1 // SPDX-License-Identifier: GPL-2.0+
2 // Copyright (c) 2016-2017 Hisilicon Limited.
4 #include <linux/dma-mapping.h>
5 #include <linux/etherdevice.h>
6 #include <linux/interrupt.h>
7 #ifdef CONFIG_RFS_ACCEL
8 #include <linux/cpu_rmap.h>
10 #include <linux/if_vlan.h>
11 #include <linux/irq.h>
13 #include <linux/ipv6.h>
14 #include <linux/module.h>
15 #include <linux/pci.h>
16 #include <linux/aer.h>
17 #include <linux/skbuff.h>
18 #include <linux/sctp.h>
20 #include <net/ip6_checksum.h>
21 #include <net/pkt_cls.h>
23 #include <net/vxlan.h>
24 #include <net/geneve.h>
27 #include "hns3_enet.h"
28 /* All hns3 tracepoints are defined by the include below, which
29 * must be included exactly once across the whole kernel with
30 * CREATE_TRACE_POINTS defined
32 #define CREATE_TRACE_POINTS
33 #include "hns3_trace.h"
35 #define hns3_set_field(origin, shift, val) ((origin) |= (val) << (shift))
36 #define hns3_tx_bd_count(S) DIV_ROUND_UP(S, HNS3_MAX_BD_SIZE)
38 #define hns3_rl_err(fmt, ...) \
40 if (net_ratelimit()) \
41 netdev_err(fmt, ##__VA_ARGS__); \
44 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force);
46 static const char hns3_driver_name[] = "hns3";
47 static const char hns3_driver_string[] =
48 "Hisilicon Ethernet Network Driver for Hip08 Family";
49 static const char hns3_copyright[] = "Copyright (c) 2017 Huawei Corporation.";
50 static struct hnae3_client client;
52 static int debug = -1;
53 module_param(debug, int, 0);
54 MODULE_PARM_DESC(debug, " Network interface message level setting");
56 static unsigned int tx_spare_buf_size;
57 module_param(tx_spare_buf_size, uint, 0400);
58 MODULE_PARM_DESC(tx_spare_buf_size, "Size used to allocate tx spare buffer");
60 static unsigned int tx_sgl = 1;
61 module_param(tx_sgl, uint, 0600);
62 MODULE_PARM_DESC(tx_sgl, "Minimum number of frags when using dma_map_sg() to optimize the IOMMU mapping");
64 static bool page_pool_enabled = true;
65 module_param(page_pool_enabled, bool, 0400);
67 #define HNS3_SGL_SIZE(nfrag) (sizeof(struct scatterlist) * (nfrag) + \
68 sizeof(struct sg_table))
69 #define HNS3_MAX_SGL_SIZE ALIGN(HNS3_SGL_SIZE(HNS3_MAX_TSO_BD_NUM), \
70 dma_get_cache_alignment())
72 #define DEFAULT_MSG_LEVEL (NETIF_MSG_PROBE | NETIF_MSG_LINK | \
73 NETIF_MSG_IFDOWN | NETIF_MSG_IFUP)
75 #define HNS3_INNER_VLAN_TAG 1
76 #define HNS3_OUTER_VLAN_TAG 2
78 #define HNS3_MIN_TX_LEN 33U
79 #define HNS3_MIN_TUN_PKT_LEN 65U
81 /* hns3_pci_tbl - PCI Device ID Table
83 * Last entry must be all 0s
85 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
86 * Class, Class Mask, private data (not used) }
88 static const struct pci_device_id hns3_pci_tbl[] = {
89 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
90 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
91 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA),
92 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
93 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC),
94 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
95 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA),
96 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
97 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC),
98 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
99 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC),
100 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
101 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_200G_RDMA),
102 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
103 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_VF), 0},
104 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_RDMA_DCB_PFC_VF),
105 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
106 /* required last entry */
109 MODULE_DEVICE_TABLE(pci, hns3_pci_tbl);
111 #define HNS3_RX_PTYPE_ENTRY(ptype, l, s, t) \
118 #define HNS3_RX_PTYPE_UNUSED_ENTRY(ptype) \
119 { ptype, 0, CHECKSUM_NONE, HNS3_L3_TYPE_PARSE_FAIL, 0 }
121 static const struct hns3_rx_ptype hns3_rx_ptype_tbl[] = {
122 HNS3_RX_PTYPE_UNUSED_ENTRY(0),
123 HNS3_RX_PTYPE_ENTRY(1, 0, COMPLETE, ARP),
124 HNS3_RX_PTYPE_ENTRY(2, 0, COMPLETE, RARP),
125 HNS3_RX_PTYPE_ENTRY(3, 0, COMPLETE, LLDP),
126 HNS3_RX_PTYPE_ENTRY(4, 0, COMPLETE, PARSE_FAIL),
127 HNS3_RX_PTYPE_ENTRY(5, 0, COMPLETE, PARSE_FAIL),
128 HNS3_RX_PTYPE_ENTRY(6, 0, COMPLETE, PARSE_FAIL),
129 HNS3_RX_PTYPE_ENTRY(7, 0, COMPLETE, CNM),
130 HNS3_RX_PTYPE_ENTRY(8, 0, NONE, PARSE_FAIL),
131 HNS3_RX_PTYPE_UNUSED_ENTRY(9),
132 HNS3_RX_PTYPE_UNUSED_ENTRY(10),
133 HNS3_RX_PTYPE_UNUSED_ENTRY(11),
134 HNS3_RX_PTYPE_UNUSED_ENTRY(12),
135 HNS3_RX_PTYPE_UNUSED_ENTRY(13),
136 HNS3_RX_PTYPE_UNUSED_ENTRY(14),
137 HNS3_RX_PTYPE_UNUSED_ENTRY(15),
138 HNS3_RX_PTYPE_ENTRY(16, 0, COMPLETE, PARSE_FAIL),
139 HNS3_RX_PTYPE_ENTRY(17, 0, COMPLETE, IPV4),
140 HNS3_RX_PTYPE_ENTRY(18, 0, COMPLETE, IPV4),
141 HNS3_RX_PTYPE_ENTRY(19, 0, UNNECESSARY, IPV4),
142 HNS3_RX_PTYPE_ENTRY(20, 0, UNNECESSARY, IPV4),
143 HNS3_RX_PTYPE_ENTRY(21, 0, NONE, IPV4),
144 HNS3_RX_PTYPE_ENTRY(22, 0, UNNECESSARY, IPV4),
145 HNS3_RX_PTYPE_ENTRY(23, 0, NONE, IPV4),
146 HNS3_RX_PTYPE_ENTRY(24, 0, NONE, IPV4),
147 HNS3_RX_PTYPE_ENTRY(25, 0, UNNECESSARY, IPV4),
148 HNS3_RX_PTYPE_UNUSED_ENTRY(26),
149 HNS3_RX_PTYPE_UNUSED_ENTRY(27),
150 HNS3_RX_PTYPE_UNUSED_ENTRY(28),
151 HNS3_RX_PTYPE_ENTRY(29, 0, COMPLETE, PARSE_FAIL),
152 HNS3_RX_PTYPE_ENTRY(30, 0, COMPLETE, PARSE_FAIL),
153 HNS3_RX_PTYPE_ENTRY(31, 0, COMPLETE, IPV4),
154 HNS3_RX_PTYPE_ENTRY(32, 0, COMPLETE, IPV4),
155 HNS3_RX_PTYPE_ENTRY(33, 1, UNNECESSARY, IPV4),
156 HNS3_RX_PTYPE_ENTRY(34, 1, UNNECESSARY, IPV4),
157 HNS3_RX_PTYPE_ENTRY(35, 1, UNNECESSARY, IPV4),
158 HNS3_RX_PTYPE_ENTRY(36, 0, COMPLETE, IPV4),
159 HNS3_RX_PTYPE_ENTRY(37, 0, COMPLETE, IPV4),
160 HNS3_RX_PTYPE_UNUSED_ENTRY(38),
161 HNS3_RX_PTYPE_ENTRY(39, 0, COMPLETE, IPV6),
162 HNS3_RX_PTYPE_ENTRY(40, 0, COMPLETE, IPV6),
163 HNS3_RX_PTYPE_ENTRY(41, 1, UNNECESSARY, IPV6),
164 HNS3_RX_PTYPE_ENTRY(42, 1, UNNECESSARY, IPV6),
165 HNS3_RX_PTYPE_ENTRY(43, 1, UNNECESSARY, IPV6),
166 HNS3_RX_PTYPE_ENTRY(44, 0, COMPLETE, IPV6),
167 HNS3_RX_PTYPE_ENTRY(45, 0, COMPLETE, IPV6),
168 HNS3_RX_PTYPE_UNUSED_ENTRY(46),
169 HNS3_RX_PTYPE_UNUSED_ENTRY(47),
170 HNS3_RX_PTYPE_UNUSED_ENTRY(48),
171 HNS3_RX_PTYPE_UNUSED_ENTRY(49),
172 HNS3_RX_PTYPE_UNUSED_ENTRY(50),
173 HNS3_RX_PTYPE_UNUSED_ENTRY(51),
174 HNS3_RX_PTYPE_UNUSED_ENTRY(52),
175 HNS3_RX_PTYPE_UNUSED_ENTRY(53),
176 HNS3_RX_PTYPE_UNUSED_ENTRY(54),
177 HNS3_RX_PTYPE_UNUSED_ENTRY(55),
178 HNS3_RX_PTYPE_UNUSED_ENTRY(56),
179 HNS3_RX_PTYPE_UNUSED_ENTRY(57),
180 HNS3_RX_PTYPE_UNUSED_ENTRY(58),
181 HNS3_RX_PTYPE_UNUSED_ENTRY(59),
182 HNS3_RX_PTYPE_UNUSED_ENTRY(60),
183 HNS3_RX_PTYPE_UNUSED_ENTRY(61),
184 HNS3_RX_PTYPE_UNUSED_ENTRY(62),
185 HNS3_RX_PTYPE_UNUSED_ENTRY(63),
186 HNS3_RX_PTYPE_UNUSED_ENTRY(64),
187 HNS3_RX_PTYPE_UNUSED_ENTRY(65),
188 HNS3_RX_PTYPE_UNUSED_ENTRY(66),
189 HNS3_RX_PTYPE_UNUSED_ENTRY(67),
190 HNS3_RX_PTYPE_UNUSED_ENTRY(68),
191 HNS3_RX_PTYPE_UNUSED_ENTRY(69),
192 HNS3_RX_PTYPE_UNUSED_ENTRY(70),
193 HNS3_RX_PTYPE_UNUSED_ENTRY(71),
194 HNS3_RX_PTYPE_UNUSED_ENTRY(72),
195 HNS3_RX_PTYPE_UNUSED_ENTRY(73),
196 HNS3_RX_PTYPE_UNUSED_ENTRY(74),
197 HNS3_RX_PTYPE_UNUSED_ENTRY(75),
198 HNS3_RX_PTYPE_UNUSED_ENTRY(76),
199 HNS3_RX_PTYPE_UNUSED_ENTRY(77),
200 HNS3_RX_PTYPE_UNUSED_ENTRY(78),
201 HNS3_RX_PTYPE_UNUSED_ENTRY(79),
202 HNS3_RX_PTYPE_UNUSED_ENTRY(80),
203 HNS3_RX_PTYPE_UNUSED_ENTRY(81),
204 HNS3_RX_PTYPE_UNUSED_ENTRY(82),
205 HNS3_RX_PTYPE_UNUSED_ENTRY(83),
206 HNS3_RX_PTYPE_UNUSED_ENTRY(84),
207 HNS3_RX_PTYPE_UNUSED_ENTRY(85),
208 HNS3_RX_PTYPE_UNUSED_ENTRY(86),
209 HNS3_RX_PTYPE_UNUSED_ENTRY(87),
210 HNS3_RX_PTYPE_UNUSED_ENTRY(88),
211 HNS3_RX_PTYPE_UNUSED_ENTRY(89),
212 HNS3_RX_PTYPE_UNUSED_ENTRY(90),
213 HNS3_RX_PTYPE_UNUSED_ENTRY(91),
214 HNS3_RX_PTYPE_UNUSED_ENTRY(92),
215 HNS3_RX_PTYPE_UNUSED_ENTRY(93),
216 HNS3_RX_PTYPE_UNUSED_ENTRY(94),
217 HNS3_RX_PTYPE_UNUSED_ENTRY(95),
218 HNS3_RX_PTYPE_UNUSED_ENTRY(96),
219 HNS3_RX_PTYPE_UNUSED_ENTRY(97),
220 HNS3_RX_PTYPE_UNUSED_ENTRY(98),
221 HNS3_RX_PTYPE_UNUSED_ENTRY(99),
222 HNS3_RX_PTYPE_UNUSED_ENTRY(100),
223 HNS3_RX_PTYPE_UNUSED_ENTRY(101),
224 HNS3_RX_PTYPE_UNUSED_ENTRY(102),
225 HNS3_RX_PTYPE_UNUSED_ENTRY(103),
226 HNS3_RX_PTYPE_UNUSED_ENTRY(104),
227 HNS3_RX_PTYPE_UNUSED_ENTRY(105),
228 HNS3_RX_PTYPE_UNUSED_ENTRY(106),
229 HNS3_RX_PTYPE_UNUSED_ENTRY(107),
230 HNS3_RX_PTYPE_UNUSED_ENTRY(108),
231 HNS3_RX_PTYPE_UNUSED_ENTRY(109),
232 HNS3_RX_PTYPE_UNUSED_ENTRY(110),
233 HNS3_RX_PTYPE_ENTRY(111, 0, COMPLETE, IPV6),
234 HNS3_RX_PTYPE_ENTRY(112, 0, COMPLETE, IPV6),
235 HNS3_RX_PTYPE_ENTRY(113, 0, UNNECESSARY, IPV6),
236 HNS3_RX_PTYPE_ENTRY(114, 0, UNNECESSARY, IPV6),
237 HNS3_RX_PTYPE_ENTRY(115, 0, NONE, IPV6),
238 HNS3_RX_PTYPE_ENTRY(116, 0, UNNECESSARY, IPV6),
239 HNS3_RX_PTYPE_ENTRY(117, 0, NONE, IPV6),
240 HNS3_RX_PTYPE_ENTRY(118, 0, NONE, IPV6),
241 HNS3_RX_PTYPE_ENTRY(119, 0, UNNECESSARY, IPV6),
242 HNS3_RX_PTYPE_UNUSED_ENTRY(120),
243 HNS3_RX_PTYPE_UNUSED_ENTRY(121),
244 HNS3_RX_PTYPE_UNUSED_ENTRY(122),
245 HNS3_RX_PTYPE_ENTRY(123, 0, COMPLETE, PARSE_FAIL),
246 HNS3_RX_PTYPE_ENTRY(124, 0, COMPLETE, PARSE_FAIL),
247 HNS3_RX_PTYPE_ENTRY(125, 0, COMPLETE, IPV4),
248 HNS3_RX_PTYPE_ENTRY(126, 0, COMPLETE, IPV4),
249 HNS3_RX_PTYPE_ENTRY(127, 1, UNNECESSARY, IPV4),
250 HNS3_RX_PTYPE_ENTRY(128, 1, UNNECESSARY, IPV4),
251 HNS3_RX_PTYPE_ENTRY(129, 1, UNNECESSARY, IPV4),
252 HNS3_RX_PTYPE_ENTRY(130, 0, COMPLETE, IPV4),
253 HNS3_RX_PTYPE_ENTRY(131, 0, COMPLETE, IPV4),
254 HNS3_RX_PTYPE_UNUSED_ENTRY(132),
255 HNS3_RX_PTYPE_ENTRY(133, 0, COMPLETE, IPV6),
256 HNS3_RX_PTYPE_ENTRY(134, 0, COMPLETE, IPV6),
257 HNS3_RX_PTYPE_ENTRY(135, 1, UNNECESSARY, IPV6),
258 HNS3_RX_PTYPE_ENTRY(136, 1, UNNECESSARY, IPV6),
259 HNS3_RX_PTYPE_ENTRY(137, 1, UNNECESSARY, IPV6),
260 HNS3_RX_PTYPE_ENTRY(138, 0, COMPLETE, IPV6),
261 HNS3_RX_PTYPE_ENTRY(139, 0, COMPLETE, IPV6),
262 HNS3_RX_PTYPE_UNUSED_ENTRY(140),
263 HNS3_RX_PTYPE_UNUSED_ENTRY(141),
264 HNS3_RX_PTYPE_UNUSED_ENTRY(142),
265 HNS3_RX_PTYPE_UNUSED_ENTRY(143),
266 HNS3_RX_PTYPE_UNUSED_ENTRY(144),
267 HNS3_RX_PTYPE_UNUSED_ENTRY(145),
268 HNS3_RX_PTYPE_UNUSED_ENTRY(146),
269 HNS3_RX_PTYPE_UNUSED_ENTRY(147),
270 HNS3_RX_PTYPE_UNUSED_ENTRY(148),
271 HNS3_RX_PTYPE_UNUSED_ENTRY(149),
272 HNS3_RX_PTYPE_UNUSED_ENTRY(150),
273 HNS3_RX_PTYPE_UNUSED_ENTRY(151),
274 HNS3_RX_PTYPE_UNUSED_ENTRY(152),
275 HNS3_RX_PTYPE_UNUSED_ENTRY(153),
276 HNS3_RX_PTYPE_UNUSED_ENTRY(154),
277 HNS3_RX_PTYPE_UNUSED_ENTRY(155),
278 HNS3_RX_PTYPE_UNUSED_ENTRY(156),
279 HNS3_RX_PTYPE_UNUSED_ENTRY(157),
280 HNS3_RX_PTYPE_UNUSED_ENTRY(158),
281 HNS3_RX_PTYPE_UNUSED_ENTRY(159),
282 HNS3_RX_PTYPE_UNUSED_ENTRY(160),
283 HNS3_RX_PTYPE_UNUSED_ENTRY(161),
284 HNS3_RX_PTYPE_UNUSED_ENTRY(162),
285 HNS3_RX_PTYPE_UNUSED_ENTRY(163),
286 HNS3_RX_PTYPE_UNUSED_ENTRY(164),
287 HNS3_RX_PTYPE_UNUSED_ENTRY(165),
288 HNS3_RX_PTYPE_UNUSED_ENTRY(166),
289 HNS3_RX_PTYPE_UNUSED_ENTRY(167),
290 HNS3_RX_PTYPE_UNUSED_ENTRY(168),
291 HNS3_RX_PTYPE_UNUSED_ENTRY(169),
292 HNS3_RX_PTYPE_UNUSED_ENTRY(170),
293 HNS3_RX_PTYPE_UNUSED_ENTRY(171),
294 HNS3_RX_PTYPE_UNUSED_ENTRY(172),
295 HNS3_RX_PTYPE_UNUSED_ENTRY(173),
296 HNS3_RX_PTYPE_UNUSED_ENTRY(174),
297 HNS3_RX_PTYPE_UNUSED_ENTRY(175),
298 HNS3_RX_PTYPE_UNUSED_ENTRY(176),
299 HNS3_RX_PTYPE_UNUSED_ENTRY(177),
300 HNS3_RX_PTYPE_UNUSED_ENTRY(178),
301 HNS3_RX_PTYPE_UNUSED_ENTRY(179),
302 HNS3_RX_PTYPE_UNUSED_ENTRY(180),
303 HNS3_RX_PTYPE_UNUSED_ENTRY(181),
304 HNS3_RX_PTYPE_UNUSED_ENTRY(182),
305 HNS3_RX_PTYPE_UNUSED_ENTRY(183),
306 HNS3_RX_PTYPE_UNUSED_ENTRY(184),
307 HNS3_RX_PTYPE_UNUSED_ENTRY(185),
308 HNS3_RX_PTYPE_UNUSED_ENTRY(186),
309 HNS3_RX_PTYPE_UNUSED_ENTRY(187),
310 HNS3_RX_PTYPE_UNUSED_ENTRY(188),
311 HNS3_RX_PTYPE_UNUSED_ENTRY(189),
312 HNS3_RX_PTYPE_UNUSED_ENTRY(190),
313 HNS3_RX_PTYPE_UNUSED_ENTRY(191),
314 HNS3_RX_PTYPE_UNUSED_ENTRY(192),
315 HNS3_RX_PTYPE_UNUSED_ENTRY(193),
316 HNS3_RX_PTYPE_UNUSED_ENTRY(194),
317 HNS3_RX_PTYPE_UNUSED_ENTRY(195),
318 HNS3_RX_PTYPE_UNUSED_ENTRY(196),
319 HNS3_RX_PTYPE_UNUSED_ENTRY(197),
320 HNS3_RX_PTYPE_UNUSED_ENTRY(198),
321 HNS3_RX_PTYPE_UNUSED_ENTRY(199),
322 HNS3_RX_PTYPE_UNUSED_ENTRY(200),
323 HNS3_RX_PTYPE_UNUSED_ENTRY(201),
324 HNS3_RX_PTYPE_UNUSED_ENTRY(202),
325 HNS3_RX_PTYPE_UNUSED_ENTRY(203),
326 HNS3_RX_PTYPE_UNUSED_ENTRY(204),
327 HNS3_RX_PTYPE_UNUSED_ENTRY(205),
328 HNS3_RX_PTYPE_UNUSED_ENTRY(206),
329 HNS3_RX_PTYPE_UNUSED_ENTRY(207),
330 HNS3_RX_PTYPE_UNUSED_ENTRY(208),
331 HNS3_RX_PTYPE_UNUSED_ENTRY(209),
332 HNS3_RX_PTYPE_UNUSED_ENTRY(210),
333 HNS3_RX_PTYPE_UNUSED_ENTRY(211),
334 HNS3_RX_PTYPE_UNUSED_ENTRY(212),
335 HNS3_RX_PTYPE_UNUSED_ENTRY(213),
336 HNS3_RX_PTYPE_UNUSED_ENTRY(214),
337 HNS3_RX_PTYPE_UNUSED_ENTRY(215),
338 HNS3_RX_PTYPE_UNUSED_ENTRY(216),
339 HNS3_RX_PTYPE_UNUSED_ENTRY(217),
340 HNS3_RX_PTYPE_UNUSED_ENTRY(218),
341 HNS3_RX_PTYPE_UNUSED_ENTRY(219),
342 HNS3_RX_PTYPE_UNUSED_ENTRY(220),
343 HNS3_RX_PTYPE_UNUSED_ENTRY(221),
344 HNS3_RX_PTYPE_UNUSED_ENTRY(222),
345 HNS3_RX_PTYPE_UNUSED_ENTRY(223),
346 HNS3_RX_PTYPE_UNUSED_ENTRY(224),
347 HNS3_RX_PTYPE_UNUSED_ENTRY(225),
348 HNS3_RX_PTYPE_UNUSED_ENTRY(226),
349 HNS3_RX_PTYPE_UNUSED_ENTRY(227),
350 HNS3_RX_PTYPE_UNUSED_ENTRY(228),
351 HNS3_RX_PTYPE_UNUSED_ENTRY(229),
352 HNS3_RX_PTYPE_UNUSED_ENTRY(230),
353 HNS3_RX_PTYPE_UNUSED_ENTRY(231),
354 HNS3_RX_PTYPE_UNUSED_ENTRY(232),
355 HNS3_RX_PTYPE_UNUSED_ENTRY(233),
356 HNS3_RX_PTYPE_UNUSED_ENTRY(234),
357 HNS3_RX_PTYPE_UNUSED_ENTRY(235),
358 HNS3_RX_PTYPE_UNUSED_ENTRY(236),
359 HNS3_RX_PTYPE_UNUSED_ENTRY(237),
360 HNS3_RX_PTYPE_UNUSED_ENTRY(238),
361 HNS3_RX_PTYPE_UNUSED_ENTRY(239),
362 HNS3_RX_PTYPE_UNUSED_ENTRY(240),
363 HNS3_RX_PTYPE_UNUSED_ENTRY(241),
364 HNS3_RX_PTYPE_UNUSED_ENTRY(242),
365 HNS3_RX_PTYPE_UNUSED_ENTRY(243),
366 HNS3_RX_PTYPE_UNUSED_ENTRY(244),
367 HNS3_RX_PTYPE_UNUSED_ENTRY(245),
368 HNS3_RX_PTYPE_UNUSED_ENTRY(246),
369 HNS3_RX_PTYPE_UNUSED_ENTRY(247),
370 HNS3_RX_PTYPE_UNUSED_ENTRY(248),
371 HNS3_RX_PTYPE_UNUSED_ENTRY(249),
372 HNS3_RX_PTYPE_UNUSED_ENTRY(250),
373 HNS3_RX_PTYPE_UNUSED_ENTRY(251),
374 HNS3_RX_PTYPE_UNUSED_ENTRY(252),
375 HNS3_RX_PTYPE_UNUSED_ENTRY(253),
376 HNS3_RX_PTYPE_UNUSED_ENTRY(254),
377 HNS3_RX_PTYPE_UNUSED_ENTRY(255),
380 #define HNS3_INVALID_PTYPE \
381 ARRAY_SIZE(hns3_rx_ptype_tbl)
383 static irqreturn_t hns3_irq_handle(int irq, void *vector)
385 struct hns3_enet_tqp_vector *tqp_vector = vector;
387 napi_schedule_irqoff(&tqp_vector->napi);
388 tqp_vector->event_cnt++;
393 static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv)
395 struct hns3_enet_tqp_vector *tqp_vectors;
398 for (i = 0; i < priv->vector_num; i++) {
399 tqp_vectors = &priv->tqp_vector[i];
401 if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED)
404 /* clear the affinity mask */
405 irq_set_affinity_hint(tqp_vectors->vector_irq, NULL);
407 /* release the irq resource */
408 free_irq(tqp_vectors->vector_irq, tqp_vectors);
409 tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED;
413 static int hns3_nic_init_irq(struct hns3_nic_priv *priv)
415 struct hns3_enet_tqp_vector *tqp_vectors;
416 int txrx_int_idx = 0;
422 for (i = 0; i < priv->vector_num; i++) {
423 tqp_vectors = &priv->tqp_vector[i];
425 if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED)
428 if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) {
429 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
430 "%s-%s-%s-%d", hns3_driver_name,
431 pci_name(priv->ae_handle->pdev),
432 "TxRx", txrx_int_idx++);
434 } else if (tqp_vectors->rx_group.ring) {
435 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
436 "%s-%s-%s-%d", hns3_driver_name,
437 pci_name(priv->ae_handle->pdev),
439 } else if (tqp_vectors->tx_group.ring) {
440 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
441 "%s-%s-%s-%d", hns3_driver_name,
442 pci_name(priv->ae_handle->pdev),
445 /* Skip this unused q_vector */
449 tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0';
451 irq_set_status_flags(tqp_vectors->vector_irq, IRQ_NOAUTOEN);
452 ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0,
453 tqp_vectors->name, tqp_vectors);
455 netdev_err(priv->netdev, "request irq(%d) fail\n",
456 tqp_vectors->vector_irq);
457 hns3_nic_uninit_irq(priv);
461 irq_set_affinity_hint(tqp_vectors->vector_irq,
462 &tqp_vectors->affinity_mask);
464 tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED;
470 static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector,
473 writel(mask_en, tqp_vector->mask_addr);
476 static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector)
478 napi_enable(&tqp_vector->napi);
479 enable_irq(tqp_vector->vector_irq);
482 hns3_mask_vector_irq(tqp_vector, 1);
485 static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector)
488 hns3_mask_vector_irq(tqp_vector, 0);
490 disable_irq(tqp_vector->vector_irq);
491 napi_disable(&tqp_vector->napi);
492 cancel_work_sync(&tqp_vector->rx_group.dim.work);
493 cancel_work_sync(&tqp_vector->tx_group.dim.work);
496 void hns3_set_vector_coalesce_rl(struct hns3_enet_tqp_vector *tqp_vector,
499 u32 rl_reg = hns3_rl_usec_to_reg(rl_value);
501 /* this defines the configuration for RL (Interrupt Rate Limiter).
502 * Rl defines rate of interrupts i.e. number of interrupts-per-second
503 * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing
505 if (rl_reg > 0 && !tqp_vector->tx_group.coal.adapt_enable &&
506 !tqp_vector->rx_group.coal.adapt_enable)
507 /* According to the hardware, the range of rl_reg is
508 * 0-59 and the unit is 4.
510 rl_reg |= HNS3_INT_RL_ENABLE_MASK;
512 writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
515 void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
520 if (tqp_vector->rx_group.coal.unit_1us)
521 new_val = gl_value | HNS3_INT_GL_1US;
523 new_val = hns3_gl_usec_to_reg(gl_value);
525 writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
528 void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
533 if (tqp_vector->tx_group.coal.unit_1us)
534 new_val = gl_value | HNS3_INT_GL_1US;
536 new_val = hns3_gl_usec_to_reg(gl_value);
538 writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
541 void hns3_set_vector_coalesce_tx_ql(struct hns3_enet_tqp_vector *tqp_vector,
544 writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_TX_QL_OFFSET);
547 void hns3_set_vector_coalesce_rx_ql(struct hns3_enet_tqp_vector *tqp_vector,
550 writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_RX_QL_OFFSET);
553 static void hns3_vector_coalesce_init(struct hns3_enet_tqp_vector *tqp_vector,
554 struct hns3_nic_priv *priv)
556 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
557 struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
558 struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
559 struct hns3_enet_coalesce *ptx_coal = &priv->tx_coal;
560 struct hns3_enet_coalesce *prx_coal = &priv->rx_coal;
562 tx_coal->adapt_enable = ptx_coal->adapt_enable;
563 rx_coal->adapt_enable = prx_coal->adapt_enable;
565 tx_coal->int_gl = ptx_coal->int_gl;
566 rx_coal->int_gl = prx_coal->int_gl;
568 rx_coal->flow_level = prx_coal->flow_level;
569 tx_coal->flow_level = ptx_coal->flow_level;
571 /* device version above V3(include V3), GL can configure 1us
572 * unit, so uses 1us unit.
574 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3) {
575 tx_coal->unit_1us = 1;
576 rx_coal->unit_1us = 1;
579 if (ae_dev->dev_specs.int_ql_max) {
580 tx_coal->ql_enable = 1;
581 rx_coal->ql_enable = 1;
582 tx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
583 rx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
584 tx_coal->int_ql = ptx_coal->int_ql;
585 rx_coal->int_ql = prx_coal->int_ql;
590 hns3_vector_coalesce_init_hw(struct hns3_enet_tqp_vector *tqp_vector,
591 struct hns3_nic_priv *priv)
593 struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
594 struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
595 struct hnae3_handle *h = priv->ae_handle;
597 hns3_set_vector_coalesce_tx_gl(tqp_vector, tx_coal->int_gl);
598 hns3_set_vector_coalesce_rx_gl(tqp_vector, rx_coal->int_gl);
599 hns3_set_vector_coalesce_rl(tqp_vector, h->kinfo.int_rl_setting);
601 if (tx_coal->ql_enable)
602 hns3_set_vector_coalesce_tx_ql(tqp_vector, tx_coal->int_ql);
604 if (rx_coal->ql_enable)
605 hns3_set_vector_coalesce_rx_ql(tqp_vector, rx_coal->int_ql);
608 static int hns3_nic_set_real_num_queue(struct net_device *netdev)
610 struct hnae3_handle *h = hns3_get_handle(netdev);
611 struct hnae3_knic_private_info *kinfo = &h->kinfo;
612 struct hnae3_tc_info *tc_info = &kinfo->tc_info;
613 unsigned int queue_size = kinfo->num_tqps;
616 if (tc_info->num_tc <= 1 && !tc_info->mqprio_active) {
617 netdev_reset_tc(netdev);
619 ret = netdev_set_num_tc(netdev, tc_info->num_tc);
622 "netdev_set_num_tc fail, ret=%d!\n", ret);
626 for (i = 0; i < tc_info->num_tc; i++)
627 netdev_set_tc_queue(netdev, i, tc_info->tqp_count[i],
628 tc_info->tqp_offset[i]);
631 ret = netif_set_real_num_tx_queues(netdev, queue_size);
634 "netif_set_real_num_tx_queues fail, ret=%d!\n", ret);
638 ret = netif_set_real_num_rx_queues(netdev, queue_size);
641 "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
648 u16 hns3_get_max_available_channels(struct hnae3_handle *h)
650 u16 alloc_tqps, max_rss_size, rss_size;
652 h->ae_algo->ops->get_tqps_and_rss_info(h, &alloc_tqps, &max_rss_size);
653 rss_size = alloc_tqps / h->kinfo.tc_info.num_tc;
655 return min_t(u16, rss_size, max_rss_size);
658 static void hns3_tqp_enable(struct hnae3_queue *tqp)
662 rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
663 rcb_reg |= BIT(HNS3_RING_EN_B);
664 hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
667 static void hns3_tqp_disable(struct hnae3_queue *tqp)
671 rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
672 rcb_reg &= ~BIT(HNS3_RING_EN_B);
673 hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
676 static void hns3_free_rx_cpu_rmap(struct net_device *netdev)
678 #ifdef CONFIG_RFS_ACCEL
679 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
680 netdev->rx_cpu_rmap = NULL;
684 static int hns3_set_rx_cpu_rmap(struct net_device *netdev)
686 #ifdef CONFIG_RFS_ACCEL
687 struct hns3_nic_priv *priv = netdev_priv(netdev);
688 struct hns3_enet_tqp_vector *tqp_vector;
691 if (!netdev->rx_cpu_rmap) {
692 netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->vector_num);
693 if (!netdev->rx_cpu_rmap)
697 for (i = 0; i < priv->vector_num; i++) {
698 tqp_vector = &priv->tqp_vector[i];
699 ret = irq_cpu_rmap_add(netdev->rx_cpu_rmap,
700 tqp_vector->vector_irq);
702 hns3_free_rx_cpu_rmap(netdev);
710 static int hns3_nic_net_up(struct net_device *netdev)
712 struct hns3_nic_priv *priv = netdev_priv(netdev);
713 struct hnae3_handle *h = priv->ae_handle;
717 ret = hns3_nic_reset_all_ring(h);
721 clear_bit(HNS3_NIC_STATE_DOWN, &priv->state);
723 /* enable the vectors */
724 for (i = 0; i < priv->vector_num; i++)
725 hns3_vector_enable(&priv->tqp_vector[i]);
728 for (j = 0; j < h->kinfo.num_tqps; j++)
729 hns3_tqp_enable(h->kinfo.tqp[j]);
731 /* start the ae_dev */
732 ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0;
734 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
736 hns3_tqp_disable(h->kinfo.tqp[j]);
738 for (j = i - 1; j >= 0; j--)
739 hns3_vector_disable(&priv->tqp_vector[j]);
745 static void hns3_config_xps(struct hns3_nic_priv *priv)
749 for (i = 0; i < priv->vector_num; i++) {
750 struct hns3_enet_tqp_vector *tqp_vector = &priv->tqp_vector[i];
751 struct hns3_enet_ring *ring = tqp_vector->tx_group.ring;
756 ret = netif_set_xps_queue(priv->netdev,
757 &tqp_vector->affinity_mask,
758 ring->tqp->tqp_index);
760 netdev_warn(priv->netdev,
761 "set xps queue failed: %d", ret);
768 static int hns3_nic_net_open(struct net_device *netdev)
770 struct hns3_nic_priv *priv = netdev_priv(netdev);
771 struct hnae3_handle *h = hns3_get_handle(netdev);
772 struct hnae3_knic_private_info *kinfo;
775 if (hns3_nic_resetting(netdev))
778 if (!test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
779 netdev_warn(netdev, "net open repeatedly!\n");
783 netif_carrier_off(netdev);
785 ret = hns3_nic_set_real_num_queue(netdev);
789 ret = hns3_nic_net_up(netdev);
791 netdev_err(netdev, "net up fail, ret=%d!\n", ret);
796 for (i = 0; i < HNAE3_MAX_USER_PRIO; i++)
797 netdev_set_prio_tc_map(netdev, i, kinfo->tc_info.prio_tc[i]);
799 if (h->ae_algo->ops->set_timer_task)
800 h->ae_algo->ops->set_timer_task(priv->ae_handle, true);
802 hns3_config_xps(priv);
804 netif_dbg(h, drv, netdev, "net open\n");
809 static void hns3_reset_tx_queue(struct hnae3_handle *h)
811 struct net_device *ndev = h->kinfo.netdev;
812 struct hns3_nic_priv *priv = netdev_priv(ndev);
813 struct netdev_queue *dev_queue;
816 for (i = 0; i < h->kinfo.num_tqps; i++) {
817 dev_queue = netdev_get_tx_queue(ndev,
818 priv->ring[i].queue_index);
819 netdev_tx_reset_queue(dev_queue);
823 static void hns3_nic_net_down(struct net_device *netdev)
825 struct hns3_nic_priv *priv = netdev_priv(netdev);
826 struct hnae3_handle *h = hns3_get_handle(netdev);
827 const struct hnae3_ae_ops *ops;
830 /* disable vectors */
831 for (i = 0; i < priv->vector_num; i++)
832 hns3_vector_disable(&priv->tqp_vector[i]);
835 for (i = 0; i < h->kinfo.num_tqps; i++)
836 hns3_tqp_disable(h->kinfo.tqp[i]);
839 ops = priv->ae_handle->ae_algo->ops;
841 ops->stop(priv->ae_handle);
843 /* delay ring buffer clearing to hns3_reset_notify_uninit_enet
844 * during reset process, because driver may not be able
845 * to disable the ring through firmware when downing the netdev.
847 if (!hns3_nic_resetting(netdev))
848 hns3_clear_all_ring(priv->ae_handle, false);
850 hns3_reset_tx_queue(priv->ae_handle);
853 static int hns3_nic_net_stop(struct net_device *netdev)
855 struct hns3_nic_priv *priv = netdev_priv(netdev);
856 struct hnae3_handle *h = hns3_get_handle(netdev);
858 if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
861 netif_dbg(h, drv, netdev, "net stop\n");
863 if (h->ae_algo->ops->set_timer_task)
864 h->ae_algo->ops->set_timer_task(priv->ae_handle, false);
866 netif_carrier_off(netdev);
867 netif_tx_disable(netdev);
869 hns3_nic_net_down(netdev);
874 static int hns3_nic_uc_sync(struct net_device *netdev,
875 const unsigned char *addr)
877 struct hnae3_handle *h = hns3_get_handle(netdev);
879 if (h->ae_algo->ops->add_uc_addr)
880 return h->ae_algo->ops->add_uc_addr(h, addr);
885 static int hns3_nic_uc_unsync(struct net_device *netdev,
886 const unsigned char *addr)
888 struct hnae3_handle *h = hns3_get_handle(netdev);
890 /* need ignore the request of removing device address, because
891 * we store the device address and other addresses of uc list
892 * in the function's mac filter list.
894 if (ether_addr_equal(addr, netdev->dev_addr))
897 if (h->ae_algo->ops->rm_uc_addr)
898 return h->ae_algo->ops->rm_uc_addr(h, addr);
903 static int hns3_nic_mc_sync(struct net_device *netdev,
904 const unsigned char *addr)
906 struct hnae3_handle *h = hns3_get_handle(netdev);
908 if (h->ae_algo->ops->add_mc_addr)
909 return h->ae_algo->ops->add_mc_addr(h, addr);
914 static int hns3_nic_mc_unsync(struct net_device *netdev,
915 const unsigned char *addr)
917 struct hnae3_handle *h = hns3_get_handle(netdev);
919 if (h->ae_algo->ops->rm_mc_addr)
920 return h->ae_algo->ops->rm_mc_addr(h, addr);
925 static u8 hns3_get_netdev_flags(struct net_device *netdev)
929 if (netdev->flags & IFF_PROMISC)
930 flags = HNAE3_USER_UPE | HNAE3_USER_MPE | HNAE3_BPE;
931 else if (netdev->flags & IFF_ALLMULTI)
932 flags = HNAE3_USER_MPE;
937 static void hns3_nic_set_rx_mode(struct net_device *netdev)
939 struct hnae3_handle *h = hns3_get_handle(netdev);
942 new_flags = hns3_get_netdev_flags(netdev);
944 __dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync);
945 __dev_mc_sync(netdev, hns3_nic_mc_sync, hns3_nic_mc_unsync);
947 /* User mode Promisc mode enable and vlan filtering is disabled to
948 * let all packets in.
950 h->netdev_flags = new_flags;
951 hns3_request_update_promisc_mode(h);
954 void hns3_request_update_promisc_mode(struct hnae3_handle *handle)
956 const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
958 if (ops->request_update_promisc_mode)
959 ops->request_update_promisc_mode(handle);
962 static u32 hns3_tx_spare_space(struct hns3_enet_ring *ring)
964 struct hns3_tx_spare *tx_spare = ring->tx_spare;
967 /* This smp_load_acquire() pairs with smp_store_release() in
968 * hns3_tx_spare_update() called in tx desc cleaning process.
970 ntc = smp_load_acquire(&tx_spare->last_to_clean);
971 ntu = tx_spare->next_to_use;
974 return ntc - ntu - 1;
976 /* The free tx buffer is divided into two part, so pick the
979 return max(ntc, tx_spare->len - ntu) - 1;
982 static void hns3_tx_spare_update(struct hns3_enet_ring *ring)
984 struct hns3_tx_spare *tx_spare = ring->tx_spare;
987 tx_spare->last_to_clean == tx_spare->next_to_clean)
990 /* This smp_store_release() pairs with smp_load_acquire() in
991 * hns3_tx_spare_space() called in xmit process.
993 smp_store_release(&tx_spare->last_to_clean,
994 tx_spare->next_to_clean);
997 static bool hns3_can_use_tx_bounce(struct hns3_enet_ring *ring,
1001 u32 len = skb->len <= ring->tx_copybreak ? skb->len :
1004 if (len > ring->tx_copybreak)
1007 if (ALIGN(len, dma_get_cache_alignment()) > space) {
1008 u64_stats_update_begin(&ring->syncp);
1009 ring->stats.tx_spare_full++;
1010 u64_stats_update_end(&ring->syncp);
1017 static bool hns3_can_use_tx_sgl(struct hns3_enet_ring *ring,
1018 struct sk_buff *skb,
1021 if (skb->len <= ring->tx_copybreak || !tx_sgl ||
1022 (!skb_has_frag_list(skb) &&
1023 skb_shinfo(skb)->nr_frags < tx_sgl))
1026 if (space < HNS3_MAX_SGL_SIZE) {
1027 u64_stats_update_begin(&ring->syncp);
1028 ring->stats.tx_spare_full++;
1029 u64_stats_update_end(&ring->syncp);
1036 static void hns3_init_tx_spare_buffer(struct hns3_enet_ring *ring)
1038 struct hns3_tx_spare *tx_spare;
1044 alloc_size = tx_spare_buf_size ? tx_spare_buf_size :
1045 ring->tqp->handle->kinfo.tx_spare_buf_size;
1049 order = get_order(alloc_size);
1050 tx_spare = devm_kzalloc(ring_to_dev(ring), sizeof(*tx_spare),
1053 /* The driver still work without the tx spare buffer */
1054 dev_warn(ring_to_dev(ring), "failed to allocate hns3_tx_spare\n");
1058 page = alloc_pages_node(dev_to_node(ring_to_dev(ring)),
1061 dev_warn(ring_to_dev(ring), "failed to allocate tx spare pages\n");
1062 devm_kfree(ring_to_dev(ring), tx_spare);
1066 dma = dma_map_page(ring_to_dev(ring), page, 0,
1067 PAGE_SIZE << order, DMA_TO_DEVICE);
1068 if (dma_mapping_error(ring_to_dev(ring), dma)) {
1069 dev_warn(ring_to_dev(ring), "failed to map pages for tx spare\n");
1071 devm_kfree(ring_to_dev(ring), tx_spare);
1075 tx_spare->dma = dma;
1076 tx_spare->buf = page_address(page);
1077 tx_spare->len = PAGE_SIZE << order;
1078 ring->tx_spare = tx_spare;
1081 /* Use hns3_tx_spare_space() to make sure there is enough buffer
1082 * before calling below function to allocate tx buffer.
1084 static void *hns3_tx_spare_alloc(struct hns3_enet_ring *ring,
1085 unsigned int size, dma_addr_t *dma,
1088 struct hns3_tx_spare *tx_spare = ring->tx_spare;
1089 u32 ntu = tx_spare->next_to_use;
1091 size = ALIGN(size, dma_get_cache_alignment());
1094 /* Tx spare buffer wraps back here because the end of
1095 * freed tx buffer is not enough.
1097 if (ntu + size > tx_spare->len) {
1098 *cb_len += (tx_spare->len - ntu);
1102 tx_spare->next_to_use = ntu + size;
1103 if (tx_spare->next_to_use == tx_spare->len)
1104 tx_spare->next_to_use = 0;
1106 *dma = tx_spare->dma + ntu;
1108 return tx_spare->buf + ntu;
1111 static void hns3_tx_spare_rollback(struct hns3_enet_ring *ring, u32 len)
1113 struct hns3_tx_spare *tx_spare = ring->tx_spare;
1115 if (len > tx_spare->next_to_use) {
1116 len -= tx_spare->next_to_use;
1117 tx_spare->next_to_use = tx_spare->len - len;
1119 tx_spare->next_to_use -= len;
1123 static void hns3_tx_spare_reclaim_cb(struct hns3_enet_ring *ring,
1124 struct hns3_desc_cb *cb)
1126 struct hns3_tx_spare *tx_spare = ring->tx_spare;
1127 u32 ntc = tx_spare->next_to_clean;
1128 u32 len = cb->length;
1130 tx_spare->next_to_clean += len;
1132 if (tx_spare->next_to_clean >= tx_spare->len) {
1133 tx_spare->next_to_clean -= tx_spare->len;
1135 if (tx_spare->next_to_clean) {
1137 len = tx_spare->next_to_clean;
1141 /* This tx spare buffer is only really reclaimed after calling
1142 * hns3_tx_spare_update(), so it is still safe to use the info in
1143 * the tx buffer to do the dma sync or sg unmapping after
1144 * tx_spare->next_to_clean is moved forword.
1146 if (cb->type & (DESC_TYPE_BOUNCE_HEAD | DESC_TYPE_BOUNCE_ALL)) {
1147 dma_addr_t dma = tx_spare->dma + ntc;
1149 dma_sync_single_for_cpu(ring_to_dev(ring), dma, len,
1152 struct sg_table *sgt = tx_spare->buf + ntc;
1154 dma_unmap_sg(ring_to_dev(ring), sgt->sgl, sgt->orig_nents,
1159 static int hns3_set_tso(struct sk_buff *skb, u32 *paylen_fdop_ol4cs,
1160 u16 *mss, u32 *type_cs_vlan_tso, u32 *send_bytes)
1162 u32 l4_offset, hdr_len;
1163 union l3_hdr_info l3;
1164 union l4_hdr_info l4;
1168 if (!skb_is_gso(skb))
1171 ret = skb_cow_head(skb, 0);
1172 if (unlikely(ret < 0))
1175 l3.hdr = skb_network_header(skb);
1176 l4.hdr = skb_transport_header(skb);
1178 /* Software should clear the IPv4's checksum field when tso is
1181 if (l3.v4->version == 4)
1185 if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
1187 SKB_GSO_UDP_TUNNEL |
1188 SKB_GSO_UDP_TUNNEL_CSUM)) {
1189 /* reset l3&l4 pointers from outer to inner headers */
1190 l3.hdr = skb_inner_network_header(skb);
1191 l4.hdr = skb_inner_transport_header(skb);
1193 /* Software should clear the IPv4's checksum field when
1196 if (l3.v4->version == 4)
1200 /* normal or tunnel packet */
1201 l4_offset = l4.hdr - skb->data;
1203 /* remove payload length from inner pseudo checksum when tso */
1204 l4_paylen = skb->len - l4_offset;
1206 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) {
1207 hdr_len = sizeof(*l4.udp) + l4_offset;
1208 csum_replace_by_diff(&l4.udp->check,
1209 (__force __wsum)htonl(l4_paylen));
1211 hdr_len = (l4.tcp->doff << 2) + l4_offset;
1212 csum_replace_by_diff(&l4.tcp->check,
1213 (__force __wsum)htonl(l4_paylen));
1216 *send_bytes = (skb_shinfo(skb)->gso_segs - 1) * hdr_len + skb->len;
1218 /* find the txbd field values */
1219 *paylen_fdop_ol4cs = skb->len - hdr_len;
1220 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_TSO_B, 1);
1222 /* offload outer UDP header checksum */
1223 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)
1224 hns3_set_field(*paylen_fdop_ol4cs, HNS3_TXD_OL4CS_B, 1);
1226 /* get MSS for TSO */
1227 *mss = skb_shinfo(skb)->gso_size;
1229 trace_hns3_tso(skb);
1234 static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
1237 union l3_hdr_info l3;
1238 unsigned char *l4_hdr;
1239 unsigned char *exthdr;
1243 /* find outer header point */
1244 l3.hdr = skb_network_header(skb);
1245 l4_hdr = skb_transport_header(skb);
1247 if (skb->protocol == htons(ETH_P_IPV6)) {
1248 exthdr = l3.hdr + sizeof(*l3.v6);
1249 l4_proto_tmp = l3.v6->nexthdr;
1250 if (l4_hdr != exthdr)
1251 ipv6_skip_exthdr(skb, exthdr - skb->data,
1252 &l4_proto_tmp, &frag_off);
1253 } else if (skb->protocol == htons(ETH_P_IP)) {
1254 l4_proto_tmp = l3.v4->protocol;
1259 *ol4_proto = l4_proto_tmp;
1262 if (!skb->encapsulation) {
1267 /* find inner header point */
1268 l3.hdr = skb_inner_network_header(skb);
1269 l4_hdr = skb_inner_transport_header(skb);
1271 if (l3.v6->version == 6) {
1272 exthdr = l3.hdr + sizeof(*l3.v6);
1273 l4_proto_tmp = l3.v6->nexthdr;
1274 if (l4_hdr != exthdr)
1275 ipv6_skip_exthdr(skb, exthdr - skb->data,
1276 &l4_proto_tmp, &frag_off);
1277 } else if (l3.v4->version == 4) {
1278 l4_proto_tmp = l3.v4->protocol;
1281 *il4_proto = l4_proto_tmp;
1286 /* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
1287 * and it is udp packet, which has a dest port as the IANA assigned.
1288 * the hardware is expected to do the checksum offload, but the
1289 * hardware will not do the checksum offload when udp dest port is
1290 * 4789, 4790 or 6081.
1292 static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
1294 struct hns3_nic_priv *priv = netdev_priv(skb->dev);
1295 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
1296 union l4_hdr_info l4;
1298 /* device version above V3(include V3), the hardware can
1299 * do this checksum offload.
1301 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
1304 l4.hdr = skb_transport_header(skb);
1306 if (!(!skb->encapsulation &&
1307 (l4.udp->dest == htons(IANA_VXLAN_UDP_PORT) ||
1308 l4.udp->dest == htons(GENEVE_UDP_PORT) ||
1309 l4.udp->dest == htons(4790))))
1315 static void hns3_set_outer_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
1316 u32 *ol_type_vlan_len_msec)
1318 u32 l2_len, l3_len, l4_len;
1319 unsigned char *il2_hdr;
1320 union l3_hdr_info l3;
1321 union l4_hdr_info l4;
1323 l3.hdr = skb_network_header(skb);
1324 l4.hdr = skb_transport_header(skb);
1326 /* compute OL2 header size, defined in 2 Bytes */
1327 l2_len = l3.hdr - skb->data;
1328 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L2LEN_S, l2_len >> 1);
1330 /* compute OL3 header size, defined in 4 Bytes */
1331 l3_len = l4.hdr - l3.hdr;
1332 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_S, l3_len >> 2);
1334 il2_hdr = skb_inner_mac_header(skb);
1335 /* compute OL4 header size, defined in 4 Bytes */
1336 l4_len = il2_hdr - l4.hdr;
1337 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L4LEN_S, l4_len >> 2);
1339 /* define outer network header type */
1340 if (skb->protocol == htons(ETH_P_IP)) {
1341 if (skb_is_gso(skb))
1342 hns3_set_field(*ol_type_vlan_len_msec,
1344 HNS3_OL3T_IPV4_CSUM);
1346 hns3_set_field(*ol_type_vlan_len_msec,
1348 HNS3_OL3T_IPV4_NO_CSUM);
1349 } else if (skb->protocol == htons(ETH_P_IPV6)) {
1350 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_S,
1354 if (ol4_proto == IPPROTO_UDP)
1355 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
1356 HNS3_TUN_MAC_IN_UDP);
1357 else if (ol4_proto == IPPROTO_GRE)
1358 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
1362 static int hns3_set_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
1363 u8 il4_proto, u32 *type_cs_vlan_tso,
1364 u32 *ol_type_vlan_len_msec)
1366 unsigned char *l2_hdr = skb->data;
1367 u32 l4_proto = ol4_proto;
1368 union l4_hdr_info l4;
1369 union l3_hdr_info l3;
1372 l4.hdr = skb_transport_header(skb);
1373 l3.hdr = skb_network_header(skb);
1375 /* handle encapsulation skb */
1376 if (skb->encapsulation) {
1377 /* If this is a not UDP/GRE encapsulation skb */
1378 if (!(ol4_proto == IPPROTO_UDP || ol4_proto == IPPROTO_GRE)) {
1379 /* drop the skb tunnel packet if hardware don't support,
1380 * because hardware can't calculate csum when TSO.
1382 if (skb_is_gso(skb))
1385 /* the stack computes the IP header already,
1386 * driver calculate l4 checksum when not TSO.
1388 return skb_checksum_help(skb);
1391 hns3_set_outer_l2l3l4(skb, ol4_proto, ol_type_vlan_len_msec);
1393 /* switch to inner header */
1394 l2_hdr = skb_inner_mac_header(skb);
1395 l3.hdr = skb_inner_network_header(skb);
1396 l4.hdr = skb_inner_transport_header(skb);
1397 l4_proto = il4_proto;
1400 if (l3.v4->version == 4) {
1401 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
1404 /* the stack computes the IP header already, the only time we
1405 * need the hardware to recompute it is in the case of TSO.
1407 if (skb_is_gso(skb))
1408 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3CS_B, 1);
1409 } else if (l3.v6->version == 6) {
1410 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
1414 /* compute inner(/normal) L2 header size, defined in 2 Bytes */
1415 l2_len = l3.hdr - l2_hdr;
1416 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S, l2_len >> 1);
1418 /* compute inner(/normal) L3 header size, defined in 4 Bytes */
1419 l3_len = l4.hdr - l3.hdr;
1420 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_S, l3_len >> 2);
1422 /* compute inner(/normal) L4 header size, defined in 4 Bytes */
1425 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1426 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1428 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1432 if (hns3_tunnel_csum_bug(skb)) {
1433 int ret = skb_put_padto(skb, HNS3_MIN_TUN_PKT_LEN);
1435 return ret ? ret : skb_checksum_help(skb);
1438 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1439 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1441 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1442 (sizeof(struct udphdr) >> 2));
1445 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1446 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1448 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1449 (sizeof(struct sctphdr) >> 2));
1452 /* drop the skb tunnel packet if hardware don't support,
1453 * because hardware can't calculate csum when TSO.
1455 if (skb_is_gso(skb))
1458 /* the stack computes the IP header already,
1459 * driver calculate l4 checksum when not TSO.
1461 return skb_checksum_help(skb);
1467 static int hns3_handle_vtags(struct hns3_enet_ring *tx_ring,
1468 struct sk_buff *skb)
1470 struct hnae3_handle *handle = tx_ring->tqp->handle;
1471 struct hnae3_ae_dev *ae_dev;
1472 struct vlan_ethhdr *vhdr;
1475 if (!(skb->protocol == htons(ETH_P_8021Q) ||
1476 skb_vlan_tag_present(skb)))
1479 /* For HW limitation on HNAE3_DEVICE_VERSION_V2, if port based insert
1480 * VLAN enabled, only one VLAN header is allowed in skb, otherwise it
1481 * will cause RAS error.
1483 ae_dev = pci_get_drvdata(handle->pdev);
1484 if (unlikely(skb_vlan_tagged_multi(skb) &&
1485 ae_dev->dev_version <= HNAE3_DEVICE_VERSION_V2 &&
1486 handle->port_base_vlan_state ==
1487 HNAE3_PORT_BASE_VLAN_ENABLE))
1490 if (skb->protocol == htons(ETH_P_8021Q) &&
1491 !(handle->kinfo.netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
1492 /* When HW VLAN acceleration is turned off, and the stack
1493 * sets the protocol to 802.1q, the driver just need to
1494 * set the protocol to the encapsulated ethertype.
1496 skb->protocol = vlan_get_protocol(skb);
1500 if (skb_vlan_tag_present(skb)) {
1501 /* Based on hw strategy, use out_vtag in two layer tag case,
1502 * and use inner_vtag in one tag case.
1504 if (skb->protocol == htons(ETH_P_8021Q) &&
1505 handle->port_base_vlan_state ==
1506 HNAE3_PORT_BASE_VLAN_DISABLE)
1507 rc = HNS3_OUTER_VLAN_TAG;
1509 rc = HNS3_INNER_VLAN_TAG;
1511 skb->protocol = vlan_get_protocol(skb);
1515 rc = skb_cow_head(skb, 0);
1516 if (unlikely(rc < 0))
1519 vhdr = (struct vlan_ethhdr *)skb->data;
1520 vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority << VLAN_PRIO_SHIFT)
1523 skb->protocol = vlan_get_protocol(skb);
1527 /* check if the hardware is capable of checksum offloading */
1528 static bool hns3_check_hw_tx_csum(struct sk_buff *skb)
1530 struct hns3_nic_priv *priv = netdev_priv(skb->dev);
1532 /* Kindly note, due to backward compatibility of the TX descriptor,
1533 * HW checksum of the non-IP packets and GSO packets is handled at
1534 * different place in the following code
1536 if (skb_csum_is_sctp(skb) || skb_is_gso(skb) ||
1537 !test_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state))
1543 static int hns3_fill_skb_desc(struct hns3_enet_ring *ring,
1544 struct sk_buff *skb, struct hns3_desc *desc,
1545 struct hns3_desc_cb *desc_cb)
1547 u32 ol_type_vlan_len_msec = 0;
1548 u32 paylen_ol4cs = skb->len;
1549 u32 type_cs_vlan_tso = 0;
1550 u16 mss_hw_csum = 0;
1555 ret = hns3_handle_vtags(ring, skb);
1556 if (unlikely(ret < 0)) {
1557 u64_stats_update_begin(&ring->syncp);
1558 ring->stats.tx_vlan_err++;
1559 u64_stats_update_end(&ring->syncp);
1561 } else if (ret == HNS3_INNER_VLAN_TAG) {
1562 inner_vtag = skb_vlan_tag_get(skb);
1563 inner_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1565 hns3_set_field(type_cs_vlan_tso, HNS3_TXD_VLAN_B, 1);
1566 } else if (ret == HNS3_OUTER_VLAN_TAG) {
1567 out_vtag = skb_vlan_tag_get(skb);
1568 out_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1570 hns3_set_field(ol_type_vlan_len_msec, HNS3_TXD_OVLAN_B,
1574 desc_cb->send_bytes = skb->len;
1576 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1577 u8 ol4_proto, il4_proto;
1579 if (hns3_check_hw_tx_csum(skb)) {
1580 /* set checksum start and offset, defined in 2 Bytes */
1581 hns3_set_field(type_cs_vlan_tso, HNS3_TXD_CSUM_START_S,
1582 skb_checksum_start_offset(skb) >> 1);
1583 hns3_set_field(ol_type_vlan_len_msec,
1584 HNS3_TXD_CSUM_OFFSET_S,
1585 skb->csum_offset >> 1);
1586 mss_hw_csum |= BIT(HNS3_TXD_HW_CS_B);
1587 goto out_hw_tx_csum;
1590 skb_reset_mac_len(skb);
1592 ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
1593 if (unlikely(ret < 0)) {
1594 u64_stats_update_begin(&ring->syncp);
1595 ring->stats.tx_l4_proto_err++;
1596 u64_stats_update_end(&ring->syncp);
1600 ret = hns3_set_l2l3l4(skb, ol4_proto, il4_proto,
1602 &ol_type_vlan_len_msec);
1603 if (unlikely(ret < 0)) {
1604 u64_stats_update_begin(&ring->syncp);
1605 ring->stats.tx_l2l3l4_err++;
1606 u64_stats_update_end(&ring->syncp);
1610 ret = hns3_set_tso(skb, &paylen_ol4cs, &mss_hw_csum,
1611 &type_cs_vlan_tso, &desc_cb->send_bytes);
1612 if (unlikely(ret < 0)) {
1613 u64_stats_update_begin(&ring->syncp);
1614 ring->stats.tx_tso_err++;
1615 u64_stats_update_end(&ring->syncp);
1622 desc->tx.ol_type_vlan_len_msec =
1623 cpu_to_le32(ol_type_vlan_len_msec);
1624 desc->tx.type_cs_vlan_tso_len = cpu_to_le32(type_cs_vlan_tso);
1625 desc->tx.paylen_ol4cs = cpu_to_le32(paylen_ol4cs);
1626 desc->tx.mss_hw_csum = cpu_to_le16(mss_hw_csum);
1627 desc->tx.vlan_tag = cpu_to_le16(inner_vtag);
1628 desc->tx.outer_vlan_tag = cpu_to_le16(out_vtag);
1633 static int hns3_fill_desc(struct hns3_enet_ring *ring, dma_addr_t dma,
1636 #define HNS3_LIKELY_BD_NUM 1
1638 struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1639 unsigned int frag_buf_num;
1642 if (likely(size <= HNS3_MAX_BD_SIZE)) {
1643 desc->addr = cpu_to_le64(dma);
1644 desc->tx.send_size = cpu_to_le16(size);
1645 desc->tx.bdtp_fe_sc_vld_ra_ri =
1646 cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1648 trace_hns3_tx_desc(ring, ring->next_to_use);
1649 ring_ptr_move_fw(ring, next_to_use);
1650 return HNS3_LIKELY_BD_NUM;
1653 frag_buf_num = hns3_tx_bd_count(size);
1654 sizeoflast = size % HNS3_MAX_BD_SIZE;
1655 sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1657 /* When frag size is bigger than hardware limit, split this frag */
1658 for (k = 0; k < frag_buf_num; k++) {
1659 /* now, fill the descriptor */
1660 desc->addr = cpu_to_le64(dma + HNS3_MAX_BD_SIZE * k);
1661 desc->tx.send_size = cpu_to_le16((k == frag_buf_num - 1) ?
1662 (u16)sizeoflast : (u16)HNS3_MAX_BD_SIZE);
1663 desc->tx.bdtp_fe_sc_vld_ra_ri =
1664 cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1666 trace_hns3_tx_desc(ring, ring->next_to_use);
1667 /* move ring pointer to next */
1668 ring_ptr_move_fw(ring, next_to_use);
1670 desc = &ring->desc[ring->next_to_use];
1673 return frag_buf_num;
1676 static int hns3_map_and_fill_desc(struct hns3_enet_ring *ring, void *priv,
1679 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1680 struct device *dev = ring_to_dev(ring);
1684 if (type & (DESC_TYPE_FRAGLIST_SKB | DESC_TYPE_SKB)) {
1685 struct sk_buff *skb = (struct sk_buff *)priv;
1687 size = skb_headlen(skb);
1691 dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1692 } else if (type & DESC_TYPE_BOUNCE_HEAD) {
1693 /* Head data has been filled in hns3_handle_tx_bounce(),
1694 * just return 0 here.
1698 skb_frag_t *frag = (skb_frag_t *)priv;
1700 size = skb_frag_size(frag);
1704 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1707 if (unlikely(dma_mapping_error(dev, dma))) {
1708 u64_stats_update_begin(&ring->syncp);
1709 ring->stats.sw_err_cnt++;
1710 u64_stats_update_end(&ring->syncp);
1714 desc_cb->priv = priv;
1715 desc_cb->length = size;
1717 desc_cb->type = type;
1719 return hns3_fill_desc(ring, dma, size);
1722 static unsigned int hns3_skb_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1723 unsigned int bd_num)
1728 size = skb_headlen(skb);
1729 while (size > HNS3_MAX_BD_SIZE) {
1730 bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1731 size -= HNS3_MAX_BD_SIZE;
1733 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1738 bd_size[bd_num++] = size;
1739 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1743 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1744 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1745 size = skb_frag_size(frag);
1749 while (size > HNS3_MAX_BD_SIZE) {
1750 bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1751 size -= HNS3_MAX_BD_SIZE;
1753 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1757 bd_size[bd_num++] = size;
1758 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1765 static unsigned int hns3_tx_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1766 u8 max_non_tso_bd_num, unsigned int bd_num,
1767 unsigned int recursion_level)
1769 #define HNS3_MAX_RECURSION_LEVEL 24
1771 struct sk_buff *frag_skb;
1773 /* If the total len is within the max bd limit */
1774 if (likely(skb->len <= HNS3_MAX_BD_SIZE && !recursion_level &&
1775 !skb_has_frag_list(skb) &&
1776 skb_shinfo(skb)->nr_frags < max_non_tso_bd_num))
1777 return skb_shinfo(skb)->nr_frags + 1U;
1779 if (unlikely(recursion_level >= HNS3_MAX_RECURSION_LEVEL))
1782 bd_num = hns3_skb_bd_num(skb, bd_size, bd_num);
1783 if (!skb_has_frag_list(skb) || bd_num > HNS3_MAX_TSO_BD_NUM)
1786 skb_walk_frags(skb, frag_skb) {
1787 bd_num = hns3_tx_bd_num(frag_skb, bd_size, max_non_tso_bd_num,
1788 bd_num, recursion_level + 1);
1789 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1796 static unsigned int hns3_gso_hdr_len(struct sk_buff *skb)
1798 if (!skb->encapsulation)
1799 return skb_transport_offset(skb) + tcp_hdrlen(skb);
1801 return skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb);
1804 /* HW need every continuous max_non_tso_bd_num buffer data to be larger
1805 * than MSS, we simplify it by ensuring skb_headlen + the first continuous
1806 * max_non_tso_bd_num - 1 frags to be larger than gso header len + mss,
1807 * and the remaining continuous max_non_tso_bd_num - 1 frags to be larger
1808 * than MSS except the last max_non_tso_bd_num - 1 frags.
1810 static bool hns3_skb_need_linearized(struct sk_buff *skb, unsigned int *bd_size,
1811 unsigned int bd_num, u8 max_non_tso_bd_num)
1813 unsigned int tot_len = 0;
1816 for (i = 0; i < max_non_tso_bd_num - 1U; i++)
1817 tot_len += bd_size[i];
1819 /* ensure the first max_non_tso_bd_num frags is greater than
1822 if (tot_len + bd_size[max_non_tso_bd_num - 1U] <
1823 skb_shinfo(skb)->gso_size + hns3_gso_hdr_len(skb))
1826 /* ensure every continuous max_non_tso_bd_num - 1 buffer is greater
1827 * than mss except the last one.
1829 for (i = 0; i < bd_num - max_non_tso_bd_num; i++) {
1830 tot_len -= bd_size[i];
1831 tot_len += bd_size[i + max_non_tso_bd_num - 1U];
1833 if (tot_len < skb_shinfo(skb)->gso_size)
1840 void hns3_shinfo_pack(struct skb_shared_info *shinfo, __u32 *size)
1844 for (i = 0; i < MAX_SKB_FRAGS; i++)
1845 size[i] = skb_frag_size(&shinfo->frags[i]);
1848 static int hns3_skb_linearize(struct hns3_enet_ring *ring,
1849 struct sk_buff *skb,
1850 u8 max_non_tso_bd_num,
1851 unsigned int bd_num)
1853 /* 'bd_num == UINT_MAX' means the skb' fraglist has a
1854 * recursion level of over HNS3_MAX_RECURSION_LEVEL.
1856 if (bd_num == UINT_MAX) {
1857 u64_stats_update_begin(&ring->syncp);
1858 ring->stats.over_max_recursion++;
1859 u64_stats_update_end(&ring->syncp);
1863 /* The skb->len has exceeded the hw limitation, linearization
1866 if (skb->len > HNS3_MAX_TSO_SIZE ||
1867 (!skb_is_gso(skb) && skb->len >
1868 HNS3_MAX_NON_TSO_SIZE(max_non_tso_bd_num))) {
1869 u64_stats_update_begin(&ring->syncp);
1870 ring->stats.hw_limitation++;
1871 u64_stats_update_end(&ring->syncp);
1875 if (__skb_linearize(skb)) {
1876 u64_stats_update_begin(&ring->syncp);
1877 ring->stats.sw_err_cnt++;
1878 u64_stats_update_end(&ring->syncp);
1885 static int hns3_nic_maybe_stop_tx(struct hns3_enet_ring *ring,
1886 struct net_device *netdev,
1887 struct sk_buff *skb)
1889 struct hns3_nic_priv *priv = netdev_priv(netdev);
1890 u8 max_non_tso_bd_num = priv->max_non_tso_bd_num;
1891 unsigned int bd_size[HNS3_MAX_TSO_BD_NUM + 1U];
1892 unsigned int bd_num;
1894 bd_num = hns3_tx_bd_num(skb, bd_size, max_non_tso_bd_num, 0, 0);
1895 if (unlikely(bd_num > max_non_tso_bd_num)) {
1896 if (bd_num <= HNS3_MAX_TSO_BD_NUM && skb_is_gso(skb) &&
1897 !hns3_skb_need_linearized(skb, bd_size, bd_num,
1898 max_non_tso_bd_num)) {
1899 trace_hns3_over_max_bd(skb);
1903 if (hns3_skb_linearize(ring, skb, max_non_tso_bd_num,
1907 bd_num = hns3_tx_bd_count(skb->len);
1909 u64_stats_update_begin(&ring->syncp);
1910 ring->stats.tx_copy++;
1911 u64_stats_update_end(&ring->syncp);
1915 if (likely(ring_space(ring) >= bd_num))
1918 netif_stop_subqueue(netdev, ring->queue_index);
1919 smp_mb(); /* Memory barrier before checking ring_space */
1921 /* Start queue in case hns3_clean_tx_ring has just made room
1922 * available and has not seen the queue stopped state performed
1923 * by netif_stop_subqueue above.
1925 if (ring_space(ring) >= bd_num && netif_carrier_ok(netdev) &&
1926 !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
1927 netif_start_subqueue(netdev, ring->queue_index);
1931 u64_stats_update_begin(&ring->syncp);
1932 ring->stats.tx_busy++;
1933 u64_stats_update_end(&ring->syncp);
1938 static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
1940 struct device *dev = ring_to_dev(ring);
1943 for (i = 0; i < ring->desc_num; i++) {
1944 struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1945 struct hns3_desc_cb *desc_cb;
1947 memset(desc, 0, sizeof(*desc));
1949 /* check if this is where we started */
1950 if (ring->next_to_use == next_to_use_orig)
1954 ring_ptr_move_bw(ring, next_to_use);
1956 desc_cb = &ring->desc_cb[ring->next_to_use];
1961 /* unmap the descriptor dma address */
1962 if (desc_cb->type & (DESC_TYPE_SKB | DESC_TYPE_FRAGLIST_SKB))
1963 dma_unmap_single(dev, desc_cb->dma, desc_cb->length,
1965 else if (desc_cb->type &
1966 (DESC_TYPE_BOUNCE_HEAD | DESC_TYPE_BOUNCE_ALL))
1967 hns3_tx_spare_rollback(ring, desc_cb->length);
1968 else if (desc_cb->length)
1969 dma_unmap_page(dev, desc_cb->dma, desc_cb->length,
1972 desc_cb->length = 0;
1974 desc_cb->type = DESC_TYPE_UNKNOWN;
1978 static int hns3_fill_skb_to_desc(struct hns3_enet_ring *ring,
1979 struct sk_buff *skb, unsigned int type)
1981 struct sk_buff *frag_skb;
1982 int i, ret, bd_num = 0;
1984 ret = hns3_map_and_fill_desc(ring, skb, type);
1985 if (unlikely(ret < 0))
1990 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1991 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1993 ret = hns3_map_and_fill_desc(ring, frag, DESC_TYPE_PAGE);
1994 if (unlikely(ret < 0))
2000 skb_walk_frags(skb, frag_skb) {
2001 ret = hns3_fill_skb_to_desc(ring, frag_skb,
2002 DESC_TYPE_FRAGLIST_SKB);
2003 if (unlikely(ret < 0))
2012 static void hns3_tx_doorbell(struct hns3_enet_ring *ring, int num,
2015 ring->pending_buf += num;
2018 u64_stats_update_begin(&ring->syncp);
2019 ring->stats.tx_more++;
2020 u64_stats_update_end(&ring->syncp);
2024 if (!ring->pending_buf)
2027 writel(ring->pending_buf,
2028 ring->tqp->io_base + HNS3_RING_TX_RING_TAIL_REG);
2029 ring->pending_buf = 0;
2030 WRITE_ONCE(ring->last_to_use, ring->next_to_use);
2033 static void hns3_tsyn(struct net_device *netdev, struct sk_buff *skb,
2034 struct hns3_desc *desc)
2036 struct hnae3_handle *h = hns3_get_handle(netdev);
2038 if (!(h->ae_algo->ops->set_tx_hwts_info &&
2039 h->ae_algo->ops->set_tx_hwts_info(h, skb)))
2042 desc->tx.bdtp_fe_sc_vld_ra_ri |= cpu_to_le16(BIT(HNS3_TXD_TSYN_B));
2045 static int hns3_handle_tx_bounce(struct hns3_enet_ring *ring,
2046 struct sk_buff *skb)
2048 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
2049 unsigned int type = DESC_TYPE_BOUNCE_HEAD;
2050 unsigned int size = skb_headlen(skb);
2057 if (skb->len <= ring->tx_copybreak) {
2059 type = DESC_TYPE_BOUNCE_ALL;
2062 /* hns3_can_use_tx_bounce() is called to ensure the below
2063 * function can always return the tx buffer.
2065 buf = hns3_tx_spare_alloc(ring, size, &dma, &cb_len);
2067 ret = skb_copy_bits(skb, 0, buf, size);
2068 if (unlikely(ret < 0)) {
2069 hns3_tx_spare_rollback(ring, cb_len);
2070 u64_stats_update_begin(&ring->syncp);
2071 ring->stats.copy_bits_err++;
2072 u64_stats_update_end(&ring->syncp);
2076 desc_cb->priv = skb;
2077 desc_cb->length = cb_len;
2079 desc_cb->type = type;
2081 bd_num += hns3_fill_desc(ring, dma, size);
2083 if (type == DESC_TYPE_BOUNCE_HEAD) {
2084 ret = hns3_fill_skb_to_desc(ring, skb,
2085 DESC_TYPE_BOUNCE_HEAD);
2086 if (unlikely(ret < 0))
2092 dma_sync_single_for_device(ring_to_dev(ring), dma, size,
2095 u64_stats_update_begin(&ring->syncp);
2096 ring->stats.tx_bounce++;
2097 u64_stats_update_end(&ring->syncp);
2101 static int hns3_handle_tx_sgl(struct hns3_enet_ring *ring,
2102 struct sk_buff *skb)
2104 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
2105 u32 nfrag = skb_shinfo(skb)->nr_frags + 1;
2106 struct sg_table *sgt;
2112 if (skb_has_frag_list(skb))
2113 nfrag = HNS3_MAX_TSO_BD_NUM;
2115 /* hns3_can_use_tx_sgl() is called to ensure the below
2116 * function can always return the tx buffer.
2118 sgt = hns3_tx_spare_alloc(ring, HNS3_SGL_SIZE(nfrag),
2121 /* scatterlist follows by the sg table */
2122 sgt->sgl = (struct scatterlist *)(sgt + 1);
2123 sg_init_table(sgt->sgl, nfrag);
2124 nents = skb_to_sgvec(skb, sgt->sgl, 0, skb->len);
2125 if (unlikely(nents < 0)) {
2126 hns3_tx_spare_rollback(ring, cb_len);
2127 u64_stats_update_begin(&ring->syncp);
2128 ring->stats.skb2sgl_err++;
2129 u64_stats_update_end(&ring->syncp);
2133 sgt->orig_nents = nents;
2134 sgt->nents = dma_map_sg(ring_to_dev(ring), sgt->sgl, sgt->orig_nents,
2136 if (unlikely(!sgt->nents)) {
2137 hns3_tx_spare_rollback(ring, cb_len);
2138 u64_stats_update_begin(&ring->syncp);
2139 ring->stats.map_sg_err++;
2140 u64_stats_update_end(&ring->syncp);
2144 desc_cb->priv = skb;
2145 desc_cb->length = cb_len;
2147 desc_cb->type = DESC_TYPE_SGL_SKB;
2149 for (i = 0; i < sgt->nents; i++)
2150 bd_num += hns3_fill_desc(ring, sg_dma_address(sgt->sgl + i),
2151 sg_dma_len(sgt->sgl + i));
2153 u64_stats_update_begin(&ring->syncp);
2154 ring->stats.tx_sgl++;
2155 u64_stats_update_end(&ring->syncp);
2160 static int hns3_handle_desc_filling(struct hns3_enet_ring *ring,
2161 struct sk_buff *skb)
2165 if (!ring->tx_spare)
2168 space = hns3_tx_spare_space(ring);
2170 if (hns3_can_use_tx_sgl(ring, skb, space))
2171 return hns3_handle_tx_sgl(ring, skb);
2173 if (hns3_can_use_tx_bounce(ring, skb, space))
2174 return hns3_handle_tx_bounce(ring, skb);
2177 return hns3_fill_skb_to_desc(ring, skb, DESC_TYPE_SKB);
2180 netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
2182 struct hns3_nic_priv *priv = netdev_priv(netdev);
2183 struct hns3_enet_ring *ring = &priv->ring[skb->queue_mapping];
2184 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
2185 struct netdev_queue *dev_queue;
2186 int pre_ntu, next_to_use_head;
2190 /* Hardware can only handle short frames above 32 bytes */
2191 if (skb_put_padto(skb, HNS3_MIN_TX_LEN)) {
2192 hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
2194 u64_stats_update_begin(&ring->syncp);
2195 ring->stats.sw_err_cnt++;
2196 u64_stats_update_end(&ring->syncp);
2198 return NETDEV_TX_OK;
2201 /* Prefetch the data used later */
2202 prefetch(skb->data);
2204 ret = hns3_nic_maybe_stop_tx(ring, netdev, skb);
2205 if (unlikely(ret <= 0)) {
2206 if (ret == -EBUSY) {
2207 hns3_tx_doorbell(ring, 0, true);
2208 return NETDEV_TX_BUSY;
2211 hns3_rl_err(netdev, "xmit error: %d!\n", ret);
2215 next_to_use_head = ring->next_to_use;
2217 ret = hns3_fill_skb_desc(ring, skb, &ring->desc[ring->next_to_use],
2219 if (unlikely(ret < 0))
2222 /* 'ret < 0' means filling error, 'ret == 0' means skb->len is
2223 * zero, which is unlikely, and 'ret > 0' means how many tx desc
2224 * need to be notified to the hw.
2226 ret = hns3_handle_desc_filling(ring, skb);
2227 if (unlikely(ret <= 0))
2230 pre_ntu = ring->next_to_use ? (ring->next_to_use - 1) :
2231 (ring->desc_num - 1);
2233 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP))
2234 hns3_tsyn(netdev, skb, &ring->desc[pre_ntu]);
2236 ring->desc[pre_ntu].tx.bdtp_fe_sc_vld_ra_ri |=
2237 cpu_to_le16(BIT(HNS3_TXD_FE_B));
2238 trace_hns3_tx_desc(ring, pre_ntu);
2240 skb_tx_timestamp(skb);
2242 /* Complete translate all packets */
2243 dev_queue = netdev_get_tx_queue(netdev, ring->queue_index);
2244 doorbell = __netdev_tx_sent_queue(dev_queue, desc_cb->send_bytes,
2245 netdev_xmit_more());
2246 hns3_tx_doorbell(ring, ret, doorbell);
2248 return NETDEV_TX_OK;
2251 hns3_clear_desc(ring, next_to_use_head);
2254 dev_kfree_skb_any(skb);
2255 hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
2256 return NETDEV_TX_OK;
2259 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
2261 struct hnae3_handle *h = hns3_get_handle(netdev);
2262 struct sockaddr *mac_addr = p;
2265 if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
2266 return -EADDRNOTAVAIL;
2268 if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
2269 netdev_info(netdev, "already using mac address %pM\n",
2274 /* For VF device, if there is a perm_addr, then the user will not
2275 * be allowed to change the address.
2277 if (!hns3_is_phys_func(h->pdev) &&
2278 !is_zero_ether_addr(netdev->perm_addr)) {
2279 netdev_err(netdev, "has permanent MAC %pM, user MAC %pM not allow\n",
2280 netdev->perm_addr, mac_addr->sa_data);
2284 ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
2286 netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
2290 ether_addr_copy(netdev->dev_addr, mac_addr->sa_data);
2295 static int hns3_nic_do_ioctl(struct net_device *netdev,
2296 struct ifreq *ifr, int cmd)
2298 struct hnae3_handle *h = hns3_get_handle(netdev);
2300 if (!netif_running(netdev))
2303 if (!h->ae_algo->ops->do_ioctl)
2306 return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
2309 static int hns3_nic_set_features(struct net_device *netdev,
2310 netdev_features_t features)
2312 netdev_features_t changed = netdev->features ^ features;
2313 struct hns3_nic_priv *priv = netdev_priv(netdev);
2314 struct hnae3_handle *h = priv->ae_handle;
2318 if (changed & (NETIF_F_GRO_HW) && h->ae_algo->ops->set_gro_en) {
2319 enable = !!(features & NETIF_F_GRO_HW);
2320 ret = h->ae_algo->ops->set_gro_en(h, enable);
2325 if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
2326 h->ae_algo->ops->enable_hw_strip_rxvtag) {
2327 enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
2328 ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, enable);
2333 if ((changed & NETIF_F_NTUPLE) && h->ae_algo->ops->enable_fd) {
2334 enable = !!(features & NETIF_F_NTUPLE);
2335 h->ae_algo->ops->enable_fd(h, enable);
2338 if ((netdev->features & NETIF_F_HW_TC) > (features & NETIF_F_HW_TC) &&
2339 h->ae_algo->ops->cls_flower_active(h)) {
2341 "there are offloaded TC filters active, cannot disable HW TC offload");
2345 if ((changed & NETIF_F_HW_VLAN_CTAG_FILTER) &&
2346 h->ae_algo->ops->enable_vlan_filter) {
2347 enable = !!(features & NETIF_F_HW_VLAN_CTAG_FILTER);
2348 ret = h->ae_algo->ops->enable_vlan_filter(h, enable);
2353 netdev->features = features;
2357 static netdev_features_t hns3_features_check(struct sk_buff *skb,
2358 struct net_device *dev,
2359 netdev_features_t features)
2361 #define HNS3_MAX_HDR_LEN 480U
2362 #define HNS3_MAX_L4_HDR_LEN 60U
2366 if (skb->ip_summed != CHECKSUM_PARTIAL)
2369 if (skb->encapsulation)
2370 len = skb_inner_transport_header(skb) - skb->data;
2372 len = skb_transport_header(skb) - skb->data;
2374 /* Assume L4 is 60 byte as TCP is the only protocol with a
2375 * a flexible value, and it's max len is 60 bytes.
2377 len += HNS3_MAX_L4_HDR_LEN;
2379 /* Hardware only supports checksum on the skb with a max header
2382 if (len > HNS3_MAX_HDR_LEN)
2383 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2388 static void hns3_nic_get_stats64(struct net_device *netdev,
2389 struct rtnl_link_stats64 *stats)
2391 struct hns3_nic_priv *priv = netdev_priv(netdev);
2392 int queue_num = priv->ae_handle->kinfo.num_tqps;
2393 struct hnae3_handle *handle = priv->ae_handle;
2394 struct hns3_enet_ring *ring;
2395 u64 rx_length_errors = 0;
2396 u64 rx_crc_errors = 0;
2397 u64 rx_multicast = 0;
2409 if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
2412 handle->ae_algo->ops->update_stats(handle, &netdev->stats);
2414 for (idx = 0; idx < queue_num; idx++) {
2415 /* fetch the tx stats */
2416 ring = &priv->ring[idx];
2418 start = u64_stats_fetch_begin_irq(&ring->syncp);
2419 tx_bytes += ring->stats.tx_bytes;
2420 tx_pkts += ring->stats.tx_pkts;
2421 tx_drop += ring->stats.sw_err_cnt;
2422 tx_drop += ring->stats.tx_vlan_err;
2423 tx_drop += ring->stats.tx_l4_proto_err;
2424 tx_drop += ring->stats.tx_l2l3l4_err;
2425 tx_drop += ring->stats.tx_tso_err;
2426 tx_drop += ring->stats.over_max_recursion;
2427 tx_drop += ring->stats.hw_limitation;
2428 tx_drop += ring->stats.copy_bits_err;
2429 tx_drop += ring->stats.skb2sgl_err;
2430 tx_drop += ring->stats.map_sg_err;
2431 tx_errors += ring->stats.sw_err_cnt;
2432 tx_errors += ring->stats.tx_vlan_err;
2433 tx_errors += ring->stats.tx_l4_proto_err;
2434 tx_errors += ring->stats.tx_l2l3l4_err;
2435 tx_errors += ring->stats.tx_tso_err;
2436 tx_errors += ring->stats.over_max_recursion;
2437 tx_errors += ring->stats.hw_limitation;
2438 tx_errors += ring->stats.copy_bits_err;
2439 tx_errors += ring->stats.skb2sgl_err;
2440 tx_errors += ring->stats.map_sg_err;
2441 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
2443 /* fetch the rx stats */
2444 ring = &priv->ring[idx + queue_num];
2446 start = u64_stats_fetch_begin_irq(&ring->syncp);
2447 rx_bytes += ring->stats.rx_bytes;
2448 rx_pkts += ring->stats.rx_pkts;
2449 rx_drop += ring->stats.l2_err;
2450 rx_errors += ring->stats.l2_err;
2451 rx_errors += ring->stats.l3l4_csum_err;
2452 rx_crc_errors += ring->stats.l2_err;
2453 rx_multicast += ring->stats.rx_multicast;
2454 rx_length_errors += ring->stats.err_pkt_len;
2455 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
2458 stats->tx_bytes = tx_bytes;
2459 stats->tx_packets = tx_pkts;
2460 stats->rx_bytes = rx_bytes;
2461 stats->rx_packets = rx_pkts;
2463 stats->rx_errors = rx_errors;
2464 stats->multicast = rx_multicast;
2465 stats->rx_length_errors = rx_length_errors;
2466 stats->rx_crc_errors = rx_crc_errors;
2467 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
2469 stats->tx_errors = tx_errors;
2470 stats->rx_dropped = rx_drop;
2471 stats->tx_dropped = tx_drop;
2472 stats->collisions = netdev->stats.collisions;
2473 stats->rx_over_errors = netdev->stats.rx_over_errors;
2474 stats->rx_frame_errors = netdev->stats.rx_frame_errors;
2475 stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
2476 stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
2477 stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
2478 stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
2479 stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
2480 stats->tx_window_errors = netdev->stats.tx_window_errors;
2481 stats->rx_compressed = netdev->stats.rx_compressed;
2482 stats->tx_compressed = netdev->stats.tx_compressed;
2485 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
2487 struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
2488 struct hnae3_knic_private_info *kinfo;
2489 u8 tc = mqprio_qopt->qopt.num_tc;
2490 u16 mode = mqprio_qopt->mode;
2491 u8 hw = mqprio_qopt->qopt.hw;
2492 struct hnae3_handle *h;
2494 if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
2495 mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
2498 if (tc > HNAE3_MAX_TC)
2504 h = hns3_get_handle(netdev);
2507 netif_dbg(h, drv, netdev, "setup tc: num_tc=%u\n", tc);
2509 return (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
2510 kinfo->dcb_ops->setup_tc(h, mqprio_qopt) : -EOPNOTSUPP;
2513 static int hns3_setup_tc_cls_flower(struct hns3_nic_priv *priv,
2514 struct flow_cls_offload *flow)
2516 int tc = tc_classid_to_hwtc(priv->netdev, flow->classid);
2517 struct hnae3_handle *h = hns3_get_handle(priv->netdev);
2519 switch (flow->command) {
2520 case FLOW_CLS_REPLACE:
2521 if (h->ae_algo->ops->add_cls_flower)
2522 return h->ae_algo->ops->add_cls_flower(h, flow, tc);
2524 case FLOW_CLS_DESTROY:
2525 if (h->ae_algo->ops->del_cls_flower)
2526 return h->ae_algo->ops->del_cls_flower(h, flow);
2535 static int hns3_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
2538 struct hns3_nic_priv *priv = cb_priv;
2540 if (!tc_cls_can_offload_and_chain0(priv->netdev, type_data))
2544 case TC_SETUP_CLSFLOWER:
2545 return hns3_setup_tc_cls_flower(priv, type_data);
2551 static LIST_HEAD(hns3_block_cb_list);
2553 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
2556 struct hns3_nic_priv *priv = netdev_priv(dev);
2560 case TC_SETUP_QDISC_MQPRIO:
2561 ret = hns3_setup_tc(dev, type_data);
2563 case TC_SETUP_BLOCK:
2564 ret = flow_block_cb_setup_simple(type_data,
2565 &hns3_block_cb_list,
2566 hns3_setup_tc_block_cb,
2576 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
2577 __be16 proto, u16 vid)
2579 struct hnae3_handle *h = hns3_get_handle(netdev);
2582 if (h->ae_algo->ops->set_vlan_filter)
2583 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
2588 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
2589 __be16 proto, u16 vid)
2591 struct hnae3_handle *h = hns3_get_handle(netdev);
2594 if (h->ae_algo->ops->set_vlan_filter)
2595 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
2600 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
2601 u8 qos, __be16 vlan_proto)
2603 struct hnae3_handle *h = hns3_get_handle(netdev);
2606 netif_dbg(h, drv, netdev,
2607 "set vf vlan: vf=%d, vlan=%u, qos=%u, vlan_proto=0x%x\n",
2608 vf, vlan, qos, ntohs(vlan_proto));
2610 if (h->ae_algo->ops->set_vf_vlan_filter)
2611 ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
2617 static int hns3_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
2619 struct hnae3_handle *handle = hns3_get_handle(netdev);
2621 if (hns3_nic_resetting(netdev))
2624 if (!handle->ae_algo->ops->set_vf_spoofchk)
2627 return handle->ae_algo->ops->set_vf_spoofchk(handle, vf, enable);
2630 static int hns3_set_vf_trust(struct net_device *netdev, int vf, bool enable)
2632 struct hnae3_handle *handle = hns3_get_handle(netdev);
2634 if (!handle->ae_algo->ops->set_vf_trust)
2637 return handle->ae_algo->ops->set_vf_trust(handle, vf, enable);
2640 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
2642 struct hnae3_handle *h = hns3_get_handle(netdev);
2645 if (hns3_nic_resetting(netdev))
2648 if (!h->ae_algo->ops->set_mtu)
2651 netif_dbg(h, drv, netdev,
2652 "change mtu from %u to %d\n", netdev->mtu, new_mtu);
2654 ret = h->ae_algo->ops->set_mtu(h, new_mtu);
2656 netdev_err(netdev, "failed to change MTU in hardware %d\n",
2659 netdev->mtu = new_mtu;
2664 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
2666 struct hns3_nic_priv *priv = netdev_priv(ndev);
2667 struct hnae3_handle *h = hns3_get_handle(ndev);
2668 struct hns3_enet_ring *tx_ring;
2669 struct napi_struct *napi;
2670 int timeout_queue = 0;
2671 int hw_head, hw_tail;
2672 int fbd_num, fbd_oft;
2673 int ebd_num, ebd_oft;
2678 /* Find the stopped queue the same way the stack does */
2679 for (i = 0; i < ndev->num_tx_queues; i++) {
2680 struct netdev_queue *q;
2681 unsigned long trans_start;
2683 q = netdev_get_tx_queue(ndev, i);
2684 trans_start = q->trans_start;
2685 if (netif_xmit_stopped(q) &&
2687 (trans_start + ndev->watchdog_timeo))) {
2689 netdev_info(ndev, "queue state: 0x%lx, delta msecs: %u\n",
2691 jiffies_to_msecs(jiffies - trans_start));
2696 if (i == ndev->num_tx_queues) {
2698 "no netdev TX timeout queue found, timeout count: %llu\n",
2699 priv->tx_timeout_count);
2703 priv->tx_timeout_count++;
2705 tx_ring = &priv->ring[timeout_queue];
2706 napi = &tx_ring->tqp_vector->napi;
2709 "tx_timeout count: %llu, queue id: %d, SW_NTU: 0x%x, SW_NTC: 0x%x, napi state: %lu\n",
2710 priv->tx_timeout_count, timeout_queue, tx_ring->next_to_use,
2711 tx_ring->next_to_clean, napi->state);
2714 "tx_pkts: %llu, tx_bytes: %llu, sw_err_cnt: %llu, tx_pending: %d\n",
2715 tx_ring->stats.tx_pkts, tx_ring->stats.tx_bytes,
2716 tx_ring->stats.sw_err_cnt, tx_ring->pending_buf);
2719 "seg_pkt_cnt: %llu, tx_more: %llu, restart_queue: %llu, tx_busy: %llu\n",
2720 tx_ring->stats.seg_pkt_cnt, tx_ring->stats.tx_more,
2721 tx_ring->stats.restart_queue, tx_ring->stats.tx_busy);
2723 /* When mac received many pause frames continuous, it's unable to send
2724 * packets, which may cause tx timeout
2726 if (h->ae_algo->ops->get_mac_stats) {
2727 struct hns3_mac_stats mac_stats;
2729 h->ae_algo->ops->get_mac_stats(h, &mac_stats);
2730 netdev_info(ndev, "tx_pause_cnt: %llu, rx_pause_cnt: %llu\n",
2731 mac_stats.tx_pause_cnt, mac_stats.rx_pause_cnt);
2734 hw_head = readl_relaxed(tx_ring->tqp->io_base +
2735 HNS3_RING_TX_RING_HEAD_REG);
2736 hw_tail = readl_relaxed(tx_ring->tqp->io_base +
2737 HNS3_RING_TX_RING_TAIL_REG);
2738 fbd_num = readl_relaxed(tx_ring->tqp->io_base +
2739 HNS3_RING_TX_RING_FBDNUM_REG);
2740 fbd_oft = readl_relaxed(tx_ring->tqp->io_base +
2741 HNS3_RING_TX_RING_OFFSET_REG);
2742 ebd_num = readl_relaxed(tx_ring->tqp->io_base +
2743 HNS3_RING_TX_RING_EBDNUM_REG);
2744 ebd_oft = readl_relaxed(tx_ring->tqp->io_base +
2745 HNS3_RING_TX_RING_EBD_OFFSET_REG);
2746 bd_num = readl_relaxed(tx_ring->tqp->io_base +
2747 HNS3_RING_TX_RING_BD_NUM_REG);
2748 bd_err = readl_relaxed(tx_ring->tqp->io_base +
2749 HNS3_RING_TX_RING_BD_ERR_REG);
2750 ring_en = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_EN_REG);
2751 tc = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_TX_RING_TC_REG);
2754 "BD_NUM: 0x%x HW_HEAD: 0x%x, HW_TAIL: 0x%x, BD_ERR: 0x%x, INT: 0x%x\n",
2755 bd_num, hw_head, hw_tail, bd_err,
2756 readl(tx_ring->tqp_vector->mask_addr));
2758 "RING_EN: 0x%x, TC: 0x%x, FBD_NUM: 0x%x FBD_OFT: 0x%x, EBD_NUM: 0x%x, EBD_OFT: 0x%x\n",
2759 ring_en, tc, fbd_num, fbd_oft, ebd_num, ebd_oft);
2764 static void hns3_nic_net_timeout(struct net_device *ndev, unsigned int txqueue)
2766 struct hns3_nic_priv *priv = netdev_priv(ndev);
2767 struct hnae3_handle *h = priv->ae_handle;
2769 if (!hns3_get_tx_timeo_queue_info(ndev))
2772 /* request the reset, and let the hclge to determine
2773 * which reset level should be done
2775 if (h->ae_algo->ops->reset_event)
2776 h->ae_algo->ops->reset_event(h->pdev, h);
2779 #ifdef CONFIG_RFS_ACCEL
2780 static int hns3_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
2781 u16 rxq_index, u32 flow_id)
2783 struct hnae3_handle *h = hns3_get_handle(dev);
2784 struct flow_keys fkeys;
2786 if (!h->ae_algo->ops->add_arfs_entry)
2789 if (skb->encapsulation)
2790 return -EPROTONOSUPPORT;
2792 if (!skb_flow_dissect_flow_keys(skb, &fkeys, 0))
2793 return -EPROTONOSUPPORT;
2795 if ((fkeys.basic.n_proto != htons(ETH_P_IP) &&
2796 fkeys.basic.n_proto != htons(ETH_P_IPV6)) ||
2797 (fkeys.basic.ip_proto != IPPROTO_TCP &&
2798 fkeys.basic.ip_proto != IPPROTO_UDP))
2799 return -EPROTONOSUPPORT;
2801 return h->ae_algo->ops->add_arfs_entry(h, rxq_index, flow_id, &fkeys);
2805 static int hns3_nic_get_vf_config(struct net_device *ndev, int vf,
2806 struct ifla_vf_info *ivf)
2808 struct hnae3_handle *h = hns3_get_handle(ndev);
2810 if (!h->ae_algo->ops->get_vf_config)
2813 return h->ae_algo->ops->get_vf_config(h, vf, ivf);
2816 static int hns3_nic_set_vf_link_state(struct net_device *ndev, int vf,
2819 struct hnae3_handle *h = hns3_get_handle(ndev);
2821 if (!h->ae_algo->ops->set_vf_link_state)
2824 return h->ae_algo->ops->set_vf_link_state(h, vf, link_state);
2827 static int hns3_nic_set_vf_rate(struct net_device *ndev, int vf,
2828 int min_tx_rate, int max_tx_rate)
2830 struct hnae3_handle *h = hns3_get_handle(ndev);
2832 if (!h->ae_algo->ops->set_vf_rate)
2835 return h->ae_algo->ops->set_vf_rate(h, vf, min_tx_rate, max_tx_rate,
2839 static int hns3_nic_set_vf_mac(struct net_device *netdev, int vf_id, u8 *mac)
2841 struct hnae3_handle *h = hns3_get_handle(netdev);
2843 if (!h->ae_algo->ops->set_vf_mac)
2846 if (is_multicast_ether_addr(mac)) {
2848 "Invalid MAC:%pM specified. Could not set MAC\n",
2853 return h->ae_algo->ops->set_vf_mac(h, vf_id, mac);
2856 static const struct net_device_ops hns3_nic_netdev_ops = {
2857 .ndo_open = hns3_nic_net_open,
2858 .ndo_stop = hns3_nic_net_stop,
2859 .ndo_start_xmit = hns3_nic_net_xmit,
2860 .ndo_tx_timeout = hns3_nic_net_timeout,
2861 .ndo_set_mac_address = hns3_nic_net_set_mac_address,
2862 .ndo_eth_ioctl = hns3_nic_do_ioctl,
2863 .ndo_change_mtu = hns3_nic_change_mtu,
2864 .ndo_set_features = hns3_nic_set_features,
2865 .ndo_features_check = hns3_features_check,
2866 .ndo_get_stats64 = hns3_nic_get_stats64,
2867 .ndo_setup_tc = hns3_nic_setup_tc,
2868 .ndo_set_rx_mode = hns3_nic_set_rx_mode,
2869 .ndo_vlan_rx_add_vid = hns3_vlan_rx_add_vid,
2870 .ndo_vlan_rx_kill_vid = hns3_vlan_rx_kill_vid,
2871 .ndo_set_vf_vlan = hns3_ndo_set_vf_vlan,
2872 .ndo_set_vf_spoofchk = hns3_set_vf_spoofchk,
2873 .ndo_set_vf_trust = hns3_set_vf_trust,
2874 #ifdef CONFIG_RFS_ACCEL
2875 .ndo_rx_flow_steer = hns3_rx_flow_steer,
2877 .ndo_get_vf_config = hns3_nic_get_vf_config,
2878 .ndo_set_vf_link_state = hns3_nic_set_vf_link_state,
2879 .ndo_set_vf_rate = hns3_nic_set_vf_rate,
2880 .ndo_set_vf_mac = hns3_nic_set_vf_mac,
2883 bool hns3_is_phys_func(struct pci_dev *pdev)
2885 u32 dev_id = pdev->device;
2888 case HNAE3_DEV_ID_GE:
2889 case HNAE3_DEV_ID_25GE:
2890 case HNAE3_DEV_ID_25GE_RDMA:
2891 case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
2892 case HNAE3_DEV_ID_50GE_RDMA:
2893 case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
2894 case HNAE3_DEV_ID_100G_RDMA_MACSEC:
2895 case HNAE3_DEV_ID_200G_RDMA:
2897 case HNAE3_DEV_ID_VF:
2898 case HNAE3_DEV_ID_RDMA_DCB_PFC_VF:
2901 dev_warn(&pdev->dev, "un-recognized pci device-id %u",
2908 static void hns3_disable_sriov(struct pci_dev *pdev)
2910 /* If our VFs are assigned we cannot shut down SR-IOV
2911 * without causing issues, so just leave the hardware
2912 * available but disabled
2914 if (pci_vfs_assigned(pdev)) {
2915 dev_warn(&pdev->dev,
2916 "disabling driver while VFs are assigned\n");
2920 pci_disable_sriov(pdev);
2923 /* hns3_probe - Device initialization routine
2924 * @pdev: PCI device information struct
2925 * @ent: entry in hns3_pci_tbl
2927 * hns3_probe initializes a PF identified by a pci_dev structure.
2928 * The OS initialization, configuring of the PF private structure,
2929 * and a hardware reset occur.
2931 * Returns 0 on success, negative on failure
2933 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2935 struct hnae3_ae_dev *ae_dev;
2938 ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev), GFP_KERNEL);
2942 ae_dev->pdev = pdev;
2943 ae_dev->flag = ent->driver_data;
2944 pci_set_drvdata(pdev, ae_dev);
2946 ret = hnae3_register_ae_dev(ae_dev);
2948 pci_set_drvdata(pdev, NULL);
2953 /* hns3_remove - Device removal routine
2954 * @pdev: PCI device information struct
2956 static void hns3_remove(struct pci_dev *pdev)
2958 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2960 if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
2961 hns3_disable_sriov(pdev);
2963 hnae3_unregister_ae_dev(ae_dev);
2964 pci_set_drvdata(pdev, NULL);
2968 * hns3_pci_sriov_configure
2969 * @pdev: pointer to a pci_dev structure
2970 * @num_vfs: number of VFs to allocate
2972 * Enable or change the number of VFs. Called when the user updates the number
2975 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
2979 if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
2980 dev_warn(&pdev->dev, "Can not config SRIOV\n");
2985 ret = pci_enable_sriov(pdev, num_vfs);
2987 dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
2990 } else if (!pci_vfs_assigned(pdev)) {
2991 pci_disable_sriov(pdev);
2993 dev_warn(&pdev->dev,
2994 "Unable to free VFs because some are assigned to VMs.\n");
3000 static void hns3_shutdown(struct pci_dev *pdev)
3002 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3004 hnae3_unregister_ae_dev(ae_dev);
3005 pci_set_drvdata(pdev, NULL);
3007 if (system_state == SYSTEM_POWER_OFF)
3008 pci_set_power_state(pdev, PCI_D3hot);
3011 static int __maybe_unused hns3_suspend(struct device *dev)
3013 struct hnae3_ae_dev *ae_dev = dev_get_drvdata(dev);
3015 if (ae_dev && hns3_is_phys_func(ae_dev->pdev)) {
3016 dev_info(dev, "Begin to suspend.\n");
3017 if (ae_dev->ops && ae_dev->ops->reset_prepare)
3018 ae_dev->ops->reset_prepare(ae_dev, HNAE3_FUNC_RESET);
3024 static int __maybe_unused hns3_resume(struct device *dev)
3026 struct hnae3_ae_dev *ae_dev = dev_get_drvdata(dev);
3028 if (ae_dev && hns3_is_phys_func(ae_dev->pdev)) {
3029 dev_info(dev, "Begin to resume.\n");
3030 if (ae_dev->ops && ae_dev->ops->reset_done)
3031 ae_dev->ops->reset_done(ae_dev);
3037 static pci_ers_result_t hns3_error_detected(struct pci_dev *pdev,
3038 pci_channel_state_t state)
3040 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3041 pci_ers_result_t ret;
3043 dev_info(&pdev->dev, "PCI error detected, state(=%u)!!\n", state);
3045 if (state == pci_channel_io_perm_failure)
3046 return PCI_ERS_RESULT_DISCONNECT;
3048 if (!ae_dev || !ae_dev->ops) {
3050 "Can't recover - error happened before device initialized\n");
3051 return PCI_ERS_RESULT_NONE;
3054 if (ae_dev->ops->handle_hw_ras_error)
3055 ret = ae_dev->ops->handle_hw_ras_error(ae_dev);
3057 return PCI_ERS_RESULT_NONE;
3062 static pci_ers_result_t hns3_slot_reset(struct pci_dev *pdev)
3064 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3065 const struct hnae3_ae_ops *ops;
3066 enum hnae3_reset_type reset_type;
3067 struct device *dev = &pdev->dev;
3069 if (!ae_dev || !ae_dev->ops)
3070 return PCI_ERS_RESULT_NONE;
3073 /* request the reset */
3074 if (ops->reset_event && ops->get_reset_level &&
3075 ops->set_default_reset_request) {
3076 if (ae_dev->hw_err_reset_req) {
3077 reset_type = ops->get_reset_level(ae_dev,
3078 &ae_dev->hw_err_reset_req);
3079 ops->set_default_reset_request(ae_dev, reset_type);
3080 dev_info(dev, "requesting reset due to PCI error\n");
3081 ops->reset_event(pdev, NULL);
3084 return PCI_ERS_RESULT_RECOVERED;
3087 return PCI_ERS_RESULT_DISCONNECT;
3090 static void hns3_reset_prepare(struct pci_dev *pdev)
3092 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3094 dev_info(&pdev->dev, "FLR prepare\n");
3095 if (ae_dev && ae_dev->ops && ae_dev->ops->reset_prepare)
3096 ae_dev->ops->reset_prepare(ae_dev, HNAE3_FLR_RESET);
3099 static void hns3_reset_done(struct pci_dev *pdev)
3101 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3103 dev_info(&pdev->dev, "FLR done\n");
3104 if (ae_dev && ae_dev->ops && ae_dev->ops->reset_done)
3105 ae_dev->ops->reset_done(ae_dev);
3108 static const struct pci_error_handlers hns3_err_handler = {
3109 .error_detected = hns3_error_detected,
3110 .slot_reset = hns3_slot_reset,
3111 .reset_prepare = hns3_reset_prepare,
3112 .reset_done = hns3_reset_done,
3115 static SIMPLE_DEV_PM_OPS(hns3_pm_ops, hns3_suspend, hns3_resume);
3117 static struct pci_driver hns3_driver = {
3118 .name = hns3_driver_name,
3119 .id_table = hns3_pci_tbl,
3120 .probe = hns3_probe,
3121 .remove = hns3_remove,
3122 .shutdown = hns3_shutdown,
3123 .driver.pm = &hns3_pm_ops,
3124 .sriov_configure = hns3_pci_sriov_configure,
3125 .err_handler = &hns3_err_handler,
3128 /* set default feature to hns3 */
3129 static void hns3_set_default_feature(struct net_device *netdev)
3131 struct hnae3_handle *h = hns3_get_handle(netdev);
3132 struct pci_dev *pdev = h->pdev;
3133 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3135 netdev->priv_flags |= IFF_UNICAST_FLT;
3137 netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
3139 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
3140 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3141 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
3142 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
3143 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
3144 NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
3146 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V2) {
3147 netdev->features |= NETIF_F_GRO_HW;
3149 if (!(h->flags & HNAE3_SUPPORT_VF))
3150 netdev->features |= NETIF_F_NTUPLE;
3153 if (test_bit(HNAE3_DEV_SUPPORT_UDP_GSO_B, ae_dev->caps))
3154 netdev->features |= NETIF_F_GSO_UDP_L4;
3156 if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps))
3157 netdev->features |= NETIF_F_HW_CSUM;
3159 netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3161 if (test_bit(HNAE3_DEV_SUPPORT_UDP_TUNNEL_CSUM_B, ae_dev->caps))
3162 netdev->features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
3164 if (test_bit(HNAE3_DEV_SUPPORT_FD_FORWARD_TC_B, ae_dev->caps))
3165 netdev->features |= NETIF_F_HW_TC;
3167 netdev->hw_features |= netdev->features;
3168 if (!test_bit(HNAE3_DEV_SUPPORT_VLAN_FLTR_MDF_B, ae_dev->caps))
3169 netdev->hw_features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
3171 netdev->vlan_features |= netdev->features &
3172 ~(NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_CTAG_TX |
3173 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_GRO_HW | NETIF_F_NTUPLE |
3176 netdev->hw_enc_features |= netdev->vlan_features | NETIF_F_TSO_MANGLEID;
3179 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
3180 struct hns3_desc_cb *cb)
3182 unsigned int order = hns3_page_order(ring);
3185 if (ring->page_pool) {
3186 p = page_pool_dev_alloc_frag(ring->page_pool,
3188 hns3_buf_size(ring));
3193 cb->buf = page_address(p);
3194 cb->dma = page_pool_get_dma_addr(p);
3195 cb->type = DESC_TYPE_PP_FRAG;
3200 p = dev_alloc_pages(order);
3205 cb->page_offset = 0;
3207 cb->buf = page_address(p);
3208 cb->length = hns3_page_size(ring);
3209 cb->type = DESC_TYPE_PAGE;
3210 page_ref_add(p, USHRT_MAX - 1);
3211 cb->pagecnt_bias = USHRT_MAX;
3216 static void hns3_free_buffer(struct hns3_enet_ring *ring,
3217 struct hns3_desc_cb *cb, int budget)
3219 if (cb->type & (DESC_TYPE_SKB | DESC_TYPE_BOUNCE_HEAD |
3220 DESC_TYPE_BOUNCE_ALL | DESC_TYPE_SGL_SKB))
3221 napi_consume_skb(cb->priv, budget);
3222 else if (!HNAE3_IS_TX_RING(ring)) {
3223 if (cb->type & DESC_TYPE_PAGE && cb->pagecnt_bias)
3224 __page_frag_cache_drain(cb->priv, cb->pagecnt_bias);
3225 else if (cb->type & DESC_TYPE_PP_FRAG)
3226 page_pool_put_full_page(ring->page_pool, cb->priv,
3229 memset(cb, 0, sizeof(*cb));
3232 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
3234 cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
3235 cb->length, ring_to_dma_dir(ring));
3237 if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
3243 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
3244 struct hns3_desc_cb *cb)
3246 if (cb->type & (DESC_TYPE_SKB | DESC_TYPE_FRAGLIST_SKB))
3247 dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
3248 ring_to_dma_dir(ring));
3249 else if ((cb->type & DESC_TYPE_PAGE) && cb->length)
3250 dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
3251 ring_to_dma_dir(ring));
3252 else if (cb->type & (DESC_TYPE_BOUNCE_ALL | DESC_TYPE_BOUNCE_HEAD |
3254 hns3_tx_spare_reclaim_cb(ring, cb);
3257 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
3259 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
3260 ring->desc[i].addr = 0;
3263 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i,
3266 struct hns3_desc_cb *cb = &ring->desc_cb[i];
3268 if (!ring->desc_cb[i].dma)
3271 hns3_buffer_detach(ring, i);
3272 hns3_free_buffer(ring, cb, budget);
3275 static void hns3_free_buffers(struct hns3_enet_ring *ring)
3279 for (i = 0; i < ring->desc_num; i++)
3280 hns3_free_buffer_detach(ring, i, 0);
3283 /* free desc along with its attached buffer */
3284 static void hns3_free_desc(struct hns3_enet_ring *ring)
3286 int size = ring->desc_num * sizeof(ring->desc[0]);
3288 hns3_free_buffers(ring);
3291 dma_free_coherent(ring_to_dev(ring), size,
3292 ring->desc, ring->desc_dma_addr);
3297 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
3299 int size = ring->desc_num * sizeof(ring->desc[0]);
3301 ring->desc = dma_alloc_coherent(ring_to_dev(ring), size,
3302 &ring->desc_dma_addr, GFP_KERNEL);
3309 static int hns3_alloc_and_map_buffer(struct hns3_enet_ring *ring,
3310 struct hns3_desc_cb *cb)
3314 ret = hns3_alloc_buffer(ring, cb);
3315 if (ret || ring->page_pool)
3318 ret = hns3_map_buffer(ring, cb);
3325 hns3_free_buffer(ring, cb, 0);
3330 static int hns3_alloc_and_attach_buffer(struct hns3_enet_ring *ring, int i)
3332 int ret = hns3_alloc_and_map_buffer(ring, &ring->desc_cb[i]);
3337 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3338 ring->desc_cb[i].page_offset);
3343 /* Allocate memory for raw pkg, and map with dma */
3344 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
3348 for (i = 0; i < ring->desc_num; i++) {
3349 ret = hns3_alloc_and_attach_buffer(ring, i);
3351 goto out_buffer_fail;
3357 for (j = i - 1; j >= 0; j--)
3358 hns3_free_buffer_detach(ring, j, 0);
3362 /* detach a in-used buffer and replace with a reserved one */
3363 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
3364 struct hns3_desc_cb *res_cb)
3366 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
3367 ring->desc_cb[i] = *res_cb;
3368 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3369 ring->desc_cb[i].page_offset);
3370 ring->desc[i].rx.bd_base_info = 0;
3373 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
3375 ring->desc_cb[i].reuse_flag = 0;
3376 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3377 ring->desc_cb[i].page_offset);
3378 ring->desc[i].rx.bd_base_info = 0;
3380 dma_sync_single_for_device(ring_to_dev(ring),
3381 ring->desc_cb[i].dma + ring->desc_cb[i].page_offset,
3382 hns3_buf_size(ring),
3386 static bool hns3_nic_reclaim_desc(struct hns3_enet_ring *ring,
3387 int *bytes, int *pkts, int budget)
3389 /* pair with ring->last_to_use update in hns3_tx_doorbell(),
3390 * smp_store_release() is not used in hns3_tx_doorbell() because
3391 * the doorbell operation already have the needed barrier operation.
3393 int ltu = smp_load_acquire(&ring->last_to_use);
3394 int ntc = ring->next_to_clean;
3395 struct hns3_desc_cb *desc_cb;
3396 bool reclaimed = false;
3397 struct hns3_desc *desc;
3399 while (ltu != ntc) {
3400 desc = &ring->desc[ntc];
3402 if (le16_to_cpu(desc->tx.bdtp_fe_sc_vld_ra_ri) &
3403 BIT(HNS3_TXD_VLD_B))
3406 desc_cb = &ring->desc_cb[ntc];
3408 if (desc_cb->type & (DESC_TYPE_SKB | DESC_TYPE_BOUNCE_ALL |
3409 DESC_TYPE_BOUNCE_HEAD |
3410 DESC_TYPE_SGL_SKB)) {
3412 (*bytes) += desc_cb->send_bytes;
3415 /* desc_cb will be cleaned, after hnae3_free_buffer_detach */
3416 hns3_free_buffer_detach(ring, ntc, budget);
3418 if (++ntc == ring->desc_num)
3421 /* Issue prefetch for next Tx descriptor */
3422 prefetch(&ring->desc_cb[ntc]);
3426 if (unlikely(!reclaimed))
3429 /* This smp_store_release() pairs with smp_load_acquire() in
3430 * ring_space called by hns3_nic_net_xmit.
3432 smp_store_release(&ring->next_to_clean, ntc);
3434 hns3_tx_spare_update(ring);
3439 void hns3_clean_tx_ring(struct hns3_enet_ring *ring, int budget)
3441 struct net_device *netdev = ring_to_netdev(ring);
3442 struct hns3_nic_priv *priv = netdev_priv(netdev);
3443 struct netdev_queue *dev_queue;
3449 if (unlikely(!hns3_nic_reclaim_desc(ring, &bytes, &pkts, budget)))
3452 ring->tqp_vector->tx_group.total_bytes += bytes;
3453 ring->tqp_vector->tx_group.total_packets += pkts;
3455 u64_stats_update_begin(&ring->syncp);
3456 ring->stats.tx_bytes += bytes;
3457 ring->stats.tx_pkts += pkts;
3458 u64_stats_update_end(&ring->syncp);
3460 dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
3461 netdev_tx_completed_queue(dev_queue, pkts, bytes);
3463 if (unlikely(netif_carrier_ok(netdev) &&
3464 ring_space(ring) > HNS3_MAX_TSO_BD_NUM)) {
3465 /* Make sure that anybody stopping the queue after this
3466 * sees the new next_to_clean.
3469 if (netif_tx_queue_stopped(dev_queue) &&
3470 !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
3471 netif_tx_wake_queue(dev_queue);
3472 ring->stats.restart_queue++;
3477 static int hns3_desc_unused(struct hns3_enet_ring *ring)
3479 int ntc = ring->next_to_clean;
3480 int ntu = ring->next_to_use;
3482 return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
3485 static void hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring,
3488 struct hns3_desc_cb *desc_cb;
3489 struct hns3_desc_cb res_cbs;
3492 for (i = 0; i < cleand_count; i++) {
3493 desc_cb = &ring->desc_cb[ring->next_to_use];
3494 if (desc_cb->reuse_flag) {
3495 u64_stats_update_begin(&ring->syncp);
3496 ring->stats.reuse_pg_cnt++;
3497 u64_stats_update_end(&ring->syncp);
3499 hns3_reuse_buffer(ring, ring->next_to_use);
3501 ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
3503 u64_stats_update_begin(&ring->syncp);
3504 ring->stats.sw_err_cnt++;
3505 u64_stats_update_end(&ring->syncp);
3507 hns3_rl_err(ring_to_netdev(ring),
3508 "alloc rx buffer failed: %d\n",
3512 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
3514 u64_stats_update_begin(&ring->syncp);
3515 ring->stats.non_reuse_pg++;
3516 u64_stats_update_end(&ring->syncp);
3519 ring_ptr_move_fw(ring, next_to_use);
3522 writel(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
3525 static bool hns3_can_reuse_page(struct hns3_desc_cb *cb)
3527 return page_count(cb->priv) == cb->pagecnt_bias;
3530 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
3531 struct hns3_enet_ring *ring, int pull_len,
3532 struct hns3_desc_cb *desc_cb)
3534 struct hns3_desc *desc = &ring->desc[ring->next_to_clean];
3535 u32 frag_offset = desc_cb->page_offset + pull_len;
3536 int size = le16_to_cpu(desc->rx.size);
3537 u32 truesize = hns3_buf_size(ring);
3538 u32 frag_size = size - pull_len;
3541 if (ring->page_pool) {
3542 skb_add_rx_frag(skb, i, desc_cb->priv, frag_offset,
3543 frag_size, truesize);
3547 /* Avoid re-using remote or pfmem page */
3548 if (unlikely(!dev_page_is_reusable(desc_cb->priv)))
3551 reused = hns3_can_reuse_page(desc_cb);
3553 /* Rx page can be reused when:
3554 * 1. Rx page is only owned by the driver when page_offset
3555 * is zero, which means 0 @ truesize will be used by
3556 * stack after skb_add_rx_frag() is called, and the rest
3557 * of rx page can be reused by driver.
3559 * 2. Rx page is only owned by the driver when page_offset
3560 * is non-zero, which means page_offset @ truesize will
3561 * be used by stack after skb_add_rx_frag() is called,
3562 * and 0 @ truesize can be reused by driver.
3564 if ((!desc_cb->page_offset && reused) ||
3565 ((desc_cb->page_offset + truesize + truesize) <=
3566 hns3_page_size(ring) && desc_cb->page_offset)) {
3567 desc_cb->page_offset += truesize;
3568 desc_cb->reuse_flag = 1;
3569 } else if (desc_cb->page_offset && reused) {
3570 desc_cb->page_offset = 0;
3571 desc_cb->reuse_flag = 1;
3572 } else if (frag_size <= ring->rx_copybreak) {
3573 void *frag = napi_alloc_frag(frag_size);
3575 if (unlikely(!frag)) {
3576 u64_stats_update_begin(&ring->syncp);
3577 ring->stats.frag_alloc_err++;
3578 u64_stats_update_end(&ring->syncp);
3580 hns3_rl_err(ring_to_netdev(ring),
3581 "failed to allocate rx frag\n");
3585 desc_cb->reuse_flag = 1;
3586 memcpy(frag, desc_cb->buf + frag_offset, frag_size);
3587 skb_add_rx_frag(skb, i, virt_to_page(frag),
3588 offset_in_page(frag), frag_size, frag_size);
3590 u64_stats_update_begin(&ring->syncp);
3591 ring->stats.frag_alloc++;
3592 u64_stats_update_end(&ring->syncp);
3597 desc_cb->pagecnt_bias--;
3599 if (unlikely(!desc_cb->pagecnt_bias)) {
3600 page_ref_add(desc_cb->priv, USHRT_MAX);
3601 desc_cb->pagecnt_bias = USHRT_MAX;
3604 skb_add_rx_frag(skb, i, desc_cb->priv, frag_offset,
3605 frag_size, truesize);
3607 if (unlikely(!desc_cb->reuse_flag))
3608 __page_frag_cache_drain(desc_cb->priv, desc_cb->pagecnt_bias);
3611 static int hns3_gro_complete(struct sk_buff *skb, u32 l234info)
3613 __be16 type = skb->protocol;
3617 while (eth_type_vlan(type)) {
3618 struct vlan_hdr *vh;
3620 if ((depth + VLAN_HLEN) > skb_headlen(skb))
3623 vh = (struct vlan_hdr *)(skb->data + depth);
3624 type = vh->h_vlan_encapsulated_proto;
3628 skb_set_network_header(skb, depth);
3630 if (type == htons(ETH_P_IP)) {
3631 const struct iphdr *iph = ip_hdr(skb);
3633 depth += sizeof(struct iphdr);
3634 skb_set_transport_header(skb, depth);
3636 th->check = ~tcp_v4_check(skb->len - depth, iph->saddr,
3638 } else if (type == htons(ETH_P_IPV6)) {
3639 const struct ipv6hdr *iph = ipv6_hdr(skb);
3641 depth += sizeof(struct ipv6hdr);
3642 skb_set_transport_header(skb, depth);
3644 th->check = ~tcp_v6_check(skb->len - depth, &iph->saddr,
3647 hns3_rl_err(skb->dev,
3648 "Error: FW GRO supports only IPv4/IPv6, not 0x%04x, depth: %d\n",
3649 be16_to_cpu(type), depth);
3653 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
3655 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
3657 if (l234info & BIT(HNS3_RXD_GRO_FIXID_B))
3658 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_FIXEDID;
3660 skb->csum_start = (unsigned char *)th - skb->head;
3661 skb->csum_offset = offsetof(struct tcphdr, check);
3662 skb->ip_summed = CHECKSUM_PARTIAL;
3664 trace_hns3_gro(skb);
3669 static bool hns3_checksum_complete(struct hns3_enet_ring *ring,
3670 struct sk_buff *skb, u32 ptype, u16 csum)
3672 if (ptype == HNS3_INVALID_PTYPE ||
3673 hns3_rx_ptype_tbl[ptype].ip_summed != CHECKSUM_COMPLETE)
3676 u64_stats_update_begin(&ring->syncp);
3677 ring->stats.csum_complete++;
3678 u64_stats_update_end(&ring->syncp);
3679 skb->ip_summed = CHECKSUM_COMPLETE;
3680 skb->csum = csum_unfold((__force __sum16)csum);
3685 static void hns3_rx_handle_csum(struct sk_buff *skb, u32 l234info,
3686 u32 ol_info, u32 ptype)
3688 int l3_type, l4_type;
3691 if (ptype != HNS3_INVALID_PTYPE) {
3692 skb->csum_level = hns3_rx_ptype_tbl[ptype].csum_level;
3693 skb->ip_summed = hns3_rx_ptype_tbl[ptype].ip_summed;
3698 ol4_type = hnae3_get_field(ol_info, HNS3_RXD_OL4ID_M,
3701 case HNS3_OL4_TYPE_MAC_IN_UDP:
3702 case HNS3_OL4_TYPE_NVGRE:
3703 skb->csum_level = 1;
3705 case HNS3_OL4_TYPE_NO_TUN:
3706 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
3708 l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
3710 /* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
3711 if ((l3_type == HNS3_L3_TYPE_IPV4 ||
3712 l3_type == HNS3_L3_TYPE_IPV6) &&
3713 (l4_type == HNS3_L4_TYPE_UDP ||
3714 l4_type == HNS3_L4_TYPE_TCP ||
3715 l4_type == HNS3_L4_TYPE_SCTP))
3716 skb->ip_summed = CHECKSUM_UNNECESSARY;
3723 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
3724 u32 l234info, u32 bd_base_info, u32 ol_info,
3727 struct net_device *netdev = ring_to_netdev(ring);
3728 struct hns3_nic_priv *priv = netdev_priv(netdev);
3729 u32 ptype = HNS3_INVALID_PTYPE;
3731 skb->ip_summed = CHECKSUM_NONE;
3733 skb_checksum_none_assert(skb);
3735 if (!(netdev->features & NETIF_F_RXCSUM))
3738 if (test_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state))
3739 ptype = hnae3_get_field(ol_info, HNS3_RXD_PTYPE_M,
3742 if (hns3_checksum_complete(ring, skb, ptype, csum))
3745 /* check if hardware has done checksum */
3746 if (!(bd_base_info & BIT(HNS3_RXD_L3L4P_B)))
3749 if (unlikely(l234info & (BIT(HNS3_RXD_L3E_B) | BIT(HNS3_RXD_L4E_B) |
3750 BIT(HNS3_RXD_OL3E_B) |
3751 BIT(HNS3_RXD_OL4E_B)))) {
3752 u64_stats_update_begin(&ring->syncp);
3753 ring->stats.l3l4_csum_err++;
3754 u64_stats_update_end(&ring->syncp);
3759 hns3_rx_handle_csum(skb, l234info, ol_info, ptype);
3762 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
3764 if (skb_has_frag_list(skb))
3765 napi_gro_flush(&ring->tqp_vector->napi, false);
3767 napi_gro_receive(&ring->tqp_vector->napi, skb);
3770 static bool hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
3771 struct hns3_desc *desc, u32 l234info,
3774 struct hnae3_handle *handle = ring->tqp->handle;
3775 struct pci_dev *pdev = ring->tqp->handle->pdev;
3776 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3778 if (unlikely(ae_dev->dev_version < HNAE3_DEVICE_VERSION_V2)) {
3779 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3780 if (!(*vlan_tag & VLAN_VID_MASK))
3781 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3783 return (*vlan_tag != 0);
3786 #define HNS3_STRP_OUTER_VLAN 0x1
3787 #define HNS3_STRP_INNER_VLAN 0x2
3788 #define HNS3_STRP_BOTH 0x3
3790 /* Hardware always insert VLAN tag into RX descriptor when
3791 * remove the tag from packet, driver needs to determine
3792 * reporting which tag to stack.
3794 switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
3795 HNS3_RXD_STRP_TAGP_S)) {
3796 case HNS3_STRP_OUTER_VLAN:
3797 if (handle->port_base_vlan_state !=
3798 HNAE3_PORT_BASE_VLAN_DISABLE)
3801 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3803 case HNS3_STRP_INNER_VLAN:
3804 if (handle->port_base_vlan_state !=
3805 HNAE3_PORT_BASE_VLAN_DISABLE)
3808 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3810 case HNS3_STRP_BOTH:
3811 if (handle->port_base_vlan_state ==
3812 HNAE3_PORT_BASE_VLAN_DISABLE)
3813 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3815 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3823 static void hns3_rx_ring_move_fw(struct hns3_enet_ring *ring)
3825 ring->desc[ring->next_to_clean].rx.bd_base_info &=
3826 cpu_to_le32(~BIT(HNS3_RXD_VLD_B));
3827 ring->next_to_clean += 1;
3829 if (unlikely(ring->next_to_clean == ring->desc_num))
3830 ring->next_to_clean = 0;
3833 static int hns3_alloc_skb(struct hns3_enet_ring *ring, unsigned int length,
3836 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
3837 struct net_device *netdev = ring_to_netdev(ring);
3838 struct sk_buff *skb;
3840 ring->skb = napi_alloc_skb(&ring->tqp_vector->napi, HNS3_RX_HEAD_SIZE);
3842 if (unlikely(!skb)) {
3843 hns3_rl_err(netdev, "alloc rx skb fail\n");
3845 u64_stats_update_begin(&ring->syncp);
3846 ring->stats.sw_err_cnt++;
3847 u64_stats_update_end(&ring->syncp);
3852 trace_hns3_rx_desc(ring);
3853 prefetchw(skb->data);
3855 ring->pending_buf = 1;
3857 ring->tail_skb = NULL;
3858 if (length <= HNS3_RX_HEAD_SIZE) {
3859 memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
3861 /* We can reuse buffer as-is, just make sure it is reusable */
3862 if (dev_page_is_reusable(desc_cb->priv))
3863 desc_cb->reuse_flag = 1;
3864 else if (desc_cb->type & DESC_TYPE_PP_FRAG)
3865 page_pool_put_full_page(ring->page_pool, desc_cb->priv,
3867 else /* This page cannot be reused so discard it */
3868 __page_frag_cache_drain(desc_cb->priv,
3869 desc_cb->pagecnt_bias);
3871 hns3_rx_ring_move_fw(ring);
3875 if (ring->page_pool)
3876 skb_mark_for_recycle(skb);
3878 u64_stats_update_begin(&ring->syncp);
3879 ring->stats.seg_pkt_cnt++;
3880 u64_stats_update_end(&ring->syncp);
3882 ring->pull_len = eth_get_headlen(netdev, va, HNS3_RX_HEAD_SIZE);
3883 __skb_put(skb, ring->pull_len);
3884 hns3_nic_reuse_page(skb, ring->frag_num++, ring, ring->pull_len,
3886 hns3_rx_ring_move_fw(ring);
3891 static int hns3_add_frag(struct hns3_enet_ring *ring)
3893 struct sk_buff *skb = ring->skb;
3894 struct sk_buff *head_skb = skb;
3895 struct sk_buff *new_skb;
3896 struct hns3_desc_cb *desc_cb;
3897 struct hns3_desc *desc;
3901 desc = &ring->desc[ring->next_to_clean];
3902 desc_cb = &ring->desc_cb[ring->next_to_clean];
3903 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
3904 /* make sure HW write desc complete */
3906 if (!(bd_base_info & BIT(HNS3_RXD_VLD_B)))
3909 if (unlikely(ring->frag_num >= MAX_SKB_FRAGS)) {
3910 new_skb = napi_alloc_skb(&ring->tqp_vector->napi, 0);
3911 if (unlikely(!new_skb)) {
3912 hns3_rl_err(ring_to_netdev(ring),
3913 "alloc rx fraglist skb fail\n");
3917 if (ring->page_pool)
3918 skb_mark_for_recycle(new_skb);
3922 if (ring->tail_skb) {
3923 ring->tail_skb->next = new_skb;
3924 ring->tail_skb = new_skb;
3926 skb_shinfo(skb)->frag_list = new_skb;
3927 ring->tail_skb = new_skb;
3931 if (ring->tail_skb) {
3932 head_skb->truesize += hns3_buf_size(ring);
3933 head_skb->data_len += le16_to_cpu(desc->rx.size);
3934 head_skb->len += le16_to_cpu(desc->rx.size);
3935 skb = ring->tail_skb;
3938 dma_sync_single_for_cpu(ring_to_dev(ring),
3939 desc_cb->dma + desc_cb->page_offset,
3940 hns3_buf_size(ring),
3943 hns3_nic_reuse_page(skb, ring->frag_num++, ring, 0, desc_cb);
3944 trace_hns3_rx_desc(ring);
3945 hns3_rx_ring_move_fw(ring);
3946 ring->pending_buf++;
3947 } while (!(bd_base_info & BIT(HNS3_RXD_FE_B)));
3952 static int hns3_set_gro_and_checksum(struct hns3_enet_ring *ring,
3953 struct sk_buff *skb, u32 l234info,
3954 u32 bd_base_info, u32 ol_info, u16 csum)
3956 struct net_device *netdev = ring_to_netdev(ring);
3957 struct hns3_nic_priv *priv = netdev_priv(netdev);
3960 skb_shinfo(skb)->gso_size = hnae3_get_field(bd_base_info,
3961 HNS3_RXD_GRO_SIZE_M,
3962 HNS3_RXD_GRO_SIZE_S);
3963 /* if there is no HW GRO, do not set gro params */
3964 if (!skb_shinfo(skb)->gso_size) {
3965 hns3_rx_checksum(ring, skb, l234info, bd_base_info, ol_info,
3970 NAPI_GRO_CB(skb)->count = hnae3_get_field(l234info,
3971 HNS3_RXD_GRO_COUNT_M,
3972 HNS3_RXD_GRO_COUNT_S);
3974 if (test_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state)) {
3975 u32 ptype = hnae3_get_field(ol_info, HNS3_RXD_PTYPE_M,
3978 l3_type = hns3_rx_ptype_tbl[ptype].l3_type;
3980 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
3984 if (l3_type == HNS3_L3_TYPE_IPV4)
3985 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
3986 else if (l3_type == HNS3_L3_TYPE_IPV6)
3987 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
3991 return hns3_gro_complete(skb, l234info);
3994 static void hns3_set_rx_skb_rss_type(struct hns3_enet_ring *ring,
3995 struct sk_buff *skb, u32 rss_hash)
3997 struct hnae3_handle *handle = ring->tqp->handle;
3998 enum pkt_hash_types rss_type;
4001 rss_type = handle->kinfo.rss_type;
4003 rss_type = PKT_HASH_TYPE_NONE;
4005 skb_set_hash(skb, rss_hash, rss_type);
4008 static int hns3_handle_bdinfo(struct hns3_enet_ring *ring, struct sk_buff *skb)
4010 struct net_device *netdev = ring_to_netdev(ring);
4011 enum hns3_pkt_l2t_type l2_frame_type;
4012 u32 bd_base_info, l234info, ol_info;
4013 struct hns3_desc *desc;
4018 /* bdinfo handled below is only valid on the last BD of the
4019 * current packet, and ring->next_to_clean indicates the first
4020 * descriptor of next packet, so need - 1 below.
4022 pre_ntc = ring->next_to_clean ? (ring->next_to_clean - 1) :
4023 (ring->desc_num - 1);
4024 desc = &ring->desc[pre_ntc];
4025 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
4026 l234info = le32_to_cpu(desc->rx.l234_info);
4027 ol_info = le32_to_cpu(desc->rx.ol_info);
4028 csum = le16_to_cpu(desc->csum);
4030 if (unlikely(bd_base_info & BIT(HNS3_RXD_TS_VLD_B))) {
4031 struct hnae3_handle *h = hns3_get_handle(netdev);
4032 u32 nsec = le32_to_cpu(desc->ts_nsec);
4033 u32 sec = le32_to_cpu(desc->ts_sec);
4035 if (h->ae_algo->ops->get_rx_hwts)
4036 h->ae_algo->ops->get_rx_hwts(h, skb, nsec, sec);
4039 /* Based on hw strategy, the tag offloaded will be stored at
4040 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
4041 * in one layer tag case.
4043 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
4046 if (hns3_parse_vlan_tag(ring, desc, l234info, &vlan_tag))
4047 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
4051 if (unlikely(!desc->rx.pkt_len || (l234info & (BIT(HNS3_RXD_TRUNCAT_B) |
4052 BIT(HNS3_RXD_L2E_B))))) {
4053 u64_stats_update_begin(&ring->syncp);
4054 if (l234info & BIT(HNS3_RXD_L2E_B))
4055 ring->stats.l2_err++;
4057 ring->stats.err_pkt_len++;
4058 u64_stats_update_end(&ring->syncp);
4065 /* Do update ip stack process */
4066 skb->protocol = eth_type_trans(skb, netdev);
4068 /* This is needed in order to enable forwarding support */
4069 ret = hns3_set_gro_and_checksum(ring, skb, l234info,
4070 bd_base_info, ol_info, csum);
4071 if (unlikely(ret)) {
4072 u64_stats_update_begin(&ring->syncp);
4073 ring->stats.rx_err_cnt++;
4074 u64_stats_update_end(&ring->syncp);
4078 l2_frame_type = hnae3_get_field(l234info, HNS3_RXD_DMAC_M,
4081 u64_stats_update_begin(&ring->syncp);
4082 ring->stats.rx_pkts++;
4083 ring->stats.rx_bytes += len;
4085 if (l2_frame_type == HNS3_L2_TYPE_MULTICAST)
4086 ring->stats.rx_multicast++;
4088 u64_stats_update_end(&ring->syncp);
4090 ring->tqp_vector->rx_group.total_bytes += len;
4092 hns3_set_rx_skb_rss_type(ring, skb, le32_to_cpu(desc->rx.rss_hash));
4096 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring)
4098 struct sk_buff *skb = ring->skb;
4099 struct hns3_desc_cb *desc_cb;
4100 struct hns3_desc *desc;
4101 unsigned int length;
4105 desc = &ring->desc[ring->next_to_clean];
4106 desc_cb = &ring->desc_cb[ring->next_to_clean];
4111 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
4112 /* Check valid BD */
4113 if (unlikely(!(bd_base_info & BIT(HNS3_RXD_VLD_B))))
4117 length = le16_to_cpu(desc->rx.size);
4119 ring->va = desc_cb->buf + desc_cb->page_offset;
4121 dma_sync_single_for_cpu(ring_to_dev(ring),
4122 desc_cb->dma + desc_cb->page_offset,
4123 hns3_buf_size(ring),
4126 /* Prefetch first cache line of first page.
4127 * Idea is to cache few bytes of the header of the packet.
4128 * Our L1 Cache line size is 64B so need to prefetch twice to make
4129 * it 128B. But in actual we can have greater size of caches with
4130 * 128B Level 1 cache lines. In such a case, single fetch would
4131 * suffice to cache in the relevant part of the header.
4133 net_prefetch(ring->va);
4135 ret = hns3_alloc_skb(ring, length, ring->va);
4138 if (ret < 0) /* alloc buffer fail */
4140 if (!(bd_base_info & BIT(HNS3_RXD_FE_B))) { /* need add frag */
4141 ret = hns3_add_frag(ring);
4146 ret = hns3_add_frag(ring);
4151 /* As the head data may be changed when GRO enable, copy
4152 * the head data in after other data rx completed
4154 if (skb->len > HNS3_RX_HEAD_SIZE)
4155 memcpy(skb->data, ring->va,
4156 ALIGN(ring->pull_len, sizeof(long)));
4158 ret = hns3_handle_bdinfo(ring, skb);
4159 if (unlikely(ret)) {
4160 dev_kfree_skb_any(skb);
4164 skb_record_rx_queue(skb, ring->tqp->tqp_index);
4168 int hns3_clean_rx_ring(struct hns3_enet_ring *ring, int budget,
4169 void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
4171 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
4172 int unused_count = hns3_desc_unused(ring);
4176 unused_count -= ring->pending_buf;
4178 while (recv_pkts < budget) {
4179 /* Reuse or realloc buffers */
4180 if (unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
4181 hns3_nic_alloc_rx_buffers(ring, unused_count);
4182 unused_count = hns3_desc_unused(ring) -
4187 err = hns3_handle_rx_bd(ring);
4188 /* Do not get FE for the packet or failed to alloc skb */
4189 if (unlikely(!ring->skb || err == -ENXIO)) {
4191 } else if (likely(!err)) {
4192 rx_fn(ring, ring->skb);
4196 unused_count += ring->pending_buf;
4198 ring->pending_buf = 0;
4202 /* Make all data has been write before submit */
4203 if (unused_count > 0)
4204 hns3_nic_alloc_rx_buffers(ring, unused_count);
4209 static void hns3_update_rx_int_coalesce(struct hns3_enet_tqp_vector *tqp_vector)
4211 struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
4212 struct dim_sample sample = {};
4214 if (!rx_group->coal.adapt_enable)
4217 dim_update_sample(tqp_vector->event_cnt, rx_group->total_packets,
4218 rx_group->total_bytes, &sample);
4219 net_dim(&rx_group->dim, sample);
4222 static void hns3_update_tx_int_coalesce(struct hns3_enet_tqp_vector *tqp_vector)
4224 struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
4225 struct dim_sample sample = {};
4227 if (!tx_group->coal.adapt_enable)
4230 dim_update_sample(tqp_vector->event_cnt, tx_group->total_packets,
4231 tx_group->total_bytes, &sample);
4232 net_dim(&tx_group->dim, sample);
4235 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
4237 struct hns3_nic_priv *priv = netdev_priv(napi->dev);
4238 struct hns3_enet_ring *ring;
4239 int rx_pkt_total = 0;
4241 struct hns3_enet_tqp_vector *tqp_vector =
4242 container_of(napi, struct hns3_enet_tqp_vector, napi);
4243 bool clean_complete = true;
4244 int rx_budget = budget;
4246 if (unlikely(test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
4247 napi_complete(napi);
4251 /* Since the actual Tx work is minimal, we can give the Tx a larger
4252 * budget and be more aggressive about cleaning up the Tx descriptors.
4254 hns3_for_each_ring(ring, tqp_vector->tx_group)
4255 hns3_clean_tx_ring(ring, budget);
4257 /* make sure rx ring budget not smaller than 1 */
4258 if (tqp_vector->num_tqps > 1)
4259 rx_budget = max(budget / tqp_vector->num_tqps, 1);
4261 hns3_for_each_ring(ring, tqp_vector->rx_group) {
4262 int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
4264 if (rx_cleaned >= rx_budget)
4265 clean_complete = false;
4267 rx_pkt_total += rx_cleaned;
4270 tqp_vector->rx_group.total_packets += rx_pkt_total;
4272 if (!clean_complete)
4275 if (napi_complete(napi) &&
4276 likely(!test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
4277 hns3_update_rx_int_coalesce(tqp_vector);
4278 hns3_update_tx_int_coalesce(tqp_vector);
4280 hns3_mask_vector_irq(tqp_vector, 1);
4283 return rx_pkt_total;
4286 static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
4287 struct hnae3_ring_chain_node *head)
4289 struct pci_dev *pdev = tqp_vector->handle->pdev;
4290 struct hnae3_ring_chain_node *cur_chain = head;
4291 struct hnae3_ring_chain_node *chain;
4292 struct hns3_enet_ring *tx_ring;
4293 struct hns3_enet_ring *rx_ring;
4295 tx_ring = tqp_vector->tx_group.ring;
4297 cur_chain->tqp_index = tx_ring->tqp->tqp_index;
4298 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
4299 HNAE3_RING_TYPE_TX);
4300 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
4301 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_TX);
4303 cur_chain->next = NULL;
4305 while (tx_ring->next) {
4306 tx_ring = tx_ring->next;
4308 chain = devm_kzalloc(&pdev->dev, sizeof(*chain),
4311 goto err_free_chain;
4313 cur_chain->next = chain;
4314 chain->tqp_index = tx_ring->tqp->tqp_index;
4315 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
4316 HNAE3_RING_TYPE_TX);
4317 hnae3_set_field(chain->int_gl_idx,
4318 HNAE3_RING_GL_IDX_M,
4319 HNAE3_RING_GL_IDX_S,
4326 rx_ring = tqp_vector->rx_group.ring;
4327 if (!tx_ring && rx_ring) {
4328 cur_chain->next = NULL;
4329 cur_chain->tqp_index = rx_ring->tqp->tqp_index;
4330 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
4331 HNAE3_RING_TYPE_RX);
4332 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
4333 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
4335 rx_ring = rx_ring->next;
4339 chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
4341 goto err_free_chain;
4343 cur_chain->next = chain;
4344 chain->tqp_index = rx_ring->tqp->tqp_index;
4345 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
4346 HNAE3_RING_TYPE_RX);
4347 hnae3_set_field(chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
4348 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
4352 rx_ring = rx_ring->next;
4358 cur_chain = head->next;
4360 chain = cur_chain->next;
4361 devm_kfree(&pdev->dev, cur_chain);
4369 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
4370 struct hnae3_ring_chain_node *head)
4372 struct pci_dev *pdev = tqp_vector->handle->pdev;
4373 struct hnae3_ring_chain_node *chain_tmp, *chain;
4378 chain_tmp = chain->next;
4379 devm_kfree(&pdev->dev, chain);
4384 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
4385 struct hns3_enet_ring *ring)
4387 ring->next = group->ring;
4393 static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
4395 struct pci_dev *pdev = priv->ae_handle->pdev;
4396 struct hns3_enet_tqp_vector *tqp_vector;
4397 int num_vectors = priv->vector_num;
4401 numa_node = dev_to_node(&pdev->dev);
4403 for (vector_i = 0; vector_i < num_vectors; vector_i++) {
4404 tqp_vector = &priv->tqp_vector[vector_i];
4405 cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
4406 &tqp_vector->affinity_mask);
4410 static void hns3_rx_dim_work(struct work_struct *work)
4412 struct dim *dim = container_of(work, struct dim, work);
4413 struct hns3_enet_ring_group *group = container_of(dim,
4414 struct hns3_enet_ring_group, dim);
4415 struct hns3_enet_tqp_vector *tqp_vector = group->ring->tqp_vector;
4416 struct dim_cq_moder cur_moder =
4417 net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
4419 hns3_set_vector_coalesce_rx_gl(group->ring->tqp_vector, cur_moder.usec);
4420 tqp_vector->rx_group.coal.int_gl = cur_moder.usec;
4422 if (cur_moder.pkts < tqp_vector->rx_group.coal.int_ql_max) {
4423 hns3_set_vector_coalesce_rx_ql(tqp_vector, cur_moder.pkts);
4424 tqp_vector->rx_group.coal.int_ql = cur_moder.pkts;
4427 dim->state = DIM_START_MEASURE;
4430 static void hns3_tx_dim_work(struct work_struct *work)
4432 struct dim *dim = container_of(work, struct dim, work);
4433 struct hns3_enet_ring_group *group = container_of(dim,
4434 struct hns3_enet_ring_group, dim);
4435 struct hns3_enet_tqp_vector *tqp_vector = group->ring->tqp_vector;
4436 struct dim_cq_moder cur_moder =
4437 net_dim_get_tx_moderation(dim->mode, dim->profile_ix);
4439 hns3_set_vector_coalesce_tx_gl(tqp_vector, cur_moder.usec);
4440 tqp_vector->tx_group.coal.int_gl = cur_moder.usec;
4442 if (cur_moder.pkts < tqp_vector->tx_group.coal.int_ql_max) {
4443 hns3_set_vector_coalesce_tx_ql(tqp_vector, cur_moder.pkts);
4444 tqp_vector->tx_group.coal.int_ql = cur_moder.pkts;
4447 dim->state = DIM_START_MEASURE;
4450 static void hns3_nic_init_dim(struct hns3_enet_tqp_vector *tqp_vector)
4452 INIT_WORK(&tqp_vector->rx_group.dim.work, hns3_rx_dim_work);
4453 INIT_WORK(&tqp_vector->tx_group.dim.work, hns3_tx_dim_work);
4456 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
4458 struct hnae3_handle *h = priv->ae_handle;
4459 struct hns3_enet_tqp_vector *tqp_vector;
4463 hns3_nic_set_cpumask(priv);
4465 for (i = 0; i < priv->vector_num; i++) {
4466 tqp_vector = &priv->tqp_vector[i];
4467 hns3_vector_coalesce_init_hw(tqp_vector, priv);
4468 tqp_vector->num_tqps = 0;
4469 hns3_nic_init_dim(tqp_vector);
4472 for (i = 0; i < h->kinfo.num_tqps; i++) {
4473 u16 vector_i = i % priv->vector_num;
4474 u16 tqp_num = h->kinfo.num_tqps;
4476 tqp_vector = &priv->tqp_vector[vector_i];
4478 hns3_add_ring_to_group(&tqp_vector->tx_group,
4481 hns3_add_ring_to_group(&tqp_vector->rx_group,
4482 &priv->ring[i + tqp_num]);
4484 priv->ring[i].tqp_vector = tqp_vector;
4485 priv->ring[i + tqp_num].tqp_vector = tqp_vector;
4486 tqp_vector->num_tqps++;
4489 for (i = 0; i < priv->vector_num; i++) {
4490 struct hnae3_ring_chain_node vector_ring_chain;
4492 tqp_vector = &priv->tqp_vector[i];
4494 tqp_vector->rx_group.total_bytes = 0;
4495 tqp_vector->rx_group.total_packets = 0;
4496 tqp_vector->tx_group.total_bytes = 0;
4497 tqp_vector->tx_group.total_packets = 0;
4498 tqp_vector->handle = h;
4500 ret = hns3_get_vector_ring_chain(tqp_vector,
4501 &vector_ring_chain);
4505 ret = h->ae_algo->ops->map_ring_to_vector(h,
4506 tqp_vector->vector_irq, &vector_ring_chain);
4508 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
4513 netif_napi_add(priv->netdev, &tqp_vector->napi,
4514 hns3_nic_common_poll, NAPI_POLL_WEIGHT);
4521 netif_napi_del(&priv->tqp_vector[i].napi);
4526 static void hns3_nic_init_coal_cfg(struct hns3_nic_priv *priv)
4528 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
4529 struct hns3_enet_coalesce *tx_coal = &priv->tx_coal;
4530 struct hns3_enet_coalesce *rx_coal = &priv->rx_coal;
4532 /* initialize the configuration for interrupt coalescing.
4533 * 1. GL (Interrupt Gap Limiter)
4534 * 2. RL (Interrupt Rate Limiter)
4535 * 3. QL (Interrupt Quantity Limiter)
4537 * Default: enable interrupt coalescing self-adaptive and GL
4539 tx_coal->adapt_enable = 1;
4540 rx_coal->adapt_enable = 1;
4542 tx_coal->int_gl = HNS3_INT_GL_50K;
4543 rx_coal->int_gl = HNS3_INT_GL_50K;
4545 rx_coal->flow_level = HNS3_FLOW_LOW;
4546 tx_coal->flow_level = HNS3_FLOW_LOW;
4548 if (ae_dev->dev_specs.int_ql_max) {
4549 tx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
4550 rx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
4554 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
4556 struct hnae3_handle *h = priv->ae_handle;
4557 struct hns3_enet_tqp_vector *tqp_vector;
4558 struct hnae3_vector_info *vector;
4559 struct pci_dev *pdev = h->pdev;
4560 u16 tqp_num = h->kinfo.num_tqps;
4565 /* RSS size, cpu online and vector_num should be the same */
4566 /* Should consider 2p/4p later */
4567 vector_num = min_t(u16, num_online_cpus(), tqp_num);
4569 vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
4574 /* save the actual available vector number */
4575 vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
4577 priv->vector_num = vector_num;
4578 priv->tqp_vector = (struct hns3_enet_tqp_vector *)
4579 devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
4581 if (!priv->tqp_vector) {
4586 for (i = 0; i < priv->vector_num; i++) {
4587 tqp_vector = &priv->tqp_vector[i];
4588 tqp_vector->idx = i;
4589 tqp_vector->mask_addr = vector[i].io_addr;
4590 tqp_vector->vector_irq = vector[i].vector;
4591 hns3_vector_coalesce_init(tqp_vector, priv);
4595 devm_kfree(&pdev->dev, vector);
4599 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
4605 static void hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
4607 struct hnae3_ring_chain_node vector_ring_chain;
4608 struct hnae3_handle *h = priv->ae_handle;
4609 struct hns3_enet_tqp_vector *tqp_vector;
4612 for (i = 0; i < priv->vector_num; i++) {
4613 tqp_vector = &priv->tqp_vector[i];
4615 if (!tqp_vector->rx_group.ring && !tqp_vector->tx_group.ring)
4618 /* Since the mapping can be overwritten, when fail to get the
4619 * chain between vector and ring, we should go on to deal with
4620 * the remaining options.
4622 if (hns3_get_vector_ring_chain(tqp_vector, &vector_ring_chain))
4623 dev_warn(priv->dev, "failed to get ring chain\n");
4625 h->ae_algo->ops->unmap_ring_from_vector(h,
4626 tqp_vector->vector_irq, &vector_ring_chain);
4628 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
4630 hns3_clear_ring_group(&tqp_vector->rx_group);
4631 hns3_clear_ring_group(&tqp_vector->tx_group);
4632 netif_napi_del(&priv->tqp_vector[i].napi);
4636 static void hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
4638 struct hnae3_handle *h = priv->ae_handle;
4639 struct pci_dev *pdev = h->pdev;
4642 for (i = 0; i < priv->vector_num; i++) {
4643 struct hns3_enet_tqp_vector *tqp_vector;
4645 tqp_vector = &priv->tqp_vector[i];
4646 ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
4651 devm_kfree(&pdev->dev, priv->tqp_vector);
4654 static void hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
4655 unsigned int ring_type)
4657 int queue_num = priv->ae_handle->kinfo.num_tqps;
4658 struct hns3_enet_ring *ring;
4661 if (ring_type == HNAE3_RING_TYPE_TX) {
4662 ring = &priv->ring[q->tqp_index];
4663 desc_num = priv->ae_handle->kinfo.num_tx_desc;
4664 ring->queue_index = q->tqp_index;
4665 ring->tx_copybreak = priv->tx_copybreak;
4666 ring->last_to_use = 0;
4668 ring = &priv->ring[q->tqp_index + queue_num];
4669 desc_num = priv->ae_handle->kinfo.num_rx_desc;
4670 ring->queue_index = q->tqp_index;
4671 ring->rx_copybreak = priv->rx_copybreak;
4674 hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
4678 ring->desc_cb = NULL;
4679 ring->dev = priv->dev;
4680 ring->desc_dma_addr = 0;
4681 ring->buf_size = q->buf_size;
4682 ring->desc_num = desc_num;
4683 ring->next_to_use = 0;
4684 ring->next_to_clean = 0;
4687 static void hns3_queue_to_ring(struct hnae3_queue *tqp,
4688 struct hns3_nic_priv *priv)
4690 hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
4691 hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
4694 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
4696 struct hnae3_handle *h = priv->ae_handle;
4697 struct pci_dev *pdev = h->pdev;
4700 priv->ring = devm_kzalloc(&pdev->dev,
4701 array3_size(h->kinfo.num_tqps,
4702 sizeof(*priv->ring), 2),
4707 for (i = 0; i < h->kinfo.num_tqps; i++)
4708 hns3_queue_to_ring(h->kinfo.tqp[i], priv);
4713 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
4718 devm_kfree(priv->dev, priv->ring);
4722 static void hns3_alloc_page_pool(struct hns3_enet_ring *ring)
4724 struct page_pool_params pp_params = {
4725 .flags = PP_FLAG_DMA_MAP | PP_FLAG_PAGE_FRAG |
4726 PP_FLAG_DMA_SYNC_DEV,
4727 .order = hns3_page_order(ring),
4728 .pool_size = ring->desc_num * hns3_buf_size(ring) /
4729 (PAGE_SIZE << hns3_page_order(ring)),
4730 .nid = dev_to_node(ring_to_dev(ring)),
4731 .dev = ring_to_dev(ring),
4732 .dma_dir = DMA_FROM_DEVICE,
4734 .max_len = PAGE_SIZE << hns3_page_order(ring),
4737 ring->page_pool = page_pool_create(&pp_params);
4738 if (IS_ERR(ring->page_pool)) {
4739 dev_warn(ring_to_dev(ring), "page pool creation failed: %ld\n",
4740 PTR_ERR(ring->page_pool));
4741 ring->page_pool = NULL;
4745 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
4749 if (ring->desc_num <= 0 || ring->buf_size <= 0)
4752 ring->desc_cb = devm_kcalloc(ring_to_dev(ring), ring->desc_num,
4753 sizeof(ring->desc_cb[0]), GFP_KERNEL);
4754 if (!ring->desc_cb) {
4759 ret = hns3_alloc_desc(ring);
4761 goto out_with_desc_cb;
4763 if (!HNAE3_IS_TX_RING(ring)) {
4764 if (page_pool_enabled)
4765 hns3_alloc_page_pool(ring);
4767 ret = hns3_alloc_ring_buffers(ring);
4771 hns3_init_tx_spare_buffer(ring);
4777 hns3_free_desc(ring);
4779 devm_kfree(ring_to_dev(ring), ring->desc_cb);
4780 ring->desc_cb = NULL;
4785 void hns3_fini_ring(struct hns3_enet_ring *ring)
4787 hns3_free_desc(ring);
4788 devm_kfree(ring_to_dev(ring), ring->desc_cb);
4789 ring->desc_cb = NULL;
4790 ring->next_to_clean = 0;
4791 ring->next_to_use = 0;
4792 ring->last_to_use = 0;
4793 ring->pending_buf = 0;
4794 if (!HNAE3_IS_TX_RING(ring) && ring->skb) {
4795 dev_kfree_skb_any(ring->skb);
4797 } else if (HNAE3_IS_TX_RING(ring) && ring->tx_spare) {
4798 struct hns3_tx_spare *tx_spare = ring->tx_spare;
4800 dma_unmap_page(ring_to_dev(ring), tx_spare->dma, tx_spare->len,
4802 free_pages((unsigned long)tx_spare->buf,
4803 get_order(tx_spare->len));
4804 devm_kfree(ring_to_dev(ring), tx_spare);
4805 ring->tx_spare = NULL;
4808 if (!HNAE3_IS_TX_RING(ring) && ring->page_pool) {
4809 page_pool_destroy(ring->page_pool);
4810 ring->page_pool = NULL;
4814 static int hns3_buf_size2type(u32 buf_size)
4820 bd_size_type = HNS3_BD_SIZE_512_TYPE;
4823 bd_size_type = HNS3_BD_SIZE_1024_TYPE;
4826 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4829 bd_size_type = HNS3_BD_SIZE_4096_TYPE;
4832 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4835 return bd_size_type;
4838 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
4840 dma_addr_t dma = ring->desc_dma_addr;
4841 struct hnae3_queue *q = ring->tqp;
4843 if (!HNAE3_IS_TX_RING(ring)) {
4844 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG, (u32)dma);
4845 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
4846 (u32)((dma >> 31) >> 1));
4848 hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
4849 hns3_buf_size2type(ring->buf_size));
4850 hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
4851 ring->desc_num / 8 - 1);
4853 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
4855 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
4856 (u32)((dma >> 31) >> 1));
4858 hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
4859 ring->desc_num / 8 - 1);
4863 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
4865 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
4866 struct hnae3_tc_info *tc_info = &kinfo->tc_info;
4869 for (i = 0; i < tc_info->num_tc; i++) {
4872 for (j = 0; j < tc_info->tqp_count[i]; j++) {
4873 struct hnae3_queue *q;
4875 q = priv->ring[tc_info->tqp_offset[i] + j].tqp;
4876 hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG, i);
4881 int hns3_init_all_ring(struct hns3_nic_priv *priv)
4883 struct hnae3_handle *h = priv->ae_handle;
4884 int ring_num = h->kinfo.num_tqps * 2;
4888 for (i = 0; i < ring_num; i++) {
4889 ret = hns3_alloc_ring_memory(&priv->ring[i]);
4892 "Alloc ring memory fail! ret=%d\n", ret);
4893 goto out_when_alloc_ring_memory;
4896 u64_stats_init(&priv->ring[i].syncp);
4901 out_when_alloc_ring_memory:
4902 for (j = i - 1; j >= 0; j--)
4903 hns3_fini_ring(&priv->ring[j]);
4908 static void hns3_uninit_all_ring(struct hns3_nic_priv *priv)
4910 struct hnae3_handle *h = priv->ae_handle;
4913 for (i = 0; i < h->kinfo.num_tqps; i++) {
4914 hns3_fini_ring(&priv->ring[i]);
4915 hns3_fini_ring(&priv->ring[i + h->kinfo.num_tqps]);
4919 /* Set mac addr if it is configured. or leave it to the AE driver */
4920 static int hns3_init_mac_addr(struct net_device *netdev)
4922 struct hns3_nic_priv *priv = netdev_priv(netdev);
4923 struct hnae3_handle *h = priv->ae_handle;
4924 u8 mac_addr_temp[ETH_ALEN];
4927 if (h->ae_algo->ops->get_mac_addr)
4928 h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
4930 /* Check if the MAC address is valid, if not get a random one */
4931 if (!is_valid_ether_addr(mac_addr_temp)) {
4932 eth_hw_addr_random(netdev);
4933 dev_warn(priv->dev, "using random MAC address %pM\n",
4935 } else if (!ether_addr_equal(netdev->dev_addr, mac_addr_temp)) {
4936 ether_addr_copy(netdev->dev_addr, mac_addr_temp);
4937 ether_addr_copy(netdev->perm_addr, mac_addr_temp);
4942 if (h->ae_algo->ops->set_mac_addr)
4943 ret = h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
4948 static int hns3_init_phy(struct net_device *netdev)
4950 struct hnae3_handle *h = hns3_get_handle(netdev);
4953 if (h->ae_algo->ops->mac_connect_phy)
4954 ret = h->ae_algo->ops->mac_connect_phy(h);
4959 static void hns3_uninit_phy(struct net_device *netdev)
4961 struct hnae3_handle *h = hns3_get_handle(netdev);
4963 if (h->ae_algo->ops->mac_disconnect_phy)
4964 h->ae_algo->ops->mac_disconnect_phy(h);
4967 static int hns3_client_start(struct hnae3_handle *handle)
4969 if (!handle->ae_algo->ops->client_start)
4972 return handle->ae_algo->ops->client_start(handle);
4975 static void hns3_client_stop(struct hnae3_handle *handle)
4977 if (!handle->ae_algo->ops->client_stop)
4980 handle->ae_algo->ops->client_stop(handle);
4983 static void hns3_info_show(struct hns3_nic_priv *priv)
4985 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
4987 dev_info(priv->dev, "MAC address: %pM\n", priv->netdev->dev_addr);
4988 dev_info(priv->dev, "Task queue pairs numbers: %u\n", kinfo->num_tqps);
4989 dev_info(priv->dev, "RSS size: %u\n", kinfo->rss_size);
4990 dev_info(priv->dev, "Allocated RSS size: %u\n", kinfo->req_rss_size);
4991 dev_info(priv->dev, "RX buffer length: %u\n", kinfo->rx_buf_len);
4992 dev_info(priv->dev, "Desc num per TX queue: %u\n", kinfo->num_tx_desc);
4993 dev_info(priv->dev, "Desc num per RX queue: %u\n", kinfo->num_rx_desc);
4994 dev_info(priv->dev, "Total number of enabled TCs: %u\n",
4995 kinfo->tc_info.num_tc);
4996 dev_info(priv->dev, "Max mtu size: %u\n", priv->netdev->max_mtu);
4999 static void hns3_set_cq_period_mode(struct hns3_nic_priv *priv,
5000 enum dim_cq_period_mode mode, bool is_tx)
5002 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
5003 struct hnae3_handle *handle = priv->ae_handle;
5007 priv->tx_cqe_mode = mode;
5009 for (i = 0; i < priv->vector_num; i++)
5010 priv->tqp_vector[i].tx_group.dim.mode = mode;
5012 priv->rx_cqe_mode = mode;
5014 for (i = 0; i < priv->vector_num; i++)
5015 priv->tqp_vector[i].rx_group.dim.mode = mode;
5018 /* only device version above V3(include V3), GL can switch CQ/EQ
5021 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3) {
5025 new_mode = (mode == DIM_CQ_PERIOD_MODE_START_FROM_CQE) ?
5026 HNS3_CQ_MODE_CQE : HNS3_CQ_MODE_EQE;
5027 reg = is_tx ? HNS3_GL1_CQ_MODE_REG : HNS3_GL0_CQ_MODE_REG;
5029 writel(new_mode, handle->kinfo.io_base + reg);
5033 void hns3_cq_period_mode_init(struct hns3_nic_priv *priv,
5034 enum dim_cq_period_mode tx_mode,
5035 enum dim_cq_period_mode rx_mode)
5037 hns3_set_cq_period_mode(priv, tx_mode, true);
5038 hns3_set_cq_period_mode(priv, rx_mode, false);
5041 static void hns3_state_init(struct hnae3_handle *handle)
5043 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(handle->pdev);
5044 struct net_device *netdev = handle->kinfo.netdev;
5045 struct hns3_nic_priv *priv = netdev_priv(netdev);
5047 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
5049 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
5050 set_bit(HNAE3_PFLAG_LIMIT_PROMISC, &handle->supported_pflags);
5052 if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps))
5053 set_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state);
5055 if (hnae3_ae_dev_rxd_adv_layout_supported(ae_dev))
5056 set_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state);
5059 static int hns3_client_init(struct hnae3_handle *handle)
5061 struct pci_dev *pdev = handle->pdev;
5062 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
5063 u16 alloc_tqps, max_rss_size;
5064 struct hns3_nic_priv *priv;
5065 struct net_device *netdev;
5068 handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
5070 netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
5074 priv = netdev_priv(netdev);
5075 priv->dev = &pdev->dev;
5076 priv->netdev = netdev;
5077 priv->ae_handle = handle;
5078 priv->tx_timeout_count = 0;
5079 priv->max_non_tso_bd_num = ae_dev->dev_specs.max_non_tso_bd_num;
5080 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
5082 handle->msg_enable = netif_msg_init(debug, DEFAULT_MSG_LEVEL);
5084 handle->kinfo.netdev = netdev;
5085 handle->priv = (void *)priv;
5087 hns3_init_mac_addr(netdev);
5089 hns3_set_default_feature(netdev);
5091 netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
5092 netdev->priv_flags |= IFF_UNICAST_FLT;
5093 netdev->netdev_ops = &hns3_nic_netdev_ops;
5094 SET_NETDEV_DEV(netdev, &pdev->dev);
5095 hns3_ethtool_set_ops(netdev);
5097 /* Carrier off reporting is important to ethtool even BEFORE open */
5098 netif_carrier_off(netdev);
5100 ret = hns3_get_ring_config(priv);
5103 goto out_get_ring_cfg;
5106 hns3_nic_init_coal_cfg(priv);
5108 ret = hns3_nic_alloc_vector_data(priv);
5111 goto out_alloc_vector_data;
5114 ret = hns3_nic_init_vector_data(priv);
5117 goto out_init_vector_data;
5120 ret = hns3_init_all_ring(priv);
5126 hns3_cq_period_mode_init(priv, DIM_CQ_PERIOD_MODE_START_FROM_EQE,
5127 DIM_CQ_PERIOD_MODE_START_FROM_EQE);
5129 ret = hns3_init_phy(netdev);
5133 /* the device can work without cpu rmap, only aRFS needs it */
5134 ret = hns3_set_rx_cpu_rmap(netdev);
5136 dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
5138 ret = hns3_nic_init_irq(priv);
5140 dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
5141 hns3_free_rx_cpu_rmap(netdev);
5142 goto out_init_irq_fail;
5145 ret = hns3_client_start(handle);
5147 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
5148 goto out_client_start;
5151 hns3_dcbnl_setup(handle);
5153 ret = hns3_dbg_init(handle);
5155 dev_err(priv->dev, "failed to init debugfs, ret = %d\n",
5157 goto out_client_start;
5160 netdev->max_mtu = HNS3_MAX_MTU(ae_dev->dev_specs.max_frm_size);
5162 hns3_state_init(handle);
5164 ret = register_netdev(netdev);
5166 dev_err(priv->dev, "probe register netdev fail!\n");
5167 goto out_reg_netdev_fail;
5170 if (netif_msg_drv(handle))
5171 hns3_info_show(priv);
5175 out_reg_netdev_fail:
5176 hns3_dbg_uninit(handle);
5178 hns3_free_rx_cpu_rmap(netdev);
5179 hns3_nic_uninit_irq(priv);
5181 hns3_uninit_phy(netdev);
5183 hns3_uninit_all_ring(priv);
5185 hns3_nic_uninit_vector_data(priv);
5186 out_init_vector_data:
5187 hns3_nic_dealloc_vector_data(priv);
5188 out_alloc_vector_data:
5191 priv->ae_handle = NULL;
5192 free_netdev(netdev);
5196 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
5198 struct net_device *netdev = handle->kinfo.netdev;
5199 struct hns3_nic_priv *priv = netdev_priv(netdev);
5201 if (netdev->reg_state != NETREG_UNINITIALIZED)
5202 unregister_netdev(netdev);
5204 hns3_client_stop(handle);
5206 hns3_uninit_phy(netdev);
5208 if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5209 netdev_warn(netdev, "already uninitialized\n");
5210 goto out_netdev_free;
5213 hns3_free_rx_cpu_rmap(netdev);
5215 hns3_nic_uninit_irq(priv);
5217 hns3_clear_all_ring(handle, true);
5219 hns3_nic_uninit_vector_data(priv);
5221 hns3_nic_dealloc_vector_data(priv);
5223 hns3_uninit_all_ring(priv);
5225 hns3_put_ring_config(priv);
5228 hns3_dbg_uninit(handle);
5229 free_netdev(netdev);
5232 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
5234 struct net_device *netdev = handle->kinfo.netdev;
5240 netif_tx_wake_all_queues(netdev);
5241 netif_carrier_on(netdev);
5242 if (netif_msg_link(handle))
5243 netdev_info(netdev, "link up\n");
5245 netif_carrier_off(netdev);
5246 netif_tx_stop_all_queues(netdev);
5247 if (netif_msg_link(handle))
5248 netdev_info(netdev, "link down\n");
5252 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
5254 while (ring->next_to_clean != ring->next_to_use) {
5255 ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
5256 hns3_free_buffer_detach(ring, ring->next_to_clean, 0);
5257 ring_ptr_move_fw(ring, next_to_clean);
5260 ring->pending_buf = 0;
5263 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
5265 struct hns3_desc_cb res_cbs;
5268 while (ring->next_to_use != ring->next_to_clean) {
5269 /* When a buffer is not reused, it's memory has been
5270 * freed in hns3_handle_rx_bd or will be freed by
5271 * stack, so we need to replace the buffer here.
5273 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
5274 ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
5276 u64_stats_update_begin(&ring->syncp);
5277 ring->stats.sw_err_cnt++;
5278 u64_stats_update_end(&ring->syncp);
5279 /* if alloc new buffer fail, exit directly
5280 * and reclear in up flow.
5282 netdev_warn(ring_to_netdev(ring),
5283 "reserve buffer map failed, ret = %d\n",
5287 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
5289 ring_ptr_move_fw(ring, next_to_use);
5292 /* Free the pending skb in rx ring */
5294 dev_kfree_skb_any(ring->skb);
5296 ring->pending_buf = 0;
5302 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
5304 while (ring->next_to_use != ring->next_to_clean) {
5305 /* When a buffer is not reused, it's memory has been
5306 * freed in hns3_handle_rx_bd or will be freed by
5307 * stack, so only need to unmap the buffer here.
5309 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
5310 hns3_unmap_buffer(ring,
5311 &ring->desc_cb[ring->next_to_use]);
5312 ring->desc_cb[ring->next_to_use].dma = 0;
5315 ring_ptr_move_fw(ring, next_to_use);
5319 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force)
5321 struct net_device *ndev = h->kinfo.netdev;
5322 struct hns3_nic_priv *priv = netdev_priv(ndev);
5325 for (i = 0; i < h->kinfo.num_tqps; i++) {
5326 struct hns3_enet_ring *ring;
5328 ring = &priv->ring[i];
5329 hns3_clear_tx_ring(ring);
5331 ring = &priv->ring[i + h->kinfo.num_tqps];
5332 /* Continue to clear other rings even if clearing some
5336 hns3_force_clear_rx_ring(ring);
5338 hns3_clear_rx_ring(ring);
5342 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
5344 struct net_device *ndev = h->kinfo.netdev;
5345 struct hns3_nic_priv *priv = netdev_priv(ndev);
5346 struct hns3_enet_ring *rx_ring;
5350 ret = h->ae_algo->ops->reset_queue(h);
5354 for (i = 0; i < h->kinfo.num_tqps; i++) {
5355 hns3_init_ring_hw(&priv->ring[i]);
5357 /* We need to clear tx ring here because self test will
5358 * use the ring and will not run down before up
5360 hns3_clear_tx_ring(&priv->ring[i]);
5361 priv->ring[i].next_to_clean = 0;
5362 priv->ring[i].next_to_use = 0;
5363 priv->ring[i].last_to_use = 0;
5365 rx_ring = &priv->ring[i + h->kinfo.num_tqps];
5366 hns3_init_ring_hw(rx_ring);
5367 ret = hns3_clear_rx_ring(rx_ring);
5371 /* We can not know the hardware head and tail when this
5372 * function is called in reset flow, so we reuse all desc.
5374 for (j = 0; j < rx_ring->desc_num; j++)
5375 hns3_reuse_buffer(rx_ring, j);
5377 rx_ring->next_to_clean = 0;
5378 rx_ring->next_to_use = 0;
5381 hns3_init_tx_ring_tc(priv);
5386 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
5388 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
5389 struct net_device *ndev = kinfo->netdev;
5390 struct hns3_nic_priv *priv = netdev_priv(ndev);
5392 if (test_and_set_bit(HNS3_NIC_STATE_RESETTING, &priv->state))
5395 if (!netif_running(ndev))
5398 return hns3_nic_net_stop(ndev);
5401 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
5403 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
5404 struct hns3_nic_priv *priv = netdev_priv(kinfo->netdev);
5407 if (!test_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5408 netdev_err(kinfo->netdev, "device is not initialized yet\n");
5412 clear_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
5414 if (netif_running(kinfo->netdev)) {
5415 ret = hns3_nic_net_open(kinfo->netdev);
5417 set_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
5418 netdev_err(kinfo->netdev,
5419 "net up fail, ret=%d!\n", ret);
5427 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
5429 struct net_device *netdev = handle->kinfo.netdev;
5430 struct hns3_nic_priv *priv = netdev_priv(netdev);
5433 /* Carrier off reporting is important to ethtool even BEFORE open */
5434 netif_carrier_off(netdev);
5436 ret = hns3_get_ring_config(priv);
5440 ret = hns3_nic_alloc_vector_data(priv);
5444 ret = hns3_nic_init_vector_data(priv);
5446 goto err_dealloc_vector;
5448 ret = hns3_init_all_ring(priv);
5450 goto err_uninit_vector;
5452 hns3_cq_period_mode_init(priv, priv->tx_cqe_mode, priv->rx_cqe_mode);
5454 /* the device can work without cpu rmap, only aRFS needs it */
5455 ret = hns3_set_rx_cpu_rmap(netdev);
5457 dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
5459 ret = hns3_nic_init_irq(priv);
5461 dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
5462 hns3_free_rx_cpu_rmap(netdev);
5463 goto err_init_irq_fail;
5466 if (!hns3_is_phys_func(handle->pdev))
5467 hns3_init_mac_addr(netdev);
5469 ret = hns3_client_start(handle);
5471 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
5472 goto err_client_start_fail;
5475 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
5479 err_client_start_fail:
5480 hns3_free_rx_cpu_rmap(netdev);
5481 hns3_nic_uninit_irq(priv);
5483 hns3_uninit_all_ring(priv);
5485 hns3_nic_uninit_vector_data(priv);
5487 hns3_nic_dealloc_vector_data(priv);
5489 hns3_put_ring_config(priv);
5494 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
5496 struct net_device *netdev = handle->kinfo.netdev;
5497 struct hns3_nic_priv *priv = netdev_priv(netdev);
5499 if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5500 netdev_warn(netdev, "already uninitialized\n");
5504 hns3_free_rx_cpu_rmap(netdev);
5505 hns3_nic_uninit_irq(priv);
5506 hns3_clear_all_ring(handle, true);
5507 hns3_reset_tx_queue(priv->ae_handle);
5509 hns3_nic_uninit_vector_data(priv);
5511 hns3_nic_dealloc_vector_data(priv);
5513 hns3_uninit_all_ring(priv);
5515 hns3_put_ring_config(priv);
5520 static int hns3_reset_notify(struct hnae3_handle *handle,
5521 enum hnae3_reset_notify_type type)
5526 case HNAE3_UP_CLIENT:
5527 ret = hns3_reset_notify_up_enet(handle);
5529 case HNAE3_DOWN_CLIENT:
5530 ret = hns3_reset_notify_down_enet(handle);
5532 case HNAE3_INIT_CLIENT:
5533 ret = hns3_reset_notify_init_enet(handle);
5535 case HNAE3_UNINIT_CLIENT:
5536 ret = hns3_reset_notify_uninit_enet(handle);
5545 static int hns3_change_channels(struct hnae3_handle *handle, u32 new_tqp_num,
5546 bool rxfh_configured)
5550 ret = handle->ae_algo->ops->set_channels(handle, new_tqp_num,
5553 dev_err(&handle->pdev->dev,
5554 "Change tqp num(%u) fail.\n", new_tqp_num);
5558 ret = hns3_reset_notify(handle, HNAE3_INIT_CLIENT);
5562 ret = hns3_reset_notify(handle, HNAE3_UP_CLIENT);
5564 hns3_reset_notify(handle, HNAE3_UNINIT_CLIENT);
5569 int hns3_set_channels(struct net_device *netdev,
5570 struct ethtool_channels *ch)
5572 struct hnae3_handle *h = hns3_get_handle(netdev);
5573 struct hnae3_knic_private_info *kinfo = &h->kinfo;
5574 bool rxfh_configured = netif_is_rxfh_configured(netdev);
5575 u32 new_tqp_num = ch->combined_count;
5579 if (hns3_nic_resetting(netdev))
5582 if (ch->rx_count || ch->tx_count)
5585 if (kinfo->tc_info.mqprio_active) {
5586 dev_err(&netdev->dev,
5587 "it's not allowed to set channels via ethtool when MQPRIO mode is on\n");
5591 if (new_tqp_num > hns3_get_max_available_channels(h) ||
5593 dev_err(&netdev->dev,
5594 "Change tqps fail, the tqp range is from 1 to %u",
5595 hns3_get_max_available_channels(h));
5599 if (kinfo->rss_size == new_tqp_num)
5602 netif_dbg(h, drv, netdev,
5603 "set channels: tqp_num=%u, rxfh=%d\n",
5604 new_tqp_num, rxfh_configured);
5606 ret = hns3_reset_notify(h, HNAE3_DOWN_CLIENT);
5610 ret = hns3_reset_notify(h, HNAE3_UNINIT_CLIENT);
5614 org_tqp_num = h->kinfo.num_tqps;
5615 ret = hns3_change_channels(h, new_tqp_num, rxfh_configured);
5620 "Change channels fail, revert to old value\n");
5621 ret1 = hns3_change_channels(h, org_tqp_num, rxfh_configured);
5624 "revert to old channel fail\n");
5634 static const struct hns3_hw_error_info hns3_hw_err[] = {
5635 { .type = HNAE3_PPU_POISON_ERROR,
5636 .msg = "PPU poison" },
5637 { .type = HNAE3_CMDQ_ECC_ERROR,
5638 .msg = "IMP CMDQ error" },
5639 { .type = HNAE3_IMP_RD_POISON_ERROR,
5640 .msg = "IMP RD poison" },
5641 { .type = HNAE3_ROCEE_AXI_RESP_ERROR,
5642 .msg = "ROCEE AXI RESP error" },
5645 static void hns3_process_hw_error(struct hnae3_handle *handle,
5646 enum hnae3_hw_error_type type)
5650 for (i = 0; i < ARRAY_SIZE(hns3_hw_err); i++) {
5651 if (hns3_hw_err[i].type == type) {
5652 dev_err(&handle->pdev->dev, "Detected %s!\n",
5653 hns3_hw_err[i].msg);
5659 static const struct hnae3_client_ops client_ops = {
5660 .init_instance = hns3_client_init,
5661 .uninit_instance = hns3_client_uninit,
5662 .link_status_change = hns3_link_status_change,
5663 .reset_notify = hns3_reset_notify,
5664 .process_hw_error = hns3_process_hw_error,
5667 /* hns3_init_module - Driver registration routine
5668 * hns3_init_module is the first routine called when the driver is
5669 * loaded. All it does is register with the PCI subsystem.
5671 static int __init hns3_init_module(void)
5675 pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
5676 pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
5678 client.type = HNAE3_CLIENT_KNIC;
5679 snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH, "%s",
5682 client.ops = &client_ops;
5684 INIT_LIST_HEAD(&client.node);
5686 hns3_dbg_register_debugfs(hns3_driver_name);
5688 ret = hnae3_register_client(&client);
5690 goto err_reg_client;
5692 ret = pci_register_driver(&hns3_driver);
5694 goto err_reg_driver;
5699 hnae3_unregister_client(&client);
5701 hns3_dbg_unregister_debugfs();
5704 module_init(hns3_init_module);
5706 /* hns3_exit_module - Driver exit cleanup routine
5707 * hns3_exit_module is called just before the driver is removed
5710 static void __exit hns3_exit_module(void)
5712 pci_unregister_driver(&hns3_driver);
5713 hnae3_unregister_client(&client);
5714 hns3_dbg_unregister_debugfs();
5716 module_exit(hns3_exit_module);
5718 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
5719 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
5720 MODULE_LICENSE("GPL");
5721 MODULE_ALIAS("pci:hns-nic");