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