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 #include <linux/if_vlan.h>
8 #include <linux/ip.h>
9 #include <linux/ipv6.h>
10 #include <linux/module.h>
11 #include <linux/pci.h>
12 #include <linux/aer.h>
13 #include <linux/skbuff.h>
14 #include <linux/sctp.h>
15 #include <linux/vermagic.h>
16 #include <net/gre.h>
17 #include <net/pkt_cls.h>
18 #include <net/vxlan.h>
19 
20 #include "hnae3.h"
21 #include "hns3_enet.h"
22 
23 static void hns3_clear_all_ring(struct hnae3_handle *h);
24 static void hns3_force_clear_all_rx_ring(struct hnae3_handle *h);
25 static void hns3_remove_hw_addr(struct net_device *netdev);
26 
27 static const char hns3_driver_name[] = "hns3";
28 const char hns3_driver_version[] = VERMAGIC_STRING;
29 static const char hns3_driver_string[] =
30 			"Hisilicon Ethernet Network Driver for Hip08 Family";
31 static const char hns3_copyright[] = "Copyright (c) 2017 Huawei Corporation.";
32 static struct hnae3_client client;
33 
34 /* hns3_pci_tbl - PCI Device ID Table
35  *
36  * Last entry must be all 0s
37  *
38  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
39  *   Class, Class Mask, private data (not used) }
40  */
41 static const struct pci_device_id hns3_pci_tbl[] = {
42 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
43 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
44 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA),
45 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
46 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC),
47 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
48 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA),
49 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
50 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC),
51 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
52 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC),
53 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
54 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_VF), 0},
55 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF),
56 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
57 	/* required last entry */
58 	{0, }
59 };
60 MODULE_DEVICE_TABLE(pci, hns3_pci_tbl);
61 
62 static irqreturn_t hns3_irq_handle(int irq, void *vector)
63 {
64 	struct hns3_enet_tqp_vector *tqp_vector = vector;
65 
66 	napi_schedule(&tqp_vector->napi);
67 
68 	return IRQ_HANDLED;
69 }
70 
71 /* This callback function is used to set affinity changes to the irq affinity
72  * masks when the irq_set_affinity_notifier function is used.
73  */
74 static void hns3_nic_irq_affinity_notify(struct irq_affinity_notify *notify,
75 					 const cpumask_t *mask)
76 {
77 	struct hns3_enet_tqp_vector *tqp_vectors =
78 		container_of(notify, struct hns3_enet_tqp_vector,
79 			     affinity_notify);
80 
81 	tqp_vectors->affinity_mask = *mask;
82 }
83 
84 static void hns3_nic_irq_affinity_release(struct kref *ref)
85 {
86 }
87 
88 static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv)
89 {
90 	struct hns3_enet_tqp_vector *tqp_vectors;
91 	unsigned int i;
92 
93 	for (i = 0; i < priv->vector_num; i++) {
94 		tqp_vectors = &priv->tqp_vector[i];
95 
96 		if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED)
97 			continue;
98 
99 		/* clear the affinity notifier and affinity mask */
100 		irq_set_affinity_notifier(tqp_vectors->vector_irq, NULL);
101 		irq_set_affinity_hint(tqp_vectors->vector_irq, NULL);
102 
103 		/* release the irq resource */
104 		free_irq(tqp_vectors->vector_irq, tqp_vectors);
105 		tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED;
106 	}
107 }
108 
109 static int hns3_nic_init_irq(struct hns3_nic_priv *priv)
110 {
111 	struct hns3_enet_tqp_vector *tqp_vectors;
112 	int txrx_int_idx = 0;
113 	int rx_int_idx = 0;
114 	int tx_int_idx = 0;
115 	unsigned int i;
116 	int ret;
117 
118 	for (i = 0; i < priv->vector_num; i++) {
119 		tqp_vectors = &priv->tqp_vector[i];
120 
121 		if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED)
122 			continue;
123 
124 		if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) {
125 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
126 				 "%s-%s-%d", priv->netdev->name, "TxRx",
127 				 txrx_int_idx++);
128 			txrx_int_idx++;
129 		} else if (tqp_vectors->rx_group.ring) {
130 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
131 				 "%s-%s-%d", priv->netdev->name, "Rx",
132 				 rx_int_idx++);
133 		} else if (tqp_vectors->tx_group.ring) {
134 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
135 				 "%s-%s-%d", priv->netdev->name, "Tx",
136 				 tx_int_idx++);
137 		} else {
138 			/* Skip this unused q_vector */
139 			continue;
140 		}
141 
142 		tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0';
143 
144 		ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0,
145 				  tqp_vectors->name,
146 				       tqp_vectors);
147 		if (ret) {
148 			netdev_err(priv->netdev, "request irq(%d) fail\n",
149 				   tqp_vectors->vector_irq);
150 			return ret;
151 		}
152 
153 		tqp_vectors->affinity_notify.notify =
154 					hns3_nic_irq_affinity_notify;
155 		tqp_vectors->affinity_notify.release =
156 					hns3_nic_irq_affinity_release;
157 		irq_set_affinity_notifier(tqp_vectors->vector_irq,
158 					  &tqp_vectors->affinity_notify);
159 		irq_set_affinity_hint(tqp_vectors->vector_irq,
160 				      &tqp_vectors->affinity_mask);
161 
162 		tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED;
163 	}
164 
165 	return 0;
166 }
167 
168 static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector,
169 				 u32 mask_en)
170 {
171 	writel(mask_en, tqp_vector->mask_addr);
172 }
173 
174 static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector)
175 {
176 	napi_enable(&tqp_vector->napi);
177 
178 	/* enable vector */
179 	hns3_mask_vector_irq(tqp_vector, 1);
180 }
181 
182 static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector)
183 {
184 	/* disable vector */
185 	hns3_mask_vector_irq(tqp_vector, 0);
186 
187 	disable_irq(tqp_vector->vector_irq);
188 	napi_disable(&tqp_vector->napi);
189 }
190 
191 void hns3_set_vector_coalesce_rl(struct hns3_enet_tqp_vector *tqp_vector,
192 				 u32 rl_value)
193 {
194 	u32 rl_reg = hns3_rl_usec_to_reg(rl_value);
195 
196 	/* this defines the configuration for RL (Interrupt Rate Limiter).
197 	 * Rl defines rate of interrupts i.e. number of interrupts-per-second
198 	 * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing
199 	 */
200 
201 	if (rl_reg > 0 && !tqp_vector->tx_group.coal.gl_adapt_enable &&
202 	    !tqp_vector->rx_group.coal.gl_adapt_enable)
203 		/* According to the hardware, the range of rl_reg is
204 		 * 0-59 and the unit is 4.
205 		 */
206 		rl_reg |=  HNS3_INT_RL_ENABLE_MASK;
207 
208 	writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
209 }
210 
211 void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
212 				    u32 gl_value)
213 {
214 	u32 rx_gl_reg = hns3_gl_usec_to_reg(gl_value);
215 
216 	writel(rx_gl_reg, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
217 }
218 
219 void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
220 				    u32 gl_value)
221 {
222 	u32 tx_gl_reg = hns3_gl_usec_to_reg(gl_value);
223 
224 	writel(tx_gl_reg, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
225 }
226 
227 static void hns3_vector_gl_rl_init(struct hns3_enet_tqp_vector *tqp_vector,
228 				   struct hns3_nic_priv *priv)
229 {
230 	/* initialize the configuration for interrupt coalescing.
231 	 * 1. GL (Interrupt Gap Limiter)
232 	 * 2. RL (Interrupt Rate Limiter)
233 	 */
234 
235 	/* Default: enable interrupt coalescing self-adaptive and GL */
236 	tqp_vector->tx_group.coal.gl_adapt_enable = 1;
237 	tqp_vector->rx_group.coal.gl_adapt_enable = 1;
238 
239 	tqp_vector->tx_group.coal.int_gl = HNS3_INT_GL_50K;
240 	tqp_vector->rx_group.coal.int_gl = HNS3_INT_GL_50K;
241 
242 	tqp_vector->int_adapt_down = HNS3_INT_ADAPT_DOWN_START;
243 	tqp_vector->rx_group.coal.flow_level = HNS3_FLOW_LOW;
244 	tqp_vector->tx_group.coal.flow_level = HNS3_FLOW_LOW;
245 }
246 
247 static void hns3_vector_gl_rl_init_hw(struct hns3_enet_tqp_vector *tqp_vector,
248 				      struct hns3_nic_priv *priv)
249 {
250 	struct hnae3_handle *h = priv->ae_handle;
251 
252 	hns3_set_vector_coalesce_tx_gl(tqp_vector,
253 				       tqp_vector->tx_group.coal.int_gl);
254 	hns3_set_vector_coalesce_rx_gl(tqp_vector,
255 				       tqp_vector->rx_group.coal.int_gl);
256 	hns3_set_vector_coalesce_rl(tqp_vector, h->kinfo.int_rl_setting);
257 }
258 
259 static int hns3_nic_set_real_num_queue(struct net_device *netdev)
260 {
261 	struct hnae3_handle *h = hns3_get_handle(netdev);
262 	struct hnae3_knic_private_info *kinfo = &h->kinfo;
263 	unsigned int queue_size = kinfo->rss_size * kinfo->num_tc;
264 	int i, ret;
265 
266 	if (kinfo->num_tc <= 1) {
267 		netdev_reset_tc(netdev);
268 	} else {
269 		ret = netdev_set_num_tc(netdev, kinfo->num_tc);
270 		if (ret) {
271 			netdev_err(netdev,
272 				   "netdev_set_num_tc fail, ret=%d!\n", ret);
273 			return ret;
274 		}
275 
276 		for (i = 0; i < HNAE3_MAX_TC; i++) {
277 			if (!kinfo->tc_info[i].enable)
278 				continue;
279 
280 			netdev_set_tc_queue(netdev,
281 					    kinfo->tc_info[i].tc,
282 					    kinfo->tc_info[i].tqp_count,
283 					    kinfo->tc_info[i].tqp_offset);
284 		}
285 	}
286 
287 	ret = netif_set_real_num_tx_queues(netdev, queue_size);
288 	if (ret) {
289 		netdev_err(netdev,
290 			   "netif_set_real_num_tx_queues fail, ret=%d!\n",
291 			   ret);
292 		return ret;
293 	}
294 
295 	ret = netif_set_real_num_rx_queues(netdev, queue_size);
296 	if (ret) {
297 		netdev_err(netdev,
298 			   "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
299 		return ret;
300 	}
301 
302 	return 0;
303 }
304 
305 static u16 hns3_get_max_available_channels(struct hnae3_handle *h)
306 {
307 	u16 alloc_tqps, max_rss_size, rss_size;
308 
309 	h->ae_algo->ops->get_tqps_and_rss_info(h, &alloc_tqps, &max_rss_size);
310 	rss_size = alloc_tqps / h->kinfo.num_tc;
311 
312 	return min_t(u16, rss_size, max_rss_size);
313 }
314 
315 static int hns3_nic_net_up(struct net_device *netdev)
316 {
317 	struct hns3_nic_priv *priv = netdev_priv(netdev);
318 	struct hnae3_handle *h = priv->ae_handle;
319 	int i, j;
320 	int ret;
321 
322 	ret = hns3_nic_reset_all_ring(h);
323 	if (ret)
324 		return ret;
325 
326 	/* get irq resource for all vectors */
327 	ret = hns3_nic_init_irq(priv);
328 	if (ret) {
329 		netdev_err(netdev, "hns init irq failed! ret=%d\n", ret);
330 		return ret;
331 	}
332 
333 	/* enable the vectors */
334 	for (i = 0; i < priv->vector_num; i++)
335 		hns3_vector_enable(&priv->tqp_vector[i]);
336 
337 	/* start the ae_dev */
338 	ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0;
339 	if (ret)
340 		goto out_start_err;
341 
342 	clear_bit(HNS3_NIC_STATE_DOWN, &priv->state);
343 
344 	return 0;
345 
346 out_start_err:
347 	for (j = i - 1; j >= 0; j--)
348 		hns3_vector_disable(&priv->tqp_vector[j]);
349 
350 	hns3_nic_uninit_irq(priv);
351 
352 	return ret;
353 }
354 
355 static int hns3_nic_net_open(struct net_device *netdev)
356 {
357 	struct hns3_nic_priv *priv = netdev_priv(netdev);
358 	struct hnae3_handle *h = hns3_get_handle(netdev);
359 	struct hnae3_knic_private_info *kinfo;
360 	int i, ret;
361 
362 	netif_carrier_off(netdev);
363 
364 	ret = hns3_nic_set_real_num_queue(netdev);
365 	if (ret)
366 		return ret;
367 
368 	ret = hns3_nic_net_up(netdev);
369 	if (ret) {
370 		netdev_err(netdev,
371 			   "hns net up fail, ret=%d!\n", ret);
372 		return ret;
373 	}
374 
375 	kinfo = &h->kinfo;
376 	for (i = 0; i < HNAE3_MAX_USER_PRIO; i++) {
377 		netdev_set_prio_tc_map(netdev, i,
378 				       kinfo->prio_tc[i]);
379 	}
380 
381 	priv->ae_handle->last_reset_time = jiffies;
382 	return 0;
383 }
384 
385 static void hns3_nic_net_down(struct net_device *netdev)
386 {
387 	struct hns3_nic_priv *priv = netdev_priv(netdev);
388 	const struct hnae3_ae_ops *ops;
389 	int i;
390 
391 	if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
392 		return;
393 
394 	/* disable vectors */
395 	for (i = 0; i < priv->vector_num; i++)
396 		hns3_vector_disable(&priv->tqp_vector[i]);
397 
398 	/* stop ae_dev */
399 	ops = priv->ae_handle->ae_algo->ops;
400 	if (ops->stop)
401 		ops->stop(priv->ae_handle);
402 
403 	/* free irq resources */
404 	hns3_nic_uninit_irq(priv);
405 
406 	hns3_clear_all_ring(priv->ae_handle);
407 }
408 
409 static int hns3_nic_net_stop(struct net_device *netdev)
410 {
411 	netif_tx_stop_all_queues(netdev);
412 	netif_carrier_off(netdev);
413 
414 	hns3_nic_net_down(netdev);
415 
416 	return 0;
417 }
418 
419 static int hns3_nic_uc_sync(struct net_device *netdev,
420 			    const unsigned char *addr)
421 {
422 	struct hnae3_handle *h = hns3_get_handle(netdev);
423 
424 	if (h->ae_algo->ops->add_uc_addr)
425 		return h->ae_algo->ops->add_uc_addr(h, addr);
426 
427 	return 0;
428 }
429 
430 static int hns3_nic_uc_unsync(struct net_device *netdev,
431 			      const unsigned char *addr)
432 {
433 	struct hnae3_handle *h = hns3_get_handle(netdev);
434 
435 	if (h->ae_algo->ops->rm_uc_addr)
436 		return h->ae_algo->ops->rm_uc_addr(h, addr);
437 
438 	return 0;
439 }
440 
441 static int hns3_nic_mc_sync(struct net_device *netdev,
442 			    const unsigned char *addr)
443 {
444 	struct hnae3_handle *h = hns3_get_handle(netdev);
445 
446 	if (h->ae_algo->ops->add_mc_addr)
447 		return h->ae_algo->ops->add_mc_addr(h, addr);
448 
449 	return 0;
450 }
451 
452 static int hns3_nic_mc_unsync(struct net_device *netdev,
453 			      const unsigned char *addr)
454 {
455 	struct hnae3_handle *h = hns3_get_handle(netdev);
456 
457 	if (h->ae_algo->ops->rm_mc_addr)
458 		return h->ae_algo->ops->rm_mc_addr(h, addr);
459 
460 	return 0;
461 }
462 
463 static u8 hns3_get_netdev_flags(struct net_device *netdev)
464 {
465 	u8 flags = 0;
466 
467 	if (netdev->flags & IFF_PROMISC) {
468 		flags = HNAE3_USER_UPE | HNAE3_USER_MPE;
469 	} else {
470 		flags |= HNAE3_VLAN_FLTR;
471 		if (netdev->flags & IFF_ALLMULTI)
472 			flags |= HNAE3_USER_MPE;
473 	}
474 
475 	return flags;
476 }
477 
478 static void hns3_nic_set_rx_mode(struct net_device *netdev)
479 {
480 	struct hnae3_handle *h = hns3_get_handle(netdev);
481 	u8 new_flags;
482 	int ret;
483 
484 	new_flags = hns3_get_netdev_flags(netdev);
485 
486 	ret = __dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync);
487 	if (ret) {
488 		netdev_err(netdev, "sync uc address fail\n");
489 		if (ret == -ENOSPC)
490 			new_flags |= HNAE3_OVERFLOW_UPE;
491 	}
492 
493 	if (netdev->flags & IFF_MULTICAST) {
494 		ret = __dev_mc_sync(netdev, hns3_nic_mc_sync,
495 				    hns3_nic_mc_unsync);
496 		if (ret) {
497 			netdev_err(netdev, "sync mc address fail\n");
498 			if (ret == -ENOSPC)
499 				new_flags |= HNAE3_OVERFLOW_MPE;
500 		}
501 	}
502 
503 	hns3_update_promisc_mode(netdev, new_flags);
504 	/* User mode Promisc mode enable and vlan filtering is disabled to
505 	 * let all packets in. MAC-VLAN Table overflow Promisc enabled and
506 	 * vlan fitering is enabled
507 	 */
508 	hns3_enable_vlan_filter(netdev, new_flags & HNAE3_VLAN_FLTR);
509 	h->netdev_flags = new_flags;
510 }
511 
512 int hns3_update_promisc_mode(struct net_device *netdev, u8 promisc_flags)
513 {
514 	struct hns3_nic_priv *priv = netdev_priv(netdev);
515 	struct hnae3_handle *h = priv->ae_handle;
516 
517 	if (h->ae_algo->ops->set_promisc_mode) {
518 		return h->ae_algo->ops->set_promisc_mode(h,
519 						promisc_flags & HNAE3_UPE,
520 						promisc_flags & HNAE3_MPE);
521 	}
522 
523 	return 0;
524 }
525 
526 void hns3_enable_vlan_filter(struct net_device *netdev, bool enable)
527 {
528 	struct hns3_nic_priv *priv = netdev_priv(netdev);
529 	struct hnae3_handle *h = priv->ae_handle;
530 	bool last_state;
531 
532 	if (h->pdev->revision >= 0x21 && h->ae_algo->ops->enable_vlan_filter) {
533 		last_state = h->netdev_flags & HNAE3_VLAN_FLTR ? true : false;
534 		if (enable != last_state) {
535 			netdev_info(netdev,
536 				    "%s vlan filter\n",
537 				    enable ? "enable" : "disable");
538 			h->ae_algo->ops->enable_vlan_filter(h, enable);
539 		}
540 	}
541 }
542 
543 static int hns3_set_tso(struct sk_buff *skb, u32 *paylen,
544 			u16 *mss, u32 *type_cs_vlan_tso)
545 {
546 	u32 l4_offset, hdr_len;
547 	union l3_hdr_info l3;
548 	union l4_hdr_info l4;
549 	u32 l4_paylen;
550 	int ret;
551 
552 	if (!skb_is_gso(skb))
553 		return 0;
554 
555 	ret = skb_cow_head(skb, 0);
556 	if (ret)
557 		return ret;
558 
559 	l3.hdr = skb_network_header(skb);
560 	l4.hdr = skb_transport_header(skb);
561 
562 	/* Software should clear the IPv4's checksum field when tso is
563 	 * needed.
564 	 */
565 	if (l3.v4->version == 4)
566 		l3.v4->check = 0;
567 
568 	/* tunnel packet.*/
569 	if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
570 					 SKB_GSO_GRE_CSUM |
571 					 SKB_GSO_UDP_TUNNEL |
572 					 SKB_GSO_UDP_TUNNEL_CSUM)) {
573 		if ((!(skb_shinfo(skb)->gso_type &
574 		    SKB_GSO_PARTIAL)) &&
575 		    (skb_shinfo(skb)->gso_type &
576 		    SKB_GSO_UDP_TUNNEL_CSUM)) {
577 			/* Software should clear the udp's checksum
578 			 * field when tso is needed.
579 			 */
580 			l4.udp->check = 0;
581 		}
582 		/* reset l3&l4 pointers from outer to inner headers */
583 		l3.hdr = skb_inner_network_header(skb);
584 		l4.hdr = skb_inner_transport_header(skb);
585 
586 		/* Software should clear the IPv4's checksum field when
587 		 * tso is needed.
588 		 */
589 		if (l3.v4->version == 4)
590 			l3.v4->check = 0;
591 	}
592 
593 	/* normal or tunnel packet*/
594 	l4_offset = l4.hdr - skb->data;
595 	hdr_len = (l4.tcp->doff * 4) + l4_offset;
596 
597 	/* remove payload length from inner pseudo checksum when tso*/
598 	l4_paylen = skb->len - l4_offset;
599 	csum_replace_by_diff(&l4.tcp->check,
600 			     (__force __wsum)htonl(l4_paylen));
601 
602 	/* find the txbd field values */
603 	*paylen = skb->len - hdr_len;
604 	hnae3_set_bit(*type_cs_vlan_tso,
605 		      HNS3_TXD_TSO_B, 1);
606 
607 	/* get MSS for TSO */
608 	*mss = skb_shinfo(skb)->gso_size;
609 
610 	return 0;
611 }
612 
613 static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
614 				u8 *il4_proto)
615 {
616 	union {
617 		struct iphdr *v4;
618 		struct ipv6hdr *v6;
619 		unsigned char *hdr;
620 	} l3;
621 	unsigned char *l4_hdr;
622 	unsigned char *exthdr;
623 	u8 l4_proto_tmp;
624 	__be16 frag_off;
625 
626 	/* find outer header point */
627 	l3.hdr = skb_network_header(skb);
628 	l4_hdr = skb_transport_header(skb);
629 
630 	if (skb->protocol == htons(ETH_P_IPV6)) {
631 		exthdr = l3.hdr + sizeof(*l3.v6);
632 		l4_proto_tmp = l3.v6->nexthdr;
633 		if (l4_hdr != exthdr)
634 			ipv6_skip_exthdr(skb, exthdr - skb->data,
635 					 &l4_proto_tmp, &frag_off);
636 	} else if (skb->protocol == htons(ETH_P_IP)) {
637 		l4_proto_tmp = l3.v4->protocol;
638 	} else {
639 		return -EINVAL;
640 	}
641 
642 	*ol4_proto = l4_proto_tmp;
643 
644 	/* tunnel packet */
645 	if (!skb->encapsulation) {
646 		*il4_proto = 0;
647 		return 0;
648 	}
649 
650 	/* find inner header point */
651 	l3.hdr = skb_inner_network_header(skb);
652 	l4_hdr = skb_inner_transport_header(skb);
653 
654 	if (l3.v6->version == 6) {
655 		exthdr = l3.hdr + sizeof(*l3.v6);
656 		l4_proto_tmp = l3.v6->nexthdr;
657 		if (l4_hdr != exthdr)
658 			ipv6_skip_exthdr(skb, exthdr - skb->data,
659 					 &l4_proto_tmp, &frag_off);
660 	} else if (l3.v4->version == 4) {
661 		l4_proto_tmp = l3.v4->protocol;
662 	}
663 
664 	*il4_proto = l4_proto_tmp;
665 
666 	return 0;
667 }
668 
669 static void hns3_set_l2l3l4_len(struct sk_buff *skb, u8 ol4_proto,
670 				u8 il4_proto, u32 *type_cs_vlan_tso,
671 				u32 *ol_type_vlan_len_msec)
672 {
673 	union {
674 		struct iphdr *v4;
675 		struct ipv6hdr *v6;
676 		unsigned char *hdr;
677 	} l3;
678 	union {
679 		struct tcphdr *tcp;
680 		struct udphdr *udp;
681 		struct gre_base_hdr *gre;
682 		unsigned char *hdr;
683 	} l4;
684 	unsigned char *l2_hdr;
685 	u8 l4_proto = ol4_proto;
686 	u32 ol2_len;
687 	u32 ol3_len;
688 	u32 ol4_len;
689 	u32 l2_len;
690 	u32 l3_len;
691 
692 	l3.hdr = skb_network_header(skb);
693 	l4.hdr = skb_transport_header(skb);
694 
695 	/* compute L2 header size for normal packet, defined in 2 Bytes */
696 	l2_len = l3.hdr - skb->data;
697 	hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_M,
698 			HNS3_TXD_L2LEN_S, l2_len >> 1);
699 
700 	/* tunnel packet*/
701 	if (skb->encapsulation) {
702 		/* compute OL2 header size, defined in 2 Bytes */
703 		ol2_len = l2_len;
704 		hnae3_set_field(*ol_type_vlan_len_msec,
705 				HNS3_TXD_L2LEN_M,
706 				HNS3_TXD_L2LEN_S, ol2_len >> 1);
707 
708 		/* compute OL3 header size, defined in 4 Bytes */
709 		ol3_len = l4.hdr - l3.hdr;
710 		hnae3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_M,
711 				HNS3_TXD_L3LEN_S, ol3_len >> 2);
712 
713 		/* MAC in UDP, MAC in GRE (0x6558)*/
714 		if ((ol4_proto == IPPROTO_UDP) || (ol4_proto == IPPROTO_GRE)) {
715 			/* switch MAC header ptr from outer to inner header.*/
716 			l2_hdr = skb_inner_mac_header(skb);
717 
718 			/* compute OL4 header size, defined in 4 Bytes. */
719 			ol4_len = l2_hdr - l4.hdr;
720 			hnae3_set_field(*ol_type_vlan_len_msec,
721 					HNS3_TXD_L4LEN_M, HNS3_TXD_L4LEN_S,
722 					ol4_len >> 2);
723 
724 			/* switch IP header ptr from outer to inner header */
725 			l3.hdr = skb_inner_network_header(skb);
726 
727 			/* compute inner l2 header size, defined in 2 Bytes. */
728 			l2_len = l3.hdr - l2_hdr;
729 			hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_M,
730 					HNS3_TXD_L2LEN_S, l2_len >> 1);
731 		} else {
732 			/* skb packet types not supported by hardware,
733 			 * txbd len fild doesn't be filled.
734 			 */
735 			return;
736 		}
737 
738 		/* switch L4 header pointer from outer to inner */
739 		l4.hdr = skb_inner_transport_header(skb);
740 
741 		l4_proto = il4_proto;
742 	}
743 
744 	/* compute inner(/normal) L3 header size, defined in 4 Bytes */
745 	l3_len = l4.hdr - l3.hdr;
746 	hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_M,
747 			HNS3_TXD_L3LEN_S, l3_len >> 2);
748 
749 	/* compute inner(/normal) L4 header size, defined in 4 Bytes */
750 	switch (l4_proto) {
751 	case IPPROTO_TCP:
752 		hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_M,
753 				HNS3_TXD_L4LEN_S, l4.tcp->doff);
754 		break;
755 	case IPPROTO_SCTP:
756 		hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_M,
757 				HNS3_TXD_L4LEN_S,
758 				(sizeof(struct sctphdr) >> 2));
759 		break;
760 	case IPPROTO_UDP:
761 		hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_M,
762 				HNS3_TXD_L4LEN_S,
763 				(sizeof(struct udphdr) >> 2));
764 		break;
765 	default:
766 		/* skb packet types not supported by hardware,
767 		 * txbd len fild doesn't be filled.
768 		 */
769 		return;
770 	}
771 }
772 
773 /* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
774  * and it is udp packet, which has a dest port as the IANA assigned.
775  * the hardware is expected to do the checksum offload, but the
776  * hardware will not do the checksum offload when udp dest port is
777  * 4789.
778  */
779 static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
780 {
781 #define IANA_VXLAN_PORT	4789
782 	union {
783 		struct tcphdr *tcp;
784 		struct udphdr *udp;
785 		struct gre_base_hdr *gre;
786 		unsigned char *hdr;
787 	} l4;
788 
789 	l4.hdr = skb_transport_header(skb);
790 
791 	if (!(!skb->encapsulation && l4.udp->dest == htons(IANA_VXLAN_PORT)))
792 		return false;
793 
794 	skb_checksum_help(skb);
795 
796 	return true;
797 }
798 
799 static int hns3_set_l3l4_type_csum(struct sk_buff *skb, u8 ol4_proto,
800 				   u8 il4_proto, u32 *type_cs_vlan_tso,
801 				   u32 *ol_type_vlan_len_msec)
802 {
803 	union {
804 		struct iphdr *v4;
805 		struct ipv6hdr *v6;
806 		unsigned char *hdr;
807 	} l3;
808 	u32 l4_proto = ol4_proto;
809 
810 	l3.hdr = skb_network_header(skb);
811 
812 	/* define OL3 type and tunnel type(OL4).*/
813 	if (skb->encapsulation) {
814 		/* define outer network header type.*/
815 		if (skb->protocol == htons(ETH_P_IP)) {
816 			if (skb_is_gso(skb))
817 				hnae3_set_field(*ol_type_vlan_len_msec,
818 						HNS3_TXD_OL3T_M,
819 						HNS3_TXD_OL3T_S,
820 						HNS3_OL3T_IPV4_CSUM);
821 			else
822 				hnae3_set_field(*ol_type_vlan_len_msec,
823 						HNS3_TXD_OL3T_M,
824 						HNS3_TXD_OL3T_S,
825 						HNS3_OL3T_IPV4_NO_CSUM);
826 
827 		} else if (skb->protocol == htons(ETH_P_IPV6)) {
828 			hnae3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_M,
829 					HNS3_TXD_OL3T_S, HNS3_OL3T_IPV6);
830 		}
831 
832 		/* define tunnel type(OL4).*/
833 		switch (l4_proto) {
834 		case IPPROTO_UDP:
835 			hnae3_set_field(*ol_type_vlan_len_msec,
836 					HNS3_TXD_TUNTYPE_M,
837 					HNS3_TXD_TUNTYPE_S,
838 					HNS3_TUN_MAC_IN_UDP);
839 			break;
840 		case IPPROTO_GRE:
841 			hnae3_set_field(*ol_type_vlan_len_msec,
842 					HNS3_TXD_TUNTYPE_M,
843 					HNS3_TXD_TUNTYPE_S,
844 					HNS3_TUN_NVGRE);
845 			break;
846 		default:
847 			/* drop the skb tunnel packet if hardware don't support,
848 			 * because hardware can't calculate csum when TSO.
849 			 */
850 			if (skb_is_gso(skb))
851 				return -EDOM;
852 
853 			/* the stack computes the IP header already,
854 			 * driver calculate l4 checksum when not TSO.
855 			 */
856 			skb_checksum_help(skb);
857 			return 0;
858 		}
859 
860 		l3.hdr = skb_inner_network_header(skb);
861 		l4_proto = il4_proto;
862 	}
863 
864 	if (l3.v4->version == 4) {
865 		hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_M,
866 				HNS3_TXD_L3T_S, HNS3_L3T_IPV4);
867 
868 		/* the stack computes the IP header already, the only time we
869 		 * need the hardware to recompute it is in the case of TSO.
870 		 */
871 		if (skb_is_gso(skb))
872 			hnae3_set_bit(*type_cs_vlan_tso, HNS3_TXD_L3CS_B, 1);
873 	} else if (l3.v6->version == 6) {
874 		hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_M,
875 				HNS3_TXD_L3T_S, HNS3_L3T_IPV6);
876 	}
877 
878 	switch (l4_proto) {
879 	case IPPROTO_TCP:
880 		hnae3_set_bit(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
881 		hnae3_set_field(*type_cs_vlan_tso,
882 				HNS3_TXD_L4T_M,
883 				HNS3_TXD_L4T_S,
884 				HNS3_L4T_TCP);
885 		break;
886 	case IPPROTO_UDP:
887 		if (hns3_tunnel_csum_bug(skb))
888 			break;
889 
890 		hnae3_set_bit(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
891 		hnae3_set_field(*type_cs_vlan_tso,
892 				HNS3_TXD_L4T_M,
893 				HNS3_TXD_L4T_S,
894 				HNS3_L4T_UDP);
895 		break;
896 	case IPPROTO_SCTP:
897 		hnae3_set_bit(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
898 		hnae3_set_field(*type_cs_vlan_tso,
899 				HNS3_TXD_L4T_M,
900 				HNS3_TXD_L4T_S,
901 				HNS3_L4T_SCTP);
902 		break;
903 	default:
904 		/* drop the skb tunnel packet if hardware don't support,
905 		 * because hardware can't calculate csum when TSO.
906 		 */
907 		if (skb_is_gso(skb))
908 			return -EDOM;
909 
910 		/* the stack computes the IP header already,
911 		 * driver calculate l4 checksum when not TSO.
912 		 */
913 		skb_checksum_help(skb);
914 		return 0;
915 	}
916 
917 	return 0;
918 }
919 
920 static void hns3_set_txbd_baseinfo(u16 *bdtp_fe_sc_vld_ra_ri, int frag_end)
921 {
922 	/* Config bd buffer end */
923 	hnae3_set_field(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_BDTYPE_M,
924 			HNS3_TXD_BDTYPE_S, 0);
925 	hnae3_set_bit(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_FE_B, !!frag_end);
926 	hnae3_set_bit(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_VLD_B, 1);
927 	hnae3_set_field(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_SC_M, HNS3_TXD_SC_S, 0);
928 }
929 
930 static int hns3_fill_desc_vtags(struct sk_buff *skb,
931 				struct hns3_enet_ring *tx_ring,
932 				u32 *inner_vlan_flag,
933 				u32 *out_vlan_flag,
934 				u16 *inner_vtag,
935 				u16 *out_vtag)
936 {
937 #define HNS3_TX_VLAN_PRIO_SHIFT 13
938 
939 	if (skb->protocol == htons(ETH_P_8021Q) &&
940 	    !(tx_ring->tqp->handle->kinfo.netdev->features &
941 	    NETIF_F_HW_VLAN_CTAG_TX)) {
942 		/* When HW VLAN acceleration is turned off, and the stack
943 		 * sets the protocol to 802.1q, the driver just need to
944 		 * set the protocol to the encapsulated ethertype.
945 		 */
946 		skb->protocol = vlan_get_protocol(skb);
947 		return 0;
948 	}
949 
950 	if (skb_vlan_tag_present(skb)) {
951 		u16 vlan_tag;
952 
953 		vlan_tag = skb_vlan_tag_get(skb);
954 		vlan_tag |= (skb->priority & 0x7) << HNS3_TX_VLAN_PRIO_SHIFT;
955 
956 		/* Based on hw strategy, use out_vtag in two layer tag case,
957 		 * and use inner_vtag in one tag case.
958 		 */
959 		if (skb->protocol == htons(ETH_P_8021Q)) {
960 			hnae3_set_bit(*out_vlan_flag, HNS3_TXD_OVLAN_B, 1);
961 			*out_vtag = vlan_tag;
962 		} else {
963 			hnae3_set_bit(*inner_vlan_flag, HNS3_TXD_VLAN_B, 1);
964 			*inner_vtag = vlan_tag;
965 		}
966 	} else if (skb->protocol == htons(ETH_P_8021Q)) {
967 		struct vlan_ethhdr *vhdr;
968 		int rc;
969 
970 		rc = skb_cow_head(skb, 0);
971 		if (rc < 0)
972 			return rc;
973 		vhdr = (struct vlan_ethhdr *)skb->data;
974 		vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority & 0x7)
975 					<< HNS3_TX_VLAN_PRIO_SHIFT);
976 	}
977 
978 	skb->protocol = vlan_get_protocol(skb);
979 	return 0;
980 }
981 
982 static int hns3_fill_desc(struct hns3_enet_ring *ring, void *priv,
983 			  int size, int frag_end, enum hns_desc_type type)
984 {
985 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
986 	struct hns3_desc *desc = &ring->desc[ring->next_to_use];
987 	struct device *dev = ring_to_dev(ring);
988 	u32 ol_type_vlan_len_msec = 0;
989 	u16 bdtp_fe_sc_vld_ra_ri = 0;
990 	struct skb_frag_struct *frag;
991 	unsigned int frag_buf_num;
992 	u32 type_cs_vlan_tso = 0;
993 	struct sk_buff *skb;
994 	u16 inner_vtag = 0;
995 	u16 out_vtag = 0;
996 	unsigned int k;
997 	int sizeoflast;
998 	u32 paylen = 0;
999 	dma_addr_t dma;
1000 	u16 mss = 0;
1001 	u8 ol4_proto;
1002 	u8 il4_proto;
1003 	int ret;
1004 
1005 	if (type == DESC_TYPE_SKB) {
1006 		skb = (struct sk_buff *)priv;
1007 		paylen = skb->len;
1008 
1009 		ret = hns3_fill_desc_vtags(skb, ring, &type_cs_vlan_tso,
1010 					   &ol_type_vlan_len_msec,
1011 					   &inner_vtag, &out_vtag);
1012 		if (unlikely(ret))
1013 			return ret;
1014 
1015 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1016 			skb_reset_mac_len(skb);
1017 
1018 			ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
1019 			if (ret)
1020 				return ret;
1021 			hns3_set_l2l3l4_len(skb, ol4_proto, il4_proto,
1022 					    &type_cs_vlan_tso,
1023 					    &ol_type_vlan_len_msec);
1024 			ret = hns3_set_l3l4_type_csum(skb, ol4_proto, il4_proto,
1025 						      &type_cs_vlan_tso,
1026 						      &ol_type_vlan_len_msec);
1027 			if (ret)
1028 				return ret;
1029 
1030 			ret = hns3_set_tso(skb, &paylen, &mss,
1031 					   &type_cs_vlan_tso);
1032 			if (ret)
1033 				return ret;
1034 		}
1035 
1036 		/* Set txbd */
1037 		desc->tx.ol_type_vlan_len_msec =
1038 			cpu_to_le32(ol_type_vlan_len_msec);
1039 		desc->tx.type_cs_vlan_tso_len =
1040 			cpu_to_le32(type_cs_vlan_tso);
1041 		desc->tx.paylen = cpu_to_le32(paylen);
1042 		desc->tx.mss = cpu_to_le16(mss);
1043 		desc->tx.vlan_tag = cpu_to_le16(inner_vtag);
1044 		desc->tx.outer_vlan_tag = cpu_to_le16(out_vtag);
1045 
1046 		dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1047 	} else {
1048 		frag = (struct skb_frag_struct *)priv;
1049 		dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1050 	}
1051 
1052 	if (dma_mapping_error(ring->dev, dma)) {
1053 		ring->stats.sw_err_cnt++;
1054 		return -ENOMEM;
1055 	}
1056 
1057 	desc_cb->length = size;
1058 
1059 	frag_buf_num = (size + HNS3_MAX_BD_SIZE - 1) / HNS3_MAX_BD_SIZE;
1060 	sizeoflast = size % HNS3_MAX_BD_SIZE;
1061 	sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1062 
1063 	/* When frag size is bigger than hardware limit, split this frag */
1064 	for (k = 0; k < frag_buf_num; k++) {
1065 		/* The txbd's baseinfo of DESC_TYPE_PAGE & DESC_TYPE_SKB */
1066 		desc_cb->priv = priv;
1067 		desc_cb->dma = dma + HNS3_MAX_BD_SIZE * k;
1068 		desc_cb->type = (type == DESC_TYPE_SKB && !k) ?
1069 					DESC_TYPE_SKB : DESC_TYPE_PAGE;
1070 
1071 		/* now, fill the descriptor */
1072 		desc->addr = cpu_to_le64(dma + HNS3_MAX_BD_SIZE * k);
1073 		desc->tx.send_size = cpu_to_le16((k == frag_buf_num - 1) ?
1074 				(u16)sizeoflast : (u16)HNS3_MAX_BD_SIZE);
1075 		hns3_set_txbd_baseinfo(&bdtp_fe_sc_vld_ra_ri,
1076 				       frag_end && (k == frag_buf_num - 1) ?
1077 						1 : 0);
1078 		desc->tx.bdtp_fe_sc_vld_ra_ri =
1079 				cpu_to_le16(bdtp_fe_sc_vld_ra_ri);
1080 
1081 		/* move ring pointer to next.*/
1082 		ring_ptr_move_fw(ring, next_to_use);
1083 
1084 		desc_cb = &ring->desc_cb[ring->next_to_use];
1085 		desc = &ring->desc[ring->next_to_use];
1086 	}
1087 
1088 	return 0;
1089 }
1090 
1091 static int hns3_nic_maybe_stop_tso(struct sk_buff **out_skb, int *bnum,
1092 				   struct hns3_enet_ring *ring)
1093 {
1094 	struct sk_buff *skb = *out_skb;
1095 	struct skb_frag_struct *frag;
1096 	int bdnum_for_frag;
1097 	int frag_num;
1098 	int buf_num;
1099 	int size;
1100 	int i;
1101 
1102 	size = skb_headlen(skb);
1103 	buf_num = (size + HNS3_MAX_BD_SIZE - 1) / HNS3_MAX_BD_SIZE;
1104 
1105 	frag_num = skb_shinfo(skb)->nr_frags;
1106 	for (i = 0; i < frag_num; i++) {
1107 		frag = &skb_shinfo(skb)->frags[i];
1108 		size = skb_frag_size(frag);
1109 		bdnum_for_frag =
1110 			(size + HNS3_MAX_BD_SIZE - 1) / HNS3_MAX_BD_SIZE;
1111 		if (bdnum_for_frag > HNS3_MAX_BD_PER_FRAG)
1112 			return -ENOMEM;
1113 
1114 		buf_num += bdnum_for_frag;
1115 	}
1116 
1117 	if (buf_num > ring_space(ring))
1118 		return -EBUSY;
1119 
1120 	*bnum = buf_num;
1121 	return 0;
1122 }
1123 
1124 static int hns3_nic_maybe_stop_tx(struct sk_buff **out_skb, int *bnum,
1125 				  struct hns3_enet_ring *ring)
1126 {
1127 	struct sk_buff *skb = *out_skb;
1128 	int buf_num;
1129 
1130 	/* No. of segments (plus a header) */
1131 	buf_num = skb_shinfo(skb)->nr_frags + 1;
1132 
1133 	if (unlikely(ring_space(ring) < buf_num))
1134 		return -EBUSY;
1135 
1136 	*bnum = buf_num;
1137 
1138 	return 0;
1139 }
1140 
1141 static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
1142 {
1143 	struct device *dev = ring_to_dev(ring);
1144 	unsigned int i;
1145 
1146 	for (i = 0; i < ring->desc_num; i++) {
1147 		/* check if this is where we started */
1148 		if (ring->next_to_use == next_to_use_orig)
1149 			break;
1150 
1151 		/* unmap the descriptor dma address */
1152 		if (ring->desc_cb[ring->next_to_use].type == DESC_TYPE_SKB)
1153 			dma_unmap_single(dev,
1154 					 ring->desc_cb[ring->next_to_use].dma,
1155 					ring->desc_cb[ring->next_to_use].length,
1156 					DMA_TO_DEVICE);
1157 		else if (ring->desc_cb[ring->next_to_use].length)
1158 			dma_unmap_page(dev,
1159 				       ring->desc_cb[ring->next_to_use].dma,
1160 				       ring->desc_cb[ring->next_to_use].length,
1161 				       DMA_TO_DEVICE);
1162 
1163 		ring->desc_cb[ring->next_to_use].length = 0;
1164 
1165 		/* rollback one */
1166 		ring_ptr_move_bw(ring, next_to_use);
1167 	}
1168 }
1169 
1170 netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
1171 {
1172 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1173 	struct hns3_nic_ring_data *ring_data =
1174 		&tx_ring_data(priv, skb->queue_mapping);
1175 	struct hns3_enet_ring *ring = ring_data->ring;
1176 	struct netdev_queue *dev_queue;
1177 	struct skb_frag_struct *frag;
1178 	int next_to_use_head;
1179 	int next_to_use_frag;
1180 	int buf_num;
1181 	int seg_num;
1182 	int size;
1183 	int ret;
1184 	int i;
1185 
1186 	/* Prefetch the data used later */
1187 	prefetch(skb->data);
1188 
1189 	switch (priv->ops.maybe_stop_tx(&skb, &buf_num, ring)) {
1190 	case -EBUSY:
1191 		u64_stats_update_begin(&ring->syncp);
1192 		ring->stats.tx_busy++;
1193 		u64_stats_update_end(&ring->syncp);
1194 
1195 		goto out_net_tx_busy;
1196 	case -ENOMEM:
1197 		u64_stats_update_begin(&ring->syncp);
1198 		ring->stats.sw_err_cnt++;
1199 		u64_stats_update_end(&ring->syncp);
1200 		netdev_err(netdev, "no memory to xmit!\n");
1201 
1202 		goto out_err_tx_ok;
1203 	default:
1204 		break;
1205 	}
1206 
1207 	/* No. of segments (plus a header) */
1208 	seg_num = skb_shinfo(skb)->nr_frags + 1;
1209 	/* Fill the first part */
1210 	size = skb_headlen(skb);
1211 
1212 	next_to_use_head = ring->next_to_use;
1213 
1214 	ret = priv->ops.fill_desc(ring, skb, size, seg_num == 1 ? 1 : 0,
1215 				  DESC_TYPE_SKB);
1216 	if (ret)
1217 		goto head_fill_err;
1218 
1219 	next_to_use_frag = ring->next_to_use;
1220 	/* Fill the fragments */
1221 	for (i = 1; i < seg_num; i++) {
1222 		frag = &skb_shinfo(skb)->frags[i - 1];
1223 		size = skb_frag_size(frag);
1224 
1225 		ret = priv->ops.fill_desc(ring, frag, size,
1226 					  seg_num - 1 == i ? 1 : 0,
1227 					  DESC_TYPE_PAGE);
1228 
1229 		if (ret)
1230 			goto frag_fill_err;
1231 	}
1232 
1233 	/* Complete translate all packets */
1234 	dev_queue = netdev_get_tx_queue(netdev, ring_data->queue_index);
1235 	netdev_tx_sent_queue(dev_queue, skb->len);
1236 
1237 	wmb(); /* Commit all data before submit */
1238 
1239 	hnae3_queue_xmit(ring->tqp, buf_num);
1240 
1241 	return NETDEV_TX_OK;
1242 
1243 frag_fill_err:
1244 	hns3_clear_desc(ring, next_to_use_frag);
1245 
1246 head_fill_err:
1247 	hns3_clear_desc(ring, next_to_use_head);
1248 
1249 out_err_tx_ok:
1250 	dev_kfree_skb_any(skb);
1251 	return NETDEV_TX_OK;
1252 
1253 out_net_tx_busy:
1254 	netif_stop_subqueue(netdev, ring_data->queue_index);
1255 	smp_mb(); /* Commit all data before submit */
1256 
1257 	return NETDEV_TX_BUSY;
1258 }
1259 
1260 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
1261 {
1262 	struct hnae3_handle *h = hns3_get_handle(netdev);
1263 	struct sockaddr *mac_addr = p;
1264 	int ret;
1265 
1266 	if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1267 		return -EADDRNOTAVAIL;
1268 
1269 	if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
1270 		netdev_info(netdev, "already using mac address %pM\n",
1271 			    mac_addr->sa_data);
1272 		return 0;
1273 	}
1274 
1275 	ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
1276 	if (ret) {
1277 		netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
1278 		return ret;
1279 	}
1280 
1281 	ether_addr_copy(netdev->dev_addr, mac_addr->sa_data);
1282 
1283 	return 0;
1284 }
1285 
1286 static int hns3_nic_do_ioctl(struct net_device *netdev,
1287 			     struct ifreq *ifr, int cmd)
1288 {
1289 	struct hnae3_handle *h = hns3_get_handle(netdev);
1290 
1291 	if (!netif_running(netdev))
1292 		return -EINVAL;
1293 
1294 	if (!h->ae_algo->ops->do_ioctl)
1295 		return -EOPNOTSUPP;
1296 
1297 	return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
1298 }
1299 
1300 static int hns3_nic_set_features(struct net_device *netdev,
1301 				 netdev_features_t features)
1302 {
1303 	netdev_features_t changed = netdev->features ^ features;
1304 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1305 	struct hnae3_handle *h = priv->ae_handle;
1306 	int ret;
1307 
1308 	if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) {
1309 		if (features & (NETIF_F_TSO | NETIF_F_TSO6))
1310 			priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tso;
1311 		else
1312 			priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tx;
1313 	}
1314 
1315 	if ((changed & NETIF_F_HW_VLAN_CTAG_FILTER) &&
1316 	    h->ae_algo->ops->enable_vlan_filter) {
1317 		if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
1318 			h->ae_algo->ops->enable_vlan_filter(h, true);
1319 		else
1320 			h->ae_algo->ops->enable_vlan_filter(h, false);
1321 	}
1322 
1323 	if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
1324 	    h->ae_algo->ops->enable_hw_strip_rxvtag) {
1325 		if (features & NETIF_F_HW_VLAN_CTAG_RX)
1326 			ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, true);
1327 		else
1328 			ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, false);
1329 
1330 		if (ret)
1331 			return ret;
1332 	}
1333 
1334 	if ((changed & NETIF_F_NTUPLE) && h->ae_algo->ops->enable_fd) {
1335 		if (features & NETIF_F_NTUPLE)
1336 			h->ae_algo->ops->enable_fd(h, true);
1337 		else
1338 			h->ae_algo->ops->enable_fd(h, false);
1339 	}
1340 
1341 	netdev->features = features;
1342 	return 0;
1343 }
1344 
1345 static void hns3_nic_get_stats64(struct net_device *netdev,
1346 				 struct rtnl_link_stats64 *stats)
1347 {
1348 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1349 	int queue_num = priv->ae_handle->kinfo.num_tqps;
1350 	struct hnae3_handle *handle = priv->ae_handle;
1351 	struct hns3_enet_ring *ring;
1352 	unsigned int start;
1353 	unsigned int idx;
1354 	u64 tx_bytes = 0;
1355 	u64 rx_bytes = 0;
1356 	u64 tx_pkts = 0;
1357 	u64 rx_pkts = 0;
1358 	u64 tx_drop = 0;
1359 	u64 rx_drop = 0;
1360 
1361 	if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
1362 		return;
1363 
1364 	handle->ae_algo->ops->update_stats(handle, &netdev->stats);
1365 
1366 	for (idx = 0; idx < queue_num; idx++) {
1367 		/* fetch the tx stats */
1368 		ring = priv->ring_data[idx].ring;
1369 		do {
1370 			start = u64_stats_fetch_begin_irq(&ring->syncp);
1371 			tx_bytes += ring->stats.tx_bytes;
1372 			tx_pkts += ring->stats.tx_pkts;
1373 			tx_drop += ring->stats.tx_busy;
1374 			tx_drop += ring->stats.sw_err_cnt;
1375 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1376 
1377 		/* fetch the rx stats */
1378 		ring = priv->ring_data[idx + queue_num].ring;
1379 		do {
1380 			start = u64_stats_fetch_begin_irq(&ring->syncp);
1381 			rx_bytes += ring->stats.rx_bytes;
1382 			rx_pkts += ring->stats.rx_pkts;
1383 			rx_drop += ring->stats.non_vld_descs;
1384 			rx_drop += ring->stats.err_pkt_len;
1385 			rx_drop += ring->stats.l2_err;
1386 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1387 	}
1388 
1389 	stats->tx_bytes = tx_bytes;
1390 	stats->tx_packets = tx_pkts;
1391 	stats->rx_bytes = rx_bytes;
1392 	stats->rx_packets = rx_pkts;
1393 
1394 	stats->rx_errors = netdev->stats.rx_errors;
1395 	stats->multicast = netdev->stats.multicast;
1396 	stats->rx_length_errors = netdev->stats.rx_length_errors;
1397 	stats->rx_crc_errors = netdev->stats.rx_crc_errors;
1398 	stats->rx_missed_errors = netdev->stats.rx_missed_errors;
1399 
1400 	stats->tx_errors = netdev->stats.tx_errors;
1401 	stats->rx_dropped = rx_drop + netdev->stats.rx_dropped;
1402 	stats->tx_dropped = tx_drop + netdev->stats.tx_dropped;
1403 	stats->collisions = netdev->stats.collisions;
1404 	stats->rx_over_errors = netdev->stats.rx_over_errors;
1405 	stats->rx_frame_errors = netdev->stats.rx_frame_errors;
1406 	stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
1407 	stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
1408 	stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
1409 	stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
1410 	stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
1411 	stats->tx_window_errors = netdev->stats.tx_window_errors;
1412 	stats->rx_compressed = netdev->stats.rx_compressed;
1413 	stats->tx_compressed = netdev->stats.tx_compressed;
1414 }
1415 
1416 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
1417 {
1418 	struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
1419 	struct hnae3_handle *h = hns3_get_handle(netdev);
1420 	struct hnae3_knic_private_info *kinfo = &h->kinfo;
1421 	u8 *prio_tc = mqprio_qopt->qopt.prio_tc_map;
1422 	u8 tc = mqprio_qopt->qopt.num_tc;
1423 	u16 mode = mqprio_qopt->mode;
1424 	u8 hw = mqprio_qopt->qopt.hw;
1425 	bool if_running;
1426 	int ret;
1427 
1428 	if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
1429 	       mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
1430 		return -EOPNOTSUPP;
1431 
1432 	if (tc > HNAE3_MAX_TC)
1433 		return -EINVAL;
1434 
1435 	if (!netdev)
1436 		return -EINVAL;
1437 
1438 	if_running = netif_running(netdev);
1439 	if (if_running) {
1440 		hns3_nic_net_stop(netdev);
1441 		msleep(100);
1442 	}
1443 
1444 	ret = (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
1445 		kinfo->dcb_ops->setup_tc(h, tc, prio_tc) : -EOPNOTSUPP;
1446 	if (ret)
1447 		goto out;
1448 
1449 	ret = hns3_nic_set_real_num_queue(netdev);
1450 
1451 out:
1452 	if (if_running)
1453 		hns3_nic_net_open(netdev);
1454 
1455 	return ret;
1456 }
1457 
1458 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
1459 			     void *type_data)
1460 {
1461 	if (type != TC_SETUP_QDISC_MQPRIO)
1462 		return -EOPNOTSUPP;
1463 
1464 	return hns3_setup_tc(dev, type_data);
1465 }
1466 
1467 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
1468 				__be16 proto, u16 vid)
1469 {
1470 	struct hnae3_handle *h = hns3_get_handle(netdev);
1471 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1472 	int ret = -EIO;
1473 
1474 	if (h->ae_algo->ops->set_vlan_filter)
1475 		ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
1476 
1477 	if (!ret)
1478 		set_bit(vid, priv->active_vlans);
1479 
1480 	return ret;
1481 }
1482 
1483 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
1484 				 __be16 proto, u16 vid)
1485 {
1486 	struct hnae3_handle *h = hns3_get_handle(netdev);
1487 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1488 	int ret = -EIO;
1489 
1490 	if (h->ae_algo->ops->set_vlan_filter)
1491 		ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
1492 
1493 	if (!ret)
1494 		clear_bit(vid, priv->active_vlans);
1495 
1496 	return ret;
1497 }
1498 
1499 static int hns3_restore_vlan(struct net_device *netdev)
1500 {
1501 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1502 	int ret = 0;
1503 	u16 vid;
1504 
1505 	for_each_set_bit(vid, priv->active_vlans, VLAN_N_VID) {
1506 		ret = hns3_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
1507 		if (ret) {
1508 			netdev_err(netdev, "Restore vlan: %d filter, ret:%d\n",
1509 				   vid, ret);
1510 			return ret;
1511 		}
1512 	}
1513 
1514 	return ret;
1515 }
1516 
1517 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
1518 				u8 qos, __be16 vlan_proto)
1519 {
1520 	struct hnae3_handle *h = hns3_get_handle(netdev);
1521 	int ret = -EIO;
1522 
1523 	if (h->ae_algo->ops->set_vf_vlan_filter)
1524 		ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
1525 						   qos, vlan_proto);
1526 
1527 	return ret;
1528 }
1529 
1530 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
1531 {
1532 	struct hnae3_handle *h = hns3_get_handle(netdev);
1533 	bool if_running = netif_running(netdev);
1534 	int ret;
1535 
1536 	if (!h->ae_algo->ops->set_mtu)
1537 		return -EOPNOTSUPP;
1538 
1539 	/* if this was called with netdev up then bring netdevice down */
1540 	if (if_running) {
1541 		(void)hns3_nic_net_stop(netdev);
1542 		msleep(100);
1543 	}
1544 
1545 	ret = h->ae_algo->ops->set_mtu(h, new_mtu);
1546 	if (ret)
1547 		netdev_err(netdev, "failed to change MTU in hardware %d\n",
1548 			   ret);
1549 	else
1550 		netdev->mtu = new_mtu;
1551 
1552 	/* if the netdev was running earlier, bring it up again */
1553 	if (if_running && hns3_nic_net_open(netdev))
1554 		ret = -EINVAL;
1555 
1556 	return ret;
1557 }
1558 
1559 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
1560 {
1561 	struct hns3_nic_priv *priv = netdev_priv(ndev);
1562 	struct hns3_enet_ring *tx_ring = NULL;
1563 	int timeout_queue = 0;
1564 	int hw_head, hw_tail;
1565 	int i;
1566 
1567 	/* Find the stopped queue the same way the stack does */
1568 	for (i = 0; i < ndev->real_num_tx_queues; i++) {
1569 		struct netdev_queue *q;
1570 		unsigned long trans_start;
1571 
1572 		q = netdev_get_tx_queue(ndev, i);
1573 		trans_start = q->trans_start;
1574 		if (netif_xmit_stopped(q) &&
1575 		    time_after(jiffies,
1576 			       (trans_start + ndev->watchdog_timeo))) {
1577 			timeout_queue = i;
1578 			break;
1579 		}
1580 	}
1581 
1582 	if (i == ndev->num_tx_queues) {
1583 		netdev_info(ndev,
1584 			    "no netdev TX timeout queue found, timeout count: %llu\n",
1585 			    priv->tx_timeout_count);
1586 		return false;
1587 	}
1588 
1589 	tx_ring = priv->ring_data[timeout_queue].ring;
1590 
1591 	hw_head = readl_relaxed(tx_ring->tqp->io_base +
1592 				HNS3_RING_TX_RING_HEAD_REG);
1593 	hw_tail = readl_relaxed(tx_ring->tqp->io_base +
1594 				HNS3_RING_TX_RING_TAIL_REG);
1595 	netdev_info(ndev,
1596 		    "tx_timeout count: %llu, queue id: %d, SW_NTU: 0x%x, SW_NTC: 0x%x, HW_HEAD: 0x%x, HW_TAIL: 0x%x, INT: 0x%x\n",
1597 		    priv->tx_timeout_count,
1598 		    timeout_queue,
1599 		    tx_ring->next_to_use,
1600 		    tx_ring->next_to_clean,
1601 		    hw_head,
1602 		    hw_tail,
1603 		    readl(tx_ring->tqp_vector->mask_addr));
1604 
1605 	return true;
1606 }
1607 
1608 static void hns3_nic_net_timeout(struct net_device *ndev)
1609 {
1610 	struct hns3_nic_priv *priv = netdev_priv(ndev);
1611 	struct hnae3_handle *h = priv->ae_handle;
1612 
1613 	if (!hns3_get_tx_timeo_queue_info(ndev))
1614 		return;
1615 
1616 	priv->tx_timeout_count++;
1617 
1618 	if (time_before(jiffies, (h->last_reset_time + ndev->watchdog_timeo)))
1619 		return;
1620 
1621 	/* request the reset */
1622 	if (h->ae_algo->ops->reset_event)
1623 		h->ae_algo->ops->reset_event(h->pdev, h);
1624 }
1625 
1626 static const struct net_device_ops hns3_nic_netdev_ops = {
1627 	.ndo_open		= hns3_nic_net_open,
1628 	.ndo_stop		= hns3_nic_net_stop,
1629 	.ndo_start_xmit		= hns3_nic_net_xmit,
1630 	.ndo_tx_timeout		= hns3_nic_net_timeout,
1631 	.ndo_set_mac_address	= hns3_nic_net_set_mac_address,
1632 	.ndo_do_ioctl		= hns3_nic_do_ioctl,
1633 	.ndo_change_mtu		= hns3_nic_change_mtu,
1634 	.ndo_set_features	= hns3_nic_set_features,
1635 	.ndo_get_stats64	= hns3_nic_get_stats64,
1636 	.ndo_setup_tc		= hns3_nic_setup_tc,
1637 	.ndo_set_rx_mode	= hns3_nic_set_rx_mode,
1638 	.ndo_vlan_rx_add_vid	= hns3_vlan_rx_add_vid,
1639 	.ndo_vlan_rx_kill_vid	= hns3_vlan_rx_kill_vid,
1640 	.ndo_set_vf_vlan	= hns3_ndo_set_vf_vlan,
1641 };
1642 
1643 static bool hns3_is_phys_func(struct pci_dev *pdev)
1644 {
1645 	u32 dev_id = pdev->device;
1646 
1647 	switch (dev_id) {
1648 	case HNAE3_DEV_ID_GE:
1649 	case HNAE3_DEV_ID_25GE:
1650 	case HNAE3_DEV_ID_25GE_RDMA:
1651 	case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
1652 	case HNAE3_DEV_ID_50GE_RDMA:
1653 	case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
1654 	case HNAE3_DEV_ID_100G_RDMA_MACSEC:
1655 		return true;
1656 	case HNAE3_DEV_ID_100G_VF:
1657 	case HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF:
1658 		return false;
1659 	default:
1660 		dev_warn(&pdev->dev, "un-recognized pci device-id %d",
1661 			 dev_id);
1662 	}
1663 
1664 	return false;
1665 }
1666 
1667 static void hns3_disable_sriov(struct pci_dev *pdev)
1668 {
1669 	/* If our VFs are assigned we cannot shut down SR-IOV
1670 	 * without causing issues, so just leave the hardware
1671 	 * available but disabled
1672 	 */
1673 	if (pci_vfs_assigned(pdev)) {
1674 		dev_warn(&pdev->dev,
1675 			 "disabling driver while VFs are assigned\n");
1676 		return;
1677 	}
1678 
1679 	pci_disable_sriov(pdev);
1680 }
1681 
1682 static void hns3_get_dev_capability(struct pci_dev *pdev,
1683 				    struct hnae3_ae_dev *ae_dev)
1684 {
1685 	if (pdev->revision >= 0x21)
1686 		hnae3_set_bit(ae_dev->flag, HNAE3_DEV_SUPPORT_FD_B, 1);
1687 }
1688 
1689 /* hns3_probe - Device initialization routine
1690  * @pdev: PCI device information struct
1691  * @ent: entry in hns3_pci_tbl
1692  *
1693  * hns3_probe initializes a PF identified by a pci_dev structure.
1694  * The OS initialization, configuring of the PF private structure,
1695  * and a hardware reset occur.
1696  *
1697  * Returns 0 on success, negative on failure
1698  */
1699 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1700 {
1701 	struct hnae3_ae_dev *ae_dev;
1702 	int ret;
1703 
1704 	ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev),
1705 			      GFP_KERNEL);
1706 	if (!ae_dev) {
1707 		ret = -ENOMEM;
1708 		return ret;
1709 	}
1710 
1711 	ae_dev->pdev = pdev;
1712 	ae_dev->flag = ent->driver_data;
1713 	ae_dev->dev_type = HNAE3_DEV_KNIC;
1714 	ae_dev->reset_type = HNAE3_NONE_RESET;
1715 	hns3_get_dev_capability(pdev, ae_dev);
1716 	pci_set_drvdata(pdev, ae_dev);
1717 
1718 	hnae3_register_ae_dev(ae_dev);
1719 
1720 	return 0;
1721 }
1722 
1723 /* hns3_remove - Device removal routine
1724  * @pdev: PCI device information struct
1725  */
1726 static void hns3_remove(struct pci_dev *pdev)
1727 {
1728 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1729 
1730 	if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
1731 		hns3_disable_sriov(pdev);
1732 
1733 	hnae3_unregister_ae_dev(ae_dev);
1734 }
1735 
1736 /**
1737  * hns3_pci_sriov_configure
1738  * @pdev: pointer to a pci_dev structure
1739  * @num_vfs: number of VFs to allocate
1740  *
1741  * Enable or change the number of VFs. Called when the user updates the number
1742  * of VFs in sysfs.
1743  **/
1744 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
1745 {
1746 	int ret;
1747 
1748 	if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
1749 		dev_warn(&pdev->dev, "Can not config SRIOV\n");
1750 		return -EINVAL;
1751 	}
1752 
1753 	if (num_vfs) {
1754 		ret = pci_enable_sriov(pdev, num_vfs);
1755 		if (ret)
1756 			dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
1757 		else
1758 			return num_vfs;
1759 	} else if (!pci_vfs_assigned(pdev)) {
1760 		pci_disable_sriov(pdev);
1761 	} else {
1762 		dev_warn(&pdev->dev,
1763 			 "Unable to free VFs because some are assigned to VMs.\n");
1764 	}
1765 
1766 	return 0;
1767 }
1768 
1769 static void hns3_shutdown(struct pci_dev *pdev)
1770 {
1771 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1772 
1773 	hnae3_unregister_ae_dev(ae_dev);
1774 	devm_kfree(&pdev->dev, ae_dev);
1775 	pci_set_drvdata(pdev, NULL);
1776 
1777 	if (system_state == SYSTEM_POWER_OFF)
1778 		pci_set_power_state(pdev, PCI_D3hot);
1779 }
1780 
1781 static pci_ers_result_t hns3_error_detected(struct pci_dev *pdev,
1782 					    pci_channel_state_t state)
1783 {
1784 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1785 	pci_ers_result_t ret;
1786 
1787 	dev_info(&pdev->dev, "PCI error detected, state(=%d)!!\n", state);
1788 
1789 	if (state == pci_channel_io_perm_failure)
1790 		return PCI_ERS_RESULT_DISCONNECT;
1791 
1792 	if (!ae_dev) {
1793 		dev_err(&pdev->dev,
1794 			"Can't recover - error happened during device init\n");
1795 		return PCI_ERS_RESULT_NONE;
1796 	}
1797 
1798 	if (ae_dev->ops->process_hw_error)
1799 		ret = ae_dev->ops->process_hw_error(ae_dev);
1800 	else
1801 		return PCI_ERS_RESULT_NONE;
1802 
1803 	return ret;
1804 }
1805 
1806 static pci_ers_result_t hns3_slot_reset(struct pci_dev *pdev)
1807 {
1808 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1809 	struct device *dev = &pdev->dev;
1810 
1811 	dev_info(dev, "requesting reset due to PCI error\n");
1812 
1813 	/* request the reset */
1814 	if (ae_dev->ops->reset_event) {
1815 		ae_dev->ops->reset_event(pdev, NULL);
1816 		return PCI_ERS_RESULT_RECOVERED;
1817 	}
1818 
1819 	return PCI_ERS_RESULT_DISCONNECT;
1820 }
1821 
1822 static const struct pci_error_handlers hns3_err_handler = {
1823 	.error_detected = hns3_error_detected,
1824 	.slot_reset     = hns3_slot_reset,
1825 };
1826 
1827 static struct pci_driver hns3_driver = {
1828 	.name     = hns3_driver_name,
1829 	.id_table = hns3_pci_tbl,
1830 	.probe    = hns3_probe,
1831 	.remove   = hns3_remove,
1832 	.shutdown = hns3_shutdown,
1833 	.sriov_configure = hns3_pci_sriov_configure,
1834 	.err_handler    = &hns3_err_handler,
1835 };
1836 
1837 /* set default feature to hns3 */
1838 static void hns3_set_default_feature(struct net_device *netdev)
1839 {
1840 	struct hnae3_handle *h = hns3_get_handle(netdev);
1841 	struct pci_dev *pdev = h->pdev;
1842 
1843 	netdev->priv_flags |= IFF_UNICAST_FLT;
1844 
1845 	netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1846 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1847 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1848 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1849 		NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1850 
1851 	netdev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
1852 
1853 	netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
1854 
1855 	netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1856 		NETIF_F_HW_VLAN_CTAG_FILTER |
1857 		NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
1858 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1859 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1860 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1861 		NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1862 
1863 	netdev->vlan_features |=
1864 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
1865 		NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO |
1866 		NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1867 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1868 		NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1869 
1870 	netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1871 		NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
1872 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1873 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1874 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1875 		NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1876 
1877 	if (pdev->revision >= 0x21) {
1878 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
1879 
1880 		if (!(h->flags & HNAE3_SUPPORT_VF)) {
1881 			netdev->hw_features |= NETIF_F_NTUPLE;
1882 			netdev->features |= NETIF_F_NTUPLE;
1883 		}
1884 	}
1885 }
1886 
1887 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
1888 			     struct hns3_desc_cb *cb)
1889 {
1890 	unsigned int order = hnae3_page_order(ring);
1891 	struct page *p;
1892 
1893 	p = dev_alloc_pages(order);
1894 	if (!p)
1895 		return -ENOMEM;
1896 
1897 	cb->priv = p;
1898 	cb->page_offset = 0;
1899 	cb->reuse_flag = 0;
1900 	cb->buf  = page_address(p);
1901 	cb->length = hnae3_page_size(ring);
1902 	cb->type = DESC_TYPE_PAGE;
1903 
1904 	return 0;
1905 }
1906 
1907 static void hns3_free_buffer(struct hns3_enet_ring *ring,
1908 			     struct hns3_desc_cb *cb)
1909 {
1910 	if (cb->type == DESC_TYPE_SKB)
1911 		dev_kfree_skb_any((struct sk_buff *)cb->priv);
1912 	else if (!HNAE3_IS_TX_RING(ring))
1913 		put_page((struct page *)cb->priv);
1914 	memset(cb, 0, sizeof(*cb));
1915 }
1916 
1917 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
1918 {
1919 	cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
1920 			       cb->length, ring_to_dma_dir(ring));
1921 
1922 	if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
1923 		return -EIO;
1924 
1925 	return 0;
1926 }
1927 
1928 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
1929 			      struct hns3_desc_cb *cb)
1930 {
1931 	if (cb->type == DESC_TYPE_SKB)
1932 		dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
1933 				 ring_to_dma_dir(ring));
1934 	else if (cb->length)
1935 		dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
1936 			       ring_to_dma_dir(ring));
1937 }
1938 
1939 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
1940 {
1941 	hns3_unmap_buffer(ring, &ring->desc_cb[i]);
1942 	ring->desc[i].addr = 0;
1943 }
1944 
1945 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i)
1946 {
1947 	struct hns3_desc_cb *cb = &ring->desc_cb[i];
1948 
1949 	if (!ring->desc_cb[i].dma)
1950 		return;
1951 
1952 	hns3_buffer_detach(ring, i);
1953 	hns3_free_buffer(ring, cb);
1954 }
1955 
1956 static void hns3_free_buffers(struct hns3_enet_ring *ring)
1957 {
1958 	int i;
1959 
1960 	for (i = 0; i < ring->desc_num; i++)
1961 		hns3_free_buffer_detach(ring, i);
1962 }
1963 
1964 /* free desc along with its attached buffer */
1965 static void hns3_free_desc(struct hns3_enet_ring *ring)
1966 {
1967 	int size = ring->desc_num * sizeof(ring->desc[0]);
1968 
1969 	hns3_free_buffers(ring);
1970 
1971 	if (ring->desc) {
1972 		dma_free_coherent(ring_to_dev(ring), size,
1973 				  ring->desc, ring->desc_dma_addr);
1974 		ring->desc = NULL;
1975 	}
1976 }
1977 
1978 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
1979 {
1980 	int size = ring->desc_num * sizeof(ring->desc[0]);
1981 
1982 	ring->desc = dma_zalloc_coherent(ring_to_dev(ring), size,
1983 					 &ring->desc_dma_addr,
1984 					 GFP_KERNEL);
1985 	if (!ring->desc)
1986 		return -ENOMEM;
1987 
1988 	return 0;
1989 }
1990 
1991 static int hns3_reserve_buffer_map(struct hns3_enet_ring *ring,
1992 				   struct hns3_desc_cb *cb)
1993 {
1994 	int ret;
1995 
1996 	ret = hns3_alloc_buffer(ring, cb);
1997 	if (ret)
1998 		goto out;
1999 
2000 	ret = hns3_map_buffer(ring, cb);
2001 	if (ret)
2002 		goto out_with_buf;
2003 
2004 	return 0;
2005 
2006 out_with_buf:
2007 	hns3_free_buffer(ring, cb);
2008 out:
2009 	return ret;
2010 }
2011 
2012 static int hns3_alloc_buffer_attach(struct hns3_enet_ring *ring, int i)
2013 {
2014 	int ret = hns3_reserve_buffer_map(ring, &ring->desc_cb[i]);
2015 
2016 	if (ret)
2017 		return ret;
2018 
2019 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2020 
2021 	return 0;
2022 }
2023 
2024 /* Allocate memory for raw pkg, and map with dma */
2025 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
2026 {
2027 	int i, j, ret;
2028 
2029 	for (i = 0; i < ring->desc_num; i++) {
2030 		ret = hns3_alloc_buffer_attach(ring, i);
2031 		if (ret)
2032 			goto out_buffer_fail;
2033 	}
2034 
2035 	return 0;
2036 
2037 out_buffer_fail:
2038 	for (j = i - 1; j >= 0; j--)
2039 		hns3_free_buffer_detach(ring, j);
2040 	return ret;
2041 }
2042 
2043 /* detach a in-used buffer and replace with a reserved one  */
2044 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
2045 				struct hns3_desc_cb *res_cb)
2046 {
2047 	hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2048 	ring->desc_cb[i] = *res_cb;
2049 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2050 	ring->desc[i].rx.bd_base_info = 0;
2051 }
2052 
2053 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
2054 {
2055 	ring->desc_cb[i].reuse_flag = 0;
2056 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma
2057 		+ ring->desc_cb[i].page_offset);
2058 	ring->desc[i].rx.bd_base_info = 0;
2059 }
2060 
2061 static void hns3_nic_reclaim_one_desc(struct hns3_enet_ring *ring, int *bytes,
2062 				      int *pkts)
2063 {
2064 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
2065 
2066 	(*pkts) += (desc_cb->type == DESC_TYPE_SKB);
2067 	(*bytes) += desc_cb->length;
2068 	/* desc_cb will be cleaned, after hnae3_free_buffer_detach*/
2069 	hns3_free_buffer_detach(ring, ring->next_to_clean);
2070 
2071 	ring_ptr_move_fw(ring, next_to_clean);
2072 }
2073 
2074 static int is_valid_clean_head(struct hns3_enet_ring *ring, int h)
2075 {
2076 	int u = ring->next_to_use;
2077 	int c = ring->next_to_clean;
2078 
2079 	if (unlikely(h > ring->desc_num))
2080 		return 0;
2081 
2082 	return u > c ? (h > c && h <= u) : (h > c || h <= u);
2083 }
2084 
2085 void hns3_clean_tx_ring(struct hns3_enet_ring *ring)
2086 {
2087 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2088 	struct hns3_nic_priv *priv = netdev_priv(netdev);
2089 	struct netdev_queue *dev_queue;
2090 	int bytes, pkts;
2091 	int head;
2092 
2093 	head = readl_relaxed(ring->tqp->io_base + HNS3_RING_TX_RING_HEAD_REG);
2094 	rmb(); /* Make sure head is ready before touch any data */
2095 
2096 	if (is_ring_empty(ring) || head == ring->next_to_clean)
2097 		return; /* no data to poll */
2098 
2099 	if (unlikely(!is_valid_clean_head(ring, head))) {
2100 		netdev_err(netdev, "wrong head (%d, %d-%d)\n", head,
2101 			   ring->next_to_use, ring->next_to_clean);
2102 
2103 		u64_stats_update_begin(&ring->syncp);
2104 		ring->stats.io_err_cnt++;
2105 		u64_stats_update_end(&ring->syncp);
2106 		return;
2107 	}
2108 
2109 	bytes = 0;
2110 	pkts = 0;
2111 	while (head != ring->next_to_clean) {
2112 		hns3_nic_reclaim_one_desc(ring, &bytes, &pkts);
2113 		/* Issue prefetch for next Tx descriptor */
2114 		prefetch(&ring->desc_cb[ring->next_to_clean]);
2115 	}
2116 
2117 	ring->tqp_vector->tx_group.total_bytes += bytes;
2118 	ring->tqp_vector->tx_group.total_packets += pkts;
2119 
2120 	u64_stats_update_begin(&ring->syncp);
2121 	ring->stats.tx_bytes += bytes;
2122 	ring->stats.tx_pkts += pkts;
2123 	u64_stats_update_end(&ring->syncp);
2124 
2125 	dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
2126 	netdev_tx_completed_queue(dev_queue, pkts, bytes);
2127 
2128 	if (unlikely(pkts && netif_carrier_ok(netdev) &&
2129 		     (ring_space(ring) > HNS3_MAX_BD_PER_PKT))) {
2130 		/* Make sure that anybody stopping the queue after this
2131 		 * sees the new next_to_clean.
2132 		 */
2133 		smp_mb();
2134 		if (netif_tx_queue_stopped(dev_queue) &&
2135 		    !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
2136 			netif_tx_wake_queue(dev_queue);
2137 			ring->stats.restart_queue++;
2138 		}
2139 	}
2140 }
2141 
2142 static int hns3_desc_unused(struct hns3_enet_ring *ring)
2143 {
2144 	int ntc = ring->next_to_clean;
2145 	int ntu = ring->next_to_use;
2146 
2147 	return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
2148 }
2149 
2150 static void
2151 hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring, int cleand_count)
2152 {
2153 	struct hns3_desc_cb *desc_cb;
2154 	struct hns3_desc_cb res_cbs;
2155 	int i, ret;
2156 
2157 	for (i = 0; i < cleand_count; i++) {
2158 		desc_cb = &ring->desc_cb[ring->next_to_use];
2159 		if (desc_cb->reuse_flag) {
2160 			u64_stats_update_begin(&ring->syncp);
2161 			ring->stats.reuse_pg_cnt++;
2162 			u64_stats_update_end(&ring->syncp);
2163 
2164 			hns3_reuse_buffer(ring, ring->next_to_use);
2165 		} else {
2166 			ret = hns3_reserve_buffer_map(ring, &res_cbs);
2167 			if (ret) {
2168 				u64_stats_update_begin(&ring->syncp);
2169 				ring->stats.sw_err_cnt++;
2170 				u64_stats_update_end(&ring->syncp);
2171 
2172 				netdev_err(ring->tqp->handle->kinfo.netdev,
2173 					   "hnae reserve buffer map failed.\n");
2174 				break;
2175 			}
2176 			hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
2177 		}
2178 
2179 		ring_ptr_move_fw(ring, next_to_use);
2180 	}
2181 
2182 	wmb(); /* Make all data has been write before submit */
2183 	writel_relaxed(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
2184 }
2185 
2186 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
2187 				struct hns3_enet_ring *ring, int pull_len,
2188 				struct hns3_desc_cb *desc_cb)
2189 {
2190 	struct hns3_desc *desc;
2191 	u32 truesize;
2192 	int size;
2193 	int last_offset;
2194 	bool twobufs;
2195 
2196 	twobufs = ((PAGE_SIZE < 8192) &&
2197 		hnae3_buf_size(ring) == HNS3_BUFFER_SIZE_2048);
2198 
2199 	desc = &ring->desc[ring->next_to_clean];
2200 	size = le16_to_cpu(desc->rx.size);
2201 
2202 	truesize = hnae3_buf_size(ring);
2203 
2204 	if (!twobufs)
2205 		last_offset = hnae3_page_size(ring) - hnae3_buf_size(ring);
2206 
2207 	skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
2208 			size - pull_len, truesize);
2209 
2210 	 /* Avoid re-using remote pages,flag default unreuse */
2211 	if (unlikely(page_to_nid(desc_cb->priv) != numa_node_id()))
2212 		return;
2213 
2214 	if (twobufs) {
2215 		/* If we are only owner of page we can reuse it */
2216 		if (likely(page_count(desc_cb->priv) == 1)) {
2217 			/* Flip page offset to other buffer */
2218 			desc_cb->page_offset ^= truesize;
2219 
2220 			desc_cb->reuse_flag = 1;
2221 			/* bump ref count on page before it is given*/
2222 			get_page(desc_cb->priv);
2223 		}
2224 		return;
2225 	}
2226 
2227 	/* Move offset up to the next cache line */
2228 	desc_cb->page_offset += truesize;
2229 
2230 	if (desc_cb->page_offset <= last_offset) {
2231 		desc_cb->reuse_flag = 1;
2232 		/* Bump ref count on page before it is given*/
2233 		get_page(desc_cb->priv);
2234 	}
2235 }
2236 
2237 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
2238 			     struct hns3_desc *desc)
2239 {
2240 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2241 	int l3_type, l4_type;
2242 	u32 bd_base_info;
2243 	int ol4_type;
2244 	u32 l234info;
2245 
2246 	bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2247 	l234info = le32_to_cpu(desc->rx.l234_info);
2248 
2249 	skb->ip_summed = CHECKSUM_NONE;
2250 
2251 	skb_checksum_none_assert(skb);
2252 
2253 	if (!(netdev->features & NETIF_F_RXCSUM))
2254 		return;
2255 
2256 	/* check if hardware has done checksum */
2257 	if (!hnae3_get_bit(bd_base_info, HNS3_RXD_L3L4P_B))
2258 		return;
2259 
2260 	if (unlikely(hnae3_get_bit(l234info, HNS3_RXD_L3E_B) ||
2261 		     hnae3_get_bit(l234info, HNS3_RXD_L4E_B) ||
2262 		     hnae3_get_bit(l234info, HNS3_RXD_OL3E_B) ||
2263 		     hnae3_get_bit(l234info, HNS3_RXD_OL4E_B))) {
2264 		u64_stats_update_begin(&ring->syncp);
2265 		ring->stats.l3l4_csum_err++;
2266 		u64_stats_update_end(&ring->syncp);
2267 
2268 		return;
2269 	}
2270 
2271 	l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
2272 				  HNS3_RXD_L3ID_S);
2273 	l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
2274 				  HNS3_RXD_L4ID_S);
2275 
2276 	ol4_type = hnae3_get_field(l234info, HNS3_RXD_OL4ID_M,
2277 				   HNS3_RXD_OL4ID_S);
2278 	switch (ol4_type) {
2279 	case HNS3_OL4_TYPE_MAC_IN_UDP:
2280 	case HNS3_OL4_TYPE_NVGRE:
2281 		skb->csum_level = 1;
2282 		/* fall through */
2283 	case HNS3_OL4_TYPE_NO_TUN:
2284 		/* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
2285 		if ((l3_type == HNS3_L3_TYPE_IPV4 ||
2286 		     l3_type == HNS3_L3_TYPE_IPV6) &&
2287 		    (l4_type == HNS3_L4_TYPE_UDP ||
2288 		     l4_type == HNS3_L4_TYPE_TCP ||
2289 		     l4_type == HNS3_L4_TYPE_SCTP))
2290 			skb->ip_summed = CHECKSUM_UNNECESSARY;
2291 		break;
2292 	default:
2293 		break;
2294 	}
2295 }
2296 
2297 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
2298 {
2299 	napi_gro_receive(&ring->tqp_vector->napi, skb);
2300 }
2301 
2302 static bool hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
2303 				struct hns3_desc *desc, u32 l234info,
2304 				u16 *vlan_tag)
2305 {
2306 	struct pci_dev *pdev = ring->tqp->handle->pdev;
2307 
2308 	if (pdev->revision == 0x20) {
2309 		*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2310 		if (!(*vlan_tag & VLAN_VID_MASK))
2311 			*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2312 
2313 		return (*vlan_tag != 0);
2314 	}
2315 
2316 #define HNS3_STRP_OUTER_VLAN	0x1
2317 #define HNS3_STRP_INNER_VLAN	0x2
2318 
2319 	switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
2320 				HNS3_RXD_STRP_TAGP_S)) {
2321 	case HNS3_STRP_OUTER_VLAN:
2322 		*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2323 		return true;
2324 	case HNS3_STRP_INNER_VLAN:
2325 		*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2326 		return true;
2327 	default:
2328 		return false;
2329 	}
2330 }
2331 
2332 static void hns3_set_rx_skb_rss_type(struct hns3_enet_ring *ring,
2333 				     struct sk_buff *skb)
2334 {
2335 	struct hns3_desc *desc = &ring->desc[ring->next_to_clean];
2336 	struct hnae3_handle *handle = ring->tqp->handle;
2337 	enum pkt_hash_types rss_type;
2338 
2339 	if (le32_to_cpu(desc->rx.rss_hash))
2340 		rss_type = handle->kinfo.rss_type;
2341 	else
2342 		rss_type = PKT_HASH_TYPE_NONE;
2343 
2344 	skb_set_hash(skb, le32_to_cpu(desc->rx.rss_hash), rss_type);
2345 }
2346 
2347 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring,
2348 			     struct sk_buff **out_skb, int *out_bnum)
2349 {
2350 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2351 	struct hns3_desc_cb *desc_cb;
2352 	struct hns3_desc *desc;
2353 	struct sk_buff *skb;
2354 	unsigned char *va;
2355 	u32 bd_base_info;
2356 	int pull_len;
2357 	u32 l234info;
2358 	int length;
2359 	int bnum;
2360 
2361 	desc = &ring->desc[ring->next_to_clean];
2362 	desc_cb = &ring->desc_cb[ring->next_to_clean];
2363 
2364 	prefetch(desc);
2365 
2366 	length = le16_to_cpu(desc->rx.size);
2367 	bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2368 
2369 	/* Check valid BD */
2370 	if (unlikely(!hnae3_get_bit(bd_base_info, HNS3_RXD_VLD_B)))
2371 		return -EFAULT;
2372 
2373 	va = (unsigned char *)desc_cb->buf + desc_cb->page_offset;
2374 
2375 	/* Prefetch first cache line of first page
2376 	 * Idea is to cache few bytes of the header of the packet. Our L1 Cache
2377 	 * line size is 64B so need to prefetch twice to make it 128B. But in
2378 	 * actual we can have greater size of caches with 128B Level 1 cache
2379 	 * lines. In such a case, single fetch would suffice to cache in the
2380 	 * relevant part of the header.
2381 	 */
2382 	prefetch(va);
2383 #if L1_CACHE_BYTES < 128
2384 	prefetch(va + L1_CACHE_BYTES);
2385 #endif
2386 
2387 	skb = *out_skb = napi_alloc_skb(&ring->tqp_vector->napi,
2388 					HNS3_RX_HEAD_SIZE);
2389 	if (unlikely(!skb)) {
2390 		netdev_err(netdev, "alloc rx skb fail\n");
2391 
2392 		u64_stats_update_begin(&ring->syncp);
2393 		ring->stats.sw_err_cnt++;
2394 		u64_stats_update_end(&ring->syncp);
2395 
2396 		return -ENOMEM;
2397 	}
2398 
2399 	prefetchw(skb->data);
2400 
2401 	bnum = 1;
2402 	if (length <= HNS3_RX_HEAD_SIZE) {
2403 		memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
2404 
2405 		/* We can reuse buffer as-is, just make sure it is local */
2406 		if (likely(page_to_nid(desc_cb->priv) == numa_node_id()))
2407 			desc_cb->reuse_flag = 1;
2408 		else /* This page cannot be reused so discard it */
2409 			put_page(desc_cb->priv);
2410 
2411 		ring_ptr_move_fw(ring, next_to_clean);
2412 	} else {
2413 		u64_stats_update_begin(&ring->syncp);
2414 		ring->stats.seg_pkt_cnt++;
2415 		u64_stats_update_end(&ring->syncp);
2416 
2417 		pull_len = eth_get_headlen(va, HNS3_RX_HEAD_SIZE);
2418 
2419 		memcpy(__skb_put(skb, pull_len), va,
2420 		       ALIGN(pull_len, sizeof(long)));
2421 
2422 		hns3_nic_reuse_page(skb, 0, ring, pull_len, desc_cb);
2423 		ring_ptr_move_fw(ring, next_to_clean);
2424 
2425 		while (!hnae3_get_bit(bd_base_info, HNS3_RXD_FE_B)) {
2426 			desc = &ring->desc[ring->next_to_clean];
2427 			desc_cb = &ring->desc_cb[ring->next_to_clean];
2428 			bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2429 			hns3_nic_reuse_page(skb, bnum, ring, 0, desc_cb);
2430 			ring_ptr_move_fw(ring, next_to_clean);
2431 			bnum++;
2432 		}
2433 	}
2434 
2435 	*out_bnum = bnum;
2436 
2437 	l234info = le32_to_cpu(desc->rx.l234_info);
2438 
2439 	/* Based on hw strategy, the tag offloaded will be stored at
2440 	 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
2441 	 * in one layer tag case.
2442 	 */
2443 	if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
2444 		u16 vlan_tag;
2445 
2446 		if (hns3_parse_vlan_tag(ring, desc, l234info, &vlan_tag))
2447 			__vlan_hwaccel_put_tag(skb,
2448 					       htons(ETH_P_8021Q),
2449 					       vlan_tag);
2450 	}
2451 
2452 	if (unlikely(!hnae3_get_bit(bd_base_info, HNS3_RXD_VLD_B))) {
2453 		u64_stats_update_begin(&ring->syncp);
2454 		ring->stats.non_vld_descs++;
2455 		u64_stats_update_end(&ring->syncp);
2456 
2457 		dev_kfree_skb_any(skb);
2458 		return -EINVAL;
2459 	}
2460 
2461 	if (unlikely((!desc->rx.pkt_len) ||
2462 		     hnae3_get_bit(l234info, HNS3_RXD_TRUNCAT_B))) {
2463 		u64_stats_update_begin(&ring->syncp);
2464 		ring->stats.err_pkt_len++;
2465 		u64_stats_update_end(&ring->syncp);
2466 
2467 		dev_kfree_skb_any(skb);
2468 		return -EFAULT;
2469 	}
2470 
2471 	if (unlikely(hnae3_get_bit(l234info, HNS3_RXD_L2E_B))) {
2472 		u64_stats_update_begin(&ring->syncp);
2473 		ring->stats.l2_err++;
2474 		u64_stats_update_end(&ring->syncp);
2475 
2476 		dev_kfree_skb_any(skb);
2477 		return -EFAULT;
2478 	}
2479 
2480 	u64_stats_update_begin(&ring->syncp);
2481 	ring->stats.rx_pkts++;
2482 	ring->stats.rx_bytes += skb->len;
2483 	u64_stats_update_end(&ring->syncp);
2484 
2485 	ring->tqp_vector->rx_group.total_bytes += skb->len;
2486 
2487 	hns3_rx_checksum(ring, skb, desc);
2488 	hns3_set_rx_skb_rss_type(ring, skb);
2489 
2490 	return 0;
2491 }
2492 
2493 int hns3_clean_rx_ring(
2494 		struct hns3_enet_ring *ring, int budget,
2495 		void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
2496 {
2497 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
2498 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2499 	int recv_pkts, recv_bds, clean_count, err;
2500 	int unused_count = hns3_desc_unused(ring);
2501 	struct sk_buff *skb = NULL;
2502 	int num, bnum = 0;
2503 
2504 	num = readl_relaxed(ring->tqp->io_base + HNS3_RING_RX_RING_FBDNUM_REG);
2505 	rmb(); /* Make sure num taken effect before the other data is touched */
2506 
2507 	recv_pkts = 0, recv_bds = 0, clean_count = 0;
2508 	num -= unused_count;
2509 
2510 	while (recv_pkts < budget && recv_bds < num) {
2511 		/* Reuse or realloc buffers */
2512 		if (clean_count + unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
2513 			hns3_nic_alloc_rx_buffers(ring,
2514 						  clean_count + unused_count);
2515 			clean_count = 0;
2516 			unused_count = hns3_desc_unused(ring);
2517 		}
2518 
2519 		/* Poll one pkt */
2520 		err = hns3_handle_rx_bd(ring, &skb, &bnum);
2521 		if (unlikely(!skb)) /* This fault cannot be repaired */
2522 			goto out;
2523 
2524 		recv_bds += bnum;
2525 		clean_count += bnum;
2526 		if (unlikely(err)) {  /* Do jump the err */
2527 			recv_pkts++;
2528 			continue;
2529 		}
2530 
2531 		/* Do update ip stack process */
2532 		skb->protocol = eth_type_trans(skb, netdev);
2533 		rx_fn(ring, skb);
2534 
2535 		recv_pkts++;
2536 	}
2537 
2538 out:
2539 	/* Make all data has been write before submit */
2540 	if (clean_count + unused_count > 0)
2541 		hns3_nic_alloc_rx_buffers(ring,
2542 					  clean_count + unused_count);
2543 
2544 	return recv_pkts;
2545 }
2546 
2547 static bool hns3_get_new_int_gl(struct hns3_enet_ring_group *ring_group)
2548 {
2549 	struct hns3_enet_tqp_vector *tqp_vector =
2550 					ring_group->ring->tqp_vector;
2551 	enum hns3_flow_level_range new_flow_level;
2552 	int packets_per_msecs;
2553 	int bytes_per_msecs;
2554 	u32 time_passed_ms;
2555 	u16 new_int_gl;
2556 
2557 	if (!ring_group->coal.int_gl || !tqp_vector->last_jiffies)
2558 		return false;
2559 
2560 	if (ring_group->total_packets == 0) {
2561 		ring_group->coal.int_gl = HNS3_INT_GL_50K;
2562 		ring_group->coal.flow_level = HNS3_FLOW_LOW;
2563 		return true;
2564 	}
2565 
2566 	/* Simple throttlerate management
2567 	 * 0-10MB/s   lower     (50000 ints/s)
2568 	 * 10-20MB/s   middle    (20000 ints/s)
2569 	 * 20-1249MB/s high      (18000 ints/s)
2570 	 * > 40000pps  ultra     (8000 ints/s)
2571 	 */
2572 	new_flow_level = ring_group->coal.flow_level;
2573 	new_int_gl = ring_group->coal.int_gl;
2574 	time_passed_ms =
2575 		jiffies_to_msecs(jiffies - tqp_vector->last_jiffies);
2576 
2577 	if (!time_passed_ms)
2578 		return false;
2579 
2580 	do_div(ring_group->total_packets, time_passed_ms);
2581 	packets_per_msecs = ring_group->total_packets;
2582 
2583 	do_div(ring_group->total_bytes, time_passed_ms);
2584 	bytes_per_msecs = ring_group->total_bytes;
2585 
2586 #define HNS3_RX_LOW_BYTE_RATE 10000
2587 #define HNS3_RX_MID_BYTE_RATE 20000
2588 
2589 	switch (new_flow_level) {
2590 	case HNS3_FLOW_LOW:
2591 		if (bytes_per_msecs > HNS3_RX_LOW_BYTE_RATE)
2592 			new_flow_level = HNS3_FLOW_MID;
2593 		break;
2594 	case HNS3_FLOW_MID:
2595 		if (bytes_per_msecs > HNS3_RX_MID_BYTE_RATE)
2596 			new_flow_level = HNS3_FLOW_HIGH;
2597 		else if (bytes_per_msecs <= HNS3_RX_LOW_BYTE_RATE)
2598 			new_flow_level = HNS3_FLOW_LOW;
2599 		break;
2600 	case HNS3_FLOW_HIGH:
2601 	case HNS3_FLOW_ULTRA:
2602 	default:
2603 		if (bytes_per_msecs <= HNS3_RX_MID_BYTE_RATE)
2604 			new_flow_level = HNS3_FLOW_MID;
2605 		break;
2606 	}
2607 
2608 #define HNS3_RX_ULTRA_PACKET_RATE 40
2609 
2610 	if (packets_per_msecs > HNS3_RX_ULTRA_PACKET_RATE &&
2611 	    &tqp_vector->rx_group == ring_group)
2612 		new_flow_level = HNS3_FLOW_ULTRA;
2613 
2614 	switch (new_flow_level) {
2615 	case HNS3_FLOW_LOW:
2616 		new_int_gl = HNS3_INT_GL_50K;
2617 		break;
2618 	case HNS3_FLOW_MID:
2619 		new_int_gl = HNS3_INT_GL_20K;
2620 		break;
2621 	case HNS3_FLOW_HIGH:
2622 		new_int_gl = HNS3_INT_GL_18K;
2623 		break;
2624 	case HNS3_FLOW_ULTRA:
2625 		new_int_gl = HNS3_INT_GL_8K;
2626 		break;
2627 	default:
2628 		break;
2629 	}
2630 
2631 	ring_group->total_bytes = 0;
2632 	ring_group->total_packets = 0;
2633 	ring_group->coal.flow_level = new_flow_level;
2634 	if (new_int_gl != ring_group->coal.int_gl) {
2635 		ring_group->coal.int_gl = new_int_gl;
2636 		return true;
2637 	}
2638 	return false;
2639 }
2640 
2641 static void hns3_update_new_int_gl(struct hns3_enet_tqp_vector *tqp_vector)
2642 {
2643 	struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
2644 	struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
2645 	bool rx_update, tx_update;
2646 
2647 	if (tqp_vector->int_adapt_down > 0) {
2648 		tqp_vector->int_adapt_down--;
2649 		return;
2650 	}
2651 
2652 	if (rx_group->coal.gl_adapt_enable) {
2653 		rx_update = hns3_get_new_int_gl(rx_group);
2654 		if (rx_update)
2655 			hns3_set_vector_coalesce_rx_gl(tqp_vector,
2656 						       rx_group->coal.int_gl);
2657 	}
2658 
2659 	if (tx_group->coal.gl_adapt_enable) {
2660 		tx_update = hns3_get_new_int_gl(&tqp_vector->tx_group);
2661 		if (tx_update)
2662 			hns3_set_vector_coalesce_tx_gl(tqp_vector,
2663 						       tx_group->coal.int_gl);
2664 	}
2665 
2666 	tqp_vector->last_jiffies = jiffies;
2667 	tqp_vector->int_adapt_down = HNS3_INT_ADAPT_DOWN_START;
2668 }
2669 
2670 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
2671 {
2672 	struct hns3_enet_ring *ring;
2673 	int rx_pkt_total = 0;
2674 
2675 	struct hns3_enet_tqp_vector *tqp_vector =
2676 		container_of(napi, struct hns3_enet_tqp_vector, napi);
2677 	bool clean_complete = true;
2678 	int rx_budget;
2679 
2680 	/* Since the actual Tx work is minimal, we can give the Tx a larger
2681 	 * budget and be more aggressive about cleaning up the Tx descriptors.
2682 	 */
2683 	hns3_for_each_ring(ring, tqp_vector->tx_group)
2684 		hns3_clean_tx_ring(ring);
2685 
2686 	/* make sure rx ring budget not smaller than 1 */
2687 	rx_budget = max(budget / tqp_vector->num_tqps, 1);
2688 
2689 	hns3_for_each_ring(ring, tqp_vector->rx_group) {
2690 		int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
2691 						    hns3_rx_skb);
2692 
2693 		if (rx_cleaned >= rx_budget)
2694 			clean_complete = false;
2695 
2696 		rx_pkt_total += rx_cleaned;
2697 	}
2698 
2699 	tqp_vector->rx_group.total_packets += rx_pkt_total;
2700 
2701 	if (!clean_complete)
2702 		return budget;
2703 
2704 	napi_complete(napi);
2705 	hns3_update_new_int_gl(tqp_vector);
2706 	hns3_mask_vector_irq(tqp_vector, 1);
2707 
2708 	return rx_pkt_total;
2709 }
2710 
2711 static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
2712 				      struct hnae3_ring_chain_node *head)
2713 {
2714 	struct pci_dev *pdev = tqp_vector->handle->pdev;
2715 	struct hnae3_ring_chain_node *cur_chain = head;
2716 	struct hnae3_ring_chain_node *chain;
2717 	struct hns3_enet_ring *tx_ring;
2718 	struct hns3_enet_ring *rx_ring;
2719 
2720 	tx_ring = tqp_vector->tx_group.ring;
2721 	if (tx_ring) {
2722 		cur_chain->tqp_index = tx_ring->tqp->tqp_index;
2723 		hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
2724 			      HNAE3_RING_TYPE_TX);
2725 		hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
2726 				HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_TX);
2727 
2728 		cur_chain->next = NULL;
2729 
2730 		while (tx_ring->next) {
2731 			tx_ring = tx_ring->next;
2732 
2733 			chain = devm_kzalloc(&pdev->dev, sizeof(*chain),
2734 					     GFP_KERNEL);
2735 			if (!chain)
2736 				goto err_free_chain;
2737 
2738 			cur_chain->next = chain;
2739 			chain->tqp_index = tx_ring->tqp->tqp_index;
2740 			hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
2741 				      HNAE3_RING_TYPE_TX);
2742 			hnae3_set_field(chain->int_gl_idx,
2743 					HNAE3_RING_GL_IDX_M,
2744 					HNAE3_RING_GL_IDX_S,
2745 					HNAE3_RING_GL_TX);
2746 
2747 			cur_chain = chain;
2748 		}
2749 	}
2750 
2751 	rx_ring = tqp_vector->rx_group.ring;
2752 	if (!tx_ring && rx_ring) {
2753 		cur_chain->next = NULL;
2754 		cur_chain->tqp_index = rx_ring->tqp->tqp_index;
2755 		hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
2756 			      HNAE3_RING_TYPE_RX);
2757 		hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
2758 				HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
2759 
2760 		rx_ring = rx_ring->next;
2761 	}
2762 
2763 	while (rx_ring) {
2764 		chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
2765 		if (!chain)
2766 			goto err_free_chain;
2767 
2768 		cur_chain->next = chain;
2769 		chain->tqp_index = rx_ring->tqp->tqp_index;
2770 		hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
2771 			      HNAE3_RING_TYPE_RX);
2772 		hnae3_set_field(chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
2773 				HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
2774 
2775 		cur_chain = chain;
2776 
2777 		rx_ring = rx_ring->next;
2778 	}
2779 
2780 	return 0;
2781 
2782 err_free_chain:
2783 	cur_chain = head->next;
2784 	while (cur_chain) {
2785 		chain = cur_chain->next;
2786 		devm_kfree(&pdev->dev, chain);
2787 		cur_chain = chain;
2788 	}
2789 
2790 	return -ENOMEM;
2791 }
2792 
2793 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
2794 					struct hnae3_ring_chain_node *head)
2795 {
2796 	struct pci_dev *pdev = tqp_vector->handle->pdev;
2797 	struct hnae3_ring_chain_node *chain_tmp, *chain;
2798 
2799 	chain = head->next;
2800 
2801 	while (chain) {
2802 		chain_tmp = chain->next;
2803 		devm_kfree(&pdev->dev, chain);
2804 		chain = chain_tmp;
2805 	}
2806 }
2807 
2808 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
2809 				   struct hns3_enet_ring *ring)
2810 {
2811 	ring->next = group->ring;
2812 	group->ring = ring;
2813 
2814 	group->count++;
2815 }
2816 
2817 static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
2818 {
2819 	struct pci_dev *pdev = priv->ae_handle->pdev;
2820 	struct hns3_enet_tqp_vector *tqp_vector;
2821 	int num_vectors = priv->vector_num;
2822 	int numa_node;
2823 	int vector_i;
2824 
2825 	numa_node = dev_to_node(&pdev->dev);
2826 
2827 	for (vector_i = 0; vector_i < num_vectors; vector_i++) {
2828 		tqp_vector = &priv->tqp_vector[vector_i];
2829 		cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
2830 				&tqp_vector->affinity_mask);
2831 	}
2832 }
2833 
2834 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
2835 {
2836 	struct hnae3_ring_chain_node vector_ring_chain;
2837 	struct hnae3_handle *h = priv->ae_handle;
2838 	struct hns3_enet_tqp_vector *tqp_vector;
2839 	int ret = 0;
2840 	int i;
2841 
2842 	hns3_nic_set_cpumask(priv);
2843 
2844 	for (i = 0; i < priv->vector_num; i++) {
2845 		tqp_vector = &priv->tqp_vector[i];
2846 		hns3_vector_gl_rl_init_hw(tqp_vector, priv);
2847 		tqp_vector->num_tqps = 0;
2848 	}
2849 
2850 	for (i = 0; i < h->kinfo.num_tqps; i++) {
2851 		u16 vector_i = i % priv->vector_num;
2852 		u16 tqp_num = h->kinfo.num_tqps;
2853 
2854 		tqp_vector = &priv->tqp_vector[vector_i];
2855 
2856 		hns3_add_ring_to_group(&tqp_vector->tx_group,
2857 				       priv->ring_data[i].ring);
2858 
2859 		hns3_add_ring_to_group(&tqp_vector->rx_group,
2860 				       priv->ring_data[i + tqp_num].ring);
2861 
2862 		priv->ring_data[i].ring->tqp_vector = tqp_vector;
2863 		priv->ring_data[i + tqp_num].ring->tqp_vector = tqp_vector;
2864 		tqp_vector->num_tqps++;
2865 	}
2866 
2867 	for (i = 0; i < priv->vector_num; i++) {
2868 		tqp_vector = &priv->tqp_vector[i];
2869 
2870 		tqp_vector->rx_group.total_bytes = 0;
2871 		tqp_vector->rx_group.total_packets = 0;
2872 		tqp_vector->tx_group.total_bytes = 0;
2873 		tqp_vector->tx_group.total_packets = 0;
2874 		tqp_vector->handle = h;
2875 
2876 		ret = hns3_get_vector_ring_chain(tqp_vector,
2877 						 &vector_ring_chain);
2878 		if (ret)
2879 			return ret;
2880 
2881 		ret = h->ae_algo->ops->map_ring_to_vector(h,
2882 			tqp_vector->vector_irq, &vector_ring_chain);
2883 
2884 		hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
2885 
2886 		if (ret)
2887 			goto map_ring_fail;
2888 
2889 		netif_napi_add(priv->netdev, &tqp_vector->napi,
2890 			       hns3_nic_common_poll, NAPI_POLL_WEIGHT);
2891 	}
2892 
2893 	return 0;
2894 
2895 map_ring_fail:
2896 	while (i--)
2897 		netif_napi_del(&priv->tqp_vector[i].napi);
2898 
2899 	return ret;
2900 }
2901 
2902 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
2903 {
2904 	struct hnae3_handle *h = priv->ae_handle;
2905 	struct hns3_enet_tqp_vector *tqp_vector;
2906 	struct hnae3_vector_info *vector;
2907 	struct pci_dev *pdev = h->pdev;
2908 	u16 tqp_num = h->kinfo.num_tqps;
2909 	u16 vector_num;
2910 	int ret = 0;
2911 	u16 i;
2912 
2913 	/* RSS size, cpu online and vector_num should be the same */
2914 	/* Should consider 2p/4p later */
2915 	vector_num = min_t(u16, num_online_cpus(), tqp_num);
2916 	vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
2917 			      GFP_KERNEL);
2918 	if (!vector)
2919 		return -ENOMEM;
2920 
2921 	vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
2922 
2923 	priv->vector_num = vector_num;
2924 	priv->tqp_vector = (struct hns3_enet_tqp_vector *)
2925 		devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
2926 			     GFP_KERNEL);
2927 	if (!priv->tqp_vector) {
2928 		ret = -ENOMEM;
2929 		goto out;
2930 	}
2931 
2932 	for (i = 0; i < priv->vector_num; i++) {
2933 		tqp_vector = &priv->tqp_vector[i];
2934 		tqp_vector->idx = i;
2935 		tqp_vector->mask_addr = vector[i].io_addr;
2936 		tqp_vector->vector_irq = vector[i].vector;
2937 		hns3_vector_gl_rl_init(tqp_vector, priv);
2938 	}
2939 
2940 out:
2941 	devm_kfree(&pdev->dev, vector);
2942 	return ret;
2943 }
2944 
2945 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
2946 {
2947 	group->ring = NULL;
2948 	group->count = 0;
2949 }
2950 
2951 static int hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
2952 {
2953 	struct hnae3_ring_chain_node vector_ring_chain;
2954 	struct hnae3_handle *h = priv->ae_handle;
2955 	struct hns3_enet_tqp_vector *tqp_vector;
2956 	int i, ret;
2957 
2958 	for (i = 0; i < priv->vector_num; i++) {
2959 		tqp_vector = &priv->tqp_vector[i];
2960 
2961 		ret = hns3_get_vector_ring_chain(tqp_vector,
2962 						 &vector_ring_chain);
2963 		if (ret)
2964 			return ret;
2965 
2966 		ret = h->ae_algo->ops->unmap_ring_from_vector(h,
2967 			tqp_vector->vector_irq, &vector_ring_chain);
2968 		if (ret)
2969 			return ret;
2970 
2971 		hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
2972 
2973 		if (priv->tqp_vector[i].irq_init_flag == HNS3_VECTOR_INITED) {
2974 			(void)irq_set_affinity_hint(
2975 				priv->tqp_vector[i].vector_irq,
2976 						    NULL);
2977 			free_irq(priv->tqp_vector[i].vector_irq,
2978 				 &priv->tqp_vector[i]);
2979 		}
2980 
2981 		priv->ring_data[i].ring->irq_init_flag = HNS3_VECTOR_NOT_INITED;
2982 		hns3_clear_ring_group(&tqp_vector->rx_group);
2983 		hns3_clear_ring_group(&tqp_vector->tx_group);
2984 		netif_napi_del(&priv->tqp_vector[i].napi);
2985 	}
2986 
2987 	return 0;
2988 }
2989 
2990 static int hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
2991 {
2992 	struct hnae3_handle *h = priv->ae_handle;
2993 	struct pci_dev *pdev = h->pdev;
2994 	int i, ret;
2995 
2996 	for (i = 0; i < priv->vector_num; i++) {
2997 		struct hns3_enet_tqp_vector *tqp_vector;
2998 
2999 		tqp_vector = &priv->tqp_vector[i];
3000 		ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
3001 		if (ret)
3002 			return ret;
3003 	}
3004 
3005 	devm_kfree(&pdev->dev, priv->tqp_vector);
3006 	return 0;
3007 }
3008 
3009 static int hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
3010 			     int ring_type)
3011 {
3012 	struct hns3_nic_ring_data *ring_data = priv->ring_data;
3013 	int queue_num = priv->ae_handle->kinfo.num_tqps;
3014 	struct pci_dev *pdev = priv->ae_handle->pdev;
3015 	struct hns3_enet_ring *ring;
3016 
3017 	ring = devm_kzalloc(&pdev->dev, sizeof(*ring), GFP_KERNEL);
3018 	if (!ring)
3019 		return -ENOMEM;
3020 
3021 	if (ring_type == HNAE3_RING_TYPE_TX) {
3022 		ring_data[q->tqp_index].ring = ring;
3023 		ring_data[q->tqp_index].queue_index = q->tqp_index;
3024 		ring->io_base = (u8 __iomem *)q->io_base + HNS3_TX_REG_OFFSET;
3025 	} else {
3026 		ring_data[q->tqp_index + queue_num].ring = ring;
3027 		ring_data[q->tqp_index + queue_num].queue_index = q->tqp_index;
3028 		ring->io_base = q->io_base;
3029 	}
3030 
3031 	hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
3032 
3033 	ring->tqp = q;
3034 	ring->desc = NULL;
3035 	ring->desc_cb = NULL;
3036 	ring->dev = priv->dev;
3037 	ring->desc_dma_addr = 0;
3038 	ring->buf_size = q->buf_size;
3039 	ring->desc_num = q->desc_num;
3040 	ring->next_to_use = 0;
3041 	ring->next_to_clean = 0;
3042 
3043 	return 0;
3044 }
3045 
3046 static int hns3_queue_to_ring(struct hnae3_queue *tqp,
3047 			      struct hns3_nic_priv *priv)
3048 {
3049 	int ret;
3050 
3051 	ret = hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
3052 	if (ret)
3053 		return ret;
3054 
3055 	ret = hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
3056 	if (ret) {
3057 		devm_kfree(priv->dev, priv->ring_data[tqp->tqp_index].ring);
3058 		return ret;
3059 	}
3060 
3061 	return 0;
3062 }
3063 
3064 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
3065 {
3066 	struct hnae3_handle *h = priv->ae_handle;
3067 	struct pci_dev *pdev = h->pdev;
3068 	int i, ret;
3069 
3070 	priv->ring_data =  devm_kzalloc(&pdev->dev,
3071 					array3_size(h->kinfo.num_tqps,
3072 						    sizeof(*priv->ring_data),
3073 						    2),
3074 					GFP_KERNEL);
3075 	if (!priv->ring_data)
3076 		return -ENOMEM;
3077 
3078 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3079 		ret = hns3_queue_to_ring(h->kinfo.tqp[i], priv);
3080 		if (ret)
3081 			goto err;
3082 	}
3083 
3084 	return 0;
3085 err:
3086 	while (i--) {
3087 		devm_kfree(priv->dev, priv->ring_data[i].ring);
3088 		devm_kfree(priv->dev,
3089 			   priv->ring_data[i + h->kinfo.num_tqps].ring);
3090 	}
3091 
3092 	devm_kfree(&pdev->dev, priv->ring_data);
3093 	return ret;
3094 }
3095 
3096 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
3097 {
3098 	struct hnae3_handle *h = priv->ae_handle;
3099 	int i;
3100 
3101 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3102 		devm_kfree(priv->dev, priv->ring_data[i].ring);
3103 		devm_kfree(priv->dev,
3104 			   priv->ring_data[i + h->kinfo.num_tqps].ring);
3105 	}
3106 	devm_kfree(priv->dev, priv->ring_data);
3107 }
3108 
3109 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
3110 {
3111 	int ret;
3112 
3113 	if (ring->desc_num <= 0 || ring->buf_size <= 0)
3114 		return -EINVAL;
3115 
3116 	ring->desc_cb = kcalloc(ring->desc_num, sizeof(ring->desc_cb[0]),
3117 				GFP_KERNEL);
3118 	if (!ring->desc_cb) {
3119 		ret = -ENOMEM;
3120 		goto out;
3121 	}
3122 
3123 	ret = hns3_alloc_desc(ring);
3124 	if (ret)
3125 		goto out_with_desc_cb;
3126 
3127 	if (!HNAE3_IS_TX_RING(ring)) {
3128 		ret = hns3_alloc_ring_buffers(ring);
3129 		if (ret)
3130 			goto out_with_desc;
3131 	}
3132 
3133 	return 0;
3134 
3135 out_with_desc:
3136 	hns3_free_desc(ring);
3137 out_with_desc_cb:
3138 	kfree(ring->desc_cb);
3139 	ring->desc_cb = NULL;
3140 out:
3141 	return ret;
3142 }
3143 
3144 static void hns3_fini_ring(struct hns3_enet_ring *ring)
3145 {
3146 	hns3_free_desc(ring);
3147 	kfree(ring->desc_cb);
3148 	ring->desc_cb = NULL;
3149 	ring->next_to_clean = 0;
3150 	ring->next_to_use = 0;
3151 }
3152 
3153 static int hns3_buf_size2type(u32 buf_size)
3154 {
3155 	int bd_size_type;
3156 
3157 	switch (buf_size) {
3158 	case 512:
3159 		bd_size_type = HNS3_BD_SIZE_512_TYPE;
3160 		break;
3161 	case 1024:
3162 		bd_size_type = HNS3_BD_SIZE_1024_TYPE;
3163 		break;
3164 	case 2048:
3165 		bd_size_type = HNS3_BD_SIZE_2048_TYPE;
3166 		break;
3167 	case 4096:
3168 		bd_size_type = HNS3_BD_SIZE_4096_TYPE;
3169 		break;
3170 	default:
3171 		bd_size_type = HNS3_BD_SIZE_2048_TYPE;
3172 	}
3173 
3174 	return bd_size_type;
3175 }
3176 
3177 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
3178 {
3179 	dma_addr_t dma = ring->desc_dma_addr;
3180 	struct hnae3_queue *q = ring->tqp;
3181 
3182 	if (!HNAE3_IS_TX_RING(ring)) {
3183 		hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG,
3184 			       (u32)dma);
3185 		hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
3186 			       (u32)((dma >> 31) >> 1));
3187 
3188 		hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
3189 			       hns3_buf_size2type(ring->buf_size));
3190 		hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
3191 			       ring->desc_num / 8 - 1);
3192 
3193 	} else {
3194 		hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
3195 			       (u32)dma);
3196 		hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
3197 			       (u32)((dma >> 31) >> 1));
3198 
3199 		hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
3200 			       ring->desc_num / 8 - 1);
3201 	}
3202 }
3203 
3204 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
3205 {
3206 	struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
3207 	int i;
3208 
3209 	for (i = 0; i < HNAE3_MAX_TC; i++) {
3210 		struct hnae3_tc_info *tc_info = &kinfo->tc_info[i];
3211 		int j;
3212 
3213 		if (!tc_info->enable)
3214 			continue;
3215 
3216 		for (j = 0; j < tc_info->tqp_count; j++) {
3217 			struct hnae3_queue *q;
3218 
3219 			q = priv->ring_data[tc_info->tqp_offset + j].ring->tqp;
3220 			hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG,
3221 				       tc_info->tc);
3222 		}
3223 	}
3224 }
3225 
3226 int hns3_init_all_ring(struct hns3_nic_priv *priv)
3227 {
3228 	struct hnae3_handle *h = priv->ae_handle;
3229 	int ring_num = h->kinfo.num_tqps * 2;
3230 	int i, j;
3231 	int ret;
3232 
3233 	for (i = 0; i < ring_num; i++) {
3234 		ret = hns3_alloc_ring_memory(priv->ring_data[i].ring);
3235 		if (ret) {
3236 			dev_err(priv->dev,
3237 				"Alloc ring memory fail! ret=%d\n", ret);
3238 			goto out_when_alloc_ring_memory;
3239 		}
3240 
3241 		u64_stats_init(&priv->ring_data[i].ring->syncp);
3242 	}
3243 
3244 	return 0;
3245 
3246 out_when_alloc_ring_memory:
3247 	for (j = i - 1; j >= 0; j--)
3248 		hns3_fini_ring(priv->ring_data[j].ring);
3249 
3250 	return -ENOMEM;
3251 }
3252 
3253 int hns3_uninit_all_ring(struct hns3_nic_priv *priv)
3254 {
3255 	struct hnae3_handle *h = priv->ae_handle;
3256 	int i;
3257 
3258 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3259 		hns3_fini_ring(priv->ring_data[i].ring);
3260 		hns3_fini_ring(priv->ring_data[i + h->kinfo.num_tqps].ring);
3261 	}
3262 	return 0;
3263 }
3264 
3265 /* Set mac addr if it is configured. or leave it to the AE driver */
3266 static int hns3_init_mac_addr(struct net_device *netdev, bool init)
3267 {
3268 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3269 	struct hnae3_handle *h = priv->ae_handle;
3270 	u8 mac_addr_temp[ETH_ALEN];
3271 	int ret = 0;
3272 
3273 	if (h->ae_algo->ops->get_mac_addr && init) {
3274 		h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
3275 		ether_addr_copy(netdev->dev_addr, mac_addr_temp);
3276 	}
3277 
3278 	/* Check if the MAC address is valid, if not get a random one */
3279 	if (!is_valid_ether_addr(netdev->dev_addr)) {
3280 		eth_hw_addr_random(netdev);
3281 		dev_warn(priv->dev, "using random MAC address %pM\n",
3282 			 netdev->dev_addr);
3283 	}
3284 
3285 	if (h->ae_algo->ops->set_mac_addr)
3286 		ret = h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
3287 
3288 	return ret;
3289 }
3290 
3291 static int hns3_restore_fd_rules(struct net_device *netdev)
3292 {
3293 	struct hnae3_handle *h = hns3_get_handle(netdev);
3294 	int ret = 0;
3295 
3296 	if (h->ae_algo->ops->restore_fd_rules)
3297 		ret = h->ae_algo->ops->restore_fd_rules(h);
3298 
3299 	return ret;
3300 }
3301 
3302 static void hns3_del_all_fd_rules(struct net_device *netdev, bool clear_list)
3303 {
3304 	struct hnae3_handle *h = hns3_get_handle(netdev);
3305 
3306 	if (h->ae_algo->ops->del_all_fd_entries)
3307 		h->ae_algo->ops->del_all_fd_entries(h, clear_list);
3308 }
3309 
3310 static void hns3_nic_set_priv_ops(struct net_device *netdev)
3311 {
3312 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3313 
3314 	priv->ops.fill_desc = hns3_fill_desc;
3315 	if ((netdev->features & NETIF_F_TSO) ||
3316 	    (netdev->features & NETIF_F_TSO6))
3317 		priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tso;
3318 	else
3319 		priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tx;
3320 }
3321 
3322 static int hns3_client_init(struct hnae3_handle *handle)
3323 {
3324 	struct pci_dev *pdev = handle->pdev;
3325 	u16 alloc_tqps, max_rss_size;
3326 	struct hns3_nic_priv *priv;
3327 	struct net_device *netdev;
3328 	int ret;
3329 
3330 	handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
3331 						    &max_rss_size);
3332 	netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
3333 	if (!netdev)
3334 		return -ENOMEM;
3335 
3336 	priv = netdev_priv(netdev);
3337 	priv->dev = &pdev->dev;
3338 	priv->netdev = netdev;
3339 	priv->ae_handle = handle;
3340 	priv->ae_handle->last_reset_time = jiffies;
3341 	priv->tx_timeout_count = 0;
3342 
3343 	handle->kinfo.netdev = netdev;
3344 	handle->priv = (void *)priv;
3345 
3346 	hns3_init_mac_addr(netdev, true);
3347 
3348 	hns3_set_default_feature(netdev);
3349 
3350 	netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
3351 	netdev->priv_flags |= IFF_UNICAST_FLT;
3352 	netdev->netdev_ops = &hns3_nic_netdev_ops;
3353 	SET_NETDEV_DEV(netdev, &pdev->dev);
3354 	hns3_ethtool_set_ops(netdev);
3355 	hns3_nic_set_priv_ops(netdev);
3356 
3357 	/* Carrier off reporting is important to ethtool even BEFORE open */
3358 	netif_carrier_off(netdev);
3359 
3360 	if (handle->flags & HNAE3_SUPPORT_VF)
3361 		handle->reset_level = HNAE3_VF_RESET;
3362 	else
3363 		handle->reset_level = HNAE3_FUNC_RESET;
3364 
3365 	ret = hns3_get_ring_config(priv);
3366 	if (ret) {
3367 		ret = -ENOMEM;
3368 		goto out_get_ring_cfg;
3369 	}
3370 
3371 	ret = hns3_nic_alloc_vector_data(priv);
3372 	if (ret) {
3373 		ret = -ENOMEM;
3374 		goto out_alloc_vector_data;
3375 	}
3376 
3377 	ret = hns3_nic_init_vector_data(priv);
3378 	if (ret) {
3379 		ret = -ENOMEM;
3380 		goto out_init_vector_data;
3381 	}
3382 
3383 	ret = hns3_init_all_ring(priv);
3384 	if (ret) {
3385 		ret = -ENOMEM;
3386 		goto out_init_ring_data;
3387 	}
3388 
3389 	ret = register_netdev(netdev);
3390 	if (ret) {
3391 		dev_err(priv->dev, "probe register netdev fail!\n");
3392 		goto out_reg_netdev_fail;
3393 	}
3394 
3395 	hns3_dcbnl_setup(handle);
3396 
3397 	/* MTU range: (ETH_MIN_MTU(kernel default) - 9706) */
3398 	netdev->max_mtu = HNS3_MAX_MTU - (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
3399 
3400 	return ret;
3401 
3402 out_reg_netdev_fail:
3403 out_init_ring_data:
3404 	(void)hns3_nic_uninit_vector_data(priv);
3405 out_init_vector_data:
3406 	hns3_nic_dealloc_vector_data(priv);
3407 out_alloc_vector_data:
3408 	priv->ring_data = NULL;
3409 out_get_ring_cfg:
3410 	priv->ae_handle = NULL;
3411 	free_netdev(netdev);
3412 	return ret;
3413 }
3414 
3415 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
3416 {
3417 	struct net_device *netdev = handle->kinfo.netdev;
3418 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3419 	int ret;
3420 
3421 	hns3_remove_hw_addr(netdev);
3422 
3423 	if (netdev->reg_state != NETREG_UNINITIALIZED)
3424 		unregister_netdev(netdev);
3425 
3426 	hns3_del_all_fd_rules(netdev, true);
3427 
3428 	hns3_force_clear_all_rx_ring(handle);
3429 
3430 	ret = hns3_nic_uninit_vector_data(priv);
3431 	if (ret)
3432 		netdev_err(netdev, "uninit vector error\n");
3433 
3434 	ret = hns3_nic_dealloc_vector_data(priv);
3435 	if (ret)
3436 		netdev_err(netdev, "dealloc vector error\n");
3437 
3438 	ret = hns3_uninit_all_ring(priv);
3439 	if (ret)
3440 		netdev_err(netdev, "uninit ring error\n");
3441 
3442 	hns3_put_ring_config(priv);
3443 
3444 	priv->ring_data = NULL;
3445 
3446 	free_netdev(netdev);
3447 }
3448 
3449 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
3450 {
3451 	struct net_device *netdev = handle->kinfo.netdev;
3452 
3453 	if (!netdev)
3454 		return;
3455 
3456 	if (linkup) {
3457 		netif_carrier_on(netdev);
3458 		netif_tx_wake_all_queues(netdev);
3459 		netdev_info(netdev, "link up\n");
3460 	} else {
3461 		netif_carrier_off(netdev);
3462 		netif_tx_stop_all_queues(netdev);
3463 		netdev_info(netdev, "link down\n");
3464 	}
3465 }
3466 
3467 static int hns3_client_setup_tc(struct hnae3_handle *handle, u8 tc)
3468 {
3469 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
3470 	struct net_device *ndev = kinfo->netdev;
3471 	bool if_running;
3472 	int ret;
3473 
3474 	if (tc > HNAE3_MAX_TC)
3475 		return -EINVAL;
3476 
3477 	if (!ndev)
3478 		return -ENODEV;
3479 
3480 	if_running = netif_running(ndev);
3481 
3482 	if (if_running) {
3483 		(void)hns3_nic_net_stop(ndev);
3484 		msleep(100);
3485 	}
3486 
3487 	ret = (kinfo->dcb_ops && kinfo->dcb_ops->map_update) ?
3488 		kinfo->dcb_ops->map_update(handle) : -EOPNOTSUPP;
3489 	if (ret)
3490 		goto err_out;
3491 
3492 	ret = hns3_nic_set_real_num_queue(ndev);
3493 
3494 err_out:
3495 	if (if_running)
3496 		(void)hns3_nic_net_open(ndev);
3497 
3498 	return ret;
3499 }
3500 
3501 static int hns3_recover_hw_addr(struct net_device *ndev)
3502 {
3503 	struct netdev_hw_addr_list *list;
3504 	struct netdev_hw_addr *ha, *tmp;
3505 	int ret = 0;
3506 
3507 	/* go through and sync uc_addr entries to the device */
3508 	list = &ndev->uc;
3509 	list_for_each_entry_safe(ha, tmp, &list->list, list) {
3510 		ret = hns3_nic_uc_sync(ndev, ha->addr);
3511 		if (ret)
3512 			return ret;
3513 	}
3514 
3515 	/* go through and sync mc_addr entries to the device */
3516 	list = &ndev->mc;
3517 	list_for_each_entry_safe(ha, tmp, &list->list, list) {
3518 		ret = hns3_nic_mc_sync(ndev, ha->addr);
3519 		if (ret)
3520 			return ret;
3521 	}
3522 
3523 	return ret;
3524 }
3525 
3526 static void hns3_remove_hw_addr(struct net_device *netdev)
3527 {
3528 	struct netdev_hw_addr_list *list;
3529 	struct netdev_hw_addr *ha, *tmp;
3530 
3531 	hns3_nic_uc_unsync(netdev, netdev->dev_addr);
3532 
3533 	/* go through and unsync uc_addr entries to the device */
3534 	list = &netdev->uc;
3535 	list_for_each_entry_safe(ha, tmp, &list->list, list)
3536 		hns3_nic_uc_unsync(netdev, ha->addr);
3537 
3538 	/* go through and unsync mc_addr entries to the device */
3539 	list = &netdev->mc;
3540 	list_for_each_entry_safe(ha, tmp, &list->list, list)
3541 		if (ha->refcount > 1)
3542 			hns3_nic_mc_unsync(netdev, ha->addr);
3543 }
3544 
3545 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
3546 {
3547 	while (ring->next_to_clean != ring->next_to_use) {
3548 		ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
3549 		hns3_free_buffer_detach(ring, ring->next_to_clean);
3550 		ring_ptr_move_fw(ring, next_to_clean);
3551 	}
3552 }
3553 
3554 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
3555 {
3556 	struct hns3_desc_cb res_cbs;
3557 	int ret;
3558 
3559 	while (ring->next_to_use != ring->next_to_clean) {
3560 		/* When a buffer is not reused, it's memory has been
3561 		 * freed in hns3_handle_rx_bd or will be freed by
3562 		 * stack, so we need to replace the buffer here.
3563 		 */
3564 		if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
3565 			ret = hns3_reserve_buffer_map(ring, &res_cbs);
3566 			if (ret) {
3567 				u64_stats_update_begin(&ring->syncp);
3568 				ring->stats.sw_err_cnt++;
3569 				u64_stats_update_end(&ring->syncp);
3570 				/* if alloc new buffer fail, exit directly
3571 				 * and reclear in up flow.
3572 				 */
3573 				netdev_warn(ring->tqp->handle->kinfo.netdev,
3574 					    "reserve buffer map failed, ret = %d\n",
3575 					    ret);
3576 				return ret;
3577 			}
3578 			hns3_replace_buffer(ring, ring->next_to_use,
3579 					    &res_cbs);
3580 		}
3581 		ring_ptr_move_fw(ring, next_to_use);
3582 	}
3583 
3584 	return 0;
3585 }
3586 
3587 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
3588 {
3589 	while (ring->next_to_use != ring->next_to_clean) {
3590 		/* When a buffer is not reused, it's memory has been
3591 		 * freed in hns3_handle_rx_bd or will be freed by
3592 		 * stack, so only need to unmap the buffer here.
3593 		 */
3594 		if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
3595 			hns3_unmap_buffer(ring,
3596 					  &ring->desc_cb[ring->next_to_use]);
3597 			ring->desc_cb[ring->next_to_use].dma = 0;
3598 		}
3599 
3600 		ring_ptr_move_fw(ring, next_to_use);
3601 	}
3602 }
3603 
3604 static void hns3_force_clear_all_rx_ring(struct hnae3_handle *h)
3605 {
3606 	struct net_device *ndev = h->kinfo.netdev;
3607 	struct hns3_nic_priv *priv = netdev_priv(ndev);
3608 	struct hns3_enet_ring *ring;
3609 	u32 i;
3610 
3611 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3612 		ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
3613 		hns3_force_clear_rx_ring(ring);
3614 	}
3615 }
3616 
3617 static void hns3_clear_all_ring(struct hnae3_handle *h)
3618 {
3619 	struct net_device *ndev = h->kinfo.netdev;
3620 	struct hns3_nic_priv *priv = netdev_priv(ndev);
3621 	u32 i;
3622 
3623 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3624 		struct netdev_queue *dev_queue;
3625 		struct hns3_enet_ring *ring;
3626 
3627 		ring = priv->ring_data[i].ring;
3628 		hns3_clear_tx_ring(ring);
3629 		dev_queue = netdev_get_tx_queue(ndev,
3630 						priv->ring_data[i].queue_index);
3631 		netdev_tx_reset_queue(dev_queue);
3632 
3633 		ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
3634 		/* Continue to clear other rings even if clearing some
3635 		 * rings failed.
3636 		 */
3637 		hns3_clear_rx_ring(ring);
3638 	}
3639 }
3640 
3641 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
3642 {
3643 	struct net_device *ndev = h->kinfo.netdev;
3644 	struct hns3_nic_priv *priv = netdev_priv(ndev);
3645 	struct hns3_enet_ring *rx_ring;
3646 	int i, j;
3647 	int ret;
3648 
3649 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3650 		ret = h->ae_algo->ops->reset_queue(h, i);
3651 		if (ret)
3652 			return ret;
3653 
3654 		hns3_init_ring_hw(priv->ring_data[i].ring);
3655 
3656 		/* We need to clear tx ring here because self test will
3657 		 * use the ring and will not run down before up
3658 		 */
3659 		hns3_clear_tx_ring(priv->ring_data[i].ring);
3660 		priv->ring_data[i].ring->next_to_clean = 0;
3661 		priv->ring_data[i].ring->next_to_use = 0;
3662 
3663 		rx_ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
3664 		hns3_init_ring_hw(rx_ring);
3665 		ret = hns3_clear_rx_ring(rx_ring);
3666 		if (ret)
3667 			return ret;
3668 
3669 		/* We can not know the hardware head and tail when this
3670 		 * function is called in reset flow, so we reuse all desc.
3671 		 */
3672 		for (j = 0; j < rx_ring->desc_num; j++)
3673 			hns3_reuse_buffer(rx_ring, j);
3674 
3675 		rx_ring->next_to_clean = 0;
3676 		rx_ring->next_to_use = 0;
3677 	}
3678 
3679 	hns3_init_tx_ring_tc(priv);
3680 
3681 	return 0;
3682 }
3683 
3684 static void hns3_store_coal(struct hns3_nic_priv *priv)
3685 {
3686 	/* ethtool only support setting and querying one coal
3687 	 * configuation for now, so save the vector 0' coal
3688 	 * configuation here in order to restore it.
3689 	 */
3690 	memcpy(&priv->tx_coal, &priv->tqp_vector[0].tx_group.coal,
3691 	       sizeof(struct hns3_enet_coalesce));
3692 	memcpy(&priv->rx_coal, &priv->tqp_vector[0].rx_group.coal,
3693 	       sizeof(struct hns3_enet_coalesce));
3694 }
3695 
3696 static void hns3_restore_coal(struct hns3_nic_priv *priv)
3697 {
3698 	u16 vector_num = priv->vector_num;
3699 	int i;
3700 
3701 	for (i = 0; i < vector_num; i++) {
3702 		memcpy(&priv->tqp_vector[i].tx_group.coal, &priv->tx_coal,
3703 		       sizeof(struct hns3_enet_coalesce));
3704 		memcpy(&priv->tqp_vector[i].rx_group.coal, &priv->rx_coal,
3705 		       sizeof(struct hns3_enet_coalesce));
3706 	}
3707 }
3708 
3709 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
3710 {
3711 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
3712 	struct net_device *ndev = kinfo->netdev;
3713 
3714 	if (!netif_running(ndev))
3715 		return 0;
3716 
3717 	return hns3_nic_net_stop(ndev);
3718 }
3719 
3720 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
3721 {
3722 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
3723 	int ret = 0;
3724 
3725 	if (netif_running(kinfo->netdev)) {
3726 		ret = hns3_nic_net_up(kinfo->netdev);
3727 		if (ret) {
3728 			netdev_err(kinfo->netdev,
3729 				   "hns net up fail, ret=%d!\n", ret);
3730 			return ret;
3731 		}
3732 		handle->last_reset_time = jiffies;
3733 	}
3734 
3735 	return ret;
3736 }
3737 
3738 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
3739 {
3740 	struct net_device *netdev = handle->kinfo.netdev;
3741 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3742 	bool vlan_filter_enable;
3743 	int ret;
3744 
3745 	ret = hns3_init_mac_addr(netdev, false);
3746 	if (ret)
3747 		return ret;
3748 
3749 	ret = hns3_recover_hw_addr(netdev);
3750 	if (ret)
3751 		return ret;
3752 
3753 	ret = hns3_update_promisc_mode(netdev, handle->netdev_flags);
3754 	if (ret)
3755 		return ret;
3756 
3757 	vlan_filter_enable = netdev->flags & IFF_PROMISC ? false : true;
3758 	hns3_enable_vlan_filter(netdev, vlan_filter_enable);
3759 
3760 	/* Hardware table is only clear when pf resets */
3761 	if (!(handle->flags & HNAE3_SUPPORT_VF)) {
3762 		ret = hns3_restore_vlan(netdev);
3763 		if (ret)
3764 			return ret;
3765 	}
3766 
3767 	ret = hns3_restore_fd_rules(netdev);
3768 	if (ret)
3769 		return ret;
3770 
3771 	/* Carrier off reporting is important to ethtool even BEFORE open */
3772 	netif_carrier_off(netdev);
3773 
3774 	hns3_restore_coal(priv);
3775 
3776 	ret = hns3_nic_init_vector_data(priv);
3777 	if (ret)
3778 		return ret;
3779 
3780 	ret = hns3_init_all_ring(priv);
3781 	if (ret) {
3782 		hns3_nic_uninit_vector_data(priv);
3783 		priv->ring_data = NULL;
3784 	}
3785 
3786 	return ret;
3787 }
3788 
3789 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
3790 {
3791 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(handle->pdev);
3792 	struct net_device *netdev = handle->kinfo.netdev;
3793 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3794 	int ret;
3795 
3796 	hns3_force_clear_all_rx_ring(handle);
3797 
3798 	ret = hns3_nic_uninit_vector_data(priv);
3799 	if (ret) {
3800 		netdev_err(netdev, "uninit vector error\n");
3801 		return ret;
3802 	}
3803 
3804 	hns3_store_coal(priv);
3805 
3806 	ret = hns3_uninit_all_ring(priv);
3807 	if (ret)
3808 		netdev_err(netdev, "uninit ring error\n");
3809 
3810 	/* it is cumbersome for hardware to pick-and-choose entries for deletion
3811 	 * from table space. Hence, for function reset software intervention is
3812 	 * required to delete the entries
3813 	 */
3814 	if (hns3_dev_ongoing_func_reset(ae_dev)) {
3815 		hns3_remove_hw_addr(netdev);
3816 		hns3_del_all_fd_rules(netdev, false);
3817 	}
3818 
3819 	return ret;
3820 }
3821 
3822 static int hns3_reset_notify(struct hnae3_handle *handle,
3823 			     enum hnae3_reset_notify_type type)
3824 {
3825 	int ret = 0;
3826 
3827 	switch (type) {
3828 	case HNAE3_UP_CLIENT:
3829 		ret = hns3_reset_notify_up_enet(handle);
3830 		break;
3831 	case HNAE3_DOWN_CLIENT:
3832 		ret = hns3_reset_notify_down_enet(handle);
3833 		break;
3834 	case HNAE3_INIT_CLIENT:
3835 		ret = hns3_reset_notify_init_enet(handle);
3836 		break;
3837 	case HNAE3_UNINIT_CLIENT:
3838 		ret = hns3_reset_notify_uninit_enet(handle);
3839 		break;
3840 	default:
3841 		break;
3842 	}
3843 
3844 	return ret;
3845 }
3846 
3847 static int hns3_modify_tqp_num(struct net_device *netdev, u16 new_tqp_num)
3848 {
3849 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3850 	struct hnae3_handle *h = hns3_get_handle(netdev);
3851 	int ret;
3852 
3853 	ret = h->ae_algo->ops->set_channels(h, new_tqp_num);
3854 	if (ret)
3855 		return ret;
3856 
3857 	ret = hns3_get_ring_config(priv);
3858 	if (ret)
3859 		return ret;
3860 
3861 	ret = hns3_nic_alloc_vector_data(priv);
3862 	if (ret)
3863 		goto err_alloc_vector;
3864 
3865 	hns3_restore_coal(priv);
3866 
3867 	ret = hns3_nic_init_vector_data(priv);
3868 	if (ret)
3869 		goto err_uninit_vector;
3870 
3871 	ret = hns3_init_all_ring(priv);
3872 	if (ret)
3873 		goto err_put_ring;
3874 
3875 	return 0;
3876 
3877 err_put_ring:
3878 	hns3_put_ring_config(priv);
3879 err_uninit_vector:
3880 	hns3_nic_uninit_vector_data(priv);
3881 err_alloc_vector:
3882 	hns3_nic_dealloc_vector_data(priv);
3883 	return ret;
3884 }
3885 
3886 static int hns3_adjust_tqps_num(u8 num_tc, u32 new_tqp_num)
3887 {
3888 	return (new_tqp_num / num_tc) * num_tc;
3889 }
3890 
3891 int hns3_set_channels(struct net_device *netdev,
3892 		      struct ethtool_channels *ch)
3893 {
3894 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3895 	struct hnae3_handle *h = hns3_get_handle(netdev);
3896 	struct hnae3_knic_private_info *kinfo = &h->kinfo;
3897 	bool if_running = netif_running(netdev);
3898 	u32 new_tqp_num = ch->combined_count;
3899 	u16 org_tqp_num;
3900 	int ret;
3901 
3902 	if (ch->rx_count || ch->tx_count)
3903 		return -EINVAL;
3904 
3905 	if (new_tqp_num > hns3_get_max_available_channels(h) ||
3906 	    new_tqp_num < kinfo->num_tc) {
3907 		dev_err(&netdev->dev,
3908 			"Change tqps fail, the tqp range is from %d to %d",
3909 			kinfo->num_tc,
3910 			hns3_get_max_available_channels(h));
3911 		return -EINVAL;
3912 	}
3913 
3914 	new_tqp_num = hns3_adjust_tqps_num(kinfo->num_tc, new_tqp_num);
3915 	if (kinfo->num_tqps == new_tqp_num)
3916 		return 0;
3917 
3918 	if (if_running)
3919 		hns3_nic_net_stop(netdev);
3920 
3921 	ret = hns3_nic_uninit_vector_data(priv);
3922 	if (ret) {
3923 		dev_err(&netdev->dev,
3924 			"Unbind vector with tqp fail, nothing is changed");
3925 		goto open_netdev;
3926 	}
3927 
3928 	hns3_store_coal(priv);
3929 
3930 	hns3_nic_dealloc_vector_data(priv);
3931 
3932 	hns3_uninit_all_ring(priv);
3933 	hns3_put_ring_config(priv);
3934 
3935 	org_tqp_num = h->kinfo.num_tqps;
3936 	ret = hns3_modify_tqp_num(netdev, new_tqp_num);
3937 	if (ret) {
3938 		ret = hns3_modify_tqp_num(netdev, org_tqp_num);
3939 		if (ret) {
3940 			/* If revert to old tqp failed, fatal error occurred */
3941 			dev_err(&netdev->dev,
3942 				"Revert to old tqp num fail, ret=%d", ret);
3943 			return ret;
3944 		}
3945 		dev_info(&netdev->dev,
3946 			 "Change tqp num fail, Revert to old tqp num");
3947 	}
3948 
3949 open_netdev:
3950 	if (if_running)
3951 		hns3_nic_net_open(netdev);
3952 
3953 	return ret;
3954 }
3955 
3956 static const struct hnae3_client_ops client_ops = {
3957 	.init_instance = hns3_client_init,
3958 	.uninit_instance = hns3_client_uninit,
3959 	.link_status_change = hns3_link_status_change,
3960 	.setup_tc = hns3_client_setup_tc,
3961 	.reset_notify = hns3_reset_notify,
3962 };
3963 
3964 /* hns3_init_module - Driver registration routine
3965  * hns3_init_module is the first routine called when the driver is
3966  * loaded. All it does is register with the PCI subsystem.
3967  */
3968 static int __init hns3_init_module(void)
3969 {
3970 	int ret;
3971 
3972 	pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
3973 	pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
3974 
3975 	client.type = HNAE3_CLIENT_KNIC;
3976 	snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH - 1, "%s",
3977 		 hns3_driver_name);
3978 
3979 	client.ops = &client_ops;
3980 
3981 	INIT_LIST_HEAD(&client.node);
3982 
3983 	ret = hnae3_register_client(&client);
3984 	if (ret)
3985 		return ret;
3986 
3987 	ret = pci_register_driver(&hns3_driver);
3988 	if (ret)
3989 		hnae3_unregister_client(&client);
3990 
3991 	return ret;
3992 }
3993 module_init(hns3_init_module);
3994 
3995 /* hns3_exit_module - Driver exit cleanup routine
3996  * hns3_exit_module is called just before the driver is removed
3997  * from memory.
3998  */
3999 static void __exit hns3_exit_module(void)
4000 {
4001 	pci_unregister_driver(&hns3_driver);
4002 	hnae3_unregister_client(&client);
4003 }
4004 module_exit(hns3_exit_module);
4005 
4006 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
4007 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
4008 MODULE_LICENSE("GPL");
4009 MODULE_ALIAS("pci:hns-nic");
4010 MODULE_VERSION(HNS3_MOD_VERSION);
4011