projects / linux-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
net: hns3: add dql info when tx timeout
[linux-block.git] / drivers / net / ethernet / hisilicon / hns3 / hns3_enet.c
... / ...
CommitLineData
1// SPDX-License-Identifier: GPL-2.0+
2// Copyright (c) 2016-2017 Hisilicon Limited.
3
4#include <linux/dma-mapping.h>
5#include <linux/etherdevice.h>
6#include <linux/interrupt.h>
7#ifdef CONFIG_RFS_ACCEL
8#include <linux/cpu_rmap.h>
9#endif
10#include <linux/if_vlan.h>
11#include <linux/irq.h>
12#include <linux/ip.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>
19#include <net/gre.h>
20#include <net/gro.h>
21#include <net/ip6_checksum.h>
22#include <net/pkt_cls.h>
23#include <net/tcp.h>
24#include <net/vxlan.h>
25#include <net/geneve.h>
26
27#include "hnae3.h"
28#include "hns3_enet.h"
29/* All hns3 tracepoints are defined by the include below, which
30 * must be included exactly once across the whole kernel with
31 * CREATE_TRACE_POINTS defined
32 */
33#define CREATE_TRACE_POINTS
34#include "hns3_trace.h"
35
36#define hns3_set_field(origin, shift, val) ((origin) |= (val) << (shift))
37#define hns3_tx_bd_count(S) DIV_ROUND_UP(S, HNS3_MAX_BD_SIZE)
38
39#define hns3_rl_err(fmt, ...) \
40 do { \
41 if (net_ratelimit()) \
42 netdev_err(fmt, ##__VA_ARGS__); \
43 } while (0)
44
45static void hns3_clear_all_ring(struct hnae3_handle *h, bool force);
46
47static const char hns3_driver_name[] = "hns3";
48static const char hns3_driver_string[] =
49 "Hisilicon Ethernet Network Driver for Hip08 Family";
50static const char hns3_copyright[] = "Copyright (c) 2017 Huawei Corporation.";
51static struct hnae3_client client;
52
53static int debug = -1;
54module_param(debug, int, 0);
55MODULE_PARM_DESC(debug, " Network interface message level setting");
56
57static unsigned int tx_sgl = 1;
58module_param(tx_sgl, uint, 0600);
59MODULE_PARM_DESC(tx_sgl, "Minimum number of frags when using dma_map_sg() to optimize the IOMMU mapping");
60
61static bool page_pool_enabled = true;
62module_param(page_pool_enabled, bool, 0400);
63
64#define HNS3_SGL_SIZE(nfrag) (sizeof(struct scatterlist) * (nfrag) + \
65 sizeof(struct sg_table))
66#define HNS3_MAX_SGL_SIZE ALIGN(HNS3_SGL_SIZE(HNS3_MAX_TSO_BD_NUM), \
67 dma_get_cache_alignment())
68
69#define DEFAULT_MSG_LEVEL (NETIF_MSG_PROBE | NETIF_MSG_LINK | \
70 NETIF_MSG_IFDOWN | NETIF_MSG_IFUP)
71
72#define HNS3_INNER_VLAN_TAG 1
73#define HNS3_OUTER_VLAN_TAG 2
74
75#define HNS3_MIN_TX_LEN 33U
76#define HNS3_MIN_TUN_PKT_LEN 65U
77
78/* hns3_pci_tbl - PCI Device ID Table
79 *
80 * Last entry must be all 0s
81 *
82 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
83 * Class, Class Mask, private data (not used) }
84 */
85static const struct pci_device_id hns3_pci_tbl[] = {
86 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
87 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
88 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA),
89 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
90 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC),
91 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
92 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA),
93 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
94 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC),
95 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
96 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC),
97 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
98 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_200G_RDMA),
99 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
100 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_VF), 0},
101 {PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_RDMA_DCB_PFC_VF),
102 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
103 /* required last entry */
104 {0,}
105};
106MODULE_DEVICE_TABLE(pci, hns3_pci_tbl);
107
108#define HNS3_RX_PTYPE_ENTRY(ptype, l, s, t) \
109 { ptype, \
110 l, \
111 CHECKSUM_##s, \
112 HNS3_L3_TYPE_##t, \
113 1 }
114
115#define HNS3_RX_PTYPE_UNUSED_ENTRY(ptype) \
116 { ptype, 0, CHECKSUM_NONE, HNS3_L3_TYPE_PARSE_FAIL, 0 }
117
118static const struct hns3_rx_ptype hns3_rx_ptype_tbl[] = {
119 HNS3_RX_PTYPE_UNUSED_ENTRY(0),
120 HNS3_RX_PTYPE_ENTRY(1, 0, COMPLETE, ARP),
121 HNS3_RX_PTYPE_ENTRY(2, 0, COMPLETE, RARP),
122 HNS3_RX_PTYPE_ENTRY(3, 0, COMPLETE, LLDP),
123 HNS3_RX_PTYPE_ENTRY(4, 0, COMPLETE, PARSE_FAIL),
124 HNS3_RX_PTYPE_ENTRY(5, 0, COMPLETE, PARSE_FAIL),
125 HNS3_RX_PTYPE_ENTRY(6, 0, COMPLETE, PARSE_FAIL),
126 HNS3_RX_PTYPE_ENTRY(7, 0, COMPLETE, CNM),
127 HNS3_RX_PTYPE_ENTRY(8, 0, NONE, PARSE_FAIL),
128 HNS3_RX_PTYPE_UNUSED_ENTRY(9),
129 HNS3_RX_PTYPE_UNUSED_ENTRY(10),
130 HNS3_RX_PTYPE_UNUSED_ENTRY(11),
131 HNS3_RX_PTYPE_UNUSED_ENTRY(12),
132 HNS3_RX_PTYPE_UNUSED_ENTRY(13),
133 HNS3_RX_PTYPE_UNUSED_ENTRY(14),
134 HNS3_RX_PTYPE_UNUSED_ENTRY(15),
135 HNS3_RX_PTYPE_ENTRY(16, 0, COMPLETE, PARSE_FAIL),
136 HNS3_RX_PTYPE_ENTRY(17, 0, COMPLETE, IPV4),
137 HNS3_RX_PTYPE_ENTRY(18, 0, COMPLETE, IPV4),
138 HNS3_RX_PTYPE_ENTRY(19, 0, UNNECESSARY, IPV4),
139 HNS3_RX_PTYPE_ENTRY(20, 0, UNNECESSARY, IPV4),
140 HNS3_RX_PTYPE_ENTRY(21, 0, NONE, IPV4),
141 HNS3_RX_PTYPE_ENTRY(22, 0, UNNECESSARY, IPV4),
142 HNS3_RX_PTYPE_ENTRY(23, 0, NONE, IPV4),
143 HNS3_RX_PTYPE_ENTRY(24, 0, NONE, IPV4),
144 HNS3_RX_PTYPE_ENTRY(25, 0, UNNECESSARY, IPV4),
145 HNS3_RX_PTYPE_UNUSED_ENTRY(26),
146 HNS3_RX_PTYPE_UNUSED_ENTRY(27),
147 HNS3_RX_PTYPE_UNUSED_ENTRY(28),
148 HNS3_RX_PTYPE_ENTRY(29, 0, COMPLETE, PARSE_FAIL),
149 HNS3_RX_PTYPE_ENTRY(30, 0, COMPLETE, PARSE_FAIL),
150 HNS3_RX_PTYPE_ENTRY(31, 0, COMPLETE, IPV4),
151 HNS3_RX_PTYPE_ENTRY(32, 0, COMPLETE, IPV4),
152 HNS3_RX_PTYPE_ENTRY(33, 1, UNNECESSARY, IPV4),
153 HNS3_RX_PTYPE_ENTRY(34, 1, UNNECESSARY, IPV4),
154 HNS3_RX_PTYPE_ENTRY(35, 1, UNNECESSARY, IPV4),
155 HNS3_RX_PTYPE_ENTRY(36, 0, COMPLETE, IPV4),
156 HNS3_RX_PTYPE_ENTRY(37, 0, COMPLETE, IPV4),
157 HNS3_RX_PTYPE_UNUSED_ENTRY(38),
158 HNS3_RX_PTYPE_ENTRY(39, 0, COMPLETE, IPV6),
159 HNS3_RX_PTYPE_ENTRY(40, 0, COMPLETE, IPV6),
160 HNS3_RX_PTYPE_ENTRY(41, 1, UNNECESSARY, IPV6),
161 HNS3_RX_PTYPE_ENTRY(42, 1, UNNECESSARY, IPV6),
162 HNS3_RX_PTYPE_ENTRY(43, 1, UNNECESSARY, IPV6),
163 HNS3_RX_PTYPE_ENTRY(44, 0, COMPLETE, IPV6),
164 HNS3_RX_PTYPE_ENTRY(45, 0, COMPLETE, IPV6),
165 HNS3_RX_PTYPE_UNUSED_ENTRY(46),
166 HNS3_RX_PTYPE_UNUSED_ENTRY(47),
167 HNS3_RX_PTYPE_UNUSED_ENTRY(48),
168 HNS3_RX_PTYPE_UNUSED_ENTRY(49),
169 HNS3_RX_PTYPE_UNUSED_ENTRY(50),
170 HNS3_RX_PTYPE_UNUSED_ENTRY(51),
171 HNS3_RX_PTYPE_UNUSED_ENTRY(52),
172 HNS3_RX_PTYPE_UNUSED_ENTRY(53),
173 HNS3_RX_PTYPE_UNUSED_ENTRY(54),
174 HNS3_RX_PTYPE_UNUSED_ENTRY(55),
175 HNS3_RX_PTYPE_UNUSED_ENTRY(56),
176 HNS3_RX_PTYPE_UNUSED_ENTRY(57),
177 HNS3_RX_PTYPE_UNUSED_ENTRY(58),
178 HNS3_RX_PTYPE_UNUSED_ENTRY(59),
179 HNS3_RX_PTYPE_UNUSED_ENTRY(60),
180 HNS3_RX_PTYPE_UNUSED_ENTRY(61),
181 HNS3_RX_PTYPE_UNUSED_ENTRY(62),
182 HNS3_RX_PTYPE_UNUSED_ENTRY(63),
183 HNS3_RX_PTYPE_UNUSED_ENTRY(64),
184 HNS3_RX_PTYPE_UNUSED_ENTRY(65),
185 HNS3_RX_PTYPE_UNUSED_ENTRY(66),
186 HNS3_RX_PTYPE_UNUSED_ENTRY(67),
187 HNS3_RX_PTYPE_UNUSED_ENTRY(68),
188 HNS3_RX_PTYPE_UNUSED_ENTRY(69),
189 HNS3_RX_PTYPE_UNUSED_ENTRY(70),
190 HNS3_RX_PTYPE_UNUSED_ENTRY(71),
191 HNS3_RX_PTYPE_UNUSED_ENTRY(72),
192 HNS3_RX_PTYPE_UNUSED_ENTRY(73),
193 HNS3_RX_PTYPE_UNUSED_ENTRY(74),
194 HNS3_RX_PTYPE_UNUSED_ENTRY(75),
195 HNS3_RX_PTYPE_UNUSED_ENTRY(76),
196 HNS3_RX_PTYPE_UNUSED_ENTRY(77),
197 HNS3_RX_PTYPE_UNUSED_ENTRY(78),
198 HNS3_RX_PTYPE_UNUSED_ENTRY(79),
199 HNS3_RX_PTYPE_UNUSED_ENTRY(80),
200 HNS3_RX_PTYPE_UNUSED_ENTRY(81),
201 HNS3_RX_PTYPE_UNUSED_ENTRY(82),
202 HNS3_RX_PTYPE_UNUSED_ENTRY(83),
203 HNS3_RX_PTYPE_UNUSED_ENTRY(84),
204 HNS3_RX_PTYPE_UNUSED_ENTRY(85),
205 HNS3_RX_PTYPE_UNUSED_ENTRY(86),
206 HNS3_RX_PTYPE_UNUSED_ENTRY(87),
207 HNS3_RX_PTYPE_UNUSED_ENTRY(88),
208 HNS3_RX_PTYPE_UNUSED_ENTRY(89),
209 HNS3_RX_PTYPE_UNUSED_ENTRY(90),
210 HNS3_RX_PTYPE_UNUSED_ENTRY(91),
211 HNS3_RX_PTYPE_UNUSED_ENTRY(92),
212 HNS3_RX_PTYPE_UNUSED_ENTRY(93),
213 HNS3_RX_PTYPE_UNUSED_ENTRY(94),
214 HNS3_RX_PTYPE_UNUSED_ENTRY(95),
215 HNS3_RX_PTYPE_UNUSED_ENTRY(96),
216 HNS3_RX_PTYPE_UNUSED_ENTRY(97),
217 HNS3_RX_PTYPE_UNUSED_ENTRY(98),
218 HNS3_RX_PTYPE_UNUSED_ENTRY(99),
219 HNS3_RX_PTYPE_UNUSED_ENTRY(100),
220 HNS3_RX_PTYPE_UNUSED_ENTRY(101),
221 HNS3_RX_PTYPE_UNUSED_ENTRY(102),
222 HNS3_RX_PTYPE_UNUSED_ENTRY(103),
223 HNS3_RX_PTYPE_UNUSED_ENTRY(104),
224 HNS3_RX_PTYPE_UNUSED_ENTRY(105),
225 HNS3_RX_PTYPE_UNUSED_ENTRY(106),
226 HNS3_RX_PTYPE_UNUSED_ENTRY(107),
227 HNS3_RX_PTYPE_UNUSED_ENTRY(108),
228 HNS3_RX_PTYPE_UNUSED_ENTRY(109),
229 HNS3_RX_PTYPE_UNUSED_ENTRY(110),
230 HNS3_RX_PTYPE_ENTRY(111, 0, COMPLETE, IPV6),
231 HNS3_RX_PTYPE_ENTRY(112, 0, COMPLETE, IPV6),
232 HNS3_RX_PTYPE_ENTRY(113, 0, UNNECESSARY, IPV6),
233 HNS3_RX_PTYPE_ENTRY(114, 0, UNNECESSARY, IPV6),
234 HNS3_RX_PTYPE_ENTRY(115, 0, NONE, IPV6),
235 HNS3_RX_PTYPE_ENTRY(116, 0, UNNECESSARY, IPV6),
236 HNS3_RX_PTYPE_ENTRY(117, 0, NONE, IPV6),
237 HNS3_RX_PTYPE_ENTRY(118, 0, NONE, IPV6),
238 HNS3_RX_PTYPE_ENTRY(119, 0, UNNECESSARY, IPV6),
239 HNS3_RX_PTYPE_UNUSED_ENTRY(120),
240 HNS3_RX_PTYPE_UNUSED_ENTRY(121),
241 HNS3_RX_PTYPE_UNUSED_ENTRY(122),
242 HNS3_RX_PTYPE_ENTRY(123, 0, COMPLETE, PARSE_FAIL),
243 HNS3_RX_PTYPE_ENTRY(124, 0, COMPLETE, PARSE_FAIL),
244 HNS3_RX_PTYPE_ENTRY(125, 0, COMPLETE, IPV4),
245 HNS3_RX_PTYPE_ENTRY(126, 0, COMPLETE, IPV4),
246 HNS3_RX_PTYPE_ENTRY(127, 1, UNNECESSARY, IPV4),
247 HNS3_RX_PTYPE_ENTRY(128, 1, UNNECESSARY, IPV4),
248 HNS3_RX_PTYPE_ENTRY(129, 1, UNNECESSARY, IPV4),
249 HNS3_RX_PTYPE_ENTRY(130, 0, COMPLETE, IPV4),
250 HNS3_RX_PTYPE_ENTRY(131, 0, COMPLETE, IPV4),
251 HNS3_RX_PTYPE_UNUSED_ENTRY(132),
252 HNS3_RX_PTYPE_ENTRY(133, 0, COMPLETE, IPV6),
253 HNS3_RX_PTYPE_ENTRY(134, 0, COMPLETE, IPV6),
254 HNS3_RX_PTYPE_ENTRY(135, 1, UNNECESSARY, IPV6),
255 HNS3_RX_PTYPE_ENTRY(136, 1, UNNECESSARY, IPV6),
256 HNS3_RX_PTYPE_ENTRY(137, 1, UNNECESSARY, IPV6),
257 HNS3_RX_PTYPE_ENTRY(138, 0, COMPLETE, IPV6),
258 HNS3_RX_PTYPE_ENTRY(139, 0, COMPLETE, IPV6),
259 HNS3_RX_PTYPE_UNUSED_ENTRY(140),
260 HNS3_RX_PTYPE_UNUSED_ENTRY(141),
261 HNS3_RX_PTYPE_UNUSED_ENTRY(142),
262 HNS3_RX_PTYPE_UNUSED_ENTRY(143),
263 HNS3_RX_PTYPE_UNUSED_ENTRY(144),
264 HNS3_RX_PTYPE_UNUSED_ENTRY(145),
265 HNS3_RX_PTYPE_UNUSED_ENTRY(146),
266 HNS3_RX_PTYPE_UNUSED_ENTRY(147),
267 HNS3_RX_PTYPE_UNUSED_ENTRY(148),
268 HNS3_RX_PTYPE_UNUSED_ENTRY(149),
269 HNS3_RX_PTYPE_UNUSED_ENTRY(150),
270 HNS3_RX_PTYPE_UNUSED_ENTRY(151),
271 HNS3_RX_PTYPE_UNUSED_ENTRY(152),
272 HNS3_RX_PTYPE_UNUSED_ENTRY(153),
273 HNS3_RX_PTYPE_UNUSED_ENTRY(154),
274 HNS3_RX_PTYPE_UNUSED_ENTRY(155),
275 HNS3_RX_PTYPE_UNUSED_ENTRY(156),
276 HNS3_RX_PTYPE_UNUSED_ENTRY(157),
277 HNS3_RX_PTYPE_UNUSED_ENTRY(158),
278 HNS3_RX_PTYPE_UNUSED_ENTRY(159),
279 HNS3_RX_PTYPE_UNUSED_ENTRY(160),
280 HNS3_RX_PTYPE_UNUSED_ENTRY(161),
281 HNS3_RX_PTYPE_UNUSED_ENTRY(162),
282 HNS3_RX_PTYPE_UNUSED_ENTRY(163),
283 HNS3_RX_PTYPE_UNUSED_ENTRY(164),
284 HNS3_RX_PTYPE_UNUSED_ENTRY(165),
285 HNS3_RX_PTYPE_UNUSED_ENTRY(166),
286 HNS3_RX_PTYPE_UNUSED_ENTRY(167),
287 HNS3_RX_PTYPE_UNUSED_ENTRY(168),
288 HNS3_RX_PTYPE_UNUSED_ENTRY(169),
289 HNS3_RX_PTYPE_UNUSED_ENTRY(170),
290 HNS3_RX_PTYPE_UNUSED_ENTRY(171),
291 HNS3_RX_PTYPE_UNUSED_ENTRY(172),
292 HNS3_RX_PTYPE_UNUSED_ENTRY(173),
293 HNS3_RX_PTYPE_UNUSED_ENTRY(174),
294 HNS3_RX_PTYPE_UNUSED_ENTRY(175),
295 HNS3_RX_PTYPE_UNUSED_ENTRY(176),
296 HNS3_RX_PTYPE_UNUSED_ENTRY(177),
297 HNS3_RX_PTYPE_UNUSED_ENTRY(178),
298 HNS3_RX_PTYPE_UNUSED_ENTRY(179),
299 HNS3_RX_PTYPE_UNUSED_ENTRY(180),
300 HNS3_RX_PTYPE_UNUSED_ENTRY(181),
301 HNS3_RX_PTYPE_UNUSED_ENTRY(182),
302 HNS3_RX_PTYPE_UNUSED_ENTRY(183),
303 HNS3_RX_PTYPE_UNUSED_ENTRY(184),
304 HNS3_RX_PTYPE_UNUSED_ENTRY(185),
305 HNS3_RX_PTYPE_UNUSED_ENTRY(186),
306 HNS3_RX_PTYPE_UNUSED_ENTRY(187),
307 HNS3_RX_PTYPE_UNUSED_ENTRY(188),
308 HNS3_RX_PTYPE_UNUSED_ENTRY(189),
309 HNS3_RX_PTYPE_UNUSED_ENTRY(190),
310 HNS3_RX_PTYPE_UNUSED_ENTRY(191),
311 HNS3_RX_PTYPE_UNUSED_ENTRY(192),
312 HNS3_RX_PTYPE_UNUSED_ENTRY(193),
313 HNS3_RX_PTYPE_UNUSED_ENTRY(194),
314 HNS3_RX_PTYPE_UNUSED_ENTRY(195),
315 HNS3_RX_PTYPE_UNUSED_ENTRY(196),
316 HNS3_RX_PTYPE_UNUSED_ENTRY(197),
317 HNS3_RX_PTYPE_UNUSED_ENTRY(198),
318 HNS3_RX_PTYPE_UNUSED_ENTRY(199),
319 HNS3_RX_PTYPE_UNUSED_ENTRY(200),
320 HNS3_RX_PTYPE_UNUSED_ENTRY(201),
321 HNS3_RX_PTYPE_UNUSED_ENTRY(202),
322 HNS3_RX_PTYPE_UNUSED_ENTRY(203),
323 HNS3_RX_PTYPE_UNUSED_ENTRY(204),
324 HNS3_RX_PTYPE_UNUSED_ENTRY(205),
325 HNS3_RX_PTYPE_UNUSED_ENTRY(206),
326 HNS3_RX_PTYPE_UNUSED_ENTRY(207),
327 HNS3_RX_PTYPE_UNUSED_ENTRY(208),
328 HNS3_RX_PTYPE_UNUSED_ENTRY(209),
329 HNS3_RX_PTYPE_UNUSED_ENTRY(210),
330 HNS3_RX_PTYPE_UNUSED_ENTRY(211),
331 HNS3_RX_PTYPE_UNUSED_ENTRY(212),
332 HNS3_RX_PTYPE_UNUSED_ENTRY(213),
333 HNS3_RX_PTYPE_UNUSED_ENTRY(214),
334 HNS3_RX_PTYPE_UNUSED_ENTRY(215),
335 HNS3_RX_PTYPE_UNUSED_ENTRY(216),
336 HNS3_RX_PTYPE_UNUSED_ENTRY(217),
337 HNS3_RX_PTYPE_UNUSED_ENTRY(218),
338 HNS3_RX_PTYPE_UNUSED_ENTRY(219),
339 HNS3_RX_PTYPE_UNUSED_ENTRY(220),
340 HNS3_RX_PTYPE_UNUSED_ENTRY(221),
341 HNS3_RX_PTYPE_UNUSED_ENTRY(222),
342 HNS3_RX_PTYPE_UNUSED_ENTRY(223),
343 HNS3_RX_PTYPE_UNUSED_ENTRY(224),
344 HNS3_RX_PTYPE_UNUSED_ENTRY(225),
345 HNS3_RX_PTYPE_UNUSED_ENTRY(226),
346 HNS3_RX_PTYPE_UNUSED_ENTRY(227),
347 HNS3_RX_PTYPE_UNUSED_ENTRY(228),
348 HNS3_RX_PTYPE_UNUSED_ENTRY(229),
349 HNS3_RX_PTYPE_UNUSED_ENTRY(230),
350 HNS3_RX_PTYPE_UNUSED_ENTRY(231),
351 HNS3_RX_PTYPE_UNUSED_ENTRY(232),
352 HNS3_RX_PTYPE_UNUSED_ENTRY(233),
353 HNS3_RX_PTYPE_UNUSED_ENTRY(234),
354 HNS3_RX_PTYPE_UNUSED_ENTRY(235),
355 HNS3_RX_PTYPE_UNUSED_ENTRY(236),
356 HNS3_RX_PTYPE_UNUSED_ENTRY(237),
357 HNS3_RX_PTYPE_UNUSED_ENTRY(238),
358 HNS3_RX_PTYPE_UNUSED_ENTRY(239),
359 HNS3_RX_PTYPE_UNUSED_ENTRY(240),
360 HNS3_RX_PTYPE_UNUSED_ENTRY(241),
361 HNS3_RX_PTYPE_UNUSED_ENTRY(242),
362 HNS3_RX_PTYPE_UNUSED_ENTRY(243),
363 HNS3_RX_PTYPE_UNUSED_ENTRY(244),
364 HNS3_RX_PTYPE_UNUSED_ENTRY(245),
365 HNS3_RX_PTYPE_UNUSED_ENTRY(246),
366 HNS3_RX_PTYPE_UNUSED_ENTRY(247),
367 HNS3_RX_PTYPE_UNUSED_ENTRY(248),
368 HNS3_RX_PTYPE_UNUSED_ENTRY(249),
369 HNS3_RX_PTYPE_UNUSED_ENTRY(250),
370 HNS3_RX_PTYPE_UNUSED_ENTRY(251),
371 HNS3_RX_PTYPE_UNUSED_ENTRY(252),
372 HNS3_RX_PTYPE_UNUSED_ENTRY(253),
373 HNS3_RX_PTYPE_UNUSED_ENTRY(254),
374 HNS3_RX_PTYPE_UNUSED_ENTRY(255),
375};
376
377#define HNS3_INVALID_PTYPE \
378 ARRAY_SIZE(hns3_rx_ptype_tbl)
379
380static irqreturn_t hns3_irq_handle(int irq, void *vector)
381{
382 struct hns3_enet_tqp_vector *tqp_vector = vector;
383
384 napi_schedule_irqoff(&tqp_vector->napi);
385 tqp_vector->event_cnt++;
386
387 return IRQ_HANDLED;
388}
389
390static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv)
391{
392 struct hns3_enet_tqp_vector *tqp_vectors;
393 unsigned int i;
394
395 for (i = 0; i < priv->vector_num; i++) {
396 tqp_vectors = &priv->tqp_vector[i];
397
398 if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED)
399 continue;
400
401 /* clear the affinity mask */
402 irq_set_affinity_hint(tqp_vectors->vector_irq, NULL);
403
404 /* release the irq resource */
405 free_irq(tqp_vectors->vector_irq, tqp_vectors);
406 tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED;
407 }
408}
409
410static int hns3_nic_init_irq(struct hns3_nic_priv *priv)
411{
412 struct hns3_enet_tqp_vector *tqp_vectors;
413 int txrx_int_idx = 0;
414 int rx_int_idx = 0;
415 int tx_int_idx = 0;
416 unsigned int i;
417 int ret;
418
419 for (i = 0; i < priv->vector_num; i++) {
420 tqp_vectors = &priv->tqp_vector[i];
421
422 if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED)
423 continue;
424
425 if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) {
426 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
427 "%s-%s-%s-%d", hns3_driver_name,
428 pci_name(priv->ae_handle->pdev),
429 "TxRx", txrx_int_idx++);
430 txrx_int_idx++;
431 } else if (tqp_vectors->rx_group.ring) {
432 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
433 "%s-%s-%s-%d", hns3_driver_name,
434 pci_name(priv->ae_handle->pdev),
435 "Rx", rx_int_idx++);
436 } else if (tqp_vectors->tx_group.ring) {
437 snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
438 "%s-%s-%s-%d", hns3_driver_name,
439 pci_name(priv->ae_handle->pdev),
440 "Tx", tx_int_idx++);
441 } else {
442 /* Skip this unused q_vector */
443 continue;
444 }
445
446 tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0';
447
448 irq_set_status_flags(tqp_vectors->vector_irq, IRQ_NOAUTOEN);
449 ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0,
450 tqp_vectors->name, tqp_vectors);
451 if (ret) {
452 netdev_err(priv->netdev, "request irq(%d) fail\n",
453 tqp_vectors->vector_irq);
454 hns3_nic_uninit_irq(priv);
455 return ret;
456 }
457
458 irq_set_affinity_hint(tqp_vectors->vector_irq,
459 &tqp_vectors->affinity_mask);
460
461 tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED;
462 }
463
464 return 0;
465}
466
467static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector,
468 u32 mask_en)
469{
470 writel(mask_en, tqp_vector->mask_addr);
471}
472
473static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector)
474{
475 napi_enable(&tqp_vector->napi);
476 enable_irq(tqp_vector->vector_irq);
477
478 /* enable vector */
479 hns3_mask_vector_irq(tqp_vector, 1);
480}
481
482static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector)
483{
484 /* disable vector */
485 hns3_mask_vector_irq(tqp_vector, 0);
486
487 disable_irq(tqp_vector->vector_irq);
488 napi_disable(&tqp_vector->napi);
489 cancel_work_sync(&tqp_vector->rx_group.dim.work);
490 cancel_work_sync(&tqp_vector->tx_group.dim.work);
491}
492
493void hns3_set_vector_coalesce_rl(struct hns3_enet_tqp_vector *tqp_vector,
494 u32 rl_value)
495{
496 u32 rl_reg = hns3_rl_usec_to_reg(rl_value);
497
498 /* this defines the configuration for RL (Interrupt Rate Limiter).
499 * Rl defines rate of interrupts i.e. number of interrupts-per-second
500 * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing
501 */
502 if (rl_reg > 0 && !tqp_vector->tx_group.coal.adapt_enable &&
503 !tqp_vector->rx_group.coal.adapt_enable)
504 /* According to the hardware, the range of rl_reg is
505 * 0-59 and the unit is 4.
506 */
507 rl_reg |= HNS3_INT_RL_ENABLE_MASK;
508
509 writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
510}
511
512void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
513 u32 gl_value)
514{
515 u32 new_val;
516
517 if (tqp_vector->rx_group.coal.unit_1us)
518 new_val = gl_value | HNS3_INT_GL_1US;
519 else
520 new_val = hns3_gl_usec_to_reg(gl_value);
521
522 writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
523}
524
525void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
526 u32 gl_value)
527{
528 u32 new_val;
529
530 if (tqp_vector->tx_group.coal.unit_1us)
531 new_val = gl_value | HNS3_INT_GL_1US;
532 else
533 new_val = hns3_gl_usec_to_reg(gl_value);
534
535 writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
536}
537
538void hns3_set_vector_coalesce_tx_ql(struct hns3_enet_tqp_vector *tqp_vector,
539 u32 ql_value)
540{
541 writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_TX_QL_OFFSET);
542}
543
544void hns3_set_vector_coalesce_rx_ql(struct hns3_enet_tqp_vector *tqp_vector,
545 u32 ql_value)
546{
547 writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_RX_QL_OFFSET);
548}
549
550static void hns3_vector_coalesce_init(struct hns3_enet_tqp_vector *tqp_vector,
551 struct hns3_nic_priv *priv)
552{
553 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
554 struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
555 struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
556 struct hns3_enet_coalesce *ptx_coal = &priv->tx_coal;
557 struct hns3_enet_coalesce *prx_coal = &priv->rx_coal;
558
559 tx_coal->adapt_enable = ptx_coal->adapt_enable;
560 rx_coal->adapt_enable = prx_coal->adapt_enable;
561
562 tx_coal->int_gl = ptx_coal->int_gl;
563 rx_coal->int_gl = prx_coal->int_gl;
564
565 rx_coal->flow_level = prx_coal->flow_level;
566 tx_coal->flow_level = ptx_coal->flow_level;
567
568 /* device version above V3(include V3), GL can configure 1us
569 * unit, so uses 1us unit.
570 */
571 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3) {
572 tx_coal->unit_1us = 1;
573 rx_coal->unit_1us = 1;
574 }
575
576 if (ae_dev->dev_specs.int_ql_max) {
577 tx_coal->ql_enable = 1;
578 rx_coal->ql_enable = 1;
579 tx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
580 rx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
581 tx_coal->int_ql = ptx_coal->int_ql;
582 rx_coal->int_ql = prx_coal->int_ql;
583 }
584}
585
586static void
587hns3_vector_coalesce_init_hw(struct hns3_enet_tqp_vector *tqp_vector,
588 struct hns3_nic_priv *priv)
589{
590 struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
591 struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
592 struct hnae3_handle *h = priv->ae_handle;
593
594 hns3_set_vector_coalesce_tx_gl(tqp_vector, tx_coal->int_gl);
595 hns3_set_vector_coalesce_rx_gl(tqp_vector, rx_coal->int_gl);
596 hns3_set_vector_coalesce_rl(tqp_vector, h->kinfo.int_rl_setting);
597
598 if (tx_coal->ql_enable)
599 hns3_set_vector_coalesce_tx_ql(tqp_vector, tx_coal->int_ql);
600
601 if (rx_coal->ql_enable)
602 hns3_set_vector_coalesce_rx_ql(tqp_vector, rx_coal->int_ql);
603}
604
605static int hns3_nic_set_real_num_queue(struct net_device *netdev)
606{
607 struct hnae3_handle *h = hns3_get_handle(netdev);
608 struct hnae3_knic_private_info *kinfo = &h->kinfo;
609 struct hnae3_tc_info *tc_info = &kinfo->tc_info;
610 unsigned int queue_size = kinfo->num_tqps;
611 int i, ret;
612
613 if (tc_info->num_tc <= 1 && !tc_info->mqprio_active) {
614 netdev_reset_tc(netdev);
615 } else {
616 ret = netdev_set_num_tc(netdev, tc_info->num_tc);
617 if (ret) {
618 netdev_err(netdev,
619 "netdev_set_num_tc fail, ret=%d!\n", ret);
620 return ret;
621 }
622
623 for (i = 0; i < tc_info->num_tc; i++)
624 netdev_set_tc_queue(netdev, i, tc_info->tqp_count[i],
625 tc_info->tqp_offset[i]);
626 }
627
628 ret = netif_set_real_num_tx_queues(netdev, queue_size);
629 if (ret) {
630 netdev_err(netdev,
631 "netif_set_real_num_tx_queues fail, ret=%d!\n", ret);
632 return ret;
633 }
634
635 ret = netif_set_real_num_rx_queues(netdev, queue_size);
636 if (ret) {
637 netdev_err(netdev,
638 "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
639 return ret;
640 }
641
642 return 0;
643}
644
645u16 hns3_get_max_available_channels(struct hnae3_handle *h)
646{
647 u16 alloc_tqps, max_rss_size, rss_size;
648
649 h->ae_algo->ops->get_tqps_and_rss_info(h, &alloc_tqps, &max_rss_size);
650 rss_size = alloc_tqps / h->kinfo.tc_info.num_tc;
651
652 return min_t(u16, rss_size, max_rss_size);
653}
654
655static void hns3_tqp_enable(struct hnae3_queue *tqp)
656{
657 u32 rcb_reg;
658
659 rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
660 rcb_reg |= BIT(HNS3_RING_EN_B);
661 hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
662}
663
664static void hns3_tqp_disable(struct hnae3_queue *tqp)
665{
666 u32 rcb_reg;
667
668 rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
669 rcb_reg &= ~BIT(HNS3_RING_EN_B);
670 hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
671}
672
673static void hns3_free_rx_cpu_rmap(struct net_device *netdev)
674{
675#ifdef CONFIG_RFS_ACCEL
676 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
677 netdev->rx_cpu_rmap = NULL;
678#endif
679}
680
681static int hns3_set_rx_cpu_rmap(struct net_device *netdev)
682{
683#ifdef CONFIG_RFS_ACCEL
684 struct hns3_nic_priv *priv = netdev_priv(netdev);
685 struct hns3_enet_tqp_vector *tqp_vector;
686 int i, ret;
687
688 if (!netdev->rx_cpu_rmap) {
689 netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->vector_num);
690 if (!netdev->rx_cpu_rmap)
691 return -ENOMEM;
692 }
693
694 for (i = 0; i < priv->vector_num; i++) {
695 tqp_vector = &priv->tqp_vector[i];
696 ret = irq_cpu_rmap_add(netdev->rx_cpu_rmap,
697 tqp_vector->vector_irq);
698 if (ret) {
699 hns3_free_rx_cpu_rmap(netdev);
700 return ret;
701 }
702 }
703#endif
704 return 0;
705}
706
707static int hns3_nic_net_up(struct net_device *netdev)
708{
709 struct hns3_nic_priv *priv = netdev_priv(netdev);
710 struct hnae3_handle *h = priv->ae_handle;
711 int i, j;
712 int ret;
713
714 ret = hns3_nic_reset_all_ring(h);
715 if (ret)
716 return ret;
717
718 clear_bit(HNS3_NIC_STATE_DOWN, &priv->state);
719
720 /* enable the vectors */
721 for (i = 0; i < priv->vector_num; i++)
722 hns3_vector_enable(&priv->tqp_vector[i]);
723
724 /* enable rcb */
725 for (j = 0; j < h->kinfo.num_tqps; j++)
726 hns3_tqp_enable(h->kinfo.tqp[j]);
727
728 /* start the ae_dev */
729 ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0;
730 if (ret) {
731 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
732 while (j--)
733 hns3_tqp_disable(h->kinfo.tqp[j]);
734
735 for (j = i - 1; j >= 0; j--)
736 hns3_vector_disable(&priv->tqp_vector[j]);
737 }
738
739 return ret;
740}
741
742static void hns3_config_xps(struct hns3_nic_priv *priv)
743{
744 int i;
745
746 for (i = 0; i < priv->vector_num; i++) {
747 struct hns3_enet_tqp_vector *tqp_vector = &priv->tqp_vector[i];
748 struct hns3_enet_ring *ring = tqp_vector->tx_group.ring;
749
750 while (ring) {
751 int ret;
752
753 ret = netif_set_xps_queue(priv->netdev,
754 &tqp_vector->affinity_mask,
755 ring->tqp->tqp_index);
756 if (ret)
757 netdev_warn(priv->netdev,
758 "set xps queue failed: %d", ret);
759
760 ring = ring->next;
761 }
762 }
763}
764
765static int hns3_nic_net_open(struct net_device *netdev)
766{
767 struct hns3_nic_priv *priv = netdev_priv(netdev);
768 struct hnae3_handle *h = hns3_get_handle(netdev);
769 struct hnae3_knic_private_info *kinfo;
770 int i, ret;
771
772 if (hns3_nic_resetting(netdev))
773 return -EBUSY;
774
775 if (!test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
776 netdev_warn(netdev, "net open repeatedly!\n");
777 return 0;
778 }
779
780 netif_carrier_off(netdev);
781
782 ret = hns3_nic_set_real_num_queue(netdev);
783 if (ret)
784 return ret;
785
786 ret = hns3_nic_net_up(netdev);
787 if (ret) {
788 netdev_err(netdev, "net up fail, ret=%d!\n", ret);
789 return ret;
790 }
791
792 kinfo = &h->kinfo;
793 for (i = 0; i < HNAE3_MAX_USER_PRIO; i++)
794 netdev_set_prio_tc_map(netdev, i, kinfo->tc_info.prio_tc[i]);
795
796 if (h->ae_algo->ops->set_timer_task)
797 h->ae_algo->ops->set_timer_task(priv->ae_handle, true);
798
799 hns3_config_xps(priv);
800
801 netif_dbg(h, drv, netdev, "net open\n");
802
803 return 0;
804}
805
806static void hns3_reset_tx_queue(struct hnae3_handle *h)
807{
808 struct net_device *ndev = h->kinfo.netdev;
809 struct hns3_nic_priv *priv = netdev_priv(ndev);
810 struct netdev_queue *dev_queue;
811 u32 i;
812
813 for (i = 0; i < h->kinfo.num_tqps; i++) {
814 dev_queue = netdev_get_tx_queue(ndev,
815 priv->ring[i].queue_index);
816 netdev_tx_reset_queue(dev_queue);
817 }
818}
819
820static void hns3_nic_net_down(struct net_device *netdev)
821{
822 struct hns3_nic_priv *priv = netdev_priv(netdev);
823 struct hnae3_handle *h = hns3_get_handle(netdev);
824 const struct hnae3_ae_ops *ops;
825 int i;
826
827 /* disable vectors */
828 for (i = 0; i < priv->vector_num; i++)
829 hns3_vector_disable(&priv->tqp_vector[i]);
830
831 /* disable rcb */
832 for (i = 0; i < h->kinfo.num_tqps; i++)
833 hns3_tqp_disable(h->kinfo.tqp[i]);
834
835 /* stop ae_dev */
836 ops = priv->ae_handle->ae_algo->ops;
837 if (ops->stop)
838 ops->stop(priv->ae_handle);
839
840 /* delay ring buffer clearing to hns3_reset_notify_uninit_enet
841 * during reset process, because driver may not be able
842 * to disable the ring through firmware when downing the netdev.
843 */
844 if (!hns3_nic_resetting(netdev))
845 hns3_clear_all_ring(priv->ae_handle, false);
846
847 hns3_reset_tx_queue(priv->ae_handle);
848}
849
850static int hns3_nic_net_stop(struct net_device *netdev)
851{
852 struct hns3_nic_priv *priv = netdev_priv(netdev);
853 struct hnae3_handle *h = hns3_get_handle(netdev);
854
855 if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
856 return 0;
857
858 netif_dbg(h, drv, netdev, "net stop\n");
859
860 if (h->ae_algo->ops->set_timer_task)
861 h->ae_algo->ops->set_timer_task(priv->ae_handle, false);
862
863 netif_carrier_off(netdev);
864 netif_tx_disable(netdev);
865
866 hns3_nic_net_down(netdev);
867
868 return 0;
869}
870
871static int hns3_nic_uc_sync(struct net_device *netdev,
872 const unsigned char *addr)
873{
874 struct hnae3_handle *h = hns3_get_handle(netdev);
875
876 if (h->ae_algo->ops->add_uc_addr)
877 return h->ae_algo->ops->add_uc_addr(h, addr);
878
879 return 0;
880}
881
882static int hns3_nic_uc_unsync(struct net_device *netdev,
883 const unsigned char *addr)
884{
885 struct hnae3_handle *h = hns3_get_handle(netdev);
886
887 /* need ignore the request of removing device address, because
888 * we store the device address and other addresses of uc list
889 * in the function's mac filter list.
890 */
891 if (ether_addr_equal(addr, netdev->dev_addr))
892 return 0;
893
894 if (h->ae_algo->ops->rm_uc_addr)
895 return h->ae_algo->ops->rm_uc_addr(h, addr);
896
897 return 0;
898}
899
900static int hns3_nic_mc_sync(struct net_device *netdev,
901 const unsigned char *addr)
902{
903 struct hnae3_handle *h = hns3_get_handle(netdev);
904
905 if (h->ae_algo->ops->add_mc_addr)
906 return h->ae_algo->ops->add_mc_addr(h, addr);
907
908 return 0;
909}
910
911static int hns3_nic_mc_unsync(struct net_device *netdev,
912 const unsigned char *addr)
913{
914 struct hnae3_handle *h = hns3_get_handle(netdev);
915
916 if (h->ae_algo->ops->rm_mc_addr)
917 return h->ae_algo->ops->rm_mc_addr(h, addr);
918
919 return 0;
920}
921
922static u8 hns3_get_netdev_flags(struct net_device *netdev)
923{
924 u8 flags = 0;
925
926 if (netdev->flags & IFF_PROMISC)
927 flags = HNAE3_USER_UPE | HNAE3_USER_MPE | HNAE3_BPE;
928 else if (netdev->flags & IFF_ALLMULTI)
929 flags = HNAE3_USER_MPE;
930
931 return flags;
932}
933
934static void hns3_nic_set_rx_mode(struct net_device *netdev)
935{
936 struct hnae3_handle *h = hns3_get_handle(netdev);
937 u8 new_flags;
938
939 new_flags = hns3_get_netdev_flags(netdev);
940
941 __dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync);
942 __dev_mc_sync(netdev, hns3_nic_mc_sync, hns3_nic_mc_unsync);
943
944 /* User mode Promisc mode enable and vlan filtering is disabled to
945 * let all packets in.
946 */
947 h->netdev_flags = new_flags;
948 hns3_request_update_promisc_mode(h);
949}
950
951void hns3_request_update_promisc_mode(struct hnae3_handle *handle)
952{
953 const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
954
955 if (ops->request_update_promisc_mode)
956 ops->request_update_promisc_mode(handle);
957}
958
959static u32 hns3_tx_spare_space(struct hns3_enet_ring *ring)
960{
961 struct hns3_tx_spare *tx_spare = ring->tx_spare;
962 u32 ntc, ntu;
963
964 /* This smp_load_acquire() pairs with smp_store_release() in
965 * hns3_tx_spare_update() called in tx desc cleaning process.
966 */
967 ntc = smp_load_acquire(&tx_spare->last_to_clean);
968 ntu = tx_spare->next_to_use;
969
970 if (ntc > ntu)
971 return ntc - ntu - 1;
972
973 /* The free tx buffer is divided into two part, so pick the
974 * larger one.
975 */
976 return max(ntc, tx_spare->len - ntu) - 1;
977}
978
979static void hns3_tx_spare_update(struct hns3_enet_ring *ring)
980{
981 struct hns3_tx_spare *tx_spare = ring->tx_spare;
982
983 if (!tx_spare ||
984 tx_spare->last_to_clean == tx_spare->next_to_clean)
985 return;
986
987 /* This smp_store_release() pairs with smp_load_acquire() in
988 * hns3_tx_spare_space() called in xmit process.
989 */
990 smp_store_release(&tx_spare->last_to_clean,
991 tx_spare->next_to_clean);
992}
993
994static bool hns3_can_use_tx_bounce(struct hns3_enet_ring *ring,
995 struct sk_buff *skb,
996 u32 space)
997{
998 u32 len = skb->len <= ring->tx_copybreak ? skb->len :
999 skb_headlen(skb);
1000
1001 if (len > ring->tx_copybreak)
1002 return false;
1003
1004 if (ALIGN(len, dma_get_cache_alignment()) > space) {
1005 u64_stats_update_begin(&ring->syncp);
1006 ring->stats.tx_spare_full++;
1007 u64_stats_update_end(&ring->syncp);
1008 return false;
1009 }
1010
1011 return true;
1012}
1013
1014static bool hns3_can_use_tx_sgl(struct hns3_enet_ring *ring,
1015 struct sk_buff *skb,
1016 u32 space)
1017{
1018 if (skb->len <= ring->tx_copybreak || !tx_sgl ||
1019 (!skb_has_frag_list(skb) &&
1020 skb_shinfo(skb)->nr_frags < tx_sgl))
1021 return false;
1022
1023 if (space < HNS3_MAX_SGL_SIZE) {
1024 u64_stats_update_begin(&ring->syncp);
1025 ring->stats.tx_spare_full++;
1026 u64_stats_update_end(&ring->syncp);
1027 return false;
1028 }
1029
1030 return true;
1031}
1032
1033static void hns3_init_tx_spare_buffer(struct hns3_enet_ring *ring)
1034{
1035 struct hns3_tx_spare *tx_spare;
1036 struct page *page;
1037 u32 alloc_size;
1038 dma_addr_t dma;
1039 int order;
1040
1041 alloc_size = ring->tqp->handle->kinfo.tx_spare_buf_size;
1042 if (!alloc_size)
1043 return;
1044
1045 order = get_order(alloc_size);
1046 tx_spare = devm_kzalloc(ring_to_dev(ring), sizeof(*tx_spare),
1047 GFP_KERNEL);
1048 if (!tx_spare) {
1049 /* The driver still work without the tx spare buffer */
1050 dev_warn(ring_to_dev(ring), "failed to allocate hns3_tx_spare\n");
1051 return;
1052 }
1053
1054 page = alloc_pages_node(dev_to_node(ring_to_dev(ring)),
1055 GFP_KERNEL, order);
1056 if (!page) {
1057 dev_warn(ring_to_dev(ring), "failed to allocate tx spare pages\n");
1058 devm_kfree(ring_to_dev(ring), tx_spare);
1059 return;
1060 }
1061
1062 dma = dma_map_page(ring_to_dev(ring), page, 0,
1063 PAGE_SIZE << order, DMA_TO_DEVICE);
1064 if (dma_mapping_error(ring_to_dev(ring), dma)) {
1065 dev_warn(ring_to_dev(ring), "failed to map pages for tx spare\n");
1066 put_page(page);
1067 devm_kfree(ring_to_dev(ring), tx_spare);
1068 return;
1069 }
1070
1071 tx_spare->dma = dma;
1072 tx_spare->buf = page_address(page);
1073 tx_spare->len = PAGE_SIZE << order;
1074 ring->tx_spare = tx_spare;
1075}
1076
1077/* Use hns3_tx_spare_space() to make sure there is enough buffer
1078 * before calling below function to allocate tx buffer.
1079 */
1080static void *hns3_tx_spare_alloc(struct hns3_enet_ring *ring,
1081 unsigned int size, dma_addr_t *dma,
1082 u32 *cb_len)
1083{
1084 struct hns3_tx_spare *tx_spare = ring->tx_spare;
1085 u32 ntu = tx_spare->next_to_use;
1086
1087 size = ALIGN(size, dma_get_cache_alignment());
1088 *cb_len = size;
1089
1090 /* Tx spare buffer wraps back here because the end of
1091 * freed tx buffer is not enough.
1092 */
1093 if (ntu + size > tx_spare->len) {
1094 *cb_len += (tx_spare->len - ntu);
1095 ntu = 0;
1096 }
1097
1098 tx_spare->next_to_use = ntu + size;
1099 if (tx_spare->next_to_use == tx_spare->len)
1100 tx_spare->next_to_use = 0;
1101
1102 *dma = tx_spare->dma + ntu;
1103
1104 return tx_spare->buf + ntu;
1105}
1106
1107static void hns3_tx_spare_rollback(struct hns3_enet_ring *ring, u32 len)
1108{
1109 struct hns3_tx_spare *tx_spare = ring->tx_spare;
1110
1111 if (len > tx_spare->next_to_use) {
1112 len -= tx_spare->next_to_use;
1113 tx_spare->next_to_use = tx_spare->len - len;
1114 } else {
1115 tx_spare->next_to_use -= len;
1116 }
1117}
1118
1119static void hns3_tx_spare_reclaim_cb(struct hns3_enet_ring *ring,
1120 struct hns3_desc_cb *cb)
1121{
1122 struct hns3_tx_spare *tx_spare = ring->tx_spare;
1123 u32 ntc = tx_spare->next_to_clean;
1124 u32 len = cb->length;
1125
1126 tx_spare->next_to_clean += len;
1127
1128 if (tx_spare->next_to_clean >= tx_spare->len) {
1129 tx_spare->next_to_clean -= tx_spare->len;
1130
1131 if (tx_spare->next_to_clean) {
1132 ntc = 0;
1133 len = tx_spare->next_to_clean;
1134 }
1135 }
1136
1137 /* This tx spare buffer is only really reclaimed after calling
1138 * hns3_tx_spare_update(), so it is still safe to use the info in
1139 * the tx buffer to do the dma sync or sg unmapping after
1140 * tx_spare->next_to_clean is moved forword.
1141 */
1142 if (cb->type & (DESC_TYPE_BOUNCE_HEAD | DESC_TYPE_BOUNCE_ALL)) {
1143 dma_addr_t dma = tx_spare->dma + ntc;
1144
1145 dma_sync_single_for_cpu(ring_to_dev(ring), dma, len,
1146 DMA_TO_DEVICE);
1147 } else {
1148 struct sg_table *sgt = tx_spare->buf + ntc;
1149
1150 dma_unmap_sg(ring_to_dev(ring), sgt->sgl, sgt->orig_nents,
1151 DMA_TO_DEVICE);
1152 }
1153}
1154
1155static int hns3_set_tso(struct sk_buff *skb, u32 *paylen_fdop_ol4cs,
1156 u16 *mss, u32 *type_cs_vlan_tso, u32 *send_bytes)
1157{
1158 u32 l4_offset, hdr_len;
1159 union l3_hdr_info l3;
1160 union l4_hdr_info l4;
1161 u32 l4_paylen;
1162 int ret;
1163
1164 if (!skb_is_gso(skb))
1165 return 0;
1166
1167 ret = skb_cow_head(skb, 0);
1168 if (unlikely(ret < 0))
1169 return ret;
1170
1171 l3.hdr = skb_network_header(skb);
1172 l4.hdr = skb_transport_header(skb);
1173
1174 /* Software should clear the IPv4's checksum field when tso is
1175 * needed.
1176 */
1177 if (l3.v4->version == 4)
1178 l3.v4->check = 0;
1179
1180 /* tunnel packet */
1181 if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
1182 SKB_GSO_GRE_CSUM |
1183 SKB_GSO_UDP_TUNNEL |
1184 SKB_GSO_UDP_TUNNEL_CSUM)) {
1185 /* reset l3&l4 pointers from outer to inner headers */
1186 l3.hdr = skb_inner_network_header(skb);
1187 l4.hdr = skb_inner_transport_header(skb);
1188
1189 /* Software should clear the IPv4's checksum field when
1190 * tso is needed.
1191 */
1192 if (l3.v4->version == 4)
1193 l3.v4->check = 0;
1194 }
1195
1196 /* normal or tunnel packet */
1197 l4_offset = l4.hdr - skb->data;
1198
1199 /* remove payload length from inner pseudo checksum when tso */
1200 l4_paylen = skb->len - l4_offset;
1201
1202 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) {
1203 hdr_len = sizeof(*l4.udp) + l4_offset;
1204 csum_replace_by_diff(&l4.udp->check,
1205 (__force __wsum)htonl(l4_paylen));
1206 } else {
1207 hdr_len = (l4.tcp->doff << 2) + l4_offset;
1208 csum_replace_by_diff(&l4.tcp->check,
1209 (__force __wsum)htonl(l4_paylen));
1210 }
1211
1212 *send_bytes = (skb_shinfo(skb)->gso_segs - 1) * hdr_len + skb->len;
1213
1214 /* find the txbd field values */
1215 *paylen_fdop_ol4cs = skb->len - hdr_len;
1216 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_TSO_B, 1);
1217
1218 /* offload outer UDP header checksum */
1219 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)
1220 hns3_set_field(*paylen_fdop_ol4cs, HNS3_TXD_OL4CS_B, 1);
1221
1222 /* get MSS for TSO */
1223 *mss = skb_shinfo(skb)->gso_size;
1224
1225 trace_hns3_tso(skb);
1226
1227 return 0;
1228}
1229
1230static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
1231 u8 *il4_proto)
1232{
1233 union l3_hdr_info l3;
1234 unsigned char *l4_hdr;
1235 unsigned char *exthdr;
1236 u8 l4_proto_tmp;
1237 __be16 frag_off;
1238
1239 /* find outer header point */
1240 l3.hdr = skb_network_header(skb);
1241 l4_hdr = skb_transport_header(skb);
1242
1243 if (skb->protocol == htons(ETH_P_IPV6)) {
1244 exthdr = l3.hdr + sizeof(*l3.v6);
1245 l4_proto_tmp = l3.v6->nexthdr;
1246 if (l4_hdr != exthdr)
1247 ipv6_skip_exthdr(skb, exthdr - skb->data,
1248 &l4_proto_tmp, &frag_off);
1249 } else if (skb->protocol == htons(ETH_P_IP)) {
1250 l4_proto_tmp = l3.v4->protocol;
1251 } else {
1252 return -EINVAL;
1253 }
1254
1255 *ol4_proto = l4_proto_tmp;
1256
1257 /* tunnel packet */
1258 if (!skb->encapsulation) {
1259 *il4_proto = 0;
1260 return 0;
1261 }
1262
1263 /* find inner header point */
1264 l3.hdr = skb_inner_network_header(skb);
1265 l4_hdr = skb_inner_transport_header(skb);
1266
1267 if (l3.v6->version == 6) {
1268 exthdr = l3.hdr + sizeof(*l3.v6);
1269 l4_proto_tmp = l3.v6->nexthdr;
1270 if (l4_hdr != exthdr)
1271 ipv6_skip_exthdr(skb, exthdr - skb->data,
1272 &l4_proto_tmp, &frag_off);
1273 } else if (l3.v4->version == 4) {
1274 l4_proto_tmp = l3.v4->protocol;
1275 }
1276
1277 *il4_proto = l4_proto_tmp;
1278
1279 return 0;
1280}
1281
1282/* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
1283 * and it is udp packet, which has a dest port as the IANA assigned.
1284 * the hardware is expected to do the checksum offload, but the
1285 * hardware will not do the checksum offload when udp dest port is
1286 * 4789, 4790 or 6081.
1287 */
1288static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
1289{
1290 struct hns3_nic_priv *priv = netdev_priv(skb->dev);
1291 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
1292 union l4_hdr_info l4;
1293
1294 /* device version above V3(include V3), the hardware can
1295 * do this checksum offload.
1296 */
1297 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
1298 return false;
1299
1300 l4.hdr = skb_transport_header(skb);
1301
1302 if (!(!skb->encapsulation &&
1303 (l4.udp->dest == htons(IANA_VXLAN_UDP_PORT) ||
1304 l4.udp->dest == htons(GENEVE_UDP_PORT) ||
1305 l4.udp->dest == htons(4790))))
1306 return false;
1307
1308 return true;
1309}
1310
1311static void hns3_set_outer_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
1312 u32 *ol_type_vlan_len_msec)
1313{
1314 u32 l2_len, l3_len, l4_len;
1315 unsigned char *il2_hdr;
1316 union l3_hdr_info l3;
1317 union l4_hdr_info l4;
1318
1319 l3.hdr = skb_network_header(skb);
1320 l4.hdr = skb_transport_header(skb);
1321
1322 /* compute OL2 header size, defined in 2 Bytes */
1323 l2_len = l3.hdr - skb->data;
1324 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L2LEN_S, l2_len >> 1);
1325
1326 /* compute OL3 header size, defined in 4 Bytes */
1327 l3_len = l4.hdr - l3.hdr;
1328 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_S, l3_len >> 2);
1329
1330 il2_hdr = skb_inner_mac_header(skb);
1331 /* compute OL4 header size, defined in 4 Bytes */
1332 l4_len = il2_hdr - l4.hdr;
1333 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L4LEN_S, l4_len >> 2);
1334
1335 /* define outer network header type */
1336 if (skb->protocol == htons(ETH_P_IP)) {
1337 if (skb_is_gso(skb))
1338 hns3_set_field(*ol_type_vlan_len_msec,
1339 HNS3_TXD_OL3T_S,
1340 HNS3_OL3T_IPV4_CSUM);
1341 else
1342 hns3_set_field(*ol_type_vlan_len_msec,
1343 HNS3_TXD_OL3T_S,
1344 HNS3_OL3T_IPV4_NO_CSUM);
1345 } else if (skb->protocol == htons(ETH_P_IPV6)) {
1346 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_S,
1347 HNS3_OL3T_IPV6);
1348 }
1349
1350 if (ol4_proto == IPPROTO_UDP)
1351 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
1352 HNS3_TUN_MAC_IN_UDP);
1353 else if (ol4_proto == IPPROTO_GRE)
1354 hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
1355 HNS3_TUN_NVGRE);
1356}
1357
1358static int hns3_set_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
1359 u8 il4_proto, u32 *type_cs_vlan_tso,
1360 u32 *ol_type_vlan_len_msec)
1361{
1362 unsigned char *l2_hdr = skb->data;
1363 u32 l4_proto = ol4_proto;
1364 union l4_hdr_info l4;
1365 union l3_hdr_info l3;
1366 u32 l2_len, l3_len;
1367
1368 l4.hdr = skb_transport_header(skb);
1369 l3.hdr = skb_network_header(skb);
1370
1371 /* handle encapsulation skb */
1372 if (skb->encapsulation) {
1373 /* If this is a not UDP/GRE encapsulation skb */
1374 if (!(ol4_proto == IPPROTO_UDP || ol4_proto == IPPROTO_GRE)) {
1375 /* drop the skb tunnel packet if hardware don't support,
1376 * because hardware can't calculate csum when TSO.
1377 */
1378 if (skb_is_gso(skb))
1379 return -EDOM;
1380
1381 /* the stack computes the IP header already,
1382 * driver calculate l4 checksum when not TSO.
1383 */
1384 return skb_checksum_help(skb);
1385 }
1386
1387 hns3_set_outer_l2l3l4(skb, ol4_proto, ol_type_vlan_len_msec);
1388
1389 /* switch to inner header */
1390 l2_hdr = skb_inner_mac_header(skb);
1391 l3.hdr = skb_inner_network_header(skb);
1392 l4.hdr = skb_inner_transport_header(skb);
1393 l4_proto = il4_proto;
1394 }
1395
1396 if (l3.v4->version == 4) {
1397 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
1398 HNS3_L3T_IPV4);
1399
1400 /* the stack computes the IP header already, the only time we
1401 * need the hardware to recompute it is in the case of TSO.
1402 */
1403 if (skb_is_gso(skb))
1404 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3CS_B, 1);
1405 } else if (l3.v6->version == 6) {
1406 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
1407 HNS3_L3T_IPV6);
1408 }
1409
1410 /* compute inner(/normal) L2 header size, defined in 2 Bytes */
1411 l2_len = l3.hdr - l2_hdr;
1412 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S, l2_len >> 1);
1413
1414 /* compute inner(/normal) L3 header size, defined in 4 Bytes */
1415 l3_len = l4.hdr - l3.hdr;
1416 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_S, l3_len >> 2);
1417
1418 /* compute inner(/normal) L4 header size, defined in 4 Bytes */
1419 switch (l4_proto) {
1420 case IPPROTO_TCP:
1421 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1422 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1423 HNS3_L4T_TCP);
1424 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1425 l4.tcp->doff);
1426 break;
1427 case IPPROTO_UDP:
1428 if (hns3_tunnel_csum_bug(skb)) {
1429 int ret = skb_put_padto(skb, HNS3_MIN_TUN_PKT_LEN);
1430
1431 return ret ? ret : skb_checksum_help(skb);
1432 }
1433
1434 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1435 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1436 HNS3_L4T_UDP);
1437 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1438 (sizeof(struct udphdr) >> 2));
1439 break;
1440 case IPPROTO_SCTP:
1441 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1442 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1443 HNS3_L4T_SCTP);
1444 hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1445 (sizeof(struct sctphdr) >> 2));
1446 break;
1447 default:
1448 /* drop the skb tunnel packet if hardware don't support,
1449 * because hardware can't calculate csum when TSO.
1450 */
1451 if (skb_is_gso(skb))
1452 return -EDOM;
1453
1454 /* the stack computes the IP header already,
1455 * driver calculate l4 checksum when not TSO.
1456 */
1457 return skb_checksum_help(skb);
1458 }
1459
1460 return 0;
1461}
1462
1463static int hns3_handle_vtags(struct hns3_enet_ring *tx_ring,
1464 struct sk_buff *skb)
1465{
1466 struct hnae3_handle *handle = tx_ring->tqp->handle;
1467 struct hnae3_ae_dev *ae_dev;
1468 struct vlan_ethhdr *vhdr;
1469 int rc;
1470
1471 if (!(skb->protocol == htons(ETH_P_8021Q) ||
1472 skb_vlan_tag_present(skb)))
1473 return 0;
1474
1475 /* For HW limitation on HNAE3_DEVICE_VERSION_V2, if port based insert
1476 * VLAN enabled, only one VLAN header is allowed in skb, otherwise it
1477 * will cause RAS error.
1478 */
1479 ae_dev = pci_get_drvdata(handle->pdev);
1480 if (unlikely(skb_vlan_tagged_multi(skb) &&
1481 ae_dev->dev_version <= HNAE3_DEVICE_VERSION_V2 &&
1482 handle->port_base_vlan_state ==
1483 HNAE3_PORT_BASE_VLAN_ENABLE))
1484 return -EINVAL;
1485
1486 if (skb->protocol == htons(ETH_P_8021Q) &&
1487 !(handle->kinfo.netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
1488 /* When HW VLAN acceleration is turned off, and the stack
1489 * sets the protocol to 802.1q, the driver just need to
1490 * set the protocol to the encapsulated ethertype.
1491 */
1492 skb->protocol = vlan_get_protocol(skb);
1493 return 0;
1494 }
1495
1496 if (skb_vlan_tag_present(skb)) {
1497 /* Based on hw strategy, use out_vtag in two layer tag case,
1498 * and use inner_vtag in one tag case.
1499 */
1500 if (skb->protocol == htons(ETH_P_8021Q) &&
1501 handle->port_base_vlan_state ==
1502 HNAE3_PORT_BASE_VLAN_DISABLE)
1503 rc = HNS3_OUTER_VLAN_TAG;
1504 else
1505 rc = HNS3_INNER_VLAN_TAG;
1506
1507 skb->protocol = vlan_get_protocol(skb);
1508 return rc;
1509 }
1510
1511 rc = skb_cow_head(skb, 0);
1512 if (unlikely(rc < 0))
1513 return rc;
1514
1515 vhdr = (struct vlan_ethhdr *)skb->data;
1516 vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority << VLAN_PRIO_SHIFT)
1517 & VLAN_PRIO_MASK);
1518
1519 skb->protocol = vlan_get_protocol(skb);
1520 return 0;
1521}
1522
1523/* check if the hardware is capable of checksum offloading */
1524static bool hns3_check_hw_tx_csum(struct sk_buff *skb)
1525{
1526 struct hns3_nic_priv *priv = netdev_priv(skb->dev);
1527
1528 /* Kindly note, due to backward compatibility of the TX descriptor,
1529 * HW checksum of the non-IP packets and GSO packets is handled at
1530 * different place in the following code
1531 */
1532 if (skb_csum_is_sctp(skb) || skb_is_gso(skb) ||
1533 !test_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state))
1534 return false;
1535
1536 return true;
1537}
1538
1539static int hns3_fill_skb_desc(struct hns3_enet_ring *ring,
1540 struct sk_buff *skb, struct hns3_desc *desc,
1541 struct hns3_desc_cb *desc_cb)
1542{
1543 u32 ol_type_vlan_len_msec = 0;
1544 u32 paylen_ol4cs = skb->len;
1545 u32 type_cs_vlan_tso = 0;
1546 u16 mss_hw_csum = 0;
1547 u16 inner_vtag = 0;
1548 u16 out_vtag = 0;
1549 int ret;
1550
1551 ret = hns3_handle_vtags(ring, skb);
1552 if (unlikely(ret < 0)) {
1553 u64_stats_update_begin(&ring->syncp);
1554 ring->stats.tx_vlan_err++;
1555 u64_stats_update_end(&ring->syncp);
1556 return ret;
1557 } else if (ret == HNS3_INNER_VLAN_TAG) {
1558 inner_vtag = skb_vlan_tag_get(skb);
1559 inner_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1560 VLAN_PRIO_MASK;
1561 hns3_set_field(type_cs_vlan_tso, HNS3_TXD_VLAN_B, 1);
1562 } else if (ret == HNS3_OUTER_VLAN_TAG) {
1563 out_vtag = skb_vlan_tag_get(skb);
1564 out_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1565 VLAN_PRIO_MASK;
1566 hns3_set_field(ol_type_vlan_len_msec, HNS3_TXD_OVLAN_B,
1567 1);
1568 }
1569
1570 desc_cb->send_bytes = skb->len;
1571
1572 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1573 u8 ol4_proto, il4_proto;
1574
1575 if (hns3_check_hw_tx_csum(skb)) {
1576 /* set checksum start and offset, defined in 2 Bytes */
1577 hns3_set_field(type_cs_vlan_tso, HNS3_TXD_CSUM_START_S,
1578 skb_checksum_start_offset(skb) >> 1);
1579 hns3_set_field(ol_type_vlan_len_msec,
1580 HNS3_TXD_CSUM_OFFSET_S,
1581 skb->csum_offset >> 1);
1582 mss_hw_csum |= BIT(HNS3_TXD_HW_CS_B);
1583 goto out_hw_tx_csum;
1584 }
1585
1586 skb_reset_mac_len(skb);
1587
1588 ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
1589 if (unlikely(ret < 0)) {
1590 u64_stats_update_begin(&ring->syncp);
1591 ring->stats.tx_l4_proto_err++;
1592 u64_stats_update_end(&ring->syncp);
1593 return ret;
1594 }
1595
1596 ret = hns3_set_l2l3l4(skb, ol4_proto, il4_proto,
1597 &type_cs_vlan_tso,
1598 &ol_type_vlan_len_msec);
1599 if (unlikely(ret < 0)) {
1600 u64_stats_update_begin(&ring->syncp);
1601 ring->stats.tx_l2l3l4_err++;
1602 u64_stats_update_end(&ring->syncp);
1603 return ret;
1604 }
1605
1606 ret = hns3_set_tso(skb, &paylen_ol4cs, &mss_hw_csum,
1607 &type_cs_vlan_tso, &desc_cb->send_bytes);
1608 if (unlikely(ret < 0)) {
1609 u64_stats_update_begin(&ring->syncp);
1610 ring->stats.tx_tso_err++;
1611 u64_stats_update_end(&ring->syncp);
1612 return ret;
1613 }
1614 }
1615
1616out_hw_tx_csum:
1617 /* Set txbd */
1618 desc->tx.ol_type_vlan_len_msec =
1619 cpu_to_le32(ol_type_vlan_len_msec);
1620 desc->tx.type_cs_vlan_tso_len = cpu_to_le32(type_cs_vlan_tso);
1621 desc->tx.paylen_ol4cs = cpu_to_le32(paylen_ol4cs);
1622 desc->tx.mss_hw_csum = cpu_to_le16(mss_hw_csum);
1623 desc->tx.vlan_tag = cpu_to_le16(inner_vtag);
1624 desc->tx.outer_vlan_tag = cpu_to_le16(out_vtag);
1625
1626 return 0;
1627}
1628
1629static int hns3_fill_desc(struct hns3_enet_ring *ring, dma_addr_t dma,
1630 unsigned int size)
1631{
1632#define HNS3_LIKELY_BD_NUM 1
1633
1634 struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1635 unsigned int frag_buf_num;
1636 int k, sizeoflast;
1637
1638 if (likely(size <= HNS3_MAX_BD_SIZE)) {
1639 desc->addr = cpu_to_le64(dma);
1640 desc->tx.send_size = cpu_to_le16(size);
1641 desc->tx.bdtp_fe_sc_vld_ra_ri =
1642 cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1643
1644 trace_hns3_tx_desc(ring, ring->next_to_use);
1645 ring_ptr_move_fw(ring, next_to_use);
1646 return HNS3_LIKELY_BD_NUM;
1647 }
1648
1649 frag_buf_num = hns3_tx_bd_count(size);
1650 sizeoflast = size % HNS3_MAX_BD_SIZE;
1651 sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1652
1653 /* When frag size is bigger than hardware limit, split this frag */
1654 for (k = 0; k < frag_buf_num; k++) {
1655 /* now, fill the descriptor */
1656 desc->addr = cpu_to_le64(dma + HNS3_MAX_BD_SIZE * k);
1657 desc->tx.send_size = cpu_to_le16((k == frag_buf_num - 1) ?
1658 (u16)sizeoflast : (u16)HNS3_MAX_BD_SIZE);
1659 desc->tx.bdtp_fe_sc_vld_ra_ri =
1660 cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1661
1662 trace_hns3_tx_desc(ring, ring->next_to_use);
1663 /* move ring pointer to next */
1664 ring_ptr_move_fw(ring, next_to_use);
1665
1666 desc = &ring->desc[ring->next_to_use];
1667 }
1668
1669 return frag_buf_num;
1670}
1671
1672static int hns3_map_and_fill_desc(struct hns3_enet_ring *ring, void *priv,
1673 unsigned int type)
1674{
1675 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1676 struct device *dev = ring_to_dev(ring);
1677 unsigned int size;
1678 dma_addr_t dma;
1679
1680 if (type & (DESC_TYPE_FRAGLIST_SKB | DESC_TYPE_SKB)) {
1681 struct sk_buff *skb = (struct sk_buff *)priv;
1682
1683 size = skb_headlen(skb);
1684 if (!size)
1685 return 0;
1686
1687 dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1688 } else if (type & DESC_TYPE_BOUNCE_HEAD) {
1689 /* Head data has been filled in hns3_handle_tx_bounce(),
1690 * just return 0 here.
1691 */
1692 return 0;
1693 } else {
1694 skb_frag_t *frag = (skb_frag_t *)priv;
1695
1696 size = skb_frag_size(frag);
1697 if (!size)
1698 return 0;
1699
1700 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1701 }
1702
1703 if (unlikely(dma_mapping_error(dev, dma))) {
1704 u64_stats_update_begin(&ring->syncp);
1705 ring->stats.sw_err_cnt++;
1706 u64_stats_update_end(&ring->syncp);
1707 return -ENOMEM;
1708 }
1709
1710 desc_cb->priv = priv;
1711 desc_cb->length = size;
1712 desc_cb->dma = dma;
1713 desc_cb->type = type;
1714
1715 return hns3_fill_desc(ring, dma, size);
1716}
1717
1718static unsigned int hns3_skb_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1719 unsigned int bd_num)
1720{
1721 unsigned int size;
1722 int i;
1723
1724 size = skb_headlen(skb);
1725 while (size > HNS3_MAX_BD_SIZE) {
1726 bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1727 size -= HNS3_MAX_BD_SIZE;
1728
1729 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1730 return bd_num;
1731 }
1732
1733 if (size) {
1734 bd_size[bd_num++] = size;
1735 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1736 return bd_num;
1737 }
1738
1739 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1740 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1741 size = skb_frag_size(frag);
1742 if (!size)
1743 continue;
1744
1745 while (size > HNS3_MAX_BD_SIZE) {
1746 bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1747 size -= HNS3_MAX_BD_SIZE;
1748
1749 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1750 return bd_num;
1751 }
1752
1753 bd_size[bd_num++] = size;
1754 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1755 return bd_num;
1756 }
1757
1758 return bd_num;
1759}
1760
1761static unsigned int hns3_tx_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1762 u8 max_non_tso_bd_num, unsigned int bd_num,
1763 unsigned int recursion_level)
1764{
1765#define HNS3_MAX_RECURSION_LEVEL 24
1766
1767 struct sk_buff *frag_skb;
1768
1769 /* If the total len is within the max bd limit */
1770 if (likely(skb->len <= HNS3_MAX_BD_SIZE && !recursion_level &&
1771 !skb_has_frag_list(skb) &&
1772 skb_shinfo(skb)->nr_frags < max_non_tso_bd_num))
1773 return skb_shinfo(skb)->nr_frags + 1U;
1774
1775 if (unlikely(recursion_level >= HNS3_MAX_RECURSION_LEVEL))
1776 return UINT_MAX;
1777
1778 bd_num = hns3_skb_bd_num(skb, bd_size, bd_num);
1779 if (!skb_has_frag_list(skb) || bd_num > HNS3_MAX_TSO_BD_NUM)
1780 return bd_num;
1781
1782 skb_walk_frags(skb, frag_skb) {
1783 bd_num = hns3_tx_bd_num(frag_skb, bd_size, max_non_tso_bd_num,
1784 bd_num, recursion_level + 1);
1785 if (bd_num > HNS3_MAX_TSO_BD_NUM)
1786 return bd_num;
1787 }
1788
1789 return bd_num;
1790}
1791
1792static unsigned int hns3_gso_hdr_len(struct sk_buff *skb)
1793{
1794 if (!skb->encapsulation)
1795 return skb_transport_offset(skb) + tcp_hdrlen(skb);
1796
1797 return skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb);
1798}
1799
1800/* HW need every continuous max_non_tso_bd_num buffer data to be larger
1801 * than MSS, we simplify it by ensuring skb_headlen + the first continuous
1802 * max_non_tso_bd_num - 1 frags to be larger than gso header len + mss,
1803 * and the remaining continuous max_non_tso_bd_num - 1 frags to be larger
1804 * than MSS except the last max_non_tso_bd_num - 1 frags.
1805 */
1806static bool hns3_skb_need_linearized(struct sk_buff *skb, unsigned int *bd_size,
1807 unsigned int bd_num, u8 max_non_tso_bd_num)
1808{
1809 unsigned int tot_len = 0;
1810 int i;
1811
1812 for (i = 0; i < max_non_tso_bd_num - 1U; i++)
1813 tot_len += bd_size[i];
1814
1815 /* ensure the first max_non_tso_bd_num frags is greater than
1816 * mss + header
1817 */
1818 if (tot_len + bd_size[max_non_tso_bd_num - 1U] <
1819 skb_shinfo(skb)->gso_size + hns3_gso_hdr_len(skb))
1820 return true;
1821
1822 /* ensure every continuous max_non_tso_bd_num - 1 buffer is greater
1823 * than mss except the last one.
1824 */
1825 for (i = 0; i < bd_num - max_non_tso_bd_num; i++) {
1826 tot_len -= bd_size[i];
1827 tot_len += bd_size[i + max_non_tso_bd_num - 1U];
1828
1829 if (tot_len < skb_shinfo(skb)->gso_size)
1830 return true;
1831 }
1832
1833 return false;
1834}
1835
1836void hns3_shinfo_pack(struct skb_shared_info *shinfo, __u32 *size)
1837{
1838 int i;
1839
1840 for (i = 0; i < MAX_SKB_FRAGS; i++)
1841 size[i] = skb_frag_size(&shinfo->frags[i]);
1842}
1843
1844static int hns3_skb_linearize(struct hns3_enet_ring *ring,
1845 struct sk_buff *skb,
1846 unsigned int bd_num)
1847{
1848 /* 'bd_num == UINT_MAX' means the skb' fraglist has a
1849 * recursion level of over HNS3_MAX_RECURSION_LEVEL.
1850 */
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);
1855 return -ENOMEM;
1856 }
1857
1858 /* The skb->len has exceeded the hw limitation, linearization
1859 * will not help.
1860 */
1861 if (skb->len > HNS3_MAX_TSO_SIZE ||
1862 (!skb_is_gso(skb) && skb->len > HNS3_MAX_NON_TSO_SIZE)) {
1863 u64_stats_update_begin(&ring->syncp);
1864 ring->stats.hw_limitation++;
1865 u64_stats_update_end(&ring->syncp);
1866 return -ENOMEM;
1867 }
1868
1869 if (__skb_linearize(skb)) {
1870 u64_stats_update_begin(&ring->syncp);
1871 ring->stats.sw_err_cnt++;
1872 u64_stats_update_end(&ring->syncp);
1873 return -ENOMEM;
1874 }
1875
1876 return 0;
1877}
1878
1879static int hns3_nic_maybe_stop_tx(struct hns3_enet_ring *ring,
1880 struct net_device *netdev,
1881 struct sk_buff *skb)
1882{
1883 struct hns3_nic_priv *priv = netdev_priv(netdev);
1884 u8 max_non_tso_bd_num = priv->max_non_tso_bd_num;
1885 unsigned int bd_size[HNS3_MAX_TSO_BD_NUM + 1U];
1886 unsigned int bd_num;
1887
1888 bd_num = hns3_tx_bd_num(skb, bd_size, max_non_tso_bd_num, 0, 0);
1889 if (unlikely(bd_num > max_non_tso_bd_num)) {
1890 if (bd_num <= HNS3_MAX_TSO_BD_NUM && skb_is_gso(skb) &&
1891 !hns3_skb_need_linearized(skb, bd_size, bd_num,
1892 max_non_tso_bd_num)) {
1893 trace_hns3_over_max_bd(skb);
1894 goto out;
1895 }
1896
1897 if (hns3_skb_linearize(ring, skb, bd_num))
1898 return -ENOMEM;
1899
1900 bd_num = hns3_tx_bd_count(skb->len);
1901
1902 u64_stats_update_begin(&ring->syncp);
1903 ring->stats.tx_copy++;
1904 u64_stats_update_end(&ring->syncp);
1905 }
1906
1907out:
1908 if (likely(ring_space(ring) >= bd_num))
1909 return bd_num;
1910
1911 netif_stop_subqueue(netdev, ring->queue_index);
1912 smp_mb(); /* Memory barrier before checking ring_space */
1913
1914 /* Start queue in case hns3_clean_tx_ring has just made room
1915 * available and has not seen the queue stopped state performed
1916 * by netif_stop_subqueue above.
1917 */
1918 if (ring_space(ring) >= bd_num && netif_carrier_ok(netdev) &&
1919 !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
1920 netif_start_subqueue(netdev, ring->queue_index);
1921 return bd_num;
1922 }
1923
1924 u64_stats_update_begin(&ring->syncp);
1925 ring->stats.tx_busy++;
1926 u64_stats_update_end(&ring->syncp);
1927
1928 return -EBUSY;
1929}
1930
1931static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
1932{
1933 struct device *dev = ring_to_dev(ring);
1934 unsigned int i;
1935
1936 for (i = 0; i < ring->desc_num; i++) {
1937 struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1938 struct hns3_desc_cb *desc_cb;
1939
1940 memset(desc, 0, sizeof(*desc));
1941
1942 /* check if this is where we started */
1943 if (ring->next_to_use == next_to_use_orig)
1944 break;
1945
1946 /* rollback one */
1947 ring_ptr_move_bw(ring, next_to_use);
1948
1949 desc_cb = &ring->desc_cb[ring->next_to_use];
1950
1951 if (!desc_cb->dma)
1952 continue;
1953
1954 /* unmap the descriptor dma address */
1955 if (desc_cb->type & (DESC_TYPE_SKB | DESC_TYPE_FRAGLIST_SKB))
1956 dma_unmap_single(dev, desc_cb->dma, desc_cb->length,
1957 DMA_TO_DEVICE);
1958 else if (desc_cb->type &
1959 (DESC_TYPE_BOUNCE_HEAD | DESC_TYPE_BOUNCE_ALL))
1960 hns3_tx_spare_rollback(ring, desc_cb->length);
1961 else if (desc_cb->length)
1962 dma_unmap_page(dev, desc_cb->dma, desc_cb->length,
1963 DMA_TO_DEVICE);
1964
1965 desc_cb->length = 0;
1966 desc_cb->dma = 0;
1967 desc_cb->type = DESC_TYPE_UNKNOWN;
1968 }
1969}
1970
1971static int hns3_fill_skb_to_desc(struct hns3_enet_ring *ring,
1972 struct sk_buff *skb, unsigned int type)
1973{
1974 struct sk_buff *frag_skb;
1975 int i, ret, bd_num = 0;
1976
1977 ret = hns3_map_and_fill_desc(ring, skb, type);
1978 if (unlikely(ret < 0))
1979 return ret;
1980
1981 bd_num += ret;
1982
1983 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1984 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1985
1986 ret = hns3_map_and_fill_desc(ring, frag, DESC_TYPE_PAGE);
1987 if (unlikely(ret < 0))
1988 return ret;
1989
1990 bd_num += ret;
1991 }
1992
1993 skb_walk_frags(skb, frag_skb) {
1994 ret = hns3_fill_skb_to_desc(ring, frag_skb,
1995 DESC_TYPE_FRAGLIST_SKB);
1996 if (unlikely(ret < 0))
1997 return ret;
1998
1999 bd_num += ret;
2000 }
2001
2002 return bd_num;
2003}
2004
2005static void hns3_tx_doorbell(struct hns3_enet_ring *ring, int num,
2006 bool doorbell)
2007{
2008 ring->pending_buf += num;
2009
2010 if (!doorbell) {
2011 u64_stats_update_begin(&ring->syncp);
2012 ring->stats.tx_more++;
2013 u64_stats_update_end(&ring->syncp);
2014 return;
2015 }
2016
2017 if (!ring->pending_buf)
2018 return;
2019
2020 writel(ring->pending_buf,
2021 ring->tqp->io_base + HNS3_RING_TX_RING_TAIL_REG);
2022 ring->pending_buf = 0;
2023 WRITE_ONCE(ring->last_to_use, ring->next_to_use);
2024}
2025
2026static void hns3_tsyn(struct net_device *netdev, struct sk_buff *skb,
2027 struct hns3_desc *desc)
2028{
2029 struct hnae3_handle *h = hns3_get_handle(netdev);
2030
2031 if (!(h->ae_algo->ops->set_tx_hwts_info &&
2032 h->ae_algo->ops->set_tx_hwts_info(h, skb)))
2033 return;
2034
2035 desc->tx.bdtp_fe_sc_vld_ra_ri |= cpu_to_le16(BIT(HNS3_TXD_TSYN_B));
2036}
2037
2038static int hns3_handle_tx_bounce(struct hns3_enet_ring *ring,
2039 struct sk_buff *skb)
2040{
2041 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
2042 unsigned int type = DESC_TYPE_BOUNCE_HEAD;
2043 unsigned int size = skb_headlen(skb);
2044 dma_addr_t dma;
2045 int bd_num = 0;
2046 u32 cb_len;
2047 void *buf;
2048 int ret;
2049
2050 if (skb->len <= ring->tx_copybreak) {
2051 size = skb->len;
2052 type = DESC_TYPE_BOUNCE_ALL;
2053 }
2054
2055 /* hns3_can_use_tx_bounce() is called to ensure the below
2056 * function can always return the tx buffer.
2057 */
2058 buf = hns3_tx_spare_alloc(ring, size, &dma, &cb_len);
2059
2060 ret = skb_copy_bits(skb, 0, buf, size);
2061 if (unlikely(ret < 0)) {
2062 hns3_tx_spare_rollback(ring, cb_len);
2063 u64_stats_update_begin(&ring->syncp);
2064 ring->stats.copy_bits_err++;
2065 u64_stats_update_end(&ring->syncp);
2066 return ret;
2067 }
2068
2069 desc_cb->priv = skb;
2070 desc_cb->length = cb_len;
2071 desc_cb->dma = dma;
2072 desc_cb->type = type;
2073
2074 bd_num += hns3_fill_desc(ring, dma, size);
2075
2076 if (type == DESC_TYPE_BOUNCE_HEAD) {
2077 ret = hns3_fill_skb_to_desc(ring, skb,
2078 DESC_TYPE_BOUNCE_HEAD);
2079 if (unlikely(ret < 0))
2080 return ret;
2081
2082 bd_num += ret;
2083 }
2084
2085 dma_sync_single_for_device(ring_to_dev(ring), dma, size,
2086 DMA_TO_DEVICE);
2087
2088 u64_stats_update_begin(&ring->syncp);
2089 ring->stats.tx_bounce++;
2090 u64_stats_update_end(&ring->syncp);
2091 return bd_num;
2092}
2093
2094static int hns3_handle_tx_sgl(struct hns3_enet_ring *ring,
2095 struct sk_buff *skb)
2096{
2097 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
2098 u32 nfrag = skb_shinfo(skb)->nr_frags + 1;
2099 struct sg_table *sgt;
2100 int i, bd_num = 0;
2101 dma_addr_t dma;
2102 u32 cb_len;
2103 int nents;
2104
2105 if (skb_has_frag_list(skb))
2106 nfrag = HNS3_MAX_TSO_BD_NUM;
2107
2108 /* hns3_can_use_tx_sgl() is called to ensure the below
2109 * function can always return the tx buffer.
2110 */
2111 sgt = hns3_tx_spare_alloc(ring, HNS3_SGL_SIZE(nfrag),
2112 &dma, &cb_len);
2113
2114 /* scatterlist follows by the sg table */
2115 sgt->sgl = (struct scatterlist *)(sgt + 1);
2116 sg_init_table(sgt->sgl, nfrag);
2117 nents = skb_to_sgvec(skb, sgt->sgl, 0, skb->len);
2118 if (unlikely(nents < 0)) {
2119 hns3_tx_spare_rollback(ring, cb_len);
2120 u64_stats_update_begin(&ring->syncp);
2121 ring->stats.skb2sgl_err++;
2122 u64_stats_update_end(&ring->syncp);
2123 return -ENOMEM;
2124 }
2125
2126 sgt->orig_nents = nents;
2127 sgt->nents = dma_map_sg(ring_to_dev(ring), sgt->sgl, sgt->orig_nents,
2128 DMA_TO_DEVICE);
2129 if (unlikely(!sgt->nents)) {
2130 hns3_tx_spare_rollback(ring, cb_len);
2131 u64_stats_update_begin(&ring->syncp);
2132 ring->stats.map_sg_err++;
2133 u64_stats_update_end(&ring->syncp);
2134 return -ENOMEM;
2135 }
2136
2137 desc_cb->priv = skb;
2138 desc_cb->length = cb_len;
2139 desc_cb->dma = dma;
2140 desc_cb->type = DESC_TYPE_SGL_SKB;
2141
2142 for (i = 0; i < sgt->nents; i++)
2143 bd_num += hns3_fill_desc(ring, sg_dma_address(sgt->sgl + i),
2144 sg_dma_len(sgt->sgl + i));
2145
2146 u64_stats_update_begin(&ring->syncp);
2147 ring->stats.tx_sgl++;
2148 u64_stats_update_end(&ring->syncp);
2149
2150 return bd_num;
2151}
2152
2153static int hns3_handle_desc_filling(struct hns3_enet_ring *ring,
2154 struct sk_buff *skb)
2155{
2156 u32 space;
2157
2158 if (!ring->tx_spare)
2159 goto out;
2160
2161 space = hns3_tx_spare_space(ring);
2162
2163 if (hns3_can_use_tx_sgl(ring, skb, space))
2164 return hns3_handle_tx_sgl(ring, skb);
2165
2166 if (hns3_can_use_tx_bounce(ring, skb, space))
2167 return hns3_handle_tx_bounce(ring, skb);
2168
2169out:
2170 return hns3_fill_skb_to_desc(ring, skb, DESC_TYPE_SKB);
2171}
2172
2173netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
2174{
2175 struct hns3_nic_priv *priv = netdev_priv(netdev);
2176 struct hns3_enet_ring *ring = &priv->ring[skb->queue_mapping];
2177 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
2178 struct netdev_queue *dev_queue;
2179 int pre_ntu, next_to_use_head;
2180 bool doorbell;
2181 int ret;
2182
2183 /* Hardware can only handle short frames above 32 bytes */
2184 if (skb_put_padto(skb, HNS3_MIN_TX_LEN)) {
2185 hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
2186
2187 u64_stats_update_begin(&ring->syncp);
2188 ring->stats.sw_err_cnt++;
2189 u64_stats_update_end(&ring->syncp);
2190
2191 return NETDEV_TX_OK;
2192 }
2193
2194 /* Prefetch the data used later */
2195 prefetch(skb->data);
2196
2197 ret = hns3_nic_maybe_stop_tx(ring, netdev, skb);
2198 if (unlikely(ret <= 0)) {
2199 if (ret == -EBUSY) {
2200 hns3_tx_doorbell(ring, 0, true);
2201 return NETDEV_TX_BUSY;
2202 }
2203
2204 hns3_rl_err(netdev, "xmit error: %d!\n", ret);
2205 goto out_err_tx_ok;
2206 }
2207
2208 next_to_use_head = ring->next_to_use;
2209
2210 ret = hns3_fill_skb_desc(ring, skb, &ring->desc[ring->next_to_use],
2211 desc_cb);
2212 if (unlikely(ret < 0))
2213 goto fill_err;
2214
2215 /* 'ret < 0' means filling error, 'ret == 0' means skb->len is
2216 * zero, which is unlikely, and 'ret > 0' means how many tx desc
2217 * need to be notified to the hw.
2218 */
2219 ret = hns3_handle_desc_filling(ring, skb);
2220 if (unlikely(ret <= 0))
2221 goto fill_err;
2222
2223 pre_ntu = ring->next_to_use ? (ring->next_to_use - 1) :
2224 (ring->desc_num - 1);
2225
2226 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP))
2227 hns3_tsyn(netdev, skb, &ring->desc[pre_ntu]);
2228
2229 ring->desc[pre_ntu].tx.bdtp_fe_sc_vld_ra_ri |=
2230 cpu_to_le16(BIT(HNS3_TXD_FE_B));
2231 trace_hns3_tx_desc(ring, pre_ntu);
2232
2233 skb_tx_timestamp(skb);
2234
2235 /* Complete translate all packets */
2236 dev_queue = netdev_get_tx_queue(netdev, ring->queue_index);
2237 doorbell = __netdev_tx_sent_queue(dev_queue, desc_cb->send_bytes,
2238 netdev_xmit_more());
2239 hns3_tx_doorbell(ring, ret, doorbell);
2240
2241 return NETDEV_TX_OK;
2242
2243fill_err:
2244 hns3_clear_desc(ring, next_to_use_head);
2245
2246out_err_tx_ok:
2247 dev_kfree_skb_any(skb);
2248 hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
2249 return NETDEV_TX_OK;
2250}
2251
2252static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
2253{
2254 char format_mac_addr_perm[HNAE3_FORMAT_MAC_ADDR_LEN];
2255 char format_mac_addr_sa[HNAE3_FORMAT_MAC_ADDR_LEN];
2256 struct hnae3_handle *h = hns3_get_handle(netdev);
2257 struct sockaddr *mac_addr = p;
2258 int ret;
2259
2260 if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
2261 return -EADDRNOTAVAIL;
2262
2263 if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
2264 hnae3_format_mac_addr(format_mac_addr_sa, mac_addr->sa_data);
2265 netdev_info(netdev, "already using mac address %s\n",
2266 format_mac_addr_sa);
2267 return 0;
2268 }
2269
2270 /* For VF device, if there is a perm_addr, then the user will not
2271 * be allowed to change the address.
2272 */
2273 if (!hns3_is_phys_func(h->pdev) &&
2274 !is_zero_ether_addr(netdev->perm_addr)) {
2275 hnae3_format_mac_addr(format_mac_addr_perm, netdev->perm_addr);
2276 hnae3_format_mac_addr(format_mac_addr_sa, mac_addr->sa_data);
2277 netdev_err(netdev, "has permanent MAC %s, user MAC %s not allow\n",
2278 format_mac_addr_perm, format_mac_addr_sa);
2279 return -EPERM;
2280 }
2281
2282 ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
2283 if (ret) {
2284 netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
2285 return ret;
2286 }
2287
2288 eth_hw_addr_set(netdev, mac_addr->sa_data);
2289
2290 return 0;
2291}
2292
2293static int hns3_nic_do_ioctl(struct net_device *netdev,
2294 struct ifreq *ifr, int cmd)
2295{
2296 struct hnae3_handle *h = hns3_get_handle(netdev);
2297
2298 if (!netif_running(netdev))
2299 return -EINVAL;
2300
2301 if (!h->ae_algo->ops->do_ioctl)
2302 return -EOPNOTSUPP;
2303
2304 return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
2305}
2306
2307static int hns3_nic_set_features(struct net_device *netdev,
2308 netdev_features_t features)
2309{
2310 netdev_features_t changed = netdev->features ^ features;
2311 struct hns3_nic_priv *priv = netdev_priv(netdev);
2312 struct hnae3_handle *h = priv->ae_handle;
2313 bool enable;
2314 int ret;
2315
2316 if (changed & (NETIF_F_GRO_HW) && h->ae_algo->ops->set_gro_en) {
2317 enable = !!(features & NETIF_F_GRO_HW);
2318 ret = h->ae_algo->ops->set_gro_en(h, enable);
2319 if (ret)
2320 return ret;
2321 }
2322
2323 if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
2324 h->ae_algo->ops->enable_hw_strip_rxvtag) {
2325 enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
2326 ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, enable);
2327 if (ret)
2328 return ret;
2329 }
2330
2331 if ((changed & NETIF_F_NTUPLE) && h->ae_algo->ops->enable_fd) {
2332 enable = !!(features & NETIF_F_NTUPLE);
2333 h->ae_algo->ops->enable_fd(h, enable);
2334 }
2335
2336 if ((netdev->features & NETIF_F_HW_TC) > (features & NETIF_F_HW_TC) &&
2337 h->ae_algo->ops->cls_flower_active(h)) {
2338 netdev_err(netdev,
2339 "there are offloaded TC filters active, cannot disable HW TC offload");
2340 return -EINVAL;
2341 }
2342
2343 if ((changed & NETIF_F_HW_VLAN_CTAG_FILTER) &&
2344 h->ae_algo->ops->enable_vlan_filter) {
2345 enable = !!(features & NETIF_F_HW_VLAN_CTAG_FILTER);
2346 ret = h->ae_algo->ops->enable_vlan_filter(h, enable);
2347 if (ret)
2348 return ret;
2349 }
2350
2351 netdev->features = features;
2352 return 0;
2353}
2354
2355static netdev_features_t hns3_features_check(struct sk_buff *skb,
2356 struct net_device *dev,
2357 netdev_features_t features)
2358{
2359#define HNS3_MAX_HDR_LEN 480U
2360#define HNS3_MAX_L4_HDR_LEN 60U
2361
2362 size_t len;
2363
2364 if (skb->ip_summed != CHECKSUM_PARTIAL)
2365 return features;
2366
2367 if (skb->encapsulation)
2368 len = skb_inner_transport_header(skb) - skb->data;
2369 else
2370 len = skb_transport_header(skb) - skb->data;
2371
2372 /* Assume L4 is 60 byte as TCP is the only protocol with a
2373 * a flexible value, and it's max len is 60 bytes.
2374 */
2375 len += HNS3_MAX_L4_HDR_LEN;
2376
2377 /* Hardware only supports checksum on the skb with a max header
2378 * len of 480 bytes.
2379 */
2380 if (len > HNS3_MAX_HDR_LEN)
2381 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2382
2383 return features;
2384}
2385
2386static void hns3_nic_get_stats64(struct net_device *netdev,
2387 struct rtnl_link_stats64 *stats)
2388{
2389 struct hns3_nic_priv *priv = netdev_priv(netdev);
2390 int queue_num = priv->ae_handle->kinfo.num_tqps;
2391 struct hnae3_handle *handle = priv->ae_handle;
2392 struct hns3_enet_ring *ring;
2393 u64 rx_length_errors = 0;
2394 u64 rx_crc_errors = 0;
2395 u64 rx_multicast = 0;
2396 unsigned int start;
2397 u64 tx_errors = 0;
2398 u64 rx_errors = 0;
2399 unsigned int idx;
2400 u64 tx_bytes = 0;
2401 u64 rx_bytes = 0;
2402 u64 tx_pkts = 0;
2403 u64 rx_pkts = 0;
2404 u64 tx_drop = 0;
2405 u64 rx_drop = 0;
2406
2407 if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
2408 return;
2409
2410 handle->ae_algo->ops->update_stats(handle, &netdev->stats);
2411
2412 for (idx = 0; idx < queue_num; idx++) {
2413 /* fetch the tx stats */
2414 ring = &priv->ring[idx];
2415 do {
2416 start = u64_stats_fetch_begin_irq(&ring->syncp);
2417 tx_bytes += ring->stats.tx_bytes;
2418 tx_pkts += ring->stats.tx_pkts;
2419 tx_drop += ring->stats.sw_err_cnt;
2420 tx_drop += ring->stats.tx_vlan_err;
2421 tx_drop += ring->stats.tx_l4_proto_err;
2422 tx_drop += ring->stats.tx_l2l3l4_err;
2423 tx_drop += ring->stats.tx_tso_err;
2424 tx_drop += ring->stats.over_max_recursion;
2425 tx_drop += ring->stats.hw_limitation;
2426 tx_drop += ring->stats.copy_bits_err;
2427 tx_drop += ring->stats.skb2sgl_err;
2428 tx_drop += ring->stats.map_sg_err;
2429 tx_errors += ring->stats.sw_err_cnt;
2430 tx_errors += ring->stats.tx_vlan_err;
2431 tx_errors += ring->stats.tx_l4_proto_err;
2432 tx_errors += ring->stats.tx_l2l3l4_err;
2433 tx_errors += ring->stats.tx_tso_err;
2434 tx_errors += ring->stats.over_max_recursion;
2435 tx_errors += ring->stats.hw_limitation;
2436 tx_errors += ring->stats.copy_bits_err;
2437 tx_errors += ring->stats.skb2sgl_err;
2438 tx_errors += ring->stats.map_sg_err;
2439 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
2440
2441 /* fetch the rx stats */
2442 ring = &priv->ring[idx + queue_num];
2443 do {
2444 start = u64_stats_fetch_begin_irq(&ring->syncp);
2445 rx_bytes += ring->stats.rx_bytes;
2446 rx_pkts += ring->stats.rx_pkts;
2447 rx_drop += ring->stats.l2_err;
2448 rx_errors += ring->stats.l2_err;
2449 rx_errors += ring->stats.l3l4_csum_err;
2450 rx_crc_errors += ring->stats.l2_err;
2451 rx_multicast += ring->stats.rx_multicast;
2452 rx_length_errors += ring->stats.err_pkt_len;
2453 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
2454 }
2455
2456 stats->tx_bytes = tx_bytes;
2457 stats->tx_packets = tx_pkts;
2458 stats->rx_bytes = rx_bytes;
2459 stats->rx_packets = rx_pkts;
2460
2461 stats->rx_errors = rx_errors;
2462 stats->multicast = rx_multicast;
2463 stats->rx_length_errors = rx_length_errors;
2464 stats->rx_crc_errors = rx_crc_errors;
2465 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
2466
2467 stats->tx_errors = tx_errors;
2468 stats->rx_dropped = rx_drop;
2469 stats->tx_dropped = tx_drop;
2470 stats->collisions = netdev->stats.collisions;
2471 stats->rx_over_errors = netdev->stats.rx_over_errors;
2472 stats->rx_frame_errors = netdev->stats.rx_frame_errors;
2473 stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
2474 stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
2475 stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
2476 stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
2477 stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
2478 stats->tx_window_errors = netdev->stats.tx_window_errors;
2479 stats->rx_compressed = netdev->stats.rx_compressed;
2480 stats->tx_compressed = netdev->stats.tx_compressed;
2481}
2482
2483static int hns3_setup_tc(struct net_device *netdev, void *type_data)
2484{
2485 struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
2486 struct hnae3_knic_private_info *kinfo;
2487 u8 tc = mqprio_qopt->qopt.num_tc;
2488 u16 mode = mqprio_qopt->mode;
2489 u8 hw = mqprio_qopt->qopt.hw;
2490 struct hnae3_handle *h;
2491
2492 if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
2493 mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
2494 return -EOPNOTSUPP;
2495
2496 if (tc > HNAE3_MAX_TC)
2497 return -EINVAL;
2498
2499 if (!netdev)
2500 return -EINVAL;
2501
2502 h = hns3_get_handle(netdev);
2503 kinfo = &h->kinfo;
2504
2505 netif_dbg(h, drv, netdev, "setup tc: num_tc=%u\n", tc);
2506
2507 return (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
2508 kinfo->dcb_ops->setup_tc(h, mqprio_qopt) : -EOPNOTSUPP;
2509}
2510
2511static int hns3_setup_tc_cls_flower(struct hns3_nic_priv *priv,
2512 struct flow_cls_offload *flow)
2513{
2514 int tc = tc_classid_to_hwtc(priv->netdev, flow->classid);
2515 struct hnae3_handle *h = hns3_get_handle(priv->netdev);
2516
2517 switch (flow->command) {
2518 case FLOW_CLS_REPLACE:
2519 if (h->ae_algo->ops->add_cls_flower)
2520 return h->ae_algo->ops->add_cls_flower(h, flow, tc);
2521 break;
2522 case FLOW_CLS_DESTROY:
2523 if (h->ae_algo->ops->del_cls_flower)
2524 return h->ae_algo->ops->del_cls_flower(h, flow);
2525 break;
2526 default:
2527 break;
2528 }
2529
2530 return -EOPNOTSUPP;
2531}
2532
2533static int hns3_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
2534 void *cb_priv)
2535{
2536 struct hns3_nic_priv *priv = cb_priv;
2537
2538 if (!tc_cls_can_offload_and_chain0(priv->netdev, type_data))
2539 return -EOPNOTSUPP;
2540
2541 switch (type) {
2542 case TC_SETUP_CLSFLOWER:
2543 return hns3_setup_tc_cls_flower(priv, type_data);
2544 default:
2545 return -EOPNOTSUPP;
2546 }
2547}
2548
2549static LIST_HEAD(hns3_block_cb_list);
2550
2551static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
2552 void *type_data)
2553{
2554 struct hns3_nic_priv *priv = netdev_priv(dev);
2555 int ret;
2556
2557 switch (type) {
2558 case TC_SETUP_QDISC_MQPRIO:
2559 ret = hns3_setup_tc(dev, type_data);
2560 break;
2561 case TC_SETUP_BLOCK:
2562 ret = flow_block_cb_setup_simple(type_data,
2563 &hns3_block_cb_list,
2564 hns3_setup_tc_block_cb,
2565 priv, priv, true);
2566 break;
2567 default:
2568 return -EOPNOTSUPP;
2569 }
2570
2571 return ret;
2572}
2573
2574static int hns3_vlan_rx_add_vid(struct net_device *netdev,
2575 __be16 proto, u16 vid)
2576{
2577 struct hnae3_handle *h = hns3_get_handle(netdev);
2578 int ret = -EIO;
2579
2580 if (h->ae_algo->ops->set_vlan_filter)
2581 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
2582
2583 return ret;
2584}
2585
2586static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
2587 __be16 proto, u16 vid)
2588{
2589 struct hnae3_handle *h = hns3_get_handle(netdev);
2590 int ret = -EIO;
2591
2592 if (h->ae_algo->ops->set_vlan_filter)
2593 ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
2594
2595 return ret;
2596}
2597
2598static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
2599 u8 qos, __be16 vlan_proto)
2600{
2601 struct hnae3_handle *h = hns3_get_handle(netdev);
2602 int ret = -EIO;
2603
2604 netif_dbg(h, drv, netdev,
2605 "set vf vlan: vf=%d, vlan=%u, qos=%u, vlan_proto=0x%x\n",
2606 vf, vlan, qos, ntohs(vlan_proto));
2607
2608 if (h->ae_algo->ops->set_vf_vlan_filter)
2609 ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
2610 qos, vlan_proto);
2611
2612 return ret;
2613}
2614
2615static int hns3_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
2616{
2617 struct hnae3_handle *handle = hns3_get_handle(netdev);
2618
2619 if (hns3_nic_resetting(netdev))
2620 return -EBUSY;
2621
2622 if (!handle->ae_algo->ops->set_vf_spoofchk)
2623 return -EOPNOTSUPP;
2624
2625 return handle->ae_algo->ops->set_vf_spoofchk(handle, vf, enable);
2626}
2627
2628static int hns3_set_vf_trust(struct net_device *netdev, int vf, bool enable)
2629{
2630 struct hnae3_handle *handle = hns3_get_handle(netdev);
2631
2632 if (!handle->ae_algo->ops->set_vf_trust)
2633 return -EOPNOTSUPP;
2634
2635 return handle->ae_algo->ops->set_vf_trust(handle, vf, enable);
2636}
2637
2638static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
2639{
2640 struct hnae3_handle *h = hns3_get_handle(netdev);
2641 int ret;
2642
2643 if (hns3_nic_resetting(netdev))
2644 return -EBUSY;
2645
2646 if (!h->ae_algo->ops->set_mtu)
2647 return -EOPNOTSUPP;
2648
2649 netif_dbg(h, drv, netdev,
2650 "change mtu from %u to %d\n", netdev->mtu, new_mtu);
2651
2652 ret = h->ae_algo->ops->set_mtu(h, new_mtu);
2653 if (ret)
2654 netdev_err(netdev, "failed to change MTU in hardware %d\n",
2655 ret);
2656 else
2657 netdev->mtu = new_mtu;
2658
2659 return ret;
2660}
2661
2662static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
2663{
2664 struct hns3_nic_priv *priv = netdev_priv(ndev);
2665 struct hnae3_handle *h = hns3_get_handle(ndev);
2666 struct hns3_enet_ring *tx_ring;
2667 struct napi_struct *napi;
2668 int timeout_queue = 0;
2669 int hw_head, hw_tail;
2670 int fbd_num, fbd_oft;
2671 int ebd_num, ebd_oft;
2672 int bd_num, bd_err;
2673 int ring_en, tc;
2674 int i;
2675
2676 /* Find the stopped queue the same way the stack does */
2677 for (i = 0; i < ndev->num_tx_queues; i++) {
2678 struct netdev_queue *q;
2679 unsigned long trans_start;
2680
2681 q = netdev_get_tx_queue(ndev, i);
2682 trans_start = READ_ONCE(q->trans_start);
2683 if (netif_xmit_stopped(q) &&
2684 time_after(jiffies,
2685 (trans_start + ndev->watchdog_timeo))) {
2686#ifdef CONFIG_BQL
2687 struct dql *dql = &q->dql;
2688
2689 netdev_info(ndev, "DQL info last_cnt: %u, queued: %u, adj_limit: %u, completed: %u\n",
2690 dql->last_obj_cnt, dql->num_queued,
2691 dql->adj_limit, dql->num_completed);
2692#endif
2693 timeout_queue = i;
2694 netdev_info(ndev, "queue state: 0x%lx, delta msecs: %u\n",
2695 q->state,
2696 jiffies_to_msecs(jiffies - trans_start));
2697 break;
2698 }
2699 }
2700
2701 if (i == ndev->num_tx_queues) {
2702 netdev_info(ndev,
2703 "no netdev TX timeout queue found, timeout count: %llu\n",
2704 priv->tx_timeout_count);
2705 return false;
2706 }
2707
2708 priv->tx_timeout_count++;
2709
2710 tx_ring = &priv->ring[timeout_queue];
2711 napi = &tx_ring->tqp_vector->napi;
2712
2713 netdev_info(ndev,
2714 "tx_timeout count: %llu, queue id: %d, SW_NTU: 0x%x, SW_NTC: 0x%x, napi state: %lu\n",
2715 priv->tx_timeout_count, timeout_queue, tx_ring->next_to_use,
2716 tx_ring->next_to_clean, napi->state);
2717
2718 netdev_info(ndev,
2719 "tx_pkts: %llu, tx_bytes: %llu, sw_err_cnt: %llu, tx_pending: %d\n",
2720 tx_ring->stats.tx_pkts, tx_ring->stats.tx_bytes,
2721 tx_ring->stats.sw_err_cnt, tx_ring->pending_buf);
2722
2723 netdev_info(ndev,
2724 "seg_pkt_cnt: %llu, tx_more: %llu, restart_queue: %llu, tx_busy: %llu\n",
2725 tx_ring->stats.seg_pkt_cnt, tx_ring->stats.tx_more,
2726 tx_ring->stats.restart_queue, tx_ring->stats.tx_busy);
2727
2728 /* When mac received many pause frames continuous, it's unable to send
2729 * packets, which may cause tx timeout
2730 */
2731 if (h->ae_algo->ops->get_mac_stats) {
2732 struct hns3_mac_stats mac_stats;
2733
2734 h->ae_algo->ops->get_mac_stats(h, &mac_stats);
2735 netdev_info(ndev, "tx_pause_cnt: %llu, rx_pause_cnt: %llu\n",
2736 mac_stats.tx_pause_cnt, mac_stats.rx_pause_cnt);
2737 }
2738
2739 hw_head = readl_relaxed(tx_ring->tqp->io_base +
2740 HNS3_RING_TX_RING_HEAD_REG);
2741 hw_tail = readl_relaxed(tx_ring->tqp->io_base +
2742 HNS3_RING_TX_RING_TAIL_REG);
2743 fbd_num = readl_relaxed(tx_ring->tqp->io_base +
2744 HNS3_RING_TX_RING_FBDNUM_REG);
2745 fbd_oft = readl_relaxed(tx_ring->tqp->io_base +
2746 HNS3_RING_TX_RING_OFFSET_REG);
2747 ebd_num = readl_relaxed(tx_ring->tqp->io_base +
2748 HNS3_RING_TX_RING_EBDNUM_REG);
2749 ebd_oft = readl_relaxed(tx_ring->tqp->io_base +
2750 HNS3_RING_TX_RING_EBD_OFFSET_REG);
2751 bd_num = readl_relaxed(tx_ring->tqp->io_base +
2752 HNS3_RING_TX_RING_BD_NUM_REG);
2753 bd_err = readl_relaxed(tx_ring->tqp->io_base +
2754 HNS3_RING_TX_RING_BD_ERR_REG);
2755 ring_en = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_EN_REG);
2756 tc = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_TX_RING_TC_REG);
2757
2758 netdev_info(ndev,
2759 "BD_NUM: 0x%x HW_HEAD: 0x%x, HW_TAIL: 0x%x, BD_ERR: 0x%x, INT: 0x%x\n",
2760 bd_num, hw_head, hw_tail, bd_err,
2761 readl(tx_ring->tqp_vector->mask_addr));
2762 netdev_info(ndev,
2763 "RING_EN: 0x%x, TC: 0x%x, FBD_NUM: 0x%x FBD_OFT: 0x%x, EBD_NUM: 0x%x, EBD_OFT: 0x%x\n",
2764 ring_en, tc, fbd_num, fbd_oft, ebd_num, ebd_oft);
2765
2766 return true;
2767}
2768
2769static void hns3_nic_net_timeout(struct net_device *ndev, unsigned int txqueue)
2770{
2771 struct hns3_nic_priv *priv = netdev_priv(ndev);
2772 struct hnae3_handle *h = priv->ae_handle;
2773
2774 if (!hns3_get_tx_timeo_queue_info(ndev))
2775 return;
2776
2777 /* request the reset, and let the hclge to determine
2778 * which reset level should be done
2779 */
2780 if (h->ae_algo->ops->reset_event)
2781 h->ae_algo->ops->reset_event(h->pdev, h);
2782}
2783
2784#ifdef CONFIG_RFS_ACCEL
2785static int hns3_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
2786 u16 rxq_index, u32 flow_id)
2787{
2788 struct hnae3_handle *h = hns3_get_handle(dev);
2789 struct flow_keys fkeys;
2790
2791 if (!h->ae_algo->ops->add_arfs_entry)
2792 return -EOPNOTSUPP;
2793
2794 if (skb->encapsulation)
2795 return -EPROTONOSUPPORT;
2796
2797 if (!skb_flow_dissect_flow_keys(skb, &fkeys, 0))
2798 return -EPROTONOSUPPORT;
2799
2800 if ((fkeys.basic.n_proto != htons(ETH_P_IP) &&
2801 fkeys.basic.n_proto != htons(ETH_P_IPV6)) ||
2802 (fkeys.basic.ip_proto != IPPROTO_TCP &&
2803 fkeys.basic.ip_proto != IPPROTO_UDP))
2804 return -EPROTONOSUPPORT;
2805
2806 return h->ae_algo->ops->add_arfs_entry(h, rxq_index, flow_id, &fkeys);
2807}
2808#endif
2809
2810static int hns3_nic_get_vf_config(struct net_device *ndev, int vf,
2811 struct ifla_vf_info *ivf)
2812{
2813 struct hnae3_handle *h = hns3_get_handle(ndev);
2814
2815 if (!h->ae_algo->ops->get_vf_config)
2816 return -EOPNOTSUPP;
2817
2818 return h->ae_algo->ops->get_vf_config(h, vf, ivf);
2819}
2820
2821static int hns3_nic_set_vf_link_state(struct net_device *ndev, int vf,
2822 int link_state)
2823{
2824 struct hnae3_handle *h = hns3_get_handle(ndev);
2825
2826 if (!h->ae_algo->ops->set_vf_link_state)
2827 return -EOPNOTSUPP;
2828
2829 return h->ae_algo->ops->set_vf_link_state(h, vf, link_state);
2830}
2831
2832static int hns3_nic_set_vf_rate(struct net_device *ndev, int vf,
2833 int min_tx_rate, int max_tx_rate)
2834{
2835 struct hnae3_handle *h = hns3_get_handle(ndev);
2836
2837 if (!h->ae_algo->ops->set_vf_rate)
2838 return -EOPNOTSUPP;
2839
2840 return h->ae_algo->ops->set_vf_rate(h, vf, min_tx_rate, max_tx_rate,
2841 false);
2842}
2843
2844static int hns3_nic_set_vf_mac(struct net_device *netdev, int vf_id, u8 *mac)
2845{
2846 struct hnae3_handle *h = hns3_get_handle(netdev);
2847 char format_mac_addr[HNAE3_FORMAT_MAC_ADDR_LEN];
2848
2849 if (!h->ae_algo->ops->set_vf_mac)
2850 return -EOPNOTSUPP;
2851
2852 if (is_multicast_ether_addr(mac)) {
2853 hnae3_format_mac_addr(format_mac_addr, mac);
2854 netdev_err(netdev,
2855 "Invalid MAC:%s specified. Could not set MAC\n",
2856 format_mac_addr);
2857 return -EINVAL;
2858 }
2859
2860 return h->ae_algo->ops->set_vf_mac(h, vf_id, mac);
2861}
2862
2863static const struct net_device_ops hns3_nic_netdev_ops = {
2864 .ndo_open = hns3_nic_net_open,
2865 .ndo_stop = hns3_nic_net_stop,
2866 .ndo_start_xmit = hns3_nic_net_xmit,
2867 .ndo_tx_timeout = hns3_nic_net_timeout,
2868 .ndo_set_mac_address = hns3_nic_net_set_mac_address,
2869 .ndo_eth_ioctl = hns3_nic_do_ioctl,
2870 .ndo_change_mtu = hns3_nic_change_mtu,
2871 .ndo_set_features = hns3_nic_set_features,
2872 .ndo_features_check = hns3_features_check,
2873 .ndo_get_stats64 = hns3_nic_get_stats64,
2874 .ndo_setup_tc = hns3_nic_setup_tc,
2875 .ndo_set_rx_mode = hns3_nic_set_rx_mode,
2876 .ndo_vlan_rx_add_vid = hns3_vlan_rx_add_vid,
2877 .ndo_vlan_rx_kill_vid = hns3_vlan_rx_kill_vid,
2878 .ndo_set_vf_vlan = hns3_ndo_set_vf_vlan,
2879 .ndo_set_vf_spoofchk = hns3_set_vf_spoofchk,
2880 .ndo_set_vf_trust = hns3_set_vf_trust,
2881#ifdef CONFIG_RFS_ACCEL
2882 .ndo_rx_flow_steer = hns3_rx_flow_steer,
2883#endif
2884 .ndo_get_vf_config = hns3_nic_get_vf_config,
2885 .ndo_set_vf_link_state = hns3_nic_set_vf_link_state,
2886 .ndo_set_vf_rate = hns3_nic_set_vf_rate,
2887 .ndo_set_vf_mac = hns3_nic_set_vf_mac,
2888};
2889
2890bool hns3_is_phys_func(struct pci_dev *pdev)
2891{
2892 u32 dev_id = pdev->device;
2893
2894 switch (dev_id) {
2895 case HNAE3_DEV_ID_GE:
2896 case HNAE3_DEV_ID_25GE:
2897 case HNAE3_DEV_ID_25GE_RDMA:
2898 case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
2899 case HNAE3_DEV_ID_50GE_RDMA:
2900 case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
2901 case HNAE3_DEV_ID_100G_RDMA_MACSEC:
2902 case HNAE3_DEV_ID_200G_RDMA:
2903 return true;
2904 case HNAE3_DEV_ID_VF:
2905 case HNAE3_DEV_ID_RDMA_DCB_PFC_VF:
2906 return false;
2907 default:
2908 dev_warn(&pdev->dev, "un-recognized pci device-id %u",
2909 dev_id);
2910 }
2911
2912 return false;
2913}
2914
2915static void hns3_disable_sriov(struct pci_dev *pdev)
2916{
2917 /* If our VFs are assigned we cannot shut down SR-IOV
2918 * without causing issues, so just leave the hardware
2919 * available but disabled
2920 */
2921 if (pci_vfs_assigned(pdev)) {
2922 dev_warn(&pdev->dev,
2923 "disabling driver while VFs are assigned\n");
2924 return;
2925 }
2926
2927 pci_disable_sriov(pdev);
2928}
2929
2930/* hns3_probe - Device initialization routine
2931 * @pdev: PCI device information struct
2932 * @ent: entry in hns3_pci_tbl
2933 *
2934 * hns3_probe initializes a PF identified by a pci_dev structure.
2935 * The OS initialization, configuring of the PF private structure,
2936 * and a hardware reset occur.
2937 *
2938 * Returns 0 on success, negative on failure
2939 */
2940static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2941{
2942 struct hnae3_ae_dev *ae_dev;
2943 int ret;
2944
2945 ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev), GFP_KERNEL);
2946 if (!ae_dev)
2947 return -ENOMEM;
2948
2949 ae_dev->pdev = pdev;
2950 ae_dev->flag = ent->driver_data;
2951 pci_set_drvdata(pdev, ae_dev);
2952
2953 ret = hnae3_register_ae_dev(ae_dev);
2954 if (ret)
2955 pci_set_drvdata(pdev, NULL);
2956
2957 return ret;
2958}
2959
2960/* hns3_remove - Device removal routine
2961 * @pdev: PCI device information struct
2962 */
2963static void hns3_remove(struct pci_dev *pdev)
2964{
2965 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2966
2967 if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
2968 hns3_disable_sriov(pdev);
2969
2970 hnae3_unregister_ae_dev(ae_dev);
2971 pci_set_drvdata(pdev, NULL);
2972}
2973
2974/**
2975 * hns3_pci_sriov_configure
2976 * @pdev: pointer to a pci_dev structure
2977 * @num_vfs: number of VFs to allocate
2978 *
2979 * Enable or change the number of VFs. Called when the user updates the number
2980 * of VFs in sysfs.
2981 **/
2982static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
2983{
2984 int ret;
2985
2986 if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
2987 dev_warn(&pdev->dev, "Can not config SRIOV\n");
2988 return -EINVAL;
2989 }
2990
2991 if (num_vfs) {
2992 ret = pci_enable_sriov(pdev, num_vfs);
2993 if (ret)
2994 dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
2995 else
2996 return num_vfs;
2997 } else if (!pci_vfs_assigned(pdev)) {
2998 pci_disable_sriov(pdev);
2999 } else {
3000 dev_warn(&pdev->dev,
3001 "Unable to free VFs because some are assigned to VMs.\n");
3002 }
3003
3004 return 0;
3005}
3006
3007static void hns3_shutdown(struct pci_dev *pdev)
3008{
3009 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3010
3011 hnae3_unregister_ae_dev(ae_dev);
3012 pci_set_drvdata(pdev, NULL);
3013
3014 if (system_state == SYSTEM_POWER_OFF)
3015 pci_set_power_state(pdev, PCI_D3hot);
3016}
3017
3018static int __maybe_unused hns3_suspend(struct device *dev)
3019{
3020 struct hnae3_ae_dev *ae_dev = dev_get_drvdata(dev);
3021
3022 if (ae_dev && hns3_is_phys_func(ae_dev->pdev)) {
3023 dev_info(dev, "Begin to suspend.\n");
3024 if (ae_dev->ops && ae_dev->ops->reset_prepare)
3025 ae_dev->ops->reset_prepare(ae_dev, HNAE3_FUNC_RESET);
3026 }
3027
3028 return 0;
3029}
3030
3031static int __maybe_unused hns3_resume(struct device *dev)
3032{
3033 struct hnae3_ae_dev *ae_dev = dev_get_drvdata(dev);
3034
3035 if (ae_dev && hns3_is_phys_func(ae_dev->pdev)) {
3036 dev_info(dev, "Begin to resume.\n");
3037 if (ae_dev->ops && ae_dev->ops->reset_done)
3038 ae_dev->ops->reset_done(ae_dev);
3039 }
3040
3041 return 0;
3042}
3043
3044static pci_ers_result_t hns3_error_detected(struct pci_dev *pdev,
3045 pci_channel_state_t state)
3046{
3047 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3048 pci_ers_result_t ret;
3049
3050 dev_info(&pdev->dev, "PCI error detected, state(=%u)!!\n", state);
3051
3052 if (state == pci_channel_io_perm_failure)
3053 return PCI_ERS_RESULT_DISCONNECT;
3054
3055 if (!ae_dev || !ae_dev->ops) {
3056 dev_err(&pdev->dev,
3057 "Can't recover - error happened before device initialized\n");
3058 return PCI_ERS_RESULT_NONE;
3059 }
3060
3061 if (ae_dev->ops->handle_hw_ras_error)
3062 ret = ae_dev->ops->handle_hw_ras_error(ae_dev);
3063 else
3064 return PCI_ERS_RESULT_NONE;
3065
3066 return ret;
3067}
3068
3069static pci_ers_result_t hns3_slot_reset(struct pci_dev *pdev)
3070{
3071 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3072 const struct hnae3_ae_ops *ops;
3073 enum hnae3_reset_type reset_type;
3074 struct device *dev = &pdev->dev;
3075
3076 if (!ae_dev || !ae_dev->ops)
3077 return PCI_ERS_RESULT_NONE;
3078
3079 ops = ae_dev->ops;
3080 /* request the reset */
3081 if (ops->reset_event && ops->get_reset_level &&
3082 ops->set_default_reset_request) {
3083 if (ae_dev->hw_err_reset_req) {
3084 reset_type = ops->get_reset_level(ae_dev,
3085 &ae_dev->hw_err_reset_req);
3086 ops->set_default_reset_request(ae_dev, reset_type);
3087 dev_info(dev, "requesting reset due to PCI error\n");
3088 ops->reset_event(pdev, NULL);
3089 }
3090
3091 return PCI_ERS_RESULT_RECOVERED;
3092 }
3093
3094 return PCI_ERS_RESULT_DISCONNECT;
3095}
3096
3097static void hns3_reset_prepare(struct pci_dev *pdev)
3098{
3099 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3100
3101 dev_info(&pdev->dev, "FLR prepare\n");
3102 if (ae_dev && ae_dev->ops && ae_dev->ops->reset_prepare)
3103 ae_dev->ops->reset_prepare(ae_dev, HNAE3_FLR_RESET);
3104}
3105
3106static void hns3_reset_done(struct pci_dev *pdev)
3107{
3108 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3109
3110 dev_info(&pdev->dev, "FLR done\n");
3111 if (ae_dev && ae_dev->ops && ae_dev->ops->reset_done)
3112 ae_dev->ops->reset_done(ae_dev);
3113}
3114
3115static const struct pci_error_handlers hns3_err_handler = {
3116 .error_detected = hns3_error_detected,
3117 .slot_reset = hns3_slot_reset,
3118 .reset_prepare = hns3_reset_prepare,
3119 .reset_done = hns3_reset_done,
3120};
3121
3122static SIMPLE_DEV_PM_OPS(hns3_pm_ops, hns3_suspend, hns3_resume);
3123
3124static struct pci_driver hns3_driver = {
3125 .name = hns3_driver_name,
3126 .id_table = hns3_pci_tbl,
3127 .probe = hns3_probe,
3128 .remove = hns3_remove,
3129 .shutdown = hns3_shutdown,
3130 .driver.pm = &hns3_pm_ops,
3131 .sriov_configure = hns3_pci_sriov_configure,
3132 .err_handler = &hns3_err_handler,
3133};
3134
3135/* set default feature to hns3 */
3136static void hns3_set_default_feature(struct net_device *netdev)
3137{
3138 struct hnae3_handle *h = hns3_get_handle(netdev);
3139 struct pci_dev *pdev = h->pdev;
3140 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3141
3142 netdev->priv_flags |= IFF_UNICAST_FLT;
3143
3144 netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
3145
3146 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
3147 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3148 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
3149 NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
3150 NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
3151 NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
3152
3153 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V2) {
3154 netdev->features |= NETIF_F_GRO_HW;
3155
3156 if (!(h->flags & HNAE3_SUPPORT_VF))
3157 netdev->features |= NETIF_F_NTUPLE;
3158 }
3159
3160 if (test_bit(HNAE3_DEV_SUPPORT_UDP_GSO_B, ae_dev->caps))
3161 netdev->features |= NETIF_F_GSO_UDP_L4;
3162
3163 if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps))
3164 netdev->features |= NETIF_F_HW_CSUM;
3165 else
3166 netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3167
3168 if (test_bit(HNAE3_DEV_SUPPORT_UDP_TUNNEL_CSUM_B, ae_dev->caps))
3169 netdev->features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
3170
3171 if (test_bit(HNAE3_DEV_SUPPORT_FD_FORWARD_TC_B, ae_dev->caps))
3172 netdev->features |= NETIF_F_HW_TC;
3173
3174 netdev->hw_features |= netdev->features;
3175 if (!test_bit(HNAE3_DEV_SUPPORT_VLAN_FLTR_MDF_B, ae_dev->caps))
3176 netdev->hw_features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
3177
3178 netdev->vlan_features |= netdev->features &
3179 ~(NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_CTAG_TX |
3180 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_GRO_HW | NETIF_F_NTUPLE |
3181 NETIF_F_HW_TC);
3182
3183 netdev->hw_enc_features |= netdev->vlan_features | NETIF_F_TSO_MANGLEID;
3184}
3185
3186static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
3187 struct hns3_desc_cb *cb)
3188{
3189 unsigned int order = hns3_page_order(ring);
3190 struct page *p;
3191
3192 if (ring->page_pool) {
3193 p = page_pool_dev_alloc_frag(ring->page_pool,
3194 &cb->page_offset,
3195 hns3_buf_size(ring));
3196 if (unlikely(!p))
3197 return -ENOMEM;
3198
3199 cb->priv = p;
3200 cb->buf = page_address(p);
3201 cb->dma = page_pool_get_dma_addr(p);
3202 cb->type = DESC_TYPE_PP_FRAG;
3203 cb->reuse_flag = 0;
3204 return 0;
3205 }
3206
3207 p = dev_alloc_pages(order);
3208 if (!p)
3209 return -ENOMEM;
3210
3211 cb->priv = p;
3212 cb->page_offset = 0;
3213 cb->reuse_flag = 0;
3214 cb->buf = page_address(p);
3215 cb->length = hns3_page_size(ring);
3216 cb->type = DESC_TYPE_PAGE;
3217 page_ref_add(p, USHRT_MAX - 1);
3218 cb->pagecnt_bias = USHRT_MAX;
3219
3220 return 0;
3221}
3222
3223static void hns3_free_buffer(struct hns3_enet_ring *ring,
3224 struct hns3_desc_cb *cb, int budget)
3225{
3226 if (cb->type & (DESC_TYPE_SKB | DESC_TYPE_BOUNCE_HEAD |
3227 DESC_TYPE_BOUNCE_ALL | DESC_TYPE_SGL_SKB))
3228 napi_consume_skb(cb->priv, budget);
3229 else if (!HNAE3_IS_TX_RING(ring)) {
3230 if (cb->type & DESC_TYPE_PAGE && cb->pagecnt_bias)
3231 __page_frag_cache_drain(cb->priv, cb->pagecnt_bias);
3232 else if (cb->type & DESC_TYPE_PP_FRAG)
3233 page_pool_put_full_page(ring->page_pool, cb->priv,
3234 false);
3235 }
3236 memset(cb, 0, sizeof(*cb));
3237}
3238
3239static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
3240{
3241 cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
3242 cb->length, ring_to_dma_dir(ring));
3243
3244 if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
3245 return -EIO;
3246
3247 return 0;
3248}
3249
3250static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
3251 struct hns3_desc_cb *cb)
3252{
3253 if (cb->type & (DESC_TYPE_SKB | DESC_TYPE_FRAGLIST_SKB))
3254 dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
3255 ring_to_dma_dir(ring));
3256 else if ((cb->type & DESC_TYPE_PAGE) && cb->length)
3257 dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
3258 ring_to_dma_dir(ring));
3259 else if (cb->type & (DESC_TYPE_BOUNCE_ALL | DESC_TYPE_BOUNCE_HEAD |
3260 DESC_TYPE_SGL_SKB))
3261 hns3_tx_spare_reclaim_cb(ring, cb);
3262}
3263
3264static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
3265{
3266 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
3267 ring->desc[i].addr = 0;
3268 ring->desc_cb[i].refill = 0;
3269}
3270
3271static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i,
3272 int budget)
3273{
3274 struct hns3_desc_cb *cb = &ring->desc_cb[i];
3275
3276 if (!ring->desc_cb[i].dma)
3277 return;
3278
3279 hns3_buffer_detach(ring, i);
3280 hns3_free_buffer(ring, cb, budget);
3281}
3282
3283static void hns3_free_buffers(struct hns3_enet_ring *ring)
3284{
3285 int i;
3286
3287 for (i = 0; i < ring->desc_num; i++)
3288 hns3_free_buffer_detach(ring, i, 0);
3289}
3290
3291/* free desc along with its attached buffer */
3292static void hns3_free_desc(struct hns3_enet_ring *ring)
3293{
3294 int size = ring->desc_num * sizeof(ring->desc[0]);
3295
3296 hns3_free_buffers(ring);
3297
3298 if (ring->desc) {
3299 dma_free_coherent(ring_to_dev(ring), size,
3300 ring->desc, ring->desc_dma_addr);
3301 ring->desc = NULL;
3302 }
3303}
3304
3305static int hns3_alloc_desc(struct hns3_enet_ring *ring)
3306{
3307 int size = ring->desc_num * sizeof(ring->desc[0]);
3308
3309 ring->desc = dma_alloc_coherent(ring_to_dev(ring), size,
3310 &ring->desc_dma_addr, GFP_KERNEL);
3311 if (!ring->desc)
3312 return -ENOMEM;
3313
3314 return 0;
3315}
3316
3317static int hns3_alloc_and_map_buffer(struct hns3_enet_ring *ring,
3318 struct hns3_desc_cb *cb)
3319{
3320 int ret;
3321
3322 ret = hns3_alloc_buffer(ring, cb);
3323 if (ret || ring->page_pool)
3324 goto out;
3325
3326 ret = hns3_map_buffer(ring, cb);
3327 if (ret)
3328 goto out_with_buf;
3329
3330 return 0;
3331
3332out_with_buf:
3333 hns3_free_buffer(ring, cb, 0);
3334out:
3335 return ret;
3336}
3337
3338static int hns3_alloc_and_attach_buffer(struct hns3_enet_ring *ring, int i)
3339{
3340 int ret = hns3_alloc_and_map_buffer(ring, &ring->desc_cb[i]);
3341
3342 if (ret)
3343 return ret;
3344
3345 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3346 ring->desc_cb[i].page_offset);
3347 ring->desc_cb[i].refill = 1;
3348
3349 return 0;
3350}
3351
3352/* Allocate memory for raw pkg, and map with dma */
3353static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
3354{
3355 int i, j, ret;
3356
3357 for (i = 0; i < ring->desc_num; i++) {
3358 ret = hns3_alloc_and_attach_buffer(ring, i);
3359 if (ret)
3360 goto out_buffer_fail;
3361 }
3362
3363 return 0;
3364
3365out_buffer_fail:
3366 for (j = i - 1; j >= 0; j--)
3367 hns3_free_buffer_detach(ring, j, 0);
3368 return ret;
3369}
3370
3371/* detach a in-used buffer and replace with a reserved one */
3372static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
3373 struct hns3_desc_cb *res_cb)
3374{
3375 hns3_unmap_buffer(ring, &ring->desc_cb[i]);
3376 ring->desc_cb[i] = *res_cb;
3377 ring->desc_cb[i].refill = 1;
3378 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3379 ring->desc_cb[i].page_offset);
3380 ring->desc[i].rx.bd_base_info = 0;
3381}
3382
3383static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
3384{
3385 ring->desc_cb[i].reuse_flag = 0;
3386 ring->desc_cb[i].refill = 1;
3387 ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3388 ring->desc_cb[i].page_offset);
3389 ring->desc[i].rx.bd_base_info = 0;
3390
3391 dma_sync_single_for_device(ring_to_dev(ring),
3392 ring->desc_cb[i].dma + ring->desc_cb[i].page_offset,
3393 hns3_buf_size(ring),
3394 DMA_FROM_DEVICE);
3395}
3396
3397static bool hns3_nic_reclaim_desc(struct hns3_enet_ring *ring,
3398 int *bytes, int *pkts, int budget)
3399{
3400 /* pair with ring->last_to_use update in hns3_tx_doorbell(),
3401 * smp_store_release() is not used in hns3_tx_doorbell() because
3402 * the doorbell operation already have the needed barrier operation.
3403 */
3404 int ltu = smp_load_acquire(&ring->last_to_use);
3405 int ntc = ring->next_to_clean;
3406 struct hns3_desc_cb *desc_cb;
3407 bool reclaimed = false;
3408 struct hns3_desc *desc;
3409
3410 while (ltu != ntc) {
3411 desc = &ring->desc[ntc];
3412
3413 if (le16_to_cpu(desc->tx.bdtp_fe_sc_vld_ra_ri) &
3414 BIT(HNS3_TXD_VLD_B))
3415 break;
3416
3417 desc_cb = &ring->desc_cb[ntc];
3418
3419 if (desc_cb->type & (DESC_TYPE_SKB | DESC_TYPE_BOUNCE_ALL |
3420 DESC_TYPE_BOUNCE_HEAD |
3421 DESC_TYPE_SGL_SKB)) {
3422 (*pkts)++;
3423 (*bytes) += desc_cb->send_bytes;
3424 }
3425
3426 /* desc_cb will be cleaned, after hnae3_free_buffer_detach */
3427 hns3_free_buffer_detach(ring, ntc, budget);
3428
3429 if (++ntc == ring->desc_num)
3430 ntc = 0;
3431
3432 /* Issue prefetch for next Tx descriptor */
3433 prefetch(&ring->desc_cb[ntc]);
3434 reclaimed = true;
3435 }
3436
3437 if (unlikely(!reclaimed))
3438 return false;
3439
3440 /* This smp_store_release() pairs with smp_load_acquire() in
3441 * ring_space called by hns3_nic_net_xmit.
3442 */
3443 smp_store_release(&ring->next_to_clean, ntc);
3444
3445 hns3_tx_spare_update(ring);
3446
3447 return true;
3448}
3449
3450void hns3_clean_tx_ring(struct hns3_enet_ring *ring, int budget)
3451{
3452 struct net_device *netdev = ring_to_netdev(ring);
3453 struct hns3_nic_priv *priv = netdev_priv(netdev);
3454 struct netdev_queue *dev_queue;
3455 int bytes, pkts;
3456
3457 bytes = 0;
3458 pkts = 0;
3459
3460 if (unlikely(!hns3_nic_reclaim_desc(ring, &bytes, &pkts, budget)))
3461 return;
3462
3463 ring->tqp_vector->tx_group.total_bytes += bytes;
3464 ring->tqp_vector->tx_group.total_packets += pkts;
3465
3466 u64_stats_update_begin(&ring->syncp);
3467 ring->stats.tx_bytes += bytes;
3468 ring->stats.tx_pkts += pkts;
3469 u64_stats_update_end(&ring->syncp);
3470
3471 dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
3472 netdev_tx_completed_queue(dev_queue, pkts, bytes);
3473
3474 if (unlikely(netif_carrier_ok(netdev) &&
3475 ring_space(ring) > HNS3_MAX_TSO_BD_NUM)) {
3476 /* Make sure that anybody stopping the queue after this
3477 * sees the new next_to_clean.
3478 */
3479 smp_mb();
3480 if (netif_tx_queue_stopped(dev_queue) &&
3481 !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
3482 netif_tx_wake_queue(dev_queue);
3483 ring->stats.restart_queue++;
3484 }
3485 }
3486}
3487
3488static int hns3_desc_unused(struct hns3_enet_ring *ring)
3489{
3490 int ntc = ring->next_to_clean;
3491 int ntu = ring->next_to_use;
3492
3493 if (unlikely(ntc == ntu && !ring->desc_cb[ntc].refill))
3494 return ring->desc_num;
3495
3496 return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
3497}
3498
3499/* Return true if there is any allocation failure */
3500static bool hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring,
3501 int cleand_count)
3502{
3503 struct hns3_desc_cb *desc_cb;
3504 struct hns3_desc_cb res_cbs;
3505 int i, ret;
3506
3507 for (i = 0; i < cleand_count; i++) {
3508 desc_cb = &ring->desc_cb[ring->next_to_use];
3509 if (desc_cb->reuse_flag) {
3510 u64_stats_update_begin(&ring->syncp);
3511 ring->stats.reuse_pg_cnt++;
3512 u64_stats_update_end(&ring->syncp);
3513
3514 hns3_reuse_buffer(ring, ring->next_to_use);
3515 } else {
3516 ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
3517 if (ret) {
3518 u64_stats_update_begin(&ring->syncp);
3519 ring->stats.sw_err_cnt++;
3520 u64_stats_update_end(&ring->syncp);
3521
3522 hns3_rl_err(ring_to_netdev(ring),
3523 "alloc rx buffer failed: %d\n",
3524 ret);
3525
3526 writel(i, ring->tqp->io_base +
3527 HNS3_RING_RX_RING_HEAD_REG);
3528 return true;
3529 }
3530 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
3531
3532 u64_stats_update_begin(&ring->syncp);
3533 ring->stats.non_reuse_pg++;
3534 u64_stats_update_end(&ring->syncp);
3535 }
3536
3537 ring_ptr_move_fw(ring, next_to_use);
3538 }
3539
3540 writel(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
3541 return false;
3542}
3543
3544static bool hns3_can_reuse_page(struct hns3_desc_cb *cb)
3545{
3546 return page_count(cb->priv) == cb->pagecnt_bias;
3547}
3548
3549static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
3550 struct hns3_enet_ring *ring, int pull_len,
3551 struct hns3_desc_cb *desc_cb)
3552{
3553 struct hns3_desc *desc = &ring->desc[ring->next_to_clean];
3554 u32 frag_offset = desc_cb->page_offset + pull_len;
3555 int size = le16_to_cpu(desc->rx.size);
3556 u32 truesize = hns3_buf_size(ring);
3557 u32 frag_size = size - pull_len;
3558 bool reused;
3559
3560 if (ring->page_pool) {
3561 skb_add_rx_frag(skb, i, desc_cb->priv, frag_offset,
3562 frag_size, truesize);
3563 return;
3564 }
3565
3566 /* Avoid re-using remote or pfmem page */
3567 if (unlikely(!dev_page_is_reusable(desc_cb->priv)))
3568 goto out;
3569
3570 reused = hns3_can_reuse_page(desc_cb);
3571
3572 /* Rx page can be reused when:
3573 * 1. Rx page is only owned by the driver when page_offset
3574 * is zero, which means 0 @ truesize will be used by
3575 * stack after skb_add_rx_frag() is called, and the rest
3576 * of rx page can be reused by driver.
3577 * Or
3578 * 2. Rx page is only owned by the driver when page_offset
3579 * is non-zero, which means page_offset @ truesize will
3580 * be used by stack after skb_add_rx_frag() is called,
3581 * and 0 @ truesize can be reused by driver.
3582 */
3583 if ((!desc_cb->page_offset && reused) ||
3584 ((desc_cb->page_offset + truesize + truesize) <=
3585 hns3_page_size(ring) && desc_cb->page_offset)) {
3586 desc_cb->page_offset += truesize;
3587 desc_cb->reuse_flag = 1;
3588 } else if (desc_cb->page_offset && reused) {
3589 desc_cb->page_offset = 0;
3590 desc_cb->reuse_flag = 1;
3591 } else if (frag_size <= ring->rx_copybreak) {
3592 void *frag = napi_alloc_frag(frag_size);
3593
3594 if (unlikely(!frag)) {
3595 u64_stats_update_begin(&ring->syncp);
3596 ring->stats.frag_alloc_err++;
3597 u64_stats_update_end(&ring->syncp);
3598
3599 hns3_rl_err(ring_to_netdev(ring),
3600 "failed to allocate rx frag\n");
3601 goto out;
3602 }
3603
3604 desc_cb->reuse_flag = 1;
3605 memcpy(frag, desc_cb->buf + frag_offset, frag_size);
3606 skb_add_rx_frag(skb, i, virt_to_page(frag),
3607 offset_in_page(frag), frag_size, frag_size);
3608
3609 u64_stats_update_begin(&ring->syncp);
3610 ring->stats.frag_alloc++;
3611 u64_stats_update_end(&ring->syncp);
3612 return;
3613 }
3614
3615out:
3616 desc_cb->pagecnt_bias--;
3617
3618 if (unlikely(!desc_cb->pagecnt_bias)) {
3619 page_ref_add(desc_cb->priv, USHRT_MAX);
3620 desc_cb->pagecnt_bias = USHRT_MAX;
3621 }
3622
3623 skb_add_rx_frag(skb, i, desc_cb->priv, frag_offset,
3624 frag_size, truesize);
3625
3626 if (unlikely(!desc_cb->reuse_flag))
3627 __page_frag_cache_drain(desc_cb->priv, desc_cb->pagecnt_bias);
3628}
3629
3630static int hns3_gro_complete(struct sk_buff *skb, u32 l234info)
3631{
3632 __be16 type = skb->protocol;
3633 struct tcphdr *th;
3634 int depth = 0;
3635
3636 while (eth_type_vlan(type)) {
3637 struct vlan_hdr *vh;
3638
3639 if ((depth + VLAN_HLEN) > skb_headlen(skb))
3640 return -EFAULT;
3641
3642 vh = (struct vlan_hdr *)(skb->data + depth);
3643 type = vh->h_vlan_encapsulated_proto;
3644 depth += VLAN_HLEN;
3645 }
3646
3647 skb_set_network_header(skb, depth);
3648
3649 if (type == htons(ETH_P_IP)) {
3650 const struct iphdr *iph = ip_hdr(skb);
3651
3652 depth += sizeof(struct iphdr);
3653 skb_set_transport_header(skb, depth);
3654 th = tcp_hdr(skb);
3655 th->check = ~tcp_v4_check(skb->len - depth, iph->saddr,
3656 iph->daddr, 0);
3657 } else if (type == htons(ETH_P_IPV6)) {
3658 const struct ipv6hdr *iph = ipv6_hdr(skb);
3659
3660 depth += sizeof(struct ipv6hdr);
3661 skb_set_transport_header(skb, depth);
3662 th = tcp_hdr(skb);
3663 th->check = ~tcp_v6_check(skb->len - depth, &iph->saddr,
3664 &iph->daddr, 0);
3665 } else {
3666 hns3_rl_err(skb->dev,
3667 "Error: FW GRO supports only IPv4/IPv6, not 0x%04x, depth: %d\n",
3668 be16_to_cpu(type), depth);
3669 return -EFAULT;
3670 }
3671
3672 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
3673 if (th->cwr)
3674 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
3675
3676 if (l234info & BIT(HNS3_RXD_GRO_FIXID_B))
3677 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_FIXEDID;
3678
3679 skb->csum_start = (unsigned char *)th - skb->head;
3680 skb->csum_offset = offsetof(struct tcphdr, check);
3681 skb->ip_summed = CHECKSUM_PARTIAL;
3682
3683 trace_hns3_gro(skb);
3684
3685 return 0;
3686}
3687
3688static bool hns3_checksum_complete(struct hns3_enet_ring *ring,
3689 struct sk_buff *skb, u32 ptype, u16 csum)
3690{
3691 if (ptype == HNS3_INVALID_PTYPE ||
3692 hns3_rx_ptype_tbl[ptype].ip_summed != CHECKSUM_COMPLETE)
3693 return false;
3694
3695 u64_stats_update_begin(&ring->syncp);
3696 ring->stats.csum_complete++;
3697 u64_stats_update_end(&ring->syncp);
3698 skb->ip_summed = CHECKSUM_COMPLETE;
3699 skb->csum = csum_unfold((__force __sum16)csum);
3700
3701 return true;
3702}
3703
3704static void hns3_rx_handle_csum(struct sk_buff *skb, u32 l234info,
3705 u32 ol_info, u32 ptype)
3706{
3707 int l3_type, l4_type;
3708 int ol4_type;
3709
3710 if (ptype != HNS3_INVALID_PTYPE) {
3711 skb->csum_level = hns3_rx_ptype_tbl[ptype].csum_level;
3712 skb->ip_summed = hns3_rx_ptype_tbl[ptype].ip_summed;
3713
3714 return;
3715 }
3716
3717 ol4_type = hnae3_get_field(ol_info, HNS3_RXD_OL4ID_M,
3718 HNS3_RXD_OL4ID_S);
3719 switch (ol4_type) {
3720 case HNS3_OL4_TYPE_MAC_IN_UDP:
3721 case HNS3_OL4_TYPE_NVGRE:
3722 skb->csum_level = 1;
3723 fallthrough;
3724 case HNS3_OL4_TYPE_NO_TUN:
3725 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
3726 HNS3_RXD_L3ID_S);
3727 l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
3728 HNS3_RXD_L4ID_S);
3729 /* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
3730 if ((l3_type == HNS3_L3_TYPE_IPV4 ||
3731 l3_type == HNS3_L3_TYPE_IPV6) &&
3732 (l4_type == HNS3_L4_TYPE_UDP ||
3733 l4_type == HNS3_L4_TYPE_TCP ||
3734 l4_type == HNS3_L4_TYPE_SCTP))
3735 skb->ip_summed = CHECKSUM_UNNECESSARY;
3736 break;
3737 default:
3738 break;
3739 }
3740}
3741
3742static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
3743 u32 l234info, u32 bd_base_info, u32 ol_info,
3744 u16 csum)
3745{
3746 struct net_device *netdev = ring_to_netdev(ring);
3747 struct hns3_nic_priv *priv = netdev_priv(netdev);
3748 u32 ptype = HNS3_INVALID_PTYPE;
3749
3750 skb->ip_summed = CHECKSUM_NONE;
3751
3752 skb_checksum_none_assert(skb);
3753
3754 if (!(netdev->features & NETIF_F_RXCSUM))
3755 return;
3756
3757 if (test_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state))
3758 ptype = hnae3_get_field(ol_info, HNS3_RXD_PTYPE_M,
3759 HNS3_RXD_PTYPE_S);
3760
3761 if (hns3_checksum_complete(ring, skb, ptype, csum))
3762 return;
3763
3764 /* check if hardware has done checksum */
3765 if (!(bd_base_info & BIT(HNS3_RXD_L3L4P_B)))
3766 return;
3767
3768 if (unlikely(l234info & (BIT(HNS3_RXD_L3E_B) | BIT(HNS3_RXD_L4E_B) |
3769 BIT(HNS3_RXD_OL3E_B) |
3770 BIT(HNS3_RXD_OL4E_B)))) {
3771 u64_stats_update_begin(&ring->syncp);
3772 ring->stats.l3l4_csum_err++;
3773 u64_stats_update_end(&ring->syncp);
3774
3775 return;
3776 }
3777
3778 hns3_rx_handle_csum(skb, l234info, ol_info, ptype);
3779}
3780
3781static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
3782{
3783 if (skb_has_frag_list(skb))
3784 napi_gro_flush(&ring->tqp_vector->napi, false);
3785
3786 napi_gro_receive(&ring->tqp_vector->napi, skb);
3787}
3788
3789static bool hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
3790 struct hns3_desc *desc, u32 l234info,
3791 u16 *vlan_tag)
3792{
3793 struct hnae3_handle *handle = ring->tqp->handle;
3794 struct pci_dev *pdev = ring->tqp->handle->pdev;
3795 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3796
3797 if (unlikely(ae_dev->dev_version < HNAE3_DEVICE_VERSION_V2)) {
3798 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3799 if (!(*vlan_tag & VLAN_VID_MASK))
3800 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3801
3802 return (*vlan_tag != 0);
3803 }
3804
3805#define HNS3_STRP_OUTER_VLAN 0x1
3806#define HNS3_STRP_INNER_VLAN 0x2
3807#define HNS3_STRP_BOTH 0x3
3808
3809 /* Hardware always insert VLAN tag into RX descriptor when
3810 * remove the tag from packet, driver needs to determine
3811 * reporting which tag to stack.
3812 */
3813 switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
3814 HNS3_RXD_STRP_TAGP_S)) {
3815 case HNS3_STRP_OUTER_VLAN:
3816 if (handle->port_base_vlan_state !=
3817 HNAE3_PORT_BASE_VLAN_DISABLE)
3818 return false;
3819
3820 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3821 return true;
3822 case HNS3_STRP_INNER_VLAN:
3823 if (handle->port_base_vlan_state !=
3824 HNAE3_PORT_BASE_VLAN_DISABLE)
3825 return false;
3826
3827 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3828 return true;
3829 case HNS3_STRP_BOTH:
3830 if (handle->port_base_vlan_state ==
3831 HNAE3_PORT_BASE_VLAN_DISABLE)
3832 *vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3833 else
3834 *vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3835
3836 return true;
3837 default:
3838 return false;
3839 }
3840}
3841
3842static void hns3_rx_ring_move_fw(struct hns3_enet_ring *ring)
3843{
3844 ring->desc[ring->next_to_clean].rx.bd_base_info &=
3845 cpu_to_le32(~BIT(HNS3_RXD_VLD_B));
3846 ring->desc_cb[ring->next_to_clean].refill = 0;
3847 ring->next_to_clean += 1;
3848
3849 if (unlikely(ring->next_to_clean == ring->desc_num))
3850 ring->next_to_clean = 0;
3851}
3852
3853static int hns3_alloc_skb(struct hns3_enet_ring *ring, unsigned int length,
3854 unsigned char *va)
3855{
3856 struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
3857 struct net_device *netdev = ring_to_netdev(ring);
3858 struct sk_buff *skb;
3859
3860 ring->skb = napi_alloc_skb(&ring->tqp_vector->napi, HNS3_RX_HEAD_SIZE);
3861 skb = ring->skb;
3862 if (unlikely(!skb)) {
3863 hns3_rl_err(netdev, "alloc rx skb fail\n");
3864
3865 u64_stats_update_begin(&ring->syncp);
3866 ring->stats.sw_err_cnt++;
3867 u64_stats_update_end(&ring->syncp);
3868
3869 return -ENOMEM;
3870 }
3871
3872 trace_hns3_rx_desc(ring);
3873 prefetchw(skb->data);
3874
3875 ring->pending_buf = 1;
3876 ring->frag_num = 0;
3877 ring->tail_skb = NULL;
3878 if (length <= HNS3_RX_HEAD_SIZE) {
3879 memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
3880
3881 /* We can reuse buffer as-is, just make sure it is reusable */
3882 if (dev_page_is_reusable(desc_cb->priv))
3883 desc_cb->reuse_flag = 1;
3884 else if (desc_cb->type & DESC_TYPE_PP_FRAG)
3885 page_pool_put_full_page(ring->page_pool, desc_cb->priv,
3886 false);
3887 else /* This page cannot be reused so discard it */
3888 __page_frag_cache_drain(desc_cb->priv,
3889 desc_cb->pagecnt_bias);
3890
3891 hns3_rx_ring_move_fw(ring);
3892 return 0;
3893 }
3894
3895 if (ring->page_pool)
3896 skb_mark_for_recycle(skb);
3897
3898 u64_stats_update_begin(&ring->syncp);
3899 ring->stats.seg_pkt_cnt++;
3900 u64_stats_update_end(&ring->syncp);
3901
3902 ring->pull_len = eth_get_headlen(netdev, va, HNS3_RX_HEAD_SIZE);
3903 __skb_put(skb, ring->pull_len);
3904 hns3_nic_reuse_page(skb, ring->frag_num++, ring, ring->pull_len,
3905 desc_cb);
3906 hns3_rx_ring_move_fw(ring);
3907
3908 return 0;
3909}
3910
3911static int hns3_add_frag(struct hns3_enet_ring *ring)
3912{
3913 struct sk_buff *skb = ring->skb;
3914 struct sk_buff *head_skb = skb;
3915 struct sk_buff *new_skb;
3916 struct hns3_desc_cb *desc_cb;
3917 struct hns3_desc *desc;
3918 u32 bd_base_info;
3919
3920 do {
3921 desc = &ring->desc[ring->next_to_clean];
3922 desc_cb = &ring->desc_cb[ring->next_to_clean];
3923 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
3924 /* make sure HW write desc complete */
3925 dma_rmb();
3926 if (!(bd_base_info & BIT(HNS3_RXD_VLD_B)))
3927 return -ENXIO;
3928
3929 if (unlikely(ring->frag_num >= MAX_SKB_FRAGS)) {
3930 new_skb = napi_alloc_skb(&ring->tqp_vector->napi, 0);
3931 if (unlikely(!new_skb)) {
3932 hns3_rl_err(ring_to_netdev(ring),
3933 "alloc rx fraglist skb fail\n");
3934 return -ENXIO;
3935 }
3936
3937 if (ring->page_pool)
3938 skb_mark_for_recycle(new_skb);
3939
3940 ring->frag_num = 0;
3941
3942 if (ring->tail_skb) {
3943 ring->tail_skb->next = new_skb;
3944 ring->tail_skb = new_skb;
3945 } else {
3946 skb_shinfo(skb)->frag_list = new_skb;
3947 ring->tail_skb = new_skb;
3948 }
3949 }
3950
3951 if (ring->tail_skb) {
3952 head_skb->truesize += hns3_buf_size(ring);
3953 head_skb->data_len += le16_to_cpu(desc->rx.size);
3954 head_skb->len += le16_to_cpu(desc->rx.size);
3955 skb = ring->tail_skb;
3956 }
3957
3958 dma_sync_single_for_cpu(ring_to_dev(ring),
3959 desc_cb->dma + desc_cb->page_offset,
3960 hns3_buf_size(ring),
3961 DMA_FROM_DEVICE);
3962
3963 hns3_nic_reuse_page(skb, ring->frag_num++, ring, 0, desc_cb);
3964 trace_hns3_rx_desc(ring);
3965 hns3_rx_ring_move_fw(ring);
3966 ring->pending_buf++;
3967 } while (!(bd_base_info & BIT(HNS3_RXD_FE_B)));
3968
3969 return 0;
3970}
3971
3972static int hns3_set_gro_and_checksum(struct hns3_enet_ring *ring,
3973 struct sk_buff *skb, u32 l234info,
3974 u32 bd_base_info, u32 ol_info, u16 csum)
3975{
3976 struct net_device *netdev = ring_to_netdev(ring);
3977 struct hns3_nic_priv *priv = netdev_priv(netdev);
3978 u32 l3_type;
3979
3980 skb_shinfo(skb)->gso_size = hnae3_get_field(bd_base_info,
3981 HNS3_RXD_GRO_SIZE_M,
3982 HNS3_RXD_GRO_SIZE_S);
3983 /* if there is no HW GRO, do not set gro params */
3984 if (!skb_shinfo(skb)->gso_size) {
3985 hns3_rx_checksum(ring, skb, l234info, bd_base_info, ol_info,
3986 csum);
3987 return 0;
3988 }
3989
3990 NAPI_GRO_CB(skb)->count = hnae3_get_field(l234info,
3991 HNS3_RXD_GRO_COUNT_M,
3992 HNS3_RXD_GRO_COUNT_S);
3993
3994 if (test_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state)) {
3995 u32 ptype = hnae3_get_field(ol_info, HNS3_RXD_PTYPE_M,
3996 HNS3_RXD_PTYPE_S);
3997
3998 l3_type = hns3_rx_ptype_tbl[ptype].l3_type;
3999 } else {
4000 l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
4001 HNS3_RXD_L3ID_S);
4002 }
4003
4004 if (l3_type == HNS3_L3_TYPE_IPV4)
4005 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
4006 else if (l3_type == HNS3_L3_TYPE_IPV6)
4007 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
4008 else
4009 return -EFAULT;
4010
4011 return hns3_gro_complete(skb, l234info);
4012}
4013
4014static void hns3_set_rx_skb_rss_type(struct hns3_enet_ring *ring,
4015 struct sk_buff *skb, u32 rss_hash)
4016{
4017 struct hnae3_handle *handle = ring->tqp->handle;
4018 enum pkt_hash_types rss_type;
4019
4020 if (rss_hash)
4021 rss_type = handle->kinfo.rss_type;
4022 else
4023 rss_type = PKT_HASH_TYPE_NONE;
4024
4025 skb_set_hash(skb, rss_hash, rss_type);
4026}
4027
4028static int hns3_handle_bdinfo(struct hns3_enet_ring *ring, struct sk_buff *skb)
4029{
4030 struct net_device *netdev = ring_to_netdev(ring);
4031 enum hns3_pkt_l2t_type l2_frame_type;
4032 u32 bd_base_info, l234info, ol_info;
4033 struct hns3_desc *desc;
4034 unsigned int len;
4035 int pre_ntc, ret;
4036 u16 csum;
4037
4038 /* bdinfo handled below is only valid on the last BD of the
4039 * current packet, and ring->next_to_clean indicates the first
4040 * descriptor of next packet, so need - 1 below.
4041 */
4042 pre_ntc = ring->next_to_clean ? (ring->next_to_clean - 1) :
4043 (ring->desc_num - 1);
4044 desc = &ring->desc[pre_ntc];
4045 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
4046 l234info = le32_to_cpu(desc->rx.l234_info);
4047 ol_info = le32_to_cpu(desc->rx.ol_info);
4048 csum = le16_to_cpu(desc->csum);
4049
4050 if (unlikely(bd_base_info & BIT(HNS3_RXD_TS_VLD_B))) {
4051 struct hnae3_handle *h = hns3_get_handle(netdev);
4052 u32 nsec = le32_to_cpu(desc->ts_nsec);
4053 u32 sec = le32_to_cpu(desc->ts_sec);
4054
4055 if (h->ae_algo->ops->get_rx_hwts)
4056 h->ae_algo->ops->get_rx_hwts(h, skb, nsec, sec);
4057 }
4058
4059 /* Based on hw strategy, the tag offloaded will be stored at
4060 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
4061 * in one layer tag case.
4062 */
4063 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
4064 u16 vlan_tag;
4065
4066 if (hns3_parse_vlan_tag(ring, desc, l234info, &vlan_tag))
4067 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
4068 vlan_tag);
4069 }
4070
4071 if (unlikely(!desc->rx.pkt_len || (l234info & (BIT(HNS3_RXD_TRUNCAT_B) |
4072 BIT(HNS3_RXD_L2E_B))))) {
4073 u64_stats_update_begin(&ring->syncp);
4074 if (l234info & BIT(HNS3_RXD_L2E_B))
4075 ring->stats.l2_err++;
4076 else
4077 ring->stats.err_pkt_len++;
4078 u64_stats_update_end(&ring->syncp);
4079
4080 return -EFAULT;
4081 }
4082
4083 len = skb->len;
4084
4085 /* Do update ip stack process */
4086 skb->protocol = eth_type_trans(skb, netdev);
4087
4088 /* This is needed in order to enable forwarding support */
4089 ret = hns3_set_gro_and_checksum(ring, skb, l234info,
4090 bd_base_info, ol_info, csum);
4091 if (unlikely(ret)) {
4092 u64_stats_update_begin(&ring->syncp);
4093 ring->stats.rx_err_cnt++;
4094 u64_stats_update_end(&ring->syncp);
4095 return ret;
4096 }
4097
4098 l2_frame_type = hnae3_get_field(l234info, HNS3_RXD_DMAC_M,
4099 HNS3_RXD_DMAC_S);
4100
4101 u64_stats_update_begin(&ring->syncp);
4102 ring->stats.rx_pkts++;
4103 ring->stats.rx_bytes += len;
4104
4105 if (l2_frame_type == HNS3_L2_TYPE_MULTICAST)
4106 ring->stats.rx_multicast++;
4107
4108 u64_stats_update_end(&ring->syncp);
4109
4110 ring->tqp_vector->rx_group.total_bytes += len;
4111
4112 hns3_set_rx_skb_rss_type(ring, skb, le32_to_cpu(desc->rx.rss_hash));
4113 return 0;
4114}
4115
4116static int hns3_handle_rx_bd(struct hns3_enet_ring *ring)
4117{
4118 struct sk_buff *skb = ring->skb;
4119 struct hns3_desc_cb *desc_cb;
4120 struct hns3_desc *desc;
4121 unsigned int length;
4122 u32 bd_base_info;
4123 int ret;
4124
4125 desc = &ring->desc[ring->next_to_clean];
4126 desc_cb = &ring->desc_cb[ring->next_to_clean];
4127
4128 prefetch(desc);
4129
4130 if (!skb) {
4131 bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
4132 /* Check valid BD */
4133 if (unlikely(!(bd_base_info & BIT(HNS3_RXD_VLD_B))))
4134 return -ENXIO;
4135
4136 dma_rmb();
4137 length = le16_to_cpu(desc->rx.size);
4138
4139 ring->va = desc_cb->buf + desc_cb->page_offset;
4140
4141 dma_sync_single_for_cpu(ring_to_dev(ring),
4142 desc_cb->dma + desc_cb->page_offset,
4143 hns3_buf_size(ring),
4144 DMA_FROM_DEVICE);
4145
4146 /* Prefetch first cache line of first page.
4147 * Idea is to cache few bytes of the header of the packet.
4148 * Our L1 Cache line size is 64B so need to prefetch twice to make
4149 * it 128B. But in actual we can have greater size of caches with
4150 * 128B Level 1 cache lines. In such a case, single fetch would
4151 * suffice to cache in the relevant part of the header.
4152 */
4153 net_prefetch(ring->va);
4154
4155 ret = hns3_alloc_skb(ring, length, ring->va);
4156 skb = ring->skb;
4157
4158 if (ret < 0) /* alloc buffer fail */
4159 return ret;
4160 if (!(bd_base_info & BIT(HNS3_RXD_FE_B))) { /* need add frag */
4161 ret = hns3_add_frag(ring);
4162 if (ret)
4163 return ret;
4164 }
4165 } else {
4166 ret = hns3_add_frag(ring);
4167 if (ret)
4168 return ret;
4169 }
4170
4171 /* As the head data may be changed when GRO enable, copy
4172 * the head data in after other data rx completed
4173 */
4174 if (skb->len > HNS3_RX_HEAD_SIZE)
4175 memcpy(skb->data, ring->va,
4176 ALIGN(ring->pull_len, sizeof(long)));
4177
4178 ret = hns3_handle_bdinfo(ring, skb);
4179 if (unlikely(ret)) {
4180 dev_kfree_skb_any(skb);
4181 return ret;
4182 }
4183
4184 skb_record_rx_queue(skb, ring->tqp->tqp_index);
4185 return 0;
4186}
4187
4188int hns3_clean_rx_ring(struct hns3_enet_ring *ring, int budget,
4189 void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
4190{
4191#define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
4192 int unused_count = hns3_desc_unused(ring);
4193 bool failure = false;
4194 int recv_pkts = 0;
4195 int err;
4196
4197 unused_count -= ring->pending_buf;
4198
4199 while (recv_pkts < budget) {
4200 /* Reuse or realloc buffers */
4201 if (unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
4202 failure = failure ||
4203 hns3_nic_alloc_rx_buffers(ring, unused_count);
4204 unused_count = 0;
4205 }
4206
4207 /* Poll one pkt */
4208 err = hns3_handle_rx_bd(ring);
4209 /* Do not get FE for the packet or failed to alloc skb */
4210 if (unlikely(!ring->skb || err == -ENXIO)) {
4211 goto out;
4212 } else if (likely(!err)) {
4213 rx_fn(ring, ring->skb);
4214 recv_pkts++;
4215 }
4216
4217 unused_count += ring->pending_buf;
4218 ring->skb = NULL;
4219 ring->pending_buf = 0;
4220 }
4221
4222out:
4223 /* sync head pointer before exiting, since hardware will calculate
4224 * FBD number with head pointer
4225 */
4226 if (unused_count > 0)
4227 failure = failure ||
4228 hns3_nic_alloc_rx_buffers(ring, unused_count);
4229
4230 return failure ? budget : recv_pkts;
4231}
4232
4233static void hns3_update_rx_int_coalesce(struct hns3_enet_tqp_vector *tqp_vector)
4234{
4235 struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
4236 struct dim_sample sample = {};
4237
4238 if (!rx_group->coal.adapt_enable)
4239 return;
4240
4241 dim_update_sample(tqp_vector->event_cnt, rx_group->total_packets,
4242 rx_group->total_bytes, &sample);
4243 net_dim(&rx_group->dim, sample);
4244}
4245
4246static void hns3_update_tx_int_coalesce(struct hns3_enet_tqp_vector *tqp_vector)
4247{
4248 struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
4249 struct dim_sample sample = {};
4250
4251 if (!tx_group->coal.adapt_enable)
4252 return;
4253
4254 dim_update_sample(tqp_vector->event_cnt, tx_group->total_packets,
4255 tx_group->total_bytes, &sample);
4256 net_dim(&tx_group->dim, sample);
4257}
4258
4259static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
4260{
4261 struct hns3_nic_priv *priv = netdev_priv(napi->dev);
4262 struct hns3_enet_ring *ring;
4263 int rx_pkt_total = 0;
4264
4265 struct hns3_enet_tqp_vector *tqp_vector =
4266 container_of(napi, struct hns3_enet_tqp_vector, napi);
4267 bool clean_complete = true;
4268 int rx_budget = budget;
4269
4270 if (unlikely(test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
4271 napi_complete(napi);
4272 return 0;
4273 }
4274
4275 /* Since the actual Tx work is minimal, we can give the Tx a larger
4276 * budget and be more aggressive about cleaning up the Tx descriptors.
4277 */
4278 hns3_for_each_ring(ring, tqp_vector->tx_group)
4279 hns3_clean_tx_ring(ring, budget);
4280
4281 /* make sure rx ring budget not smaller than 1 */
4282 if (tqp_vector->num_tqps > 1)
4283 rx_budget = max(budget / tqp_vector->num_tqps, 1);
4284
4285 hns3_for_each_ring(ring, tqp_vector->rx_group) {
4286 int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
4287 hns3_rx_skb);
4288 if (rx_cleaned >= rx_budget)
4289 clean_complete = false;
4290
4291 rx_pkt_total += rx_cleaned;
4292 }
4293
4294 tqp_vector->rx_group.total_packets += rx_pkt_total;
4295
4296 if (!clean_complete)
4297 return budget;
4298
4299 if (napi_complete(napi) &&
4300 likely(!test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
4301 hns3_update_rx_int_coalesce(tqp_vector);
4302 hns3_update_tx_int_coalesce(tqp_vector);
4303
4304 hns3_mask_vector_irq(tqp_vector, 1);
4305 }
4306
4307 return rx_pkt_total;
4308}
4309
4310static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
4311 struct hnae3_ring_chain_node *head)
4312{
4313 struct pci_dev *pdev = tqp_vector->handle->pdev;
4314 struct hnae3_ring_chain_node *cur_chain = head;
4315 struct hnae3_ring_chain_node *chain;
4316 struct hns3_enet_ring *tx_ring;
4317 struct hns3_enet_ring *rx_ring;
4318
4319 tx_ring = tqp_vector->tx_group.ring;
4320 if (tx_ring) {
4321 cur_chain->tqp_index = tx_ring->tqp->tqp_index;
4322 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
4323 HNAE3_RING_TYPE_TX);
4324 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
4325 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_TX);
4326
4327 cur_chain->next = NULL;
4328
4329 while (tx_ring->next) {
4330 tx_ring = tx_ring->next;
4331
4332 chain = devm_kzalloc(&pdev->dev, sizeof(*chain),
4333 GFP_KERNEL);
4334 if (!chain)
4335 goto err_free_chain;
4336
4337 cur_chain->next = chain;
4338 chain->tqp_index = tx_ring->tqp->tqp_index;
4339 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
4340 HNAE3_RING_TYPE_TX);
4341 hnae3_set_field(chain->int_gl_idx,
4342 HNAE3_RING_GL_IDX_M,
4343 HNAE3_RING_GL_IDX_S,
4344 HNAE3_RING_GL_TX);
4345
4346 cur_chain = chain;
4347 }
4348 }
4349
4350 rx_ring = tqp_vector->rx_group.ring;
4351 if (!tx_ring && rx_ring) {
4352 cur_chain->next = NULL;
4353 cur_chain->tqp_index = rx_ring->tqp->tqp_index;
4354 hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
4355 HNAE3_RING_TYPE_RX);
4356 hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
4357 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
4358
4359 rx_ring = rx_ring->next;
4360 }
4361
4362 while (rx_ring) {
4363 chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
4364 if (!chain)
4365 goto err_free_chain;
4366
4367 cur_chain->next = chain;
4368 chain->tqp_index = rx_ring->tqp->tqp_index;
4369 hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
4370 HNAE3_RING_TYPE_RX);
4371 hnae3_set_field(chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
4372 HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
4373
4374 cur_chain = chain;
4375
4376 rx_ring = rx_ring->next;
4377 }
4378
4379 return 0;
4380
4381err_free_chain:
4382 cur_chain = head->next;
4383 while (cur_chain) {
4384 chain = cur_chain->next;
4385 devm_kfree(&pdev->dev, cur_chain);
4386 cur_chain = chain;
4387 }
4388 head->next = NULL;
4389
4390 return -ENOMEM;
4391}
4392
4393static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
4394 struct hnae3_ring_chain_node *head)
4395{
4396 struct pci_dev *pdev = tqp_vector->handle->pdev;
4397 struct hnae3_ring_chain_node *chain_tmp, *chain;
4398
4399 chain = head->next;
4400
4401 while (chain) {
4402 chain_tmp = chain->next;
4403 devm_kfree(&pdev->dev, chain);
4404 chain = chain_tmp;
4405 }
4406}
4407
4408static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
4409 struct hns3_enet_ring *ring)
4410{
4411 ring->next = group->ring;
4412 group->ring = ring;
4413
4414 group->count++;
4415}
4416
4417static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
4418{
4419 struct pci_dev *pdev = priv->ae_handle->pdev;
4420 struct hns3_enet_tqp_vector *tqp_vector;
4421 int num_vectors = priv->vector_num;
4422 int numa_node;
4423 int vector_i;
4424
4425 numa_node = dev_to_node(&pdev->dev);
4426
4427 for (vector_i = 0; vector_i < num_vectors; vector_i++) {
4428 tqp_vector = &priv->tqp_vector[vector_i];
4429 cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
4430 &tqp_vector->affinity_mask);
4431 }
4432}
4433
4434static void hns3_rx_dim_work(struct work_struct *work)
4435{
4436 struct dim *dim = container_of(work, struct dim, work);
4437 struct hns3_enet_ring_group *group = container_of(dim,
4438 struct hns3_enet_ring_group, dim);
4439 struct hns3_enet_tqp_vector *tqp_vector = group->ring->tqp_vector;
4440 struct dim_cq_moder cur_moder =
4441 net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
4442
4443 hns3_set_vector_coalesce_rx_gl(group->ring->tqp_vector, cur_moder.usec);
4444 tqp_vector->rx_group.coal.int_gl = cur_moder.usec;
4445
4446 if (cur_moder.pkts < tqp_vector->rx_group.coal.int_ql_max) {
4447 hns3_set_vector_coalesce_rx_ql(tqp_vector, cur_moder.pkts);
4448 tqp_vector->rx_group.coal.int_ql = cur_moder.pkts;
4449 }
4450
4451 dim->state = DIM_START_MEASURE;
4452}
4453
4454static void hns3_tx_dim_work(struct work_struct *work)
4455{
4456 struct dim *dim = container_of(work, struct dim, work);
4457 struct hns3_enet_ring_group *group = container_of(dim,
4458 struct hns3_enet_ring_group, dim);
4459 struct hns3_enet_tqp_vector *tqp_vector = group->ring->tqp_vector;
4460 struct dim_cq_moder cur_moder =
4461 net_dim_get_tx_moderation(dim->mode, dim->profile_ix);
4462
4463 hns3_set_vector_coalesce_tx_gl(tqp_vector, cur_moder.usec);
4464 tqp_vector->tx_group.coal.int_gl = cur_moder.usec;
4465
4466 if (cur_moder.pkts < tqp_vector->tx_group.coal.int_ql_max) {
4467 hns3_set_vector_coalesce_tx_ql(tqp_vector, cur_moder.pkts);
4468 tqp_vector->tx_group.coal.int_ql = cur_moder.pkts;
4469 }
4470
4471 dim->state = DIM_START_MEASURE;
4472}
4473
4474static void hns3_nic_init_dim(struct hns3_enet_tqp_vector *tqp_vector)
4475{
4476 INIT_WORK(&tqp_vector->rx_group.dim.work, hns3_rx_dim_work);
4477 INIT_WORK(&tqp_vector->tx_group.dim.work, hns3_tx_dim_work);
4478}
4479
4480static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
4481{
4482 struct hnae3_handle *h = priv->ae_handle;
4483 struct hns3_enet_tqp_vector *tqp_vector;
4484 int ret;
4485 int i;
4486
4487 hns3_nic_set_cpumask(priv);
4488
4489 for (i = 0; i < priv->vector_num; i++) {
4490 tqp_vector = &priv->tqp_vector[i];
4491 hns3_vector_coalesce_init_hw(tqp_vector, priv);
4492 tqp_vector->num_tqps = 0;
4493 hns3_nic_init_dim(tqp_vector);
4494 }
4495
4496 for (i = 0; i < h->kinfo.num_tqps; i++) {
4497 u16 vector_i = i % priv->vector_num;
4498 u16 tqp_num = h->kinfo.num_tqps;
4499
4500 tqp_vector = &priv->tqp_vector[vector_i];
4501
4502 hns3_add_ring_to_group(&tqp_vector->tx_group,
4503 &priv->ring[i]);
4504
4505 hns3_add_ring_to_group(&tqp_vector->rx_group,
4506 &priv->ring[i + tqp_num]);
4507
4508 priv->ring[i].tqp_vector = tqp_vector;
4509 priv->ring[i + tqp_num].tqp_vector = tqp_vector;
4510 tqp_vector->num_tqps++;
4511 }
4512
4513 for (i = 0; i < priv->vector_num; i++) {
4514 struct hnae3_ring_chain_node vector_ring_chain;
4515
4516 tqp_vector = &priv->tqp_vector[i];
4517
4518 tqp_vector->rx_group.total_bytes = 0;
4519 tqp_vector->rx_group.total_packets = 0;
4520 tqp_vector->tx_group.total_bytes = 0;
4521 tqp_vector->tx_group.total_packets = 0;
4522 tqp_vector->handle = h;
4523
4524 ret = hns3_get_vector_ring_chain(tqp_vector,
4525 &vector_ring_chain);
4526 if (ret)
4527 goto map_ring_fail;
4528
4529 ret = h->ae_algo->ops->map_ring_to_vector(h,
4530 tqp_vector->vector_irq, &vector_ring_chain);
4531
4532 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
4533
4534 if (ret)
4535 goto map_ring_fail;
4536
4537 netif_napi_add(priv->netdev, &tqp_vector->napi,
4538 hns3_nic_common_poll, NAPI_POLL_WEIGHT);
4539 }
4540
4541 return 0;
4542
4543map_ring_fail:
4544 while (i--)
4545 netif_napi_del(&priv->tqp_vector[i].napi);
4546
4547 return ret;
4548}
4549
4550static void hns3_nic_init_coal_cfg(struct hns3_nic_priv *priv)
4551{
4552 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
4553 struct hns3_enet_coalesce *tx_coal = &priv->tx_coal;
4554 struct hns3_enet_coalesce *rx_coal = &priv->rx_coal;
4555
4556 /* initialize the configuration for interrupt coalescing.
4557 * 1. GL (Interrupt Gap Limiter)
4558 * 2. RL (Interrupt Rate Limiter)
4559 * 3. QL (Interrupt Quantity Limiter)
4560 *
4561 * Default: enable interrupt coalescing self-adaptive and GL
4562 */
4563 tx_coal->adapt_enable = 1;
4564 rx_coal->adapt_enable = 1;
4565
4566 tx_coal->int_gl = HNS3_INT_GL_50K;
4567 rx_coal->int_gl = HNS3_INT_GL_50K;
4568
4569 rx_coal->flow_level = HNS3_FLOW_LOW;
4570 tx_coal->flow_level = HNS3_FLOW_LOW;
4571
4572 if (ae_dev->dev_specs.int_ql_max) {
4573 tx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
4574 rx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
4575 }
4576}
4577
4578static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
4579{
4580 struct hnae3_handle *h = priv->ae_handle;
4581 struct hns3_enet_tqp_vector *tqp_vector;
4582 struct hnae3_vector_info *vector;
4583 struct pci_dev *pdev = h->pdev;
4584 u16 tqp_num = h->kinfo.num_tqps;
4585 u16 vector_num;
4586 int ret = 0;
4587 u16 i;
4588
4589 /* RSS size, cpu online and vector_num should be the same */
4590 /* Should consider 2p/4p later */
4591 vector_num = min_t(u16, num_online_cpus(), tqp_num);
4592
4593 vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
4594 GFP_KERNEL);
4595 if (!vector)
4596 return -ENOMEM;
4597
4598 /* save the actual available vector number */
4599 vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
4600
4601 priv->vector_num = vector_num;
4602 priv->tqp_vector = (struct hns3_enet_tqp_vector *)
4603 devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
4604 GFP_KERNEL);
4605 if (!priv->tqp_vector) {
4606 ret = -ENOMEM;
4607 goto out;
4608 }
4609
4610 for (i = 0; i < priv->vector_num; i++) {
4611 tqp_vector = &priv->tqp_vector[i];
4612 tqp_vector->idx = i;
4613 tqp_vector->mask_addr = vector[i].io_addr;
4614 tqp_vector->vector_irq = vector[i].vector;
4615 hns3_vector_coalesce_init(tqp_vector, priv);
4616 }
4617
4618out:
4619 devm_kfree(&pdev->dev, vector);
4620 return ret;
4621}
4622
4623static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
4624{
4625 group->ring = NULL;
4626 group->count = 0;
4627}
4628
4629static void hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
4630{
4631 struct hnae3_ring_chain_node vector_ring_chain;
4632 struct hnae3_handle *h = priv->ae_handle;
4633 struct hns3_enet_tqp_vector *tqp_vector;
4634 int i;
4635
4636 for (i = 0; i < priv->vector_num; i++) {
4637 tqp_vector = &priv->tqp_vector[i];
4638
4639 if (!tqp_vector->rx_group.ring && !tqp_vector->tx_group.ring)
4640 continue;
4641
4642 /* Since the mapping can be overwritten, when fail to get the
4643 * chain between vector and ring, we should go on to deal with
4644 * the remaining options.
4645 */
4646 if (hns3_get_vector_ring_chain(tqp_vector, &vector_ring_chain))
4647 dev_warn(priv->dev, "failed to get ring chain\n");
4648
4649 h->ae_algo->ops->unmap_ring_from_vector(h,
4650 tqp_vector->vector_irq, &vector_ring_chain);
4651
4652 hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
4653
4654 hns3_clear_ring_group(&tqp_vector->rx_group);
4655 hns3_clear_ring_group(&tqp_vector->tx_group);
4656 netif_napi_del(&priv->tqp_vector[i].napi);
4657 }
4658}
4659
4660static void hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
4661{
4662 struct hnae3_handle *h = priv->ae_handle;
4663 struct pci_dev *pdev = h->pdev;
4664 int i, ret;
4665
4666 for (i = 0; i < priv->vector_num; i++) {
4667 struct hns3_enet_tqp_vector *tqp_vector;
4668
4669 tqp_vector = &priv->tqp_vector[i];
4670 ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
4671 if (ret)
4672 return;
4673 }
4674
4675 devm_kfree(&pdev->dev, priv->tqp_vector);
4676}
4677
4678static void hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
4679 unsigned int ring_type)
4680{
4681 int queue_num = priv->ae_handle->kinfo.num_tqps;
4682 struct hns3_enet_ring *ring;
4683 int desc_num;
4684
4685 if (ring_type == HNAE3_RING_TYPE_TX) {
4686 ring = &priv->ring[q->tqp_index];
4687 desc_num = priv->ae_handle->kinfo.num_tx_desc;
4688 ring->queue_index = q->tqp_index;
4689 ring->tx_copybreak = priv->tx_copybreak;
4690 ring->last_to_use = 0;
4691 } else {
4692 ring = &priv->ring[q->tqp_index + queue_num];
4693 desc_num = priv->ae_handle->kinfo.num_rx_desc;
4694 ring->queue_index = q->tqp_index;
4695 ring->rx_copybreak = priv->rx_copybreak;
4696 }
4697
4698 hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
4699
4700 ring->tqp = q;
4701 ring->desc = NULL;
4702 ring->desc_cb = NULL;
4703 ring->dev = priv->dev;
4704 ring->desc_dma_addr = 0;
4705 ring->buf_size = q->buf_size;
4706 ring->desc_num = desc_num;
4707 ring->next_to_use = 0;
4708 ring->next_to_clean = 0;
4709}
4710
4711static void hns3_queue_to_ring(struct hnae3_queue *tqp,
4712 struct hns3_nic_priv *priv)
4713{
4714 hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
4715 hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
4716}
4717
4718static int hns3_get_ring_config(struct hns3_nic_priv *priv)
4719{
4720 struct hnae3_handle *h = priv->ae_handle;
4721 struct pci_dev *pdev = h->pdev;
4722 int i;
4723
4724 priv->ring = devm_kzalloc(&pdev->dev,
4725 array3_size(h->kinfo.num_tqps,
4726 sizeof(*priv->ring), 2),
4727 GFP_KERNEL);
4728 if (!priv->ring)
4729 return -ENOMEM;
4730
4731 for (i = 0; i < h->kinfo.num_tqps; i++)
4732 hns3_queue_to_ring(h->kinfo.tqp[i], priv);
4733
4734 return 0;
4735}
4736
4737static void hns3_put_ring_config(struct hns3_nic_priv *priv)
4738{
4739 if (!priv->ring)
4740 return;
4741
4742 devm_kfree(priv->dev, priv->ring);
4743 priv->ring = NULL;
4744}
4745
4746static void hns3_alloc_page_pool(struct hns3_enet_ring *ring)
4747{
4748 struct page_pool_params pp_params = {
4749 .flags = PP_FLAG_DMA_MAP | PP_FLAG_PAGE_FRAG |
4750 PP_FLAG_DMA_SYNC_DEV,
4751 .order = hns3_page_order(ring),
4752 .pool_size = ring->desc_num * hns3_buf_size(ring) /
4753 (PAGE_SIZE << hns3_page_order(ring)),
4754 .nid = dev_to_node(ring_to_dev(ring)),
4755 .dev = ring_to_dev(ring),
4756 .dma_dir = DMA_FROM_DEVICE,
4757 .offset = 0,
4758 .max_len = PAGE_SIZE << hns3_page_order(ring),
4759 };
4760
4761 ring->page_pool = page_pool_create(&pp_params);
4762 if (IS_ERR(ring->page_pool)) {
4763 dev_warn(ring_to_dev(ring), "page pool creation failed: %ld\n",
4764 PTR_ERR(ring->page_pool));
4765 ring->page_pool = NULL;
4766 }
4767}
4768
4769static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
4770{
4771 int ret;
4772
4773 if (ring->desc_num <= 0 || ring->buf_size <= 0)
4774 return -EINVAL;
4775
4776 ring->desc_cb = devm_kcalloc(ring_to_dev(ring), ring->desc_num,
4777 sizeof(ring->desc_cb[0]), GFP_KERNEL);
4778 if (!ring->desc_cb) {
4779 ret = -ENOMEM;
4780 goto out;
4781 }
4782
4783 ret = hns3_alloc_desc(ring);
4784 if (ret)
4785 goto out_with_desc_cb;
4786
4787 if (!HNAE3_IS_TX_RING(ring)) {
4788 if (page_pool_enabled)
4789 hns3_alloc_page_pool(ring);
4790
4791 ret = hns3_alloc_ring_buffers(ring);
4792 if (ret)
4793 goto out_with_desc;
4794 } else {
4795 hns3_init_tx_spare_buffer(ring);
4796 }
4797
4798 return 0;
4799
4800out_with_desc:
4801 hns3_free_desc(ring);
4802out_with_desc_cb:
4803 devm_kfree(ring_to_dev(ring), ring->desc_cb);
4804 ring->desc_cb = NULL;
4805out:
4806 return ret;
4807}
4808
4809void hns3_fini_ring(struct hns3_enet_ring *ring)
4810{
4811 hns3_free_desc(ring);
4812 devm_kfree(ring_to_dev(ring), ring->desc_cb);
4813 ring->desc_cb = NULL;
4814 ring->next_to_clean = 0;
4815 ring->next_to_use = 0;
4816 ring->last_to_use = 0;
4817 ring->pending_buf = 0;
4818 if (!HNAE3_IS_TX_RING(ring) && ring->skb) {
4819 dev_kfree_skb_any(ring->skb);
4820 ring->skb = NULL;
4821 } else if (HNAE3_IS_TX_RING(ring) && ring->tx_spare) {
4822 struct hns3_tx_spare *tx_spare = ring->tx_spare;
4823
4824 dma_unmap_page(ring_to_dev(ring), tx_spare->dma, tx_spare->len,
4825 DMA_TO_DEVICE);
4826 free_pages((unsigned long)tx_spare->buf,
4827 get_order(tx_spare->len));
4828 devm_kfree(ring_to_dev(ring), tx_spare);
4829 ring->tx_spare = NULL;
4830 }
4831
4832 if (!HNAE3_IS_TX_RING(ring) && ring->page_pool) {
4833 page_pool_destroy(ring->page_pool);
4834 ring->page_pool = NULL;
4835 }
4836}
4837
4838static int hns3_buf_size2type(u32 buf_size)
4839{
4840 int bd_size_type;
4841
4842 switch (buf_size) {
4843 case 512:
4844 bd_size_type = HNS3_BD_SIZE_512_TYPE;
4845 break;
4846 case 1024:
4847 bd_size_type = HNS3_BD_SIZE_1024_TYPE;
4848 break;
4849 case 2048:
4850 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4851 break;
4852 case 4096:
4853 bd_size_type = HNS3_BD_SIZE_4096_TYPE;
4854 break;
4855 default:
4856 bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4857 }
4858
4859 return bd_size_type;
4860}
4861
4862static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
4863{
4864 dma_addr_t dma = ring->desc_dma_addr;
4865 struct hnae3_queue *q = ring->tqp;
4866
4867 if (!HNAE3_IS_TX_RING(ring)) {
4868 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG, (u32)dma);
4869 hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
4870 (u32)((dma >> 31) >> 1));
4871
4872 hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
4873 hns3_buf_size2type(ring->buf_size));
4874 hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
4875 ring->desc_num / 8 - 1);
4876 } else {
4877 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
4878 (u32)dma);
4879 hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
4880 (u32)((dma >> 31) >> 1));
4881
4882 hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
4883 ring->desc_num / 8 - 1);
4884 }
4885}
4886
4887static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
4888{
4889 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
4890 struct hnae3_tc_info *tc_info = &kinfo->tc_info;
4891 int i;
4892
4893 for (i = 0; i < tc_info->num_tc; i++) {
4894 int j;
4895
4896 for (j = 0; j < tc_info->tqp_count[i]; j++) {
4897 struct hnae3_queue *q;
4898
4899 q = priv->ring[tc_info->tqp_offset[i] + j].tqp;
4900 hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG, i);
4901 }
4902 }
4903}
4904
4905int hns3_init_all_ring(struct hns3_nic_priv *priv)
4906{
4907 struct hnae3_handle *h = priv->ae_handle;
4908 int ring_num = h->kinfo.num_tqps * 2;
4909 int i, j;
4910 int ret;
4911
4912 for (i = 0; i < ring_num; i++) {
4913 ret = hns3_alloc_ring_memory(&priv->ring[i]);
4914 if (ret) {
4915 dev_err(priv->dev,
4916 "Alloc ring memory fail! ret=%d\n", ret);
4917 goto out_when_alloc_ring_memory;
4918 }
4919
4920 u64_stats_init(&priv->ring[i].syncp);
4921 }
4922
4923 return 0;
4924
4925out_when_alloc_ring_memory:
4926 for (j = i - 1; j >= 0; j--)
4927 hns3_fini_ring(&priv->ring[j]);
4928
4929 return -ENOMEM;
4930}
4931
4932static void hns3_uninit_all_ring(struct hns3_nic_priv *priv)
4933{
4934 struct hnae3_handle *h = priv->ae_handle;
4935 int i;
4936
4937 for (i = 0; i < h->kinfo.num_tqps; i++) {
4938 hns3_fini_ring(&priv->ring[i]);
4939 hns3_fini_ring(&priv->ring[i + h->kinfo.num_tqps]);
4940 }
4941}
4942
4943/* Set mac addr if it is configured. or leave it to the AE driver */
4944static int hns3_init_mac_addr(struct net_device *netdev)
4945{
4946 struct hns3_nic_priv *priv = netdev_priv(netdev);
4947 char format_mac_addr[HNAE3_FORMAT_MAC_ADDR_LEN];
4948 struct hnae3_handle *h = priv->ae_handle;
4949 u8 mac_addr_temp[ETH_ALEN];
4950 int ret = 0;
4951
4952 if (h->ae_algo->ops->get_mac_addr)
4953 h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
4954
4955 /* Check if the MAC address is valid, if not get a random one */
4956 if (!is_valid_ether_addr(mac_addr_temp)) {
4957 eth_hw_addr_random(netdev);
4958 hnae3_format_mac_addr(format_mac_addr, netdev->dev_addr);
4959 dev_warn(priv->dev, "using random MAC address %s\n",
4960 format_mac_addr);
4961 } else if (!ether_addr_equal(netdev->dev_addr, mac_addr_temp)) {
4962 eth_hw_addr_set(netdev, mac_addr_temp);
4963 ether_addr_copy(netdev->perm_addr, mac_addr_temp);
4964 } else {
4965 return 0;
4966 }
4967
4968 if (h->ae_algo->ops->set_mac_addr)
4969 ret = h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
4970
4971 return ret;
4972}
4973
4974static int hns3_init_phy(struct net_device *netdev)
4975{
4976 struct hnae3_handle *h = hns3_get_handle(netdev);
4977 int ret = 0;
4978
4979 if (h->ae_algo->ops->mac_connect_phy)
4980 ret = h->ae_algo->ops->mac_connect_phy(h);
4981
4982 return ret;
4983}
4984
4985static void hns3_uninit_phy(struct net_device *netdev)
4986{
4987 struct hnae3_handle *h = hns3_get_handle(netdev);
4988
4989 if (h->ae_algo->ops->mac_disconnect_phy)
4990 h->ae_algo->ops->mac_disconnect_phy(h);
4991}
4992
4993static int hns3_client_start(struct hnae3_handle *handle)
4994{
4995 if (!handle->ae_algo->ops->client_start)
4996 return 0;
4997
4998 return handle->ae_algo->ops->client_start(handle);
4999}
5000
5001static void hns3_client_stop(struct hnae3_handle *handle)
5002{
5003 if (!handle->ae_algo->ops->client_stop)
5004 return;
5005
5006 handle->ae_algo->ops->client_stop(handle);
5007}
5008
5009static void hns3_info_show(struct hns3_nic_priv *priv)
5010{
5011 struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
5012 char format_mac_addr[HNAE3_FORMAT_MAC_ADDR_LEN];
5013
5014 hnae3_format_mac_addr(format_mac_addr, priv->netdev->dev_addr);
5015 dev_info(priv->dev, "MAC address: %s\n", format_mac_addr);
5016 dev_info(priv->dev, "Task queue pairs numbers: %u\n", kinfo->num_tqps);
5017 dev_info(priv->dev, "RSS size: %u\n", kinfo->rss_size);
5018 dev_info(priv->dev, "Allocated RSS size: %u\n", kinfo->req_rss_size);
5019 dev_info(priv->dev, "RX buffer length: %u\n", kinfo->rx_buf_len);
5020 dev_info(priv->dev, "Desc num per TX queue: %u\n", kinfo->num_tx_desc);
5021 dev_info(priv->dev, "Desc num per RX queue: %u\n", kinfo->num_rx_desc);
5022 dev_info(priv->dev, "Total number of enabled TCs: %u\n",
5023 kinfo->tc_info.num_tc);
5024 dev_info(priv->dev, "Max mtu size: %u\n", priv->netdev->max_mtu);
5025}
5026
5027static void hns3_set_cq_period_mode(struct hns3_nic_priv *priv,
5028 enum dim_cq_period_mode mode, bool is_tx)
5029{
5030 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
5031 struct hnae3_handle *handle = priv->ae_handle;
5032 int i;
5033
5034 if (is_tx) {
5035 priv->tx_cqe_mode = mode;
5036
5037 for (i = 0; i < priv->vector_num; i++)
5038 priv->tqp_vector[i].tx_group.dim.mode = mode;
5039 } else {
5040 priv->rx_cqe_mode = mode;
5041
5042 for (i = 0; i < priv->vector_num; i++)
5043 priv->tqp_vector[i].rx_group.dim.mode = mode;
5044 }
5045
5046 /* only device version above V3(include V3), GL can switch CQ/EQ
5047 * period mode.
5048 */
5049 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3) {
5050 u32 new_mode;
5051 u64 reg;
5052
5053 new_mode = (mode == DIM_CQ_PERIOD_MODE_START_FROM_CQE) ?
5054 HNS3_CQ_MODE_CQE : HNS3_CQ_MODE_EQE;
5055 reg = is_tx ? HNS3_GL1_CQ_MODE_REG : HNS3_GL0_CQ_MODE_REG;
5056
5057 writel(new_mode, handle->kinfo.io_base + reg);
5058 }
5059}
5060
5061void hns3_cq_period_mode_init(struct hns3_nic_priv *priv,
5062 enum dim_cq_period_mode tx_mode,
5063 enum dim_cq_period_mode rx_mode)
5064{
5065 hns3_set_cq_period_mode(priv, tx_mode, true);
5066 hns3_set_cq_period_mode(priv, rx_mode, false);
5067}
5068
5069static void hns3_state_init(struct hnae3_handle *handle)
5070{
5071 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(handle->pdev);
5072 struct net_device *netdev = handle->kinfo.netdev;
5073 struct hns3_nic_priv *priv = netdev_priv(netdev);
5074
5075 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
5076
5077 if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
5078 set_bit(HNAE3_PFLAG_LIMIT_PROMISC, &handle->supported_pflags);
5079
5080 if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps))
5081 set_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state);
5082
5083 if (hnae3_ae_dev_rxd_adv_layout_supported(ae_dev))
5084 set_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state);
5085}
5086
5087static int hns3_client_init(struct hnae3_handle *handle)
5088{
5089 struct pci_dev *pdev = handle->pdev;
5090 struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
5091 u16 alloc_tqps, max_rss_size;
5092 struct hns3_nic_priv *priv;
5093 struct net_device *netdev;
5094 int ret;
5095
5096 handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
5097 &max_rss_size);
5098 netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
5099 if (!netdev)
5100 return -ENOMEM;
5101
5102 priv = netdev_priv(netdev);
5103 priv->dev = &pdev->dev;
5104 priv->netdev = netdev;
5105 priv->ae_handle = handle;
5106 priv->tx_timeout_count = 0;
5107 priv->max_non_tso_bd_num = ae_dev->dev_specs.max_non_tso_bd_num;
5108 set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
5109
5110 handle->msg_enable = netif_msg_init(debug, DEFAULT_MSG_LEVEL);
5111
5112 handle->kinfo.netdev = netdev;
5113 handle->priv = (void *)priv;
5114
5115 hns3_init_mac_addr(netdev);
5116
5117 hns3_set_default_feature(netdev);
5118
5119 netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
5120 netdev->priv_flags |= IFF_UNICAST_FLT;
5121 netdev->netdev_ops = &hns3_nic_netdev_ops;
5122 SET_NETDEV_DEV(netdev, &pdev->dev);
5123 hns3_ethtool_set_ops(netdev);
5124
5125 /* Carrier off reporting is important to ethtool even BEFORE open */
5126 netif_carrier_off(netdev);
5127
5128 ret = hns3_get_ring_config(priv);
5129 if (ret) {
5130 ret = -ENOMEM;
5131 goto out_get_ring_cfg;
5132 }
5133
5134 hns3_nic_init_coal_cfg(priv);
5135
5136 ret = hns3_nic_alloc_vector_data(priv);
5137 if (ret) {
5138 ret = -ENOMEM;
5139 goto out_alloc_vector_data;
5140 }
5141
5142 ret = hns3_nic_init_vector_data(priv);
5143 if (ret) {
5144 ret = -ENOMEM;
5145 goto out_init_vector_data;
5146 }
5147
5148 ret = hns3_init_all_ring(priv);
5149 if (ret) {
5150 ret = -ENOMEM;
5151 goto out_init_ring;
5152 }
5153
5154 hns3_cq_period_mode_init(priv, DIM_CQ_PERIOD_MODE_START_FROM_EQE,
5155 DIM_CQ_PERIOD_MODE_START_FROM_EQE);
5156
5157 ret = hns3_init_phy(netdev);
5158 if (ret)
5159 goto out_init_phy;
5160
5161 /* the device can work without cpu rmap, only aRFS needs it */
5162 ret = hns3_set_rx_cpu_rmap(netdev);
5163 if (ret)
5164 dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
5165
5166 ret = hns3_nic_init_irq(priv);
5167 if (ret) {
5168 dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
5169 hns3_free_rx_cpu_rmap(netdev);
5170 goto out_init_irq_fail;
5171 }
5172
5173 ret = hns3_client_start(handle);
5174 if (ret) {
5175 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
5176 goto out_client_start;
5177 }
5178
5179 hns3_dcbnl_setup(handle);
5180
5181 ret = hns3_dbg_init(handle);
5182 if (ret) {
5183 dev_err(priv->dev, "failed to init debugfs, ret = %d\n",
5184 ret);
5185 goto out_client_start;
5186 }
5187
5188 netdev->max_mtu = HNS3_MAX_MTU(ae_dev->dev_specs.max_frm_size);
5189
5190 hns3_state_init(handle);
5191
5192 ret = register_netdev(netdev);
5193 if (ret) {
5194 dev_err(priv->dev, "probe register netdev fail!\n");
5195 goto out_reg_netdev_fail;
5196 }
5197
5198 if (netif_msg_drv(handle))
5199 hns3_info_show(priv);
5200
5201 return ret;
5202
5203out_reg_netdev_fail:
5204 hns3_dbg_uninit(handle);
5205out_client_start:
5206 hns3_free_rx_cpu_rmap(netdev);
5207 hns3_nic_uninit_irq(priv);
5208out_init_irq_fail:
5209 hns3_uninit_phy(netdev);
5210out_init_phy:
5211 hns3_uninit_all_ring(priv);
5212out_init_ring:
5213 hns3_nic_uninit_vector_data(priv);
5214out_init_vector_data:
5215 hns3_nic_dealloc_vector_data(priv);
5216out_alloc_vector_data:
5217 priv->ring = NULL;
5218out_get_ring_cfg:
5219 priv->ae_handle = NULL;
5220 free_netdev(netdev);
5221 return ret;
5222}
5223
5224static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
5225{
5226 struct net_device *netdev = handle->kinfo.netdev;
5227 struct hns3_nic_priv *priv = netdev_priv(netdev);
5228
5229 if (netdev->reg_state != NETREG_UNINITIALIZED)
5230 unregister_netdev(netdev);
5231
5232 hns3_client_stop(handle);
5233
5234 hns3_uninit_phy(netdev);
5235
5236 if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5237 netdev_warn(netdev, "already uninitialized\n");
5238 goto out_netdev_free;
5239 }
5240
5241 hns3_free_rx_cpu_rmap(netdev);
5242
5243 hns3_nic_uninit_irq(priv);
5244
5245 hns3_clear_all_ring(handle, true);
5246
5247 hns3_nic_uninit_vector_data(priv);
5248
5249 hns3_nic_dealloc_vector_data(priv);
5250
5251 hns3_uninit_all_ring(priv);
5252
5253 hns3_put_ring_config(priv);
5254
5255out_netdev_free:
5256 hns3_dbg_uninit(handle);
5257 free_netdev(netdev);
5258}
5259
5260static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
5261{
5262 struct net_device *netdev = handle->kinfo.netdev;
5263
5264 if (!netdev)
5265 return;
5266
5267 if (linkup) {
5268 netif_tx_wake_all_queues(netdev);
5269 netif_carrier_on(netdev);
5270 if (netif_msg_link(handle))
5271 netdev_info(netdev, "link up\n");
5272 } else {
5273 netif_carrier_off(netdev);
5274 netif_tx_stop_all_queues(netdev);
5275 if (netif_msg_link(handle))
5276 netdev_info(netdev, "link down\n");
5277 }
5278}
5279
5280static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
5281{
5282 while (ring->next_to_clean != ring->next_to_use) {
5283 ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
5284 hns3_free_buffer_detach(ring, ring->next_to_clean, 0);
5285 ring_ptr_move_fw(ring, next_to_clean);
5286 }
5287
5288 ring->pending_buf = 0;
5289}
5290
5291static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
5292{
5293 struct hns3_desc_cb res_cbs;
5294 int ret;
5295
5296 while (ring->next_to_use != ring->next_to_clean) {
5297 /* When a buffer is not reused, it's memory has been
5298 * freed in hns3_handle_rx_bd or will be freed by
5299 * stack, so we need to replace the buffer here.
5300 */
5301 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
5302 ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
5303 if (ret) {
5304 u64_stats_update_begin(&ring->syncp);
5305 ring->stats.sw_err_cnt++;
5306 u64_stats_update_end(&ring->syncp);
5307 /* if alloc new buffer fail, exit directly
5308 * and reclear in up flow.
5309 */
5310 netdev_warn(ring_to_netdev(ring),
5311 "reserve buffer map failed, ret = %d\n",
5312 ret);
5313 return ret;
5314 }
5315 hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
5316 }
5317 ring_ptr_move_fw(ring, next_to_use);
5318 }
5319
5320 /* Free the pending skb in rx ring */
5321 if (ring->skb) {
5322 dev_kfree_skb_any(ring->skb);
5323 ring->skb = NULL;
5324 ring->pending_buf = 0;
5325 }
5326
5327 return 0;
5328}
5329
5330static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
5331{
5332 while (ring->next_to_use != ring->next_to_clean) {
5333 /* When a buffer is not reused, it's memory has been
5334 * freed in hns3_handle_rx_bd or will be freed by
5335 * stack, so only need to unmap the buffer here.
5336 */
5337 if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
5338 hns3_unmap_buffer(ring,
5339 &ring->desc_cb[ring->next_to_use]);
5340 ring->desc_cb[ring->next_to_use].dma = 0;
5341 }
5342
5343 ring_ptr_move_fw(ring, next_to_use);
5344 }
5345}
5346
5347static void hns3_clear_all_ring(struct hnae3_handle *h, bool force)
5348{
5349 struct net_device *ndev = h->kinfo.netdev;
5350 struct hns3_nic_priv *priv = netdev_priv(ndev);
5351 u32 i;
5352
5353 for (i = 0; i < h->kinfo.num_tqps; i++) {
5354 struct hns3_enet_ring *ring;
5355
5356 ring = &priv->ring[i];
5357 hns3_clear_tx_ring(ring);
5358
5359 ring = &priv->ring[i + h->kinfo.num_tqps];
5360 /* Continue to clear other rings even if clearing some
5361 * rings failed.
5362 */
5363 if (force)
5364 hns3_force_clear_rx_ring(ring);
5365 else
5366 hns3_clear_rx_ring(ring);
5367 }
5368}
5369
5370int hns3_nic_reset_all_ring(struct hnae3_handle *h)
5371{
5372 struct net_device *ndev = h->kinfo.netdev;
5373 struct hns3_nic_priv *priv = netdev_priv(ndev);
5374 struct hns3_enet_ring *rx_ring;
5375 int i, j;
5376 int ret;
5377
5378 ret = h->ae_algo->ops->reset_queue(h);
5379 if (ret)
5380 return ret;
5381
5382 for (i = 0; i < h->kinfo.num_tqps; i++) {
5383 hns3_init_ring_hw(&priv->ring[i]);
5384
5385 /* We need to clear tx ring here because self test will
5386 * use the ring and will not run down before up
5387 */
5388 hns3_clear_tx_ring(&priv->ring[i]);
5389 priv->ring[i].next_to_clean = 0;
5390 priv->ring[i].next_to_use = 0;
5391 priv->ring[i].last_to_use = 0;
5392
5393 rx_ring = &priv->ring[i + h->kinfo.num_tqps];
5394 hns3_init_ring_hw(rx_ring);
5395 ret = hns3_clear_rx_ring(rx_ring);
5396 if (ret)
5397 return ret;
5398
5399 /* We can not know the hardware head and tail when this
5400 * function is called in reset flow, so we reuse all desc.
5401 */
5402 for (j = 0; j < rx_ring->desc_num; j++)
5403 hns3_reuse_buffer(rx_ring, j);
5404
5405 rx_ring->next_to_clean = 0;
5406 rx_ring->next_to_use = 0;
5407 }
5408
5409 hns3_init_tx_ring_tc(priv);
5410
5411 return 0;
5412}
5413
5414static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
5415{
5416 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
5417 struct net_device *ndev = kinfo->netdev;
5418 struct hns3_nic_priv *priv = netdev_priv(ndev);
5419
5420 if (test_and_set_bit(HNS3_NIC_STATE_RESETTING, &priv->state))
5421 return 0;
5422
5423 if (!netif_running(ndev))
5424 return 0;
5425
5426 return hns3_nic_net_stop(ndev);
5427}
5428
5429static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
5430{
5431 struct hnae3_knic_private_info *kinfo = &handle->kinfo;
5432 struct hns3_nic_priv *priv = netdev_priv(kinfo->netdev);
5433 int ret = 0;
5434
5435 if (!test_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5436 netdev_err(kinfo->netdev, "device is not initialized yet\n");
5437 return -EFAULT;
5438 }
5439
5440 clear_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
5441
5442 if (netif_running(kinfo->netdev)) {
5443 ret = hns3_nic_net_open(kinfo->netdev);
5444 if (ret) {
5445 set_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
5446 netdev_err(kinfo->netdev,
5447 "net up fail, ret=%d!\n", ret);
5448 return ret;
5449 }
5450 }
5451
5452 return ret;
5453}
5454
5455static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
5456{
5457 struct net_device *netdev = handle->kinfo.netdev;
5458 struct hns3_nic_priv *priv = netdev_priv(netdev);
5459 int ret;
5460
5461 /* Carrier off reporting is important to ethtool even BEFORE open */
5462 netif_carrier_off(netdev);
5463
5464 ret = hns3_get_ring_config(priv);
5465 if (ret)
5466 return ret;
5467
5468 ret = hns3_nic_alloc_vector_data(priv);
5469 if (ret)
5470 goto err_put_ring;
5471
5472 ret = hns3_nic_init_vector_data(priv);
5473 if (ret)
5474 goto err_dealloc_vector;
5475
5476 ret = hns3_init_all_ring(priv);
5477 if (ret)
5478 goto err_uninit_vector;
5479
5480 hns3_cq_period_mode_init(priv, priv->tx_cqe_mode, priv->rx_cqe_mode);
5481
5482 /* the device can work without cpu rmap, only aRFS needs it */
5483 ret = hns3_set_rx_cpu_rmap(netdev);
5484 if (ret)
5485 dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
5486
5487 ret = hns3_nic_init_irq(priv);
5488 if (ret) {
5489 dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
5490 hns3_free_rx_cpu_rmap(netdev);
5491 goto err_init_irq_fail;
5492 }
5493
5494 if (!hns3_is_phys_func(handle->pdev))
5495 hns3_init_mac_addr(netdev);
5496
5497 ret = hns3_client_start(handle);
5498 if (ret) {
5499 dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
5500 goto err_client_start_fail;
5501 }
5502
5503 set_bit(HNS3_NIC_STATE_INITED, &priv->state);
5504
5505 return ret;
5506
5507err_client_start_fail:
5508 hns3_free_rx_cpu_rmap(netdev);
5509 hns3_nic_uninit_irq(priv);
5510err_init_irq_fail:
5511 hns3_uninit_all_ring(priv);
5512err_uninit_vector:
5513 hns3_nic_uninit_vector_data(priv);
5514err_dealloc_vector:
5515 hns3_nic_dealloc_vector_data(priv);
5516err_put_ring:
5517 hns3_put_ring_config(priv);
5518
5519 return ret;
5520}
5521
5522static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
5523{
5524 struct net_device *netdev = handle->kinfo.netdev;
5525 struct hns3_nic_priv *priv = netdev_priv(netdev);
5526
5527 if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5528 netdev_warn(netdev, "already uninitialized\n");
5529 return 0;
5530 }
5531
5532 hns3_free_rx_cpu_rmap(netdev);
5533 hns3_nic_uninit_irq(priv);
5534 hns3_clear_all_ring(handle, true);
5535 hns3_reset_tx_queue(priv->ae_handle);
5536
5537 hns3_nic_uninit_vector_data(priv);
5538
5539 hns3_nic_dealloc_vector_data(priv);
5540
5541 hns3_uninit_all_ring(priv);
5542
5543 hns3_put_ring_config(priv);
5544
5545 return 0;
5546}
5547
5548int hns3_reset_notify(struct hnae3_handle *handle,
5549 enum hnae3_reset_notify_type type)
5550{
5551 int ret = 0;
5552
5553 switch (type) {
5554 case HNAE3_UP_CLIENT:
5555 ret = hns3_reset_notify_up_enet(handle);
5556 break;
5557 case HNAE3_DOWN_CLIENT:
5558 ret = hns3_reset_notify_down_enet(handle);
5559 break;
5560 case HNAE3_INIT_CLIENT:
5561 ret = hns3_reset_notify_init_enet(handle);
5562 break;
5563 case HNAE3_UNINIT_CLIENT:
5564 ret = hns3_reset_notify_uninit_enet(handle);
5565 break;
5566 default:
5567 break;
5568 }
5569
5570 return ret;
5571}
5572
5573static int hns3_change_channels(struct hnae3_handle *handle, u32 new_tqp_num,
5574 bool rxfh_configured)
5575{
5576 int ret;
5577
5578 ret = handle->ae_algo->ops->set_channels(handle, new_tqp_num,
5579 rxfh_configured);
5580 if (ret) {
5581 dev_err(&handle->pdev->dev,
5582 "Change tqp num(%u) fail.\n", new_tqp_num);
5583 return ret;
5584 }
5585
5586 ret = hns3_reset_notify(handle, HNAE3_INIT_CLIENT);
5587 if (ret)
5588 return ret;
5589
5590 ret = hns3_reset_notify(handle, HNAE3_UP_CLIENT);
5591 if (ret)
5592 hns3_reset_notify(handle, HNAE3_UNINIT_CLIENT);
5593
5594 return ret;
5595}
5596
5597int hns3_set_channels(struct net_device *netdev,
5598 struct ethtool_channels *ch)
5599{
5600 struct hnae3_handle *h = hns3_get_handle(netdev);
5601 struct hnae3_knic_private_info *kinfo = &h->kinfo;
5602 bool rxfh_configured = netif_is_rxfh_configured(netdev);
5603 u32 new_tqp_num = ch->combined_count;
5604 u16 org_tqp_num;
5605 int ret;
5606
5607 if (hns3_nic_resetting(netdev))
5608 return -EBUSY;
5609
5610 if (ch->rx_count || ch->tx_count)
5611 return -EINVAL;
5612
5613 if (kinfo->tc_info.mqprio_active) {
5614 dev_err(&netdev->dev,
5615 "it's not allowed to set channels via ethtool when MQPRIO mode is on\n");
5616 return -EINVAL;
5617 }
5618
5619 if (new_tqp_num > hns3_get_max_available_channels(h) ||
5620 new_tqp_num < 1) {
5621 dev_err(&netdev->dev,
5622 "Change tqps fail, the tqp range is from 1 to %u",
5623 hns3_get_max_available_channels(h));
5624 return -EINVAL;
5625 }
5626
5627 if (kinfo->rss_size == new_tqp_num)
5628 return 0;
5629
5630 netif_dbg(h, drv, netdev,
5631 "set channels: tqp_num=%u, rxfh=%d\n",
5632 new_tqp_num, rxfh_configured);
5633
5634 ret = hns3_reset_notify(h, HNAE3_DOWN_CLIENT);
5635 if (ret)
5636 return ret;
5637
5638 ret = hns3_reset_notify(h, HNAE3_UNINIT_CLIENT);
5639 if (ret)
5640 return ret;
5641
5642 org_tqp_num = h->kinfo.num_tqps;
5643 ret = hns3_change_channels(h, new_tqp_num, rxfh_configured);
5644 if (ret) {
5645 int ret1;
5646
5647 netdev_warn(netdev,
5648 "Change channels fail, revert to old value\n");
5649 ret1 = hns3_change_channels(h, org_tqp_num, rxfh_configured);
5650 if (ret1) {
5651 netdev_err(netdev,
5652 "revert to old channel fail\n");
5653 return ret1;
5654 }
5655
5656 return ret;
5657 }
5658
5659 return 0;
5660}
5661
5662static const struct hns3_hw_error_info hns3_hw_err[] = {
5663 { .type = HNAE3_PPU_POISON_ERROR,
5664 .msg = "PPU poison" },
5665 { .type = HNAE3_CMDQ_ECC_ERROR,
5666 .msg = "IMP CMDQ error" },
5667 { .type = HNAE3_IMP_RD_POISON_ERROR,
5668 .msg = "IMP RD poison" },
5669 { .type = HNAE3_ROCEE_AXI_RESP_ERROR,
5670 .msg = "ROCEE AXI RESP error" },
5671};
5672
5673static void hns3_process_hw_error(struct hnae3_handle *handle,
5674 enum hnae3_hw_error_type type)
5675{
5676 int i;
5677
5678 for (i = 0; i < ARRAY_SIZE(hns3_hw_err); i++) {
5679 if (hns3_hw_err[i].type == type) {
5680 dev_err(&handle->pdev->dev, "Detected %s!\n",
5681 hns3_hw_err[i].msg);
5682 break;
5683 }
5684 }
5685}
5686
5687static const struct hnae3_client_ops client_ops = {
5688 .init_instance = hns3_client_init,
5689 .uninit_instance = hns3_client_uninit,
5690 .link_status_change = hns3_link_status_change,
5691 .reset_notify = hns3_reset_notify,
5692 .process_hw_error = hns3_process_hw_error,
5693};
5694
5695/* hns3_init_module - Driver registration routine
5696 * hns3_init_module is the first routine called when the driver is
5697 * loaded. All it does is register with the PCI subsystem.
5698 */
5699static int __init hns3_init_module(void)
5700{
5701 int ret;
5702
5703 pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
5704 pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
5705
5706 client.type = HNAE3_CLIENT_KNIC;
5707 snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH, "%s",
5708 hns3_driver_name);
5709
5710 client.ops = &client_ops;
5711
5712 INIT_LIST_HEAD(&client.node);
5713
5714 hns3_dbg_register_debugfs(hns3_driver_name);
5715
5716 ret = hnae3_register_client(&client);
5717 if (ret)
5718 goto err_reg_client;
5719
5720 ret = pci_register_driver(&hns3_driver);
5721 if (ret)
5722 goto err_reg_driver;
5723
5724 return ret;
5725
5726err_reg_driver:
5727 hnae3_unregister_client(&client);
5728err_reg_client:
5729 hns3_dbg_unregister_debugfs();
5730 return ret;
5731}
5732module_init(hns3_init_module);
5733
5734/* hns3_exit_module - Driver exit cleanup routine
5735 * hns3_exit_module is called just before the driver is removed
5736 * from memory.
5737 */
5738static void __exit hns3_exit_module(void)
5739{
5740 pci_unregister_driver(&hns3_driver);
5741 hnae3_unregister_client(&client);
5742 hns3_dbg_unregister_debugfs();
5743}
5744module_exit(hns3_exit_module);
5745
5746MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
5747MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
5748MODULE_LICENSE("GPL");
5749MODULE_ALIAS("pci:hns-nic");
Linux 6.x block layer and io_uring tree(s)