1 // SPDX-License-Identifier: GPL-2.0+
2 // Copyright (c) 2016-2017 Hisilicon Limited.
4 #include <linux/dma-mapping.h>
5 #include <linux/etherdevice.h>
6 #include <linux/interrupt.h>
7 #ifdef CONFIG_RFS_ACCEL
8 #include <linux/cpu_rmap.h>
10 #include <linux/if_vlan.h>
11 #include <linux/irq.h>
13 #include <linux/ipv6.h>
14 #include <linux/module.h>
15 #include <linux/pci.h>
16 #include <linux/aer.h>
17 #include <linux/skbuff.h>
18 #include <linux/sctp.h>
20 #include <net/ip6_checksum.h>
21 #include <net/pkt_cls.h>
23 #include <net/vxlan.h>
24 #include <net/geneve.h>
27 #include "hns3_enet.h"
28 /* All hns3 tracepoints are defined by the include below, which
29 * must be included exactly once across the whole kernel with
30 * CREATE_TRACE_POINTS defined
32 #define CREATE_TRACE_POINTS
33 #include "hns3_trace.h"
35 #define hns3_set_field(origin, shift, val) ((origin) |= (val) << (shift))
36 #define hns3_tx_bd_count(S) DIV_ROUND_UP(S, HNS3_MAX_BD_SIZE)
38 #define hns3_rl_err(fmt, ...) \
40 if (net_ratelimit()) \
41 netdev_err(fmt, ##__VA_ARGS__); \
44 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force);
46 static const char hns3_driver_name[] = "hns3";
47 static const char hns3_driver_string[] =
48 "Hisilicon Ethernet Network Driver for Hip08 Family";
49 static const char hns3_copyright[] = "Copyright (c) 2017 Huawei Corporation.";
50 static struct hnae3_client client;
52 static int debug = -1;
53 module_param(debug, int, 0);
54 MODULE_PARM_DESC(debug, " Network interface message level setting");
56 static unsigned int tx_spare_buf_size;
57 module_param(tx_spare_buf_size, uint, 0400);
58 MODULE_PARM_DESC(tx_spare_buf_size, "Size used to allocate tx spare buffer");
60 static unsigned int tx_sgl = 1;
61 module_param(tx_sgl, uint, 0600);
62 MODULE_PARM_DESC(tx_sgl, "Minimum number of frags when using dma_map_sg() to optimize the IOMMU mapping");
64 static bool page_pool_enabled = true;
65 module_param(page_pool_enabled, bool, 0400);
67 #define HNS3_SGL_SIZE(nfrag) (sizeof(struct scatterlist) * (nfrag) + \
68 sizeof(struct sg_table))
69 #define HNS3_MAX_SGL_SIZE ALIGN(HNS3_SGL_SIZE(HNS3_MAX_TSO_BD_NUM), \
70 dma_get_cache_alignment())
72 #define DEFAULT_MSG_LEVEL (NETIF_MSG_PROBE | NETIF_MSG_LINK | \
73 NETIF_MSG_IFDOWN | NETIF_MSG_IFUP)
75 #define HNS3_INNER_VLAN_TAG 1
76 #define HNS3_OUTER_VLAN_TAG 2
78 #define HNS3_MIN_TX_LEN 33U
80 /* hns3_pci_tbl - PCI Device ID Table
82 * Last entry must be all 0s
84 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
85 * Class, Class Mask, private data (not used) }
87 static const struct pci_device_id hns3_pci_tbl[] = {
88 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
89 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
90 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA),
91 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
92 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC),
93 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
94 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA),
95 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
96 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC),
97 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
98 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC),
99 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
100 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_200G_RDMA),
101 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
102 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_VF), 0},
103 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_RDMA_DCB_PFC_VF),
104 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
105 /* required last entry */
108 MODULE_DEVICE_TABLE(pci, hns3_pci_tbl);
110 #define HNS3_RX_PTYPE_ENTRY(ptype, l, s, t) \
117 #define HNS3_RX_PTYPE_UNUSED_ENTRY(ptype) \
118 { ptype, 0, CHECKSUM_NONE, HNS3_L3_TYPE_PARSE_FAIL, 0 }
120 static const struct hns3_rx_ptype hns3_rx_ptype_tbl[] = {
121 HNS3_RX_PTYPE_UNUSED_ENTRY(0),
122 HNS3_RX_PTYPE_ENTRY(1, 0, COMPLETE, ARP),
123 HNS3_RX_PTYPE_ENTRY(2, 0, COMPLETE, RARP),
124 HNS3_RX_PTYPE_ENTRY(3, 0, COMPLETE, LLDP),
125 HNS3_RX_PTYPE_ENTRY(4, 0, COMPLETE, PARSE_FAIL),
126 HNS3_RX_PTYPE_ENTRY(5, 0, COMPLETE, PARSE_FAIL),
127 HNS3_RX_PTYPE_ENTRY(6, 0, COMPLETE, PARSE_FAIL),
128 HNS3_RX_PTYPE_ENTRY(7, 0, COMPLETE, CNM),
129 HNS3_RX_PTYPE_ENTRY(8, 0, NONE, PARSE_FAIL),
130 HNS3_RX_PTYPE_UNUSED_ENTRY(9),
131 HNS3_RX_PTYPE_UNUSED_ENTRY(10),
132 HNS3_RX_PTYPE_UNUSED_ENTRY(11),
133 HNS3_RX_PTYPE_UNUSED_ENTRY(12),
134 HNS3_RX_PTYPE_UNUSED_ENTRY(13),
135 HNS3_RX_PTYPE_UNUSED_ENTRY(14),
136 HNS3_RX_PTYPE_UNUSED_ENTRY(15),
137 HNS3_RX_PTYPE_ENTRY(16, 0, COMPLETE, PARSE_FAIL),
138 HNS3_RX_PTYPE_ENTRY(17, 0, COMPLETE, IPV4),
139 HNS3_RX_PTYPE_ENTRY(18, 0, COMPLETE, IPV4),
140 HNS3_RX_PTYPE_ENTRY(19, 0, UNNECESSARY, IPV4),
141 HNS3_RX_PTYPE_ENTRY(20, 0, UNNECESSARY, IPV4),
142 HNS3_RX_PTYPE_ENTRY(21, 0, NONE, IPV4),
143 HNS3_RX_PTYPE_ENTRY(22, 0, UNNECESSARY, IPV4),
144 HNS3_RX_PTYPE_ENTRY(23, 0, NONE, IPV4),
145 HNS3_RX_PTYPE_ENTRY(24, 0, NONE, IPV4),
146 HNS3_RX_PTYPE_ENTRY(25, 0, UNNECESSARY, IPV4),
147 HNS3_RX_PTYPE_UNUSED_ENTRY(26),
148 HNS3_RX_PTYPE_UNUSED_ENTRY(27),
149 HNS3_RX_PTYPE_UNUSED_ENTRY(28),
150 HNS3_RX_PTYPE_ENTRY(29, 0, COMPLETE, PARSE_FAIL),
151 HNS3_RX_PTYPE_ENTRY(30, 0, COMPLETE, PARSE_FAIL),
152 HNS3_RX_PTYPE_ENTRY(31, 0, COMPLETE, IPV4),
153 HNS3_RX_PTYPE_ENTRY(32, 0, COMPLETE, IPV4),
154 HNS3_RX_PTYPE_ENTRY(33, 1, UNNECESSARY, IPV4),
155 HNS3_RX_PTYPE_ENTRY(34, 1, UNNECESSARY, IPV4),
156 HNS3_RX_PTYPE_ENTRY(35, 1, UNNECESSARY, IPV4),
157 HNS3_RX_PTYPE_ENTRY(36, 0, COMPLETE, IPV4),
158 HNS3_RX_PTYPE_ENTRY(37, 0, COMPLETE, IPV4),
159 HNS3_RX_PTYPE_UNUSED_ENTRY(38),
160 HNS3_RX_PTYPE_ENTRY(39, 0, COMPLETE, IPV6),
161 HNS3_RX_PTYPE_ENTRY(40, 0, COMPLETE, IPV6),
162 HNS3_RX_PTYPE_ENTRY(41, 1, UNNECESSARY, IPV6),
163 HNS3_RX_PTYPE_ENTRY(42, 1, UNNECESSARY, IPV6),
164 HNS3_RX_PTYPE_ENTRY(43, 1, UNNECESSARY, IPV6),
165 HNS3_RX_PTYPE_ENTRY(44, 0, COMPLETE, IPV6),
166 HNS3_RX_PTYPE_ENTRY(45, 0, COMPLETE, IPV6),
167 HNS3_RX_PTYPE_UNUSED_ENTRY(46),
168 HNS3_RX_PTYPE_UNUSED_ENTRY(47),
169 HNS3_RX_PTYPE_UNUSED_ENTRY(48),
170 HNS3_RX_PTYPE_UNUSED_ENTRY(49),
171 HNS3_RX_PTYPE_UNUSED_ENTRY(50),
172 HNS3_RX_PTYPE_UNUSED_ENTRY(51),
173 HNS3_RX_PTYPE_UNUSED_ENTRY(52),
174 HNS3_RX_PTYPE_UNUSED_ENTRY(53),
175 HNS3_RX_PTYPE_UNUSED_ENTRY(54),
176 HNS3_RX_PTYPE_UNUSED_ENTRY(55),
177 HNS3_RX_PTYPE_UNUSED_ENTRY(56),
178 HNS3_RX_PTYPE_UNUSED_ENTRY(57),
179 HNS3_RX_PTYPE_UNUSED_ENTRY(58),
180 HNS3_RX_PTYPE_UNUSED_ENTRY(59),
181 HNS3_RX_PTYPE_UNUSED_ENTRY(60),
182 HNS3_RX_PTYPE_UNUSED_ENTRY(61),
183 HNS3_RX_PTYPE_UNUSED_ENTRY(62),
184 HNS3_RX_PTYPE_UNUSED_ENTRY(63),
185 HNS3_RX_PTYPE_UNUSED_ENTRY(64),
186 HNS3_RX_PTYPE_UNUSED_ENTRY(65),
187 HNS3_RX_PTYPE_UNUSED_ENTRY(66),
188 HNS3_RX_PTYPE_UNUSED_ENTRY(67),
189 HNS3_RX_PTYPE_UNUSED_ENTRY(68),
190 HNS3_RX_PTYPE_UNUSED_ENTRY(69),
191 HNS3_RX_PTYPE_UNUSED_ENTRY(70),
192 HNS3_RX_PTYPE_UNUSED_ENTRY(71),
193 HNS3_RX_PTYPE_UNUSED_ENTRY(72),
194 HNS3_RX_PTYPE_UNUSED_ENTRY(73),
195 HNS3_RX_PTYPE_UNUSED_ENTRY(74),
196 HNS3_RX_PTYPE_UNUSED_ENTRY(75),
197 HNS3_RX_PTYPE_UNUSED_ENTRY(76),
198 HNS3_RX_PTYPE_UNUSED_ENTRY(77),
199 HNS3_RX_PTYPE_UNUSED_ENTRY(78),
200 HNS3_RX_PTYPE_UNUSED_ENTRY(79),
201 HNS3_RX_PTYPE_UNUSED_ENTRY(80),
202 HNS3_RX_PTYPE_UNUSED_ENTRY(81),
203 HNS3_RX_PTYPE_UNUSED_ENTRY(82),
204 HNS3_RX_PTYPE_UNUSED_ENTRY(83),
205 HNS3_RX_PTYPE_UNUSED_ENTRY(84),
206 HNS3_RX_PTYPE_UNUSED_ENTRY(85),
207 HNS3_RX_PTYPE_UNUSED_ENTRY(86),
208 HNS3_RX_PTYPE_UNUSED_ENTRY(87),
209 HNS3_RX_PTYPE_UNUSED_ENTRY(88),
210 HNS3_RX_PTYPE_UNUSED_ENTRY(89),
211 HNS3_RX_PTYPE_UNUSED_ENTRY(90),
212 HNS3_RX_PTYPE_UNUSED_ENTRY(91),
213 HNS3_RX_PTYPE_UNUSED_ENTRY(92),
214 HNS3_RX_PTYPE_UNUSED_ENTRY(93),
215 HNS3_RX_PTYPE_UNUSED_ENTRY(94),
216 HNS3_RX_PTYPE_UNUSED_ENTRY(95),
217 HNS3_RX_PTYPE_UNUSED_ENTRY(96),
218 HNS3_RX_PTYPE_UNUSED_ENTRY(97),
219 HNS3_RX_PTYPE_UNUSED_ENTRY(98),
220 HNS3_RX_PTYPE_UNUSED_ENTRY(99),
221 HNS3_RX_PTYPE_UNUSED_ENTRY(100),
222 HNS3_RX_PTYPE_UNUSED_ENTRY(101),
223 HNS3_RX_PTYPE_UNUSED_ENTRY(102),
224 HNS3_RX_PTYPE_UNUSED_ENTRY(103),
225 HNS3_RX_PTYPE_UNUSED_ENTRY(104),
226 HNS3_RX_PTYPE_UNUSED_ENTRY(105),
227 HNS3_RX_PTYPE_UNUSED_ENTRY(106),
228 HNS3_RX_PTYPE_UNUSED_ENTRY(107),
229 HNS3_RX_PTYPE_UNUSED_ENTRY(108),
230 HNS3_RX_PTYPE_UNUSED_ENTRY(109),
231 HNS3_RX_PTYPE_UNUSED_ENTRY(110),
232 HNS3_RX_PTYPE_ENTRY(111, 0, COMPLETE, IPV6),
233 HNS3_RX_PTYPE_ENTRY(112, 0, COMPLETE, IPV6),
234 HNS3_RX_PTYPE_ENTRY(113, 0, UNNECESSARY, IPV6),
235 HNS3_RX_PTYPE_ENTRY(114, 0, UNNECESSARY, IPV6),
236 HNS3_RX_PTYPE_ENTRY(115, 0, NONE, IPV6),
237 HNS3_RX_PTYPE_ENTRY(116, 0, UNNECESSARY, IPV6),
238 HNS3_RX_PTYPE_ENTRY(117, 0, NONE, IPV6),
239 HNS3_RX_PTYPE_ENTRY(118, 0, NONE, IPV6),
240 HNS3_RX_PTYPE_ENTRY(119, 0, UNNECESSARY, IPV6),
241 HNS3_RX_PTYPE_UNUSED_ENTRY(120),
242 HNS3_RX_PTYPE_UNUSED_ENTRY(121),
243 HNS3_RX_PTYPE_UNUSED_ENTRY(122),
244 HNS3_RX_PTYPE_ENTRY(123, 0, COMPLETE, PARSE_FAIL),
245 HNS3_RX_PTYPE_ENTRY(124, 0, COMPLETE, PARSE_FAIL),
246 HNS3_RX_PTYPE_ENTRY(125, 0, COMPLETE, IPV4),
247 HNS3_RX_PTYPE_ENTRY(126, 0, COMPLETE, IPV4),
248 HNS3_RX_PTYPE_ENTRY(127, 1, UNNECESSARY, IPV4),
249 HNS3_RX_PTYPE_ENTRY(128, 1, UNNECESSARY, IPV4),
250 HNS3_RX_PTYPE_ENTRY(129, 1, UNNECESSARY, IPV4),
251 HNS3_RX_PTYPE_ENTRY(130, 0, COMPLETE, IPV4),
252 HNS3_RX_PTYPE_ENTRY(131, 0, COMPLETE, IPV4),
253 HNS3_RX_PTYPE_UNUSED_ENTRY(132),
254 HNS3_RX_PTYPE_ENTRY(133, 0, COMPLETE, IPV6),
255 HNS3_RX_PTYPE_ENTRY(134, 0, COMPLETE, IPV6),
256 HNS3_RX_PTYPE_ENTRY(135, 1, UNNECESSARY, IPV6),
257 HNS3_RX_PTYPE_ENTRY(136, 1, UNNECESSARY, IPV6),
258 HNS3_RX_PTYPE_ENTRY(137, 1, UNNECESSARY, IPV6),
259 HNS3_RX_PTYPE_ENTRY(138, 0, COMPLETE, IPV6),
260 HNS3_RX_PTYPE_ENTRY(139, 0, COMPLETE, IPV6),
261 HNS3_RX_PTYPE_UNUSED_ENTRY(140),
262 HNS3_RX_PTYPE_UNUSED_ENTRY(141),
263 HNS3_RX_PTYPE_UNUSED_ENTRY(142),
264 HNS3_RX_PTYPE_UNUSED_ENTRY(143),
265 HNS3_RX_PTYPE_UNUSED_ENTRY(144),
266 HNS3_RX_PTYPE_UNUSED_ENTRY(145),
267 HNS3_RX_PTYPE_UNUSED_ENTRY(146),
268 HNS3_RX_PTYPE_UNUSED_ENTRY(147),
269 HNS3_RX_PTYPE_UNUSED_ENTRY(148),
270 HNS3_RX_PTYPE_UNUSED_ENTRY(149),
271 HNS3_RX_PTYPE_UNUSED_ENTRY(150),
272 HNS3_RX_PTYPE_UNUSED_ENTRY(151),
273 HNS3_RX_PTYPE_UNUSED_ENTRY(152),
274 HNS3_RX_PTYPE_UNUSED_ENTRY(153),
275 HNS3_RX_PTYPE_UNUSED_ENTRY(154),
276 HNS3_RX_PTYPE_UNUSED_ENTRY(155),
277 HNS3_RX_PTYPE_UNUSED_ENTRY(156),
278 HNS3_RX_PTYPE_UNUSED_ENTRY(157),
279 HNS3_RX_PTYPE_UNUSED_ENTRY(158),
280 HNS3_RX_PTYPE_UNUSED_ENTRY(159),
281 HNS3_RX_PTYPE_UNUSED_ENTRY(160),
282 HNS3_RX_PTYPE_UNUSED_ENTRY(161),
283 HNS3_RX_PTYPE_UNUSED_ENTRY(162),
284 HNS3_RX_PTYPE_UNUSED_ENTRY(163),
285 HNS3_RX_PTYPE_UNUSED_ENTRY(164),
286 HNS3_RX_PTYPE_UNUSED_ENTRY(165),
287 HNS3_RX_PTYPE_UNUSED_ENTRY(166),
288 HNS3_RX_PTYPE_UNUSED_ENTRY(167),
289 HNS3_RX_PTYPE_UNUSED_ENTRY(168),
290 HNS3_RX_PTYPE_UNUSED_ENTRY(169),
291 HNS3_RX_PTYPE_UNUSED_ENTRY(170),
292 HNS3_RX_PTYPE_UNUSED_ENTRY(171),
293 HNS3_RX_PTYPE_UNUSED_ENTRY(172),
294 HNS3_RX_PTYPE_UNUSED_ENTRY(173),
295 HNS3_RX_PTYPE_UNUSED_ENTRY(174),
296 HNS3_RX_PTYPE_UNUSED_ENTRY(175),
297 HNS3_RX_PTYPE_UNUSED_ENTRY(176),
298 HNS3_RX_PTYPE_UNUSED_ENTRY(177),
299 HNS3_RX_PTYPE_UNUSED_ENTRY(178),
300 HNS3_RX_PTYPE_UNUSED_ENTRY(179),
301 HNS3_RX_PTYPE_UNUSED_ENTRY(180),
302 HNS3_RX_PTYPE_UNUSED_ENTRY(181),
303 HNS3_RX_PTYPE_UNUSED_ENTRY(182),
304 HNS3_RX_PTYPE_UNUSED_ENTRY(183),
305 HNS3_RX_PTYPE_UNUSED_ENTRY(184),
306 HNS3_RX_PTYPE_UNUSED_ENTRY(185),
307 HNS3_RX_PTYPE_UNUSED_ENTRY(186),
308 HNS3_RX_PTYPE_UNUSED_ENTRY(187),
309 HNS3_RX_PTYPE_UNUSED_ENTRY(188),
310 HNS3_RX_PTYPE_UNUSED_ENTRY(189),
311 HNS3_RX_PTYPE_UNUSED_ENTRY(190),
312 HNS3_RX_PTYPE_UNUSED_ENTRY(191),
313 HNS3_RX_PTYPE_UNUSED_ENTRY(192),
314 HNS3_RX_PTYPE_UNUSED_ENTRY(193),
315 HNS3_RX_PTYPE_UNUSED_ENTRY(194),
316 HNS3_RX_PTYPE_UNUSED_ENTRY(195),
317 HNS3_RX_PTYPE_UNUSED_ENTRY(196),
318 HNS3_RX_PTYPE_UNUSED_ENTRY(197),
319 HNS3_RX_PTYPE_UNUSED_ENTRY(198),
320 HNS3_RX_PTYPE_UNUSED_ENTRY(199),
321 HNS3_RX_PTYPE_UNUSED_ENTRY(200),
322 HNS3_RX_PTYPE_UNUSED_ENTRY(201),
323 HNS3_RX_PTYPE_UNUSED_ENTRY(202),
324 HNS3_RX_PTYPE_UNUSED_ENTRY(203),
325 HNS3_RX_PTYPE_UNUSED_ENTRY(204),
326 HNS3_RX_PTYPE_UNUSED_ENTRY(205),
327 HNS3_RX_PTYPE_UNUSED_ENTRY(206),
328 HNS3_RX_PTYPE_UNUSED_ENTRY(207),
329 HNS3_RX_PTYPE_UNUSED_ENTRY(208),
330 HNS3_RX_PTYPE_UNUSED_ENTRY(209),
331 HNS3_RX_PTYPE_UNUSED_ENTRY(210),
332 HNS3_RX_PTYPE_UNUSED_ENTRY(211),
333 HNS3_RX_PTYPE_UNUSED_ENTRY(212),
334 HNS3_RX_PTYPE_UNUSED_ENTRY(213),
335 HNS3_RX_PTYPE_UNUSED_ENTRY(214),
336 HNS3_RX_PTYPE_UNUSED_ENTRY(215),
337 HNS3_RX_PTYPE_UNUSED_ENTRY(216),
338 HNS3_RX_PTYPE_UNUSED_ENTRY(217),
339 HNS3_RX_PTYPE_UNUSED_ENTRY(218),
340 HNS3_RX_PTYPE_UNUSED_ENTRY(219),
341 HNS3_RX_PTYPE_UNUSED_ENTRY(220),
342 HNS3_RX_PTYPE_UNUSED_ENTRY(221),
343 HNS3_RX_PTYPE_UNUSED_ENTRY(222),
344 HNS3_RX_PTYPE_UNUSED_ENTRY(223),
345 HNS3_RX_PTYPE_UNUSED_ENTRY(224),
346 HNS3_RX_PTYPE_UNUSED_ENTRY(225),
347 HNS3_RX_PTYPE_UNUSED_ENTRY(226),
348 HNS3_RX_PTYPE_UNUSED_ENTRY(227),
349 HNS3_RX_PTYPE_UNUSED_ENTRY(228),
350 HNS3_RX_PTYPE_UNUSED_ENTRY(229),
351 HNS3_RX_PTYPE_UNUSED_ENTRY(230),
352 HNS3_RX_PTYPE_UNUSED_ENTRY(231),
353 HNS3_RX_PTYPE_UNUSED_ENTRY(232),
354 HNS3_RX_PTYPE_UNUSED_ENTRY(233),
355 HNS3_RX_PTYPE_UNUSED_ENTRY(234),
356 HNS3_RX_PTYPE_UNUSED_ENTRY(235),
357 HNS3_RX_PTYPE_UNUSED_ENTRY(236),
358 HNS3_RX_PTYPE_UNUSED_ENTRY(237),
359 HNS3_RX_PTYPE_UNUSED_ENTRY(238),
360 HNS3_RX_PTYPE_UNUSED_ENTRY(239),
361 HNS3_RX_PTYPE_UNUSED_ENTRY(240),
362 HNS3_RX_PTYPE_UNUSED_ENTRY(241),
363 HNS3_RX_PTYPE_UNUSED_ENTRY(242),
364 HNS3_RX_PTYPE_UNUSED_ENTRY(243),
365 HNS3_RX_PTYPE_UNUSED_ENTRY(244),
366 HNS3_RX_PTYPE_UNUSED_ENTRY(245),
367 HNS3_RX_PTYPE_UNUSED_ENTRY(246),
368 HNS3_RX_PTYPE_UNUSED_ENTRY(247),
369 HNS3_RX_PTYPE_UNUSED_ENTRY(248),
370 HNS3_RX_PTYPE_UNUSED_ENTRY(249),
371 HNS3_RX_PTYPE_UNUSED_ENTRY(250),
372 HNS3_RX_PTYPE_UNUSED_ENTRY(251),
373 HNS3_RX_PTYPE_UNUSED_ENTRY(252),
374 HNS3_RX_PTYPE_UNUSED_ENTRY(253),
375 HNS3_RX_PTYPE_UNUSED_ENTRY(254),
376 HNS3_RX_PTYPE_UNUSED_ENTRY(255),
379 #define HNS3_INVALID_PTYPE \
380 ARRAY_SIZE(hns3_rx_ptype_tbl)
382 static irqreturn_t hns3_irq_handle(int irq, void *vector)
384 struct hns3_enet_tqp_vector *tqp_vector = vector;
386 napi_schedule_irqoff(&tqp_vector->napi);
387 tqp_vector->event_cnt++;
392 static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv)
394 struct hns3_enet_tqp_vector *tqp_vectors;
397 for (i = 0; i < priv->vector_num; i++) {
398 tqp_vectors = &priv->tqp_vector[i];
400 if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED)
403 /* clear the affinity mask */
404 irq_set_affinity_hint(tqp_vectors->vector_irq, NULL);
406 /* release the irq resource */
407 free_irq(tqp_vectors->vector_irq, tqp_vectors);
408 tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED;
412 static int hns3_nic_init_irq(struct hns3_nic_priv *priv)
414 struct hns3_enet_tqp_vector *tqp_vectors;
415 int txrx_int_idx = 0;
421 for (i = 0; i < priv->vector_num; i++) {
422 tqp_vectors = &priv->tqp_vector[i];
424 if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED)
427 if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) {
428 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
429 "%s-%s-%s-%d", hns3_driver_name,
430 pci_name(priv->ae_handle->pdev),
431 "TxRx", txrx_int_idx++);
433 } else if (tqp_vectors->rx_group.ring) {
434 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
435 "%s-%s-%s-%d", hns3_driver_name,
436 pci_name(priv->ae_handle->pdev),
438 } else if (tqp_vectors->tx_group.ring) {
439 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
440 "%s-%s-%s-%d", hns3_driver_name,
441 pci_name(priv->ae_handle->pdev),
444 /* Skip this unused q_vector */
448 tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0';
450 irq_set_status_flags(tqp_vectors->vector_irq, IRQ_NOAUTOEN);
451 ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0,
452 tqp_vectors->name, tqp_vectors);
454 netdev_err(priv->netdev, "request irq(%d) fail\n",
455 tqp_vectors->vector_irq);
456 hns3_nic_uninit_irq(priv);
460 irq_set_affinity_hint(tqp_vectors->vector_irq,
461 &tqp_vectors->affinity_mask);
463 tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED;
469 static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector,
472 writel(mask_en, tqp_vector->mask_addr);
475 static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector)
477 napi_enable(&tqp_vector->napi);
478 enable_irq(tqp_vector->vector_irq);
481 hns3_mask_vector_irq(tqp_vector, 1);
484 static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector)
487 hns3_mask_vector_irq(tqp_vector, 0);
489 disable_irq(tqp_vector->vector_irq);
490 napi_disable(&tqp_vector->napi);
491 cancel_work_sync(&tqp_vector->rx_group.dim.work);
492 cancel_work_sync(&tqp_vector->tx_group.dim.work);
495 void hns3_set_vector_coalesce_rl(struct hns3_enet_tqp_vector *tqp_vector,
498 u32 rl_reg = hns3_rl_usec_to_reg(rl_value);
500 /* this defines the configuration for RL (Interrupt Rate Limiter).
501 * Rl defines rate of interrupts i.e. number of interrupts-per-second
502 * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing
504 if (rl_reg > 0 && !tqp_vector->tx_group.coal.adapt_enable &&
505 !tqp_vector->rx_group.coal.adapt_enable)
506 /* According to the hardware, the range of rl_reg is
507 * 0-59 and the unit is 4.
509 rl_reg |= HNS3_INT_RL_ENABLE_MASK;
511 writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
514 void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
519 if (tqp_vector->rx_group.coal.unit_1us)
520 new_val = gl_value | HNS3_INT_GL_1US;
522 new_val = hns3_gl_usec_to_reg(gl_value);
524 writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
527 void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
532 if (tqp_vector->tx_group.coal.unit_1us)
533 new_val = gl_value | HNS3_INT_GL_1US;
535 new_val = hns3_gl_usec_to_reg(gl_value);
537 writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
540 void hns3_set_vector_coalesce_tx_ql(struct hns3_enet_tqp_vector *tqp_vector,
543 writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_TX_QL_OFFSET);
546 void hns3_set_vector_coalesce_rx_ql(struct hns3_enet_tqp_vector *tqp_vector,
549 writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_RX_QL_OFFSET);
552 static void hns3_vector_coalesce_init(struct hns3_enet_tqp_vector *tqp_vector,
553 struct hns3_nic_priv *priv)
555 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
556 struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
557 struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
558 struct hns3_enet_coalesce *ptx_coal = &priv->tx_coal;
559 struct hns3_enet_coalesce *prx_coal = &priv->rx_coal;
561 tx_coal->adapt_enable = ptx_coal->adapt_enable;
562 rx_coal->adapt_enable = prx_coal->adapt_enable;
564 tx_coal->int_gl = ptx_coal->int_gl;
565 rx_coal->int_gl = prx_coal->int_gl;
567 rx_coal->flow_level = prx_coal->flow_level;
568 tx_coal->flow_level = ptx_coal->flow_level;
570 /* device version above V3(include V3), GL can configure 1us
571 * unit, so uses 1us unit.
573 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3) {
574 tx_coal->unit_1us = 1;
575 rx_coal->unit_1us = 1;
578 if (ae_dev->dev_specs.int_ql_max) {
579 tx_coal->ql_enable = 1;
580 rx_coal->ql_enable = 1;
581 tx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
582 rx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
583 tx_coal->int_ql = ptx_coal->int_ql;
584 rx_coal->int_ql = prx_coal->int_ql;
589 hns3_vector_coalesce_init_hw(struct hns3_enet_tqp_vector *tqp_vector,
590 struct hns3_nic_priv *priv)
592 struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
593 struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
594 struct hnae3_handle *h = priv->ae_handle;
596 hns3_set_vector_coalesce_tx_gl(tqp_vector, tx_coal->int_gl);
597 hns3_set_vector_coalesce_rx_gl(tqp_vector, rx_coal->int_gl);
598 hns3_set_vector_coalesce_rl(tqp_vector, h->kinfo.int_rl_setting);
600 if (tx_coal->ql_enable)
601 hns3_set_vector_coalesce_tx_ql(tqp_vector, tx_coal->int_ql);
603 if (rx_coal->ql_enable)
604 hns3_set_vector_coalesce_rx_ql(tqp_vector, rx_coal->int_ql);
607 static int hns3_nic_set_real_num_queue(struct net_device *netdev)
609 struct hnae3_handle *h = hns3_get_handle(netdev);
610 struct hnae3_knic_private_info *kinfo = &h->kinfo;
611 struct hnae3_tc_info *tc_info = &kinfo->tc_info;
612 unsigned int queue_size = kinfo->num_tqps;
615 if (tc_info->num_tc <= 1 && !tc_info->mqprio_active) {
616 netdev_reset_tc(netdev);
618 ret = netdev_set_num_tc(netdev, tc_info->num_tc);
621 "netdev_set_num_tc fail, ret=%d!\n", ret);
625 for (i = 0; i < HNAE3_MAX_TC; i++) {
626 if (!test_bit(i, &tc_info->tc_en))
629 netdev_set_tc_queue(netdev, i, tc_info->tqp_count[i],
630 tc_info->tqp_offset[i]);
634 ret = netif_set_real_num_tx_queues(netdev, queue_size);
637 "netif_set_real_num_tx_queues fail, ret=%d!\n", ret);
641 ret = netif_set_real_num_rx_queues(netdev, queue_size);
644 "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
651 u16 hns3_get_max_available_channels(struct hnae3_handle *h)
653 u16 alloc_tqps, max_rss_size, rss_size;
655 h->ae_algo->ops->get_tqps_and_rss_info(h, &alloc_tqps, &max_rss_size);
656 rss_size = alloc_tqps / h->kinfo.tc_info.num_tc;
658 return min_t(u16, rss_size, max_rss_size);
661 static void hns3_tqp_enable(struct hnae3_queue *tqp)
665 rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
666 rcb_reg |= BIT(HNS3_RING_EN_B);
667 hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
670 static void hns3_tqp_disable(struct hnae3_queue *tqp)
674 rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
675 rcb_reg &= ~BIT(HNS3_RING_EN_B);
676 hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
679 static void hns3_free_rx_cpu_rmap(struct net_device *netdev)
681 #ifdef CONFIG_RFS_ACCEL
682 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
683 netdev->rx_cpu_rmap = NULL;
687 static int hns3_set_rx_cpu_rmap(struct net_device *netdev)
689 #ifdef CONFIG_RFS_ACCEL
690 struct hns3_nic_priv *priv = netdev_priv(netdev);
691 struct hns3_enet_tqp_vector *tqp_vector;
694 if (!netdev->rx_cpu_rmap) {
695 netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->vector_num);
696 if (!netdev->rx_cpu_rmap)
700 for (i = 0; i < priv->vector_num; i++) {
701 tqp_vector = &priv->tqp_vector[i];
702 ret = irq_cpu_rmap_add(netdev->rx_cpu_rmap,
703 tqp_vector->vector_irq);
705 hns3_free_rx_cpu_rmap(netdev);
713 static int hns3_nic_net_up(struct net_device *netdev)
715 struct hns3_nic_priv *priv = netdev_priv(netdev);
716 struct hnae3_handle *h = priv->ae_handle;
720 ret = hns3_nic_reset_all_ring(h);
724 clear_bit(HNS3_NIC_STATE_DOWN, &priv->state);
726 /* enable the vectors */
727 for (i = 0; i < priv->vector_num; i++)
728 hns3_vector_enable(&priv->tqp_vector[i]);
731 for (j = 0; j < h->kinfo.num_tqps; j++)
732 hns3_tqp_enable(h->kinfo.tqp[j]);
734 /* start the ae_dev */
735 ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0;
737 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
739 hns3_tqp_disable(h->kinfo.tqp[j]);
741 for (j = i - 1; j >= 0; j--)
742 hns3_vector_disable(&priv->tqp_vector[j]);
748 static void hns3_config_xps(struct hns3_nic_priv *priv)
752 for (i = 0; i < priv->vector_num; i++) {
753 struct hns3_enet_tqp_vector *tqp_vector = &priv->tqp_vector[i];
754 struct hns3_enet_ring *ring = tqp_vector->tx_group.ring;
759 ret = netif_set_xps_queue(priv->netdev,
760 &tqp_vector->affinity_mask,
761 ring->tqp->tqp_index);
763 netdev_warn(priv->netdev,
764 "set xps queue failed: %d", ret);
771 static int hns3_nic_net_open(struct net_device *netdev)
773 struct hns3_nic_priv *priv = netdev_priv(netdev);
774 struct hnae3_handle *h = hns3_get_handle(netdev);
775 struct hnae3_knic_private_info *kinfo;
778 if (hns3_nic_resetting(netdev))
781 netif_carrier_off(netdev);
783 ret = hns3_nic_set_real_num_queue(netdev);
787 ret = hns3_nic_net_up(netdev);
789 netdev_err(netdev, "net up fail, ret=%d!\n", ret);
794 for (i = 0; i < HNAE3_MAX_USER_PRIO; i++)
795 netdev_set_prio_tc_map(netdev, i, kinfo->tc_info.prio_tc[i]);
797 if (h->ae_algo->ops->set_timer_task)
798 h->ae_algo->ops->set_timer_task(priv->ae_handle, true);
800 hns3_config_xps(priv);
802 netif_dbg(h, drv, netdev, "net open\n");
807 static void hns3_reset_tx_queue(struct hnae3_handle *h)
809 struct net_device *ndev = h->kinfo.netdev;
810 struct hns3_nic_priv *priv = netdev_priv(ndev);
811 struct netdev_queue *dev_queue;
814 for (i = 0; i < h->kinfo.num_tqps; i++) {
815 dev_queue = netdev_get_tx_queue(ndev,
816 priv->ring[i].queue_index);
817 netdev_tx_reset_queue(dev_queue);
821 static void hns3_nic_net_down(struct net_device *netdev)
823 struct hns3_nic_priv *priv = netdev_priv(netdev);
824 struct hnae3_handle *h = hns3_get_handle(netdev);
825 const struct hnae3_ae_ops *ops;
828 /* disable vectors */
829 for (i = 0; i < priv->vector_num; i++)
830 hns3_vector_disable(&priv->tqp_vector[i]);
833 for (i = 0; i < h->kinfo.num_tqps; i++)
834 hns3_tqp_disable(h->kinfo.tqp[i]);
837 ops = priv->ae_handle->ae_algo->ops;
839 ops->stop(priv->ae_handle);
841 /* delay ring buffer clearing to hns3_reset_notify_uninit_enet
842 * during reset process, because driver may not be able
843 * to disable the ring through firmware when downing the netdev.
845 if (!hns3_nic_resetting(netdev))
846 hns3_clear_all_ring(priv->ae_handle, false);
848 hns3_reset_tx_queue(priv->ae_handle);
851 static int hns3_nic_net_stop(struct net_device *netdev)
853 struct hns3_nic_priv *priv = netdev_priv(netdev);
854 struct hnae3_handle *h = hns3_get_handle(netdev);
856 if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
859 netif_dbg(h, drv, netdev, "net stop\n");
861 if (h->ae_algo->ops->set_timer_task)
862 h->ae_algo->ops->set_timer_task(priv->ae_handle, false);
864 netif_carrier_off(netdev);
865 netif_tx_disable(netdev);
867 hns3_nic_net_down(netdev);
872 static int hns3_nic_uc_sync(struct net_device *netdev,
873 const unsigned char *addr)
875 struct hnae3_handle *h = hns3_get_handle(netdev);
877 if (h->ae_algo->ops->add_uc_addr)
878 return h->ae_algo->ops->add_uc_addr(h, addr);
883 static int hns3_nic_uc_unsync(struct net_device *netdev,
884 const unsigned char *addr)
886 struct hnae3_handle *h = hns3_get_handle(netdev);
888 /* need ignore the request of removing device address, because
889 * we store the device address and other addresses of uc list
890 * in the function's mac filter list.
892 if (ether_addr_equal(addr, netdev->dev_addr))
895 if (h->ae_algo->ops->rm_uc_addr)
896 return h->ae_algo->ops->rm_uc_addr(h, addr);
901 static int hns3_nic_mc_sync(struct net_device *netdev,
902 const unsigned char *addr)
904 struct hnae3_handle *h = hns3_get_handle(netdev);
906 if (h->ae_algo->ops->add_mc_addr)
907 return h->ae_algo->ops->add_mc_addr(h, addr);
912 static int hns3_nic_mc_unsync(struct net_device *netdev,
913 const unsigned char *addr)
915 struct hnae3_handle *h = hns3_get_handle(netdev);
917 if (h->ae_algo->ops->rm_mc_addr)
918 return h->ae_algo->ops->rm_mc_addr(h, addr);
923 static u8 hns3_get_netdev_flags(struct net_device *netdev)
927 if (netdev->flags & IFF_PROMISC)
928 flags = HNAE3_USER_UPE | HNAE3_USER_MPE | HNAE3_BPE;
929 else if (netdev->flags & IFF_ALLMULTI)
930 flags = HNAE3_USER_MPE;
935 static void hns3_nic_set_rx_mode(struct net_device *netdev)
937 struct hnae3_handle *h = hns3_get_handle(netdev);
940 new_flags = hns3_get_netdev_flags(netdev);
942 __dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync);
943 __dev_mc_sync(netdev, hns3_nic_mc_sync, hns3_nic_mc_unsync);
945 /* User mode Promisc mode enable and vlan filtering is disabled to
946 * let all packets in.
948 h->netdev_flags = new_flags;
949 hns3_request_update_promisc_mode(h);
952 void hns3_request_update_promisc_mode(struct hnae3_handle *handle)
954 const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
956 if (ops->request_update_promisc_mode)
957 ops->request_update_promisc_mode(handle);
960 static u32 hns3_tx_spare_space(struct hns3_enet_ring *ring)
962 struct hns3_tx_spare *tx_spare = ring->tx_spare;
965 /* This smp_load_acquire() pairs with smp_store_release() in
966 * hns3_tx_spare_update() called in tx desc cleaning process.
968 ntc = smp_load_acquire(&tx_spare->last_to_clean);
969 ntu = tx_spare->next_to_use;
972 return ntc - ntu - 1;
974 /* The free tx buffer is divided into two part, so pick the
977 return max(ntc, tx_spare->len - ntu) - 1;
980 static void hns3_tx_spare_update(struct hns3_enet_ring *ring)
982 struct hns3_tx_spare *tx_spare = ring->tx_spare;
985 tx_spare->last_to_clean == tx_spare->next_to_clean)
988 /* This smp_store_release() pairs with smp_load_acquire() in
989 * hns3_tx_spare_space() called in xmit process.
991 smp_store_release(&tx_spare->last_to_clean,
992 tx_spare->next_to_clean);
995 static bool hns3_can_use_tx_bounce(struct hns3_enet_ring *ring,
999 u32 len = skb->len <= ring->tx_copybreak ? skb->len :
1002 if (len > ring->tx_copybreak)
1005 if (ALIGN(len, dma_get_cache_alignment()) > space) {
1006 u64_stats_update_begin(&ring->syncp);
1007 ring->stats.tx_spare_full++;
1008 u64_stats_update_end(&ring->syncp);
1015 static bool hns3_can_use_tx_sgl(struct hns3_enet_ring *ring,
1016 struct sk_buff *skb,
1019 if (skb->len <= ring->tx_copybreak || !tx_sgl ||
1020 (!skb_has_frag_list(skb) &&
1021 skb_shinfo(skb)->nr_frags < tx_sgl))
1024 if (space < HNS3_MAX_SGL_SIZE) {
1025 u64_stats_update_begin(&ring->syncp);
1026 ring->stats.tx_spare_full++;
1027 u64_stats_update_end(&ring->syncp);
1034 static void hns3_init_tx_spare_buffer(struct hns3_enet_ring *ring)
1036 struct hns3_tx_spare *tx_spare;
1042 alloc_size = tx_spare_buf_size ? tx_spare_buf_size :
1043 ring->tqp->handle->kinfo.tx_spare_buf_size;
1047 order = get_order(alloc_size);
1048 tx_spare = devm_kzalloc(ring_to_dev(ring), sizeof(*tx_spare),
1051 /* The driver still work without the tx spare buffer */
1052 dev_warn(ring_to_dev(ring), "failed to allocate hns3_tx_spare\n");
1056 page = alloc_pages_node(dev_to_node(ring_to_dev(ring)),
1059 dev_warn(ring_to_dev(ring), "failed to allocate tx spare pages\n");
1060 devm_kfree(ring_to_dev(ring), tx_spare);
1064 dma = dma_map_page(ring_to_dev(ring), page, 0,
1065 PAGE_SIZE << order, DMA_TO_DEVICE);
1066 if (dma_mapping_error(ring_to_dev(ring), dma)) {
1067 dev_warn(ring_to_dev(ring), "failed to map pages for tx spare\n");
1069 devm_kfree(ring_to_dev(ring), tx_spare);
1073 tx_spare->dma = dma;
1074 tx_spare->buf = page_address(page);
1075 tx_spare->len = PAGE_SIZE << order;
1076 ring->tx_spare = tx_spare;
1079 /* Use hns3_tx_spare_space() to make sure there is enough buffer
1080 * before calling below function to allocate tx buffer.
1082 static void *hns3_tx_spare_alloc(struct hns3_enet_ring *ring,
1083 unsigned int size, dma_addr_t *dma,
1086 struct hns3_tx_spare *tx_spare = ring->tx_spare;
1087 u32 ntu = tx_spare->next_to_use;
1089 size = ALIGN(size, dma_get_cache_alignment());
1092 /* Tx spare buffer wraps back here because the end of
1093 * freed tx buffer is not enough.
1095 if (ntu + size > tx_spare->len) {
1096 *cb_len += (tx_spare->len - ntu);
1100 tx_spare->next_to_use = ntu + size;
1101 if (tx_spare->next_to_use == tx_spare->len)
1102 tx_spare->next_to_use = 0;
1104 *dma = tx_spare->dma + ntu;
1106 return tx_spare->buf + ntu;
1109 static void hns3_tx_spare_rollback(struct hns3_enet_ring *ring, u32 len)
1111 struct hns3_tx_spare *tx_spare = ring->tx_spare;
1113 if (len > tx_spare->next_to_use) {
1114 len -= tx_spare->next_to_use;
1115 tx_spare->next_to_use = tx_spare->len - len;
1117 tx_spare->next_to_use -= len;
1121 static void hns3_tx_spare_reclaim_cb(struct hns3_enet_ring *ring,
1122 struct hns3_desc_cb *cb)
1124 struct hns3_tx_spare *tx_spare = ring->tx_spare;
1125 u32 ntc = tx_spare->next_to_clean;
1126 u32 len = cb->length;
1128 tx_spare->next_to_clean += len;
1130 if (tx_spare->next_to_clean >= tx_spare->len) {
1131 tx_spare->next_to_clean -= tx_spare->len;
1133 if (tx_spare->next_to_clean) {
1135 len = tx_spare->next_to_clean;
1139 /* This tx spare buffer is only really reclaimed after calling
1140 * hns3_tx_spare_update(), so it is still safe to use the info in
1141 * the tx buffer to do the dma sync or sg unmapping after
1142 * tx_spare->next_to_clean is moved forword.
1144 if (cb->type & (DESC_TYPE_BOUNCE_HEAD | DESC_TYPE_BOUNCE_ALL)) {
1145 dma_addr_t dma = tx_spare->dma + ntc;
1147 dma_sync_single_for_cpu(ring_to_dev(ring), dma, len,
1150 struct sg_table *sgt = tx_spare->buf + ntc;
1152 dma_unmap_sg(ring_to_dev(ring), sgt->sgl, sgt->orig_nents,
1157 static int hns3_set_tso(struct sk_buff *skb, u32 *paylen_fdop_ol4cs,
1158 u16 *mss, u32 *type_cs_vlan_tso, u32 *send_bytes)
1160 u32 l4_offset, hdr_len;
1161 union l3_hdr_info l3;
1162 union l4_hdr_info l4;
1166 if (!skb_is_gso(skb))
1169 ret = skb_cow_head(skb, 0);
1170 if (unlikely(ret < 0))
1173 l3.hdr = skb_network_header(skb);
1174 l4.hdr = skb_transport_header(skb);
1176 /* Software should clear the IPv4's checksum field when tso is
1179 if (l3.v4->version == 4)
1183 if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
1185 SKB_GSO_UDP_TUNNEL |
1186 SKB_GSO_UDP_TUNNEL_CSUM)) {
1187 /* reset l3&l4 pointers from outer to inner headers */
1188 l3.hdr = skb_inner_network_header(skb);
1189 l4.hdr = skb_inner_transport_header(skb);
1191 /* Software should clear the IPv4's checksum field when
1194 if (l3.v4->version == 4)
1198 /* normal or tunnel packet */
1199 l4_offset = l4.hdr - skb->data;
1201 /* remove payload length from inner pseudo checksum when tso */
1202 l4_paylen = skb->len - l4_offset;
1204 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) {
1205 hdr_len = sizeof(*l4.udp) + l4_offset;
1206 csum_replace_by_diff(&l4.udp->check,
1207 (__force __wsum)htonl(l4_paylen));
1209 hdr_len = (l4.tcp->doff << 2) + l4_offset;
1210 csum_replace_by_diff(&l4.tcp->check,
1211 (__force __wsum)htonl(l4_paylen));
1214 *send_bytes = (skb_shinfo(skb)->gso_segs - 1) * hdr_len + skb->len;
1216 /* find the txbd field values */
1217 *paylen_fdop_ol4cs = skb->len - hdr_len;
1218 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_TSO_B, 1);
1220 /* offload outer UDP header checksum */
1221 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)
1222 hns3_set_field(*paylen_fdop_ol4cs, HNS3_TXD_OL4CS_B, 1);
1224 /* get MSS for TSO */
1225 *mss = skb_shinfo(skb)->gso_size;
1227 trace_hns3_tso(skb);
1232 static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
1235 union l3_hdr_info l3;
1236 unsigned char *l4_hdr;
1237 unsigned char *exthdr;
1241 /* find outer header point */
1242 l3.hdr = skb_network_header(skb);
1243 l4_hdr = skb_transport_header(skb);
1245 if (skb->protocol == htons(ETH_P_IPV6)) {
1246 exthdr = l3.hdr + sizeof(*l3.v6);
1247 l4_proto_tmp = l3.v6->nexthdr;
1248 if (l4_hdr != exthdr)
1249 ipv6_skip_exthdr(skb, exthdr - skb->data,
1250 &l4_proto_tmp, &frag_off);
1251 } else if (skb->protocol == htons(ETH_P_IP)) {
1252 l4_proto_tmp = l3.v4->protocol;
1257 *ol4_proto = l4_proto_tmp;
1260 if (!skb->encapsulation) {
1265 /* find inner header point */
1266 l3.hdr = skb_inner_network_header(skb);
1267 l4_hdr = skb_inner_transport_header(skb);
1269 if (l3.v6->version == 6) {
1270 exthdr = l3.hdr + sizeof(*l3.v6);
1271 l4_proto_tmp = l3.v6->nexthdr;
1272 if (l4_hdr != exthdr)
1273 ipv6_skip_exthdr(skb, exthdr - skb->data,
1274 &l4_proto_tmp, &frag_off);
1275 } else if (l3.v4->version == 4) {
1276 l4_proto_tmp = l3.v4->protocol;
1279 *il4_proto = l4_proto_tmp;
1284 /* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
1285 * and it is udp packet, which has a dest port as the IANA assigned.
1286 * the hardware is expected to do the checksum offload, but the
1287 * hardware will not do the checksum offload when udp dest port is
1288 * 4789, 4790 or 6081.
1290 static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
1292 struct hns3_nic_priv *priv = netdev_priv(skb->dev);
1293 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
1294 union l4_hdr_info l4;
1296 /* device version above V3(include V3), the hardware can
1297 * do this checksum offload.
1299 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
1302 l4.hdr = skb_transport_header(skb);
1304 if (!(!skb->encapsulation &&
1305 (l4.udp->dest == htons(IANA_VXLAN_UDP_PORT) ||
1306 l4.udp->dest == htons(GENEVE_UDP_PORT) ||
1307 l4.udp->dest == htons(4790))))
1313 static void hns3_set_outer_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
1314 u32 *ol_type_vlan_len_msec)
1316 u32 l2_len, l3_len, l4_len;
1317 unsigned char *il2_hdr;
1318 union l3_hdr_info l3;
1319 union l4_hdr_info l4;
1321 l3.hdr = skb_network_header(skb);
1322 l4.hdr = skb_transport_header(skb);
1324 /* compute OL2 header size, defined in 2 Bytes */
1325 l2_len = l3.hdr - skb->data;
1326 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L2LEN_S, l2_len >> 1);
1328 /* compute OL3 header size, defined in 4 Bytes */
1329 l3_len = l4.hdr - l3.hdr;
1330 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_S, l3_len >> 2);
1332 il2_hdr = skb_inner_mac_header(skb);
1333 /* compute OL4 header size, defined in 4 Bytes */
1334 l4_len = il2_hdr - l4.hdr;
1335 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L4LEN_S, l4_len >> 2);
1337 /* define outer network header type */
1338 if (skb->protocol == htons(ETH_P_IP)) {
1339 if (skb_is_gso(skb))
1340 hns3_set_field(*ol_type_vlan_len_msec,
1342 HNS3_OL3T_IPV4_CSUM);
1344 hns3_set_field(*ol_type_vlan_len_msec,
1346 HNS3_OL3T_IPV4_NO_CSUM);
1347 } else if (skb->protocol == htons(ETH_P_IPV6)) {
1348 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_S,
1352 if (ol4_proto == IPPROTO_UDP)
1353 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
1354 HNS3_TUN_MAC_IN_UDP);
1355 else if (ol4_proto == IPPROTO_GRE)
1356 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
1360 static int hns3_set_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
1361 u8 il4_proto, u32 *type_cs_vlan_tso,
1362 u32 *ol_type_vlan_len_msec)
1364 unsigned char *l2_hdr = skb->data;
1365 u32 l4_proto = ol4_proto;
1366 union l4_hdr_info l4;
1367 union l3_hdr_info l3;
1370 l4.hdr = skb_transport_header(skb);
1371 l3.hdr = skb_network_header(skb);
1373 /* handle encapsulation skb */
1374 if (skb->encapsulation) {
1375 /* If this is a not UDP/GRE encapsulation skb */
1376 if (!(ol4_proto == IPPROTO_UDP || ol4_proto == IPPROTO_GRE)) {
1377 /* drop the skb tunnel packet if hardware don't support,
1378 * because hardware can't calculate csum when TSO.
1380 if (skb_is_gso(skb))
1383 /* the stack computes the IP header already,
1384 * driver calculate l4 checksum when not TSO.
1386 return skb_checksum_help(skb);
1389 hns3_set_outer_l2l3l4(skb, ol4_proto, ol_type_vlan_len_msec);
1391 /* switch to inner header */
1392 l2_hdr = skb_inner_mac_header(skb);
1393 l3.hdr = skb_inner_network_header(skb);
1394 l4.hdr = skb_inner_transport_header(skb);
1395 l4_proto = il4_proto;
1398 if (l3.v4->version == 4) {
1399 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
1402 /* the stack computes the IP header already, the only time we
1403 * need the hardware to recompute it is in the case of TSO.
1405 if (skb_is_gso(skb))
1406 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3CS_B, 1);
1407 } else if (l3.v6->version == 6) {
1408 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
1412 /* compute inner(/normal) L2 header size, defined in 2 Bytes */
1413 l2_len = l3.hdr - l2_hdr;
1414 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S, l2_len >> 1);
1416 /* compute inner(/normal) L3 header size, defined in 4 Bytes */
1417 l3_len = l4.hdr - l3.hdr;
1418 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_S, l3_len >> 2);
1420 /* compute inner(/normal) L4 header size, defined in 4 Bytes */
1423 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1424 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1426 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1430 if (hns3_tunnel_csum_bug(skb))
1431 return skb_checksum_help(skb);
1433 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1434 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1436 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1437 (sizeof(struct udphdr) >> 2));
1440 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1441 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1443 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1444 (sizeof(struct sctphdr) >> 2));
1447 /* drop the skb tunnel packet if hardware don't support,
1448 * because hardware can't calculate csum when TSO.
1450 if (skb_is_gso(skb))
1453 /* the stack computes the IP header already,
1454 * driver calculate l4 checksum when not TSO.
1456 return skb_checksum_help(skb);
1462 static int hns3_handle_vtags(struct hns3_enet_ring *tx_ring,
1463 struct sk_buff *skb)
1465 struct hnae3_handle *handle = tx_ring->tqp->handle;
1466 struct hnae3_ae_dev *ae_dev;
1467 struct vlan_ethhdr *vhdr;
1470 if (!(skb->protocol == htons(ETH_P_8021Q) ||
1471 skb_vlan_tag_present(skb)))
1474 /* For HW limitation on HNAE3_DEVICE_VERSION_V2, if port based insert
1475 * VLAN enabled, only one VLAN header is allowed in skb, otherwise it
1476 * will cause RAS error.
1478 ae_dev = pci_get_drvdata(handle->pdev);
1479 if (unlikely(skb_vlan_tagged_multi(skb) &&
1480 ae_dev->dev_version <= HNAE3_DEVICE_VERSION_V2 &&
1481 handle->port_base_vlan_state ==
1482 HNAE3_PORT_BASE_VLAN_ENABLE))
1485 if (skb->protocol == htons(ETH_P_8021Q) &&
1486 !(handle->kinfo.netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
1487 /* When HW VLAN acceleration is turned off, and the stack
1488 * sets the protocol to 802.1q, the driver just need to
1489 * set the protocol to the encapsulated ethertype.
1491 skb->protocol = vlan_get_protocol(skb);
1495 if (skb_vlan_tag_present(skb)) {
1496 /* Based on hw strategy, use out_vtag in two layer tag case,
1497 * and use inner_vtag in one tag case.
1499 if (skb->protocol == htons(ETH_P_8021Q) &&
1500 handle->port_base_vlan_state ==
1501 HNAE3_PORT_BASE_VLAN_DISABLE)
1502 rc = HNS3_OUTER_VLAN_TAG;
1504 rc = HNS3_INNER_VLAN_TAG;
1506 skb->protocol = vlan_get_protocol(skb);
1510 rc = skb_cow_head(skb, 0);
1511 if (unlikely(rc < 0))
1514 vhdr = (struct vlan_ethhdr *)skb->data;
1515 vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority << VLAN_PRIO_SHIFT)
1518 skb->protocol = vlan_get_protocol(skb);
1522 /* check if the hardware is capable of checksum offloading */
1523 static bool hns3_check_hw_tx_csum(struct sk_buff *skb)
1525 struct hns3_nic_priv *priv = netdev_priv(skb->dev);
1527 /* Kindly note, due to backward compatibility of the TX descriptor,
1528 * HW checksum of the non-IP packets and GSO packets is handled at
1529 * different place in the following code
1531 if (skb_csum_is_sctp(skb) || skb_is_gso(skb) ||
1532 !test_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state))
1538 static int hns3_fill_skb_desc(struct hns3_enet_ring *ring,
1539 struct sk_buff *skb, struct hns3_desc *desc,
1540 struct hns3_desc_cb *desc_cb)
1542 u32 ol_type_vlan_len_msec = 0;
1543 u32 paylen_ol4cs = skb->len;
1544 u32 type_cs_vlan_tso = 0;
1545 u16 mss_hw_csum = 0;
1550 ret = hns3_handle_vtags(ring, skb);
1551 if (unlikely(ret < 0)) {
1552 u64_stats_update_begin(&ring->syncp);
1553 ring->stats.tx_vlan_err++;
1554 u64_stats_update_end(&ring->syncp);
1556 } else if (ret == HNS3_INNER_VLAN_TAG) {
1557 inner_vtag = skb_vlan_tag_get(skb);
1558 inner_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1560 hns3_set_field(type_cs_vlan_tso, HNS3_TXD_VLAN_B, 1);
1561 } else if (ret == HNS3_OUTER_VLAN_TAG) {
1562 out_vtag = skb_vlan_tag_get(skb);
1563 out_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1565 hns3_set_field(ol_type_vlan_len_msec, HNS3_TXD_OVLAN_B,
1569 desc_cb->send_bytes = skb->len;
1571 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1572 u8 ol4_proto, il4_proto;
1574 if (hns3_check_hw_tx_csum(skb)) {
1575 /* set checksum start and offset, defined in 2 Bytes */
1576 hns3_set_field(type_cs_vlan_tso, HNS3_TXD_CSUM_START_S,
1577 skb_checksum_start_offset(skb) >> 1);
1578 hns3_set_field(ol_type_vlan_len_msec,
1579 HNS3_TXD_CSUM_OFFSET_S,
1580 skb->csum_offset >> 1);
1581 mss_hw_csum |= BIT(HNS3_TXD_HW_CS_B);
1582 goto out_hw_tx_csum;
1585 skb_reset_mac_len(skb);
1587 ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
1588 if (unlikely(ret < 0)) {
1589 u64_stats_update_begin(&ring->syncp);
1590 ring->stats.tx_l4_proto_err++;
1591 u64_stats_update_end(&ring->syncp);
1595 ret = hns3_set_l2l3l4(skb, ol4_proto, il4_proto,
1597 &ol_type_vlan_len_msec);
1598 if (unlikely(ret < 0)) {
1599 u64_stats_update_begin(&ring->syncp);
1600 ring->stats.tx_l2l3l4_err++;
1601 u64_stats_update_end(&ring->syncp);
1605 ret = hns3_set_tso(skb, &paylen_ol4cs, &mss_hw_csum,
1606 &type_cs_vlan_tso, &desc_cb->send_bytes);
1607 if (unlikely(ret < 0)) {
1608 u64_stats_update_begin(&ring->syncp);
1609 ring->stats.tx_tso_err++;
1610 u64_stats_update_end(&ring->syncp);
1617 desc->tx.ol_type_vlan_len_msec =
1618 cpu_to_le32(ol_type_vlan_len_msec);
1619 desc->tx.type_cs_vlan_tso_len = cpu_to_le32(type_cs_vlan_tso);
1620 desc->tx.paylen_ol4cs = cpu_to_le32(paylen_ol4cs);
1621 desc->tx.mss_hw_csum = cpu_to_le16(mss_hw_csum);
1622 desc->tx.vlan_tag = cpu_to_le16(inner_vtag);
1623 desc->tx.outer_vlan_tag = cpu_to_le16(out_vtag);
1628 static int hns3_fill_desc(struct hns3_enet_ring *ring, dma_addr_t dma,
1631 #define HNS3_LIKELY_BD_NUM 1
1633 struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1634 unsigned int frag_buf_num;
1637 if (likely(size <= HNS3_MAX_BD_SIZE)) {
1638 desc->addr = cpu_to_le64(dma);
1639 desc->tx.send_size = cpu_to_le16(size);
1640 desc->tx.bdtp_fe_sc_vld_ra_ri =
1641 cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1643 trace_hns3_tx_desc(ring, ring->next_to_use);
1644 ring_ptr_move_fw(ring, next_to_use);
1645 return HNS3_LIKELY_BD_NUM;
1648 frag_buf_num = hns3_tx_bd_count(size);
1649 sizeoflast = size % HNS3_MAX_BD_SIZE;
1650 sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1652 /* When frag size is bigger than hardware limit, split this frag */
1653 for (k = 0; k < frag_buf_num; k++) {
1654 /* now, fill the descriptor */
1655 desc->addr = cpu_to_le64(dma + HNS3_MAX_BD_SIZE * k);
1656 desc->tx.send_size = cpu_to_le16((k == frag_buf_num - 1) ?
1657 (u16)sizeoflast : (u16)HNS3_MAX_BD_SIZE);
1658 desc->tx.bdtp_fe_sc_vld_ra_ri =
1659 cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1661 trace_hns3_tx_desc(ring, ring->next_to_use);
1662 /* move ring pointer to next */
1663 ring_ptr_move_fw(ring, next_to_use);
1665 desc = &ring->desc[ring->next_to_use];
1668 return frag_buf_num;
1671 static int hns3_map_and_fill_desc(struct hns3_enet_ring *ring, void *priv,
1674 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1675 struct device *dev = ring_to_dev(ring);
1679 if (type & (DESC_TYPE_FRAGLIST_SKB | DESC_TYPE_SKB)) {
1680 struct sk_buff *skb = (struct sk_buff *)priv;
1682 size = skb_headlen(skb);
1686 dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1687 } else if (type & DESC_TYPE_BOUNCE_HEAD) {
1688 /* Head data has been filled in hns3_handle_tx_bounce(),
1689 * just return 0 here.
1693 skb_frag_t *frag = (skb_frag_t *)priv;
1695 size = skb_frag_size(frag);
1699 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1702 if (unlikely(dma_mapping_error(dev, dma))) {
1703 u64_stats_update_begin(&ring->syncp);
1704 ring->stats.sw_err_cnt++;
1705 u64_stats_update_end(&ring->syncp);
1709 desc_cb->priv = priv;
1710 desc_cb->length = size;
1712 desc_cb->type = type;
1714 return hns3_fill_desc(ring, dma, size);
1717 static unsigned int hns3_skb_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1718 unsigned int bd_num)
1723 size = skb_headlen(skb);
1724 while (size > HNS3_MAX_BD_SIZE) {
1725 bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1726 size -= HNS3_MAX_BD_SIZE;
1728 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1733 bd_size[bd_num++] = size;
1734 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1738 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1739 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1740 size = skb_frag_size(frag);
1744 while (size > HNS3_MAX_BD_SIZE) {
1745 bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1746 size -= HNS3_MAX_BD_SIZE;
1748 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1752 bd_size[bd_num++] = size;
1753 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1760 static unsigned int hns3_tx_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1761 u8 max_non_tso_bd_num, unsigned int bd_num,
1762 unsigned int recursion_level)
1764 #define HNS3_MAX_RECURSION_LEVEL 24
1766 struct sk_buff *frag_skb;
1768 /* If the total len is within the max bd limit */
1769 if (likely(skb->len <= HNS3_MAX_BD_SIZE && !recursion_level &&
1770 !skb_has_frag_list(skb) &&
1771 skb_shinfo(skb)->nr_frags < max_non_tso_bd_num))
1772 return skb_shinfo(skb)->nr_frags + 1U;
1774 if (unlikely(recursion_level >= HNS3_MAX_RECURSION_LEVEL))
1777 bd_num = hns3_skb_bd_num(skb, bd_size, bd_num);
1778 if (!skb_has_frag_list(skb) || bd_num > HNS3_MAX_TSO_BD_NUM)
1781 skb_walk_frags(skb, frag_skb) {
1782 bd_num = hns3_tx_bd_num(frag_skb, bd_size, max_non_tso_bd_num,
1783 bd_num, recursion_level + 1);
1784 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1791 static unsigned int hns3_gso_hdr_len(struct sk_buff *skb)
1793 if (!skb->encapsulation)
1794 return skb_transport_offset(skb) + tcp_hdrlen(skb);
1796 return skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb);
1799 /* HW need every continuous max_non_tso_bd_num buffer data to be larger
1800 * than MSS, we simplify it by ensuring skb_headlen + the first continuous
1801 * max_non_tso_bd_num - 1 frags to be larger than gso header len + mss,
1802 * and the remaining continuous max_non_tso_bd_num - 1 frags to be larger
1803 * than MSS except the last max_non_tso_bd_num - 1 frags.
1805 static bool hns3_skb_need_linearized(struct sk_buff *skb, unsigned int *bd_size,
1806 unsigned int bd_num, u8 max_non_tso_bd_num)
1808 unsigned int tot_len = 0;
1811 for (i = 0; i < max_non_tso_bd_num - 1U; i++)
1812 tot_len += bd_size[i];
1814 /* ensure the first max_non_tso_bd_num frags is greater than
1817 if (tot_len + bd_size[max_non_tso_bd_num - 1U] <
1818 skb_shinfo(skb)->gso_size + hns3_gso_hdr_len(skb))
1821 /* ensure every continuous max_non_tso_bd_num - 1 buffer is greater
1822 * than mss except the last one.
1824 for (i = 0; i < bd_num - max_non_tso_bd_num; i++) {
1825 tot_len -= bd_size[i];
1826 tot_len += bd_size[i + max_non_tso_bd_num - 1U];
1828 if (tot_len < skb_shinfo(skb)->gso_size)
1835 void hns3_shinfo_pack(struct skb_shared_info *shinfo, __u32 *size)
1839 for (i = 0; i < MAX_SKB_FRAGS; i++)
1840 size[i] = skb_frag_size(&shinfo->frags[i]);
1843 static int hns3_skb_linearize(struct hns3_enet_ring *ring,
1844 struct sk_buff *skb,
1845 u8 max_non_tso_bd_num,
1846 unsigned int bd_num)
1848 /* 'bd_num == UINT_MAX' means the skb' fraglist has a
1849 * recursion level of over HNS3_MAX_RECURSION_LEVEL.
1851 if (bd_num == UINT_MAX) {
1852 u64_stats_update_begin(&ring->syncp);
1853 ring->stats.over_max_recursion++;
1854 u64_stats_update_end(&ring->syncp);
1858 /* The skb->len has exceeded the hw limitation, linearization
1861 if (skb->len > HNS3_MAX_TSO_SIZE ||
1862 (!skb_is_gso(skb) && skb->len >
1863 HNS3_MAX_NON_TSO_SIZE(max_non_tso_bd_num))) {
1864 u64_stats_update_begin(&ring->syncp);
1865 ring->stats.hw_limitation++;
1866 u64_stats_update_end(&ring->syncp);
1870 if (__skb_linearize(skb)) {
1871 u64_stats_update_begin(&ring->syncp);
1872 ring->stats.sw_err_cnt++;
1873 u64_stats_update_end(&ring->syncp);
1880 static int hns3_nic_maybe_stop_tx(struct hns3_enet_ring *ring,
1881 struct net_device *netdev,
1882 struct sk_buff *skb)
1884 struct hns3_nic_priv *priv = netdev_priv(netdev);
1885 u8 max_non_tso_bd_num = priv->max_non_tso_bd_num;
1886 unsigned int bd_size[HNS3_MAX_TSO_BD_NUM + 1U];
1887 unsigned int bd_num;
1889 bd_num = hns3_tx_bd_num(skb, bd_size, max_non_tso_bd_num, 0, 0);
1890 if (unlikely(bd_num > max_non_tso_bd_num)) {
1891 if (bd_num <= HNS3_MAX_TSO_BD_NUM && skb_is_gso(skb) &&
1892 !hns3_skb_need_linearized(skb, bd_size, bd_num,
1893 max_non_tso_bd_num)) {
1894 trace_hns3_over_max_bd(skb);
1898 if (hns3_skb_linearize(ring, skb, max_non_tso_bd_num,
1902 bd_num = hns3_tx_bd_count(skb->len);
1904 u64_stats_update_begin(&ring->syncp);
1905 ring->stats.tx_copy++;
1906 u64_stats_update_end(&ring->syncp);
1910 if (likely(ring_space(ring) >= bd_num))
1913 netif_stop_subqueue(netdev, ring->queue_index);
1914 smp_mb(); /* Memory barrier before checking ring_space */
1916 /* Start queue in case hns3_clean_tx_ring has just made room
1917 * available and has not seen the queue stopped state performed
1918 * by netif_stop_subqueue above.
1920 if (ring_space(ring) >= bd_num && netif_carrier_ok(netdev) &&
1921 !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
1922 netif_start_subqueue(netdev, ring->queue_index);
1926 u64_stats_update_begin(&ring->syncp);
1927 ring->stats.tx_busy++;
1928 u64_stats_update_end(&ring->syncp);
1933 static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
1935 struct device *dev = ring_to_dev(ring);
1938 for (i = 0; i < ring->desc_num; i++) {
1939 struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1940 struct hns3_desc_cb *desc_cb;
1942 memset(desc, 0, sizeof(*desc));
1944 /* check if this is where we started */
1945 if (ring->next_to_use == next_to_use_orig)
1949 ring_ptr_move_bw(ring, next_to_use);
1951 desc_cb = &ring->desc_cb[ring->next_to_use];
1956 /* unmap the descriptor dma address */
1957 if (desc_cb->type & (DESC_TYPE_SKB | DESC_TYPE_FRAGLIST_SKB))
1958 dma_unmap_single(dev, desc_cb->dma, desc_cb->length,
1960 else if (desc_cb->type &
1961 (DESC_TYPE_BOUNCE_HEAD | DESC_TYPE_BOUNCE_ALL))
1962 hns3_tx_spare_rollback(ring, desc_cb->length);
1963 else if (desc_cb->length)
1964 dma_unmap_page(dev, desc_cb->dma, desc_cb->length,
1967 desc_cb->length = 0;
1969 desc_cb->type = DESC_TYPE_UNKNOWN;
1973 static int hns3_fill_skb_to_desc(struct hns3_enet_ring *ring,
1974 struct sk_buff *skb, unsigned int type)
1976 struct sk_buff *frag_skb;
1977 int i, ret, bd_num = 0;
1979 ret = hns3_map_and_fill_desc(ring, skb, type);
1980 if (unlikely(ret < 0))
1985 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1986 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1988 ret = hns3_map_and_fill_desc(ring, frag, DESC_TYPE_PAGE);
1989 if (unlikely(ret < 0))
1995 skb_walk_frags(skb, frag_skb) {
1996 ret = hns3_fill_skb_to_desc(ring, frag_skb,
1997 DESC_TYPE_FRAGLIST_SKB);
1998 if (unlikely(ret < 0))
2007 static void hns3_tx_doorbell(struct hns3_enet_ring *ring, int num,
2010 ring->pending_buf += num;
2013 u64_stats_update_begin(&ring->syncp);
2014 ring->stats.tx_more++;
2015 u64_stats_update_end(&ring->syncp);
2019 if (!ring->pending_buf)
2022 writel(ring->pending_buf,
2023 ring->tqp->io_base + HNS3_RING_TX_RING_TAIL_REG);
2024 ring->pending_buf = 0;
2025 WRITE_ONCE(ring->last_to_use, ring->next_to_use);
2028 static void hns3_tsyn(struct net_device *netdev, struct sk_buff *skb,
2029 struct hns3_desc *desc)
2031 struct hnae3_handle *h = hns3_get_handle(netdev);
2033 if (!(h->ae_algo->ops->set_tx_hwts_info &&
2034 h->ae_algo->ops->set_tx_hwts_info(h, skb)))
2037 desc->tx.bdtp_fe_sc_vld_ra_ri |= cpu_to_le16(BIT(HNS3_TXD_TSYN_B));
2040 static int hns3_handle_tx_bounce(struct hns3_enet_ring *ring,
2041 struct sk_buff *skb)
2043 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
2044 unsigned int type = DESC_TYPE_BOUNCE_HEAD;
2045 unsigned int size = skb_headlen(skb);
2052 if (skb->len <= ring->tx_copybreak) {
2054 type = DESC_TYPE_BOUNCE_ALL;
2057 /* hns3_can_use_tx_bounce() is called to ensure the below
2058 * function can always return the tx buffer.
2060 buf = hns3_tx_spare_alloc(ring, size, &dma, &cb_len);
2062 ret = skb_copy_bits(skb, 0, buf, size);
2063 if (unlikely(ret < 0)) {
2064 hns3_tx_spare_rollback(ring, cb_len);
2065 u64_stats_update_begin(&ring->syncp);
2066 ring->stats.copy_bits_err++;
2067 u64_stats_update_end(&ring->syncp);
2071 desc_cb->priv = skb;
2072 desc_cb->length = cb_len;
2074 desc_cb->type = type;
2076 bd_num += hns3_fill_desc(ring, dma, size);
2078 if (type == DESC_TYPE_BOUNCE_HEAD) {
2079 ret = hns3_fill_skb_to_desc(ring, skb,
2080 DESC_TYPE_BOUNCE_HEAD);
2081 if (unlikely(ret < 0))
2087 dma_sync_single_for_device(ring_to_dev(ring), dma, size,
2090 u64_stats_update_begin(&ring->syncp);
2091 ring->stats.tx_bounce++;
2092 u64_stats_update_end(&ring->syncp);
2096 static int hns3_handle_tx_sgl(struct hns3_enet_ring *ring,
2097 struct sk_buff *skb)
2099 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
2100 u32 nfrag = skb_shinfo(skb)->nr_frags + 1;
2101 struct sg_table *sgt;
2107 if (skb_has_frag_list(skb))
2108 nfrag = HNS3_MAX_TSO_BD_NUM;
2110 /* hns3_can_use_tx_sgl() is called to ensure the below
2111 * function can always return the tx buffer.
2113 sgt = hns3_tx_spare_alloc(ring, HNS3_SGL_SIZE(nfrag),
2116 /* scatterlist follows by the sg table */
2117 sgt->sgl = (struct scatterlist *)(sgt + 1);
2118 sg_init_table(sgt->sgl, nfrag);
2119 nents = skb_to_sgvec(skb, sgt->sgl, 0, skb->len);
2120 if (unlikely(nents < 0)) {
2121 hns3_tx_spare_rollback(ring, cb_len);
2122 u64_stats_update_begin(&ring->syncp);
2123 ring->stats.skb2sgl_err++;
2124 u64_stats_update_end(&ring->syncp);
2128 sgt->orig_nents = nents;
2129 sgt->nents = dma_map_sg(ring_to_dev(ring), sgt->sgl, sgt->orig_nents,
2131 if (unlikely(!sgt->nents)) {
2132 hns3_tx_spare_rollback(ring, cb_len);
2133 u64_stats_update_begin(&ring->syncp);
2134 ring->stats.map_sg_err++;
2135 u64_stats_update_end(&ring->syncp);
2139 desc_cb->priv = skb;
2140 desc_cb->length = cb_len;
2142 desc_cb->type = DESC_TYPE_SGL_SKB;
2144 for (i = 0; i < sgt->nents; i++)
2145 bd_num += hns3_fill_desc(ring, sg_dma_address(sgt->sgl + i),
2146 sg_dma_len(sgt->sgl + i));
2148 u64_stats_update_begin(&ring->syncp);
2149 ring->stats.tx_sgl++;
2150 u64_stats_update_end(&ring->syncp);
2155 static int hns3_handle_desc_filling(struct hns3_enet_ring *ring,
2156 struct sk_buff *skb)
2160 if (!ring->tx_spare)
2163 space = hns3_tx_spare_space(ring);
2165 if (hns3_can_use_tx_sgl(ring, skb, space))
2166 return hns3_handle_tx_sgl(ring, skb);
2168 if (hns3_can_use_tx_bounce(ring, skb, space))
2169 return hns3_handle_tx_bounce(ring, skb);
2172 return hns3_fill_skb_to_desc(ring, skb, DESC_TYPE_SKB);
2175 netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
2177 struct hns3_nic_priv *priv = netdev_priv(netdev);
2178 struct hns3_enet_ring *ring = &priv->ring[skb->queue_mapping];
2179 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
2180 struct netdev_queue *dev_queue;
2181 int pre_ntu, next_to_use_head;
2185 /* Hardware can only handle short frames above 32 bytes */
2186 if (skb_put_padto(skb, HNS3_MIN_TX_LEN)) {
2187 hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
2189 u64_stats_update_begin(&ring->syncp);
2190 ring->stats.sw_err_cnt++;
2191 u64_stats_update_end(&ring->syncp);
2193 return NETDEV_TX_OK;
2196 /* Prefetch the data used later */
2197 prefetch(skb->data);
2199 ret = hns3_nic_maybe_stop_tx(ring, netdev, skb);
2200 if (unlikely(ret <= 0)) {
2201 if (ret == -EBUSY) {
2202 hns3_tx_doorbell(ring, 0, true);
2203 return NETDEV_TX_BUSY;
2206 hns3_rl_err(netdev, "xmit error: %d!\n", ret);
2210 next_to_use_head = ring->next_to_use;
2212 ret = hns3_fill_skb_desc(ring, skb, &ring->desc[ring->next_to_use],
2214 if (unlikely(ret < 0))
2217 /* 'ret < 0' means filling error, 'ret == 0' means skb->len is
2218 * zero, which is unlikely, and 'ret > 0' means how many tx desc
2219 * need to be notified to the hw.
2221 ret = hns3_handle_desc_filling(ring, skb);
2222 if (unlikely(ret <= 0))
2225 pre_ntu = ring->next_to_use ? (ring->next_to_use - 1) :
2226 (ring->desc_num - 1);
2228 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP))
2229 hns3_tsyn(netdev, skb, &ring->desc[pre_ntu]);
2231 ring->desc[pre_ntu].tx.bdtp_fe_sc_vld_ra_ri |=
2232 cpu_to_le16(BIT(HNS3_TXD_FE_B));
2233 trace_hns3_tx_desc(ring, pre_ntu);
2235 skb_tx_timestamp(skb);
2237 /* Complete translate all packets */
2238 dev_queue = netdev_get_tx_queue(netdev, ring->queue_index);
2239 doorbell = __netdev_tx_sent_queue(dev_queue, desc_cb->send_bytes,
2240 netdev_xmit_more());
2241 hns3_tx_doorbell(ring, ret, doorbell);
2243 return NETDEV_TX_OK;
2246 hns3_clear_desc(ring, next_to_use_head);
2249 dev_kfree_skb_any(skb);
2250 hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
2251 return NETDEV_TX_OK;
2254 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
2256 struct hnae3_handle *h = hns3_get_handle(netdev);
2257 struct sockaddr *mac_addr = p;
2260 if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
2261 return -EADDRNOTAVAIL;
2263 if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
2264 netdev_info(netdev, "already using mac address %pM\n",
2269 /* For VF device, if there is a perm_addr, then the user will not
2270 * be allowed to change the address.
2272 if (!hns3_is_phys_func(h->pdev) &&
2273 !is_zero_ether_addr(netdev->perm_addr)) {
2274 netdev_err(netdev, "has permanent MAC %pM, user MAC %pM not allow\n",
2275 netdev->perm_addr, mac_addr->sa_data);
2279 ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
2281 netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
2285 ether_addr_copy(netdev->dev_addr, mac_addr->sa_data);
2290 static int hns3_nic_do_ioctl(struct net_device *netdev,
2291 struct ifreq *ifr, int cmd)
2293 struct hnae3_handle *h = hns3_get_handle(netdev);
2295 if (!netif_running(netdev))
2298 if (!h->ae_algo->ops->do_ioctl)
2301 return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
2304 static int hns3_nic_set_features(struct net_device *netdev,
2305 netdev_features_t features)
2307 netdev_features_t changed = netdev->features ^ features;
2308 struct hns3_nic_priv *priv = netdev_priv(netdev);
2309 struct hnae3_handle *h = priv->ae_handle;
2313 if (changed & (NETIF_F_GRO_HW) && h->ae_algo->ops->set_gro_en) {
2314 enable = !!(features & NETIF_F_GRO_HW);
2315 ret = h->ae_algo->ops->set_gro_en(h, enable);
2320 if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
2321 h->ae_algo->ops->enable_hw_strip_rxvtag) {
2322 enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
2323 ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, enable);
2328 if ((changed & NETIF_F_NTUPLE) && h->ae_algo->ops->enable_fd) {
2329 enable = !!(features & NETIF_F_NTUPLE);
2330 h->ae_algo->ops->enable_fd(h, enable);
2333 if ((netdev->features & NETIF_F_HW_TC) > (features & NETIF_F_HW_TC) &&
2334 h->ae_algo->ops->cls_flower_active(h)) {
2336 "there are offloaded TC filters active, cannot disable HW TC offload");
2340 if ((changed & NETIF_F_HW_VLAN_CTAG_FILTER) &&
2341 h->ae_algo->ops->enable_vlan_filter) {
2342 enable = !!(features & NETIF_F_HW_VLAN_CTAG_FILTER);
2343 ret = h->ae_algo->ops->enable_vlan_filter(h, enable);
2348 netdev->features = features;
2352 static netdev_features_t hns3_features_check(struct sk_buff *skb,
2353 struct net_device *dev,
2354 netdev_features_t features)
2356 #define HNS3_MAX_HDR_LEN 480U
2357 #define HNS3_MAX_L4_HDR_LEN 60U
2361 if (skb->ip_summed != CHECKSUM_PARTIAL)
2364 if (skb->encapsulation)
2365 len = skb_inner_transport_header(skb) - skb->data;
2367 len = skb_transport_header(skb) - skb->data;
2369 /* Assume L4 is 60 byte as TCP is the only protocol with a
2370 * a flexible value, and it's max len is 60 bytes.
2372 len += HNS3_MAX_L4_HDR_LEN;
2374 /* Hardware only supports checksum on the skb with a max header
2377 if (len > HNS3_MAX_HDR_LEN)
2378 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2383 static void hns3_nic_get_stats64(struct net_device *netdev,
2384 struct rtnl_link_stats64 *stats)
2386 struct hns3_nic_priv *priv = netdev_priv(netdev);
2387 int queue_num = priv->ae_handle->kinfo.num_tqps;
2388 struct hnae3_handle *handle = priv->ae_handle;
2389 struct hns3_enet_ring *ring;
2390 u64 rx_length_errors = 0;
2391 u64 rx_crc_errors = 0;
2392 u64 rx_multicast = 0;
2404 if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
2407 handle->ae_algo->ops->update_stats(handle, &netdev->stats);
2409 for (idx = 0; idx < queue_num; idx++) {
2410 /* fetch the tx stats */
2411 ring = &priv->ring[idx];
2413 start = u64_stats_fetch_begin_irq(&ring->syncp);
2414 tx_bytes += ring->stats.tx_bytes;
2415 tx_pkts += ring->stats.tx_pkts;
2416 tx_drop += ring->stats.sw_err_cnt;
2417 tx_drop += ring->stats.tx_vlan_err;
2418 tx_drop += ring->stats.tx_l4_proto_err;
2419 tx_drop += ring->stats.tx_l2l3l4_err;
2420 tx_drop += ring->stats.tx_tso_err;
2421 tx_drop += ring->stats.over_max_recursion;
2422 tx_drop += ring->stats.hw_limitation;
2423 tx_drop += ring->stats.copy_bits_err;
2424 tx_drop += ring->stats.skb2sgl_err;
2425 tx_drop += ring->stats.map_sg_err;
2426 tx_errors += ring->stats.sw_err_cnt;
2427 tx_errors += ring->stats.tx_vlan_err;
2428 tx_errors += ring->stats.tx_l4_proto_err;
2429 tx_errors += ring->stats.tx_l2l3l4_err;
2430 tx_errors += ring->stats.tx_tso_err;
2431 tx_errors += ring->stats.over_max_recursion;
2432 tx_errors += ring->stats.hw_limitation;
2433 tx_errors += ring->stats.copy_bits_err;
2434 tx_errors += ring->stats.skb2sgl_err;
2435 tx_errors += ring->stats.map_sg_err;
2436 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
2438 /* fetch the rx stats */
2439 ring = &priv->ring[idx + queue_num];
2441 start = u64_stats_fetch_begin_irq(&ring->syncp);
2442 rx_bytes += ring->stats.rx_bytes;
2443 rx_pkts += ring->stats.rx_pkts;
2444 rx_drop += ring->stats.l2_err;
2445 rx_errors += ring->stats.l2_err;
2446 rx_errors += ring->stats.l3l4_csum_err;
2447 rx_crc_errors += ring->stats.l2_err;
2448 rx_multicast += ring->stats.rx_multicast;
2449 rx_length_errors += ring->stats.err_pkt_len;
2450 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
2453 stats->tx_bytes = tx_bytes;
2454 stats->tx_packets = tx_pkts;
2455 stats->rx_bytes = rx_bytes;
2456 stats->rx_packets = rx_pkts;
2458 stats->rx_errors = rx_errors;
2459 stats->multicast = rx_multicast;
2460 stats->rx_length_errors = rx_length_errors;
2461 stats->rx_crc_errors = rx_crc_errors;
2462 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
2464 stats->tx_errors = tx_errors;
2465 stats->rx_dropped = rx_drop;
2466 stats->tx_dropped = tx_drop;
2467 stats->collisions = netdev->stats.collisions;
2468 stats->rx_over_errors = netdev->stats.rx_over_errors;
2469 stats->rx_frame_errors = netdev->stats.rx_frame_errors;
2470 stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
2471 stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
2472 stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
2473 stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
2474 stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
2475 stats->tx_window_errors = netdev->stats.tx_window_errors;
2476 stats->rx_compressed = netdev->stats.rx_compressed;
2477 stats->tx_compressed = netdev->stats.tx_compressed;
2480 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
2482 struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
2483 struct hnae3_knic_private_info *kinfo;
2484 u8 tc = mqprio_qopt->qopt.num_tc;
2485 u16 mode = mqprio_qopt->mode;
2486 u8 hw = mqprio_qopt->qopt.hw;
2487 struct hnae3_handle *h;
2489 if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
2490 mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
2493 if (tc > HNAE3_MAX_TC)
2499 h = hns3_get_handle(netdev);
2502 netif_dbg(h, drv, netdev, "setup tc: num_tc=%u\n", tc);
2504 return (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
2505 kinfo->dcb_ops->setup_tc(h, mqprio_qopt) : -EOPNOTSUPP;
2508 static int hns3_setup_tc_cls_flower(struct hns3_nic_priv *priv,
2509 struct flow_cls_offload *flow)
2511 int tc = tc_classid_to_hwtc(priv->netdev, flow->classid);
2512 struct hnae3_handle *h = hns3_get_handle(priv->netdev);
2514 switch (flow->command) {
2515 case FLOW_CLS_REPLACE:
2516 if (h->ae_algo->ops->add_cls_flower)
2517 return h->ae_algo->ops->add_cls_flower(h, flow, tc);
2519 case FLOW_CLS_DESTROY:
2520 if (h->ae_algo->ops->del_cls_flower)
2521 return h->ae_algo->ops->del_cls_flower(h, flow);
2530 static int hns3_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
2533 struct hns3_nic_priv *priv = cb_priv;
2535 if (!tc_cls_can_offload_and_chain0(priv->netdev, type_data))
2539 case TC_SETUP_CLSFLOWER:
2540 return hns3_setup_tc_cls_flower(priv, type_data);
2546 static LIST_HEAD(hns3_block_cb_list);
2548 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
2551 struct hns3_nic_priv *priv = netdev_priv(dev);
2555 case TC_SETUP_QDISC_MQPRIO:
2556 ret = hns3_setup_tc(dev, type_data);
2558 case TC_SETUP_BLOCK:
2559 ret = flow_block_cb_setup_simple(type_data,
2560 &hns3_block_cb_list,
2561 hns3_setup_tc_block_cb,
2571 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
2572 __be16 proto, u16 vid)
2574 struct hnae3_handle *h = hns3_get_handle(netdev);
2577 if (h->ae_algo->ops->set_vlan_filter)
2578 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
2583 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
2584 __be16 proto, u16 vid)
2586 struct hnae3_handle *h = hns3_get_handle(netdev);
2589 if (h->ae_algo->ops->set_vlan_filter)
2590 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
2595 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
2596 u8 qos, __be16 vlan_proto)
2598 struct hnae3_handle *h = hns3_get_handle(netdev);
2601 netif_dbg(h, drv, netdev,
2602 "set vf vlan: vf=%d, vlan=%u, qos=%u, vlan_proto=0x%x\n",
2603 vf, vlan, qos, ntohs(vlan_proto));
2605 if (h->ae_algo->ops->set_vf_vlan_filter)
2606 ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
2612 static int hns3_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
2614 struct hnae3_handle *handle = hns3_get_handle(netdev);
2616 if (hns3_nic_resetting(netdev))
2619 if (!handle->ae_algo->ops->set_vf_spoofchk)
2622 return handle->ae_algo->ops->set_vf_spoofchk(handle, vf, enable);
2625 static int hns3_set_vf_trust(struct net_device *netdev, int vf, bool enable)
2627 struct hnae3_handle *handle = hns3_get_handle(netdev);
2629 if (!handle->ae_algo->ops->set_vf_trust)
2632 return handle->ae_algo->ops->set_vf_trust(handle, vf, enable);
2635 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
2637 struct hnae3_handle *h = hns3_get_handle(netdev);
2640 if (hns3_nic_resetting(netdev))
2643 if (!h->ae_algo->ops->set_mtu)
2646 netif_dbg(h, drv, netdev,
2647 "change mtu from %u to %d\n", netdev->mtu, new_mtu);
2649 ret = h->ae_algo->ops->set_mtu(h, new_mtu);
2651 netdev_err(netdev, "failed to change MTU in hardware %d\n",
2654 netdev->mtu = new_mtu;
2659 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
2661 struct hns3_nic_priv *priv = netdev_priv(ndev);
2662 struct hnae3_handle *h = hns3_get_handle(ndev);
2663 struct hns3_enet_ring *tx_ring;
2664 struct napi_struct *napi;
2665 int timeout_queue = 0;
2666 int hw_head, hw_tail;
2667 int fbd_num, fbd_oft;
2668 int ebd_num, ebd_oft;
2673 /* Find the stopped queue the same way the stack does */
2674 for (i = 0; i < ndev->num_tx_queues; i++) {
2675 struct netdev_queue *q;
2676 unsigned long trans_start;
2678 q = netdev_get_tx_queue(ndev, i);
2679 trans_start = q->trans_start;
2680 if (netif_xmit_stopped(q) &&
2682 (trans_start + ndev->watchdog_timeo))) {
2684 netdev_info(ndev, "queue state: 0x%lx, delta msecs: %u\n",
2686 jiffies_to_msecs(jiffies - trans_start));
2691 if (i == ndev->num_tx_queues) {
2693 "no netdev TX timeout queue found, timeout count: %llu\n",
2694 priv->tx_timeout_count);
2698 priv->tx_timeout_count++;
2700 tx_ring = &priv->ring[timeout_queue];
2701 napi = &tx_ring->tqp_vector->napi;
2704 "tx_timeout count: %llu, queue id: %d, SW_NTU: 0x%x, SW_NTC: 0x%x, napi state: %lu\n",
2705 priv->tx_timeout_count, timeout_queue, tx_ring->next_to_use,
2706 tx_ring->next_to_clean, napi->state);
2709 "tx_pkts: %llu, tx_bytes: %llu, sw_err_cnt: %llu, tx_pending: %d\n",
2710 tx_ring->stats.tx_pkts, tx_ring->stats.tx_bytes,
2711 tx_ring->stats.sw_err_cnt, tx_ring->pending_buf);
2714 "seg_pkt_cnt: %llu, tx_more: %llu, restart_queue: %llu, tx_busy: %llu\n",
2715 tx_ring->stats.seg_pkt_cnt, tx_ring->stats.tx_more,
2716 tx_ring->stats.restart_queue, tx_ring->stats.tx_busy);
2718 /* When mac received many pause frames continuous, it's unable to send
2719 * packets, which may cause tx timeout
2721 if (h->ae_algo->ops->get_mac_stats) {
2722 struct hns3_mac_stats mac_stats;
2724 h->ae_algo->ops->get_mac_stats(h, &mac_stats);
2725 netdev_info(ndev, "tx_pause_cnt: %llu, rx_pause_cnt: %llu\n",
2726 mac_stats.tx_pause_cnt, mac_stats.rx_pause_cnt);
2729 hw_head = readl_relaxed(tx_ring->tqp->io_base +
2730 HNS3_RING_TX_RING_HEAD_REG);
2731 hw_tail = readl_relaxed(tx_ring->tqp->io_base +
2732 HNS3_RING_TX_RING_TAIL_REG);
2733 fbd_num = readl_relaxed(tx_ring->tqp->io_base +
2734 HNS3_RING_TX_RING_FBDNUM_REG);
2735 fbd_oft = readl_relaxed(tx_ring->tqp->io_base +
2736 HNS3_RING_TX_RING_OFFSET_REG);
2737 ebd_num = readl_relaxed(tx_ring->tqp->io_base +
2738 HNS3_RING_TX_RING_EBDNUM_REG);
2739 ebd_oft = readl_relaxed(tx_ring->tqp->io_base +
2740 HNS3_RING_TX_RING_EBD_OFFSET_REG);
2741 bd_num = readl_relaxed(tx_ring->tqp->io_base +
2742 HNS3_RING_TX_RING_BD_NUM_REG);
2743 bd_err = readl_relaxed(tx_ring->tqp->io_base +
2744 HNS3_RING_TX_RING_BD_ERR_REG);
2745 ring_en = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_EN_REG);
2746 tc = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_TX_RING_TC_REG);
2749 "BD_NUM: 0x%x HW_HEAD: 0x%x, HW_TAIL: 0x%x, BD_ERR: 0x%x, INT: 0x%x\n",
2750 bd_num, hw_head, hw_tail, bd_err,
2751 readl(tx_ring->tqp_vector->mask_addr));
2753 "RING_EN: 0x%x, TC: 0x%x, FBD_NUM: 0x%x FBD_OFT: 0x%x, EBD_NUM: 0x%x, EBD_OFT: 0x%x\n",
2754 ring_en, tc, fbd_num, fbd_oft, ebd_num, ebd_oft);
2759 static void hns3_nic_net_timeout(struct net_device *ndev, unsigned int txqueue)
2761 struct hns3_nic_priv *priv = netdev_priv(ndev);
2762 struct hnae3_handle *h = priv->ae_handle;
2764 if (!hns3_get_tx_timeo_queue_info(ndev))
2767 /* request the reset, and let the hclge to determine
2768 * which reset level should be done
2770 if (h->ae_algo->ops->reset_event)
2771 h->ae_algo->ops->reset_event(h->pdev, h);
2774 #ifdef CONFIG_RFS_ACCEL
2775 static int hns3_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
2776 u16 rxq_index, u32 flow_id)
2778 struct hnae3_handle *h = hns3_get_handle(dev);
2779 struct flow_keys fkeys;
2781 if (!h->ae_algo->ops->add_arfs_entry)
2784 if (skb->encapsulation)
2785 return -EPROTONOSUPPORT;
2787 if (!skb_flow_dissect_flow_keys(skb, &fkeys, 0))
2788 return -EPROTONOSUPPORT;
2790 if ((fkeys.basic.n_proto != htons(ETH_P_IP) &&
2791 fkeys.basic.n_proto != htons(ETH_P_IPV6)) ||
2792 (fkeys.basic.ip_proto != IPPROTO_TCP &&
2793 fkeys.basic.ip_proto != IPPROTO_UDP))
2794 return -EPROTONOSUPPORT;
2796 return h->ae_algo->ops->add_arfs_entry(h, rxq_index, flow_id, &fkeys);
2800 static int hns3_nic_get_vf_config(struct net_device *ndev, int vf,
2801 struct ifla_vf_info *ivf)
2803 struct hnae3_handle *h = hns3_get_handle(ndev);
2805 if (!h->ae_algo->ops->get_vf_config)
2808 return h->ae_algo->ops->get_vf_config(h, vf, ivf);
2811 static int hns3_nic_set_vf_link_state(struct net_device *ndev, int vf,
2814 struct hnae3_handle *h = hns3_get_handle(ndev);
2816 if (!h->ae_algo->ops->set_vf_link_state)
2819 return h->ae_algo->ops->set_vf_link_state(h, vf, link_state);
2822 static int hns3_nic_set_vf_rate(struct net_device *ndev, int vf,
2823 int min_tx_rate, int max_tx_rate)
2825 struct hnae3_handle *h = hns3_get_handle(ndev);
2827 if (!h->ae_algo->ops->set_vf_rate)
2830 return h->ae_algo->ops->set_vf_rate(h, vf, min_tx_rate, max_tx_rate,
2834 static int hns3_nic_set_vf_mac(struct net_device *netdev, int vf_id, u8 *mac)
2836 struct hnae3_handle *h = hns3_get_handle(netdev);
2838 if (!h->ae_algo->ops->set_vf_mac)
2841 if (is_multicast_ether_addr(mac)) {
2843 "Invalid MAC:%pM specified. Could not set MAC\n",
2848 return h->ae_algo->ops->set_vf_mac(h, vf_id, mac);
2851 static const struct net_device_ops hns3_nic_netdev_ops = {
2852 .ndo_open = hns3_nic_net_open,
2853 .ndo_stop = hns3_nic_net_stop,
2854 .ndo_start_xmit = hns3_nic_net_xmit,
2855 .ndo_tx_timeout = hns3_nic_net_timeout,
2856 .ndo_set_mac_address = hns3_nic_net_set_mac_address,
2857 .ndo_eth_ioctl = hns3_nic_do_ioctl,
2858 .ndo_change_mtu = hns3_nic_change_mtu,
2859 .ndo_set_features = hns3_nic_set_features,
2860 .ndo_features_check = hns3_features_check,
2861 .ndo_get_stats64 = hns3_nic_get_stats64,
2862 .ndo_setup_tc = hns3_nic_setup_tc,
2863 .ndo_set_rx_mode = hns3_nic_set_rx_mode,
2864 .ndo_vlan_rx_add_vid = hns3_vlan_rx_add_vid,
2865 .ndo_vlan_rx_kill_vid = hns3_vlan_rx_kill_vid,
2866 .ndo_set_vf_vlan = hns3_ndo_set_vf_vlan,
2867 .ndo_set_vf_spoofchk = hns3_set_vf_spoofchk,
2868 .ndo_set_vf_trust = hns3_set_vf_trust,
2869 #ifdef CONFIG_RFS_ACCEL
2870 .ndo_rx_flow_steer = hns3_rx_flow_steer,
2872 .ndo_get_vf_config = hns3_nic_get_vf_config,
2873 .ndo_set_vf_link_state = hns3_nic_set_vf_link_state,
2874 .ndo_set_vf_rate = hns3_nic_set_vf_rate,
2875 .ndo_set_vf_mac = hns3_nic_set_vf_mac,
2878 bool hns3_is_phys_func(struct pci_dev *pdev)
2880 u32 dev_id = pdev->device;
2883 case HNAE3_DEV_ID_GE:
2884 case HNAE3_DEV_ID_25GE:
2885 case HNAE3_DEV_ID_25GE_RDMA:
2886 case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
2887 case HNAE3_DEV_ID_50GE_RDMA:
2888 case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
2889 case HNAE3_DEV_ID_100G_RDMA_MACSEC:
2890 case HNAE3_DEV_ID_200G_RDMA:
2892 case HNAE3_DEV_ID_VF:
2893 case HNAE3_DEV_ID_RDMA_DCB_PFC_VF:
2896 dev_warn(&pdev->dev, "un-recognized pci device-id %u",
2903 static void hns3_disable_sriov(struct pci_dev *pdev)
2905 /* If our VFs are assigned we cannot shut down SR-IOV
2906 * without causing issues, so just leave the hardware
2907 * available but disabled
2909 if (pci_vfs_assigned(pdev)) {
2910 dev_warn(&pdev->dev,
2911 "disabling driver while VFs are assigned\n");
2915 pci_disable_sriov(pdev);
2918 /* hns3_probe - Device initialization routine
2919 * @pdev: PCI device information struct
2920 * @ent: entry in hns3_pci_tbl
2922 * hns3_probe initializes a PF identified by a pci_dev structure.
2923 * The OS initialization, configuring of the PF private structure,
2924 * and a hardware reset occur.
2926 * Returns 0 on success, negative on failure
2928 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2930 struct hnae3_ae_dev *ae_dev;
2933 ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev), GFP_KERNEL);
2937 ae_dev->pdev = pdev;
2938 ae_dev->flag = ent->driver_data;
2939 pci_set_drvdata(pdev, ae_dev);
2941 ret = hnae3_register_ae_dev(ae_dev);
2943 pci_set_drvdata(pdev, NULL);
2948 /* hns3_remove - Device removal routine
2949 * @pdev: PCI device information struct
2951 static void hns3_remove(struct pci_dev *pdev)
2953 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2955 if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
2956 hns3_disable_sriov(pdev);
2958 hnae3_unregister_ae_dev(ae_dev);
2959 pci_set_drvdata(pdev, NULL);
2963 * hns3_pci_sriov_configure
2964 * @pdev: pointer to a pci_dev structure
2965 * @num_vfs: number of VFs to allocate
2967 * Enable or change the number of VFs. Called when the user updates the number
2970 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
2974 if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
2975 dev_warn(&pdev->dev, "Can not config SRIOV\n");
2980 ret = pci_enable_sriov(pdev, num_vfs);
2982 dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
2985 } else if (!pci_vfs_assigned(pdev)) {
2986 pci_disable_sriov(pdev);
2988 dev_warn(&pdev->dev,
2989 "Unable to free VFs because some are assigned to VMs.\n");
2995 static void hns3_shutdown(struct pci_dev *pdev)
2997 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2999 hnae3_unregister_ae_dev(ae_dev);
3000 pci_set_drvdata(pdev, NULL);
3002 if (system_state == SYSTEM_POWER_OFF)
3003 pci_set_power_state(pdev, PCI_D3hot);
3006 static int __maybe_unused hns3_suspend(struct device *dev)
3008 struct hnae3_ae_dev *ae_dev = dev_get_drvdata(dev);
3010 if (ae_dev && hns3_is_phys_func(ae_dev->pdev)) {
3011 dev_info(dev, "Begin to suspend.\n");
3012 if (ae_dev->ops && ae_dev->ops->reset_prepare)
3013 ae_dev->ops->reset_prepare(ae_dev, HNAE3_FUNC_RESET);
3019 static int __maybe_unused hns3_resume(struct device *dev)
3021 struct hnae3_ae_dev *ae_dev = dev_get_drvdata(dev);
3023 if (ae_dev && hns3_is_phys_func(ae_dev->pdev)) {
3024 dev_info(dev, "Begin to resume.\n");
3025 if (ae_dev->ops && ae_dev->ops->reset_done)
3026 ae_dev->ops->reset_done(ae_dev);
3032 static pci_ers_result_t hns3_error_detected(struct pci_dev *pdev,
3033 pci_channel_state_t state)
3035 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3036 pci_ers_result_t ret;
3038 dev_info(&pdev->dev, "PCI error detected, state(=%u)!!\n", state);
3040 if (state == pci_channel_io_perm_failure)
3041 return PCI_ERS_RESULT_DISCONNECT;
3043 if (!ae_dev || !ae_dev->ops) {
3045 "Can't recover - error happened before device initialized\n");
3046 return PCI_ERS_RESULT_NONE;
3049 if (ae_dev->ops->handle_hw_ras_error)
3050 ret = ae_dev->ops->handle_hw_ras_error(ae_dev);
3052 return PCI_ERS_RESULT_NONE;
3057 static pci_ers_result_t hns3_slot_reset(struct pci_dev *pdev)
3059 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3060 const struct hnae3_ae_ops *ops;
3061 enum hnae3_reset_type reset_type;
3062 struct device *dev = &pdev->dev;
3064 if (!ae_dev || !ae_dev->ops)
3065 return PCI_ERS_RESULT_NONE;
3068 /* request the reset */
3069 if (ops->reset_event && ops->get_reset_level &&
3070 ops->set_default_reset_request) {
3071 if (ae_dev->hw_err_reset_req) {
3072 reset_type = ops->get_reset_level(ae_dev,
3073 &ae_dev->hw_err_reset_req);
3074 ops->set_default_reset_request(ae_dev, reset_type);
3075 dev_info(dev, "requesting reset due to PCI error\n");
3076 ops->reset_event(pdev, NULL);
3079 return PCI_ERS_RESULT_RECOVERED;
3082 return PCI_ERS_RESULT_DISCONNECT;
3085 static void hns3_reset_prepare(struct pci_dev *pdev)
3087 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3089 dev_info(&pdev->dev, "FLR prepare\n");
3090 if (ae_dev && ae_dev->ops && ae_dev->ops->reset_prepare)
3091 ae_dev->ops->reset_prepare(ae_dev, HNAE3_FLR_RESET);
3094 static void hns3_reset_done(struct pci_dev *pdev)
3096 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3098 dev_info(&pdev->dev, "FLR done\n");
3099 if (ae_dev && ae_dev->ops && ae_dev->ops->reset_done)
3100 ae_dev->ops->reset_done(ae_dev);
3103 static const struct pci_error_handlers hns3_err_handler = {
3104 .error_detected = hns3_error_detected,
3105 .slot_reset = hns3_slot_reset,
3106 .reset_prepare = hns3_reset_prepare,
3107 .reset_done = hns3_reset_done,
3110 static SIMPLE_DEV_PM_OPS(hns3_pm_ops, hns3_suspend, hns3_resume);
3112 static struct pci_driver hns3_driver = {
3113 .name = hns3_driver_name,
3114 .id_table = hns3_pci_tbl,
3115 .probe = hns3_probe,
3116 .remove = hns3_remove,
3117 .shutdown = hns3_shutdown,
3118 .driver.pm = &hns3_pm_ops,
3119 .sriov_configure = hns3_pci_sriov_configure,
3120 .err_handler = &hns3_err_handler,
3123 /* set default feature to hns3 */
3124 static void hns3_set_default_feature(struct net_device *netdev)
3126 struct hnae3_handle *h = hns3_get_handle(netdev);
3127 struct pci_dev *pdev = h->pdev;
3128 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3130 netdev->priv_flags |= IFF_UNICAST_FLT;
3132 netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
3134 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
3135 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3136 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
3137 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
3138 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
3139 NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
3141 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V2) {
3142 netdev->features |= NETIF_F_GRO_HW;
3144 if (!(h->flags & HNAE3_SUPPORT_VF))
3145 netdev->features |= NETIF_F_NTUPLE;
3148 if (test_bit(HNAE3_DEV_SUPPORT_UDP_GSO_B, ae_dev->caps))
3149 netdev->features |= NETIF_F_GSO_UDP_L4;
3151 if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps))
3152 netdev->features |= NETIF_F_HW_CSUM;
3154 netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3156 if (test_bit(HNAE3_DEV_SUPPORT_UDP_TUNNEL_CSUM_B, ae_dev->caps))
3157 netdev->features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
3159 if (test_bit(HNAE3_DEV_SUPPORT_FD_FORWARD_TC_B, ae_dev->caps))
3160 netdev->features |= NETIF_F_HW_TC;
3162 netdev->hw_features |= netdev->features;
3163 if (!test_bit(HNAE3_DEV_SUPPORT_VLAN_FLTR_MDF_B, ae_dev->caps))
3164 netdev->hw_features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
3166 netdev->vlan_features |= netdev->features &
3167 ~(NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_CTAG_TX |
3168 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_GRO_HW | NETIF_F_NTUPLE |
3171 netdev->hw_enc_features |= netdev->vlan_features | NETIF_F_TSO_MANGLEID;
3174 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
3175 struct hns3_desc_cb *cb)
3177 unsigned int order = hns3_page_order(ring);
3180 if (ring->page_pool) {
3181 p = page_pool_dev_alloc_frag(ring->page_pool,
3183 hns3_buf_size(ring));
3188 cb->buf = page_address(p);
3189 cb->dma = page_pool_get_dma_addr(p);
3190 cb->type = DESC_TYPE_PP_FRAG;
3195 p = dev_alloc_pages(order);
3200 cb->page_offset = 0;
3202 cb->buf = page_address(p);
3203 cb->length = hns3_page_size(ring);
3204 cb->type = DESC_TYPE_PAGE;
3205 page_ref_add(p, USHRT_MAX - 1);
3206 cb->pagecnt_bias = USHRT_MAX;
3211 static void hns3_free_buffer(struct hns3_enet_ring *ring,
3212 struct hns3_desc_cb *cb, int budget)
3214 if (cb->type & (DESC_TYPE_SKB | DESC_TYPE_BOUNCE_HEAD |
3215 DESC_TYPE_BOUNCE_ALL | DESC_TYPE_SGL_SKB))
3216 napi_consume_skb(cb->priv, budget);
3217 else if (!HNAE3_IS_TX_RING(ring)) {
3218 if (cb->type & DESC_TYPE_PAGE && cb->pagecnt_bias)
3219 __page_frag_cache_drain(cb->priv, cb->pagecnt_bias);
3220 else if (cb->type & DESC_TYPE_PP_FRAG)
3221 page_pool_put_full_page(ring->page_pool, cb->priv,
3224 memset(cb, 0, sizeof(*cb));
3227 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
3229 cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
3230 cb->length, ring_to_dma_dir(ring));
3232 if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
3238 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
3239 struct hns3_desc_cb *cb)
3241 if (cb->type & (DESC_TYPE_SKB | DESC_TYPE_FRAGLIST_SKB))
3242 dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
3243 ring_to_dma_dir(ring));
3244 else if ((cb->type & DESC_TYPE_PAGE) && cb->length)
3245 dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
3246 ring_to_dma_dir(ring));
3247 else if (cb->type & (DESC_TYPE_BOUNCE_ALL | DESC_TYPE_BOUNCE_HEAD |
3249 hns3_tx_spare_reclaim_cb(ring, cb);
3252 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
3254 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
3255 ring->desc[i].addr = 0;
3258 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i,
3261 struct hns3_desc_cb *cb = &ring->desc_cb[i];
3263 if (!ring->desc_cb[i].dma)
3266 hns3_buffer_detach(ring, i);
3267 hns3_free_buffer(ring, cb, budget);
3270 static void hns3_free_buffers(struct hns3_enet_ring *ring)
3274 for (i = 0; i < ring->desc_num; i++)
3275 hns3_free_buffer_detach(ring, i, 0);
3278 /* free desc along with its attached buffer */
3279 static void hns3_free_desc(struct hns3_enet_ring *ring)
3281 int size = ring->desc_num * sizeof(ring->desc[0]);
3283 hns3_free_buffers(ring);
3286 dma_free_coherent(ring_to_dev(ring), size,
3287 ring->desc, ring->desc_dma_addr);
3292 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
3294 int size = ring->desc_num * sizeof(ring->desc[0]);
3296 ring->desc = dma_alloc_coherent(ring_to_dev(ring), size,
3297 &ring->desc_dma_addr, GFP_KERNEL);
3304 static int hns3_alloc_and_map_buffer(struct hns3_enet_ring *ring,
3305 struct hns3_desc_cb *cb)
3309 ret = hns3_alloc_buffer(ring, cb);
3310 if (ret || ring->page_pool)
3313 ret = hns3_map_buffer(ring, cb);
3320 hns3_free_buffer(ring, cb, 0);
3325 static int hns3_alloc_and_attach_buffer(struct hns3_enet_ring *ring, int i)
3327 int ret = hns3_alloc_and_map_buffer(ring, &ring->desc_cb[i]);
3332 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3333 ring->desc_cb[i].page_offset);
3338 /* Allocate memory for raw pkg, and map with dma */
3339 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
3343 for (i = 0; i < ring->desc_num; i++) {
3344 ret = hns3_alloc_and_attach_buffer(ring, i);
3346 goto out_buffer_fail;
3352 for (j = i - 1; j >= 0; j--)
3353 hns3_free_buffer_detach(ring, j, 0);
3357 /* detach a in-used buffer and replace with a reserved one */
3358 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
3359 struct hns3_desc_cb *res_cb)
3361 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
3362 ring->desc_cb[i] = *res_cb;
3363 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3364 ring->desc_cb[i].page_offset);
3365 ring->desc[i].rx.bd_base_info = 0;
3368 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
3370 ring->desc_cb[i].reuse_flag = 0;
3371 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3372 ring->desc_cb[i].page_offset);
3373 ring->desc[i].rx.bd_base_info = 0;
3375 dma_sync_single_for_device(ring_to_dev(ring),
3376 ring->desc_cb[i].dma + ring->desc_cb[i].page_offset,
3377 hns3_buf_size(ring),
3381 static bool hns3_nic_reclaim_desc(struct hns3_enet_ring *ring,
3382 int *bytes, int *pkts, int budget)
3384 /* pair with ring->last_to_use update in hns3_tx_doorbell(),
3385 * smp_store_release() is not used in hns3_tx_doorbell() because
3386 * the doorbell operation already have the needed barrier operation.
3388 int ltu = smp_load_acquire(&ring->last_to_use);
3389 int ntc = ring->next_to_clean;
3390 struct hns3_desc_cb *desc_cb;
3391 bool reclaimed = false;
3392 struct hns3_desc *desc;
3394 while (ltu != ntc) {
3395 desc = &ring->desc[ntc];
3397 if (le16_to_cpu(desc->tx.bdtp_fe_sc_vld_ra_ri) &
3398 BIT(HNS3_TXD_VLD_B))
3401 desc_cb = &ring->desc_cb[ntc];
3403 if (desc_cb->type & (DESC_TYPE_SKB | DESC_TYPE_BOUNCE_ALL |
3404 DESC_TYPE_BOUNCE_HEAD |
3405 DESC_TYPE_SGL_SKB)) {
3407 (*bytes) += desc_cb->send_bytes;
3410 /* desc_cb will be cleaned, after hnae3_free_buffer_detach */
3411 hns3_free_buffer_detach(ring, ntc, budget);
3413 if (++ntc == ring->desc_num)
3416 /* Issue prefetch for next Tx descriptor */
3417 prefetch(&ring->desc_cb[ntc]);
3421 if (unlikely(!reclaimed))
3424 /* This smp_store_release() pairs with smp_load_acquire() in
3425 * ring_space called by hns3_nic_net_xmit.
3427 smp_store_release(&ring->next_to_clean, ntc);
3429 hns3_tx_spare_update(ring);
3434 void hns3_clean_tx_ring(struct hns3_enet_ring *ring, int budget)
3436 struct net_device *netdev = ring_to_netdev(ring);
3437 struct hns3_nic_priv *priv = netdev_priv(netdev);
3438 struct netdev_queue *dev_queue;
3444 if (unlikely(!hns3_nic_reclaim_desc(ring, &bytes, &pkts, budget)))
3447 ring->tqp_vector->tx_group.total_bytes += bytes;
3448 ring->tqp_vector->tx_group.total_packets += pkts;
3450 u64_stats_update_begin(&ring->syncp);
3451 ring->stats.tx_bytes += bytes;
3452 ring->stats.tx_pkts += pkts;
3453 u64_stats_update_end(&ring->syncp);
3455 dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
3456 netdev_tx_completed_queue(dev_queue, pkts, bytes);
3458 if (unlikely(netif_carrier_ok(netdev) &&
3459 ring_space(ring) > HNS3_MAX_TSO_BD_NUM)) {
3460 /* Make sure that anybody stopping the queue after this
3461 * sees the new next_to_clean.
3464 if (netif_tx_queue_stopped(dev_queue) &&
3465 !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
3466 netif_tx_wake_queue(dev_queue);
3467 ring->stats.restart_queue++;
3472 static int hns3_desc_unused(struct hns3_enet_ring *ring)
3474 int ntc = ring->next_to_clean;
3475 int ntu = ring->next_to_use;
3477 return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
3480 static void hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring,
3483 struct hns3_desc_cb *desc_cb;
3484 struct hns3_desc_cb res_cbs;
3487 for (i = 0; i < cleand_count; i++) {
3488 desc_cb = &ring->desc_cb[ring->next_to_use];
3489 if (desc_cb->reuse_flag) {
3490 u64_stats_update_begin(&ring->syncp);
3491 ring->stats.reuse_pg_cnt++;
3492 u64_stats_update_end(&ring->syncp);
3494 hns3_reuse_buffer(ring, ring->next_to_use);
3496 ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
3498 u64_stats_update_begin(&ring->syncp);
3499 ring->stats.sw_err_cnt++;
3500 u64_stats_update_end(&ring->syncp);
3502 hns3_rl_err(ring_to_netdev(ring),
3503 "alloc rx buffer failed: %d\n",
3507 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
3509 u64_stats_update_begin(&ring->syncp);
3510 ring->stats.non_reuse_pg++;
3511 u64_stats_update_end(&ring->syncp);
3514 ring_ptr_move_fw(ring, next_to_use);
3517 writel(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
3520 static bool hns3_can_reuse_page(struct hns3_desc_cb *cb)
3522 return page_count(cb->priv) == cb->pagecnt_bias;
3525 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
3526 struct hns3_enet_ring *ring, int pull_len,
3527 struct hns3_desc_cb *desc_cb)
3529 struct hns3_desc *desc = &ring->desc[ring->next_to_clean];
3530 u32 frag_offset = desc_cb->page_offset + pull_len;
3531 int size = le16_to_cpu(desc->rx.size);
3532 u32 truesize = hns3_buf_size(ring);
3533 u32 frag_size = size - pull_len;
3536 if (ring->page_pool) {
3537 skb_add_rx_frag(skb, i, desc_cb->priv, frag_offset,
3538 frag_size, truesize);
3542 /* Avoid re-using remote or pfmem page */
3543 if (unlikely(!dev_page_is_reusable(desc_cb->priv)))
3546 reused = hns3_can_reuse_page(desc_cb);
3548 /* Rx page can be reused when:
3549 * 1. Rx page is only owned by the driver when page_offset
3550 * is zero, which means 0 @ truesize will be used by
3551 * stack after skb_add_rx_frag() is called, and the rest
3552 * of rx page can be reused by driver.
3554 * 2. Rx page is only owned by the driver when page_offset
3555 * is non-zero, which means page_offset @ truesize will
3556 * be used by stack after skb_add_rx_frag() is called,
3557 * and 0 @ truesize can be reused by driver.
3559 if ((!desc_cb->page_offset && reused) ||
3560 ((desc_cb->page_offset + truesize + truesize) <=
3561 hns3_page_size(ring) && desc_cb->page_offset)) {
3562 desc_cb->page_offset += truesize;
3563 desc_cb->reuse_flag = 1;
3564 } else if (desc_cb->page_offset && reused) {
3565 desc_cb->page_offset = 0;
3566 desc_cb->reuse_flag = 1;
3567 } else if (frag_size <= ring->rx_copybreak) {
3568 void *frag = napi_alloc_frag(frag_size);
3570 if (unlikely(!frag)) {
3571 u64_stats_update_begin(&ring->syncp);
3572 ring->stats.frag_alloc_err++;
3573 u64_stats_update_end(&ring->syncp);
3575 hns3_rl_err(ring_to_netdev(ring),
3576 "failed to allocate rx frag\n");
3580 desc_cb->reuse_flag = 1;
3581 memcpy(frag, desc_cb->buf + frag_offset, frag_size);
3582 skb_add_rx_frag(skb, i, virt_to_page(frag),
3583 offset_in_page(frag), frag_size, frag_size);
3585 u64_stats_update_begin(&ring->syncp);
3586 ring->stats.frag_alloc++;
3587 u64_stats_update_end(&ring->syncp);
3592 desc_cb->pagecnt_bias--;
3594 if (unlikely(!desc_cb->pagecnt_bias)) {
3595 page_ref_add(desc_cb->priv, USHRT_MAX);
3596 desc_cb->pagecnt_bias = USHRT_MAX;
3599 skb_add_rx_frag(skb, i, desc_cb->priv, frag_offset,
3600 frag_size, truesize);
3602 if (unlikely(!desc_cb->reuse_flag))
3603 __page_frag_cache_drain(desc_cb->priv, desc_cb->pagecnt_bias);
3606 static int hns3_gro_complete(struct sk_buff *skb, u32 l234info)
3608 __be16 type = skb->protocol;
3612 while (eth_type_vlan(type)) {
3613 struct vlan_hdr *vh;
3615 if ((depth + VLAN_HLEN) > skb_headlen(skb))
3618 vh = (struct vlan_hdr *)(skb->data + depth);
3619 type = vh->h_vlan_encapsulated_proto;
3623 skb_set_network_header(skb, depth);
3625 if (type == htons(ETH_P_IP)) {
3626 const struct iphdr *iph = ip_hdr(skb);
3628 depth += sizeof(struct iphdr);
3629 skb_set_transport_header(skb, depth);
3631 th->check = ~tcp_v4_check(skb->len - depth, iph->saddr,
3633 } else if (type == htons(ETH_P_IPV6)) {
3634 const struct ipv6hdr *iph = ipv6_hdr(skb);
3636 depth += sizeof(struct ipv6hdr);
3637 skb_set_transport_header(skb, depth);
3639 th->check = ~tcp_v6_check(skb->len - depth, &iph->saddr,
3642 hns3_rl_err(skb->dev,
3643 "Error: FW GRO supports only IPv4/IPv6, not 0x%04x, depth: %d\n",
3644 be16_to_cpu(type), depth);
3648 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
3650 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
3652 if (l234info & BIT(HNS3_RXD_GRO_FIXID_B))
3653 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_FIXEDID;
3655 skb->csum_start = (unsigned char *)th - skb->head;
3656 skb->csum_offset = offsetof(struct tcphdr, check);
3657 skb->ip_summed = CHECKSUM_PARTIAL;
3659 trace_hns3_gro(skb);
3664 static bool hns3_checksum_complete(struct hns3_enet_ring *ring,
3665 struct sk_buff *skb, u32 ptype, u16 csum)
3667 if (ptype == HNS3_INVALID_PTYPE ||
3668 hns3_rx_ptype_tbl[ptype].ip_summed != CHECKSUM_COMPLETE)
3671 u64_stats_update_begin(&ring->syncp);
3672 ring->stats.csum_complete++;
3673 u64_stats_update_end(&ring->syncp);
3674 skb->ip_summed = CHECKSUM_COMPLETE;
3675 skb->csum = csum_unfold((__force __sum16)csum);
3680 static void hns3_rx_handle_csum(struct sk_buff *skb, u32 l234info,
3681 u32 ol_info, u32 ptype)
3683 int l3_type, l4_type;
3686 if (ptype != HNS3_INVALID_PTYPE) {
3687 skb->csum_level = hns3_rx_ptype_tbl[ptype].csum_level;
3688 skb->ip_summed = hns3_rx_ptype_tbl[ptype].ip_summed;
3693 ol4_type = hnae3_get_field(ol_info, HNS3_RXD_OL4ID_M,
3696 case HNS3_OL4_TYPE_MAC_IN_UDP:
3697 case HNS3_OL4_TYPE_NVGRE:
3698 skb->csum_level = 1;
3700 case HNS3_OL4_TYPE_NO_TUN:
3701 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
3703 l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
3705 /* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
3706 if ((l3_type == HNS3_L3_TYPE_IPV4 ||
3707 l3_type == HNS3_L3_TYPE_IPV6) &&
3708 (l4_type == HNS3_L4_TYPE_UDP ||
3709 l4_type == HNS3_L4_TYPE_TCP ||
3710 l4_type == HNS3_L4_TYPE_SCTP))
3711 skb->ip_summed = CHECKSUM_UNNECESSARY;
3718 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
3719 u32 l234info, u32 bd_base_info, u32 ol_info,
3722 struct net_device *netdev = ring_to_netdev(ring);
3723 struct hns3_nic_priv *priv = netdev_priv(netdev);
3724 u32 ptype = HNS3_INVALID_PTYPE;
3726 skb->ip_summed = CHECKSUM_NONE;
3728 skb_checksum_none_assert(skb);
3730 if (!(netdev->features & NETIF_F_RXCSUM))
3733 if (test_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state))
3734 ptype = hnae3_get_field(ol_info, HNS3_RXD_PTYPE_M,
3737 if (hns3_checksum_complete(ring, skb, ptype, csum))
3740 /* check if hardware has done checksum */
3741 if (!(bd_base_info & BIT(HNS3_RXD_L3L4P_B)))
3744 if (unlikely(l234info & (BIT(HNS3_RXD_L3E_B) | BIT(HNS3_RXD_L4E_B) |
3745 BIT(HNS3_RXD_OL3E_B) |
3746 BIT(HNS3_RXD_OL4E_B)))) {
3747 u64_stats_update_begin(&ring->syncp);
3748 ring->stats.l3l4_csum_err++;
3749 u64_stats_update_end(&ring->syncp);
3754 hns3_rx_handle_csum(skb, l234info, ol_info, ptype);
3757 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
3759 if (skb_has_frag_list(skb))
3760 napi_gro_flush(&ring->tqp_vector->napi, false);
3762 napi_gro_receive(&ring->tqp_vector->napi, skb);
3765 static bool hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
3766 struct hns3_desc *desc, u32 l234info,
3769 struct hnae3_handle *handle = ring->tqp->handle;
3770 struct pci_dev *pdev = ring->tqp->handle->pdev;
3771 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3773 if (unlikely(ae_dev->dev_version < HNAE3_DEVICE_VERSION_V2)) {
3774 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3775 if (!(*vlan_tag & VLAN_VID_MASK))
3776 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3778 return (*vlan_tag != 0);
3781 #define HNS3_STRP_OUTER_VLAN 0x1
3782 #define HNS3_STRP_INNER_VLAN 0x2
3783 #define HNS3_STRP_BOTH 0x3
3785 /* Hardware always insert VLAN tag into RX descriptor when
3786 * remove the tag from packet, driver needs to determine
3787 * reporting which tag to stack.
3789 switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
3790 HNS3_RXD_STRP_TAGP_S)) {
3791 case HNS3_STRP_OUTER_VLAN:
3792 if (handle->port_base_vlan_state !=
3793 HNAE3_PORT_BASE_VLAN_DISABLE)
3796 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3798 case HNS3_STRP_INNER_VLAN:
3799 if (handle->port_base_vlan_state !=
3800 HNAE3_PORT_BASE_VLAN_DISABLE)
3803 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3805 case HNS3_STRP_BOTH:
3806 if (handle->port_base_vlan_state ==
3807 HNAE3_PORT_BASE_VLAN_DISABLE)
3808 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3810 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3818 static void hns3_rx_ring_move_fw(struct hns3_enet_ring *ring)
3820 ring->desc[ring->next_to_clean].rx.bd_base_info &=
3821 cpu_to_le32(~BIT(HNS3_RXD_VLD_B));
3822 ring->next_to_clean += 1;
3824 if (unlikely(ring->next_to_clean == ring->desc_num))
3825 ring->next_to_clean = 0;
3828 static int hns3_alloc_skb(struct hns3_enet_ring *ring, unsigned int length,
3831 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
3832 struct net_device *netdev = ring_to_netdev(ring);
3833 struct sk_buff *skb;
3835 ring->skb = napi_alloc_skb(&ring->tqp_vector->napi, HNS3_RX_HEAD_SIZE);
3837 if (unlikely(!skb)) {
3838 hns3_rl_err(netdev, "alloc rx skb fail\n");
3840 u64_stats_update_begin(&ring->syncp);
3841 ring->stats.sw_err_cnt++;
3842 u64_stats_update_end(&ring->syncp);
3847 trace_hns3_rx_desc(ring);
3848 prefetchw(skb->data);
3850 ring->pending_buf = 1;
3852 ring->tail_skb = NULL;
3853 if (length <= HNS3_RX_HEAD_SIZE) {
3854 memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
3856 /* We can reuse buffer as-is, just make sure it is reusable */
3857 if (dev_page_is_reusable(desc_cb->priv))
3858 desc_cb->reuse_flag = 1;
3859 else if (desc_cb->type & DESC_TYPE_PP_FRAG)
3860 page_pool_put_full_page(ring->page_pool, desc_cb->priv,
3862 else /* This page cannot be reused so discard it */
3863 __page_frag_cache_drain(desc_cb->priv,
3864 desc_cb->pagecnt_bias);
3866 hns3_rx_ring_move_fw(ring);
3870 if (ring->page_pool)
3871 skb_mark_for_recycle(skb);
3873 u64_stats_update_begin(&ring->syncp);
3874 ring->stats.seg_pkt_cnt++;
3875 u64_stats_update_end(&ring->syncp);
3877 ring->pull_len = eth_get_headlen(netdev, va, HNS3_RX_HEAD_SIZE);
3878 __skb_put(skb, ring->pull_len);
3879 hns3_nic_reuse_page(skb, ring->frag_num++, ring, ring->pull_len,
3881 hns3_rx_ring_move_fw(ring);
3886 static int hns3_add_frag(struct hns3_enet_ring *ring)
3888 struct sk_buff *skb = ring->skb;
3889 struct sk_buff *head_skb = skb;
3890 struct sk_buff *new_skb;
3891 struct hns3_desc_cb *desc_cb;
3892 struct hns3_desc *desc;
3896 desc = &ring->desc[ring->next_to_clean];
3897 desc_cb = &ring->desc_cb[ring->next_to_clean];
3898 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
3899 /* make sure HW write desc complete */
3901 if (!(bd_base_info & BIT(HNS3_RXD_VLD_B)))
3904 if (unlikely(ring->frag_num >= MAX_SKB_FRAGS)) {
3905 new_skb = napi_alloc_skb(&ring->tqp_vector->napi, 0);
3906 if (unlikely(!new_skb)) {
3907 hns3_rl_err(ring_to_netdev(ring),
3908 "alloc rx fraglist skb fail\n");
3912 if (ring->page_pool)
3913 skb_mark_for_recycle(new_skb);
3917 if (ring->tail_skb) {
3918 ring->tail_skb->next = new_skb;
3919 ring->tail_skb = new_skb;
3921 skb_shinfo(skb)->frag_list = new_skb;
3922 ring->tail_skb = new_skb;
3926 if (ring->tail_skb) {
3927 head_skb->truesize += hns3_buf_size(ring);
3928 head_skb->data_len += le16_to_cpu(desc->rx.size);
3929 head_skb->len += le16_to_cpu(desc->rx.size);
3930 skb = ring->tail_skb;
3933 dma_sync_single_for_cpu(ring_to_dev(ring),
3934 desc_cb->dma + desc_cb->page_offset,
3935 hns3_buf_size(ring),
3938 hns3_nic_reuse_page(skb, ring->frag_num++, ring, 0, desc_cb);
3939 trace_hns3_rx_desc(ring);
3940 hns3_rx_ring_move_fw(ring);
3941 ring->pending_buf++;
3942 } while (!(bd_base_info & BIT(HNS3_RXD_FE_B)));
3947 static int hns3_set_gro_and_checksum(struct hns3_enet_ring *ring,
3948 struct sk_buff *skb, u32 l234info,
3949 u32 bd_base_info, u32 ol_info, u16 csum)
3951 struct net_device *netdev = ring_to_netdev(ring);
3952 struct hns3_nic_priv *priv = netdev_priv(netdev);
3955 skb_shinfo(skb)->gso_size = hnae3_get_field(bd_base_info,
3956 HNS3_RXD_GRO_SIZE_M,
3957 HNS3_RXD_GRO_SIZE_S);
3958 /* if there is no HW GRO, do not set gro params */
3959 if (!skb_shinfo(skb)->gso_size) {
3960 hns3_rx_checksum(ring, skb, l234info, bd_base_info, ol_info,
3965 NAPI_GRO_CB(skb)->count = hnae3_get_field(l234info,
3966 HNS3_RXD_GRO_COUNT_M,
3967 HNS3_RXD_GRO_COUNT_S);
3969 if (test_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state)) {
3970 u32 ptype = hnae3_get_field(ol_info, HNS3_RXD_PTYPE_M,
3973 l3_type = hns3_rx_ptype_tbl[ptype].l3_type;
3975 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
3979 if (l3_type == HNS3_L3_TYPE_IPV4)
3980 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
3981 else if (l3_type == HNS3_L3_TYPE_IPV6)
3982 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
3986 return hns3_gro_complete(skb, l234info);
3989 static void hns3_set_rx_skb_rss_type(struct hns3_enet_ring *ring,
3990 struct sk_buff *skb, u32 rss_hash)
3992 struct hnae3_handle *handle = ring->tqp->handle;
3993 enum pkt_hash_types rss_type;
3996 rss_type = handle->kinfo.rss_type;
3998 rss_type = PKT_HASH_TYPE_NONE;
4000 skb_set_hash(skb, rss_hash, rss_type);
4003 static int hns3_handle_bdinfo(struct hns3_enet_ring *ring, struct sk_buff *skb)
4005 struct net_device *netdev = ring_to_netdev(ring);
4006 enum hns3_pkt_l2t_type l2_frame_type;
4007 u32 bd_base_info, l234info, ol_info;
4008 struct hns3_desc *desc;
4013 /* bdinfo handled below is only valid on the last BD of the
4014 * current packet, and ring->next_to_clean indicates the first
4015 * descriptor of next packet, so need - 1 below.
4017 pre_ntc = ring->next_to_clean ? (ring->next_to_clean - 1) :
4018 (ring->desc_num - 1);
4019 desc = &ring->desc[pre_ntc];
4020 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
4021 l234info = le32_to_cpu(desc->rx.l234_info);
4022 ol_info = le32_to_cpu(desc->rx.ol_info);
4023 csum = le16_to_cpu(desc->csum);
4025 if (unlikely(bd_base_info & BIT(HNS3_RXD_TS_VLD_B))) {
4026 struct hnae3_handle *h = hns3_get_handle(netdev);
4027 u32 nsec = le32_to_cpu(desc->ts_nsec);
4028 u32 sec = le32_to_cpu(desc->ts_sec);
4030 if (h->ae_algo->ops->get_rx_hwts)
4031 h->ae_algo->ops->get_rx_hwts(h, skb, nsec, sec);
4034 /* Based on hw strategy, the tag offloaded will be stored at
4035 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
4036 * in one layer tag case.
4038 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
4041 if (hns3_parse_vlan_tag(ring, desc, l234info, &vlan_tag))
4042 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
4046 if (unlikely(!desc->rx.pkt_len || (l234info & (BIT(HNS3_RXD_TRUNCAT_B) |
4047 BIT(HNS3_RXD_L2E_B))))) {
4048 u64_stats_update_begin(&ring->syncp);
4049 if (l234info & BIT(HNS3_RXD_L2E_B))
4050 ring->stats.l2_err++;
4052 ring->stats.err_pkt_len++;
4053 u64_stats_update_end(&ring->syncp);
4060 /* Do update ip stack process */
4061 skb->protocol = eth_type_trans(skb, netdev);
4063 /* This is needed in order to enable forwarding support */
4064 ret = hns3_set_gro_and_checksum(ring, skb, l234info,
4065 bd_base_info, ol_info, csum);
4066 if (unlikely(ret)) {
4067 u64_stats_update_begin(&ring->syncp);
4068 ring->stats.rx_err_cnt++;
4069 u64_stats_update_end(&ring->syncp);
4073 l2_frame_type = hnae3_get_field(l234info, HNS3_RXD_DMAC_M,
4076 u64_stats_update_begin(&ring->syncp);
4077 ring->stats.rx_pkts++;
4078 ring->stats.rx_bytes += len;
4080 if (l2_frame_type == HNS3_L2_TYPE_MULTICAST)
4081 ring->stats.rx_multicast++;
4083 u64_stats_update_end(&ring->syncp);
4085 ring->tqp_vector->rx_group.total_bytes += len;
4087 hns3_set_rx_skb_rss_type(ring, skb, le32_to_cpu(desc->rx.rss_hash));
4091 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring)
4093 struct sk_buff *skb = ring->skb;
4094 struct hns3_desc_cb *desc_cb;
4095 struct hns3_desc *desc;
4096 unsigned int length;
4100 desc = &ring->desc[ring->next_to_clean];
4101 desc_cb = &ring->desc_cb[ring->next_to_clean];
4106 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
4107 /* Check valid BD */
4108 if (unlikely(!(bd_base_info & BIT(HNS3_RXD_VLD_B))))
4112 length = le16_to_cpu(desc->rx.size);
4114 ring->va = desc_cb->buf + desc_cb->page_offset;
4116 dma_sync_single_for_cpu(ring_to_dev(ring),
4117 desc_cb->dma + desc_cb->page_offset,
4118 hns3_buf_size(ring),
4121 /* Prefetch first cache line of first page.
4122 * Idea is to cache few bytes of the header of the packet.
4123 * Our L1 Cache line size is 64B so need to prefetch twice to make
4124 * it 128B. But in actual we can have greater size of caches with
4125 * 128B Level 1 cache lines. In such a case, single fetch would
4126 * suffice to cache in the relevant part of the header.
4128 net_prefetch(ring->va);
4130 ret = hns3_alloc_skb(ring, length, ring->va);
4133 if (ret < 0) /* alloc buffer fail */
4135 if (!(bd_base_info & BIT(HNS3_RXD_FE_B))) { /* need add frag */
4136 ret = hns3_add_frag(ring);
4141 ret = hns3_add_frag(ring);
4146 /* As the head data may be changed when GRO enable, copy
4147 * the head data in after other data rx completed
4149 if (skb->len > HNS3_RX_HEAD_SIZE)
4150 memcpy(skb->data, ring->va,
4151 ALIGN(ring->pull_len, sizeof(long)));
4153 ret = hns3_handle_bdinfo(ring, skb);
4154 if (unlikely(ret)) {
4155 dev_kfree_skb_any(skb);
4159 skb_record_rx_queue(skb, ring->tqp->tqp_index);
4163 int hns3_clean_rx_ring(struct hns3_enet_ring *ring, int budget,
4164 void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
4166 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
4167 int unused_count = hns3_desc_unused(ring);
4171 unused_count -= ring->pending_buf;
4173 while (recv_pkts < budget) {
4174 /* Reuse or realloc buffers */
4175 if (unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
4176 hns3_nic_alloc_rx_buffers(ring, unused_count);
4177 unused_count = hns3_desc_unused(ring) -
4182 err = hns3_handle_rx_bd(ring);
4183 /* Do not get FE for the packet or failed to alloc skb */
4184 if (unlikely(!ring->skb || err == -ENXIO)) {
4186 } else if (likely(!err)) {
4187 rx_fn(ring, ring->skb);
4191 unused_count += ring->pending_buf;
4193 ring->pending_buf = 0;
4197 /* Make all data has been write before submit */
4198 if (unused_count > 0)
4199 hns3_nic_alloc_rx_buffers(ring, unused_count);
4204 static void hns3_update_rx_int_coalesce(struct hns3_enet_tqp_vector *tqp_vector)
4206 struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
4207 struct dim_sample sample = {};
4209 if (!rx_group->coal.adapt_enable)
4212 dim_update_sample(tqp_vector->event_cnt, rx_group->total_packets,
4213 rx_group->total_bytes, &sample);
4214 net_dim(&rx_group->dim, sample);
4217 static void hns3_update_tx_int_coalesce(struct hns3_enet_tqp_vector *tqp_vector)
4219 struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
4220 struct dim_sample sample = {};
4222 if (!tx_group->coal.adapt_enable)
4225 dim_update_sample(tqp_vector->event_cnt, tx_group->total_packets,
4226 tx_group->total_bytes, &sample);
4227 net_dim(&tx_group->dim, sample);
4230 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
4232 struct hns3_nic_priv *priv = netdev_priv(napi->dev);
4233 struct hns3_enet_ring *ring;
4234 int rx_pkt_total = 0;
4236 struct hns3_enet_tqp_vector *tqp_vector =
4237 container_of(napi, struct hns3_enet_tqp_vector, napi);
4238 bool clean_complete = true;
4239 int rx_budget = budget;
4241 if (unlikely(test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
4242 napi_complete(napi);
4246 /* Since the actual Tx work is minimal, we can give the Tx a larger
4247 * budget and be more aggressive about cleaning up the Tx descriptors.
4249 hns3_for_each_ring(ring, tqp_vector->tx_group)
4250 hns3_clean_tx_ring(ring, budget);
4252 /* make sure rx ring budget not smaller than 1 */
4253 if (tqp_vector->num_tqps > 1)
4254 rx_budget = max(budget / tqp_vector->num_tqps, 1);
4256 hns3_for_each_ring(ring, tqp_vector->rx_group) {
4257 int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
4259 if (rx_cleaned >= rx_budget)
4260 clean_complete = false;
4262 rx_pkt_total += rx_cleaned;
4265 tqp_vector->rx_group.total_packets += rx_pkt_total;
4267 if (!clean_complete)
4270 if (napi_complete(napi) &&
4271 likely(!test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
4272 hns3_update_rx_int_coalesce(tqp_vector);
4273 hns3_update_tx_int_coalesce(tqp_vector);
4275 hns3_mask_vector_irq(tqp_vector, 1);
4278 return rx_pkt_total;
4281 static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
4282 struct hnae3_ring_chain_node *head)
4284 struct pci_dev *pdev = tqp_vector->handle->pdev;
4285 struct hnae3_ring_chain_node *cur_chain = head;
4286 struct hnae3_ring_chain_node *chain;
4287 struct hns3_enet_ring *tx_ring;
4288 struct hns3_enet_ring *rx_ring;
4290 tx_ring = tqp_vector->tx_group.ring;
4292 cur_chain->tqp_index = tx_ring->tqp->tqp_index;
4293 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
4294 HNAE3_RING_TYPE_TX);
4295 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
4296 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_TX);
4298 cur_chain->next = NULL;
4300 while (tx_ring->next) {
4301 tx_ring = tx_ring->next;
4303 chain = devm_kzalloc(&pdev->dev, sizeof(*chain),
4306 goto err_free_chain;
4308 cur_chain->next = chain;
4309 chain->tqp_index = tx_ring->tqp->tqp_index;
4310 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
4311 HNAE3_RING_TYPE_TX);
4312 hnae3_set_field(chain->int_gl_idx,
4313 HNAE3_RING_GL_IDX_M,
4314 HNAE3_RING_GL_IDX_S,
4321 rx_ring = tqp_vector->rx_group.ring;
4322 if (!tx_ring && rx_ring) {
4323 cur_chain->next = NULL;
4324 cur_chain->tqp_index = rx_ring->tqp->tqp_index;
4325 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
4326 HNAE3_RING_TYPE_RX);
4327 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
4328 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
4330 rx_ring = rx_ring->next;
4334 chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
4336 goto err_free_chain;
4338 cur_chain->next = chain;
4339 chain->tqp_index = rx_ring->tqp->tqp_index;
4340 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
4341 HNAE3_RING_TYPE_RX);
4342 hnae3_set_field(chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
4343 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
4347 rx_ring = rx_ring->next;
4353 cur_chain = head->next;
4355 chain = cur_chain->next;
4356 devm_kfree(&pdev->dev, cur_chain);
4364 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
4365 struct hnae3_ring_chain_node *head)
4367 struct pci_dev *pdev = tqp_vector->handle->pdev;
4368 struct hnae3_ring_chain_node *chain_tmp, *chain;
4373 chain_tmp = chain->next;
4374 devm_kfree(&pdev->dev, chain);
4379 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
4380 struct hns3_enet_ring *ring)
4382 ring->next = group->ring;
4388 static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
4390 struct pci_dev *pdev = priv->ae_handle->pdev;
4391 struct hns3_enet_tqp_vector *tqp_vector;
4392 int num_vectors = priv->vector_num;
4396 numa_node = dev_to_node(&pdev->dev);
4398 for (vector_i = 0; vector_i < num_vectors; vector_i++) {
4399 tqp_vector = &priv->tqp_vector[vector_i];
4400 cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
4401 &tqp_vector->affinity_mask);
4405 static void hns3_rx_dim_work(struct work_struct *work)
4407 struct dim *dim = container_of(work, struct dim, work);
4408 struct hns3_enet_ring_group *group = container_of(dim,
4409 struct hns3_enet_ring_group, dim);
4410 struct hns3_enet_tqp_vector *tqp_vector = group->ring->tqp_vector;
4411 struct dim_cq_moder cur_moder =
4412 net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
4414 hns3_set_vector_coalesce_rx_gl(group->ring->tqp_vector, cur_moder.usec);
4415 tqp_vector->rx_group.coal.int_gl = cur_moder.usec;
4417 if (cur_moder.pkts < tqp_vector->rx_group.coal.int_ql_max) {
4418 hns3_set_vector_coalesce_rx_ql(tqp_vector, cur_moder.pkts);
4419 tqp_vector->rx_group.coal.int_ql = cur_moder.pkts;
4422 dim->state = DIM_START_MEASURE;
4425 static void hns3_tx_dim_work(struct work_struct *work)
4427 struct dim *dim = container_of(work, struct dim, work);
4428 struct hns3_enet_ring_group *group = container_of(dim,
4429 struct hns3_enet_ring_group, dim);
4430 struct hns3_enet_tqp_vector *tqp_vector = group->ring->tqp_vector;
4431 struct dim_cq_moder cur_moder =
4432 net_dim_get_tx_moderation(dim->mode, dim->profile_ix);
4434 hns3_set_vector_coalesce_tx_gl(tqp_vector, cur_moder.usec);
4435 tqp_vector->tx_group.coal.int_gl = cur_moder.usec;
4437 if (cur_moder.pkts < tqp_vector->tx_group.coal.int_ql_max) {
4438 hns3_set_vector_coalesce_tx_ql(tqp_vector, cur_moder.pkts);
4439 tqp_vector->tx_group.coal.int_ql = cur_moder.pkts;
4442 dim->state = DIM_START_MEASURE;
4445 static void hns3_nic_init_dim(struct hns3_enet_tqp_vector *tqp_vector)
4447 INIT_WORK(&tqp_vector->rx_group.dim.work, hns3_rx_dim_work);
4448 INIT_WORK(&tqp_vector->tx_group.dim.work, hns3_tx_dim_work);
4451 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
4453 struct hnae3_handle *h = priv->ae_handle;
4454 struct hns3_enet_tqp_vector *tqp_vector;
4458 hns3_nic_set_cpumask(priv);
4460 for (i = 0; i < priv->vector_num; i++) {
4461 tqp_vector = &priv->tqp_vector[i];
4462 hns3_vector_coalesce_init_hw(tqp_vector, priv);
4463 tqp_vector->num_tqps = 0;
4464 hns3_nic_init_dim(tqp_vector);
4467 for (i = 0; i < h->kinfo.num_tqps; i++) {
4468 u16 vector_i = i % priv->vector_num;
4469 u16 tqp_num = h->kinfo.num_tqps;
4471 tqp_vector = &priv->tqp_vector[vector_i];
4473 hns3_add_ring_to_group(&tqp_vector->tx_group,
4476 hns3_add_ring_to_group(&tqp_vector->rx_group,
4477 &priv->ring[i + tqp_num]);
4479 priv->ring[i].tqp_vector = tqp_vector;
4480 priv->ring[i + tqp_num].tqp_vector = tqp_vector;
4481 tqp_vector->num_tqps++;
4484 for (i = 0; i < priv->vector_num; i++) {
4485 struct hnae3_ring_chain_node vector_ring_chain;
4487 tqp_vector = &priv->tqp_vector[i];
4489 tqp_vector->rx_group.total_bytes = 0;
4490 tqp_vector->rx_group.total_packets = 0;
4491 tqp_vector->tx_group.total_bytes = 0;
4492 tqp_vector->tx_group.total_packets = 0;
4493 tqp_vector->handle = h;
4495 ret = hns3_get_vector_ring_chain(tqp_vector,
4496 &vector_ring_chain);
4500 ret = h->ae_algo->ops->map_ring_to_vector(h,
4501 tqp_vector->vector_irq, &vector_ring_chain);
4503 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
4508 netif_napi_add(priv->netdev, &tqp_vector->napi,
4509 hns3_nic_common_poll, NAPI_POLL_WEIGHT);
4516 netif_napi_del(&priv->tqp_vector[i].napi);
4521 static void hns3_nic_init_coal_cfg(struct hns3_nic_priv *priv)
4523 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
4524 struct hns3_enet_coalesce *tx_coal = &priv->tx_coal;
4525 struct hns3_enet_coalesce *rx_coal = &priv->rx_coal;
4527 /* initialize the configuration for interrupt coalescing.
4528 * 1. GL (Interrupt Gap Limiter)
4529 * 2. RL (Interrupt Rate Limiter)
4530 * 3. QL (Interrupt Quantity Limiter)
4532 * Default: enable interrupt coalescing self-adaptive and GL
4534 tx_coal->adapt_enable = 1;
4535 rx_coal->adapt_enable = 1;
4537 tx_coal->int_gl = HNS3_INT_GL_50K;
4538 rx_coal->int_gl = HNS3_INT_GL_50K;
4540 rx_coal->flow_level = HNS3_FLOW_LOW;
4541 tx_coal->flow_level = HNS3_FLOW_LOW;
4543 if (ae_dev->dev_specs.int_ql_max) {
4544 tx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
4545 rx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
4549 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
4551 struct hnae3_handle *h = priv->ae_handle;
4552 struct hns3_enet_tqp_vector *tqp_vector;
4553 struct hnae3_vector_info *vector;
4554 struct pci_dev *pdev = h->pdev;
4555 u16 tqp_num = h->kinfo.num_tqps;
4560 /* RSS size, cpu online and vector_num should be the same */
4561 /* Should consider 2p/4p later */
4562 vector_num = min_t(u16, num_online_cpus(), tqp_num);
4564 vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
4569 /* save the actual available vector number */
4570 vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
4572 priv->vector_num = vector_num;
4573 priv->tqp_vector = (struct hns3_enet_tqp_vector *)
4574 devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
4576 if (!priv->tqp_vector) {
4581 for (i = 0; i < priv->vector_num; i++) {
4582 tqp_vector = &priv->tqp_vector[i];
4583 tqp_vector->idx = i;
4584 tqp_vector->mask_addr = vector[i].io_addr;
4585 tqp_vector->vector_irq = vector[i].vector;
4586 hns3_vector_coalesce_init(tqp_vector, priv);
4590 devm_kfree(&pdev->dev, vector);
4594 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
4600 static void hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
4602 struct hnae3_ring_chain_node vector_ring_chain;
4603 struct hnae3_handle *h = priv->ae_handle;
4604 struct hns3_enet_tqp_vector *tqp_vector;
4607 for (i = 0; i < priv->vector_num; i++) {
4608 tqp_vector = &priv->tqp_vector[i];
4610 if (!tqp_vector->rx_group.ring && !tqp_vector->tx_group.ring)
4613 /* Since the mapping can be overwritten, when fail to get the
4614 * chain between vector and ring, we should go on to deal with
4615 * the remaining options.
4617 if (hns3_get_vector_ring_chain(tqp_vector, &vector_ring_chain))
4618 dev_warn(priv->dev, "failed to get ring chain\n");
4620 h->ae_algo->ops->unmap_ring_from_vector(h,
4621 tqp_vector->vector_irq, &vector_ring_chain);
4623 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
4625 hns3_clear_ring_group(&tqp_vector->rx_group);
4626 hns3_clear_ring_group(&tqp_vector->tx_group);
4627 netif_napi_del(&priv->tqp_vector[i].napi);
4631 static void hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
4633 struct hnae3_handle *h = priv->ae_handle;
4634 struct pci_dev *pdev = h->pdev;
4637 for (i = 0; i < priv->vector_num; i++) {
4638 struct hns3_enet_tqp_vector *tqp_vector;
4640 tqp_vector = &priv->tqp_vector[i];
4641 ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
4646 devm_kfree(&pdev->dev, priv->tqp_vector);
4649 static void hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
4650 unsigned int ring_type)
4652 int queue_num = priv->ae_handle->kinfo.num_tqps;
4653 struct hns3_enet_ring *ring;
4656 if (ring_type == HNAE3_RING_TYPE_TX) {
4657 ring = &priv->ring[q->tqp_index];
4658 desc_num = priv->ae_handle->kinfo.num_tx_desc;
4659 ring->queue_index = q->tqp_index;
4660 ring->tx_copybreak = priv->tx_copybreak;
4661 ring->last_to_use = 0;
4663 ring = &priv->ring[q->tqp_index + queue_num];
4664 desc_num = priv->ae_handle->kinfo.num_rx_desc;
4665 ring->queue_index = q->tqp_index;
4666 ring->rx_copybreak = priv->rx_copybreak;
4669 hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
4673 ring->desc_cb = NULL;
4674 ring->dev = priv->dev;
4675 ring->desc_dma_addr = 0;
4676 ring->buf_size = q->buf_size;
4677 ring->desc_num = desc_num;
4678 ring->next_to_use = 0;
4679 ring->next_to_clean = 0;
4682 static void hns3_queue_to_ring(struct hnae3_queue *tqp,
4683 struct hns3_nic_priv *priv)
4685 hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
4686 hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
4689 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
4691 struct hnae3_handle *h = priv->ae_handle;
4692 struct pci_dev *pdev = h->pdev;
4695 priv->ring = devm_kzalloc(&pdev->dev,
4696 array3_size(h->kinfo.num_tqps,
4697 sizeof(*priv->ring), 2),
4702 for (i = 0; i < h->kinfo.num_tqps; i++)
4703 hns3_queue_to_ring(h->kinfo.tqp[i], priv);
4708 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
4713 devm_kfree(priv->dev, priv->ring);
4717 static void hns3_alloc_page_pool(struct hns3_enet_ring *ring)
4719 struct page_pool_params pp_params = {
4720 .flags = PP_FLAG_DMA_MAP | PP_FLAG_PAGE_FRAG |
4721 PP_FLAG_DMA_SYNC_DEV,
4722 .order = hns3_page_order(ring),
4723 .pool_size = ring->desc_num * hns3_buf_size(ring) /
4724 (PAGE_SIZE << hns3_page_order(ring)),
4725 .nid = dev_to_node(ring_to_dev(ring)),
4726 .dev = ring_to_dev(ring),
4727 .dma_dir = DMA_FROM_DEVICE,
4729 .max_len = PAGE_SIZE << hns3_page_order(ring),
4732 ring->page_pool = page_pool_create(&pp_params);
4733 if (IS_ERR(ring->page_pool)) {
4734 dev_warn(ring_to_dev(ring), "page pool creation failed: %ld\n",
4735 PTR_ERR(ring->page_pool));
4736 ring->page_pool = NULL;
4740 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
4744 if (ring->desc_num <= 0 || ring->buf_size <= 0)
4747 ring->desc_cb = devm_kcalloc(ring_to_dev(ring), ring->desc_num,
4748 sizeof(ring->desc_cb[0]), GFP_KERNEL);
4749 if (!ring->desc_cb) {
4754 ret = hns3_alloc_desc(ring);
4756 goto out_with_desc_cb;
4758 if (!HNAE3_IS_TX_RING(ring)) {
4759 if (page_pool_enabled)
4760 hns3_alloc_page_pool(ring);
4762 ret = hns3_alloc_ring_buffers(ring);
4766 hns3_init_tx_spare_buffer(ring);
4772 hns3_free_desc(ring);
4774 devm_kfree(ring_to_dev(ring), ring->desc_cb);
4775 ring->desc_cb = NULL;
4780 void hns3_fini_ring(struct hns3_enet_ring *ring)
4782 hns3_free_desc(ring);
4783 devm_kfree(ring_to_dev(ring), ring->desc_cb);
4784 ring->desc_cb = NULL;
4785 ring->next_to_clean = 0;
4786 ring->next_to_use = 0;
4787 ring->last_to_use = 0;
4788 ring->pending_buf = 0;
4789 if (!HNAE3_IS_TX_RING(ring) && ring->skb) {
4790 dev_kfree_skb_any(ring->skb);
4792 } else if (HNAE3_IS_TX_RING(ring) && ring->tx_spare) {
4793 struct hns3_tx_spare *tx_spare = ring->tx_spare;
4795 dma_unmap_page(ring_to_dev(ring), tx_spare->dma, tx_spare->len,
4797 free_pages((unsigned long)tx_spare->buf,
4798 get_order(tx_spare->len));
4799 devm_kfree(ring_to_dev(ring), tx_spare);
4800 ring->tx_spare = NULL;
4803 if (!HNAE3_IS_TX_RING(ring) && ring->page_pool) {
4804 page_pool_destroy(ring->page_pool);
4805 ring->page_pool = NULL;
4809 static int hns3_buf_size2type(u32 buf_size)
4815 bd_size_type = HNS3_BD_SIZE_512_TYPE;
4818 bd_size_type = HNS3_BD_SIZE_1024_TYPE;
4821 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4824 bd_size_type = HNS3_BD_SIZE_4096_TYPE;
4827 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4830 return bd_size_type;
4833 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
4835 dma_addr_t dma = ring->desc_dma_addr;
4836 struct hnae3_queue *q = ring->tqp;
4838 if (!HNAE3_IS_TX_RING(ring)) {
4839 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG, (u32)dma);
4840 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
4841 (u32)((dma >> 31) >> 1));
4843 hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
4844 hns3_buf_size2type(ring->buf_size));
4845 hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
4846 ring->desc_num / 8 - 1);
4848 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
4850 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
4851 (u32)((dma >> 31) >> 1));
4853 hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
4854 ring->desc_num / 8 - 1);
4858 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
4860 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
4861 struct hnae3_tc_info *tc_info = &kinfo->tc_info;
4864 for (i = 0; i < HNAE3_MAX_TC; i++) {
4867 if (!test_bit(i, &tc_info->tc_en))
4870 for (j = 0; j < tc_info->tqp_count[i]; j++) {
4871 struct hnae3_queue *q;
4873 q = priv->ring[tc_info->tqp_offset[i] + j].tqp;
4874 hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG, i);
4879 int hns3_init_all_ring(struct hns3_nic_priv *priv)
4881 struct hnae3_handle *h = priv->ae_handle;
4882 int ring_num = h->kinfo.num_tqps * 2;
4886 for (i = 0; i < ring_num; i++) {
4887 ret = hns3_alloc_ring_memory(&priv->ring[i]);
4890 "Alloc ring memory fail! ret=%d\n", ret);
4891 goto out_when_alloc_ring_memory;
4894 u64_stats_init(&priv->ring[i].syncp);
4899 out_when_alloc_ring_memory:
4900 for (j = i - 1; j >= 0; j--)
4901 hns3_fini_ring(&priv->ring[j]);
4906 static void hns3_uninit_all_ring(struct hns3_nic_priv *priv)
4908 struct hnae3_handle *h = priv->ae_handle;
4911 for (i = 0; i < h->kinfo.num_tqps; i++) {
4912 hns3_fini_ring(&priv->ring[i]);
4913 hns3_fini_ring(&priv->ring[i + h->kinfo.num_tqps]);
4917 /* Set mac addr if it is configured. or leave it to the AE driver */
4918 static int hns3_init_mac_addr(struct net_device *netdev)
4920 struct hns3_nic_priv *priv = netdev_priv(netdev);
4921 struct hnae3_handle *h = priv->ae_handle;
4922 u8 mac_addr_temp[ETH_ALEN];
4925 if (h->ae_algo->ops->get_mac_addr)
4926 h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
4928 /* Check if the MAC address is valid, if not get a random one */
4929 if (!is_valid_ether_addr(mac_addr_temp)) {
4930 eth_hw_addr_random(netdev);
4931 dev_warn(priv->dev, "using random MAC address %pM\n",
4933 } else if (!ether_addr_equal(netdev->dev_addr, mac_addr_temp)) {
4934 ether_addr_copy(netdev->dev_addr, mac_addr_temp);
4935 ether_addr_copy(netdev->perm_addr, mac_addr_temp);
4940 if (h->ae_algo->ops->set_mac_addr)
4941 ret = h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
4946 static int hns3_init_phy(struct net_device *netdev)
4948 struct hnae3_handle *h = hns3_get_handle(netdev);
4951 if (h->ae_algo->ops->mac_connect_phy)
4952 ret = h->ae_algo->ops->mac_connect_phy(h);
4957 static void hns3_uninit_phy(struct net_device *netdev)
4959 struct hnae3_handle *h = hns3_get_handle(netdev);
4961 if (h->ae_algo->ops->mac_disconnect_phy)
4962 h->ae_algo->ops->mac_disconnect_phy(h);
4965 static int hns3_client_start(struct hnae3_handle *handle)
4967 if (!handle->ae_algo->ops->client_start)
4970 return handle->ae_algo->ops->client_start(handle);
4973 static void hns3_client_stop(struct hnae3_handle *handle)
4975 if (!handle->ae_algo->ops->client_stop)
4978 handle->ae_algo->ops->client_stop(handle);
4981 static void hns3_info_show(struct hns3_nic_priv *priv)
4983 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
4985 dev_info(priv->dev, "MAC address: %pM\n", priv->netdev->dev_addr);
4986 dev_info(priv->dev, "Task queue pairs numbers: %u\n", kinfo->num_tqps);
4987 dev_info(priv->dev, "RSS size: %u\n", kinfo->rss_size);
4988 dev_info(priv->dev, "Allocated RSS size: %u\n", kinfo->req_rss_size);
4989 dev_info(priv->dev, "RX buffer length: %u\n", kinfo->rx_buf_len);
4990 dev_info(priv->dev, "Desc num per TX queue: %u\n", kinfo->num_tx_desc);
4991 dev_info(priv->dev, "Desc num per RX queue: %u\n", kinfo->num_rx_desc);
4992 dev_info(priv->dev, "Total number of enabled TCs: %u\n",
4993 kinfo->tc_info.num_tc);
4994 dev_info(priv->dev, "Max mtu size: %u\n", priv->netdev->max_mtu);
4997 static void hns3_set_cq_period_mode(struct hns3_nic_priv *priv,
4998 enum dim_cq_period_mode mode, bool is_tx)
5000 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
5001 struct hnae3_handle *handle = priv->ae_handle;
5005 priv->tx_cqe_mode = mode;
5007 for (i = 0; i < priv->vector_num; i++)
5008 priv->tqp_vector[i].tx_group.dim.mode = mode;
5010 priv->rx_cqe_mode = mode;
5012 for (i = 0; i < priv->vector_num; i++)
5013 priv->tqp_vector[i].rx_group.dim.mode = mode;
5016 /* only device version above V3(include V3), GL can switch CQ/EQ
5019 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3) {
5023 new_mode = (mode == DIM_CQ_PERIOD_MODE_START_FROM_CQE) ?
5024 HNS3_CQ_MODE_CQE : HNS3_CQ_MODE_EQE;
5025 reg = is_tx ? HNS3_GL1_CQ_MODE_REG : HNS3_GL0_CQ_MODE_REG;
5027 writel(new_mode, handle->kinfo.io_base + reg);
5031 void hns3_cq_period_mode_init(struct hns3_nic_priv *priv,
5032 enum dim_cq_period_mode tx_mode,
5033 enum dim_cq_period_mode rx_mode)
5035 hns3_set_cq_period_mode(priv, tx_mode, true);
5036 hns3_set_cq_period_mode(priv, rx_mode, false);
5039 static void hns3_state_init(struct hnae3_handle *handle)
5041 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(handle->pdev);
5042 struct net_device *netdev = handle->kinfo.netdev;
5043 struct hns3_nic_priv *priv = netdev_priv(netdev);
5045 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
5047 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
5048 set_bit(HNAE3_PFLAG_LIMIT_PROMISC, &handle->supported_pflags);
5050 if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps))
5051 set_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state);
5053 if (hnae3_ae_dev_rxd_adv_layout_supported(ae_dev))
5054 set_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state);
5057 static int hns3_client_init(struct hnae3_handle *handle)
5059 struct pci_dev *pdev = handle->pdev;
5060 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
5061 u16 alloc_tqps, max_rss_size;
5062 struct hns3_nic_priv *priv;
5063 struct net_device *netdev;
5066 handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
5068 netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
5072 priv = netdev_priv(netdev);
5073 priv->dev = &pdev->dev;
5074 priv->netdev = netdev;
5075 priv->ae_handle = handle;
5076 priv->tx_timeout_count = 0;
5077 priv->max_non_tso_bd_num = ae_dev->dev_specs.max_non_tso_bd_num;
5078 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
5080 handle->msg_enable = netif_msg_init(debug, DEFAULT_MSG_LEVEL);
5082 handle->kinfo.netdev = netdev;
5083 handle->priv = (void *)priv;
5085 hns3_init_mac_addr(netdev);
5087 hns3_set_default_feature(netdev);
5089 netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
5090 netdev->priv_flags |= IFF_UNICAST_FLT;
5091 netdev->netdev_ops = &hns3_nic_netdev_ops;
5092 SET_NETDEV_DEV(netdev, &pdev->dev);
5093 hns3_ethtool_set_ops(netdev);
5095 /* Carrier off reporting is important to ethtool even BEFORE open */
5096 netif_carrier_off(netdev);
5098 ret = hns3_get_ring_config(priv);
5101 goto out_get_ring_cfg;
5104 hns3_nic_init_coal_cfg(priv);
5106 ret = hns3_nic_alloc_vector_data(priv);
5109 goto out_alloc_vector_data;
5112 ret = hns3_nic_init_vector_data(priv);
5115 goto out_init_vector_data;
5118 ret = hns3_init_all_ring(priv);
5124 hns3_cq_period_mode_init(priv, DIM_CQ_PERIOD_MODE_START_FROM_EQE,
5125 DIM_CQ_PERIOD_MODE_START_FROM_EQE);
5127 ret = hns3_init_phy(netdev);
5131 /* the device can work without cpu rmap, only aRFS needs it */
5132 ret = hns3_set_rx_cpu_rmap(netdev);
5134 dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
5136 ret = hns3_nic_init_irq(priv);
5138 dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
5139 hns3_free_rx_cpu_rmap(netdev);
5140 goto out_init_irq_fail;
5143 ret = hns3_client_start(handle);
5145 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
5146 goto out_client_start;
5149 hns3_dcbnl_setup(handle);
5151 ret = hns3_dbg_init(handle);
5153 dev_err(priv->dev, "failed to init debugfs, ret = %d\n",
5155 goto out_client_start;
5158 netdev->max_mtu = HNS3_MAX_MTU(ae_dev->dev_specs.max_frm_size);
5160 hns3_state_init(handle);
5162 ret = register_netdev(netdev);
5164 dev_err(priv->dev, "probe register netdev fail!\n");
5165 goto out_reg_netdev_fail;
5168 if (netif_msg_drv(handle))
5169 hns3_info_show(priv);
5173 out_reg_netdev_fail:
5174 hns3_dbg_uninit(handle);
5176 hns3_free_rx_cpu_rmap(netdev);
5177 hns3_nic_uninit_irq(priv);
5179 hns3_uninit_phy(netdev);
5181 hns3_uninit_all_ring(priv);
5183 hns3_nic_uninit_vector_data(priv);
5184 out_init_vector_data:
5185 hns3_nic_dealloc_vector_data(priv);
5186 out_alloc_vector_data:
5189 priv->ae_handle = NULL;
5190 free_netdev(netdev);
5194 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
5196 struct net_device *netdev = handle->kinfo.netdev;
5197 struct hns3_nic_priv *priv = netdev_priv(netdev);
5199 if (netdev->reg_state != NETREG_UNINITIALIZED)
5200 unregister_netdev(netdev);
5202 hns3_client_stop(handle);
5204 hns3_uninit_phy(netdev);
5206 if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5207 netdev_warn(netdev, "already uninitialized\n");
5208 goto out_netdev_free;
5211 hns3_free_rx_cpu_rmap(netdev);
5213 hns3_nic_uninit_irq(priv);
5215 hns3_clear_all_ring(handle, true);
5217 hns3_nic_uninit_vector_data(priv);
5219 hns3_nic_dealloc_vector_data(priv);
5221 hns3_uninit_all_ring(priv);
5223 hns3_put_ring_config(priv);
5226 hns3_dbg_uninit(handle);
5227 free_netdev(netdev);
5230 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
5232 struct net_device *netdev = handle->kinfo.netdev;
5238 netif_tx_wake_all_queues(netdev);
5239 netif_carrier_on(netdev);
5240 if (netif_msg_link(handle))
5241 netdev_info(netdev, "link up\n");
5243 netif_carrier_off(netdev);
5244 netif_tx_stop_all_queues(netdev);
5245 if (netif_msg_link(handle))
5246 netdev_info(netdev, "link down\n");
5250 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
5252 while (ring->next_to_clean != ring->next_to_use) {
5253 ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
5254 hns3_free_buffer_detach(ring, ring->next_to_clean, 0);
5255 ring_ptr_move_fw(ring, next_to_clean);
5258 ring->pending_buf = 0;
5261 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
5263 struct hns3_desc_cb res_cbs;
5266 while (ring->next_to_use != ring->next_to_clean) {
5267 /* When a buffer is not reused, it's memory has been
5268 * freed in hns3_handle_rx_bd or will be freed by
5269 * stack, so we need to replace the buffer here.
5271 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
5272 ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
5274 u64_stats_update_begin(&ring->syncp);
5275 ring->stats.sw_err_cnt++;
5276 u64_stats_update_end(&ring->syncp);
5277 /* if alloc new buffer fail, exit directly
5278 * and reclear in up flow.
5280 netdev_warn(ring_to_netdev(ring),
5281 "reserve buffer map failed, ret = %d\n",
5285 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
5287 ring_ptr_move_fw(ring, next_to_use);
5290 /* Free the pending skb in rx ring */
5292 dev_kfree_skb_any(ring->skb);
5294 ring->pending_buf = 0;
5300 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
5302 while (ring->next_to_use != ring->next_to_clean) {
5303 /* When a buffer is not reused, it's memory has been
5304 * freed in hns3_handle_rx_bd or will be freed by
5305 * stack, so only need to unmap the buffer here.
5307 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
5308 hns3_unmap_buffer(ring,
5309 &ring->desc_cb[ring->next_to_use]);
5310 ring->desc_cb[ring->next_to_use].dma = 0;
5313 ring_ptr_move_fw(ring, next_to_use);
5317 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force)
5319 struct net_device *ndev = h->kinfo.netdev;
5320 struct hns3_nic_priv *priv = netdev_priv(ndev);
5323 for (i = 0; i < h->kinfo.num_tqps; i++) {
5324 struct hns3_enet_ring *ring;
5326 ring = &priv->ring[i];
5327 hns3_clear_tx_ring(ring);
5329 ring = &priv->ring[i + h->kinfo.num_tqps];
5330 /* Continue to clear other rings even if clearing some
5334 hns3_force_clear_rx_ring(ring);
5336 hns3_clear_rx_ring(ring);
5340 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
5342 struct net_device *ndev = h->kinfo.netdev;
5343 struct hns3_nic_priv *priv = netdev_priv(ndev);
5344 struct hns3_enet_ring *rx_ring;
5348 ret = h->ae_algo->ops->reset_queue(h);
5352 for (i = 0; i < h->kinfo.num_tqps; i++) {
5353 hns3_init_ring_hw(&priv->ring[i]);
5355 /* We need to clear tx ring here because self test will
5356 * use the ring and will not run down before up
5358 hns3_clear_tx_ring(&priv->ring[i]);
5359 priv->ring[i].next_to_clean = 0;
5360 priv->ring[i].next_to_use = 0;
5361 priv->ring[i].last_to_use = 0;
5363 rx_ring = &priv->ring[i + h->kinfo.num_tqps];
5364 hns3_init_ring_hw(rx_ring);
5365 ret = hns3_clear_rx_ring(rx_ring);
5369 /* We can not know the hardware head and tail when this
5370 * function is called in reset flow, so we reuse all desc.
5372 for (j = 0; j < rx_ring->desc_num; j++)
5373 hns3_reuse_buffer(rx_ring, j);
5375 rx_ring->next_to_clean = 0;
5376 rx_ring->next_to_use = 0;
5379 hns3_init_tx_ring_tc(priv);
5384 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
5386 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
5387 struct net_device *ndev = kinfo->netdev;
5388 struct hns3_nic_priv *priv = netdev_priv(ndev);
5390 if (test_and_set_bit(HNS3_NIC_STATE_RESETTING, &priv->state))
5393 if (!netif_running(ndev))
5396 return hns3_nic_net_stop(ndev);
5399 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
5401 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
5402 struct hns3_nic_priv *priv = netdev_priv(kinfo->netdev);
5405 if (!test_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5406 netdev_err(kinfo->netdev, "device is not initialized yet\n");
5410 clear_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
5412 if (netif_running(kinfo->netdev)) {
5413 ret = hns3_nic_net_open(kinfo->netdev);
5415 set_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
5416 netdev_err(kinfo->netdev,
5417 "net up fail, ret=%d!\n", ret);
5425 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
5427 struct net_device *netdev = handle->kinfo.netdev;
5428 struct hns3_nic_priv *priv = netdev_priv(netdev);
5431 /* Carrier off reporting is important to ethtool even BEFORE open */
5432 netif_carrier_off(netdev);
5434 ret = hns3_get_ring_config(priv);
5438 ret = hns3_nic_alloc_vector_data(priv);
5442 ret = hns3_nic_init_vector_data(priv);
5444 goto err_dealloc_vector;
5446 ret = hns3_init_all_ring(priv);
5448 goto err_uninit_vector;
5450 hns3_cq_period_mode_init(priv, priv->tx_cqe_mode, priv->rx_cqe_mode);
5452 /* the device can work without cpu rmap, only aRFS needs it */
5453 ret = hns3_set_rx_cpu_rmap(netdev);
5455 dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
5457 ret = hns3_nic_init_irq(priv);
5459 dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
5460 hns3_free_rx_cpu_rmap(netdev);
5461 goto err_init_irq_fail;
5464 if (!hns3_is_phys_func(handle->pdev))
5465 hns3_init_mac_addr(netdev);
5467 ret = hns3_client_start(handle);
5469 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
5470 goto err_client_start_fail;
5473 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
5477 err_client_start_fail:
5478 hns3_free_rx_cpu_rmap(netdev);
5479 hns3_nic_uninit_irq(priv);
5481 hns3_uninit_all_ring(priv);
5483 hns3_nic_uninit_vector_data(priv);
5485 hns3_nic_dealloc_vector_data(priv);
5487 hns3_put_ring_config(priv);
5492 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
5494 struct net_device *netdev = handle->kinfo.netdev;
5495 struct hns3_nic_priv *priv = netdev_priv(netdev);
5497 if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5498 netdev_warn(netdev, "already uninitialized\n");
5502 hns3_free_rx_cpu_rmap(netdev);
5503 hns3_nic_uninit_irq(priv);
5504 hns3_clear_all_ring(handle, true);
5505 hns3_reset_tx_queue(priv->ae_handle);
5507 hns3_nic_uninit_vector_data(priv);
5509 hns3_nic_dealloc_vector_data(priv);
5511 hns3_uninit_all_ring(priv);
5513 hns3_put_ring_config(priv);
5518 static int hns3_reset_notify(struct hnae3_handle *handle,
5519 enum hnae3_reset_notify_type type)
5524 case HNAE3_UP_CLIENT:
5525 ret = hns3_reset_notify_up_enet(handle);
5527 case HNAE3_DOWN_CLIENT:
5528 ret = hns3_reset_notify_down_enet(handle);
5530 case HNAE3_INIT_CLIENT:
5531 ret = hns3_reset_notify_init_enet(handle);
5533 case HNAE3_UNINIT_CLIENT:
5534 ret = hns3_reset_notify_uninit_enet(handle);
5543 static int hns3_change_channels(struct hnae3_handle *handle, u32 new_tqp_num,
5544 bool rxfh_configured)
5548 ret = handle->ae_algo->ops->set_channels(handle, new_tqp_num,
5551 dev_err(&handle->pdev->dev,
5552 "Change tqp num(%u) fail.\n", new_tqp_num);
5556 ret = hns3_reset_notify(handle, HNAE3_INIT_CLIENT);
5560 ret = hns3_reset_notify(handle, HNAE3_UP_CLIENT);
5562 hns3_reset_notify(handle, HNAE3_UNINIT_CLIENT);
5567 int hns3_set_channels(struct net_device *netdev,
5568 struct ethtool_channels *ch)
5570 struct hnae3_handle *h = hns3_get_handle(netdev);
5571 struct hnae3_knic_private_info *kinfo = &h->kinfo;
5572 bool rxfh_configured = netif_is_rxfh_configured(netdev);
5573 u32 new_tqp_num = ch->combined_count;
5577 if (hns3_nic_resetting(netdev))
5580 if (ch->rx_count || ch->tx_count)
5583 if (kinfo->tc_info.mqprio_active) {
5584 dev_err(&netdev->dev,
5585 "it's not allowed to set channels via ethtool when MQPRIO mode is on\n");
5589 if (new_tqp_num > hns3_get_max_available_channels(h) ||
5591 dev_err(&netdev->dev,
5592 "Change tqps fail, the tqp range is from 1 to %u",
5593 hns3_get_max_available_channels(h));
5597 if (kinfo->rss_size == new_tqp_num)
5600 netif_dbg(h, drv, netdev,
5601 "set channels: tqp_num=%u, rxfh=%d\n",
5602 new_tqp_num, rxfh_configured);
5604 ret = hns3_reset_notify(h, HNAE3_DOWN_CLIENT);
5608 ret = hns3_reset_notify(h, HNAE3_UNINIT_CLIENT);
5612 org_tqp_num = h->kinfo.num_tqps;
5613 ret = hns3_change_channels(h, new_tqp_num, rxfh_configured);
5618 "Change channels fail, revert to old value\n");
5619 ret1 = hns3_change_channels(h, org_tqp_num, rxfh_configured);
5622 "revert to old channel fail\n");
5632 static const struct hns3_hw_error_info hns3_hw_err[] = {
5633 { .type = HNAE3_PPU_POISON_ERROR,
5634 .msg = "PPU poison" },
5635 { .type = HNAE3_CMDQ_ECC_ERROR,
5636 .msg = "IMP CMDQ error" },
5637 { .type = HNAE3_IMP_RD_POISON_ERROR,
5638 .msg = "IMP RD poison" },
5639 { .type = HNAE3_ROCEE_AXI_RESP_ERROR,
5640 .msg = "ROCEE AXI RESP error" },
5643 static void hns3_process_hw_error(struct hnae3_handle *handle,
5644 enum hnae3_hw_error_type type)
5648 for (i = 0; i < ARRAY_SIZE(hns3_hw_err); i++) {
5649 if (hns3_hw_err[i].type == type) {
5650 dev_err(&handle->pdev->dev, "Detected %s!\n",
5651 hns3_hw_err[i].msg);
5657 static const struct hnae3_client_ops client_ops = {
5658 .init_instance = hns3_client_init,
5659 .uninit_instance = hns3_client_uninit,
5660 .link_status_change = hns3_link_status_change,
5661 .reset_notify = hns3_reset_notify,
5662 .process_hw_error = hns3_process_hw_error,
5665 /* hns3_init_module - Driver registration routine
5666 * hns3_init_module is the first routine called when the driver is
5667 * loaded. All it does is register with the PCI subsystem.
5669 static int __init hns3_init_module(void)
5673 pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
5674 pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
5676 client.type = HNAE3_CLIENT_KNIC;
5677 snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH, "%s",
5680 client.ops = &client_ops;
5682 INIT_LIST_HEAD(&client.node);
5684 hns3_dbg_register_debugfs(hns3_driver_name);
5686 ret = hnae3_register_client(&client);
5688 goto err_reg_client;
5690 ret = pci_register_driver(&hns3_driver);
5692 goto err_reg_driver;
5697 hnae3_unregister_client(&client);
5699 hns3_dbg_unregister_debugfs();
5702 module_init(hns3_init_module);
5704 /* hns3_exit_module - Driver exit cleanup routine
5705 * hns3_exit_module is called just before the driver is removed
5708 static void __exit hns3_exit_module(void)
5710 pci_unregister_driver(&hns3_driver);
5711 hnae3_unregister_client(&client);
5712 hns3_dbg_unregister_debugfs();
5714 module_exit(hns3_exit_module);
5716 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
5717 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
5718 MODULE_LICENSE("GPL");
5719 MODULE_ALIAS("pci:hns-nic");