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 #define HNS3_SGL_SIZE(nfrag) (sizeof(struct scatterlist) * (nfrag) + \
65 sizeof(struct sg_table))
66 #define HNS3_MAX_SGL_SIZE ALIGN(HNS3_SGL_SIZE(HNS3_MAX_TSO_BD_NUM), \
67 dma_get_cache_alignment())
69 #define DEFAULT_MSG_LEVEL (NETIF_MSG_PROBE | NETIF_MSG_LINK | \
70 NETIF_MSG_IFDOWN | NETIF_MSG_IFUP)
72 #define HNS3_INNER_VLAN_TAG 1
73 #define HNS3_OUTER_VLAN_TAG 2
75 #define HNS3_MIN_TX_LEN 33U
77 /* hns3_pci_tbl - PCI Device ID Table
79 * Last entry must be all 0s
81 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
82 * Class, Class Mask, private data (not used) }
84 static const struct pci_device_id hns3_pci_tbl[] = {
85 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
86 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
87 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA),
88 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
89 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC),
90 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
91 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA),
92 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
93 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC),
94 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
95 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC),
96 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
97 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_200G_RDMA),
98 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
99 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_VF), 0},
100 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_RDMA_DCB_PFC_VF),
101 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
102 /* required last entry */
105 MODULE_DEVICE_TABLE(pci, hns3_pci_tbl);
107 #define HNS3_RX_PTYPE_ENTRY(ptype, l, s, t) \
114 #define HNS3_RX_PTYPE_UNUSED_ENTRY(ptype) \
115 { ptype, 0, CHECKSUM_NONE, HNS3_L3_TYPE_PARSE_FAIL, 0 }
117 static const struct hns3_rx_ptype hns3_rx_ptype_tbl[] = {
118 HNS3_RX_PTYPE_UNUSED_ENTRY(0),
119 HNS3_RX_PTYPE_ENTRY(1, 0, COMPLETE, ARP),
120 HNS3_RX_PTYPE_ENTRY(2, 0, COMPLETE, RARP),
121 HNS3_RX_PTYPE_ENTRY(3, 0, COMPLETE, LLDP),
122 HNS3_RX_PTYPE_ENTRY(4, 0, COMPLETE, PARSE_FAIL),
123 HNS3_RX_PTYPE_ENTRY(5, 0, COMPLETE, PARSE_FAIL),
124 HNS3_RX_PTYPE_ENTRY(6, 0, COMPLETE, PARSE_FAIL),
125 HNS3_RX_PTYPE_ENTRY(7, 0, COMPLETE, CNM),
126 HNS3_RX_PTYPE_ENTRY(8, 0, NONE, PARSE_FAIL),
127 HNS3_RX_PTYPE_UNUSED_ENTRY(9),
128 HNS3_RX_PTYPE_UNUSED_ENTRY(10),
129 HNS3_RX_PTYPE_UNUSED_ENTRY(11),
130 HNS3_RX_PTYPE_UNUSED_ENTRY(12),
131 HNS3_RX_PTYPE_UNUSED_ENTRY(13),
132 HNS3_RX_PTYPE_UNUSED_ENTRY(14),
133 HNS3_RX_PTYPE_UNUSED_ENTRY(15),
134 HNS3_RX_PTYPE_ENTRY(16, 0, COMPLETE, PARSE_FAIL),
135 HNS3_RX_PTYPE_ENTRY(17, 0, COMPLETE, IPV4),
136 HNS3_RX_PTYPE_ENTRY(18, 0, COMPLETE, IPV4),
137 HNS3_RX_PTYPE_ENTRY(19, 0, UNNECESSARY, IPV4),
138 HNS3_RX_PTYPE_ENTRY(20, 0, UNNECESSARY, IPV4),
139 HNS3_RX_PTYPE_ENTRY(21, 0, NONE, IPV4),
140 HNS3_RX_PTYPE_ENTRY(22, 0, UNNECESSARY, IPV4),
141 HNS3_RX_PTYPE_ENTRY(23, 0, NONE, IPV4),
142 HNS3_RX_PTYPE_ENTRY(24, 0, NONE, IPV4),
143 HNS3_RX_PTYPE_ENTRY(25, 0, UNNECESSARY, IPV4),
144 HNS3_RX_PTYPE_UNUSED_ENTRY(26),
145 HNS3_RX_PTYPE_UNUSED_ENTRY(27),
146 HNS3_RX_PTYPE_UNUSED_ENTRY(28),
147 HNS3_RX_PTYPE_ENTRY(29, 0, COMPLETE, PARSE_FAIL),
148 HNS3_RX_PTYPE_ENTRY(30, 0, COMPLETE, PARSE_FAIL),
149 HNS3_RX_PTYPE_ENTRY(31, 0, COMPLETE, IPV4),
150 HNS3_RX_PTYPE_ENTRY(32, 0, COMPLETE, IPV4),
151 HNS3_RX_PTYPE_ENTRY(33, 1, UNNECESSARY, IPV4),
152 HNS3_RX_PTYPE_ENTRY(34, 1, UNNECESSARY, IPV4),
153 HNS3_RX_PTYPE_ENTRY(35, 1, UNNECESSARY, IPV4),
154 HNS3_RX_PTYPE_ENTRY(36, 0, COMPLETE, IPV4),
155 HNS3_RX_PTYPE_ENTRY(37, 0, COMPLETE, IPV4),
156 HNS3_RX_PTYPE_UNUSED_ENTRY(38),
157 HNS3_RX_PTYPE_ENTRY(39, 0, COMPLETE, IPV6),
158 HNS3_RX_PTYPE_ENTRY(40, 0, COMPLETE, IPV6),
159 HNS3_RX_PTYPE_ENTRY(41, 1, UNNECESSARY, IPV6),
160 HNS3_RX_PTYPE_ENTRY(42, 1, UNNECESSARY, IPV6),
161 HNS3_RX_PTYPE_ENTRY(43, 1, UNNECESSARY, IPV6),
162 HNS3_RX_PTYPE_ENTRY(44, 0, COMPLETE, IPV6),
163 HNS3_RX_PTYPE_ENTRY(45, 0, COMPLETE, IPV6),
164 HNS3_RX_PTYPE_UNUSED_ENTRY(46),
165 HNS3_RX_PTYPE_UNUSED_ENTRY(47),
166 HNS3_RX_PTYPE_UNUSED_ENTRY(48),
167 HNS3_RX_PTYPE_UNUSED_ENTRY(49),
168 HNS3_RX_PTYPE_UNUSED_ENTRY(50),
169 HNS3_RX_PTYPE_UNUSED_ENTRY(51),
170 HNS3_RX_PTYPE_UNUSED_ENTRY(52),
171 HNS3_RX_PTYPE_UNUSED_ENTRY(53),
172 HNS3_RX_PTYPE_UNUSED_ENTRY(54),
173 HNS3_RX_PTYPE_UNUSED_ENTRY(55),
174 HNS3_RX_PTYPE_UNUSED_ENTRY(56),
175 HNS3_RX_PTYPE_UNUSED_ENTRY(57),
176 HNS3_RX_PTYPE_UNUSED_ENTRY(58),
177 HNS3_RX_PTYPE_UNUSED_ENTRY(59),
178 HNS3_RX_PTYPE_UNUSED_ENTRY(60),
179 HNS3_RX_PTYPE_UNUSED_ENTRY(61),
180 HNS3_RX_PTYPE_UNUSED_ENTRY(62),
181 HNS3_RX_PTYPE_UNUSED_ENTRY(63),
182 HNS3_RX_PTYPE_UNUSED_ENTRY(64),
183 HNS3_RX_PTYPE_UNUSED_ENTRY(65),
184 HNS3_RX_PTYPE_UNUSED_ENTRY(66),
185 HNS3_RX_PTYPE_UNUSED_ENTRY(67),
186 HNS3_RX_PTYPE_UNUSED_ENTRY(68),
187 HNS3_RX_PTYPE_UNUSED_ENTRY(69),
188 HNS3_RX_PTYPE_UNUSED_ENTRY(70),
189 HNS3_RX_PTYPE_UNUSED_ENTRY(71),
190 HNS3_RX_PTYPE_UNUSED_ENTRY(72),
191 HNS3_RX_PTYPE_UNUSED_ENTRY(73),
192 HNS3_RX_PTYPE_UNUSED_ENTRY(74),
193 HNS3_RX_PTYPE_UNUSED_ENTRY(75),
194 HNS3_RX_PTYPE_UNUSED_ENTRY(76),
195 HNS3_RX_PTYPE_UNUSED_ENTRY(77),
196 HNS3_RX_PTYPE_UNUSED_ENTRY(78),
197 HNS3_RX_PTYPE_UNUSED_ENTRY(79),
198 HNS3_RX_PTYPE_UNUSED_ENTRY(80),
199 HNS3_RX_PTYPE_UNUSED_ENTRY(81),
200 HNS3_RX_PTYPE_UNUSED_ENTRY(82),
201 HNS3_RX_PTYPE_UNUSED_ENTRY(83),
202 HNS3_RX_PTYPE_UNUSED_ENTRY(84),
203 HNS3_RX_PTYPE_UNUSED_ENTRY(85),
204 HNS3_RX_PTYPE_UNUSED_ENTRY(86),
205 HNS3_RX_PTYPE_UNUSED_ENTRY(87),
206 HNS3_RX_PTYPE_UNUSED_ENTRY(88),
207 HNS3_RX_PTYPE_UNUSED_ENTRY(89),
208 HNS3_RX_PTYPE_UNUSED_ENTRY(90),
209 HNS3_RX_PTYPE_UNUSED_ENTRY(91),
210 HNS3_RX_PTYPE_UNUSED_ENTRY(92),
211 HNS3_RX_PTYPE_UNUSED_ENTRY(93),
212 HNS3_RX_PTYPE_UNUSED_ENTRY(94),
213 HNS3_RX_PTYPE_UNUSED_ENTRY(95),
214 HNS3_RX_PTYPE_UNUSED_ENTRY(96),
215 HNS3_RX_PTYPE_UNUSED_ENTRY(97),
216 HNS3_RX_PTYPE_UNUSED_ENTRY(98),
217 HNS3_RX_PTYPE_UNUSED_ENTRY(99),
218 HNS3_RX_PTYPE_UNUSED_ENTRY(100),
219 HNS3_RX_PTYPE_UNUSED_ENTRY(101),
220 HNS3_RX_PTYPE_UNUSED_ENTRY(102),
221 HNS3_RX_PTYPE_UNUSED_ENTRY(103),
222 HNS3_RX_PTYPE_UNUSED_ENTRY(104),
223 HNS3_RX_PTYPE_UNUSED_ENTRY(105),
224 HNS3_RX_PTYPE_UNUSED_ENTRY(106),
225 HNS3_RX_PTYPE_UNUSED_ENTRY(107),
226 HNS3_RX_PTYPE_UNUSED_ENTRY(108),
227 HNS3_RX_PTYPE_UNUSED_ENTRY(109),
228 HNS3_RX_PTYPE_UNUSED_ENTRY(110),
229 HNS3_RX_PTYPE_ENTRY(111, 0, COMPLETE, IPV6),
230 HNS3_RX_PTYPE_ENTRY(112, 0, COMPLETE, IPV6),
231 HNS3_RX_PTYPE_ENTRY(113, 0, UNNECESSARY, IPV6),
232 HNS3_RX_PTYPE_ENTRY(114, 0, UNNECESSARY, IPV6),
233 HNS3_RX_PTYPE_ENTRY(115, 0, NONE, IPV6),
234 HNS3_RX_PTYPE_ENTRY(116, 0, UNNECESSARY, IPV6),
235 HNS3_RX_PTYPE_ENTRY(117, 0, NONE, IPV6),
236 HNS3_RX_PTYPE_ENTRY(118, 0, NONE, IPV6),
237 HNS3_RX_PTYPE_ENTRY(119, 0, UNNECESSARY, IPV6),
238 HNS3_RX_PTYPE_UNUSED_ENTRY(120),
239 HNS3_RX_PTYPE_UNUSED_ENTRY(121),
240 HNS3_RX_PTYPE_UNUSED_ENTRY(122),
241 HNS3_RX_PTYPE_ENTRY(123, 0, COMPLETE, PARSE_FAIL),
242 HNS3_RX_PTYPE_ENTRY(124, 0, COMPLETE, PARSE_FAIL),
243 HNS3_RX_PTYPE_ENTRY(125, 0, COMPLETE, IPV4),
244 HNS3_RX_PTYPE_ENTRY(126, 0, COMPLETE, IPV4),
245 HNS3_RX_PTYPE_ENTRY(127, 1, UNNECESSARY, IPV4),
246 HNS3_RX_PTYPE_ENTRY(128, 1, UNNECESSARY, IPV4),
247 HNS3_RX_PTYPE_ENTRY(129, 1, UNNECESSARY, IPV4),
248 HNS3_RX_PTYPE_ENTRY(130, 0, COMPLETE, IPV4),
249 HNS3_RX_PTYPE_ENTRY(131, 0, COMPLETE, IPV4),
250 HNS3_RX_PTYPE_UNUSED_ENTRY(132),
251 HNS3_RX_PTYPE_ENTRY(133, 0, COMPLETE, IPV6),
252 HNS3_RX_PTYPE_ENTRY(134, 0, COMPLETE, IPV6),
253 HNS3_RX_PTYPE_ENTRY(135, 1, UNNECESSARY, IPV6),
254 HNS3_RX_PTYPE_ENTRY(136, 1, UNNECESSARY, IPV6),
255 HNS3_RX_PTYPE_ENTRY(137, 1, UNNECESSARY, IPV6),
256 HNS3_RX_PTYPE_ENTRY(138, 0, COMPLETE, IPV6),
257 HNS3_RX_PTYPE_ENTRY(139, 0, COMPLETE, IPV6),
258 HNS3_RX_PTYPE_UNUSED_ENTRY(140),
259 HNS3_RX_PTYPE_UNUSED_ENTRY(141),
260 HNS3_RX_PTYPE_UNUSED_ENTRY(142),
261 HNS3_RX_PTYPE_UNUSED_ENTRY(143),
262 HNS3_RX_PTYPE_UNUSED_ENTRY(144),
263 HNS3_RX_PTYPE_UNUSED_ENTRY(145),
264 HNS3_RX_PTYPE_UNUSED_ENTRY(146),
265 HNS3_RX_PTYPE_UNUSED_ENTRY(147),
266 HNS3_RX_PTYPE_UNUSED_ENTRY(148),
267 HNS3_RX_PTYPE_UNUSED_ENTRY(149),
268 HNS3_RX_PTYPE_UNUSED_ENTRY(150),
269 HNS3_RX_PTYPE_UNUSED_ENTRY(151),
270 HNS3_RX_PTYPE_UNUSED_ENTRY(152),
271 HNS3_RX_PTYPE_UNUSED_ENTRY(153),
272 HNS3_RX_PTYPE_UNUSED_ENTRY(154),
273 HNS3_RX_PTYPE_UNUSED_ENTRY(155),
274 HNS3_RX_PTYPE_UNUSED_ENTRY(156),
275 HNS3_RX_PTYPE_UNUSED_ENTRY(157),
276 HNS3_RX_PTYPE_UNUSED_ENTRY(158),
277 HNS3_RX_PTYPE_UNUSED_ENTRY(159),
278 HNS3_RX_PTYPE_UNUSED_ENTRY(160),
279 HNS3_RX_PTYPE_UNUSED_ENTRY(161),
280 HNS3_RX_PTYPE_UNUSED_ENTRY(162),
281 HNS3_RX_PTYPE_UNUSED_ENTRY(163),
282 HNS3_RX_PTYPE_UNUSED_ENTRY(164),
283 HNS3_RX_PTYPE_UNUSED_ENTRY(165),
284 HNS3_RX_PTYPE_UNUSED_ENTRY(166),
285 HNS3_RX_PTYPE_UNUSED_ENTRY(167),
286 HNS3_RX_PTYPE_UNUSED_ENTRY(168),
287 HNS3_RX_PTYPE_UNUSED_ENTRY(169),
288 HNS3_RX_PTYPE_UNUSED_ENTRY(170),
289 HNS3_RX_PTYPE_UNUSED_ENTRY(171),
290 HNS3_RX_PTYPE_UNUSED_ENTRY(172),
291 HNS3_RX_PTYPE_UNUSED_ENTRY(173),
292 HNS3_RX_PTYPE_UNUSED_ENTRY(174),
293 HNS3_RX_PTYPE_UNUSED_ENTRY(175),
294 HNS3_RX_PTYPE_UNUSED_ENTRY(176),
295 HNS3_RX_PTYPE_UNUSED_ENTRY(177),
296 HNS3_RX_PTYPE_UNUSED_ENTRY(178),
297 HNS3_RX_PTYPE_UNUSED_ENTRY(179),
298 HNS3_RX_PTYPE_UNUSED_ENTRY(180),
299 HNS3_RX_PTYPE_UNUSED_ENTRY(181),
300 HNS3_RX_PTYPE_UNUSED_ENTRY(182),
301 HNS3_RX_PTYPE_UNUSED_ENTRY(183),
302 HNS3_RX_PTYPE_UNUSED_ENTRY(184),
303 HNS3_RX_PTYPE_UNUSED_ENTRY(185),
304 HNS3_RX_PTYPE_UNUSED_ENTRY(186),
305 HNS3_RX_PTYPE_UNUSED_ENTRY(187),
306 HNS3_RX_PTYPE_UNUSED_ENTRY(188),
307 HNS3_RX_PTYPE_UNUSED_ENTRY(189),
308 HNS3_RX_PTYPE_UNUSED_ENTRY(190),
309 HNS3_RX_PTYPE_UNUSED_ENTRY(191),
310 HNS3_RX_PTYPE_UNUSED_ENTRY(192),
311 HNS3_RX_PTYPE_UNUSED_ENTRY(193),
312 HNS3_RX_PTYPE_UNUSED_ENTRY(194),
313 HNS3_RX_PTYPE_UNUSED_ENTRY(195),
314 HNS3_RX_PTYPE_UNUSED_ENTRY(196),
315 HNS3_RX_PTYPE_UNUSED_ENTRY(197),
316 HNS3_RX_PTYPE_UNUSED_ENTRY(198),
317 HNS3_RX_PTYPE_UNUSED_ENTRY(199),
318 HNS3_RX_PTYPE_UNUSED_ENTRY(200),
319 HNS3_RX_PTYPE_UNUSED_ENTRY(201),
320 HNS3_RX_PTYPE_UNUSED_ENTRY(202),
321 HNS3_RX_PTYPE_UNUSED_ENTRY(203),
322 HNS3_RX_PTYPE_UNUSED_ENTRY(204),
323 HNS3_RX_PTYPE_UNUSED_ENTRY(205),
324 HNS3_RX_PTYPE_UNUSED_ENTRY(206),
325 HNS3_RX_PTYPE_UNUSED_ENTRY(207),
326 HNS3_RX_PTYPE_UNUSED_ENTRY(208),
327 HNS3_RX_PTYPE_UNUSED_ENTRY(209),
328 HNS3_RX_PTYPE_UNUSED_ENTRY(210),
329 HNS3_RX_PTYPE_UNUSED_ENTRY(211),
330 HNS3_RX_PTYPE_UNUSED_ENTRY(212),
331 HNS3_RX_PTYPE_UNUSED_ENTRY(213),
332 HNS3_RX_PTYPE_UNUSED_ENTRY(214),
333 HNS3_RX_PTYPE_UNUSED_ENTRY(215),
334 HNS3_RX_PTYPE_UNUSED_ENTRY(216),
335 HNS3_RX_PTYPE_UNUSED_ENTRY(217),
336 HNS3_RX_PTYPE_UNUSED_ENTRY(218),
337 HNS3_RX_PTYPE_UNUSED_ENTRY(219),
338 HNS3_RX_PTYPE_UNUSED_ENTRY(220),
339 HNS3_RX_PTYPE_UNUSED_ENTRY(221),
340 HNS3_RX_PTYPE_UNUSED_ENTRY(222),
341 HNS3_RX_PTYPE_UNUSED_ENTRY(223),
342 HNS3_RX_PTYPE_UNUSED_ENTRY(224),
343 HNS3_RX_PTYPE_UNUSED_ENTRY(225),
344 HNS3_RX_PTYPE_UNUSED_ENTRY(226),
345 HNS3_RX_PTYPE_UNUSED_ENTRY(227),
346 HNS3_RX_PTYPE_UNUSED_ENTRY(228),
347 HNS3_RX_PTYPE_UNUSED_ENTRY(229),
348 HNS3_RX_PTYPE_UNUSED_ENTRY(230),
349 HNS3_RX_PTYPE_UNUSED_ENTRY(231),
350 HNS3_RX_PTYPE_UNUSED_ENTRY(232),
351 HNS3_RX_PTYPE_UNUSED_ENTRY(233),
352 HNS3_RX_PTYPE_UNUSED_ENTRY(234),
353 HNS3_RX_PTYPE_UNUSED_ENTRY(235),
354 HNS3_RX_PTYPE_UNUSED_ENTRY(236),
355 HNS3_RX_PTYPE_UNUSED_ENTRY(237),
356 HNS3_RX_PTYPE_UNUSED_ENTRY(238),
357 HNS3_RX_PTYPE_UNUSED_ENTRY(239),
358 HNS3_RX_PTYPE_UNUSED_ENTRY(240),
359 HNS3_RX_PTYPE_UNUSED_ENTRY(241),
360 HNS3_RX_PTYPE_UNUSED_ENTRY(242),
361 HNS3_RX_PTYPE_UNUSED_ENTRY(243),
362 HNS3_RX_PTYPE_UNUSED_ENTRY(244),
363 HNS3_RX_PTYPE_UNUSED_ENTRY(245),
364 HNS3_RX_PTYPE_UNUSED_ENTRY(246),
365 HNS3_RX_PTYPE_UNUSED_ENTRY(247),
366 HNS3_RX_PTYPE_UNUSED_ENTRY(248),
367 HNS3_RX_PTYPE_UNUSED_ENTRY(249),
368 HNS3_RX_PTYPE_UNUSED_ENTRY(250),
369 HNS3_RX_PTYPE_UNUSED_ENTRY(251),
370 HNS3_RX_PTYPE_UNUSED_ENTRY(252),
371 HNS3_RX_PTYPE_UNUSED_ENTRY(253),
372 HNS3_RX_PTYPE_UNUSED_ENTRY(254),
373 HNS3_RX_PTYPE_UNUSED_ENTRY(255),
376 #define HNS3_INVALID_PTYPE \
377 ARRAY_SIZE(hns3_rx_ptype_tbl)
379 static irqreturn_t hns3_irq_handle(int irq, void *vector)
381 struct hns3_enet_tqp_vector *tqp_vector = vector;
383 napi_schedule_irqoff(&tqp_vector->napi);
384 tqp_vector->event_cnt++;
389 static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv)
391 struct hns3_enet_tqp_vector *tqp_vectors;
394 for (i = 0; i < priv->vector_num; i++) {
395 tqp_vectors = &priv->tqp_vector[i];
397 if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED)
400 /* clear the affinity mask */
401 irq_set_affinity_hint(tqp_vectors->vector_irq, NULL);
403 /* release the irq resource */
404 free_irq(tqp_vectors->vector_irq, tqp_vectors);
405 tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED;
409 static int hns3_nic_init_irq(struct hns3_nic_priv *priv)
411 struct hns3_enet_tqp_vector *tqp_vectors;
412 int txrx_int_idx = 0;
418 for (i = 0; i < priv->vector_num; i++) {
419 tqp_vectors = &priv->tqp_vector[i];
421 if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED)
424 if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) {
425 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
426 "%s-%s-%s-%d", hns3_driver_name,
427 pci_name(priv->ae_handle->pdev),
428 "TxRx", txrx_int_idx++);
430 } else if (tqp_vectors->rx_group.ring) {
431 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
432 "%s-%s-%s-%d", hns3_driver_name,
433 pci_name(priv->ae_handle->pdev),
435 } else if (tqp_vectors->tx_group.ring) {
436 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
437 "%s-%s-%s-%d", hns3_driver_name,
438 pci_name(priv->ae_handle->pdev),
441 /* Skip this unused q_vector */
445 tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0';
447 irq_set_status_flags(tqp_vectors->vector_irq, IRQ_NOAUTOEN);
448 ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0,
449 tqp_vectors->name, tqp_vectors);
451 netdev_err(priv->netdev, "request irq(%d) fail\n",
452 tqp_vectors->vector_irq);
453 hns3_nic_uninit_irq(priv);
457 irq_set_affinity_hint(tqp_vectors->vector_irq,
458 &tqp_vectors->affinity_mask);
460 tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED;
466 static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector,
469 writel(mask_en, tqp_vector->mask_addr);
472 static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector)
474 napi_enable(&tqp_vector->napi);
475 enable_irq(tqp_vector->vector_irq);
478 hns3_mask_vector_irq(tqp_vector, 1);
481 static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector)
484 hns3_mask_vector_irq(tqp_vector, 0);
486 disable_irq(tqp_vector->vector_irq);
487 napi_disable(&tqp_vector->napi);
488 cancel_work_sync(&tqp_vector->rx_group.dim.work);
489 cancel_work_sync(&tqp_vector->tx_group.dim.work);
492 void hns3_set_vector_coalesce_rl(struct hns3_enet_tqp_vector *tqp_vector,
495 u32 rl_reg = hns3_rl_usec_to_reg(rl_value);
497 /* this defines the configuration for RL (Interrupt Rate Limiter).
498 * Rl defines rate of interrupts i.e. number of interrupts-per-second
499 * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing
501 if (rl_reg > 0 && !tqp_vector->tx_group.coal.adapt_enable &&
502 !tqp_vector->rx_group.coal.adapt_enable)
503 /* According to the hardware, the range of rl_reg is
504 * 0-59 and the unit is 4.
506 rl_reg |= HNS3_INT_RL_ENABLE_MASK;
508 writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
511 void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
516 if (tqp_vector->rx_group.coal.unit_1us)
517 new_val = gl_value | HNS3_INT_GL_1US;
519 new_val = hns3_gl_usec_to_reg(gl_value);
521 writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
524 void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
529 if (tqp_vector->tx_group.coal.unit_1us)
530 new_val = gl_value | HNS3_INT_GL_1US;
532 new_val = hns3_gl_usec_to_reg(gl_value);
534 writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
537 void hns3_set_vector_coalesce_tx_ql(struct hns3_enet_tqp_vector *tqp_vector,
540 writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_TX_QL_OFFSET);
543 void hns3_set_vector_coalesce_rx_ql(struct hns3_enet_tqp_vector *tqp_vector,
546 writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_RX_QL_OFFSET);
549 static void hns3_vector_coalesce_init(struct hns3_enet_tqp_vector *tqp_vector,
550 struct hns3_nic_priv *priv)
552 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
553 struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
554 struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
555 struct hns3_enet_coalesce *ptx_coal = &priv->tx_coal;
556 struct hns3_enet_coalesce *prx_coal = &priv->rx_coal;
558 tx_coal->adapt_enable = ptx_coal->adapt_enable;
559 rx_coal->adapt_enable = prx_coal->adapt_enable;
561 tx_coal->int_gl = ptx_coal->int_gl;
562 rx_coal->int_gl = prx_coal->int_gl;
564 rx_coal->flow_level = prx_coal->flow_level;
565 tx_coal->flow_level = ptx_coal->flow_level;
567 /* device version above V3(include V3), GL can configure 1us
568 * unit, so uses 1us unit.
570 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3) {
571 tx_coal->unit_1us = 1;
572 rx_coal->unit_1us = 1;
575 if (ae_dev->dev_specs.int_ql_max) {
576 tx_coal->ql_enable = 1;
577 rx_coal->ql_enable = 1;
578 tx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
579 rx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
580 tx_coal->int_ql = ptx_coal->int_ql;
581 rx_coal->int_ql = prx_coal->int_ql;
586 hns3_vector_coalesce_init_hw(struct hns3_enet_tqp_vector *tqp_vector,
587 struct hns3_nic_priv *priv)
589 struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
590 struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
591 struct hnae3_handle *h = priv->ae_handle;
593 hns3_set_vector_coalesce_tx_gl(tqp_vector, tx_coal->int_gl);
594 hns3_set_vector_coalesce_rx_gl(tqp_vector, rx_coal->int_gl);
595 hns3_set_vector_coalesce_rl(tqp_vector, h->kinfo.int_rl_setting);
597 if (tx_coal->ql_enable)
598 hns3_set_vector_coalesce_tx_ql(tqp_vector, tx_coal->int_ql);
600 if (rx_coal->ql_enable)
601 hns3_set_vector_coalesce_rx_ql(tqp_vector, rx_coal->int_ql);
604 static int hns3_nic_set_real_num_queue(struct net_device *netdev)
606 struct hnae3_handle *h = hns3_get_handle(netdev);
607 struct hnae3_knic_private_info *kinfo = &h->kinfo;
608 struct hnae3_tc_info *tc_info = &kinfo->tc_info;
609 unsigned int queue_size = kinfo->num_tqps;
612 if (tc_info->num_tc <= 1 && !tc_info->mqprio_active) {
613 netdev_reset_tc(netdev);
615 ret = netdev_set_num_tc(netdev, tc_info->num_tc);
618 "netdev_set_num_tc fail, ret=%d!\n", ret);
622 for (i = 0; i < HNAE3_MAX_TC; i++) {
623 if (!test_bit(i, &tc_info->tc_en))
626 netdev_set_tc_queue(netdev, i, tc_info->tqp_count[i],
627 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 netif_carrier_off(netdev);
780 ret = hns3_nic_set_real_num_queue(netdev);
784 ret = hns3_nic_net_up(netdev);
786 netdev_err(netdev, "net up fail, ret=%d!\n", ret);
791 for (i = 0; i < HNAE3_MAX_USER_PRIO; i++)
792 netdev_set_prio_tc_map(netdev, i, kinfo->tc_info.prio_tc[i]);
794 if (h->ae_algo->ops->set_timer_task)
795 h->ae_algo->ops->set_timer_task(priv->ae_handle, true);
797 hns3_config_xps(priv);
799 netif_dbg(h, drv, netdev, "net open\n");
804 static void hns3_reset_tx_queue(struct hnae3_handle *h)
806 struct net_device *ndev = h->kinfo.netdev;
807 struct hns3_nic_priv *priv = netdev_priv(ndev);
808 struct netdev_queue *dev_queue;
811 for (i = 0; i < h->kinfo.num_tqps; i++) {
812 dev_queue = netdev_get_tx_queue(ndev,
813 priv->ring[i].queue_index);
814 netdev_tx_reset_queue(dev_queue);
818 static void hns3_nic_net_down(struct net_device *netdev)
820 struct hns3_nic_priv *priv = netdev_priv(netdev);
821 struct hnae3_handle *h = hns3_get_handle(netdev);
822 const struct hnae3_ae_ops *ops;
825 /* disable vectors */
826 for (i = 0; i < priv->vector_num; i++)
827 hns3_vector_disable(&priv->tqp_vector[i]);
830 for (i = 0; i < h->kinfo.num_tqps; i++)
831 hns3_tqp_disable(h->kinfo.tqp[i]);
834 ops = priv->ae_handle->ae_algo->ops;
836 ops->stop(priv->ae_handle);
838 /* delay ring buffer clearing to hns3_reset_notify_uninit_enet
839 * during reset process, because driver may not be able
840 * to disable the ring through firmware when downing the netdev.
842 if (!hns3_nic_resetting(netdev))
843 hns3_clear_all_ring(priv->ae_handle, false);
845 hns3_reset_tx_queue(priv->ae_handle);
848 static int hns3_nic_net_stop(struct net_device *netdev)
850 struct hns3_nic_priv *priv = netdev_priv(netdev);
851 struct hnae3_handle *h = hns3_get_handle(netdev);
853 if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
856 netif_dbg(h, drv, netdev, "net stop\n");
858 if (h->ae_algo->ops->set_timer_task)
859 h->ae_algo->ops->set_timer_task(priv->ae_handle, false);
861 netif_carrier_off(netdev);
862 netif_tx_disable(netdev);
864 hns3_nic_net_down(netdev);
869 static int hns3_nic_uc_sync(struct net_device *netdev,
870 const unsigned char *addr)
872 struct hnae3_handle *h = hns3_get_handle(netdev);
874 if (h->ae_algo->ops->add_uc_addr)
875 return h->ae_algo->ops->add_uc_addr(h, addr);
880 static int hns3_nic_uc_unsync(struct net_device *netdev,
881 const unsigned char *addr)
883 struct hnae3_handle *h = hns3_get_handle(netdev);
885 /* need ignore the request of removing device address, because
886 * we store the device address and other addresses of uc list
887 * in the function's mac filter list.
889 if (ether_addr_equal(addr, netdev->dev_addr))
892 if (h->ae_algo->ops->rm_uc_addr)
893 return h->ae_algo->ops->rm_uc_addr(h, addr);
898 static int hns3_nic_mc_sync(struct net_device *netdev,
899 const unsigned char *addr)
901 struct hnae3_handle *h = hns3_get_handle(netdev);
903 if (h->ae_algo->ops->add_mc_addr)
904 return h->ae_algo->ops->add_mc_addr(h, addr);
909 static int hns3_nic_mc_unsync(struct net_device *netdev,
910 const unsigned char *addr)
912 struct hnae3_handle *h = hns3_get_handle(netdev);
914 if (h->ae_algo->ops->rm_mc_addr)
915 return h->ae_algo->ops->rm_mc_addr(h, addr);
920 static u8 hns3_get_netdev_flags(struct net_device *netdev)
924 if (netdev->flags & IFF_PROMISC)
925 flags = HNAE3_USER_UPE | HNAE3_USER_MPE | HNAE3_BPE;
926 else if (netdev->flags & IFF_ALLMULTI)
927 flags = HNAE3_USER_MPE;
932 static void hns3_nic_set_rx_mode(struct net_device *netdev)
934 struct hnae3_handle *h = hns3_get_handle(netdev);
937 new_flags = hns3_get_netdev_flags(netdev);
939 __dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync);
940 __dev_mc_sync(netdev, hns3_nic_mc_sync, hns3_nic_mc_unsync);
942 /* User mode Promisc mode enable and vlan filtering is disabled to
943 * let all packets in.
945 h->netdev_flags = new_flags;
946 hns3_request_update_promisc_mode(h);
949 void hns3_request_update_promisc_mode(struct hnae3_handle *handle)
951 const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
953 if (ops->request_update_promisc_mode)
954 ops->request_update_promisc_mode(handle);
957 static u32 hns3_tx_spare_space(struct hns3_enet_ring *ring)
959 struct hns3_tx_spare *tx_spare = ring->tx_spare;
962 /* This smp_load_acquire() pairs with smp_store_release() in
963 * hns3_tx_spare_update() called in tx desc cleaning process.
965 ntc = smp_load_acquire(&tx_spare->last_to_clean);
966 ntu = tx_spare->next_to_use;
969 return ntc - ntu - 1;
971 /* The free tx buffer is divided into two part, so pick the
974 return max(ntc, tx_spare->len - ntu) - 1;
977 static void hns3_tx_spare_update(struct hns3_enet_ring *ring)
979 struct hns3_tx_spare *tx_spare = ring->tx_spare;
982 tx_spare->last_to_clean == tx_spare->next_to_clean)
985 /* This smp_store_release() pairs with smp_load_acquire() in
986 * hns3_tx_spare_space() called in xmit process.
988 smp_store_release(&tx_spare->last_to_clean,
989 tx_spare->next_to_clean);
992 static bool hns3_can_use_tx_bounce(struct hns3_enet_ring *ring,
996 u32 len = skb->len <= ring->tx_copybreak ? skb->len :
999 if (len > ring->tx_copybreak)
1002 if (ALIGN(len, dma_get_cache_alignment()) > space) {
1003 u64_stats_update_begin(&ring->syncp);
1004 ring->stats.tx_spare_full++;
1005 u64_stats_update_end(&ring->syncp);
1012 static bool hns3_can_use_tx_sgl(struct hns3_enet_ring *ring,
1013 struct sk_buff *skb,
1016 if (skb->len <= ring->tx_copybreak || !tx_sgl ||
1017 (!skb_has_frag_list(skb) &&
1018 skb_shinfo(skb)->nr_frags < tx_sgl))
1021 if (space < HNS3_MAX_SGL_SIZE) {
1022 u64_stats_update_begin(&ring->syncp);
1023 ring->stats.tx_spare_full++;
1024 u64_stats_update_end(&ring->syncp);
1031 static void hns3_init_tx_spare_buffer(struct hns3_enet_ring *ring)
1033 struct hns3_tx_spare *tx_spare;
1039 alloc_size = tx_spare_buf_size ? tx_spare_buf_size :
1040 ring->tqp->handle->kinfo.tx_spare_buf_size;
1044 order = get_order(alloc_size);
1045 tx_spare = devm_kzalloc(ring_to_dev(ring), sizeof(*tx_spare),
1048 /* The driver still work without the tx spare buffer */
1049 dev_warn(ring_to_dev(ring), "failed to allocate hns3_tx_spare\n");
1053 page = alloc_pages_node(dev_to_node(ring_to_dev(ring)),
1056 dev_warn(ring_to_dev(ring), "failed to allocate tx spare pages\n");
1057 devm_kfree(ring_to_dev(ring), tx_spare);
1061 dma = dma_map_page(ring_to_dev(ring), page, 0,
1062 PAGE_SIZE << order, DMA_TO_DEVICE);
1063 if (dma_mapping_error(ring_to_dev(ring), dma)) {
1064 dev_warn(ring_to_dev(ring), "failed to map pages for tx spare\n");
1066 devm_kfree(ring_to_dev(ring), tx_spare);
1070 tx_spare->dma = dma;
1071 tx_spare->buf = page_address(page);
1072 tx_spare->len = PAGE_SIZE << order;
1073 ring->tx_spare = tx_spare;
1076 /* Use hns3_tx_spare_space() to make sure there is enough buffer
1077 * before calling below function to allocate tx buffer.
1079 static void *hns3_tx_spare_alloc(struct hns3_enet_ring *ring,
1080 unsigned int size, dma_addr_t *dma,
1083 struct hns3_tx_spare *tx_spare = ring->tx_spare;
1084 u32 ntu = tx_spare->next_to_use;
1086 size = ALIGN(size, dma_get_cache_alignment());
1089 /* Tx spare buffer wraps back here because the end of
1090 * freed tx buffer is not enough.
1092 if (ntu + size > tx_spare->len) {
1093 *cb_len += (tx_spare->len - ntu);
1097 tx_spare->next_to_use = ntu + size;
1098 if (tx_spare->next_to_use == tx_spare->len)
1099 tx_spare->next_to_use = 0;
1101 *dma = tx_spare->dma + ntu;
1103 return tx_spare->buf + ntu;
1106 static void hns3_tx_spare_rollback(struct hns3_enet_ring *ring, u32 len)
1108 struct hns3_tx_spare *tx_spare = ring->tx_spare;
1110 if (len > tx_spare->next_to_use) {
1111 len -= tx_spare->next_to_use;
1112 tx_spare->next_to_use = tx_spare->len - len;
1114 tx_spare->next_to_use -= len;
1118 static void hns3_tx_spare_reclaim_cb(struct hns3_enet_ring *ring,
1119 struct hns3_desc_cb *cb)
1121 struct hns3_tx_spare *tx_spare = ring->tx_spare;
1122 u32 ntc = tx_spare->next_to_clean;
1123 u32 len = cb->length;
1125 tx_spare->next_to_clean += len;
1127 if (tx_spare->next_to_clean >= tx_spare->len) {
1128 tx_spare->next_to_clean -= tx_spare->len;
1130 if (tx_spare->next_to_clean) {
1132 len = tx_spare->next_to_clean;
1136 /* This tx spare buffer is only really reclaimed after calling
1137 * hns3_tx_spare_update(), so it is still safe to use the info in
1138 * the tx buffer to do the dma sync or sg unmapping after
1139 * tx_spare->next_to_clean is moved forword.
1141 if (cb->type & (DESC_TYPE_BOUNCE_HEAD | DESC_TYPE_BOUNCE_ALL)) {
1142 dma_addr_t dma = tx_spare->dma + ntc;
1144 dma_sync_single_for_cpu(ring_to_dev(ring), dma, len,
1147 struct sg_table *sgt = tx_spare->buf + ntc;
1149 dma_unmap_sg(ring_to_dev(ring), sgt->sgl, sgt->orig_nents,
1154 static int hns3_set_tso(struct sk_buff *skb, u32 *paylen_fdop_ol4cs,
1155 u16 *mss, u32 *type_cs_vlan_tso, u32 *send_bytes)
1157 u32 l4_offset, hdr_len;
1158 union l3_hdr_info l3;
1159 union l4_hdr_info l4;
1163 if (!skb_is_gso(skb))
1166 ret = skb_cow_head(skb, 0);
1167 if (unlikely(ret < 0))
1170 l3.hdr = skb_network_header(skb);
1171 l4.hdr = skb_transport_header(skb);
1173 /* Software should clear the IPv4's checksum field when tso is
1176 if (l3.v4->version == 4)
1180 if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
1182 SKB_GSO_UDP_TUNNEL |
1183 SKB_GSO_UDP_TUNNEL_CSUM)) {
1184 /* reset l3&l4 pointers from outer to inner headers */
1185 l3.hdr = skb_inner_network_header(skb);
1186 l4.hdr = skb_inner_transport_header(skb);
1188 /* Software should clear the IPv4's checksum field when
1191 if (l3.v4->version == 4)
1195 /* normal or tunnel packet */
1196 l4_offset = l4.hdr - skb->data;
1198 /* remove payload length from inner pseudo checksum when tso */
1199 l4_paylen = skb->len - l4_offset;
1201 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) {
1202 hdr_len = sizeof(*l4.udp) + l4_offset;
1203 csum_replace_by_diff(&l4.udp->check,
1204 (__force __wsum)htonl(l4_paylen));
1206 hdr_len = (l4.tcp->doff << 2) + l4_offset;
1207 csum_replace_by_diff(&l4.tcp->check,
1208 (__force __wsum)htonl(l4_paylen));
1211 *send_bytes = (skb_shinfo(skb)->gso_segs - 1) * hdr_len + skb->len;
1213 /* find the txbd field values */
1214 *paylen_fdop_ol4cs = skb->len - hdr_len;
1215 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_TSO_B, 1);
1217 /* offload outer UDP header checksum */
1218 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)
1219 hns3_set_field(*paylen_fdop_ol4cs, HNS3_TXD_OL4CS_B, 1);
1221 /* get MSS for TSO */
1222 *mss = skb_shinfo(skb)->gso_size;
1224 trace_hns3_tso(skb);
1229 static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
1232 union l3_hdr_info l3;
1233 unsigned char *l4_hdr;
1234 unsigned char *exthdr;
1238 /* find outer header point */
1239 l3.hdr = skb_network_header(skb);
1240 l4_hdr = skb_transport_header(skb);
1242 if (skb->protocol == htons(ETH_P_IPV6)) {
1243 exthdr = l3.hdr + sizeof(*l3.v6);
1244 l4_proto_tmp = l3.v6->nexthdr;
1245 if (l4_hdr != exthdr)
1246 ipv6_skip_exthdr(skb, exthdr - skb->data,
1247 &l4_proto_tmp, &frag_off);
1248 } else if (skb->protocol == htons(ETH_P_IP)) {
1249 l4_proto_tmp = l3.v4->protocol;
1254 *ol4_proto = l4_proto_tmp;
1257 if (!skb->encapsulation) {
1262 /* find inner header point */
1263 l3.hdr = skb_inner_network_header(skb);
1264 l4_hdr = skb_inner_transport_header(skb);
1266 if (l3.v6->version == 6) {
1267 exthdr = l3.hdr + sizeof(*l3.v6);
1268 l4_proto_tmp = l3.v6->nexthdr;
1269 if (l4_hdr != exthdr)
1270 ipv6_skip_exthdr(skb, exthdr - skb->data,
1271 &l4_proto_tmp, &frag_off);
1272 } else if (l3.v4->version == 4) {
1273 l4_proto_tmp = l3.v4->protocol;
1276 *il4_proto = l4_proto_tmp;
1281 /* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
1282 * and it is udp packet, which has a dest port as the IANA assigned.
1283 * the hardware is expected to do the checksum offload, but the
1284 * hardware will not do the checksum offload when udp dest port is
1285 * 4789, 4790 or 6081.
1287 static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
1289 struct hns3_nic_priv *priv = netdev_priv(skb->dev);
1290 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
1291 union l4_hdr_info l4;
1293 /* device version above V3(include V3), the hardware can
1294 * do this checksum offload.
1296 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
1299 l4.hdr = skb_transport_header(skb);
1301 if (!(!skb->encapsulation &&
1302 (l4.udp->dest == htons(IANA_VXLAN_UDP_PORT) ||
1303 l4.udp->dest == htons(GENEVE_UDP_PORT) ||
1304 l4.udp->dest == htons(4790))))
1310 static void hns3_set_outer_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
1311 u32 *ol_type_vlan_len_msec)
1313 u32 l2_len, l3_len, l4_len;
1314 unsigned char *il2_hdr;
1315 union l3_hdr_info l3;
1316 union l4_hdr_info l4;
1318 l3.hdr = skb_network_header(skb);
1319 l4.hdr = skb_transport_header(skb);
1321 /* compute OL2 header size, defined in 2 Bytes */
1322 l2_len = l3.hdr - skb->data;
1323 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L2LEN_S, l2_len >> 1);
1325 /* compute OL3 header size, defined in 4 Bytes */
1326 l3_len = l4.hdr - l3.hdr;
1327 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_S, l3_len >> 2);
1329 il2_hdr = skb_inner_mac_header(skb);
1330 /* compute OL4 header size, defined in 4 Bytes */
1331 l4_len = il2_hdr - l4.hdr;
1332 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L4LEN_S, l4_len >> 2);
1334 /* define outer network header type */
1335 if (skb->protocol == htons(ETH_P_IP)) {
1336 if (skb_is_gso(skb))
1337 hns3_set_field(*ol_type_vlan_len_msec,
1339 HNS3_OL3T_IPV4_CSUM);
1341 hns3_set_field(*ol_type_vlan_len_msec,
1343 HNS3_OL3T_IPV4_NO_CSUM);
1344 } else if (skb->protocol == htons(ETH_P_IPV6)) {
1345 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_S,
1349 if (ol4_proto == IPPROTO_UDP)
1350 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
1351 HNS3_TUN_MAC_IN_UDP);
1352 else if (ol4_proto == IPPROTO_GRE)
1353 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
1357 static int hns3_set_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
1358 u8 il4_proto, u32 *type_cs_vlan_tso,
1359 u32 *ol_type_vlan_len_msec)
1361 unsigned char *l2_hdr = skb->data;
1362 u32 l4_proto = ol4_proto;
1363 union l4_hdr_info l4;
1364 union l3_hdr_info l3;
1367 l4.hdr = skb_transport_header(skb);
1368 l3.hdr = skb_network_header(skb);
1370 /* handle encapsulation skb */
1371 if (skb->encapsulation) {
1372 /* If this is a not UDP/GRE encapsulation skb */
1373 if (!(ol4_proto == IPPROTO_UDP || ol4_proto == IPPROTO_GRE)) {
1374 /* drop the skb tunnel packet if hardware don't support,
1375 * because hardware can't calculate csum when TSO.
1377 if (skb_is_gso(skb))
1380 /* the stack computes the IP header already,
1381 * driver calculate l4 checksum when not TSO.
1383 return skb_checksum_help(skb);
1386 hns3_set_outer_l2l3l4(skb, ol4_proto, ol_type_vlan_len_msec);
1388 /* switch to inner header */
1389 l2_hdr = skb_inner_mac_header(skb);
1390 l3.hdr = skb_inner_network_header(skb);
1391 l4.hdr = skb_inner_transport_header(skb);
1392 l4_proto = il4_proto;
1395 if (l3.v4->version == 4) {
1396 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
1399 /* the stack computes the IP header already, the only time we
1400 * need the hardware to recompute it is in the case of TSO.
1402 if (skb_is_gso(skb))
1403 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3CS_B, 1);
1404 } else if (l3.v6->version == 6) {
1405 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
1409 /* compute inner(/normal) L2 header size, defined in 2 Bytes */
1410 l2_len = l3.hdr - l2_hdr;
1411 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S, l2_len >> 1);
1413 /* compute inner(/normal) L3 header size, defined in 4 Bytes */
1414 l3_len = l4.hdr - l3.hdr;
1415 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_S, l3_len >> 2);
1417 /* compute inner(/normal) L4 header size, defined in 4 Bytes */
1420 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1421 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1423 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1427 if (hns3_tunnel_csum_bug(skb))
1428 return skb_checksum_help(skb);
1430 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1431 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1433 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1434 (sizeof(struct udphdr) >> 2));
1437 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1438 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1440 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1441 (sizeof(struct sctphdr) >> 2));
1444 /* drop the skb tunnel packet if hardware don't support,
1445 * because hardware can't calculate csum when TSO.
1447 if (skb_is_gso(skb))
1450 /* the stack computes the IP header already,
1451 * driver calculate l4 checksum when not TSO.
1453 return skb_checksum_help(skb);
1459 static int hns3_handle_vtags(struct hns3_enet_ring *tx_ring,
1460 struct sk_buff *skb)
1462 struct hnae3_handle *handle = tx_ring->tqp->handle;
1463 struct hnae3_ae_dev *ae_dev;
1464 struct vlan_ethhdr *vhdr;
1467 if (!(skb->protocol == htons(ETH_P_8021Q) ||
1468 skb_vlan_tag_present(skb)))
1471 /* For HW limitation on HNAE3_DEVICE_VERSION_V2, if port based insert
1472 * VLAN enabled, only one VLAN header is allowed in skb, otherwise it
1473 * will cause RAS error.
1475 ae_dev = pci_get_drvdata(handle->pdev);
1476 if (unlikely(skb_vlan_tagged_multi(skb) &&
1477 ae_dev->dev_version <= HNAE3_DEVICE_VERSION_V2 &&
1478 handle->port_base_vlan_state ==
1479 HNAE3_PORT_BASE_VLAN_ENABLE))
1482 if (skb->protocol == htons(ETH_P_8021Q) &&
1483 !(handle->kinfo.netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
1484 /* When HW VLAN acceleration is turned off, and the stack
1485 * sets the protocol to 802.1q, the driver just need to
1486 * set the protocol to the encapsulated ethertype.
1488 skb->protocol = vlan_get_protocol(skb);
1492 if (skb_vlan_tag_present(skb)) {
1493 /* Based on hw strategy, use out_vtag in two layer tag case,
1494 * and use inner_vtag in one tag case.
1496 if (skb->protocol == htons(ETH_P_8021Q) &&
1497 handle->port_base_vlan_state ==
1498 HNAE3_PORT_BASE_VLAN_DISABLE)
1499 rc = HNS3_OUTER_VLAN_TAG;
1501 rc = HNS3_INNER_VLAN_TAG;
1503 skb->protocol = vlan_get_protocol(skb);
1507 rc = skb_cow_head(skb, 0);
1508 if (unlikely(rc < 0))
1511 vhdr = (struct vlan_ethhdr *)skb->data;
1512 vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority << VLAN_PRIO_SHIFT)
1515 skb->protocol = vlan_get_protocol(skb);
1519 /* check if the hardware is capable of checksum offloading */
1520 static bool hns3_check_hw_tx_csum(struct sk_buff *skb)
1522 struct hns3_nic_priv *priv = netdev_priv(skb->dev);
1524 /* Kindly note, due to backward compatibility of the TX descriptor,
1525 * HW checksum of the non-IP packets and GSO packets is handled at
1526 * different place in the following code
1528 if (skb_csum_is_sctp(skb) || skb_is_gso(skb) ||
1529 !test_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state))
1535 static int hns3_fill_skb_desc(struct hns3_enet_ring *ring,
1536 struct sk_buff *skb, struct hns3_desc *desc,
1537 struct hns3_desc_cb *desc_cb)
1539 u32 ol_type_vlan_len_msec = 0;
1540 u32 paylen_ol4cs = skb->len;
1541 u32 type_cs_vlan_tso = 0;
1542 u16 mss_hw_csum = 0;
1547 ret = hns3_handle_vtags(ring, skb);
1548 if (unlikely(ret < 0)) {
1549 u64_stats_update_begin(&ring->syncp);
1550 ring->stats.tx_vlan_err++;
1551 u64_stats_update_end(&ring->syncp);
1553 } else if (ret == HNS3_INNER_VLAN_TAG) {
1554 inner_vtag = skb_vlan_tag_get(skb);
1555 inner_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1557 hns3_set_field(type_cs_vlan_tso, HNS3_TXD_VLAN_B, 1);
1558 } else if (ret == HNS3_OUTER_VLAN_TAG) {
1559 out_vtag = skb_vlan_tag_get(skb);
1560 out_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1562 hns3_set_field(ol_type_vlan_len_msec, HNS3_TXD_OVLAN_B,
1566 desc_cb->send_bytes = skb->len;
1568 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1569 u8 ol4_proto, il4_proto;
1571 if (hns3_check_hw_tx_csum(skb)) {
1572 /* set checksum start and offset, defined in 2 Bytes */
1573 hns3_set_field(type_cs_vlan_tso, HNS3_TXD_CSUM_START_S,
1574 skb_checksum_start_offset(skb) >> 1);
1575 hns3_set_field(ol_type_vlan_len_msec,
1576 HNS3_TXD_CSUM_OFFSET_S,
1577 skb->csum_offset >> 1);
1578 mss_hw_csum |= BIT(HNS3_TXD_HW_CS_B);
1579 goto out_hw_tx_csum;
1582 skb_reset_mac_len(skb);
1584 ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
1585 if (unlikely(ret < 0)) {
1586 u64_stats_update_begin(&ring->syncp);
1587 ring->stats.tx_l4_proto_err++;
1588 u64_stats_update_end(&ring->syncp);
1592 ret = hns3_set_l2l3l4(skb, ol4_proto, il4_proto,
1594 &ol_type_vlan_len_msec);
1595 if (unlikely(ret < 0)) {
1596 u64_stats_update_begin(&ring->syncp);
1597 ring->stats.tx_l2l3l4_err++;
1598 u64_stats_update_end(&ring->syncp);
1602 ret = hns3_set_tso(skb, &paylen_ol4cs, &mss_hw_csum,
1603 &type_cs_vlan_tso, &desc_cb->send_bytes);
1604 if (unlikely(ret < 0)) {
1605 u64_stats_update_begin(&ring->syncp);
1606 ring->stats.tx_tso_err++;
1607 u64_stats_update_end(&ring->syncp);
1614 desc->tx.ol_type_vlan_len_msec =
1615 cpu_to_le32(ol_type_vlan_len_msec);
1616 desc->tx.type_cs_vlan_tso_len = cpu_to_le32(type_cs_vlan_tso);
1617 desc->tx.paylen_ol4cs = cpu_to_le32(paylen_ol4cs);
1618 desc->tx.mss_hw_csum = cpu_to_le16(mss_hw_csum);
1619 desc->tx.vlan_tag = cpu_to_le16(inner_vtag);
1620 desc->tx.outer_vlan_tag = cpu_to_le16(out_vtag);
1625 static int hns3_fill_desc(struct hns3_enet_ring *ring, dma_addr_t dma,
1628 #define HNS3_LIKELY_BD_NUM 1
1630 struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1631 unsigned int frag_buf_num;
1634 if (likely(size <= HNS3_MAX_BD_SIZE)) {
1635 desc->addr = cpu_to_le64(dma);
1636 desc->tx.send_size = cpu_to_le16(size);
1637 desc->tx.bdtp_fe_sc_vld_ra_ri =
1638 cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1640 trace_hns3_tx_desc(ring, ring->next_to_use);
1641 ring_ptr_move_fw(ring, next_to_use);
1642 return HNS3_LIKELY_BD_NUM;
1645 frag_buf_num = hns3_tx_bd_count(size);
1646 sizeoflast = size % HNS3_MAX_BD_SIZE;
1647 sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1649 /* When frag size is bigger than hardware limit, split this frag */
1650 for (k = 0; k < frag_buf_num; k++) {
1651 /* now, fill the descriptor */
1652 desc->addr = cpu_to_le64(dma + HNS3_MAX_BD_SIZE * k);
1653 desc->tx.send_size = cpu_to_le16((k == frag_buf_num - 1) ?
1654 (u16)sizeoflast : (u16)HNS3_MAX_BD_SIZE);
1655 desc->tx.bdtp_fe_sc_vld_ra_ri =
1656 cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1658 trace_hns3_tx_desc(ring, ring->next_to_use);
1659 /* move ring pointer to next */
1660 ring_ptr_move_fw(ring, next_to_use);
1662 desc = &ring->desc[ring->next_to_use];
1665 return frag_buf_num;
1668 static int hns3_map_and_fill_desc(struct hns3_enet_ring *ring, void *priv,
1671 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1672 struct device *dev = ring_to_dev(ring);
1676 if (type & (DESC_TYPE_FRAGLIST_SKB | DESC_TYPE_SKB)) {
1677 struct sk_buff *skb = (struct sk_buff *)priv;
1679 size = skb_headlen(skb);
1683 dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1684 } else if (type & DESC_TYPE_BOUNCE_HEAD) {
1685 /* Head data has been filled in hns3_handle_tx_bounce(),
1686 * just return 0 here.
1690 skb_frag_t *frag = (skb_frag_t *)priv;
1692 size = skb_frag_size(frag);
1696 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1699 if (unlikely(dma_mapping_error(dev, dma))) {
1700 u64_stats_update_begin(&ring->syncp);
1701 ring->stats.sw_err_cnt++;
1702 u64_stats_update_end(&ring->syncp);
1706 desc_cb->priv = priv;
1707 desc_cb->length = size;
1709 desc_cb->type = type;
1711 return hns3_fill_desc(ring, dma, size);
1714 static unsigned int hns3_skb_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1715 unsigned int bd_num)
1720 size = skb_headlen(skb);
1721 while (size > HNS3_MAX_BD_SIZE) {
1722 bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1723 size -= HNS3_MAX_BD_SIZE;
1725 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1730 bd_size[bd_num++] = size;
1731 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1735 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1736 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1737 size = skb_frag_size(frag);
1741 while (size > HNS3_MAX_BD_SIZE) {
1742 bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1743 size -= HNS3_MAX_BD_SIZE;
1745 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1749 bd_size[bd_num++] = size;
1750 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1757 static unsigned int hns3_tx_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1758 u8 max_non_tso_bd_num, unsigned int bd_num,
1759 unsigned int recursion_level)
1761 #define HNS3_MAX_RECURSION_LEVEL 24
1763 struct sk_buff *frag_skb;
1765 /* If the total len is within the max bd limit */
1766 if (likely(skb->len <= HNS3_MAX_BD_SIZE && !recursion_level &&
1767 !skb_has_frag_list(skb) &&
1768 skb_shinfo(skb)->nr_frags < max_non_tso_bd_num))
1769 return skb_shinfo(skb)->nr_frags + 1U;
1771 if (unlikely(recursion_level >= HNS3_MAX_RECURSION_LEVEL))
1774 bd_num = hns3_skb_bd_num(skb, bd_size, bd_num);
1775 if (!skb_has_frag_list(skb) || bd_num > HNS3_MAX_TSO_BD_NUM)
1778 skb_walk_frags(skb, frag_skb) {
1779 bd_num = hns3_tx_bd_num(frag_skb, bd_size, max_non_tso_bd_num,
1780 bd_num, recursion_level + 1);
1781 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1788 static unsigned int hns3_gso_hdr_len(struct sk_buff *skb)
1790 if (!skb->encapsulation)
1791 return skb_transport_offset(skb) + tcp_hdrlen(skb);
1793 return skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb);
1796 /* HW need every continuous max_non_tso_bd_num buffer data to be larger
1797 * than MSS, we simplify it by ensuring skb_headlen + the first continuous
1798 * max_non_tso_bd_num - 1 frags to be larger than gso header len + mss,
1799 * and the remaining continuous max_non_tso_bd_num - 1 frags to be larger
1800 * than MSS except the last max_non_tso_bd_num - 1 frags.
1802 static bool hns3_skb_need_linearized(struct sk_buff *skb, unsigned int *bd_size,
1803 unsigned int bd_num, u8 max_non_tso_bd_num)
1805 unsigned int tot_len = 0;
1808 for (i = 0; i < max_non_tso_bd_num - 1U; i++)
1809 tot_len += bd_size[i];
1811 /* ensure the first max_non_tso_bd_num frags is greater than
1814 if (tot_len + bd_size[max_non_tso_bd_num - 1U] <
1815 skb_shinfo(skb)->gso_size + hns3_gso_hdr_len(skb))
1818 /* ensure every continuous max_non_tso_bd_num - 1 buffer is greater
1819 * than mss except the last one.
1821 for (i = 0; i < bd_num - max_non_tso_bd_num; i++) {
1822 tot_len -= bd_size[i];
1823 tot_len += bd_size[i + max_non_tso_bd_num - 1U];
1825 if (tot_len < skb_shinfo(skb)->gso_size)
1832 void hns3_shinfo_pack(struct skb_shared_info *shinfo, __u32 *size)
1836 for (i = 0; i < MAX_SKB_FRAGS; i++)
1837 size[i] = skb_frag_size(&shinfo->frags[i]);
1840 static int hns3_skb_linearize(struct hns3_enet_ring *ring,
1841 struct sk_buff *skb,
1842 u8 max_non_tso_bd_num,
1843 unsigned int bd_num)
1845 /* 'bd_num == UINT_MAX' means the skb' fraglist has a
1846 * recursion level of over HNS3_MAX_RECURSION_LEVEL.
1848 if (bd_num == UINT_MAX) {
1849 u64_stats_update_begin(&ring->syncp);
1850 ring->stats.over_max_recursion++;
1851 u64_stats_update_end(&ring->syncp);
1855 /* The skb->len has exceeded the hw limitation, linearization
1858 if (skb->len > HNS3_MAX_TSO_SIZE ||
1859 (!skb_is_gso(skb) && skb->len >
1860 HNS3_MAX_NON_TSO_SIZE(max_non_tso_bd_num))) {
1861 u64_stats_update_begin(&ring->syncp);
1862 ring->stats.hw_limitation++;
1863 u64_stats_update_end(&ring->syncp);
1867 if (__skb_linearize(skb)) {
1868 u64_stats_update_begin(&ring->syncp);
1869 ring->stats.sw_err_cnt++;
1870 u64_stats_update_end(&ring->syncp);
1877 static int hns3_nic_maybe_stop_tx(struct hns3_enet_ring *ring,
1878 struct net_device *netdev,
1879 struct sk_buff *skb)
1881 struct hns3_nic_priv *priv = netdev_priv(netdev);
1882 u8 max_non_tso_bd_num = priv->max_non_tso_bd_num;
1883 unsigned int bd_size[HNS3_MAX_TSO_BD_NUM + 1U];
1884 unsigned int bd_num;
1886 bd_num = hns3_tx_bd_num(skb, bd_size, max_non_tso_bd_num, 0, 0);
1887 if (unlikely(bd_num > max_non_tso_bd_num)) {
1888 if (bd_num <= HNS3_MAX_TSO_BD_NUM && skb_is_gso(skb) &&
1889 !hns3_skb_need_linearized(skb, bd_size, bd_num,
1890 max_non_tso_bd_num)) {
1891 trace_hns3_over_max_bd(skb);
1895 if (hns3_skb_linearize(ring, skb, max_non_tso_bd_num,
1899 bd_num = hns3_tx_bd_count(skb->len);
1901 u64_stats_update_begin(&ring->syncp);
1902 ring->stats.tx_copy++;
1903 u64_stats_update_end(&ring->syncp);
1907 if (likely(ring_space(ring) >= bd_num))
1910 netif_stop_subqueue(netdev, ring->queue_index);
1911 smp_mb(); /* Memory barrier before checking ring_space */
1913 /* Start queue in case hns3_clean_tx_ring has just made room
1914 * available and has not seen the queue stopped state performed
1915 * by netif_stop_subqueue above.
1917 if (ring_space(ring) >= bd_num && netif_carrier_ok(netdev) &&
1918 !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
1919 netif_start_subqueue(netdev, ring->queue_index);
1923 u64_stats_update_begin(&ring->syncp);
1924 ring->stats.tx_busy++;
1925 u64_stats_update_end(&ring->syncp);
1930 static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
1932 struct device *dev = ring_to_dev(ring);
1935 for (i = 0; i < ring->desc_num; i++) {
1936 struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1937 struct hns3_desc_cb *desc_cb;
1939 memset(desc, 0, sizeof(*desc));
1941 /* check if this is where we started */
1942 if (ring->next_to_use == next_to_use_orig)
1946 ring_ptr_move_bw(ring, next_to_use);
1948 desc_cb = &ring->desc_cb[ring->next_to_use];
1953 /* unmap the descriptor dma address */
1954 if (desc_cb->type & (DESC_TYPE_SKB | DESC_TYPE_FRAGLIST_SKB))
1955 dma_unmap_single(dev, desc_cb->dma, desc_cb->length,
1957 else if (desc_cb->type &
1958 (DESC_TYPE_BOUNCE_HEAD | DESC_TYPE_BOUNCE_ALL))
1959 hns3_tx_spare_rollback(ring, desc_cb->length);
1960 else if (desc_cb->length)
1961 dma_unmap_page(dev, desc_cb->dma, desc_cb->length,
1964 desc_cb->length = 0;
1966 desc_cb->type = DESC_TYPE_UNKNOWN;
1970 static int hns3_fill_skb_to_desc(struct hns3_enet_ring *ring,
1971 struct sk_buff *skb, unsigned int type)
1973 struct sk_buff *frag_skb;
1974 int i, ret, bd_num = 0;
1976 ret = hns3_map_and_fill_desc(ring, skb, type);
1977 if (unlikely(ret < 0))
1982 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1983 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1985 ret = hns3_map_and_fill_desc(ring, frag, DESC_TYPE_PAGE);
1986 if (unlikely(ret < 0))
1992 skb_walk_frags(skb, frag_skb) {
1993 ret = hns3_fill_skb_to_desc(ring, frag_skb,
1994 DESC_TYPE_FRAGLIST_SKB);
1995 if (unlikely(ret < 0))
2004 static void hns3_tx_doorbell(struct hns3_enet_ring *ring, int num,
2007 ring->pending_buf += num;
2010 u64_stats_update_begin(&ring->syncp);
2011 ring->stats.tx_more++;
2012 u64_stats_update_end(&ring->syncp);
2016 if (!ring->pending_buf)
2019 writel(ring->pending_buf,
2020 ring->tqp->io_base + HNS3_RING_TX_RING_TAIL_REG);
2021 ring->pending_buf = 0;
2022 WRITE_ONCE(ring->last_to_use, ring->next_to_use);
2025 static void hns3_tsyn(struct net_device *netdev, struct sk_buff *skb,
2026 struct hns3_desc *desc)
2028 struct hnae3_handle *h = hns3_get_handle(netdev);
2030 if (!(h->ae_algo->ops->set_tx_hwts_info &&
2031 h->ae_algo->ops->set_tx_hwts_info(h, skb)))
2034 desc->tx.bdtp_fe_sc_vld_ra_ri |= cpu_to_le16(BIT(HNS3_TXD_TSYN_B));
2037 static int hns3_handle_tx_bounce(struct hns3_enet_ring *ring,
2038 struct sk_buff *skb)
2040 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
2041 unsigned int type = DESC_TYPE_BOUNCE_HEAD;
2042 unsigned int size = skb_headlen(skb);
2049 if (skb->len <= ring->tx_copybreak) {
2051 type = DESC_TYPE_BOUNCE_ALL;
2054 /* hns3_can_use_tx_bounce() is called to ensure the below
2055 * function can always return the tx buffer.
2057 buf = hns3_tx_spare_alloc(ring, size, &dma, &cb_len);
2059 ret = skb_copy_bits(skb, 0, buf, size);
2060 if (unlikely(ret < 0)) {
2061 hns3_tx_spare_rollback(ring, cb_len);
2062 u64_stats_update_begin(&ring->syncp);
2063 ring->stats.copy_bits_err++;
2064 u64_stats_update_end(&ring->syncp);
2068 desc_cb->priv = skb;
2069 desc_cb->length = cb_len;
2071 desc_cb->type = type;
2073 bd_num += hns3_fill_desc(ring, dma, size);
2075 if (type == DESC_TYPE_BOUNCE_HEAD) {
2076 ret = hns3_fill_skb_to_desc(ring, skb,
2077 DESC_TYPE_BOUNCE_HEAD);
2078 if (unlikely(ret < 0))
2084 dma_sync_single_for_device(ring_to_dev(ring), dma, size,
2087 u64_stats_update_begin(&ring->syncp);
2088 ring->stats.tx_bounce++;
2089 u64_stats_update_end(&ring->syncp);
2093 static int hns3_handle_tx_sgl(struct hns3_enet_ring *ring,
2094 struct sk_buff *skb)
2096 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
2097 u32 nfrag = skb_shinfo(skb)->nr_frags + 1;
2098 struct sg_table *sgt;
2104 if (skb_has_frag_list(skb))
2105 nfrag = HNS3_MAX_TSO_BD_NUM;
2107 /* hns3_can_use_tx_sgl() is called to ensure the below
2108 * function can always return the tx buffer.
2110 sgt = hns3_tx_spare_alloc(ring, HNS3_SGL_SIZE(nfrag),
2113 /* scatterlist follows by the sg table */
2114 sgt->sgl = (struct scatterlist *)(sgt + 1);
2115 sg_init_table(sgt->sgl, nfrag);
2116 nents = skb_to_sgvec(skb, sgt->sgl, 0, skb->len);
2117 if (unlikely(nents < 0)) {
2118 hns3_tx_spare_rollback(ring, cb_len);
2119 u64_stats_update_begin(&ring->syncp);
2120 ring->stats.skb2sgl_err++;
2121 u64_stats_update_end(&ring->syncp);
2125 sgt->orig_nents = nents;
2126 sgt->nents = dma_map_sg(ring_to_dev(ring), sgt->sgl, sgt->orig_nents,
2128 if (unlikely(!sgt->nents)) {
2129 hns3_tx_spare_rollback(ring, cb_len);
2130 u64_stats_update_begin(&ring->syncp);
2131 ring->stats.map_sg_err++;
2132 u64_stats_update_end(&ring->syncp);
2136 desc_cb->priv = skb;
2137 desc_cb->length = cb_len;
2139 desc_cb->type = DESC_TYPE_SGL_SKB;
2141 for (i = 0; i < sgt->nents; i++)
2142 bd_num += hns3_fill_desc(ring, sg_dma_address(sgt->sgl + i),
2143 sg_dma_len(sgt->sgl + i));
2145 u64_stats_update_begin(&ring->syncp);
2146 ring->stats.tx_sgl++;
2147 u64_stats_update_end(&ring->syncp);
2152 static int hns3_handle_desc_filling(struct hns3_enet_ring *ring,
2153 struct sk_buff *skb)
2157 if (!ring->tx_spare)
2160 space = hns3_tx_spare_space(ring);
2162 if (hns3_can_use_tx_sgl(ring, skb, space))
2163 return hns3_handle_tx_sgl(ring, skb);
2165 if (hns3_can_use_tx_bounce(ring, skb, space))
2166 return hns3_handle_tx_bounce(ring, skb);
2169 return hns3_fill_skb_to_desc(ring, skb, DESC_TYPE_SKB);
2172 netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
2174 struct hns3_nic_priv *priv = netdev_priv(netdev);
2175 struct hns3_enet_ring *ring = &priv->ring[skb->queue_mapping];
2176 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
2177 struct netdev_queue *dev_queue;
2178 int pre_ntu, next_to_use_head;
2182 /* Hardware can only handle short frames above 32 bytes */
2183 if (skb_put_padto(skb, HNS3_MIN_TX_LEN)) {
2184 hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
2186 u64_stats_update_begin(&ring->syncp);
2187 ring->stats.sw_err_cnt++;
2188 u64_stats_update_end(&ring->syncp);
2190 return NETDEV_TX_OK;
2193 /* Prefetch the data used later */
2194 prefetch(skb->data);
2196 ret = hns3_nic_maybe_stop_tx(ring, netdev, skb);
2197 if (unlikely(ret <= 0)) {
2198 if (ret == -EBUSY) {
2199 hns3_tx_doorbell(ring, 0, true);
2200 return NETDEV_TX_BUSY;
2203 hns3_rl_err(netdev, "xmit error: %d!\n", ret);
2207 next_to_use_head = ring->next_to_use;
2209 ret = hns3_fill_skb_desc(ring, skb, &ring->desc[ring->next_to_use],
2211 if (unlikely(ret < 0))
2214 /* 'ret < 0' means filling error, 'ret == 0' means skb->len is
2215 * zero, which is unlikely, and 'ret > 0' means how many tx desc
2216 * need to be notified to the hw.
2218 ret = hns3_handle_desc_filling(ring, skb);
2219 if (unlikely(ret <= 0))
2222 pre_ntu = ring->next_to_use ? (ring->next_to_use - 1) :
2223 (ring->desc_num - 1);
2225 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP))
2226 hns3_tsyn(netdev, skb, &ring->desc[pre_ntu]);
2228 ring->desc[pre_ntu].tx.bdtp_fe_sc_vld_ra_ri |=
2229 cpu_to_le16(BIT(HNS3_TXD_FE_B));
2230 trace_hns3_tx_desc(ring, pre_ntu);
2232 skb_tx_timestamp(skb);
2234 /* Complete translate all packets */
2235 dev_queue = netdev_get_tx_queue(netdev, ring->queue_index);
2236 doorbell = __netdev_tx_sent_queue(dev_queue, desc_cb->send_bytes,
2237 netdev_xmit_more());
2238 hns3_tx_doorbell(ring, ret, doorbell);
2240 return NETDEV_TX_OK;
2243 hns3_clear_desc(ring, next_to_use_head);
2246 dev_kfree_skb_any(skb);
2247 hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
2248 return NETDEV_TX_OK;
2251 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
2253 struct hnae3_handle *h = hns3_get_handle(netdev);
2254 struct sockaddr *mac_addr = p;
2257 if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
2258 return -EADDRNOTAVAIL;
2260 if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
2261 netdev_info(netdev, "already using mac address %pM\n",
2266 /* For VF device, if there is a perm_addr, then the user will not
2267 * be allowed to change the address.
2269 if (!hns3_is_phys_func(h->pdev) &&
2270 !is_zero_ether_addr(netdev->perm_addr)) {
2271 netdev_err(netdev, "has permanent MAC %pM, user MAC %pM not allow\n",
2272 netdev->perm_addr, mac_addr->sa_data);
2276 ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
2278 netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
2282 ether_addr_copy(netdev->dev_addr, mac_addr->sa_data);
2287 static int hns3_nic_do_ioctl(struct net_device *netdev,
2288 struct ifreq *ifr, int cmd)
2290 struct hnae3_handle *h = hns3_get_handle(netdev);
2292 if (!netif_running(netdev))
2295 if (!h->ae_algo->ops->do_ioctl)
2298 return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
2301 static int hns3_nic_set_features(struct net_device *netdev,
2302 netdev_features_t features)
2304 netdev_features_t changed = netdev->features ^ features;
2305 struct hns3_nic_priv *priv = netdev_priv(netdev);
2306 struct hnae3_handle *h = priv->ae_handle;
2310 if (changed & (NETIF_F_GRO_HW) && h->ae_algo->ops->set_gro_en) {
2311 enable = !!(features & NETIF_F_GRO_HW);
2312 ret = h->ae_algo->ops->set_gro_en(h, enable);
2317 if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
2318 h->ae_algo->ops->enable_hw_strip_rxvtag) {
2319 enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
2320 ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, enable);
2325 if ((changed & NETIF_F_NTUPLE) && h->ae_algo->ops->enable_fd) {
2326 enable = !!(features & NETIF_F_NTUPLE);
2327 h->ae_algo->ops->enable_fd(h, enable);
2330 if ((netdev->features & NETIF_F_HW_TC) > (features & NETIF_F_HW_TC) &&
2331 h->ae_algo->ops->cls_flower_active(h)) {
2333 "there are offloaded TC filters active, cannot disable HW TC offload");
2337 if ((changed & NETIF_F_HW_VLAN_CTAG_FILTER) &&
2338 h->ae_algo->ops->enable_vlan_filter) {
2339 enable = !!(features & NETIF_F_HW_VLAN_CTAG_FILTER);
2340 ret = h->ae_algo->ops->enable_vlan_filter(h, enable);
2345 netdev->features = features;
2349 static netdev_features_t hns3_features_check(struct sk_buff *skb,
2350 struct net_device *dev,
2351 netdev_features_t features)
2353 #define HNS3_MAX_HDR_LEN 480U
2354 #define HNS3_MAX_L4_HDR_LEN 60U
2358 if (skb->ip_summed != CHECKSUM_PARTIAL)
2361 if (skb->encapsulation)
2362 len = skb_inner_transport_header(skb) - skb->data;
2364 len = skb_transport_header(skb) - skb->data;
2366 /* Assume L4 is 60 byte as TCP is the only protocol with a
2367 * a flexible value, and it's max len is 60 bytes.
2369 len += HNS3_MAX_L4_HDR_LEN;
2371 /* Hardware only supports checksum on the skb with a max header
2374 if (len > HNS3_MAX_HDR_LEN)
2375 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2380 static void hns3_nic_get_stats64(struct net_device *netdev,
2381 struct rtnl_link_stats64 *stats)
2383 struct hns3_nic_priv *priv = netdev_priv(netdev);
2384 int queue_num = priv->ae_handle->kinfo.num_tqps;
2385 struct hnae3_handle *handle = priv->ae_handle;
2386 struct hns3_enet_ring *ring;
2387 u64 rx_length_errors = 0;
2388 u64 rx_crc_errors = 0;
2389 u64 rx_multicast = 0;
2401 if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
2404 handle->ae_algo->ops->update_stats(handle, &netdev->stats);
2406 for (idx = 0; idx < queue_num; idx++) {
2407 /* fetch the tx stats */
2408 ring = &priv->ring[idx];
2410 start = u64_stats_fetch_begin_irq(&ring->syncp);
2411 tx_bytes += ring->stats.tx_bytes;
2412 tx_pkts += ring->stats.tx_pkts;
2413 tx_drop += ring->stats.sw_err_cnt;
2414 tx_drop += ring->stats.tx_vlan_err;
2415 tx_drop += ring->stats.tx_l4_proto_err;
2416 tx_drop += ring->stats.tx_l2l3l4_err;
2417 tx_drop += ring->stats.tx_tso_err;
2418 tx_drop += ring->stats.over_max_recursion;
2419 tx_drop += ring->stats.hw_limitation;
2420 tx_drop += ring->stats.copy_bits_err;
2421 tx_drop += ring->stats.skb2sgl_err;
2422 tx_drop += ring->stats.map_sg_err;
2423 tx_errors += ring->stats.sw_err_cnt;
2424 tx_errors += ring->stats.tx_vlan_err;
2425 tx_errors += ring->stats.tx_l4_proto_err;
2426 tx_errors += ring->stats.tx_l2l3l4_err;
2427 tx_errors += ring->stats.tx_tso_err;
2428 tx_errors += ring->stats.over_max_recursion;
2429 tx_errors += ring->stats.hw_limitation;
2430 tx_errors += ring->stats.copy_bits_err;
2431 tx_errors += ring->stats.skb2sgl_err;
2432 tx_errors += ring->stats.map_sg_err;
2433 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
2435 /* fetch the rx stats */
2436 ring = &priv->ring[idx + queue_num];
2438 start = u64_stats_fetch_begin_irq(&ring->syncp);
2439 rx_bytes += ring->stats.rx_bytes;
2440 rx_pkts += ring->stats.rx_pkts;
2441 rx_drop += ring->stats.l2_err;
2442 rx_errors += ring->stats.l2_err;
2443 rx_errors += ring->stats.l3l4_csum_err;
2444 rx_crc_errors += ring->stats.l2_err;
2445 rx_multicast += ring->stats.rx_multicast;
2446 rx_length_errors += ring->stats.err_pkt_len;
2447 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
2450 stats->tx_bytes = tx_bytes;
2451 stats->tx_packets = tx_pkts;
2452 stats->rx_bytes = rx_bytes;
2453 stats->rx_packets = rx_pkts;
2455 stats->rx_errors = rx_errors;
2456 stats->multicast = rx_multicast;
2457 stats->rx_length_errors = rx_length_errors;
2458 stats->rx_crc_errors = rx_crc_errors;
2459 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
2461 stats->tx_errors = tx_errors;
2462 stats->rx_dropped = rx_drop;
2463 stats->tx_dropped = tx_drop;
2464 stats->collisions = netdev->stats.collisions;
2465 stats->rx_over_errors = netdev->stats.rx_over_errors;
2466 stats->rx_frame_errors = netdev->stats.rx_frame_errors;
2467 stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
2468 stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
2469 stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
2470 stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
2471 stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
2472 stats->tx_window_errors = netdev->stats.tx_window_errors;
2473 stats->rx_compressed = netdev->stats.rx_compressed;
2474 stats->tx_compressed = netdev->stats.tx_compressed;
2477 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
2479 struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
2480 struct hnae3_knic_private_info *kinfo;
2481 u8 tc = mqprio_qopt->qopt.num_tc;
2482 u16 mode = mqprio_qopt->mode;
2483 u8 hw = mqprio_qopt->qopt.hw;
2484 struct hnae3_handle *h;
2486 if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
2487 mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
2490 if (tc > HNAE3_MAX_TC)
2496 h = hns3_get_handle(netdev);
2499 netif_dbg(h, drv, netdev, "setup tc: num_tc=%u\n", tc);
2501 return (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
2502 kinfo->dcb_ops->setup_tc(h, mqprio_qopt) : -EOPNOTSUPP;
2505 static int hns3_setup_tc_cls_flower(struct hns3_nic_priv *priv,
2506 struct flow_cls_offload *flow)
2508 int tc = tc_classid_to_hwtc(priv->netdev, flow->classid);
2509 struct hnae3_handle *h = hns3_get_handle(priv->netdev);
2511 switch (flow->command) {
2512 case FLOW_CLS_REPLACE:
2513 if (h->ae_algo->ops->add_cls_flower)
2514 return h->ae_algo->ops->add_cls_flower(h, flow, tc);
2516 case FLOW_CLS_DESTROY:
2517 if (h->ae_algo->ops->del_cls_flower)
2518 return h->ae_algo->ops->del_cls_flower(h, flow);
2527 static int hns3_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
2530 struct hns3_nic_priv *priv = cb_priv;
2532 if (!tc_cls_can_offload_and_chain0(priv->netdev, type_data))
2536 case TC_SETUP_CLSFLOWER:
2537 return hns3_setup_tc_cls_flower(priv, type_data);
2543 static LIST_HEAD(hns3_block_cb_list);
2545 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
2548 struct hns3_nic_priv *priv = netdev_priv(dev);
2552 case TC_SETUP_QDISC_MQPRIO:
2553 ret = hns3_setup_tc(dev, type_data);
2555 case TC_SETUP_BLOCK:
2556 ret = flow_block_cb_setup_simple(type_data,
2557 &hns3_block_cb_list,
2558 hns3_setup_tc_block_cb,
2568 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
2569 __be16 proto, u16 vid)
2571 struct hnae3_handle *h = hns3_get_handle(netdev);
2574 if (h->ae_algo->ops->set_vlan_filter)
2575 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
2580 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
2581 __be16 proto, u16 vid)
2583 struct hnae3_handle *h = hns3_get_handle(netdev);
2586 if (h->ae_algo->ops->set_vlan_filter)
2587 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
2592 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
2593 u8 qos, __be16 vlan_proto)
2595 struct hnae3_handle *h = hns3_get_handle(netdev);
2598 netif_dbg(h, drv, netdev,
2599 "set vf vlan: vf=%d, vlan=%u, qos=%u, vlan_proto=0x%x\n",
2600 vf, vlan, qos, ntohs(vlan_proto));
2602 if (h->ae_algo->ops->set_vf_vlan_filter)
2603 ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
2609 static int hns3_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
2611 struct hnae3_handle *handle = hns3_get_handle(netdev);
2613 if (hns3_nic_resetting(netdev))
2616 if (!handle->ae_algo->ops->set_vf_spoofchk)
2619 return handle->ae_algo->ops->set_vf_spoofchk(handle, vf, enable);
2622 static int hns3_set_vf_trust(struct net_device *netdev, int vf, bool enable)
2624 struct hnae3_handle *handle = hns3_get_handle(netdev);
2626 if (!handle->ae_algo->ops->set_vf_trust)
2629 return handle->ae_algo->ops->set_vf_trust(handle, vf, enable);
2632 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
2634 struct hnae3_handle *h = hns3_get_handle(netdev);
2637 if (hns3_nic_resetting(netdev))
2640 if (!h->ae_algo->ops->set_mtu)
2643 netif_dbg(h, drv, netdev,
2644 "change mtu from %u to %d\n", netdev->mtu, new_mtu);
2646 ret = h->ae_algo->ops->set_mtu(h, new_mtu);
2648 netdev_err(netdev, "failed to change MTU in hardware %d\n",
2651 netdev->mtu = new_mtu;
2656 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
2658 struct hns3_nic_priv *priv = netdev_priv(ndev);
2659 struct hnae3_handle *h = hns3_get_handle(ndev);
2660 struct hns3_enet_ring *tx_ring;
2661 struct napi_struct *napi;
2662 int timeout_queue = 0;
2663 int hw_head, hw_tail;
2664 int fbd_num, fbd_oft;
2665 int ebd_num, ebd_oft;
2670 /* Find the stopped queue the same way the stack does */
2671 for (i = 0; i < ndev->num_tx_queues; i++) {
2672 struct netdev_queue *q;
2673 unsigned long trans_start;
2675 q = netdev_get_tx_queue(ndev, i);
2676 trans_start = q->trans_start;
2677 if (netif_xmit_stopped(q) &&
2679 (trans_start + ndev->watchdog_timeo))) {
2681 netdev_info(ndev, "queue state: 0x%lx, delta msecs: %u\n",
2683 jiffies_to_msecs(jiffies - trans_start));
2688 if (i == ndev->num_tx_queues) {
2690 "no netdev TX timeout queue found, timeout count: %llu\n",
2691 priv->tx_timeout_count);
2695 priv->tx_timeout_count++;
2697 tx_ring = &priv->ring[timeout_queue];
2698 napi = &tx_ring->tqp_vector->napi;
2701 "tx_timeout count: %llu, queue id: %d, SW_NTU: 0x%x, SW_NTC: 0x%x, napi state: %lu\n",
2702 priv->tx_timeout_count, timeout_queue, tx_ring->next_to_use,
2703 tx_ring->next_to_clean, napi->state);
2706 "tx_pkts: %llu, tx_bytes: %llu, sw_err_cnt: %llu, tx_pending: %d\n",
2707 tx_ring->stats.tx_pkts, tx_ring->stats.tx_bytes,
2708 tx_ring->stats.sw_err_cnt, tx_ring->pending_buf);
2711 "seg_pkt_cnt: %llu, tx_more: %llu, restart_queue: %llu, tx_busy: %llu\n",
2712 tx_ring->stats.seg_pkt_cnt, tx_ring->stats.tx_more,
2713 tx_ring->stats.restart_queue, tx_ring->stats.tx_busy);
2715 /* When mac received many pause frames continuous, it's unable to send
2716 * packets, which may cause tx timeout
2718 if (h->ae_algo->ops->get_mac_stats) {
2719 struct hns3_mac_stats mac_stats;
2721 h->ae_algo->ops->get_mac_stats(h, &mac_stats);
2722 netdev_info(ndev, "tx_pause_cnt: %llu, rx_pause_cnt: %llu\n",
2723 mac_stats.tx_pause_cnt, mac_stats.rx_pause_cnt);
2726 hw_head = readl_relaxed(tx_ring->tqp->io_base +
2727 HNS3_RING_TX_RING_HEAD_REG);
2728 hw_tail = readl_relaxed(tx_ring->tqp->io_base +
2729 HNS3_RING_TX_RING_TAIL_REG);
2730 fbd_num = readl_relaxed(tx_ring->tqp->io_base +
2731 HNS3_RING_TX_RING_FBDNUM_REG);
2732 fbd_oft = readl_relaxed(tx_ring->tqp->io_base +
2733 HNS3_RING_TX_RING_OFFSET_REG);
2734 ebd_num = readl_relaxed(tx_ring->tqp->io_base +
2735 HNS3_RING_TX_RING_EBDNUM_REG);
2736 ebd_oft = readl_relaxed(tx_ring->tqp->io_base +
2737 HNS3_RING_TX_RING_EBD_OFFSET_REG);
2738 bd_num = readl_relaxed(tx_ring->tqp->io_base +
2739 HNS3_RING_TX_RING_BD_NUM_REG);
2740 bd_err = readl_relaxed(tx_ring->tqp->io_base +
2741 HNS3_RING_TX_RING_BD_ERR_REG);
2742 ring_en = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_EN_REG);
2743 tc = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_TX_RING_TC_REG);
2746 "BD_NUM: 0x%x HW_HEAD: 0x%x, HW_TAIL: 0x%x, BD_ERR: 0x%x, INT: 0x%x\n",
2747 bd_num, hw_head, hw_tail, bd_err,
2748 readl(tx_ring->tqp_vector->mask_addr));
2750 "RING_EN: 0x%x, TC: 0x%x, FBD_NUM: 0x%x FBD_OFT: 0x%x, EBD_NUM: 0x%x, EBD_OFT: 0x%x\n",
2751 ring_en, tc, fbd_num, fbd_oft, ebd_num, ebd_oft);
2756 static void hns3_nic_net_timeout(struct net_device *ndev, unsigned int txqueue)
2758 struct hns3_nic_priv *priv = netdev_priv(ndev);
2759 struct hnae3_handle *h = priv->ae_handle;
2761 if (!hns3_get_tx_timeo_queue_info(ndev))
2764 /* request the reset, and let the hclge to determine
2765 * which reset level should be done
2767 if (h->ae_algo->ops->reset_event)
2768 h->ae_algo->ops->reset_event(h->pdev, h);
2771 #ifdef CONFIG_RFS_ACCEL
2772 static int hns3_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
2773 u16 rxq_index, u32 flow_id)
2775 struct hnae3_handle *h = hns3_get_handle(dev);
2776 struct flow_keys fkeys;
2778 if (!h->ae_algo->ops->add_arfs_entry)
2781 if (skb->encapsulation)
2782 return -EPROTONOSUPPORT;
2784 if (!skb_flow_dissect_flow_keys(skb, &fkeys, 0))
2785 return -EPROTONOSUPPORT;
2787 if ((fkeys.basic.n_proto != htons(ETH_P_IP) &&
2788 fkeys.basic.n_proto != htons(ETH_P_IPV6)) ||
2789 (fkeys.basic.ip_proto != IPPROTO_TCP &&
2790 fkeys.basic.ip_proto != IPPROTO_UDP))
2791 return -EPROTONOSUPPORT;
2793 return h->ae_algo->ops->add_arfs_entry(h, rxq_index, flow_id, &fkeys);
2797 static int hns3_nic_get_vf_config(struct net_device *ndev, int vf,
2798 struct ifla_vf_info *ivf)
2800 struct hnae3_handle *h = hns3_get_handle(ndev);
2802 if (!h->ae_algo->ops->get_vf_config)
2805 return h->ae_algo->ops->get_vf_config(h, vf, ivf);
2808 static int hns3_nic_set_vf_link_state(struct net_device *ndev, int vf,
2811 struct hnae3_handle *h = hns3_get_handle(ndev);
2813 if (!h->ae_algo->ops->set_vf_link_state)
2816 return h->ae_algo->ops->set_vf_link_state(h, vf, link_state);
2819 static int hns3_nic_set_vf_rate(struct net_device *ndev, int vf,
2820 int min_tx_rate, int max_tx_rate)
2822 struct hnae3_handle *h = hns3_get_handle(ndev);
2824 if (!h->ae_algo->ops->set_vf_rate)
2827 return h->ae_algo->ops->set_vf_rate(h, vf, min_tx_rate, max_tx_rate,
2831 static int hns3_nic_set_vf_mac(struct net_device *netdev, int vf_id, u8 *mac)
2833 struct hnae3_handle *h = hns3_get_handle(netdev);
2835 if (!h->ae_algo->ops->set_vf_mac)
2838 if (is_multicast_ether_addr(mac)) {
2840 "Invalid MAC:%pM specified. Could not set MAC\n",
2845 return h->ae_algo->ops->set_vf_mac(h, vf_id, mac);
2848 static const struct net_device_ops hns3_nic_netdev_ops = {
2849 .ndo_open = hns3_nic_net_open,
2850 .ndo_stop = hns3_nic_net_stop,
2851 .ndo_start_xmit = hns3_nic_net_xmit,
2852 .ndo_tx_timeout = hns3_nic_net_timeout,
2853 .ndo_set_mac_address = hns3_nic_net_set_mac_address,
2854 .ndo_eth_ioctl = hns3_nic_do_ioctl,
2855 .ndo_change_mtu = hns3_nic_change_mtu,
2856 .ndo_set_features = hns3_nic_set_features,
2857 .ndo_features_check = hns3_features_check,
2858 .ndo_get_stats64 = hns3_nic_get_stats64,
2859 .ndo_setup_tc = hns3_nic_setup_tc,
2860 .ndo_set_rx_mode = hns3_nic_set_rx_mode,
2861 .ndo_vlan_rx_add_vid = hns3_vlan_rx_add_vid,
2862 .ndo_vlan_rx_kill_vid = hns3_vlan_rx_kill_vid,
2863 .ndo_set_vf_vlan = hns3_ndo_set_vf_vlan,
2864 .ndo_set_vf_spoofchk = hns3_set_vf_spoofchk,
2865 .ndo_set_vf_trust = hns3_set_vf_trust,
2866 #ifdef CONFIG_RFS_ACCEL
2867 .ndo_rx_flow_steer = hns3_rx_flow_steer,
2869 .ndo_get_vf_config = hns3_nic_get_vf_config,
2870 .ndo_set_vf_link_state = hns3_nic_set_vf_link_state,
2871 .ndo_set_vf_rate = hns3_nic_set_vf_rate,
2872 .ndo_set_vf_mac = hns3_nic_set_vf_mac,
2875 bool hns3_is_phys_func(struct pci_dev *pdev)
2877 u32 dev_id = pdev->device;
2880 case HNAE3_DEV_ID_GE:
2881 case HNAE3_DEV_ID_25GE:
2882 case HNAE3_DEV_ID_25GE_RDMA:
2883 case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
2884 case HNAE3_DEV_ID_50GE_RDMA:
2885 case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
2886 case HNAE3_DEV_ID_100G_RDMA_MACSEC:
2887 case HNAE3_DEV_ID_200G_RDMA:
2889 case HNAE3_DEV_ID_VF:
2890 case HNAE3_DEV_ID_RDMA_DCB_PFC_VF:
2893 dev_warn(&pdev->dev, "un-recognized pci device-id %u",
2900 static void hns3_disable_sriov(struct pci_dev *pdev)
2902 /* If our VFs are assigned we cannot shut down SR-IOV
2903 * without causing issues, so just leave the hardware
2904 * available but disabled
2906 if (pci_vfs_assigned(pdev)) {
2907 dev_warn(&pdev->dev,
2908 "disabling driver while VFs are assigned\n");
2912 pci_disable_sriov(pdev);
2915 /* hns3_probe - Device initialization routine
2916 * @pdev: PCI device information struct
2917 * @ent: entry in hns3_pci_tbl
2919 * hns3_probe initializes a PF identified by a pci_dev structure.
2920 * The OS initialization, configuring of the PF private structure,
2921 * and a hardware reset occur.
2923 * Returns 0 on success, negative on failure
2925 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2927 struct hnae3_ae_dev *ae_dev;
2930 ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev), GFP_KERNEL);
2934 ae_dev->pdev = pdev;
2935 ae_dev->flag = ent->driver_data;
2936 pci_set_drvdata(pdev, ae_dev);
2938 ret = hnae3_register_ae_dev(ae_dev);
2940 pci_set_drvdata(pdev, NULL);
2945 /* hns3_remove - Device removal routine
2946 * @pdev: PCI device information struct
2948 static void hns3_remove(struct pci_dev *pdev)
2950 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2952 if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
2953 hns3_disable_sriov(pdev);
2955 hnae3_unregister_ae_dev(ae_dev);
2956 pci_set_drvdata(pdev, NULL);
2960 * hns3_pci_sriov_configure
2961 * @pdev: pointer to a pci_dev structure
2962 * @num_vfs: number of VFs to allocate
2964 * Enable or change the number of VFs. Called when the user updates the number
2967 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
2971 if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
2972 dev_warn(&pdev->dev, "Can not config SRIOV\n");
2977 ret = pci_enable_sriov(pdev, num_vfs);
2979 dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
2982 } else if (!pci_vfs_assigned(pdev)) {
2983 pci_disable_sriov(pdev);
2985 dev_warn(&pdev->dev,
2986 "Unable to free VFs because some are assigned to VMs.\n");
2992 static void hns3_shutdown(struct pci_dev *pdev)
2994 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2996 hnae3_unregister_ae_dev(ae_dev);
2997 pci_set_drvdata(pdev, NULL);
2999 if (system_state == SYSTEM_POWER_OFF)
3000 pci_set_power_state(pdev, PCI_D3hot);
3003 static int __maybe_unused hns3_suspend(struct device *dev)
3005 struct hnae3_ae_dev *ae_dev = dev_get_drvdata(dev);
3007 if (ae_dev && hns3_is_phys_func(ae_dev->pdev)) {
3008 dev_info(dev, "Begin to suspend.\n");
3009 if (ae_dev->ops && ae_dev->ops->reset_prepare)
3010 ae_dev->ops->reset_prepare(ae_dev, HNAE3_FUNC_RESET);
3016 static int __maybe_unused hns3_resume(struct device *dev)
3018 struct hnae3_ae_dev *ae_dev = dev_get_drvdata(dev);
3020 if (ae_dev && hns3_is_phys_func(ae_dev->pdev)) {
3021 dev_info(dev, "Begin to resume.\n");
3022 if (ae_dev->ops && ae_dev->ops->reset_done)
3023 ae_dev->ops->reset_done(ae_dev);
3029 static pci_ers_result_t hns3_error_detected(struct pci_dev *pdev,
3030 pci_channel_state_t state)
3032 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3033 pci_ers_result_t ret;
3035 dev_info(&pdev->dev, "PCI error detected, state(=%u)!!\n", state);
3037 if (state == pci_channel_io_perm_failure)
3038 return PCI_ERS_RESULT_DISCONNECT;
3040 if (!ae_dev || !ae_dev->ops) {
3042 "Can't recover - error happened before device initialized\n");
3043 return PCI_ERS_RESULT_NONE;
3046 if (ae_dev->ops->handle_hw_ras_error)
3047 ret = ae_dev->ops->handle_hw_ras_error(ae_dev);
3049 return PCI_ERS_RESULT_NONE;
3054 static pci_ers_result_t hns3_slot_reset(struct pci_dev *pdev)
3056 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3057 const struct hnae3_ae_ops *ops;
3058 enum hnae3_reset_type reset_type;
3059 struct device *dev = &pdev->dev;
3061 if (!ae_dev || !ae_dev->ops)
3062 return PCI_ERS_RESULT_NONE;
3065 /* request the reset */
3066 if (ops->reset_event && ops->get_reset_level &&
3067 ops->set_default_reset_request) {
3068 if (ae_dev->hw_err_reset_req) {
3069 reset_type = ops->get_reset_level(ae_dev,
3070 &ae_dev->hw_err_reset_req);
3071 ops->set_default_reset_request(ae_dev, reset_type);
3072 dev_info(dev, "requesting reset due to PCI error\n");
3073 ops->reset_event(pdev, NULL);
3076 return PCI_ERS_RESULT_RECOVERED;
3079 return PCI_ERS_RESULT_DISCONNECT;
3082 static void hns3_reset_prepare(struct pci_dev *pdev)
3084 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3086 dev_info(&pdev->dev, "FLR prepare\n");
3087 if (ae_dev && ae_dev->ops && ae_dev->ops->reset_prepare)
3088 ae_dev->ops->reset_prepare(ae_dev, HNAE3_FLR_RESET);
3091 static void hns3_reset_done(struct pci_dev *pdev)
3093 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3095 dev_info(&pdev->dev, "FLR done\n");
3096 if (ae_dev && ae_dev->ops && ae_dev->ops->reset_done)
3097 ae_dev->ops->reset_done(ae_dev);
3100 static const struct pci_error_handlers hns3_err_handler = {
3101 .error_detected = hns3_error_detected,
3102 .slot_reset = hns3_slot_reset,
3103 .reset_prepare = hns3_reset_prepare,
3104 .reset_done = hns3_reset_done,
3107 static SIMPLE_DEV_PM_OPS(hns3_pm_ops, hns3_suspend, hns3_resume);
3109 static struct pci_driver hns3_driver = {
3110 .name = hns3_driver_name,
3111 .id_table = hns3_pci_tbl,
3112 .probe = hns3_probe,
3113 .remove = hns3_remove,
3114 .shutdown = hns3_shutdown,
3115 .driver.pm = &hns3_pm_ops,
3116 .sriov_configure = hns3_pci_sriov_configure,
3117 .err_handler = &hns3_err_handler,
3120 /* set default feature to hns3 */
3121 static void hns3_set_default_feature(struct net_device *netdev)
3123 struct hnae3_handle *h = hns3_get_handle(netdev);
3124 struct pci_dev *pdev = h->pdev;
3125 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3127 netdev->priv_flags |= IFF_UNICAST_FLT;
3129 netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
3131 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
3132 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3133 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
3134 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
3135 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
3136 NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
3138 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V2) {
3139 netdev->features |= NETIF_F_GRO_HW;
3141 if (!(h->flags & HNAE3_SUPPORT_VF))
3142 netdev->features |= NETIF_F_NTUPLE;
3145 if (test_bit(HNAE3_DEV_SUPPORT_UDP_GSO_B, ae_dev->caps))
3146 netdev->features |= NETIF_F_GSO_UDP_L4;
3148 if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps))
3149 netdev->features |= NETIF_F_HW_CSUM;
3151 netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3153 if (test_bit(HNAE3_DEV_SUPPORT_UDP_TUNNEL_CSUM_B, ae_dev->caps))
3154 netdev->features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
3156 if (test_bit(HNAE3_DEV_SUPPORT_FD_FORWARD_TC_B, ae_dev->caps))
3157 netdev->features |= NETIF_F_HW_TC;
3159 netdev->hw_features |= netdev->features;
3160 if (!test_bit(HNAE3_DEV_SUPPORT_VLAN_FLTR_MDF_B, ae_dev->caps))
3161 netdev->hw_features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
3163 netdev->vlan_features |= netdev->features &
3164 ~(NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_CTAG_TX |
3165 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_GRO_HW | NETIF_F_NTUPLE |
3168 netdev->hw_enc_features |= netdev->vlan_features | NETIF_F_TSO_MANGLEID;
3171 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
3172 struct hns3_desc_cb *cb)
3174 unsigned int order = hns3_page_order(ring);
3177 if (ring->page_pool) {
3178 p = page_pool_dev_alloc_frag(ring->page_pool,
3180 hns3_buf_size(ring));
3185 cb->buf = page_address(p);
3186 cb->dma = page_pool_get_dma_addr(p);
3187 cb->type = DESC_TYPE_PP_FRAG;
3192 p = dev_alloc_pages(order);
3197 cb->page_offset = 0;
3199 cb->buf = page_address(p);
3200 cb->length = hns3_page_size(ring);
3201 cb->type = DESC_TYPE_PAGE;
3202 page_ref_add(p, USHRT_MAX - 1);
3203 cb->pagecnt_bias = USHRT_MAX;
3208 static void hns3_free_buffer(struct hns3_enet_ring *ring,
3209 struct hns3_desc_cb *cb, int budget)
3211 if (cb->type & (DESC_TYPE_SKB | DESC_TYPE_BOUNCE_HEAD |
3212 DESC_TYPE_BOUNCE_ALL | DESC_TYPE_SGL_SKB))
3213 napi_consume_skb(cb->priv, budget);
3214 else if (!HNAE3_IS_TX_RING(ring)) {
3215 if (cb->type & DESC_TYPE_PAGE && cb->pagecnt_bias)
3216 __page_frag_cache_drain(cb->priv, cb->pagecnt_bias);
3217 else if (cb->type & DESC_TYPE_PP_FRAG)
3218 page_pool_put_full_page(ring->page_pool, cb->priv,
3221 memset(cb, 0, sizeof(*cb));
3224 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
3226 cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
3227 cb->length, ring_to_dma_dir(ring));
3229 if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
3235 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
3236 struct hns3_desc_cb *cb)
3238 if (cb->type & (DESC_TYPE_SKB | DESC_TYPE_FRAGLIST_SKB))
3239 dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
3240 ring_to_dma_dir(ring));
3241 else if ((cb->type & DESC_TYPE_PAGE) && cb->length)
3242 dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
3243 ring_to_dma_dir(ring));
3244 else if (cb->type & (DESC_TYPE_BOUNCE_ALL | DESC_TYPE_BOUNCE_HEAD |
3246 hns3_tx_spare_reclaim_cb(ring, cb);
3249 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
3251 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
3252 ring->desc[i].addr = 0;
3255 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i,
3258 struct hns3_desc_cb *cb = &ring->desc_cb[i];
3260 if (!ring->desc_cb[i].dma)
3263 hns3_buffer_detach(ring, i);
3264 hns3_free_buffer(ring, cb, budget);
3267 static void hns3_free_buffers(struct hns3_enet_ring *ring)
3271 for (i = 0; i < ring->desc_num; i++)
3272 hns3_free_buffer_detach(ring, i, 0);
3275 /* free desc along with its attached buffer */
3276 static void hns3_free_desc(struct hns3_enet_ring *ring)
3278 int size = ring->desc_num * sizeof(ring->desc[0]);
3280 hns3_free_buffers(ring);
3283 dma_free_coherent(ring_to_dev(ring), size,
3284 ring->desc, ring->desc_dma_addr);
3289 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
3291 int size = ring->desc_num * sizeof(ring->desc[0]);
3293 ring->desc = dma_alloc_coherent(ring_to_dev(ring), size,
3294 &ring->desc_dma_addr, GFP_KERNEL);
3301 static int hns3_alloc_and_map_buffer(struct hns3_enet_ring *ring,
3302 struct hns3_desc_cb *cb)
3306 ret = hns3_alloc_buffer(ring, cb);
3307 if (ret || ring->page_pool)
3310 ret = hns3_map_buffer(ring, cb);
3317 hns3_free_buffer(ring, cb, 0);
3322 static int hns3_alloc_and_attach_buffer(struct hns3_enet_ring *ring, int i)
3324 int ret = hns3_alloc_and_map_buffer(ring, &ring->desc_cb[i]);
3329 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3330 ring->desc_cb[i].page_offset);
3335 /* Allocate memory for raw pkg, and map with dma */
3336 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
3340 for (i = 0; i < ring->desc_num; i++) {
3341 ret = hns3_alloc_and_attach_buffer(ring, i);
3343 goto out_buffer_fail;
3349 for (j = i - 1; j >= 0; j--)
3350 hns3_free_buffer_detach(ring, j, 0);
3354 /* detach a in-used buffer and replace with a reserved one */
3355 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
3356 struct hns3_desc_cb *res_cb)
3358 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
3359 ring->desc_cb[i] = *res_cb;
3360 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3361 ring->desc_cb[i].page_offset);
3362 ring->desc[i].rx.bd_base_info = 0;
3365 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
3367 ring->desc_cb[i].reuse_flag = 0;
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;
3372 dma_sync_single_for_device(ring_to_dev(ring),
3373 ring->desc_cb[i].dma + ring->desc_cb[i].page_offset,
3374 hns3_buf_size(ring),
3378 static bool hns3_nic_reclaim_desc(struct hns3_enet_ring *ring,
3379 int *bytes, int *pkts, int budget)
3381 /* pair with ring->last_to_use update in hns3_tx_doorbell(),
3382 * smp_store_release() is not used in hns3_tx_doorbell() because
3383 * the doorbell operation already have the needed barrier operation.
3385 int ltu = smp_load_acquire(&ring->last_to_use);
3386 int ntc = ring->next_to_clean;
3387 struct hns3_desc_cb *desc_cb;
3388 bool reclaimed = false;
3389 struct hns3_desc *desc;
3391 while (ltu != ntc) {
3392 desc = &ring->desc[ntc];
3394 if (le16_to_cpu(desc->tx.bdtp_fe_sc_vld_ra_ri) &
3395 BIT(HNS3_TXD_VLD_B))
3398 desc_cb = &ring->desc_cb[ntc];
3400 if (desc_cb->type & (DESC_TYPE_SKB | DESC_TYPE_BOUNCE_ALL |
3401 DESC_TYPE_BOUNCE_HEAD |
3402 DESC_TYPE_SGL_SKB)) {
3404 (*bytes) += desc_cb->send_bytes;
3407 /* desc_cb will be cleaned, after hnae3_free_buffer_detach */
3408 hns3_free_buffer_detach(ring, ntc, budget);
3410 if (++ntc == ring->desc_num)
3413 /* Issue prefetch for next Tx descriptor */
3414 prefetch(&ring->desc_cb[ntc]);
3418 if (unlikely(!reclaimed))
3421 /* This smp_store_release() pairs with smp_load_acquire() in
3422 * ring_space called by hns3_nic_net_xmit.
3424 smp_store_release(&ring->next_to_clean, ntc);
3426 hns3_tx_spare_update(ring);
3431 void hns3_clean_tx_ring(struct hns3_enet_ring *ring, int budget)
3433 struct net_device *netdev = ring_to_netdev(ring);
3434 struct hns3_nic_priv *priv = netdev_priv(netdev);
3435 struct netdev_queue *dev_queue;
3441 if (unlikely(!hns3_nic_reclaim_desc(ring, &bytes, &pkts, budget)))
3444 ring->tqp_vector->tx_group.total_bytes += bytes;
3445 ring->tqp_vector->tx_group.total_packets += pkts;
3447 u64_stats_update_begin(&ring->syncp);
3448 ring->stats.tx_bytes += bytes;
3449 ring->stats.tx_pkts += pkts;
3450 u64_stats_update_end(&ring->syncp);
3452 dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
3453 netdev_tx_completed_queue(dev_queue, pkts, bytes);
3455 if (unlikely(netif_carrier_ok(netdev) &&
3456 ring_space(ring) > HNS3_MAX_TSO_BD_NUM)) {
3457 /* Make sure that anybody stopping the queue after this
3458 * sees the new next_to_clean.
3461 if (netif_tx_queue_stopped(dev_queue) &&
3462 !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
3463 netif_tx_wake_queue(dev_queue);
3464 ring->stats.restart_queue++;
3469 static int hns3_desc_unused(struct hns3_enet_ring *ring)
3471 int ntc = ring->next_to_clean;
3472 int ntu = ring->next_to_use;
3474 return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
3477 static void hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring,
3480 struct hns3_desc_cb *desc_cb;
3481 struct hns3_desc_cb res_cbs;
3484 for (i = 0; i < cleand_count; i++) {
3485 desc_cb = &ring->desc_cb[ring->next_to_use];
3486 if (desc_cb->reuse_flag) {
3487 u64_stats_update_begin(&ring->syncp);
3488 ring->stats.reuse_pg_cnt++;
3489 u64_stats_update_end(&ring->syncp);
3491 hns3_reuse_buffer(ring, ring->next_to_use);
3493 ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
3495 u64_stats_update_begin(&ring->syncp);
3496 ring->stats.sw_err_cnt++;
3497 u64_stats_update_end(&ring->syncp);
3499 hns3_rl_err(ring_to_netdev(ring),
3500 "alloc rx buffer failed: %d\n",
3504 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
3506 u64_stats_update_begin(&ring->syncp);
3507 ring->stats.non_reuse_pg++;
3508 u64_stats_update_end(&ring->syncp);
3511 ring_ptr_move_fw(ring, next_to_use);
3514 writel(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
3517 static bool hns3_can_reuse_page(struct hns3_desc_cb *cb)
3519 return page_count(cb->priv) == cb->pagecnt_bias;
3522 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
3523 struct hns3_enet_ring *ring, int pull_len,
3524 struct hns3_desc_cb *desc_cb)
3526 struct hns3_desc *desc = &ring->desc[ring->next_to_clean];
3527 u32 frag_offset = desc_cb->page_offset + pull_len;
3528 int size = le16_to_cpu(desc->rx.size);
3529 u32 truesize = hns3_buf_size(ring);
3530 u32 frag_size = size - pull_len;
3533 if (ring->page_pool) {
3534 skb_add_rx_frag(skb, i, desc_cb->priv, frag_offset,
3535 frag_size, truesize);
3539 /* Avoid re-using remote or pfmem page */
3540 if (unlikely(!dev_page_is_reusable(desc_cb->priv)))
3543 reused = hns3_can_reuse_page(desc_cb);
3545 /* Rx page can be reused when:
3546 * 1. Rx page is only owned by the driver when page_offset
3547 * is zero, which means 0 @ truesize will be used by
3548 * stack after skb_add_rx_frag() is called, and the rest
3549 * of rx page can be reused by driver.
3551 * 2. Rx page is only owned by the driver when page_offset
3552 * is non-zero, which means page_offset @ truesize will
3553 * be used by stack after skb_add_rx_frag() is called,
3554 * and 0 @ truesize can be reused by driver.
3556 if ((!desc_cb->page_offset && reused) ||
3557 ((desc_cb->page_offset + truesize + truesize) <=
3558 hns3_page_size(ring) && desc_cb->page_offset)) {
3559 desc_cb->page_offset += truesize;
3560 desc_cb->reuse_flag = 1;
3561 } else if (desc_cb->page_offset && reused) {
3562 desc_cb->page_offset = 0;
3563 desc_cb->reuse_flag = 1;
3564 } else if (frag_size <= ring->rx_copybreak) {
3565 void *frag = napi_alloc_frag(frag_size);
3567 if (unlikely(!frag)) {
3568 u64_stats_update_begin(&ring->syncp);
3569 ring->stats.frag_alloc_err++;
3570 u64_stats_update_end(&ring->syncp);
3572 hns3_rl_err(ring_to_netdev(ring),
3573 "failed to allocate rx frag\n");
3577 desc_cb->reuse_flag = 1;
3578 memcpy(frag, desc_cb->buf + frag_offset, frag_size);
3579 skb_add_rx_frag(skb, i, virt_to_page(frag),
3580 offset_in_page(frag), frag_size, frag_size);
3582 u64_stats_update_begin(&ring->syncp);
3583 ring->stats.frag_alloc++;
3584 u64_stats_update_end(&ring->syncp);
3589 desc_cb->pagecnt_bias--;
3591 if (unlikely(!desc_cb->pagecnt_bias)) {
3592 page_ref_add(desc_cb->priv, USHRT_MAX);
3593 desc_cb->pagecnt_bias = USHRT_MAX;
3596 skb_add_rx_frag(skb, i, desc_cb->priv, frag_offset,
3597 frag_size, truesize);
3599 if (unlikely(!desc_cb->reuse_flag))
3600 __page_frag_cache_drain(desc_cb->priv, desc_cb->pagecnt_bias);
3603 static int hns3_gro_complete(struct sk_buff *skb, u32 l234info)
3605 __be16 type = skb->protocol;
3609 while (eth_type_vlan(type)) {
3610 struct vlan_hdr *vh;
3612 if ((depth + VLAN_HLEN) > skb_headlen(skb))
3615 vh = (struct vlan_hdr *)(skb->data + depth);
3616 type = vh->h_vlan_encapsulated_proto;
3620 skb_set_network_header(skb, depth);
3622 if (type == htons(ETH_P_IP)) {
3623 const struct iphdr *iph = ip_hdr(skb);
3625 depth += sizeof(struct iphdr);
3626 skb_set_transport_header(skb, depth);
3628 th->check = ~tcp_v4_check(skb->len - depth, iph->saddr,
3630 } else if (type == htons(ETH_P_IPV6)) {
3631 const struct ipv6hdr *iph = ipv6_hdr(skb);
3633 depth += sizeof(struct ipv6hdr);
3634 skb_set_transport_header(skb, depth);
3636 th->check = ~tcp_v6_check(skb->len - depth, &iph->saddr,
3639 hns3_rl_err(skb->dev,
3640 "Error: FW GRO supports only IPv4/IPv6, not 0x%04x, depth: %d\n",
3641 be16_to_cpu(type), depth);
3645 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
3647 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
3649 if (l234info & BIT(HNS3_RXD_GRO_FIXID_B))
3650 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_FIXEDID;
3652 skb->csum_start = (unsigned char *)th - skb->head;
3653 skb->csum_offset = offsetof(struct tcphdr, check);
3654 skb->ip_summed = CHECKSUM_PARTIAL;
3656 trace_hns3_gro(skb);
3661 static bool hns3_checksum_complete(struct hns3_enet_ring *ring,
3662 struct sk_buff *skb, u32 ptype, u16 csum)
3664 if (ptype == HNS3_INVALID_PTYPE ||
3665 hns3_rx_ptype_tbl[ptype].ip_summed != CHECKSUM_COMPLETE)
3668 u64_stats_update_begin(&ring->syncp);
3669 ring->stats.csum_complete++;
3670 u64_stats_update_end(&ring->syncp);
3671 skb->ip_summed = CHECKSUM_COMPLETE;
3672 skb->csum = csum_unfold((__force __sum16)csum);
3677 static void hns3_rx_handle_csum(struct sk_buff *skb, u32 l234info,
3678 u32 ol_info, u32 ptype)
3680 int l3_type, l4_type;
3683 if (ptype != HNS3_INVALID_PTYPE) {
3684 skb->csum_level = hns3_rx_ptype_tbl[ptype].csum_level;
3685 skb->ip_summed = hns3_rx_ptype_tbl[ptype].ip_summed;
3690 ol4_type = hnae3_get_field(ol_info, HNS3_RXD_OL4ID_M,
3693 case HNS3_OL4_TYPE_MAC_IN_UDP:
3694 case HNS3_OL4_TYPE_NVGRE:
3695 skb->csum_level = 1;
3697 case HNS3_OL4_TYPE_NO_TUN:
3698 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
3700 l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
3702 /* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
3703 if ((l3_type == HNS3_L3_TYPE_IPV4 ||
3704 l3_type == HNS3_L3_TYPE_IPV6) &&
3705 (l4_type == HNS3_L4_TYPE_UDP ||
3706 l4_type == HNS3_L4_TYPE_TCP ||
3707 l4_type == HNS3_L4_TYPE_SCTP))
3708 skb->ip_summed = CHECKSUM_UNNECESSARY;
3715 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
3716 u32 l234info, u32 bd_base_info, u32 ol_info,
3719 struct net_device *netdev = ring_to_netdev(ring);
3720 struct hns3_nic_priv *priv = netdev_priv(netdev);
3721 u32 ptype = HNS3_INVALID_PTYPE;
3723 skb->ip_summed = CHECKSUM_NONE;
3725 skb_checksum_none_assert(skb);
3727 if (!(netdev->features & NETIF_F_RXCSUM))
3730 if (test_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state))
3731 ptype = hnae3_get_field(ol_info, HNS3_RXD_PTYPE_M,
3734 if (hns3_checksum_complete(ring, skb, ptype, csum))
3737 /* check if hardware has done checksum */
3738 if (!(bd_base_info & BIT(HNS3_RXD_L3L4P_B)))
3741 if (unlikely(l234info & (BIT(HNS3_RXD_L3E_B) | BIT(HNS3_RXD_L4E_B) |
3742 BIT(HNS3_RXD_OL3E_B) |
3743 BIT(HNS3_RXD_OL4E_B)))) {
3744 u64_stats_update_begin(&ring->syncp);
3745 ring->stats.l3l4_csum_err++;
3746 u64_stats_update_end(&ring->syncp);
3751 hns3_rx_handle_csum(skb, l234info, ol_info, ptype);
3754 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
3756 if (skb_has_frag_list(skb))
3757 napi_gro_flush(&ring->tqp_vector->napi, false);
3759 napi_gro_receive(&ring->tqp_vector->napi, skb);
3762 static bool hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
3763 struct hns3_desc *desc, u32 l234info,
3766 struct hnae3_handle *handle = ring->tqp->handle;
3767 struct pci_dev *pdev = ring->tqp->handle->pdev;
3768 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3770 if (unlikely(ae_dev->dev_version < HNAE3_DEVICE_VERSION_V2)) {
3771 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3772 if (!(*vlan_tag & VLAN_VID_MASK))
3773 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3775 return (*vlan_tag != 0);
3778 #define HNS3_STRP_OUTER_VLAN 0x1
3779 #define HNS3_STRP_INNER_VLAN 0x2
3780 #define HNS3_STRP_BOTH 0x3
3782 /* Hardware always insert VLAN tag into RX descriptor when
3783 * remove the tag from packet, driver needs to determine
3784 * reporting which tag to stack.
3786 switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
3787 HNS3_RXD_STRP_TAGP_S)) {
3788 case HNS3_STRP_OUTER_VLAN:
3789 if (handle->port_base_vlan_state !=
3790 HNAE3_PORT_BASE_VLAN_DISABLE)
3793 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3795 case HNS3_STRP_INNER_VLAN:
3796 if (handle->port_base_vlan_state !=
3797 HNAE3_PORT_BASE_VLAN_DISABLE)
3800 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3802 case HNS3_STRP_BOTH:
3803 if (handle->port_base_vlan_state ==
3804 HNAE3_PORT_BASE_VLAN_DISABLE)
3805 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3807 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3815 static void hns3_rx_ring_move_fw(struct hns3_enet_ring *ring)
3817 ring->desc[ring->next_to_clean].rx.bd_base_info &=
3818 cpu_to_le32(~BIT(HNS3_RXD_VLD_B));
3819 ring->next_to_clean += 1;
3821 if (unlikely(ring->next_to_clean == ring->desc_num))
3822 ring->next_to_clean = 0;
3825 static int hns3_alloc_skb(struct hns3_enet_ring *ring, unsigned int length,
3828 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
3829 struct net_device *netdev = ring_to_netdev(ring);
3830 struct sk_buff *skb;
3832 ring->skb = napi_alloc_skb(&ring->tqp_vector->napi, HNS3_RX_HEAD_SIZE);
3834 if (unlikely(!skb)) {
3835 hns3_rl_err(netdev, "alloc rx skb fail\n");
3837 u64_stats_update_begin(&ring->syncp);
3838 ring->stats.sw_err_cnt++;
3839 u64_stats_update_end(&ring->syncp);
3844 trace_hns3_rx_desc(ring);
3845 prefetchw(skb->data);
3847 ring->pending_buf = 1;
3849 ring->tail_skb = NULL;
3850 if (length <= HNS3_RX_HEAD_SIZE) {
3851 memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
3853 /* We can reuse buffer as-is, just make sure it is reusable */
3854 if (dev_page_is_reusable(desc_cb->priv))
3855 desc_cb->reuse_flag = 1;
3856 else if (desc_cb->type & DESC_TYPE_PP_FRAG)
3857 page_pool_put_full_page(ring->page_pool, desc_cb->priv,
3859 else /* This page cannot be reused so discard it */
3860 __page_frag_cache_drain(desc_cb->priv,
3861 desc_cb->pagecnt_bias);
3863 hns3_rx_ring_move_fw(ring);
3867 if (ring->page_pool)
3868 skb_mark_for_recycle(skb);
3870 u64_stats_update_begin(&ring->syncp);
3871 ring->stats.seg_pkt_cnt++;
3872 u64_stats_update_end(&ring->syncp);
3874 ring->pull_len = eth_get_headlen(netdev, va, HNS3_RX_HEAD_SIZE);
3875 __skb_put(skb, ring->pull_len);
3876 hns3_nic_reuse_page(skb, ring->frag_num++, ring, ring->pull_len,
3878 hns3_rx_ring_move_fw(ring);
3883 static int hns3_add_frag(struct hns3_enet_ring *ring)
3885 struct sk_buff *skb = ring->skb;
3886 struct sk_buff *head_skb = skb;
3887 struct sk_buff *new_skb;
3888 struct hns3_desc_cb *desc_cb;
3889 struct hns3_desc *desc;
3893 desc = &ring->desc[ring->next_to_clean];
3894 desc_cb = &ring->desc_cb[ring->next_to_clean];
3895 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
3896 /* make sure HW write desc complete */
3898 if (!(bd_base_info & BIT(HNS3_RXD_VLD_B)))
3901 if (unlikely(ring->frag_num >= MAX_SKB_FRAGS)) {
3902 new_skb = napi_alloc_skb(&ring->tqp_vector->napi, 0);
3903 if (unlikely(!new_skb)) {
3904 hns3_rl_err(ring_to_netdev(ring),
3905 "alloc rx fraglist skb fail\n");
3909 if (ring->page_pool)
3910 skb_mark_for_recycle(new_skb);
3914 if (ring->tail_skb) {
3915 ring->tail_skb->next = new_skb;
3916 ring->tail_skb = new_skb;
3918 skb_shinfo(skb)->frag_list = new_skb;
3919 ring->tail_skb = new_skb;
3923 if (ring->tail_skb) {
3924 head_skb->truesize += hns3_buf_size(ring);
3925 head_skb->data_len += le16_to_cpu(desc->rx.size);
3926 head_skb->len += le16_to_cpu(desc->rx.size);
3927 skb = ring->tail_skb;
3930 dma_sync_single_for_cpu(ring_to_dev(ring),
3931 desc_cb->dma + desc_cb->page_offset,
3932 hns3_buf_size(ring),
3935 hns3_nic_reuse_page(skb, ring->frag_num++, ring, 0, desc_cb);
3936 trace_hns3_rx_desc(ring);
3937 hns3_rx_ring_move_fw(ring);
3938 ring->pending_buf++;
3939 } while (!(bd_base_info & BIT(HNS3_RXD_FE_B)));
3944 static int hns3_set_gro_and_checksum(struct hns3_enet_ring *ring,
3945 struct sk_buff *skb, u32 l234info,
3946 u32 bd_base_info, u32 ol_info, u16 csum)
3948 struct net_device *netdev = ring_to_netdev(ring);
3949 struct hns3_nic_priv *priv = netdev_priv(netdev);
3952 skb_shinfo(skb)->gso_size = hnae3_get_field(bd_base_info,
3953 HNS3_RXD_GRO_SIZE_M,
3954 HNS3_RXD_GRO_SIZE_S);
3955 /* if there is no HW GRO, do not set gro params */
3956 if (!skb_shinfo(skb)->gso_size) {
3957 hns3_rx_checksum(ring, skb, l234info, bd_base_info, ol_info,
3962 NAPI_GRO_CB(skb)->count = hnae3_get_field(l234info,
3963 HNS3_RXD_GRO_COUNT_M,
3964 HNS3_RXD_GRO_COUNT_S);
3966 if (test_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state)) {
3967 u32 ptype = hnae3_get_field(ol_info, HNS3_RXD_PTYPE_M,
3970 l3_type = hns3_rx_ptype_tbl[ptype].l3_type;
3972 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
3976 if (l3_type == HNS3_L3_TYPE_IPV4)
3977 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
3978 else if (l3_type == HNS3_L3_TYPE_IPV6)
3979 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
3983 return hns3_gro_complete(skb, l234info);
3986 static void hns3_set_rx_skb_rss_type(struct hns3_enet_ring *ring,
3987 struct sk_buff *skb, u32 rss_hash)
3989 struct hnae3_handle *handle = ring->tqp->handle;
3990 enum pkt_hash_types rss_type;
3993 rss_type = handle->kinfo.rss_type;
3995 rss_type = PKT_HASH_TYPE_NONE;
3997 skb_set_hash(skb, rss_hash, rss_type);
4000 static int hns3_handle_bdinfo(struct hns3_enet_ring *ring, struct sk_buff *skb)
4002 struct net_device *netdev = ring_to_netdev(ring);
4003 enum hns3_pkt_l2t_type l2_frame_type;
4004 u32 bd_base_info, l234info, ol_info;
4005 struct hns3_desc *desc;
4010 /* bdinfo handled below is only valid on the last BD of the
4011 * current packet, and ring->next_to_clean indicates the first
4012 * descriptor of next packet, so need - 1 below.
4014 pre_ntc = ring->next_to_clean ? (ring->next_to_clean - 1) :
4015 (ring->desc_num - 1);
4016 desc = &ring->desc[pre_ntc];
4017 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
4018 l234info = le32_to_cpu(desc->rx.l234_info);
4019 ol_info = le32_to_cpu(desc->rx.ol_info);
4020 csum = le16_to_cpu(desc->csum);
4022 if (unlikely(bd_base_info & BIT(HNS3_RXD_TS_VLD_B))) {
4023 struct hnae3_handle *h = hns3_get_handle(netdev);
4024 u32 nsec = le32_to_cpu(desc->ts_nsec);
4025 u32 sec = le32_to_cpu(desc->ts_sec);
4027 if (h->ae_algo->ops->get_rx_hwts)
4028 h->ae_algo->ops->get_rx_hwts(h, skb, nsec, sec);
4031 /* Based on hw strategy, the tag offloaded will be stored at
4032 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
4033 * in one layer tag case.
4035 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
4038 if (hns3_parse_vlan_tag(ring, desc, l234info, &vlan_tag))
4039 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
4043 if (unlikely(!desc->rx.pkt_len || (l234info & (BIT(HNS3_RXD_TRUNCAT_B) |
4044 BIT(HNS3_RXD_L2E_B))))) {
4045 u64_stats_update_begin(&ring->syncp);
4046 if (l234info & BIT(HNS3_RXD_L2E_B))
4047 ring->stats.l2_err++;
4049 ring->stats.err_pkt_len++;
4050 u64_stats_update_end(&ring->syncp);
4057 /* Do update ip stack process */
4058 skb->protocol = eth_type_trans(skb, netdev);
4060 /* This is needed in order to enable forwarding support */
4061 ret = hns3_set_gro_and_checksum(ring, skb, l234info,
4062 bd_base_info, ol_info, csum);
4063 if (unlikely(ret)) {
4064 u64_stats_update_begin(&ring->syncp);
4065 ring->stats.rx_err_cnt++;
4066 u64_stats_update_end(&ring->syncp);
4070 l2_frame_type = hnae3_get_field(l234info, HNS3_RXD_DMAC_M,
4073 u64_stats_update_begin(&ring->syncp);
4074 ring->stats.rx_pkts++;
4075 ring->stats.rx_bytes += len;
4077 if (l2_frame_type == HNS3_L2_TYPE_MULTICAST)
4078 ring->stats.rx_multicast++;
4080 u64_stats_update_end(&ring->syncp);
4082 ring->tqp_vector->rx_group.total_bytes += len;
4084 hns3_set_rx_skb_rss_type(ring, skb, le32_to_cpu(desc->rx.rss_hash));
4088 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring)
4090 struct sk_buff *skb = ring->skb;
4091 struct hns3_desc_cb *desc_cb;
4092 struct hns3_desc *desc;
4093 unsigned int length;
4097 desc = &ring->desc[ring->next_to_clean];
4098 desc_cb = &ring->desc_cb[ring->next_to_clean];
4103 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
4104 /* Check valid BD */
4105 if (unlikely(!(bd_base_info & BIT(HNS3_RXD_VLD_B))))
4109 length = le16_to_cpu(desc->rx.size);
4111 ring->va = desc_cb->buf + desc_cb->page_offset;
4113 dma_sync_single_for_cpu(ring_to_dev(ring),
4114 desc_cb->dma + desc_cb->page_offset,
4115 hns3_buf_size(ring),
4118 /* Prefetch first cache line of first page.
4119 * Idea is to cache few bytes of the header of the packet.
4120 * Our L1 Cache line size is 64B so need to prefetch twice to make
4121 * it 128B. But in actual we can have greater size of caches with
4122 * 128B Level 1 cache lines. In such a case, single fetch would
4123 * suffice to cache in the relevant part of the header.
4125 net_prefetch(ring->va);
4127 ret = hns3_alloc_skb(ring, length, ring->va);
4130 if (ret < 0) /* alloc buffer fail */
4132 if (!(bd_base_info & BIT(HNS3_RXD_FE_B))) { /* need add frag */
4133 ret = hns3_add_frag(ring);
4138 ret = hns3_add_frag(ring);
4143 /* As the head data may be changed when GRO enable, copy
4144 * the head data in after other data rx completed
4146 if (skb->len > HNS3_RX_HEAD_SIZE)
4147 memcpy(skb->data, ring->va,
4148 ALIGN(ring->pull_len, sizeof(long)));
4150 ret = hns3_handle_bdinfo(ring, skb);
4151 if (unlikely(ret)) {
4152 dev_kfree_skb_any(skb);
4156 skb_record_rx_queue(skb, ring->tqp->tqp_index);
4160 int hns3_clean_rx_ring(struct hns3_enet_ring *ring, int budget,
4161 void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
4163 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
4164 int unused_count = hns3_desc_unused(ring);
4168 unused_count -= ring->pending_buf;
4170 while (recv_pkts < budget) {
4171 /* Reuse or realloc buffers */
4172 if (unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
4173 hns3_nic_alloc_rx_buffers(ring, unused_count);
4174 unused_count = hns3_desc_unused(ring) -
4179 err = hns3_handle_rx_bd(ring);
4180 /* Do not get FE for the packet or failed to alloc skb */
4181 if (unlikely(!ring->skb || err == -ENXIO)) {
4183 } else if (likely(!err)) {
4184 rx_fn(ring, ring->skb);
4188 unused_count += ring->pending_buf;
4190 ring->pending_buf = 0;
4194 /* Make all data has been write before submit */
4195 if (unused_count > 0)
4196 hns3_nic_alloc_rx_buffers(ring, unused_count);
4201 static void hns3_update_rx_int_coalesce(struct hns3_enet_tqp_vector *tqp_vector)
4203 struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
4204 struct dim_sample sample = {};
4206 if (!rx_group->coal.adapt_enable)
4209 dim_update_sample(tqp_vector->event_cnt, rx_group->total_packets,
4210 rx_group->total_bytes, &sample);
4211 net_dim(&rx_group->dim, sample);
4214 static void hns3_update_tx_int_coalesce(struct hns3_enet_tqp_vector *tqp_vector)
4216 struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
4217 struct dim_sample sample = {};
4219 if (!tx_group->coal.adapt_enable)
4222 dim_update_sample(tqp_vector->event_cnt, tx_group->total_packets,
4223 tx_group->total_bytes, &sample);
4224 net_dim(&tx_group->dim, sample);
4227 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
4229 struct hns3_nic_priv *priv = netdev_priv(napi->dev);
4230 struct hns3_enet_ring *ring;
4231 int rx_pkt_total = 0;
4233 struct hns3_enet_tqp_vector *tqp_vector =
4234 container_of(napi, struct hns3_enet_tqp_vector, napi);
4235 bool clean_complete = true;
4236 int rx_budget = budget;
4238 if (unlikely(test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
4239 napi_complete(napi);
4243 /* Since the actual Tx work is minimal, we can give the Tx a larger
4244 * budget and be more aggressive about cleaning up the Tx descriptors.
4246 hns3_for_each_ring(ring, tqp_vector->tx_group)
4247 hns3_clean_tx_ring(ring, budget);
4249 /* make sure rx ring budget not smaller than 1 */
4250 if (tqp_vector->num_tqps > 1)
4251 rx_budget = max(budget / tqp_vector->num_tqps, 1);
4253 hns3_for_each_ring(ring, tqp_vector->rx_group) {
4254 int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
4256 if (rx_cleaned >= rx_budget)
4257 clean_complete = false;
4259 rx_pkt_total += rx_cleaned;
4262 tqp_vector->rx_group.total_packets += rx_pkt_total;
4264 if (!clean_complete)
4267 if (napi_complete(napi) &&
4268 likely(!test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
4269 hns3_update_rx_int_coalesce(tqp_vector);
4270 hns3_update_tx_int_coalesce(tqp_vector);
4272 hns3_mask_vector_irq(tqp_vector, 1);
4275 return rx_pkt_total;
4278 static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
4279 struct hnae3_ring_chain_node *head)
4281 struct pci_dev *pdev = tqp_vector->handle->pdev;
4282 struct hnae3_ring_chain_node *cur_chain = head;
4283 struct hnae3_ring_chain_node *chain;
4284 struct hns3_enet_ring *tx_ring;
4285 struct hns3_enet_ring *rx_ring;
4287 tx_ring = tqp_vector->tx_group.ring;
4289 cur_chain->tqp_index = tx_ring->tqp->tqp_index;
4290 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
4291 HNAE3_RING_TYPE_TX);
4292 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
4293 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_TX);
4295 cur_chain->next = NULL;
4297 while (tx_ring->next) {
4298 tx_ring = tx_ring->next;
4300 chain = devm_kzalloc(&pdev->dev, sizeof(*chain),
4303 goto err_free_chain;
4305 cur_chain->next = chain;
4306 chain->tqp_index = tx_ring->tqp->tqp_index;
4307 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
4308 HNAE3_RING_TYPE_TX);
4309 hnae3_set_field(chain->int_gl_idx,
4310 HNAE3_RING_GL_IDX_M,
4311 HNAE3_RING_GL_IDX_S,
4318 rx_ring = tqp_vector->rx_group.ring;
4319 if (!tx_ring && rx_ring) {
4320 cur_chain->next = NULL;
4321 cur_chain->tqp_index = rx_ring->tqp->tqp_index;
4322 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
4323 HNAE3_RING_TYPE_RX);
4324 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
4325 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
4327 rx_ring = rx_ring->next;
4331 chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
4333 goto err_free_chain;
4335 cur_chain->next = chain;
4336 chain->tqp_index = rx_ring->tqp->tqp_index;
4337 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
4338 HNAE3_RING_TYPE_RX);
4339 hnae3_set_field(chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
4340 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
4344 rx_ring = rx_ring->next;
4350 cur_chain = head->next;
4352 chain = cur_chain->next;
4353 devm_kfree(&pdev->dev, cur_chain);
4361 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
4362 struct hnae3_ring_chain_node *head)
4364 struct pci_dev *pdev = tqp_vector->handle->pdev;
4365 struct hnae3_ring_chain_node *chain_tmp, *chain;
4370 chain_tmp = chain->next;
4371 devm_kfree(&pdev->dev, chain);
4376 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
4377 struct hns3_enet_ring *ring)
4379 ring->next = group->ring;
4385 static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
4387 struct pci_dev *pdev = priv->ae_handle->pdev;
4388 struct hns3_enet_tqp_vector *tqp_vector;
4389 int num_vectors = priv->vector_num;
4393 numa_node = dev_to_node(&pdev->dev);
4395 for (vector_i = 0; vector_i < num_vectors; vector_i++) {
4396 tqp_vector = &priv->tqp_vector[vector_i];
4397 cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
4398 &tqp_vector->affinity_mask);
4402 static void hns3_rx_dim_work(struct work_struct *work)
4404 struct dim *dim = container_of(work, struct dim, work);
4405 struct hns3_enet_ring_group *group = container_of(dim,
4406 struct hns3_enet_ring_group, dim);
4407 struct hns3_enet_tqp_vector *tqp_vector = group->ring->tqp_vector;
4408 struct dim_cq_moder cur_moder =
4409 net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
4411 hns3_set_vector_coalesce_rx_gl(group->ring->tqp_vector, cur_moder.usec);
4412 tqp_vector->rx_group.coal.int_gl = cur_moder.usec;
4414 if (cur_moder.pkts < tqp_vector->rx_group.coal.int_ql_max) {
4415 hns3_set_vector_coalesce_rx_ql(tqp_vector, cur_moder.pkts);
4416 tqp_vector->rx_group.coal.int_ql = cur_moder.pkts;
4419 dim->state = DIM_START_MEASURE;
4422 static void hns3_tx_dim_work(struct work_struct *work)
4424 struct dim *dim = container_of(work, struct dim, work);
4425 struct hns3_enet_ring_group *group = container_of(dim,
4426 struct hns3_enet_ring_group, dim);
4427 struct hns3_enet_tqp_vector *tqp_vector = group->ring->tqp_vector;
4428 struct dim_cq_moder cur_moder =
4429 net_dim_get_tx_moderation(dim->mode, dim->profile_ix);
4431 hns3_set_vector_coalesce_tx_gl(tqp_vector, cur_moder.usec);
4432 tqp_vector->tx_group.coal.int_gl = cur_moder.usec;
4434 if (cur_moder.pkts < tqp_vector->tx_group.coal.int_ql_max) {
4435 hns3_set_vector_coalesce_tx_ql(tqp_vector, cur_moder.pkts);
4436 tqp_vector->tx_group.coal.int_ql = cur_moder.pkts;
4439 dim->state = DIM_START_MEASURE;
4442 static void hns3_nic_init_dim(struct hns3_enet_tqp_vector *tqp_vector)
4444 INIT_WORK(&tqp_vector->rx_group.dim.work, hns3_rx_dim_work);
4445 INIT_WORK(&tqp_vector->tx_group.dim.work, hns3_tx_dim_work);
4448 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
4450 struct hnae3_handle *h = priv->ae_handle;
4451 struct hns3_enet_tqp_vector *tqp_vector;
4455 hns3_nic_set_cpumask(priv);
4457 for (i = 0; i < priv->vector_num; i++) {
4458 tqp_vector = &priv->tqp_vector[i];
4459 hns3_vector_coalesce_init_hw(tqp_vector, priv);
4460 tqp_vector->num_tqps = 0;
4461 hns3_nic_init_dim(tqp_vector);
4464 for (i = 0; i < h->kinfo.num_tqps; i++) {
4465 u16 vector_i = i % priv->vector_num;
4466 u16 tqp_num = h->kinfo.num_tqps;
4468 tqp_vector = &priv->tqp_vector[vector_i];
4470 hns3_add_ring_to_group(&tqp_vector->tx_group,
4473 hns3_add_ring_to_group(&tqp_vector->rx_group,
4474 &priv->ring[i + tqp_num]);
4476 priv->ring[i].tqp_vector = tqp_vector;
4477 priv->ring[i + tqp_num].tqp_vector = tqp_vector;
4478 tqp_vector->num_tqps++;
4481 for (i = 0; i < priv->vector_num; i++) {
4482 struct hnae3_ring_chain_node vector_ring_chain;
4484 tqp_vector = &priv->tqp_vector[i];
4486 tqp_vector->rx_group.total_bytes = 0;
4487 tqp_vector->rx_group.total_packets = 0;
4488 tqp_vector->tx_group.total_bytes = 0;
4489 tqp_vector->tx_group.total_packets = 0;
4490 tqp_vector->handle = h;
4492 ret = hns3_get_vector_ring_chain(tqp_vector,
4493 &vector_ring_chain);
4497 ret = h->ae_algo->ops->map_ring_to_vector(h,
4498 tqp_vector->vector_irq, &vector_ring_chain);
4500 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
4505 netif_napi_add(priv->netdev, &tqp_vector->napi,
4506 hns3_nic_common_poll, NAPI_POLL_WEIGHT);
4513 netif_napi_del(&priv->tqp_vector[i].napi);
4518 static void hns3_nic_init_coal_cfg(struct hns3_nic_priv *priv)
4520 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
4521 struct hns3_enet_coalesce *tx_coal = &priv->tx_coal;
4522 struct hns3_enet_coalesce *rx_coal = &priv->rx_coal;
4524 /* initialize the configuration for interrupt coalescing.
4525 * 1. GL (Interrupt Gap Limiter)
4526 * 2. RL (Interrupt Rate Limiter)
4527 * 3. QL (Interrupt Quantity Limiter)
4529 * Default: enable interrupt coalescing self-adaptive and GL
4531 tx_coal->adapt_enable = 1;
4532 rx_coal->adapt_enable = 1;
4534 tx_coal->int_gl = HNS3_INT_GL_50K;
4535 rx_coal->int_gl = HNS3_INT_GL_50K;
4537 rx_coal->flow_level = HNS3_FLOW_LOW;
4538 tx_coal->flow_level = HNS3_FLOW_LOW;
4540 if (ae_dev->dev_specs.int_ql_max) {
4541 tx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
4542 rx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
4546 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
4548 struct hnae3_handle *h = priv->ae_handle;
4549 struct hns3_enet_tqp_vector *tqp_vector;
4550 struct hnae3_vector_info *vector;
4551 struct pci_dev *pdev = h->pdev;
4552 u16 tqp_num = h->kinfo.num_tqps;
4557 /* RSS size, cpu online and vector_num should be the same */
4558 /* Should consider 2p/4p later */
4559 vector_num = min_t(u16, num_online_cpus(), tqp_num);
4561 vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
4566 /* save the actual available vector number */
4567 vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
4569 priv->vector_num = vector_num;
4570 priv->tqp_vector = (struct hns3_enet_tqp_vector *)
4571 devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
4573 if (!priv->tqp_vector) {
4578 for (i = 0; i < priv->vector_num; i++) {
4579 tqp_vector = &priv->tqp_vector[i];
4580 tqp_vector->idx = i;
4581 tqp_vector->mask_addr = vector[i].io_addr;
4582 tqp_vector->vector_irq = vector[i].vector;
4583 hns3_vector_coalesce_init(tqp_vector, priv);
4587 devm_kfree(&pdev->dev, vector);
4591 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
4597 static void hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
4599 struct hnae3_ring_chain_node vector_ring_chain;
4600 struct hnae3_handle *h = priv->ae_handle;
4601 struct hns3_enet_tqp_vector *tqp_vector;
4604 for (i = 0; i < priv->vector_num; i++) {
4605 tqp_vector = &priv->tqp_vector[i];
4607 if (!tqp_vector->rx_group.ring && !tqp_vector->tx_group.ring)
4610 /* Since the mapping can be overwritten, when fail to get the
4611 * chain between vector and ring, we should go on to deal with
4612 * the remaining options.
4614 if (hns3_get_vector_ring_chain(tqp_vector, &vector_ring_chain))
4615 dev_warn(priv->dev, "failed to get ring chain\n");
4617 h->ae_algo->ops->unmap_ring_from_vector(h,
4618 tqp_vector->vector_irq, &vector_ring_chain);
4620 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
4622 hns3_clear_ring_group(&tqp_vector->rx_group);
4623 hns3_clear_ring_group(&tqp_vector->tx_group);
4624 netif_napi_del(&priv->tqp_vector[i].napi);
4628 static void hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
4630 struct hnae3_handle *h = priv->ae_handle;
4631 struct pci_dev *pdev = h->pdev;
4634 for (i = 0; i < priv->vector_num; i++) {
4635 struct hns3_enet_tqp_vector *tqp_vector;
4637 tqp_vector = &priv->tqp_vector[i];
4638 ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
4643 devm_kfree(&pdev->dev, priv->tqp_vector);
4646 static void hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
4647 unsigned int ring_type)
4649 int queue_num = priv->ae_handle->kinfo.num_tqps;
4650 struct hns3_enet_ring *ring;
4653 if (ring_type == HNAE3_RING_TYPE_TX) {
4654 ring = &priv->ring[q->tqp_index];
4655 desc_num = priv->ae_handle->kinfo.num_tx_desc;
4656 ring->queue_index = q->tqp_index;
4657 ring->tx_copybreak = priv->tx_copybreak;
4658 ring->last_to_use = 0;
4660 ring = &priv->ring[q->tqp_index + queue_num];
4661 desc_num = priv->ae_handle->kinfo.num_rx_desc;
4662 ring->queue_index = q->tqp_index;
4663 ring->rx_copybreak = priv->rx_copybreak;
4666 hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
4670 ring->desc_cb = NULL;
4671 ring->dev = priv->dev;
4672 ring->desc_dma_addr = 0;
4673 ring->buf_size = q->buf_size;
4674 ring->desc_num = desc_num;
4675 ring->next_to_use = 0;
4676 ring->next_to_clean = 0;
4679 static void hns3_queue_to_ring(struct hnae3_queue *tqp,
4680 struct hns3_nic_priv *priv)
4682 hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
4683 hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
4686 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
4688 struct hnae3_handle *h = priv->ae_handle;
4689 struct pci_dev *pdev = h->pdev;
4692 priv->ring = devm_kzalloc(&pdev->dev,
4693 array3_size(h->kinfo.num_tqps,
4694 sizeof(*priv->ring), 2),
4699 for (i = 0; i < h->kinfo.num_tqps; i++)
4700 hns3_queue_to_ring(h->kinfo.tqp[i], priv);
4705 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
4710 devm_kfree(priv->dev, priv->ring);
4714 static void hns3_alloc_page_pool(struct hns3_enet_ring *ring)
4716 struct page_pool_params pp_params = {
4717 .flags = PP_FLAG_DMA_MAP | PP_FLAG_PAGE_FRAG |
4718 PP_FLAG_DMA_SYNC_DEV,
4719 .order = hns3_page_order(ring),
4720 .pool_size = ring->desc_num * hns3_buf_size(ring) /
4721 (PAGE_SIZE << hns3_page_order(ring)),
4722 .nid = dev_to_node(ring_to_dev(ring)),
4723 .dev = ring_to_dev(ring),
4724 .dma_dir = DMA_FROM_DEVICE,
4726 .max_len = PAGE_SIZE << hns3_page_order(ring),
4729 ring->page_pool = page_pool_create(&pp_params);
4730 if (IS_ERR(ring->page_pool)) {
4731 dev_warn(ring_to_dev(ring), "page pool creation failed: %ld\n",
4732 PTR_ERR(ring->page_pool));
4733 ring->page_pool = NULL;
4737 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
4741 if (ring->desc_num <= 0 || ring->buf_size <= 0)
4744 ring->desc_cb = devm_kcalloc(ring_to_dev(ring), ring->desc_num,
4745 sizeof(ring->desc_cb[0]), GFP_KERNEL);
4746 if (!ring->desc_cb) {
4751 ret = hns3_alloc_desc(ring);
4753 goto out_with_desc_cb;
4755 if (!HNAE3_IS_TX_RING(ring)) {
4756 hns3_alloc_page_pool(ring);
4758 ret = hns3_alloc_ring_buffers(ring);
4762 hns3_init_tx_spare_buffer(ring);
4768 hns3_free_desc(ring);
4770 devm_kfree(ring_to_dev(ring), ring->desc_cb);
4771 ring->desc_cb = NULL;
4776 void hns3_fini_ring(struct hns3_enet_ring *ring)
4778 hns3_free_desc(ring);
4779 devm_kfree(ring_to_dev(ring), ring->desc_cb);
4780 ring->desc_cb = NULL;
4781 ring->next_to_clean = 0;
4782 ring->next_to_use = 0;
4783 ring->last_to_use = 0;
4784 ring->pending_buf = 0;
4785 if (!HNAE3_IS_TX_RING(ring) && ring->skb) {
4786 dev_kfree_skb_any(ring->skb);
4788 } else if (HNAE3_IS_TX_RING(ring) && ring->tx_spare) {
4789 struct hns3_tx_spare *tx_spare = ring->tx_spare;
4791 dma_unmap_page(ring_to_dev(ring), tx_spare->dma, tx_spare->len,
4793 free_pages((unsigned long)tx_spare->buf,
4794 get_order(tx_spare->len));
4795 devm_kfree(ring_to_dev(ring), tx_spare);
4796 ring->tx_spare = NULL;
4799 if (!HNAE3_IS_TX_RING(ring) && ring->page_pool) {
4800 page_pool_destroy(ring->page_pool);
4801 ring->page_pool = NULL;
4805 static int hns3_buf_size2type(u32 buf_size)
4811 bd_size_type = HNS3_BD_SIZE_512_TYPE;
4814 bd_size_type = HNS3_BD_SIZE_1024_TYPE;
4817 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4820 bd_size_type = HNS3_BD_SIZE_4096_TYPE;
4823 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4826 return bd_size_type;
4829 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
4831 dma_addr_t dma = ring->desc_dma_addr;
4832 struct hnae3_queue *q = ring->tqp;
4834 if (!HNAE3_IS_TX_RING(ring)) {
4835 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG, (u32)dma);
4836 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
4837 (u32)((dma >> 31) >> 1));
4839 hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
4840 hns3_buf_size2type(ring->buf_size));
4841 hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
4842 ring->desc_num / 8 - 1);
4844 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
4846 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
4847 (u32)((dma >> 31) >> 1));
4849 hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
4850 ring->desc_num / 8 - 1);
4854 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
4856 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
4857 struct hnae3_tc_info *tc_info = &kinfo->tc_info;
4860 for (i = 0; i < HNAE3_MAX_TC; i++) {
4863 if (!test_bit(i, &tc_info->tc_en))
4866 for (j = 0; j < tc_info->tqp_count[i]; j++) {
4867 struct hnae3_queue *q;
4869 q = priv->ring[tc_info->tqp_offset[i] + j].tqp;
4870 hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG, i);
4875 int hns3_init_all_ring(struct hns3_nic_priv *priv)
4877 struct hnae3_handle *h = priv->ae_handle;
4878 int ring_num = h->kinfo.num_tqps * 2;
4882 for (i = 0; i < ring_num; i++) {
4883 ret = hns3_alloc_ring_memory(&priv->ring[i]);
4886 "Alloc ring memory fail! ret=%d\n", ret);
4887 goto out_when_alloc_ring_memory;
4890 u64_stats_init(&priv->ring[i].syncp);
4895 out_when_alloc_ring_memory:
4896 for (j = i - 1; j >= 0; j--)
4897 hns3_fini_ring(&priv->ring[j]);
4902 static void hns3_uninit_all_ring(struct hns3_nic_priv *priv)
4904 struct hnae3_handle *h = priv->ae_handle;
4907 for (i = 0; i < h->kinfo.num_tqps; i++) {
4908 hns3_fini_ring(&priv->ring[i]);
4909 hns3_fini_ring(&priv->ring[i + h->kinfo.num_tqps]);
4913 /* Set mac addr if it is configured. or leave it to the AE driver */
4914 static int hns3_init_mac_addr(struct net_device *netdev)
4916 struct hns3_nic_priv *priv = netdev_priv(netdev);
4917 struct hnae3_handle *h = priv->ae_handle;
4918 u8 mac_addr_temp[ETH_ALEN];
4921 if (h->ae_algo->ops->get_mac_addr)
4922 h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
4924 /* Check if the MAC address is valid, if not get a random one */
4925 if (!is_valid_ether_addr(mac_addr_temp)) {
4926 eth_hw_addr_random(netdev);
4927 dev_warn(priv->dev, "using random MAC address %pM\n",
4929 } else if (!ether_addr_equal(netdev->dev_addr, mac_addr_temp)) {
4930 ether_addr_copy(netdev->dev_addr, mac_addr_temp);
4931 ether_addr_copy(netdev->perm_addr, mac_addr_temp);
4936 if (h->ae_algo->ops->set_mac_addr)
4937 ret = h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
4942 static int hns3_init_phy(struct net_device *netdev)
4944 struct hnae3_handle *h = hns3_get_handle(netdev);
4947 if (h->ae_algo->ops->mac_connect_phy)
4948 ret = h->ae_algo->ops->mac_connect_phy(h);
4953 static void hns3_uninit_phy(struct net_device *netdev)
4955 struct hnae3_handle *h = hns3_get_handle(netdev);
4957 if (h->ae_algo->ops->mac_disconnect_phy)
4958 h->ae_algo->ops->mac_disconnect_phy(h);
4961 static int hns3_client_start(struct hnae3_handle *handle)
4963 if (!handle->ae_algo->ops->client_start)
4966 return handle->ae_algo->ops->client_start(handle);
4969 static void hns3_client_stop(struct hnae3_handle *handle)
4971 if (!handle->ae_algo->ops->client_stop)
4974 handle->ae_algo->ops->client_stop(handle);
4977 static void hns3_info_show(struct hns3_nic_priv *priv)
4979 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
4981 dev_info(priv->dev, "MAC address: %pM\n", priv->netdev->dev_addr);
4982 dev_info(priv->dev, "Task queue pairs numbers: %u\n", kinfo->num_tqps);
4983 dev_info(priv->dev, "RSS size: %u\n", kinfo->rss_size);
4984 dev_info(priv->dev, "Allocated RSS size: %u\n", kinfo->req_rss_size);
4985 dev_info(priv->dev, "RX buffer length: %u\n", kinfo->rx_buf_len);
4986 dev_info(priv->dev, "Desc num per TX queue: %u\n", kinfo->num_tx_desc);
4987 dev_info(priv->dev, "Desc num per RX queue: %u\n", kinfo->num_rx_desc);
4988 dev_info(priv->dev, "Total number of enabled TCs: %u\n",
4989 kinfo->tc_info.num_tc);
4990 dev_info(priv->dev, "Max mtu size: %u\n", priv->netdev->max_mtu);
4993 static void hns3_set_cq_period_mode(struct hns3_nic_priv *priv,
4994 enum dim_cq_period_mode mode, bool is_tx)
4996 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
4997 struct hnae3_handle *handle = priv->ae_handle;
5001 priv->tx_cqe_mode = mode;
5003 for (i = 0; i < priv->vector_num; i++)
5004 priv->tqp_vector[i].tx_group.dim.mode = mode;
5006 priv->rx_cqe_mode = mode;
5008 for (i = 0; i < priv->vector_num; i++)
5009 priv->tqp_vector[i].rx_group.dim.mode = mode;
5012 /* only device version above V3(include V3), GL can switch CQ/EQ
5015 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3) {
5019 new_mode = (mode == DIM_CQ_PERIOD_MODE_START_FROM_CQE) ?
5020 HNS3_CQ_MODE_CQE : HNS3_CQ_MODE_EQE;
5021 reg = is_tx ? HNS3_GL1_CQ_MODE_REG : HNS3_GL0_CQ_MODE_REG;
5023 writel(new_mode, handle->kinfo.io_base + reg);
5027 void hns3_cq_period_mode_init(struct hns3_nic_priv *priv,
5028 enum dim_cq_period_mode tx_mode,
5029 enum dim_cq_period_mode rx_mode)
5031 hns3_set_cq_period_mode(priv, tx_mode, true);
5032 hns3_set_cq_period_mode(priv, rx_mode, false);
5035 static void hns3_state_init(struct hnae3_handle *handle)
5037 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(handle->pdev);
5038 struct net_device *netdev = handle->kinfo.netdev;
5039 struct hns3_nic_priv *priv = netdev_priv(netdev);
5041 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
5043 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
5044 set_bit(HNAE3_PFLAG_LIMIT_PROMISC, &handle->supported_pflags);
5046 if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps))
5047 set_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state);
5049 if (hnae3_ae_dev_rxd_adv_layout_supported(ae_dev))
5050 set_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state);
5053 static int hns3_client_init(struct hnae3_handle *handle)
5055 struct pci_dev *pdev = handle->pdev;
5056 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
5057 u16 alloc_tqps, max_rss_size;
5058 struct hns3_nic_priv *priv;
5059 struct net_device *netdev;
5062 handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
5064 netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
5068 priv = netdev_priv(netdev);
5069 priv->dev = &pdev->dev;
5070 priv->netdev = netdev;
5071 priv->ae_handle = handle;
5072 priv->tx_timeout_count = 0;
5073 priv->max_non_tso_bd_num = ae_dev->dev_specs.max_non_tso_bd_num;
5074 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
5076 handle->msg_enable = netif_msg_init(debug, DEFAULT_MSG_LEVEL);
5078 handle->kinfo.netdev = netdev;
5079 handle->priv = (void *)priv;
5081 hns3_init_mac_addr(netdev);
5083 hns3_set_default_feature(netdev);
5085 netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
5086 netdev->priv_flags |= IFF_UNICAST_FLT;
5087 netdev->netdev_ops = &hns3_nic_netdev_ops;
5088 SET_NETDEV_DEV(netdev, &pdev->dev);
5089 hns3_ethtool_set_ops(netdev);
5091 /* Carrier off reporting is important to ethtool even BEFORE open */
5092 netif_carrier_off(netdev);
5094 ret = hns3_get_ring_config(priv);
5097 goto out_get_ring_cfg;
5100 hns3_nic_init_coal_cfg(priv);
5102 ret = hns3_nic_alloc_vector_data(priv);
5105 goto out_alloc_vector_data;
5108 ret = hns3_nic_init_vector_data(priv);
5111 goto out_init_vector_data;
5114 ret = hns3_init_all_ring(priv);
5120 hns3_cq_period_mode_init(priv, DIM_CQ_PERIOD_MODE_START_FROM_EQE,
5121 DIM_CQ_PERIOD_MODE_START_FROM_EQE);
5123 ret = hns3_init_phy(netdev);
5127 /* the device can work without cpu rmap, only aRFS needs it */
5128 ret = hns3_set_rx_cpu_rmap(netdev);
5130 dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
5132 ret = hns3_nic_init_irq(priv);
5134 dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
5135 hns3_free_rx_cpu_rmap(netdev);
5136 goto out_init_irq_fail;
5139 ret = hns3_client_start(handle);
5141 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
5142 goto out_client_start;
5145 hns3_dcbnl_setup(handle);
5147 ret = hns3_dbg_init(handle);
5149 dev_err(priv->dev, "failed to init debugfs, ret = %d\n",
5151 goto out_client_start;
5154 netdev->max_mtu = HNS3_MAX_MTU(ae_dev->dev_specs.max_frm_size);
5156 hns3_state_init(handle);
5158 ret = register_netdev(netdev);
5160 dev_err(priv->dev, "probe register netdev fail!\n");
5161 goto out_reg_netdev_fail;
5164 if (netif_msg_drv(handle))
5165 hns3_info_show(priv);
5169 out_reg_netdev_fail:
5170 hns3_dbg_uninit(handle);
5172 hns3_free_rx_cpu_rmap(netdev);
5173 hns3_nic_uninit_irq(priv);
5175 hns3_uninit_phy(netdev);
5177 hns3_uninit_all_ring(priv);
5179 hns3_nic_uninit_vector_data(priv);
5180 out_init_vector_data:
5181 hns3_nic_dealloc_vector_data(priv);
5182 out_alloc_vector_data:
5185 priv->ae_handle = NULL;
5186 free_netdev(netdev);
5190 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
5192 struct net_device *netdev = handle->kinfo.netdev;
5193 struct hns3_nic_priv *priv = netdev_priv(netdev);
5195 if (netdev->reg_state != NETREG_UNINITIALIZED)
5196 unregister_netdev(netdev);
5198 hns3_client_stop(handle);
5200 hns3_uninit_phy(netdev);
5202 if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5203 netdev_warn(netdev, "already uninitialized\n");
5204 goto out_netdev_free;
5207 hns3_free_rx_cpu_rmap(netdev);
5209 hns3_nic_uninit_irq(priv);
5211 hns3_clear_all_ring(handle, true);
5213 hns3_nic_uninit_vector_data(priv);
5215 hns3_nic_dealloc_vector_data(priv);
5217 hns3_uninit_all_ring(priv);
5219 hns3_put_ring_config(priv);
5222 hns3_dbg_uninit(handle);
5223 free_netdev(netdev);
5226 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
5228 struct net_device *netdev = handle->kinfo.netdev;
5234 netif_tx_wake_all_queues(netdev);
5235 netif_carrier_on(netdev);
5236 if (netif_msg_link(handle))
5237 netdev_info(netdev, "link up\n");
5239 netif_carrier_off(netdev);
5240 netif_tx_stop_all_queues(netdev);
5241 if (netif_msg_link(handle))
5242 netdev_info(netdev, "link down\n");
5246 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
5248 while (ring->next_to_clean != ring->next_to_use) {
5249 ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
5250 hns3_free_buffer_detach(ring, ring->next_to_clean, 0);
5251 ring_ptr_move_fw(ring, next_to_clean);
5254 ring->pending_buf = 0;
5257 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
5259 struct hns3_desc_cb res_cbs;
5262 while (ring->next_to_use != ring->next_to_clean) {
5263 /* When a buffer is not reused, it's memory has been
5264 * freed in hns3_handle_rx_bd or will be freed by
5265 * stack, so we need to replace the buffer here.
5267 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
5268 ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
5270 u64_stats_update_begin(&ring->syncp);
5271 ring->stats.sw_err_cnt++;
5272 u64_stats_update_end(&ring->syncp);
5273 /* if alloc new buffer fail, exit directly
5274 * and reclear in up flow.
5276 netdev_warn(ring_to_netdev(ring),
5277 "reserve buffer map failed, ret = %d\n",
5281 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
5283 ring_ptr_move_fw(ring, next_to_use);
5286 /* Free the pending skb in rx ring */
5288 dev_kfree_skb_any(ring->skb);
5290 ring->pending_buf = 0;
5296 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
5298 while (ring->next_to_use != ring->next_to_clean) {
5299 /* When a buffer is not reused, it's memory has been
5300 * freed in hns3_handle_rx_bd or will be freed by
5301 * stack, so only need to unmap the buffer here.
5303 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
5304 hns3_unmap_buffer(ring,
5305 &ring->desc_cb[ring->next_to_use]);
5306 ring->desc_cb[ring->next_to_use].dma = 0;
5309 ring_ptr_move_fw(ring, next_to_use);
5313 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force)
5315 struct net_device *ndev = h->kinfo.netdev;
5316 struct hns3_nic_priv *priv = netdev_priv(ndev);
5319 for (i = 0; i < h->kinfo.num_tqps; i++) {
5320 struct hns3_enet_ring *ring;
5322 ring = &priv->ring[i];
5323 hns3_clear_tx_ring(ring);
5325 ring = &priv->ring[i + h->kinfo.num_tqps];
5326 /* Continue to clear other rings even if clearing some
5330 hns3_force_clear_rx_ring(ring);
5332 hns3_clear_rx_ring(ring);
5336 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
5338 struct net_device *ndev = h->kinfo.netdev;
5339 struct hns3_nic_priv *priv = netdev_priv(ndev);
5340 struct hns3_enet_ring *rx_ring;
5344 ret = h->ae_algo->ops->reset_queue(h);
5348 for (i = 0; i < h->kinfo.num_tqps; i++) {
5349 hns3_init_ring_hw(&priv->ring[i]);
5351 /* We need to clear tx ring here because self test will
5352 * use the ring and will not run down before up
5354 hns3_clear_tx_ring(&priv->ring[i]);
5355 priv->ring[i].next_to_clean = 0;
5356 priv->ring[i].next_to_use = 0;
5357 priv->ring[i].last_to_use = 0;
5359 rx_ring = &priv->ring[i + h->kinfo.num_tqps];
5360 hns3_init_ring_hw(rx_ring);
5361 ret = hns3_clear_rx_ring(rx_ring);
5365 /* We can not know the hardware head and tail when this
5366 * function is called in reset flow, so we reuse all desc.
5368 for (j = 0; j < rx_ring->desc_num; j++)
5369 hns3_reuse_buffer(rx_ring, j);
5371 rx_ring->next_to_clean = 0;
5372 rx_ring->next_to_use = 0;
5375 hns3_init_tx_ring_tc(priv);
5380 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
5382 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
5383 struct net_device *ndev = kinfo->netdev;
5384 struct hns3_nic_priv *priv = netdev_priv(ndev);
5386 if (test_and_set_bit(HNS3_NIC_STATE_RESETTING, &priv->state))
5389 if (!netif_running(ndev))
5392 return hns3_nic_net_stop(ndev);
5395 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
5397 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
5398 struct hns3_nic_priv *priv = netdev_priv(kinfo->netdev);
5401 if (!test_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5402 netdev_err(kinfo->netdev, "device is not initialized yet\n");
5406 clear_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
5408 if (netif_running(kinfo->netdev)) {
5409 ret = hns3_nic_net_open(kinfo->netdev);
5411 set_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
5412 netdev_err(kinfo->netdev,
5413 "net up fail, ret=%d!\n", ret);
5421 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
5423 struct net_device *netdev = handle->kinfo.netdev;
5424 struct hns3_nic_priv *priv = netdev_priv(netdev);
5427 /* Carrier off reporting is important to ethtool even BEFORE open */
5428 netif_carrier_off(netdev);
5430 ret = hns3_get_ring_config(priv);
5434 ret = hns3_nic_alloc_vector_data(priv);
5438 ret = hns3_nic_init_vector_data(priv);
5440 goto err_dealloc_vector;
5442 ret = hns3_init_all_ring(priv);
5444 goto err_uninit_vector;
5446 hns3_cq_period_mode_init(priv, priv->tx_cqe_mode, priv->rx_cqe_mode);
5448 /* the device can work without cpu rmap, only aRFS needs it */
5449 ret = hns3_set_rx_cpu_rmap(netdev);
5451 dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
5453 ret = hns3_nic_init_irq(priv);
5455 dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
5456 hns3_free_rx_cpu_rmap(netdev);
5457 goto err_init_irq_fail;
5460 if (!hns3_is_phys_func(handle->pdev))
5461 hns3_init_mac_addr(netdev);
5463 ret = hns3_client_start(handle);
5465 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
5466 goto err_client_start_fail;
5469 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
5473 err_client_start_fail:
5474 hns3_free_rx_cpu_rmap(netdev);
5475 hns3_nic_uninit_irq(priv);
5477 hns3_uninit_all_ring(priv);
5479 hns3_nic_uninit_vector_data(priv);
5481 hns3_nic_dealloc_vector_data(priv);
5483 hns3_put_ring_config(priv);
5488 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
5490 struct net_device *netdev = handle->kinfo.netdev;
5491 struct hns3_nic_priv *priv = netdev_priv(netdev);
5493 if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5494 netdev_warn(netdev, "already uninitialized\n");
5498 hns3_free_rx_cpu_rmap(netdev);
5499 hns3_nic_uninit_irq(priv);
5500 hns3_clear_all_ring(handle, true);
5501 hns3_reset_tx_queue(priv->ae_handle);
5503 hns3_nic_uninit_vector_data(priv);
5505 hns3_nic_dealloc_vector_data(priv);
5507 hns3_uninit_all_ring(priv);
5509 hns3_put_ring_config(priv);
5514 static int hns3_reset_notify(struct hnae3_handle *handle,
5515 enum hnae3_reset_notify_type type)
5520 case HNAE3_UP_CLIENT:
5521 ret = hns3_reset_notify_up_enet(handle);
5523 case HNAE3_DOWN_CLIENT:
5524 ret = hns3_reset_notify_down_enet(handle);
5526 case HNAE3_INIT_CLIENT:
5527 ret = hns3_reset_notify_init_enet(handle);
5529 case HNAE3_UNINIT_CLIENT:
5530 ret = hns3_reset_notify_uninit_enet(handle);
5539 static int hns3_change_channels(struct hnae3_handle *handle, u32 new_tqp_num,
5540 bool rxfh_configured)
5544 ret = handle->ae_algo->ops->set_channels(handle, new_tqp_num,
5547 dev_err(&handle->pdev->dev,
5548 "Change tqp num(%u) fail.\n", new_tqp_num);
5552 ret = hns3_reset_notify(handle, HNAE3_INIT_CLIENT);
5556 ret = hns3_reset_notify(handle, HNAE3_UP_CLIENT);
5558 hns3_reset_notify(handle, HNAE3_UNINIT_CLIENT);
5563 int hns3_set_channels(struct net_device *netdev,
5564 struct ethtool_channels *ch)
5566 struct hnae3_handle *h = hns3_get_handle(netdev);
5567 struct hnae3_knic_private_info *kinfo = &h->kinfo;
5568 bool rxfh_configured = netif_is_rxfh_configured(netdev);
5569 u32 new_tqp_num = ch->combined_count;
5573 if (hns3_nic_resetting(netdev))
5576 if (ch->rx_count || ch->tx_count)
5579 if (kinfo->tc_info.mqprio_active) {
5580 dev_err(&netdev->dev,
5581 "it's not allowed to set channels via ethtool when MQPRIO mode is on\n");
5585 if (new_tqp_num > hns3_get_max_available_channels(h) ||
5587 dev_err(&netdev->dev,
5588 "Change tqps fail, the tqp range is from 1 to %u",
5589 hns3_get_max_available_channels(h));
5593 if (kinfo->rss_size == new_tqp_num)
5596 netif_dbg(h, drv, netdev,
5597 "set channels: tqp_num=%u, rxfh=%d\n",
5598 new_tqp_num, rxfh_configured);
5600 ret = hns3_reset_notify(h, HNAE3_DOWN_CLIENT);
5604 ret = hns3_reset_notify(h, HNAE3_UNINIT_CLIENT);
5608 org_tqp_num = h->kinfo.num_tqps;
5609 ret = hns3_change_channels(h, new_tqp_num, rxfh_configured);
5614 "Change channels fail, revert to old value\n");
5615 ret1 = hns3_change_channels(h, org_tqp_num, rxfh_configured);
5618 "revert to old channel fail\n");
5628 static const struct hns3_hw_error_info hns3_hw_err[] = {
5629 { .type = HNAE3_PPU_POISON_ERROR,
5630 .msg = "PPU poison" },
5631 { .type = HNAE3_CMDQ_ECC_ERROR,
5632 .msg = "IMP CMDQ error" },
5633 { .type = HNAE3_IMP_RD_POISON_ERROR,
5634 .msg = "IMP RD poison" },
5635 { .type = HNAE3_ROCEE_AXI_RESP_ERROR,
5636 .msg = "ROCEE AXI RESP error" },
5639 static void hns3_process_hw_error(struct hnae3_handle *handle,
5640 enum hnae3_hw_error_type type)
5644 for (i = 0; i < ARRAY_SIZE(hns3_hw_err); i++) {
5645 if (hns3_hw_err[i].type == type) {
5646 dev_err(&handle->pdev->dev, "Detected %s!\n",
5647 hns3_hw_err[i].msg);
5653 static const struct hnae3_client_ops client_ops = {
5654 .init_instance = hns3_client_init,
5655 .uninit_instance = hns3_client_uninit,
5656 .link_status_change = hns3_link_status_change,
5657 .reset_notify = hns3_reset_notify,
5658 .process_hw_error = hns3_process_hw_error,
5661 /* hns3_init_module - Driver registration routine
5662 * hns3_init_module is the first routine called when the driver is
5663 * loaded. All it does is register with the PCI subsystem.
5665 static int __init hns3_init_module(void)
5669 pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
5670 pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
5672 client.type = HNAE3_CLIENT_KNIC;
5673 snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH, "%s",
5676 client.ops = &client_ops;
5678 INIT_LIST_HEAD(&client.node);
5680 hns3_dbg_register_debugfs(hns3_driver_name);
5682 ret = hnae3_register_client(&client);
5684 goto err_reg_client;
5686 ret = pci_register_driver(&hns3_driver);
5688 goto err_reg_driver;
5693 hnae3_unregister_client(&client);
5695 hns3_dbg_unregister_debugfs();
5698 module_init(hns3_init_module);
5700 /* hns3_exit_module - Driver exit cleanup routine
5701 * hns3_exit_module is called just before the driver is removed
5704 static void __exit hns3_exit_module(void)
5706 pci_unregister_driver(&hns3_driver);
5707 hnae3_unregister_client(&client);
5708 hns3_dbg_unregister_debugfs();
5710 module_exit(hns3_exit_module);
5712 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
5713 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
5714 MODULE_LICENSE("GPL");
5715 MODULE_ALIAS("pci:hns-nic");