xref: /openbmc/linux/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c (revision dd2934a95701576203b2f61e8ded4e4a2f9183ea)
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/skbuff.h>
13 #include <linux/sctp.h>
14 #include <linux/vermagic.h>
15 #include <net/gre.h>
16 #include <net/pkt_cls.h>
17 #include <net/vxlan.h>
18 
19 #include "hnae3.h"
20 #include "hns3_enet.h"
21 
22 static void hns3_clear_all_ring(struct hnae3_handle *h);
23 static void hns3_force_clear_all_rx_ring(struct hnae3_handle *h);
24 
25 static const char hns3_driver_name[] = "hns3";
26 const char hns3_driver_version[] = VERMAGIC_STRING;
27 static const char hns3_driver_string[] =
28 			"Hisilicon Ethernet Network Driver for Hip08 Family";
29 static const char hns3_copyright[] = "Copyright (c) 2017 Huawei Corporation.";
30 static struct hnae3_client client;
31 
32 /* hns3_pci_tbl - PCI Device ID Table
33  *
34  * Last entry must be all 0s
35  *
36  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
37  *   Class, Class Mask, private data (not used) }
38  */
39 static const struct pci_device_id hns3_pci_tbl[] = {
40 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
41 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
42 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA),
43 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
44 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC),
45 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
46 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA),
47 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
48 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC),
49 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
50 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC),
51 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
52 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_VF), 0},
53 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF),
54 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
55 	/* required last entry */
56 	{0, }
57 };
58 MODULE_DEVICE_TABLE(pci, hns3_pci_tbl);
59 
60 static irqreturn_t hns3_irq_handle(int irq, void *vector)
61 {
62 	struct hns3_enet_tqp_vector *tqp_vector = vector;
63 
64 	napi_schedule(&tqp_vector->napi);
65 
66 	return IRQ_HANDLED;
67 }
68 
69 static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv)
70 {
71 	struct hns3_enet_tqp_vector *tqp_vectors;
72 	unsigned int i;
73 
74 	for (i = 0; i < priv->vector_num; i++) {
75 		tqp_vectors = &priv->tqp_vector[i];
76 
77 		if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED)
78 			continue;
79 
80 		/* release the irq resource */
81 		free_irq(tqp_vectors->vector_irq, tqp_vectors);
82 		tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED;
83 	}
84 }
85 
86 static int hns3_nic_init_irq(struct hns3_nic_priv *priv)
87 {
88 	struct hns3_enet_tqp_vector *tqp_vectors;
89 	int txrx_int_idx = 0;
90 	int rx_int_idx = 0;
91 	int tx_int_idx = 0;
92 	unsigned int i;
93 	int ret;
94 
95 	for (i = 0; i < priv->vector_num; i++) {
96 		tqp_vectors = &priv->tqp_vector[i];
97 
98 		if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED)
99 			continue;
100 
101 		if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) {
102 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
103 				 "%s-%s-%d", priv->netdev->name, "TxRx",
104 				 txrx_int_idx++);
105 			txrx_int_idx++;
106 		} else if (tqp_vectors->rx_group.ring) {
107 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
108 				 "%s-%s-%d", priv->netdev->name, "Rx",
109 				 rx_int_idx++);
110 		} else if (tqp_vectors->tx_group.ring) {
111 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
112 				 "%s-%s-%d", priv->netdev->name, "Tx",
113 				 tx_int_idx++);
114 		} else {
115 			/* Skip this unused q_vector */
116 			continue;
117 		}
118 
119 		tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0';
120 
121 		ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0,
122 				  tqp_vectors->name,
123 				       tqp_vectors);
124 		if (ret) {
125 			netdev_err(priv->netdev, "request irq(%d) fail\n",
126 				   tqp_vectors->vector_irq);
127 			return ret;
128 		}
129 
130 		tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED;
131 	}
132 
133 	return 0;
134 }
135 
136 static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector,
137 				 u32 mask_en)
138 {
139 	writel(mask_en, tqp_vector->mask_addr);
140 }
141 
142 static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector)
143 {
144 	napi_enable(&tqp_vector->napi);
145 
146 	/* enable vector */
147 	hns3_mask_vector_irq(tqp_vector, 1);
148 }
149 
150 static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector)
151 {
152 	/* disable vector */
153 	hns3_mask_vector_irq(tqp_vector, 0);
154 
155 	disable_irq(tqp_vector->vector_irq);
156 	napi_disable(&tqp_vector->napi);
157 }
158 
159 void hns3_set_vector_coalesce_rl(struct hns3_enet_tqp_vector *tqp_vector,
160 				 u32 rl_value)
161 {
162 	u32 rl_reg = hns3_rl_usec_to_reg(rl_value);
163 
164 	/* this defines the configuration for RL (Interrupt Rate Limiter).
165 	 * Rl defines rate of interrupts i.e. number of interrupts-per-second
166 	 * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing
167 	 */
168 
169 	if (rl_reg > 0 && !tqp_vector->tx_group.coal.gl_adapt_enable &&
170 	    !tqp_vector->rx_group.coal.gl_adapt_enable)
171 		/* According to the hardware, the range of rl_reg is
172 		 * 0-59 and the unit is 4.
173 		 */
174 		rl_reg |=  HNS3_INT_RL_ENABLE_MASK;
175 
176 	writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
177 }
178 
179 void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
180 				    u32 gl_value)
181 {
182 	u32 rx_gl_reg = hns3_gl_usec_to_reg(gl_value);
183 
184 	writel(rx_gl_reg, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
185 }
186 
187 void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
188 				    u32 gl_value)
189 {
190 	u32 tx_gl_reg = hns3_gl_usec_to_reg(gl_value);
191 
192 	writel(tx_gl_reg, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
193 }
194 
195 static void hns3_vector_gl_rl_init(struct hns3_enet_tqp_vector *tqp_vector,
196 				   struct hns3_nic_priv *priv)
197 {
198 	struct hnae3_handle *h = priv->ae_handle;
199 
200 	/* initialize the configuration for interrupt coalescing.
201 	 * 1. GL (Interrupt Gap Limiter)
202 	 * 2. RL (Interrupt Rate Limiter)
203 	 */
204 
205 	/* Default: enable interrupt coalescing self-adaptive and GL */
206 	tqp_vector->tx_group.coal.gl_adapt_enable = 1;
207 	tqp_vector->rx_group.coal.gl_adapt_enable = 1;
208 
209 	tqp_vector->tx_group.coal.int_gl = HNS3_INT_GL_50K;
210 	tqp_vector->rx_group.coal.int_gl = HNS3_INT_GL_50K;
211 
212 	/* Default: disable RL */
213 	h->kinfo.int_rl_setting = 0;
214 
215 	tqp_vector->int_adapt_down = HNS3_INT_ADAPT_DOWN_START;
216 	tqp_vector->rx_group.coal.flow_level = HNS3_FLOW_LOW;
217 	tqp_vector->tx_group.coal.flow_level = HNS3_FLOW_LOW;
218 }
219 
220 static void hns3_vector_gl_rl_init_hw(struct hns3_enet_tqp_vector *tqp_vector,
221 				      struct hns3_nic_priv *priv)
222 {
223 	struct hnae3_handle *h = priv->ae_handle;
224 
225 	hns3_set_vector_coalesce_tx_gl(tqp_vector,
226 				       tqp_vector->tx_group.coal.int_gl);
227 	hns3_set_vector_coalesce_rx_gl(tqp_vector,
228 				       tqp_vector->rx_group.coal.int_gl);
229 	hns3_set_vector_coalesce_rl(tqp_vector, h->kinfo.int_rl_setting);
230 }
231 
232 static int hns3_nic_set_real_num_queue(struct net_device *netdev)
233 {
234 	struct hnae3_handle *h = hns3_get_handle(netdev);
235 	struct hnae3_knic_private_info *kinfo = &h->kinfo;
236 	unsigned int queue_size = kinfo->rss_size * kinfo->num_tc;
237 	int i, ret;
238 
239 	if (kinfo->num_tc <= 1) {
240 		netdev_reset_tc(netdev);
241 	} else {
242 		ret = netdev_set_num_tc(netdev, kinfo->num_tc);
243 		if (ret) {
244 			netdev_err(netdev,
245 				   "netdev_set_num_tc fail, ret=%d!\n", ret);
246 			return ret;
247 		}
248 
249 		for (i = 0; i < HNAE3_MAX_TC; i++) {
250 			if (!kinfo->tc_info[i].enable)
251 				continue;
252 
253 			netdev_set_tc_queue(netdev,
254 					    kinfo->tc_info[i].tc,
255 					    kinfo->tc_info[i].tqp_count,
256 					    kinfo->tc_info[i].tqp_offset);
257 		}
258 	}
259 
260 	ret = netif_set_real_num_tx_queues(netdev, queue_size);
261 	if (ret) {
262 		netdev_err(netdev,
263 			   "netif_set_real_num_tx_queues fail, ret=%d!\n",
264 			   ret);
265 		return ret;
266 	}
267 
268 	ret = netif_set_real_num_rx_queues(netdev, queue_size);
269 	if (ret) {
270 		netdev_err(netdev,
271 			   "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
272 		return ret;
273 	}
274 
275 	return 0;
276 }
277 
278 static u16 hns3_get_max_available_channels(struct hnae3_handle *h)
279 {
280 	u16 free_tqps, max_rss_size, max_tqps;
281 
282 	h->ae_algo->ops->get_tqps_and_rss_info(h, &free_tqps, &max_rss_size);
283 	max_tqps = h->kinfo.num_tc * max_rss_size;
284 
285 	return min_t(u16, max_tqps, (free_tqps + h->kinfo.num_tqps));
286 }
287 
288 static int hns3_nic_net_up(struct net_device *netdev)
289 {
290 	struct hns3_nic_priv *priv = netdev_priv(netdev);
291 	struct hnae3_handle *h = priv->ae_handle;
292 	int i, j;
293 	int ret;
294 
295 	ret = hns3_nic_reset_all_ring(h);
296 	if (ret)
297 		return ret;
298 
299 	/* get irq resource for all vectors */
300 	ret = hns3_nic_init_irq(priv);
301 	if (ret) {
302 		netdev_err(netdev, "hns init irq failed! ret=%d\n", ret);
303 		return ret;
304 	}
305 
306 	/* enable the vectors */
307 	for (i = 0; i < priv->vector_num; i++)
308 		hns3_vector_enable(&priv->tqp_vector[i]);
309 
310 	/* start the ae_dev */
311 	ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0;
312 	if (ret)
313 		goto out_start_err;
314 
315 	clear_bit(HNS3_NIC_STATE_DOWN, &priv->state);
316 
317 	return 0;
318 
319 out_start_err:
320 	for (j = i - 1; j >= 0; j--)
321 		hns3_vector_disable(&priv->tqp_vector[j]);
322 
323 	hns3_nic_uninit_irq(priv);
324 
325 	return ret;
326 }
327 
328 static int hns3_nic_net_open(struct net_device *netdev)
329 {
330 	struct hns3_nic_priv *priv = netdev_priv(netdev);
331 	struct hnae3_handle *h = hns3_get_handle(netdev);
332 	struct hnae3_knic_private_info *kinfo;
333 	int i, ret;
334 
335 	netif_carrier_off(netdev);
336 
337 	ret = hns3_nic_set_real_num_queue(netdev);
338 	if (ret)
339 		return ret;
340 
341 	ret = hns3_nic_net_up(netdev);
342 	if (ret) {
343 		netdev_err(netdev,
344 			   "hns net up fail, ret=%d!\n", ret);
345 		return ret;
346 	}
347 
348 	kinfo = &h->kinfo;
349 	for (i = 0; i < HNAE3_MAX_USER_PRIO; i++) {
350 		netdev_set_prio_tc_map(netdev, i,
351 				       kinfo->prio_tc[i]);
352 	}
353 
354 	priv->ae_handle->last_reset_time = jiffies;
355 	return 0;
356 }
357 
358 static void hns3_nic_net_down(struct net_device *netdev)
359 {
360 	struct hns3_nic_priv *priv = netdev_priv(netdev);
361 	const struct hnae3_ae_ops *ops;
362 	int i;
363 
364 	if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
365 		return;
366 
367 	/* disable vectors */
368 	for (i = 0; i < priv->vector_num; i++)
369 		hns3_vector_disable(&priv->tqp_vector[i]);
370 
371 	/* stop ae_dev */
372 	ops = priv->ae_handle->ae_algo->ops;
373 	if (ops->stop)
374 		ops->stop(priv->ae_handle);
375 
376 	/* free irq resources */
377 	hns3_nic_uninit_irq(priv);
378 
379 	hns3_clear_all_ring(priv->ae_handle);
380 }
381 
382 static int hns3_nic_net_stop(struct net_device *netdev)
383 {
384 	netif_tx_stop_all_queues(netdev);
385 	netif_carrier_off(netdev);
386 
387 	hns3_nic_net_down(netdev);
388 
389 	return 0;
390 }
391 
392 static int hns3_nic_uc_sync(struct net_device *netdev,
393 			    const unsigned char *addr)
394 {
395 	struct hnae3_handle *h = hns3_get_handle(netdev);
396 
397 	if (h->ae_algo->ops->add_uc_addr)
398 		return h->ae_algo->ops->add_uc_addr(h, addr);
399 
400 	return 0;
401 }
402 
403 static int hns3_nic_uc_unsync(struct net_device *netdev,
404 			      const unsigned char *addr)
405 {
406 	struct hnae3_handle *h = hns3_get_handle(netdev);
407 
408 	if (h->ae_algo->ops->rm_uc_addr)
409 		return h->ae_algo->ops->rm_uc_addr(h, addr);
410 
411 	return 0;
412 }
413 
414 static int hns3_nic_mc_sync(struct net_device *netdev,
415 			    const unsigned char *addr)
416 {
417 	struct hnae3_handle *h = hns3_get_handle(netdev);
418 
419 	if (h->ae_algo->ops->add_mc_addr)
420 		return h->ae_algo->ops->add_mc_addr(h, addr);
421 
422 	return 0;
423 }
424 
425 static int hns3_nic_mc_unsync(struct net_device *netdev,
426 			      const unsigned char *addr)
427 {
428 	struct hnae3_handle *h = hns3_get_handle(netdev);
429 
430 	if (h->ae_algo->ops->rm_mc_addr)
431 		return h->ae_algo->ops->rm_mc_addr(h, addr);
432 
433 	return 0;
434 }
435 
436 static void hns3_nic_set_rx_mode(struct net_device *netdev)
437 {
438 	struct hnae3_handle *h = hns3_get_handle(netdev);
439 
440 	if (h->ae_algo->ops->set_promisc_mode) {
441 		if (netdev->flags & IFF_PROMISC)
442 			h->ae_algo->ops->set_promisc_mode(h, true, true);
443 		else if (netdev->flags & IFF_ALLMULTI)
444 			h->ae_algo->ops->set_promisc_mode(h, false, true);
445 		else
446 			h->ae_algo->ops->set_promisc_mode(h, false, false);
447 	}
448 	if (__dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync))
449 		netdev_err(netdev, "sync uc address fail\n");
450 	if (netdev->flags & IFF_MULTICAST) {
451 		if (__dev_mc_sync(netdev, hns3_nic_mc_sync, hns3_nic_mc_unsync))
452 			netdev_err(netdev, "sync mc address fail\n");
453 
454 		if (h->ae_algo->ops->update_mta_status)
455 			h->ae_algo->ops->update_mta_status(h);
456 	}
457 }
458 
459 static int hns3_set_tso(struct sk_buff *skb, u32 *paylen,
460 			u16 *mss, u32 *type_cs_vlan_tso)
461 {
462 	u32 l4_offset, hdr_len;
463 	union l3_hdr_info l3;
464 	union l4_hdr_info l4;
465 	u32 l4_paylen;
466 	int ret;
467 
468 	if (!skb_is_gso(skb))
469 		return 0;
470 
471 	ret = skb_cow_head(skb, 0);
472 	if (ret)
473 		return ret;
474 
475 	l3.hdr = skb_network_header(skb);
476 	l4.hdr = skb_transport_header(skb);
477 
478 	/* Software should clear the IPv4's checksum field when tso is
479 	 * needed.
480 	 */
481 	if (l3.v4->version == 4)
482 		l3.v4->check = 0;
483 
484 	/* tunnel packet.*/
485 	if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
486 					 SKB_GSO_GRE_CSUM |
487 					 SKB_GSO_UDP_TUNNEL |
488 					 SKB_GSO_UDP_TUNNEL_CSUM)) {
489 		if ((!(skb_shinfo(skb)->gso_type &
490 		    SKB_GSO_PARTIAL)) &&
491 		    (skb_shinfo(skb)->gso_type &
492 		    SKB_GSO_UDP_TUNNEL_CSUM)) {
493 			/* Software should clear the udp's checksum
494 			 * field when tso is needed.
495 			 */
496 			l4.udp->check = 0;
497 		}
498 		/* reset l3&l4 pointers from outer to inner headers */
499 		l3.hdr = skb_inner_network_header(skb);
500 		l4.hdr = skb_inner_transport_header(skb);
501 
502 		/* Software should clear the IPv4's checksum field when
503 		 * tso is needed.
504 		 */
505 		if (l3.v4->version == 4)
506 			l3.v4->check = 0;
507 	}
508 
509 	/* normal or tunnel packet*/
510 	l4_offset = l4.hdr - skb->data;
511 	hdr_len = (l4.tcp->doff * 4) + l4_offset;
512 
513 	/* remove payload length from inner pseudo checksum when tso*/
514 	l4_paylen = skb->len - l4_offset;
515 	csum_replace_by_diff(&l4.tcp->check,
516 			     (__force __wsum)htonl(l4_paylen));
517 
518 	/* find the txbd field values */
519 	*paylen = skb->len - hdr_len;
520 	hnae3_set_bit(*type_cs_vlan_tso,
521 		      HNS3_TXD_TSO_B, 1);
522 
523 	/* get MSS for TSO */
524 	*mss = skb_shinfo(skb)->gso_size;
525 
526 	return 0;
527 }
528 
529 static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
530 				u8 *il4_proto)
531 {
532 	union {
533 		struct iphdr *v4;
534 		struct ipv6hdr *v6;
535 		unsigned char *hdr;
536 	} l3;
537 	unsigned char *l4_hdr;
538 	unsigned char *exthdr;
539 	u8 l4_proto_tmp;
540 	__be16 frag_off;
541 
542 	/* find outer header point */
543 	l3.hdr = skb_network_header(skb);
544 	l4_hdr = skb_transport_header(skb);
545 
546 	if (skb->protocol == htons(ETH_P_IPV6)) {
547 		exthdr = l3.hdr + sizeof(*l3.v6);
548 		l4_proto_tmp = l3.v6->nexthdr;
549 		if (l4_hdr != exthdr)
550 			ipv6_skip_exthdr(skb, exthdr - skb->data,
551 					 &l4_proto_tmp, &frag_off);
552 	} else if (skb->protocol == htons(ETH_P_IP)) {
553 		l4_proto_tmp = l3.v4->protocol;
554 	} else {
555 		return -EINVAL;
556 	}
557 
558 	*ol4_proto = l4_proto_tmp;
559 
560 	/* tunnel packet */
561 	if (!skb->encapsulation) {
562 		*il4_proto = 0;
563 		return 0;
564 	}
565 
566 	/* find inner header point */
567 	l3.hdr = skb_inner_network_header(skb);
568 	l4_hdr = skb_inner_transport_header(skb);
569 
570 	if (l3.v6->version == 6) {
571 		exthdr = l3.hdr + sizeof(*l3.v6);
572 		l4_proto_tmp = l3.v6->nexthdr;
573 		if (l4_hdr != exthdr)
574 			ipv6_skip_exthdr(skb, exthdr - skb->data,
575 					 &l4_proto_tmp, &frag_off);
576 	} else if (l3.v4->version == 4) {
577 		l4_proto_tmp = l3.v4->protocol;
578 	}
579 
580 	*il4_proto = l4_proto_tmp;
581 
582 	return 0;
583 }
584 
585 static void hns3_set_l2l3l4_len(struct sk_buff *skb, u8 ol4_proto,
586 				u8 il4_proto, u32 *type_cs_vlan_tso,
587 				u32 *ol_type_vlan_len_msec)
588 {
589 	union {
590 		struct iphdr *v4;
591 		struct ipv6hdr *v6;
592 		unsigned char *hdr;
593 	} l3;
594 	union {
595 		struct tcphdr *tcp;
596 		struct udphdr *udp;
597 		struct gre_base_hdr *gre;
598 		unsigned char *hdr;
599 	} l4;
600 	unsigned char *l2_hdr;
601 	u8 l4_proto = ol4_proto;
602 	u32 ol2_len;
603 	u32 ol3_len;
604 	u32 ol4_len;
605 	u32 l2_len;
606 	u32 l3_len;
607 
608 	l3.hdr = skb_network_header(skb);
609 	l4.hdr = skb_transport_header(skb);
610 
611 	/* compute L2 header size for normal packet, defined in 2 Bytes */
612 	l2_len = l3.hdr - skb->data;
613 	hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_M,
614 			HNS3_TXD_L2LEN_S, l2_len >> 1);
615 
616 	/* tunnel packet*/
617 	if (skb->encapsulation) {
618 		/* compute OL2 header size, defined in 2 Bytes */
619 		ol2_len = l2_len;
620 		hnae3_set_field(*ol_type_vlan_len_msec,
621 				HNS3_TXD_L2LEN_M,
622 				HNS3_TXD_L2LEN_S, ol2_len >> 1);
623 
624 		/* compute OL3 header size, defined in 4 Bytes */
625 		ol3_len = l4.hdr - l3.hdr;
626 		hnae3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_M,
627 				HNS3_TXD_L3LEN_S, ol3_len >> 2);
628 
629 		/* MAC in UDP, MAC in GRE (0x6558)*/
630 		if ((ol4_proto == IPPROTO_UDP) || (ol4_proto == IPPROTO_GRE)) {
631 			/* switch MAC header ptr from outer to inner header.*/
632 			l2_hdr = skb_inner_mac_header(skb);
633 
634 			/* compute OL4 header size, defined in 4 Bytes. */
635 			ol4_len = l2_hdr - l4.hdr;
636 			hnae3_set_field(*ol_type_vlan_len_msec,
637 					HNS3_TXD_L4LEN_M, HNS3_TXD_L4LEN_S,
638 					ol4_len >> 2);
639 
640 			/* switch IP header ptr from outer to inner header */
641 			l3.hdr = skb_inner_network_header(skb);
642 
643 			/* compute inner l2 header size, defined in 2 Bytes. */
644 			l2_len = l3.hdr - l2_hdr;
645 			hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_M,
646 					HNS3_TXD_L2LEN_S, l2_len >> 1);
647 		} else {
648 			/* skb packet types not supported by hardware,
649 			 * txbd len fild doesn't be filled.
650 			 */
651 			return;
652 		}
653 
654 		/* switch L4 header pointer from outer to inner */
655 		l4.hdr = skb_inner_transport_header(skb);
656 
657 		l4_proto = il4_proto;
658 	}
659 
660 	/* compute inner(/normal) L3 header size, defined in 4 Bytes */
661 	l3_len = l4.hdr - l3.hdr;
662 	hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_M,
663 			HNS3_TXD_L3LEN_S, l3_len >> 2);
664 
665 	/* compute inner(/normal) L4 header size, defined in 4 Bytes */
666 	switch (l4_proto) {
667 	case IPPROTO_TCP:
668 		hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_M,
669 				HNS3_TXD_L4LEN_S, l4.tcp->doff);
670 		break;
671 	case IPPROTO_SCTP:
672 		hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_M,
673 				HNS3_TXD_L4LEN_S,
674 				(sizeof(struct sctphdr) >> 2));
675 		break;
676 	case IPPROTO_UDP:
677 		hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_M,
678 				HNS3_TXD_L4LEN_S,
679 				(sizeof(struct udphdr) >> 2));
680 		break;
681 	default:
682 		/* skb packet types not supported by hardware,
683 		 * txbd len fild doesn't be filled.
684 		 */
685 		return;
686 	}
687 }
688 
689 /* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
690  * and it is udp packet, which has a dest port as the IANA assigned.
691  * the hardware is expected to do the checksum offload, but the
692  * hardware will not do the checksum offload when udp dest port is
693  * 4789.
694  */
695 static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
696 {
697 #define IANA_VXLAN_PORT	4789
698 	union {
699 		struct tcphdr *tcp;
700 		struct udphdr *udp;
701 		struct gre_base_hdr *gre;
702 		unsigned char *hdr;
703 	} l4;
704 
705 	l4.hdr = skb_transport_header(skb);
706 
707 	if (!(!skb->encapsulation && l4.udp->dest == htons(IANA_VXLAN_PORT)))
708 		return false;
709 
710 	skb_checksum_help(skb);
711 
712 	return true;
713 }
714 
715 static int hns3_set_l3l4_type_csum(struct sk_buff *skb, u8 ol4_proto,
716 				   u8 il4_proto, u32 *type_cs_vlan_tso,
717 				   u32 *ol_type_vlan_len_msec)
718 {
719 	union {
720 		struct iphdr *v4;
721 		struct ipv6hdr *v6;
722 		unsigned char *hdr;
723 	} l3;
724 	u32 l4_proto = ol4_proto;
725 
726 	l3.hdr = skb_network_header(skb);
727 
728 	/* define OL3 type and tunnel type(OL4).*/
729 	if (skb->encapsulation) {
730 		/* define outer network header type.*/
731 		if (skb->protocol == htons(ETH_P_IP)) {
732 			if (skb_is_gso(skb))
733 				hnae3_set_field(*ol_type_vlan_len_msec,
734 						HNS3_TXD_OL3T_M,
735 						HNS3_TXD_OL3T_S,
736 						HNS3_OL3T_IPV4_CSUM);
737 			else
738 				hnae3_set_field(*ol_type_vlan_len_msec,
739 						HNS3_TXD_OL3T_M,
740 						HNS3_TXD_OL3T_S,
741 						HNS3_OL3T_IPV4_NO_CSUM);
742 
743 		} else if (skb->protocol == htons(ETH_P_IPV6)) {
744 			hnae3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_M,
745 					HNS3_TXD_OL3T_S, HNS3_OL3T_IPV6);
746 		}
747 
748 		/* define tunnel type(OL4).*/
749 		switch (l4_proto) {
750 		case IPPROTO_UDP:
751 			hnae3_set_field(*ol_type_vlan_len_msec,
752 					HNS3_TXD_TUNTYPE_M,
753 					HNS3_TXD_TUNTYPE_S,
754 					HNS3_TUN_MAC_IN_UDP);
755 			break;
756 		case IPPROTO_GRE:
757 			hnae3_set_field(*ol_type_vlan_len_msec,
758 					HNS3_TXD_TUNTYPE_M,
759 					HNS3_TXD_TUNTYPE_S,
760 					HNS3_TUN_NVGRE);
761 			break;
762 		default:
763 			/* drop the skb tunnel packet if hardware don't support,
764 			 * because hardware can't calculate csum when TSO.
765 			 */
766 			if (skb_is_gso(skb))
767 				return -EDOM;
768 
769 			/* the stack computes the IP header already,
770 			 * driver calculate l4 checksum when not TSO.
771 			 */
772 			skb_checksum_help(skb);
773 			return 0;
774 		}
775 
776 		l3.hdr = skb_inner_network_header(skb);
777 		l4_proto = il4_proto;
778 	}
779 
780 	if (l3.v4->version == 4) {
781 		hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_M,
782 				HNS3_TXD_L3T_S, HNS3_L3T_IPV4);
783 
784 		/* the stack computes the IP header already, the only time we
785 		 * need the hardware to recompute it is in the case of TSO.
786 		 */
787 		if (skb_is_gso(skb))
788 			hnae3_set_bit(*type_cs_vlan_tso, HNS3_TXD_L3CS_B, 1);
789 	} else if (l3.v6->version == 6) {
790 		hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_M,
791 				HNS3_TXD_L3T_S, HNS3_L3T_IPV6);
792 	}
793 
794 	switch (l4_proto) {
795 	case IPPROTO_TCP:
796 		hnae3_set_bit(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
797 		hnae3_set_field(*type_cs_vlan_tso,
798 				HNS3_TXD_L4T_M,
799 				HNS3_TXD_L4T_S,
800 				HNS3_L4T_TCP);
801 		break;
802 	case IPPROTO_UDP:
803 		if (hns3_tunnel_csum_bug(skb))
804 			break;
805 
806 		hnae3_set_bit(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
807 		hnae3_set_field(*type_cs_vlan_tso,
808 				HNS3_TXD_L4T_M,
809 				HNS3_TXD_L4T_S,
810 				HNS3_L4T_UDP);
811 		break;
812 	case IPPROTO_SCTP:
813 		hnae3_set_bit(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
814 		hnae3_set_field(*type_cs_vlan_tso,
815 				HNS3_TXD_L4T_M,
816 				HNS3_TXD_L4T_S,
817 				HNS3_L4T_SCTP);
818 		break;
819 	default:
820 		/* drop the skb tunnel packet if hardware don't support,
821 		 * because hardware can't calculate csum when TSO.
822 		 */
823 		if (skb_is_gso(skb))
824 			return -EDOM;
825 
826 		/* the stack computes the IP header already,
827 		 * driver calculate l4 checksum when not TSO.
828 		 */
829 		skb_checksum_help(skb);
830 		return 0;
831 	}
832 
833 	return 0;
834 }
835 
836 static void hns3_set_txbd_baseinfo(u16 *bdtp_fe_sc_vld_ra_ri, int frag_end)
837 {
838 	/* Config bd buffer end */
839 	hnae3_set_field(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_BDTYPE_M,
840 			HNS3_TXD_BDTYPE_S, 0);
841 	hnae3_set_bit(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_FE_B, !!frag_end);
842 	hnae3_set_bit(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_VLD_B, 1);
843 	hnae3_set_field(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_SC_M, HNS3_TXD_SC_S, 0);
844 }
845 
846 static int hns3_fill_desc_vtags(struct sk_buff *skb,
847 				struct hns3_enet_ring *tx_ring,
848 				u32 *inner_vlan_flag,
849 				u32 *out_vlan_flag,
850 				u16 *inner_vtag,
851 				u16 *out_vtag)
852 {
853 #define HNS3_TX_VLAN_PRIO_SHIFT 13
854 
855 	if (skb->protocol == htons(ETH_P_8021Q) &&
856 	    !(tx_ring->tqp->handle->kinfo.netdev->features &
857 	    NETIF_F_HW_VLAN_CTAG_TX)) {
858 		/* When HW VLAN acceleration is turned off, and the stack
859 		 * sets the protocol to 802.1q, the driver just need to
860 		 * set the protocol to the encapsulated ethertype.
861 		 */
862 		skb->protocol = vlan_get_protocol(skb);
863 		return 0;
864 	}
865 
866 	if (skb_vlan_tag_present(skb)) {
867 		u16 vlan_tag;
868 
869 		vlan_tag = skb_vlan_tag_get(skb);
870 		vlan_tag |= (skb->priority & 0x7) << HNS3_TX_VLAN_PRIO_SHIFT;
871 
872 		/* Based on hw strategy, use out_vtag in two layer tag case,
873 		 * and use inner_vtag in one tag case.
874 		 */
875 		if (skb->protocol == htons(ETH_P_8021Q)) {
876 			hnae3_set_bit(*out_vlan_flag, HNS3_TXD_OVLAN_B, 1);
877 			*out_vtag = vlan_tag;
878 		} else {
879 			hnae3_set_bit(*inner_vlan_flag, HNS3_TXD_VLAN_B, 1);
880 			*inner_vtag = vlan_tag;
881 		}
882 	} else if (skb->protocol == htons(ETH_P_8021Q)) {
883 		struct vlan_ethhdr *vhdr;
884 		int rc;
885 
886 		rc = skb_cow_head(skb, 0);
887 		if (rc < 0)
888 			return rc;
889 		vhdr = (struct vlan_ethhdr *)skb->data;
890 		vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority & 0x7)
891 					<< HNS3_TX_VLAN_PRIO_SHIFT);
892 	}
893 
894 	skb->protocol = vlan_get_protocol(skb);
895 	return 0;
896 }
897 
898 static int hns3_fill_desc(struct hns3_enet_ring *ring, void *priv,
899 			  int size, dma_addr_t dma, int frag_end,
900 			  enum hns_desc_type type)
901 {
902 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
903 	struct hns3_desc *desc = &ring->desc[ring->next_to_use];
904 	u32 ol_type_vlan_len_msec = 0;
905 	u16 bdtp_fe_sc_vld_ra_ri = 0;
906 	u32 type_cs_vlan_tso = 0;
907 	struct sk_buff *skb;
908 	u16 inner_vtag = 0;
909 	u16 out_vtag = 0;
910 	u32 paylen = 0;
911 	u16 mss = 0;
912 	u8 ol4_proto;
913 	u8 il4_proto;
914 	int ret;
915 
916 	/* The txbd's baseinfo of DESC_TYPE_PAGE & DESC_TYPE_SKB */
917 	desc_cb->priv = priv;
918 	desc_cb->length = size;
919 	desc_cb->dma = dma;
920 	desc_cb->type = type;
921 
922 	/* now, fill the descriptor */
923 	desc->addr = cpu_to_le64(dma);
924 	desc->tx.send_size = cpu_to_le16((u16)size);
925 	hns3_set_txbd_baseinfo(&bdtp_fe_sc_vld_ra_ri, frag_end);
926 	desc->tx.bdtp_fe_sc_vld_ra_ri = cpu_to_le16(bdtp_fe_sc_vld_ra_ri);
927 
928 	if (type == DESC_TYPE_SKB) {
929 		skb = (struct sk_buff *)priv;
930 		paylen = skb->len;
931 
932 		ret = hns3_fill_desc_vtags(skb, ring, &type_cs_vlan_tso,
933 					   &ol_type_vlan_len_msec,
934 					   &inner_vtag, &out_vtag);
935 		if (unlikely(ret))
936 			return ret;
937 
938 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
939 			skb_reset_mac_len(skb);
940 
941 			ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
942 			if (ret)
943 				return ret;
944 			hns3_set_l2l3l4_len(skb, ol4_proto, il4_proto,
945 					    &type_cs_vlan_tso,
946 					    &ol_type_vlan_len_msec);
947 			ret = hns3_set_l3l4_type_csum(skb, ol4_proto, il4_proto,
948 						      &type_cs_vlan_tso,
949 						      &ol_type_vlan_len_msec);
950 			if (ret)
951 				return ret;
952 
953 			ret = hns3_set_tso(skb, &paylen, &mss,
954 					   &type_cs_vlan_tso);
955 			if (ret)
956 				return ret;
957 		}
958 
959 		/* Set txbd */
960 		desc->tx.ol_type_vlan_len_msec =
961 			cpu_to_le32(ol_type_vlan_len_msec);
962 		desc->tx.type_cs_vlan_tso_len =
963 			cpu_to_le32(type_cs_vlan_tso);
964 		desc->tx.paylen = cpu_to_le32(paylen);
965 		desc->tx.mss = cpu_to_le16(mss);
966 		desc->tx.vlan_tag = cpu_to_le16(inner_vtag);
967 		desc->tx.outer_vlan_tag = cpu_to_le16(out_vtag);
968 	}
969 
970 	/* move ring pointer to next.*/
971 	ring_ptr_move_fw(ring, next_to_use);
972 
973 	return 0;
974 }
975 
976 static int hns3_fill_desc_tso(struct hns3_enet_ring *ring, void *priv,
977 			      int size, dma_addr_t dma, int frag_end,
978 			      enum hns_desc_type type)
979 {
980 	unsigned int frag_buf_num;
981 	unsigned int k;
982 	int sizeoflast;
983 	int ret;
984 
985 	frag_buf_num = (size + HNS3_MAX_BD_SIZE - 1) / HNS3_MAX_BD_SIZE;
986 	sizeoflast = size % HNS3_MAX_BD_SIZE;
987 	sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
988 
989 	/* When the frag size is bigger than hardware, split this frag */
990 	for (k = 0; k < frag_buf_num; k++) {
991 		ret = hns3_fill_desc(ring, priv,
992 				     (k == frag_buf_num - 1) ?
993 				sizeoflast : HNS3_MAX_BD_SIZE,
994 				dma + HNS3_MAX_BD_SIZE * k,
995 				frag_end && (k == frag_buf_num - 1) ? 1 : 0,
996 				(type == DESC_TYPE_SKB && !k) ?
997 					DESC_TYPE_SKB : DESC_TYPE_PAGE);
998 		if (ret)
999 			return ret;
1000 	}
1001 
1002 	return 0;
1003 }
1004 
1005 static int hns3_nic_maybe_stop_tso(struct sk_buff **out_skb, int *bnum,
1006 				   struct hns3_enet_ring *ring)
1007 {
1008 	struct sk_buff *skb = *out_skb;
1009 	struct skb_frag_struct *frag;
1010 	int bdnum_for_frag;
1011 	int frag_num;
1012 	int buf_num;
1013 	int size;
1014 	int i;
1015 
1016 	size = skb_headlen(skb);
1017 	buf_num = (size + HNS3_MAX_BD_SIZE - 1) / HNS3_MAX_BD_SIZE;
1018 
1019 	frag_num = skb_shinfo(skb)->nr_frags;
1020 	for (i = 0; i < frag_num; i++) {
1021 		frag = &skb_shinfo(skb)->frags[i];
1022 		size = skb_frag_size(frag);
1023 		bdnum_for_frag =
1024 			(size + HNS3_MAX_BD_SIZE - 1) / HNS3_MAX_BD_SIZE;
1025 		if (bdnum_for_frag > HNS3_MAX_BD_PER_FRAG)
1026 			return -ENOMEM;
1027 
1028 		buf_num += bdnum_for_frag;
1029 	}
1030 
1031 	if (buf_num > ring_space(ring))
1032 		return -EBUSY;
1033 
1034 	*bnum = buf_num;
1035 	return 0;
1036 }
1037 
1038 static int hns3_nic_maybe_stop_tx(struct sk_buff **out_skb, int *bnum,
1039 				  struct hns3_enet_ring *ring)
1040 {
1041 	struct sk_buff *skb = *out_skb;
1042 	int buf_num;
1043 
1044 	/* No. of segments (plus a header) */
1045 	buf_num = skb_shinfo(skb)->nr_frags + 1;
1046 
1047 	if (buf_num > ring_space(ring))
1048 		return -EBUSY;
1049 
1050 	*bnum = buf_num;
1051 
1052 	return 0;
1053 }
1054 
1055 static void hns_nic_dma_unmap(struct hns3_enet_ring *ring, int next_to_use_orig)
1056 {
1057 	struct device *dev = ring_to_dev(ring);
1058 	unsigned int i;
1059 
1060 	for (i = 0; i < ring->desc_num; i++) {
1061 		/* check if this is where we started */
1062 		if (ring->next_to_use == next_to_use_orig)
1063 			break;
1064 
1065 		/* unmap the descriptor dma address */
1066 		if (ring->desc_cb[ring->next_to_use].type == DESC_TYPE_SKB)
1067 			dma_unmap_single(dev,
1068 					 ring->desc_cb[ring->next_to_use].dma,
1069 					ring->desc_cb[ring->next_to_use].length,
1070 					DMA_TO_DEVICE);
1071 		else
1072 			dma_unmap_page(dev,
1073 				       ring->desc_cb[ring->next_to_use].dma,
1074 				       ring->desc_cb[ring->next_to_use].length,
1075 				       DMA_TO_DEVICE);
1076 
1077 		/* rollback one */
1078 		ring_ptr_move_bw(ring, next_to_use);
1079 	}
1080 }
1081 
1082 netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
1083 {
1084 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1085 	struct hns3_nic_ring_data *ring_data =
1086 		&tx_ring_data(priv, skb->queue_mapping);
1087 	struct hns3_enet_ring *ring = ring_data->ring;
1088 	struct device *dev = priv->dev;
1089 	struct netdev_queue *dev_queue;
1090 	struct skb_frag_struct *frag;
1091 	int next_to_use_head;
1092 	int next_to_use_frag;
1093 	dma_addr_t dma;
1094 	int buf_num;
1095 	int seg_num;
1096 	int size;
1097 	int ret;
1098 	int i;
1099 
1100 	/* Prefetch the data used later */
1101 	prefetch(skb->data);
1102 
1103 	switch (priv->ops.maybe_stop_tx(&skb, &buf_num, ring)) {
1104 	case -EBUSY:
1105 		u64_stats_update_begin(&ring->syncp);
1106 		ring->stats.tx_busy++;
1107 		u64_stats_update_end(&ring->syncp);
1108 
1109 		goto out_net_tx_busy;
1110 	case -ENOMEM:
1111 		u64_stats_update_begin(&ring->syncp);
1112 		ring->stats.sw_err_cnt++;
1113 		u64_stats_update_end(&ring->syncp);
1114 		netdev_err(netdev, "no memory to xmit!\n");
1115 
1116 		goto out_err_tx_ok;
1117 	default:
1118 		break;
1119 	}
1120 
1121 	/* No. of segments (plus a header) */
1122 	seg_num = skb_shinfo(skb)->nr_frags + 1;
1123 	/* Fill the first part */
1124 	size = skb_headlen(skb);
1125 
1126 	next_to_use_head = ring->next_to_use;
1127 
1128 	dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1129 	if (dma_mapping_error(dev, dma)) {
1130 		netdev_err(netdev, "TX head DMA map failed\n");
1131 		ring->stats.sw_err_cnt++;
1132 		goto out_err_tx_ok;
1133 	}
1134 
1135 	ret = priv->ops.fill_desc(ring, skb, size, dma, seg_num == 1 ? 1 : 0,
1136 			   DESC_TYPE_SKB);
1137 	if (ret)
1138 		goto head_dma_map_err;
1139 
1140 	next_to_use_frag = ring->next_to_use;
1141 	/* Fill the fragments */
1142 	for (i = 1; i < seg_num; i++) {
1143 		frag = &skb_shinfo(skb)->frags[i - 1];
1144 		size = skb_frag_size(frag);
1145 		dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1146 		if (dma_mapping_error(dev, dma)) {
1147 			netdev_err(netdev, "TX frag(%d) DMA map failed\n", i);
1148 			ring->stats.sw_err_cnt++;
1149 			goto frag_dma_map_err;
1150 		}
1151 		ret = priv->ops.fill_desc(ring, skb_frag_page(frag), size, dma,
1152 				    seg_num - 1 == i ? 1 : 0,
1153 				    DESC_TYPE_PAGE);
1154 
1155 		if (ret)
1156 			goto frag_dma_map_err;
1157 	}
1158 
1159 	/* Complete translate all packets */
1160 	dev_queue = netdev_get_tx_queue(netdev, ring_data->queue_index);
1161 	netdev_tx_sent_queue(dev_queue, skb->len);
1162 
1163 	wmb(); /* Commit all data before submit */
1164 
1165 	hnae3_queue_xmit(ring->tqp, buf_num);
1166 
1167 	return NETDEV_TX_OK;
1168 
1169 frag_dma_map_err:
1170 	hns_nic_dma_unmap(ring, next_to_use_frag);
1171 
1172 head_dma_map_err:
1173 	hns_nic_dma_unmap(ring, next_to_use_head);
1174 
1175 out_err_tx_ok:
1176 	dev_kfree_skb_any(skb);
1177 	return NETDEV_TX_OK;
1178 
1179 out_net_tx_busy:
1180 	netif_stop_subqueue(netdev, ring_data->queue_index);
1181 	smp_mb(); /* Commit all data before submit */
1182 
1183 	return NETDEV_TX_BUSY;
1184 }
1185 
1186 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
1187 {
1188 	struct hnae3_handle *h = hns3_get_handle(netdev);
1189 	struct sockaddr *mac_addr = p;
1190 	int ret;
1191 
1192 	if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1193 		return -EADDRNOTAVAIL;
1194 
1195 	if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
1196 		netdev_info(netdev, "already using mac address %pM\n",
1197 			    mac_addr->sa_data);
1198 		return 0;
1199 	}
1200 
1201 	ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
1202 	if (ret) {
1203 		netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
1204 		return ret;
1205 	}
1206 
1207 	ether_addr_copy(netdev->dev_addr, mac_addr->sa_data);
1208 
1209 	return 0;
1210 }
1211 
1212 static int hns3_nic_set_features(struct net_device *netdev,
1213 				 netdev_features_t features)
1214 {
1215 	netdev_features_t changed = netdev->features ^ features;
1216 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1217 	struct hnae3_handle *h = priv->ae_handle;
1218 	int ret;
1219 
1220 	if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) {
1221 		if (features & (NETIF_F_TSO | NETIF_F_TSO6)) {
1222 			priv->ops.fill_desc = hns3_fill_desc_tso;
1223 			priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tso;
1224 		} else {
1225 			priv->ops.fill_desc = hns3_fill_desc;
1226 			priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tx;
1227 		}
1228 	}
1229 
1230 	if ((changed & NETIF_F_HW_VLAN_CTAG_FILTER) &&
1231 	    h->ae_algo->ops->enable_vlan_filter) {
1232 		if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
1233 			h->ae_algo->ops->enable_vlan_filter(h, true);
1234 		else
1235 			h->ae_algo->ops->enable_vlan_filter(h, false);
1236 	}
1237 
1238 	if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
1239 	    h->ae_algo->ops->enable_hw_strip_rxvtag) {
1240 		if (features & NETIF_F_HW_VLAN_CTAG_RX)
1241 			ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, true);
1242 		else
1243 			ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, false);
1244 
1245 		if (ret)
1246 			return ret;
1247 	}
1248 
1249 	netdev->features = features;
1250 	return 0;
1251 }
1252 
1253 static void hns3_nic_get_stats64(struct net_device *netdev,
1254 				 struct rtnl_link_stats64 *stats)
1255 {
1256 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1257 	int queue_num = priv->ae_handle->kinfo.num_tqps;
1258 	struct hnae3_handle *handle = priv->ae_handle;
1259 	struct hns3_enet_ring *ring;
1260 	unsigned int start;
1261 	unsigned int idx;
1262 	u64 tx_bytes = 0;
1263 	u64 rx_bytes = 0;
1264 	u64 tx_pkts = 0;
1265 	u64 rx_pkts = 0;
1266 	u64 tx_drop = 0;
1267 	u64 rx_drop = 0;
1268 
1269 	if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
1270 		return;
1271 
1272 	handle->ae_algo->ops->update_stats(handle, &netdev->stats);
1273 
1274 	for (idx = 0; idx < queue_num; idx++) {
1275 		/* fetch the tx stats */
1276 		ring = priv->ring_data[idx].ring;
1277 		do {
1278 			start = u64_stats_fetch_begin_irq(&ring->syncp);
1279 			tx_bytes += ring->stats.tx_bytes;
1280 			tx_pkts += ring->stats.tx_pkts;
1281 			tx_drop += ring->stats.tx_busy;
1282 			tx_drop += ring->stats.sw_err_cnt;
1283 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1284 
1285 		/* fetch the rx stats */
1286 		ring = priv->ring_data[idx + queue_num].ring;
1287 		do {
1288 			start = u64_stats_fetch_begin_irq(&ring->syncp);
1289 			rx_bytes += ring->stats.rx_bytes;
1290 			rx_pkts += ring->stats.rx_pkts;
1291 			rx_drop += ring->stats.non_vld_descs;
1292 			rx_drop += ring->stats.err_pkt_len;
1293 			rx_drop += ring->stats.l2_err;
1294 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1295 	}
1296 
1297 	stats->tx_bytes = tx_bytes;
1298 	stats->tx_packets = tx_pkts;
1299 	stats->rx_bytes = rx_bytes;
1300 	stats->rx_packets = rx_pkts;
1301 
1302 	stats->rx_errors = netdev->stats.rx_errors;
1303 	stats->multicast = netdev->stats.multicast;
1304 	stats->rx_length_errors = netdev->stats.rx_length_errors;
1305 	stats->rx_crc_errors = netdev->stats.rx_crc_errors;
1306 	stats->rx_missed_errors = netdev->stats.rx_missed_errors;
1307 
1308 	stats->tx_errors = netdev->stats.tx_errors;
1309 	stats->rx_dropped = rx_drop + netdev->stats.rx_dropped;
1310 	stats->tx_dropped = tx_drop + netdev->stats.tx_dropped;
1311 	stats->collisions = netdev->stats.collisions;
1312 	stats->rx_over_errors = netdev->stats.rx_over_errors;
1313 	stats->rx_frame_errors = netdev->stats.rx_frame_errors;
1314 	stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
1315 	stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
1316 	stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
1317 	stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
1318 	stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
1319 	stats->tx_window_errors = netdev->stats.tx_window_errors;
1320 	stats->rx_compressed = netdev->stats.rx_compressed;
1321 	stats->tx_compressed = netdev->stats.tx_compressed;
1322 }
1323 
1324 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
1325 {
1326 	struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
1327 	struct hnae3_handle *h = hns3_get_handle(netdev);
1328 	struct hnae3_knic_private_info *kinfo = &h->kinfo;
1329 	u8 *prio_tc = mqprio_qopt->qopt.prio_tc_map;
1330 	u8 tc = mqprio_qopt->qopt.num_tc;
1331 	u16 mode = mqprio_qopt->mode;
1332 	u8 hw = mqprio_qopt->qopt.hw;
1333 	bool if_running;
1334 	int ret;
1335 
1336 	if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
1337 	       mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
1338 		return -EOPNOTSUPP;
1339 
1340 	if (tc > HNAE3_MAX_TC)
1341 		return -EINVAL;
1342 
1343 	if (!netdev)
1344 		return -EINVAL;
1345 
1346 	if_running = netif_running(netdev);
1347 	if (if_running) {
1348 		hns3_nic_net_stop(netdev);
1349 		msleep(100);
1350 	}
1351 
1352 	ret = (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
1353 		kinfo->dcb_ops->setup_tc(h, tc, prio_tc) : -EOPNOTSUPP;
1354 	if (ret)
1355 		goto out;
1356 
1357 	ret = hns3_nic_set_real_num_queue(netdev);
1358 
1359 out:
1360 	if (if_running)
1361 		hns3_nic_net_open(netdev);
1362 
1363 	return ret;
1364 }
1365 
1366 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
1367 			     void *type_data)
1368 {
1369 	if (type != TC_SETUP_QDISC_MQPRIO)
1370 		return -EOPNOTSUPP;
1371 
1372 	return hns3_setup_tc(dev, type_data);
1373 }
1374 
1375 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
1376 				__be16 proto, u16 vid)
1377 {
1378 	struct hnae3_handle *h = hns3_get_handle(netdev);
1379 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1380 	int ret = -EIO;
1381 
1382 	if (h->ae_algo->ops->set_vlan_filter)
1383 		ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
1384 
1385 	if (!ret)
1386 		set_bit(vid, priv->active_vlans);
1387 
1388 	return ret;
1389 }
1390 
1391 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
1392 				 __be16 proto, u16 vid)
1393 {
1394 	struct hnae3_handle *h = hns3_get_handle(netdev);
1395 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1396 	int ret = -EIO;
1397 
1398 	if (h->ae_algo->ops->set_vlan_filter)
1399 		ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
1400 
1401 	if (!ret)
1402 		clear_bit(vid, priv->active_vlans);
1403 
1404 	return ret;
1405 }
1406 
1407 static void hns3_restore_vlan(struct net_device *netdev)
1408 {
1409 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1410 	u16 vid;
1411 	int ret;
1412 
1413 	for_each_set_bit(vid, priv->active_vlans, VLAN_N_VID) {
1414 		ret = hns3_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
1415 		if (ret)
1416 			netdev_warn(netdev, "Restore vlan: %d filter, ret:%d\n",
1417 				    vid, ret);
1418 	}
1419 }
1420 
1421 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
1422 				u8 qos, __be16 vlan_proto)
1423 {
1424 	struct hnae3_handle *h = hns3_get_handle(netdev);
1425 	int ret = -EIO;
1426 
1427 	if (h->ae_algo->ops->set_vf_vlan_filter)
1428 		ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
1429 						   qos, vlan_proto);
1430 
1431 	return ret;
1432 }
1433 
1434 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
1435 {
1436 	struct hnae3_handle *h = hns3_get_handle(netdev);
1437 	bool if_running = netif_running(netdev);
1438 	int ret;
1439 
1440 	if (!h->ae_algo->ops->set_mtu)
1441 		return -EOPNOTSUPP;
1442 
1443 	/* if this was called with netdev up then bring netdevice down */
1444 	if (if_running) {
1445 		(void)hns3_nic_net_stop(netdev);
1446 		msleep(100);
1447 	}
1448 
1449 	ret = h->ae_algo->ops->set_mtu(h, new_mtu);
1450 	if (ret) {
1451 		netdev_err(netdev, "failed to change MTU in hardware %d\n",
1452 			   ret);
1453 		return ret;
1454 	}
1455 
1456 	netdev->mtu = new_mtu;
1457 
1458 	/* if the netdev was running earlier, bring it up again */
1459 	if (if_running && hns3_nic_net_open(netdev))
1460 		ret = -EINVAL;
1461 
1462 	return ret;
1463 }
1464 
1465 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
1466 {
1467 	struct hns3_nic_priv *priv = netdev_priv(ndev);
1468 	struct hns3_enet_ring *tx_ring = NULL;
1469 	int timeout_queue = 0;
1470 	int hw_head, hw_tail;
1471 	int i;
1472 
1473 	/* Find the stopped queue the same way the stack does */
1474 	for (i = 0; i < ndev->real_num_tx_queues; i++) {
1475 		struct netdev_queue *q;
1476 		unsigned long trans_start;
1477 
1478 		q = netdev_get_tx_queue(ndev, i);
1479 		trans_start = q->trans_start;
1480 		if (netif_xmit_stopped(q) &&
1481 		    time_after(jiffies,
1482 			       (trans_start + ndev->watchdog_timeo))) {
1483 			timeout_queue = i;
1484 			break;
1485 		}
1486 	}
1487 
1488 	if (i == ndev->num_tx_queues) {
1489 		netdev_info(ndev,
1490 			    "no netdev TX timeout queue found, timeout count: %llu\n",
1491 			    priv->tx_timeout_count);
1492 		return false;
1493 	}
1494 
1495 	tx_ring = priv->ring_data[timeout_queue].ring;
1496 
1497 	hw_head = readl_relaxed(tx_ring->tqp->io_base +
1498 				HNS3_RING_TX_RING_HEAD_REG);
1499 	hw_tail = readl_relaxed(tx_ring->tqp->io_base +
1500 				HNS3_RING_TX_RING_TAIL_REG);
1501 	netdev_info(ndev,
1502 		    "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",
1503 		    priv->tx_timeout_count,
1504 		    timeout_queue,
1505 		    tx_ring->next_to_use,
1506 		    tx_ring->next_to_clean,
1507 		    hw_head,
1508 		    hw_tail,
1509 		    readl(tx_ring->tqp_vector->mask_addr));
1510 
1511 	return true;
1512 }
1513 
1514 static void hns3_nic_net_timeout(struct net_device *ndev)
1515 {
1516 	struct hns3_nic_priv *priv = netdev_priv(ndev);
1517 	struct hnae3_handle *h = priv->ae_handle;
1518 
1519 	if (!hns3_get_tx_timeo_queue_info(ndev))
1520 		return;
1521 
1522 	priv->tx_timeout_count++;
1523 
1524 	if (time_before(jiffies, (h->last_reset_time + ndev->watchdog_timeo)))
1525 		return;
1526 
1527 	/* request the reset */
1528 	if (h->ae_algo->ops->reset_event)
1529 		h->ae_algo->ops->reset_event(h);
1530 }
1531 
1532 static const struct net_device_ops hns3_nic_netdev_ops = {
1533 	.ndo_open		= hns3_nic_net_open,
1534 	.ndo_stop		= hns3_nic_net_stop,
1535 	.ndo_start_xmit		= hns3_nic_net_xmit,
1536 	.ndo_tx_timeout		= hns3_nic_net_timeout,
1537 	.ndo_set_mac_address	= hns3_nic_net_set_mac_address,
1538 	.ndo_change_mtu		= hns3_nic_change_mtu,
1539 	.ndo_set_features	= hns3_nic_set_features,
1540 	.ndo_get_stats64	= hns3_nic_get_stats64,
1541 	.ndo_setup_tc		= hns3_nic_setup_tc,
1542 	.ndo_set_rx_mode	= hns3_nic_set_rx_mode,
1543 	.ndo_vlan_rx_add_vid	= hns3_vlan_rx_add_vid,
1544 	.ndo_vlan_rx_kill_vid	= hns3_vlan_rx_kill_vid,
1545 	.ndo_set_vf_vlan	= hns3_ndo_set_vf_vlan,
1546 };
1547 
1548 static bool hns3_is_phys_func(struct pci_dev *pdev)
1549 {
1550 	u32 dev_id = pdev->device;
1551 
1552 	switch (dev_id) {
1553 	case HNAE3_DEV_ID_GE:
1554 	case HNAE3_DEV_ID_25GE:
1555 	case HNAE3_DEV_ID_25GE_RDMA:
1556 	case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
1557 	case HNAE3_DEV_ID_50GE_RDMA:
1558 	case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
1559 	case HNAE3_DEV_ID_100G_RDMA_MACSEC:
1560 		return true;
1561 	case HNAE3_DEV_ID_100G_VF:
1562 	case HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF:
1563 		return false;
1564 	default:
1565 		dev_warn(&pdev->dev, "un-recognized pci device-id %d",
1566 			 dev_id);
1567 	}
1568 
1569 	return false;
1570 }
1571 
1572 static void hns3_disable_sriov(struct pci_dev *pdev)
1573 {
1574 	/* If our VFs are assigned we cannot shut down SR-IOV
1575 	 * without causing issues, so just leave the hardware
1576 	 * available but disabled
1577 	 */
1578 	if (pci_vfs_assigned(pdev)) {
1579 		dev_warn(&pdev->dev,
1580 			 "disabling driver while VFs are assigned\n");
1581 		return;
1582 	}
1583 
1584 	pci_disable_sriov(pdev);
1585 }
1586 
1587 /* hns3_probe - Device initialization routine
1588  * @pdev: PCI device information struct
1589  * @ent: entry in hns3_pci_tbl
1590  *
1591  * hns3_probe initializes a PF identified by a pci_dev structure.
1592  * The OS initialization, configuring of the PF private structure,
1593  * and a hardware reset occur.
1594  *
1595  * Returns 0 on success, negative on failure
1596  */
1597 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1598 {
1599 	struct hnae3_ae_dev *ae_dev;
1600 	int ret;
1601 
1602 	ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev),
1603 			      GFP_KERNEL);
1604 	if (!ae_dev) {
1605 		ret = -ENOMEM;
1606 		return ret;
1607 	}
1608 
1609 	ae_dev->pdev = pdev;
1610 	ae_dev->flag = ent->driver_data;
1611 	ae_dev->dev_type = HNAE3_DEV_KNIC;
1612 	pci_set_drvdata(pdev, ae_dev);
1613 
1614 	hnae3_register_ae_dev(ae_dev);
1615 
1616 	return 0;
1617 }
1618 
1619 /* hns3_remove - Device removal routine
1620  * @pdev: PCI device information struct
1621  */
1622 static void hns3_remove(struct pci_dev *pdev)
1623 {
1624 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1625 
1626 	if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
1627 		hns3_disable_sriov(pdev);
1628 
1629 	hnae3_unregister_ae_dev(ae_dev);
1630 }
1631 
1632 /**
1633  * hns3_pci_sriov_configure
1634  * @pdev: pointer to a pci_dev structure
1635  * @num_vfs: number of VFs to allocate
1636  *
1637  * Enable or change the number of VFs. Called when the user updates the number
1638  * of VFs in sysfs.
1639  **/
1640 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
1641 {
1642 	int ret;
1643 
1644 	if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
1645 		dev_warn(&pdev->dev, "Can not config SRIOV\n");
1646 		return -EINVAL;
1647 	}
1648 
1649 	if (num_vfs) {
1650 		ret = pci_enable_sriov(pdev, num_vfs);
1651 		if (ret)
1652 			dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
1653 		else
1654 			return num_vfs;
1655 	} else if (!pci_vfs_assigned(pdev)) {
1656 		pci_disable_sriov(pdev);
1657 	} else {
1658 		dev_warn(&pdev->dev,
1659 			 "Unable to free VFs because some are assigned to VMs.\n");
1660 	}
1661 
1662 	return 0;
1663 }
1664 
1665 static void hns3_shutdown(struct pci_dev *pdev)
1666 {
1667 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1668 
1669 	hnae3_unregister_ae_dev(ae_dev);
1670 	devm_kfree(&pdev->dev, ae_dev);
1671 	pci_set_drvdata(pdev, NULL);
1672 
1673 	if (system_state == SYSTEM_POWER_OFF)
1674 		pci_set_power_state(pdev, PCI_D3hot);
1675 }
1676 
1677 static struct pci_driver hns3_driver = {
1678 	.name     = hns3_driver_name,
1679 	.id_table = hns3_pci_tbl,
1680 	.probe    = hns3_probe,
1681 	.remove   = hns3_remove,
1682 	.shutdown = hns3_shutdown,
1683 	.sriov_configure = hns3_pci_sriov_configure,
1684 };
1685 
1686 /* set default feature to hns3 */
1687 static void hns3_set_default_feature(struct net_device *netdev)
1688 {
1689 	struct hnae3_handle *h = hns3_get_handle(netdev);
1690 	struct pci_dev *pdev = h->pdev;
1691 
1692 	netdev->priv_flags |= IFF_UNICAST_FLT;
1693 
1694 	netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1695 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1696 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1697 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1698 		NETIF_F_GSO_UDP_TUNNEL_CSUM;
1699 
1700 	netdev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
1701 
1702 	netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
1703 
1704 	netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1705 		NETIF_F_HW_VLAN_CTAG_FILTER |
1706 		NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
1707 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1708 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1709 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1710 		NETIF_F_GSO_UDP_TUNNEL_CSUM;
1711 
1712 	netdev->vlan_features |=
1713 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
1714 		NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO |
1715 		NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1716 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1717 		NETIF_F_GSO_UDP_TUNNEL_CSUM;
1718 
1719 	netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1720 		NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
1721 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1722 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1723 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1724 		NETIF_F_GSO_UDP_TUNNEL_CSUM;
1725 
1726 	if (pdev->revision != 0x20)
1727 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
1728 }
1729 
1730 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
1731 			     struct hns3_desc_cb *cb)
1732 {
1733 	unsigned int order = hnae3_page_order(ring);
1734 	struct page *p;
1735 
1736 	p = dev_alloc_pages(order);
1737 	if (!p)
1738 		return -ENOMEM;
1739 
1740 	cb->priv = p;
1741 	cb->page_offset = 0;
1742 	cb->reuse_flag = 0;
1743 	cb->buf  = page_address(p);
1744 	cb->length = hnae3_page_size(ring);
1745 	cb->type = DESC_TYPE_PAGE;
1746 
1747 	return 0;
1748 }
1749 
1750 static void hns3_free_buffer(struct hns3_enet_ring *ring,
1751 			     struct hns3_desc_cb *cb)
1752 {
1753 	if (cb->type == DESC_TYPE_SKB)
1754 		dev_kfree_skb_any((struct sk_buff *)cb->priv);
1755 	else if (!HNAE3_IS_TX_RING(ring))
1756 		put_page((struct page *)cb->priv);
1757 	memset(cb, 0, sizeof(*cb));
1758 }
1759 
1760 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
1761 {
1762 	cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
1763 			       cb->length, ring_to_dma_dir(ring));
1764 
1765 	if (dma_mapping_error(ring_to_dev(ring), cb->dma))
1766 		return -EIO;
1767 
1768 	return 0;
1769 }
1770 
1771 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
1772 			      struct hns3_desc_cb *cb)
1773 {
1774 	if (cb->type == DESC_TYPE_SKB)
1775 		dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
1776 				 ring_to_dma_dir(ring));
1777 	else
1778 		dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
1779 			       ring_to_dma_dir(ring));
1780 }
1781 
1782 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
1783 {
1784 	hns3_unmap_buffer(ring, &ring->desc_cb[i]);
1785 	ring->desc[i].addr = 0;
1786 }
1787 
1788 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i)
1789 {
1790 	struct hns3_desc_cb *cb = &ring->desc_cb[i];
1791 
1792 	if (!ring->desc_cb[i].dma)
1793 		return;
1794 
1795 	hns3_buffer_detach(ring, i);
1796 	hns3_free_buffer(ring, cb);
1797 }
1798 
1799 static void hns3_free_buffers(struct hns3_enet_ring *ring)
1800 {
1801 	int i;
1802 
1803 	for (i = 0; i < ring->desc_num; i++)
1804 		hns3_free_buffer_detach(ring, i);
1805 }
1806 
1807 /* free desc along with its attached buffer */
1808 static void hns3_free_desc(struct hns3_enet_ring *ring)
1809 {
1810 	int size = ring->desc_num * sizeof(ring->desc[0]);
1811 
1812 	hns3_free_buffers(ring);
1813 
1814 	if (ring->desc) {
1815 		dma_free_coherent(ring_to_dev(ring), size,
1816 				  ring->desc, ring->desc_dma_addr);
1817 		ring->desc = NULL;
1818 	}
1819 }
1820 
1821 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
1822 {
1823 	int size = ring->desc_num * sizeof(ring->desc[0]);
1824 
1825 	ring->desc = dma_zalloc_coherent(ring_to_dev(ring), size,
1826 					 &ring->desc_dma_addr,
1827 					 GFP_KERNEL);
1828 	if (!ring->desc)
1829 		return -ENOMEM;
1830 
1831 	return 0;
1832 }
1833 
1834 static int hns3_reserve_buffer_map(struct hns3_enet_ring *ring,
1835 				   struct hns3_desc_cb *cb)
1836 {
1837 	int ret;
1838 
1839 	ret = hns3_alloc_buffer(ring, cb);
1840 	if (ret)
1841 		goto out;
1842 
1843 	ret = hns3_map_buffer(ring, cb);
1844 	if (ret)
1845 		goto out_with_buf;
1846 
1847 	return 0;
1848 
1849 out_with_buf:
1850 	hns3_free_buffer(ring, cb);
1851 out:
1852 	return ret;
1853 }
1854 
1855 static int hns3_alloc_buffer_attach(struct hns3_enet_ring *ring, int i)
1856 {
1857 	int ret = hns3_reserve_buffer_map(ring, &ring->desc_cb[i]);
1858 
1859 	if (ret)
1860 		return ret;
1861 
1862 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
1863 
1864 	return 0;
1865 }
1866 
1867 /* Allocate memory for raw pkg, and map with dma */
1868 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
1869 {
1870 	int i, j, ret;
1871 
1872 	for (i = 0; i < ring->desc_num; i++) {
1873 		ret = hns3_alloc_buffer_attach(ring, i);
1874 		if (ret)
1875 			goto out_buffer_fail;
1876 	}
1877 
1878 	return 0;
1879 
1880 out_buffer_fail:
1881 	for (j = i - 1; j >= 0; j--)
1882 		hns3_free_buffer_detach(ring, j);
1883 	return ret;
1884 }
1885 
1886 /* detach a in-used buffer and replace with a reserved one  */
1887 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
1888 				struct hns3_desc_cb *res_cb)
1889 {
1890 	hns3_unmap_buffer(ring, &ring->desc_cb[i]);
1891 	ring->desc_cb[i] = *res_cb;
1892 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
1893 	ring->desc[i].rx.bd_base_info = 0;
1894 }
1895 
1896 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
1897 {
1898 	ring->desc_cb[i].reuse_flag = 0;
1899 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma
1900 		+ ring->desc_cb[i].page_offset);
1901 	ring->desc[i].rx.bd_base_info = 0;
1902 }
1903 
1904 static void hns3_nic_reclaim_one_desc(struct hns3_enet_ring *ring, int *bytes,
1905 				      int *pkts)
1906 {
1907 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
1908 
1909 	(*pkts) += (desc_cb->type == DESC_TYPE_SKB);
1910 	(*bytes) += desc_cb->length;
1911 	/* desc_cb will be cleaned, after hnae3_free_buffer_detach*/
1912 	hns3_free_buffer_detach(ring, ring->next_to_clean);
1913 
1914 	ring_ptr_move_fw(ring, next_to_clean);
1915 }
1916 
1917 static int is_valid_clean_head(struct hns3_enet_ring *ring, int h)
1918 {
1919 	int u = ring->next_to_use;
1920 	int c = ring->next_to_clean;
1921 
1922 	if (unlikely(h > ring->desc_num))
1923 		return 0;
1924 
1925 	return u > c ? (h > c && h <= u) : (h > c || h <= u);
1926 }
1927 
1928 bool hns3_clean_tx_ring(struct hns3_enet_ring *ring, int budget)
1929 {
1930 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
1931 	struct netdev_queue *dev_queue;
1932 	int bytes, pkts;
1933 	int head;
1934 
1935 	head = readl_relaxed(ring->tqp->io_base + HNS3_RING_TX_RING_HEAD_REG);
1936 	rmb(); /* Make sure head is ready before touch any data */
1937 
1938 	if (is_ring_empty(ring) || head == ring->next_to_clean)
1939 		return true; /* no data to poll */
1940 
1941 	if (unlikely(!is_valid_clean_head(ring, head))) {
1942 		netdev_err(netdev, "wrong head (%d, %d-%d)\n", head,
1943 			   ring->next_to_use, ring->next_to_clean);
1944 
1945 		u64_stats_update_begin(&ring->syncp);
1946 		ring->stats.io_err_cnt++;
1947 		u64_stats_update_end(&ring->syncp);
1948 		return true;
1949 	}
1950 
1951 	bytes = 0;
1952 	pkts = 0;
1953 	while (head != ring->next_to_clean && budget) {
1954 		hns3_nic_reclaim_one_desc(ring, &bytes, &pkts);
1955 		/* Issue prefetch for next Tx descriptor */
1956 		prefetch(&ring->desc_cb[ring->next_to_clean]);
1957 		budget--;
1958 	}
1959 
1960 	ring->tqp_vector->tx_group.total_bytes += bytes;
1961 	ring->tqp_vector->tx_group.total_packets += pkts;
1962 
1963 	u64_stats_update_begin(&ring->syncp);
1964 	ring->stats.tx_bytes += bytes;
1965 	ring->stats.tx_pkts += pkts;
1966 	u64_stats_update_end(&ring->syncp);
1967 
1968 	dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
1969 	netdev_tx_completed_queue(dev_queue, pkts, bytes);
1970 
1971 	if (unlikely(pkts && netif_carrier_ok(netdev) &&
1972 		     (ring_space(ring) > HNS3_MAX_BD_PER_PKT))) {
1973 		/* Make sure that anybody stopping the queue after this
1974 		 * sees the new next_to_clean.
1975 		 */
1976 		smp_mb();
1977 		if (netif_tx_queue_stopped(dev_queue)) {
1978 			netif_tx_wake_queue(dev_queue);
1979 			ring->stats.restart_queue++;
1980 		}
1981 	}
1982 
1983 	return !!budget;
1984 }
1985 
1986 static int hns3_desc_unused(struct hns3_enet_ring *ring)
1987 {
1988 	int ntc = ring->next_to_clean;
1989 	int ntu = ring->next_to_use;
1990 
1991 	return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
1992 }
1993 
1994 static void
1995 hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring, int cleand_count)
1996 {
1997 	struct hns3_desc_cb *desc_cb;
1998 	struct hns3_desc_cb res_cbs;
1999 	int i, ret;
2000 
2001 	for (i = 0; i < cleand_count; i++) {
2002 		desc_cb = &ring->desc_cb[ring->next_to_use];
2003 		if (desc_cb->reuse_flag) {
2004 			u64_stats_update_begin(&ring->syncp);
2005 			ring->stats.reuse_pg_cnt++;
2006 			u64_stats_update_end(&ring->syncp);
2007 
2008 			hns3_reuse_buffer(ring, ring->next_to_use);
2009 		} else {
2010 			ret = hns3_reserve_buffer_map(ring, &res_cbs);
2011 			if (ret) {
2012 				u64_stats_update_begin(&ring->syncp);
2013 				ring->stats.sw_err_cnt++;
2014 				u64_stats_update_end(&ring->syncp);
2015 
2016 				netdev_err(ring->tqp->handle->kinfo.netdev,
2017 					   "hnae reserve buffer map failed.\n");
2018 				break;
2019 			}
2020 			hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
2021 		}
2022 
2023 		ring_ptr_move_fw(ring, next_to_use);
2024 	}
2025 
2026 	wmb(); /* Make all data has been write before submit */
2027 	writel_relaxed(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
2028 }
2029 
2030 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
2031 				struct hns3_enet_ring *ring, int pull_len,
2032 				struct hns3_desc_cb *desc_cb)
2033 {
2034 	struct hns3_desc *desc;
2035 	u32 truesize;
2036 	int size;
2037 	int last_offset;
2038 	bool twobufs;
2039 
2040 	twobufs = ((PAGE_SIZE < 8192) &&
2041 		hnae3_buf_size(ring) == HNS3_BUFFER_SIZE_2048);
2042 
2043 	desc = &ring->desc[ring->next_to_clean];
2044 	size = le16_to_cpu(desc->rx.size);
2045 
2046 	truesize = hnae3_buf_size(ring);
2047 
2048 	if (!twobufs)
2049 		last_offset = hnae3_page_size(ring) - hnae3_buf_size(ring);
2050 
2051 	skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
2052 			size - pull_len, truesize);
2053 
2054 	 /* Avoid re-using remote pages,flag default unreuse */
2055 	if (unlikely(page_to_nid(desc_cb->priv) != numa_node_id()))
2056 		return;
2057 
2058 	if (twobufs) {
2059 		/* If we are only owner of page we can reuse it */
2060 		if (likely(page_count(desc_cb->priv) == 1)) {
2061 			/* Flip page offset to other buffer */
2062 			desc_cb->page_offset ^= truesize;
2063 
2064 			desc_cb->reuse_flag = 1;
2065 			/* bump ref count on page before it is given*/
2066 			get_page(desc_cb->priv);
2067 		}
2068 		return;
2069 	}
2070 
2071 	/* Move offset up to the next cache line */
2072 	desc_cb->page_offset += truesize;
2073 
2074 	if (desc_cb->page_offset <= last_offset) {
2075 		desc_cb->reuse_flag = 1;
2076 		/* Bump ref count on page before it is given*/
2077 		get_page(desc_cb->priv);
2078 	}
2079 }
2080 
2081 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
2082 			     struct hns3_desc *desc)
2083 {
2084 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2085 	int l3_type, l4_type;
2086 	u32 bd_base_info;
2087 	int ol4_type;
2088 	u32 l234info;
2089 
2090 	bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2091 	l234info = le32_to_cpu(desc->rx.l234_info);
2092 
2093 	skb->ip_summed = CHECKSUM_NONE;
2094 
2095 	skb_checksum_none_assert(skb);
2096 
2097 	if (!(netdev->features & NETIF_F_RXCSUM))
2098 		return;
2099 
2100 	/* check if hardware has done checksum */
2101 	if (!hnae3_get_bit(bd_base_info, HNS3_RXD_L3L4P_B))
2102 		return;
2103 
2104 	if (unlikely(hnae3_get_bit(l234info, HNS3_RXD_L3E_B) ||
2105 		     hnae3_get_bit(l234info, HNS3_RXD_L4E_B) ||
2106 		     hnae3_get_bit(l234info, HNS3_RXD_OL3E_B) ||
2107 		     hnae3_get_bit(l234info, HNS3_RXD_OL4E_B))) {
2108 		netdev_err(netdev, "L3/L4 error pkt\n");
2109 		u64_stats_update_begin(&ring->syncp);
2110 		ring->stats.l3l4_csum_err++;
2111 		u64_stats_update_end(&ring->syncp);
2112 
2113 		return;
2114 	}
2115 
2116 	l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
2117 				  HNS3_RXD_L3ID_S);
2118 	l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
2119 				  HNS3_RXD_L4ID_S);
2120 
2121 	ol4_type = hnae3_get_field(l234info, HNS3_RXD_OL4ID_M,
2122 				   HNS3_RXD_OL4ID_S);
2123 	switch (ol4_type) {
2124 	case HNS3_OL4_TYPE_MAC_IN_UDP:
2125 	case HNS3_OL4_TYPE_NVGRE:
2126 		skb->csum_level = 1;
2127 		/* fall through */
2128 	case HNS3_OL4_TYPE_NO_TUN:
2129 		/* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
2130 		if ((l3_type == HNS3_L3_TYPE_IPV4 ||
2131 		     l3_type == HNS3_L3_TYPE_IPV6) &&
2132 		    (l4_type == HNS3_L4_TYPE_UDP ||
2133 		     l4_type == HNS3_L4_TYPE_TCP ||
2134 		     l4_type == HNS3_L4_TYPE_SCTP))
2135 			skb->ip_summed = CHECKSUM_UNNECESSARY;
2136 		break;
2137 	}
2138 }
2139 
2140 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
2141 {
2142 	napi_gro_receive(&ring->tqp_vector->napi, skb);
2143 }
2144 
2145 static u16 hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
2146 			       struct hns3_desc *desc, u32 l234info)
2147 {
2148 	struct pci_dev *pdev = ring->tqp->handle->pdev;
2149 	u16 vlan_tag;
2150 
2151 	if (pdev->revision == 0x20) {
2152 		vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2153 		if (!(vlan_tag & VLAN_VID_MASK))
2154 			vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2155 
2156 		return vlan_tag;
2157 	}
2158 
2159 #define HNS3_STRP_OUTER_VLAN	0x1
2160 #define HNS3_STRP_INNER_VLAN	0x2
2161 
2162 	switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
2163 				HNS3_RXD_STRP_TAGP_S)) {
2164 	case HNS3_STRP_OUTER_VLAN:
2165 		vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2166 		break;
2167 	case HNS3_STRP_INNER_VLAN:
2168 		vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2169 		break;
2170 	default:
2171 		vlan_tag = 0;
2172 		break;
2173 	}
2174 
2175 	return vlan_tag;
2176 }
2177 
2178 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring,
2179 			     struct sk_buff **out_skb, int *out_bnum)
2180 {
2181 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2182 	struct hns3_desc_cb *desc_cb;
2183 	struct hns3_desc *desc;
2184 	struct sk_buff *skb;
2185 	unsigned char *va;
2186 	u32 bd_base_info;
2187 	int pull_len;
2188 	u32 l234info;
2189 	int length;
2190 	int bnum;
2191 
2192 	desc = &ring->desc[ring->next_to_clean];
2193 	desc_cb = &ring->desc_cb[ring->next_to_clean];
2194 
2195 	prefetch(desc);
2196 
2197 	length = le16_to_cpu(desc->rx.size);
2198 	bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2199 
2200 	/* Check valid BD */
2201 	if (unlikely(!hnae3_get_bit(bd_base_info, HNS3_RXD_VLD_B)))
2202 		return -EFAULT;
2203 
2204 	va = (unsigned char *)desc_cb->buf + desc_cb->page_offset;
2205 
2206 	/* Prefetch first cache line of first page
2207 	 * Idea is to cache few bytes of the header of the packet. Our L1 Cache
2208 	 * line size is 64B so need to prefetch twice to make it 128B. But in
2209 	 * actual we can have greater size of caches with 128B Level 1 cache
2210 	 * lines. In such a case, single fetch would suffice to cache in the
2211 	 * relevant part of the header.
2212 	 */
2213 	prefetch(va);
2214 #if L1_CACHE_BYTES < 128
2215 	prefetch(va + L1_CACHE_BYTES);
2216 #endif
2217 
2218 	skb = *out_skb = napi_alloc_skb(&ring->tqp_vector->napi,
2219 					HNS3_RX_HEAD_SIZE);
2220 	if (unlikely(!skb)) {
2221 		netdev_err(netdev, "alloc rx skb fail\n");
2222 
2223 		u64_stats_update_begin(&ring->syncp);
2224 		ring->stats.sw_err_cnt++;
2225 		u64_stats_update_end(&ring->syncp);
2226 
2227 		return -ENOMEM;
2228 	}
2229 
2230 	prefetchw(skb->data);
2231 
2232 	bnum = 1;
2233 	if (length <= HNS3_RX_HEAD_SIZE) {
2234 		memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
2235 
2236 		/* We can reuse buffer as-is, just make sure it is local */
2237 		if (likely(page_to_nid(desc_cb->priv) == numa_node_id()))
2238 			desc_cb->reuse_flag = 1;
2239 		else /* This page cannot be reused so discard it */
2240 			put_page(desc_cb->priv);
2241 
2242 		ring_ptr_move_fw(ring, next_to_clean);
2243 	} else {
2244 		u64_stats_update_begin(&ring->syncp);
2245 		ring->stats.seg_pkt_cnt++;
2246 		u64_stats_update_end(&ring->syncp);
2247 
2248 		pull_len = eth_get_headlen(va, HNS3_RX_HEAD_SIZE);
2249 
2250 		memcpy(__skb_put(skb, pull_len), va,
2251 		       ALIGN(pull_len, sizeof(long)));
2252 
2253 		hns3_nic_reuse_page(skb, 0, ring, pull_len, desc_cb);
2254 		ring_ptr_move_fw(ring, next_to_clean);
2255 
2256 		while (!hnae3_get_bit(bd_base_info, HNS3_RXD_FE_B)) {
2257 			desc = &ring->desc[ring->next_to_clean];
2258 			desc_cb = &ring->desc_cb[ring->next_to_clean];
2259 			bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2260 			hns3_nic_reuse_page(skb, bnum, ring, 0, desc_cb);
2261 			ring_ptr_move_fw(ring, next_to_clean);
2262 			bnum++;
2263 		}
2264 	}
2265 
2266 	*out_bnum = bnum;
2267 
2268 	l234info = le32_to_cpu(desc->rx.l234_info);
2269 
2270 	/* Based on hw strategy, the tag offloaded will be stored at
2271 	 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
2272 	 * in one layer tag case.
2273 	 */
2274 	if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
2275 		u16 vlan_tag;
2276 
2277 		vlan_tag = hns3_parse_vlan_tag(ring, desc, l234info);
2278 		if (vlan_tag & VLAN_VID_MASK)
2279 			__vlan_hwaccel_put_tag(skb,
2280 					       htons(ETH_P_8021Q),
2281 					       vlan_tag);
2282 	}
2283 
2284 	if (unlikely(!hnae3_get_bit(bd_base_info, HNS3_RXD_VLD_B))) {
2285 		netdev_err(netdev, "no valid bd,%016llx,%016llx\n",
2286 			   ((u64 *)desc)[0], ((u64 *)desc)[1]);
2287 		u64_stats_update_begin(&ring->syncp);
2288 		ring->stats.non_vld_descs++;
2289 		u64_stats_update_end(&ring->syncp);
2290 
2291 		dev_kfree_skb_any(skb);
2292 		return -EINVAL;
2293 	}
2294 
2295 	if (unlikely((!desc->rx.pkt_len) ||
2296 		     hnae3_get_bit(l234info, HNS3_RXD_TRUNCAT_B))) {
2297 		netdev_err(netdev, "truncated pkt\n");
2298 		u64_stats_update_begin(&ring->syncp);
2299 		ring->stats.err_pkt_len++;
2300 		u64_stats_update_end(&ring->syncp);
2301 
2302 		dev_kfree_skb_any(skb);
2303 		return -EFAULT;
2304 	}
2305 
2306 	if (unlikely(hnae3_get_bit(l234info, HNS3_RXD_L2E_B))) {
2307 		netdev_err(netdev, "L2 error pkt\n");
2308 		u64_stats_update_begin(&ring->syncp);
2309 		ring->stats.l2_err++;
2310 		u64_stats_update_end(&ring->syncp);
2311 
2312 		dev_kfree_skb_any(skb);
2313 		return -EFAULT;
2314 	}
2315 
2316 	u64_stats_update_begin(&ring->syncp);
2317 	ring->stats.rx_pkts++;
2318 	ring->stats.rx_bytes += skb->len;
2319 	u64_stats_update_end(&ring->syncp);
2320 
2321 	ring->tqp_vector->rx_group.total_bytes += skb->len;
2322 
2323 	hns3_rx_checksum(ring, skb, desc);
2324 	return 0;
2325 }
2326 
2327 int hns3_clean_rx_ring(
2328 		struct hns3_enet_ring *ring, int budget,
2329 		void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
2330 {
2331 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
2332 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2333 	int recv_pkts, recv_bds, clean_count, err;
2334 	int unused_count = hns3_desc_unused(ring);
2335 	struct sk_buff *skb = NULL;
2336 	int num, bnum = 0;
2337 
2338 	num = readl_relaxed(ring->tqp->io_base + HNS3_RING_RX_RING_FBDNUM_REG);
2339 	rmb(); /* Make sure num taken effect before the other data is touched */
2340 
2341 	recv_pkts = 0, recv_bds = 0, clean_count = 0;
2342 	num -= unused_count;
2343 
2344 	while (recv_pkts < budget && recv_bds < num) {
2345 		/* Reuse or realloc buffers */
2346 		if (clean_count + unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
2347 			hns3_nic_alloc_rx_buffers(ring,
2348 						  clean_count + unused_count);
2349 			clean_count = 0;
2350 			unused_count = hns3_desc_unused(ring);
2351 		}
2352 
2353 		/* Poll one pkt */
2354 		err = hns3_handle_rx_bd(ring, &skb, &bnum);
2355 		if (unlikely(!skb)) /* This fault cannot be repaired */
2356 			goto out;
2357 
2358 		recv_bds += bnum;
2359 		clean_count += bnum;
2360 		if (unlikely(err)) {  /* Do jump the err */
2361 			recv_pkts++;
2362 			continue;
2363 		}
2364 
2365 		/* Do update ip stack process */
2366 		skb->protocol = eth_type_trans(skb, netdev);
2367 		rx_fn(ring, skb);
2368 
2369 		recv_pkts++;
2370 	}
2371 
2372 out:
2373 	/* Make all data has been write before submit */
2374 	if (clean_count + unused_count > 0)
2375 		hns3_nic_alloc_rx_buffers(ring,
2376 					  clean_count + unused_count);
2377 
2378 	return recv_pkts;
2379 }
2380 
2381 static bool hns3_get_new_int_gl(struct hns3_enet_ring_group *ring_group)
2382 {
2383 	struct hns3_enet_tqp_vector *tqp_vector =
2384 					ring_group->ring->tqp_vector;
2385 	enum hns3_flow_level_range new_flow_level;
2386 	int packets_per_msecs;
2387 	int bytes_per_msecs;
2388 	u32 time_passed_ms;
2389 	u16 new_int_gl;
2390 
2391 	if (!ring_group->coal.int_gl || !tqp_vector->last_jiffies)
2392 		return false;
2393 
2394 	if (ring_group->total_packets == 0) {
2395 		ring_group->coal.int_gl = HNS3_INT_GL_50K;
2396 		ring_group->coal.flow_level = HNS3_FLOW_LOW;
2397 		return true;
2398 	}
2399 
2400 	/* Simple throttlerate management
2401 	 * 0-10MB/s   lower     (50000 ints/s)
2402 	 * 10-20MB/s   middle    (20000 ints/s)
2403 	 * 20-1249MB/s high      (18000 ints/s)
2404 	 * > 40000pps  ultra     (8000 ints/s)
2405 	 */
2406 	new_flow_level = ring_group->coal.flow_level;
2407 	new_int_gl = ring_group->coal.int_gl;
2408 	time_passed_ms =
2409 		jiffies_to_msecs(jiffies - tqp_vector->last_jiffies);
2410 
2411 	if (!time_passed_ms)
2412 		return false;
2413 
2414 	do_div(ring_group->total_packets, time_passed_ms);
2415 	packets_per_msecs = ring_group->total_packets;
2416 
2417 	do_div(ring_group->total_bytes, time_passed_ms);
2418 	bytes_per_msecs = ring_group->total_bytes;
2419 
2420 #define HNS3_RX_LOW_BYTE_RATE 10000
2421 #define HNS3_RX_MID_BYTE_RATE 20000
2422 
2423 	switch (new_flow_level) {
2424 	case HNS3_FLOW_LOW:
2425 		if (bytes_per_msecs > HNS3_RX_LOW_BYTE_RATE)
2426 			new_flow_level = HNS3_FLOW_MID;
2427 		break;
2428 	case HNS3_FLOW_MID:
2429 		if (bytes_per_msecs > HNS3_RX_MID_BYTE_RATE)
2430 			new_flow_level = HNS3_FLOW_HIGH;
2431 		else if (bytes_per_msecs <= HNS3_RX_LOW_BYTE_RATE)
2432 			new_flow_level = HNS3_FLOW_LOW;
2433 		break;
2434 	case HNS3_FLOW_HIGH:
2435 	case HNS3_FLOW_ULTRA:
2436 	default:
2437 		if (bytes_per_msecs <= HNS3_RX_MID_BYTE_RATE)
2438 			new_flow_level = HNS3_FLOW_MID;
2439 		break;
2440 	}
2441 
2442 #define HNS3_RX_ULTRA_PACKET_RATE 40
2443 
2444 	if (packets_per_msecs > HNS3_RX_ULTRA_PACKET_RATE &&
2445 	    &tqp_vector->rx_group == ring_group)
2446 		new_flow_level = HNS3_FLOW_ULTRA;
2447 
2448 	switch (new_flow_level) {
2449 	case HNS3_FLOW_LOW:
2450 		new_int_gl = HNS3_INT_GL_50K;
2451 		break;
2452 	case HNS3_FLOW_MID:
2453 		new_int_gl = HNS3_INT_GL_20K;
2454 		break;
2455 	case HNS3_FLOW_HIGH:
2456 		new_int_gl = HNS3_INT_GL_18K;
2457 		break;
2458 	case HNS3_FLOW_ULTRA:
2459 		new_int_gl = HNS3_INT_GL_8K;
2460 		break;
2461 	default:
2462 		break;
2463 	}
2464 
2465 	ring_group->total_bytes = 0;
2466 	ring_group->total_packets = 0;
2467 	ring_group->coal.flow_level = new_flow_level;
2468 	if (new_int_gl != ring_group->coal.int_gl) {
2469 		ring_group->coal.int_gl = new_int_gl;
2470 		return true;
2471 	}
2472 	return false;
2473 }
2474 
2475 static void hns3_update_new_int_gl(struct hns3_enet_tqp_vector *tqp_vector)
2476 {
2477 	struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
2478 	struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
2479 	bool rx_update, tx_update;
2480 
2481 	if (tqp_vector->int_adapt_down > 0) {
2482 		tqp_vector->int_adapt_down--;
2483 		return;
2484 	}
2485 
2486 	if (rx_group->coal.gl_adapt_enable) {
2487 		rx_update = hns3_get_new_int_gl(rx_group);
2488 		if (rx_update)
2489 			hns3_set_vector_coalesce_rx_gl(tqp_vector,
2490 						       rx_group->coal.int_gl);
2491 	}
2492 
2493 	if (tx_group->coal.gl_adapt_enable) {
2494 		tx_update = hns3_get_new_int_gl(&tqp_vector->tx_group);
2495 		if (tx_update)
2496 			hns3_set_vector_coalesce_tx_gl(tqp_vector,
2497 						       tx_group->coal.int_gl);
2498 	}
2499 
2500 	tqp_vector->last_jiffies = jiffies;
2501 	tqp_vector->int_adapt_down = HNS3_INT_ADAPT_DOWN_START;
2502 }
2503 
2504 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
2505 {
2506 	struct hns3_enet_ring *ring;
2507 	int rx_pkt_total = 0;
2508 
2509 	struct hns3_enet_tqp_vector *tqp_vector =
2510 		container_of(napi, struct hns3_enet_tqp_vector, napi);
2511 	bool clean_complete = true;
2512 	int rx_budget;
2513 
2514 	/* Since the actual Tx work is minimal, we can give the Tx a larger
2515 	 * budget and be more aggressive about cleaning up the Tx descriptors.
2516 	 */
2517 	hns3_for_each_ring(ring, tqp_vector->tx_group) {
2518 		if (!hns3_clean_tx_ring(ring, budget))
2519 			clean_complete = false;
2520 	}
2521 
2522 	/* make sure rx ring budget not smaller than 1 */
2523 	rx_budget = max(budget / tqp_vector->num_tqps, 1);
2524 
2525 	hns3_for_each_ring(ring, tqp_vector->rx_group) {
2526 		int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
2527 						    hns3_rx_skb);
2528 
2529 		if (rx_cleaned >= rx_budget)
2530 			clean_complete = false;
2531 
2532 		rx_pkt_total += rx_cleaned;
2533 	}
2534 
2535 	tqp_vector->rx_group.total_packets += rx_pkt_total;
2536 
2537 	if (!clean_complete)
2538 		return budget;
2539 
2540 	napi_complete(napi);
2541 	hns3_update_new_int_gl(tqp_vector);
2542 	hns3_mask_vector_irq(tqp_vector, 1);
2543 
2544 	return rx_pkt_total;
2545 }
2546 
2547 static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
2548 				      struct hnae3_ring_chain_node *head)
2549 {
2550 	struct pci_dev *pdev = tqp_vector->handle->pdev;
2551 	struct hnae3_ring_chain_node *cur_chain = head;
2552 	struct hnae3_ring_chain_node *chain;
2553 	struct hns3_enet_ring *tx_ring;
2554 	struct hns3_enet_ring *rx_ring;
2555 
2556 	tx_ring = tqp_vector->tx_group.ring;
2557 	if (tx_ring) {
2558 		cur_chain->tqp_index = tx_ring->tqp->tqp_index;
2559 		hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
2560 			      HNAE3_RING_TYPE_TX);
2561 		hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
2562 				HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_TX);
2563 
2564 		cur_chain->next = NULL;
2565 
2566 		while (tx_ring->next) {
2567 			tx_ring = tx_ring->next;
2568 
2569 			chain = devm_kzalloc(&pdev->dev, sizeof(*chain),
2570 					     GFP_KERNEL);
2571 			if (!chain)
2572 				return -ENOMEM;
2573 
2574 			cur_chain->next = chain;
2575 			chain->tqp_index = tx_ring->tqp->tqp_index;
2576 			hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
2577 				      HNAE3_RING_TYPE_TX);
2578 			hnae3_set_field(chain->int_gl_idx,
2579 					HNAE3_RING_GL_IDX_M,
2580 					HNAE3_RING_GL_IDX_S,
2581 					HNAE3_RING_GL_TX);
2582 
2583 			cur_chain = chain;
2584 		}
2585 	}
2586 
2587 	rx_ring = tqp_vector->rx_group.ring;
2588 	if (!tx_ring && rx_ring) {
2589 		cur_chain->next = NULL;
2590 		cur_chain->tqp_index = rx_ring->tqp->tqp_index;
2591 		hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
2592 			      HNAE3_RING_TYPE_RX);
2593 		hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
2594 				HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
2595 
2596 		rx_ring = rx_ring->next;
2597 	}
2598 
2599 	while (rx_ring) {
2600 		chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
2601 		if (!chain)
2602 			return -ENOMEM;
2603 
2604 		cur_chain->next = chain;
2605 		chain->tqp_index = rx_ring->tqp->tqp_index;
2606 		hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
2607 			      HNAE3_RING_TYPE_RX);
2608 		hnae3_set_field(chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
2609 				HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
2610 
2611 		cur_chain = chain;
2612 
2613 		rx_ring = rx_ring->next;
2614 	}
2615 
2616 	return 0;
2617 }
2618 
2619 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
2620 					struct hnae3_ring_chain_node *head)
2621 {
2622 	struct pci_dev *pdev = tqp_vector->handle->pdev;
2623 	struct hnae3_ring_chain_node *chain_tmp, *chain;
2624 
2625 	chain = head->next;
2626 
2627 	while (chain) {
2628 		chain_tmp = chain->next;
2629 		devm_kfree(&pdev->dev, chain);
2630 		chain = chain_tmp;
2631 	}
2632 }
2633 
2634 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
2635 				   struct hns3_enet_ring *ring)
2636 {
2637 	ring->next = group->ring;
2638 	group->ring = ring;
2639 
2640 	group->count++;
2641 }
2642 
2643 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
2644 {
2645 	struct hnae3_ring_chain_node vector_ring_chain;
2646 	struct hnae3_handle *h = priv->ae_handle;
2647 	struct hns3_enet_tqp_vector *tqp_vector;
2648 	int ret = 0;
2649 	u16 i;
2650 
2651 	for (i = 0; i < priv->vector_num; i++) {
2652 		tqp_vector = &priv->tqp_vector[i];
2653 		hns3_vector_gl_rl_init_hw(tqp_vector, priv);
2654 		tqp_vector->num_tqps = 0;
2655 	}
2656 
2657 	for (i = 0; i < h->kinfo.num_tqps; i++) {
2658 		u16 vector_i = i % priv->vector_num;
2659 		u16 tqp_num = h->kinfo.num_tqps;
2660 
2661 		tqp_vector = &priv->tqp_vector[vector_i];
2662 
2663 		hns3_add_ring_to_group(&tqp_vector->tx_group,
2664 				       priv->ring_data[i].ring);
2665 
2666 		hns3_add_ring_to_group(&tqp_vector->rx_group,
2667 				       priv->ring_data[i + tqp_num].ring);
2668 
2669 		priv->ring_data[i].ring->tqp_vector = tqp_vector;
2670 		priv->ring_data[i + tqp_num].ring->tqp_vector = tqp_vector;
2671 		tqp_vector->num_tqps++;
2672 	}
2673 
2674 	for (i = 0; i < priv->vector_num; i++) {
2675 		tqp_vector = &priv->tqp_vector[i];
2676 
2677 		tqp_vector->rx_group.total_bytes = 0;
2678 		tqp_vector->rx_group.total_packets = 0;
2679 		tqp_vector->tx_group.total_bytes = 0;
2680 		tqp_vector->tx_group.total_packets = 0;
2681 		tqp_vector->handle = h;
2682 
2683 		ret = hns3_get_vector_ring_chain(tqp_vector,
2684 						 &vector_ring_chain);
2685 		if (ret)
2686 			return ret;
2687 
2688 		ret = h->ae_algo->ops->map_ring_to_vector(h,
2689 			tqp_vector->vector_irq, &vector_ring_chain);
2690 
2691 		hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
2692 
2693 		if (ret)
2694 			return ret;
2695 
2696 		netif_napi_add(priv->netdev, &tqp_vector->napi,
2697 			       hns3_nic_common_poll, NAPI_POLL_WEIGHT);
2698 	}
2699 
2700 	return 0;
2701 }
2702 
2703 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
2704 {
2705 	struct hnae3_handle *h = priv->ae_handle;
2706 	struct hns3_enet_tqp_vector *tqp_vector;
2707 	struct hnae3_vector_info *vector;
2708 	struct pci_dev *pdev = h->pdev;
2709 	u16 tqp_num = h->kinfo.num_tqps;
2710 	u16 vector_num;
2711 	int ret = 0;
2712 	u16 i;
2713 
2714 	/* RSS size, cpu online and vector_num should be the same */
2715 	/* Should consider 2p/4p later */
2716 	vector_num = min_t(u16, num_online_cpus(), tqp_num);
2717 	vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
2718 			      GFP_KERNEL);
2719 	if (!vector)
2720 		return -ENOMEM;
2721 
2722 	vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
2723 
2724 	priv->vector_num = vector_num;
2725 	priv->tqp_vector = (struct hns3_enet_tqp_vector *)
2726 		devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
2727 			     GFP_KERNEL);
2728 	if (!priv->tqp_vector) {
2729 		ret = -ENOMEM;
2730 		goto out;
2731 	}
2732 
2733 	for (i = 0; i < priv->vector_num; i++) {
2734 		tqp_vector = &priv->tqp_vector[i];
2735 		tqp_vector->idx = i;
2736 		tqp_vector->mask_addr = vector[i].io_addr;
2737 		tqp_vector->vector_irq = vector[i].vector;
2738 		hns3_vector_gl_rl_init(tqp_vector, priv);
2739 	}
2740 
2741 out:
2742 	devm_kfree(&pdev->dev, vector);
2743 	return ret;
2744 }
2745 
2746 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
2747 {
2748 	group->ring = NULL;
2749 	group->count = 0;
2750 }
2751 
2752 static int hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
2753 {
2754 	struct hnae3_ring_chain_node vector_ring_chain;
2755 	struct hnae3_handle *h = priv->ae_handle;
2756 	struct hns3_enet_tqp_vector *tqp_vector;
2757 	int i, ret;
2758 
2759 	for (i = 0; i < priv->vector_num; i++) {
2760 		tqp_vector = &priv->tqp_vector[i];
2761 
2762 		ret = hns3_get_vector_ring_chain(tqp_vector,
2763 						 &vector_ring_chain);
2764 		if (ret)
2765 			return ret;
2766 
2767 		ret = h->ae_algo->ops->unmap_ring_from_vector(h,
2768 			tqp_vector->vector_irq, &vector_ring_chain);
2769 		if (ret)
2770 			return ret;
2771 
2772 		hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
2773 
2774 		if (priv->tqp_vector[i].irq_init_flag == HNS3_VECTOR_INITED) {
2775 			(void)irq_set_affinity_hint(
2776 				priv->tqp_vector[i].vector_irq,
2777 						    NULL);
2778 			free_irq(priv->tqp_vector[i].vector_irq,
2779 				 &priv->tqp_vector[i]);
2780 		}
2781 
2782 		priv->ring_data[i].ring->irq_init_flag = HNS3_VECTOR_NOT_INITED;
2783 		hns3_clear_ring_group(&tqp_vector->rx_group);
2784 		hns3_clear_ring_group(&tqp_vector->tx_group);
2785 		netif_napi_del(&priv->tqp_vector[i].napi);
2786 	}
2787 
2788 	return 0;
2789 }
2790 
2791 static int hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
2792 {
2793 	struct hnae3_handle *h = priv->ae_handle;
2794 	struct pci_dev *pdev = h->pdev;
2795 	int i, ret;
2796 
2797 	for (i = 0; i < priv->vector_num; i++) {
2798 		struct hns3_enet_tqp_vector *tqp_vector;
2799 
2800 		tqp_vector = &priv->tqp_vector[i];
2801 		ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
2802 		if (ret)
2803 			return ret;
2804 	}
2805 
2806 	devm_kfree(&pdev->dev, priv->tqp_vector);
2807 	return 0;
2808 }
2809 
2810 static int hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
2811 			     int ring_type)
2812 {
2813 	struct hns3_nic_ring_data *ring_data = priv->ring_data;
2814 	int queue_num = priv->ae_handle->kinfo.num_tqps;
2815 	struct pci_dev *pdev = priv->ae_handle->pdev;
2816 	struct hns3_enet_ring *ring;
2817 
2818 	ring = devm_kzalloc(&pdev->dev, sizeof(*ring), GFP_KERNEL);
2819 	if (!ring)
2820 		return -ENOMEM;
2821 
2822 	if (ring_type == HNAE3_RING_TYPE_TX) {
2823 		ring_data[q->tqp_index].ring = ring;
2824 		ring_data[q->tqp_index].queue_index = q->tqp_index;
2825 		ring->io_base = (u8 __iomem *)q->io_base + HNS3_TX_REG_OFFSET;
2826 	} else {
2827 		ring_data[q->tqp_index + queue_num].ring = ring;
2828 		ring_data[q->tqp_index + queue_num].queue_index = q->tqp_index;
2829 		ring->io_base = q->io_base;
2830 	}
2831 
2832 	hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
2833 
2834 	ring->tqp = q;
2835 	ring->desc = NULL;
2836 	ring->desc_cb = NULL;
2837 	ring->dev = priv->dev;
2838 	ring->desc_dma_addr = 0;
2839 	ring->buf_size = q->buf_size;
2840 	ring->desc_num = q->desc_num;
2841 	ring->next_to_use = 0;
2842 	ring->next_to_clean = 0;
2843 
2844 	return 0;
2845 }
2846 
2847 static int hns3_queue_to_ring(struct hnae3_queue *tqp,
2848 			      struct hns3_nic_priv *priv)
2849 {
2850 	int ret;
2851 
2852 	ret = hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
2853 	if (ret)
2854 		return ret;
2855 
2856 	ret = hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
2857 	if (ret)
2858 		return ret;
2859 
2860 	return 0;
2861 }
2862 
2863 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
2864 {
2865 	struct hnae3_handle *h = priv->ae_handle;
2866 	struct pci_dev *pdev = h->pdev;
2867 	int i, ret;
2868 
2869 	priv->ring_data =  devm_kzalloc(&pdev->dev,
2870 					array3_size(h->kinfo.num_tqps,
2871 						    sizeof(*priv->ring_data),
2872 						    2),
2873 					GFP_KERNEL);
2874 	if (!priv->ring_data)
2875 		return -ENOMEM;
2876 
2877 	for (i = 0; i < h->kinfo.num_tqps; i++) {
2878 		ret = hns3_queue_to_ring(h->kinfo.tqp[i], priv);
2879 		if (ret)
2880 			goto err;
2881 	}
2882 
2883 	return 0;
2884 err:
2885 	devm_kfree(&pdev->dev, priv->ring_data);
2886 	return ret;
2887 }
2888 
2889 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
2890 {
2891 	struct hnae3_handle *h = priv->ae_handle;
2892 	int i;
2893 
2894 	for (i = 0; i < h->kinfo.num_tqps; i++) {
2895 		devm_kfree(priv->dev, priv->ring_data[i].ring);
2896 		devm_kfree(priv->dev,
2897 			   priv->ring_data[i + h->kinfo.num_tqps].ring);
2898 	}
2899 	devm_kfree(priv->dev, priv->ring_data);
2900 }
2901 
2902 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
2903 {
2904 	int ret;
2905 
2906 	if (ring->desc_num <= 0 || ring->buf_size <= 0)
2907 		return -EINVAL;
2908 
2909 	ring->desc_cb = kcalloc(ring->desc_num, sizeof(ring->desc_cb[0]),
2910 				GFP_KERNEL);
2911 	if (!ring->desc_cb) {
2912 		ret = -ENOMEM;
2913 		goto out;
2914 	}
2915 
2916 	ret = hns3_alloc_desc(ring);
2917 	if (ret)
2918 		goto out_with_desc_cb;
2919 
2920 	if (!HNAE3_IS_TX_RING(ring)) {
2921 		ret = hns3_alloc_ring_buffers(ring);
2922 		if (ret)
2923 			goto out_with_desc;
2924 	}
2925 
2926 	return 0;
2927 
2928 out_with_desc:
2929 	hns3_free_desc(ring);
2930 out_with_desc_cb:
2931 	kfree(ring->desc_cb);
2932 	ring->desc_cb = NULL;
2933 out:
2934 	return ret;
2935 }
2936 
2937 static void hns3_fini_ring(struct hns3_enet_ring *ring)
2938 {
2939 	hns3_free_desc(ring);
2940 	kfree(ring->desc_cb);
2941 	ring->desc_cb = NULL;
2942 	ring->next_to_clean = 0;
2943 	ring->next_to_use = 0;
2944 }
2945 
2946 static int hns3_buf_size2type(u32 buf_size)
2947 {
2948 	int bd_size_type;
2949 
2950 	switch (buf_size) {
2951 	case 512:
2952 		bd_size_type = HNS3_BD_SIZE_512_TYPE;
2953 		break;
2954 	case 1024:
2955 		bd_size_type = HNS3_BD_SIZE_1024_TYPE;
2956 		break;
2957 	case 2048:
2958 		bd_size_type = HNS3_BD_SIZE_2048_TYPE;
2959 		break;
2960 	case 4096:
2961 		bd_size_type = HNS3_BD_SIZE_4096_TYPE;
2962 		break;
2963 	default:
2964 		bd_size_type = HNS3_BD_SIZE_2048_TYPE;
2965 	}
2966 
2967 	return bd_size_type;
2968 }
2969 
2970 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
2971 {
2972 	dma_addr_t dma = ring->desc_dma_addr;
2973 	struct hnae3_queue *q = ring->tqp;
2974 
2975 	if (!HNAE3_IS_TX_RING(ring)) {
2976 		hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG,
2977 			       (u32)dma);
2978 		hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
2979 			       (u32)((dma >> 31) >> 1));
2980 
2981 		hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
2982 			       hns3_buf_size2type(ring->buf_size));
2983 		hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
2984 			       ring->desc_num / 8 - 1);
2985 
2986 	} else {
2987 		hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
2988 			       (u32)dma);
2989 		hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
2990 			       (u32)((dma >> 31) >> 1));
2991 
2992 		hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
2993 			       ring->desc_num / 8 - 1);
2994 	}
2995 }
2996 
2997 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
2998 {
2999 	struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
3000 	int i;
3001 
3002 	for (i = 0; i < HNAE3_MAX_TC; i++) {
3003 		struct hnae3_tc_info *tc_info = &kinfo->tc_info[i];
3004 		int j;
3005 
3006 		if (!tc_info->enable)
3007 			continue;
3008 
3009 		for (j = 0; j < tc_info->tqp_count; j++) {
3010 			struct hnae3_queue *q;
3011 
3012 			q = priv->ring_data[tc_info->tqp_offset + j].ring->tqp;
3013 			hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG,
3014 				       tc_info->tc);
3015 		}
3016 	}
3017 }
3018 
3019 int hns3_init_all_ring(struct hns3_nic_priv *priv)
3020 {
3021 	struct hnae3_handle *h = priv->ae_handle;
3022 	int ring_num = h->kinfo.num_tqps * 2;
3023 	int i, j;
3024 	int ret;
3025 
3026 	for (i = 0; i < ring_num; i++) {
3027 		ret = hns3_alloc_ring_memory(priv->ring_data[i].ring);
3028 		if (ret) {
3029 			dev_err(priv->dev,
3030 				"Alloc ring memory fail! ret=%d\n", ret);
3031 			goto out_when_alloc_ring_memory;
3032 		}
3033 
3034 		u64_stats_init(&priv->ring_data[i].ring->syncp);
3035 	}
3036 
3037 	return 0;
3038 
3039 out_when_alloc_ring_memory:
3040 	for (j = i - 1; j >= 0; j--)
3041 		hns3_fini_ring(priv->ring_data[j].ring);
3042 
3043 	return -ENOMEM;
3044 }
3045 
3046 int hns3_uninit_all_ring(struct hns3_nic_priv *priv)
3047 {
3048 	struct hnae3_handle *h = priv->ae_handle;
3049 	int i;
3050 
3051 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3052 		if (h->ae_algo->ops->reset_queue)
3053 			h->ae_algo->ops->reset_queue(h, i);
3054 
3055 		hns3_fini_ring(priv->ring_data[i].ring);
3056 		hns3_fini_ring(priv->ring_data[i + h->kinfo.num_tqps].ring);
3057 	}
3058 	return 0;
3059 }
3060 
3061 /* Set mac addr if it is configured. or leave it to the AE driver */
3062 static void hns3_init_mac_addr(struct net_device *netdev, bool init)
3063 {
3064 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3065 	struct hnae3_handle *h = priv->ae_handle;
3066 	u8 mac_addr_temp[ETH_ALEN];
3067 
3068 	if (h->ae_algo->ops->get_mac_addr && init) {
3069 		h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
3070 		ether_addr_copy(netdev->dev_addr, mac_addr_temp);
3071 	}
3072 
3073 	/* Check if the MAC address is valid, if not get a random one */
3074 	if (!is_valid_ether_addr(netdev->dev_addr)) {
3075 		eth_hw_addr_random(netdev);
3076 		dev_warn(priv->dev, "using random MAC address %pM\n",
3077 			 netdev->dev_addr);
3078 	}
3079 
3080 	if (h->ae_algo->ops->set_mac_addr)
3081 		h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
3082 
3083 }
3084 
3085 static void hns3_uninit_mac_addr(struct net_device *netdev)
3086 {
3087 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3088 	struct hnae3_handle *h = priv->ae_handle;
3089 
3090 	if (h->ae_algo->ops->rm_uc_addr)
3091 		h->ae_algo->ops->rm_uc_addr(h, netdev->dev_addr);
3092 }
3093 
3094 static void hns3_nic_set_priv_ops(struct net_device *netdev)
3095 {
3096 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3097 
3098 	if ((netdev->features & NETIF_F_TSO) ||
3099 	    (netdev->features & NETIF_F_TSO6)) {
3100 		priv->ops.fill_desc = hns3_fill_desc_tso;
3101 		priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tso;
3102 	} else {
3103 		priv->ops.fill_desc = hns3_fill_desc;
3104 		priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tx;
3105 	}
3106 }
3107 
3108 static int hns3_client_init(struct hnae3_handle *handle)
3109 {
3110 	struct pci_dev *pdev = handle->pdev;
3111 	struct hns3_nic_priv *priv;
3112 	struct net_device *netdev;
3113 	int ret;
3114 
3115 	netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv),
3116 				   hns3_get_max_available_channels(handle));
3117 	if (!netdev)
3118 		return -ENOMEM;
3119 
3120 	priv = netdev_priv(netdev);
3121 	priv->dev = &pdev->dev;
3122 	priv->netdev = netdev;
3123 	priv->ae_handle = handle;
3124 	priv->ae_handle->last_reset_time = jiffies;
3125 	priv->tx_timeout_count = 0;
3126 
3127 	handle->kinfo.netdev = netdev;
3128 	handle->priv = (void *)priv;
3129 
3130 	hns3_init_mac_addr(netdev, true);
3131 
3132 	hns3_set_default_feature(netdev);
3133 
3134 	netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
3135 	netdev->priv_flags |= IFF_UNICAST_FLT;
3136 	netdev->netdev_ops = &hns3_nic_netdev_ops;
3137 	SET_NETDEV_DEV(netdev, &pdev->dev);
3138 	hns3_ethtool_set_ops(netdev);
3139 	hns3_nic_set_priv_ops(netdev);
3140 
3141 	/* Carrier off reporting is important to ethtool even BEFORE open */
3142 	netif_carrier_off(netdev);
3143 
3144 	if (handle->flags & HNAE3_SUPPORT_VF)
3145 		handle->reset_level = HNAE3_VF_RESET;
3146 	else
3147 		handle->reset_level = HNAE3_FUNC_RESET;
3148 
3149 	ret = hns3_get_ring_config(priv);
3150 	if (ret) {
3151 		ret = -ENOMEM;
3152 		goto out_get_ring_cfg;
3153 	}
3154 
3155 	ret = hns3_nic_alloc_vector_data(priv);
3156 	if (ret) {
3157 		ret = -ENOMEM;
3158 		goto out_alloc_vector_data;
3159 	}
3160 
3161 	ret = hns3_nic_init_vector_data(priv);
3162 	if (ret) {
3163 		ret = -ENOMEM;
3164 		goto out_init_vector_data;
3165 	}
3166 
3167 	ret = hns3_init_all_ring(priv);
3168 	if (ret) {
3169 		ret = -ENOMEM;
3170 		goto out_init_ring_data;
3171 	}
3172 
3173 	ret = register_netdev(netdev);
3174 	if (ret) {
3175 		dev_err(priv->dev, "probe register netdev fail!\n");
3176 		goto out_reg_netdev_fail;
3177 	}
3178 
3179 	hns3_dcbnl_setup(handle);
3180 
3181 	/* MTU range: (ETH_MIN_MTU(kernel default) - 9706) */
3182 	netdev->max_mtu = HNS3_MAX_MTU - (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
3183 
3184 	return ret;
3185 
3186 out_reg_netdev_fail:
3187 out_init_ring_data:
3188 	(void)hns3_nic_uninit_vector_data(priv);
3189 out_init_vector_data:
3190 	hns3_nic_dealloc_vector_data(priv);
3191 out_alloc_vector_data:
3192 	priv->ring_data = NULL;
3193 out_get_ring_cfg:
3194 	priv->ae_handle = NULL;
3195 	free_netdev(netdev);
3196 	return ret;
3197 }
3198 
3199 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
3200 {
3201 	struct net_device *netdev = handle->kinfo.netdev;
3202 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3203 	int ret;
3204 
3205 	if (netdev->reg_state != NETREG_UNINITIALIZED)
3206 		unregister_netdev(netdev);
3207 
3208 	hns3_force_clear_all_rx_ring(handle);
3209 
3210 	ret = hns3_nic_uninit_vector_data(priv);
3211 	if (ret)
3212 		netdev_err(netdev, "uninit vector error\n");
3213 
3214 	ret = hns3_nic_dealloc_vector_data(priv);
3215 	if (ret)
3216 		netdev_err(netdev, "dealloc vector error\n");
3217 
3218 	ret = hns3_uninit_all_ring(priv);
3219 	if (ret)
3220 		netdev_err(netdev, "uninit ring error\n");
3221 
3222 	hns3_put_ring_config(priv);
3223 
3224 	priv->ring_data = NULL;
3225 
3226 	hns3_uninit_mac_addr(netdev);
3227 
3228 	free_netdev(netdev);
3229 }
3230 
3231 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
3232 {
3233 	struct net_device *netdev = handle->kinfo.netdev;
3234 
3235 	if (!netdev)
3236 		return;
3237 
3238 	if (linkup) {
3239 		netif_carrier_on(netdev);
3240 		netif_tx_wake_all_queues(netdev);
3241 		netdev_info(netdev, "link up\n");
3242 	} else {
3243 		netif_carrier_off(netdev);
3244 		netif_tx_stop_all_queues(netdev);
3245 		netdev_info(netdev, "link down\n");
3246 	}
3247 }
3248 
3249 static int hns3_client_setup_tc(struct hnae3_handle *handle, u8 tc)
3250 {
3251 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
3252 	struct net_device *ndev = kinfo->netdev;
3253 	bool if_running;
3254 	int ret;
3255 
3256 	if (tc > HNAE3_MAX_TC)
3257 		return -EINVAL;
3258 
3259 	if (!ndev)
3260 		return -ENODEV;
3261 
3262 	if_running = netif_running(ndev);
3263 
3264 	if (if_running) {
3265 		(void)hns3_nic_net_stop(ndev);
3266 		msleep(100);
3267 	}
3268 
3269 	ret = (kinfo->dcb_ops && kinfo->dcb_ops->map_update) ?
3270 		kinfo->dcb_ops->map_update(handle) : -EOPNOTSUPP;
3271 	if (ret)
3272 		goto err_out;
3273 
3274 	ret = hns3_nic_set_real_num_queue(ndev);
3275 
3276 err_out:
3277 	if (if_running)
3278 		(void)hns3_nic_net_open(ndev);
3279 
3280 	return ret;
3281 }
3282 
3283 static void hns3_recover_hw_addr(struct net_device *ndev)
3284 {
3285 	struct netdev_hw_addr_list *list;
3286 	struct netdev_hw_addr *ha, *tmp;
3287 
3288 	/* go through and sync uc_addr entries to the device */
3289 	list = &ndev->uc;
3290 	list_for_each_entry_safe(ha, tmp, &list->list, list)
3291 		hns3_nic_uc_sync(ndev, ha->addr);
3292 
3293 	/* go through and sync mc_addr entries to the device */
3294 	list = &ndev->mc;
3295 	list_for_each_entry_safe(ha, tmp, &list->list, list)
3296 		hns3_nic_mc_sync(ndev, ha->addr);
3297 }
3298 
3299 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
3300 {
3301 	while (ring->next_to_clean != ring->next_to_use) {
3302 		ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
3303 		hns3_free_buffer_detach(ring, ring->next_to_clean);
3304 		ring_ptr_move_fw(ring, next_to_clean);
3305 	}
3306 }
3307 
3308 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
3309 {
3310 	struct hns3_desc_cb res_cbs;
3311 	int ret;
3312 
3313 	while (ring->next_to_use != ring->next_to_clean) {
3314 		/* When a buffer is not reused, it's memory has been
3315 		 * freed in hns3_handle_rx_bd or will be freed by
3316 		 * stack, so we need to replace the buffer here.
3317 		 */
3318 		if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
3319 			ret = hns3_reserve_buffer_map(ring, &res_cbs);
3320 			if (ret) {
3321 				u64_stats_update_begin(&ring->syncp);
3322 				ring->stats.sw_err_cnt++;
3323 				u64_stats_update_end(&ring->syncp);
3324 				/* if alloc new buffer fail, exit directly
3325 				 * and reclear in up flow.
3326 				 */
3327 				netdev_warn(ring->tqp->handle->kinfo.netdev,
3328 					    "reserve buffer map failed, ret = %d\n",
3329 					    ret);
3330 				return ret;
3331 			}
3332 			hns3_replace_buffer(ring, ring->next_to_use,
3333 					    &res_cbs);
3334 		}
3335 		ring_ptr_move_fw(ring, next_to_use);
3336 	}
3337 
3338 	return 0;
3339 }
3340 
3341 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
3342 {
3343 	while (ring->next_to_use != ring->next_to_clean) {
3344 		/* When a buffer is not reused, it's memory has been
3345 		 * freed in hns3_handle_rx_bd or will be freed by
3346 		 * stack, so only need to unmap the buffer here.
3347 		 */
3348 		if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
3349 			hns3_unmap_buffer(ring,
3350 					  &ring->desc_cb[ring->next_to_use]);
3351 			ring->desc_cb[ring->next_to_use].dma = 0;
3352 		}
3353 
3354 		ring_ptr_move_fw(ring, next_to_use);
3355 	}
3356 }
3357 
3358 static void hns3_force_clear_all_rx_ring(struct hnae3_handle *h)
3359 {
3360 	struct net_device *ndev = h->kinfo.netdev;
3361 	struct hns3_nic_priv *priv = netdev_priv(ndev);
3362 	struct hns3_enet_ring *ring;
3363 	u32 i;
3364 
3365 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3366 		ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
3367 		hns3_force_clear_rx_ring(ring);
3368 	}
3369 }
3370 
3371 static void hns3_clear_all_ring(struct hnae3_handle *h)
3372 {
3373 	struct net_device *ndev = h->kinfo.netdev;
3374 	struct hns3_nic_priv *priv = netdev_priv(ndev);
3375 	u32 i;
3376 
3377 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3378 		struct netdev_queue *dev_queue;
3379 		struct hns3_enet_ring *ring;
3380 
3381 		ring = priv->ring_data[i].ring;
3382 		hns3_clear_tx_ring(ring);
3383 		dev_queue = netdev_get_tx_queue(ndev,
3384 						priv->ring_data[i].queue_index);
3385 		netdev_tx_reset_queue(dev_queue);
3386 
3387 		ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
3388 		/* Continue to clear other rings even if clearing some
3389 		 * rings failed.
3390 		 */
3391 		hns3_clear_rx_ring(ring);
3392 	}
3393 }
3394 
3395 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
3396 {
3397 	struct net_device *ndev = h->kinfo.netdev;
3398 	struct hns3_nic_priv *priv = netdev_priv(ndev);
3399 	struct hns3_enet_ring *rx_ring;
3400 	int i, j;
3401 	int ret;
3402 
3403 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3404 		h->ae_algo->ops->reset_queue(h, i);
3405 		hns3_init_ring_hw(priv->ring_data[i].ring);
3406 
3407 		/* We need to clear tx ring here because self test will
3408 		 * use the ring and will not run down before up
3409 		 */
3410 		hns3_clear_tx_ring(priv->ring_data[i].ring);
3411 		priv->ring_data[i].ring->next_to_clean = 0;
3412 		priv->ring_data[i].ring->next_to_use = 0;
3413 
3414 		rx_ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
3415 		hns3_init_ring_hw(rx_ring);
3416 		ret = hns3_clear_rx_ring(rx_ring);
3417 		if (ret)
3418 			return ret;
3419 
3420 		/* We can not know the hardware head and tail when this
3421 		 * function is called in reset flow, so we reuse all desc.
3422 		 */
3423 		for (j = 0; j < rx_ring->desc_num; j++)
3424 			hns3_reuse_buffer(rx_ring, j);
3425 
3426 		rx_ring->next_to_clean = 0;
3427 		rx_ring->next_to_use = 0;
3428 	}
3429 
3430 	hns3_init_tx_ring_tc(priv);
3431 
3432 	return 0;
3433 }
3434 
3435 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
3436 {
3437 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
3438 	struct net_device *ndev = kinfo->netdev;
3439 
3440 	if (!netif_running(ndev))
3441 		return 0;
3442 
3443 	return hns3_nic_net_stop(ndev);
3444 }
3445 
3446 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
3447 {
3448 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
3449 	int ret = 0;
3450 
3451 	if (netif_running(kinfo->netdev)) {
3452 		ret = hns3_nic_net_up(kinfo->netdev);
3453 		if (ret) {
3454 			netdev_err(kinfo->netdev,
3455 				   "hns net up fail, ret=%d!\n", ret);
3456 			return ret;
3457 		}
3458 		handle->last_reset_time = jiffies;
3459 	}
3460 
3461 	return ret;
3462 }
3463 
3464 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
3465 {
3466 	struct net_device *netdev = handle->kinfo.netdev;
3467 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3468 	int ret;
3469 
3470 	hns3_init_mac_addr(netdev, false);
3471 	hns3_nic_set_rx_mode(netdev);
3472 	hns3_recover_hw_addr(netdev);
3473 
3474 	/* Hardware table is only clear when pf resets */
3475 	if (!(handle->flags & HNAE3_SUPPORT_VF))
3476 		hns3_restore_vlan(netdev);
3477 
3478 	/* Carrier off reporting is important to ethtool even BEFORE open */
3479 	netif_carrier_off(netdev);
3480 
3481 	ret = hns3_nic_init_vector_data(priv);
3482 	if (ret)
3483 		return ret;
3484 
3485 	ret = hns3_init_all_ring(priv);
3486 	if (ret) {
3487 		hns3_nic_uninit_vector_data(priv);
3488 		priv->ring_data = NULL;
3489 	}
3490 
3491 	return ret;
3492 }
3493 
3494 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
3495 {
3496 	struct net_device *netdev = handle->kinfo.netdev;
3497 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3498 	int ret;
3499 
3500 	hns3_force_clear_all_rx_ring(handle);
3501 
3502 	ret = hns3_nic_uninit_vector_data(priv);
3503 	if (ret) {
3504 		netdev_err(netdev, "uninit vector error\n");
3505 		return ret;
3506 	}
3507 
3508 	ret = hns3_uninit_all_ring(priv);
3509 	if (ret)
3510 		netdev_err(netdev, "uninit ring error\n");
3511 
3512 	hns3_uninit_mac_addr(netdev);
3513 
3514 	return ret;
3515 }
3516 
3517 static int hns3_reset_notify(struct hnae3_handle *handle,
3518 			     enum hnae3_reset_notify_type type)
3519 {
3520 	int ret = 0;
3521 
3522 	switch (type) {
3523 	case HNAE3_UP_CLIENT:
3524 		ret = hns3_reset_notify_up_enet(handle);
3525 		break;
3526 	case HNAE3_DOWN_CLIENT:
3527 		ret = hns3_reset_notify_down_enet(handle);
3528 		break;
3529 	case HNAE3_INIT_CLIENT:
3530 		ret = hns3_reset_notify_init_enet(handle);
3531 		break;
3532 	case HNAE3_UNINIT_CLIENT:
3533 		ret = hns3_reset_notify_uninit_enet(handle);
3534 		break;
3535 	default:
3536 		break;
3537 	}
3538 
3539 	return ret;
3540 }
3541 
3542 static void hns3_restore_coal(struct hns3_nic_priv *priv,
3543 			      struct hns3_enet_coalesce *tx,
3544 			      struct hns3_enet_coalesce *rx)
3545 {
3546 	u16 vector_num = priv->vector_num;
3547 	int i;
3548 
3549 	for (i = 0; i < vector_num; i++) {
3550 		memcpy(&priv->tqp_vector[i].tx_group.coal, tx,
3551 		       sizeof(struct hns3_enet_coalesce));
3552 		memcpy(&priv->tqp_vector[i].rx_group.coal, rx,
3553 		       sizeof(struct hns3_enet_coalesce));
3554 	}
3555 }
3556 
3557 static int hns3_modify_tqp_num(struct net_device *netdev, u16 new_tqp_num,
3558 			       struct hns3_enet_coalesce *tx,
3559 			       struct hns3_enet_coalesce *rx)
3560 {
3561 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3562 	struct hnae3_handle *h = hns3_get_handle(netdev);
3563 	int ret;
3564 
3565 	ret = h->ae_algo->ops->set_channels(h, new_tqp_num);
3566 	if (ret)
3567 		return ret;
3568 
3569 	ret = hns3_get_ring_config(priv);
3570 	if (ret)
3571 		return ret;
3572 
3573 	ret = hns3_nic_alloc_vector_data(priv);
3574 	if (ret)
3575 		goto err_alloc_vector;
3576 
3577 	hns3_restore_coal(priv, tx, rx);
3578 
3579 	ret = hns3_nic_init_vector_data(priv);
3580 	if (ret)
3581 		goto err_uninit_vector;
3582 
3583 	ret = hns3_init_all_ring(priv);
3584 	if (ret)
3585 		goto err_put_ring;
3586 
3587 	return 0;
3588 
3589 err_put_ring:
3590 	hns3_put_ring_config(priv);
3591 err_uninit_vector:
3592 	hns3_nic_uninit_vector_data(priv);
3593 err_alloc_vector:
3594 	hns3_nic_dealloc_vector_data(priv);
3595 	return ret;
3596 }
3597 
3598 static int hns3_adjust_tqps_num(u8 num_tc, u32 new_tqp_num)
3599 {
3600 	return (new_tqp_num / num_tc) * num_tc;
3601 }
3602 
3603 int hns3_set_channels(struct net_device *netdev,
3604 		      struct ethtool_channels *ch)
3605 {
3606 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3607 	struct hnae3_handle *h = hns3_get_handle(netdev);
3608 	struct hnae3_knic_private_info *kinfo = &h->kinfo;
3609 	struct hns3_enet_coalesce tx_coal, rx_coal;
3610 	bool if_running = netif_running(netdev);
3611 	u32 new_tqp_num = ch->combined_count;
3612 	u16 org_tqp_num;
3613 	int ret;
3614 
3615 	if (ch->rx_count || ch->tx_count)
3616 		return -EINVAL;
3617 
3618 	if (new_tqp_num > hns3_get_max_available_channels(h) ||
3619 	    new_tqp_num < kinfo->num_tc) {
3620 		dev_err(&netdev->dev,
3621 			"Change tqps fail, the tqp range is from %d to %d",
3622 			kinfo->num_tc,
3623 			hns3_get_max_available_channels(h));
3624 		return -EINVAL;
3625 	}
3626 
3627 	new_tqp_num = hns3_adjust_tqps_num(kinfo->num_tc, new_tqp_num);
3628 	if (kinfo->num_tqps == new_tqp_num)
3629 		return 0;
3630 
3631 	if (if_running)
3632 		hns3_nic_net_stop(netdev);
3633 
3634 	ret = hns3_nic_uninit_vector_data(priv);
3635 	if (ret) {
3636 		dev_err(&netdev->dev,
3637 			"Unbind vector with tqp fail, nothing is changed");
3638 		goto open_netdev;
3639 	}
3640 
3641 	/* Changing the tqp num may also change the vector num,
3642 	 * ethtool only support setting and querying one coal
3643 	 * configuation for now, so save the vector 0' coal
3644 	 * configuation here in order to restore it.
3645 	 */
3646 	memcpy(&tx_coal, &priv->tqp_vector[0].tx_group.coal,
3647 	       sizeof(struct hns3_enet_coalesce));
3648 	memcpy(&rx_coal, &priv->tqp_vector[0].rx_group.coal,
3649 	       sizeof(struct hns3_enet_coalesce));
3650 
3651 	hns3_nic_dealloc_vector_data(priv);
3652 
3653 	hns3_uninit_all_ring(priv);
3654 	hns3_put_ring_config(priv);
3655 
3656 	org_tqp_num = h->kinfo.num_tqps;
3657 	ret = hns3_modify_tqp_num(netdev, new_tqp_num, &tx_coal, &rx_coal);
3658 	if (ret) {
3659 		ret = hns3_modify_tqp_num(netdev, org_tqp_num,
3660 					  &tx_coal, &rx_coal);
3661 		if (ret) {
3662 			/* If revert to old tqp failed, fatal error occurred */
3663 			dev_err(&netdev->dev,
3664 				"Revert to old tqp num fail, ret=%d", ret);
3665 			return ret;
3666 		}
3667 		dev_info(&netdev->dev,
3668 			 "Change tqp num fail, Revert to old tqp num");
3669 	}
3670 
3671 open_netdev:
3672 	if (if_running)
3673 		hns3_nic_net_open(netdev);
3674 
3675 	return ret;
3676 }
3677 
3678 static const struct hnae3_client_ops client_ops = {
3679 	.init_instance = hns3_client_init,
3680 	.uninit_instance = hns3_client_uninit,
3681 	.link_status_change = hns3_link_status_change,
3682 	.setup_tc = hns3_client_setup_tc,
3683 	.reset_notify = hns3_reset_notify,
3684 };
3685 
3686 /* hns3_init_module - Driver registration routine
3687  * hns3_init_module is the first routine called when the driver is
3688  * loaded. All it does is register with the PCI subsystem.
3689  */
3690 static int __init hns3_init_module(void)
3691 {
3692 	int ret;
3693 
3694 	pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
3695 	pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
3696 
3697 	client.type = HNAE3_CLIENT_KNIC;
3698 	snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH - 1, "%s",
3699 		 hns3_driver_name);
3700 
3701 	client.ops = &client_ops;
3702 
3703 	INIT_LIST_HEAD(&client.node);
3704 
3705 	ret = hnae3_register_client(&client);
3706 	if (ret)
3707 		return ret;
3708 
3709 	ret = pci_register_driver(&hns3_driver);
3710 	if (ret)
3711 		hnae3_unregister_client(&client);
3712 
3713 	return ret;
3714 }
3715 module_init(hns3_init_module);
3716 
3717 /* hns3_exit_module - Driver exit cleanup routine
3718  * hns3_exit_module is called just before the driver is removed
3719  * from memory.
3720  */
3721 static void __exit hns3_exit_module(void)
3722 {
3723 	pci_unregister_driver(&hns3_driver);
3724 	hnae3_unregister_client(&client);
3725 }
3726 module_exit(hns3_exit_module);
3727 
3728 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
3729 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
3730 MODULE_LICENSE("GPL");
3731 MODULE_ALIAS("pci:hns-nic");
3732 MODULE_VERSION(HNS3_MOD_VERSION);
3733