xref: /openbmc/linux/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c (revision 7fc38225363dd8f19e667ad7c77b63bc4a5c065d)
1 // SPDX-License-Identifier: GPL-2.0+
2 // Copyright (c) 2016-2017 Hisilicon Limited.
3 
4 #include <linux/dma-mapping.h>
5 #include <linux/etherdevice.h>
6 #include <linux/interrupt.h>
7 #include <linux/if_vlan.h>
8 #include <linux/ip.h>
9 #include <linux/ipv6.h>
10 #include <linux/module.h>
11 #include <linux/pci.h>
12 #include <linux/aer.h>
13 #include <linux/skbuff.h>
14 #include <linux/sctp.h>
15 #include <linux/vermagic.h>
16 #include <net/gre.h>
17 #include <net/pkt_cls.h>
18 #include <net/tcp.h>
19 #include <net/vxlan.h>
20 
21 #include "hnae3.h"
22 #include "hns3_enet.h"
23 
24 static void hns3_clear_all_ring(struct hnae3_handle *h);
25 static void hns3_force_clear_all_rx_ring(struct hnae3_handle *h);
26 static void hns3_remove_hw_addr(struct net_device *netdev);
27 
28 static const char hns3_driver_name[] = "hns3";
29 const char hns3_driver_version[] = VERMAGIC_STRING;
30 static const char hns3_driver_string[] =
31 			"Hisilicon Ethernet Network Driver for Hip08 Family";
32 static const char hns3_copyright[] = "Copyright (c) 2017 Huawei Corporation.";
33 static struct hnae3_client client;
34 
35 /* hns3_pci_tbl - PCI Device ID Table
36  *
37  * Last entry must be all 0s
38  *
39  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
40  *   Class, Class Mask, private data (not used) }
41  */
42 static const struct pci_device_id hns3_pci_tbl[] = {
43 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
44 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
45 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA),
46 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
47 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC),
48 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
49 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA),
50 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
51 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC),
52 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
53 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC),
54 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
55 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_VF), 0},
56 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF),
57 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
58 	/* required last entry */
59 	{0, }
60 };
61 MODULE_DEVICE_TABLE(pci, hns3_pci_tbl);
62 
63 static irqreturn_t hns3_irq_handle(int irq, void *vector)
64 {
65 	struct hns3_enet_tqp_vector *tqp_vector = vector;
66 
67 	napi_schedule(&tqp_vector->napi);
68 
69 	return IRQ_HANDLED;
70 }
71 
72 /* This callback function is used to set affinity changes to the irq affinity
73  * masks when the irq_set_affinity_notifier function is used.
74  */
75 static void hns3_nic_irq_affinity_notify(struct irq_affinity_notify *notify,
76 					 const cpumask_t *mask)
77 {
78 	struct hns3_enet_tqp_vector *tqp_vectors =
79 		container_of(notify, struct hns3_enet_tqp_vector,
80 			     affinity_notify);
81 
82 	tqp_vectors->affinity_mask = *mask;
83 }
84 
85 static void hns3_nic_irq_affinity_release(struct kref *ref)
86 {
87 }
88 
89 static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv)
90 {
91 	struct hns3_enet_tqp_vector *tqp_vectors;
92 	unsigned int i;
93 
94 	for (i = 0; i < priv->vector_num; i++) {
95 		tqp_vectors = &priv->tqp_vector[i];
96 
97 		if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED)
98 			continue;
99 
100 		/* clear the affinity notifier and affinity mask */
101 		irq_set_affinity_notifier(tqp_vectors->vector_irq, NULL);
102 		irq_set_affinity_hint(tqp_vectors->vector_irq, NULL);
103 
104 		/* release the irq resource */
105 		free_irq(tqp_vectors->vector_irq, tqp_vectors);
106 		tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED;
107 	}
108 }
109 
110 static int hns3_nic_init_irq(struct hns3_nic_priv *priv)
111 {
112 	struct hns3_enet_tqp_vector *tqp_vectors;
113 	int txrx_int_idx = 0;
114 	int rx_int_idx = 0;
115 	int tx_int_idx = 0;
116 	unsigned int i;
117 	int ret;
118 
119 	for (i = 0; i < priv->vector_num; i++) {
120 		tqp_vectors = &priv->tqp_vector[i];
121 
122 		if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED)
123 			continue;
124 
125 		if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) {
126 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
127 				 "%s-%s-%d", priv->netdev->name, "TxRx",
128 				 txrx_int_idx++);
129 			txrx_int_idx++;
130 		} else if (tqp_vectors->rx_group.ring) {
131 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
132 				 "%s-%s-%d", priv->netdev->name, "Rx",
133 				 rx_int_idx++);
134 		} else if (tqp_vectors->tx_group.ring) {
135 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
136 				 "%s-%s-%d", priv->netdev->name, "Tx",
137 				 tx_int_idx++);
138 		} else {
139 			/* Skip this unused q_vector */
140 			continue;
141 		}
142 
143 		tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0';
144 
145 		ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0,
146 				  tqp_vectors->name,
147 				       tqp_vectors);
148 		if (ret) {
149 			netdev_err(priv->netdev, "request irq(%d) fail\n",
150 				   tqp_vectors->vector_irq);
151 			return ret;
152 		}
153 
154 		tqp_vectors->affinity_notify.notify =
155 					hns3_nic_irq_affinity_notify;
156 		tqp_vectors->affinity_notify.release =
157 					hns3_nic_irq_affinity_release;
158 		irq_set_affinity_notifier(tqp_vectors->vector_irq,
159 					  &tqp_vectors->affinity_notify);
160 		irq_set_affinity_hint(tqp_vectors->vector_irq,
161 				      &tqp_vectors->affinity_mask);
162 
163 		tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED;
164 	}
165 
166 	return 0;
167 }
168 
169 static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector,
170 				 u32 mask_en)
171 {
172 	writel(mask_en, tqp_vector->mask_addr);
173 }
174 
175 static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector)
176 {
177 	napi_enable(&tqp_vector->napi);
178 
179 	/* enable vector */
180 	hns3_mask_vector_irq(tqp_vector, 1);
181 }
182 
183 static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector)
184 {
185 	/* disable vector */
186 	hns3_mask_vector_irq(tqp_vector, 0);
187 
188 	disable_irq(tqp_vector->vector_irq);
189 	napi_disable(&tqp_vector->napi);
190 }
191 
192 void hns3_set_vector_coalesce_rl(struct hns3_enet_tqp_vector *tqp_vector,
193 				 u32 rl_value)
194 {
195 	u32 rl_reg = hns3_rl_usec_to_reg(rl_value);
196 
197 	/* this defines the configuration for RL (Interrupt Rate Limiter).
198 	 * Rl defines rate of interrupts i.e. number of interrupts-per-second
199 	 * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing
200 	 */
201 
202 	if (rl_reg > 0 && !tqp_vector->tx_group.coal.gl_adapt_enable &&
203 	    !tqp_vector->rx_group.coal.gl_adapt_enable)
204 		/* According to the hardware, the range of rl_reg is
205 		 * 0-59 and the unit is 4.
206 		 */
207 		rl_reg |=  HNS3_INT_RL_ENABLE_MASK;
208 
209 	writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
210 }
211 
212 void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
213 				    u32 gl_value)
214 {
215 	u32 rx_gl_reg = hns3_gl_usec_to_reg(gl_value);
216 
217 	writel(rx_gl_reg, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
218 }
219 
220 void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
221 				    u32 gl_value)
222 {
223 	u32 tx_gl_reg = hns3_gl_usec_to_reg(gl_value);
224 
225 	writel(tx_gl_reg, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
226 }
227 
228 static void hns3_vector_gl_rl_init(struct hns3_enet_tqp_vector *tqp_vector,
229 				   struct hns3_nic_priv *priv)
230 {
231 	/* initialize the configuration for interrupt coalescing.
232 	 * 1. GL (Interrupt Gap Limiter)
233 	 * 2. RL (Interrupt Rate Limiter)
234 	 */
235 
236 	/* Default: enable interrupt coalescing self-adaptive and GL */
237 	tqp_vector->tx_group.coal.gl_adapt_enable = 1;
238 	tqp_vector->rx_group.coal.gl_adapt_enable = 1;
239 
240 	tqp_vector->tx_group.coal.int_gl = HNS3_INT_GL_50K;
241 	tqp_vector->rx_group.coal.int_gl = HNS3_INT_GL_50K;
242 
243 	tqp_vector->rx_group.coal.flow_level = HNS3_FLOW_LOW;
244 	tqp_vector->tx_group.coal.flow_level = HNS3_FLOW_LOW;
245 }
246 
247 static void hns3_vector_gl_rl_init_hw(struct hns3_enet_tqp_vector *tqp_vector,
248 				      struct hns3_nic_priv *priv)
249 {
250 	struct hnae3_handle *h = priv->ae_handle;
251 
252 	hns3_set_vector_coalesce_tx_gl(tqp_vector,
253 				       tqp_vector->tx_group.coal.int_gl);
254 	hns3_set_vector_coalesce_rx_gl(tqp_vector,
255 				       tqp_vector->rx_group.coal.int_gl);
256 	hns3_set_vector_coalesce_rl(tqp_vector, h->kinfo.int_rl_setting);
257 }
258 
259 static int hns3_nic_set_real_num_queue(struct net_device *netdev)
260 {
261 	struct hnae3_handle *h = hns3_get_handle(netdev);
262 	struct hnae3_knic_private_info *kinfo = &h->kinfo;
263 	unsigned int queue_size = kinfo->rss_size * kinfo->num_tc;
264 	int i, ret;
265 
266 	if (kinfo->num_tc <= 1) {
267 		netdev_reset_tc(netdev);
268 	} else {
269 		ret = netdev_set_num_tc(netdev, kinfo->num_tc);
270 		if (ret) {
271 			netdev_err(netdev,
272 				   "netdev_set_num_tc fail, ret=%d!\n", ret);
273 			return ret;
274 		}
275 
276 		for (i = 0; i < HNAE3_MAX_TC; i++) {
277 			if (!kinfo->tc_info[i].enable)
278 				continue;
279 
280 			netdev_set_tc_queue(netdev,
281 					    kinfo->tc_info[i].tc,
282 					    kinfo->tc_info[i].tqp_count,
283 					    kinfo->tc_info[i].tqp_offset);
284 		}
285 	}
286 
287 	ret = netif_set_real_num_tx_queues(netdev, queue_size);
288 	if (ret) {
289 		netdev_err(netdev,
290 			   "netif_set_real_num_tx_queues fail, ret=%d!\n",
291 			   ret);
292 		return ret;
293 	}
294 
295 	ret = netif_set_real_num_rx_queues(netdev, queue_size);
296 	if (ret) {
297 		netdev_err(netdev,
298 			   "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
299 		return ret;
300 	}
301 
302 	return 0;
303 }
304 
305 static u16 hns3_get_max_available_channels(struct hnae3_handle *h)
306 {
307 	u16 alloc_tqps, max_rss_size, rss_size;
308 
309 	h->ae_algo->ops->get_tqps_and_rss_info(h, &alloc_tqps, &max_rss_size);
310 	rss_size = alloc_tqps / h->kinfo.num_tc;
311 
312 	return min_t(u16, rss_size, max_rss_size);
313 }
314 
315 static void hns3_tqp_enable(struct hnae3_queue *tqp)
316 {
317 	u32 rcb_reg;
318 
319 	rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
320 	rcb_reg |= BIT(HNS3_RING_EN_B);
321 	hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
322 }
323 
324 static void hns3_tqp_disable(struct hnae3_queue *tqp)
325 {
326 	u32 rcb_reg;
327 
328 	rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
329 	rcb_reg &= ~BIT(HNS3_RING_EN_B);
330 	hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
331 }
332 
333 static int hns3_nic_net_up(struct net_device *netdev)
334 {
335 	struct hns3_nic_priv *priv = netdev_priv(netdev);
336 	struct hnae3_handle *h = priv->ae_handle;
337 	int i, j;
338 	int ret;
339 
340 	ret = hns3_nic_reset_all_ring(h);
341 	if (ret)
342 		return ret;
343 
344 	/* get irq resource for all vectors */
345 	ret = hns3_nic_init_irq(priv);
346 	if (ret) {
347 		netdev_err(netdev, "hns init irq failed! ret=%d\n", ret);
348 		return ret;
349 	}
350 
351 	clear_bit(HNS3_NIC_STATE_DOWN, &priv->state);
352 
353 	/* enable the vectors */
354 	for (i = 0; i < priv->vector_num; i++)
355 		hns3_vector_enable(&priv->tqp_vector[i]);
356 
357 	/* enable rcb */
358 	for (j = 0; j < h->kinfo.num_tqps; j++)
359 		hns3_tqp_enable(h->kinfo.tqp[j]);
360 
361 	/* start the ae_dev */
362 	ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0;
363 	if (ret)
364 		goto out_start_err;
365 
366 	return 0;
367 
368 out_start_err:
369 	set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
370 	while (j--)
371 		hns3_tqp_disable(h->kinfo.tqp[j]);
372 
373 	for (j = i - 1; j >= 0; j--)
374 		hns3_vector_disable(&priv->tqp_vector[j]);
375 
376 	hns3_nic_uninit_irq(priv);
377 
378 	return ret;
379 }
380 
381 static int hns3_nic_net_open(struct net_device *netdev)
382 {
383 	struct hns3_nic_priv *priv = netdev_priv(netdev);
384 	struct hnae3_handle *h = hns3_get_handle(netdev);
385 	struct hnae3_knic_private_info *kinfo;
386 	int i, ret;
387 
388 	if (hns3_nic_resetting(netdev))
389 		return -EBUSY;
390 
391 	netif_carrier_off(netdev);
392 
393 	ret = hns3_nic_set_real_num_queue(netdev);
394 	if (ret)
395 		return ret;
396 
397 	ret = hns3_nic_net_up(netdev);
398 	if (ret) {
399 		netdev_err(netdev,
400 			   "hns net up fail, ret=%d!\n", ret);
401 		return ret;
402 	}
403 
404 	kinfo = &h->kinfo;
405 	for (i = 0; i < HNAE3_MAX_USER_PRIO; i++) {
406 		netdev_set_prio_tc_map(netdev, i,
407 				       kinfo->prio_tc[i]);
408 	}
409 
410 	if (h->ae_algo->ops->set_timer_task)
411 		h->ae_algo->ops->set_timer_task(priv->ae_handle, true);
412 
413 	return 0;
414 }
415 
416 static void hns3_nic_net_down(struct net_device *netdev)
417 {
418 	struct hns3_nic_priv *priv = netdev_priv(netdev);
419 	struct hnae3_handle *h = hns3_get_handle(netdev);
420 	const struct hnae3_ae_ops *ops;
421 	int i;
422 
423 	/* disable vectors */
424 	for (i = 0; i < priv->vector_num; i++)
425 		hns3_vector_disable(&priv->tqp_vector[i]);
426 
427 	/* disable rcb */
428 	for (i = 0; i < h->kinfo.num_tqps; i++)
429 		hns3_tqp_disable(h->kinfo.tqp[i]);
430 
431 	/* stop ae_dev */
432 	ops = priv->ae_handle->ae_algo->ops;
433 	if (ops->stop)
434 		ops->stop(priv->ae_handle);
435 
436 	/* free irq resources */
437 	hns3_nic_uninit_irq(priv);
438 
439 	hns3_clear_all_ring(priv->ae_handle);
440 }
441 
442 static int hns3_nic_net_stop(struct net_device *netdev)
443 {
444 	struct hns3_nic_priv *priv = netdev_priv(netdev);
445 	struct hnae3_handle *h = hns3_get_handle(netdev);
446 
447 	if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
448 		return 0;
449 
450 	if (h->ae_algo->ops->set_timer_task)
451 		h->ae_algo->ops->set_timer_task(priv->ae_handle, false);
452 
453 	netif_tx_stop_all_queues(netdev);
454 	netif_carrier_off(netdev);
455 
456 	hns3_nic_net_down(netdev);
457 
458 	return 0;
459 }
460 
461 static int hns3_nic_uc_sync(struct net_device *netdev,
462 			    const unsigned char *addr)
463 {
464 	struct hnae3_handle *h = hns3_get_handle(netdev);
465 
466 	if (h->ae_algo->ops->add_uc_addr)
467 		return h->ae_algo->ops->add_uc_addr(h, addr);
468 
469 	return 0;
470 }
471 
472 static int hns3_nic_uc_unsync(struct net_device *netdev,
473 			      const unsigned char *addr)
474 {
475 	struct hnae3_handle *h = hns3_get_handle(netdev);
476 
477 	if (h->ae_algo->ops->rm_uc_addr)
478 		return h->ae_algo->ops->rm_uc_addr(h, addr);
479 
480 	return 0;
481 }
482 
483 static int hns3_nic_mc_sync(struct net_device *netdev,
484 			    const unsigned char *addr)
485 {
486 	struct hnae3_handle *h = hns3_get_handle(netdev);
487 
488 	if (h->ae_algo->ops->add_mc_addr)
489 		return h->ae_algo->ops->add_mc_addr(h, addr);
490 
491 	return 0;
492 }
493 
494 static int hns3_nic_mc_unsync(struct net_device *netdev,
495 			      const unsigned char *addr)
496 {
497 	struct hnae3_handle *h = hns3_get_handle(netdev);
498 
499 	if (h->ae_algo->ops->rm_mc_addr)
500 		return h->ae_algo->ops->rm_mc_addr(h, addr);
501 
502 	return 0;
503 }
504 
505 static u8 hns3_get_netdev_flags(struct net_device *netdev)
506 {
507 	u8 flags = 0;
508 
509 	if (netdev->flags & IFF_PROMISC) {
510 		flags = HNAE3_USER_UPE | HNAE3_USER_MPE | HNAE3_BPE;
511 	} else {
512 		flags |= HNAE3_VLAN_FLTR;
513 		if (netdev->flags & IFF_ALLMULTI)
514 			flags |= HNAE3_USER_MPE;
515 	}
516 
517 	return flags;
518 }
519 
520 static void hns3_nic_set_rx_mode(struct net_device *netdev)
521 {
522 	struct hnae3_handle *h = hns3_get_handle(netdev);
523 	u8 new_flags;
524 	int ret;
525 
526 	new_flags = hns3_get_netdev_flags(netdev);
527 
528 	ret = __dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync);
529 	if (ret) {
530 		netdev_err(netdev, "sync uc address fail\n");
531 		if (ret == -ENOSPC)
532 			new_flags |= HNAE3_OVERFLOW_UPE;
533 	}
534 
535 	if (netdev->flags & IFF_MULTICAST) {
536 		ret = __dev_mc_sync(netdev, hns3_nic_mc_sync,
537 				    hns3_nic_mc_unsync);
538 		if (ret) {
539 			netdev_err(netdev, "sync mc address fail\n");
540 			if (ret == -ENOSPC)
541 				new_flags |= HNAE3_OVERFLOW_MPE;
542 		}
543 	}
544 
545 	/* User mode Promisc mode enable and vlan filtering is disabled to
546 	 * let all packets in. MAC-VLAN Table overflow Promisc enabled and
547 	 * vlan fitering is enabled
548 	 */
549 	hns3_enable_vlan_filter(netdev, new_flags & HNAE3_VLAN_FLTR);
550 	h->netdev_flags = new_flags;
551 	hns3_update_promisc_mode(netdev, new_flags);
552 }
553 
554 int hns3_update_promisc_mode(struct net_device *netdev, u8 promisc_flags)
555 {
556 	struct hns3_nic_priv *priv = netdev_priv(netdev);
557 	struct hnae3_handle *h = priv->ae_handle;
558 
559 	if (h->ae_algo->ops->set_promisc_mode) {
560 		return h->ae_algo->ops->set_promisc_mode(h,
561 						promisc_flags & HNAE3_UPE,
562 						promisc_flags & HNAE3_MPE);
563 	}
564 
565 	return 0;
566 }
567 
568 void hns3_enable_vlan_filter(struct net_device *netdev, bool enable)
569 {
570 	struct hns3_nic_priv *priv = netdev_priv(netdev);
571 	struct hnae3_handle *h = priv->ae_handle;
572 	bool last_state;
573 
574 	if (h->pdev->revision >= 0x21 && h->ae_algo->ops->enable_vlan_filter) {
575 		last_state = h->netdev_flags & HNAE3_VLAN_FLTR ? true : false;
576 		if (enable != last_state) {
577 			netdev_info(netdev,
578 				    "%s vlan filter\n",
579 				    enable ? "enable" : "disable");
580 			h->ae_algo->ops->enable_vlan_filter(h, enable);
581 		}
582 	}
583 }
584 
585 static int hns3_set_tso(struct sk_buff *skb, u32 *paylen,
586 			u16 *mss, u32 *type_cs_vlan_tso)
587 {
588 	u32 l4_offset, hdr_len;
589 	union l3_hdr_info l3;
590 	union l4_hdr_info l4;
591 	u32 l4_paylen;
592 	int ret;
593 
594 	if (!skb_is_gso(skb))
595 		return 0;
596 
597 	ret = skb_cow_head(skb, 0);
598 	if (ret)
599 		return ret;
600 
601 	l3.hdr = skb_network_header(skb);
602 	l4.hdr = skb_transport_header(skb);
603 
604 	/* Software should clear the IPv4's checksum field when tso is
605 	 * needed.
606 	 */
607 	if (l3.v4->version == 4)
608 		l3.v4->check = 0;
609 
610 	/* tunnel packet.*/
611 	if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
612 					 SKB_GSO_GRE_CSUM |
613 					 SKB_GSO_UDP_TUNNEL |
614 					 SKB_GSO_UDP_TUNNEL_CSUM)) {
615 		if ((!(skb_shinfo(skb)->gso_type &
616 		    SKB_GSO_PARTIAL)) &&
617 		    (skb_shinfo(skb)->gso_type &
618 		    SKB_GSO_UDP_TUNNEL_CSUM)) {
619 			/* Software should clear the udp's checksum
620 			 * field when tso is needed.
621 			 */
622 			l4.udp->check = 0;
623 		}
624 		/* reset l3&l4 pointers from outer to inner headers */
625 		l3.hdr = skb_inner_network_header(skb);
626 		l4.hdr = skb_inner_transport_header(skb);
627 
628 		/* Software should clear the IPv4's checksum field when
629 		 * tso is needed.
630 		 */
631 		if (l3.v4->version == 4)
632 			l3.v4->check = 0;
633 	}
634 
635 	/* normal or tunnel packet*/
636 	l4_offset = l4.hdr - skb->data;
637 	hdr_len = (l4.tcp->doff * 4) + l4_offset;
638 
639 	/* remove payload length from inner pseudo checksum when tso*/
640 	l4_paylen = skb->len - l4_offset;
641 	csum_replace_by_diff(&l4.tcp->check,
642 			     (__force __wsum)htonl(l4_paylen));
643 
644 	/* find the txbd field values */
645 	*paylen = skb->len - hdr_len;
646 	hnae3_set_bit(*type_cs_vlan_tso,
647 		      HNS3_TXD_TSO_B, 1);
648 
649 	/* get MSS for TSO */
650 	*mss = skb_shinfo(skb)->gso_size;
651 
652 	return 0;
653 }
654 
655 static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
656 				u8 *il4_proto)
657 {
658 	union {
659 		struct iphdr *v4;
660 		struct ipv6hdr *v6;
661 		unsigned char *hdr;
662 	} l3;
663 	unsigned char *l4_hdr;
664 	unsigned char *exthdr;
665 	u8 l4_proto_tmp;
666 	__be16 frag_off;
667 
668 	/* find outer header point */
669 	l3.hdr = skb_network_header(skb);
670 	l4_hdr = skb_transport_header(skb);
671 
672 	if (skb->protocol == htons(ETH_P_IPV6)) {
673 		exthdr = l3.hdr + sizeof(*l3.v6);
674 		l4_proto_tmp = l3.v6->nexthdr;
675 		if (l4_hdr != exthdr)
676 			ipv6_skip_exthdr(skb, exthdr - skb->data,
677 					 &l4_proto_tmp, &frag_off);
678 	} else if (skb->protocol == htons(ETH_P_IP)) {
679 		l4_proto_tmp = l3.v4->protocol;
680 	} else {
681 		return -EINVAL;
682 	}
683 
684 	*ol4_proto = l4_proto_tmp;
685 
686 	/* tunnel packet */
687 	if (!skb->encapsulation) {
688 		*il4_proto = 0;
689 		return 0;
690 	}
691 
692 	/* find inner header point */
693 	l3.hdr = skb_inner_network_header(skb);
694 	l4_hdr = skb_inner_transport_header(skb);
695 
696 	if (l3.v6->version == 6) {
697 		exthdr = l3.hdr + sizeof(*l3.v6);
698 		l4_proto_tmp = l3.v6->nexthdr;
699 		if (l4_hdr != exthdr)
700 			ipv6_skip_exthdr(skb, exthdr - skb->data,
701 					 &l4_proto_tmp, &frag_off);
702 	} else if (l3.v4->version == 4) {
703 		l4_proto_tmp = l3.v4->protocol;
704 	}
705 
706 	*il4_proto = l4_proto_tmp;
707 
708 	return 0;
709 }
710 
711 static void hns3_set_l2l3l4_len(struct sk_buff *skb, u8 ol4_proto,
712 				u8 il4_proto, u32 *type_cs_vlan_tso,
713 				u32 *ol_type_vlan_len_msec)
714 {
715 	union {
716 		struct iphdr *v4;
717 		struct ipv6hdr *v6;
718 		unsigned char *hdr;
719 	} l3;
720 	union {
721 		struct tcphdr *tcp;
722 		struct udphdr *udp;
723 		struct gre_base_hdr *gre;
724 		unsigned char *hdr;
725 	} l4;
726 	unsigned char *l2_hdr;
727 	u8 l4_proto = ol4_proto;
728 	u32 ol2_len;
729 	u32 ol3_len;
730 	u32 ol4_len;
731 	u32 l2_len;
732 	u32 l3_len;
733 
734 	l3.hdr = skb_network_header(skb);
735 	l4.hdr = skb_transport_header(skb);
736 
737 	/* compute L2 header size for normal packet, defined in 2 Bytes */
738 	l2_len = l3.hdr - skb->data;
739 	hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_M,
740 			HNS3_TXD_L2LEN_S, l2_len >> 1);
741 
742 	/* tunnel packet*/
743 	if (skb->encapsulation) {
744 		/* compute OL2 header size, defined in 2 Bytes */
745 		ol2_len = l2_len;
746 		hnae3_set_field(*ol_type_vlan_len_msec,
747 				HNS3_TXD_L2LEN_M,
748 				HNS3_TXD_L2LEN_S, ol2_len >> 1);
749 
750 		/* compute OL3 header size, defined in 4 Bytes */
751 		ol3_len = l4.hdr - l3.hdr;
752 		hnae3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_M,
753 				HNS3_TXD_L3LEN_S, ol3_len >> 2);
754 
755 		/* MAC in UDP, MAC in GRE (0x6558)*/
756 		if ((ol4_proto == IPPROTO_UDP) || (ol4_proto == IPPROTO_GRE)) {
757 			/* switch MAC header ptr from outer to inner header.*/
758 			l2_hdr = skb_inner_mac_header(skb);
759 
760 			/* compute OL4 header size, defined in 4 Bytes. */
761 			ol4_len = l2_hdr - l4.hdr;
762 			hnae3_set_field(*ol_type_vlan_len_msec,
763 					HNS3_TXD_L4LEN_M, HNS3_TXD_L4LEN_S,
764 					ol4_len >> 2);
765 
766 			/* switch IP header ptr from outer to inner header */
767 			l3.hdr = skb_inner_network_header(skb);
768 
769 			/* compute inner l2 header size, defined in 2 Bytes. */
770 			l2_len = l3.hdr - l2_hdr;
771 			hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_M,
772 					HNS3_TXD_L2LEN_S, l2_len >> 1);
773 		} else {
774 			/* skb packet types not supported by hardware,
775 			 * txbd len fild doesn't be filled.
776 			 */
777 			return;
778 		}
779 
780 		/* switch L4 header pointer from outer to inner */
781 		l4.hdr = skb_inner_transport_header(skb);
782 
783 		l4_proto = il4_proto;
784 	}
785 
786 	/* compute inner(/normal) L3 header size, defined in 4 Bytes */
787 	l3_len = l4.hdr - l3.hdr;
788 	hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_M,
789 			HNS3_TXD_L3LEN_S, l3_len >> 2);
790 
791 	/* compute inner(/normal) L4 header size, defined in 4 Bytes */
792 	switch (l4_proto) {
793 	case IPPROTO_TCP:
794 		hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_M,
795 				HNS3_TXD_L4LEN_S, l4.tcp->doff);
796 		break;
797 	case IPPROTO_SCTP:
798 		hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_M,
799 				HNS3_TXD_L4LEN_S,
800 				(sizeof(struct sctphdr) >> 2));
801 		break;
802 	case IPPROTO_UDP:
803 		hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_M,
804 				HNS3_TXD_L4LEN_S,
805 				(sizeof(struct udphdr) >> 2));
806 		break;
807 	default:
808 		/* skb packet types not supported by hardware,
809 		 * txbd len fild doesn't be filled.
810 		 */
811 		return;
812 	}
813 }
814 
815 /* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
816  * and it is udp packet, which has a dest port as the IANA assigned.
817  * the hardware is expected to do the checksum offload, but the
818  * hardware will not do the checksum offload when udp dest port is
819  * 4789.
820  */
821 static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
822 {
823 #define IANA_VXLAN_PORT	4789
824 	union {
825 		struct tcphdr *tcp;
826 		struct udphdr *udp;
827 		struct gre_base_hdr *gre;
828 		unsigned char *hdr;
829 	} l4;
830 
831 	l4.hdr = skb_transport_header(skb);
832 
833 	if (!(!skb->encapsulation && l4.udp->dest == htons(IANA_VXLAN_PORT)))
834 		return false;
835 
836 	skb_checksum_help(skb);
837 
838 	return true;
839 }
840 
841 static int hns3_set_l3l4_type_csum(struct sk_buff *skb, u8 ol4_proto,
842 				   u8 il4_proto, u32 *type_cs_vlan_tso,
843 				   u32 *ol_type_vlan_len_msec)
844 {
845 	union {
846 		struct iphdr *v4;
847 		struct ipv6hdr *v6;
848 		unsigned char *hdr;
849 	} l3;
850 	u32 l4_proto = ol4_proto;
851 
852 	l3.hdr = skb_network_header(skb);
853 
854 	/* define OL3 type and tunnel type(OL4).*/
855 	if (skb->encapsulation) {
856 		/* define outer network header type.*/
857 		if (skb->protocol == htons(ETH_P_IP)) {
858 			if (skb_is_gso(skb))
859 				hnae3_set_field(*ol_type_vlan_len_msec,
860 						HNS3_TXD_OL3T_M,
861 						HNS3_TXD_OL3T_S,
862 						HNS3_OL3T_IPV4_CSUM);
863 			else
864 				hnae3_set_field(*ol_type_vlan_len_msec,
865 						HNS3_TXD_OL3T_M,
866 						HNS3_TXD_OL3T_S,
867 						HNS3_OL3T_IPV4_NO_CSUM);
868 
869 		} else if (skb->protocol == htons(ETH_P_IPV6)) {
870 			hnae3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_M,
871 					HNS3_TXD_OL3T_S, HNS3_OL3T_IPV6);
872 		}
873 
874 		/* define tunnel type(OL4).*/
875 		switch (l4_proto) {
876 		case IPPROTO_UDP:
877 			hnae3_set_field(*ol_type_vlan_len_msec,
878 					HNS3_TXD_TUNTYPE_M,
879 					HNS3_TXD_TUNTYPE_S,
880 					HNS3_TUN_MAC_IN_UDP);
881 			break;
882 		case IPPROTO_GRE:
883 			hnae3_set_field(*ol_type_vlan_len_msec,
884 					HNS3_TXD_TUNTYPE_M,
885 					HNS3_TXD_TUNTYPE_S,
886 					HNS3_TUN_NVGRE);
887 			break;
888 		default:
889 			/* drop the skb tunnel packet if hardware don't support,
890 			 * because hardware can't calculate csum when TSO.
891 			 */
892 			if (skb_is_gso(skb))
893 				return -EDOM;
894 
895 			/* the stack computes the IP header already,
896 			 * driver calculate l4 checksum when not TSO.
897 			 */
898 			skb_checksum_help(skb);
899 			return 0;
900 		}
901 
902 		l3.hdr = skb_inner_network_header(skb);
903 		l4_proto = il4_proto;
904 	}
905 
906 	if (l3.v4->version == 4) {
907 		hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_M,
908 				HNS3_TXD_L3T_S, HNS3_L3T_IPV4);
909 
910 		/* the stack computes the IP header already, the only time we
911 		 * need the hardware to recompute it is in the case of TSO.
912 		 */
913 		if (skb_is_gso(skb))
914 			hnae3_set_bit(*type_cs_vlan_tso, HNS3_TXD_L3CS_B, 1);
915 	} else if (l3.v6->version == 6) {
916 		hnae3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_M,
917 				HNS3_TXD_L3T_S, HNS3_L3T_IPV6);
918 	}
919 
920 	switch (l4_proto) {
921 	case IPPROTO_TCP:
922 		hnae3_set_bit(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
923 		hnae3_set_field(*type_cs_vlan_tso,
924 				HNS3_TXD_L4T_M,
925 				HNS3_TXD_L4T_S,
926 				HNS3_L4T_TCP);
927 		break;
928 	case IPPROTO_UDP:
929 		if (hns3_tunnel_csum_bug(skb))
930 			break;
931 
932 		hnae3_set_bit(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
933 		hnae3_set_field(*type_cs_vlan_tso,
934 				HNS3_TXD_L4T_M,
935 				HNS3_TXD_L4T_S,
936 				HNS3_L4T_UDP);
937 		break;
938 	case IPPROTO_SCTP:
939 		hnae3_set_bit(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
940 		hnae3_set_field(*type_cs_vlan_tso,
941 				HNS3_TXD_L4T_M,
942 				HNS3_TXD_L4T_S,
943 				HNS3_L4T_SCTP);
944 		break;
945 	default:
946 		/* drop the skb tunnel packet if hardware don't support,
947 		 * because hardware can't calculate csum when TSO.
948 		 */
949 		if (skb_is_gso(skb))
950 			return -EDOM;
951 
952 		/* the stack computes the IP header already,
953 		 * driver calculate l4 checksum when not TSO.
954 		 */
955 		skb_checksum_help(skb);
956 		return 0;
957 	}
958 
959 	return 0;
960 }
961 
962 static void hns3_set_txbd_baseinfo(u16 *bdtp_fe_sc_vld_ra_ri, int frag_end)
963 {
964 	/* Config bd buffer end */
965 	hnae3_set_field(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_BDTYPE_M,
966 			HNS3_TXD_BDTYPE_S, 0);
967 	hnae3_set_bit(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_FE_B, !!frag_end);
968 	hnae3_set_bit(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_VLD_B, 1);
969 	hnae3_set_field(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_SC_M, HNS3_TXD_SC_S, 0);
970 }
971 
972 static int hns3_fill_desc_vtags(struct sk_buff *skb,
973 				struct hns3_enet_ring *tx_ring,
974 				u32 *inner_vlan_flag,
975 				u32 *out_vlan_flag,
976 				u16 *inner_vtag,
977 				u16 *out_vtag)
978 {
979 #define HNS3_TX_VLAN_PRIO_SHIFT 13
980 
981 	if (skb->protocol == htons(ETH_P_8021Q) &&
982 	    !(tx_ring->tqp->handle->kinfo.netdev->features &
983 	    NETIF_F_HW_VLAN_CTAG_TX)) {
984 		/* When HW VLAN acceleration is turned off, and the stack
985 		 * sets the protocol to 802.1q, the driver just need to
986 		 * set the protocol to the encapsulated ethertype.
987 		 */
988 		skb->protocol = vlan_get_protocol(skb);
989 		return 0;
990 	}
991 
992 	if (skb_vlan_tag_present(skb)) {
993 		u16 vlan_tag;
994 
995 		vlan_tag = skb_vlan_tag_get(skb);
996 		vlan_tag |= (skb->priority & 0x7) << HNS3_TX_VLAN_PRIO_SHIFT;
997 
998 		/* Based on hw strategy, use out_vtag in two layer tag case,
999 		 * and use inner_vtag in one tag case.
1000 		 */
1001 		if (skb->protocol == htons(ETH_P_8021Q)) {
1002 			hnae3_set_bit(*out_vlan_flag, HNS3_TXD_OVLAN_B, 1);
1003 			*out_vtag = vlan_tag;
1004 		} else {
1005 			hnae3_set_bit(*inner_vlan_flag, HNS3_TXD_VLAN_B, 1);
1006 			*inner_vtag = vlan_tag;
1007 		}
1008 	} else if (skb->protocol == htons(ETH_P_8021Q)) {
1009 		struct vlan_ethhdr *vhdr;
1010 		int rc;
1011 
1012 		rc = skb_cow_head(skb, 0);
1013 		if (rc < 0)
1014 			return rc;
1015 		vhdr = (struct vlan_ethhdr *)skb->data;
1016 		vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority & 0x7)
1017 					<< HNS3_TX_VLAN_PRIO_SHIFT);
1018 	}
1019 
1020 	skb->protocol = vlan_get_protocol(skb);
1021 	return 0;
1022 }
1023 
1024 static int hns3_fill_desc(struct hns3_enet_ring *ring, void *priv,
1025 			  int size, int frag_end, enum hns_desc_type type)
1026 {
1027 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1028 	struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1029 	struct device *dev = ring_to_dev(ring);
1030 	u32 ol_type_vlan_len_msec = 0;
1031 	u16 bdtp_fe_sc_vld_ra_ri = 0;
1032 	struct skb_frag_struct *frag;
1033 	unsigned int frag_buf_num;
1034 	u32 type_cs_vlan_tso = 0;
1035 	struct sk_buff *skb;
1036 	u16 inner_vtag = 0;
1037 	u16 out_vtag = 0;
1038 	unsigned int k;
1039 	int sizeoflast;
1040 	u32 paylen = 0;
1041 	dma_addr_t dma;
1042 	u16 mss = 0;
1043 	u8 ol4_proto;
1044 	u8 il4_proto;
1045 	int ret;
1046 
1047 	if (type == DESC_TYPE_SKB) {
1048 		skb = (struct sk_buff *)priv;
1049 		paylen = skb->len;
1050 
1051 		ret = hns3_fill_desc_vtags(skb, ring, &type_cs_vlan_tso,
1052 					   &ol_type_vlan_len_msec,
1053 					   &inner_vtag, &out_vtag);
1054 		if (unlikely(ret))
1055 			return ret;
1056 
1057 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1058 			skb_reset_mac_len(skb);
1059 
1060 			ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
1061 			if (ret)
1062 				return ret;
1063 			hns3_set_l2l3l4_len(skb, ol4_proto, il4_proto,
1064 					    &type_cs_vlan_tso,
1065 					    &ol_type_vlan_len_msec);
1066 			ret = hns3_set_l3l4_type_csum(skb, ol4_proto, il4_proto,
1067 						      &type_cs_vlan_tso,
1068 						      &ol_type_vlan_len_msec);
1069 			if (ret)
1070 				return ret;
1071 
1072 			ret = hns3_set_tso(skb, &paylen, &mss,
1073 					   &type_cs_vlan_tso);
1074 			if (ret)
1075 				return ret;
1076 		}
1077 
1078 		/* Set txbd */
1079 		desc->tx.ol_type_vlan_len_msec =
1080 			cpu_to_le32(ol_type_vlan_len_msec);
1081 		desc->tx.type_cs_vlan_tso_len =
1082 			cpu_to_le32(type_cs_vlan_tso);
1083 		desc->tx.paylen = cpu_to_le32(paylen);
1084 		desc->tx.mss = cpu_to_le16(mss);
1085 		desc->tx.vlan_tag = cpu_to_le16(inner_vtag);
1086 		desc->tx.outer_vlan_tag = cpu_to_le16(out_vtag);
1087 
1088 		dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1089 	} else {
1090 		frag = (struct skb_frag_struct *)priv;
1091 		dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1092 	}
1093 
1094 	if (dma_mapping_error(ring->dev, dma)) {
1095 		ring->stats.sw_err_cnt++;
1096 		return -ENOMEM;
1097 	}
1098 
1099 	desc_cb->length = size;
1100 
1101 	frag_buf_num = (size + HNS3_MAX_BD_SIZE - 1) / HNS3_MAX_BD_SIZE;
1102 	sizeoflast = size % HNS3_MAX_BD_SIZE;
1103 	sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1104 
1105 	/* When frag size is bigger than hardware limit, split this frag */
1106 	for (k = 0; k < frag_buf_num; k++) {
1107 		/* The txbd's baseinfo of DESC_TYPE_PAGE & DESC_TYPE_SKB */
1108 		desc_cb->priv = priv;
1109 		desc_cb->dma = dma + HNS3_MAX_BD_SIZE * k;
1110 		desc_cb->type = (type == DESC_TYPE_SKB && !k) ?
1111 					DESC_TYPE_SKB : DESC_TYPE_PAGE;
1112 
1113 		/* now, fill the descriptor */
1114 		desc->addr = cpu_to_le64(dma + HNS3_MAX_BD_SIZE * k);
1115 		desc->tx.send_size = cpu_to_le16((k == frag_buf_num - 1) ?
1116 				(u16)sizeoflast : (u16)HNS3_MAX_BD_SIZE);
1117 		hns3_set_txbd_baseinfo(&bdtp_fe_sc_vld_ra_ri,
1118 				       frag_end && (k == frag_buf_num - 1) ?
1119 						1 : 0);
1120 		desc->tx.bdtp_fe_sc_vld_ra_ri =
1121 				cpu_to_le16(bdtp_fe_sc_vld_ra_ri);
1122 
1123 		/* move ring pointer to next.*/
1124 		ring_ptr_move_fw(ring, next_to_use);
1125 
1126 		desc_cb = &ring->desc_cb[ring->next_to_use];
1127 		desc = &ring->desc[ring->next_to_use];
1128 	}
1129 
1130 	return 0;
1131 }
1132 
1133 static int hns3_nic_maybe_stop_tso(struct sk_buff **out_skb, int *bnum,
1134 				   struct hns3_enet_ring *ring)
1135 {
1136 	struct sk_buff *skb = *out_skb;
1137 	struct sk_buff *new_skb = NULL;
1138 	struct skb_frag_struct *frag;
1139 	int bdnum_for_frag;
1140 	int frag_num;
1141 	int buf_num;
1142 	int size;
1143 	int i;
1144 
1145 	size = skb_headlen(skb);
1146 	buf_num = (size + HNS3_MAX_BD_SIZE - 1) / HNS3_MAX_BD_SIZE;
1147 
1148 	frag_num = skb_shinfo(skb)->nr_frags;
1149 	for (i = 0; i < frag_num; i++) {
1150 		frag = &skb_shinfo(skb)->frags[i];
1151 		size = skb_frag_size(frag);
1152 		bdnum_for_frag =
1153 			(size + HNS3_MAX_BD_SIZE - 1) / HNS3_MAX_BD_SIZE;
1154 		if (bdnum_for_frag > HNS3_MAX_BD_PER_FRAG)
1155 			return -ENOMEM;
1156 
1157 		buf_num += bdnum_for_frag;
1158 	}
1159 
1160 	if (unlikely(buf_num > HNS3_MAX_BD_PER_FRAG)) {
1161 		buf_num = (skb->len + HNS3_MAX_BD_SIZE - 1) / HNS3_MAX_BD_SIZE;
1162 		if (ring_space(ring) < buf_num)
1163 			return -EBUSY;
1164 		/* manual split the send packet */
1165 		new_skb = skb_copy(skb, GFP_ATOMIC);
1166 		if (!new_skb)
1167 			return -ENOMEM;
1168 		dev_kfree_skb_any(skb);
1169 		*out_skb = new_skb;
1170 	}
1171 
1172 	if (unlikely(ring_space(ring) < buf_num))
1173 		return -EBUSY;
1174 
1175 	*bnum = buf_num;
1176 	return 0;
1177 }
1178 
1179 static int hns3_nic_maybe_stop_tx(struct sk_buff **out_skb, int *bnum,
1180 				  struct hns3_enet_ring *ring)
1181 {
1182 	struct sk_buff *skb = *out_skb;
1183 	struct sk_buff *new_skb = NULL;
1184 	int buf_num;
1185 
1186 	/* No. of segments (plus a header) */
1187 	buf_num = skb_shinfo(skb)->nr_frags + 1;
1188 
1189 	if (unlikely(buf_num > HNS3_MAX_BD_PER_FRAG)) {
1190 		buf_num = (skb->len + HNS3_MAX_BD_SIZE - 1) / HNS3_MAX_BD_SIZE;
1191 		if (ring_space(ring) < buf_num)
1192 			return -EBUSY;
1193 		/* manual split the send packet */
1194 		new_skb = skb_copy(skb, GFP_ATOMIC);
1195 		if (!new_skb)
1196 			return -ENOMEM;
1197 		dev_kfree_skb_any(skb);
1198 		*out_skb = new_skb;
1199 	}
1200 
1201 	if (unlikely(ring_space(ring) < buf_num))
1202 		return -EBUSY;
1203 
1204 	*bnum = buf_num;
1205 
1206 	return 0;
1207 }
1208 
1209 static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
1210 {
1211 	struct device *dev = ring_to_dev(ring);
1212 	unsigned int i;
1213 
1214 	for (i = 0; i < ring->desc_num; i++) {
1215 		/* check if this is where we started */
1216 		if (ring->next_to_use == next_to_use_orig)
1217 			break;
1218 
1219 		/* unmap the descriptor dma address */
1220 		if (ring->desc_cb[ring->next_to_use].type == DESC_TYPE_SKB)
1221 			dma_unmap_single(dev,
1222 					 ring->desc_cb[ring->next_to_use].dma,
1223 					ring->desc_cb[ring->next_to_use].length,
1224 					DMA_TO_DEVICE);
1225 		else if (ring->desc_cb[ring->next_to_use].length)
1226 			dma_unmap_page(dev,
1227 				       ring->desc_cb[ring->next_to_use].dma,
1228 				       ring->desc_cb[ring->next_to_use].length,
1229 				       DMA_TO_DEVICE);
1230 
1231 		ring->desc_cb[ring->next_to_use].length = 0;
1232 
1233 		/* rollback one */
1234 		ring_ptr_move_bw(ring, next_to_use);
1235 	}
1236 }
1237 
1238 netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
1239 {
1240 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1241 	struct hns3_nic_ring_data *ring_data =
1242 		&tx_ring_data(priv, skb->queue_mapping);
1243 	struct hns3_enet_ring *ring = ring_data->ring;
1244 	struct netdev_queue *dev_queue;
1245 	struct skb_frag_struct *frag;
1246 	int next_to_use_head;
1247 	int next_to_use_frag;
1248 	int buf_num;
1249 	int seg_num;
1250 	int size;
1251 	int ret;
1252 	int i;
1253 
1254 	/* Prefetch the data used later */
1255 	prefetch(skb->data);
1256 
1257 	switch (priv->ops.maybe_stop_tx(&skb, &buf_num, ring)) {
1258 	case -EBUSY:
1259 		u64_stats_update_begin(&ring->syncp);
1260 		ring->stats.tx_busy++;
1261 		u64_stats_update_end(&ring->syncp);
1262 
1263 		goto out_net_tx_busy;
1264 	case -ENOMEM:
1265 		u64_stats_update_begin(&ring->syncp);
1266 		ring->stats.sw_err_cnt++;
1267 		u64_stats_update_end(&ring->syncp);
1268 		netdev_err(netdev, "no memory to xmit!\n");
1269 
1270 		goto out_err_tx_ok;
1271 	default:
1272 		break;
1273 	}
1274 
1275 	/* No. of segments (plus a header) */
1276 	seg_num = skb_shinfo(skb)->nr_frags + 1;
1277 	/* Fill the first part */
1278 	size = skb_headlen(skb);
1279 
1280 	next_to_use_head = ring->next_to_use;
1281 
1282 	ret = priv->ops.fill_desc(ring, skb, size, seg_num == 1 ? 1 : 0,
1283 				  DESC_TYPE_SKB);
1284 	if (ret)
1285 		goto head_fill_err;
1286 
1287 	next_to_use_frag = ring->next_to_use;
1288 	/* Fill the fragments */
1289 	for (i = 1; i < seg_num; i++) {
1290 		frag = &skb_shinfo(skb)->frags[i - 1];
1291 		size = skb_frag_size(frag);
1292 
1293 		ret = priv->ops.fill_desc(ring, frag, size,
1294 					  seg_num - 1 == i ? 1 : 0,
1295 					  DESC_TYPE_PAGE);
1296 
1297 		if (ret)
1298 			goto frag_fill_err;
1299 	}
1300 
1301 	/* Complete translate all packets */
1302 	dev_queue = netdev_get_tx_queue(netdev, ring_data->queue_index);
1303 	netdev_tx_sent_queue(dev_queue, skb->len);
1304 
1305 	wmb(); /* Commit all data before submit */
1306 
1307 	hnae3_queue_xmit(ring->tqp, buf_num);
1308 
1309 	return NETDEV_TX_OK;
1310 
1311 frag_fill_err:
1312 	hns3_clear_desc(ring, next_to_use_frag);
1313 
1314 head_fill_err:
1315 	hns3_clear_desc(ring, next_to_use_head);
1316 
1317 out_err_tx_ok:
1318 	dev_kfree_skb_any(skb);
1319 	return NETDEV_TX_OK;
1320 
1321 out_net_tx_busy:
1322 	netif_stop_subqueue(netdev, ring_data->queue_index);
1323 	smp_mb(); /* Commit all data before submit */
1324 
1325 	return NETDEV_TX_BUSY;
1326 }
1327 
1328 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
1329 {
1330 	struct hnae3_handle *h = hns3_get_handle(netdev);
1331 	struct sockaddr *mac_addr = p;
1332 	int ret;
1333 
1334 	if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1335 		return -EADDRNOTAVAIL;
1336 
1337 	if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
1338 		netdev_info(netdev, "already using mac address %pM\n",
1339 			    mac_addr->sa_data);
1340 		return 0;
1341 	}
1342 
1343 	ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
1344 	if (ret) {
1345 		netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
1346 		return ret;
1347 	}
1348 
1349 	ether_addr_copy(netdev->dev_addr, mac_addr->sa_data);
1350 
1351 	return 0;
1352 }
1353 
1354 static int hns3_nic_do_ioctl(struct net_device *netdev,
1355 			     struct ifreq *ifr, int cmd)
1356 {
1357 	struct hnae3_handle *h = hns3_get_handle(netdev);
1358 
1359 	if (!netif_running(netdev))
1360 		return -EINVAL;
1361 
1362 	if (!h->ae_algo->ops->do_ioctl)
1363 		return -EOPNOTSUPP;
1364 
1365 	return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
1366 }
1367 
1368 static int hns3_nic_set_features(struct net_device *netdev,
1369 				 netdev_features_t features)
1370 {
1371 	netdev_features_t changed = netdev->features ^ features;
1372 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1373 	struct hnae3_handle *h = priv->ae_handle;
1374 	int ret;
1375 
1376 	if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) {
1377 		if (features & (NETIF_F_TSO | NETIF_F_TSO6))
1378 			priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tso;
1379 		else
1380 			priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tx;
1381 	}
1382 
1383 	if (changed & (NETIF_F_GRO_HW) && h->ae_algo->ops->set_gro_en) {
1384 		if (features & NETIF_F_GRO_HW)
1385 			ret = h->ae_algo->ops->set_gro_en(h, true);
1386 		else
1387 			ret = h->ae_algo->ops->set_gro_en(h, false);
1388 		if (ret)
1389 			return ret;
1390 	}
1391 
1392 	if ((changed & NETIF_F_HW_VLAN_CTAG_FILTER) &&
1393 	    h->ae_algo->ops->enable_vlan_filter) {
1394 		if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
1395 			h->ae_algo->ops->enable_vlan_filter(h, true);
1396 		else
1397 			h->ae_algo->ops->enable_vlan_filter(h, false);
1398 	}
1399 
1400 	if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
1401 	    h->ae_algo->ops->enable_hw_strip_rxvtag) {
1402 		if (features & NETIF_F_HW_VLAN_CTAG_RX)
1403 			ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, true);
1404 		else
1405 			ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, false);
1406 
1407 		if (ret)
1408 			return ret;
1409 	}
1410 
1411 	if ((changed & NETIF_F_NTUPLE) && h->ae_algo->ops->enable_fd) {
1412 		if (features & NETIF_F_NTUPLE)
1413 			h->ae_algo->ops->enable_fd(h, true);
1414 		else
1415 			h->ae_algo->ops->enable_fd(h, false);
1416 	}
1417 
1418 	netdev->features = features;
1419 	return 0;
1420 }
1421 
1422 static void hns3_nic_get_stats64(struct net_device *netdev,
1423 				 struct rtnl_link_stats64 *stats)
1424 {
1425 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1426 	int queue_num = priv->ae_handle->kinfo.num_tqps;
1427 	struct hnae3_handle *handle = priv->ae_handle;
1428 	struct hns3_enet_ring *ring;
1429 	u64 rx_length_errors = 0;
1430 	u64 rx_crc_errors = 0;
1431 	u64 rx_multicast = 0;
1432 	unsigned int start;
1433 	u64 tx_errors = 0;
1434 	u64 rx_errors = 0;
1435 	unsigned int idx;
1436 	u64 tx_bytes = 0;
1437 	u64 rx_bytes = 0;
1438 	u64 tx_pkts = 0;
1439 	u64 rx_pkts = 0;
1440 	u64 tx_drop = 0;
1441 	u64 rx_drop = 0;
1442 
1443 	if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
1444 		return;
1445 
1446 	handle->ae_algo->ops->update_stats(handle, &netdev->stats);
1447 
1448 	for (idx = 0; idx < queue_num; idx++) {
1449 		/* fetch the tx stats */
1450 		ring = priv->ring_data[idx].ring;
1451 		do {
1452 			start = u64_stats_fetch_begin_irq(&ring->syncp);
1453 			tx_bytes += ring->stats.tx_bytes;
1454 			tx_pkts += ring->stats.tx_pkts;
1455 			tx_drop += ring->stats.sw_err_cnt;
1456 			tx_errors += ring->stats.sw_err_cnt;
1457 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1458 
1459 		/* fetch the rx stats */
1460 		ring = priv->ring_data[idx + queue_num].ring;
1461 		do {
1462 			start = u64_stats_fetch_begin_irq(&ring->syncp);
1463 			rx_bytes += ring->stats.rx_bytes;
1464 			rx_pkts += ring->stats.rx_pkts;
1465 			rx_drop += ring->stats.non_vld_descs;
1466 			rx_drop += ring->stats.l2_err;
1467 			rx_errors += ring->stats.non_vld_descs;
1468 			rx_errors += ring->stats.l2_err;
1469 			rx_crc_errors += ring->stats.l2_err;
1470 			rx_crc_errors += ring->stats.l3l4_csum_err;
1471 			rx_multicast += ring->stats.rx_multicast;
1472 			rx_length_errors += ring->stats.err_pkt_len;
1473 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1474 	}
1475 
1476 	stats->tx_bytes = tx_bytes;
1477 	stats->tx_packets = tx_pkts;
1478 	stats->rx_bytes = rx_bytes;
1479 	stats->rx_packets = rx_pkts;
1480 
1481 	stats->rx_errors = rx_errors;
1482 	stats->multicast = rx_multicast;
1483 	stats->rx_length_errors = rx_length_errors;
1484 	stats->rx_crc_errors = rx_crc_errors;
1485 	stats->rx_missed_errors = netdev->stats.rx_missed_errors;
1486 
1487 	stats->tx_errors = tx_errors;
1488 	stats->rx_dropped = rx_drop;
1489 	stats->tx_dropped = tx_drop;
1490 	stats->collisions = netdev->stats.collisions;
1491 	stats->rx_over_errors = netdev->stats.rx_over_errors;
1492 	stats->rx_frame_errors = netdev->stats.rx_frame_errors;
1493 	stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
1494 	stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
1495 	stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
1496 	stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
1497 	stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
1498 	stats->tx_window_errors = netdev->stats.tx_window_errors;
1499 	stats->rx_compressed = netdev->stats.rx_compressed;
1500 	stats->tx_compressed = netdev->stats.tx_compressed;
1501 }
1502 
1503 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
1504 {
1505 	struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
1506 	struct hnae3_handle *h = hns3_get_handle(netdev);
1507 	struct hnae3_knic_private_info *kinfo = &h->kinfo;
1508 	u8 *prio_tc = mqprio_qopt->qopt.prio_tc_map;
1509 	u8 tc = mqprio_qopt->qopt.num_tc;
1510 	u16 mode = mqprio_qopt->mode;
1511 	u8 hw = mqprio_qopt->qopt.hw;
1512 
1513 	if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
1514 	       mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
1515 		return -EOPNOTSUPP;
1516 
1517 	if (tc > HNAE3_MAX_TC)
1518 		return -EINVAL;
1519 
1520 	if (!netdev)
1521 		return -EINVAL;
1522 
1523 	return (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
1524 		kinfo->dcb_ops->setup_tc(h, tc, prio_tc) : -EOPNOTSUPP;
1525 }
1526 
1527 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
1528 			     void *type_data)
1529 {
1530 	if (type != TC_SETUP_QDISC_MQPRIO)
1531 		return -EOPNOTSUPP;
1532 
1533 	return hns3_setup_tc(dev, type_data);
1534 }
1535 
1536 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
1537 				__be16 proto, u16 vid)
1538 {
1539 	struct hnae3_handle *h = hns3_get_handle(netdev);
1540 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1541 	int ret = -EIO;
1542 
1543 	if (h->ae_algo->ops->set_vlan_filter)
1544 		ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
1545 
1546 	if (!ret)
1547 		set_bit(vid, priv->active_vlans);
1548 
1549 	return ret;
1550 }
1551 
1552 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
1553 				 __be16 proto, u16 vid)
1554 {
1555 	struct hnae3_handle *h = hns3_get_handle(netdev);
1556 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1557 	int ret = -EIO;
1558 
1559 	if (h->ae_algo->ops->set_vlan_filter)
1560 		ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
1561 
1562 	if (!ret)
1563 		clear_bit(vid, priv->active_vlans);
1564 
1565 	return ret;
1566 }
1567 
1568 static int hns3_restore_vlan(struct net_device *netdev)
1569 {
1570 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1571 	int ret = 0;
1572 	u16 vid;
1573 
1574 	for_each_set_bit(vid, priv->active_vlans, VLAN_N_VID) {
1575 		ret = hns3_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
1576 		if (ret) {
1577 			netdev_err(netdev, "Restore vlan: %d filter, ret:%d\n",
1578 				   vid, ret);
1579 			return ret;
1580 		}
1581 	}
1582 
1583 	return ret;
1584 }
1585 
1586 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
1587 				u8 qos, __be16 vlan_proto)
1588 {
1589 	struct hnae3_handle *h = hns3_get_handle(netdev);
1590 	int ret = -EIO;
1591 
1592 	if (h->ae_algo->ops->set_vf_vlan_filter)
1593 		ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
1594 						   qos, vlan_proto);
1595 
1596 	return ret;
1597 }
1598 
1599 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
1600 {
1601 	struct hnae3_handle *h = hns3_get_handle(netdev);
1602 	int ret;
1603 
1604 	if (!h->ae_algo->ops->set_mtu)
1605 		return -EOPNOTSUPP;
1606 
1607 	ret = h->ae_algo->ops->set_mtu(h, new_mtu);
1608 	if (ret)
1609 		netdev_err(netdev, "failed to change MTU in hardware %d\n",
1610 			   ret);
1611 	else
1612 		netdev->mtu = new_mtu;
1613 
1614 	return ret;
1615 }
1616 
1617 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
1618 {
1619 	struct hns3_nic_priv *priv = netdev_priv(ndev);
1620 	struct hns3_enet_ring *tx_ring = NULL;
1621 	int timeout_queue = 0;
1622 	int hw_head, hw_tail;
1623 	int i;
1624 
1625 	/* Find the stopped queue the same way the stack does */
1626 	for (i = 0; i < ndev->real_num_tx_queues; i++) {
1627 		struct netdev_queue *q;
1628 		unsigned long trans_start;
1629 
1630 		q = netdev_get_tx_queue(ndev, i);
1631 		trans_start = q->trans_start;
1632 		if (netif_xmit_stopped(q) &&
1633 		    time_after(jiffies,
1634 			       (trans_start + ndev->watchdog_timeo))) {
1635 			timeout_queue = i;
1636 			break;
1637 		}
1638 	}
1639 
1640 	if (i == ndev->num_tx_queues) {
1641 		netdev_info(ndev,
1642 			    "no netdev TX timeout queue found, timeout count: %llu\n",
1643 			    priv->tx_timeout_count);
1644 		return false;
1645 	}
1646 
1647 	tx_ring = priv->ring_data[timeout_queue].ring;
1648 
1649 	hw_head = readl_relaxed(tx_ring->tqp->io_base +
1650 				HNS3_RING_TX_RING_HEAD_REG);
1651 	hw_tail = readl_relaxed(tx_ring->tqp->io_base +
1652 				HNS3_RING_TX_RING_TAIL_REG);
1653 	netdev_info(ndev,
1654 		    "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",
1655 		    priv->tx_timeout_count,
1656 		    timeout_queue,
1657 		    tx_ring->next_to_use,
1658 		    tx_ring->next_to_clean,
1659 		    hw_head,
1660 		    hw_tail,
1661 		    readl(tx_ring->tqp_vector->mask_addr));
1662 
1663 	return true;
1664 }
1665 
1666 static void hns3_nic_net_timeout(struct net_device *ndev)
1667 {
1668 	struct hns3_nic_priv *priv = netdev_priv(ndev);
1669 	struct hnae3_handle *h = priv->ae_handle;
1670 
1671 	if (!hns3_get_tx_timeo_queue_info(ndev))
1672 		return;
1673 
1674 	priv->tx_timeout_count++;
1675 
1676 	/* request the reset, and let the hclge to determine
1677 	 * which reset level should be done
1678 	 */
1679 	if (h->ae_algo->ops->reset_event)
1680 		h->ae_algo->ops->reset_event(h->pdev, h);
1681 }
1682 
1683 static const struct net_device_ops hns3_nic_netdev_ops = {
1684 	.ndo_open		= hns3_nic_net_open,
1685 	.ndo_stop		= hns3_nic_net_stop,
1686 	.ndo_start_xmit		= hns3_nic_net_xmit,
1687 	.ndo_tx_timeout		= hns3_nic_net_timeout,
1688 	.ndo_set_mac_address	= hns3_nic_net_set_mac_address,
1689 	.ndo_do_ioctl		= hns3_nic_do_ioctl,
1690 	.ndo_change_mtu		= hns3_nic_change_mtu,
1691 	.ndo_set_features	= hns3_nic_set_features,
1692 	.ndo_get_stats64	= hns3_nic_get_stats64,
1693 	.ndo_setup_tc		= hns3_nic_setup_tc,
1694 	.ndo_set_rx_mode	= hns3_nic_set_rx_mode,
1695 	.ndo_vlan_rx_add_vid	= hns3_vlan_rx_add_vid,
1696 	.ndo_vlan_rx_kill_vid	= hns3_vlan_rx_kill_vid,
1697 	.ndo_set_vf_vlan	= hns3_ndo_set_vf_vlan,
1698 };
1699 
1700 static bool hns3_is_phys_func(struct pci_dev *pdev)
1701 {
1702 	u32 dev_id = pdev->device;
1703 
1704 	switch (dev_id) {
1705 	case HNAE3_DEV_ID_GE:
1706 	case HNAE3_DEV_ID_25GE:
1707 	case HNAE3_DEV_ID_25GE_RDMA:
1708 	case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
1709 	case HNAE3_DEV_ID_50GE_RDMA:
1710 	case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
1711 	case HNAE3_DEV_ID_100G_RDMA_MACSEC:
1712 		return true;
1713 	case HNAE3_DEV_ID_100G_VF:
1714 	case HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF:
1715 		return false;
1716 	default:
1717 		dev_warn(&pdev->dev, "un-recognized pci device-id %d",
1718 			 dev_id);
1719 	}
1720 
1721 	return false;
1722 }
1723 
1724 static void hns3_disable_sriov(struct pci_dev *pdev)
1725 {
1726 	/* If our VFs are assigned we cannot shut down SR-IOV
1727 	 * without causing issues, so just leave the hardware
1728 	 * available but disabled
1729 	 */
1730 	if (pci_vfs_assigned(pdev)) {
1731 		dev_warn(&pdev->dev,
1732 			 "disabling driver while VFs are assigned\n");
1733 		return;
1734 	}
1735 
1736 	pci_disable_sriov(pdev);
1737 }
1738 
1739 static void hns3_get_dev_capability(struct pci_dev *pdev,
1740 				    struct hnae3_ae_dev *ae_dev)
1741 {
1742 	if (pdev->revision >= 0x21) {
1743 		hnae3_set_bit(ae_dev->flag, HNAE3_DEV_SUPPORT_FD_B, 1);
1744 		hnae3_set_bit(ae_dev->flag, HNAE3_DEV_SUPPORT_GRO_B, 1);
1745 	}
1746 }
1747 
1748 /* hns3_probe - Device initialization routine
1749  * @pdev: PCI device information struct
1750  * @ent: entry in hns3_pci_tbl
1751  *
1752  * hns3_probe initializes a PF identified by a pci_dev structure.
1753  * The OS initialization, configuring of the PF private structure,
1754  * and a hardware reset occur.
1755  *
1756  * Returns 0 on success, negative on failure
1757  */
1758 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1759 {
1760 	struct hnae3_ae_dev *ae_dev;
1761 	int ret;
1762 
1763 	ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev),
1764 			      GFP_KERNEL);
1765 	if (!ae_dev) {
1766 		ret = -ENOMEM;
1767 		return ret;
1768 	}
1769 
1770 	ae_dev->pdev = pdev;
1771 	ae_dev->flag = ent->driver_data;
1772 	ae_dev->dev_type = HNAE3_DEV_KNIC;
1773 	ae_dev->reset_type = HNAE3_NONE_RESET;
1774 	hns3_get_dev_capability(pdev, ae_dev);
1775 	pci_set_drvdata(pdev, ae_dev);
1776 
1777 	hnae3_register_ae_dev(ae_dev);
1778 
1779 	return 0;
1780 }
1781 
1782 /* hns3_remove - Device removal routine
1783  * @pdev: PCI device information struct
1784  */
1785 static void hns3_remove(struct pci_dev *pdev)
1786 {
1787 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1788 
1789 	if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
1790 		hns3_disable_sriov(pdev);
1791 
1792 	hnae3_unregister_ae_dev(ae_dev);
1793 	pci_set_drvdata(pdev, NULL);
1794 }
1795 
1796 /**
1797  * hns3_pci_sriov_configure
1798  * @pdev: pointer to a pci_dev structure
1799  * @num_vfs: number of VFs to allocate
1800  *
1801  * Enable or change the number of VFs. Called when the user updates the number
1802  * of VFs in sysfs.
1803  **/
1804 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
1805 {
1806 	int ret;
1807 
1808 	if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
1809 		dev_warn(&pdev->dev, "Can not config SRIOV\n");
1810 		return -EINVAL;
1811 	}
1812 
1813 	if (num_vfs) {
1814 		ret = pci_enable_sriov(pdev, num_vfs);
1815 		if (ret)
1816 			dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
1817 		else
1818 			return num_vfs;
1819 	} else if (!pci_vfs_assigned(pdev)) {
1820 		pci_disable_sriov(pdev);
1821 	} else {
1822 		dev_warn(&pdev->dev,
1823 			 "Unable to free VFs because some are assigned to VMs.\n");
1824 	}
1825 
1826 	return 0;
1827 }
1828 
1829 static void hns3_shutdown(struct pci_dev *pdev)
1830 {
1831 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1832 
1833 	hnae3_unregister_ae_dev(ae_dev);
1834 	devm_kfree(&pdev->dev, ae_dev);
1835 	pci_set_drvdata(pdev, NULL);
1836 
1837 	if (system_state == SYSTEM_POWER_OFF)
1838 		pci_set_power_state(pdev, PCI_D3hot);
1839 }
1840 
1841 static pci_ers_result_t hns3_error_detected(struct pci_dev *pdev,
1842 					    pci_channel_state_t state)
1843 {
1844 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1845 	pci_ers_result_t ret;
1846 
1847 	dev_info(&pdev->dev, "PCI error detected, state(=%d)!!\n", state);
1848 
1849 	if (state == pci_channel_io_perm_failure)
1850 		return PCI_ERS_RESULT_DISCONNECT;
1851 
1852 	if (!ae_dev) {
1853 		dev_err(&pdev->dev,
1854 			"Can't recover - error happened during device init\n");
1855 		return PCI_ERS_RESULT_NONE;
1856 	}
1857 
1858 	if (ae_dev->ops->handle_hw_ras_error)
1859 		ret = ae_dev->ops->handle_hw_ras_error(ae_dev);
1860 	else
1861 		return PCI_ERS_RESULT_NONE;
1862 
1863 	return ret;
1864 }
1865 
1866 static pci_ers_result_t hns3_slot_reset(struct pci_dev *pdev)
1867 {
1868 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1869 	struct device *dev = &pdev->dev;
1870 
1871 	dev_info(dev, "requesting reset due to PCI error\n");
1872 
1873 	/* request the reset */
1874 	if (ae_dev->ops->reset_event) {
1875 		ae_dev->ops->reset_event(pdev, NULL);
1876 		return PCI_ERS_RESULT_RECOVERED;
1877 	}
1878 
1879 	return PCI_ERS_RESULT_DISCONNECT;
1880 }
1881 
1882 static void hns3_reset_prepare(struct pci_dev *pdev)
1883 {
1884 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1885 
1886 	dev_info(&pdev->dev, "hns3 flr prepare\n");
1887 	if (ae_dev && ae_dev->ops && ae_dev->ops->flr_prepare)
1888 		ae_dev->ops->flr_prepare(ae_dev);
1889 }
1890 
1891 static void hns3_reset_done(struct pci_dev *pdev)
1892 {
1893 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1894 
1895 	dev_info(&pdev->dev, "hns3 flr done\n");
1896 	if (ae_dev && ae_dev->ops && ae_dev->ops->flr_done)
1897 		ae_dev->ops->flr_done(ae_dev);
1898 }
1899 
1900 static const struct pci_error_handlers hns3_err_handler = {
1901 	.error_detected = hns3_error_detected,
1902 	.slot_reset     = hns3_slot_reset,
1903 	.reset_prepare	= hns3_reset_prepare,
1904 	.reset_done	= hns3_reset_done,
1905 };
1906 
1907 static struct pci_driver hns3_driver = {
1908 	.name     = hns3_driver_name,
1909 	.id_table = hns3_pci_tbl,
1910 	.probe    = hns3_probe,
1911 	.remove   = hns3_remove,
1912 	.shutdown = hns3_shutdown,
1913 	.sriov_configure = hns3_pci_sriov_configure,
1914 	.err_handler    = &hns3_err_handler,
1915 };
1916 
1917 /* set default feature to hns3 */
1918 static void hns3_set_default_feature(struct net_device *netdev)
1919 {
1920 	struct hnae3_handle *h = hns3_get_handle(netdev);
1921 	struct pci_dev *pdev = h->pdev;
1922 
1923 	netdev->priv_flags |= IFF_UNICAST_FLT;
1924 
1925 	netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1926 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1927 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1928 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1929 		NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1930 
1931 	netdev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
1932 
1933 	netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
1934 
1935 	netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1936 		NETIF_F_HW_VLAN_CTAG_FILTER |
1937 		NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
1938 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1939 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1940 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1941 		NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1942 
1943 	netdev->vlan_features |=
1944 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
1945 		NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO |
1946 		NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1947 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1948 		NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1949 
1950 	netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1951 		NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
1952 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1953 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
1954 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
1955 		NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
1956 
1957 	if (pdev->revision >= 0x21) {
1958 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER |
1959 			NETIF_F_GRO_HW;
1960 		netdev->features |= NETIF_F_GRO_HW;
1961 
1962 		if (!(h->flags & HNAE3_SUPPORT_VF)) {
1963 			netdev->hw_features |= NETIF_F_NTUPLE;
1964 			netdev->features |= NETIF_F_NTUPLE;
1965 		}
1966 	}
1967 }
1968 
1969 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
1970 			     struct hns3_desc_cb *cb)
1971 {
1972 	unsigned int order = hnae3_page_order(ring);
1973 	struct page *p;
1974 
1975 	p = dev_alloc_pages(order);
1976 	if (!p)
1977 		return -ENOMEM;
1978 
1979 	cb->priv = p;
1980 	cb->page_offset = 0;
1981 	cb->reuse_flag = 0;
1982 	cb->buf  = page_address(p);
1983 	cb->length = hnae3_page_size(ring);
1984 	cb->type = DESC_TYPE_PAGE;
1985 
1986 	return 0;
1987 }
1988 
1989 static void hns3_free_buffer(struct hns3_enet_ring *ring,
1990 			     struct hns3_desc_cb *cb)
1991 {
1992 	if (cb->type == DESC_TYPE_SKB)
1993 		dev_kfree_skb_any((struct sk_buff *)cb->priv);
1994 	else if (!HNAE3_IS_TX_RING(ring))
1995 		put_page((struct page *)cb->priv);
1996 	memset(cb, 0, sizeof(*cb));
1997 }
1998 
1999 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
2000 {
2001 	cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
2002 			       cb->length, ring_to_dma_dir(ring));
2003 
2004 	if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
2005 		return -EIO;
2006 
2007 	return 0;
2008 }
2009 
2010 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
2011 			      struct hns3_desc_cb *cb)
2012 {
2013 	if (cb->type == DESC_TYPE_SKB)
2014 		dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
2015 				 ring_to_dma_dir(ring));
2016 	else if (cb->length)
2017 		dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
2018 			       ring_to_dma_dir(ring));
2019 }
2020 
2021 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
2022 {
2023 	hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2024 	ring->desc[i].addr = 0;
2025 }
2026 
2027 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i)
2028 {
2029 	struct hns3_desc_cb *cb = &ring->desc_cb[i];
2030 
2031 	if (!ring->desc_cb[i].dma)
2032 		return;
2033 
2034 	hns3_buffer_detach(ring, i);
2035 	hns3_free_buffer(ring, cb);
2036 }
2037 
2038 static void hns3_free_buffers(struct hns3_enet_ring *ring)
2039 {
2040 	int i;
2041 
2042 	for (i = 0; i < ring->desc_num; i++)
2043 		hns3_free_buffer_detach(ring, i);
2044 }
2045 
2046 /* free desc along with its attached buffer */
2047 static void hns3_free_desc(struct hns3_enet_ring *ring)
2048 {
2049 	int size = ring->desc_num * sizeof(ring->desc[0]);
2050 
2051 	hns3_free_buffers(ring);
2052 
2053 	if (ring->desc) {
2054 		dma_free_coherent(ring_to_dev(ring), size,
2055 				  ring->desc, ring->desc_dma_addr);
2056 		ring->desc = NULL;
2057 	}
2058 }
2059 
2060 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
2061 {
2062 	int size = ring->desc_num * sizeof(ring->desc[0]);
2063 
2064 	ring->desc = dma_alloc_coherent(ring_to_dev(ring), size,
2065 					&ring->desc_dma_addr, GFP_KERNEL);
2066 	if (!ring->desc)
2067 		return -ENOMEM;
2068 
2069 	return 0;
2070 }
2071 
2072 static int hns3_reserve_buffer_map(struct hns3_enet_ring *ring,
2073 				   struct hns3_desc_cb *cb)
2074 {
2075 	int ret;
2076 
2077 	ret = hns3_alloc_buffer(ring, cb);
2078 	if (ret)
2079 		goto out;
2080 
2081 	ret = hns3_map_buffer(ring, cb);
2082 	if (ret)
2083 		goto out_with_buf;
2084 
2085 	return 0;
2086 
2087 out_with_buf:
2088 	hns3_free_buffer(ring, cb);
2089 out:
2090 	return ret;
2091 }
2092 
2093 static int hns3_alloc_buffer_attach(struct hns3_enet_ring *ring, int i)
2094 {
2095 	int ret = hns3_reserve_buffer_map(ring, &ring->desc_cb[i]);
2096 
2097 	if (ret)
2098 		return ret;
2099 
2100 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2101 
2102 	return 0;
2103 }
2104 
2105 /* Allocate memory for raw pkg, and map with dma */
2106 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
2107 {
2108 	int i, j, ret;
2109 
2110 	for (i = 0; i < ring->desc_num; i++) {
2111 		ret = hns3_alloc_buffer_attach(ring, i);
2112 		if (ret)
2113 			goto out_buffer_fail;
2114 	}
2115 
2116 	return 0;
2117 
2118 out_buffer_fail:
2119 	for (j = i - 1; j >= 0; j--)
2120 		hns3_free_buffer_detach(ring, j);
2121 	return ret;
2122 }
2123 
2124 /* detach a in-used buffer and replace with a reserved one  */
2125 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
2126 				struct hns3_desc_cb *res_cb)
2127 {
2128 	hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2129 	ring->desc_cb[i] = *res_cb;
2130 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2131 	ring->desc[i].rx.bd_base_info = 0;
2132 }
2133 
2134 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
2135 {
2136 	ring->desc_cb[i].reuse_flag = 0;
2137 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma
2138 		+ ring->desc_cb[i].page_offset);
2139 	ring->desc[i].rx.bd_base_info = 0;
2140 }
2141 
2142 static void hns3_nic_reclaim_one_desc(struct hns3_enet_ring *ring, int *bytes,
2143 				      int *pkts)
2144 {
2145 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
2146 
2147 	(*pkts) += (desc_cb->type == DESC_TYPE_SKB);
2148 	(*bytes) += desc_cb->length;
2149 	/* desc_cb will be cleaned, after hnae3_free_buffer_detach*/
2150 	hns3_free_buffer_detach(ring, ring->next_to_clean);
2151 
2152 	ring_ptr_move_fw(ring, next_to_clean);
2153 }
2154 
2155 static int is_valid_clean_head(struct hns3_enet_ring *ring, int h)
2156 {
2157 	int u = ring->next_to_use;
2158 	int c = ring->next_to_clean;
2159 
2160 	if (unlikely(h > ring->desc_num))
2161 		return 0;
2162 
2163 	return u > c ? (h > c && h <= u) : (h > c || h <= u);
2164 }
2165 
2166 void hns3_clean_tx_ring(struct hns3_enet_ring *ring)
2167 {
2168 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2169 	struct hns3_nic_priv *priv = netdev_priv(netdev);
2170 	struct netdev_queue *dev_queue;
2171 	int bytes, pkts;
2172 	int head;
2173 
2174 	head = readl_relaxed(ring->tqp->io_base + HNS3_RING_TX_RING_HEAD_REG);
2175 	rmb(); /* Make sure head is ready before touch any data */
2176 
2177 	if (is_ring_empty(ring) || head == ring->next_to_clean)
2178 		return; /* no data to poll */
2179 
2180 	if (unlikely(!is_valid_clean_head(ring, head))) {
2181 		netdev_err(netdev, "wrong head (%d, %d-%d)\n", head,
2182 			   ring->next_to_use, ring->next_to_clean);
2183 
2184 		u64_stats_update_begin(&ring->syncp);
2185 		ring->stats.io_err_cnt++;
2186 		u64_stats_update_end(&ring->syncp);
2187 		return;
2188 	}
2189 
2190 	bytes = 0;
2191 	pkts = 0;
2192 	while (head != ring->next_to_clean) {
2193 		hns3_nic_reclaim_one_desc(ring, &bytes, &pkts);
2194 		/* Issue prefetch for next Tx descriptor */
2195 		prefetch(&ring->desc_cb[ring->next_to_clean]);
2196 	}
2197 
2198 	ring->tqp_vector->tx_group.total_bytes += bytes;
2199 	ring->tqp_vector->tx_group.total_packets += pkts;
2200 
2201 	u64_stats_update_begin(&ring->syncp);
2202 	ring->stats.tx_bytes += bytes;
2203 	ring->stats.tx_pkts += pkts;
2204 	u64_stats_update_end(&ring->syncp);
2205 
2206 	dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
2207 	netdev_tx_completed_queue(dev_queue, pkts, bytes);
2208 
2209 	if (unlikely(pkts && netif_carrier_ok(netdev) &&
2210 		     (ring_space(ring) > HNS3_MAX_BD_PER_PKT))) {
2211 		/* Make sure that anybody stopping the queue after this
2212 		 * sees the new next_to_clean.
2213 		 */
2214 		smp_mb();
2215 		if (netif_tx_queue_stopped(dev_queue) &&
2216 		    !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
2217 			netif_tx_wake_queue(dev_queue);
2218 			ring->stats.restart_queue++;
2219 		}
2220 	}
2221 }
2222 
2223 static int hns3_desc_unused(struct hns3_enet_ring *ring)
2224 {
2225 	int ntc = ring->next_to_clean;
2226 	int ntu = ring->next_to_use;
2227 
2228 	return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
2229 }
2230 
2231 static void
2232 hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring, int cleand_count)
2233 {
2234 	struct hns3_desc_cb *desc_cb;
2235 	struct hns3_desc_cb res_cbs;
2236 	int i, ret;
2237 
2238 	for (i = 0; i < cleand_count; i++) {
2239 		desc_cb = &ring->desc_cb[ring->next_to_use];
2240 		if (desc_cb->reuse_flag) {
2241 			u64_stats_update_begin(&ring->syncp);
2242 			ring->stats.reuse_pg_cnt++;
2243 			u64_stats_update_end(&ring->syncp);
2244 
2245 			hns3_reuse_buffer(ring, ring->next_to_use);
2246 		} else {
2247 			ret = hns3_reserve_buffer_map(ring, &res_cbs);
2248 			if (ret) {
2249 				u64_stats_update_begin(&ring->syncp);
2250 				ring->stats.sw_err_cnt++;
2251 				u64_stats_update_end(&ring->syncp);
2252 
2253 				netdev_err(ring->tqp->handle->kinfo.netdev,
2254 					   "hnae reserve buffer map failed.\n");
2255 				break;
2256 			}
2257 			hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
2258 		}
2259 
2260 		ring_ptr_move_fw(ring, next_to_use);
2261 	}
2262 
2263 	wmb(); /* Make all data has been write before submit */
2264 	writel_relaxed(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
2265 }
2266 
2267 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
2268 				struct hns3_enet_ring *ring, int pull_len,
2269 				struct hns3_desc_cb *desc_cb)
2270 {
2271 	struct hns3_desc *desc;
2272 	u32 truesize;
2273 	int size;
2274 	int last_offset;
2275 	bool twobufs;
2276 
2277 	twobufs = ((PAGE_SIZE < 8192) &&
2278 		hnae3_buf_size(ring) == HNS3_BUFFER_SIZE_2048);
2279 
2280 	desc = &ring->desc[ring->next_to_clean];
2281 	size = le16_to_cpu(desc->rx.size);
2282 
2283 	truesize = hnae3_buf_size(ring);
2284 
2285 	if (!twobufs)
2286 		last_offset = hnae3_page_size(ring) - hnae3_buf_size(ring);
2287 
2288 	skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
2289 			size - pull_len, truesize);
2290 
2291 	 /* Avoid re-using remote pages,flag default unreuse */
2292 	if (unlikely(page_to_nid(desc_cb->priv) != numa_node_id()))
2293 		return;
2294 
2295 	if (twobufs) {
2296 		/* If we are only owner of page we can reuse it */
2297 		if (likely(page_count(desc_cb->priv) == 1)) {
2298 			/* Flip page offset to other buffer */
2299 			desc_cb->page_offset ^= truesize;
2300 
2301 			desc_cb->reuse_flag = 1;
2302 			/* bump ref count on page before it is given*/
2303 			get_page(desc_cb->priv);
2304 		}
2305 		return;
2306 	}
2307 
2308 	/* Move offset up to the next cache line */
2309 	desc_cb->page_offset += truesize;
2310 
2311 	if (desc_cb->page_offset <= last_offset) {
2312 		desc_cb->reuse_flag = 1;
2313 		/* Bump ref count on page before it is given*/
2314 		get_page(desc_cb->priv);
2315 	}
2316 }
2317 
2318 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
2319 			     struct hns3_desc *desc)
2320 {
2321 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2322 	int l3_type, l4_type;
2323 	u32 bd_base_info;
2324 	int ol4_type;
2325 	u32 l234info;
2326 
2327 	bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2328 	l234info = le32_to_cpu(desc->rx.l234_info);
2329 
2330 	skb->ip_summed = CHECKSUM_NONE;
2331 
2332 	skb_checksum_none_assert(skb);
2333 
2334 	if (!(netdev->features & NETIF_F_RXCSUM))
2335 		return;
2336 
2337 	/* We MUST enable hardware checksum before enabling hardware GRO */
2338 	if (skb_shinfo(skb)->gso_size) {
2339 		skb->ip_summed = CHECKSUM_UNNECESSARY;
2340 		return;
2341 	}
2342 
2343 	/* check if hardware has done checksum */
2344 	if (!hnae3_get_bit(bd_base_info, HNS3_RXD_L3L4P_B))
2345 		return;
2346 
2347 	if (unlikely(hnae3_get_bit(l234info, HNS3_RXD_L3E_B) ||
2348 		     hnae3_get_bit(l234info, HNS3_RXD_L4E_B) ||
2349 		     hnae3_get_bit(l234info, HNS3_RXD_OL3E_B) ||
2350 		     hnae3_get_bit(l234info, HNS3_RXD_OL4E_B))) {
2351 		u64_stats_update_begin(&ring->syncp);
2352 		ring->stats.l3l4_csum_err++;
2353 		u64_stats_update_end(&ring->syncp);
2354 
2355 		return;
2356 	}
2357 
2358 	l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
2359 				  HNS3_RXD_L3ID_S);
2360 	l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
2361 				  HNS3_RXD_L4ID_S);
2362 
2363 	ol4_type = hnae3_get_field(l234info, HNS3_RXD_OL4ID_M,
2364 				   HNS3_RXD_OL4ID_S);
2365 	switch (ol4_type) {
2366 	case HNS3_OL4_TYPE_MAC_IN_UDP:
2367 	case HNS3_OL4_TYPE_NVGRE:
2368 		skb->csum_level = 1;
2369 		/* fall through */
2370 	case HNS3_OL4_TYPE_NO_TUN:
2371 		/* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
2372 		if ((l3_type == HNS3_L3_TYPE_IPV4 ||
2373 		     l3_type == HNS3_L3_TYPE_IPV6) &&
2374 		    (l4_type == HNS3_L4_TYPE_UDP ||
2375 		     l4_type == HNS3_L4_TYPE_TCP ||
2376 		     l4_type == HNS3_L4_TYPE_SCTP))
2377 			skb->ip_summed = CHECKSUM_UNNECESSARY;
2378 		break;
2379 	default:
2380 		break;
2381 	}
2382 }
2383 
2384 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
2385 {
2386 	if (skb_has_frag_list(skb))
2387 		napi_gro_flush(&ring->tqp_vector->napi, false);
2388 
2389 	napi_gro_receive(&ring->tqp_vector->napi, skb);
2390 }
2391 
2392 static bool hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
2393 				struct hns3_desc *desc, u32 l234info,
2394 				u16 *vlan_tag)
2395 {
2396 	struct pci_dev *pdev = ring->tqp->handle->pdev;
2397 
2398 	if (pdev->revision == 0x20) {
2399 		*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2400 		if (!(*vlan_tag & VLAN_VID_MASK))
2401 			*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2402 
2403 		return (*vlan_tag != 0);
2404 	}
2405 
2406 #define HNS3_STRP_OUTER_VLAN	0x1
2407 #define HNS3_STRP_INNER_VLAN	0x2
2408 
2409 	switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
2410 				HNS3_RXD_STRP_TAGP_S)) {
2411 	case HNS3_STRP_OUTER_VLAN:
2412 		*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2413 		return true;
2414 	case HNS3_STRP_INNER_VLAN:
2415 		*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2416 		return true;
2417 	default:
2418 		return false;
2419 	}
2420 }
2421 
2422 static int hns3_alloc_skb(struct hns3_enet_ring *ring, int length,
2423 			  unsigned char *va)
2424 {
2425 #define HNS3_NEED_ADD_FRAG	1
2426 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
2427 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2428 	struct sk_buff *skb;
2429 
2430 	ring->skb = napi_alloc_skb(&ring->tqp_vector->napi, HNS3_RX_HEAD_SIZE);
2431 	skb = ring->skb;
2432 	if (unlikely(!skb)) {
2433 		netdev_err(netdev, "alloc rx skb fail\n");
2434 
2435 		u64_stats_update_begin(&ring->syncp);
2436 		ring->stats.sw_err_cnt++;
2437 		u64_stats_update_end(&ring->syncp);
2438 
2439 		return -ENOMEM;
2440 	}
2441 
2442 	prefetchw(skb->data);
2443 
2444 	ring->pending_buf = 1;
2445 	ring->frag_num = 0;
2446 	ring->tail_skb = NULL;
2447 	if (length <= HNS3_RX_HEAD_SIZE) {
2448 		memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
2449 
2450 		/* We can reuse buffer as-is, just make sure it is local */
2451 		if (likely(page_to_nid(desc_cb->priv) == numa_node_id()))
2452 			desc_cb->reuse_flag = 1;
2453 		else /* This page cannot be reused so discard it */
2454 			put_page(desc_cb->priv);
2455 
2456 		ring_ptr_move_fw(ring, next_to_clean);
2457 		return 0;
2458 	}
2459 	u64_stats_update_begin(&ring->syncp);
2460 	ring->stats.seg_pkt_cnt++;
2461 	u64_stats_update_end(&ring->syncp);
2462 
2463 	ring->pull_len = eth_get_headlen(va, HNS3_RX_HEAD_SIZE);
2464 	__skb_put(skb, ring->pull_len);
2465 	hns3_nic_reuse_page(skb, ring->frag_num++, ring, ring->pull_len,
2466 			    desc_cb);
2467 	ring_ptr_move_fw(ring, next_to_clean);
2468 
2469 	return HNS3_NEED_ADD_FRAG;
2470 }
2471 
2472 static int hns3_add_frag(struct hns3_enet_ring *ring, struct hns3_desc *desc,
2473 			 struct sk_buff **out_skb, bool pending)
2474 {
2475 	struct sk_buff *skb = *out_skb;
2476 	struct sk_buff *head_skb = *out_skb;
2477 	struct sk_buff *new_skb;
2478 	struct hns3_desc_cb *desc_cb;
2479 	struct hns3_desc *pre_desc;
2480 	u32 bd_base_info;
2481 	int pre_bd;
2482 
2483 	/* if there is pending bd, the SW param next_to_clean has moved
2484 	 * to next and the next is NULL
2485 	 */
2486 	if (pending) {
2487 		pre_bd = (ring->next_to_clean - 1 + ring->desc_num) %
2488 			ring->desc_num;
2489 		pre_desc = &ring->desc[pre_bd];
2490 		bd_base_info = le32_to_cpu(pre_desc->rx.bd_base_info);
2491 	} else {
2492 		bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2493 	}
2494 
2495 	while (!hnae3_get_bit(bd_base_info, HNS3_RXD_FE_B)) {
2496 		desc = &ring->desc[ring->next_to_clean];
2497 		desc_cb = &ring->desc_cb[ring->next_to_clean];
2498 		bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2499 		if (!hnae3_get_bit(bd_base_info, HNS3_RXD_VLD_B))
2500 			return -ENXIO;
2501 
2502 		if (unlikely(ring->frag_num >= MAX_SKB_FRAGS)) {
2503 			new_skb = napi_alloc_skb(&ring->tqp_vector->napi,
2504 						 HNS3_RX_HEAD_SIZE);
2505 			if (unlikely(!new_skb)) {
2506 				netdev_err(ring->tqp->handle->kinfo.netdev,
2507 					   "alloc rx skb frag fail\n");
2508 				return -ENXIO;
2509 			}
2510 			ring->frag_num = 0;
2511 
2512 			if (ring->tail_skb) {
2513 				ring->tail_skb->next = new_skb;
2514 				ring->tail_skb = new_skb;
2515 			} else {
2516 				skb_shinfo(skb)->frag_list = new_skb;
2517 				ring->tail_skb = new_skb;
2518 			}
2519 		}
2520 
2521 		if (ring->tail_skb) {
2522 			head_skb->truesize += hnae3_buf_size(ring);
2523 			head_skb->data_len += le16_to_cpu(desc->rx.size);
2524 			head_skb->len += le16_to_cpu(desc->rx.size);
2525 			skb = ring->tail_skb;
2526 		}
2527 
2528 		hns3_nic_reuse_page(skb, ring->frag_num++, ring, 0, desc_cb);
2529 		ring_ptr_move_fw(ring, next_to_clean);
2530 		ring->pending_buf++;
2531 	}
2532 
2533 	return 0;
2534 }
2535 
2536 static void hns3_set_gro_param(struct sk_buff *skb, u32 l234info,
2537 			       u32 bd_base_info)
2538 {
2539 	u16 gro_count;
2540 	u32 l3_type;
2541 
2542 	gro_count = hnae3_get_field(l234info, HNS3_RXD_GRO_COUNT_M,
2543 				    HNS3_RXD_GRO_COUNT_S);
2544 	/* if there is no HW GRO, do not set gro params */
2545 	if (!gro_count)
2546 		return;
2547 
2548 	/* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
2549 	 * to skb_shinfo(skb)->gso_segs
2550 	 */
2551 	NAPI_GRO_CB(skb)->count = gro_count;
2552 
2553 	l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
2554 				  HNS3_RXD_L3ID_S);
2555 	if (l3_type == HNS3_L3_TYPE_IPV4)
2556 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2557 	else if (l3_type == HNS3_L3_TYPE_IPV6)
2558 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
2559 	else
2560 		return;
2561 
2562 	skb_shinfo(skb)->gso_size = hnae3_get_field(bd_base_info,
2563 						    HNS3_RXD_GRO_SIZE_M,
2564 						    HNS3_RXD_GRO_SIZE_S);
2565 	if (skb_shinfo(skb)->gso_size)
2566 		tcp_gro_complete(skb);
2567 }
2568 
2569 static void hns3_set_rx_skb_rss_type(struct hns3_enet_ring *ring,
2570 				     struct sk_buff *skb)
2571 {
2572 	struct hnae3_handle *handle = ring->tqp->handle;
2573 	enum pkt_hash_types rss_type;
2574 	struct hns3_desc *desc;
2575 	int last_bd;
2576 
2577 	/* When driver handle the rss type, ring->next_to_clean indicates the
2578 	 * first descriptor of next packet, need -1 here.
2579 	 */
2580 	last_bd = (ring->next_to_clean - 1 + ring->desc_num) % ring->desc_num;
2581 	desc = &ring->desc[last_bd];
2582 
2583 	if (le32_to_cpu(desc->rx.rss_hash))
2584 		rss_type = handle->kinfo.rss_type;
2585 	else
2586 		rss_type = PKT_HASH_TYPE_NONE;
2587 
2588 	skb_set_hash(skb, le32_to_cpu(desc->rx.rss_hash), rss_type);
2589 }
2590 
2591 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring,
2592 			     struct sk_buff **out_skb)
2593 {
2594 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2595 	enum hns3_pkt_l2t_type l2_frame_type;
2596 	struct sk_buff *skb = ring->skb;
2597 	struct hns3_desc_cb *desc_cb;
2598 	struct hns3_desc *desc;
2599 	u32 bd_base_info;
2600 	u32 l234info;
2601 	int length;
2602 	int ret;
2603 
2604 	desc = &ring->desc[ring->next_to_clean];
2605 	desc_cb = &ring->desc_cb[ring->next_to_clean];
2606 
2607 	prefetch(desc);
2608 
2609 	length = le16_to_cpu(desc->rx.size);
2610 	bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2611 
2612 	/* Check valid BD */
2613 	if (unlikely(!hnae3_get_bit(bd_base_info, HNS3_RXD_VLD_B)))
2614 		return -ENXIO;
2615 
2616 	if (!skb)
2617 		ring->va = (unsigned char *)desc_cb->buf + desc_cb->page_offset;
2618 
2619 	/* Prefetch first cache line of first page
2620 	 * Idea is to cache few bytes of the header of the packet. Our L1 Cache
2621 	 * line size is 64B so need to prefetch twice to make it 128B. But in
2622 	 * actual we can have greater size of caches with 128B Level 1 cache
2623 	 * lines. In such a case, single fetch would suffice to cache in the
2624 	 * relevant part of the header.
2625 	 */
2626 	prefetch(ring->va);
2627 #if L1_CACHE_BYTES < 128
2628 	prefetch(ring->va + L1_CACHE_BYTES);
2629 #endif
2630 
2631 	if (!skb) {
2632 		ret = hns3_alloc_skb(ring, length, ring->va);
2633 		*out_skb = skb = ring->skb;
2634 
2635 		if (ret < 0) /* alloc buffer fail */
2636 			return ret;
2637 		if (ret > 0) { /* need add frag */
2638 			ret = hns3_add_frag(ring, desc, &skb, false);
2639 			if (ret)
2640 				return ret;
2641 
2642 			/* As the head data may be changed when GRO enable, copy
2643 			 * the head data in after other data rx completed
2644 			 */
2645 			memcpy(skb->data, ring->va,
2646 			       ALIGN(ring->pull_len, sizeof(long)));
2647 		}
2648 	} else {
2649 		ret = hns3_add_frag(ring, desc, &skb, true);
2650 		if (ret)
2651 			return ret;
2652 
2653 		/* As the head data may be changed when GRO enable, copy
2654 		 * the head data in after other data rx completed
2655 		 */
2656 		memcpy(skb->data, ring->va,
2657 		       ALIGN(ring->pull_len, sizeof(long)));
2658 	}
2659 
2660 	l234info = le32_to_cpu(desc->rx.l234_info);
2661 	bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2662 
2663 	/* Based on hw strategy, the tag offloaded will be stored at
2664 	 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
2665 	 * in one layer tag case.
2666 	 */
2667 	if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
2668 		u16 vlan_tag;
2669 
2670 		if (hns3_parse_vlan_tag(ring, desc, l234info, &vlan_tag))
2671 			__vlan_hwaccel_put_tag(skb,
2672 					       htons(ETH_P_8021Q),
2673 					       vlan_tag);
2674 	}
2675 
2676 	if (unlikely(!hnae3_get_bit(bd_base_info, HNS3_RXD_VLD_B))) {
2677 		u64_stats_update_begin(&ring->syncp);
2678 		ring->stats.non_vld_descs++;
2679 		u64_stats_update_end(&ring->syncp);
2680 
2681 		dev_kfree_skb_any(skb);
2682 		return -EINVAL;
2683 	}
2684 
2685 	if (unlikely((!desc->rx.pkt_len) ||
2686 		     hnae3_get_bit(l234info, HNS3_RXD_TRUNCAT_B))) {
2687 		u64_stats_update_begin(&ring->syncp);
2688 		ring->stats.err_pkt_len++;
2689 		u64_stats_update_end(&ring->syncp);
2690 
2691 		dev_kfree_skb_any(skb);
2692 		return -EFAULT;
2693 	}
2694 
2695 	if (unlikely(hnae3_get_bit(l234info, HNS3_RXD_L2E_B))) {
2696 		u64_stats_update_begin(&ring->syncp);
2697 		ring->stats.l2_err++;
2698 		u64_stats_update_end(&ring->syncp);
2699 
2700 		dev_kfree_skb_any(skb);
2701 		return -EFAULT;
2702 	}
2703 
2704 	l2_frame_type = hnae3_get_field(l234info, HNS3_RXD_DMAC_M,
2705 					HNS3_RXD_DMAC_S);
2706 	u64_stats_update_begin(&ring->syncp);
2707 	if (l2_frame_type == HNS3_L2_TYPE_MULTICAST)
2708 		ring->stats.rx_multicast++;
2709 
2710 	ring->stats.rx_pkts++;
2711 	ring->stats.rx_bytes += skb->len;
2712 	u64_stats_update_end(&ring->syncp);
2713 
2714 	ring->tqp_vector->rx_group.total_bytes += skb->len;
2715 
2716 	/* This is needed in order to enable forwarding support */
2717 	hns3_set_gro_param(skb, l234info, bd_base_info);
2718 
2719 	hns3_rx_checksum(ring, skb, desc);
2720 	*out_skb = skb;
2721 	hns3_set_rx_skb_rss_type(ring, skb);
2722 
2723 	return 0;
2724 }
2725 
2726 int hns3_clean_rx_ring(
2727 		struct hns3_enet_ring *ring, int budget,
2728 		void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
2729 {
2730 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
2731 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2732 	int recv_pkts, recv_bds, clean_count, err;
2733 	int unused_count = hns3_desc_unused(ring) - ring->pending_buf;
2734 	struct sk_buff *skb = ring->skb;
2735 	int num;
2736 
2737 	num = readl_relaxed(ring->tqp->io_base + HNS3_RING_RX_RING_FBDNUM_REG);
2738 	rmb(); /* Make sure num taken effect before the other data is touched */
2739 
2740 	recv_pkts = 0, recv_bds = 0, clean_count = 0;
2741 	num -= unused_count;
2742 
2743 	while (recv_pkts < budget && recv_bds < num) {
2744 		/* Reuse or realloc buffers */
2745 		if (clean_count + unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
2746 			hns3_nic_alloc_rx_buffers(ring,
2747 						  clean_count + unused_count);
2748 			clean_count = 0;
2749 			unused_count = hns3_desc_unused(ring) -
2750 					ring->pending_buf;
2751 		}
2752 
2753 		/* Poll one pkt */
2754 		err = hns3_handle_rx_bd(ring, &skb);
2755 		if (unlikely(!skb)) /* This fault cannot be repaired */
2756 			goto out;
2757 
2758 		if (err == -ENXIO) { /* Do not get FE for the packet */
2759 			goto out;
2760 		} else if (unlikely(err)) {  /* Do jump the err */
2761 			recv_bds += ring->pending_buf;
2762 			clean_count += ring->pending_buf;
2763 			ring->skb = NULL;
2764 			ring->pending_buf = 0;
2765 			continue;
2766 		}
2767 
2768 		/* Do update ip stack process */
2769 		skb->protocol = eth_type_trans(skb, netdev);
2770 		rx_fn(ring, skb);
2771 		recv_bds += ring->pending_buf;
2772 		clean_count += ring->pending_buf;
2773 		ring->skb = NULL;
2774 		ring->pending_buf = 0;
2775 
2776 		recv_pkts++;
2777 	}
2778 
2779 out:
2780 	/* Make all data has been write before submit */
2781 	if (clean_count + unused_count > 0)
2782 		hns3_nic_alloc_rx_buffers(ring,
2783 					  clean_count + unused_count);
2784 
2785 	return recv_pkts;
2786 }
2787 
2788 static bool hns3_get_new_int_gl(struct hns3_enet_ring_group *ring_group)
2789 {
2790 	struct hns3_enet_tqp_vector *tqp_vector =
2791 					ring_group->ring->tqp_vector;
2792 	enum hns3_flow_level_range new_flow_level;
2793 	int packets_per_msecs;
2794 	int bytes_per_msecs;
2795 	u32 time_passed_ms;
2796 	u16 new_int_gl;
2797 
2798 	if (!ring_group->coal.int_gl || !tqp_vector->last_jiffies)
2799 		return false;
2800 
2801 	if (ring_group->total_packets == 0) {
2802 		ring_group->coal.int_gl = HNS3_INT_GL_50K;
2803 		ring_group->coal.flow_level = HNS3_FLOW_LOW;
2804 		return true;
2805 	}
2806 
2807 	/* Simple throttlerate management
2808 	 * 0-10MB/s   lower     (50000 ints/s)
2809 	 * 10-20MB/s   middle    (20000 ints/s)
2810 	 * 20-1249MB/s high      (18000 ints/s)
2811 	 * > 40000pps  ultra     (8000 ints/s)
2812 	 */
2813 	new_flow_level = ring_group->coal.flow_level;
2814 	new_int_gl = ring_group->coal.int_gl;
2815 	time_passed_ms =
2816 		jiffies_to_msecs(jiffies - tqp_vector->last_jiffies);
2817 
2818 	if (!time_passed_ms)
2819 		return false;
2820 
2821 	do_div(ring_group->total_packets, time_passed_ms);
2822 	packets_per_msecs = ring_group->total_packets;
2823 
2824 	do_div(ring_group->total_bytes, time_passed_ms);
2825 	bytes_per_msecs = ring_group->total_bytes;
2826 
2827 #define HNS3_RX_LOW_BYTE_RATE 10000
2828 #define HNS3_RX_MID_BYTE_RATE 20000
2829 
2830 	switch (new_flow_level) {
2831 	case HNS3_FLOW_LOW:
2832 		if (bytes_per_msecs > HNS3_RX_LOW_BYTE_RATE)
2833 			new_flow_level = HNS3_FLOW_MID;
2834 		break;
2835 	case HNS3_FLOW_MID:
2836 		if (bytes_per_msecs > HNS3_RX_MID_BYTE_RATE)
2837 			new_flow_level = HNS3_FLOW_HIGH;
2838 		else if (bytes_per_msecs <= HNS3_RX_LOW_BYTE_RATE)
2839 			new_flow_level = HNS3_FLOW_LOW;
2840 		break;
2841 	case HNS3_FLOW_HIGH:
2842 	case HNS3_FLOW_ULTRA:
2843 	default:
2844 		if (bytes_per_msecs <= HNS3_RX_MID_BYTE_RATE)
2845 			new_flow_level = HNS3_FLOW_MID;
2846 		break;
2847 	}
2848 
2849 #define HNS3_RX_ULTRA_PACKET_RATE 40
2850 
2851 	if (packets_per_msecs > HNS3_RX_ULTRA_PACKET_RATE &&
2852 	    &tqp_vector->rx_group == ring_group)
2853 		new_flow_level = HNS3_FLOW_ULTRA;
2854 
2855 	switch (new_flow_level) {
2856 	case HNS3_FLOW_LOW:
2857 		new_int_gl = HNS3_INT_GL_50K;
2858 		break;
2859 	case HNS3_FLOW_MID:
2860 		new_int_gl = HNS3_INT_GL_20K;
2861 		break;
2862 	case HNS3_FLOW_HIGH:
2863 		new_int_gl = HNS3_INT_GL_18K;
2864 		break;
2865 	case HNS3_FLOW_ULTRA:
2866 		new_int_gl = HNS3_INT_GL_8K;
2867 		break;
2868 	default:
2869 		break;
2870 	}
2871 
2872 	ring_group->total_bytes = 0;
2873 	ring_group->total_packets = 0;
2874 	ring_group->coal.flow_level = new_flow_level;
2875 	if (new_int_gl != ring_group->coal.int_gl) {
2876 		ring_group->coal.int_gl = new_int_gl;
2877 		return true;
2878 	}
2879 	return false;
2880 }
2881 
2882 static void hns3_update_new_int_gl(struct hns3_enet_tqp_vector *tqp_vector)
2883 {
2884 	struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
2885 	struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
2886 	bool rx_update, tx_update;
2887 
2888 	/* update param every 1000ms */
2889 	if (time_before(jiffies,
2890 			tqp_vector->last_jiffies + msecs_to_jiffies(1000)))
2891 		return;
2892 
2893 	if (rx_group->coal.gl_adapt_enable) {
2894 		rx_update = hns3_get_new_int_gl(rx_group);
2895 		if (rx_update)
2896 			hns3_set_vector_coalesce_rx_gl(tqp_vector,
2897 						       rx_group->coal.int_gl);
2898 	}
2899 
2900 	if (tx_group->coal.gl_adapt_enable) {
2901 		tx_update = hns3_get_new_int_gl(&tqp_vector->tx_group);
2902 		if (tx_update)
2903 			hns3_set_vector_coalesce_tx_gl(tqp_vector,
2904 						       tx_group->coal.int_gl);
2905 	}
2906 
2907 	tqp_vector->last_jiffies = jiffies;
2908 }
2909 
2910 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
2911 {
2912 	struct hns3_nic_priv *priv = netdev_priv(napi->dev);
2913 	struct hns3_enet_ring *ring;
2914 	int rx_pkt_total = 0;
2915 
2916 	struct hns3_enet_tqp_vector *tqp_vector =
2917 		container_of(napi, struct hns3_enet_tqp_vector, napi);
2918 	bool clean_complete = true;
2919 	int rx_budget;
2920 
2921 	if (unlikely(test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
2922 		napi_complete(napi);
2923 		return 0;
2924 	}
2925 
2926 	/* Since the actual Tx work is minimal, we can give the Tx a larger
2927 	 * budget and be more aggressive about cleaning up the Tx descriptors.
2928 	 */
2929 	hns3_for_each_ring(ring, tqp_vector->tx_group)
2930 		hns3_clean_tx_ring(ring);
2931 
2932 	/* make sure rx ring budget not smaller than 1 */
2933 	rx_budget = max(budget / tqp_vector->num_tqps, 1);
2934 
2935 	hns3_for_each_ring(ring, tqp_vector->rx_group) {
2936 		int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
2937 						    hns3_rx_skb);
2938 
2939 		if (rx_cleaned >= rx_budget)
2940 			clean_complete = false;
2941 
2942 		rx_pkt_total += rx_cleaned;
2943 	}
2944 
2945 	tqp_vector->rx_group.total_packets += rx_pkt_total;
2946 
2947 	if (!clean_complete)
2948 		return budget;
2949 
2950 	if (napi_complete(napi) &&
2951 	    likely(!test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
2952 		hns3_update_new_int_gl(tqp_vector);
2953 		hns3_mask_vector_irq(tqp_vector, 1);
2954 	}
2955 
2956 	return rx_pkt_total;
2957 }
2958 
2959 static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
2960 				      struct hnae3_ring_chain_node *head)
2961 {
2962 	struct pci_dev *pdev = tqp_vector->handle->pdev;
2963 	struct hnae3_ring_chain_node *cur_chain = head;
2964 	struct hnae3_ring_chain_node *chain;
2965 	struct hns3_enet_ring *tx_ring;
2966 	struct hns3_enet_ring *rx_ring;
2967 
2968 	tx_ring = tqp_vector->tx_group.ring;
2969 	if (tx_ring) {
2970 		cur_chain->tqp_index = tx_ring->tqp->tqp_index;
2971 		hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
2972 			      HNAE3_RING_TYPE_TX);
2973 		hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
2974 				HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_TX);
2975 
2976 		cur_chain->next = NULL;
2977 
2978 		while (tx_ring->next) {
2979 			tx_ring = tx_ring->next;
2980 
2981 			chain = devm_kzalloc(&pdev->dev, sizeof(*chain),
2982 					     GFP_KERNEL);
2983 			if (!chain)
2984 				goto err_free_chain;
2985 
2986 			cur_chain->next = chain;
2987 			chain->tqp_index = tx_ring->tqp->tqp_index;
2988 			hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
2989 				      HNAE3_RING_TYPE_TX);
2990 			hnae3_set_field(chain->int_gl_idx,
2991 					HNAE3_RING_GL_IDX_M,
2992 					HNAE3_RING_GL_IDX_S,
2993 					HNAE3_RING_GL_TX);
2994 
2995 			cur_chain = chain;
2996 		}
2997 	}
2998 
2999 	rx_ring = tqp_vector->rx_group.ring;
3000 	if (!tx_ring && rx_ring) {
3001 		cur_chain->next = NULL;
3002 		cur_chain->tqp_index = rx_ring->tqp->tqp_index;
3003 		hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
3004 			      HNAE3_RING_TYPE_RX);
3005 		hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3006 				HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
3007 
3008 		rx_ring = rx_ring->next;
3009 	}
3010 
3011 	while (rx_ring) {
3012 		chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
3013 		if (!chain)
3014 			goto err_free_chain;
3015 
3016 		cur_chain->next = chain;
3017 		chain->tqp_index = rx_ring->tqp->tqp_index;
3018 		hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
3019 			      HNAE3_RING_TYPE_RX);
3020 		hnae3_set_field(chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3021 				HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
3022 
3023 		cur_chain = chain;
3024 
3025 		rx_ring = rx_ring->next;
3026 	}
3027 
3028 	return 0;
3029 
3030 err_free_chain:
3031 	cur_chain = head->next;
3032 	while (cur_chain) {
3033 		chain = cur_chain->next;
3034 		devm_kfree(&pdev->dev, cur_chain);
3035 		cur_chain = chain;
3036 	}
3037 	head->next = NULL;
3038 
3039 	return -ENOMEM;
3040 }
3041 
3042 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
3043 					struct hnae3_ring_chain_node *head)
3044 {
3045 	struct pci_dev *pdev = tqp_vector->handle->pdev;
3046 	struct hnae3_ring_chain_node *chain_tmp, *chain;
3047 
3048 	chain = head->next;
3049 
3050 	while (chain) {
3051 		chain_tmp = chain->next;
3052 		devm_kfree(&pdev->dev, chain);
3053 		chain = chain_tmp;
3054 	}
3055 }
3056 
3057 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
3058 				   struct hns3_enet_ring *ring)
3059 {
3060 	ring->next = group->ring;
3061 	group->ring = ring;
3062 
3063 	group->count++;
3064 }
3065 
3066 static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
3067 {
3068 	struct pci_dev *pdev = priv->ae_handle->pdev;
3069 	struct hns3_enet_tqp_vector *tqp_vector;
3070 	int num_vectors = priv->vector_num;
3071 	int numa_node;
3072 	int vector_i;
3073 
3074 	numa_node = dev_to_node(&pdev->dev);
3075 
3076 	for (vector_i = 0; vector_i < num_vectors; vector_i++) {
3077 		tqp_vector = &priv->tqp_vector[vector_i];
3078 		cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
3079 				&tqp_vector->affinity_mask);
3080 	}
3081 }
3082 
3083 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
3084 {
3085 	struct hnae3_ring_chain_node vector_ring_chain;
3086 	struct hnae3_handle *h = priv->ae_handle;
3087 	struct hns3_enet_tqp_vector *tqp_vector;
3088 	int ret = 0;
3089 	int i;
3090 
3091 	hns3_nic_set_cpumask(priv);
3092 
3093 	for (i = 0; i < priv->vector_num; i++) {
3094 		tqp_vector = &priv->tqp_vector[i];
3095 		hns3_vector_gl_rl_init_hw(tqp_vector, priv);
3096 		tqp_vector->num_tqps = 0;
3097 	}
3098 
3099 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3100 		u16 vector_i = i % priv->vector_num;
3101 		u16 tqp_num = h->kinfo.num_tqps;
3102 
3103 		tqp_vector = &priv->tqp_vector[vector_i];
3104 
3105 		hns3_add_ring_to_group(&tqp_vector->tx_group,
3106 				       priv->ring_data[i].ring);
3107 
3108 		hns3_add_ring_to_group(&tqp_vector->rx_group,
3109 				       priv->ring_data[i + tqp_num].ring);
3110 
3111 		priv->ring_data[i].ring->tqp_vector = tqp_vector;
3112 		priv->ring_data[i + tqp_num].ring->tqp_vector = tqp_vector;
3113 		tqp_vector->num_tqps++;
3114 	}
3115 
3116 	for (i = 0; i < priv->vector_num; i++) {
3117 		tqp_vector = &priv->tqp_vector[i];
3118 
3119 		tqp_vector->rx_group.total_bytes = 0;
3120 		tqp_vector->rx_group.total_packets = 0;
3121 		tqp_vector->tx_group.total_bytes = 0;
3122 		tqp_vector->tx_group.total_packets = 0;
3123 		tqp_vector->handle = h;
3124 
3125 		ret = hns3_get_vector_ring_chain(tqp_vector,
3126 						 &vector_ring_chain);
3127 		if (ret)
3128 			goto map_ring_fail;
3129 
3130 		ret = h->ae_algo->ops->map_ring_to_vector(h,
3131 			tqp_vector->vector_irq, &vector_ring_chain);
3132 
3133 		hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3134 
3135 		if (ret)
3136 			goto map_ring_fail;
3137 
3138 		netif_napi_add(priv->netdev, &tqp_vector->napi,
3139 			       hns3_nic_common_poll, NAPI_POLL_WEIGHT);
3140 	}
3141 
3142 	return 0;
3143 
3144 map_ring_fail:
3145 	while (i--)
3146 		netif_napi_del(&priv->tqp_vector[i].napi);
3147 
3148 	return ret;
3149 }
3150 
3151 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
3152 {
3153 #define HNS3_VECTOR_PF_MAX_NUM		64
3154 
3155 	struct hnae3_handle *h = priv->ae_handle;
3156 	struct hns3_enet_tqp_vector *tqp_vector;
3157 	struct hnae3_vector_info *vector;
3158 	struct pci_dev *pdev = h->pdev;
3159 	u16 tqp_num = h->kinfo.num_tqps;
3160 	u16 vector_num;
3161 	int ret = 0;
3162 	u16 i;
3163 
3164 	/* RSS size, cpu online and vector_num should be the same */
3165 	/* Should consider 2p/4p later */
3166 	vector_num = min_t(u16, num_online_cpus(), tqp_num);
3167 	vector_num = min_t(u16, vector_num, HNS3_VECTOR_PF_MAX_NUM);
3168 
3169 	vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
3170 			      GFP_KERNEL);
3171 	if (!vector)
3172 		return -ENOMEM;
3173 
3174 	vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
3175 
3176 	priv->vector_num = vector_num;
3177 	priv->tqp_vector = (struct hns3_enet_tqp_vector *)
3178 		devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
3179 			     GFP_KERNEL);
3180 	if (!priv->tqp_vector) {
3181 		ret = -ENOMEM;
3182 		goto out;
3183 	}
3184 
3185 	for (i = 0; i < priv->vector_num; i++) {
3186 		tqp_vector = &priv->tqp_vector[i];
3187 		tqp_vector->idx = i;
3188 		tqp_vector->mask_addr = vector[i].io_addr;
3189 		tqp_vector->vector_irq = vector[i].vector;
3190 		hns3_vector_gl_rl_init(tqp_vector, priv);
3191 	}
3192 
3193 out:
3194 	devm_kfree(&pdev->dev, vector);
3195 	return ret;
3196 }
3197 
3198 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
3199 {
3200 	group->ring = NULL;
3201 	group->count = 0;
3202 }
3203 
3204 static int hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
3205 {
3206 	struct hnae3_ring_chain_node vector_ring_chain;
3207 	struct hnae3_handle *h = priv->ae_handle;
3208 	struct hns3_enet_tqp_vector *tqp_vector;
3209 	int i, ret;
3210 
3211 	for (i = 0; i < priv->vector_num; i++) {
3212 		tqp_vector = &priv->tqp_vector[i];
3213 
3214 		if (!tqp_vector->rx_group.ring && !tqp_vector->tx_group.ring)
3215 			continue;
3216 
3217 		ret = hns3_get_vector_ring_chain(tqp_vector,
3218 						 &vector_ring_chain);
3219 		if (ret)
3220 			return ret;
3221 
3222 		ret = h->ae_algo->ops->unmap_ring_from_vector(h,
3223 			tqp_vector->vector_irq, &vector_ring_chain);
3224 		if (ret)
3225 			return ret;
3226 
3227 		hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3228 
3229 		if (tqp_vector->irq_init_flag == HNS3_VECTOR_INITED) {
3230 			irq_set_affinity_notifier(tqp_vector->vector_irq,
3231 						  NULL);
3232 			irq_set_affinity_hint(tqp_vector->vector_irq, NULL);
3233 			free_irq(tqp_vector->vector_irq, tqp_vector);
3234 			tqp_vector->irq_init_flag = HNS3_VECTOR_NOT_INITED;
3235 		}
3236 
3237 		hns3_clear_ring_group(&tqp_vector->rx_group);
3238 		hns3_clear_ring_group(&tqp_vector->tx_group);
3239 		netif_napi_del(&priv->tqp_vector[i].napi);
3240 	}
3241 
3242 	return 0;
3243 }
3244 
3245 static int hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
3246 {
3247 	struct hnae3_handle *h = priv->ae_handle;
3248 	struct pci_dev *pdev = h->pdev;
3249 	int i, ret;
3250 
3251 	for (i = 0; i < priv->vector_num; i++) {
3252 		struct hns3_enet_tqp_vector *tqp_vector;
3253 
3254 		tqp_vector = &priv->tqp_vector[i];
3255 		ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
3256 		if (ret)
3257 			return ret;
3258 	}
3259 
3260 	devm_kfree(&pdev->dev, priv->tqp_vector);
3261 	return 0;
3262 }
3263 
3264 static int hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
3265 			     int ring_type)
3266 {
3267 	struct hns3_nic_ring_data *ring_data = priv->ring_data;
3268 	int queue_num = priv->ae_handle->kinfo.num_tqps;
3269 	int desc_num = priv->ae_handle->kinfo.num_desc;
3270 	struct pci_dev *pdev = priv->ae_handle->pdev;
3271 	struct hns3_enet_ring *ring;
3272 
3273 	ring = devm_kzalloc(&pdev->dev, sizeof(*ring), GFP_KERNEL);
3274 	if (!ring)
3275 		return -ENOMEM;
3276 
3277 	if (ring_type == HNAE3_RING_TYPE_TX) {
3278 		ring_data[q->tqp_index].ring = ring;
3279 		ring_data[q->tqp_index].queue_index = q->tqp_index;
3280 		ring->io_base = (u8 __iomem *)q->io_base + HNS3_TX_REG_OFFSET;
3281 	} else {
3282 		ring_data[q->tqp_index + queue_num].ring = ring;
3283 		ring_data[q->tqp_index + queue_num].queue_index = q->tqp_index;
3284 		ring->io_base = q->io_base;
3285 	}
3286 
3287 	hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
3288 
3289 	ring->tqp = q;
3290 	ring->desc = NULL;
3291 	ring->desc_cb = NULL;
3292 	ring->dev = priv->dev;
3293 	ring->desc_dma_addr = 0;
3294 	ring->buf_size = q->buf_size;
3295 	ring->desc_num = desc_num;
3296 	ring->next_to_use = 0;
3297 	ring->next_to_clean = 0;
3298 
3299 	return 0;
3300 }
3301 
3302 static int hns3_queue_to_ring(struct hnae3_queue *tqp,
3303 			      struct hns3_nic_priv *priv)
3304 {
3305 	int ret;
3306 
3307 	ret = hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
3308 	if (ret)
3309 		return ret;
3310 
3311 	ret = hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
3312 	if (ret) {
3313 		devm_kfree(priv->dev, priv->ring_data[tqp->tqp_index].ring);
3314 		return ret;
3315 	}
3316 
3317 	return 0;
3318 }
3319 
3320 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
3321 {
3322 	struct hnae3_handle *h = priv->ae_handle;
3323 	struct pci_dev *pdev = h->pdev;
3324 	int i, ret;
3325 
3326 	priv->ring_data =  devm_kzalloc(&pdev->dev,
3327 					array3_size(h->kinfo.num_tqps,
3328 						    sizeof(*priv->ring_data),
3329 						    2),
3330 					GFP_KERNEL);
3331 	if (!priv->ring_data)
3332 		return -ENOMEM;
3333 
3334 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3335 		ret = hns3_queue_to_ring(h->kinfo.tqp[i], priv);
3336 		if (ret)
3337 			goto err;
3338 	}
3339 
3340 	return 0;
3341 err:
3342 	while (i--) {
3343 		devm_kfree(priv->dev, priv->ring_data[i].ring);
3344 		devm_kfree(priv->dev,
3345 			   priv->ring_data[i + h->kinfo.num_tqps].ring);
3346 	}
3347 
3348 	devm_kfree(&pdev->dev, priv->ring_data);
3349 	return ret;
3350 }
3351 
3352 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
3353 {
3354 	struct hnae3_handle *h = priv->ae_handle;
3355 	int i;
3356 
3357 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3358 		devm_kfree(priv->dev, priv->ring_data[i].ring);
3359 		devm_kfree(priv->dev,
3360 			   priv->ring_data[i + h->kinfo.num_tqps].ring);
3361 	}
3362 	devm_kfree(priv->dev, priv->ring_data);
3363 }
3364 
3365 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
3366 {
3367 	int ret;
3368 
3369 	if (ring->desc_num <= 0 || ring->buf_size <= 0)
3370 		return -EINVAL;
3371 
3372 	ring->desc_cb = kcalloc(ring->desc_num, sizeof(ring->desc_cb[0]),
3373 				GFP_KERNEL);
3374 	if (!ring->desc_cb) {
3375 		ret = -ENOMEM;
3376 		goto out;
3377 	}
3378 
3379 	ret = hns3_alloc_desc(ring);
3380 	if (ret)
3381 		goto out_with_desc_cb;
3382 
3383 	if (!HNAE3_IS_TX_RING(ring)) {
3384 		ret = hns3_alloc_ring_buffers(ring);
3385 		if (ret)
3386 			goto out_with_desc;
3387 	}
3388 
3389 	return 0;
3390 
3391 out_with_desc:
3392 	hns3_free_desc(ring);
3393 out_with_desc_cb:
3394 	kfree(ring->desc_cb);
3395 	ring->desc_cb = NULL;
3396 out:
3397 	return ret;
3398 }
3399 
3400 static void hns3_fini_ring(struct hns3_enet_ring *ring)
3401 {
3402 	hns3_free_desc(ring);
3403 	kfree(ring->desc_cb);
3404 	ring->desc_cb = NULL;
3405 	ring->next_to_clean = 0;
3406 	ring->next_to_use = 0;
3407 	ring->pending_buf = 0;
3408 	if (ring->skb) {
3409 		dev_kfree_skb_any(ring->skb);
3410 		ring->skb = NULL;
3411 	}
3412 }
3413 
3414 static int hns3_buf_size2type(u32 buf_size)
3415 {
3416 	int bd_size_type;
3417 
3418 	switch (buf_size) {
3419 	case 512:
3420 		bd_size_type = HNS3_BD_SIZE_512_TYPE;
3421 		break;
3422 	case 1024:
3423 		bd_size_type = HNS3_BD_SIZE_1024_TYPE;
3424 		break;
3425 	case 2048:
3426 		bd_size_type = HNS3_BD_SIZE_2048_TYPE;
3427 		break;
3428 	case 4096:
3429 		bd_size_type = HNS3_BD_SIZE_4096_TYPE;
3430 		break;
3431 	default:
3432 		bd_size_type = HNS3_BD_SIZE_2048_TYPE;
3433 	}
3434 
3435 	return bd_size_type;
3436 }
3437 
3438 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
3439 {
3440 	dma_addr_t dma = ring->desc_dma_addr;
3441 	struct hnae3_queue *q = ring->tqp;
3442 
3443 	if (!HNAE3_IS_TX_RING(ring)) {
3444 		hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG,
3445 			       (u32)dma);
3446 		hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
3447 			       (u32)((dma >> 31) >> 1));
3448 
3449 		hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
3450 			       hns3_buf_size2type(ring->buf_size));
3451 		hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
3452 			       ring->desc_num / 8 - 1);
3453 
3454 	} else {
3455 		hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
3456 			       (u32)dma);
3457 		hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
3458 			       (u32)((dma >> 31) >> 1));
3459 
3460 		hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
3461 			       ring->desc_num / 8 - 1);
3462 	}
3463 }
3464 
3465 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
3466 {
3467 	struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
3468 	int i;
3469 
3470 	for (i = 0; i < HNAE3_MAX_TC; i++) {
3471 		struct hnae3_tc_info *tc_info = &kinfo->tc_info[i];
3472 		int j;
3473 
3474 		if (!tc_info->enable)
3475 			continue;
3476 
3477 		for (j = 0; j < tc_info->tqp_count; j++) {
3478 			struct hnae3_queue *q;
3479 
3480 			q = priv->ring_data[tc_info->tqp_offset + j].ring->tqp;
3481 			hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG,
3482 				       tc_info->tc);
3483 		}
3484 	}
3485 }
3486 
3487 int hns3_init_all_ring(struct hns3_nic_priv *priv)
3488 {
3489 	struct hnae3_handle *h = priv->ae_handle;
3490 	int ring_num = h->kinfo.num_tqps * 2;
3491 	int i, j;
3492 	int ret;
3493 
3494 	for (i = 0; i < ring_num; i++) {
3495 		ret = hns3_alloc_ring_memory(priv->ring_data[i].ring);
3496 		if (ret) {
3497 			dev_err(priv->dev,
3498 				"Alloc ring memory fail! ret=%d\n", ret);
3499 			goto out_when_alloc_ring_memory;
3500 		}
3501 
3502 		u64_stats_init(&priv->ring_data[i].ring->syncp);
3503 	}
3504 
3505 	return 0;
3506 
3507 out_when_alloc_ring_memory:
3508 	for (j = i - 1; j >= 0; j--)
3509 		hns3_fini_ring(priv->ring_data[j].ring);
3510 
3511 	return -ENOMEM;
3512 }
3513 
3514 int hns3_uninit_all_ring(struct hns3_nic_priv *priv)
3515 {
3516 	struct hnae3_handle *h = priv->ae_handle;
3517 	int i;
3518 
3519 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3520 		hns3_fini_ring(priv->ring_data[i].ring);
3521 		hns3_fini_ring(priv->ring_data[i + h->kinfo.num_tqps].ring);
3522 	}
3523 	return 0;
3524 }
3525 
3526 /* Set mac addr if it is configured. or leave it to the AE driver */
3527 static int hns3_init_mac_addr(struct net_device *netdev, bool init)
3528 {
3529 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3530 	struct hnae3_handle *h = priv->ae_handle;
3531 	u8 mac_addr_temp[ETH_ALEN];
3532 	int ret = 0;
3533 
3534 	if (h->ae_algo->ops->get_mac_addr && init) {
3535 		h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
3536 		ether_addr_copy(netdev->dev_addr, mac_addr_temp);
3537 	}
3538 
3539 	/* Check if the MAC address is valid, if not get a random one */
3540 	if (!is_valid_ether_addr(netdev->dev_addr)) {
3541 		eth_hw_addr_random(netdev);
3542 		dev_warn(priv->dev, "using random MAC address %pM\n",
3543 			 netdev->dev_addr);
3544 	}
3545 
3546 	if (h->ae_algo->ops->set_mac_addr)
3547 		ret = h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
3548 
3549 	return ret;
3550 }
3551 
3552 static int hns3_restore_fd_rules(struct net_device *netdev)
3553 {
3554 	struct hnae3_handle *h = hns3_get_handle(netdev);
3555 	int ret = 0;
3556 
3557 	if (h->ae_algo->ops->restore_fd_rules)
3558 		ret = h->ae_algo->ops->restore_fd_rules(h);
3559 
3560 	return ret;
3561 }
3562 
3563 static void hns3_del_all_fd_rules(struct net_device *netdev, bool clear_list)
3564 {
3565 	struct hnae3_handle *h = hns3_get_handle(netdev);
3566 
3567 	if (h->ae_algo->ops->del_all_fd_entries)
3568 		h->ae_algo->ops->del_all_fd_entries(h, clear_list);
3569 }
3570 
3571 static void hns3_nic_set_priv_ops(struct net_device *netdev)
3572 {
3573 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3574 
3575 	priv->ops.fill_desc = hns3_fill_desc;
3576 	if ((netdev->features & NETIF_F_TSO) ||
3577 	    (netdev->features & NETIF_F_TSO6))
3578 		priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tso;
3579 	else
3580 		priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tx;
3581 }
3582 
3583 static int hns3_client_start(struct hnae3_handle *handle)
3584 {
3585 	if (!handle->ae_algo->ops->client_start)
3586 		return 0;
3587 
3588 	return handle->ae_algo->ops->client_start(handle);
3589 }
3590 
3591 static void hns3_client_stop(struct hnae3_handle *handle)
3592 {
3593 	if (!handle->ae_algo->ops->client_stop)
3594 		return;
3595 
3596 	handle->ae_algo->ops->client_stop(handle);
3597 }
3598 
3599 static int hns3_client_init(struct hnae3_handle *handle)
3600 {
3601 	struct pci_dev *pdev = handle->pdev;
3602 	u16 alloc_tqps, max_rss_size;
3603 	struct hns3_nic_priv *priv;
3604 	struct net_device *netdev;
3605 	int ret;
3606 
3607 	handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
3608 						    &max_rss_size);
3609 	netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
3610 	if (!netdev)
3611 		return -ENOMEM;
3612 
3613 	priv = netdev_priv(netdev);
3614 	priv->dev = &pdev->dev;
3615 	priv->netdev = netdev;
3616 	priv->ae_handle = handle;
3617 	priv->tx_timeout_count = 0;
3618 	set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
3619 
3620 	handle->kinfo.netdev = netdev;
3621 	handle->priv = (void *)priv;
3622 
3623 	hns3_init_mac_addr(netdev, true);
3624 
3625 	hns3_set_default_feature(netdev);
3626 
3627 	netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
3628 	netdev->priv_flags |= IFF_UNICAST_FLT;
3629 	netdev->netdev_ops = &hns3_nic_netdev_ops;
3630 	SET_NETDEV_DEV(netdev, &pdev->dev);
3631 	hns3_ethtool_set_ops(netdev);
3632 	hns3_nic_set_priv_ops(netdev);
3633 
3634 	/* Carrier off reporting is important to ethtool even BEFORE open */
3635 	netif_carrier_off(netdev);
3636 
3637 	ret = hns3_get_ring_config(priv);
3638 	if (ret) {
3639 		ret = -ENOMEM;
3640 		goto out_get_ring_cfg;
3641 	}
3642 
3643 	ret = hns3_nic_alloc_vector_data(priv);
3644 	if (ret) {
3645 		ret = -ENOMEM;
3646 		goto out_alloc_vector_data;
3647 	}
3648 
3649 	ret = hns3_nic_init_vector_data(priv);
3650 	if (ret) {
3651 		ret = -ENOMEM;
3652 		goto out_init_vector_data;
3653 	}
3654 
3655 	ret = hns3_init_all_ring(priv);
3656 	if (ret) {
3657 		ret = -ENOMEM;
3658 		goto out_init_ring_data;
3659 	}
3660 
3661 	ret = register_netdev(netdev);
3662 	if (ret) {
3663 		dev_err(priv->dev, "probe register netdev fail!\n");
3664 		goto out_reg_netdev_fail;
3665 	}
3666 
3667 	ret = hns3_client_start(handle);
3668 	if (ret) {
3669 		dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
3670 			goto out_reg_netdev_fail;
3671 	}
3672 
3673 	hns3_dcbnl_setup(handle);
3674 
3675 	hns3_dbg_init(handle);
3676 
3677 	/* MTU range: (ETH_MIN_MTU(kernel default) - 9702) */
3678 	netdev->max_mtu = HNS3_MAX_MTU;
3679 
3680 	set_bit(HNS3_NIC_STATE_INITED, &priv->state);
3681 
3682 	return ret;
3683 
3684 out_reg_netdev_fail:
3685 out_init_ring_data:
3686 	(void)hns3_nic_uninit_vector_data(priv);
3687 out_init_vector_data:
3688 	hns3_nic_dealloc_vector_data(priv);
3689 out_alloc_vector_data:
3690 	priv->ring_data = NULL;
3691 out_get_ring_cfg:
3692 	priv->ae_handle = NULL;
3693 	free_netdev(netdev);
3694 	return ret;
3695 }
3696 
3697 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
3698 {
3699 	struct net_device *netdev = handle->kinfo.netdev;
3700 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3701 	int ret;
3702 
3703 	hns3_client_stop(handle);
3704 
3705 	hns3_remove_hw_addr(netdev);
3706 
3707 	if (netdev->reg_state != NETREG_UNINITIALIZED)
3708 		unregister_netdev(netdev);
3709 
3710 	if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
3711 		netdev_warn(netdev, "already uninitialized\n");
3712 		goto out_netdev_free;
3713 	}
3714 
3715 	hns3_del_all_fd_rules(netdev, true);
3716 
3717 	hns3_force_clear_all_rx_ring(handle);
3718 
3719 	ret = hns3_nic_uninit_vector_data(priv);
3720 	if (ret)
3721 		netdev_err(netdev, "uninit vector error\n");
3722 
3723 	ret = hns3_nic_dealloc_vector_data(priv);
3724 	if (ret)
3725 		netdev_err(netdev, "dealloc vector error\n");
3726 
3727 	ret = hns3_uninit_all_ring(priv);
3728 	if (ret)
3729 		netdev_err(netdev, "uninit ring error\n");
3730 
3731 	hns3_put_ring_config(priv);
3732 
3733 	hns3_dbg_uninit(handle);
3734 
3735 	priv->ring_data = NULL;
3736 
3737 out_netdev_free:
3738 	free_netdev(netdev);
3739 }
3740 
3741 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
3742 {
3743 	struct net_device *netdev = handle->kinfo.netdev;
3744 
3745 	if (!netdev)
3746 		return;
3747 
3748 	if (linkup) {
3749 		netif_carrier_on(netdev);
3750 		netif_tx_wake_all_queues(netdev);
3751 		netdev_info(netdev, "link up\n");
3752 	} else {
3753 		netif_carrier_off(netdev);
3754 		netif_tx_stop_all_queues(netdev);
3755 		netdev_info(netdev, "link down\n");
3756 	}
3757 }
3758 
3759 static int hns3_client_setup_tc(struct hnae3_handle *handle, u8 tc)
3760 {
3761 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
3762 	struct net_device *ndev = kinfo->netdev;
3763 
3764 	if (tc > HNAE3_MAX_TC)
3765 		return -EINVAL;
3766 
3767 	if (!ndev)
3768 		return -ENODEV;
3769 
3770 	return hns3_nic_set_real_num_queue(ndev);
3771 }
3772 
3773 static int hns3_recover_hw_addr(struct net_device *ndev)
3774 {
3775 	struct netdev_hw_addr_list *list;
3776 	struct netdev_hw_addr *ha, *tmp;
3777 	int ret = 0;
3778 
3779 	/* go through and sync uc_addr entries to the device */
3780 	list = &ndev->uc;
3781 	list_for_each_entry_safe(ha, tmp, &list->list, list) {
3782 		ret = hns3_nic_uc_sync(ndev, ha->addr);
3783 		if (ret)
3784 			return ret;
3785 	}
3786 
3787 	/* go through and sync mc_addr entries to the device */
3788 	list = &ndev->mc;
3789 	list_for_each_entry_safe(ha, tmp, &list->list, list) {
3790 		ret = hns3_nic_mc_sync(ndev, ha->addr);
3791 		if (ret)
3792 			return ret;
3793 	}
3794 
3795 	return ret;
3796 }
3797 
3798 static void hns3_remove_hw_addr(struct net_device *netdev)
3799 {
3800 	struct netdev_hw_addr_list *list;
3801 	struct netdev_hw_addr *ha, *tmp;
3802 
3803 	hns3_nic_uc_unsync(netdev, netdev->dev_addr);
3804 
3805 	/* go through and unsync uc_addr entries to the device */
3806 	list = &netdev->uc;
3807 	list_for_each_entry_safe(ha, tmp, &list->list, list)
3808 		hns3_nic_uc_unsync(netdev, ha->addr);
3809 
3810 	/* go through and unsync mc_addr entries to the device */
3811 	list = &netdev->mc;
3812 	list_for_each_entry_safe(ha, tmp, &list->list, list)
3813 		if (ha->refcount > 1)
3814 			hns3_nic_mc_unsync(netdev, ha->addr);
3815 }
3816 
3817 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
3818 {
3819 	while (ring->next_to_clean != ring->next_to_use) {
3820 		ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
3821 		hns3_free_buffer_detach(ring, ring->next_to_clean);
3822 		ring_ptr_move_fw(ring, next_to_clean);
3823 	}
3824 }
3825 
3826 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
3827 {
3828 	struct hns3_desc_cb res_cbs;
3829 	int ret;
3830 
3831 	while (ring->next_to_use != ring->next_to_clean) {
3832 		/* When a buffer is not reused, it's memory has been
3833 		 * freed in hns3_handle_rx_bd or will be freed by
3834 		 * stack, so we need to replace the buffer here.
3835 		 */
3836 		if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
3837 			ret = hns3_reserve_buffer_map(ring, &res_cbs);
3838 			if (ret) {
3839 				u64_stats_update_begin(&ring->syncp);
3840 				ring->stats.sw_err_cnt++;
3841 				u64_stats_update_end(&ring->syncp);
3842 				/* if alloc new buffer fail, exit directly
3843 				 * and reclear in up flow.
3844 				 */
3845 				netdev_warn(ring->tqp->handle->kinfo.netdev,
3846 					    "reserve buffer map failed, ret = %d\n",
3847 					    ret);
3848 				return ret;
3849 			}
3850 			hns3_replace_buffer(ring, ring->next_to_use,
3851 					    &res_cbs);
3852 		}
3853 		ring_ptr_move_fw(ring, next_to_use);
3854 	}
3855 
3856 	return 0;
3857 }
3858 
3859 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
3860 {
3861 	while (ring->next_to_use != ring->next_to_clean) {
3862 		/* When a buffer is not reused, it's memory has been
3863 		 * freed in hns3_handle_rx_bd or will be freed by
3864 		 * stack, so only need to unmap the buffer here.
3865 		 */
3866 		if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
3867 			hns3_unmap_buffer(ring,
3868 					  &ring->desc_cb[ring->next_to_use]);
3869 			ring->desc_cb[ring->next_to_use].dma = 0;
3870 		}
3871 
3872 		ring_ptr_move_fw(ring, next_to_use);
3873 	}
3874 }
3875 
3876 static void hns3_force_clear_all_rx_ring(struct hnae3_handle *h)
3877 {
3878 	struct net_device *ndev = h->kinfo.netdev;
3879 	struct hns3_nic_priv *priv = netdev_priv(ndev);
3880 	struct hns3_enet_ring *ring;
3881 	u32 i;
3882 
3883 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3884 		ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
3885 		hns3_force_clear_rx_ring(ring);
3886 	}
3887 }
3888 
3889 static void hns3_clear_all_ring(struct hnae3_handle *h)
3890 {
3891 	struct net_device *ndev = h->kinfo.netdev;
3892 	struct hns3_nic_priv *priv = netdev_priv(ndev);
3893 	u32 i;
3894 
3895 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3896 		struct netdev_queue *dev_queue;
3897 		struct hns3_enet_ring *ring;
3898 
3899 		ring = priv->ring_data[i].ring;
3900 		hns3_clear_tx_ring(ring);
3901 		dev_queue = netdev_get_tx_queue(ndev,
3902 						priv->ring_data[i].queue_index);
3903 		netdev_tx_reset_queue(dev_queue);
3904 
3905 		ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
3906 		/* Continue to clear other rings even if clearing some
3907 		 * rings failed.
3908 		 */
3909 		hns3_clear_rx_ring(ring);
3910 	}
3911 }
3912 
3913 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
3914 {
3915 	struct net_device *ndev = h->kinfo.netdev;
3916 	struct hns3_nic_priv *priv = netdev_priv(ndev);
3917 	struct hns3_enet_ring *rx_ring;
3918 	int i, j;
3919 	int ret;
3920 
3921 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3922 		ret = h->ae_algo->ops->reset_queue(h, i);
3923 		if (ret)
3924 			return ret;
3925 
3926 		hns3_init_ring_hw(priv->ring_data[i].ring);
3927 
3928 		/* We need to clear tx ring here because self test will
3929 		 * use the ring and will not run down before up
3930 		 */
3931 		hns3_clear_tx_ring(priv->ring_data[i].ring);
3932 		priv->ring_data[i].ring->next_to_clean = 0;
3933 		priv->ring_data[i].ring->next_to_use = 0;
3934 
3935 		rx_ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
3936 		hns3_init_ring_hw(rx_ring);
3937 		ret = hns3_clear_rx_ring(rx_ring);
3938 		if (ret)
3939 			return ret;
3940 
3941 		/* We can not know the hardware head and tail when this
3942 		 * function is called in reset flow, so we reuse all desc.
3943 		 */
3944 		for (j = 0; j < rx_ring->desc_num; j++)
3945 			hns3_reuse_buffer(rx_ring, j);
3946 
3947 		rx_ring->next_to_clean = 0;
3948 		rx_ring->next_to_use = 0;
3949 	}
3950 
3951 	hns3_init_tx_ring_tc(priv);
3952 
3953 	return 0;
3954 }
3955 
3956 static void hns3_store_coal(struct hns3_nic_priv *priv)
3957 {
3958 	/* ethtool only support setting and querying one coal
3959 	 * configuation for now, so save the vector 0' coal
3960 	 * configuation here in order to restore it.
3961 	 */
3962 	memcpy(&priv->tx_coal, &priv->tqp_vector[0].tx_group.coal,
3963 	       sizeof(struct hns3_enet_coalesce));
3964 	memcpy(&priv->rx_coal, &priv->tqp_vector[0].rx_group.coal,
3965 	       sizeof(struct hns3_enet_coalesce));
3966 }
3967 
3968 static void hns3_restore_coal(struct hns3_nic_priv *priv)
3969 {
3970 	u16 vector_num = priv->vector_num;
3971 	int i;
3972 
3973 	for (i = 0; i < vector_num; i++) {
3974 		memcpy(&priv->tqp_vector[i].tx_group.coal, &priv->tx_coal,
3975 		       sizeof(struct hns3_enet_coalesce));
3976 		memcpy(&priv->tqp_vector[i].rx_group.coal, &priv->rx_coal,
3977 		       sizeof(struct hns3_enet_coalesce));
3978 	}
3979 }
3980 
3981 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
3982 {
3983 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(handle->pdev);
3984 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
3985 	struct net_device *ndev = kinfo->netdev;
3986 	struct hns3_nic_priv *priv = netdev_priv(ndev);
3987 
3988 	if (test_and_set_bit(HNS3_NIC_STATE_RESETTING, &priv->state))
3989 		return 0;
3990 
3991 	/* it is cumbersome for hardware to pick-and-choose entries for deletion
3992 	 * from table space. Hence, for function reset software intervention is
3993 	 * required to delete the entries
3994 	 */
3995 	if (hns3_dev_ongoing_func_reset(ae_dev)) {
3996 		hns3_remove_hw_addr(ndev);
3997 		hns3_del_all_fd_rules(ndev, false);
3998 	}
3999 
4000 	if (!netif_running(ndev))
4001 		return 0;
4002 
4003 	return hns3_nic_net_stop(ndev);
4004 }
4005 
4006 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
4007 {
4008 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4009 	struct hns3_nic_priv *priv = netdev_priv(kinfo->netdev);
4010 	int ret = 0;
4011 
4012 	clear_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4013 
4014 	if (netif_running(kinfo->netdev)) {
4015 		ret = hns3_nic_net_open(kinfo->netdev);
4016 		if (ret) {
4017 			set_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4018 			netdev_err(kinfo->netdev,
4019 				   "hns net up fail, ret=%d!\n", ret);
4020 			return ret;
4021 		}
4022 	}
4023 
4024 	return ret;
4025 }
4026 
4027 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
4028 {
4029 	struct net_device *netdev = handle->kinfo.netdev;
4030 	struct hns3_nic_priv *priv = netdev_priv(netdev);
4031 	int ret;
4032 
4033 	/* Carrier off reporting is important to ethtool even BEFORE open */
4034 	netif_carrier_off(netdev);
4035 
4036 	ret = hns3_get_ring_config(priv);
4037 	if (ret)
4038 		return ret;
4039 
4040 	ret = hns3_nic_alloc_vector_data(priv);
4041 	if (ret)
4042 		goto err_put_ring;
4043 
4044 	hns3_restore_coal(priv);
4045 
4046 	ret = hns3_nic_init_vector_data(priv);
4047 	if (ret)
4048 		goto err_dealloc_vector;
4049 
4050 	ret = hns3_init_all_ring(priv);
4051 	if (ret)
4052 		goto err_uninit_vector;
4053 
4054 	set_bit(HNS3_NIC_STATE_INITED, &priv->state);
4055 
4056 	return ret;
4057 
4058 err_uninit_vector:
4059 	hns3_nic_uninit_vector_data(priv);
4060 	priv->ring_data = NULL;
4061 err_dealloc_vector:
4062 	hns3_nic_dealloc_vector_data(priv);
4063 err_put_ring:
4064 	hns3_put_ring_config(priv);
4065 	priv->ring_data = NULL;
4066 
4067 	return ret;
4068 }
4069 
4070 static int hns3_reset_notify_restore_enet(struct hnae3_handle *handle)
4071 {
4072 	struct net_device *netdev = handle->kinfo.netdev;
4073 	bool vlan_filter_enable;
4074 	int ret;
4075 
4076 	ret = hns3_init_mac_addr(netdev, false);
4077 	if (ret)
4078 		return ret;
4079 
4080 	ret = hns3_recover_hw_addr(netdev);
4081 	if (ret)
4082 		return ret;
4083 
4084 	ret = hns3_update_promisc_mode(netdev, handle->netdev_flags);
4085 	if (ret)
4086 		return ret;
4087 
4088 	vlan_filter_enable = netdev->flags & IFF_PROMISC ? false : true;
4089 	hns3_enable_vlan_filter(netdev, vlan_filter_enable);
4090 
4091 	/* Hardware table is only clear when pf resets */
4092 	if (!(handle->flags & HNAE3_SUPPORT_VF)) {
4093 		ret = hns3_restore_vlan(netdev);
4094 		if (ret)
4095 			return ret;
4096 	}
4097 
4098 	return hns3_restore_fd_rules(netdev);
4099 }
4100 
4101 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
4102 {
4103 	struct net_device *netdev = handle->kinfo.netdev;
4104 	struct hns3_nic_priv *priv = netdev_priv(netdev);
4105 	int ret;
4106 
4107 	if (!test_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
4108 		netdev_warn(netdev, "already uninitialized\n");
4109 		return 0;
4110 	}
4111 
4112 	hns3_force_clear_all_rx_ring(handle);
4113 
4114 	ret = hns3_nic_uninit_vector_data(priv);
4115 	if (ret) {
4116 		netdev_err(netdev, "uninit vector error\n");
4117 		return ret;
4118 	}
4119 
4120 	hns3_store_coal(priv);
4121 
4122 	ret = hns3_nic_dealloc_vector_data(priv);
4123 	if (ret)
4124 		netdev_err(netdev, "dealloc vector error\n");
4125 
4126 	ret = hns3_uninit_all_ring(priv);
4127 	if (ret)
4128 		netdev_err(netdev, "uninit ring error\n");
4129 
4130 	hns3_put_ring_config(priv);
4131 	priv->ring_data = NULL;
4132 
4133 	clear_bit(HNS3_NIC_STATE_INITED, &priv->state);
4134 
4135 	return ret;
4136 }
4137 
4138 static int hns3_reset_notify(struct hnae3_handle *handle,
4139 			     enum hnae3_reset_notify_type type)
4140 {
4141 	int ret = 0;
4142 
4143 	switch (type) {
4144 	case HNAE3_UP_CLIENT:
4145 		ret = hns3_reset_notify_up_enet(handle);
4146 		break;
4147 	case HNAE3_DOWN_CLIENT:
4148 		ret = hns3_reset_notify_down_enet(handle);
4149 		break;
4150 	case HNAE3_INIT_CLIENT:
4151 		ret = hns3_reset_notify_init_enet(handle);
4152 		break;
4153 	case HNAE3_UNINIT_CLIENT:
4154 		ret = hns3_reset_notify_uninit_enet(handle);
4155 		break;
4156 	case HNAE3_RESTORE_CLIENT:
4157 		ret = hns3_reset_notify_restore_enet(handle);
4158 		break;
4159 	default:
4160 		break;
4161 	}
4162 
4163 	return ret;
4164 }
4165 
4166 int hns3_set_channels(struct net_device *netdev,
4167 		      struct ethtool_channels *ch)
4168 {
4169 	struct hnae3_handle *h = hns3_get_handle(netdev);
4170 	struct hnae3_knic_private_info *kinfo = &h->kinfo;
4171 	bool rxfh_configured = netif_is_rxfh_configured(netdev);
4172 	u32 new_tqp_num = ch->combined_count;
4173 	u16 org_tqp_num;
4174 	int ret;
4175 
4176 	if (ch->rx_count || ch->tx_count)
4177 		return -EINVAL;
4178 
4179 	if (new_tqp_num > hns3_get_max_available_channels(h) ||
4180 	    new_tqp_num < 1) {
4181 		dev_err(&netdev->dev,
4182 			"Change tqps fail, the tqp range is from 1 to %d",
4183 			hns3_get_max_available_channels(h));
4184 		return -EINVAL;
4185 	}
4186 
4187 	if (kinfo->rss_size == new_tqp_num)
4188 		return 0;
4189 
4190 	ret = hns3_reset_notify(h, HNAE3_DOWN_CLIENT);
4191 	if (ret)
4192 		return ret;
4193 
4194 	ret = hns3_reset_notify(h, HNAE3_UNINIT_CLIENT);
4195 	if (ret)
4196 		return ret;
4197 
4198 	org_tqp_num = h->kinfo.num_tqps;
4199 	ret = h->ae_algo->ops->set_channels(h, new_tqp_num, rxfh_configured);
4200 	if (ret) {
4201 		ret = h->ae_algo->ops->set_channels(h, org_tqp_num,
4202 						    rxfh_configured);
4203 		if (ret) {
4204 			/* If revert to old tqp failed, fatal error occurred */
4205 			dev_err(&netdev->dev,
4206 				"Revert to old tqp num fail, ret=%d", ret);
4207 			return ret;
4208 		}
4209 		dev_info(&netdev->dev,
4210 			 "Change tqp num fail, Revert to old tqp num");
4211 	}
4212 	ret = hns3_reset_notify(h, HNAE3_INIT_CLIENT);
4213 	if (ret)
4214 		return ret;
4215 
4216 	return hns3_reset_notify(h, HNAE3_UP_CLIENT);
4217 }
4218 
4219 static const struct hnae3_client_ops client_ops = {
4220 	.init_instance = hns3_client_init,
4221 	.uninit_instance = hns3_client_uninit,
4222 	.link_status_change = hns3_link_status_change,
4223 	.setup_tc = hns3_client_setup_tc,
4224 	.reset_notify = hns3_reset_notify,
4225 };
4226 
4227 /* hns3_init_module - Driver registration routine
4228  * hns3_init_module is the first routine called when the driver is
4229  * loaded. All it does is register with the PCI subsystem.
4230  */
4231 static int __init hns3_init_module(void)
4232 {
4233 	int ret;
4234 
4235 	pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
4236 	pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
4237 
4238 	client.type = HNAE3_CLIENT_KNIC;
4239 	snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH - 1, "%s",
4240 		 hns3_driver_name);
4241 
4242 	client.ops = &client_ops;
4243 
4244 	INIT_LIST_HEAD(&client.node);
4245 
4246 	hns3_dbg_register_debugfs(hns3_driver_name);
4247 
4248 	ret = hnae3_register_client(&client);
4249 	if (ret)
4250 		goto err_reg_client;
4251 
4252 	ret = pci_register_driver(&hns3_driver);
4253 	if (ret)
4254 		goto err_reg_driver;
4255 
4256 	return ret;
4257 
4258 err_reg_driver:
4259 	hnae3_unregister_client(&client);
4260 err_reg_client:
4261 	hns3_dbg_unregister_debugfs();
4262 	return ret;
4263 }
4264 module_init(hns3_init_module);
4265 
4266 /* hns3_exit_module - Driver exit cleanup routine
4267  * hns3_exit_module is called just before the driver is removed
4268  * from memory.
4269  */
4270 static void __exit hns3_exit_module(void)
4271 {
4272 	pci_unregister_driver(&hns3_driver);
4273 	hnae3_unregister_client(&client);
4274 	hns3_dbg_unregister_debugfs();
4275 }
4276 module_exit(hns3_exit_module);
4277 
4278 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
4279 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
4280 MODULE_LICENSE("GPL");
4281 MODULE_ALIAS("pci:hns-nic");
4282 MODULE_VERSION(HNS3_MOD_VERSION);
4283