xref: /linux/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c (revision ba95c7452439756d4f6dceb5a188b7c31dbbe5b6)
1 // SPDX-License-Identifier: GPL-2.0+
2 // Copyright (c) 2016-2017 Hisilicon Limited.
3 
4 #include <linux/dma-mapping.h>
5 #include <linux/etherdevice.h>
6 #include <linux/interrupt.h>
7 #ifdef CONFIG_RFS_ACCEL
8 #include <linux/cpu_rmap.h>
9 #endif
10 #include <linux/if_vlan.h>
11 #include <linux/ip.h>
12 #include <linux/ipv6.h>
13 #include <linux/module.h>
14 #include <linux/pci.h>
15 #include <linux/aer.h>
16 #include <linux/skbuff.h>
17 #include <linux/sctp.h>
18 #include <linux/vermagic.h>
19 #include <net/gre.h>
20 #include <net/ip6_checksum.h>
21 #include <net/pkt_cls.h>
22 #include <net/tcp.h>
23 #include <net/vxlan.h>
24 
25 #include "hnae3.h"
26 #include "hns3_enet.h"
27 
28 #define hns3_set_field(origin, shift, val)	((origin) |= ((val) << (shift)))
29 #define hns3_tx_bd_count(S)	DIV_ROUND_UP(S, HNS3_MAX_BD_SIZE)
30 
31 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force);
32 static void hns3_remove_hw_addr(struct net_device *netdev);
33 
34 static const char hns3_driver_name[] = "hns3";
35 const char hns3_driver_version[] = VERMAGIC_STRING;
36 static const char hns3_driver_string[] =
37 			"Hisilicon Ethernet Network Driver for Hip08 Family";
38 static const char hns3_copyright[] = "Copyright (c) 2017 Huawei Corporation.";
39 static struct hnae3_client client;
40 
41 static int debug = -1;
42 module_param(debug, int, 0);
43 MODULE_PARM_DESC(debug, " Network interface message level setting");
44 
45 #define DEFAULT_MSG_LEVEL (NETIF_MSG_PROBE | NETIF_MSG_LINK | \
46 			   NETIF_MSG_IFDOWN | NETIF_MSG_IFUP)
47 
48 /* hns3_pci_tbl - PCI Device ID Table
49  *
50  * Last entry must be all 0s
51  *
52  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
53  *   Class, Class Mask, private data (not used) }
54  */
55 static const struct pci_device_id hns3_pci_tbl[] = {
56 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
57 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
58 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA),
59 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
60 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC),
61 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
62 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA),
63 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
64 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC),
65 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
66 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC),
67 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
68 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_VF), 0},
69 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF),
70 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
71 	/* required last entry */
72 	{0, }
73 };
74 MODULE_DEVICE_TABLE(pci, hns3_pci_tbl);
75 
76 static irqreturn_t hns3_irq_handle(int irq, void *vector)
77 {
78 	struct hns3_enet_tqp_vector *tqp_vector = vector;
79 
80 	napi_schedule_irqoff(&tqp_vector->napi);
81 
82 	return IRQ_HANDLED;
83 }
84 
85 static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv)
86 {
87 	struct hns3_enet_tqp_vector *tqp_vectors;
88 	unsigned int i;
89 
90 	for (i = 0; i < priv->vector_num; i++) {
91 		tqp_vectors = &priv->tqp_vector[i];
92 
93 		if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED)
94 			continue;
95 
96 		/* clear the affinity mask */
97 		irq_set_affinity_hint(tqp_vectors->vector_irq, NULL);
98 
99 		/* release the irq resource */
100 		free_irq(tqp_vectors->vector_irq, tqp_vectors);
101 		tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED;
102 	}
103 }
104 
105 static int hns3_nic_init_irq(struct hns3_nic_priv *priv)
106 {
107 	struct hns3_enet_tqp_vector *tqp_vectors;
108 	int txrx_int_idx = 0;
109 	int rx_int_idx = 0;
110 	int tx_int_idx = 0;
111 	unsigned int i;
112 	int ret;
113 
114 	for (i = 0; i < priv->vector_num; i++) {
115 		tqp_vectors = &priv->tqp_vector[i];
116 
117 		if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED)
118 			continue;
119 
120 		if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) {
121 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
122 				 "%s-%s-%d", priv->netdev->name, "TxRx",
123 				 txrx_int_idx++);
124 			txrx_int_idx++;
125 		} else if (tqp_vectors->rx_group.ring) {
126 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
127 				 "%s-%s-%d", priv->netdev->name, "Rx",
128 				 rx_int_idx++);
129 		} else if (tqp_vectors->tx_group.ring) {
130 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1,
131 				 "%s-%s-%d", priv->netdev->name, "Tx",
132 				 tx_int_idx++);
133 		} else {
134 			/* Skip this unused q_vector */
135 			continue;
136 		}
137 
138 		tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0';
139 
140 		ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0,
141 				  tqp_vectors->name, tqp_vectors);
142 		if (ret) {
143 			netdev_err(priv->netdev, "request irq(%d) fail\n",
144 				   tqp_vectors->vector_irq);
145 			hns3_nic_uninit_irq(priv);
146 			return ret;
147 		}
148 
149 		irq_set_affinity_hint(tqp_vectors->vector_irq,
150 				      &tqp_vectors->affinity_mask);
151 
152 		tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED;
153 	}
154 
155 	return 0;
156 }
157 
158 static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector,
159 				 u32 mask_en)
160 {
161 	writel(mask_en, tqp_vector->mask_addr);
162 }
163 
164 static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector)
165 {
166 	napi_enable(&tqp_vector->napi);
167 
168 	/* enable vector */
169 	hns3_mask_vector_irq(tqp_vector, 1);
170 }
171 
172 static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector)
173 {
174 	/* disable vector */
175 	hns3_mask_vector_irq(tqp_vector, 0);
176 
177 	disable_irq(tqp_vector->vector_irq);
178 	napi_disable(&tqp_vector->napi);
179 }
180 
181 void hns3_set_vector_coalesce_rl(struct hns3_enet_tqp_vector *tqp_vector,
182 				 u32 rl_value)
183 {
184 	u32 rl_reg = hns3_rl_usec_to_reg(rl_value);
185 
186 	/* this defines the configuration for RL (Interrupt Rate Limiter).
187 	 * Rl defines rate of interrupts i.e. number of interrupts-per-second
188 	 * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing
189 	 */
190 
191 	if (rl_reg > 0 && !tqp_vector->tx_group.coal.gl_adapt_enable &&
192 	    !tqp_vector->rx_group.coal.gl_adapt_enable)
193 		/* According to the hardware, the range of rl_reg is
194 		 * 0-59 and the unit is 4.
195 		 */
196 		rl_reg |=  HNS3_INT_RL_ENABLE_MASK;
197 
198 	writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
199 }
200 
201 void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
202 				    u32 gl_value)
203 {
204 	u32 rx_gl_reg = hns3_gl_usec_to_reg(gl_value);
205 
206 	writel(rx_gl_reg, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
207 }
208 
209 void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
210 				    u32 gl_value)
211 {
212 	u32 tx_gl_reg = hns3_gl_usec_to_reg(gl_value);
213 
214 	writel(tx_gl_reg, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
215 }
216 
217 static void hns3_vector_gl_rl_init(struct hns3_enet_tqp_vector *tqp_vector,
218 				   struct hns3_nic_priv *priv)
219 {
220 	/* initialize the configuration for interrupt coalescing.
221 	 * 1. GL (Interrupt Gap Limiter)
222 	 * 2. RL (Interrupt Rate Limiter)
223 	 */
224 
225 	/* Default: enable interrupt coalescing self-adaptive and GL */
226 	tqp_vector->tx_group.coal.gl_adapt_enable = 1;
227 	tqp_vector->rx_group.coal.gl_adapt_enable = 1;
228 
229 	tqp_vector->tx_group.coal.int_gl = HNS3_INT_GL_50K;
230 	tqp_vector->rx_group.coal.int_gl = HNS3_INT_GL_50K;
231 
232 	tqp_vector->rx_group.coal.flow_level = HNS3_FLOW_LOW;
233 	tqp_vector->tx_group.coal.flow_level = HNS3_FLOW_LOW;
234 }
235 
236 static void hns3_vector_gl_rl_init_hw(struct hns3_enet_tqp_vector *tqp_vector,
237 				      struct hns3_nic_priv *priv)
238 {
239 	struct hnae3_handle *h = priv->ae_handle;
240 
241 	hns3_set_vector_coalesce_tx_gl(tqp_vector,
242 				       tqp_vector->tx_group.coal.int_gl);
243 	hns3_set_vector_coalesce_rx_gl(tqp_vector,
244 				       tqp_vector->rx_group.coal.int_gl);
245 	hns3_set_vector_coalesce_rl(tqp_vector, h->kinfo.int_rl_setting);
246 }
247 
248 static int hns3_nic_set_real_num_queue(struct net_device *netdev)
249 {
250 	struct hnae3_handle *h = hns3_get_handle(netdev);
251 	struct hnae3_knic_private_info *kinfo = &h->kinfo;
252 	unsigned int queue_size = kinfo->rss_size * kinfo->num_tc;
253 	int i, ret;
254 
255 	if (kinfo->num_tc <= 1) {
256 		netdev_reset_tc(netdev);
257 	} else {
258 		ret = netdev_set_num_tc(netdev, kinfo->num_tc);
259 		if (ret) {
260 			netdev_err(netdev,
261 				   "netdev_set_num_tc fail, ret=%d!\n", ret);
262 			return ret;
263 		}
264 
265 		for (i = 0; i < HNAE3_MAX_TC; i++) {
266 			if (!kinfo->tc_info[i].enable)
267 				continue;
268 
269 			netdev_set_tc_queue(netdev,
270 					    kinfo->tc_info[i].tc,
271 					    kinfo->tc_info[i].tqp_count,
272 					    kinfo->tc_info[i].tqp_offset);
273 		}
274 	}
275 
276 	ret = netif_set_real_num_tx_queues(netdev, queue_size);
277 	if (ret) {
278 		netdev_err(netdev,
279 			   "netif_set_real_num_tx_queues fail, ret=%d!\n", ret);
280 		return ret;
281 	}
282 
283 	ret = netif_set_real_num_rx_queues(netdev, queue_size);
284 	if (ret) {
285 		netdev_err(netdev,
286 			   "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
287 		return ret;
288 	}
289 
290 	return 0;
291 }
292 
293 static u16 hns3_get_max_available_channels(struct hnae3_handle *h)
294 {
295 	u16 alloc_tqps, max_rss_size, rss_size;
296 
297 	h->ae_algo->ops->get_tqps_and_rss_info(h, &alloc_tqps, &max_rss_size);
298 	rss_size = alloc_tqps / h->kinfo.num_tc;
299 
300 	return min_t(u16, rss_size, max_rss_size);
301 }
302 
303 static void hns3_tqp_enable(struct hnae3_queue *tqp)
304 {
305 	u32 rcb_reg;
306 
307 	rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
308 	rcb_reg |= BIT(HNS3_RING_EN_B);
309 	hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
310 }
311 
312 static void hns3_tqp_disable(struct hnae3_queue *tqp)
313 {
314 	u32 rcb_reg;
315 
316 	rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
317 	rcb_reg &= ~BIT(HNS3_RING_EN_B);
318 	hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
319 }
320 
321 static void hns3_free_rx_cpu_rmap(struct net_device *netdev)
322 {
323 #ifdef CONFIG_RFS_ACCEL
324 	free_irq_cpu_rmap(netdev->rx_cpu_rmap);
325 	netdev->rx_cpu_rmap = NULL;
326 #endif
327 }
328 
329 static int hns3_set_rx_cpu_rmap(struct net_device *netdev)
330 {
331 #ifdef CONFIG_RFS_ACCEL
332 	struct hns3_nic_priv *priv = netdev_priv(netdev);
333 	struct hns3_enet_tqp_vector *tqp_vector;
334 	int i, ret;
335 
336 	if (!netdev->rx_cpu_rmap) {
337 		netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->vector_num);
338 		if (!netdev->rx_cpu_rmap)
339 			return -ENOMEM;
340 	}
341 
342 	for (i = 0; i < priv->vector_num; i++) {
343 		tqp_vector = &priv->tqp_vector[i];
344 		ret = irq_cpu_rmap_add(netdev->rx_cpu_rmap,
345 				       tqp_vector->vector_irq);
346 		if (ret) {
347 			hns3_free_rx_cpu_rmap(netdev);
348 			return ret;
349 		}
350 	}
351 #endif
352 	return 0;
353 }
354 
355 static int hns3_nic_net_up(struct net_device *netdev)
356 {
357 	struct hns3_nic_priv *priv = netdev_priv(netdev);
358 	struct hnae3_handle *h = priv->ae_handle;
359 	int i, j;
360 	int ret;
361 
362 	ret = hns3_nic_reset_all_ring(h);
363 	if (ret)
364 		return ret;
365 
366 	/* the device can work without cpu rmap, only aRFS needs it */
367 	ret = hns3_set_rx_cpu_rmap(netdev);
368 	if (ret)
369 		netdev_warn(netdev, "set rx cpu rmap fail, ret=%d!\n", ret);
370 
371 	/* get irq resource for all vectors */
372 	ret = hns3_nic_init_irq(priv);
373 	if (ret) {
374 		netdev_err(netdev, "init irq failed! ret=%d\n", ret);
375 		goto free_rmap;
376 	}
377 
378 	clear_bit(HNS3_NIC_STATE_DOWN, &priv->state);
379 
380 	/* enable the vectors */
381 	for (i = 0; i < priv->vector_num; i++)
382 		hns3_vector_enable(&priv->tqp_vector[i]);
383 
384 	/* enable rcb */
385 	for (j = 0; j < h->kinfo.num_tqps; j++)
386 		hns3_tqp_enable(h->kinfo.tqp[j]);
387 
388 	/* start the ae_dev */
389 	ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0;
390 	if (ret)
391 		goto out_start_err;
392 
393 	return 0;
394 
395 out_start_err:
396 	set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
397 	while (j--)
398 		hns3_tqp_disable(h->kinfo.tqp[j]);
399 
400 	for (j = i - 1; j >= 0; j--)
401 		hns3_vector_disable(&priv->tqp_vector[j]);
402 
403 	hns3_nic_uninit_irq(priv);
404 free_rmap:
405 	hns3_free_rx_cpu_rmap(netdev);
406 	return ret;
407 }
408 
409 static void hns3_config_xps(struct hns3_nic_priv *priv)
410 {
411 	int i;
412 
413 	for (i = 0; i < priv->vector_num; i++) {
414 		struct hns3_enet_tqp_vector *tqp_vector = &priv->tqp_vector[i];
415 		struct hns3_enet_ring *ring = tqp_vector->tx_group.ring;
416 
417 		while (ring) {
418 			int ret;
419 
420 			ret = netif_set_xps_queue(priv->netdev,
421 						  &tqp_vector->affinity_mask,
422 						  ring->tqp->tqp_index);
423 			if (ret)
424 				netdev_warn(priv->netdev,
425 					    "set xps queue failed: %d", ret);
426 
427 			ring = ring->next;
428 		}
429 	}
430 }
431 
432 static int hns3_nic_net_open(struct net_device *netdev)
433 {
434 	struct hns3_nic_priv *priv = netdev_priv(netdev);
435 	struct hnae3_handle *h = hns3_get_handle(netdev);
436 	struct hnae3_knic_private_info *kinfo;
437 	int i, ret;
438 
439 	if (hns3_nic_resetting(netdev))
440 		return -EBUSY;
441 
442 	netif_carrier_off(netdev);
443 
444 	ret = hns3_nic_set_real_num_queue(netdev);
445 	if (ret)
446 		return ret;
447 
448 	ret = hns3_nic_net_up(netdev);
449 	if (ret) {
450 		netdev_err(netdev, "net up fail, ret=%d!\n", ret);
451 		return ret;
452 	}
453 
454 	kinfo = &h->kinfo;
455 	for (i = 0; i < HNAE3_MAX_USER_PRIO; i++)
456 		netdev_set_prio_tc_map(netdev, i, kinfo->prio_tc[i]);
457 
458 	if (h->ae_algo->ops->set_timer_task)
459 		h->ae_algo->ops->set_timer_task(priv->ae_handle, true);
460 
461 	hns3_config_xps(priv);
462 	return 0;
463 }
464 
465 static void hns3_reset_tx_queue(struct hnae3_handle *h)
466 {
467 	struct net_device *ndev = h->kinfo.netdev;
468 	struct hns3_nic_priv *priv = netdev_priv(ndev);
469 	struct netdev_queue *dev_queue;
470 	u32 i;
471 
472 	for (i = 0; i < h->kinfo.num_tqps; i++) {
473 		dev_queue = netdev_get_tx_queue(ndev,
474 						priv->ring_data[i].queue_index);
475 		netdev_tx_reset_queue(dev_queue);
476 	}
477 }
478 
479 static void hns3_nic_net_down(struct net_device *netdev)
480 {
481 	struct hns3_nic_priv *priv = netdev_priv(netdev);
482 	struct hnae3_handle *h = hns3_get_handle(netdev);
483 	const struct hnae3_ae_ops *ops;
484 	int i;
485 
486 	/* disable vectors */
487 	for (i = 0; i < priv->vector_num; i++)
488 		hns3_vector_disable(&priv->tqp_vector[i]);
489 
490 	/* disable rcb */
491 	for (i = 0; i < h->kinfo.num_tqps; i++)
492 		hns3_tqp_disable(h->kinfo.tqp[i]);
493 
494 	/* stop ae_dev */
495 	ops = priv->ae_handle->ae_algo->ops;
496 	if (ops->stop)
497 		ops->stop(priv->ae_handle);
498 
499 	hns3_free_rx_cpu_rmap(netdev);
500 
501 	/* free irq resources */
502 	hns3_nic_uninit_irq(priv);
503 
504 	/* delay ring buffer clearing to hns3_reset_notify_uninit_enet
505 	 * during reset process, because driver may not be able
506 	 * to disable the ring through firmware when downing the netdev.
507 	 */
508 	if (!hns3_nic_resetting(netdev))
509 		hns3_clear_all_ring(priv->ae_handle, false);
510 
511 	hns3_reset_tx_queue(priv->ae_handle);
512 }
513 
514 static int hns3_nic_net_stop(struct net_device *netdev)
515 {
516 	struct hns3_nic_priv *priv = netdev_priv(netdev);
517 	struct hnae3_handle *h = hns3_get_handle(netdev);
518 
519 	if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
520 		return 0;
521 
522 	if (h->ae_algo->ops->set_timer_task)
523 		h->ae_algo->ops->set_timer_task(priv->ae_handle, false);
524 
525 	netif_tx_stop_all_queues(netdev);
526 	netif_carrier_off(netdev);
527 
528 	hns3_nic_net_down(netdev);
529 
530 	return 0;
531 }
532 
533 static int hns3_nic_uc_sync(struct net_device *netdev,
534 			    const unsigned char *addr)
535 {
536 	struct hnae3_handle *h = hns3_get_handle(netdev);
537 
538 	if (h->ae_algo->ops->add_uc_addr)
539 		return h->ae_algo->ops->add_uc_addr(h, addr);
540 
541 	return 0;
542 }
543 
544 static int hns3_nic_uc_unsync(struct net_device *netdev,
545 			      const unsigned char *addr)
546 {
547 	struct hnae3_handle *h = hns3_get_handle(netdev);
548 
549 	if (h->ae_algo->ops->rm_uc_addr)
550 		return h->ae_algo->ops->rm_uc_addr(h, addr);
551 
552 	return 0;
553 }
554 
555 static int hns3_nic_mc_sync(struct net_device *netdev,
556 			    const unsigned char *addr)
557 {
558 	struct hnae3_handle *h = hns3_get_handle(netdev);
559 
560 	if (h->ae_algo->ops->add_mc_addr)
561 		return h->ae_algo->ops->add_mc_addr(h, addr);
562 
563 	return 0;
564 }
565 
566 static int hns3_nic_mc_unsync(struct net_device *netdev,
567 			      const unsigned char *addr)
568 {
569 	struct hnae3_handle *h = hns3_get_handle(netdev);
570 
571 	if (h->ae_algo->ops->rm_mc_addr)
572 		return h->ae_algo->ops->rm_mc_addr(h, addr);
573 
574 	return 0;
575 }
576 
577 static u8 hns3_get_netdev_flags(struct net_device *netdev)
578 {
579 	u8 flags = 0;
580 
581 	if (netdev->flags & IFF_PROMISC) {
582 		flags = HNAE3_USER_UPE | HNAE3_USER_MPE | HNAE3_BPE;
583 	} else {
584 		flags |= HNAE3_VLAN_FLTR;
585 		if (netdev->flags & IFF_ALLMULTI)
586 			flags |= HNAE3_USER_MPE;
587 	}
588 
589 	return flags;
590 }
591 
592 static void hns3_nic_set_rx_mode(struct net_device *netdev)
593 {
594 	struct hnae3_handle *h = hns3_get_handle(netdev);
595 	u8 new_flags;
596 	int ret;
597 
598 	new_flags = hns3_get_netdev_flags(netdev);
599 
600 	ret = __dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync);
601 	if (ret) {
602 		netdev_err(netdev, "sync uc address fail\n");
603 		if (ret == -ENOSPC)
604 			new_flags |= HNAE3_OVERFLOW_UPE;
605 	}
606 
607 	if (netdev->flags & IFF_MULTICAST) {
608 		ret = __dev_mc_sync(netdev, hns3_nic_mc_sync,
609 				    hns3_nic_mc_unsync);
610 		if (ret) {
611 			netdev_err(netdev, "sync mc address fail\n");
612 			if (ret == -ENOSPC)
613 				new_flags |= HNAE3_OVERFLOW_MPE;
614 		}
615 	}
616 
617 	/* User mode Promisc mode enable and vlan filtering is disabled to
618 	 * let all packets in. MAC-VLAN Table overflow Promisc enabled and
619 	 * vlan fitering is enabled
620 	 */
621 	hns3_enable_vlan_filter(netdev, new_flags & HNAE3_VLAN_FLTR);
622 	h->netdev_flags = new_flags;
623 	hns3_update_promisc_mode(netdev, new_flags);
624 }
625 
626 int hns3_update_promisc_mode(struct net_device *netdev, u8 promisc_flags)
627 {
628 	struct hns3_nic_priv *priv = netdev_priv(netdev);
629 	struct hnae3_handle *h = priv->ae_handle;
630 
631 	if (h->ae_algo->ops->set_promisc_mode) {
632 		return h->ae_algo->ops->set_promisc_mode(h,
633 						promisc_flags & HNAE3_UPE,
634 						promisc_flags & HNAE3_MPE);
635 	}
636 
637 	return 0;
638 }
639 
640 void hns3_enable_vlan_filter(struct net_device *netdev, bool enable)
641 {
642 	struct hns3_nic_priv *priv = netdev_priv(netdev);
643 	struct hnae3_handle *h = priv->ae_handle;
644 	bool last_state;
645 
646 	if (h->pdev->revision >= 0x21 && h->ae_algo->ops->enable_vlan_filter) {
647 		last_state = h->netdev_flags & HNAE3_VLAN_FLTR ? true : false;
648 		if (enable != last_state) {
649 			netdev_info(netdev,
650 				    "%s vlan filter\n",
651 				    enable ? "enable" : "disable");
652 			h->ae_algo->ops->enable_vlan_filter(h, enable);
653 		}
654 	}
655 }
656 
657 static int hns3_set_tso(struct sk_buff *skb, u32 *paylen,
658 			u16 *mss, u32 *type_cs_vlan_tso)
659 {
660 	u32 l4_offset, hdr_len;
661 	union l3_hdr_info l3;
662 	union l4_hdr_info l4;
663 	u32 l4_paylen;
664 	int ret;
665 
666 	if (!skb_is_gso(skb))
667 		return 0;
668 
669 	ret = skb_cow_head(skb, 0);
670 	if (unlikely(ret))
671 		return ret;
672 
673 	l3.hdr = skb_network_header(skb);
674 	l4.hdr = skb_transport_header(skb);
675 
676 	/* Software should clear the IPv4's checksum field when tso is
677 	 * needed.
678 	 */
679 	if (l3.v4->version == 4)
680 		l3.v4->check = 0;
681 
682 	/* tunnel packet */
683 	if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
684 					 SKB_GSO_GRE_CSUM |
685 					 SKB_GSO_UDP_TUNNEL |
686 					 SKB_GSO_UDP_TUNNEL_CSUM)) {
687 		if ((!(skb_shinfo(skb)->gso_type &
688 		    SKB_GSO_PARTIAL)) &&
689 		    (skb_shinfo(skb)->gso_type &
690 		    SKB_GSO_UDP_TUNNEL_CSUM)) {
691 			/* Software should clear the udp's checksum
692 			 * field when tso is needed.
693 			 */
694 			l4.udp->check = 0;
695 		}
696 		/* reset l3&l4 pointers from outer to inner headers */
697 		l3.hdr = skb_inner_network_header(skb);
698 		l4.hdr = skb_inner_transport_header(skb);
699 
700 		/* Software should clear the IPv4's checksum field when
701 		 * tso is needed.
702 		 */
703 		if (l3.v4->version == 4)
704 			l3.v4->check = 0;
705 	}
706 
707 	/* normal or tunnel packet */
708 	l4_offset = l4.hdr - skb->data;
709 	hdr_len = (l4.tcp->doff << 2) + l4_offset;
710 
711 	/* remove payload length from inner pseudo checksum when tso */
712 	l4_paylen = skb->len - l4_offset;
713 	csum_replace_by_diff(&l4.tcp->check,
714 			     (__force __wsum)htonl(l4_paylen));
715 
716 	/* find the txbd field values */
717 	*paylen = skb->len - hdr_len;
718 	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_TSO_B, 1);
719 
720 	/* get MSS for TSO */
721 	*mss = skb_shinfo(skb)->gso_size;
722 
723 	return 0;
724 }
725 
726 static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
727 				u8 *il4_proto)
728 {
729 	union l3_hdr_info l3;
730 	unsigned char *l4_hdr;
731 	unsigned char *exthdr;
732 	u8 l4_proto_tmp;
733 	__be16 frag_off;
734 
735 	/* find outer header point */
736 	l3.hdr = skb_network_header(skb);
737 	l4_hdr = skb_transport_header(skb);
738 
739 	if (skb->protocol == htons(ETH_P_IPV6)) {
740 		exthdr = l3.hdr + sizeof(*l3.v6);
741 		l4_proto_tmp = l3.v6->nexthdr;
742 		if (l4_hdr != exthdr)
743 			ipv6_skip_exthdr(skb, exthdr - skb->data,
744 					 &l4_proto_tmp, &frag_off);
745 	} else if (skb->protocol == htons(ETH_P_IP)) {
746 		l4_proto_tmp = l3.v4->protocol;
747 	} else {
748 		return -EINVAL;
749 	}
750 
751 	*ol4_proto = l4_proto_tmp;
752 
753 	/* tunnel packet */
754 	if (!skb->encapsulation) {
755 		*il4_proto = 0;
756 		return 0;
757 	}
758 
759 	/* find inner header point */
760 	l3.hdr = skb_inner_network_header(skb);
761 	l4_hdr = skb_inner_transport_header(skb);
762 
763 	if (l3.v6->version == 6) {
764 		exthdr = l3.hdr + sizeof(*l3.v6);
765 		l4_proto_tmp = l3.v6->nexthdr;
766 		if (l4_hdr != exthdr)
767 			ipv6_skip_exthdr(skb, exthdr - skb->data,
768 					 &l4_proto_tmp, &frag_off);
769 	} else if (l3.v4->version == 4) {
770 		l4_proto_tmp = l3.v4->protocol;
771 	}
772 
773 	*il4_proto = l4_proto_tmp;
774 
775 	return 0;
776 }
777 
778 /* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
779  * and it is udp packet, which has a dest port as the IANA assigned.
780  * the hardware is expected to do the checksum offload, but the
781  * hardware will not do the checksum offload when udp dest port is
782  * 4789.
783  */
784 static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
785 {
786 	union l4_hdr_info l4;
787 
788 	l4.hdr = skb_transport_header(skb);
789 
790 	if (!(!skb->encapsulation &&
791 	      l4.udp->dest == htons(IANA_VXLAN_UDP_PORT)))
792 		return false;
793 
794 	skb_checksum_help(skb);
795 
796 	return true;
797 }
798 
799 static void hns3_set_outer_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
800 				  u32 *ol_type_vlan_len_msec)
801 {
802 	u32 l2_len, l3_len, l4_len;
803 	unsigned char *il2_hdr;
804 	union l3_hdr_info l3;
805 	union l4_hdr_info l4;
806 
807 	l3.hdr = skb_network_header(skb);
808 	l4.hdr = skb_transport_header(skb);
809 
810 	/* compute OL2 header size, defined in 2 Bytes */
811 	l2_len = l3.hdr - skb->data;
812 	hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L2LEN_S, l2_len >> 1);
813 
814 	/* compute OL3 header size, defined in 4 Bytes */
815 	l3_len = l4.hdr - l3.hdr;
816 	hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_S, l3_len >> 2);
817 
818 	il2_hdr = skb_inner_mac_header(skb);
819 	/* compute OL4 header size, defined in 4 Bytes */
820 	l4_len = il2_hdr - l4.hdr;
821 	hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L4LEN_S, l4_len >> 2);
822 
823 	/* define outer network header type */
824 	if (skb->protocol == htons(ETH_P_IP)) {
825 		if (skb_is_gso(skb))
826 			hns3_set_field(*ol_type_vlan_len_msec,
827 				       HNS3_TXD_OL3T_S,
828 				       HNS3_OL3T_IPV4_CSUM);
829 		else
830 			hns3_set_field(*ol_type_vlan_len_msec,
831 				       HNS3_TXD_OL3T_S,
832 				       HNS3_OL3T_IPV4_NO_CSUM);
833 
834 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
835 		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_S,
836 			       HNS3_OL3T_IPV6);
837 	}
838 
839 	if (ol4_proto == IPPROTO_UDP)
840 		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
841 			       HNS3_TUN_MAC_IN_UDP);
842 	else if (ol4_proto == IPPROTO_GRE)
843 		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
844 			       HNS3_TUN_NVGRE);
845 }
846 
847 static int hns3_set_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
848 			   u8 il4_proto, u32 *type_cs_vlan_tso,
849 			   u32 *ol_type_vlan_len_msec)
850 {
851 	unsigned char *l2_hdr = skb->data;
852 	u32 l4_proto = ol4_proto;
853 	union l4_hdr_info l4;
854 	union l3_hdr_info l3;
855 	u32 l2_len, l3_len;
856 
857 	l4.hdr = skb_transport_header(skb);
858 	l3.hdr = skb_network_header(skb);
859 
860 	/* handle encapsulation skb */
861 	if (skb->encapsulation) {
862 		/* If this is a not UDP/GRE encapsulation skb */
863 		if (!(ol4_proto == IPPROTO_UDP || ol4_proto == IPPROTO_GRE)) {
864 			/* drop the skb tunnel packet if hardware don't support,
865 			 * because hardware can't calculate csum when TSO.
866 			 */
867 			if (skb_is_gso(skb))
868 				return -EDOM;
869 
870 			/* the stack computes the IP header already,
871 			 * driver calculate l4 checksum when not TSO.
872 			 */
873 			skb_checksum_help(skb);
874 			return 0;
875 		}
876 
877 		hns3_set_outer_l2l3l4(skb, ol4_proto, ol_type_vlan_len_msec);
878 
879 		/* switch to inner header */
880 		l2_hdr = skb_inner_mac_header(skb);
881 		l3.hdr = skb_inner_network_header(skb);
882 		l4.hdr = skb_inner_transport_header(skb);
883 		l4_proto = il4_proto;
884 	}
885 
886 	if (l3.v4->version == 4) {
887 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
888 			       HNS3_L3T_IPV4);
889 
890 		/* the stack computes the IP header already, the only time we
891 		 * need the hardware to recompute it is in the case of TSO.
892 		 */
893 		if (skb_is_gso(skb))
894 			hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3CS_B, 1);
895 	} else if (l3.v6->version == 6) {
896 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
897 			       HNS3_L3T_IPV6);
898 	}
899 
900 	/* compute inner(/normal) L2 header size, defined in 2 Bytes */
901 	l2_len = l3.hdr - l2_hdr;
902 	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S, l2_len >> 1);
903 
904 	/* compute inner(/normal) L3 header size, defined in 4 Bytes */
905 	l3_len = l4.hdr - l3.hdr;
906 	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_S, l3_len >> 2);
907 
908 	/* compute inner(/normal) L4 header size, defined in 4 Bytes */
909 	switch (l4_proto) {
910 	case IPPROTO_TCP:
911 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
912 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
913 			       HNS3_L4T_TCP);
914 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
915 			       l4.tcp->doff);
916 		break;
917 	case IPPROTO_UDP:
918 		if (hns3_tunnel_csum_bug(skb))
919 			break;
920 
921 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
922 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
923 			       HNS3_L4T_UDP);
924 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
925 			       (sizeof(struct udphdr) >> 2));
926 		break;
927 	case IPPROTO_SCTP:
928 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
929 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
930 			       HNS3_L4T_SCTP);
931 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
932 			       (sizeof(struct sctphdr) >> 2));
933 		break;
934 	default:
935 		/* drop the skb tunnel packet if hardware don't support,
936 		 * because hardware can't calculate csum when TSO.
937 		 */
938 		if (skb_is_gso(skb))
939 			return -EDOM;
940 
941 		/* the stack computes the IP header already,
942 		 * driver calculate l4 checksum when not TSO.
943 		 */
944 		skb_checksum_help(skb);
945 		return 0;
946 	}
947 
948 	return 0;
949 }
950 
951 static void hns3_set_txbd_baseinfo(u16 *bdtp_fe_sc_vld_ra_ri, int frag_end)
952 {
953 	/* Config bd buffer end */
954 	hns3_set_field(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_FE_B, !!frag_end);
955 	hns3_set_field(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_VLD_B, 1);
956 }
957 
958 static int hns3_fill_desc_vtags(struct sk_buff *skb,
959 				struct hns3_enet_ring *tx_ring,
960 				u32 *inner_vlan_flag,
961 				u32 *out_vlan_flag,
962 				u16 *inner_vtag,
963 				u16 *out_vtag)
964 {
965 #define HNS3_TX_VLAN_PRIO_SHIFT 13
966 
967 	struct hnae3_handle *handle = tx_ring->tqp->handle;
968 
969 	/* Since HW limitation, if port based insert VLAN enabled, only one VLAN
970 	 * header is allowed in skb, otherwise it will cause RAS error.
971 	 */
972 	if (unlikely(skb_vlan_tagged_multi(skb) &&
973 		     handle->port_base_vlan_state ==
974 		     HNAE3_PORT_BASE_VLAN_ENABLE))
975 		return -EINVAL;
976 
977 	if (skb->protocol == htons(ETH_P_8021Q) &&
978 	    !(tx_ring->tqp->handle->kinfo.netdev->features &
979 	    NETIF_F_HW_VLAN_CTAG_TX)) {
980 		/* When HW VLAN acceleration is turned off, and the stack
981 		 * sets the protocol to 802.1q, the driver just need to
982 		 * set the protocol to the encapsulated ethertype.
983 		 */
984 		skb->protocol = vlan_get_protocol(skb);
985 		return 0;
986 	}
987 
988 	if (skb_vlan_tag_present(skb)) {
989 		u16 vlan_tag;
990 
991 		vlan_tag = skb_vlan_tag_get(skb);
992 		vlan_tag |= (skb->priority & 0x7) << HNS3_TX_VLAN_PRIO_SHIFT;
993 
994 		/* Based on hw strategy, use out_vtag in two layer tag case,
995 		 * and use inner_vtag in one tag case.
996 		 */
997 		if (skb->protocol == htons(ETH_P_8021Q)) {
998 			if (handle->port_base_vlan_state ==
999 			    HNAE3_PORT_BASE_VLAN_DISABLE){
1000 				hns3_set_field(*out_vlan_flag,
1001 					       HNS3_TXD_OVLAN_B, 1);
1002 				*out_vtag = vlan_tag;
1003 			} else {
1004 				hns3_set_field(*inner_vlan_flag,
1005 					       HNS3_TXD_VLAN_B, 1);
1006 				*inner_vtag = vlan_tag;
1007 			}
1008 		} else {
1009 			hns3_set_field(*inner_vlan_flag, HNS3_TXD_VLAN_B, 1);
1010 			*inner_vtag = vlan_tag;
1011 		}
1012 	} else if (skb->protocol == htons(ETH_P_8021Q)) {
1013 		struct vlan_ethhdr *vhdr;
1014 		int rc;
1015 
1016 		rc = skb_cow_head(skb, 0);
1017 		if (unlikely(rc < 0))
1018 			return rc;
1019 		vhdr = (struct vlan_ethhdr *)skb->data;
1020 		vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority & 0x7)
1021 					<< HNS3_TX_VLAN_PRIO_SHIFT);
1022 	}
1023 
1024 	skb->protocol = vlan_get_protocol(skb);
1025 	return 0;
1026 }
1027 
1028 static int hns3_fill_desc(struct hns3_enet_ring *ring, void *priv,
1029 			  unsigned int size, int frag_end,
1030 			  enum hns_desc_type type)
1031 {
1032 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1033 	struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1034 	struct device *dev = ring_to_dev(ring);
1035 	struct skb_frag_struct *frag;
1036 	unsigned int frag_buf_num;
1037 	int k, sizeoflast;
1038 	dma_addr_t dma;
1039 
1040 	if (type == DESC_TYPE_SKB) {
1041 		struct sk_buff *skb = (struct sk_buff *)priv;
1042 		u32 ol_type_vlan_len_msec = 0;
1043 		u32 type_cs_vlan_tso = 0;
1044 		u32 paylen = skb->len;
1045 		u16 inner_vtag = 0;
1046 		u16 out_vtag = 0;
1047 		u16 mss = 0;
1048 		int ret;
1049 
1050 		ret = hns3_fill_desc_vtags(skb, ring, &type_cs_vlan_tso,
1051 					   &ol_type_vlan_len_msec,
1052 					   &inner_vtag, &out_vtag);
1053 		if (unlikely(ret))
1054 			return ret;
1055 
1056 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1057 			u8 ol4_proto, il4_proto;
1058 
1059 			skb_reset_mac_len(skb);
1060 
1061 			ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
1062 			if (unlikely(ret))
1063 				return ret;
1064 
1065 			ret = hns3_set_l2l3l4(skb, ol4_proto, il4_proto,
1066 					      &type_cs_vlan_tso,
1067 					      &ol_type_vlan_len_msec);
1068 			if (unlikely(ret))
1069 				return ret;
1070 
1071 			ret = hns3_set_tso(skb, &paylen, &mss,
1072 					   &type_cs_vlan_tso);
1073 			if (unlikely(ret))
1074 				return ret;
1075 		}
1076 
1077 		/* Set txbd */
1078 		desc->tx.ol_type_vlan_len_msec =
1079 			cpu_to_le32(ol_type_vlan_len_msec);
1080 		desc->tx.type_cs_vlan_tso_len =	cpu_to_le32(type_cs_vlan_tso);
1081 		desc->tx.paylen = cpu_to_le32(paylen);
1082 		desc->tx.mss = cpu_to_le16(mss);
1083 		desc->tx.vlan_tag = cpu_to_le16(inner_vtag);
1084 		desc->tx.outer_vlan_tag = cpu_to_le16(out_vtag);
1085 
1086 		dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1087 	} else {
1088 		frag = (struct skb_frag_struct *)priv;
1089 		dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1090 	}
1091 
1092 	if (unlikely(dma_mapping_error(dev, dma))) {
1093 		ring->stats.sw_err_cnt++;
1094 		return -ENOMEM;
1095 	}
1096 
1097 	desc_cb->length = size;
1098 
1099 	if (likely(size <= HNS3_MAX_BD_SIZE)) {
1100 		u16 bdtp_fe_sc_vld_ra_ri = 0;
1101 
1102 		desc_cb->priv = priv;
1103 		desc_cb->dma = dma;
1104 		desc_cb->type = type;
1105 		desc->addr = cpu_to_le64(dma);
1106 		desc->tx.send_size = cpu_to_le16(size);
1107 		hns3_set_txbd_baseinfo(&bdtp_fe_sc_vld_ra_ri, frag_end);
1108 		desc->tx.bdtp_fe_sc_vld_ra_ri =
1109 			cpu_to_le16(bdtp_fe_sc_vld_ra_ri);
1110 
1111 		ring_ptr_move_fw(ring, next_to_use);
1112 		return 0;
1113 	}
1114 
1115 	frag_buf_num = hns3_tx_bd_count(size);
1116 	sizeoflast = size & HNS3_TX_LAST_SIZE_M;
1117 	sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1118 
1119 	/* When frag size is bigger than hardware limit, split this frag */
1120 	for (k = 0; k < frag_buf_num; k++) {
1121 		u16 bdtp_fe_sc_vld_ra_ri = 0;
1122 
1123 		/* The txbd's baseinfo of DESC_TYPE_PAGE & DESC_TYPE_SKB */
1124 		desc_cb->priv = priv;
1125 		desc_cb->dma = dma + HNS3_MAX_BD_SIZE * k;
1126 		desc_cb->type = (type == DESC_TYPE_SKB && !k) ?
1127 				DESC_TYPE_SKB : DESC_TYPE_PAGE;
1128 
1129 		/* now, fill the descriptor */
1130 		desc->addr = cpu_to_le64(dma + HNS3_MAX_BD_SIZE * k);
1131 		desc->tx.send_size = cpu_to_le16((k == frag_buf_num - 1) ?
1132 				     (u16)sizeoflast : (u16)HNS3_MAX_BD_SIZE);
1133 		hns3_set_txbd_baseinfo(&bdtp_fe_sc_vld_ra_ri,
1134 				       frag_end && (k == frag_buf_num - 1) ?
1135 						1 : 0);
1136 		desc->tx.bdtp_fe_sc_vld_ra_ri =
1137 				cpu_to_le16(bdtp_fe_sc_vld_ra_ri);
1138 
1139 		/* move ring pointer to next */
1140 		ring_ptr_move_fw(ring, next_to_use);
1141 
1142 		desc_cb = &ring->desc_cb[ring->next_to_use];
1143 		desc = &ring->desc[ring->next_to_use];
1144 	}
1145 
1146 	return 0;
1147 }
1148 
1149 static int hns3_nic_bd_num(struct sk_buff *skb)
1150 {
1151 	int size = skb_headlen(skb);
1152 	int i, bd_num;
1153 
1154 	/* if the total len is within the max bd limit */
1155 	if (likely(skb->len <= HNS3_MAX_BD_SIZE))
1156 		return skb_shinfo(skb)->nr_frags + 1;
1157 
1158 	bd_num = hns3_tx_bd_count(size);
1159 
1160 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1161 		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
1162 		int frag_bd_num;
1163 
1164 		size = skb_frag_size(frag);
1165 		frag_bd_num = hns3_tx_bd_count(size);
1166 
1167 		if (unlikely(frag_bd_num > HNS3_MAX_BD_PER_FRAG))
1168 			return -ENOMEM;
1169 
1170 		bd_num += frag_bd_num;
1171 	}
1172 
1173 	return bd_num;
1174 }
1175 
1176 static unsigned int hns3_gso_hdr_len(struct sk_buff *skb)
1177 {
1178 	if (!skb->encapsulation)
1179 		return skb_transport_offset(skb) + tcp_hdrlen(skb);
1180 
1181 	return skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb);
1182 }
1183 
1184 /* HW need every continuous 8 buffer data to be larger than MSS,
1185  * we simplify it by ensuring skb_headlen + the first continuous
1186  * 7 frags to to be larger than gso header len + mss, and the remaining
1187  * continuous 7 frags to be larger than MSS except the last 7 frags.
1188  */
1189 static bool hns3_skb_need_linearized(struct sk_buff *skb)
1190 {
1191 	int bd_limit = HNS3_MAX_BD_PER_FRAG - 1;
1192 	unsigned int tot_len = 0;
1193 	int i;
1194 
1195 	for (i = 0; i < bd_limit; i++)
1196 		tot_len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1197 
1198 	/* ensure headlen + the first 7 frags is greater than mss + header
1199 	 * and the first 7 frags is greater than mss.
1200 	 */
1201 	if (((tot_len + skb_headlen(skb)) < (skb_shinfo(skb)->gso_size +
1202 	    hns3_gso_hdr_len(skb))) || (tot_len < skb_shinfo(skb)->gso_size))
1203 		return true;
1204 
1205 	/* ensure the remaining continuous 7 buffer is greater than mss */
1206 	for (i = 0; i < (skb_shinfo(skb)->nr_frags - bd_limit - 1); i++) {
1207 		tot_len -= skb_frag_size(&skb_shinfo(skb)->frags[i]);
1208 		tot_len += skb_frag_size(&skb_shinfo(skb)->frags[i + bd_limit]);
1209 
1210 		if (tot_len < skb_shinfo(skb)->gso_size)
1211 			return true;
1212 	}
1213 
1214 	return false;
1215 }
1216 
1217 static int hns3_nic_maybe_stop_tx(struct hns3_enet_ring *ring,
1218 				  struct sk_buff **out_skb)
1219 {
1220 	struct sk_buff *skb = *out_skb;
1221 	int bd_num;
1222 
1223 	bd_num = hns3_nic_bd_num(skb);
1224 	if (bd_num < 0)
1225 		return bd_num;
1226 
1227 	if (unlikely(bd_num > HNS3_MAX_BD_PER_FRAG)) {
1228 		struct sk_buff *new_skb;
1229 
1230 		if (skb_is_gso(skb) && !hns3_skb_need_linearized(skb))
1231 			goto out;
1232 
1233 		bd_num = hns3_tx_bd_count(skb->len);
1234 		if (unlikely(ring_space(ring) < bd_num))
1235 			return -EBUSY;
1236 		/* manual split the send packet */
1237 		new_skb = skb_copy(skb, GFP_ATOMIC);
1238 		if (!new_skb)
1239 			return -ENOMEM;
1240 		dev_kfree_skb_any(skb);
1241 		*out_skb = new_skb;
1242 
1243 		u64_stats_update_begin(&ring->syncp);
1244 		ring->stats.tx_copy++;
1245 		u64_stats_update_end(&ring->syncp);
1246 	}
1247 
1248 out:
1249 	if (unlikely(ring_space(ring) < bd_num))
1250 		return -EBUSY;
1251 
1252 	return bd_num;
1253 }
1254 
1255 static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
1256 {
1257 	struct device *dev = ring_to_dev(ring);
1258 	unsigned int i;
1259 
1260 	for (i = 0; i < ring->desc_num; i++) {
1261 		/* check if this is where we started */
1262 		if (ring->next_to_use == next_to_use_orig)
1263 			break;
1264 
1265 		/* rollback one */
1266 		ring_ptr_move_bw(ring, next_to_use);
1267 
1268 		/* unmap the descriptor dma address */
1269 		if (ring->desc_cb[ring->next_to_use].type == DESC_TYPE_SKB)
1270 			dma_unmap_single(dev,
1271 					 ring->desc_cb[ring->next_to_use].dma,
1272 					ring->desc_cb[ring->next_to_use].length,
1273 					DMA_TO_DEVICE);
1274 		else if (ring->desc_cb[ring->next_to_use].length)
1275 			dma_unmap_page(dev,
1276 				       ring->desc_cb[ring->next_to_use].dma,
1277 				       ring->desc_cb[ring->next_to_use].length,
1278 				       DMA_TO_DEVICE);
1279 
1280 		ring->desc_cb[ring->next_to_use].length = 0;
1281 		ring->desc_cb[ring->next_to_use].dma = 0;
1282 	}
1283 }
1284 
1285 netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
1286 {
1287 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1288 	struct hns3_nic_ring_data *ring_data =
1289 		&tx_ring_data(priv, skb->queue_mapping);
1290 	struct hns3_enet_ring *ring = ring_data->ring;
1291 	struct netdev_queue *dev_queue;
1292 	struct skb_frag_struct *frag;
1293 	int next_to_use_head;
1294 	int buf_num;
1295 	int seg_num;
1296 	int size;
1297 	int ret;
1298 	int i;
1299 
1300 	/* Prefetch the data used later */
1301 	prefetch(skb->data);
1302 
1303 	buf_num = hns3_nic_maybe_stop_tx(ring, &skb);
1304 	if (unlikely(buf_num <= 0)) {
1305 		if (buf_num == -EBUSY) {
1306 			u64_stats_update_begin(&ring->syncp);
1307 			ring->stats.tx_busy++;
1308 			u64_stats_update_end(&ring->syncp);
1309 			goto out_net_tx_busy;
1310 		} else if (buf_num == -ENOMEM) {
1311 			u64_stats_update_begin(&ring->syncp);
1312 			ring->stats.sw_err_cnt++;
1313 			u64_stats_update_end(&ring->syncp);
1314 		}
1315 
1316 		if (net_ratelimit())
1317 			netdev_err(netdev, "xmit error: %d!\n", buf_num);
1318 
1319 		goto out_err_tx_ok;
1320 	}
1321 
1322 	/* No. of segments (plus a header) */
1323 	seg_num = skb_shinfo(skb)->nr_frags + 1;
1324 	/* Fill the first part */
1325 	size = skb_headlen(skb);
1326 
1327 	next_to_use_head = ring->next_to_use;
1328 
1329 	ret = hns3_fill_desc(ring, skb, size, seg_num == 1 ? 1 : 0,
1330 			     DESC_TYPE_SKB);
1331 	if (unlikely(ret))
1332 		goto fill_err;
1333 
1334 	/* Fill the fragments */
1335 	for (i = 1; i < seg_num; i++) {
1336 		frag = &skb_shinfo(skb)->frags[i - 1];
1337 		size = skb_frag_size(frag);
1338 
1339 		ret = hns3_fill_desc(ring, frag, size,
1340 				     seg_num - 1 == i ? 1 : 0,
1341 				     DESC_TYPE_PAGE);
1342 
1343 		if (unlikely(ret))
1344 			goto fill_err;
1345 	}
1346 
1347 	/* Complete translate all packets */
1348 	dev_queue = netdev_get_tx_queue(netdev, ring_data->queue_index);
1349 	netdev_tx_sent_queue(dev_queue, skb->len);
1350 
1351 	wmb(); /* Commit all data before submit */
1352 
1353 	hnae3_queue_xmit(ring->tqp, buf_num);
1354 
1355 	return NETDEV_TX_OK;
1356 
1357 fill_err:
1358 	hns3_clear_desc(ring, next_to_use_head);
1359 
1360 out_err_tx_ok:
1361 	dev_kfree_skb_any(skb);
1362 	return NETDEV_TX_OK;
1363 
1364 out_net_tx_busy:
1365 	netif_stop_subqueue(netdev, ring_data->queue_index);
1366 	smp_mb(); /* Commit all data before submit */
1367 
1368 	return NETDEV_TX_BUSY;
1369 }
1370 
1371 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
1372 {
1373 	struct hnae3_handle *h = hns3_get_handle(netdev);
1374 	struct sockaddr *mac_addr = p;
1375 	int ret;
1376 
1377 	if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1378 		return -EADDRNOTAVAIL;
1379 
1380 	if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
1381 		netdev_info(netdev, "already using mac address %pM\n",
1382 			    mac_addr->sa_data);
1383 		return 0;
1384 	}
1385 
1386 	ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
1387 	if (ret) {
1388 		netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
1389 		return ret;
1390 	}
1391 
1392 	ether_addr_copy(netdev->dev_addr, mac_addr->sa_data);
1393 
1394 	return 0;
1395 }
1396 
1397 static int hns3_nic_do_ioctl(struct net_device *netdev,
1398 			     struct ifreq *ifr, int cmd)
1399 {
1400 	struct hnae3_handle *h = hns3_get_handle(netdev);
1401 
1402 	if (!netif_running(netdev))
1403 		return -EINVAL;
1404 
1405 	if (!h->ae_algo->ops->do_ioctl)
1406 		return -EOPNOTSUPP;
1407 
1408 	return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
1409 }
1410 
1411 static int hns3_nic_set_features(struct net_device *netdev,
1412 				 netdev_features_t features)
1413 {
1414 	netdev_features_t changed = netdev->features ^ features;
1415 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1416 	struct hnae3_handle *h = priv->ae_handle;
1417 	bool enable;
1418 	int ret;
1419 
1420 	if (changed & (NETIF_F_GRO_HW) && h->ae_algo->ops->set_gro_en) {
1421 		enable = !!(features & NETIF_F_GRO_HW);
1422 		ret = h->ae_algo->ops->set_gro_en(h, enable);
1423 		if (ret)
1424 			return ret;
1425 	}
1426 
1427 	if ((changed & NETIF_F_HW_VLAN_CTAG_FILTER) &&
1428 	    h->ae_algo->ops->enable_vlan_filter) {
1429 		enable = !!(features & NETIF_F_HW_VLAN_CTAG_FILTER);
1430 		h->ae_algo->ops->enable_vlan_filter(h, enable);
1431 	}
1432 
1433 	if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
1434 	    h->ae_algo->ops->enable_hw_strip_rxvtag) {
1435 		enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
1436 		ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, enable);
1437 		if (ret)
1438 			return ret;
1439 	}
1440 
1441 	if ((changed & NETIF_F_NTUPLE) && h->ae_algo->ops->enable_fd) {
1442 		enable = !!(features & NETIF_F_NTUPLE);
1443 		h->ae_algo->ops->enable_fd(h, enable);
1444 	}
1445 
1446 	netdev->features = features;
1447 	return 0;
1448 }
1449 
1450 static void hns3_nic_get_stats64(struct net_device *netdev,
1451 				 struct rtnl_link_stats64 *stats)
1452 {
1453 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1454 	int queue_num = priv->ae_handle->kinfo.num_tqps;
1455 	struct hnae3_handle *handle = priv->ae_handle;
1456 	struct hns3_enet_ring *ring;
1457 	u64 rx_length_errors = 0;
1458 	u64 rx_crc_errors = 0;
1459 	u64 rx_multicast = 0;
1460 	unsigned int start;
1461 	u64 tx_errors = 0;
1462 	u64 rx_errors = 0;
1463 	unsigned int idx;
1464 	u64 tx_bytes = 0;
1465 	u64 rx_bytes = 0;
1466 	u64 tx_pkts = 0;
1467 	u64 rx_pkts = 0;
1468 	u64 tx_drop = 0;
1469 	u64 rx_drop = 0;
1470 
1471 	if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
1472 		return;
1473 
1474 	handle->ae_algo->ops->update_stats(handle, &netdev->stats);
1475 
1476 	for (idx = 0; idx < queue_num; idx++) {
1477 		/* fetch the tx stats */
1478 		ring = priv->ring_data[idx].ring;
1479 		do {
1480 			start = u64_stats_fetch_begin_irq(&ring->syncp);
1481 			tx_bytes += ring->stats.tx_bytes;
1482 			tx_pkts += ring->stats.tx_pkts;
1483 			tx_drop += ring->stats.sw_err_cnt;
1484 			tx_errors += ring->stats.sw_err_cnt;
1485 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1486 
1487 		/* fetch the rx stats */
1488 		ring = priv->ring_data[idx + queue_num].ring;
1489 		do {
1490 			start = u64_stats_fetch_begin_irq(&ring->syncp);
1491 			rx_bytes += ring->stats.rx_bytes;
1492 			rx_pkts += ring->stats.rx_pkts;
1493 			rx_drop += ring->stats.l2_err;
1494 			rx_errors += ring->stats.l2_err;
1495 			rx_errors += ring->stats.l3l4_csum_err;
1496 			rx_crc_errors += ring->stats.l2_err;
1497 			rx_multicast += ring->stats.rx_multicast;
1498 			rx_length_errors += ring->stats.err_pkt_len;
1499 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1500 	}
1501 
1502 	stats->tx_bytes = tx_bytes;
1503 	stats->tx_packets = tx_pkts;
1504 	stats->rx_bytes = rx_bytes;
1505 	stats->rx_packets = rx_pkts;
1506 
1507 	stats->rx_errors = rx_errors;
1508 	stats->multicast = rx_multicast;
1509 	stats->rx_length_errors = rx_length_errors;
1510 	stats->rx_crc_errors = rx_crc_errors;
1511 	stats->rx_missed_errors = netdev->stats.rx_missed_errors;
1512 
1513 	stats->tx_errors = tx_errors;
1514 	stats->rx_dropped = rx_drop;
1515 	stats->tx_dropped = tx_drop;
1516 	stats->collisions = netdev->stats.collisions;
1517 	stats->rx_over_errors = netdev->stats.rx_over_errors;
1518 	stats->rx_frame_errors = netdev->stats.rx_frame_errors;
1519 	stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
1520 	stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
1521 	stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
1522 	stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
1523 	stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
1524 	stats->tx_window_errors = netdev->stats.tx_window_errors;
1525 	stats->rx_compressed = netdev->stats.rx_compressed;
1526 	stats->tx_compressed = netdev->stats.tx_compressed;
1527 }
1528 
1529 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
1530 {
1531 	struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
1532 	u8 *prio_tc = mqprio_qopt->qopt.prio_tc_map;
1533 	struct hnae3_knic_private_info *kinfo;
1534 	u8 tc = mqprio_qopt->qopt.num_tc;
1535 	u16 mode = mqprio_qopt->mode;
1536 	u8 hw = mqprio_qopt->qopt.hw;
1537 	struct hnae3_handle *h;
1538 
1539 	if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
1540 	       mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
1541 		return -EOPNOTSUPP;
1542 
1543 	if (tc > HNAE3_MAX_TC)
1544 		return -EINVAL;
1545 
1546 	if (!netdev)
1547 		return -EINVAL;
1548 
1549 	h = hns3_get_handle(netdev);
1550 	kinfo = &h->kinfo;
1551 
1552 	return (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
1553 		kinfo->dcb_ops->setup_tc(h, tc, prio_tc) : -EOPNOTSUPP;
1554 }
1555 
1556 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
1557 			     void *type_data)
1558 {
1559 	if (type != TC_SETUP_QDISC_MQPRIO)
1560 		return -EOPNOTSUPP;
1561 
1562 	return hns3_setup_tc(dev, type_data);
1563 }
1564 
1565 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
1566 				__be16 proto, u16 vid)
1567 {
1568 	struct hnae3_handle *h = hns3_get_handle(netdev);
1569 	int ret = -EIO;
1570 
1571 	if (h->ae_algo->ops->set_vlan_filter)
1572 		ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
1573 
1574 	return ret;
1575 }
1576 
1577 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
1578 				 __be16 proto, u16 vid)
1579 {
1580 	struct hnae3_handle *h = hns3_get_handle(netdev);
1581 	int ret = -EIO;
1582 
1583 	if (h->ae_algo->ops->set_vlan_filter)
1584 		ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
1585 
1586 	return ret;
1587 }
1588 
1589 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
1590 				u8 qos, __be16 vlan_proto)
1591 {
1592 	struct hnae3_handle *h = hns3_get_handle(netdev);
1593 	int ret = -EIO;
1594 
1595 	if (h->ae_algo->ops->set_vf_vlan_filter)
1596 		ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
1597 							  qos, vlan_proto);
1598 
1599 	return ret;
1600 }
1601 
1602 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
1603 {
1604 	struct hnae3_handle *h = hns3_get_handle(netdev);
1605 	int ret;
1606 
1607 	if (hns3_nic_resetting(netdev))
1608 		return -EBUSY;
1609 
1610 	if (!h->ae_algo->ops->set_mtu)
1611 		return -EOPNOTSUPP;
1612 
1613 	ret = h->ae_algo->ops->set_mtu(h, new_mtu);
1614 	if (ret)
1615 		netdev_err(netdev, "failed to change MTU in hardware %d\n",
1616 			   ret);
1617 	else
1618 		netdev->mtu = new_mtu;
1619 
1620 	return ret;
1621 }
1622 
1623 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
1624 {
1625 	struct hns3_nic_priv *priv = netdev_priv(ndev);
1626 	struct hnae3_handle *h = hns3_get_handle(ndev);
1627 	struct hns3_enet_ring *tx_ring = NULL;
1628 	struct napi_struct *napi;
1629 	int timeout_queue = 0;
1630 	int hw_head, hw_tail;
1631 	int fbd_num, fbd_oft;
1632 	int ebd_num, ebd_oft;
1633 	int bd_num, bd_err;
1634 	int ring_en, tc;
1635 	int i;
1636 
1637 	/* Find the stopped queue the same way the stack does */
1638 	for (i = 0; i < ndev->num_tx_queues; i++) {
1639 		struct netdev_queue *q;
1640 		unsigned long trans_start;
1641 
1642 		q = netdev_get_tx_queue(ndev, i);
1643 		trans_start = q->trans_start;
1644 		if (netif_xmit_stopped(q) &&
1645 		    time_after(jiffies,
1646 			       (trans_start + ndev->watchdog_timeo))) {
1647 			timeout_queue = i;
1648 			break;
1649 		}
1650 	}
1651 
1652 	if (i == ndev->num_tx_queues) {
1653 		netdev_info(ndev,
1654 			    "no netdev TX timeout queue found, timeout count: %llu\n",
1655 			    priv->tx_timeout_count);
1656 		return false;
1657 	}
1658 
1659 	priv->tx_timeout_count++;
1660 
1661 	tx_ring = priv->ring_data[timeout_queue].ring;
1662 	napi = &tx_ring->tqp_vector->napi;
1663 
1664 	netdev_info(ndev,
1665 		    "tx_timeout count: %llu, queue id: %d, SW_NTU: 0x%x, SW_NTC: 0x%x, napi state: %lu\n",
1666 		    priv->tx_timeout_count, timeout_queue, tx_ring->next_to_use,
1667 		    tx_ring->next_to_clean, napi->state);
1668 
1669 	netdev_info(ndev,
1670 		    "tx_pkts: %llu, tx_bytes: %llu, io_err_cnt: %llu, sw_err_cnt: %llu\n",
1671 		    tx_ring->stats.tx_pkts, tx_ring->stats.tx_bytes,
1672 		    tx_ring->stats.io_err_cnt, tx_ring->stats.sw_err_cnt);
1673 
1674 	netdev_info(ndev,
1675 		    "seg_pkt_cnt: %llu, tx_err_cnt: %llu, restart_queue: %llu, tx_busy: %llu\n",
1676 		    tx_ring->stats.seg_pkt_cnt, tx_ring->stats.tx_err_cnt,
1677 		    tx_ring->stats.restart_queue, tx_ring->stats.tx_busy);
1678 
1679 	/* When mac received many pause frames continuous, it's unable to send
1680 	 * packets, which may cause tx timeout
1681 	 */
1682 	if (h->ae_algo->ops->update_stats &&
1683 	    h->ae_algo->ops->get_mac_pause_stats) {
1684 		u64 tx_pause_cnt, rx_pause_cnt;
1685 
1686 		h->ae_algo->ops->update_stats(h, &ndev->stats);
1687 		h->ae_algo->ops->get_mac_pause_stats(h, &tx_pause_cnt,
1688 						     &rx_pause_cnt);
1689 		netdev_info(ndev, "tx_pause_cnt: %llu, rx_pause_cnt: %llu\n",
1690 			    tx_pause_cnt, rx_pause_cnt);
1691 	}
1692 
1693 	hw_head = readl_relaxed(tx_ring->tqp->io_base +
1694 				HNS3_RING_TX_RING_HEAD_REG);
1695 	hw_tail = readl_relaxed(tx_ring->tqp->io_base +
1696 				HNS3_RING_TX_RING_TAIL_REG);
1697 	fbd_num = readl_relaxed(tx_ring->tqp->io_base +
1698 				HNS3_RING_TX_RING_FBDNUM_REG);
1699 	fbd_oft = readl_relaxed(tx_ring->tqp->io_base +
1700 				HNS3_RING_TX_RING_OFFSET_REG);
1701 	ebd_num = readl_relaxed(tx_ring->tqp->io_base +
1702 				HNS3_RING_TX_RING_EBDNUM_REG);
1703 	ebd_oft = readl_relaxed(tx_ring->tqp->io_base +
1704 				HNS3_RING_TX_RING_EBD_OFFSET_REG);
1705 	bd_num = readl_relaxed(tx_ring->tqp->io_base +
1706 			       HNS3_RING_TX_RING_BD_NUM_REG);
1707 	bd_err = readl_relaxed(tx_ring->tqp->io_base +
1708 			       HNS3_RING_TX_RING_BD_ERR_REG);
1709 	ring_en = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_EN_REG);
1710 	tc = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_TX_RING_TC_REG);
1711 
1712 	netdev_info(ndev,
1713 		    "BD_NUM: 0x%x HW_HEAD: 0x%x, HW_TAIL: 0x%x, BD_ERR: 0x%x, INT: 0x%x\n",
1714 		    bd_num, hw_head, hw_tail, bd_err,
1715 		    readl(tx_ring->tqp_vector->mask_addr));
1716 	netdev_info(ndev,
1717 		    "RING_EN: 0x%x, TC: 0x%x, FBD_NUM: 0x%x FBD_OFT: 0x%x, EBD_NUM: 0x%x, EBD_OFT: 0x%x\n",
1718 		    ring_en, tc, fbd_num, fbd_oft, ebd_num, ebd_oft);
1719 
1720 	return true;
1721 }
1722 
1723 static void hns3_nic_net_timeout(struct net_device *ndev)
1724 {
1725 	struct hns3_nic_priv *priv = netdev_priv(ndev);
1726 	struct hnae3_handle *h = priv->ae_handle;
1727 
1728 	if (!hns3_get_tx_timeo_queue_info(ndev))
1729 		return;
1730 
1731 	/* request the reset, and let the hclge to determine
1732 	 * which reset level should be done
1733 	 */
1734 	if (h->ae_algo->ops->reset_event)
1735 		h->ae_algo->ops->reset_event(h->pdev, h);
1736 }
1737 
1738 #ifdef CONFIG_RFS_ACCEL
1739 static int hns3_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
1740 			      u16 rxq_index, u32 flow_id)
1741 {
1742 	struct hnae3_handle *h = hns3_get_handle(dev);
1743 	struct flow_keys fkeys;
1744 
1745 	if (!h->ae_algo->ops->add_arfs_entry)
1746 		return -EOPNOTSUPP;
1747 
1748 	if (skb->encapsulation)
1749 		return -EPROTONOSUPPORT;
1750 
1751 	if (!skb_flow_dissect_flow_keys(skb, &fkeys, 0))
1752 		return -EPROTONOSUPPORT;
1753 
1754 	if ((fkeys.basic.n_proto != htons(ETH_P_IP) &&
1755 	     fkeys.basic.n_proto != htons(ETH_P_IPV6)) ||
1756 	    (fkeys.basic.ip_proto != IPPROTO_TCP &&
1757 	     fkeys.basic.ip_proto != IPPROTO_UDP))
1758 		return -EPROTONOSUPPORT;
1759 
1760 	return h->ae_algo->ops->add_arfs_entry(h, rxq_index, flow_id, &fkeys);
1761 }
1762 #endif
1763 
1764 static const struct net_device_ops hns3_nic_netdev_ops = {
1765 	.ndo_open		= hns3_nic_net_open,
1766 	.ndo_stop		= hns3_nic_net_stop,
1767 	.ndo_start_xmit		= hns3_nic_net_xmit,
1768 	.ndo_tx_timeout		= hns3_nic_net_timeout,
1769 	.ndo_set_mac_address	= hns3_nic_net_set_mac_address,
1770 	.ndo_do_ioctl		= hns3_nic_do_ioctl,
1771 	.ndo_change_mtu		= hns3_nic_change_mtu,
1772 	.ndo_set_features	= hns3_nic_set_features,
1773 	.ndo_get_stats64	= hns3_nic_get_stats64,
1774 	.ndo_setup_tc		= hns3_nic_setup_tc,
1775 	.ndo_set_rx_mode	= hns3_nic_set_rx_mode,
1776 	.ndo_vlan_rx_add_vid	= hns3_vlan_rx_add_vid,
1777 	.ndo_vlan_rx_kill_vid	= hns3_vlan_rx_kill_vid,
1778 	.ndo_set_vf_vlan	= hns3_ndo_set_vf_vlan,
1779 #ifdef CONFIG_RFS_ACCEL
1780 	.ndo_rx_flow_steer	= hns3_rx_flow_steer,
1781 #endif
1782 
1783 };
1784 
1785 bool hns3_is_phys_func(struct pci_dev *pdev)
1786 {
1787 	u32 dev_id = pdev->device;
1788 
1789 	switch (dev_id) {
1790 	case HNAE3_DEV_ID_GE:
1791 	case HNAE3_DEV_ID_25GE:
1792 	case HNAE3_DEV_ID_25GE_RDMA:
1793 	case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
1794 	case HNAE3_DEV_ID_50GE_RDMA:
1795 	case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
1796 	case HNAE3_DEV_ID_100G_RDMA_MACSEC:
1797 		return true;
1798 	case HNAE3_DEV_ID_100G_VF:
1799 	case HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF:
1800 		return false;
1801 	default:
1802 		dev_warn(&pdev->dev, "un-recognized pci device-id %d",
1803 			 dev_id);
1804 	}
1805 
1806 	return false;
1807 }
1808 
1809 static void hns3_disable_sriov(struct pci_dev *pdev)
1810 {
1811 	/* If our VFs are assigned we cannot shut down SR-IOV
1812 	 * without causing issues, so just leave the hardware
1813 	 * available but disabled
1814 	 */
1815 	if (pci_vfs_assigned(pdev)) {
1816 		dev_warn(&pdev->dev,
1817 			 "disabling driver while VFs are assigned\n");
1818 		return;
1819 	}
1820 
1821 	pci_disable_sriov(pdev);
1822 }
1823 
1824 static void hns3_get_dev_capability(struct pci_dev *pdev,
1825 				    struct hnae3_ae_dev *ae_dev)
1826 {
1827 	if (pdev->revision >= 0x21) {
1828 		hnae3_set_bit(ae_dev->flag, HNAE3_DEV_SUPPORT_FD_B, 1);
1829 		hnae3_set_bit(ae_dev->flag, HNAE3_DEV_SUPPORT_GRO_B, 1);
1830 	}
1831 }
1832 
1833 /* hns3_probe - Device initialization routine
1834  * @pdev: PCI device information struct
1835  * @ent: entry in hns3_pci_tbl
1836  *
1837  * hns3_probe initializes a PF identified by a pci_dev structure.
1838  * The OS initialization, configuring of the PF private structure,
1839  * and a hardware reset occur.
1840  *
1841  * Returns 0 on success, negative on failure
1842  */
1843 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1844 {
1845 	struct hnae3_ae_dev *ae_dev;
1846 	int ret;
1847 
1848 	ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev), GFP_KERNEL);
1849 	if (!ae_dev) {
1850 		ret = -ENOMEM;
1851 		return ret;
1852 	}
1853 
1854 	ae_dev->pdev = pdev;
1855 	ae_dev->flag = ent->driver_data;
1856 	ae_dev->reset_type = HNAE3_NONE_RESET;
1857 	hns3_get_dev_capability(pdev, ae_dev);
1858 	pci_set_drvdata(pdev, ae_dev);
1859 
1860 	ret = hnae3_register_ae_dev(ae_dev);
1861 	if (ret) {
1862 		devm_kfree(&pdev->dev, ae_dev);
1863 		pci_set_drvdata(pdev, NULL);
1864 	}
1865 
1866 	return ret;
1867 }
1868 
1869 /* hns3_remove - Device removal routine
1870  * @pdev: PCI device information struct
1871  */
1872 static void hns3_remove(struct pci_dev *pdev)
1873 {
1874 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1875 
1876 	if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
1877 		hns3_disable_sriov(pdev);
1878 
1879 	hnae3_unregister_ae_dev(ae_dev);
1880 	pci_set_drvdata(pdev, NULL);
1881 }
1882 
1883 /**
1884  * hns3_pci_sriov_configure
1885  * @pdev: pointer to a pci_dev structure
1886  * @num_vfs: number of VFs to allocate
1887  *
1888  * Enable or change the number of VFs. Called when the user updates the number
1889  * of VFs in sysfs.
1890  **/
1891 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
1892 {
1893 	int ret;
1894 
1895 	if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
1896 		dev_warn(&pdev->dev, "Can not config SRIOV\n");
1897 		return -EINVAL;
1898 	}
1899 
1900 	if (num_vfs) {
1901 		ret = pci_enable_sriov(pdev, num_vfs);
1902 		if (ret)
1903 			dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
1904 		else
1905 			return num_vfs;
1906 	} else if (!pci_vfs_assigned(pdev)) {
1907 		pci_disable_sriov(pdev);
1908 	} else {
1909 		dev_warn(&pdev->dev,
1910 			 "Unable to free VFs because some are assigned to VMs.\n");
1911 	}
1912 
1913 	return 0;
1914 }
1915 
1916 static void hns3_shutdown(struct pci_dev *pdev)
1917 {
1918 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1919 
1920 	hnae3_unregister_ae_dev(ae_dev);
1921 	devm_kfree(&pdev->dev, ae_dev);
1922 	pci_set_drvdata(pdev, NULL);
1923 
1924 	if (system_state == SYSTEM_POWER_OFF)
1925 		pci_set_power_state(pdev, PCI_D3hot);
1926 }
1927 
1928 static pci_ers_result_t hns3_error_detected(struct pci_dev *pdev,
1929 					    pci_channel_state_t state)
1930 {
1931 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1932 	pci_ers_result_t ret;
1933 
1934 	dev_info(&pdev->dev, "PCI error detected, state(=%d)!!\n", state);
1935 
1936 	if (state == pci_channel_io_perm_failure)
1937 		return PCI_ERS_RESULT_DISCONNECT;
1938 
1939 	if (!ae_dev || !ae_dev->ops) {
1940 		dev_err(&pdev->dev,
1941 			"Can't recover - error happened before device initialized\n");
1942 		return PCI_ERS_RESULT_NONE;
1943 	}
1944 
1945 	if (ae_dev->ops->handle_hw_ras_error)
1946 		ret = ae_dev->ops->handle_hw_ras_error(ae_dev);
1947 	else
1948 		return PCI_ERS_RESULT_NONE;
1949 
1950 	return ret;
1951 }
1952 
1953 static pci_ers_result_t hns3_slot_reset(struct pci_dev *pdev)
1954 {
1955 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1956 	const struct hnae3_ae_ops *ops;
1957 	enum hnae3_reset_type reset_type;
1958 	struct device *dev = &pdev->dev;
1959 
1960 	if (!ae_dev || !ae_dev->ops)
1961 		return PCI_ERS_RESULT_NONE;
1962 
1963 	ops = ae_dev->ops;
1964 	/* request the reset */
1965 	if (ops->reset_event) {
1966 		if (ae_dev->hw_err_reset_req) {
1967 			reset_type = ops->get_reset_level(ae_dev,
1968 						&ae_dev->hw_err_reset_req);
1969 			ops->set_default_reset_request(ae_dev, reset_type);
1970 			dev_info(dev, "requesting reset due to PCI error\n");
1971 			ops->reset_event(pdev, NULL);
1972 		}
1973 
1974 		return PCI_ERS_RESULT_RECOVERED;
1975 	}
1976 
1977 	return PCI_ERS_RESULT_DISCONNECT;
1978 }
1979 
1980 static void hns3_reset_prepare(struct pci_dev *pdev)
1981 {
1982 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1983 
1984 	dev_info(&pdev->dev, "hns3 flr prepare\n");
1985 	if (ae_dev && ae_dev->ops && ae_dev->ops->flr_prepare)
1986 		ae_dev->ops->flr_prepare(ae_dev);
1987 }
1988 
1989 static void hns3_reset_done(struct pci_dev *pdev)
1990 {
1991 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
1992 
1993 	dev_info(&pdev->dev, "hns3 flr done\n");
1994 	if (ae_dev && ae_dev->ops && ae_dev->ops->flr_done)
1995 		ae_dev->ops->flr_done(ae_dev);
1996 }
1997 
1998 static const struct pci_error_handlers hns3_err_handler = {
1999 	.error_detected = hns3_error_detected,
2000 	.slot_reset     = hns3_slot_reset,
2001 	.reset_prepare	= hns3_reset_prepare,
2002 	.reset_done	= hns3_reset_done,
2003 };
2004 
2005 static struct pci_driver hns3_driver = {
2006 	.name     = hns3_driver_name,
2007 	.id_table = hns3_pci_tbl,
2008 	.probe    = hns3_probe,
2009 	.remove   = hns3_remove,
2010 	.shutdown = hns3_shutdown,
2011 	.sriov_configure = hns3_pci_sriov_configure,
2012 	.err_handler    = &hns3_err_handler,
2013 };
2014 
2015 /* set default feature to hns3 */
2016 static void hns3_set_default_feature(struct net_device *netdev)
2017 {
2018 	struct hnae3_handle *h = hns3_get_handle(netdev);
2019 	struct pci_dev *pdev = h->pdev;
2020 
2021 	netdev->priv_flags |= IFF_UNICAST_FLT;
2022 
2023 	netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2024 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2025 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2026 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2027 		NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
2028 
2029 	netdev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
2030 
2031 	netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
2032 
2033 	netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2034 		NETIF_F_HW_VLAN_CTAG_FILTER |
2035 		NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
2036 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2037 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2038 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2039 		NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
2040 
2041 	netdev->vlan_features |=
2042 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
2043 		NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO |
2044 		NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2045 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2046 		NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
2047 
2048 	netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2049 		NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
2050 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2051 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2052 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2053 		NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_SCTP_CRC;
2054 
2055 	if (pdev->revision >= 0x21) {
2056 		netdev->hw_features |= NETIF_F_GRO_HW;
2057 		netdev->features |= NETIF_F_GRO_HW;
2058 
2059 		if (!(h->flags & HNAE3_SUPPORT_VF)) {
2060 			netdev->hw_features |= NETIF_F_NTUPLE;
2061 			netdev->features |= NETIF_F_NTUPLE;
2062 		}
2063 	}
2064 }
2065 
2066 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
2067 			     struct hns3_desc_cb *cb)
2068 {
2069 	unsigned int order = hnae3_page_order(ring);
2070 	struct page *p;
2071 
2072 	p = dev_alloc_pages(order);
2073 	if (!p)
2074 		return -ENOMEM;
2075 
2076 	cb->priv = p;
2077 	cb->page_offset = 0;
2078 	cb->reuse_flag = 0;
2079 	cb->buf  = page_address(p);
2080 	cb->length = hnae3_page_size(ring);
2081 	cb->type = DESC_TYPE_PAGE;
2082 
2083 	return 0;
2084 }
2085 
2086 static void hns3_free_buffer(struct hns3_enet_ring *ring,
2087 			     struct hns3_desc_cb *cb)
2088 {
2089 	if (cb->type == DESC_TYPE_SKB)
2090 		dev_kfree_skb_any((struct sk_buff *)cb->priv);
2091 	else if (!HNAE3_IS_TX_RING(ring))
2092 		put_page((struct page *)cb->priv);
2093 	memset(cb, 0, sizeof(*cb));
2094 }
2095 
2096 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
2097 {
2098 	cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
2099 			       cb->length, ring_to_dma_dir(ring));
2100 
2101 	if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
2102 		return -EIO;
2103 
2104 	return 0;
2105 }
2106 
2107 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
2108 			      struct hns3_desc_cb *cb)
2109 {
2110 	if (cb->type == DESC_TYPE_SKB)
2111 		dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
2112 				 ring_to_dma_dir(ring));
2113 	else if (cb->length)
2114 		dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
2115 			       ring_to_dma_dir(ring));
2116 }
2117 
2118 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
2119 {
2120 	hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2121 	ring->desc[i].addr = 0;
2122 }
2123 
2124 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i)
2125 {
2126 	struct hns3_desc_cb *cb = &ring->desc_cb[i];
2127 
2128 	if (!ring->desc_cb[i].dma)
2129 		return;
2130 
2131 	hns3_buffer_detach(ring, i);
2132 	hns3_free_buffer(ring, cb);
2133 }
2134 
2135 static void hns3_free_buffers(struct hns3_enet_ring *ring)
2136 {
2137 	int i;
2138 
2139 	for (i = 0; i < ring->desc_num; i++)
2140 		hns3_free_buffer_detach(ring, i);
2141 }
2142 
2143 /* free desc along with its attached buffer */
2144 static void hns3_free_desc(struct hns3_enet_ring *ring)
2145 {
2146 	int size = ring->desc_num * sizeof(ring->desc[0]);
2147 
2148 	hns3_free_buffers(ring);
2149 
2150 	if (ring->desc) {
2151 		dma_free_coherent(ring_to_dev(ring), size,
2152 				  ring->desc, ring->desc_dma_addr);
2153 		ring->desc = NULL;
2154 	}
2155 }
2156 
2157 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
2158 {
2159 	int size = ring->desc_num * sizeof(ring->desc[0]);
2160 
2161 	ring->desc = dma_alloc_coherent(ring_to_dev(ring), size,
2162 					&ring->desc_dma_addr, GFP_KERNEL);
2163 	if (!ring->desc)
2164 		return -ENOMEM;
2165 
2166 	return 0;
2167 }
2168 
2169 static int hns3_reserve_buffer_map(struct hns3_enet_ring *ring,
2170 				   struct hns3_desc_cb *cb)
2171 {
2172 	int ret;
2173 
2174 	ret = hns3_alloc_buffer(ring, cb);
2175 	if (ret)
2176 		goto out;
2177 
2178 	ret = hns3_map_buffer(ring, cb);
2179 	if (ret)
2180 		goto out_with_buf;
2181 
2182 	return 0;
2183 
2184 out_with_buf:
2185 	hns3_free_buffer(ring, cb);
2186 out:
2187 	return ret;
2188 }
2189 
2190 static int hns3_alloc_buffer_attach(struct hns3_enet_ring *ring, int i)
2191 {
2192 	int ret = hns3_reserve_buffer_map(ring, &ring->desc_cb[i]);
2193 
2194 	if (ret)
2195 		return ret;
2196 
2197 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2198 
2199 	return 0;
2200 }
2201 
2202 /* Allocate memory for raw pkg, and map with dma */
2203 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
2204 {
2205 	int i, j, ret;
2206 
2207 	for (i = 0; i < ring->desc_num; i++) {
2208 		ret = hns3_alloc_buffer_attach(ring, i);
2209 		if (ret)
2210 			goto out_buffer_fail;
2211 	}
2212 
2213 	return 0;
2214 
2215 out_buffer_fail:
2216 	for (j = i - 1; j >= 0; j--)
2217 		hns3_free_buffer_detach(ring, j);
2218 	return ret;
2219 }
2220 
2221 /* detach a in-used buffer and replace with a reserved one */
2222 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
2223 				struct hns3_desc_cb *res_cb)
2224 {
2225 	hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2226 	ring->desc_cb[i] = *res_cb;
2227 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2228 	ring->desc[i].rx.bd_base_info = 0;
2229 }
2230 
2231 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
2232 {
2233 	ring->desc_cb[i].reuse_flag = 0;
2234 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
2235 					 ring->desc_cb[i].page_offset);
2236 	ring->desc[i].rx.bd_base_info = 0;
2237 }
2238 
2239 static void hns3_nic_reclaim_desc(struct hns3_enet_ring *ring, int head,
2240 				  int *bytes, int *pkts)
2241 {
2242 	int ntc = ring->next_to_clean;
2243 	struct hns3_desc_cb *desc_cb;
2244 
2245 	while (head != ntc) {
2246 		desc_cb = &ring->desc_cb[ntc];
2247 		(*pkts) += (desc_cb->type == DESC_TYPE_SKB);
2248 		(*bytes) += desc_cb->length;
2249 		/* desc_cb will be cleaned, after hnae3_free_buffer_detach */
2250 		hns3_free_buffer_detach(ring, ntc);
2251 
2252 		if (++ntc == ring->desc_num)
2253 			ntc = 0;
2254 
2255 		/* Issue prefetch for next Tx descriptor */
2256 		prefetch(&ring->desc_cb[ntc]);
2257 	}
2258 
2259 	/* This smp_store_release() pairs with smp_load_acquire() in
2260 	 * ring_space called by hns3_nic_net_xmit.
2261 	 */
2262 	smp_store_release(&ring->next_to_clean, ntc);
2263 }
2264 
2265 static int is_valid_clean_head(struct hns3_enet_ring *ring, int h)
2266 {
2267 	int u = ring->next_to_use;
2268 	int c = ring->next_to_clean;
2269 
2270 	if (unlikely(h > ring->desc_num))
2271 		return 0;
2272 
2273 	return u > c ? (h > c && h <= u) : (h > c || h <= u);
2274 }
2275 
2276 void hns3_clean_tx_ring(struct hns3_enet_ring *ring)
2277 {
2278 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2279 	struct hns3_nic_priv *priv = netdev_priv(netdev);
2280 	struct netdev_queue *dev_queue;
2281 	int bytes, pkts;
2282 	int head;
2283 
2284 	head = readl_relaxed(ring->tqp->io_base + HNS3_RING_TX_RING_HEAD_REG);
2285 	rmb(); /* Make sure head is ready before touch any data */
2286 
2287 	if (is_ring_empty(ring) || head == ring->next_to_clean)
2288 		return; /* no data to poll */
2289 
2290 	if (unlikely(!is_valid_clean_head(ring, head))) {
2291 		netdev_err(netdev, "wrong head (%d, %d-%d)\n", head,
2292 			   ring->next_to_use, ring->next_to_clean);
2293 
2294 		u64_stats_update_begin(&ring->syncp);
2295 		ring->stats.io_err_cnt++;
2296 		u64_stats_update_end(&ring->syncp);
2297 		return;
2298 	}
2299 
2300 	bytes = 0;
2301 	pkts = 0;
2302 	hns3_nic_reclaim_desc(ring, head, &bytes, &pkts);
2303 
2304 	ring->tqp_vector->tx_group.total_bytes += bytes;
2305 	ring->tqp_vector->tx_group.total_packets += pkts;
2306 
2307 	u64_stats_update_begin(&ring->syncp);
2308 	ring->stats.tx_bytes += bytes;
2309 	ring->stats.tx_pkts += pkts;
2310 	u64_stats_update_end(&ring->syncp);
2311 
2312 	dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
2313 	netdev_tx_completed_queue(dev_queue, pkts, bytes);
2314 
2315 	if (unlikely(pkts && netif_carrier_ok(netdev) &&
2316 		     (ring_space(ring) > HNS3_MAX_BD_PER_PKT))) {
2317 		/* Make sure that anybody stopping the queue after this
2318 		 * sees the new next_to_clean.
2319 		 */
2320 		smp_mb();
2321 		if (netif_tx_queue_stopped(dev_queue) &&
2322 		    !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
2323 			netif_tx_wake_queue(dev_queue);
2324 			ring->stats.restart_queue++;
2325 		}
2326 	}
2327 }
2328 
2329 static int hns3_desc_unused(struct hns3_enet_ring *ring)
2330 {
2331 	int ntc = ring->next_to_clean;
2332 	int ntu = ring->next_to_use;
2333 
2334 	return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
2335 }
2336 
2337 static void hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring,
2338 				      int cleand_count)
2339 {
2340 	struct hns3_desc_cb *desc_cb;
2341 	struct hns3_desc_cb res_cbs;
2342 	int i, ret;
2343 
2344 	for (i = 0; i < cleand_count; i++) {
2345 		desc_cb = &ring->desc_cb[ring->next_to_use];
2346 		if (desc_cb->reuse_flag) {
2347 			u64_stats_update_begin(&ring->syncp);
2348 			ring->stats.reuse_pg_cnt++;
2349 			u64_stats_update_end(&ring->syncp);
2350 
2351 			hns3_reuse_buffer(ring, ring->next_to_use);
2352 		} else {
2353 			ret = hns3_reserve_buffer_map(ring, &res_cbs);
2354 			if (ret) {
2355 				u64_stats_update_begin(&ring->syncp);
2356 				ring->stats.sw_err_cnt++;
2357 				u64_stats_update_end(&ring->syncp);
2358 
2359 				netdev_err(ring->tqp->handle->kinfo.netdev,
2360 					   "hnae reserve buffer map failed.\n");
2361 				break;
2362 			}
2363 			hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
2364 
2365 			u64_stats_update_begin(&ring->syncp);
2366 			ring->stats.non_reuse_pg++;
2367 			u64_stats_update_end(&ring->syncp);
2368 		}
2369 
2370 		ring_ptr_move_fw(ring, next_to_use);
2371 	}
2372 
2373 	wmb(); /* Make all data has been write before submit */
2374 	writel_relaxed(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
2375 }
2376 
2377 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
2378 				struct hns3_enet_ring *ring, int pull_len,
2379 				struct hns3_desc_cb *desc_cb)
2380 {
2381 	struct hns3_desc *desc = &ring->desc[ring->next_to_clean];
2382 	int size = le16_to_cpu(desc->rx.size);
2383 	u32 truesize = hnae3_buf_size(ring);
2384 
2385 	skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
2386 			size - pull_len, truesize);
2387 
2388 	/* Avoid re-using remote pages, or the stack is still using the page
2389 	 * when page_offset rollback to zero, flag default unreuse
2390 	 */
2391 	if (unlikely(page_to_nid(desc_cb->priv) != numa_mem_id()) ||
2392 	    (!desc_cb->page_offset && page_count(desc_cb->priv) > 1))
2393 		return;
2394 
2395 	/* Move offset up to the next cache line */
2396 	desc_cb->page_offset += truesize;
2397 
2398 	if (desc_cb->page_offset + truesize <= hnae3_page_size(ring)) {
2399 		desc_cb->reuse_flag = 1;
2400 		/* Bump ref count on page before it is given */
2401 		get_page(desc_cb->priv);
2402 	} else if (page_count(desc_cb->priv) == 1) {
2403 		desc_cb->reuse_flag = 1;
2404 		desc_cb->page_offset = 0;
2405 		get_page(desc_cb->priv);
2406 	}
2407 }
2408 
2409 static int hns3_gro_complete(struct sk_buff *skb, u32 l234info)
2410 {
2411 	__be16 type = skb->protocol;
2412 	struct tcphdr *th;
2413 	int depth = 0;
2414 
2415 	while (eth_type_vlan(type)) {
2416 		struct vlan_hdr *vh;
2417 
2418 		if ((depth + VLAN_HLEN) > skb_headlen(skb))
2419 			return -EFAULT;
2420 
2421 		vh = (struct vlan_hdr *)(skb->data + depth);
2422 		type = vh->h_vlan_encapsulated_proto;
2423 		depth += VLAN_HLEN;
2424 	}
2425 
2426 	skb_set_network_header(skb, depth);
2427 
2428 	if (type == htons(ETH_P_IP)) {
2429 		const struct iphdr *iph = ip_hdr(skb);
2430 
2431 		depth += sizeof(struct iphdr);
2432 		skb_set_transport_header(skb, depth);
2433 		th = tcp_hdr(skb);
2434 		th->check = ~tcp_v4_check(skb->len - depth, iph->saddr,
2435 					  iph->daddr, 0);
2436 	} else if (type == htons(ETH_P_IPV6)) {
2437 		const struct ipv6hdr *iph = ipv6_hdr(skb);
2438 
2439 		depth += sizeof(struct ipv6hdr);
2440 		skb_set_transport_header(skb, depth);
2441 		th = tcp_hdr(skb);
2442 		th->check = ~tcp_v6_check(skb->len - depth, &iph->saddr,
2443 					  &iph->daddr, 0);
2444 	} else {
2445 		netdev_err(skb->dev,
2446 			   "Error: FW GRO supports only IPv4/IPv6, not 0x%04x, depth: %d\n",
2447 			   be16_to_cpu(type), depth);
2448 		return -EFAULT;
2449 	}
2450 
2451 	skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
2452 	if (th->cwr)
2453 		skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
2454 
2455 	if (l234info & BIT(HNS3_RXD_GRO_FIXID_B))
2456 		skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_FIXEDID;
2457 
2458 	skb->csum_start = (unsigned char *)th - skb->head;
2459 	skb->csum_offset = offsetof(struct tcphdr, check);
2460 	skb->ip_summed = CHECKSUM_PARTIAL;
2461 	return 0;
2462 }
2463 
2464 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
2465 			     u32 l234info, u32 bd_base_info, u32 ol_info)
2466 {
2467 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2468 	int l3_type, l4_type;
2469 	int ol4_type;
2470 
2471 	skb->ip_summed = CHECKSUM_NONE;
2472 
2473 	skb_checksum_none_assert(skb);
2474 
2475 	if (!(netdev->features & NETIF_F_RXCSUM))
2476 		return;
2477 
2478 	/* check if hardware has done checksum */
2479 	if (!(bd_base_info & BIT(HNS3_RXD_L3L4P_B)))
2480 		return;
2481 
2482 	if (unlikely(l234info & (BIT(HNS3_RXD_L3E_B) | BIT(HNS3_RXD_L4E_B) |
2483 				 BIT(HNS3_RXD_OL3E_B) |
2484 				 BIT(HNS3_RXD_OL4E_B)))) {
2485 		u64_stats_update_begin(&ring->syncp);
2486 		ring->stats.l3l4_csum_err++;
2487 		u64_stats_update_end(&ring->syncp);
2488 
2489 		return;
2490 	}
2491 
2492 	ol4_type = hnae3_get_field(ol_info, HNS3_RXD_OL4ID_M,
2493 				   HNS3_RXD_OL4ID_S);
2494 	switch (ol4_type) {
2495 	case HNS3_OL4_TYPE_MAC_IN_UDP:
2496 	case HNS3_OL4_TYPE_NVGRE:
2497 		skb->csum_level = 1;
2498 		/* fall through */
2499 	case HNS3_OL4_TYPE_NO_TUN:
2500 		l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
2501 					  HNS3_RXD_L3ID_S);
2502 		l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
2503 					  HNS3_RXD_L4ID_S);
2504 
2505 		/* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
2506 		if ((l3_type == HNS3_L3_TYPE_IPV4 ||
2507 		     l3_type == HNS3_L3_TYPE_IPV6) &&
2508 		    (l4_type == HNS3_L4_TYPE_UDP ||
2509 		     l4_type == HNS3_L4_TYPE_TCP ||
2510 		     l4_type == HNS3_L4_TYPE_SCTP))
2511 			skb->ip_summed = CHECKSUM_UNNECESSARY;
2512 		break;
2513 	default:
2514 		break;
2515 	}
2516 }
2517 
2518 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
2519 {
2520 	if (skb_has_frag_list(skb))
2521 		napi_gro_flush(&ring->tqp_vector->napi, false);
2522 
2523 	napi_gro_receive(&ring->tqp_vector->napi, skb);
2524 }
2525 
2526 static bool hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
2527 				struct hns3_desc *desc, u32 l234info,
2528 				u16 *vlan_tag)
2529 {
2530 	struct hnae3_handle *handle = ring->tqp->handle;
2531 	struct pci_dev *pdev = ring->tqp->handle->pdev;
2532 
2533 	if (pdev->revision == 0x20) {
2534 		*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2535 		if (!(*vlan_tag & VLAN_VID_MASK))
2536 			*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2537 
2538 		return (*vlan_tag != 0);
2539 	}
2540 
2541 #define HNS3_STRP_OUTER_VLAN	0x1
2542 #define HNS3_STRP_INNER_VLAN	0x2
2543 #define HNS3_STRP_BOTH		0x3
2544 
2545 	/* Hardware always insert VLAN tag into RX descriptor when
2546 	 * remove the tag from packet, driver needs to determine
2547 	 * reporting which tag to stack.
2548 	 */
2549 	switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
2550 				HNS3_RXD_STRP_TAGP_S)) {
2551 	case HNS3_STRP_OUTER_VLAN:
2552 		if (handle->port_base_vlan_state !=
2553 				HNAE3_PORT_BASE_VLAN_DISABLE)
2554 			return false;
2555 
2556 		*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2557 		return true;
2558 	case HNS3_STRP_INNER_VLAN:
2559 		if (handle->port_base_vlan_state !=
2560 				HNAE3_PORT_BASE_VLAN_DISABLE)
2561 			return false;
2562 
2563 		*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2564 		return true;
2565 	case HNS3_STRP_BOTH:
2566 		if (handle->port_base_vlan_state ==
2567 				HNAE3_PORT_BASE_VLAN_DISABLE)
2568 			*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
2569 		else
2570 			*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
2571 
2572 		return true;
2573 	default:
2574 		return false;
2575 	}
2576 }
2577 
2578 static int hns3_alloc_skb(struct hns3_enet_ring *ring, int length,
2579 			  unsigned char *va)
2580 {
2581 #define HNS3_NEED_ADD_FRAG	1
2582 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
2583 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2584 	struct sk_buff *skb;
2585 
2586 	ring->skb = napi_alloc_skb(&ring->tqp_vector->napi, HNS3_RX_HEAD_SIZE);
2587 	skb = ring->skb;
2588 	if (unlikely(!skb)) {
2589 		netdev_err(netdev, "alloc rx skb fail\n");
2590 
2591 		u64_stats_update_begin(&ring->syncp);
2592 		ring->stats.sw_err_cnt++;
2593 		u64_stats_update_end(&ring->syncp);
2594 
2595 		return -ENOMEM;
2596 	}
2597 
2598 	prefetchw(skb->data);
2599 
2600 	ring->pending_buf = 1;
2601 	ring->frag_num = 0;
2602 	ring->tail_skb = NULL;
2603 	if (length <= HNS3_RX_HEAD_SIZE) {
2604 		memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
2605 
2606 		/* We can reuse buffer as-is, just make sure it is local */
2607 		if (likely(page_to_nid(desc_cb->priv) == numa_mem_id()))
2608 			desc_cb->reuse_flag = 1;
2609 		else /* This page cannot be reused so discard it */
2610 			put_page(desc_cb->priv);
2611 
2612 		ring_ptr_move_fw(ring, next_to_clean);
2613 		return 0;
2614 	}
2615 	u64_stats_update_begin(&ring->syncp);
2616 	ring->stats.seg_pkt_cnt++;
2617 	u64_stats_update_end(&ring->syncp);
2618 
2619 	ring->pull_len = eth_get_headlen(netdev, va, HNS3_RX_HEAD_SIZE);
2620 	__skb_put(skb, ring->pull_len);
2621 	hns3_nic_reuse_page(skb, ring->frag_num++, ring, ring->pull_len,
2622 			    desc_cb);
2623 	ring_ptr_move_fw(ring, next_to_clean);
2624 
2625 	return HNS3_NEED_ADD_FRAG;
2626 }
2627 
2628 static int hns3_add_frag(struct hns3_enet_ring *ring, struct hns3_desc *desc,
2629 			 struct sk_buff **out_skb, bool pending)
2630 {
2631 	struct sk_buff *skb = *out_skb;
2632 	struct sk_buff *head_skb = *out_skb;
2633 	struct sk_buff *new_skb;
2634 	struct hns3_desc_cb *desc_cb;
2635 	struct hns3_desc *pre_desc;
2636 	u32 bd_base_info;
2637 	int pre_bd;
2638 
2639 	/* if there is pending bd, the SW param next_to_clean has moved
2640 	 * to next and the next is NULL
2641 	 */
2642 	if (pending) {
2643 		pre_bd = (ring->next_to_clean - 1 + ring->desc_num) %
2644 			 ring->desc_num;
2645 		pre_desc = &ring->desc[pre_bd];
2646 		bd_base_info = le32_to_cpu(pre_desc->rx.bd_base_info);
2647 	} else {
2648 		bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2649 	}
2650 
2651 	while (!(bd_base_info & BIT(HNS3_RXD_FE_B))) {
2652 		desc = &ring->desc[ring->next_to_clean];
2653 		desc_cb = &ring->desc_cb[ring->next_to_clean];
2654 		bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2655 		/* make sure HW write desc complete */
2656 		dma_rmb();
2657 		if (!(bd_base_info & BIT(HNS3_RXD_VLD_B)))
2658 			return -ENXIO;
2659 
2660 		if (unlikely(ring->frag_num >= MAX_SKB_FRAGS)) {
2661 			new_skb = napi_alloc_skb(&ring->tqp_vector->napi,
2662 						 HNS3_RX_HEAD_SIZE);
2663 			if (unlikely(!new_skb)) {
2664 				netdev_err(ring->tqp->handle->kinfo.netdev,
2665 					   "alloc rx skb frag fail\n");
2666 				return -ENXIO;
2667 			}
2668 			ring->frag_num = 0;
2669 
2670 			if (ring->tail_skb) {
2671 				ring->tail_skb->next = new_skb;
2672 				ring->tail_skb = new_skb;
2673 			} else {
2674 				skb_shinfo(skb)->frag_list = new_skb;
2675 				ring->tail_skb = new_skb;
2676 			}
2677 		}
2678 
2679 		if (ring->tail_skb) {
2680 			head_skb->truesize += hnae3_buf_size(ring);
2681 			head_skb->data_len += le16_to_cpu(desc->rx.size);
2682 			head_skb->len += le16_to_cpu(desc->rx.size);
2683 			skb = ring->tail_skb;
2684 		}
2685 
2686 		hns3_nic_reuse_page(skb, ring->frag_num++, ring, 0, desc_cb);
2687 		ring_ptr_move_fw(ring, next_to_clean);
2688 		ring->pending_buf++;
2689 	}
2690 
2691 	return 0;
2692 }
2693 
2694 static int hns3_set_gro_and_checksum(struct hns3_enet_ring *ring,
2695 				     struct sk_buff *skb, u32 l234info,
2696 				     u32 bd_base_info, u32 ol_info)
2697 {
2698 	u32 l3_type;
2699 
2700 	skb_shinfo(skb)->gso_size = hnae3_get_field(bd_base_info,
2701 						    HNS3_RXD_GRO_SIZE_M,
2702 						    HNS3_RXD_GRO_SIZE_S);
2703 	/* if there is no HW GRO, do not set gro params */
2704 	if (!skb_shinfo(skb)->gso_size) {
2705 		hns3_rx_checksum(ring, skb, l234info, bd_base_info, ol_info);
2706 		return 0;
2707 	}
2708 
2709 	NAPI_GRO_CB(skb)->count = hnae3_get_field(l234info,
2710 						  HNS3_RXD_GRO_COUNT_M,
2711 						  HNS3_RXD_GRO_COUNT_S);
2712 
2713 	l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M, HNS3_RXD_L3ID_S);
2714 	if (l3_type == HNS3_L3_TYPE_IPV4)
2715 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2716 	else if (l3_type == HNS3_L3_TYPE_IPV6)
2717 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
2718 	else
2719 		return -EFAULT;
2720 
2721 	return  hns3_gro_complete(skb, l234info);
2722 }
2723 
2724 static void hns3_set_rx_skb_rss_type(struct hns3_enet_ring *ring,
2725 				     struct sk_buff *skb, u32 rss_hash)
2726 {
2727 	struct hnae3_handle *handle = ring->tqp->handle;
2728 	enum pkt_hash_types rss_type;
2729 
2730 	if (rss_hash)
2731 		rss_type = handle->kinfo.rss_type;
2732 	else
2733 		rss_type = PKT_HASH_TYPE_NONE;
2734 
2735 	skb_set_hash(skb, rss_hash, rss_type);
2736 }
2737 
2738 static int hns3_handle_bdinfo(struct hns3_enet_ring *ring, struct sk_buff *skb)
2739 {
2740 	struct net_device *netdev = ring->tqp->handle->kinfo.netdev;
2741 	enum hns3_pkt_l2t_type l2_frame_type;
2742 	u32 bd_base_info, l234info, ol_info;
2743 	struct hns3_desc *desc;
2744 	unsigned int len;
2745 	int pre_ntc, ret;
2746 
2747 	/* bdinfo handled below is only valid on the last BD of the
2748 	 * current packet, and ring->next_to_clean indicates the first
2749 	 * descriptor of next packet, so need - 1 below.
2750 	 */
2751 	pre_ntc = ring->next_to_clean ? (ring->next_to_clean - 1) :
2752 					(ring->desc_num - 1);
2753 	desc = &ring->desc[pre_ntc];
2754 	bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2755 	l234info = le32_to_cpu(desc->rx.l234_info);
2756 	ol_info = le32_to_cpu(desc->rx.ol_info);
2757 
2758 	/* Based on hw strategy, the tag offloaded will be stored at
2759 	 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
2760 	 * in one layer tag case.
2761 	 */
2762 	if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
2763 		u16 vlan_tag;
2764 
2765 		if (hns3_parse_vlan_tag(ring, desc, l234info, &vlan_tag))
2766 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2767 					       vlan_tag);
2768 	}
2769 
2770 	if (unlikely(!desc->rx.pkt_len || (l234info & (BIT(HNS3_RXD_TRUNCAT_B) |
2771 				  BIT(HNS3_RXD_L2E_B))))) {
2772 		u64_stats_update_begin(&ring->syncp);
2773 		if (l234info & BIT(HNS3_RXD_L2E_B))
2774 			ring->stats.l2_err++;
2775 		else
2776 			ring->stats.err_pkt_len++;
2777 		u64_stats_update_end(&ring->syncp);
2778 
2779 		return -EFAULT;
2780 	}
2781 
2782 	len = skb->len;
2783 
2784 	/* Do update ip stack process */
2785 	skb->protocol = eth_type_trans(skb, netdev);
2786 
2787 	/* This is needed in order to enable forwarding support */
2788 	ret = hns3_set_gro_and_checksum(ring, skb, l234info,
2789 					bd_base_info, ol_info);
2790 	if (unlikely(ret)) {
2791 		u64_stats_update_begin(&ring->syncp);
2792 		ring->stats.rx_err_cnt++;
2793 		u64_stats_update_end(&ring->syncp);
2794 		return ret;
2795 	}
2796 
2797 	l2_frame_type = hnae3_get_field(l234info, HNS3_RXD_DMAC_M,
2798 					HNS3_RXD_DMAC_S);
2799 
2800 	u64_stats_update_begin(&ring->syncp);
2801 	ring->stats.rx_pkts++;
2802 	ring->stats.rx_bytes += len;
2803 
2804 	if (l2_frame_type == HNS3_L2_TYPE_MULTICAST)
2805 		ring->stats.rx_multicast++;
2806 
2807 	u64_stats_update_end(&ring->syncp);
2808 
2809 	ring->tqp_vector->rx_group.total_bytes += len;
2810 
2811 	hns3_set_rx_skb_rss_type(ring, skb, le32_to_cpu(desc->rx.rss_hash));
2812 	return 0;
2813 }
2814 
2815 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring,
2816 			     struct sk_buff **out_skb)
2817 {
2818 	struct sk_buff *skb = ring->skb;
2819 	struct hns3_desc_cb *desc_cb;
2820 	struct hns3_desc *desc;
2821 	u32 bd_base_info;
2822 	int length;
2823 	int ret;
2824 
2825 	desc = &ring->desc[ring->next_to_clean];
2826 	desc_cb = &ring->desc_cb[ring->next_to_clean];
2827 
2828 	prefetch(desc);
2829 
2830 	length = le16_to_cpu(desc->rx.size);
2831 	bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
2832 
2833 	/* Check valid BD */
2834 	if (unlikely(!(bd_base_info & BIT(HNS3_RXD_VLD_B))))
2835 		return -ENXIO;
2836 
2837 	if (!skb)
2838 		ring->va = (unsigned char *)desc_cb->buf + desc_cb->page_offset;
2839 
2840 	/* Prefetch first cache line of first page
2841 	 * Idea is to cache few bytes of the header of the packet. Our L1 Cache
2842 	 * line size is 64B so need to prefetch twice to make it 128B. But in
2843 	 * actual we can have greater size of caches with 128B Level 1 cache
2844 	 * lines. In such a case, single fetch would suffice to cache in the
2845 	 * relevant part of the header.
2846 	 */
2847 	prefetch(ring->va);
2848 #if L1_CACHE_BYTES < 128
2849 	prefetch(ring->va + L1_CACHE_BYTES);
2850 #endif
2851 
2852 	if (!skb) {
2853 		ret = hns3_alloc_skb(ring, length, ring->va);
2854 		*out_skb = skb = ring->skb;
2855 
2856 		if (ret < 0) /* alloc buffer fail */
2857 			return ret;
2858 		if (ret > 0) { /* need add frag */
2859 			ret = hns3_add_frag(ring, desc, &skb, false);
2860 			if (ret)
2861 				return ret;
2862 
2863 			/* As the head data may be changed when GRO enable, copy
2864 			 * the head data in after other data rx completed
2865 			 */
2866 			memcpy(skb->data, ring->va,
2867 			       ALIGN(ring->pull_len, sizeof(long)));
2868 		}
2869 	} else {
2870 		ret = hns3_add_frag(ring, desc, &skb, true);
2871 		if (ret)
2872 			return ret;
2873 
2874 		/* As the head data may be changed when GRO enable, copy
2875 		 * the head data in after other data rx completed
2876 		 */
2877 		memcpy(skb->data, ring->va,
2878 		       ALIGN(ring->pull_len, sizeof(long)));
2879 	}
2880 
2881 	ret = hns3_handle_bdinfo(ring, skb);
2882 	if (unlikely(ret)) {
2883 		dev_kfree_skb_any(skb);
2884 		return ret;
2885 	}
2886 
2887 	skb_record_rx_queue(skb, ring->tqp->tqp_index);
2888 	*out_skb = skb;
2889 
2890 	return 0;
2891 }
2892 
2893 int hns3_clean_rx_ring(struct hns3_enet_ring *ring, int budget,
2894 		       void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
2895 {
2896 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
2897 	int recv_pkts, recv_bds, clean_count, err;
2898 	int unused_count = hns3_desc_unused(ring);
2899 	struct sk_buff *skb = ring->skb;
2900 	int num;
2901 
2902 	num = readl_relaxed(ring->tqp->io_base + HNS3_RING_RX_RING_FBDNUM_REG);
2903 	rmb(); /* Make sure num taken effect before the other data is touched */
2904 
2905 	recv_pkts = 0, recv_bds = 0, clean_count = 0;
2906 	num -= unused_count;
2907 	unused_count -= ring->pending_buf;
2908 
2909 	while (recv_pkts < budget && recv_bds < num) {
2910 		/* Reuse or realloc buffers */
2911 		if (clean_count + unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
2912 			hns3_nic_alloc_rx_buffers(ring,
2913 						  clean_count + unused_count);
2914 			clean_count = 0;
2915 			unused_count = hns3_desc_unused(ring) -
2916 					ring->pending_buf;
2917 		}
2918 
2919 		/* Poll one pkt */
2920 		err = hns3_handle_rx_bd(ring, &skb);
2921 		if (unlikely(!skb)) /* This fault cannot be repaired */
2922 			goto out;
2923 
2924 		if (err == -ENXIO) { /* Do not get FE for the packet */
2925 			goto out;
2926 		} else if (unlikely(err)) {  /* Do jump the err */
2927 			recv_bds += ring->pending_buf;
2928 			clean_count += ring->pending_buf;
2929 			ring->skb = NULL;
2930 			ring->pending_buf = 0;
2931 			continue;
2932 		}
2933 
2934 		rx_fn(ring, skb);
2935 		recv_bds += ring->pending_buf;
2936 		clean_count += ring->pending_buf;
2937 		ring->skb = NULL;
2938 		ring->pending_buf = 0;
2939 
2940 		recv_pkts++;
2941 	}
2942 
2943 out:
2944 	/* Make all data has been write before submit */
2945 	if (clean_count + unused_count > 0)
2946 		hns3_nic_alloc_rx_buffers(ring, clean_count + unused_count);
2947 
2948 	return recv_pkts;
2949 }
2950 
2951 static bool hns3_get_new_flow_lvl(struct hns3_enet_ring_group *ring_group)
2952 {
2953 #define HNS3_RX_LOW_BYTE_RATE 10000
2954 #define HNS3_RX_MID_BYTE_RATE 20000
2955 #define HNS3_RX_ULTRA_PACKET_RATE 40
2956 
2957 	enum hns3_flow_level_range new_flow_level;
2958 	struct hns3_enet_tqp_vector *tqp_vector;
2959 	int packets_per_msecs, bytes_per_msecs;
2960 	u32 time_passed_ms;
2961 
2962 	tqp_vector = ring_group->ring->tqp_vector;
2963 	time_passed_ms =
2964 		jiffies_to_msecs(jiffies - tqp_vector->last_jiffies);
2965 	if (!time_passed_ms)
2966 		return false;
2967 
2968 	do_div(ring_group->total_packets, time_passed_ms);
2969 	packets_per_msecs = ring_group->total_packets;
2970 
2971 	do_div(ring_group->total_bytes, time_passed_ms);
2972 	bytes_per_msecs = ring_group->total_bytes;
2973 
2974 	new_flow_level = ring_group->coal.flow_level;
2975 
2976 	/* Simple throttlerate management
2977 	 * 0-10MB/s   lower     (50000 ints/s)
2978 	 * 10-20MB/s   middle    (20000 ints/s)
2979 	 * 20-1249MB/s high      (18000 ints/s)
2980 	 * > 40000pps  ultra     (8000 ints/s)
2981 	 */
2982 	switch (new_flow_level) {
2983 	case HNS3_FLOW_LOW:
2984 		if (bytes_per_msecs > HNS3_RX_LOW_BYTE_RATE)
2985 			new_flow_level = HNS3_FLOW_MID;
2986 		break;
2987 	case HNS3_FLOW_MID:
2988 		if (bytes_per_msecs > HNS3_RX_MID_BYTE_RATE)
2989 			new_flow_level = HNS3_FLOW_HIGH;
2990 		else if (bytes_per_msecs <= HNS3_RX_LOW_BYTE_RATE)
2991 			new_flow_level = HNS3_FLOW_LOW;
2992 		break;
2993 	case HNS3_FLOW_HIGH:
2994 	case HNS3_FLOW_ULTRA:
2995 	default:
2996 		if (bytes_per_msecs <= HNS3_RX_MID_BYTE_RATE)
2997 			new_flow_level = HNS3_FLOW_MID;
2998 		break;
2999 	}
3000 
3001 	if (packets_per_msecs > HNS3_RX_ULTRA_PACKET_RATE &&
3002 	    &tqp_vector->rx_group == ring_group)
3003 		new_flow_level = HNS3_FLOW_ULTRA;
3004 
3005 	ring_group->total_bytes = 0;
3006 	ring_group->total_packets = 0;
3007 	ring_group->coal.flow_level = new_flow_level;
3008 
3009 	return true;
3010 }
3011 
3012 static bool hns3_get_new_int_gl(struct hns3_enet_ring_group *ring_group)
3013 {
3014 	struct hns3_enet_tqp_vector *tqp_vector;
3015 	u16 new_int_gl;
3016 
3017 	if (!ring_group->ring)
3018 		return false;
3019 
3020 	tqp_vector = ring_group->ring->tqp_vector;
3021 	if (!tqp_vector->last_jiffies)
3022 		return false;
3023 
3024 	if (ring_group->total_packets == 0) {
3025 		ring_group->coal.int_gl = HNS3_INT_GL_50K;
3026 		ring_group->coal.flow_level = HNS3_FLOW_LOW;
3027 		return true;
3028 	}
3029 
3030 	if (!hns3_get_new_flow_lvl(ring_group))
3031 		return false;
3032 
3033 	new_int_gl = ring_group->coal.int_gl;
3034 	switch (ring_group->coal.flow_level) {
3035 	case HNS3_FLOW_LOW:
3036 		new_int_gl = HNS3_INT_GL_50K;
3037 		break;
3038 	case HNS3_FLOW_MID:
3039 		new_int_gl = HNS3_INT_GL_20K;
3040 		break;
3041 	case HNS3_FLOW_HIGH:
3042 		new_int_gl = HNS3_INT_GL_18K;
3043 		break;
3044 	case HNS3_FLOW_ULTRA:
3045 		new_int_gl = HNS3_INT_GL_8K;
3046 		break;
3047 	default:
3048 		break;
3049 	}
3050 
3051 	if (new_int_gl != ring_group->coal.int_gl) {
3052 		ring_group->coal.int_gl = new_int_gl;
3053 		return true;
3054 	}
3055 	return false;
3056 }
3057 
3058 static void hns3_update_new_int_gl(struct hns3_enet_tqp_vector *tqp_vector)
3059 {
3060 	struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
3061 	struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
3062 	bool rx_update, tx_update;
3063 
3064 	/* update param every 1000ms */
3065 	if (time_before(jiffies,
3066 			tqp_vector->last_jiffies + msecs_to_jiffies(1000)))
3067 		return;
3068 
3069 	if (rx_group->coal.gl_adapt_enable) {
3070 		rx_update = hns3_get_new_int_gl(rx_group);
3071 		if (rx_update)
3072 			hns3_set_vector_coalesce_rx_gl(tqp_vector,
3073 						       rx_group->coal.int_gl);
3074 	}
3075 
3076 	if (tx_group->coal.gl_adapt_enable) {
3077 		tx_update = hns3_get_new_int_gl(tx_group);
3078 		if (tx_update)
3079 			hns3_set_vector_coalesce_tx_gl(tqp_vector,
3080 						       tx_group->coal.int_gl);
3081 	}
3082 
3083 	tqp_vector->last_jiffies = jiffies;
3084 }
3085 
3086 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
3087 {
3088 	struct hns3_nic_priv *priv = netdev_priv(napi->dev);
3089 	struct hns3_enet_ring *ring;
3090 	int rx_pkt_total = 0;
3091 
3092 	struct hns3_enet_tqp_vector *tqp_vector =
3093 		container_of(napi, struct hns3_enet_tqp_vector, napi);
3094 	bool clean_complete = true;
3095 	int rx_budget = budget;
3096 
3097 	if (unlikely(test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
3098 		napi_complete(napi);
3099 		return 0;
3100 	}
3101 
3102 	/* Since the actual Tx work is minimal, we can give the Tx a larger
3103 	 * budget and be more aggressive about cleaning up the Tx descriptors.
3104 	 */
3105 	hns3_for_each_ring(ring, tqp_vector->tx_group)
3106 		hns3_clean_tx_ring(ring);
3107 
3108 	/* make sure rx ring budget not smaller than 1 */
3109 	if (tqp_vector->num_tqps > 1)
3110 		rx_budget = max(budget / tqp_vector->num_tqps, 1);
3111 
3112 	hns3_for_each_ring(ring, tqp_vector->rx_group) {
3113 		int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
3114 						    hns3_rx_skb);
3115 
3116 		if (rx_cleaned >= rx_budget)
3117 			clean_complete = false;
3118 
3119 		rx_pkt_total += rx_cleaned;
3120 	}
3121 
3122 	tqp_vector->rx_group.total_packets += rx_pkt_total;
3123 
3124 	if (!clean_complete)
3125 		return budget;
3126 
3127 	if (napi_complete(napi) &&
3128 	    likely(!test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
3129 		hns3_update_new_int_gl(tqp_vector);
3130 		hns3_mask_vector_irq(tqp_vector, 1);
3131 	}
3132 
3133 	return rx_pkt_total;
3134 }
3135 
3136 static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
3137 				      struct hnae3_ring_chain_node *head)
3138 {
3139 	struct pci_dev *pdev = tqp_vector->handle->pdev;
3140 	struct hnae3_ring_chain_node *cur_chain = head;
3141 	struct hnae3_ring_chain_node *chain;
3142 	struct hns3_enet_ring *tx_ring;
3143 	struct hns3_enet_ring *rx_ring;
3144 
3145 	tx_ring = tqp_vector->tx_group.ring;
3146 	if (tx_ring) {
3147 		cur_chain->tqp_index = tx_ring->tqp->tqp_index;
3148 		hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
3149 			      HNAE3_RING_TYPE_TX);
3150 		hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3151 				HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_TX);
3152 
3153 		cur_chain->next = NULL;
3154 
3155 		while (tx_ring->next) {
3156 			tx_ring = tx_ring->next;
3157 
3158 			chain = devm_kzalloc(&pdev->dev, sizeof(*chain),
3159 					     GFP_KERNEL);
3160 			if (!chain)
3161 				goto err_free_chain;
3162 
3163 			cur_chain->next = chain;
3164 			chain->tqp_index = tx_ring->tqp->tqp_index;
3165 			hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
3166 				      HNAE3_RING_TYPE_TX);
3167 			hnae3_set_field(chain->int_gl_idx,
3168 					HNAE3_RING_GL_IDX_M,
3169 					HNAE3_RING_GL_IDX_S,
3170 					HNAE3_RING_GL_TX);
3171 
3172 			cur_chain = chain;
3173 		}
3174 	}
3175 
3176 	rx_ring = tqp_vector->rx_group.ring;
3177 	if (!tx_ring && rx_ring) {
3178 		cur_chain->next = NULL;
3179 		cur_chain->tqp_index = rx_ring->tqp->tqp_index;
3180 		hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
3181 			      HNAE3_RING_TYPE_RX);
3182 		hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3183 				HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
3184 
3185 		rx_ring = rx_ring->next;
3186 	}
3187 
3188 	while (rx_ring) {
3189 		chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
3190 		if (!chain)
3191 			goto err_free_chain;
3192 
3193 		cur_chain->next = chain;
3194 		chain->tqp_index = rx_ring->tqp->tqp_index;
3195 		hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
3196 			      HNAE3_RING_TYPE_RX);
3197 		hnae3_set_field(chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3198 				HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
3199 
3200 		cur_chain = chain;
3201 
3202 		rx_ring = rx_ring->next;
3203 	}
3204 
3205 	return 0;
3206 
3207 err_free_chain:
3208 	cur_chain = head->next;
3209 	while (cur_chain) {
3210 		chain = cur_chain->next;
3211 		devm_kfree(&pdev->dev, cur_chain);
3212 		cur_chain = chain;
3213 	}
3214 	head->next = NULL;
3215 
3216 	return -ENOMEM;
3217 }
3218 
3219 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
3220 					struct hnae3_ring_chain_node *head)
3221 {
3222 	struct pci_dev *pdev = tqp_vector->handle->pdev;
3223 	struct hnae3_ring_chain_node *chain_tmp, *chain;
3224 
3225 	chain = head->next;
3226 
3227 	while (chain) {
3228 		chain_tmp = chain->next;
3229 		devm_kfree(&pdev->dev, chain);
3230 		chain = chain_tmp;
3231 	}
3232 }
3233 
3234 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
3235 				   struct hns3_enet_ring *ring)
3236 {
3237 	ring->next = group->ring;
3238 	group->ring = ring;
3239 
3240 	group->count++;
3241 }
3242 
3243 static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
3244 {
3245 	struct pci_dev *pdev = priv->ae_handle->pdev;
3246 	struct hns3_enet_tqp_vector *tqp_vector;
3247 	int num_vectors = priv->vector_num;
3248 	int numa_node;
3249 	int vector_i;
3250 
3251 	numa_node = dev_to_node(&pdev->dev);
3252 
3253 	for (vector_i = 0; vector_i < num_vectors; vector_i++) {
3254 		tqp_vector = &priv->tqp_vector[vector_i];
3255 		cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
3256 				&tqp_vector->affinity_mask);
3257 	}
3258 }
3259 
3260 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
3261 {
3262 	struct hnae3_ring_chain_node vector_ring_chain;
3263 	struct hnae3_handle *h = priv->ae_handle;
3264 	struct hns3_enet_tqp_vector *tqp_vector;
3265 	int ret = 0;
3266 	int i;
3267 
3268 	hns3_nic_set_cpumask(priv);
3269 
3270 	for (i = 0; i < priv->vector_num; i++) {
3271 		tqp_vector = &priv->tqp_vector[i];
3272 		hns3_vector_gl_rl_init_hw(tqp_vector, priv);
3273 		tqp_vector->num_tqps = 0;
3274 	}
3275 
3276 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3277 		u16 vector_i = i % priv->vector_num;
3278 		u16 tqp_num = h->kinfo.num_tqps;
3279 
3280 		tqp_vector = &priv->tqp_vector[vector_i];
3281 
3282 		hns3_add_ring_to_group(&tqp_vector->tx_group,
3283 				       priv->ring_data[i].ring);
3284 
3285 		hns3_add_ring_to_group(&tqp_vector->rx_group,
3286 				       priv->ring_data[i + tqp_num].ring);
3287 
3288 		priv->ring_data[i].ring->tqp_vector = tqp_vector;
3289 		priv->ring_data[i + tqp_num].ring->tqp_vector = tqp_vector;
3290 		tqp_vector->num_tqps++;
3291 	}
3292 
3293 	for (i = 0; i < priv->vector_num; i++) {
3294 		tqp_vector = &priv->tqp_vector[i];
3295 
3296 		tqp_vector->rx_group.total_bytes = 0;
3297 		tqp_vector->rx_group.total_packets = 0;
3298 		tqp_vector->tx_group.total_bytes = 0;
3299 		tqp_vector->tx_group.total_packets = 0;
3300 		tqp_vector->handle = h;
3301 
3302 		ret = hns3_get_vector_ring_chain(tqp_vector,
3303 						 &vector_ring_chain);
3304 		if (ret)
3305 			goto map_ring_fail;
3306 
3307 		ret = h->ae_algo->ops->map_ring_to_vector(h,
3308 			tqp_vector->vector_irq, &vector_ring_chain);
3309 
3310 		hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3311 
3312 		if (ret)
3313 			goto map_ring_fail;
3314 
3315 		netif_napi_add(priv->netdev, &tqp_vector->napi,
3316 			       hns3_nic_common_poll, NAPI_POLL_WEIGHT);
3317 	}
3318 
3319 	return 0;
3320 
3321 map_ring_fail:
3322 	while (i--)
3323 		netif_napi_del(&priv->tqp_vector[i].napi);
3324 
3325 	return ret;
3326 }
3327 
3328 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
3329 {
3330 #define HNS3_VECTOR_PF_MAX_NUM		64
3331 
3332 	struct hnae3_handle *h = priv->ae_handle;
3333 	struct hns3_enet_tqp_vector *tqp_vector;
3334 	struct hnae3_vector_info *vector;
3335 	struct pci_dev *pdev = h->pdev;
3336 	u16 tqp_num = h->kinfo.num_tqps;
3337 	u16 vector_num;
3338 	int ret = 0;
3339 	u16 i;
3340 
3341 	/* RSS size, cpu online and vector_num should be the same */
3342 	/* Should consider 2p/4p later */
3343 	vector_num = min_t(u16, num_online_cpus(), tqp_num);
3344 	vector_num = min_t(u16, vector_num, HNS3_VECTOR_PF_MAX_NUM);
3345 
3346 	vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
3347 			      GFP_KERNEL);
3348 	if (!vector)
3349 		return -ENOMEM;
3350 
3351 	/* save the actual available vector number */
3352 	vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
3353 
3354 	priv->vector_num = vector_num;
3355 	priv->tqp_vector = (struct hns3_enet_tqp_vector *)
3356 		devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
3357 			     GFP_KERNEL);
3358 	if (!priv->tqp_vector) {
3359 		ret = -ENOMEM;
3360 		goto out;
3361 	}
3362 
3363 	for (i = 0; i < priv->vector_num; i++) {
3364 		tqp_vector = &priv->tqp_vector[i];
3365 		tqp_vector->idx = i;
3366 		tqp_vector->mask_addr = vector[i].io_addr;
3367 		tqp_vector->vector_irq = vector[i].vector;
3368 		hns3_vector_gl_rl_init(tqp_vector, priv);
3369 	}
3370 
3371 out:
3372 	devm_kfree(&pdev->dev, vector);
3373 	return ret;
3374 }
3375 
3376 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
3377 {
3378 	group->ring = NULL;
3379 	group->count = 0;
3380 }
3381 
3382 static void hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
3383 {
3384 	struct hnae3_ring_chain_node vector_ring_chain;
3385 	struct hnae3_handle *h = priv->ae_handle;
3386 	struct hns3_enet_tqp_vector *tqp_vector;
3387 	int i;
3388 
3389 	for (i = 0; i < priv->vector_num; i++) {
3390 		tqp_vector = &priv->tqp_vector[i];
3391 
3392 		if (!tqp_vector->rx_group.ring && !tqp_vector->tx_group.ring)
3393 			continue;
3394 
3395 		hns3_get_vector_ring_chain(tqp_vector, &vector_ring_chain);
3396 
3397 		h->ae_algo->ops->unmap_ring_from_vector(h,
3398 			tqp_vector->vector_irq, &vector_ring_chain);
3399 
3400 		hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3401 
3402 		if (tqp_vector->irq_init_flag == HNS3_VECTOR_INITED) {
3403 			irq_set_affinity_hint(tqp_vector->vector_irq, NULL);
3404 			free_irq(tqp_vector->vector_irq, tqp_vector);
3405 			tqp_vector->irq_init_flag = HNS3_VECTOR_NOT_INITED;
3406 		}
3407 
3408 		hns3_clear_ring_group(&tqp_vector->rx_group);
3409 		hns3_clear_ring_group(&tqp_vector->tx_group);
3410 		netif_napi_del(&priv->tqp_vector[i].napi);
3411 	}
3412 }
3413 
3414 static int hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
3415 {
3416 	struct hnae3_handle *h = priv->ae_handle;
3417 	struct pci_dev *pdev = h->pdev;
3418 	int i, ret;
3419 
3420 	for (i = 0; i < priv->vector_num; i++) {
3421 		struct hns3_enet_tqp_vector *tqp_vector;
3422 
3423 		tqp_vector = &priv->tqp_vector[i];
3424 		ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
3425 		if (ret)
3426 			return ret;
3427 	}
3428 
3429 	devm_kfree(&pdev->dev, priv->tqp_vector);
3430 	return 0;
3431 }
3432 
3433 static int hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
3434 			     unsigned int ring_type)
3435 {
3436 	struct hns3_nic_ring_data *ring_data = priv->ring_data;
3437 	int queue_num = priv->ae_handle->kinfo.num_tqps;
3438 	struct pci_dev *pdev = priv->ae_handle->pdev;
3439 	struct hns3_enet_ring *ring;
3440 	int desc_num;
3441 
3442 	ring = devm_kzalloc(&pdev->dev, sizeof(*ring), GFP_KERNEL);
3443 	if (!ring)
3444 		return -ENOMEM;
3445 
3446 	if (ring_type == HNAE3_RING_TYPE_TX) {
3447 		desc_num = priv->ae_handle->kinfo.num_tx_desc;
3448 		ring_data[q->tqp_index].ring = ring;
3449 		ring_data[q->tqp_index].queue_index = q->tqp_index;
3450 		ring->io_base = (u8 __iomem *)q->io_base + HNS3_TX_REG_OFFSET;
3451 	} else {
3452 		desc_num = priv->ae_handle->kinfo.num_rx_desc;
3453 		ring_data[q->tqp_index + queue_num].ring = ring;
3454 		ring_data[q->tqp_index + queue_num].queue_index = q->tqp_index;
3455 		ring->io_base = q->io_base;
3456 	}
3457 
3458 	hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
3459 
3460 	ring->tqp = q;
3461 	ring->desc = NULL;
3462 	ring->desc_cb = NULL;
3463 	ring->dev = priv->dev;
3464 	ring->desc_dma_addr = 0;
3465 	ring->buf_size = q->buf_size;
3466 	ring->desc_num = desc_num;
3467 	ring->next_to_use = 0;
3468 	ring->next_to_clean = 0;
3469 
3470 	return 0;
3471 }
3472 
3473 static int hns3_queue_to_ring(struct hnae3_queue *tqp,
3474 			      struct hns3_nic_priv *priv)
3475 {
3476 	int ret;
3477 
3478 	ret = hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
3479 	if (ret)
3480 		return ret;
3481 
3482 	ret = hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
3483 	if (ret) {
3484 		devm_kfree(priv->dev, priv->ring_data[tqp->tqp_index].ring);
3485 		return ret;
3486 	}
3487 
3488 	return 0;
3489 }
3490 
3491 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
3492 {
3493 	struct hnae3_handle *h = priv->ae_handle;
3494 	struct pci_dev *pdev = h->pdev;
3495 	int i, ret;
3496 
3497 	priv->ring_data =  devm_kzalloc(&pdev->dev,
3498 					array3_size(h->kinfo.num_tqps,
3499 						    sizeof(*priv->ring_data),
3500 						    2),
3501 					GFP_KERNEL);
3502 	if (!priv->ring_data)
3503 		return -ENOMEM;
3504 
3505 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3506 		ret = hns3_queue_to_ring(h->kinfo.tqp[i], priv);
3507 		if (ret)
3508 			goto err;
3509 	}
3510 
3511 	return 0;
3512 err:
3513 	while (i--) {
3514 		devm_kfree(priv->dev, priv->ring_data[i].ring);
3515 		devm_kfree(priv->dev,
3516 			   priv->ring_data[i + h->kinfo.num_tqps].ring);
3517 	}
3518 
3519 	devm_kfree(&pdev->dev, priv->ring_data);
3520 	priv->ring_data = NULL;
3521 	return ret;
3522 }
3523 
3524 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
3525 {
3526 	struct hnae3_handle *h = priv->ae_handle;
3527 	int i;
3528 
3529 	if (!priv->ring_data)
3530 		return;
3531 
3532 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3533 		devm_kfree(priv->dev, priv->ring_data[i].ring);
3534 		devm_kfree(priv->dev,
3535 			   priv->ring_data[i + h->kinfo.num_tqps].ring);
3536 	}
3537 	devm_kfree(priv->dev, priv->ring_data);
3538 	priv->ring_data = NULL;
3539 }
3540 
3541 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
3542 {
3543 	int ret;
3544 
3545 	if (ring->desc_num <= 0 || ring->buf_size <= 0)
3546 		return -EINVAL;
3547 
3548 	ring->desc_cb = devm_kcalloc(ring_to_dev(ring), ring->desc_num,
3549 				     sizeof(ring->desc_cb[0]), GFP_KERNEL);
3550 	if (!ring->desc_cb) {
3551 		ret = -ENOMEM;
3552 		goto out;
3553 	}
3554 
3555 	ret = hns3_alloc_desc(ring);
3556 	if (ret)
3557 		goto out_with_desc_cb;
3558 
3559 	if (!HNAE3_IS_TX_RING(ring)) {
3560 		ret = hns3_alloc_ring_buffers(ring);
3561 		if (ret)
3562 			goto out_with_desc;
3563 	}
3564 
3565 	return 0;
3566 
3567 out_with_desc:
3568 	hns3_free_desc(ring);
3569 out_with_desc_cb:
3570 	devm_kfree(ring_to_dev(ring), ring->desc_cb);
3571 	ring->desc_cb = NULL;
3572 out:
3573 	return ret;
3574 }
3575 
3576 static void hns3_fini_ring(struct hns3_enet_ring *ring)
3577 {
3578 	hns3_free_desc(ring);
3579 	devm_kfree(ring_to_dev(ring), ring->desc_cb);
3580 	ring->desc_cb = NULL;
3581 	ring->next_to_clean = 0;
3582 	ring->next_to_use = 0;
3583 	ring->pending_buf = 0;
3584 	if (ring->skb) {
3585 		dev_kfree_skb_any(ring->skb);
3586 		ring->skb = NULL;
3587 	}
3588 }
3589 
3590 static int hns3_buf_size2type(u32 buf_size)
3591 {
3592 	int bd_size_type;
3593 
3594 	switch (buf_size) {
3595 	case 512:
3596 		bd_size_type = HNS3_BD_SIZE_512_TYPE;
3597 		break;
3598 	case 1024:
3599 		bd_size_type = HNS3_BD_SIZE_1024_TYPE;
3600 		break;
3601 	case 2048:
3602 		bd_size_type = HNS3_BD_SIZE_2048_TYPE;
3603 		break;
3604 	case 4096:
3605 		bd_size_type = HNS3_BD_SIZE_4096_TYPE;
3606 		break;
3607 	default:
3608 		bd_size_type = HNS3_BD_SIZE_2048_TYPE;
3609 	}
3610 
3611 	return bd_size_type;
3612 }
3613 
3614 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
3615 {
3616 	dma_addr_t dma = ring->desc_dma_addr;
3617 	struct hnae3_queue *q = ring->tqp;
3618 
3619 	if (!HNAE3_IS_TX_RING(ring)) {
3620 		hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG, (u32)dma);
3621 		hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
3622 			       (u32)((dma >> 31) >> 1));
3623 
3624 		hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
3625 			       hns3_buf_size2type(ring->buf_size));
3626 		hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
3627 			       ring->desc_num / 8 - 1);
3628 
3629 	} else {
3630 		hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
3631 			       (u32)dma);
3632 		hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
3633 			       (u32)((dma >> 31) >> 1));
3634 
3635 		hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
3636 			       ring->desc_num / 8 - 1);
3637 	}
3638 }
3639 
3640 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
3641 {
3642 	struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
3643 	int i;
3644 
3645 	for (i = 0; i < HNAE3_MAX_TC; i++) {
3646 		struct hnae3_tc_info *tc_info = &kinfo->tc_info[i];
3647 		int j;
3648 
3649 		if (!tc_info->enable)
3650 			continue;
3651 
3652 		for (j = 0; j < tc_info->tqp_count; j++) {
3653 			struct hnae3_queue *q;
3654 
3655 			q = priv->ring_data[tc_info->tqp_offset + j].ring->tqp;
3656 			hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG,
3657 				       tc_info->tc);
3658 		}
3659 	}
3660 }
3661 
3662 int hns3_init_all_ring(struct hns3_nic_priv *priv)
3663 {
3664 	struct hnae3_handle *h = priv->ae_handle;
3665 	int ring_num = h->kinfo.num_tqps * 2;
3666 	int i, j;
3667 	int ret;
3668 
3669 	for (i = 0; i < ring_num; i++) {
3670 		ret = hns3_alloc_ring_memory(priv->ring_data[i].ring);
3671 		if (ret) {
3672 			dev_err(priv->dev,
3673 				"Alloc ring memory fail! ret=%d\n", ret);
3674 			goto out_when_alloc_ring_memory;
3675 		}
3676 
3677 		u64_stats_init(&priv->ring_data[i].ring->syncp);
3678 	}
3679 
3680 	return 0;
3681 
3682 out_when_alloc_ring_memory:
3683 	for (j = i - 1; j >= 0; j--)
3684 		hns3_fini_ring(priv->ring_data[j].ring);
3685 
3686 	return -ENOMEM;
3687 }
3688 
3689 int hns3_uninit_all_ring(struct hns3_nic_priv *priv)
3690 {
3691 	struct hnae3_handle *h = priv->ae_handle;
3692 	int i;
3693 
3694 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3695 		hns3_fini_ring(priv->ring_data[i].ring);
3696 		hns3_fini_ring(priv->ring_data[i + h->kinfo.num_tqps].ring);
3697 	}
3698 	return 0;
3699 }
3700 
3701 /* Set mac addr if it is configured. or leave it to the AE driver */
3702 static int hns3_init_mac_addr(struct net_device *netdev, bool init)
3703 {
3704 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3705 	struct hnae3_handle *h = priv->ae_handle;
3706 	u8 mac_addr_temp[ETH_ALEN];
3707 	int ret = 0;
3708 
3709 	if (h->ae_algo->ops->get_mac_addr && init) {
3710 		h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
3711 		ether_addr_copy(netdev->dev_addr, mac_addr_temp);
3712 	}
3713 
3714 	/* Check if the MAC address is valid, if not get a random one */
3715 	if (!is_valid_ether_addr(netdev->dev_addr)) {
3716 		eth_hw_addr_random(netdev);
3717 		dev_warn(priv->dev, "using random MAC address %pM\n",
3718 			 netdev->dev_addr);
3719 	}
3720 
3721 	if (h->ae_algo->ops->set_mac_addr)
3722 		ret = h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
3723 
3724 	return ret;
3725 }
3726 
3727 static int hns3_init_phy(struct net_device *netdev)
3728 {
3729 	struct hnae3_handle *h = hns3_get_handle(netdev);
3730 	int ret = 0;
3731 
3732 	if (h->ae_algo->ops->mac_connect_phy)
3733 		ret = h->ae_algo->ops->mac_connect_phy(h);
3734 
3735 	return ret;
3736 }
3737 
3738 static void hns3_uninit_phy(struct net_device *netdev)
3739 {
3740 	struct hnae3_handle *h = hns3_get_handle(netdev);
3741 
3742 	if (h->ae_algo->ops->mac_disconnect_phy)
3743 		h->ae_algo->ops->mac_disconnect_phy(h);
3744 }
3745 
3746 static int hns3_restore_fd_rules(struct net_device *netdev)
3747 {
3748 	struct hnae3_handle *h = hns3_get_handle(netdev);
3749 	int ret = 0;
3750 
3751 	if (h->ae_algo->ops->restore_fd_rules)
3752 		ret = h->ae_algo->ops->restore_fd_rules(h);
3753 
3754 	return ret;
3755 }
3756 
3757 static void hns3_del_all_fd_rules(struct net_device *netdev, bool clear_list)
3758 {
3759 	struct hnae3_handle *h = hns3_get_handle(netdev);
3760 
3761 	if (h->ae_algo->ops->del_all_fd_entries)
3762 		h->ae_algo->ops->del_all_fd_entries(h, clear_list);
3763 }
3764 
3765 static int hns3_client_start(struct hnae3_handle *handle)
3766 {
3767 	if (!handle->ae_algo->ops->client_start)
3768 		return 0;
3769 
3770 	return handle->ae_algo->ops->client_start(handle);
3771 }
3772 
3773 static void hns3_client_stop(struct hnae3_handle *handle)
3774 {
3775 	if (!handle->ae_algo->ops->client_stop)
3776 		return;
3777 
3778 	handle->ae_algo->ops->client_stop(handle);
3779 }
3780 
3781 static void hns3_info_show(struct hns3_nic_priv *priv)
3782 {
3783 	struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
3784 
3785 	dev_info(priv->dev, "MAC address: %pM\n", priv->netdev->dev_addr);
3786 	dev_info(priv->dev, "Task queue pairs numbers: %d\n", kinfo->num_tqps);
3787 	dev_info(priv->dev, "RSS size: %d\n", kinfo->rss_size);
3788 	dev_info(priv->dev, "Allocated RSS size: %d\n", kinfo->req_rss_size);
3789 	dev_info(priv->dev, "RX buffer length: %d\n", kinfo->rx_buf_len);
3790 	dev_info(priv->dev, "Desc num per TX queue: %d\n", kinfo->num_tx_desc);
3791 	dev_info(priv->dev, "Desc num per RX queue: %d\n", kinfo->num_rx_desc);
3792 	dev_info(priv->dev, "Total number of enabled TCs: %d\n", kinfo->num_tc);
3793 	dev_info(priv->dev, "Max mtu size: %d\n", priv->netdev->max_mtu);
3794 }
3795 
3796 static int hns3_client_init(struct hnae3_handle *handle)
3797 {
3798 	struct pci_dev *pdev = handle->pdev;
3799 	u16 alloc_tqps, max_rss_size;
3800 	struct hns3_nic_priv *priv;
3801 	struct net_device *netdev;
3802 	int ret;
3803 
3804 	handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
3805 						    &max_rss_size);
3806 	netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
3807 	if (!netdev)
3808 		return -ENOMEM;
3809 
3810 	priv = netdev_priv(netdev);
3811 	priv->dev = &pdev->dev;
3812 	priv->netdev = netdev;
3813 	priv->ae_handle = handle;
3814 	priv->tx_timeout_count = 0;
3815 	set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
3816 
3817 	handle->msg_enable = netif_msg_init(debug, DEFAULT_MSG_LEVEL);
3818 
3819 	handle->kinfo.netdev = netdev;
3820 	handle->priv = (void *)priv;
3821 
3822 	hns3_init_mac_addr(netdev, true);
3823 
3824 	hns3_set_default_feature(netdev);
3825 
3826 	netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
3827 	netdev->priv_flags |= IFF_UNICAST_FLT;
3828 	netdev->netdev_ops = &hns3_nic_netdev_ops;
3829 	SET_NETDEV_DEV(netdev, &pdev->dev);
3830 	hns3_ethtool_set_ops(netdev);
3831 
3832 	/* Carrier off reporting is important to ethtool even BEFORE open */
3833 	netif_carrier_off(netdev);
3834 
3835 	ret = hns3_get_ring_config(priv);
3836 	if (ret) {
3837 		ret = -ENOMEM;
3838 		goto out_get_ring_cfg;
3839 	}
3840 
3841 	ret = hns3_nic_alloc_vector_data(priv);
3842 	if (ret) {
3843 		ret = -ENOMEM;
3844 		goto out_alloc_vector_data;
3845 	}
3846 
3847 	ret = hns3_nic_init_vector_data(priv);
3848 	if (ret) {
3849 		ret = -ENOMEM;
3850 		goto out_init_vector_data;
3851 	}
3852 
3853 	ret = hns3_init_all_ring(priv);
3854 	if (ret) {
3855 		ret = -ENOMEM;
3856 		goto out_init_ring_data;
3857 	}
3858 
3859 	ret = hns3_init_phy(netdev);
3860 	if (ret)
3861 		goto out_init_phy;
3862 
3863 	ret = register_netdev(netdev);
3864 	if (ret) {
3865 		dev_err(priv->dev, "probe register netdev fail!\n");
3866 		goto out_reg_netdev_fail;
3867 	}
3868 
3869 	ret = hns3_client_start(handle);
3870 	if (ret) {
3871 		dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
3872 		goto out_client_start;
3873 	}
3874 
3875 	hns3_dcbnl_setup(handle);
3876 
3877 	hns3_dbg_init(handle);
3878 
3879 	/* MTU range: (ETH_MIN_MTU(kernel default) - 9702) */
3880 	netdev->max_mtu = HNS3_MAX_MTU;
3881 
3882 	set_bit(HNS3_NIC_STATE_INITED, &priv->state);
3883 
3884 	if (netif_msg_drv(handle))
3885 		hns3_info_show(priv);
3886 
3887 	return ret;
3888 
3889 out_client_start:
3890 	unregister_netdev(netdev);
3891 out_reg_netdev_fail:
3892 	hns3_uninit_phy(netdev);
3893 out_init_phy:
3894 	hns3_uninit_all_ring(priv);
3895 out_init_ring_data:
3896 	hns3_nic_uninit_vector_data(priv);
3897 out_init_vector_data:
3898 	hns3_nic_dealloc_vector_data(priv);
3899 out_alloc_vector_data:
3900 	priv->ring_data = NULL;
3901 out_get_ring_cfg:
3902 	priv->ae_handle = NULL;
3903 	free_netdev(netdev);
3904 	return ret;
3905 }
3906 
3907 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
3908 {
3909 	struct net_device *netdev = handle->kinfo.netdev;
3910 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3911 	int ret;
3912 
3913 	hns3_remove_hw_addr(netdev);
3914 
3915 	if (netdev->reg_state != NETREG_UNINITIALIZED)
3916 		unregister_netdev(netdev);
3917 
3918 	hns3_client_stop(handle);
3919 
3920 	hns3_uninit_phy(netdev);
3921 
3922 	if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
3923 		netdev_warn(netdev, "already uninitialized\n");
3924 		goto out_netdev_free;
3925 	}
3926 
3927 	hns3_del_all_fd_rules(netdev, true);
3928 
3929 	hns3_clear_all_ring(handle, true);
3930 
3931 	hns3_nic_uninit_vector_data(priv);
3932 
3933 	ret = hns3_nic_dealloc_vector_data(priv);
3934 	if (ret)
3935 		netdev_err(netdev, "dealloc vector error\n");
3936 
3937 	ret = hns3_uninit_all_ring(priv);
3938 	if (ret)
3939 		netdev_err(netdev, "uninit ring error\n");
3940 
3941 	hns3_put_ring_config(priv);
3942 
3943 	hns3_dbg_uninit(handle);
3944 
3945 out_netdev_free:
3946 	free_netdev(netdev);
3947 }
3948 
3949 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
3950 {
3951 	struct net_device *netdev = handle->kinfo.netdev;
3952 
3953 	if (!netdev)
3954 		return;
3955 
3956 	if (linkup) {
3957 		netif_carrier_on(netdev);
3958 		netif_tx_wake_all_queues(netdev);
3959 		if (netif_msg_link(handle))
3960 			netdev_info(netdev, "link up\n");
3961 	} else {
3962 		netif_carrier_off(netdev);
3963 		netif_tx_stop_all_queues(netdev);
3964 		if (netif_msg_link(handle))
3965 			netdev_info(netdev, "link down\n");
3966 	}
3967 }
3968 
3969 static int hns3_client_setup_tc(struct hnae3_handle *handle, u8 tc)
3970 {
3971 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
3972 	struct net_device *ndev = kinfo->netdev;
3973 
3974 	if (tc > HNAE3_MAX_TC)
3975 		return -EINVAL;
3976 
3977 	if (!ndev)
3978 		return -ENODEV;
3979 
3980 	return hns3_nic_set_real_num_queue(ndev);
3981 }
3982 
3983 static int hns3_recover_hw_addr(struct net_device *ndev)
3984 {
3985 	struct netdev_hw_addr_list *list;
3986 	struct netdev_hw_addr *ha, *tmp;
3987 	int ret = 0;
3988 
3989 	netif_addr_lock_bh(ndev);
3990 	/* go through and sync uc_addr entries to the device */
3991 	list = &ndev->uc;
3992 	list_for_each_entry_safe(ha, tmp, &list->list, list) {
3993 		ret = hns3_nic_uc_sync(ndev, ha->addr);
3994 		if (ret)
3995 			goto out;
3996 	}
3997 
3998 	/* go through and sync mc_addr entries to the device */
3999 	list = &ndev->mc;
4000 	list_for_each_entry_safe(ha, tmp, &list->list, list) {
4001 		ret = hns3_nic_mc_sync(ndev, ha->addr);
4002 		if (ret)
4003 			goto out;
4004 	}
4005 
4006 out:
4007 	netif_addr_unlock_bh(ndev);
4008 	return ret;
4009 }
4010 
4011 static void hns3_remove_hw_addr(struct net_device *netdev)
4012 {
4013 	struct netdev_hw_addr_list *list;
4014 	struct netdev_hw_addr *ha, *tmp;
4015 
4016 	hns3_nic_uc_unsync(netdev, netdev->dev_addr);
4017 
4018 	netif_addr_lock_bh(netdev);
4019 	/* go through and unsync uc_addr entries to the device */
4020 	list = &netdev->uc;
4021 	list_for_each_entry_safe(ha, tmp, &list->list, list)
4022 		hns3_nic_uc_unsync(netdev, ha->addr);
4023 
4024 	/* go through and unsync mc_addr entries to the device */
4025 	list = &netdev->mc;
4026 	list_for_each_entry_safe(ha, tmp, &list->list, list)
4027 		if (ha->refcount > 1)
4028 			hns3_nic_mc_unsync(netdev, ha->addr);
4029 
4030 	netif_addr_unlock_bh(netdev);
4031 }
4032 
4033 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
4034 {
4035 	while (ring->next_to_clean != ring->next_to_use) {
4036 		ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
4037 		hns3_free_buffer_detach(ring, ring->next_to_clean);
4038 		ring_ptr_move_fw(ring, next_to_clean);
4039 	}
4040 }
4041 
4042 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
4043 {
4044 	struct hns3_desc_cb res_cbs;
4045 	int ret;
4046 
4047 	while (ring->next_to_use != ring->next_to_clean) {
4048 		/* When a buffer is not reused, it's memory has been
4049 		 * freed in hns3_handle_rx_bd or will be freed by
4050 		 * stack, so we need to replace the buffer here.
4051 		 */
4052 		if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
4053 			ret = hns3_reserve_buffer_map(ring, &res_cbs);
4054 			if (ret) {
4055 				u64_stats_update_begin(&ring->syncp);
4056 				ring->stats.sw_err_cnt++;
4057 				u64_stats_update_end(&ring->syncp);
4058 				/* if alloc new buffer fail, exit directly
4059 				 * and reclear in up flow.
4060 				 */
4061 				netdev_warn(ring->tqp->handle->kinfo.netdev,
4062 					    "reserve buffer map failed, ret = %d\n",
4063 					    ret);
4064 				return ret;
4065 			}
4066 			hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
4067 		}
4068 		ring_ptr_move_fw(ring, next_to_use);
4069 	}
4070 
4071 	/* Free the pending skb in rx ring */
4072 	if (ring->skb) {
4073 		dev_kfree_skb_any(ring->skb);
4074 		ring->skb = NULL;
4075 		ring->pending_buf = 0;
4076 	}
4077 
4078 	return 0;
4079 }
4080 
4081 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
4082 {
4083 	while (ring->next_to_use != ring->next_to_clean) {
4084 		/* When a buffer is not reused, it's memory has been
4085 		 * freed in hns3_handle_rx_bd or will be freed by
4086 		 * stack, so only need to unmap the buffer here.
4087 		 */
4088 		if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
4089 			hns3_unmap_buffer(ring,
4090 					  &ring->desc_cb[ring->next_to_use]);
4091 			ring->desc_cb[ring->next_to_use].dma = 0;
4092 		}
4093 
4094 		ring_ptr_move_fw(ring, next_to_use);
4095 	}
4096 }
4097 
4098 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force)
4099 {
4100 	struct net_device *ndev = h->kinfo.netdev;
4101 	struct hns3_nic_priv *priv = netdev_priv(ndev);
4102 	u32 i;
4103 
4104 	for (i = 0; i < h->kinfo.num_tqps; i++) {
4105 		struct hns3_enet_ring *ring;
4106 
4107 		ring = priv->ring_data[i].ring;
4108 		hns3_clear_tx_ring(ring);
4109 
4110 		ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
4111 		/* Continue to clear other rings even if clearing some
4112 		 * rings failed.
4113 		 */
4114 		if (force)
4115 			hns3_force_clear_rx_ring(ring);
4116 		else
4117 			hns3_clear_rx_ring(ring);
4118 	}
4119 }
4120 
4121 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
4122 {
4123 	struct net_device *ndev = h->kinfo.netdev;
4124 	struct hns3_nic_priv *priv = netdev_priv(ndev);
4125 	struct hns3_enet_ring *rx_ring;
4126 	int i, j;
4127 	int ret;
4128 
4129 	for (i = 0; i < h->kinfo.num_tqps; i++) {
4130 		ret = h->ae_algo->ops->reset_queue(h, i);
4131 		if (ret)
4132 			return ret;
4133 
4134 		hns3_init_ring_hw(priv->ring_data[i].ring);
4135 
4136 		/* We need to clear tx ring here because self test will
4137 		 * use the ring and will not run down before up
4138 		 */
4139 		hns3_clear_tx_ring(priv->ring_data[i].ring);
4140 		priv->ring_data[i].ring->next_to_clean = 0;
4141 		priv->ring_data[i].ring->next_to_use = 0;
4142 
4143 		rx_ring = priv->ring_data[i + h->kinfo.num_tqps].ring;
4144 		hns3_init_ring_hw(rx_ring);
4145 		ret = hns3_clear_rx_ring(rx_ring);
4146 		if (ret)
4147 			return ret;
4148 
4149 		/* We can not know the hardware head and tail when this
4150 		 * function is called in reset flow, so we reuse all desc.
4151 		 */
4152 		for (j = 0; j < rx_ring->desc_num; j++)
4153 			hns3_reuse_buffer(rx_ring, j);
4154 
4155 		rx_ring->next_to_clean = 0;
4156 		rx_ring->next_to_use = 0;
4157 	}
4158 
4159 	hns3_init_tx_ring_tc(priv);
4160 
4161 	return 0;
4162 }
4163 
4164 static void hns3_store_coal(struct hns3_nic_priv *priv)
4165 {
4166 	/* ethtool only support setting and querying one coal
4167 	 * configuation for now, so save the vector 0' coal
4168 	 * configuation here in order to restore it.
4169 	 */
4170 	memcpy(&priv->tx_coal, &priv->tqp_vector[0].tx_group.coal,
4171 	       sizeof(struct hns3_enet_coalesce));
4172 	memcpy(&priv->rx_coal, &priv->tqp_vector[0].rx_group.coal,
4173 	       sizeof(struct hns3_enet_coalesce));
4174 }
4175 
4176 static void hns3_restore_coal(struct hns3_nic_priv *priv)
4177 {
4178 	u16 vector_num = priv->vector_num;
4179 	int i;
4180 
4181 	for (i = 0; i < vector_num; i++) {
4182 		memcpy(&priv->tqp_vector[i].tx_group.coal, &priv->tx_coal,
4183 		       sizeof(struct hns3_enet_coalesce));
4184 		memcpy(&priv->tqp_vector[i].rx_group.coal, &priv->rx_coal,
4185 		       sizeof(struct hns3_enet_coalesce));
4186 	}
4187 }
4188 
4189 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
4190 {
4191 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(handle->pdev);
4192 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4193 	struct net_device *ndev = kinfo->netdev;
4194 	struct hns3_nic_priv *priv = netdev_priv(ndev);
4195 
4196 	if (test_and_set_bit(HNS3_NIC_STATE_RESETTING, &priv->state))
4197 		return 0;
4198 
4199 	/* it is cumbersome for hardware to pick-and-choose entries for deletion
4200 	 * from table space. Hence, for function reset software intervention is
4201 	 * required to delete the entries
4202 	 */
4203 	if (hns3_dev_ongoing_func_reset(ae_dev)) {
4204 		hns3_remove_hw_addr(ndev);
4205 		hns3_del_all_fd_rules(ndev, false);
4206 	}
4207 
4208 	if (!netif_running(ndev))
4209 		return 0;
4210 
4211 	return hns3_nic_net_stop(ndev);
4212 }
4213 
4214 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
4215 {
4216 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4217 	struct hns3_nic_priv *priv = netdev_priv(kinfo->netdev);
4218 	int ret = 0;
4219 
4220 	clear_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4221 
4222 	if (netif_running(kinfo->netdev)) {
4223 		ret = hns3_nic_net_open(kinfo->netdev);
4224 		if (ret) {
4225 			set_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4226 			netdev_err(kinfo->netdev,
4227 				   "net up fail, ret=%d!\n", ret);
4228 			return ret;
4229 		}
4230 	}
4231 
4232 	return ret;
4233 }
4234 
4235 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
4236 {
4237 	struct net_device *netdev = handle->kinfo.netdev;
4238 	struct hns3_nic_priv *priv = netdev_priv(netdev);
4239 	int ret;
4240 
4241 	/* Carrier off reporting is important to ethtool even BEFORE open */
4242 	netif_carrier_off(netdev);
4243 
4244 	ret = hns3_get_ring_config(priv);
4245 	if (ret)
4246 		return ret;
4247 
4248 	ret = hns3_nic_alloc_vector_data(priv);
4249 	if (ret)
4250 		goto err_put_ring;
4251 
4252 	hns3_restore_coal(priv);
4253 
4254 	ret = hns3_nic_init_vector_data(priv);
4255 	if (ret)
4256 		goto err_dealloc_vector;
4257 
4258 	ret = hns3_init_all_ring(priv);
4259 	if (ret)
4260 		goto err_uninit_vector;
4261 
4262 	ret = hns3_client_start(handle);
4263 	if (ret) {
4264 		dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
4265 		goto err_uninit_ring;
4266 	}
4267 
4268 	set_bit(HNS3_NIC_STATE_INITED, &priv->state);
4269 
4270 	return ret;
4271 
4272 err_uninit_ring:
4273 	hns3_uninit_all_ring(priv);
4274 err_uninit_vector:
4275 	hns3_nic_uninit_vector_data(priv);
4276 err_dealloc_vector:
4277 	hns3_nic_dealloc_vector_data(priv);
4278 err_put_ring:
4279 	hns3_put_ring_config(priv);
4280 
4281 	return ret;
4282 }
4283 
4284 static int hns3_reset_notify_restore_enet(struct hnae3_handle *handle)
4285 {
4286 	struct net_device *netdev = handle->kinfo.netdev;
4287 	bool vlan_filter_enable;
4288 	int ret;
4289 
4290 	ret = hns3_init_mac_addr(netdev, false);
4291 	if (ret)
4292 		return ret;
4293 
4294 	ret = hns3_recover_hw_addr(netdev);
4295 	if (ret)
4296 		return ret;
4297 
4298 	ret = hns3_update_promisc_mode(netdev, handle->netdev_flags);
4299 	if (ret)
4300 		return ret;
4301 
4302 	vlan_filter_enable = netdev->flags & IFF_PROMISC ? false : true;
4303 	hns3_enable_vlan_filter(netdev, vlan_filter_enable);
4304 
4305 	if (handle->ae_algo->ops->restore_vlan_table)
4306 		handle->ae_algo->ops->restore_vlan_table(handle);
4307 
4308 	return hns3_restore_fd_rules(netdev);
4309 }
4310 
4311 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
4312 {
4313 	struct net_device *netdev = handle->kinfo.netdev;
4314 	struct hns3_nic_priv *priv = netdev_priv(netdev);
4315 	int ret;
4316 
4317 	if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
4318 		netdev_warn(netdev, "already uninitialized\n");
4319 		return 0;
4320 	}
4321 
4322 	hns3_clear_all_ring(handle, true);
4323 	hns3_reset_tx_queue(priv->ae_handle);
4324 
4325 	hns3_nic_uninit_vector_data(priv);
4326 
4327 	hns3_store_coal(priv);
4328 
4329 	ret = hns3_nic_dealloc_vector_data(priv);
4330 	if (ret)
4331 		netdev_err(netdev, "dealloc vector error\n");
4332 
4333 	ret = hns3_uninit_all_ring(priv);
4334 	if (ret)
4335 		netdev_err(netdev, "uninit ring error\n");
4336 
4337 	hns3_put_ring_config(priv);
4338 
4339 	return ret;
4340 }
4341 
4342 static int hns3_reset_notify(struct hnae3_handle *handle,
4343 			     enum hnae3_reset_notify_type type)
4344 {
4345 	int ret = 0;
4346 
4347 	switch (type) {
4348 	case HNAE3_UP_CLIENT:
4349 		ret = hns3_reset_notify_up_enet(handle);
4350 		break;
4351 	case HNAE3_DOWN_CLIENT:
4352 		ret = hns3_reset_notify_down_enet(handle);
4353 		break;
4354 	case HNAE3_INIT_CLIENT:
4355 		ret = hns3_reset_notify_init_enet(handle);
4356 		break;
4357 	case HNAE3_UNINIT_CLIENT:
4358 		ret = hns3_reset_notify_uninit_enet(handle);
4359 		break;
4360 	case HNAE3_RESTORE_CLIENT:
4361 		ret = hns3_reset_notify_restore_enet(handle);
4362 		break;
4363 	default:
4364 		break;
4365 	}
4366 
4367 	return ret;
4368 }
4369 
4370 int hns3_set_channels(struct net_device *netdev,
4371 		      struct ethtool_channels *ch)
4372 {
4373 	struct hnae3_handle *h = hns3_get_handle(netdev);
4374 	struct hnae3_knic_private_info *kinfo = &h->kinfo;
4375 	bool rxfh_configured = netif_is_rxfh_configured(netdev);
4376 	u32 new_tqp_num = ch->combined_count;
4377 	u16 org_tqp_num;
4378 	int ret;
4379 
4380 	if (ch->rx_count || ch->tx_count)
4381 		return -EINVAL;
4382 
4383 	if (new_tqp_num > hns3_get_max_available_channels(h) ||
4384 	    new_tqp_num < 1) {
4385 		dev_err(&netdev->dev,
4386 			"Change tqps fail, the tqp range is from 1 to %d",
4387 			hns3_get_max_available_channels(h));
4388 		return -EINVAL;
4389 	}
4390 
4391 	if (kinfo->rss_size == new_tqp_num)
4392 		return 0;
4393 
4394 	ret = hns3_reset_notify(h, HNAE3_DOWN_CLIENT);
4395 	if (ret)
4396 		return ret;
4397 
4398 	ret = hns3_reset_notify(h, HNAE3_UNINIT_CLIENT);
4399 	if (ret)
4400 		return ret;
4401 
4402 	org_tqp_num = h->kinfo.num_tqps;
4403 	ret = h->ae_algo->ops->set_channels(h, new_tqp_num, rxfh_configured);
4404 	if (ret) {
4405 		ret = h->ae_algo->ops->set_channels(h, org_tqp_num,
4406 						    rxfh_configured);
4407 		if (ret) {
4408 			/* If revert to old tqp failed, fatal error occurred */
4409 			dev_err(&netdev->dev,
4410 				"Revert to old tqp num fail, ret=%d", ret);
4411 			return ret;
4412 		}
4413 		dev_info(&netdev->dev,
4414 			 "Change tqp num fail, Revert to old tqp num");
4415 	}
4416 	ret = hns3_reset_notify(h, HNAE3_INIT_CLIENT);
4417 	if (ret)
4418 		return ret;
4419 
4420 	return hns3_reset_notify(h, HNAE3_UP_CLIENT);
4421 }
4422 
4423 static const struct hnae3_client_ops client_ops = {
4424 	.init_instance = hns3_client_init,
4425 	.uninit_instance = hns3_client_uninit,
4426 	.link_status_change = hns3_link_status_change,
4427 	.setup_tc = hns3_client_setup_tc,
4428 	.reset_notify = hns3_reset_notify,
4429 };
4430 
4431 /* hns3_init_module - Driver registration routine
4432  * hns3_init_module is the first routine called when the driver is
4433  * loaded. All it does is register with the PCI subsystem.
4434  */
4435 static int __init hns3_init_module(void)
4436 {
4437 	int ret;
4438 
4439 	pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
4440 	pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
4441 
4442 	client.type = HNAE3_CLIENT_KNIC;
4443 	snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH - 1, "%s",
4444 		 hns3_driver_name);
4445 
4446 	client.ops = &client_ops;
4447 
4448 	INIT_LIST_HEAD(&client.node);
4449 
4450 	hns3_dbg_register_debugfs(hns3_driver_name);
4451 
4452 	ret = hnae3_register_client(&client);
4453 	if (ret)
4454 		goto err_reg_client;
4455 
4456 	ret = pci_register_driver(&hns3_driver);
4457 	if (ret)
4458 		goto err_reg_driver;
4459 
4460 	return ret;
4461 
4462 err_reg_driver:
4463 	hnae3_unregister_client(&client);
4464 err_reg_client:
4465 	hns3_dbg_unregister_debugfs();
4466 	return ret;
4467 }
4468 module_init(hns3_init_module);
4469 
4470 /* hns3_exit_module - Driver exit cleanup routine
4471  * hns3_exit_module is called just before the driver is removed
4472  * from memory.
4473  */
4474 static void __exit hns3_exit_module(void)
4475 {
4476 	pci_unregister_driver(&hns3_driver);
4477 	hnae3_unregister_client(&client);
4478 	hns3_dbg_unregister_debugfs();
4479 }
4480 module_exit(hns3_exit_module);
4481 
4482 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
4483 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
4484 MODULE_LICENSE("GPL");
4485 MODULE_ALIAS("pci:hns-nic");
4486 MODULE_VERSION(HNS3_MOD_VERSION);
4487