
3576         group->ring = NULL;
3577         group->count = 0;
3578 }
3579
3580 static void hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
3581 {
3582         struct hnae3_ring_chain_node vector_ring_chain;
3583         struct hnae3_handle *h = priv->ae_handle;
3584         struct hns3_enet_tqp_vector *tqp_vector;
3585         int i;
3586
3587         for (i = 0; i < priv->vector_num; i++) {
3588                 tqp_vector = &priv->tqp_vector[i];
3589
3590                 if (!tqp_vector->rx_group.ring && !tqp_vector->tx_group.ring)
3591                         continue;
3592
3593                 hns3_get_vector_ring_chain(tqp_vector, &vector_ring_chain);
3594
3595                 h->ae_algo->ops->unmap_ring_from_vector(h,
3596                         tqp_vector->vector_irq, &vector_ring_chain);
3597
3598                 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3599
3600                 if (tqp_vector->irq_init_flag == HNS3_VECTOR_INITED) {
3601                         irq_set_affinity_hint(tqp_vector->vector_irq, NULL);
3602                         free_irq(tqp_vector->vector_irq, tqp_vector);
3603                         tqp_vector->irq_init_flag = HNS3_VECTOR_NOT_INITED;
3604                 }
3605
3606                 hns3_clear_ring_group(&tqp_vector->rx_group);
3607                 hns3_clear_ring_group(&tqp_vector->tx_group);
3608                 netif_napi_del(&priv->tqp_vector[i].napi);
3609         }
3610 }
3611
3612 static int hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
3613 {
3614         struct hnae3_handle *h = priv->ae_handle;
3615         struct pci_dev *pdev = h->pdev;
3616         int i, ret;
3617
3618         for (i = 0; i < priv->vector_num; i++) {
3619                 struct hns3_enet_tqp_vector *tqp_vector;
3620
3621                 tqp_vector = &priv->tqp_vector[i];
3622                 ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
3623                 if (ret)
3624                         return ret;
3625         }
3626
3627         devm_kfree(&pdev->dev, priv->tqp_vector);
3628         return 0;
3629 }
3630
3631 static void hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
3632                               unsigned int ring_type)
3633 {
3634         int queue_num = priv->ae_handle->kinfo.num_tqps;
3635         struct hns3_enet_ring *ring;
3636         int desc_num;
3637
3638         if (ring_type == HNAE3_RING_TYPE_TX) {
3639                 ring = &priv->ring[q->tqp_index];
3640                 desc_num = priv->ae_handle->kinfo.num_tx_desc;
3641                 ring->queue_index = q->tqp_index;
3642                 ring->io_base = (u8 __iomem *)q->io_base + HNS3_TX_REG_OFFSET;
3643         } else {
3644                 ring = &priv->ring[q->tqp_index + queue_num];
3645                 desc_num = priv->ae_handle->kinfo.num_rx_desc;
3646                 ring->queue_index = q->tqp_index;
3647                 ring->io_base = q->io_base;
3648         }
3649
3650         hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
3651
3652         ring->tqp = q;
3653         ring->desc = NULL;
3654         ring->desc_cb = NULL;
3655         ring->dev = priv->dev;
3656         ring->desc_dma_addr = 0;
3657         ring->buf_size = q->buf_size;
3658         ring->desc_num = desc_num;
3659         ring->next_to_use = 0;
3660         ring->next_to_clean = 0;
3661 }
3662
3663 static void hns3_queue_to_ring(struct hnae3_queue *tqp,
3664                                struct hns3_nic_priv *priv)
3665 {
3666         hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
3667         hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
3668 }
3669
3670 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
3671 {
3672         struct hnae3_handle *h = priv->ae_handle;
3673         struct pci_dev *pdev = h->pdev;
3674         int i;
3675
3676         priv->ring = devm_kzalloc(&pdev->dev,
3677                                   array3_size(h->kinfo.num_tqps,
3678                                               sizeof(*priv->ring), 2),
3679                                   GFP_KERNEL);
3680         if (!priv->ring)
3681                 return -ENOMEM;
3682
3683         for (i = 0; i < h->kinfo.num_tqps; i++)
3684                 hns3_queue_to_ring(h->kinfo.tqp[i], priv);
3685
3686         return 0;
3687 }
3688
3689 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
3690 {
3691         if (!priv->ring)
3692                 return;
3693
3694         devm_kfree(priv->dev, priv->ring);
3695         priv->ring = NULL;
3696 }
3697
3698 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
3699 {
3700         int ret;
3701
3702         if (ring->desc_num <= 0 || ring->buf_size <= 0)
3703                 return -EINVAL;
3704
3705         ring->desc_cb = devm_kcalloc(ring_to_dev(ring), ring->desc_num,
3706                                      sizeof(ring->desc_cb[0]), GFP_KERNEL);
3707         if (!ring->desc_cb) {
3708                 ret = -ENOMEM;
3709                 goto out;
3710         }
3711
3712         ret = hns3_alloc_desc(ring);
3713         if (ret)
3714                 goto out_with_desc_cb;
3715
3716         if (!HNAE3_IS_TX_RING(ring)) {
3717                 ret = hns3_alloc_ring_buffers(ring);
3718                 if (ret)
3719                         goto out_with_desc;
3720         }
3721
3722         return 0;
3723
3724 out_with_desc:
3725         hns3_free_desc(ring);
3726 out_with_desc_cb:
3727         devm_kfree(ring_to_dev(ring), ring->desc_cb);
3728         ring->desc_cb = NULL;
3729 out:
3730         return ret;
3731 }
3732
3733 void hns3_fini_ring(struct hns3_enet_ring *ring)
3734 {
3735         hns3_free_desc(ring);
3736         devm_kfree(ring_to_dev(ring), ring->desc_cb);
3737         ring->desc_cb = NULL;
3738         ring->next_to_clean = 0;
3739         ring->next_to_use = 0;
3740         ring->pending_buf = 0;
3741         if (ring->skb) {
3742                 dev_kfree_skb_any(ring->skb);
3743                 ring->skb = NULL;
3744         }
3745 }
3746
3747 static int hns3_buf_size2type(u32 buf_size)
3748 {
3749         int bd_size_type;
3750
3751         switch (buf_size) {
3752         case 512:
3753                 bd_size_type = HNS3_BD_SIZE_512_TYPE;
3754                 break;
3755         case 1024:
3756                 bd_size_type = HNS3_BD_SIZE_1024_TYPE;
3757                 break;
3758         case 2048:
3759                 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
3760                 break;
3761         case 4096:
3762                 bd_size_type = HNS3_BD_SIZE_4096_TYPE;
3763                 break;
3764         default:
3765                 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
3766         }
3767
3768         return bd_size_type;
3769 }
3770
3771 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
3772 {
3773         dma_addr_t dma = ring->desc_dma_addr;
3774         struct hnae3_queue *q = ring->tqp;
3775
3776         if (!HNAE3_IS_TX_RING(ring)) {
3777                 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG, (u32)dma);
3778                 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
3779                                (u32)((dma >> 31) >> 1));
3780
3781                 hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
3782                                hns3_buf_size2type(ring->buf_size));
3783                 hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
3784                                ring->desc_num / 8 - 1);
3785
3786         } else {
3787                 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
3788                                (u32)dma);
3789                 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
3790                                (u32)((dma >> 31) >> 1));
3791
3792                 hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
3793                                ring->desc_num / 8 - 1);
3794         }
3795 }
3796
3797 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
3798 {
3799         struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
3800         int i;
3801
3802         for (i = 0; i < HNAE3_MAX_TC; i++) {
3803                 struct hnae3_tc_info *tc_info = &kinfo->tc_info[i];
3804                 int j;
3805
3806                 if (!tc_info->enable)
3807                         continue;
3808
3809                 for (j = 0; j < tc_info->tqp_count; j++) {
3810                         struct hnae3_queue *q;
3811
3812                         q = priv->ring[tc_info->tqp_offset + j].tqp;
3813                         hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG,
3814                                        tc_info->tc);
3815                 }
3816         }
3817 }
3818
3819 int hns3_init_all_ring(struct hns3_nic_priv *priv)
3820 {
3821         struct hnae3_handle *h = priv->ae_handle;
3822         int ring_num = h->kinfo.num_tqps * 2;
3823         int i, j;
3824         int ret;
3825
3826         for (i = 0; i < ring_num; i++) {
3827                 ret = hns3_alloc_ring_memory(&priv->ring[i]);
3828                 if (ret) {
3829                         dev_err(priv->dev,
3830                                 "Alloc ring memory fail! ret=%d\n", ret);
3831                         goto out_when_alloc_ring_memory;
3832                 }
3833
3834                 u64_stats_init(&priv->ring[i].syncp);
3835         }
3836
3837         return 0;
3838
3839 out_when_alloc_ring_memory:
3840         for (j = i - 1; j >= 0; j--)
3841                 hns3_fini_ring(&priv->ring[j]);
3842
3843         return -ENOMEM;
3844 }
3845
3846 int hns3_uninit_all_ring(struct hns3_nic_priv *priv)
3847 {
3848         struct hnae3_handle *h = priv->ae_handle;
3849         int i;
3850
3851         for (i = 0; i < h->kinfo.num_tqps; i++) {
3852                 hns3_fini_ring(&priv->ring[i]);
3853                 hns3_fini_ring(&priv->ring[i + h->kinfo.num_tqps]);
3854         }
3855         return 0;
3856 }
3857
3858 /* Set mac addr if it is configured. or leave it to the AE driver */
3859 static int hns3_init_mac_addr(struct net_device *netdev)
3860 {
3861         struct hns3_nic_priv *priv = netdev_priv(netdev);
3862         struct hnae3_handle *h = priv->ae_handle;
3863         u8 mac_addr_temp[ETH_ALEN];
3864         int ret = 0;
3865
3866         if (h->ae_algo->ops->get_mac_addr)
3867                 h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
3868
3869         /* Check if the MAC address is valid, if not get a random one */
3870         if (!is_valid_ether_addr(mac_addr_temp)) {
3871                 eth_hw_addr_random(netdev);
3872                 dev_warn(priv->dev, "using random MAC address %pM\n",
3873                          netdev->dev_addr);
3874         } else {
3875                 ether_addr_copy(netdev->dev_addr, mac_addr_temp);
3876                 ether_addr_copy(netdev->perm_addr, mac_addr_temp);
3877         }
3878
3879         if (h->ae_algo->ops->set_mac_addr)
3880                 ret = h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
3881
3882         return ret;
3883 }
3884
3885 static int hns3_init_phy(struct net_device *netdev)
3886 {
3887         struct hnae3_handle *h = hns3_get_handle(netdev);
3888         int ret = 0;
3889
3890         if (h->ae_algo->ops->mac_connect_phy)
3891                 ret = h->ae_algo->ops->mac_connect_phy(h);
3892
3893         return ret;
3894 }
3895
3896 static void hns3_uninit_phy(struct net_device *netdev)
3897 {
3898         struct hnae3_handle *h = hns3_get_handle(netdev);
3899
3900         if (h->ae_algo->ops->mac_disconnect_phy)
3901                 h->ae_algo->ops->mac_disconnect_phy(h);
3902 }
3903
3904 static int hns3_restore_fd_rules(struct net_device *netdev)
3905 {
3906         struct hnae3_handle *h = hns3_get_handle(netdev);
3907         int ret = 0;
3908
3909         if (h->ae_algo->ops->restore_fd_rules)
3910                 ret = h->ae_algo->ops->restore_fd_rules(h);
3911
3912         return ret;
3913 }
3914
3915 static void hns3_del_all_fd_rules(struct net_device *netdev, bool clear_list)
3916 {
3917         struct hnae3_handle *h = hns3_get_handle(netdev);
3918
3919         if (h->ae_algo->ops->del_all_fd_entries)
3920                 h->ae_algo->ops->del_all_fd_entries(h, clear_list);
3921 }
3922
3923 static int hns3_client_start(struct hnae3_handle *handle)
3924 {
3925         if (!handle->ae_algo->ops->client_start)
3926                 return 0;
3927
3928         return handle->ae_algo->ops->client_start(handle);
3929 }
3930
3931 static void hns3_client_stop(struct hnae3_handle *handle)
3932 {
3933         if (!handle->ae_algo->ops->client_stop)
3934                 return;
3935
3936         handle->ae_algo->ops->client_stop(handle);
3937 }
3938
3939 static void hns3_info_show(struct hns3_nic_priv *priv)
3940 {
3941         struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
3942
3943         dev_info(priv->dev, "MAC address: %pM\n", priv->netdev->dev_addr);
3944         dev_info(priv->dev, "Task queue pairs numbers: %u\n", kinfo->num_tqps);
3945         dev_info(priv->dev, "RSS size: %u\n", kinfo->rss_size);
3946         dev_info(priv->dev, "Allocated RSS size: %u\n", kinfo->req_rss_size);
3947         dev_info(priv->dev, "RX buffer length: %u\n", kinfo->rx_buf_len);
3948         dev_info(priv->dev, "Desc num per TX queue: %u\n", kinfo->num_tx_desc);
3949         dev_info(priv->dev, "Desc num per RX queue: %u\n", kinfo->num_rx_desc);
3950         dev_info(priv->dev, "Total number of enabled TCs: %u\n", kinfo->num_tc);
3951         dev_info(priv->dev, "Max mtu size: %u\n", priv->netdev->max_mtu);
3952 }
3953
3954 static int hns3_client_init(struct hnae3_handle *handle)
3955 {
3956         struct pci_dev *pdev = handle->pdev;
3957         u16 alloc_tqps, max_rss_size;
3958         struct hns3_nic_priv *priv;
3959         struct net_device *netdev;
3960         int ret;
3961
3962         handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
3963                                                     &max_rss_size);
3964         netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
3965         if (!netdev)
3966                 return -ENOMEM;
3967
3968         priv = netdev_priv(netdev);
3969         priv->dev = &pdev->dev;
3970         priv->netdev = netdev;
3971         priv->ae_handle = handle;
3972         priv->tx_timeout_count = 0;
3973         set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
3974
3975         handle->msg_enable = netif_msg_init(debug, DEFAULT_MSG_LEVEL);
3976
3977         handle->kinfo.netdev = netdev;
3978         handle->priv = (void *)priv;
3979
3980         hns3_init_mac_addr(netdev);
3981
3982         hns3_set_default_feature(netdev);
3983
3984         netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
3985         netdev->priv_flags |= IFF_UNICAST_FLT;
3986         netdev->netdev_ops = &hns3_nic_netdev_ops;
3987         SET_NETDEV_DEV(netdev, &pdev->dev);
3988         hns3_ethtool_set_ops(netdev);
3989
3990         /* Carrier off reporting is important to ethtool even BEFORE open */
3991         netif_carrier_off(netdev);
3992
3993         ret = hns3_get_ring_config(priv);
3994         if (ret) {
3995                 ret = -ENOMEM;
3996                 goto out_get_ring_cfg;
3997         }
3998
3999         ret = hns3_nic_alloc_vector_data(priv);
4000         if (ret) {
4001                 ret = -ENOMEM;
4002                 goto out_alloc_vector_data;
4003         }
4004
4005         ret = hns3_nic_init_vector_data(priv);
4006         if (ret) {
4007                 ret = -ENOMEM;
4008                 goto out_init_vector_data;
4009         }
4010
4011         ret = hns3_init_all_ring(priv);
4012         if (ret) {
4013                 ret = -ENOMEM;
4014                 goto out_init_ring;
4015         }
4016
4017         ret = hns3_init_phy(netdev);
4018         if (ret)
4019                 goto out_init_phy;
4020
4021         ret = register_netdev(netdev);
4022         if (ret) {
4023                 dev_err(priv->dev, "probe register netdev fail!\n");
4024                 goto out_reg_netdev_fail;
4025         }
4026
4027         ret = hns3_client_start(handle);
4028         if (ret) {
4029                 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
4030                 goto out_client_start;
4031         }
4032
4033         hns3_dcbnl_setup(handle);
4034
4035         hns3_dbg_init(handle);
4036
4037         /* MTU range: (ETH_MIN_MTU(kernel default) - 9702) */
4038         netdev->max_mtu = HNS3_MAX_MTU;
4039
4040         set_bit(HNS3_NIC_STATE_INITED, &priv->state);
4041
4042         if (netif_msg_drv(handle))
4043                 hns3_info_show(priv);
4044
4045         return ret;
4046
4047 out_client_start:
4048         unregister_netdev(netdev);
4049 out_reg_netdev_fail:
4050         hns3_uninit_phy(netdev);
4051 out_init_phy:
4052         hns3_uninit_all_ring(priv);
4053 out_init_ring:
4054         hns3_nic_uninit_vector_data(priv);
4055 out_init_vector_data:
4056         hns3_nic_dealloc_vector_data(priv);
4057 out_alloc_vector_data:
4058         priv->ring = NULL;
4059 out_get_ring_cfg:
4060         priv->ae_handle = NULL;
4061         free_netdev(netdev);
4062         return ret;
4063 }
4064
4065 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
4066 {
4067         struct net_device *netdev = handle->kinfo.netdev;
4068         struct hns3_nic_priv *priv = netdev_priv(netdev);
4069         int ret;
4070
4071         hns3_remove_hw_addr(netdev);
4072
4073         if (netdev->reg_state != NETREG_UNINITIALIZED)
4074                 unregister_netdev(netdev);
4075
4076         hns3_client_stop(handle);
4077
4078         hns3_uninit_phy(netdev);
4079
4080         if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
4081                 netdev_warn(netdev, "already uninitialized\n");
4082                 goto out_netdev_free;
4083         }
4084
4085         hns3_del_all_fd_rules(netdev, true);
4086
4087         hns3_clear_all_ring(handle, true);
4088
4089         hns3_nic_uninit_vector_data(priv);
4090
4091         ret = hns3_nic_dealloc_vector_data(priv);
4092         if (ret)
4093                 netdev_err(netdev, "dealloc vector error\n");
4094
4095         ret = hns3_uninit_all_ring(priv);
4096         if (ret)
4097                 netdev_err(netdev, "uninit ring error\n");
4098
4099         hns3_put_ring_config(priv);
4100
4101         hns3_dbg_uninit(handle);
4102
4103 out_netdev_free:
4104         free_netdev(netdev);
4105 }
4106
4107 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
4108 {
4109         struct net_device *netdev = handle->kinfo.netdev;
4110
4111         if (!netdev)
4112                 return;
4113
4114         if (linkup) {
4115                 netif_carrier_on(netdev);
4116                 netif_tx_wake_all_queues(netdev);
4117                 if (netif_msg_link(handle))
4118                         netdev_info(netdev, "link up\n");
4119         } else {
4120                 netif_carrier_off(netdev);
4121                 netif_tx_stop_all_queues(netdev);
4122                 if (netif_msg_link(handle))
4123                         netdev_info(netdev, "link down\n");
4124         }
4125 }
4126
4127 static int hns3_client_setup_tc(struct hnae3_handle *handle, u8 tc)
4128 {
4129         struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4130         struct net_device *ndev = kinfo->netdev;
4131
4132         if (tc > HNAE3_MAX_TC)
4133                 return -EINVAL;
4134
4135         if (!ndev)
4136                 return -ENODEV;
4137
4138         return hns3_nic_set_real_num_queue(ndev);
4139 }
4140
4141 static int hns3_recover_hw_addr(struct net_device *ndev)
4142 {
4143         struct netdev_hw_addr_list *list;
4144         struct netdev_hw_addr *ha, *tmp;
4145         int ret = 0;
4146
4147         netif_addr_lock_bh(ndev);
4148         /* go through and sync uc_addr entries to the device */
4149         list = &ndev->uc;
4150         list_for_each_entry_safe(ha, tmp, &list->list, list) {
4151                 ret = hns3_nic_uc_sync(ndev, ha->addr);
4152                 if (ret)
4153                         goto out;
4154         }
4155
4156         /* go through and sync mc_addr entries to the device */
4157         list = &ndev->mc;
4158         list_for_each_entry_safe(ha, tmp, &list->list, list) {
4159                 ret = hns3_nic_mc_sync(ndev, ha->addr);
4160                 if (ret)
4161                         goto out;
4162         }
4163
4164 out:
4165         netif_addr_unlock_bh(ndev);
4166         return ret;
4167 }
4168
4169 static void hns3_remove_hw_addr(struct net_device *netdev)
4170 {
4171         struct netdev_hw_addr_list *list;
4172         struct netdev_hw_addr *ha, *tmp;
4173
4174         hns3_nic_uc_unsync(netdev, netdev->dev_addr);
4175
4176         netif_addr_lock_bh(netdev);
4177         /* go through and unsync uc_addr entries to the device */
4178         list = &netdev->uc;
4179         list_for_each_entry_safe(ha, tmp, &list->list, list)
4180                 hns3_nic_uc_unsync(netdev, ha->addr);
4181
4182         /* go through and unsync mc_addr entries to the device */
4183         list = &netdev->mc;
4184         list_for_each_entry_safe(ha, tmp, &list->list, list)
4185                 if (ha->refcount > 1)
4186                         hns3_nic_mc_unsync(netdev, ha->addr);
4187
4188         netif_addr_unlock_bh(netdev);
4189 }
4190
4191 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
4192 {
4193         while (ring->next_to_clean != ring->next_to_use) {
4194                 ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
4195                 hns3_free_buffer_detach(ring, ring->next_to_clean);
4196                 ring_ptr_move_fw(ring, next_to_clean);
4197         }
4198 }
4199
4200 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
4201 {
4202         struct hns3_desc_cb res_cbs;
4203         int ret;
4204
4205         while (ring->next_to_use != ring->next_to_clean) {
4206                 /* When a buffer is not reused, it's memory has been
4207                  * freed in hns3_handle_rx_bd or will be freed by
4208                  * stack, so we need to replace the buffer here.
4209                  */
4210                 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
4211                         ret = hns3_reserve_buffer_map(ring, &res_cbs);
4212                         if (ret) {
4213                                 u64_stats_update_begin(&ring->syncp);
4214                                 ring->stats.sw_err_cnt++;
4215                                 u64_stats_update_end(&ring->syncp);
4216                                 /* if alloc new buffer fail, exit directly
4217                                  * and reclear in up flow.
4218                                  */
4219                                 netdev_warn(ring_to_netdev(ring),
4220                                             "reserve buffer map failed, ret = %d\n",
4221                                             ret);
4222                                 return ret;
4223                         }
4224                         hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
4225                 }
4226                 ring_ptr_move_fw(ring, next_to_use);
4227         }
4228
4229         /* Free the pending skb in rx ring */
4230         if (ring->skb) {
4231                 dev_kfree_skb_any(ring->skb);
4232                 ring->skb = NULL;
4233                 ring->pending_buf = 0;
4234         }
4235
4236         return 0;
4237 }
4238
4239 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
4240 {
4241         while (ring->next_to_use != ring->next_to_clean) {
4242                 /* When a buffer is not reused, it's memory has been
4243                  * freed in hns3_handle_rx_bd or will be freed by
4244                  * stack, so only need to unmap the buffer here.
4245                  */
4246                 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
4247                         hns3_unmap_buffer(ring,
4248                                           &ring->desc_cb[ring->next_to_use]);
4249                         ring->desc_cb[ring->next_to_use].dma = 0;
4250                 }
4251
4252                 ring_ptr_move_fw(ring, next_to_use);
4253         }
4254 }
4255
4256 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force)
4257 {
4258         struct net_device *ndev = h->kinfo.netdev;
4259         struct hns3_nic_priv *priv = netdev_priv(ndev);
4260         u32 i;
4261
4262         for (i = 0; i < h->kinfo.num_tqps; i++) {
4263                 struct hns3_enet_ring *ring;
4264
4265                 ring = &priv->ring[i];
4266                 hns3_clear_tx_ring(ring);
4267
4268                 ring = &priv->ring[i + h->kinfo.num_tqps];
4269                 /* Continue to clear other rings even if clearing some
4270                  * rings failed.
4271                  */
4272                 if (force)
4273                         hns3_force_clear_rx_ring(ring);
4274                 else
4275                         hns3_clear_rx_ring(ring);
4276         }
4277 }
4278
4279 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
4280 {
4281         struct net_device *ndev = h->kinfo.netdev;
4282         struct hns3_nic_priv *priv = netdev_priv(ndev);
4283         struct hns3_enet_ring *rx_ring;
4284         int i, j;
4285         int ret;
4286
4287         for (i = 0; i < h->kinfo.num_tqps; i++) {
4288                 ret = h->ae_algo->ops->reset_queue(h, i);
4289                 if (ret)
4290                         return ret;
4291
4292                 hns3_init_ring_hw(&priv->ring[i]);
4293
4294                 /* We need to clear tx ring here because self test will
4295                  * use the ring and will not run down before up
4296                  */
4297                 hns3_clear_tx_ring(&priv->ring[i]);
4298                 priv->ring[i].next_to_clean = 0;
4299                 priv->ring[i].next_to_use = 0;
4300
4301                 rx_ring = &priv->ring[i + h->kinfo.num_tqps];
4302                 hns3_init_ring_hw(rx_ring);
4303                 ret = hns3_clear_rx_ring(rx_ring);
4304                 if (ret)
4305                         return ret;
4306
4307                 /* We can not know the hardware head and tail when this
4308                  * function is called in reset flow, so we reuse all desc.
4309                  */
4310                 for (j = 0; j < rx_ring->desc_num; j++)
4311                         hns3_reuse_buffer(rx_ring, j);
4312
4313                 rx_ring->next_to_clean = 0;
4314                 rx_ring->next_to_use = 0;
4315         }
4316
4317         hns3_init_tx_ring_tc(priv);
4318
4319         return 0;
4320 }
4321
4322 static void hns3_store_coal(struct hns3_nic_priv *priv)
4323 {
4324         /* ethtool only support setting and querying one coal
4325          * configuration for now, so save the vector 0' coal
4326          * configuration here in order to restore it.
4327          */
4328         memcpy(&priv->tx_coal, &priv->tqp_vector[0].tx_group.coal,
4329                sizeof(struct hns3_enet_coalesce));
4330         memcpy(&priv->rx_coal, &priv->tqp_vector[0].rx_group.coal,
4331                sizeof(struct hns3_enet_coalesce));
4332 }
4333
4334 static void hns3_restore_coal(struct hns3_nic_priv *priv)
4335 {
4336         u16 vector_num = priv->vector_num;
4337         int i;
4338
4339         for (i = 0; i < vector_num; i++) {
4340                 memcpy(&priv->tqp_vector[i].tx_group.coal, &priv->tx_coal,
4341                        sizeof(struct hns3_enet_coalesce));
4342                 memcpy(&priv->tqp_vector[i].rx_group.coal, &priv->rx_coal,
4343                        sizeof(struct hns3_enet_coalesce));
4344         }
4345 }
4346
4347 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
4348 {
4349         struct hnae3_ae_dev *ae_dev = pci_get_drvdata(handle->pdev);
4350         struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4351         struct net_device *ndev = kinfo->netdev;
4352         struct hns3_nic_priv *priv = netdev_priv(ndev);
4353
4354         if (test_and_set_bit(HNS3_NIC_STATE_RESETTING, &priv->state))
4355                 return 0;
4356
4357         /* it is cumbersome for hardware to pick-and-choose entries for deletion
4358          * from table space. Hence, for function reset software intervention is
4359          * required to delete the entries
4360          */
4361         if (hns3_dev_ongoing_func_reset(ae_dev)) {
4362                 hns3_remove_hw_addr(ndev);
4363                 hns3_del_all_fd_rules(ndev, false);
4364         }
4365
4366         if (!netif_running(ndev))
4367                 return 0;
4368
4369         return hns3_nic_net_stop(ndev);
4370 }
4371
4372 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
4373 {
4374         struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4375         struct hns3_nic_priv *priv = netdev_priv(kinfo->netdev);
4376         int ret = 0;
4377
4378         clear_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4379
4380         if (netif_running(kinfo->netdev)) {
4381                 ret = hns3_nic_net_open(kinfo->netdev);
4382                 if (ret) {
4383                         set_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4384                         netdev_err(kinfo->netdev,
4385                                    "net up fail, ret=%d!\n", ret);
4386                         return ret;
4387                 }
4388         }
4389
4390         return ret;
4391 }
4392
4393 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
4394 {
4395         struct net_device *netdev = handle->kinfo.netdev;
4396         struct hns3_nic_priv *priv = netdev_priv(netdev);
4397         int ret;
4398
4399         /* Carrier off reporting is important to ethtool even BEFORE open */
4400         netif_carrier_off(netdev);
4401
4402         ret = hns3_get_ring_config(priv);
4403         if (ret)
4404                 return ret;
4405
4406         ret = hns3_nic_alloc_vector_data(priv);
4407         if (ret)
4408                 goto err_put_ring;
4409
4410         hns3_restore_coal(priv);
4411
4412         ret = hns3_nic_init_vector_data(priv);
4413         if (ret)
4414                 goto err_dealloc_vector;
4415
4416         ret = hns3_init_all_ring(priv);
4417         if (ret)
4418                 goto err_uninit_vector;
4419
4420         ret = hns3_client_start(handle);
4421         if (ret) {
4422                 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
4423                 goto err_uninit_ring;
4424         }
4425
4426         set_bit(HNS3_NIC_STATE_INITED, &priv->state);
4427
4428         return ret;
4429
4430 err_uninit_ring:
4431         hns3_uninit_all_ring(priv);
4432 err_uninit_vector:
4433         hns3_nic_uninit_vector_data(priv);
4434 err_dealloc_vector:
4435         hns3_nic_dealloc_vector_data(priv);
4436 err_put_ring:
4437         hns3_put_ring_config(priv);
4438
4439         return ret;
4440 }
4441
4442 static int hns3_reset_notify_restore_enet(struct hnae3_handle *handle)
4443 {
4444         struct net_device *netdev = handle->kinfo.netdev;
4445         bool vlan_filter_enable;
4446         int ret;
4447
4448         ret = hns3_init_mac_addr(netdev);
4449         if (ret)
4450                 return ret;
4451
4452         ret = hns3_recover_hw_addr(netdev);
4453         if (ret)
4454                 return ret;
4455
4456         ret = hns3_update_promisc_mode(netdev, handle->netdev_flags);
4457         if (ret)
4458                 return ret;
4459
4460         vlan_filter_enable = netdev->flags & IFF_PROMISC ? false : true;
4461         hns3_enable_vlan_filter(netdev, vlan_filter_enable);
4462
4463         if (handle->ae_algo->ops->restore_vlan_table)
4464                 handle->ae_algo->ops->restore_vlan_table(handle);
4465
4466         return hns3_restore_fd_rules(netdev);
4467 }
4468
4469 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
4470 {
4471         struct net_device *netdev = handle->kinfo.netdev;
4472         struct hns3_nic_priv *priv = netdev_priv(netdev);
4473         int ret;
4474
4475         if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
4476                 netdev_warn(netdev, "already uninitialized\n");
4477                 return 0;
4478         }
4479
4480         hns3_clear_all_ring(handle, true);
4481         hns3_reset_tx_queue(priv->ae_handle);
4482
4483         hns3_nic_uninit_vector_data(priv);
4484
4485         hns3_store_coal(priv);
4486
4487         ret = hns3_nic_dealloc_vector_data(priv);
4488         if (ret)
4489                 netdev_err(netdev, "dealloc vector error\n");
4490
4491         ret = hns3_uninit_all_ring(priv);
4492         if (ret)
4493                 netdev_err(netdev, "uninit ring error\n");
4494
4495         hns3_put_ring_config(priv);
4496
4497         return ret;
4498 }
4499
4500 static int hns3_reset_notify(struct hnae3_handle *handle,
4501                              enum hnae3_reset_notify_type type)
4502 {
4503         int ret = 0;
4504
4505         switch (type) {
4506         case HNAE3_UP_CLIENT:
4507                 ret = hns3_reset_notify_up_enet(handle);
4508                 break;
4509         case HNAE3_DOWN_CLIENT:
4510                 ret = hns3_reset_notify_down_enet(handle);
4511                 break;
4512         case HNAE3_INIT_CLIENT:
4513                 ret = hns3_reset_notify_init_enet(handle);
4514                 break;
4515         case HNAE3_UNINIT_CLIENT:
4516                 ret = hns3_reset_notify_uninit_enet(handle);
4517                 break;
4518         case HNAE3_RESTORE_CLIENT:
4519                 ret = hns3_reset_notify_restore_enet(handle);
4520                 break;
4521         default:
4522                 break;
4523         }
4524
4525         return ret;
4526 }
4527
4528 static int hns3_change_channels(struct hnae3_handle *handle, u32 new_tqp_num,
4529                                 bool rxfh_configured)
4530 {
4531         int ret;
4532
4533         ret = handle->ae_algo->ops->set_channels(handle, new_tqp_num,
4534                                                  rxfh_configured);
4535         if (ret) {
4536                 dev_err(&handle->pdev->dev,
4537                         "Change tqp num(%u) fail.\n", new_tqp_num);
4538                 return ret;
4539         }
4540
4541         ret = hns3_reset_notify(handle, HNAE3_INIT_CLIENT);
4542         if (ret)
4543                 return ret;
4544
4545         ret =  hns3_reset_notify(handle, HNAE3_UP_CLIENT);
4546         if (ret)
4547                 hns3_reset_notify(handle, HNAE3_UNINIT_CLIENT);
4548
4549         return ret;
4550 }
4551
4552 int hns3_set_channels(struct net_device *netdev,
4553                       struct ethtool_channels *ch)
4554 {
4555         struct hnae3_handle *h = hns3_get_handle(netdev);
4556         struct hnae3_knic_private_info *kinfo = &h->kinfo;
4557         bool rxfh_configured = netif_is_rxfh_configured(netdev);
4558         u32 new_tqp_num = ch->combined_count;
4559         u16 org_tqp_num;
4560         int ret;
4561
4562         if (hns3_nic_resetting(netdev))
4563                 return -EBUSY;
4564
4565         if (ch->rx_count || ch->tx_count)
4566                 return -EINVAL;
4567
4568         if (new_tqp_num > hns3_get_max_available_channels(h) ||
4569             new_tqp_num < 1) {
4570                 dev_err(&netdev->dev,
4571                         "Change tqps fail, the tqp range is from 1 to %u",
4572                         hns3_get_max_available_channels(h));
4573                 return -EINVAL;
4574         }
4575
4576         if (kinfo->rss_size == new_tqp_num)
4577                 return 0;
4578
4579         netif_dbg(h, drv, netdev,
4580                   "set channels: tqp_num=%u, rxfh=%d\n",
4581                   new_tqp_num, rxfh_configured);
4582
4583         ret = hns3_reset_notify(h, HNAE3_DOWN_CLIENT);
4584         if (ret)
4585                 return ret;
4586
4587         ret = hns3_reset_notify(h, HNAE3_UNINIT_CLIENT);
4588         if (ret)
4589                 return ret;
4590
4591         org_tqp_num = h->kinfo.num_tqps;
4592         ret = hns3_change_channels(h, new_tqp_num, rxfh_configured);
4593         if (ret) {
4594                 int ret1;
4595
4596                 netdev_warn(netdev,
4597                             "Change channels fail, revert to old value\n");
4598                 ret1 = hns3_change_channels(h, org_tqp_num, rxfh_configured);
4599                 if (ret1) {
4600                         netdev_err(netdev,
4601                                    "revert to old channel fail\n");
4602                         return ret1;
4603                 }
4604
4605                 return ret;
4606         }
4607
4608         return 0;
4609 }
4610
4611 static const struct hns3_hw_error_info hns3_hw_err[] = {
4612         { .type = HNAE3_PPU_POISON_ERROR,
4613           .msg = "PPU poison" },
4614         { .type = HNAE3_CMDQ_ECC_ERROR,
4615           .msg = "IMP CMDQ error" },
4616         { .type = HNAE3_IMP_RD_POISON_ERROR,
4617           .msg = "IMP RD poison" },
4618 };
4619
4620 static void hns3_process_hw_error(struct hnae3_handle *handle,
4621                                   enum hnae3_hw_error_type type)
4622 {
4623         int i;
4624
4625         for (i = 0; i < ARRAY_SIZE(hns3_hw_err); i++) {
4626                 if (hns3_hw_err[i].type == type) {
4627                         dev_err(&handle->pdev->dev, "Detected %s!\n",
4628                                 hns3_hw_err[i].msg);
4629                         break;
4630                 }
4631         }
4632 }
4633
4634 static const struct hnae3_client_ops client_ops = {
4635         .init_instance = hns3_client_init,
4636         .uninit_instance = hns3_client_uninit,
4637         .link_status_change = hns3_link_status_change,
4638         .setup_tc = hns3_client_setup_tc,
4639         .reset_notify = hns3_reset_notify,
4640         .process_hw_error = hns3_process_hw_error,
4641 };
4642
4643 /* hns3_init_module - Driver registration routine
4644  * hns3_init_module is the first routine called when the driver is
4645  * loaded. All it does is register with the PCI subsystem.
4646  */
4647 static int __init hns3_init_module(void)
4648 {
4649         int ret;
4650
4651         pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
4652         pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
4653
4654         client.type = HNAE3_CLIENT_KNIC;
4655         snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH - 1, "%s",
4656                  hns3_driver_name);
4657
4658         client.ops = &client_ops;
4659
4660         INIT_LIST_HEAD(&client.node);
4661
4662         hns3_dbg_register_debugfs(hns3_driver_name);
4663
4664         ret = hnae3_register_client(&client);
4665         if (ret)
4666                 goto err_reg_client;
4667
4668         ret = pci_register_driver(&hns3_driver);
4669         if (ret)
4670                 goto err_reg_driver;
4671
4672         return ret;
4673
4674 err_reg_driver:
4675         hnae3_unregister_client(&client);
4676 err_reg_client:
4677         hns3_dbg_unregister_debugfs();
4678         return ret;
4679 }
4680 module_init(hns3_init_module);
4681
4682 /* hns3_exit_module - Driver exit cleanup routine
4683  * hns3_exit_module is called just before the driver is removed
4684  * from memory.
4685  */
4686 static void __exit hns3_exit_module(void)
4687 {
4688         pci_unregister_driver(&hns3_driver);
4689         hnae3_unregister_client(&client);
4690         hns3_dbg_unregister_debugfs();
4691 }
4692 module_exit(hns3_exit_module);
4693
4694 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
4695 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
4696 MODULE_LICENSE("GPL");
4697 MODULE_ALIAS("pci:hns-nic");
4698 MODULE_VERSION(HNS3_MOD_VERSION);