]>
Commit | Line | Data |
---|---|---|
76ad4f0e S |
1 | /* |
2 | * Copyright (c) 2016~2017 Hisilicon Limited. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License as published by | |
6 | * the Free Software Foundation; either version 2 of the License, or | |
7 | * (at your option) any later version. | |
8 | */ | |
9 | ||
10 | #include <linux/dma-mapping.h> | |
11 | #include <linux/etherdevice.h> | |
12 | #include <linux/interrupt.h> | |
13 | #include <linux/if_vlan.h> | |
14 | #include <linux/ip.h> | |
15 | #include <linux/ipv6.h> | |
16 | #include <linux/module.h> | |
17 | #include <linux/pci.h> | |
18 | #include <linux/skbuff.h> | |
19 | #include <linux/sctp.h> | |
20 | #include <linux/vermagic.h> | |
21 | #include <net/gre.h> | |
22 | #include <net/vxlan.h> | |
23 | ||
24 | #include "hnae3.h" | |
25 | #include "hns3_enet.h" | |
26 | ||
1db9b1bf | 27 | static const char hns3_driver_name[] = "hns3"; |
76ad4f0e S |
28 | const char hns3_driver_version[] = VERMAGIC_STRING; |
29 | static const char hns3_driver_string[] = | |
30 | "Hisilicon Ethernet Network Driver for Hip08 Family"; | |
31 | static const char hns3_copyright[] = "Copyright (c) 2017 Huawei Corporation."; | |
32 | static struct hnae3_client client; | |
33 | ||
34 | /* hns3_pci_tbl - PCI Device ID Table | |
35 | * | |
36 | * Last entry must be all 0s | |
37 | * | |
38 | * { Vendor ID, Device ID, SubVendor ID, SubDevice ID, | |
39 | * Class, Class Mask, private data (not used) } | |
40 | */ | |
41 | static const struct pci_device_id hns3_pci_tbl[] = { | |
42 | {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0}, | |
43 | {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0}, | |
e92a0843 | 44 | {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA), |
2daf4a65 | 45 | HNAE3_DEV_SUPPORT_ROCE_DCB_BITS}, |
e92a0843 | 46 | {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC), |
2daf4a65 | 47 | HNAE3_DEV_SUPPORT_ROCE_DCB_BITS}, |
e92a0843 | 48 | {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA), |
2daf4a65 | 49 | HNAE3_DEV_SUPPORT_ROCE_DCB_BITS}, |
e92a0843 | 50 | {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC), |
2daf4a65 | 51 | HNAE3_DEV_SUPPORT_ROCE_DCB_BITS}, |
e92a0843 | 52 | {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC), |
2daf4a65 | 53 | HNAE3_DEV_SUPPORT_ROCE_DCB_BITS}, |
76ad4f0e S |
54 | /* required last entry */ |
55 | {0, } | |
56 | }; | |
57 | MODULE_DEVICE_TABLE(pci, hns3_pci_tbl); | |
58 | ||
59 | static irqreturn_t hns3_irq_handle(int irq, void *dev) | |
60 | { | |
61 | struct hns3_enet_tqp_vector *tqp_vector = dev; | |
62 | ||
63 | napi_schedule(&tqp_vector->napi); | |
64 | ||
65 | return IRQ_HANDLED; | |
66 | } | |
67 | ||
68 | static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv) | |
69 | { | |
70 | struct hns3_enet_tqp_vector *tqp_vectors; | |
71 | unsigned int i; | |
72 | ||
73 | for (i = 0; i < priv->vector_num; i++) { | |
74 | tqp_vectors = &priv->tqp_vector[i]; | |
75 | ||
76 | if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED) | |
77 | continue; | |
78 | ||
79 | /* release the irq resource */ | |
80 | free_irq(tqp_vectors->vector_irq, tqp_vectors); | |
81 | tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED; | |
82 | } | |
83 | } | |
84 | ||
85 | static int hns3_nic_init_irq(struct hns3_nic_priv *priv) | |
86 | { | |
87 | struct hns3_enet_tqp_vector *tqp_vectors; | |
88 | int txrx_int_idx = 0; | |
89 | int rx_int_idx = 0; | |
90 | int tx_int_idx = 0; | |
91 | unsigned int i; | |
92 | int ret; | |
93 | ||
94 | for (i = 0; i < priv->vector_num; i++) { | |
95 | tqp_vectors = &priv->tqp_vector[i]; | |
96 | ||
97 | if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED) | |
98 | continue; | |
99 | ||
100 | if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) { | |
101 | snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1, | |
102 | "%s-%s-%d", priv->netdev->name, "TxRx", | |
103 | txrx_int_idx++); | |
104 | txrx_int_idx++; | |
105 | } else if (tqp_vectors->rx_group.ring) { | |
106 | snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1, | |
107 | "%s-%s-%d", priv->netdev->name, "Rx", | |
108 | rx_int_idx++); | |
109 | } else if (tqp_vectors->tx_group.ring) { | |
110 | snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN - 1, | |
111 | "%s-%s-%d", priv->netdev->name, "Tx", | |
112 | tx_int_idx++); | |
113 | } else { | |
114 | /* Skip this unused q_vector */ | |
115 | continue; | |
116 | } | |
117 | ||
118 | tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0'; | |
119 | ||
120 | ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0, | |
121 | tqp_vectors->name, | |
122 | tqp_vectors); | |
123 | if (ret) { | |
124 | netdev_err(priv->netdev, "request irq(%d) fail\n", | |
125 | tqp_vectors->vector_irq); | |
126 | return ret; | |
127 | } | |
128 | ||
129 | tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED; | |
130 | } | |
131 | ||
132 | return 0; | |
133 | } | |
134 | ||
135 | static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector, | |
136 | u32 mask_en) | |
137 | { | |
138 | writel(mask_en, tqp_vector->mask_addr); | |
139 | } | |
140 | ||
141 | static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector) | |
142 | { | |
143 | napi_enable(&tqp_vector->napi); | |
144 | ||
145 | /* enable vector */ | |
146 | hns3_mask_vector_irq(tqp_vector, 1); | |
147 | } | |
148 | ||
149 | static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector) | |
150 | { | |
151 | /* disable vector */ | |
152 | hns3_mask_vector_irq(tqp_vector, 0); | |
153 | ||
154 | disable_irq(tqp_vector->vector_irq); | |
155 | napi_disable(&tqp_vector->napi); | |
156 | } | |
157 | ||
158 | static void hns3_set_vector_coalesc_gl(struct hns3_enet_tqp_vector *tqp_vector, | |
159 | u32 gl_value) | |
160 | { | |
161 | /* this defines the configuration for GL (Interrupt Gap Limiter) | |
162 | * GL defines inter interrupt gap. | |
163 | * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing | |
164 | */ | |
165 | writel(gl_value, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET); | |
166 | writel(gl_value, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET); | |
167 | writel(gl_value, tqp_vector->mask_addr + HNS3_VECTOR_GL2_OFFSET); | |
168 | } | |
169 | ||
170 | static void hns3_set_vector_coalesc_rl(struct hns3_enet_tqp_vector *tqp_vector, | |
171 | u32 rl_value) | |
172 | { | |
173 | /* this defines the configuration for RL (Interrupt Rate Limiter). | |
174 | * Rl defines rate of interrupts i.e. number of interrupts-per-second | |
175 | * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing | |
176 | */ | |
177 | writel(rl_value, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET); | |
178 | } | |
179 | ||
180 | static void hns3_vector_gl_rl_init(struct hns3_enet_tqp_vector *tqp_vector) | |
181 | { | |
182 | /* initialize the configuration for interrupt coalescing. | |
183 | * 1. GL (Interrupt Gap Limiter) | |
184 | * 2. RL (Interrupt Rate Limiter) | |
185 | */ | |
186 | ||
187 | /* Default :enable interrupt coalesce */ | |
188 | tqp_vector->rx_group.int_gl = HNS3_INT_GL_50K; | |
189 | tqp_vector->tx_group.int_gl = HNS3_INT_GL_50K; | |
190 | hns3_set_vector_coalesc_gl(tqp_vector, HNS3_INT_GL_50K); | |
191 | /* for now we are disabling Interrupt RL - we | |
192 | * will re-enable later | |
193 | */ | |
194 | hns3_set_vector_coalesc_rl(tqp_vector, 0); | |
195 | tqp_vector->rx_group.flow_level = HNS3_FLOW_LOW; | |
196 | tqp_vector->tx_group.flow_level = HNS3_FLOW_LOW; | |
197 | } | |
198 | ||
9df8f79a YL |
199 | static int hns3_nic_set_real_num_queue(struct net_device *netdev) |
200 | { | |
9780cb97 | 201 | struct hnae3_handle *h = hns3_get_handle(netdev); |
9df8f79a YL |
202 | struct hnae3_knic_private_info *kinfo = &h->kinfo; |
203 | unsigned int queue_size = kinfo->rss_size * kinfo->num_tc; | |
204 | int ret; | |
205 | ||
206 | ret = netif_set_real_num_tx_queues(netdev, queue_size); | |
207 | if (ret) { | |
208 | netdev_err(netdev, | |
209 | "netif_set_real_num_tx_queues fail, ret=%d!\n", | |
210 | ret); | |
211 | return ret; | |
212 | } | |
213 | ||
214 | ret = netif_set_real_num_rx_queues(netdev, queue_size); | |
215 | if (ret) { | |
216 | netdev_err(netdev, | |
217 | "netif_set_real_num_rx_queues fail, ret=%d!\n", ret); | |
218 | return ret; | |
219 | } | |
220 | ||
221 | return 0; | |
222 | } | |
223 | ||
76ad4f0e S |
224 | static int hns3_nic_net_up(struct net_device *netdev) |
225 | { | |
226 | struct hns3_nic_priv *priv = netdev_priv(netdev); | |
227 | struct hnae3_handle *h = priv->ae_handle; | |
228 | int i, j; | |
229 | int ret; | |
230 | ||
231 | /* get irq resource for all vectors */ | |
232 | ret = hns3_nic_init_irq(priv); | |
233 | if (ret) { | |
234 | netdev_err(netdev, "hns init irq failed! ret=%d\n", ret); | |
235 | return ret; | |
236 | } | |
237 | ||
238 | /* enable the vectors */ | |
239 | for (i = 0; i < priv->vector_num; i++) | |
240 | hns3_vector_enable(&priv->tqp_vector[i]); | |
241 | ||
242 | /* start the ae_dev */ | |
243 | ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0; | |
244 | if (ret) | |
245 | goto out_start_err; | |
246 | ||
247 | return 0; | |
248 | ||
249 | out_start_err: | |
250 | for (j = i - 1; j >= 0; j--) | |
251 | hns3_vector_disable(&priv->tqp_vector[j]); | |
252 | ||
253 | hns3_nic_uninit_irq(priv); | |
254 | ||
255 | return ret; | |
256 | } | |
257 | ||
258 | static int hns3_nic_net_open(struct net_device *netdev) | |
259 | { | |
76ad4f0e S |
260 | int ret; |
261 | ||
262 | netif_carrier_off(netdev); | |
263 | ||
9df8f79a YL |
264 | ret = hns3_nic_set_real_num_queue(netdev); |
265 | if (ret) | |
76ad4f0e | 266 | return ret; |
76ad4f0e S |
267 | |
268 | ret = hns3_nic_net_up(netdev); | |
269 | if (ret) { | |
270 | netdev_err(netdev, | |
271 | "hns net up fail, ret=%d!\n", ret); | |
272 | return ret; | |
273 | } | |
274 | ||
275 | return 0; | |
276 | } | |
277 | ||
278 | static void hns3_nic_net_down(struct net_device *netdev) | |
279 | { | |
280 | struct hns3_nic_priv *priv = netdev_priv(netdev); | |
281 | const struct hnae3_ae_ops *ops; | |
282 | int i; | |
283 | ||
284 | /* stop ae_dev */ | |
285 | ops = priv->ae_handle->ae_algo->ops; | |
286 | if (ops->stop) | |
287 | ops->stop(priv->ae_handle); | |
288 | ||
289 | /* disable vectors */ | |
290 | for (i = 0; i < priv->vector_num; i++) | |
291 | hns3_vector_disable(&priv->tqp_vector[i]); | |
292 | ||
293 | /* free irq resources */ | |
294 | hns3_nic_uninit_irq(priv); | |
295 | } | |
296 | ||
297 | static int hns3_nic_net_stop(struct net_device *netdev) | |
298 | { | |
299 | netif_tx_stop_all_queues(netdev); | |
300 | netif_carrier_off(netdev); | |
301 | ||
302 | hns3_nic_net_down(netdev); | |
303 | ||
304 | return 0; | |
305 | } | |
306 | ||
76ad4f0e S |
307 | static int hns3_nic_uc_sync(struct net_device *netdev, |
308 | const unsigned char *addr) | |
309 | { | |
9780cb97 | 310 | struct hnae3_handle *h = hns3_get_handle(netdev); |
76ad4f0e S |
311 | |
312 | if (h->ae_algo->ops->add_uc_addr) | |
313 | return h->ae_algo->ops->add_uc_addr(h, addr); | |
314 | ||
315 | return 0; | |
316 | } | |
317 | ||
318 | static int hns3_nic_uc_unsync(struct net_device *netdev, | |
319 | const unsigned char *addr) | |
320 | { | |
9780cb97 | 321 | struct hnae3_handle *h = hns3_get_handle(netdev); |
76ad4f0e S |
322 | |
323 | if (h->ae_algo->ops->rm_uc_addr) | |
324 | return h->ae_algo->ops->rm_uc_addr(h, addr); | |
325 | ||
326 | return 0; | |
327 | } | |
328 | ||
329 | static int hns3_nic_mc_sync(struct net_device *netdev, | |
330 | const unsigned char *addr) | |
331 | { | |
9780cb97 | 332 | struct hnae3_handle *h = hns3_get_handle(netdev); |
76ad4f0e | 333 | |
720a8478 | 334 | if (h->ae_algo->ops->add_mc_addr) |
76ad4f0e S |
335 | return h->ae_algo->ops->add_mc_addr(h, addr); |
336 | ||
337 | return 0; | |
338 | } | |
339 | ||
340 | static int hns3_nic_mc_unsync(struct net_device *netdev, | |
341 | const unsigned char *addr) | |
342 | { | |
9780cb97 | 343 | struct hnae3_handle *h = hns3_get_handle(netdev); |
76ad4f0e | 344 | |
720a8478 | 345 | if (h->ae_algo->ops->rm_mc_addr) |
76ad4f0e S |
346 | return h->ae_algo->ops->rm_mc_addr(h, addr); |
347 | ||
348 | return 0; | |
349 | } | |
350 | ||
1db9b1bf | 351 | static void hns3_nic_set_rx_mode(struct net_device *netdev) |
76ad4f0e | 352 | { |
9780cb97 | 353 | struct hnae3_handle *h = hns3_get_handle(netdev); |
76ad4f0e S |
354 | |
355 | if (h->ae_algo->ops->set_promisc_mode) { | |
356 | if (netdev->flags & IFF_PROMISC) | |
357 | h->ae_algo->ops->set_promisc_mode(h, 1); | |
358 | else | |
359 | h->ae_algo->ops->set_promisc_mode(h, 0); | |
360 | } | |
361 | if (__dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync)) | |
362 | netdev_err(netdev, "sync uc address fail\n"); | |
363 | if (netdev->flags & IFF_MULTICAST) | |
364 | if (__dev_mc_sync(netdev, hns3_nic_mc_sync, hns3_nic_mc_unsync)) | |
365 | netdev_err(netdev, "sync mc address fail\n"); | |
366 | } | |
367 | ||
368 | static int hns3_set_tso(struct sk_buff *skb, u32 *paylen, | |
369 | u16 *mss, u32 *type_cs_vlan_tso) | |
370 | { | |
371 | u32 l4_offset, hdr_len; | |
372 | union l3_hdr_info l3; | |
373 | union l4_hdr_info l4; | |
374 | u32 l4_paylen; | |
375 | int ret; | |
376 | ||
377 | if (!skb_is_gso(skb)) | |
378 | return 0; | |
379 | ||
380 | ret = skb_cow_head(skb, 0); | |
381 | if (ret) | |
382 | return ret; | |
383 | ||
384 | l3.hdr = skb_network_header(skb); | |
385 | l4.hdr = skb_transport_header(skb); | |
386 | ||
387 | /* Software should clear the IPv4's checksum field when tso is | |
388 | * needed. | |
389 | */ | |
390 | if (l3.v4->version == 4) | |
391 | l3.v4->check = 0; | |
392 | ||
393 | /* tunnel packet.*/ | |
394 | if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE | | |
395 | SKB_GSO_GRE_CSUM | | |
396 | SKB_GSO_UDP_TUNNEL | | |
397 | SKB_GSO_UDP_TUNNEL_CSUM)) { | |
398 | if ((!(skb_shinfo(skb)->gso_type & | |
399 | SKB_GSO_PARTIAL)) && | |
400 | (skb_shinfo(skb)->gso_type & | |
401 | SKB_GSO_UDP_TUNNEL_CSUM)) { | |
402 | /* Software should clear the udp's checksum | |
403 | * field when tso is needed. | |
404 | */ | |
405 | l4.udp->check = 0; | |
406 | } | |
407 | /* reset l3&l4 pointers from outer to inner headers */ | |
408 | l3.hdr = skb_inner_network_header(skb); | |
409 | l4.hdr = skb_inner_transport_header(skb); | |
410 | ||
411 | /* Software should clear the IPv4's checksum field when | |
412 | * tso is needed. | |
413 | */ | |
414 | if (l3.v4->version == 4) | |
415 | l3.v4->check = 0; | |
416 | } | |
417 | ||
418 | /* normal or tunnel packet*/ | |
419 | l4_offset = l4.hdr - skb->data; | |
420 | hdr_len = (l4.tcp->doff * 4) + l4_offset; | |
421 | ||
422 | /* remove payload length from inner pseudo checksum when tso*/ | |
423 | l4_paylen = skb->len - l4_offset; | |
424 | csum_replace_by_diff(&l4.tcp->check, | |
425 | (__force __wsum)htonl(l4_paylen)); | |
426 | ||
427 | /* find the txbd field values */ | |
428 | *paylen = skb->len - hdr_len; | |
429 | hnae_set_bit(*type_cs_vlan_tso, | |
430 | HNS3_TXD_TSO_B, 1); | |
431 | ||
432 | /* get MSS for TSO */ | |
433 | *mss = skb_shinfo(skb)->gso_size; | |
434 | ||
435 | return 0; | |
436 | } | |
437 | ||
1898d4e4 S |
438 | static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto, |
439 | u8 *il4_proto) | |
76ad4f0e S |
440 | { |
441 | union { | |
442 | struct iphdr *v4; | |
443 | struct ipv6hdr *v6; | |
444 | unsigned char *hdr; | |
445 | } l3; | |
446 | unsigned char *l4_hdr; | |
447 | unsigned char *exthdr; | |
448 | u8 l4_proto_tmp; | |
449 | __be16 frag_off; | |
450 | ||
451 | /* find outer header point */ | |
452 | l3.hdr = skb_network_header(skb); | |
453 | l4_hdr = skb_inner_transport_header(skb); | |
454 | ||
455 | if (skb->protocol == htons(ETH_P_IPV6)) { | |
456 | exthdr = l3.hdr + sizeof(*l3.v6); | |
457 | l4_proto_tmp = l3.v6->nexthdr; | |
458 | if (l4_hdr != exthdr) | |
459 | ipv6_skip_exthdr(skb, exthdr - skb->data, | |
460 | &l4_proto_tmp, &frag_off); | |
461 | } else if (skb->protocol == htons(ETH_P_IP)) { | |
462 | l4_proto_tmp = l3.v4->protocol; | |
1898d4e4 S |
463 | } else { |
464 | return -EINVAL; | |
76ad4f0e S |
465 | } |
466 | ||
467 | *ol4_proto = l4_proto_tmp; | |
468 | ||
469 | /* tunnel packet */ | |
470 | if (!skb->encapsulation) { | |
471 | *il4_proto = 0; | |
1898d4e4 | 472 | return 0; |
76ad4f0e S |
473 | } |
474 | ||
475 | /* find inner header point */ | |
476 | l3.hdr = skb_inner_network_header(skb); | |
477 | l4_hdr = skb_inner_transport_header(skb); | |
478 | ||
479 | if (l3.v6->version == 6) { | |
480 | exthdr = l3.hdr + sizeof(*l3.v6); | |
481 | l4_proto_tmp = l3.v6->nexthdr; | |
482 | if (l4_hdr != exthdr) | |
483 | ipv6_skip_exthdr(skb, exthdr - skb->data, | |
484 | &l4_proto_tmp, &frag_off); | |
485 | } else if (l3.v4->version == 4) { | |
486 | l4_proto_tmp = l3.v4->protocol; | |
487 | } | |
488 | ||
489 | *il4_proto = l4_proto_tmp; | |
1898d4e4 S |
490 | |
491 | return 0; | |
76ad4f0e S |
492 | } |
493 | ||
494 | static void hns3_set_l2l3l4_len(struct sk_buff *skb, u8 ol4_proto, | |
495 | u8 il4_proto, u32 *type_cs_vlan_tso, | |
496 | u32 *ol_type_vlan_len_msec) | |
497 | { | |
498 | union { | |
499 | struct iphdr *v4; | |
500 | struct ipv6hdr *v6; | |
501 | unsigned char *hdr; | |
502 | } l3; | |
503 | union { | |
504 | struct tcphdr *tcp; | |
505 | struct udphdr *udp; | |
506 | struct gre_base_hdr *gre; | |
507 | unsigned char *hdr; | |
508 | } l4; | |
509 | unsigned char *l2_hdr; | |
510 | u8 l4_proto = ol4_proto; | |
511 | u32 ol2_len; | |
512 | u32 ol3_len; | |
513 | u32 ol4_len; | |
514 | u32 l2_len; | |
515 | u32 l3_len; | |
516 | ||
517 | l3.hdr = skb_network_header(skb); | |
518 | l4.hdr = skb_transport_header(skb); | |
519 | ||
520 | /* compute L2 header size for normal packet, defined in 2 Bytes */ | |
521 | l2_len = l3.hdr - skb->data; | |
522 | hnae_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_M, | |
523 | HNS3_TXD_L2LEN_S, l2_len >> 1); | |
524 | ||
525 | /* tunnel packet*/ | |
526 | if (skb->encapsulation) { | |
527 | /* compute OL2 header size, defined in 2 Bytes */ | |
528 | ol2_len = l2_len; | |
529 | hnae_set_field(*ol_type_vlan_len_msec, | |
530 | HNS3_TXD_L2LEN_M, | |
531 | HNS3_TXD_L2LEN_S, ol2_len >> 1); | |
532 | ||
533 | /* compute OL3 header size, defined in 4 Bytes */ | |
534 | ol3_len = l4.hdr - l3.hdr; | |
535 | hnae_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_M, | |
536 | HNS3_TXD_L3LEN_S, ol3_len >> 2); | |
537 | ||
538 | /* MAC in UDP, MAC in GRE (0x6558)*/ | |
539 | if ((ol4_proto == IPPROTO_UDP) || (ol4_proto == IPPROTO_GRE)) { | |
540 | /* switch MAC header ptr from outer to inner header.*/ | |
541 | l2_hdr = skb_inner_mac_header(skb); | |
542 | ||
543 | /* compute OL4 header size, defined in 4 Bytes. */ | |
544 | ol4_len = l2_hdr - l4.hdr; | |
545 | hnae_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L4LEN_M, | |
546 | HNS3_TXD_L4LEN_S, ol4_len >> 2); | |
547 | ||
548 | /* switch IP header ptr from outer to inner header */ | |
549 | l3.hdr = skb_inner_network_header(skb); | |
550 | ||
551 | /* compute inner l2 header size, defined in 2 Bytes. */ | |
552 | l2_len = l3.hdr - l2_hdr; | |
553 | hnae_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_M, | |
554 | HNS3_TXD_L2LEN_S, l2_len >> 1); | |
555 | } else { | |
556 | /* skb packet types not supported by hardware, | |
557 | * txbd len fild doesn't be filled. | |
558 | */ | |
559 | return; | |
560 | } | |
561 | ||
562 | /* switch L4 header pointer from outer to inner */ | |
563 | l4.hdr = skb_inner_transport_header(skb); | |
564 | ||
565 | l4_proto = il4_proto; | |
566 | } | |
567 | ||
568 | /* compute inner(/normal) L3 header size, defined in 4 Bytes */ | |
569 | l3_len = l4.hdr - l3.hdr; | |
570 | hnae_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_M, | |
571 | HNS3_TXD_L3LEN_S, l3_len >> 2); | |
572 | ||
573 | /* compute inner(/normal) L4 header size, defined in 4 Bytes */ | |
574 | switch (l4_proto) { | |
575 | case IPPROTO_TCP: | |
576 | hnae_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_M, | |
577 | HNS3_TXD_L4LEN_S, l4.tcp->doff); | |
578 | break; | |
579 | case IPPROTO_SCTP: | |
580 | hnae_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_M, | |
581 | HNS3_TXD_L4LEN_S, (sizeof(struct sctphdr) >> 2)); | |
582 | break; | |
583 | case IPPROTO_UDP: | |
584 | hnae_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_M, | |
585 | HNS3_TXD_L4LEN_S, (sizeof(struct udphdr) >> 2)); | |
586 | break; | |
587 | default: | |
588 | /* skb packet types not supported by hardware, | |
589 | * txbd len fild doesn't be filled. | |
590 | */ | |
591 | return; | |
592 | } | |
593 | } | |
594 | ||
595 | static int hns3_set_l3l4_type_csum(struct sk_buff *skb, u8 ol4_proto, | |
596 | u8 il4_proto, u32 *type_cs_vlan_tso, | |
597 | u32 *ol_type_vlan_len_msec) | |
598 | { | |
599 | union { | |
600 | struct iphdr *v4; | |
601 | struct ipv6hdr *v6; | |
602 | unsigned char *hdr; | |
603 | } l3; | |
604 | u32 l4_proto = ol4_proto; | |
605 | ||
606 | l3.hdr = skb_network_header(skb); | |
607 | ||
608 | /* define OL3 type and tunnel type(OL4).*/ | |
609 | if (skb->encapsulation) { | |
610 | /* define outer network header type.*/ | |
611 | if (skb->protocol == htons(ETH_P_IP)) { | |
612 | if (skb_is_gso(skb)) | |
613 | hnae_set_field(*ol_type_vlan_len_msec, | |
614 | HNS3_TXD_OL3T_M, HNS3_TXD_OL3T_S, | |
615 | HNS3_OL3T_IPV4_CSUM); | |
616 | else | |
617 | hnae_set_field(*ol_type_vlan_len_msec, | |
618 | HNS3_TXD_OL3T_M, HNS3_TXD_OL3T_S, | |
619 | HNS3_OL3T_IPV4_NO_CSUM); | |
620 | ||
621 | } else if (skb->protocol == htons(ETH_P_IPV6)) { | |
622 | hnae_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_M, | |
623 | HNS3_TXD_OL3T_S, HNS3_OL3T_IPV6); | |
624 | } | |
625 | ||
626 | /* define tunnel type(OL4).*/ | |
627 | switch (l4_proto) { | |
628 | case IPPROTO_UDP: | |
629 | hnae_set_field(*ol_type_vlan_len_msec, | |
630 | HNS3_TXD_TUNTYPE_M, | |
631 | HNS3_TXD_TUNTYPE_S, | |
632 | HNS3_TUN_MAC_IN_UDP); | |
633 | break; | |
634 | case IPPROTO_GRE: | |
635 | hnae_set_field(*ol_type_vlan_len_msec, | |
636 | HNS3_TXD_TUNTYPE_M, | |
637 | HNS3_TXD_TUNTYPE_S, | |
638 | HNS3_TUN_NVGRE); | |
639 | break; | |
640 | default: | |
641 | /* drop the skb tunnel packet if hardware don't support, | |
642 | * because hardware can't calculate csum when TSO. | |
643 | */ | |
644 | if (skb_is_gso(skb)) | |
645 | return -EDOM; | |
646 | ||
647 | /* the stack computes the IP header already, | |
648 | * driver calculate l4 checksum when not TSO. | |
649 | */ | |
650 | skb_checksum_help(skb); | |
651 | return 0; | |
652 | } | |
653 | ||
654 | l3.hdr = skb_inner_network_header(skb); | |
655 | l4_proto = il4_proto; | |
656 | } | |
657 | ||
658 | if (l3.v4->version == 4) { | |
659 | hnae_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_M, | |
660 | HNS3_TXD_L3T_S, HNS3_L3T_IPV4); | |
661 | ||
662 | /* the stack computes the IP header already, the only time we | |
663 | * need the hardware to recompute it is in the case of TSO. | |
664 | */ | |
665 | if (skb_is_gso(skb)) | |
666 | hnae_set_bit(*type_cs_vlan_tso, HNS3_TXD_L3CS_B, 1); | |
667 | ||
668 | hnae_set_bit(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1); | |
669 | } else if (l3.v6->version == 6) { | |
670 | hnae_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_M, | |
671 | HNS3_TXD_L3T_S, HNS3_L3T_IPV6); | |
672 | hnae_set_bit(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1); | |
673 | } | |
674 | ||
675 | switch (l4_proto) { | |
676 | case IPPROTO_TCP: | |
677 | hnae_set_field(*type_cs_vlan_tso, | |
678 | HNS3_TXD_L4T_M, | |
679 | HNS3_TXD_L4T_S, | |
680 | HNS3_L4T_TCP); | |
681 | break; | |
682 | case IPPROTO_UDP: | |
683 | hnae_set_field(*type_cs_vlan_tso, | |
684 | HNS3_TXD_L4T_M, | |
685 | HNS3_TXD_L4T_S, | |
686 | HNS3_L4T_UDP); | |
687 | break; | |
688 | case IPPROTO_SCTP: | |
689 | hnae_set_field(*type_cs_vlan_tso, | |
690 | HNS3_TXD_L4T_M, | |
691 | HNS3_TXD_L4T_S, | |
692 | HNS3_L4T_SCTP); | |
693 | break; | |
694 | default: | |
695 | /* drop the skb tunnel packet if hardware don't support, | |
696 | * because hardware can't calculate csum when TSO. | |
697 | */ | |
698 | if (skb_is_gso(skb)) | |
699 | return -EDOM; | |
700 | ||
701 | /* the stack computes the IP header already, | |
702 | * driver calculate l4 checksum when not TSO. | |
703 | */ | |
704 | skb_checksum_help(skb); | |
705 | return 0; | |
706 | } | |
707 | ||
708 | return 0; | |
709 | } | |
710 | ||
711 | static void hns3_set_txbd_baseinfo(u16 *bdtp_fe_sc_vld_ra_ri, int frag_end) | |
712 | { | |
713 | /* Config bd buffer end */ | |
714 | hnae_set_field(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_BDTYPE_M, | |
715 | HNS3_TXD_BDTYPE_M, 0); | |
716 | hnae_set_bit(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_FE_B, !!frag_end); | |
717 | hnae_set_bit(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_VLD_B, 1); | |
718 | hnae_set_field(*bdtp_fe_sc_vld_ra_ri, HNS3_TXD_SC_M, HNS3_TXD_SC_S, 1); | |
719 | } | |
720 | ||
721 | static int hns3_fill_desc(struct hns3_enet_ring *ring, void *priv, | |
722 | int size, dma_addr_t dma, int frag_end, | |
723 | enum hns_desc_type type) | |
724 | { | |
725 | struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use]; | |
726 | struct hns3_desc *desc = &ring->desc[ring->next_to_use]; | |
727 | u32 ol_type_vlan_len_msec = 0; | |
728 | u16 bdtp_fe_sc_vld_ra_ri = 0; | |
729 | u32 type_cs_vlan_tso = 0; | |
730 | struct sk_buff *skb; | |
731 | u32 paylen = 0; | |
732 | u16 mss = 0; | |
733 | __be16 protocol; | |
734 | u8 ol4_proto; | |
735 | u8 il4_proto; | |
736 | int ret; | |
737 | ||
738 | /* The txbd's baseinfo of DESC_TYPE_PAGE & DESC_TYPE_SKB */ | |
739 | desc_cb->priv = priv; | |
740 | desc_cb->length = size; | |
741 | desc_cb->dma = dma; | |
742 | desc_cb->type = type; | |
743 | ||
744 | /* now, fill the descriptor */ | |
745 | desc->addr = cpu_to_le64(dma); | |
746 | desc->tx.send_size = cpu_to_le16((u16)size); | |
747 | hns3_set_txbd_baseinfo(&bdtp_fe_sc_vld_ra_ri, frag_end); | |
748 | desc->tx.bdtp_fe_sc_vld_ra_ri = cpu_to_le16(bdtp_fe_sc_vld_ra_ri); | |
749 | ||
750 | if (type == DESC_TYPE_SKB) { | |
751 | skb = (struct sk_buff *)priv; | |
a90bb9a5 | 752 | paylen = skb->len; |
76ad4f0e S |
753 | |
754 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
755 | skb_reset_mac_len(skb); | |
756 | protocol = skb->protocol; | |
757 | ||
758 | /* vlan packet*/ | |
759 | if (protocol == htons(ETH_P_8021Q)) { | |
760 | protocol = vlan_get_protocol(skb); | |
761 | skb->protocol = protocol; | |
762 | } | |
1898d4e4 S |
763 | ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto); |
764 | if (ret) | |
765 | return ret; | |
76ad4f0e S |
766 | hns3_set_l2l3l4_len(skb, ol4_proto, il4_proto, |
767 | &type_cs_vlan_tso, | |
768 | &ol_type_vlan_len_msec); | |
769 | ret = hns3_set_l3l4_type_csum(skb, ol4_proto, il4_proto, | |
770 | &type_cs_vlan_tso, | |
771 | &ol_type_vlan_len_msec); | |
772 | if (ret) | |
773 | return ret; | |
774 | ||
775 | ret = hns3_set_tso(skb, &paylen, &mss, | |
776 | &type_cs_vlan_tso); | |
777 | if (ret) | |
778 | return ret; | |
779 | } | |
780 | ||
781 | /* Set txbd */ | |
782 | desc->tx.ol_type_vlan_len_msec = | |
783 | cpu_to_le32(ol_type_vlan_len_msec); | |
784 | desc->tx.type_cs_vlan_tso_len = | |
785 | cpu_to_le32(type_cs_vlan_tso); | |
a90bb9a5 | 786 | desc->tx.paylen = cpu_to_le32(paylen); |
76ad4f0e S |
787 | desc->tx.mss = cpu_to_le16(mss); |
788 | } | |
789 | ||
790 | /* move ring pointer to next.*/ | |
791 | ring_ptr_move_fw(ring, next_to_use); | |
792 | ||
793 | return 0; | |
794 | } | |
795 | ||
796 | static int hns3_fill_desc_tso(struct hns3_enet_ring *ring, void *priv, | |
797 | int size, dma_addr_t dma, int frag_end, | |
798 | enum hns_desc_type type) | |
799 | { | |
800 | unsigned int frag_buf_num; | |
801 | unsigned int k; | |
802 | int sizeoflast; | |
803 | int ret; | |
804 | ||
805 | frag_buf_num = (size + HNS3_MAX_BD_SIZE - 1) / HNS3_MAX_BD_SIZE; | |
806 | sizeoflast = size % HNS3_MAX_BD_SIZE; | |
807 | sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE; | |
808 | ||
809 | /* When the frag size is bigger than hardware, split this frag */ | |
810 | for (k = 0; k < frag_buf_num; k++) { | |
811 | ret = hns3_fill_desc(ring, priv, | |
812 | (k == frag_buf_num - 1) ? | |
813 | sizeoflast : HNS3_MAX_BD_SIZE, | |
814 | dma + HNS3_MAX_BD_SIZE * k, | |
815 | frag_end && (k == frag_buf_num - 1) ? 1 : 0, | |
816 | (type == DESC_TYPE_SKB && !k) ? | |
817 | DESC_TYPE_SKB : DESC_TYPE_PAGE); | |
818 | if (ret) | |
819 | return ret; | |
820 | } | |
821 | ||
822 | return 0; | |
823 | } | |
824 | ||
825 | static int hns3_nic_maybe_stop_tso(struct sk_buff **out_skb, int *bnum, | |
826 | struct hns3_enet_ring *ring) | |
827 | { | |
828 | struct sk_buff *skb = *out_skb; | |
829 | struct skb_frag_struct *frag; | |
830 | int bdnum_for_frag; | |
831 | int frag_num; | |
832 | int buf_num; | |
833 | int size; | |
834 | int i; | |
835 | ||
836 | size = skb_headlen(skb); | |
837 | buf_num = (size + HNS3_MAX_BD_SIZE - 1) / HNS3_MAX_BD_SIZE; | |
838 | ||
839 | frag_num = skb_shinfo(skb)->nr_frags; | |
840 | for (i = 0; i < frag_num; i++) { | |
841 | frag = &skb_shinfo(skb)->frags[i]; | |
842 | size = skb_frag_size(frag); | |
843 | bdnum_for_frag = | |
844 | (size + HNS3_MAX_BD_SIZE - 1) / HNS3_MAX_BD_SIZE; | |
845 | if (bdnum_for_frag > HNS3_MAX_BD_PER_FRAG) | |
846 | return -ENOMEM; | |
847 | ||
848 | buf_num += bdnum_for_frag; | |
849 | } | |
850 | ||
851 | if (buf_num > ring_space(ring)) | |
852 | return -EBUSY; | |
853 | ||
854 | *bnum = buf_num; | |
855 | return 0; | |
856 | } | |
857 | ||
858 | static int hns3_nic_maybe_stop_tx(struct sk_buff **out_skb, int *bnum, | |
859 | struct hns3_enet_ring *ring) | |
860 | { | |
861 | struct sk_buff *skb = *out_skb; | |
862 | int buf_num; | |
863 | ||
864 | /* No. of segments (plus a header) */ | |
865 | buf_num = skb_shinfo(skb)->nr_frags + 1; | |
866 | ||
867 | if (buf_num > ring_space(ring)) | |
868 | return -EBUSY; | |
869 | ||
870 | *bnum = buf_num; | |
871 | ||
872 | return 0; | |
873 | } | |
874 | ||
875 | static void hns_nic_dma_unmap(struct hns3_enet_ring *ring, int next_to_use_orig) | |
876 | { | |
877 | struct device *dev = ring_to_dev(ring); | |
878 | unsigned int i; | |
879 | ||
880 | for (i = 0; i < ring->desc_num; i++) { | |
881 | /* check if this is where we started */ | |
882 | if (ring->next_to_use == next_to_use_orig) | |
883 | break; | |
884 | ||
885 | /* unmap the descriptor dma address */ | |
886 | if (ring->desc_cb[ring->next_to_use].type == DESC_TYPE_SKB) | |
887 | dma_unmap_single(dev, | |
888 | ring->desc_cb[ring->next_to_use].dma, | |
889 | ring->desc_cb[ring->next_to_use].length, | |
890 | DMA_TO_DEVICE); | |
891 | else | |
892 | dma_unmap_page(dev, | |
893 | ring->desc_cb[ring->next_to_use].dma, | |
894 | ring->desc_cb[ring->next_to_use].length, | |
895 | DMA_TO_DEVICE); | |
896 | ||
897 | /* rollback one */ | |
898 | ring_ptr_move_bw(ring, next_to_use); | |
899 | } | |
900 | } | |
901 | ||
902 | static netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, | |
903 | struct net_device *netdev) | |
904 | { | |
905 | struct hns3_nic_priv *priv = netdev_priv(netdev); | |
906 | struct hns3_nic_ring_data *ring_data = | |
907 | &tx_ring_data(priv, skb->queue_mapping); | |
908 | struct hns3_enet_ring *ring = ring_data->ring; | |
909 | struct device *dev = priv->dev; | |
910 | struct netdev_queue *dev_queue; | |
911 | struct skb_frag_struct *frag; | |
912 | int next_to_use_head; | |
913 | int next_to_use_frag; | |
914 | dma_addr_t dma; | |
915 | int buf_num; | |
916 | int seg_num; | |
917 | int size; | |
918 | int ret; | |
919 | int i; | |
920 | ||
921 | /* Prefetch the data used later */ | |
922 | prefetch(skb->data); | |
923 | ||
924 | switch (priv->ops.maybe_stop_tx(&skb, &buf_num, ring)) { | |
925 | case -EBUSY: | |
926 | u64_stats_update_begin(&ring->syncp); | |
927 | ring->stats.tx_busy++; | |
928 | u64_stats_update_end(&ring->syncp); | |
929 | ||
930 | goto out_net_tx_busy; | |
931 | case -ENOMEM: | |
932 | u64_stats_update_begin(&ring->syncp); | |
933 | ring->stats.sw_err_cnt++; | |
934 | u64_stats_update_end(&ring->syncp); | |
935 | netdev_err(netdev, "no memory to xmit!\n"); | |
936 | ||
937 | goto out_err_tx_ok; | |
938 | default: | |
939 | break; | |
940 | } | |
941 | ||
942 | /* No. of segments (plus a header) */ | |
943 | seg_num = skb_shinfo(skb)->nr_frags + 1; | |
944 | /* Fill the first part */ | |
945 | size = skb_headlen(skb); | |
946 | ||
947 | next_to_use_head = ring->next_to_use; | |
948 | ||
949 | dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE); | |
950 | if (dma_mapping_error(dev, dma)) { | |
951 | netdev_err(netdev, "TX head DMA map failed\n"); | |
952 | ring->stats.sw_err_cnt++; | |
953 | goto out_err_tx_ok; | |
954 | } | |
955 | ||
956 | ret = priv->ops.fill_desc(ring, skb, size, dma, seg_num == 1 ? 1 : 0, | |
957 | DESC_TYPE_SKB); | |
958 | if (ret) | |
959 | goto head_dma_map_err; | |
960 | ||
961 | next_to_use_frag = ring->next_to_use; | |
962 | /* Fill the fragments */ | |
963 | for (i = 1; i < seg_num; i++) { | |
964 | frag = &skb_shinfo(skb)->frags[i - 1]; | |
965 | size = skb_frag_size(frag); | |
966 | dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE); | |
967 | if (dma_mapping_error(dev, dma)) { | |
968 | netdev_err(netdev, "TX frag(%d) DMA map failed\n", i); | |
969 | ring->stats.sw_err_cnt++; | |
970 | goto frag_dma_map_err; | |
971 | } | |
972 | ret = priv->ops.fill_desc(ring, skb_frag_page(frag), size, dma, | |
973 | seg_num - 1 == i ? 1 : 0, | |
974 | DESC_TYPE_PAGE); | |
975 | ||
976 | if (ret) | |
977 | goto frag_dma_map_err; | |
978 | } | |
979 | ||
980 | /* Complete translate all packets */ | |
981 | dev_queue = netdev_get_tx_queue(netdev, ring_data->queue_index); | |
982 | netdev_tx_sent_queue(dev_queue, skb->len); | |
983 | ||
984 | wmb(); /* Commit all data before submit */ | |
985 | ||
986 | hnae_queue_xmit(ring->tqp, buf_num); | |
987 | ||
988 | return NETDEV_TX_OK; | |
989 | ||
990 | frag_dma_map_err: | |
991 | hns_nic_dma_unmap(ring, next_to_use_frag); | |
992 | ||
993 | head_dma_map_err: | |
994 | hns_nic_dma_unmap(ring, next_to_use_head); | |
995 | ||
996 | out_err_tx_ok: | |
997 | dev_kfree_skb_any(skb); | |
998 | return NETDEV_TX_OK; | |
999 | ||
1000 | out_net_tx_busy: | |
1001 | netif_stop_subqueue(netdev, ring_data->queue_index); | |
1002 | smp_mb(); /* Commit all data before submit */ | |
1003 | ||
1004 | return NETDEV_TX_BUSY; | |
1005 | } | |
1006 | ||
1007 | static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p) | |
1008 | { | |
9780cb97 | 1009 | struct hnae3_handle *h = hns3_get_handle(netdev); |
76ad4f0e S |
1010 | struct sockaddr *mac_addr = p; |
1011 | int ret; | |
1012 | ||
1013 | if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data)) | |
1014 | return -EADDRNOTAVAIL; | |
1015 | ||
1016 | ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data); | |
1017 | if (ret) { | |
1018 | netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret); | |
1019 | return ret; | |
1020 | } | |
1021 | ||
1022 | ether_addr_copy(netdev->dev_addr, mac_addr->sa_data); | |
1023 | ||
1024 | return 0; | |
1025 | } | |
1026 | ||
1027 | static int hns3_nic_set_features(struct net_device *netdev, | |
1028 | netdev_features_t features) | |
1029 | { | |
1030 | struct hns3_nic_priv *priv = netdev_priv(netdev); | |
1031 | ||
1032 | if (features & (NETIF_F_TSO | NETIF_F_TSO6)) { | |
1033 | priv->ops.fill_desc = hns3_fill_desc_tso; | |
1034 | priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tso; | |
1035 | } else { | |
1036 | priv->ops.fill_desc = hns3_fill_desc; | |
1037 | priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tx; | |
1038 | } | |
1039 | ||
1040 | netdev->features = features; | |
1041 | return 0; | |
1042 | } | |
1043 | ||
1044 | static void | |
1045 | hns3_nic_get_stats64(struct net_device *netdev, struct rtnl_link_stats64 *stats) | |
1046 | { | |
1047 | struct hns3_nic_priv *priv = netdev_priv(netdev); | |
1048 | int queue_num = priv->ae_handle->kinfo.num_tqps; | |
1049 | struct hns3_enet_ring *ring; | |
1050 | unsigned int start; | |
1051 | unsigned int idx; | |
1052 | u64 tx_bytes = 0; | |
1053 | u64 rx_bytes = 0; | |
1054 | u64 tx_pkts = 0; | |
1055 | u64 rx_pkts = 0; | |
1056 | ||
1057 | for (idx = 0; idx < queue_num; idx++) { | |
1058 | /* fetch the tx stats */ | |
1059 | ring = priv->ring_data[idx].ring; | |
1060 | do { | |
d36d36ce | 1061 | start = u64_stats_fetch_begin_irq(&ring->syncp); |
76ad4f0e S |
1062 | tx_bytes += ring->stats.tx_bytes; |
1063 | tx_pkts += ring->stats.tx_pkts; | |
1064 | } while (u64_stats_fetch_retry_irq(&ring->syncp, start)); | |
1065 | ||
1066 | /* fetch the rx stats */ | |
1067 | ring = priv->ring_data[idx + queue_num].ring; | |
1068 | do { | |
d36d36ce | 1069 | start = u64_stats_fetch_begin_irq(&ring->syncp); |
76ad4f0e S |
1070 | rx_bytes += ring->stats.rx_bytes; |
1071 | rx_pkts += ring->stats.rx_pkts; | |
1072 | } while (u64_stats_fetch_retry_irq(&ring->syncp, start)); | |
1073 | } | |
1074 | ||
1075 | stats->tx_bytes = tx_bytes; | |
1076 | stats->tx_packets = tx_pkts; | |
1077 | stats->rx_bytes = rx_bytes; | |
1078 | stats->rx_packets = rx_pkts; | |
1079 | ||
1080 | stats->rx_errors = netdev->stats.rx_errors; | |
1081 | stats->multicast = netdev->stats.multicast; | |
1082 | stats->rx_length_errors = netdev->stats.rx_length_errors; | |
1083 | stats->rx_crc_errors = netdev->stats.rx_crc_errors; | |
1084 | stats->rx_missed_errors = netdev->stats.rx_missed_errors; | |
1085 | ||
1086 | stats->tx_errors = netdev->stats.tx_errors; | |
1087 | stats->rx_dropped = netdev->stats.rx_dropped; | |
1088 | stats->tx_dropped = netdev->stats.tx_dropped; | |
1089 | stats->collisions = netdev->stats.collisions; | |
1090 | stats->rx_over_errors = netdev->stats.rx_over_errors; | |
1091 | stats->rx_frame_errors = netdev->stats.rx_frame_errors; | |
1092 | stats->rx_fifo_errors = netdev->stats.rx_fifo_errors; | |
1093 | stats->tx_aborted_errors = netdev->stats.tx_aborted_errors; | |
1094 | stats->tx_carrier_errors = netdev->stats.tx_carrier_errors; | |
1095 | stats->tx_fifo_errors = netdev->stats.tx_fifo_errors; | |
1096 | stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors; | |
1097 | stats->tx_window_errors = netdev->stats.tx_window_errors; | |
1098 | stats->rx_compressed = netdev->stats.rx_compressed; | |
1099 | stats->tx_compressed = netdev->stats.tx_compressed; | |
1100 | } | |
1101 | ||
1102 | static void hns3_add_tunnel_port(struct net_device *netdev, u16 port, | |
1103 | enum hns3_udp_tnl_type type) | |
1104 | { | |
1105 | struct hns3_nic_priv *priv = netdev_priv(netdev); | |
1106 | struct hns3_udp_tunnel *udp_tnl = &priv->udp_tnl[type]; | |
1107 | struct hnae3_handle *h = priv->ae_handle; | |
1108 | ||
1109 | if (udp_tnl->used && udp_tnl->dst_port == port) { | |
1110 | udp_tnl->used++; | |
1111 | return; | |
1112 | } | |
1113 | ||
1114 | if (udp_tnl->used) { | |
1115 | netdev_warn(netdev, | |
1116 | "UDP tunnel [%d], port [%d] offload\n", type, port); | |
1117 | return; | |
1118 | } | |
1119 | ||
1120 | udp_tnl->dst_port = port; | |
1121 | udp_tnl->used = 1; | |
1122 | /* TBD send command to hardware to add port */ | |
1123 | if (h->ae_algo->ops->add_tunnel_udp) | |
1124 | h->ae_algo->ops->add_tunnel_udp(h, port); | |
1125 | } | |
1126 | ||
1127 | static void hns3_del_tunnel_port(struct net_device *netdev, u16 port, | |
1128 | enum hns3_udp_tnl_type type) | |
1129 | { | |
1130 | struct hns3_nic_priv *priv = netdev_priv(netdev); | |
1131 | struct hns3_udp_tunnel *udp_tnl = &priv->udp_tnl[type]; | |
1132 | struct hnae3_handle *h = priv->ae_handle; | |
1133 | ||
1134 | if (!udp_tnl->used || udp_tnl->dst_port != port) { | |
1135 | netdev_warn(netdev, | |
1136 | "Invalid UDP tunnel port %d\n", port); | |
1137 | return; | |
1138 | } | |
1139 | ||
1140 | udp_tnl->used--; | |
1141 | if (udp_tnl->used) | |
1142 | return; | |
1143 | ||
1144 | udp_tnl->dst_port = 0; | |
1145 | /* TBD send command to hardware to del port */ | |
1146 | if (h->ae_algo->ops->del_tunnel_udp) | |
9537e7cb | 1147 | h->ae_algo->ops->del_tunnel_udp(h, port); |
76ad4f0e S |
1148 | } |
1149 | ||
1150 | /* hns3_nic_udp_tunnel_add - Get notifiacetion about UDP tunnel ports | |
1151 | * @netdev: This physical ports's netdev | |
1152 | * @ti: Tunnel information | |
1153 | */ | |
1154 | static void hns3_nic_udp_tunnel_add(struct net_device *netdev, | |
1155 | struct udp_tunnel_info *ti) | |
1156 | { | |
1157 | u16 port_n = ntohs(ti->port); | |
1158 | ||
1159 | switch (ti->type) { | |
1160 | case UDP_TUNNEL_TYPE_VXLAN: | |
1161 | hns3_add_tunnel_port(netdev, port_n, HNS3_UDP_TNL_VXLAN); | |
1162 | break; | |
1163 | case UDP_TUNNEL_TYPE_GENEVE: | |
1164 | hns3_add_tunnel_port(netdev, port_n, HNS3_UDP_TNL_GENEVE); | |
1165 | break; | |
1166 | default: | |
1167 | netdev_err(netdev, "unsupported tunnel type %d\n", ti->type); | |
1168 | break; | |
1169 | } | |
1170 | } | |
1171 | ||
1172 | static void hns3_nic_udp_tunnel_del(struct net_device *netdev, | |
1173 | struct udp_tunnel_info *ti) | |
1174 | { | |
1175 | u16 port_n = ntohs(ti->port); | |
1176 | ||
1177 | switch (ti->type) { | |
1178 | case UDP_TUNNEL_TYPE_VXLAN: | |
1179 | hns3_del_tunnel_port(netdev, port_n, HNS3_UDP_TNL_VXLAN); | |
1180 | break; | |
1181 | case UDP_TUNNEL_TYPE_GENEVE: | |
1182 | hns3_del_tunnel_port(netdev, port_n, HNS3_UDP_TNL_GENEVE); | |
1183 | break; | |
1184 | default: | |
1185 | break; | |
1186 | } | |
1187 | } | |
1188 | ||
1189 | static int hns3_setup_tc(struct net_device *netdev, u8 tc) | |
1190 | { | |
9780cb97 | 1191 | struct hnae3_handle *h = hns3_get_handle(netdev); |
76ad4f0e S |
1192 | struct hnae3_knic_private_info *kinfo = &h->kinfo; |
1193 | unsigned int i; | |
1194 | int ret; | |
1195 | ||
1196 | if (tc > HNAE3_MAX_TC) | |
1197 | return -EINVAL; | |
1198 | ||
1199 | if (kinfo->num_tc == tc) | |
1200 | return 0; | |
1201 | ||
1202 | if (!netdev) | |
1203 | return -EINVAL; | |
1204 | ||
1205 | if (!tc) { | |
1206 | netdev_reset_tc(netdev); | |
1207 | return 0; | |
1208 | } | |
1209 | ||
1210 | /* Set num_tc for netdev */ | |
1211 | ret = netdev_set_num_tc(netdev, tc); | |
1212 | if (ret) | |
1213 | return ret; | |
1214 | ||
1215 | /* Set per TC queues for the VSI */ | |
1216 | for (i = 0; i < HNAE3_MAX_TC; i++) { | |
1217 | if (kinfo->tc_info[i].enable) | |
1218 | netdev_set_tc_queue(netdev, | |
1219 | kinfo->tc_info[i].tc, | |
1220 | kinfo->tc_info[i].tqp_count, | |
1221 | kinfo->tc_info[i].tqp_offset); | |
1222 | } | |
1223 | ||
1224 | return 0; | |
1225 | } | |
1226 | ||
2572ac53 | 1227 | static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type, |
de4784ca | 1228 | void *type_data) |
76ad4f0e | 1229 | { |
de4784ca JP |
1230 | struct tc_mqprio_qopt *mqprio = type_data; |
1231 | ||
74897ef0 | 1232 | if (type != TC_SETUP_MQPRIO) |
38cf0426 | 1233 | return -EOPNOTSUPP; |
76ad4f0e | 1234 | |
de4784ca | 1235 | return hns3_setup_tc(dev, mqprio->num_tc); |
76ad4f0e S |
1236 | } |
1237 | ||
1238 | static int hns3_vlan_rx_add_vid(struct net_device *netdev, | |
1239 | __be16 proto, u16 vid) | |
1240 | { | |
9780cb97 | 1241 | struct hnae3_handle *h = hns3_get_handle(netdev); |
76ad4f0e S |
1242 | int ret = -EIO; |
1243 | ||
1244 | if (h->ae_algo->ops->set_vlan_filter) | |
1245 | ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false); | |
1246 | ||
1247 | return ret; | |
1248 | } | |
1249 | ||
1250 | static int hns3_vlan_rx_kill_vid(struct net_device *netdev, | |
1251 | __be16 proto, u16 vid) | |
1252 | { | |
9780cb97 | 1253 | struct hnae3_handle *h = hns3_get_handle(netdev); |
76ad4f0e S |
1254 | int ret = -EIO; |
1255 | ||
1256 | if (h->ae_algo->ops->set_vlan_filter) | |
1257 | ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true); | |
1258 | ||
1259 | return ret; | |
1260 | } | |
1261 | ||
1262 | static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, | |
1263 | u8 qos, __be16 vlan_proto) | |
1264 | { | |
9780cb97 | 1265 | struct hnae3_handle *h = hns3_get_handle(netdev); |
76ad4f0e S |
1266 | int ret = -EIO; |
1267 | ||
1268 | if (h->ae_algo->ops->set_vf_vlan_filter) | |
1269 | ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan, | |
1270 | qos, vlan_proto); | |
1271 | ||
1272 | return ret; | |
1273 | } | |
1274 | ||
a8e8b7ff S |
1275 | static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu) |
1276 | { | |
9780cb97 | 1277 | struct hnae3_handle *h = hns3_get_handle(netdev); |
a8e8b7ff S |
1278 | bool if_running = netif_running(netdev); |
1279 | int ret; | |
1280 | ||
1281 | if (!h->ae_algo->ops->set_mtu) | |
1282 | return -EOPNOTSUPP; | |
1283 | ||
1284 | /* if this was called with netdev up then bring netdevice down */ | |
1285 | if (if_running) { | |
1286 | (void)hns3_nic_net_stop(netdev); | |
1287 | msleep(100); | |
1288 | } | |
1289 | ||
1290 | ret = h->ae_algo->ops->set_mtu(h, new_mtu); | |
1291 | if (ret) { | |
1292 | netdev_err(netdev, "failed to change MTU in hardware %d\n", | |
1293 | ret); | |
1294 | return ret; | |
1295 | } | |
1296 | ||
1297 | /* if the netdev was running earlier, bring it up again */ | |
1298 | if (if_running && hns3_nic_net_open(netdev)) | |
1299 | ret = -EINVAL; | |
1300 | ||
1301 | return ret; | |
1302 | } | |
1303 | ||
76ad4f0e S |
1304 | static const struct net_device_ops hns3_nic_netdev_ops = { |
1305 | .ndo_open = hns3_nic_net_open, | |
1306 | .ndo_stop = hns3_nic_net_stop, | |
1307 | .ndo_start_xmit = hns3_nic_net_xmit, | |
1308 | .ndo_set_mac_address = hns3_nic_net_set_mac_address, | |
a8e8b7ff | 1309 | .ndo_change_mtu = hns3_nic_change_mtu, |
76ad4f0e S |
1310 | .ndo_set_features = hns3_nic_set_features, |
1311 | .ndo_get_stats64 = hns3_nic_get_stats64, | |
1312 | .ndo_setup_tc = hns3_nic_setup_tc, | |
1313 | .ndo_set_rx_mode = hns3_nic_set_rx_mode, | |
1314 | .ndo_udp_tunnel_add = hns3_nic_udp_tunnel_add, | |
1315 | .ndo_udp_tunnel_del = hns3_nic_udp_tunnel_del, | |
1316 | .ndo_vlan_rx_add_vid = hns3_vlan_rx_add_vid, | |
1317 | .ndo_vlan_rx_kill_vid = hns3_vlan_rx_kill_vid, | |
1318 | .ndo_set_vf_vlan = hns3_ndo_set_vf_vlan, | |
1319 | }; | |
1320 | ||
1321 | /* hns3_probe - Device initialization routine | |
1322 | * @pdev: PCI device information struct | |
1323 | * @ent: entry in hns3_pci_tbl | |
1324 | * | |
1325 | * hns3_probe initializes a PF identified by a pci_dev structure. | |
1326 | * The OS initialization, configuring of the PF private structure, | |
1327 | * and a hardware reset occur. | |
1328 | * | |
1329 | * Returns 0 on success, negative on failure | |
1330 | */ | |
1331 | static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent) | |
1332 | { | |
1333 | struct hnae3_ae_dev *ae_dev; | |
1334 | int ret; | |
1335 | ||
1336 | ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev), | |
1337 | GFP_KERNEL); | |
1338 | if (!ae_dev) { | |
1339 | ret = -ENOMEM; | |
1340 | return ret; | |
1341 | } | |
1342 | ||
1343 | ae_dev->pdev = pdev; | |
e92a0843 | 1344 | ae_dev->flag = ent->driver_data; |
76ad4f0e S |
1345 | ae_dev->dev_type = HNAE3_DEV_KNIC; |
1346 | pci_set_drvdata(pdev, ae_dev); | |
1347 | ||
1348 | return hnae3_register_ae_dev(ae_dev); | |
1349 | } | |
1350 | ||
1351 | /* hns3_remove - Device removal routine | |
1352 | * @pdev: PCI device information struct | |
1353 | */ | |
1354 | static void hns3_remove(struct pci_dev *pdev) | |
1355 | { | |
1356 | struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev); | |
1357 | ||
1358 | hnae3_unregister_ae_dev(ae_dev); | |
1359 | ||
1360 | devm_kfree(&pdev->dev, ae_dev); | |
1361 | ||
1362 | pci_set_drvdata(pdev, NULL); | |
1363 | } | |
1364 | ||
1365 | static struct pci_driver hns3_driver = { | |
1366 | .name = hns3_driver_name, | |
1367 | .id_table = hns3_pci_tbl, | |
1368 | .probe = hns3_probe, | |
1369 | .remove = hns3_remove, | |
1370 | }; | |
1371 | ||
1372 | /* set default feature to hns3 */ | |
1373 | static void hns3_set_default_feature(struct net_device *netdev) | |
1374 | { | |
1375 | netdev->priv_flags |= IFF_UNICAST_FLT; | |
1376 | ||
1377 | netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | | |
1378 | NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO | | |
1379 | NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE | | |
1380 | NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL | | |
1381 | NETIF_F_GSO_UDP_TUNNEL_CSUM; | |
1382 | ||
1383 | netdev->hw_enc_features |= NETIF_F_TSO_MANGLEID; | |
1384 | ||
1385 | netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM; | |
1386 | ||
1387 | netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | | |
1388 | NETIF_F_HW_VLAN_CTAG_FILTER | | |
1389 | NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO | | |
1390 | NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE | | |
1391 | NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL | | |
1392 | NETIF_F_GSO_UDP_TUNNEL_CSUM; | |
1393 | ||
1394 | netdev->vlan_features |= | |
1395 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM | | |
1396 | NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO | | |
1397 | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE | | |
1398 | NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL | | |
1399 | NETIF_F_GSO_UDP_TUNNEL_CSUM; | |
1400 | ||
1401 | netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | | |
1402 | NETIF_F_HW_VLAN_CTAG_FILTER | | |
1403 | NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO | | |
1404 | NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE | | |
1405 | NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL | | |
1406 | NETIF_F_GSO_UDP_TUNNEL_CSUM; | |
1407 | } | |
1408 | ||
1409 | static int hns3_alloc_buffer(struct hns3_enet_ring *ring, | |
1410 | struct hns3_desc_cb *cb) | |
1411 | { | |
1412 | unsigned int order = hnae_page_order(ring); | |
1413 | struct page *p; | |
1414 | ||
1415 | p = dev_alloc_pages(order); | |
1416 | if (!p) | |
1417 | return -ENOMEM; | |
1418 | ||
1419 | cb->priv = p; | |
1420 | cb->page_offset = 0; | |
1421 | cb->reuse_flag = 0; | |
1422 | cb->buf = page_address(p); | |
1423 | cb->length = hnae_page_size(ring); | |
1424 | cb->type = DESC_TYPE_PAGE; | |
1425 | ||
1426 | memset(cb->buf, 0, cb->length); | |
1427 | ||
1428 | return 0; | |
1429 | } | |
1430 | ||
1431 | static void hns3_free_buffer(struct hns3_enet_ring *ring, | |
1432 | struct hns3_desc_cb *cb) | |
1433 | { | |
1434 | if (cb->type == DESC_TYPE_SKB) | |
1435 | dev_kfree_skb_any((struct sk_buff *)cb->priv); | |
1436 | else if (!HNAE3_IS_TX_RING(ring)) | |
1437 | put_page((struct page *)cb->priv); | |
1438 | memset(cb, 0, sizeof(*cb)); | |
1439 | } | |
1440 | ||
1441 | static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb) | |
1442 | { | |
1443 | cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0, | |
1444 | cb->length, ring_to_dma_dir(ring)); | |
1445 | ||
1446 | if (dma_mapping_error(ring_to_dev(ring), cb->dma)) | |
1447 | return -EIO; | |
1448 | ||
1449 | return 0; | |
1450 | } | |
1451 | ||
1452 | static void hns3_unmap_buffer(struct hns3_enet_ring *ring, | |
1453 | struct hns3_desc_cb *cb) | |
1454 | { | |
1455 | if (cb->type == DESC_TYPE_SKB) | |
1456 | dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length, | |
1457 | ring_to_dma_dir(ring)); | |
1458 | else | |
1459 | dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length, | |
1460 | ring_to_dma_dir(ring)); | |
1461 | } | |
1462 | ||
1463 | static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i) | |
1464 | { | |
1465 | hns3_unmap_buffer(ring, &ring->desc_cb[i]); | |
1466 | ring->desc[i].addr = 0; | |
1467 | } | |
1468 | ||
1469 | static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i) | |
1470 | { | |
1471 | struct hns3_desc_cb *cb = &ring->desc_cb[i]; | |
1472 | ||
1473 | if (!ring->desc_cb[i].dma) | |
1474 | return; | |
1475 | ||
1476 | hns3_buffer_detach(ring, i); | |
1477 | hns3_free_buffer(ring, cb); | |
1478 | } | |
1479 | ||
1480 | static void hns3_free_buffers(struct hns3_enet_ring *ring) | |
1481 | { | |
1482 | int i; | |
1483 | ||
1484 | for (i = 0; i < ring->desc_num; i++) | |
1485 | hns3_free_buffer_detach(ring, i); | |
1486 | } | |
1487 | ||
1488 | /* free desc along with its attached buffer */ | |
1489 | static void hns3_free_desc(struct hns3_enet_ring *ring) | |
1490 | { | |
1491 | hns3_free_buffers(ring); | |
1492 | ||
1493 | dma_unmap_single(ring_to_dev(ring), ring->desc_dma_addr, | |
1494 | ring->desc_num * sizeof(ring->desc[0]), | |
1495 | DMA_BIDIRECTIONAL); | |
1496 | ring->desc_dma_addr = 0; | |
1497 | kfree(ring->desc); | |
1498 | ring->desc = NULL; | |
1499 | } | |
1500 | ||
1501 | static int hns3_alloc_desc(struct hns3_enet_ring *ring) | |
1502 | { | |
1503 | int size = ring->desc_num * sizeof(ring->desc[0]); | |
1504 | ||
1505 | ring->desc = kzalloc(size, GFP_KERNEL); | |
1506 | if (!ring->desc) | |
1507 | return -ENOMEM; | |
1508 | ||
1509 | ring->desc_dma_addr = dma_map_single(ring_to_dev(ring), ring->desc, | |
1510 | size, DMA_BIDIRECTIONAL); | |
1511 | if (dma_mapping_error(ring_to_dev(ring), ring->desc_dma_addr)) { | |
1512 | ring->desc_dma_addr = 0; | |
1513 | kfree(ring->desc); | |
1514 | ring->desc = NULL; | |
1515 | return -ENOMEM; | |
1516 | } | |
1517 | ||
1518 | return 0; | |
1519 | } | |
1520 | ||
1521 | static int hns3_reserve_buffer_map(struct hns3_enet_ring *ring, | |
1522 | struct hns3_desc_cb *cb) | |
1523 | { | |
1524 | int ret; | |
1525 | ||
1526 | ret = hns3_alloc_buffer(ring, cb); | |
1527 | if (ret) | |
1528 | goto out; | |
1529 | ||
1530 | ret = hns3_map_buffer(ring, cb); | |
1531 | if (ret) | |
1532 | goto out_with_buf; | |
1533 | ||
1534 | return 0; | |
1535 | ||
1536 | out_with_buf: | |
1537 | hns3_free_buffers(ring); | |
1538 | out: | |
1539 | return ret; | |
1540 | } | |
1541 | ||
1542 | static int hns3_alloc_buffer_attach(struct hns3_enet_ring *ring, int i) | |
1543 | { | |
1544 | int ret = hns3_reserve_buffer_map(ring, &ring->desc_cb[i]); | |
1545 | ||
1546 | if (ret) | |
1547 | return ret; | |
1548 | ||
1549 | ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma); | |
1550 | ||
1551 | return 0; | |
1552 | } | |
1553 | ||
1554 | /* Allocate memory for raw pkg, and map with dma */ | |
1555 | static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring) | |
1556 | { | |
1557 | int i, j, ret; | |
1558 | ||
1559 | for (i = 0; i < ring->desc_num; i++) { | |
1560 | ret = hns3_alloc_buffer_attach(ring, i); | |
1561 | if (ret) | |
1562 | goto out_buffer_fail; | |
1563 | } | |
1564 | ||
1565 | return 0; | |
1566 | ||
1567 | out_buffer_fail: | |
1568 | for (j = i - 1; j >= 0; j--) | |
1569 | hns3_free_buffer_detach(ring, j); | |
1570 | return ret; | |
1571 | } | |
1572 | ||
1573 | /* detach a in-used buffer and replace with a reserved one */ | |
1574 | static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i, | |
1575 | struct hns3_desc_cb *res_cb) | |
1576 | { | |
1577 | hns3_map_buffer(ring, &ring->desc_cb[i]); | |
1578 | ring->desc_cb[i] = *res_cb; | |
1579 | ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma); | |
1580 | } | |
1581 | ||
1582 | static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i) | |
1583 | { | |
1584 | ring->desc_cb[i].reuse_flag = 0; | |
1585 | ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma | |
1586 | + ring->desc_cb[i].page_offset); | |
1587 | } | |
1588 | ||
1589 | static void hns3_nic_reclaim_one_desc(struct hns3_enet_ring *ring, int *bytes, | |
1590 | int *pkts) | |
1591 | { | |
1592 | struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean]; | |
1593 | ||
1594 | (*pkts) += (desc_cb->type == DESC_TYPE_SKB); | |
1595 | (*bytes) += desc_cb->length; | |
1596 | /* desc_cb will be cleaned, after hnae_free_buffer_detach*/ | |
1597 | hns3_free_buffer_detach(ring, ring->next_to_clean); | |
1598 | ||
1599 | ring_ptr_move_fw(ring, next_to_clean); | |
1600 | } | |
1601 | ||
1602 | static int is_valid_clean_head(struct hns3_enet_ring *ring, int h) | |
1603 | { | |
1604 | int u = ring->next_to_use; | |
1605 | int c = ring->next_to_clean; | |
1606 | ||
1607 | if (unlikely(h > ring->desc_num)) | |
1608 | return 0; | |
1609 | ||
1610 | return u > c ? (h > c && h <= u) : (h > c || h <= u); | |
1611 | } | |
1612 | ||
1613 | int hns3_clean_tx_ring(struct hns3_enet_ring *ring, int budget) | |
1614 | { | |
1615 | struct net_device *netdev = ring->tqp->handle->kinfo.netdev; | |
1616 | struct netdev_queue *dev_queue; | |
1617 | int bytes, pkts; | |
1618 | int head; | |
1619 | ||
1620 | head = readl_relaxed(ring->tqp->io_base + HNS3_RING_TX_RING_HEAD_REG); | |
1621 | rmb(); /* Make sure head is ready before touch any data */ | |
1622 | ||
1623 | if (is_ring_empty(ring) || head == ring->next_to_clean) | |
1624 | return 0; /* no data to poll */ | |
1625 | ||
1626 | if (!is_valid_clean_head(ring, head)) { | |
1627 | netdev_err(netdev, "wrong head (%d, %d-%d)\n", head, | |
1628 | ring->next_to_use, ring->next_to_clean); | |
1629 | ||
1630 | u64_stats_update_begin(&ring->syncp); | |
1631 | ring->stats.io_err_cnt++; | |
1632 | u64_stats_update_end(&ring->syncp); | |
1633 | return -EIO; | |
1634 | } | |
1635 | ||
1636 | bytes = 0; | |
1637 | pkts = 0; | |
1638 | while (head != ring->next_to_clean && budget) { | |
1639 | hns3_nic_reclaim_one_desc(ring, &bytes, &pkts); | |
1640 | /* Issue prefetch for next Tx descriptor */ | |
1641 | prefetch(&ring->desc_cb[ring->next_to_clean]); | |
1642 | budget--; | |
1643 | } | |
1644 | ||
1645 | ring->tqp_vector->tx_group.total_bytes += bytes; | |
1646 | ring->tqp_vector->tx_group.total_packets += pkts; | |
1647 | ||
1648 | u64_stats_update_begin(&ring->syncp); | |
1649 | ring->stats.tx_bytes += bytes; | |
1650 | ring->stats.tx_pkts += pkts; | |
1651 | u64_stats_update_end(&ring->syncp); | |
1652 | ||
1653 | dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index); | |
1654 | netdev_tx_completed_queue(dev_queue, pkts, bytes); | |
1655 | ||
1656 | if (unlikely(pkts && netif_carrier_ok(netdev) && | |
1657 | (ring_space(ring) > HNS3_MAX_BD_PER_PKT))) { | |
1658 | /* Make sure that anybody stopping the queue after this | |
1659 | * sees the new next_to_clean. | |
1660 | */ | |
1661 | smp_mb(); | |
1662 | if (netif_tx_queue_stopped(dev_queue)) { | |
1663 | netif_tx_wake_queue(dev_queue); | |
1664 | ring->stats.restart_queue++; | |
1665 | } | |
1666 | } | |
1667 | ||
1668 | return !!budget; | |
1669 | } | |
1670 | ||
1671 | static int hns3_desc_unused(struct hns3_enet_ring *ring) | |
1672 | { | |
1673 | int ntc = ring->next_to_clean; | |
1674 | int ntu = ring->next_to_use; | |
1675 | ||
1676 | return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu; | |
1677 | } | |
1678 | ||
1679 | static void | |
1680 | hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring, int cleand_count) | |
1681 | { | |
1682 | struct hns3_desc_cb *desc_cb; | |
1683 | struct hns3_desc_cb res_cbs; | |
1684 | int i, ret; | |
1685 | ||
1686 | for (i = 0; i < cleand_count; i++) { | |
1687 | desc_cb = &ring->desc_cb[ring->next_to_use]; | |
1688 | if (desc_cb->reuse_flag) { | |
1689 | u64_stats_update_begin(&ring->syncp); | |
1690 | ring->stats.reuse_pg_cnt++; | |
1691 | u64_stats_update_end(&ring->syncp); | |
1692 | ||
1693 | hns3_reuse_buffer(ring, ring->next_to_use); | |
1694 | } else { | |
1695 | ret = hns3_reserve_buffer_map(ring, &res_cbs); | |
1696 | if (ret) { | |
1697 | u64_stats_update_begin(&ring->syncp); | |
1698 | ring->stats.sw_err_cnt++; | |
1699 | u64_stats_update_end(&ring->syncp); | |
1700 | ||
1701 | netdev_err(ring->tqp->handle->kinfo.netdev, | |
1702 | "hnae reserve buffer map failed.\n"); | |
1703 | break; | |
1704 | } | |
1705 | hns3_replace_buffer(ring, ring->next_to_use, &res_cbs); | |
1706 | } | |
1707 | ||
1708 | ring_ptr_move_fw(ring, next_to_use); | |
1709 | } | |
1710 | ||
1711 | wmb(); /* Make all data has been write before submit */ | |
1712 | writel_relaxed(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG); | |
1713 | } | |
1714 | ||
1715 | /* hns3_nic_get_headlen - determine size of header for LRO/GRO | |
1716 | * @data: pointer to the start of the headers | |
1717 | * @max: total length of section to find headers in | |
1718 | * | |
1719 | * This function is meant to determine the length of headers that will | |
1720 | * be recognized by hardware for LRO, GRO, and RSC offloads. The main | |
1721 | * motivation of doing this is to only perform one pull for IPv4 TCP | |
1722 | * packets so that we can do basic things like calculating the gso_size | |
1723 | * based on the average data per packet. | |
1724 | */ | |
1725 | static unsigned int hns3_nic_get_headlen(unsigned char *data, u32 flag, | |
1726 | unsigned int max_size) | |
1727 | { | |
1728 | unsigned char *network; | |
1729 | u8 hlen; | |
1730 | ||
1731 | /* This should never happen, but better safe than sorry */ | |
1732 | if (max_size < ETH_HLEN) | |
1733 | return max_size; | |
1734 | ||
1735 | /* Initialize network frame pointer */ | |
1736 | network = data; | |
1737 | ||
1738 | /* Set first protocol and move network header forward */ | |
1739 | network += ETH_HLEN; | |
1740 | ||
1741 | /* Handle any vlan tag if present */ | |
1742 | if (hnae_get_field(flag, HNS3_RXD_VLAN_M, HNS3_RXD_VLAN_S) | |
1743 | == HNS3_RX_FLAG_VLAN_PRESENT) { | |
1744 | if ((typeof(max_size))(network - data) > (max_size - VLAN_HLEN)) | |
1745 | return max_size; | |
1746 | ||
1747 | network += VLAN_HLEN; | |
1748 | } | |
1749 | ||
1750 | /* Handle L3 protocols */ | |
1751 | if (hnae_get_field(flag, HNS3_RXD_L3ID_M, HNS3_RXD_L3ID_S) | |
1752 | == HNS3_RX_FLAG_L3ID_IPV4) { | |
1753 | if ((typeof(max_size))(network - data) > | |
1754 | (max_size - sizeof(struct iphdr))) | |
1755 | return max_size; | |
1756 | ||
1757 | /* Access ihl as a u8 to avoid unaligned access on ia64 */ | |
1758 | hlen = (network[0] & 0x0F) << 2; | |
1759 | ||
1760 | /* Verify hlen meets minimum size requirements */ | |
1761 | if (hlen < sizeof(struct iphdr)) | |
1762 | return network - data; | |
1763 | ||
1764 | /* Record next protocol if header is present */ | |
1765 | } else if (hnae_get_field(flag, HNS3_RXD_L3ID_M, HNS3_RXD_L3ID_S) | |
1766 | == HNS3_RX_FLAG_L3ID_IPV6) { | |
1767 | if ((typeof(max_size))(network - data) > | |
1768 | (max_size - sizeof(struct ipv6hdr))) | |
1769 | return max_size; | |
1770 | ||
1771 | /* Record next protocol */ | |
1772 | hlen = sizeof(struct ipv6hdr); | |
1773 | } else { | |
1774 | return network - data; | |
1775 | } | |
1776 | ||
1777 | /* Relocate pointer to start of L4 header */ | |
1778 | network += hlen; | |
1779 | ||
1780 | /* Finally sort out TCP/UDP */ | |
1781 | if (hnae_get_field(flag, HNS3_RXD_L4ID_M, HNS3_RXD_L4ID_S) | |
1782 | == HNS3_RX_FLAG_L4ID_TCP) { | |
1783 | if ((typeof(max_size))(network - data) > | |
1784 | (max_size - sizeof(struct tcphdr))) | |
1785 | return max_size; | |
1786 | ||
1787 | /* Access doff as a u8 to avoid unaligned access on ia64 */ | |
1788 | hlen = (network[12] & 0xF0) >> 2; | |
1789 | ||
1790 | /* Verify hlen meets minimum size requirements */ | |
1791 | if (hlen < sizeof(struct tcphdr)) | |
1792 | return network - data; | |
1793 | ||
1794 | network += hlen; | |
1795 | } else if (hnae_get_field(flag, HNS3_RXD_L4ID_M, HNS3_RXD_L4ID_S) | |
1796 | == HNS3_RX_FLAG_L4ID_UDP) { | |
1797 | if ((typeof(max_size))(network - data) > | |
1798 | (max_size - sizeof(struct udphdr))) | |
1799 | return max_size; | |
1800 | ||
1801 | network += sizeof(struct udphdr); | |
1802 | } | |
1803 | ||
1804 | /* If everything has gone correctly network should be the | |
1805 | * data section of the packet and will be the end of the header. | |
1806 | * If not then it probably represents the end of the last recognized | |
1807 | * header. | |
1808 | */ | |
1809 | if ((typeof(max_size))(network - data) < max_size) | |
1810 | return network - data; | |
1811 | else | |
1812 | return max_size; | |
1813 | } | |
1814 | ||
1815 | static void hns3_nic_reuse_page(struct sk_buff *skb, int i, | |
1816 | struct hns3_enet_ring *ring, int pull_len, | |
1817 | struct hns3_desc_cb *desc_cb) | |
1818 | { | |
1819 | struct hns3_desc *desc; | |
1820 | int truesize, size; | |
1821 | int last_offset; | |
1822 | bool twobufs; | |
1823 | ||
1824 | twobufs = ((PAGE_SIZE < 8192) && | |
1825 | hnae_buf_size(ring) == HNS3_BUFFER_SIZE_2048); | |
1826 | ||
1827 | desc = &ring->desc[ring->next_to_clean]; | |
1828 | size = le16_to_cpu(desc->rx.size); | |
1829 | ||
1830 | if (twobufs) { | |
1831 | truesize = hnae_buf_size(ring); | |
1832 | } else { | |
1833 | truesize = ALIGN(size, L1_CACHE_BYTES); | |
1834 | last_offset = hnae_page_size(ring) - hnae_buf_size(ring); | |
1835 | } | |
1836 | ||
1837 | skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len, | |
1838 | size - pull_len, truesize - pull_len); | |
1839 | ||
1840 | /* Avoid re-using remote pages,flag default unreuse */ | |
1841 | if (unlikely(page_to_nid(desc_cb->priv) != numa_node_id())) | |
1842 | return; | |
1843 | ||
1844 | if (twobufs) { | |
1845 | /* If we are only owner of page we can reuse it */ | |
1846 | if (likely(page_count(desc_cb->priv) == 1)) { | |
1847 | /* Flip page offset to other buffer */ | |
1848 | desc_cb->page_offset ^= truesize; | |
1849 | ||
1850 | desc_cb->reuse_flag = 1; | |
1851 | /* bump ref count on page before it is given*/ | |
1852 | get_page(desc_cb->priv); | |
1853 | } | |
1854 | return; | |
1855 | } | |
1856 | ||
1857 | /* Move offset up to the next cache line */ | |
1858 | desc_cb->page_offset += truesize; | |
1859 | ||
1860 | if (desc_cb->page_offset <= last_offset) { | |
1861 | desc_cb->reuse_flag = 1; | |
1862 | /* Bump ref count on page before it is given*/ | |
1863 | get_page(desc_cb->priv); | |
1864 | } | |
1865 | } | |
1866 | ||
1867 | static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb, | |
1868 | struct hns3_desc *desc) | |
1869 | { | |
1870 | struct net_device *netdev = ring->tqp->handle->kinfo.netdev; | |
1871 | int l3_type, l4_type; | |
1872 | u32 bd_base_info; | |
1873 | int ol4_type; | |
1874 | u32 l234info; | |
1875 | ||
1876 | bd_base_info = le32_to_cpu(desc->rx.bd_base_info); | |
1877 | l234info = le32_to_cpu(desc->rx.l234_info); | |
1878 | ||
1879 | skb->ip_summed = CHECKSUM_NONE; | |
1880 | ||
1881 | skb_checksum_none_assert(skb); | |
1882 | ||
1883 | if (!(netdev->features & NETIF_F_RXCSUM)) | |
1884 | return; | |
1885 | ||
1886 | /* check if hardware has done checksum */ | |
1887 | if (!hnae_get_bit(bd_base_info, HNS3_RXD_L3L4P_B)) | |
1888 | return; | |
1889 | ||
1890 | if (unlikely(hnae_get_bit(l234info, HNS3_RXD_L3E_B) || | |
1891 | hnae_get_bit(l234info, HNS3_RXD_L4E_B) || | |
1892 | hnae_get_bit(l234info, HNS3_RXD_OL3E_B) || | |
1893 | hnae_get_bit(l234info, HNS3_RXD_OL4E_B))) { | |
1894 | netdev_err(netdev, "L3/L4 error pkt\n"); | |
1895 | u64_stats_update_begin(&ring->syncp); | |
1896 | ring->stats.l3l4_csum_err++; | |
1897 | u64_stats_update_end(&ring->syncp); | |
1898 | ||
1899 | return; | |
1900 | } | |
1901 | ||
1902 | l3_type = hnae_get_field(l234info, HNS3_RXD_L3ID_M, | |
1903 | HNS3_RXD_L3ID_S); | |
1904 | l4_type = hnae_get_field(l234info, HNS3_RXD_L4ID_M, | |
1905 | HNS3_RXD_L4ID_S); | |
1906 | ||
1907 | ol4_type = hnae_get_field(l234info, HNS3_RXD_OL4ID_M, HNS3_RXD_OL4ID_S); | |
1908 | switch (ol4_type) { | |
1909 | case HNS3_OL4_TYPE_MAC_IN_UDP: | |
1910 | case HNS3_OL4_TYPE_NVGRE: | |
1911 | skb->csum_level = 1; | |
1912 | case HNS3_OL4_TYPE_NO_TUN: | |
1913 | /* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */ | |
1914 | if (l3_type == HNS3_L3_TYPE_IPV4 || | |
1915 | (l3_type == HNS3_L3_TYPE_IPV6 && | |
1916 | (l4_type == HNS3_L4_TYPE_UDP || | |
1917 | l4_type == HNS3_L4_TYPE_TCP || | |
1918 | l4_type == HNS3_L4_TYPE_SCTP))) | |
1919 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1920 | break; | |
1921 | } | |
1922 | } | |
1923 | ||
1924 | static int hns3_handle_rx_bd(struct hns3_enet_ring *ring, | |
1925 | struct sk_buff **out_skb, int *out_bnum) | |
1926 | { | |
1927 | struct net_device *netdev = ring->tqp->handle->kinfo.netdev; | |
1928 | struct hns3_desc_cb *desc_cb; | |
1929 | struct hns3_desc *desc; | |
1930 | struct sk_buff *skb; | |
1931 | unsigned char *va; | |
1932 | u32 bd_base_info; | |
1933 | int pull_len; | |
1934 | u32 l234info; | |
1935 | int length; | |
1936 | int bnum; | |
1937 | ||
1938 | desc = &ring->desc[ring->next_to_clean]; | |
1939 | desc_cb = &ring->desc_cb[ring->next_to_clean]; | |
1940 | ||
1941 | prefetch(desc); | |
1942 | ||
1943 | length = le16_to_cpu(desc->rx.pkt_len); | |
1944 | bd_base_info = le32_to_cpu(desc->rx.bd_base_info); | |
1945 | l234info = le32_to_cpu(desc->rx.l234_info); | |
1946 | ||
1947 | /* Check valid BD */ | |
1948 | if (!hnae_get_bit(bd_base_info, HNS3_RXD_VLD_B)) | |
1949 | return -EFAULT; | |
1950 | ||
1951 | va = (unsigned char *)desc_cb->buf + desc_cb->page_offset; | |
1952 | ||
1953 | /* Prefetch first cache line of first page | |
1954 | * Idea is to cache few bytes of the header of the packet. Our L1 Cache | |
1955 | * line size is 64B so need to prefetch twice to make it 128B. But in | |
1956 | * actual we can have greater size of caches with 128B Level 1 cache | |
1957 | * lines. In such a case, single fetch would suffice to cache in the | |
1958 | * relevant part of the header. | |
1959 | */ | |
1960 | prefetch(va); | |
1961 | #if L1_CACHE_BYTES < 128 | |
1962 | prefetch(va + L1_CACHE_BYTES); | |
1963 | #endif | |
1964 | ||
1965 | skb = *out_skb = napi_alloc_skb(&ring->tqp_vector->napi, | |
1966 | HNS3_RX_HEAD_SIZE); | |
1967 | if (unlikely(!skb)) { | |
1968 | netdev_err(netdev, "alloc rx skb fail\n"); | |
1969 | ||
1970 | u64_stats_update_begin(&ring->syncp); | |
1971 | ring->stats.sw_err_cnt++; | |
1972 | u64_stats_update_end(&ring->syncp); | |
1973 | ||
1974 | return -ENOMEM; | |
1975 | } | |
1976 | ||
1977 | prefetchw(skb->data); | |
1978 | ||
1979 | bnum = 1; | |
1980 | if (length <= HNS3_RX_HEAD_SIZE) { | |
1981 | memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long))); | |
1982 | ||
1983 | /* We can reuse buffer as-is, just make sure it is local */ | |
1984 | if (likely(page_to_nid(desc_cb->priv) == numa_node_id())) | |
1985 | desc_cb->reuse_flag = 1; | |
1986 | else /* This page cannot be reused so discard it */ | |
1987 | put_page(desc_cb->priv); | |
1988 | ||
1989 | ring_ptr_move_fw(ring, next_to_clean); | |
1990 | } else { | |
1991 | u64_stats_update_begin(&ring->syncp); | |
1992 | ring->stats.seg_pkt_cnt++; | |
1993 | u64_stats_update_end(&ring->syncp); | |
1994 | ||
1995 | pull_len = hns3_nic_get_headlen(va, l234info, | |
1996 | HNS3_RX_HEAD_SIZE); | |
1997 | memcpy(__skb_put(skb, pull_len), va, | |
1998 | ALIGN(pull_len, sizeof(long))); | |
1999 | ||
2000 | hns3_nic_reuse_page(skb, 0, ring, pull_len, desc_cb); | |
2001 | ring_ptr_move_fw(ring, next_to_clean); | |
2002 | ||
2003 | while (!hnae_get_bit(bd_base_info, HNS3_RXD_FE_B)) { | |
2004 | desc = &ring->desc[ring->next_to_clean]; | |
2005 | desc_cb = &ring->desc_cb[ring->next_to_clean]; | |
2006 | bd_base_info = le32_to_cpu(desc->rx.bd_base_info); | |
2007 | hns3_nic_reuse_page(skb, bnum, ring, 0, desc_cb); | |
2008 | ring_ptr_move_fw(ring, next_to_clean); | |
2009 | bnum++; | |
2010 | } | |
2011 | } | |
2012 | ||
2013 | *out_bnum = bnum; | |
2014 | ||
2015 | if (unlikely(!hnae_get_bit(bd_base_info, HNS3_RXD_VLD_B))) { | |
2016 | netdev_err(netdev, "no valid bd,%016llx,%016llx\n", | |
2017 | ((u64 *)desc)[0], ((u64 *)desc)[1]); | |
2018 | u64_stats_update_begin(&ring->syncp); | |
2019 | ring->stats.non_vld_descs++; | |
2020 | u64_stats_update_end(&ring->syncp); | |
2021 | ||
2022 | dev_kfree_skb_any(skb); | |
2023 | return -EINVAL; | |
2024 | } | |
2025 | ||
2026 | if (unlikely((!desc->rx.pkt_len) || | |
2027 | hnae_get_bit(l234info, HNS3_RXD_TRUNCAT_B))) { | |
2028 | netdev_err(netdev, "truncated pkt\n"); | |
2029 | u64_stats_update_begin(&ring->syncp); | |
2030 | ring->stats.err_pkt_len++; | |
2031 | u64_stats_update_end(&ring->syncp); | |
2032 | ||
2033 | dev_kfree_skb_any(skb); | |
2034 | return -EFAULT; | |
2035 | } | |
2036 | ||
2037 | if (unlikely(hnae_get_bit(l234info, HNS3_RXD_L2E_B))) { | |
2038 | netdev_err(netdev, "L2 error pkt\n"); | |
2039 | u64_stats_update_begin(&ring->syncp); | |
2040 | ring->stats.l2_err++; | |
2041 | u64_stats_update_end(&ring->syncp); | |
2042 | ||
2043 | dev_kfree_skb_any(skb); | |
2044 | return -EFAULT; | |
2045 | } | |
2046 | ||
2047 | u64_stats_update_begin(&ring->syncp); | |
2048 | ring->stats.rx_pkts++; | |
2049 | ring->stats.rx_bytes += skb->len; | |
2050 | u64_stats_update_end(&ring->syncp); | |
2051 | ||
2052 | ring->tqp_vector->rx_group.total_bytes += skb->len; | |
2053 | ||
2054 | hns3_rx_checksum(ring, skb, desc); | |
2055 | return 0; | |
2056 | } | |
2057 | ||
2058 | static int hns3_clean_rx_ring(struct hns3_enet_ring *ring, int budget) | |
2059 | { | |
2060 | #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16 | |
2061 | struct net_device *netdev = ring->tqp->handle->kinfo.netdev; | |
2062 | int recv_pkts, recv_bds, clean_count, err; | |
2063 | int unused_count = hns3_desc_unused(ring); | |
2064 | struct sk_buff *skb = NULL; | |
2065 | int num, bnum = 0; | |
2066 | ||
2067 | num = readl_relaxed(ring->tqp->io_base + HNS3_RING_RX_RING_FBDNUM_REG); | |
2068 | rmb(); /* Make sure num taken effect before the other data is touched */ | |
2069 | ||
2070 | recv_pkts = 0, recv_bds = 0, clean_count = 0; | |
2071 | num -= unused_count; | |
2072 | ||
2073 | while (recv_pkts < budget && recv_bds < num) { | |
2074 | /* Reuse or realloc buffers */ | |
2075 | if (clean_count + unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) { | |
2076 | hns3_nic_alloc_rx_buffers(ring, | |
2077 | clean_count + unused_count); | |
2078 | clean_count = 0; | |
2079 | unused_count = hns3_desc_unused(ring); | |
2080 | } | |
2081 | ||
2082 | /* Poll one pkt */ | |
2083 | err = hns3_handle_rx_bd(ring, &skb, &bnum); | |
2084 | if (unlikely(!skb)) /* This fault cannot be repaired */ | |
2085 | goto out; | |
2086 | ||
2087 | recv_bds += bnum; | |
2088 | clean_count += bnum; | |
2089 | if (unlikely(err)) { /* Do jump the err */ | |
2090 | recv_pkts++; | |
2091 | continue; | |
2092 | } | |
2093 | ||
2094 | /* Do update ip stack process */ | |
2095 | skb->protocol = eth_type_trans(skb, netdev); | |
2096 | (void)napi_gro_receive(&ring->tqp_vector->napi, skb); | |
2097 | ||
2098 | recv_pkts++; | |
2099 | } | |
2100 | ||
2101 | out: | |
2102 | /* Make all data has been write before submit */ | |
2103 | if (clean_count + unused_count > 0) | |
2104 | hns3_nic_alloc_rx_buffers(ring, | |
2105 | clean_count + unused_count); | |
2106 | ||
2107 | return recv_pkts; | |
2108 | } | |
2109 | ||
2110 | static bool hns3_get_new_int_gl(struct hns3_enet_ring_group *ring_group) | |
2111 | { | |
2112 | #define HNS3_RX_ULTRA_PACKET_RATE 40000 | |
2113 | enum hns3_flow_level_range new_flow_level; | |
2114 | struct hns3_enet_tqp_vector *tqp_vector; | |
2115 | int packets_per_secs; | |
2116 | int bytes_per_usecs; | |
2117 | u16 new_int_gl; | |
2118 | int usecs; | |
2119 | ||
2120 | if (!ring_group->int_gl) | |
2121 | return false; | |
2122 | ||
2123 | if (ring_group->total_packets == 0) { | |
2124 | ring_group->int_gl = HNS3_INT_GL_50K; | |
2125 | ring_group->flow_level = HNS3_FLOW_LOW; | |
2126 | return true; | |
2127 | } | |
2128 | ||
2129 | /* Simple throttlerate management | |
2130 | * 0-10MB/s lower (50000 ints/s) | |
2131 | * 10-20MB/s middle (20000 ints/s) | |
2132 | * 20-1249MB/s high (18000 ints/s) | |
2133 | * > 40000pps ultra (8000 ints/s) | |
2134 | */ | |
2135 | new_flow_level = ring_group->flow_level; | |
2136 | new_int_gl = ring_group->int_gl; | |
2137 | tqp_vector = ring_group->ring->tqp_vector; | |
2138 | usecs = (ring_group->int_gl << 1); | |
2139 | bytes_per_usecs = ring_group->total_bytes / usecs; | |
2140 | /* 1000000 microseconds */ | |
2141 | packets_per_secs = ring_group->total_packets * 1000000 / usecs; | |
2142 | ||
2143 | switch (new_flow_level) { | |
2144 | case HNS3_FLOW_LOW: | |
2145 | if (bytes_per_usecs > 10) | |
2146 | new_flow_level = HNS3_FLOW_MID; | |
2147 | break; | |
2148 | case HNS3_FLOW_MID: | |
2149 | if (bytes_per_usecs > 20) | |
2150 | new_flow_level = HNS3_FLOW_HIGH; | |
2151 | else if (bytes_per_usecs <= 10) | |
2152 | new_flow_level = HNS3_FLOW_LOW; | |
2153 | break; | |
2154 | case HNS3_FLOW_HIGH: | |
2155 | case HNS3_FLOW_ULTRA: | |
2156 | default: | |
2157 | if (bytes_per_usecs <= 20) | |
2158 | new_flow_level = HNS3_FLOW_MID; | |
2159 | break; | |
2160 | } | |
2161 | ||
2162 | if ((packets_per_secs > HNS3_RX_ULTRA_PACKET_RATE) && | |
2163 | (&tqp_vector->rx_group == ring_group)) | |
2164 | new_flow_level = HNS3_FLOW_ULTRA; | |
2165 | ||
2166 | switch (new_flow_level) { | |
2167 | case HNS3_FLOW_LOW: | |
2168 | new_int_gl = HNS3_INT_GL_50K; | |
2169 | break; | |
2170 | case HNS3_FLOW_MID: | |
2171 | new_int_gl = HNS3_INT_GL_20K; | |
2172 | break; | |
2173 | case HNS3_FLOW_HIGH: | |
2174 | new_int_gl = HNS3_INT_GL_18K; | |
2175 | break; | |
2176 | case HNS3_FLOW_ULTRA: | |
2177 | new_int_gl = HNS3_INT_GL_8K; | |
2178 | break; | |
2179 | default: | |
2180 | break; | |
2181 | } | |
2182 | ||
2183 | ring_group->total_bytes = 0; | |
2184 | ring_group->total_packets = 0; | |
2185 | ring_group->flow_level = new_flow_level; | |
2186 | if (new_int_gl != ring_group->int_gl) { | |
2187 | ring_group->int_gl = new_int_gl; | |
2188 | return true; | |
2189 | } | |
2190 | return false; | |
2191 | } | |
2192 | ||
2193 | static void hns3_update_new_int_gl(struct hns3_enet_tqp_vector *tqp_vector) | |
2194 | { | |
2195 | u16 rx_int_gl, tx_int_gl; | |
2196 | bool rx, tx; | |
2197 | ||
2198 | rx = hns3_get_new_int_gl(&tqp_vector->rx_group); | |
2199 | tx = hns3_get_new_int_gl(&tqp_vector->tx_group); | |
2200 | rx_int_gl = tqp_vector->rx_group.int_gl; | |
2201 | tx_int_gl = tqp_vector->tx_group.int_gl; | |
2202 | if (rx && tx) { | |
2203 | if (rx_int_gl > tx_int_gl) { | |
2204 | tqp_vector->tx_group.int_gl = rx_int_gl; | |
2205 | tqp_vector->tx_group.flow_level = | |
2206 | tqp_vector->rx_group.flow_level; | |
2207 | hns3_set_vector_coalesc_gl(tqp_vector, rx_int_gl); | |
2208 | } else { | |
2209 | tqp_vector->rx_group.int_gl = tx_int_gl; | |
2210 | tqp_vector->rx_group.flow_level = | |
2211 | tqp_vector->tx_group.flow_level; | |
2212 | hns3_set_vector_coalesc_gl(tqp_vector, tx_int_gl); | |
2213 | } | |
2214 | } | |
2215 | } | |
2216 | ||
2217 | static int hns3_nic_common_poll(struct napi_struct *napi, int budget) | |
2218 | { | |
2219 | struct hns3_enet_ring *ring; | |
2220 | int rx_pkt_total = 0; | |
2221 | ||
2222 | struct hns3_enet_tqp_vector *tqp_vector = | |
2223 | container_of(napi, struct hns3_enet_tqp_vector, napi); | |
2224 | bool clean_complete = true; | |
2225 | int rx_budget; | |
2226 | ||
2227 | /* Since the actual Tx work is minimal, we can give the Tx a larger | |
2228 | * budget and be more aggressive about cleaning up the Tx descriptors. | |
2229 | */ | |
2230 | hns3_for_each_ring(ring, tqp_vector->tx_group) { | |
2231 | if (!hns3_clean_tx_ring(ring, budget)) | |
2232 | clean_complete = false; | |
2233 | } | |
2234 | ||
2235 | /* make sure rx ring budget not smaller than 1 */ | |
2236 | rx_budget = max(budget / tqp_vector->num_tqps, 1); | |
2237 | ||
2238 | hns3_for_each_ring(ring, tqp_vector->rx_group) { | |
2239 | int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget); | |
2240 | ||
2241 | if (rx_cleaned >= rx_budget) | |
2242 | clean_complete = false; | |
2243 | ||
2244 | rx_pkt_total += rx_cleaned; | |
2245 | } | |
2246 | ||
2247 | tqp_vector->rx_group.total_packets += rx_pkt_total; | |
2248 | ||
2249 | if (!clean_complete) | |
2250 | return budget; | |
2251 | ||
2252 | napi_complete(napi); | |
2253 | hns3_update_new_int_gl(tqp_vector); | |
2254 | hns3_mask_vector_irq(tqp_vector, 1); | |
2255 | ||
2256 | return rx_pkt_total; | |
2257 | } | |
2258 | ||
2259 | static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector, | |
2260 | struct hnae3_ring_chain_node *head) | |
2261 | { | |
2262 | struct pci_dev *pdev = tqp_vector->handle->pdev; | |
2263 | struct hnae3_ring_chain_node *cur_chain = head; | |
2264 | struct hnae3_ring_chain_node *chain; | |
2265 | struct hns3_enet_ring *tx_ring; | |
2266 | struct hns3_enet_ring *rx_ring; | |
2267 | ||
2268 | tx_ring = tqp_vector->tx_group.ring; | |
2269 | if (tx_ring) { | |
2270 | cur_chain->tqp_index = tx_ring->tqp->tqp_index; | |
2271 | hnae_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B, | |
2272 | HNAE3_RING_TYPE_TX); | |
2273 | ||
2274 | cur_chain->next = NULL; | |
2275 | ||
2276 | while (tx_ring->next) { | |
2277 | tx_ring = tx_ring->next; | |
2278 | ||
2279 | chain = devm_kzalloc(&pdev->dev, sizeof(*chain), | |
2280 | GFP_KERNEL); | |
2281 | if (!chain) | |
2282 | return -ENOMEM; | |
2283 | ||
2284 | cur_chain->next = chain; | |
2285 | chain->tqp_index = tx_ring->tqp->tqp_index; | |
2286 | hnae_set_bit(chain->flag, HNAE3_RING_TYPE_B, | |
2287 | HNAE3_RING_TYPE_TX); | |
2288 | ||
2289 | cur_chain = chain; | |
2290 | } | |
2291 | } | |
2292 | ||
2293 | rx_ring = tqp_vector->rx_group.ring; | |
2294 | if (!tx_ring && rx_ring) { | |
2295 | cur_chain->next = NULL; | |
2296 | cur_chain->tqp_index = rx_ring->tqp->tqp_index; | |
2297 | hnae_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B, | |
2298 | HNAE3_RING_TYPE_RX); | |
2299 | ||
2300 | rx_ring = rx_ring->next; | |
2301 | } | |
2302 | ||
2303 | while (rx_ring) { | |
2304 | chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL); | |
2305 | if (!chain) | |
2306 | return -ENOMEM; | |
2307 | ||
2308 | cur_chain->next = chain; | |
2309 | chain->tqp_index = rx_ring->tqp->tqp_index; | |
2310 | hnae_set_bit(chain->flag, HNAE3_RING_TYPE_B, | |
2311 | HNAE3_RING_TYPE_RX); | |
2312 | cur_chain = chain; | |
2313 | ||
2314 | rx_ring = rx_ring->next; | |
2315 | } | |
2316 | ||
2317 | return 0; | |
2318 | } | |
2319 | ||
2320 | static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector, | |
2321 | struct hnae3_ring_chain_node *head) | |
2322 | { | |
2323 | struct pci_dev *pdev = tqp_vector->handle->pdev; | |
2324 | struct hnae3_ring_chain_node *chain_tmp, *chain; | |
2325 | ||
2326 | chain = head->next; | |
2327 | ||
2328 | while (chain) { | |
2329 | chain_tmp = chain->next; | |
2330 | devm_kfree(&pdev->dev, chain); | |
2331 | chain = chain_tmp; | |
2332 | } | |
2333 | } | |
2334 | ||
2335 | static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group, | |
2336 | struct hns3_enet_ring *ring) | |
2337 | { | |
2338 | ring->next = group->ring; | |
2339 | group->ring = ring; | |
2340 | ||
2341 | group->count++; | |
2342 | } | |
2343 | ||
2344 | static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv) | |
2345 | { | |
2346 | struct hnae3_ring_chain_node vector_ring_chain; | |
2347 | struct hnae3_handle *h = priv->ae_handle; | |
2348 | struct hns3_enet_tqp_vector *tqp_vector; | |
2349 | struct hnae3_vector_info *vector; | |
2350 | struct pci_dev *pdev = h->pdev; | |
2351 | u16 tqp_num = h->kinfo.num_tqps; | |
2352 | u16 vector_num; | |
2353 | int ret = 0; | |
2354 | u16 i; | |
2355 | ||
2356 | /* RSS size, cpu online and vector_num should be the same */ | |
2357 | /* Should consider 2p/4p later */ | |
2358 | vector_num = min_t(u16, num_online_cpus(), tqp_num); | |
2359 | vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector), | |
2360 | GFP_KERNEL); | |
2361 | if (!vector) | |
2362 | return -ENOMEM; | |
2363 | ||
2364 | vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector); | |
2365 | ||
2366 | priv->vector_num = vector_num; | |
2367 | priv->tqp_vector = (struct hns3_enet_tqp_vector *) | |
2368 | devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector), | |
2369 | GFP_KERNEL); | |
2370 | if (!priv->tqp_vector) | |
2371 | return -ENOMEM; | |
2372 | ||
2373 | for (i = 0; i < tqp_num; i++) { | |
2374 | u16 vector_i = i % vector_num; | |
2375 | ||
2376 | tqp_vector = &priv->tqp_vector[vector_i]; | |
2377 | ||
2378 | hns3_add_ring_to_group(&tqp_vector->tx_group, | |
2379 | priv->ring_data[i].ring); | |
2380 | ||
2381 | hns3_add_ring_to_group(&tqp_vector->rx_group, | |
2382 | priv->ring_data[i + tqp_num].ring); | |
2383 | ||
2384 | tqp_vector->idx = vector_i; | |
2385 | tqp_vector->mask_addr = vector[vector_i].io_addr; | |
2386 | tqp_vector->vector_irq = vector[vector_i].vector; | |
2387 | tqp_vector->num_tqps++; | |
2388 | ||
2389 | priv->ring_data[i].ring->tqp_vector = tqp_vector; | |
2390 | priv->ring_data[i + tqp_num].ring->tqp_vector = tqp_vector; | |
2391 | } | |
2392 | ||
2393 | for (i = 0; i < vector_num; i++) { | |
2394 | tqp_vector = &priv->tqp_vector[i]; | |
2395 | ||
2396 | tqp_vector->rx_group.total_bytes = 0; | |
2397 | tqp_vector->rx_group.total_packets = 0; | |
2398 | tqp_vector->tx_group.total_bytes = 0; | |
2399 | tqp_vector->tx_group.total_packets = 0; | |
2400 | hns3_vector_gl_rl_init(tqp_vector); | |
2401 | tqp_vector->handle = h; | |
2402 | ||
2403 | ret = hns3_get_vector_ring_chain(tqp_vector, | |
2404 | &vector_ring_chain); | |
2405 | if (ret) | |
2406 | goto out; | |
2407 | ||
2408 | ret = h->ae_algo->ops->map_ring_to_vector(h, | |
2409 | tqp_vector->vector_irq, &vector_ring_chain); | |
2410 | if (ret) | |
2411 | goto out; | |
2412 | ||
2413 | hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain); | |
2414 | ||
2415 | netif_napi_add(priv->netdev, &tqp_vector->napi, | |
2416 | hns3_nic_common_poll, NAPI_POLL_WEIGHT); | |
2417 | } | |
2418 | ||
2419 | out: | |
2420 | devm_kfree(&pdev->dev, vector); | |
2421 | return ret; | |
2422 | } | |
2423 | ||
2424 | static int hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv) | |
2425 | { | |
2426 | struct hnae3_ring_chain_node vector_ring_chain; | |
2427 | struct hnae3_handle *h = priv->ae_handle; | |
2428 | struct hns3_enet_tqp_vector *tqp_vector; | |
2429 | struct pci_dev *pdev = h->pdev; | |
2430 | int i, ret; | |
2431 | ||
2432 | for (i = 0; i < priv->vector_num; i++) { | |
2433 | tqp_vector = &priv->tqp_vector[i]; | |
2434 | ||
2435 | ret = hns3_get_vector_ring_chain(tqp_vector, | |
2436 | &vector_ring_chain); | |
2437 | if (ret) | |
2438 | return ret; | |
2439 | ||
2440 | ret = h->ae_algo->ops->unmap_ring_from_vector(h, | |
2441 | tqp_vector->vector_irq, &vector_ring_chain); | |
2442 | if (ret) | |
2443 | return ret; | |
2444 | ||
2445 | hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain); | |
2446 | ||
2447 | if (priv->tqp_vector[i].irq_init_flag == HNS3_VECTOR_INITED) { | |
2448 | (void)irq_set_affinity_hint( | |
2449 | priv->tqp_vector[i].vector_irq, | |
2450 | NULL); | |
2451 | devm_free_irq(&pdev->dev, | |
2452 | priv->tqp_vector[i].vector_irq, | |
2453 | &priv->tqp_vector[i]); | |
2454 | } | |
2455 | ||
2456 | priv->ring_data[i].ring->irq_init_flag = HNS3_VECTOR_NOT_INITED; | |
2457 | ||
2458 | netif_napi_del(&priv->tqp_vector[i].napi); | |
2459 | } | |
2460 | ||
2461 | devm_kfree(&pdev->dev, priv->tqp_vector); | |
2462 | ||
2463 | return 0; | |
2464 | } | |
2465 | ||
2466 | static int hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv, | |
2467 | int ring_type) | |
2468 | { | |
2469 | struct hns3_nic_ring_data *ring_data = priv->ring_data; | |
2470 | int queue_num = priv->ae_handle->kinfo.num_tqps; | |
2471 | struct pci_dev *pdev = priv->ae_handle->pdev; | |
2472 | struct hns3_enet_ring *ring; | |
2473 | ||
2474 | ring = devm_kzalloc(&pdev->dev, sizeof(*ring), GFP_KERNEL); | |
2475 | if (!ring) | |
2476 | return -ENOMEM; | |
2477 | ||
2478 | if (ring_type == HNAE3_RING_TYPE_TX) { | |
2479 | ring_data[q->tqp_index].ring = ring; | |
2480 | ring->io_base = (u8 __iomem *)q->io_base + HNS3_TX_REG_OFFSET; | |
2481 | } else { | |
2482 | ring_data[q->tqp_index + queue_num].ring = ring; | |
2483 | ring->io_base = q->io_base; | |
2484 | } | |
2485 | ||
2486 | hnae_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type); | |
2487 | ||
2488 | ring_data[q->tqp_index].queue_index = q->tqp_index; | |
2489 | ||
2490 | ring->tqp = q; | |
2491 | ring->desc = NULL; | |
2492 | ring->desc_cb = NULL; | |
2493 | ring->dev = priv->dev; | |
2494 | ring->desc_dma_addr = 0; | |
2495 | ring->buf_size = q->buf_size; | |
2496 | ring->desc_num = q->desc_num; | |
2497 | ring->next_to_use = 0; | |
2498 | ring->next_to_clean = 0; | |
2499 | ||
2500 | return 0; | |
2501 | } | |
2502 | ||
2503 | static int hns3_queue_to_ring(struct hnae3_queue *tqp, | |
2504 | struct hns3_nic_priv *priv) | |
2505 | { | |
2506 | int ret; | |
2507 | ||
2508 | ret = hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX); | |
2509 | if (ret) | |
2510 | return ret; | |
2511 | ||
2512 | ret = hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX); | |
2513 | if (ret) | |
2514 | return ret; | |
2515 | ||
2516 | return 0; | |
2517 | } | |
2518 | ||
2519 | static int hns3_get_ring_config(struct hns3_nic_priv *priv) | |
2520 | { | |
2521 | struct hnae3_handle *h = priv->ae_handle; | |
2522 | struct pci_dev *pdev = h->pdev; | |
2523 | int i, ret; | |
2524 | ||
2525 | priv->ring_data = devm_kzalloc(&pdev->dev, h->kinfo.num_tqps * | |
2526 | sizeof(*priv->ring_data) * 2, | |
2527 | GFP_KERNEL); | |
2528 | if (!priv->ring_data) | |
2529 | return -ENOMEM; | |
2530 | ||
2531 | for (i = 0; i < h->kinfo.num_tqps; i++) { | |
2532 | ret = hns3_queue_to_ring(h->kinfo.tqp[i], priv); | |
2533 | if (ret) | |
2534 | goto err; | |
2535 | } | |
2536 | ||
2537 | return 0; | |
2538 | err: | |
2539 | devm_kfree(&pdev->dev, priv->ring_data); | |
2540 | return ret; | |
2541 | } | |
2542 | ||
2543 | static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring) | |
2544 | { | |
2545 | int ret; | |
2546 | ||
2547 | if (ring->desc_num <= 0 || ring->buf_size <= 0) | |
2548 | return -EINVAL; | |
2549 | ||
2550 | ring->desc_cb = kcalloc(ring->desc_num, sizeof(ring->desc_cb[0]), | |
2551 | GFP_KERNEL); | |
2552 | if (!ring->desc_cb) { | |
2553 | ret = -ENOMEM; | |
2554 | goto out; | |
2555 | } | |
2556 | ||
2557 | ret = hns3_alloc_desc(ring); | |
2558 | if (ret) | |
2559 | goto out_with_desc_cb; | |
2560 | ||
2561 | if (!HNAE3_IS_TX_RING(ring)) { | |
2562 | ret = hns3_alloc_ring_buffers(ring); | |
2563 | if (ret) | |
2564 | goto out_with_desc; | |
2565 | } | |
2566 | ||
2567 | return 0; | |
2568 | ||
2569 | out_with_desc: | |
2570 | hns3_free_desc(ring); | |
2571 | out_with_desc_cb: | |
2572 | kfree(ring->desc_cb); | |
2573 | ring->desc_cb = NULL; | |
2574 | out: | |
2575 | return ret; | |
2576 | } | |
2577 | ||
2578 | static void hns3_fini_ring(struct hns3_enet_ring *ring) | |
2579 | { | |
2580 | hns3_free_desc(ring); | |
2581 | kfree(ring->desc_cb); | |
2582 | ring->desc_cb = NULL; | |
2583 | ring->next_to_clean = 0; | |
2584 | ring->next_to_use = 0; | |
2585 | } | |
2586 | ||
1db9b1bf | 2587 | static int hns3_buf_size2type(u32 buf_size) |
76ad4f0e S |
2588 | { |
2589 | int bd_size_type; | |
2590 | ||
2591 | switch (buf_size) { | |
2592 | case 512: | |
2593 | bd_size_type = HNS3_BD_SIZE_512_TYPE; | |
2594 | break; | |
2595 | case 1024: | |
2596 | bd_size_type = HNS3_BD_SIZE_1024_TYPE; | |
2597 | break; | |
2598 | case 2048: | |
2599 | bd_size_type = HNS3_BD_SIZE_2048_TYPE; | |
2600 | break; | |
2601 | case 4096: | |
2602 | bd_size_type = HNS3_BD_SIZE_4096_TYPE; | |
2603 | break; | |
2604 | default: | |
2605 | bd_size_type = HNS3_BD_SIZE_2048_TYPE; | |
2606 | } | |
2607 | ||
2608 | return bd_size_type; | |
2609 | } | |
2610 | ||
2611 | static void hns3_init_ring_hw(struct hns3_enet_ring *ring) | |
2612 | { | |
2613 | dma_addr_t dma = ring->desc_dma_addr; | |
2614 | struct hnae3_queue *q = ring->tqp; | |
2615 | ||
2616 | if (!HNAE3_IS_TX_RING(ring)) { | |
2617 | hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG, | |
2618 | (u32)dma); | |
2619 | hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG, | |
2620 | (u32)((dma >> 31) >> 1)); | |
2621 | ||
2622 | hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG, | |
2623 | hns3_buf_size2type(ring->buf_size)); | |
2624 | hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG, | |
2625 | ring->desc_num / 8 - 1); | |
2626 | ||
2627 | } else { | |
2628 | hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG, | |
2629 | (u32)dma); | |
2630 | hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG, | |
2631 | (u32)((dma >> 31) >> 1)); | |
2632 | ||
2633 | hns3_write_dev(q, HNS3_RING_TX_RING_BD_LEN_REG, | |
2634 | hns3_buf_size2type(ring->buf_size)); | |
2635 | hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG, | |
2636 | ring->desc_num / 8 - 1); | |
2637 | } | |
2638 | } | |
2639 | ||
2640 | static int hns3_init_all_ring(struct hns3_nic_priv *priv) | |
2641 | { | |
2642 | struct hnae3_handle *h = priv->ae_handle; | |
2643 | int ring_num = h->kinfo.num_tqps * 2; | |
2644 | int i, j; | |
2645 | int ret; | |
2646 | ||
2647 | for (i = 0; i < ring_num; i++) { | |
2648 | ret = hns3_alloc_ring_memory(priv->ring_data[i].ring); | |
2649 | if (ret) { | |
2650 | dev_err(priv->dev, | |
2651 | "Alloc ring memory fail! ret=%d\n", ret); | |
2652 | goto out_when_alloc_ring_memory; | |
2653 | } | |
2654 | ||
2655 | hns3_init_ring_hw(priv->ring_data[i].ring); | |
2656 | ||
2657 | u64_stats_init(&priv->ring_data[i].ring->syncp); | |
2658 | } | |
2659 | ||
2660 | return 0; | |
2661 | ||
2662 | out_when_alloc_ring_memory: | |
2663 | for (j = i - 1; j >= 0; j--) | |
2664 | hns3_fini_ring(priv->ring_data[i].ring); | |
2665 | ||
2666 | return -ENOMEM; | |
2667 | } | |
2668 | ||
2669 | static int hns3_uninit_all_ring(struct hns3_nic_priv *priv) | |
2670 | { | |
2671 | struct hnae3_handle *h = priv->ae_handle; | |
2672 | int i; | |
2673 | ||
2674 | for (i = 0; i < h->kinfo.num_tqps; i++) { | |
2675 | if (h->ae_algo->ops->reset_queue) | |
2676 | h->ae_algo->ops->reset_queue(h, i); | |
2677 | ||
2678 | hns3_fini_ring(priv->ring_data[i].ring); | |
2679 | hns3_fini_ring(priv->ring_data[i + h->kinfo.num_tqps].ring); | |
2680 | } | |
2681 | ||
2682 | return 0; | |
2683 | } | |
2684 | ||
2685 | /* Set mac addr if it is configured. or leave it to the AE driver */ | |
2686 | static void hns3_init_mac_addr(struct net_device *netdev) | |
2687 | { | |
2688 | struct hns3_nic_priv *priv = netdev_priv(netdev); | |
2689 | struct hnae3_handle *h = priv->ae_handle; | |
2690 | u8 mac_addr_temp[ETH_ALEN]; | |
2691 | ||
2692 | if (h->ae_algo->ops->get_mac_addr) { | |
2693 | h->ae_algo->ops->get_mac_addr(h, mac_addr_temp); | |
2694 | ether_addr_copy(netdev->dev_addr, mac_addr_temp); | |
2695 | } | |
2696 | ||
2697 | /* Check if the MAC address is valid, if not get a random one */ | |
2698 | if (!is_valid_ether_addr(netdev->dev_addr)) { | |
2699 | eth_hw_addr_random(netdev); | |
2700 | dev_warn(priv->dev, "using random MAC address %pM\n", | |
2701 | netdev->dev_addr); | |
76ad4f0e | 2702 | } |
139e8792 L |
2703 | |
2704 | if (h->ae_algo->ops->set_mac_addr) | |
2705 | h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr); | |
2706 | ||
76ad4f0e S |
2707 | } |
2708 | ||
2709 | static void hns3_nic_set_priv_ops(struct net_device *netdev) | |
2710 | { | |
2711 | struct hns3_nic_priv *priv = netdev_priv(netdev); | |
2712 | ||
2713 | if ((netdev->features & NETIF_F_TSO) || | |
2714 | (netdev->features & NETIF_F_TSO6)) { | |
2715 | priv->ops.fill_desc = hns3_fill_desc_tso; | |
2716 | priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tso; | |
2717 | } else { | |
2718 | priv->ops.fill_desc = hns3_fill_desc; | |
2719 | priv->ops.maybe_stop_tx = hns3_nic_maybe_stop_tx; | |
2720 | } | |
2721 | } | |
2722 | ||
2723 | static int hns3_client_init(struct hnae3_handle *handle) | |
2724 | { | |
2725 | struct pci_dev *pdev = handle->pdev; | |
2726 | struct hns3_nic_priv *priv; | |
2727 | struct net_device *netdev; | |
2728 | int ret; | |
2729 | ||
2730 | netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), | |
2731 | handle->kinfo.num_tqps); | |
2732 | if (!netdev) | |
2733 | return -ENOMEM; | |
2734 | ||
2735 | priv = netdev_priv(netdev); | |
2736 | priv->dev = &pdev->dev; | |
2737 | priv->netdev = netdev; | |
2738 | priv->ae_handle = handle; | |
2739 | ||
2740 | handle->kinfo.netdev = netdev; | |
2741 | handle->priv = (void *)priv; | |
2742 | ||
2743 | hns3_init_mac_addr(netdev); | |
2744 | ||
2745 | hns3_set_default_feature(netdev); | |
2746 | ||
2747 | netdev->watchdog_timeo = HNS3_TX_TIMEOUT; | |
2748 | netdev->priv_flags |= IFF_UNICAST_FLT; | |
2749 | netdev->netdev_ops = &hns3_nic_netdev_ops; | |
2750 | SET_NETDEV_DEV(netdev, &pdev->dev); | |
2751 | hns3_ethtool_set_ops(netdev); | |
2752 | hns3_nic_set_priv_ops(netdev); | |
2753 | ||
2754 | /* Carrier off reporting is important to ethtool even BEFORE open */ | |
2755 | netif_carrier_off(netdev); | |
2756 | ||
2757 | ret = hns3_get_ring_config(priv); | |
2758 | if (ret) { | |
2759 | ret = -ENOMEM; | |
2760 | goto out_get_ring_cfg; | |
2761 | } | |
2762 | ||
2763 | ret = hns3_nic_init_vector_data(priv); | |
2764 | if (ret) { | |
2765 | ret = -ENOMEM; | |
2766 | goto out_init_vector_data; | |
2767 | } | |
2768 | ||
2769 | ret = hns3_init_all_ring(priv); | |
2770 | if (ret) { | |
2771 | ret = -ENOMEM; | |
2772 | goto out_init_ring_data; | |
2773 | } | |
2774 | ||
2775 | ret = register_netdev(netdev); | |
2776 | if (ret) { | |
2777 | dev_err(priv->dev, "probe register netdev fail!\n"); | |
2778 | goto out_reg_netdev_fail; | |
2779 | } | |
2780 | ||
986743db YL |
2781 | hns3_dcbnl_setup(handle); |
2782 | ||
a8e8b7ff S |
2783 | /* MTU range: (ETH_MIN_MTU(kernel default) - 9706) */ |
2784 | netdev->max_mtu = HNS3_MAX_MTU - (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN); | |
2785 | ||
76ad4f0e S |
2786 | return ret; |
2787 | ||
2788 | out_reg_netdev_fail: | |
2789 | out_init_ring_data: | |
2790 | (void)hns3_nic_uninit_vector_data(priv); | |
2791 | priv->ring_data = NULL; | |
2792 | out_init_vector_data: | |
2793 | out_get_ring_cfg: | |
2794 | priv->ae_handle = NULL; | |
2795 | free_netdev(netdev); | |
2796 | return ret; | |
2797 | } | |
2798 | ||
2799 | static void hns3_client_uninit(struct hnae3_handle *handle, bool reset) | |
2800 | { | |
2801 | struct net_device *netdev = handle->kinfo.netdev; | |
2802 | struct hns3_nic_priv *priv = netdev_priv(netdev); | |
2803 | int ret; | |
2804 | ||
2805 | if (netdev->reg_state != NETREG_UNINITIALIZED) | |
2806 | unregister_netdev(netdev); | |
2807 | ||
2808 | ret = hns3_nic_uninit_vector_data(priv); | |
2809 | if (ret) | |
2810 | netdev_err(netdev, "uninit vector error\n"); | |
2811 | ||
2812 | ret = hns3_uninit_all_ring(priv); | |
2813 | if (ret) | |
2814 | netdev_err(netdev, "uninit ring error\n"); | |
2815 | ||
2816 | priv->ring_data = NULL; | |
2817 | ||
2818 | free_netdev(netdev); | |
2819 | } | |
2820 | ||
2821 | static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup) | |
2822 | { | |
2823 | struct net_device *netdev = handle->kinfo.netdev; | |
2824 | ||
2825 | if (!netdev) | |
2826 | return; | |
2827 | ||
2828 | if (linkup) { | |
2829 | netif_carrier_on(netdev); | |
2830 | netif_tx_wake_all_queues(netdev); | |
2831 | netdev_info(netdev, "link up\n"); | |
2832 | } else { | |
2833 | netif_carrier_off(netdev); | |
2834 | netif_tx_stop_all_queues(netdev); | |
2835 | netdev_info(netdev, "link down\n"); | |
2836 | } | |
2837 | } | |
2838 | ||
9df8f79a YL |
2839 | static int hns3_client_setup_tc(struct hnae3_handle *handle, u8 tc) |
2840 | { | |
2841 | struct hnae3_knic_private_info *kinfo = &handle->kinfo; | |
2842 | struct net_device *ndev = kinfo->netdev; | |
075cfdd6 | 2843 | bool if_running; |
9df8f79a YL |
2844 | int ret; |
2845 | u8 i; | |
2846 | ||
2847 | if (tc > HNAE3_MAX_TC) | |
2848 | return -EINVAL; | |
2849 | ||
2850 | if (!ndev) | |
2851 | return -ENODEV; | |
2852 | ||
075cfdd6 CIK |
2853 | if_running = netif_running(ndev); |
2854 | ||
9df8f79a YL |
2855 | ret = netdev_set_num_tc(ndev, tc); |
2856 | if (ret) | |
2857 | return ret; | |
2858 | ||
2859 | if (if_running) { | |
2860 | (void)hns3_nic_net_stop(ndev); | |
2861 | msleep(100); | |
2862 | } | |
2863 | ||
2864 | ret = (kinfo->dcb_ops && kinfo->dcb_ops->map_update) ? | |
2865 | kinfo->dcb_ops->map_update(handle) : -EOPNOTSUPP; | |
2866 | if (ret) | |
2867 | goto err_out; | |
2868 | ||
2869 | if (tc <= 1) { | |
2870 | netdev_reset_tc(ndev); | |
2871 | goto out; | |
2872 | } | |
2873 | ||
2874 | for (i = 0; i < HNAE3_MAX_TC; i++) { | |
2875 | struct hnae3_tc_info *tc_info = &kinfo->tc_info[i]; | |
2876 | ||
2877 | if (tc_info->enable) | |
2878 | netdev_set_tc_queue(ndev, | |
2879 | tc_info->tc, | |
2880 | tc_info->tqp_count, | |
2881 | tc_info->tqp_offset); | |
2882 | } | |
2883 | ||
2884 | for (i = 0; i < HNAE3_MAX_USER_PRIO; i++) { | |
2885 | netdev_set_prio_tc_map(ndev, i, | |
2886 | kinfo->prio_tc[i]); | |
2887 | } | |
2888 | ||
2889 | out: | |
2890 | ret = hns3_nic_set_real_num_queue(ndev); | |
2891 | ||
2892 | err_out: | |
2893 | if (if_running) | |
2894 | (void)hns3_nic_net_open(ndev); | |
2895 | ||
2896 | return ret; | |
2897 | } | |
2898 | ||
1db9b1bf | 2899 | static const struct hnae3_client_ops client_ops = { |
76ad4f0e S |
2900 | .init_instance = hns3_client_init, |
2901 | .uninit_instance = hns3_client_uninit, | |
2902 | .link_status_change = hns3_link_status_change, | |
9df8f79a | 2903 | .setup_tc = hns3_client_setup_tc, |
76ad4f0e S |
2904 | }; |
2905 | ||
2906 | /* hns3_init_module - Driver registration routine | |
2907 | * hns3_init_module is the first routine called when the driver is | |
2908 | * loaded. All it does is register with the PCI subsystem. | |
2909 | */ | |
2910 | static int __init hns3_init_module(void) | |
2911 | { | |
2912 | int ret; | |
2913 | ||
2914 | pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string); | |
2915 | pr_info("%s: %s\n", hns3_driver_name, hns3_copyright); | |
2916 | ||
2917 | client.type = HNAE3_CLIENT_KNIC; | |
2918 | snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH - 1, "%s", | |
2919 | hns3_driver_name); | |
2920 | ||
2921 | client.ops = &client_ops; | |
2922 | ||
2923 | ret = hnae3_register_client(&client); | |
2924 | if (ret) | |
2925 | return ret; | |
2926 | ||
2927 | ret = pci_register_driver(&hns3_driver); | |
2928 | if (ret) | |
2929 | hnae3_unregister_client(&client); | |
2930 | ||
2931 | return ret; | |
2932 | } | |
2933 | module_init(hns3_init_module); | |
2934 | ||
2935 | /* hns3_exit_module - Driver exit cleanup routine | |
2936 | * hns3_exit_module is called just before the driver is removed | |
2937 | * from memory. | |
2938 | */ | |
2939 | static void __exit hns3_exit_module(void) | |
2940 | { | |
2941 | pci_unregister_driver(&hns3_driver); | |
2942 | hnae3_unregister_client(&client); | |
2943 | } | |
2944 | module_exit(hns3_exit_module); | |
2945 | ||
2946 | MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver"); | |
2947 | MODULE_AUTHOR("Huawei Tech. Co., Ltd."); | |
2948 | MODULE_LICENSE("GPL"); | |
2949 | MODULE_ALIAS("pci:hns-nic"); |