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/ip6_checksum.h>
21 #include <net/pkt_cls.h>
22 #include <net/tcp.h>
23 #include <net/vxlan.h>
24 #include <net/geneve.h>
25 
26 #include "hnae3.h"
27 #include "hns3_enet.h"
28 /* All hns3 tracepoints are defined by the include below, which
29  * must be included exactly once across the whole kernel with
30  * CREATE_TRACE_POINTS defined
31  */
32 #define CREATE_TRACE_POINTS
33 #include "hns3_trace.h"
34 
35 #define hns3_set_field(origin, shift, val)	((origin) |= (val) << (shift))
36 #define hns3_tx_bd_count(S)	DIV_ROUND_UP(S, HNS3_MAX_BD_SIZE)
37 
38 #define hns3_rl_err(fmt, ...)						\
39 	do {								\
40 		if (net_ratelimit())					\
41 			netdev_err(fmt, ##__VA_ARGS__);			\
42 	} while (0)
43 
44 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force);
45 
46 static const char hns3_driver_name[] = "hns3";
47 static const char hns3_driver_string[] =
48 			"Hisilicon Ethernet Network Driver for Hip08 Family";
49 static const char hns3_copyright[] = "Copyright (c) 2017 Huawei Corporation.";
50 static struct hnae3_client client;
51 
52 static int debug = -1;
53 module_param(debug, int, 0);
54 MODULE_PARM_DESC(debug, " Network interface message level setting");
55 
56 static unsigned int tx_spare_buf_size;
57 module_param(tx_spare_buf_size, uint, 0400);
58 MODULE_PARM_DESC(tx_spare_buf_size, "Size used to allocate tx spare buffer");
59 
60 static unsigned int tx_sgl = 1;
61 module_param(tx_sgl, uint, 0600);
62 MODULE_PARM_DESC(tx_sgl, "Minimum number of frags when using dma_map_sg() to optimize the IOMMU mapping");
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 
77 /* hns3_pci_tbl - PCI Device ID Table
78  *
79  * Last entry must be all 0s
80  *
81  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
82  *   Class, Class Mask, private data (not used) }
83  */
84 static const struct pci_device_id hns3_pci_tbl[] = {
85 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
86 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
87 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA),
88 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
89 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC),
90 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
91 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA),
92 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
93 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC),
94 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
95 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC),
96 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
97 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_200G_RDMA),
98 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
99 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_VF), 0},
100 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_RDMA_DCB_PFC_VF),
101 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
102 	/* required last entry */
103 	{0, }
104 };
105 MODULE_DEVICE_TABLE(pci, hns3_pci_tbl);
106 
107 #define HNS3_RX_PTYPE_ENTRY(ptype, l, s, t) \
108 	{	ptype, \
109 		l, \
110 		CHECKSUM_##s, \
111 		HNS3_L3_TYPE_##t, \
112 		1 }
113 
114 #define HNS3_RX_PTYPE_UNUSED_ENTRY(ptype) \
115 		{ ptype, 0, CHECKSUM_NONE, HNS3_L3_TYPE_PARSE_FAIL, 0 }
116 
117 static const struct hns3_rx_ptype hns3_rx_ptype_tbl[] = {
118 	HNS3_RX_PTYPE_UNUSED_ENTRY(0),
119 	HNS3_RX_PTYPE_ENTRY(1, 0, COMPLETE, ARP),
120 	HNS3_RX_PTYPE_ENTRY(2, 0, COMPLETE, RARP),
121 	HNS3_RX_PTYPE_ENTRY(3, 0, COMPLETE, LLDP),
122 	HNS3_RX_PTYPE_ENTRY(4, 0, COMPLETE, PARSE_FAIL),
123 	HNS3_RX_PTYPE_ENTRY(5, 0, COMPLETE, PARSE_FAIL),
124 	HNS3_RX_PTYPE_ENTRY(6, 0, COMPLETE, PARSE_FAIL),
125 	HNS3_RX_PTYPE_ENTRY(7, 0, COMPLETE, CNM),
126 	HNS3_RX_PTYPE_ENTRY(8, 0, NONE, PARSE_FAIL),
127 	HNS3_RX_PTYPE_UNUSED_ENTRY(9),
128 	HNS3_RX_PTYPE_UNUSED_ENTRY(10),
129 	HNS3_RX_PTYPE_UNUSED_ENTRY(11),
130 	HNS3_RX_PTYPE_UNUSED_ENTRY(12),
131 	HNS3_RX_PTYPE_UNUSED_ENTRY(13),
132 	HNS3_RX_PTYPE_UNUSED_ENTRY(14),
133 	HNS3_RX_PTYPE_UNUSED_ENTRY(15),
134 	HNS3_RX_PTYPE_ENTRY(16, 0, COMPLETE, PARSE_FAIL),
135 	HNS3_RX_PTYPE_ENTRY(17, 0, COMPLETE, IPV4),
136 	HNS3_RX_PTYPE_ENTRY(18, 0, COMPLETE, IPV4),
137 	HNS3_RX_PTYPE_ENTRY(19, 0, UNNECESSARY, IPV4),
138 	HNS3_RX_PTYPE_ENTRY(20, 0, UNNECESSARY, IPV4),
139 	HNS3_RX_PTYPE_ENTRY(21, 0, NONE, IPV4),
140 	HNS3_RX_PTYPE_ENTRY(22, 0, UNNECESSARY, IPV4),
141 	HNS3_RX_PTYPE_ENTRY(23, 0, NONE, IPV4),
142 	HNS3_RX_PTYPE_ENTRY(24, 0, NONE, IPV4),
143 	HNS3_RX_PTYPE_ENTRY(25, 0, UNNECESSARY, IPV4),
144 	HNS3_RX_PTYPE_UNUSED_ENTRY(26),
145 	HNS3_RX_PTYPE_UNUSED_ENTRY(27),
146 	HNS3_RX_PTYPE_UNUSED_ENTRY(28),
147 	HNS3_RX_PTYPE_ENTRY(29, 0, COMPLETE, PARSE_FAIL),
148 	HNS3_RX_PTYPE_ENTRY(30, 0, COMPLETE, PARSE_FAIL),
149 	HNS3_RX_PTYPE_ENTRY(31, 0, COMPLETE, IPV4),
150 	HNS3_RX_PTYPE_ENTRY(32, 0, COMPLETE, IPV4),
151 	HNS3_RX_PTYPE_ENTRY(33, 1, UNNECESSARY, IPV4),
152 	HNS3_RX_PTYPE_ENTRY(34, 1, UNNECESSARY, IPV4),
153 	HNS3_RX_PTYPE_ENTRY(35, 1, UNNECESSARY, IPV4),
154 	HNS3_RX_PTYPE_ENTRY(36, 0, COMPLETE, IPV4),
155 	HNS3_RX_PTYPE_ENTRY(37, 0, COMPLETE, IPV4),
156 	HNS3_RX_PTYPE_UNUSED_ENTRY(38),
157 	HNS3_RX_PTYPE_ENTRY(39, 0, COMPLETE, IPV6),
158 	HNS3_RX_PTYPE_ENTRY(40, 0, COMPLETE, IPV6),
159 	HNS3_RX_PTYPE_ENTRY(41, 1, UNNECESSARY, IPV6),
160 	HNS3_RX_PTYPE_ENTRY(42, 1, UNNECESSARY, IPV6),
161 	HNS3_RX_PTYPE_ENTRY(43, 1, UNNECESSARY, IPV6),
162 	HNS3_RX_PTYPE_ENTRY(44, 0, COMPLETE, IPV6),
163 	HNS3_RX_PTYPE_ENTRY(45, 0, COMPLETE, IPV6),
164 	HNS3_RX_PTYPE_UNUSED_ENTRY(46),
165 	HNS3_RX_PTYPE_UNUSED_ENTRY(47),
166 	HNS3_RX_PTYPE_UNUSED_ENTRY(48),
167 	HNS3_RX_PTYPE_UNUSED_ENTRY(49),
168 	HNS3_RX_PTYPE_UNUSED_ENTRY(50),
169 	HNS3_RX_PTYPE_UNUSED_ENTRY(51),
170 	HNS3_RX_PTYPE_UNUSED_ENTRY(52),
171 	HNS3_RX_PTYPE_UNUSED_ENTRY(53),
172 	HNS3_RX_PTYPE_UNUSED_ENTRY(54),
173 	HNS3_RX_PTYPE_UNUSED_ENTRY(55),
174 	HNS3_RX_PTYPE_UNUSED_ENTRY(56),
175 	HNS3_RX_PTYPE_UNUSED_ENTRY(57),
176 	HNS3_RX_PTYPE_UNUSED_ENTRY(58),
177 	HNS3_RX_PTYPE_UNUSED_ENTRY(59),
178 	HNS3_RX_PTYPE_UNUSED_ENTRY(60),
179 	HNS3_RX_PTYPE_UNUSED_ENTRY(61),
180 	HNS3_RX_PTYPE_UNUSED_ENTRY(62),
181 	HNS3_RX_PTYPE_UNUSED_ENTRY(63),
182 	HNS3_RX_PTYPE_UNUSED_ENTRY(64),
183 	HNS3_RX_PTYPE_UNUSED_ENTRY(65),
184 	HNS3_RX_PTYPE_UNUSED_ENTRY(66),
185 	HNS3_RX_PTYPE_UNUSED_ENTRY(67),
186 	HNS3_RX_PTYPE_UNUSED_ENTRY(68),
187 	HNS3_RX_PTYPE_UNUSED_ENTRY(69),
188 	HNS3_RX_PTYPE_UNUSED_ENTRY(70),
189 	HNS3_RX_PTYPE_UNUSED_ENTRY(71),
190 	HNS3_RX_PTYPE_UNUSED_ENTRY(72),
191 	HNS3_RX_PTYPE_UNUSED_ENTRY(73),
192 	HNS3_RX_PTYPE_UNUSED_ENTRY(74),
193 	HNS3_RX_PTYPE_UNUSED_ENTRY(75),
194 	HNS3_RX_PTYPE_UNUSED_ENTRY(76),
195 	HNS3_RX_PTYPE_UNUSED_ENTRY(77),
196 	HNS3_RX_PTYPE_UNUSED_ENTRY(78),
197 	HNS3_RX_PTYPE_UNUSED_ENTRY(79),
198 	HNS3_RX_PTYPE_UNUSED_ENTRY(80),
199 	HNS3_RX_PTYPE_UNUSED_ENTRY(81),
200 	HNS3_RX_PTYPE_UNUSED_ENTRY(82),
201 	HNS3_RX_PTYPE_UNUSED_ENTRY(83),
202 	HNS3_RX_PTYPE_UNUSED_ENTRY(84),
203 	HNS3_RX_PTYPE_UNUSED_ENTRY(85),
204 	HNS3_RX_PTYPE_UNUSED_ENTRY(86),
205 	HNS3_RX_PTYPE_UNUSED_ENTRY(87),
206 	HNS3_RX_PTYPE_UNUSED_ENTRY(88),
207 	HNS3_RX_PTYPE_UNUSED_ENTRY(89),
208 	HNS3_RX_PTYPE_UNUSED_ENTRY(90),
209 	HNS3_RX_PTYPE_UNUSED_ENTRY(91),
210 	HNS3_RX_PTYPE_UNUSED_ENTRY(92),
211 	HNS3_RX_PTYPE_UNUSED_ENTRY(93),
212 	HNS3_RX_PTYPE_UNUSED_ENTRY(94),
213 	HNS3_RX_PTYPE_UNUSED_ENTRY(95),
214 	HNS3_RX_PTYPE_UNUSED_ENTRY(96),
215 	HNS3_RX_PTYPE_UNUSED_ENTRY(97),
216 	HNS3_RX_PTYPE_UNUSED_ENTRY(98),
217 	HNS3_RX_PTYPE_UNUSED_ENTRY(99),
218 	HNS3_RX_PTYPE_UNUSED_ENTRY(100),
219 	HNS3_RX_PTYPE_UNUSED_ENTRY(101),
220 	HNS3_RX_PTYPE_UNUSED_ENTRY(102),
221 	HNS3_RX_PTYPE_UNUSED_ENTRY(103),
222 	HNS3_RX_PTYPE_UNUSED_ENTRY(104),
223 	HNS3_RX_PTYPE_UNUSED_ENTRY(105),
224 	HNS3_RX_PTYPE_UNUSED_ENTRY(106),
225 	HNS3_RX_PTYPE_UNUSED_ENTRY(107),
226 	HNS3_RX_PTYPE_UNUSED_ENTRY(108),
227 	HNS3_RX_PTYPE_UNUSED_ENTRY(109),
228 	HNS3_RX_PTYPE_UNUSED_ENTRY(110),
229 	HNS3_RX_PTYPE_ENTRY(111, 0, COMPLETE, IPV6),
230 	HNS3_RX_PTYPE_ENTRY(112, 0, COMPLETE, IPV6),
231 	HNS3_RX_PTYPE_ENTRY(113, 0, UNNECESSARY, IPV6),
232 	HNS3_RX_PTYPE_ENTRY(114, 0, UNNECESSARY, IPV6),
233 	HNS3_RX_PTYPE_ENTRY(115, 0, NONE, IPV6),
234 	HNS3_RX_PTYPE_ENTRY(116, 0, UNNECESSARY, IPV6),
235 	HNS3_RX_PTYPE_ENTRY(117, 0, NONE, IPV6),
236 	HNS3_RX_PTYPE_ENTRY(118, 0, NONE, IPV6),
237 	HNS3_RX_PTYPE_ENTRY(119, 0, UNNECESSARY, IPV6),
238 	HNS3_RX_PTYPE_UNUSED_ENTRY(120),
239 	HNS3_RX_PTYPE_UNUSED_ENTRY(121),
240 	HNS3_RX_PTYPE_UNUSED_ENTRY(122),
241 	HNS3_RX_PTYPE_ENTRY(123, 0, COMPLETE, PARSE_FAIL),
242 	HNS3_RX_PTYPE_ENTRY(124, 0, COMPLETE, PARSE_FAIL),
243 	HNS3_RX_PTYPE_ENTRY(125, 0, COMPLETE, IPV4),
244 	HNS3_RX_PTYPE_ENTRY(126, 0, COMPLETE, IPV4),
245 	HNS3_RX_PTYPE_ENTRY(127, 1, UNNECESSARY, IPV4),
246 	HNS3_RX_PTYPE_ENTRY(128, 1, UNNECESSARY, IPV4),
247 	HNS3_RX_PTYPE_ENTRY(129, 1, UNNECESSARY, IPV4),
248 	HNS3_RX_PTYPE_ENTRY(130, 0, COMPLETE, IPV4),
249 	HNS3_RX_PTYPE_ENTRY(131, 0, COMPLETE, IPV4),
250 	HNS3_RX_PTYPE_UNUSED_ENTRY(132),
251 	HNS3_RX_PTYPE_ENTRY(133, 0, COMPLETE, IPV6),
252 	HNS3_RX_PTYPE_ENTRY(134, 0, COMPLETE, IPV6),
253 	HNS3_RX_PTYPE_ENTRY(135, 1, UNNECESSARY, IPV6),
254 	HNS3_RX_PTYPE_ENTRY(136, 1, UNNECESSARY, IPV6),
255 	HNS3_RX_PTYPE_ENTRY(137, 1, UNNECESSARY, IPV6),
256 	HNS3_RX_PTYPE_ENTRY(138, 0, COMPLETE, IPV6),
257 	HNS3_RX_PTYPE_ENTRY(139, 0, COMPLETE, IPV6),
258 	HNS3_RX_PTYPE_UNUSED_ENTRY(140),
259 	HNS3_RX_PTYPE_UNUSED_ENTRY(141),
260 	HNS3_RX_PTYPE_UNUSED_ENTRY(142),
261 	HNS3_RX_PTYPE_UNUSED_ENTRY(143),
262 	HNS3_RX_PTYPE_UNUSED_ENTRY(144),
263 	HNS3_RX_PTYPE_UNUSED_ENTRY(145),
264 	HNS3_RX_PTYPE_UNUSED_ENTRY(146),
265 	HNS3_RX_PTYPE_UNUSED_ENTRY(147),
266 	HNS3_RX_PTYPE_UNUSED_ENTRY(148),
267 	HNS3_RX_PTYPE_UNUSED_ENTRY(149),
268 	HNS3_RX_PTYPE_UNUSED_ENTRY(150),
269 	HNS3_RX_PTYPE_UNUSED_ENTRY(151),
270 	HNS3_RX_PTYPE_UNUSED_ENTRY(152),
271 	HNS3_RX_PTYPE_UNUSED_ENTRY(153),
272 	HNS3_RX_PTYPE_UNUSED_ENTRY(154),
273 	HNS3_RX_PTYPE_UNUSED_ENTRY(155),
274 	HNS3_RX_PTYPE_UNUSED_ENTRY(156),
275 	HNS3_RX_PTYPE_UNUSED_ENTRY(157),
276 	HNS3_RX_PTYPE_UNUSED_ENTRY(158),
277 	HNS3_RX_PTYPE_UNUSED_ENTRY(159),
278 	HNS3_RX_PTYPE_UNUSED_ENTRY(160),
279 	HNS3_RX_PTYPE_UNUSED_ENTRY(161),
280 	HNS3_RX_PTYPE_UNUSED_ENTRY(162),
281 	HNS3_RX_PTYPE_UNUSED_ENTRY(163),
282 	HNS3_RX_PTYPE_UNUSED_ENTRY(164),
283 	HNS3_RX_PTYPE_UNUSED_ENTRY(165),
284 	HNS3_RX_PTYPE_UNUSED_ENTRY(166),
285 	HNS3_RX_PTYPE_UNUSED_ENTRY(167),
286 	HNS3_RX_PTYPE_UNUSED_ENTRY(168),
287 	HNS3_RX_PTYPE_UNUSED_ENTRY(169),
288 	HNS3_RX_PTYPE_UNUSED_ENTRY(170),
289 	HNS3_RX_PTYPE_UNUSED_ENTRY(171),
290 	HNS3_RX_PTYPE_UNUSED_ENTRY(172),
291 	HNS3_RX_PTYPE_UNUSED_ENTRY(173),
292 	HNS3_RX_PTYPE_UNUSED_ENTRY(174),
293 	HNS3_RX_PTYPE_UNUSED_ENTRY(175),
294 	HNS3_RX_PTYPE_UNUSED_ENTRY(176),
295 	HNS3_RX_PTYPE_UNUSED_ENTRY(177),
296 	HNS3_RX_PTYPE_UNUSED_ENTRY(178),
297 	HNS3_RX_PTYPE_UNUSED_ENTRY(179),
298 	HNS3_RX_PTYPE_UNUSED_ENTRY(180),
299 	HNS3_RX_PTYPE_UNUSED_ENTRY(181),
300 	HNS3_RX_PTYPE_UNUSED_ENTRY(182),
301 	HNS3_RX_PTYPE_UNUSED_ENTRY(183),
302 	HNS3_RX_PTYPE_UNUSED_ENTRY(184),
303 	HNS3_RX_PTYPE_UNUSED_ENTRY(185),
304 	HNS3_RX_PTYPE_UNUSED_ENTRY(186),
305 	HNS3_RX_PTYPE_UNUSED_ENTRY(187),
306 	HNS3_RX_PTYPE_UNUSED_ENTRY(188),
307 	HNS3_RX_PTYPE_UNUSED_ENTRY(189),
308 	HNS3_RX_PTYPE_UNUSED_ENTRY(190),
309 	HNS3_RX_PTYPE_UNUSED_ENTRY(191),
310 	HNS3_RX_PTYPE_UNUSED_ENTRY(192),
311 	HNS3_RX_PTYPE_UNUSED_ENTRY(193),
312 	HNS3_RX_PTYPE_UNUSED_ENTRY(194),
313 	HNS3_RX_PTYPE_UNUSED_ENTRY(195),
314 	HNS3_RX_PTYPE_UNUSED_ENTRY(196),
315 	HNS3_RX_PTYPE_UNUSED_ENTRY(197),
316 	HNS3_RX_PTYPE_UNUSED_ENTRY(198),
317 	HNS3_RX_PTYPE_UNUSED_ENTRY(199),
318 	HNS3_RX_PTYPE_UNUSED_ENTRY(200),
319 	HNS3_RX_PTYPE_UNUSED_ENTRY(201),
320 	HNS3_RX_PTYPE_UNUSED_ENTRY(202),
321 	HNS3_RX_PTYPE_UNUSED_ENTRY(203),
322 	HNS3_RX_PTYPE_UNUSED_ENTRY(204),
323 	HNS3_RX_PTYPE_UNUSED_ENTRY(205),
324 	HNS3_RX_PTYPE_UNUSED_ENTRY(206),
325 	HNS3_RX_PTYPE_UNUSED_ENTRY(207),
326 	HNS3_RX_PTYPE_UNUSED_ENTRY(208),
327 	HNS3_RX_PTYPE_UNUSED_ENTRY(209),
328 	HNS3_RX_PTYPE_UNUSED_ENTRY(210),
329 	HNS3_RX_PTYPE_UNUSED_ENTRY(211),
330 	HNS3_RX_PTYPE_UNUSED_ENTRY(212),
331 	HNS3_RX_PTYPE_UNUSED_ENTRY(213),
332 	HNS3_RX_PTYPE_UNUSED_ENTRY(214),
333 	HNS3_RX_PTYPE_UNUSED_ENTRY(215),
334 	HNS3_RX_PTYPE_UNUSED_ENTRY(216),
335 	HNS3_RX_PTYPE_UNUSED_ENTRY(217),
336 	HNS3_RX_PTYPE_UNUSED_ENTRY(218),
337 	HNS3_RX_PTYPE_UNUSED_ENTRY(219),
338 	HNS3_RX_PTYPE_UNUSED_ENTRY(220),
339 	HNS3_RX_PTYPE_UNUSED_ENTRY(221),
340 	HNS3_RX_PTYPE_UNUSED_ENTRY(222),
341 	HNS3_RX_PTYPE_UNUSED_ENTRY(223),
342 	HNS3_RX_PTYPE_UNUSED_ENTRY(224),
343 	HNS3_RX_PTYPE_UNUSED_ENTRY(225),
344 	HNS3_RX_PTYPE_UNUSED_ENTRY(226),
345 	HNS3_RX_PTYPE_UNUSED_ENTRY(227),
346 	HNS3_RX_PTYPE_UNUSED_ENTRY(228),
347 	HNS3_RX_PTYPE_UNUSED_ENTRY(229),
348 	HNS3_RX_PTYPE_UNUSED_ENTRY(230),
349 	HNS3_RX_PTYPE_UNUSED_ENTRY(231),
350 	HNS3_RX_PTYPE_UNUSED_ENTRY(232),
351 	HNS3_RX_PTYPE_UNUSED_ENTRY(233),
352 	HNS3_RX_PTYPE_UNUSED_ENTRY(234),
353 	HNS3_RX_PTYPE_UNUSED_ENTRY(235),
354 	HNS3_RX_PTYPE_UNUSED_ENTRY(236),
355 	HNS3_RX_PTYPE_UNUSED_ENTRY(237),
356 	HNS3_RX_PTYPE_UNUSED_ENTRY(238),
357 	HNS3_RX_PTYPE_UNUSED_ENTRY(239),
358 	HNS3_RX_PTYPE_UNUSED_ENTRY(240),
359 	HNS3_RX_PTYPE_UNUSED_ENTRY(241),
360 	HNS3_RX_PTYPE_UNUSED_ENTRY(242),
361 	HNS3_RX_PTYPE_UNUSED_ENTRY(243),
362 	HNS3_RX_PTYPE_UNUSED_ENTRY(244),
363 	HNS3_RX_PTYPE_UNUSED_ENTRY(245),
364 	HNS3_RX_PTYPE_UNUSED_ENTRY(246),
365 	HNS3_RX_PTYPE_UNUSED_ENTRY(247),
366 	HNS3_RX_PTYPE_UNUSED_ENTRY(248),
367 	HNS3_RX_PTYPE_UNUSED_ENTRY(249),
368 	HNS3_RX_PTYPE_UNUSED_ENTRY(250),
369 	HNS3_RX_PTYPE_UNUSED_ENTRY(251),
370 	HNS3_RX_PTYPE_UNUSED_ENTRY(252),
371 	HNS3_RX_PTYPE_UNUSED_ENTRY(253),
372 	HNS3_RX_PTYPE_UNUSED_ENTRY(254),
373 	HNS3_RX_PTYPE_UNUSED_ENTRY(255),
374 };
375 
376 #define HNS3_INVALID_PTYPE \
377 		ARRAY_SIZE(hns3_rx_ptype_tbl)
378 
379 static irqreturn_t hns3_irq_handle(int irq, void *vector)
380 {
381 	struct hns3_enet_tqp_vector *tqp_vector = vector;
382 
383 	napi_schedule_irqoff(&tqp_vector->napi);
384 	tqp_vector->event_cnt++;
385 
386 	return IRQ_HANDLED;
387 }
388 
389 static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv)
390 {
391 	struct hns3_enet_tqp_vector *tqp_vectors;
392 	unsigned int i;
393 
394 	for (i = 0; i < priv->vector_num; i++) {
395 		tqp_vectors = &priv->tqp_vector[i];
396 
397 		if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED)
398 			continue;
399 
400 		/* clear the affinity mask */
401 		irq_set_affinity_hint(tqp_vectors->vector_irq, NULL);
402 
403 		/* release the irq resource */
404 		free_irq(tqp_vectors->vector_irq, tqp_vectors);
405 		tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED;
406 	}
407 }
408 
409 static int hns3_nic_init_irq(struct hns3_nic_priv *priv)
410 {
411 	struct hns3_enet_tqp_vector *tqp_vectors;
412 	int txrx_int_idx = 0;
413 	int rx_int_idx = 0;
414 	int tx_int_idx = 0;
415 	unsigned int i;
416 	int ret;
417 
418 	for (i = 0; i < priv->vector_num; i++) {
419 		tqp_vectors = &priv->tqp_vector[i];
420 
421 		if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED)
422 			continue;
423 
424 		if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) {
425 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
426 				 "%s-%s-%s-%d", hns3_driver_name,
427 				 pci_name(priv->ae_handle->pdev),
428 				 "TxRx", txrx_int_idx++);
429 			txrx_int_idx++;
430 		} else if (tqp_vectors->rx_group.ring) {
431 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
432 				 "%s-%s-%s-%d", hns3_driver_name,
433 				 pci_name(priv->ae_handle->pdev),
434 				 "Rx", rx_int_idx++);
435 		} else if (tqp_vectors->tx_group.ring) {
436 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
437 				 "%s-%s-%s-%d", hns3_driver_name,
438 				 pci_name(priv->ae_handle->pdev),
439 				 "Tx", tx_int_idx++);
440 		} else {
441 			/* Skip this unused q_vector */
442 			continue;
443 		}
444 
445 		tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0';
446 
447 		irq_set_status_flags(tqp_vectors->vector_irq, IRQ_NOAUTOEN);
448 		ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0,
449 				  tqp_vectors->name, tqp_vectors);
450 		if (ret) {
451 			netdev_err(priv->netdev, "request irq(%d) fail\n",
452 				   tqp_vectors->vector_irq);
453 			hns3_nic_uninit_irq(priv);
454 			return ret;
455 		}
456 
457 		irq_set_affinity_hint(tqp_vectors->vector_irq,
458 				      &tqp_vectors->affinity_mask);
459 
460 		tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED;
461 	}
462 
463 	return 0;
464 }
465 
466 static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector,
467 				 u32 mask_en)
468 {
469 	writel(mask_en, tqp_vector->mask_addr);
470 }
471 
472 static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector)
473 {
474 	napi_enable(&tqp_vector->napi);
475 	enable_irq(tqp_vector->vector_irq);
476 
477 	/* enable vector */
478 	hns3_mask_vector_irq(tqp_vector, 1);
479 }
480 
481 static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector)
482 {
483 	/* disable vector */
484 	hns3_mask_vector_irq(tqp_vector, 0);
485 
486 	disable_irq(tqp_vector->vector_irq);
487 	napi_disable(&tqp_vector->napi);
488 	cancel_work_sync(&tqp_vector->rx_group.dim.work);
489 	cancel_work_sync(&tqp_vector->tx_group.dim.work);
490 }
491 
492 void hns3_set_vector_coalesce_rl(struct hns3_enet_tqp_vector *tqp_vector,
493 				 u32 rl_value)
494 {
495 	u32 rl_reg = hns3_rl_usec_to_reg(rl_value);
496 
497 	/* this defines the configuration for RL (Interrupt Rate Limiter).
498 	 * Rl defines rate of interrupts i.e. number of interrupts-per-second
499 	 * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing
500 	 */
501 	if (rl_reg > 0 && !tqp_vector->tx_group.coal.adapt_enable &&
502 	    !tqp_vector->rx_group.coal.adapt_enable)
503 		/* According to the hardware, the range of rl_reg is
504 		 * 0-59 and the unit is 4.
505 		 */
506 		rl_reg |=  HNS3_INT_RL_ENABLE_MASK;
507 
508 	writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
509 }
510 
511 void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
512 				    u32 gl_value)
513 {
514 	u32 new_val;
515 
516 	if (tqp_vector->rx_group.coal.unit_1us)
517 		new_val = gl_value | HNS3_INT_GL_1US;
518 	else
519 		new_val = hns3_gl_usec_to_reg(gl_value);
520 
521 	writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
522 }
523 
524 void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
525 				    u32 gl_value)
526 {
527 	u32 new_val;
528 
529 	if (tqp_vector->tx_group.coal.unit_1us)
530 		new_val = gl_value | HNS3_INT_GL_1US;
531 	else
532 		new_val = hns3_gl_usec_to_reg(gl_value);
533 
534 	writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
535 }
536 
537 void hns3_set_vector_coalesce_tx_ql(struct hns3_enet_tqp_vector *tqp_vector,
538 				    u32 ql_value)
539 {
540 	writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_TX_QL_OFFSET);
541 }
542 
543 void hns3_set_vector_coalesce_rx_ql(struct hns3_enet_tqp_vector *tqp_vector,
544 				    u32 ql_value)
545 {
546 	writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_RX_QL_OFFSET);
547 }
548 
549 static void hns3_vector_coalesce_init(struct hns3_enet_tqp_vector *tqp_vector,
550 				      struct hns3_nic_priv *priv)
551 {
552 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
553 	struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
554 	struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
555 	struct hns3_enet_coalesce *ptx_coal = &priv->tx_coal;
556 	struct hns3_enet_coalesce *prx_coal = &priv->rx_coal;
557 
558 	tx_coal->adapt_enable = ptx_coal->adapt_enable;
559 	rx_coal->adapt_enable = prx_coal->adapt_enable;
560 
561 	tx_coal->int_gl = ptx_coal->int_gl;
562 	rx_coal->int_gl = prx_coal->int_gl;
563 
564 	rx_coal->flow_level = prx_coal->flow_level;
565 	tx_coal->flow_level = ptx_coal->flow_level;
566 
567 	/* device version above V3(include V3), GL can configure 1us
568 	 * unit, so uses 1us unit.
569 	 */
570 	if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3) {
571 		tx_coal->unit_1us = 1;
572 		rx_coal->unit_1us = 1;
573 	}
574 
575 	if (ae_dev->dev_specs.int_ql_max) {
576 		tx_coal->ql_enable = 1;
577 		rx_coal->ql_enable = 1;
578 		tx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
579 		rx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
580 		tx_coal->int_ql = ptx_coal->int_ql;
581 		rx_coal->int_ql = prx_coal->int_ql;
582 	}
583 }
584 
585 static void
586 hns3_vector_coalesce_init_hw(struct hns3_enet_tqp_vector *tqp_vector,
587 			     struct hns3_nic_priv *priv)
588 {
589 	struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
590 	struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
591 	struct hnae3_handle *h = priv->ae_handle;
592 
593 	hns3_set_vector_coalesce_tx_gl(tqp_vector, tx_coal->int_gl);
594 	hns3_set_vector_coalesce_rx_gl(tqp_vector, rx_coal->int_gl);
595 	hns3_set_vector_coalesce_rl(tqp_vector, h->kinfo.int_rl_setting);
596 
597 	if (tx_coal->ql_enable)
598 		hns3_set_vector_coalesce_tx_ql(tqp_vector, tx_coal->int_ql);
599 
600 	if (rx_coal->ql_enable)
601 		hns3_set_vector_coalesce_rx_ql(tqp_vector, rx_coal->int_ql);
602 }
603 
604 static int hns3_nic_set_real_num_queue(struct net_device *netdev)
605 {
606 	struct hnae3_handle *h = hns3_get_handle(netdev);
607 	struct hnae3_knic_private_info *kinfo = &h->kinfo;
608 	struct hnae3_tc_info *tc_info = &kinfo->tc_info;
609 	unsigned int queue_size = kinfo->num_tqps;
610 	int i, ret;
611 
612 	if (tc_info->num_tc <= 1 && !tc_info->mqprio_active) {
613 		netdev_reset_tc(netdev);
614 	} else {
615 		ret = netdev_set_num_tc(netdev, tc_info->num_tc);
616 		if (ret) {
617 			netdev_err(netdev,
618 				   "netdev_set_num_tc fail, ret=%d!\n", ret);
619 			return ret;
620 		}
621 
622 		for (i = 0; i < HNAE3_MAX_TC; i++) {
623 			if (!test_bit(i, &tc_info->tc_en))
624 				continue;
625 
626 			netdev_set_tc_queue(netdev, i, tc_info->tqp_count[i],
627 					    tc_info->tqp_offset[i]);
628 		}
629 	}
630 
631 	ret = netif_set_real_num_tx_queues(netdev, queue_size);
632 	if (ret) {
633 		netdev_err(netdev,
634 			   "netif_set_real_num_tx_queues fail, ret=%d!\n", ret);
635 		return ret;
636 	}
637 
638 	ret = netif_set_real_num_rx_queues(netdev, queue_size);
639 	if (ret) {
640 		netdev_err(netdev,
641 			   "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
642 		return ret;
643 	}
644 
645 	return 0;
646 }
647 
648 u16 hns3_get_max_available_channels(struct hnae3_handle *h)
649 {
650 	u16 alloc_tqps, max_rss_size, rss_size;
651 
652 	h->ae_algo->ops->get_tqps_and_rss_info(h, &alloc_tqps, &max_rss_size);
653 	rss_size = alloc_tqps / h->kinfo.tc_info.num_tc;
654 
655 	return min_t(u16, rss_size, max_rss_size);
656 }
657 
658 static void hns3_tqp_enable(struct hnae3_queue *tqp)
659 {
660 	u32 rcb_reg;
661 
662 	rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
663 	rcb_reg |= BIT(HNS3_RING_EN_B);
664 	hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
665 }
666 
667 static void hns3_tqp_disable(struct hnae3_queue *tqp)
668 {
669 	u32 rcb_reg;
670 
671 	rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
672 	rcb_reg &= ~BIT(HNS3_RING_EN_B);
673 	hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
674 }
675 
676 static void hns3_free_rx_cpu_rmap(struct net_device *netdev)
677 {
678 #ifdef CONFIG_RFS_ACCEL
679 	free_irq_cpu_rmap(netdev->rx_cpu_rmap);
680 	netdev->rx_cpu_rmap = NULL;
681 #endif
682 }
683 
684 static int hns3_set_rx_cpu_rmap(struct net_device *netdev)
685 {
686 #ifdef CONFIG_RFS_ACCEL
687 	struct hns3_nic_priv *priv = netdev_priv(netdev);
688 	struct hns3_enet_tqp_vector *tqp_vector;
689 	int i, ret;
690 
691 	if (!netdev->rx_cpu_rmap) {
692 		netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->vector_num);
693 		if (!netdev->rx_cpu_rmap)
694 			return -ENOMEM;
695 	}
696 
697 	for (i = 0; i < priv->vector_num; i++) {
698 		tqp_vector = &priv->tqp_vector[i];
699 		ret = irq_cpu_rmap_add(netdev->rx_cpu_rmap,
700 				       tqp_vector->vector_irq);
701 		if (ret) {
702 			hns3_free_rx_cpu_rmap(netdev);
703 			return ret;
704 		}
705 	}
706 #endif
707 	return 0;
708 }
709 
710 static int hns3_nic_net_up(struct net_device *netdev)
711 {
712 	struct hns3_nic_priv *priv = netdev_priv(netdev);
713 	struct hnae3_handle *h = priv->ae_handle;
714 	int i, j;
715 	int ret;
716 
717 	ret = hns3_nic_reset_all_ring(h);
718 	if (ret)
719 		return ret;
720 
721 	clear_bit(HNS3_NIC_STATE_DOWN, &priv->state);
722 
723 	/* enable the vectors */
724 	for (i = 0; i < priv->vector_num; i++)
725 		hns3_vector_enable(&priv->tqp_vector[i]);
726 
727 	/* enable rcb */
728 	for (j = 0; j < h->kinfo.num_tqps; j++)
729 		hns3_tqp_enable(h->kinfo.tqp[j]);
730 
731 	/* start the ae_dev */
732 	ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0;
733 	if (ret) {
734 		set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
735 		while (j--)
736 			hns3_tqp_disable(h->kinfo.tqp[j]);
737 
738 		for (j = i - 1; j >= 0; j--)
739 			hns3_vector_disable(&priv->tqp_vector[j]);
740 	}
741 
742 	return ret;
743 }
744 
745 static void hns3_config_xps(struct hns3_nic_priv *priv)
746 {
747 	int i;
748 
749 	for (i = 0; i < priv->vector_num; i++) {
750 		struct hns3_enet_tqp_vector *tqp_vector = &priv->tqp_vector[i];
751 		struct hns3_enet_ring *ring = tqp_vector->tx_group.ring;
752 
753 		while (ring) {
754 			int ret;
755 
756 			ret = netif_set_xps_queue(priv->netdev,
757 						  &tqp_vector->affinity_mask,
758 						  ring->tqp->tqp_index);
759 			if (ret)
760 				netdev_warn(priv->netdev,
761 					    "set xps queue failed: %d", ret);
762 
763 			ring = ring->next;
764 		}
765 	}
766 }
767 
768 static int hns3_nic_net_open(struct net_device *netdev)
769 {
770 	struct hns3_nic_priv *priv = netdev_priv(netdev);
771 	struct hnae3_handle *h = hns3_get_handle(netdev);
772 	struct hnae3_knic_private_info *kinfo;
773 	int i, ret;
774 
775 	if (hns3_nic_resetting(netdev))
776 		return -EBUSY;
777 
778 	netif_carrier_off(netdev);
779 
780 	ret = hns3_nic_set_real_num_queue(netdev);
781 	if (ret)
782 		return ret;
783 
784 	ret = hns3_nic_net_up(netdev);
785 	if (ret) {
786 		netdev_err(netdev, "net up fail, ret=%d!\n", ret);
787 		return ret;
788 	}
789 
790 	kinfo = &h->kinfo;
791 	for (i = 0; i < HNAE3_MAX_USER_PRIO; i++)
792 		netdev_set_prio_tc_map(netdev, i, kinfo->tc_info.prio_tc[i]);
793 
794 	if (h->ae_algo->ops->set_timer_task)
795 		h->ae_algo->ops->set_timer_task(priv->ae_handle, true);
796 
797 	hns3_config_xps(priv);
798 
799 	netif_dbg(h, drv, netdev, "net open\n");
800 
801 	return 0;
802 }
803 
804 static void hns3_reset_tx_queue(struct hnae3_handle *h)
805 {
806 	struct net_device *ndev = h->kinfo.netdev;
807 	struct hns3_nic_priv *priv = netdev_priv(ndev);
808 	struct netdev_queue *dev_queue;
809 	u32 i;
810 
811 	for (i = 0; i < h->kinfo.num_tqps; i++) {
812 		dev_queue = netdev_get_tx_queue(ndev,
813 						priv->ring[i].queue_index);
814 		netdev_tx_reset_queue(dev_queue);
815 	}
816 }
817 
818 static void hns3_nic_net_down(struct net_device *netdev)
819 {
820 	struct hns3_nic_priv *priv = netdev_priv(netdev);
821 	struct hnae3_handle *h = hns3_get_handle(netdev);
822 	const struct hnae3_ae_ops *ops;
823 	int i;
824 
825 	/* disable vectors */
826 	for (i = 0; i < priv->vector_num; i++)
827 		hns3_vector_disable(&priv->tqp_vector[i]);
828 
829 	/* disable rcb */
830 	for (i = 0; i < h->kinfo.num_tqps; i++)
831 		hns3_tqp_disable(h->kinfo.tqp[i]);
832 
833 	/* stop ae_dev */
834 	ops = priv->ae_handle->ae_algo->ops;
835 	if (ops->stop)
836 		ops->stop(priv->ae_handle);
837 
838 	/* delay ring buffer clearing to hns3_reset_notify_uninit_enet
839 	 * during reset process, because driver may not be able
840 	 * to disable the ring through firmware when downing the netdev.
841 	 */
842 	if (!hns3_nic_resetting(netdev))
843 		hns3_clear_all_ring(priv->ae_handle, false);
844 
845 	hns3_reset_tx_queue(priv->ae_handle);
846 }
847 
848 static int hns3_nic_net_stop(struct net_device *netdev)
849 {
850 	struct hns3_nic_priv *priv = netdev_priv(netdev);
851 	struct hnae3_handle *h = hns3_get_handle(netdev);
852 
853 	if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
854 		return 0;
855 
856 	netif_dbg(h, drv, netdev, "net stop\n");
857 
858 	if (h->ae_algo->ops->set_timer_task)
859 		h->ae_algo->ops->set_timer_task(priv->ae_handle, false);
860 
861 	netif_carrier_off(netdev);
862 	netif_tx_disable(netdev);
863 
864 	hns3_nic_net_down(netdev);
865 
866 	return 0;
867 }
868 
869 static int hns3_nic_uc_sync(struct net_device *netdev,
870 			    const unsigned char *addr)
871 {
872 	struct hnae3_handle *h = hns3_get_handle(netdev);
873 
874 	if (h->ae_algo->ops->add_uc_addr)
875 		return h->ae_algo->ops->add_uc_addr(h, addr);
876 
877 	return 0;
878 }
879 
880 static int hns3_nic_uc_unsync(struct net_device *netdev,
881 			      const unsigned char *addr)
882 {
883 	struct hnae3_handle *h = hns3_get_handle(netdev);
884 
885 	/* need ignore the request of removing device address, because
886 	 * we store the device address and other addresses of uc list
887 	 * in the function's mac filter list.
888 	 */
889 	if (ether_addr_equal(addr, netdev->dev_addr))
890 		return 0;
891 
892 	if (h->ae_algo->ops->rm_uc_addr)
893 		return h->ae_algo->ops->rm_uc_addr(h, addr);
894 
895 	return 0;
896 }
897 
898 static int hns3_nic_mc_sync(struct net_device *netdev,
899 			    const unsigned char *addr)
900 {
901 	struct hnae3_handle *h = hns3_get_handle(netdev);
902 
903 	if (h->ae_algo->ops->add_mc_addr)
904 		return h->ae_algo->ops->add_mc_addr(h, addr);
905 
906 	return 0;
907 }
908 
909 static int hns3_nic_mc_unsync(struct net_device *netdev,
910 			      const unsigned char *addr)
911 {
912 	struct hnae3_handle *h = hns3_get_handle(netdev);
913 
914 	if (h->ae_algo->ops->rm_mc_addr)
915 		return h->ae_algo->ops->rm_mc_addr(h, addr);
916 
917 	return 0;
918 }
919 
920 static u8 hns3_get_netdev_flags(struct net_device *netdev)
921 {
922 	u8 flags = 0;
923 
924 	if (netdev->flags & IFF_PROMISC)
925 		flags = HNAE3_USER_UPE | HNAE3_USER_MPE | HNAE3_BPE;
926 	else if (netdev->flags & IFF_ALLMULTI)
927 		flags = HNAE3_USER_MPE;
928 
929 	return flags;
930 }
931 
932 static void hns3_nic_set_rx_mode(struct net_device *netdev)
933 {
934 	struct hnae3_handle *h = hns3_get_handle(netdev);
935 	u8 new_flags;
936 
937 	new_flags = hns3_get_netdev_flags(netdev);
938 
939 	__dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync);
940 	__dev_mc_sync(netdev, hns3_nic_mc_sync, hns3_nic_mc_unsync);
941 
942 	/* User mode Promisc mode enable and vlan filtering is disabled to
943 	 * let all packets in.
944 	 */
945 	h->netdev_flags = new_flags;
946 	hns3_request_update_promisc_mode(h);
947 }
948 
949 void hns3_request_update_promisc_mode(struct hnae3_handle *handle)
950 {
951 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
952 
953 	if (ops->request_update_promisc_mode)
954 		ops->request_update_promisc_mode(handle);
955 }
956 
957 static u32 hns3_tx_spare_space(struct hns3_enet_ring *ring)
958 {
959 	struct hns3_tx_spare *tx_spare = ring->tx_spare;
960 	u32 ntc, ntu;
961 
962 	/* This smp_load_acquire() pairs with smp_store_release() in
963 	 * hns3_tx_spare_update() called in tx desc cleaning process.
964 	 */
965 	ntc = smp_load_acquire(&tx_spare->last_to_clean);
966 	ntu = tx_spare->next_to_use;
967 
968 	if (ntc > ntu)
969 		return ntc - ntu - 1;
970 
971 	/* The free tx buffer is divided into two part, so pick the
972 	 * larger one.
973 	 */
974 	return (ntc > (tx_spare->len - ntu) ? ntc :
975 			(tx_spare->len - ntu)) - 1;
976 }
977 
978 static void hns3_tx_spare_update(struct hns3_enet_ring *ring)
979 {
980 	struct hns3_tx_spare *tx_spare = ring->tx_spare;
981 
982 	if (!tx_spare ||
983 	    tx_spare->last_to_clean == tx_spare->next_to_clean)
984 		return;
985 
986 	/* This smp_store_release() pairs with smp_load_acquire() in
987 	 * hns3_tx_spare_space() called in xmit process.
988 	 */
989 	smp_store_release(&tx_spare->last_to_clean,
990 			  tx_spare->next_to_clean);
991 }
992 
993 static bool hns3_can_use_tx_bounce(struct hns3_enet_ring *ring,
994 				   struct sk_buff *skb,
995 				   u32 space)
996 {
997 	u32 len = skb->len <= ring->tx_copybreak ? skb->len :
998 				skb_headlen(skb);
999 
1000 	if (len > ring->tx_copybreak)
1001 		return false;
1002 
1003 	if (ALIGN(len, dma_get_cache_alignment()) > space) {
1004 		u64_stats_update_begin(&ring->syncp);
1005 		ring->stats.tx_spare_full++;
1006 		u64_stats_update_end(&ring->syncp);
1007 		return false;
1008 	}
1009 
1010 	return true;
1011 }
1012 
1013 static bool hns3_can_use_tx_sgl(struct hns3_enet_ring *ring,
1014 				struct sk_buff *skb,
1015 				u32 space)
1016 {
1017 	if (skb->len <= ring->tx_copybreak || !tx_sgl ||
1018 	    (!skb_has_frag_list(skb) &&
1019 	     skb_shinfo(skb)->nr_frags < tx_sgl))
1020 		return false;
1021 
1022 	if (space < HNS3_MAX_SGL_SIZE) {
1023 		u64_stats_update_begin(&ring->syncp);
1024 		ring->stats.tx_spare_full++;
1025 		u64_stats_update_end(&ring->syncp);
1026 		return false;
1027 	}
1028 
1029 	return true;
1030 }
1031 
1032 static void hns3_init_tx_spare_buffer(struct hns3_enet_ring *ring)
1033 {
1034 	struct hns3_tx_spare *tx_spare;
1035 	struct page *page;
1036 	u32 alloc_size;
1037 	dma_addr_t dma;
1038 	int order;
1039 
1040 	alloc_size = tx_spare_buf_size ? tx_spare_buf_size :
1041 		     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  */
1080 static 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 
1107 static 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 
1119 static 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 
1155 static 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 
1230 static 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  */
1288 static 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 
1311 static 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 
1358 static 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 			return skb_checksum_help(skb);
1430 
1431 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1432 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1433 			       HNS3_L4T_UDP);
1434 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1435 			       (sizeof(struct udphdr) >> 2));
1436 		break;
1437 	case IPPROTO_SCTP:
1438 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1439 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1440 			       HNS3_L4T_SCTP);
1441 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1442 			       (sizeof(struct sctphdr) >> 2));
1443 		break;
1444 	default:
1445 		/* drop the skb tunnel packet if hardware don't support,
1446 		 * because hardware can't calculate csum when TSO.
1447 		 */
1448 		if (skb_is_gso(skb))
1449 			return -EDOM;
1450 
1451 		/* the stack computes the IP header already,
1452 		 * driver calculate l4 checksum when not TSO.
1453 		 */
1454 		return skb_checksum_help(skb);
1455 	}
1456 
1457 	return 0;
1458 }
1459 
1460 static int hns3_handle_vtags(struct hns3_enet_ring *tx_ring,
1461 			     struct sk_buff *skb)
1462 {
1463 	struct hnae3_handle *handle = tx_ring->tqp->handle;
1464 	struct hnae3_ae_dev *ae_dev;
1465 	struct vlan_ethhdr *vhdr;
1466 	int rc;
1467 
1468 	if (!(skb->protocol == htons(ETH_P_8021Q) ||
1469 	      skb_vlan_tag_present(skb)))
1470 		return 0;
1471 
1472 	/* For HW limitation on HNAE3_DEVICE_VERSION_V2, if port based insert
1473 	 * VLAN enabled, only one VLAN header is allowed in skb, otherwise it
1474 	 * will cause RAS error.
1475 	 */
1476 	ae_dev = pci_get_drvdata(handle->pdev);
1477 	if (unlikely(skb_vlan_tagged_multi(skb) &&
1478 		     ae_dev->dev_version <= HNAE3_DEVICE_VERSION_V2 &&
1479 		     handle->port_base_vlan_state ==
1480 		     HNAE3_PORT_BASE_VLAN_ENABLE))
1481 		return -EINVAL;
1482 
1483 	if (skb->protocol == htons(ETH_P_8021Q) &&
1484 	    !(handle->kinfo.netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
1485 		/* When HW VLAN acceleration is turned off, and the stack
1486 		 * sets the protocol to 802.1q, the driver just need to
1487 		 * set the protocol to the encapsulated ethertype.
1488 		 */
1489 		skb->protocol = vlan_get_protocol(skb);
1490 		return 0;
1491 	}
1492 
1493 	if (skb_vlan_tag_present(skb)) {
1494 		/* Based on hw strategy, use out_vtag in two layer tag case,
1495 		 * and use inner_vtag in one tag case.
1496 		 */
1497 		if (skb->protocol == htons(ETH_P_8021Q) &&
1498 		    handle->port_base_vlan_state ==
1499 		    HNAE3_PORT_BASE_VLAN_DISABLE)
1500 			rc = HNS3_OUTER_VLAN_TAG;
1501 		else
1502 			rc = HNS3_INNER_VLAN_TAG;
1503 
1504 		skb->protocol = vlan_get_protocol(skb);
1505 		return rc;
1506 	}
1507 
1508 	rc = skb_cow_head(skb, 0);
1509 	if (unlikely(rc < 0))
1510 		return rc;
1511 
1512 	vhdr = (struct vlan_ethhdr *)skb->data;
1513 	vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority << VLAN_PRIO_SHIFT)
1514 					 & VLAN_PRIO_MASK);
1515 
1516 	skb->protocol = vlan_get_protocol(skb);
1517 	return 0;
1518 }
1519 
1520 /* check if the hardware is capable of checksum offloading */
1521 static bool hns3_check_hw_tx_csum(struct sk_buff *skb)
1522 {
1523 	struct hns3_nic_priv *priv = netdev_priv(skb->dev);
1524 
1525 	/* Kindly note, due to backward compatibility of the TX descriptor,
1526 	 * HW checksum of the non-IP packets and GSO packets is handled at
1527 	 * different place in the following code
1528 	 */
1529 	if (skb_csum_is_sctp(skb) || skb_is_gso(skb) ||
1530 	    !test_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state))
1531 		return false;
1532 
1533 	return true;
1534 }
1535 
1536 static int hns3_fill_skb_desc(struct hns3_enet_ring *ring,
1537 			      struct sk_buff *skb, struct hns3_desc *desc,
1538 			      struct hns3_desc_cb *desc_cb)
1539 {
1540 	u32 ol_type_vlan_len_msec = 0;
1541 	u32 paylen_ol4cs = skb->len;
1542 	u32 type_cs_vlan_tso = 0;
1543 	u16 mss_hw_csum = 0;
1544 	u16 inner_vtag = 0;
1545 	u16 out_vtag = 0;
1546 	int ret;
1547 
1548 	ret = hns3_handle_vtags(ring, skb);
1549 	if (unlikely(ret < 0)) {
1550 		u64_stats_update_begin(&ring->syncp);
1551 		ring->stats.tx_vlan_err++;
1552 		u64_stats_update_end(&ring->syncp);
1553 		return ret;
1554 	} else if (ret == HNS3_INNER_VLAN_TAG) {
1555 		inner_vtag = skb_vlan_tag_get(skb);
1556 		inner_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1557 				VLAN_PRIO_MASK;
1558 		hns3_set_field(type_cs_vlan_tso, HNS3_TXD_VLAN_B, 1);
1559 	} else if (ret == HNS3_OUTER_VLAN_TAG) {
1560 		out_vtag = skb_vlan_tag_get(skb);
1561 		out_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1562 				VLAN_PRIO_MASK;
1563 		hns3_set_field(ol_type_vlan_len_msec, HNS3_TXD_OVLAN_B,
1564 			       1);
1565 	}
1566 
1567 	desc_cb->send_bytes = skb->len;
1568 
1569 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1570 		u8 ol4_proto, il4_proto;
1571 
1572 		if (hns3_check_hw_tx_csum(skb)) {
1573 			/* set checksum start and offset, defined in 2 Bytes */
1574 			hns3_set_field(type_cs_vlan_tso, HNS3_TXD_CSUM_START_S,
1575 				       skb_checksum_start_offset(skb) >> 1);
1576 			hns3_set_field(ol_type_vlan_len_msec,
1577 				       HNS3_TXD_CSUM_OFFSET_S,
1578 				       skb->csum_offset >> 1);
1579 			mss_hw_csum |= BIT(HNS3_TXD_HW_CS_B);
1580 			goto out_hw_tx_csum;
1581 		}
1582 
1583 		skb_reset_mac_len(skb);
1584 
1585 		ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
1586 		if (unlikely(ret < 0)) {
1587 			u64_stats_update_begin(&ring->syncp);
1588 			ring->stats.tx_l4_proto_err++;
1589 			u64_stats_update_end(&ring->syncp);
1590 			return ret;
1591 		}
1592 
1593 		ret = hns3_set_l2l3l4(skb, ol4_proto, il4_proto,
1594 				      &type_cs_vlan_tso,
1595 				      &ol_type_vlan_len_msec);
1596 		if (unlikely(ret < 0)) {
1597 			u64_stats_update_begin(&ring->syncp);
1598 			ring->stats.tx_l2l3l4_err++;
1599 			u64_stats_update_end(&ring->syncp);
1600 			return ret;
1601 		}
1602 
1603 		ret = hns3_set_tso(skb, &paylen_ol4cs, &mss_hw_csum,
1604 				   &type_cs_vlan_tso, &desc_cb->send_bytes);
1605 		if (unlikely(ret < 0)) {
1606 			u64_stats_update_begin(&ring->syncp);
1607 			ring->stats.tx_tso_err++;
1608 			u64_stats_update_end(&ring->syncp);
1609 			return ret;
1610 		}
1611 	}
1612 
1613 out_hw_tx_csum:
1614 	/* Set txbd */
1615 	desc->tx.ol_type_vlan_len_msec =
1616 		cpu_to_le32(ol_type_vlan_len_msec);
1617 	desc->tx.type_cs_vlan_tso_len = cpu_to_le32(type_cs_vlan_tso);
1618 	desc->tx.paylen_ol4cs = cpu_to_le32(paylen_ol4cs);
1619 	desc->tx.mss_hw_csum = cpu_to_le16(mss_hw_csum);
1620 	desc->tx.vlan_tag = cpu_to_le16(inner_vtag);
1621 	desc->tx.outer_vlan_tag = cpu_to_le16(out_vtag);
1622 
1623 	return 0;
1624 }
1625 
1626 static int hns3_fill_desc(struct hns3_enet_ring *ring, dma_addr_t dma,
1627 			  unsigned int size)
1628 {
1629 #define HNS3_LIKELY_BD_NUM	1
1630 
1631 	struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1632 	unsigned int frag_buf_num;
1633 	int k, sizeoflast;
1634 
1635 	if (likely(size <= HNS3_MAX_BD_SIZE)) {
1636 		desc->addr = cpu_to_le64(dma);
1637 		desc->tx.send_size = cpu_to_le16(size);
1638 		desc->tx.bdtp_fe_sc_vld_ra_ri =
1639 			cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1640 
1641 		trace_hns3_tx_desc(ring, ring->next_to_use);
1642 		ring_ptr_move_fw(ring, next_to_use);
1643 		return HNS3_LIKELY_BD_NUM;
1644 	}
1645 
1646 	frag_buf_num = hns3_tx_bd_count(size);
1647 	sizeoflast = size % HNS3_MAX_BD_SIZE;
1648 	sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1649 
1650 	/* When frag size is bigger than hardware limit, split this frag */
1651 	for (k = 0; k < frag_buf_num; k++) {
1652 		/* now, fill the descriptor */
1653 		desc->addr = cpu_to_le64(dma + HNS3_MAX_BD_SIZE * k);
1654 		desc->tx.send_size = cpu_to_le16((k == frag_buf_num - 1) ?
1655 				     (u16)sizeoflast : (u16)HNS3_MAX_BD_SIZE);
1656 		desc->tx.bdtp_fe_sc_vld_ra_ri =
1657 				cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1658 
1659 		trace_hns3_tx_desc(ring, ring->next_to_use);
1660 		/* move ring pointer to next */
1661 		ring_ptr_move_fw(ring, next_to_use);
1662 
1663 		desc = &ring->desc[ring->next_to_use];
1664 	}
1665 
1666 	return frag_buf_num;
1667 }
1668 
1669 static int hns3_map_and_fill_desc(struct hns3_enet_ring *ring, void *priv,
1670 				  unsigned int type)
1671 {
1672 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1673 	struct device *dev = ring_to_dev(ring);
1674 	unsigned int size;
1675 	dma_addr_t dma;
1676 
1677 	if (type & (DESC_TYPE_FRAGLIST_SKB | DESC_TYPE_SKB)) {
1678 		struct sk_buff *skb = (struct sk_buff *)priv;
1679 
1680 		size = skb_headlen(skb);
1681 		if (!size)
1682 			return 0;
1683 
1684 		dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1685 	} else if (type & DESC_TYPE_BOUNCE_HEAD) {
1686 		/* Head data has been filled in hns3_handle_tx_bounce(),
1687 		 * just return 0 here.
1688 		 */
1689 		return 0;
1690 	} else {
1691 		skb_frag_t *frag = (skb_frag_t *)priv;
1692 
1693 		size = skb_frag_size(frag);
1694 		if (!size)
1695 			return 0;
1696 
1697 		dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1698 	}
1699 
1700 	if (unlikely(dma_mapping_error(dev, dma))) {
1701 		u64_stats_update_begin(&ring->syncp);
1702 		ring->stats.sw_err_cnt++;
1703 		u64_stats_update_end(&ring->syncp);
1704 		return -ENOMEM;
1705 	}
1706 
1707 	desc_cb->priv = priv;
1708 	desc_cb->length = size;
1709 	desc_cb->dma = dma;
1710 	desc_cb->type = type;
1711 
1712 	return hns3_fill_desc(ring, dma, size);
1713 }
1714 
1715 static unsigned int hns3_skb_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1716 				    unsigned int bd_num)
1717 {
1718 	unsigned int size;
1719 	int i;
1720 
1721 	size = skb_headlen(skb);
1722 	while (size > HNS3_MAX_BD_SIZE) {
1723 		bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1724 		size -= HNS3_MAX_BD_SIZE;
1725 
1726 		if (bd_num > HNS3_MAX_TSO_BD_NUM)
1727 			return bd_num;
1728 	}
1729 
1730 	if (size) {
1731 		bd_size[bd_num++] = size;
1732 		if (bd_num > HNS3_MAX_TSO_BD_NUM)
1733 			return bd_num;
1734 	}
1735 
1736 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1737 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1738 		size = skb_frag_size(frag);
1739 		if (!size)
1740 			continue;
1741 
1742 		while (size > HNS3_MAX_BD_SIZE) {
1743 			bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1744 			size -= HNS3_MAX_BD_SIZE;
1745 
1746 			if (bd_num > HNS3_MAX_TSO_BD_NUM)
1747 				return bd_num;
1748 		}
1749 
1750 		bd_size[bd_num++] = size;
1751 		if (bd_num > HNS3_MAX_TSO_BD_NUM)
1752 			return bd_num;
1753 	}
1754 
1755 	return bd_num;
1756 }
1757 
1758 static unsigned int hns3_tx_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1759 				   u8 max_non_tso_bd_num, unsigned int bd_num,
1760 				   unsigned int recursion_level)
1761 {
1762 #define HNS3_MAX_RECURSION_LEVEL	24
1763 
1764 	struct sk_buff *frag_skb;
1765 
1766 	/* If the total len is within the max bd limit */
1767 	if (likely(skb->len <= HNS3_MAX_BD_SIZE && !recursion_level &&
1768 		   !skb_has_frag_list(skb) &&
1769 		   skb_shinfo(skb)->nr_frags < max_non_tso_bd_num))
1770 		return skb_shinfo(skb)->nr_frags + 1U;
1771 
1772 	if (unlikely(recursion_level >= HNS3_MAX_RECURSION_LEVEL))
1773 		return UINT_MAX;
1774 
1775 	bd_num = hns3_skb_bd_num(skb, bd_size, bd_num);
1776 	if (!skb_has_frag_list(skb) || bd_num > HNS3_MAX_TSO_BD_NUM)
1777 		return bd_num;
1778 
1779 	skb_walk_frags(skb, frag_skb) {
1780 		bd_num = hns3_tx_bd_num(frag_skb, bd_size, max_non_tso_bd_num,
1781 					bd_num, recursion_level + 1);
1782 		if (bd_num > HNS3_MAX_TSO_BD_NUM)
1783 			return bd_num;
1784 	}
1785 
1786 	return bd_num;
1787 }
1788 
1789 static unsigned int hns3_gso_hdr_len(struct sk_buff *skb)
1790 {
1791 	if (!skb->encapsulation)
1792 		return skb_transport_offset(skb) + tcp_hdrlen(skb);
1793 
1794 	return skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb);
1795 }
1796 
1797 /* HW need every continuous max_non_tso_bd_num buffer data to be larger
1798  * than MSS, we simplify it by ensuring skb_headlen + the first continuous
1799  * max_non_tso_bd_num - 1 frags to be larger than gso header len + mss,
1800  * and the remaining continuous max_non_tso_bd_num - 1 frags to be larger
1801  * than MSS except the last max_non_tso_bd_num - 1 frags.
1802  */
1803 static bool hns3_skb_need_linearized(struct sk_buff *skb, unsigned int *bd_size,
1804 				     unsigned int bd_num, u8 max_non_tso_bd_num)
1805 {
1806 	unsigned int tot_len = 0;
1807 	int i;
1808 
1809 	for (i = 0; i < max_non_tso_bd_num - 1U; i++)
1810 		tot_len += bd_size[i];
1811 
1812 	/* ensure the first max_non_tso_bd_num frags is greater than
1813 	 * mss + header
1814 	 */
1815 	if (tot_len + bd_size[max_non_tso_bd_num - 1U] <
1816 	    skb_shinfo(skb)->gso_size + hns3_gso_hdr_len(skb))
1817 		return true;
1818 
1819 	/* ensure every continuous max_non_tso_bd_num - 1 buffer is greater
1820 	 * than mss except the last one.
1821 	 */
1822 	for (i = 0; i < bd_num - max_non_tso_bd_num; i++) {
1823 		tot_len -= bd_size[i];
1824 		tot_len += bd_size[i + max_non_tso_bd_num - 1U];
1825 
1826 		if (tot_len < skb_shinfo(skb)->gso_size)
1827 			return true;
1828 	}
1829 
1830 	return false;
1831 }
1832 
1833 void hns3_shinfo_pack(struct skb_shared_info *shinfo, __u32 *size)
1834 {
1835 	int i;
1836 
1837 	for (i = 0; i < MAX_SKB_FRAGS; i++)
1838 		size[i] = skb_frag_size(&shinfo->frags[i]);
1839 }
1840 
1841 static int hns3_skb_linearize(struct hns3_enet_ring *ring,
1842 			      struct sk_buff *skb,
1843 			      u8 max_non_tso_bd_num,
1844 			      unsigned int bd_num)
1845 {
1846 	/* 'bd_num == UINT_MAX' means the skb' fraglist has a
1847 	 * recursion level of over HNS3_MAX_RECURSION_LEVEL.
1848 	 */
1849 	if (bd_num == UINT_MAX) {
1850 		u64_stats_update_begin(&ring->syncp);
1851 		ring->stats.over_max_recursion++;
1852 		u64_stats_update_end(&ring->syncp);
1853 		return -ENOMEM;
1854 	}
1855 
1856 	/* The skb->len has exceeded the hw limitation, linearization
1857 	 * will not help.
1858 	 */
1859 	if (skb->len > HNS3_MAX_TSO_SIZE ||
1860 	    (!skb_is_gso(skb) && skb->len >
1861 	     HNS3_MAX_NON_TSO_SIZE(max_non_tso_bd_num))) {
1862 		u64_stats_update_begin(&ring->syncp);
1863 		ring->stats.hw_limitation++;
1864 		u64_stats_update_end(&ring->syncp);
1865 		return -ENOMEM;
1866 	}
1867 
1868 	if (__skb_linearize(skb)) {
1869 		u64_stats_update_begin(&ring->syncp);
1870 		ring->stats.sw_err_cnt++;
1871 		u64_stats_update_end(&ring->syncp);
1872 		return -ENOMEM;
1873 	}
1874 
1875 	return 0;
1876 }
1877 
1878 static int hns3_nic_maybe_stop_tx(struct hns3_enet_ring *ring,
1879 				  struct net_device *netdev,
1880 				  struct sk_buff *skb)
1881 {
1882 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1883 	u8 max_non_tso_bd_num = priv->max_non_tso_bd_num;
1884 	unsigned int bd_size[HNS3_MAX_TSO_BD_NUM + 1U];
1885 	unsigned int bd_num;
1886 
1887 	bd_num = hns3_tx_bd_num(skb, bd_size, max_non_tso_bd_num, 0, 0);
1888 	if (unlikely(bd_num > max_non_tso_bd_num)) {
1889 		if (bd_num <= HNS3_MAX_TSO_BD_NUM && skb_is_gso(skb) &&
1890 		    !hns3_skb_need_linearized(skb, bd_size, bd_num,
1891 					      max_non_tso_bd_num)) {
1892 			trace_hns3_over_max_bd(skb);
1893 			goto out;
1894 		}
1895 
1896 		if (hns3_skb_linearize(ring, skb, max_non_tso_bd_num,
1897 				       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 
1907 out:
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 
1931 static 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 
1971 static 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 
2005 static 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 
2026 static 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 
2038 static 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 
2094 static 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 
2153 static 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 
2169 out:
2170 	return hns3_fill_skb_to_desc(ring, skb, DESC_TYPE_SKB);
2171 }
2172 
2173 netdev_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 
2243 fill_err:
2244 	hns3_clear_desc(ring, next_to_use_head);
2245 
2246 out_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 
2252 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
2253 {
2254 	struct hnae3_handle *h = hns3_get_handle(netdev);
2255 	struct sockaddr *mac_addr = p;
2256 	int ret;
2257 
2258 	if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
2259 		return -EADDRNOTAVAIL;
2260 
2261 	if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
2262 		netdev_info(netdev, "already using mac address %pM\n",
2263 			    mac_addr->sa_data);
2264 		return 0;
2265 	}
2266 
2267 	/* For VF device, if there is a perm_addr, then the user will not
2268 	 * be allowed to change the address.
2269 	 */
2270 	if (!hns3_is_phys_func(h->pdev) &&
2271 	    !is_zero_ether_addr(netdev->perm_addr)) {
2272 		netdev_err(netdev, "has permanent MAC %pM, user MAC %pM not allow\n",
2273 			   netdev->perm_addr, mac_addr->sa_data);
2274 		return -EPERM;
2275 	}
2276 
2277 	ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
2278 	if (ret) {
2279 		netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
2280 		return ret;
2281 	}
2282 
2283 	ether_addr_copy(netdev->dev_addr, mac_addr->sa_data);
2284 
2285 	return 0;
2286 }
2287 
2288 static int hns3_nic_do_ioctl(struct net_device *netdev,
2289 			     struct ifreq *ifr, int cmd)
2290 {
2291 	struct hnae3_handle *h = hns3_get_handle(netdev);
2292 
2293 	if (!netif_running(netdev))
2294 		return -EINVAL;
2295 
2296 	if (!h->ae_algo->ops->do_ioctl)
2297 		return -EOPNOTSUPP;
2298 
2299 	return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
2300 }
2301 
2302 static int hns3_nic_set_features(struct net_device *netdev,
2303 				 netdev_features_t features)
2304 {
2305 	netdev_features_t changed = netdev->features ^ features;
2306 	struct hns3_nic_priv *priv = netdev_priv(netdev);
2307 	struct hnae3_handle *h = priv->ae_handle;
2308 	bool enable;
2309 	int ret;
2310 
2311 	if (changed & (NETIF_F_GRO_HW) && h->ae_algo->ops->set_gro_en) {
2312 		enable = !!(features & NETIF_F_GRO_HW);
2313 		ret = h->ae_algo->ops->set_gro_en(h, enable);
2314 		if (ret)
2315 			return ret;
2316 	}
2317 
2318 	if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
2319 	    h->ae_algo->ops->enable_hw_strip_rxvtag) {
2320 		enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
2321 		ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, enable);
2322 		if (ret)
2323 			return ret;
2324 	}
2325 
2326 	if ((changed & NETIF_F_NTUPLE) && h->ae_algo->ops->enable_fd) {
2327 		enable = !!(features & NETIF_F_NTUPLE);
2328 		h->ae_algo->ops->enable_fd(h, enable);
2329 	}
2330 
2331 	if ((netdev->features & NETIF_F_HW_TC) > (features & NETIF_F_HW_TC) &&
2332 	    h->ae_algo->ops->cls_flower_active(h)) {
2333 		netdev_err(netdev,
2334 			   "there are offloaded TC filters active, cannot disable HW TC offload");
2335 		return -EINVAL;
2336 	}
2337 
2338 	if ((changed & NETIF_F_HW_VLAN_CTAG_FILTER) &&
2339 	    h->ae_algo->ops->enable_vlan_filter) {
2340 		enable = !!(features & NETIF_F_HW_VLAN_CTAG_FILTER);
2341 		ret = h->ae_algo->ops->enable_vlan_filter(h, enable);
2342 		if (ret)
2343 			return ret;
2344 	}
2345 
2346 	netdev->features = features;
2347 	return 0;
2348 }
2349 
2350 static netdev_features_t hns3_features_check(struct sk_buff *skb,
2351 					     struct net_device *dev,
2352 					     netdev_features_t features)
2353 {
2354 #define HNS3_MAX_HDR_LEN	480U
2355 #define HNS3_MAX_L4_HDR_LEN	60U
2356 
2357 	size_t len;
2358 
2359 	if (skb->ip_summed != CHECKSUM_PARTIAL)
2360 		return features;
2361 
2362 	if (skb->encapsulation)
2363 		len = skb_inner_transport_header(skb) - skb->data;
2364 	else
2365 		len = skb_transport_header(skb) - skb->data;
2366 
2367 	/* Assume L4 is 60 byte as TCP is the only protocol with a
2368 	 * a flexible value, and it's max len is 60 bytes.
2369 	 */
2370 	len += HNS3_MAX_L4_HDR_LEN;
2371 
2372 	/* Hardware only supports checksum on the skb with a max header
2373 	 * len of 480 bytes.
2374 	 */
2375 	if (len > HNS3_MAX_HDR_LEN)
2376 		features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2377 
2378 	return features;
2379 }
2380 
2381 static void hns3_nic_get_stats64(struct net_device *netdev,
2382 				 struct rtnl_link_stats64 *stats)
2383 {
2384 	struct hns3_nic_priv *priv = netdev_priv(netdev);
2385 	int queue_num = priv->ae_handle->kinfo.num_tqps;
2386 	struct hnae3_handle *handle = priv->ae_handle;
2387 	struct hns3_enet_ring *ring;
2388 	u64 rx_length_errors = 0;
2389 	u64 rx_crc_errors = 0;
2390 	u64 rx_multicast = 0;
2391 	unsigned int start;
2392 	u64 tx_errors = 0;
2393 	u64 rx_errors = 0;
2394 	unsigned int idx;
2395 	u64 tx_bytes = 0;
2396 	u64 rx_bytes = 0;
2397 	u64 tx_pkts = 0;
2398 	u64 rx_pkts = 0;
2399 	u64 tx_drop = 0;
2400 	u64 rx_drop = 0;
2401 
2402 	if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
2403 		return;
2404 
2405 	handle->ae_algo->ops->update_stats(handle, &netdev->stats);
2406 
2407 	for (idx = 0; idx < queue_num; idx++) {
2408 		/* fetch the tx stats */
2409 		ring = &priv->ring[idx];
2410 		do {
2411 			start = u64_stats_fetch_begin_irq(&ring->syncp);
2412 			tx_bytes += ring->stats.tx_bytes;
2413 			tx_pkts += ring->stats.tx_pkts;
2414 			tx_drop += ring->stats.sw_err_cnt;
2415 			tx_drop += ring->stats.tx_vlan_err;
2416 			tx_drop += ring->stats.tx_l4_proto_err;
2417 			tx_drop += ring->stats.tx_l2l3l4_err;
2418 			tx_drop += ring->stats.tx_tso_err;
2419 			tx_drop += ring->stats.over_max_recursion;
2420 			tx_drop += ring->stats.hw_limitation;
2421 			tx_drop += ring->stats.copy_bits_err;
2422 			tx_drop += ring->stats.skb2sgl_err;
2423 			tx_drop += ring->stats.map_sg_err;
2424 			tx_errors += ring->stats.sw_err_cnt;
2425 			tx_errors += ring->stats.tx_vlan_err;
2426 			tx_errors += ring->stats.tx_l4_proto_err;
2427 			tx_errors += ring->stats.tx_l2l3l4_err;
2428 			tx_errors += ring->stats.tx_tso_err;
2429 			tx_errors += ring->stats.over_max_recursion;
2430 			tx_errors += ring->stats.hw_limitation;
2431 			tx_errors += ring->stats.copy_bits_err;
2432 			tx_errors += ring->stats.skb2sgl_err;
2433 			tx_errors += ring->stats.map_sg_err;
2434 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
2435 
2436 		/* fetch the rx stats */
2437 		ring = &priv->ring[idx + queue_num];
2438 		do {
2439 			start = u64_stats_fetch_begin_irq(&ring->syncp);
2440 			rx_bytes += ring->stats.rx_bytes;
2441 			rx_pkts += ring->stats.rx_pkts;
2442 			rx_drop += ring->stats.l2_err;
2443 			rx_errors += ring->stats.l2_err;
2444 			rx_errors += ring->stats.l3l4_csum_err;
2445 			rx_crc_errors += ring->stats.l2_err;
2446 			rx_multicast += ring->stats.rx_multicast;
2447 			rx_length_errors += ring->stats.err_pkt_len;
2448 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
2449 	}
2450 
2451 	stats->tx_bytes = tx_bytes;
2452 	stats->tx_packets = tx_pkts;
2453 	stats->rx_bytes = rx_bytes;
2454 	stats->rx_packets = rx_pkts;
2455 
2456 	stats->rx_errors = rx_errors;
2457 	stats->multicast = rx_multicast;
2458 	stats->rx_length_errors = rx_length_errors;
2459 	stats->rx_crc_errors = rx_crc_errors;
2460 	stats->rx_missed_errors = netdev->stats.rx_missed_errors;
2461 
2462 	stats->tx_errors = tx_errors;
2463 	stats->rx_dropped = rx_drop;
2464 	stats->tx_dropped = tx_drop;
2465 	stats->collisions = netdev->stats.collisions;
2466 	stats->rx_over_errors = netdev->stats.rx_over_errors;
2467 	stats->rx_frame_errors = netdev->stats.rx_frame_errors;
2468 	stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
2469 	stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
2470 	stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
2471 	stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
2472 	stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
2473 	stats->tx_window_errors = netdev->stats.tx_window_errors;
2474 	stats->rx_compressed = netdev->stats.rx_compressed;
2475 	stats->tx_compressed = netdev->stats.tx_compressed;
2476 }
2477 
2478 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
2479 {
2480 	struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
2481 	struct hnae3_knic_private_info *kinfo;
2482 	u8 tc = mqprio_qopt->qopt.num_tc;
2483 	u16 mode = mqprio_qopt->mode;
2484 	u8 hw = mqprio_qopt->qopt.hw;
2485 	struct hnae3_handle *h;
2486 
2487 	if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
2488 	       mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
2489 		return -EOPNOTSUPP;
2490 
2491 	if (tc > HNAE3_MAX_TC)
2492 		return -EINVAL;
2493 
2494 	if (!netdev)
2495 		return -EINVAL;
2496 
2497 	h = hns3_get_handle(netdev);
2498 	kinfo = &h->kinfo;
2499 
2500 	netif_dbg(h, drv, netdev, "setup tc: num_tc=%u\n", tc);
2501 
2502 	return (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
2503 		kinfo->dcb_ops->setup_tc(h, mqprio_qopt) : -EOPNOTSUPP;
2504 }
2505 
2506 static int hns3_setup_tc_cls_flower(struct hns3_nic_priv *priv,
2507 				    struct flow_cls_offload *flow)
2508 {
2509 	int tc = tc_classid_to_hwtc(priv->netdev, flow->classid);
2510 	struct hnae3_handle *h = hns3_get_handle(priv->netdev);
2511 
2512 	switch (flow->command) {
2513 	case FLOW_CLS_REPLACE:
2514 		if (h->ae_algo->ops->add_cls_flower)
2515 			return h->ae_algo->ops->add_cls_flower(h, flow, tc);
2516 		break;
2517 	case FLOW_CLS_DESTROY:
2518 		if (h->ae_algo->ops->del_cls_flower)
2519 			return h->ae_algo->ops->del_cls_flower(h, flow);
2520 		break;
2521 	default:
2522 		break;
2523 	}
2524 
2525 	return -EOPNOTSUPP;
2526 }
2527 
2528 static int hns3_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
2529 				  void *cb_priv)
2530 {
2531 	struct hns3_nic_priv *priv = cb_priv;
2532 
2533 	if (!tc_cls_can_offload_and_chain0(priv->netdev, type_data))
2534 		return -EOPNOTSUPP;
2535 
2536 	switch (type) {
2537 	case TC_SETUP_CLSFLOWER:
2538 		return hns3_setup_tc_cls_flower(priv, type_data);
2539 	default:
2540 		return -EOPNOTSUPP;
2541 	}
2542 }
2543 
2544 static LIST_HEAD(hns3_block_cb_list);
2545 
2546 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
2547 			     void *type_data)
2548 {
2549 	struct hns3_nic_priv *priv = netdev_priv(dev);
2550 	int ret;
2551 
2552 	switch (type) {
2553 	case TC_SETUP_QDISC_MQPRIO:
2554 		ret = hns3_setup_tc(dev, type_data);
2555 		break;
2556 	case TC_SETUP_BLOCK:
2557 		ret = flow_block_cb_setup_simple(type_data,
2558 						 &hns3_block_cb_list,
2559 						 hns3_setup_tc_block_cb,
2560 						 priv, priv, true);
2561 		break;
2562 	default:
2563 		return -EOPNOTSUPP;
2564 	}
2565 
2566 	return ret;
2567 }
2568 
2569 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
2570 				__be16 proto, u16 vid)
2571 {
2572 	struct hnae3_handle *h = hns3_get_handle(netdev);
2573 	int ret = -EIO;
2574 
2575 	if (h->ae_algo->ops->set_vlan_filter)
2576 		ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
2577 
2578 	return ret;
2579 }
2580 
2581 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
2582 				 __be16 proto, u16 vid)
2583 {
2584 	struct hnae3_handle *h = hns3_get_handle(netdev);
2585 	int ret = -EIO;
2586 
2587 	if (h->ae_algo->ops->set_vlan_filter)
2588 		ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
2589 
2590 	return ret;
2591 }
2592 
2593 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
2594 				u8 qos, __be16 vlan_proto)
2595 {
2596 	struct hnae3_handle *h = hns3_get_handle(netdev);
2597 	int ret = -EIO;
2598 
2599 	netif_dbg(h, drv, netdev,
2600 		  "set vf vlan: vf=%d, vlan=%u, qos=%u, vlan_proto=0x%x\n",
2601 		  vf, vlan, qos, ntohs(vlan_proto));
2602 
2603 	if (h->ae_algo->ops->set_vf_vlan_filter)
2604 		ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
2605 							  qos, vlan_proto);
2606 
2607 	return ret;
2608 }
2609 
2610 static int hns3_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
2611 {
2612 	struct hnae3_handle *handle = hns3_get_handle(netdev);
2613 
2614 	if (hns3_nic_resetting(netdev))
2615 		return -EBUSY;
2616 
2617 	if (!handle->ae_algo->ops->set_vf_spoofchk)
2618 		return -EOPNOTSUPP;
2619 
2620 	return handle->ae_algo->ops->set_vf_spoofchk(handle, vf, enable);
2621 }
2622 
2623 static int hns3_set_vf_trust(struct net_device *netdev, int vf, bool enable)
2624 {
2625 	struct hnae3_handle *handle = hns3_get_handle(netdev);
2626 
2627 	if (!handle->ae_algo->ops->set_vf_trust)
2628 		return -EOPNOTSUPP;
2629 
2630 	return handle->ae_algo->ops->set_vf_trust(handle, vf, enable);
2631 }
2632 
2633 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
2634 {
2635 	struct hnae3_handle *h = hns3_get_handle(netdev);
2636 	int ret;
2637 
2638 	if (hns3_nic_resetting(netdev))
2639 		return -EBUSY;
2640 
2641 	if (!h->ae_algo->ops->set_mtu)
2642 		return -EOPNOTSUPP;
2643 
2644 	netif_dbg(h, drv, netdev,
2645 		  "change mtu from %u to %d\n", netdev->mtu, new_mtu);
2646 
2647 	ret = h->ae_algo->ops->set_mtu(h, new_mtu);
2648 	if (ret)
2649 		netdev_err(netdev, "failed to change MTU in hardware %d\n",
2650 			   ret);
2651 	else
2652 		netdev->mtu = new_mtu;
2653 
2654 	return ret;
2655 }
2656 
2657 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
2658 {
2659 	struct hns3_nic_priv *priv = netdev_priv(ndev);
2660 	struct hnae3_handle *h = hns3_get_handle(ndev);
2661 	struct hns3_enet_ring *tx_ring;
2662 	struct napi_struct *napi;
2663 	int timeout_queue = 0;
2664 	int hw_head, hw_tail;
2665 	int fbd_num, fbd_oft;
2666 	int ebd_num, ebd_oft;
2667 	int bd_num, bd_err;
2668 	int ring_en, tc;
2669 	int i;
2670 
2671 	/* Find the stopped queue the same way the stack does */
2672 	for (i = 0; i < ndev->num_tx_queues; i++) {
2673 		struct netdev_queue *q;
2674 		unsigned long trans_start;
2675 
2676 		q = netdev_get_tx_queue(ndev, i);
2677 		trans_start = q->trans_start;
2678 		if (netif_xmit_stopped(q) &&
2679 		    time_after(jiffies,
2680 			       (trans_start + ndev->watchdog_timeo))) {
2681 			timeout_queue = i;
2682 			netdev_info(ndev, "queue state: 0x%lx, delta msecs: %u\n",
2683 				    q->state,
2684 				    jiffies_to_msecs(jiffies - trans_start));
2685 			break;
2686 		}
2687 	}
2688 
2689 	if (i == ndev->num_tx_queues) {
2690 		netdev_info(ndev,
2691 			    "no netdev TX timeout queue found, timeout count: %llu\n",
2692 			    priv->tx_timeout_count);
2693 		return false;
2694 	}
2695 
2696 	priv->tx_timeout_count++;
2697 
2698 	tx_ring = &priv->ring[timeout_queue];
2699 	napi = &tx_ring->tqp_vector->napi;
2700 
2701 	netdev_info(ndev,
2702 		    "tx_timeout count: %llu, queue id: %d, SW_NTU: 0x%x, SW_NTC: 0x%x, napi state: %lu\n",
2703 		    priv->tx_timeout_count, timeout_queue, tx_ring->next_to_use,
2704 		    tx_ring->next_to_clean, napi->state);
2705 
2706 	netdev_info(ndev,
2707 		    "tx_pkts: %llu, tx_bytes: %llu, sw_err_cnt: %llu, tx_pending: %d\n",
2708 		    tx_ring->stats.tx_pkts, tx_ring->stats.tx_bytes,
2709 		    tx_ring->stats.sw_err_cnt, tx_ring->pending_buf);
2710 
2711 	netdev_info(ndev,
2712 		    "seg_pkt_cnt: %llu, tx_more: %llu, restart_queue: %llu, tx_busy: %llu\n",
2713 		    tx_ring->stats.seg_pkt_cnt, tx_ring->stats.tx_more,
2714 		    tx_ring->stats.restart_queue, tx_ring->stats.tx_busy);
2715 
2716 	/* When mac received many pause frames continuous, it's unable to send
2717 	 * packets, which may cause tx timeout
2718 	 */
2719 	if (h->ae_algo->ops->get_mac_stats) {
2720 		struct hns3_mac_stats mac_stats;
2721 
2722 		h->ae_algo->ops->get_mac_stats(h, &mac_stats);
2723 		netdev_info(ndev, "tx_pause_cnt: %llu, rx_pause_cnt: %llu\n",
2724 			    mac_stats.tx_pause_cnt, mac_stats.rx_pause_cnt);
2725 	}
2726 
2727 	hw_head = readl_relaxed(tx_ring->tqp->io_base +
2728 				HNS3_RING_TX_RING_HEAD_REG);
2729 	hw_tail = readl_relaxed(tx_ring->tqp->io_base +
2730 				HNS3_RING_TX_RING_TAIL_REG);
2731 	fbd_num = readl_relaxed(tx_ring->tqp->io_base +
2732 				HNS3_RING_TX_RING_FBDNUM_REG);
2733 	fbd_oft = readl_relaxed(tx_ring->tqp->io_base +
2734 				HNS3_RING_TX_RING_OFFSET_REG);
2735 	ebd_num = readl_relaxed(tx_ring->tqp->io_base +
2736 				HNS3_RING_TX_RING_EBDNUM_REG);
2737 	ebd_oft = readl_relaxed(tx_ring->tqp->io_base +
2738 				HNS3_RING_TX_RING_EBD_OFFSET_REG);
2739 	bd_num = readl_relaxed(tx_ring->tqp->io_base +
2740 			       HNS3_RING_TX_RING_BD_NUM_REG);
2741 	bd_err = readl_relaxed(tx_ring->tqp->io_base +
2742 			       HNS3_RING_TX_RING_BD_ERR_REG);
2743 	ring_en = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_EN_REG);
2744 	tc = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_TX_RING_TC_REG);
2745 
2746 	netdev_info(ndev,
2747 		    "BD_NUM: 0x%x HW_HEAD: 0x%x, HW_TAIL: 0x%x, BD_ERR: 0x%x, INT: 0x%x\n",
2748 		    bd_num, hw_head, hw_tail, bd_err,
2749 		    readl(tx_ring->tqp_vector->mask_addr));
2750 	netdev_info(ndev,
2751 		    "RING_EN: 0x%x, TC: 0x%x, FBD_NUM: 0x%x FBD_OFT: 0x%x, EBD_NUM: 0x%x, EBD_OFT: 0x%x\n",
2752 		    ring_en, tc, fbd_num, fbd_oft, ebd_num, ebd_oft);
2753 
2754 	return true;
2755 }
2756 
2757 static void hns3_nic_net_timeout(struct net_device *ndev, unsigned int txqueue)
2758 {
2759 	struct hns3_nic_priv *priv = netdev_priv(ndev);
2760 	struct hnae3_handle *h = priv->ae_handle;
2761 
2762 	if (!hns3_get_tx_timeo_queue_info(ndev))
2763 		return;
2764 
2765 	/* request the reset, and let the hclge to determine
2766 	 * which reset level should be done
2767 	 */
2768 	if (h->ae_algo->ops->reset_event)
2769 		h->ae_algo->ops->reset_event(h->pdev, h);
2770 }
2771 
2772 #ifdef CONFIG_RFS_ACCEL
2773 static int hns3_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
2774 			      u16 rxq_index, u32 flow_id)
2775 {
2776 	struct hnae3_handle *h = hns3_get_handle(dev);
2777 	struct flow_keys fkeys;
2778 
2779 	if (!h->ae_algo->ops->add_arfs_entry)
2780 		return -EOPNOTSUPP;
2781 
2782 	if (skb->encapsulation)
2783 		return -EPROTONOSUPPORT;
2784 
2785 	if (!skb_flow_dissect_flow_keys(skb, &fkeys, 0))
2786 		return -EPROTONOSUPPORT;
2787 
2788 	if ((fkeys.basic.n_proto != htons(ETH_P_IP) &&
2789 	     fkeys.basic.n_proto != htons(ETH_P_IPV6)) ||
2790 	    (fkeys.basic.ip_proto != IPPROTO_TCP &&
2791 	     fkeys.basic.ip_proto != IPPROTO_UDP))
2792 		return -EPROTONOSUPPORT;
2793 
2794 	return h->ae_algo->ops->add_arfs_entry(h, rxq_index, flow_id, &fkeys);
2795 }
2796 #endif
2797 
2798 static int hns3_nic_get_vf_config(struct net_device *ndev, int vf,
2799 				  struct ifla_vf_info *ivf)
2800 {
2801 	struct hnae3_handle *h = hns3_get_handle(ndev);
2802 
2803 	if (!h->ae_algo->ops->get_vf_config)
2804 		return -EOPNOTSUPP;
2805 
2806 	return h->ae_algo->ops->get_vf_config(h, vf, ivf);
2807 }
2808 
2809 static int hns3_nic_set_vf_link_state(struct net_device *ndev, int vf,
2810 				      int link_state)
2811 {
2812 	struct hnae3_handle *h = hns3_get_handle(ndev);
2813 
2814 	if (!h->ae_algo->ops->set_vf_link_state)
2815 		return -EOPNOTSUPP;
2816 
2817 	return h->ae_algo->ops->set_vf_link_state(h, vf, link_state);
2818 }
2819 
2820 static int hns3_nic_set_vf_rate(struct net_device *ndev, int vf,
2821 				int min_tx_rate, int max_tx_rate)
2822 {
2823 	struct hnae3_handle *h = hns3_get_handle(ndev);
2824 
2825 	if (!h->ae_algo->ops->set_vf_rate)
2826 		return -EOPNOTSUPP;
2827 
2828 	return h->ae_algo->ops->set_vf_rate(h, vf, min_tx_rate, max_tx_rate,
2829 					    false);
2830 }
2831 
2832 static int hns3_nic_set_vf_mac(struct net_device *netdev, int vf_id, u8 *mac)
2833 {
2834 	struct hnae3_handle *h = hns3_get_handle(netdev);
2835 
2836 	if (!h->ae_algo->ops->set_vf_mac)
2837 		return -EOPNOTSUPP;
2838 
2839 	if (is_multicast_ether_addr(mac)) {
2840 		netdev_err(netdev,
2841 			   "Invalid MAC:%pM specified. Could not set MAC\n",
2842 			   mac);
2843 		return -EINVAL;
2844 	}
2845 
2846 	return h->ae_algo->ops->set_vf_mac(h, vf_id, mac);
2847 }
2848 
2849 static const struct net_device_ops hns3_nic_netdev_ops = {
2850 	.ndo_open		= hns3_nic_net_open,
2851 	.ndo_stop		= hns3_nic_net_stop,
2852 	.ndo_start_xmit		= hns3_nic_net_xmit,
2853 	.ndo_tx_timeout		= hns3_nic_net_timeout,
2854 	.ndo_set_mac_address	= hns3_nic_net_set_mac_address,
2855 	.ndo_eth_ioctl		= hns3_nic_do_ioctl,
2856 	.ndo_change_mtu		= hns3_nic_change_mtu,
2857 	.ndo_set_features	= hns3_nic_set_features,
2858 	.ndo_features_check	= hns3_features_check,
2859 	.ndo_get_stats64	= hns3_nic_get_stats64,
2860 	.ndo_setup_tc		= hns3_nic_setup_tc,
2861 	.ndo_set_rx_mode	= hns3_nic_set_rx_mode,
2862 	.ndo_vlan_rx_add_vid	= hns3_vlan_rx_add_vid,
2863 	.ndo_vlan_rx_kill_vid	= hns3_vlan_rx_kill_vid,
2864 	.ndo_set_vf_vlan	= hns3_ndo_set_vf_vlan,
2865 	.ndo_set_vf_spoofchk	= hns3_set_vf_spoofchk,
2866 	.ndo_set_vf_trust	= hns3_set_vf_trust,
2867 #ifdef CONFIG_RFS_ACCEL
2868 	.ndo_rx_flow_steer	= hns3_rx_flow_steer,
2869 #endif
2870 	.ndo_get_vf_config	= hns3_nic_get_vf_config,
2871 	.ndo_set_vf_link_state	= hns3_nic_set_vf_link_state,
2872 	.ndo_set_vf_rate	= hns3_nic_set_vf_rate,
2873 	.ndo_set_vf_mac		= hns3_nic_set_vf_mac,
2874 };
2875 
2876 bool hns3_is_phys_func(struct pci_dev *pdev)
2877 {
2878 	u32 dev_id = pdev->device;
2879 
2880 	switch (dev_id) {
2881 	case HNAE3_DEV_ID_GE:
2882 	case HNAE3_DEV_ID_25GE:
2883 	case HNAE3_DEV_ID_25GE_RDMA:
2884 	case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
2885 	case HNAE3_DEV_ID_50GE_RDMA:
2886 	case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
2887 	case HNAE3_DEV_ID_100G_RDMA_MACSEC:
2888 	case HNAE3_DEV_ID_200G_RDMA:
2889 		return true;
2890 	case HNAE3_DEV_ID_VF:
2891 	case HNAE3_DEV_ID_RDMA_DCB_PFC_VF:
2892 		return false;
2893 	default:
2894 		dev_warn(&pdev->dev, "un-recognized pci device-id %u",
2895 			 dev_id);
2896 	}
2897 
2898 	return false;
2899 }
2900 
2901 static void hns3_disable_sriov(struct pci_dev *pdev)
2902 {
2903 	/* If our VFs are assigned we cannot shut down SR-IOV
2904 	 * without causing issues, so just leave the hardware
2905 	 * available but disabled
2906 	 */
2907 	if (pci_vfs_assigned(pdev)) {
2908 		dev_warn(&pdev->dev,
2909 			 "disabling driver while VFs are assigned\n");
2910 		return;
2911 	}
2912 
2913 	pci_disable_sriov(pdev);
2914 }
2915 
2916 /* hns3_probe - Device initialization routine
2917  * @pdev: PCI device information struct
2918  * @ent: entry in hns3_pci_tbl
2919  *
2920  * hns3_probe initializes a PF identified by a pci_dev structure.
2921  * The OS initialization, configuring of the PF private structure,
2922  * and a hardware reset occur.
2923  *
2924  * Returns 0 on success, negative on failure
2925  */
2926 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2927 {
2928 	struct hnae3_ae_dev *ae_dev;
2929 	int ret;
2930 
2931 	ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev), GFP_KERNEL);
2932 	if (!ae_dev)
2933 		return -ENOMEM;
2934 
2935 	ae_dev->pdev = pdev;
2936 	ae_dev->flag = ent->driver_data;
2937 	pci_set_drvdata(pdev, ae_dev);
2938 
2939 	ret = hnae3_register_ae_dev(ae_dev);
2940 	if (ret)
2941 		pci_set_drvdata(pdev, NULL);
2942 
2943 	return ret;
2944 }
2945 
2946 /* hns3_remove - Device removal routine
2947  * @pdev: PCI device information struct
2948  */
2949 static void hns3_remove(struct pci_dev *pdev)
2950 {
2951 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2952 
2953 	if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
2954 		hns3_disable_sriov(pdev);
2955 
2956 	hnae3_unregister_ae_dev(ae_dev);
2957 	pci_set_drvdata(pdev, NULL);
2958 }
2959 
2960 /**
2961  * hns3_pci_sriov_configure
2962  * @pdev: pointer to a pci_dev structure
2963  * @num_vfs: number of VFs to allocate
2964  *
2965  * Enable or change the number of VFs. Called when the user updates the number
2966  * of VFs in sysfs.
2967  **/
2968 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
2969 {
2970 	int ret;
2971 
2972 	if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
2973 		dev_warn(&pdev->dev, "Can not config SRIOV\n");
2974 		return -EINVAL;
2975 	}
2976 
2977 	if (num_vfs) {
2978 		ret = pci_enable_sriov(pdev, num_vfs);
2979 		if (ret)
2980 			dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
2981 		else
2982 			return num_vfs;
2983 	} else if (!pci_vfs_assigned(pdev)) {
2984 		pci_disable_sriov(pdev);
2985 	} else {
2986 		dev_warn(&pdev->dev,
2987 			 "Unable to free VFs because some are assigned to VMs.\n");
2988 	}
2989 
2990 	return 0;
2991 }
2992 
2993 static void hns3_shutdown(struct pci_dev *pdev)
2994 {
2995 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2996 
2997 	hnae3_unregister_ae_dev(ae_dev);
2998 	pci_set_drvdata(pdev, NULL);
2999 
3000 	if (system_state == SYSTEM_POWER_OFF)
3001 		pci_set_power_state(pdev, PCI_D3hot);
3002 }
3003 
3004 static int __maybe_unused hns3_suspend(struct device *dev)
3005 {
3006 	struct hnae3_ae_dev *ae_dev = dev_get_drvdata(dev);
3007 
3008 	if (ae_dev && hns3_is_phys_func(ae_dev->pdev)) {
3009 		dev_info(dev, "Begin to suspend.\n");
3010 		if (ae_dev->ops && ae_dev->ops->reset_prepare)
3011 			ae_dev->ops->reset_prepare(ae_dev, HNAE3_FUNC_RESET);
3012 	}
3013 
3014 	return 0;
3015 }
3016 
3017 static int __maybe_unused hns3_resume(struct device *dev)
3018 {
3019 	struct hnae3_ae_dev *ae_dev = dev_get_drvdata(dev);
3020 
3021 	if (ae_dev && hns3_is_phys_func(ae_dev->pdev)) {
3022 		dev_info(dev, "Begin to resume.\n");
3023 		if (ae_dev->ops && ae_dev->ops->reset_done)
3024 			ae_dev->ops->reset_done(ae_dev);
3025 	}
3026 
3027 	return 0;
3028 }
3029 
3030 static pci_ers_result_t hns3_error_detected(struct pci_dev *pdev,
3031 					    pci_channel_state_t state)
3032 {
3033 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3034 	pci_ers_result_t ret;
3035 
3036 	dev_info(&pdev->dev, "PCI error detected, state(=%u)!!\n", state);
3037 
3038 	if (state == pci_channel_io_perm_failure)
3039 		return PCI_ERS_RESULT_DISCONNECT;
3040 
3041 	if (!ae_dev || !ae_dev->ops) {
3042 		dev_err(&pdev->dev,
3043 			"Can't recover - error happened before device initialized\n");
3044 		return PCI_ERS_RESULT_NONE;
3045 	}
3046 
3047 	if (ae_dev->ops->handle_hw_ras_error)
3048 		ret = ae_dev->ops->handle_hw_ras_error(ae_dev);
3049 	else
3050 		return PCI_ERS_RESULT_NONE;
3051 
3052 	return ret;
3053 }
3054 
3055 static pci_ers_result_t hns3_slot_reset(struct pci_dev *pdev)
3056 {
3057 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3058 	const struct hnae3_ae_ops *ops;
3059 	enum hnae3_reset_type reset_type;
3060 	struct device *dev = &pdev->dev;
3061 
3062 	if (!ae_dev || !ae_dev->ops)
3063 		return PCI_ERS_RESULT_NONE;
3064 
3065 	ops = ae_dev->ops;
3066 	/* request the reset */
3067 	if (ops->reset_event && ops->get_reset_level &&
3068 	    ops->set_default_reset_request) {
3069 		if (ae_dev->hw_err_reset_req) {
3070 			reset_type = ops->get_reset_level(ae_dev,
3071 						&ae_dev->hw_err_reset_req);
3072 			ops->set_default_reset_request(ae_dev, reset_type);
3073 			dev_info(dev, "requesting reset due to PCI error\n");
3074 			ops->reset_event(pdev, NULL);
3075 		}
3076 
3077 		return PCI_ERS_RESULT_RECOVERED;
3078 	}
3079 
3080 	return PCI_ERS_RESULT_DISCONNECT;
3081 }
3082 
3083 static void hns3_reset_prepare(struct pci_dev *pdev)
3084 {
3085 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3086 
3087 	dev_info(&pdev->dev, "FLR prepare\n");
3088 	if (ae_dev && ae_dev->ops && ae_dev->ops->reset_prepare)
3089 		ae_dev->ops->reset_prepare(ae_dev, HNAE3_FLR_RESET);
3090 }
3091 
3092 static void hns3_reset_done(struct pci_dev *pdev)
3093 {
3094 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3095 
3096 	dev_info(&pdev->dev, "FLR done\n");
3097 	if (ae_dev && ae_dev->ops && ae_dev->ops->reset_done)
3098 		ae_dev->ops->reset_done(ae_dev);
3099 }
3100 
3101 static const struct pci_error_handlers hns3_err_handler = {
3102 	.error_detected = hns3_error_detected,
3103 	.slot_reset     = hns3_slot_reset,
3104 	.reset_prepare	= hns3_reset_prepare,
3105 	.reset_done	= hns3_reset_done,
3106 };
3107 
3108 static SIMPLE_DEV_PM_OPS(hns3_pm_ops, hns3_suspend, hns3_resume);
3109 
3110 static struct pci_driver hns3_driver = {
3111 	.name     = hns3_driver_name,
3112 	.id_table = hns3_pci_tbl,
3113 	.probe    = hns3_probe,
3114 	.remove   = hns3_remove,
3115 	.shutdown = hns3_shutdown,
3116 	.driver.pm  = &hns3_pm_ops,
3117 	.sriov_configure = hns3_pci_sriov_configure,
3118 	.err_handler    = &hns3_err_handler,
3119 };
3120 
3121 /* set default feature to hns3 */
3122 static void hns3_set_default_feature(struct net_device *netdev)
3123 {
3124 	struct hnae3_handle *h = hns3_get_handle(netdev);
3125 	struct pci_dev *pdev = h->pdev;
3126 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3127 
3128 	netdev->priv_flags |= IFF_UNICAST_FLT;
3129 
3130 	netdev->hw_enc_features |= NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
3131 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
3132 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
3133 		NETIF_F_SCTP_CRC | NETIF_F_TSO_MANGLEID | NETIF_F_FRAGLIST;
3134 
3135 	netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
3136 
3137 	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
3138 		NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3139 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
3140 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
3141 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
3142 		NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
3143 
3144 	netdev->vlan_features |= NETIF_F_RXCSUM |
3145 		NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO |
3146 		NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
3147 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
3148 		NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
3149 
3150 	netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX |
3151 		NETIF_F_HW_VLAN_CTAG_RX |
3152 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
3153 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
3154 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
3155 		NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
3156 
3157 	if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V2) {
3158 		netdev->hw_features |= NETIF_F_GRO_HW;
3159 		netdev->features |= NETIF_F_GRO_HW;
3160 
3161 		if (!(h->flags & HNAE3_SUPPORT_VF)) {
3162 			netdev->hw_features |= NETIF_F_NTUPLE;
3163 			netdev->features |= NETIF_F_NTUPLE;
3164 		}
3165 	}
3166 
3167 	if (test_bit(HNAE3_DEV_SUPPORT_UDP_GSO_B, ae_dev->caps)) {
3168 		netdev->hw_features |= NETIF_F_GSO_UDP_L4;
3169 		netdev->features |= NETIF_F_GSO_UDP_L4;
3170 		netdev->vlan_features |= NETIF_F_GSO_UDP_L4;
3171 		netdev->hw_enc_features |= NETIF_F_GSO_UDP_L4;
3172 	}
3173 
3174 	if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps)) {
3175 		netdev->hw_features |= NETIF_F_HW_CSUM;
3176 		netdev->features |= NETIF_F_HW_CSUM;
3177 		netdev->vlan_features |= NETIF_F_HW_CSUM;
3178 		netdev->hw_enc_features |= NETIF_F_HW_CSUM;
3179 	} else {
3180 		netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3181 		netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3182 		netdev->vlan_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3183 		netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3184 	}
3185 
3186 	if (test_bit(HNAE3_DEV_SUPPORT_UDP_TUNNEL_CSUM_B, ae_dev->caps)) {
3187 		netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
3188 		netdev->features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
3189 		netdev->vlan_features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
3190 		netdev->hw_enc_features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
3191 	}
3192 
3193 	if (test_bit(HNAE3_DEV_SUPPORT_FD_FORWARD_TC_B, ae_dev->caps)) {
3194 		netdev->hw_features |= NETIF_F_HW_TC;
3195 		netdev->features |= NETIF_F_HW_TC;
3196 	}
3197 
3198 	if (test_bit(HNAE3_DEV_SUPPORT_VLAN_FLTR_MDF_B, ae_dev->caps))
3199 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
3200 }
3201 
3202 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
3203 			     struct hns3_desc_cb *cb)
3204 {
3205 	unsigned int order = hns3_page_order(ring);
3206 	struct page *p;
3207 
3208 	if (ring->page_pool) {
3209 		p = page_pool_dev_alloc_frag(ring->page_pool,
3210 					     &cb->page_offset,
3211 					     hns3_buf_size(ring));
3212 		if (unlikely(!p))
3213 			return -ENOMEM;
3214 
3215 		cb->priv = p;
3216 		cb->buf = page_address(p);
3217 		cb->dma = page_pool_get_dma_addr(p);
3218 		cb->type = DESC_TYPE_PP_FRAG;
3219 		cb->reuse_flag = 0;
3220 		return 0;
3221 	}
3222 
3223 	p = dev_alloc_pages(order);
3224 	if (!p)
3225 		return -ENOMEM;
3226 
3227 	cb->priv = p;
3228 	cb->page_offset = 0;
3229 	cb->reuse_flag = 0;
3230 	cb->buf  = page_address(p);
3231 	cb->length = hns3_page_size(ring);
3232 	cb->type = DESC_TYPE_PAGE;
3233 	page_ref_add(p, USHRT_MAX - 1);
3234 	cb->pagecnt_bias = USHRT_MAX;
3235 
3236 	return 0;
3237 }
3238 
3239 static void hns3_free_buffer(struct hns3_enet_ring *ring,
3240 			     struct hns3_desc_cb *cb, int budget)
3241 {
3242 	if (cb->type & (DESC_TYPE_SKB | DESC_TYPE_BOUNCE_HEAD |
3243 			DESC_TYPE_BOUNCE_ALL | DESC_TYPE_SGL_SKB))
3244 		napi_consume_skb(cb->priv, budget);
3245 	else if (!HNAE3_IS_TX_RING(ring)) {
3246 		if (cb->type & DESC_TYPE_PAGE && cb->pagecnt_bias)
3247 			__page_frag_cache_drain(cb->priv, cb->pagecnt_bias);
3248 		else if (cb->type & DESC_TYPE_PP_FRAG)
3249 			page_pool_put_full_page(ring->page_pool, cb->priv,
3250 						false);
3251 	}
3252 	memset(cb, 0, sizeof(*cb));
3253 }
3254 
3255 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
3256 {
3257 	cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
3258 			       cb->length, ring_to_dma_dir(ring));
3259 
3260 	if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
3261 		return -EIO;
3262 
3263 	return 0;
3264 }
3265 
3266 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
3267 			      struct hns3_desc_cb *cb)
3268 {
3269 	if (cb->type & (DESC_TYPE_SKB | DESC_TYPE_FRAGLIST_SKB))
3270 		dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
3271 				 ring_to_dma_dir(ring));
3272 	else if ((cb->type & DESC_TYPE_PAGE) && cb->length)
3273 		dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
3274 			       ring_to_dma_dir(ring));
3275 	else if (cb->type & (DESC_TYPE_BOUNCE_ALL | DESC_TYPE_BOUNCE_HEAD |
3276 			     DESC_TYPE_SGL_SKB))
3277 		hns3_tx_spare_reclaim_cb(ring, cb);
3278 }
3279 
3280 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
3281 {
3282 	hns3_unmap_buffer(ring, &ring->desc_cb[i]);
3283 	ring->desc[i].addr = 0;
3284 }
3285 
3286 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i,
3287 				    int budget)
3288 {
3289 	struct hns3_desc_cb *cb = &ring->desc_cb[i];
3290 
3291 	if (!ring->desc_cb[i].dma)
3292 		return;
3293 
3294 	hns3_buffer_detach(ring, i);
3295 	hns3_free_buffer(ring, cb, budget);
3296 }
3297 
3298 static void hns3_free_buffers(struct hns3_enet_ring *ring)
3299 {
3300 	int i;
3301 
3302 	for (i = 0; i < ring->desc_num; i++)
3303 		hns3_free_buffer_detach(ring, i, 0);
3304 }
3305 
3306 /* free desc along with its attached buffer */
3307 static void hns3_free_desc(struct hns3_enet_ring *ring)
3308 {
3309 	int size = ring->desc_num * sizeof(ring->desc[0]);
3310 
3311 	hns3_free_buffers(ring);
3312 
3313 	if (ring->desc) {
3314 		dma_free_coherent(ring_to_dev(ring), size,
3315 				  ring->desc, ring->desc_dma_addr);
3316 		ring->desc = NULL;
3317 	}
3318 }
3319 
3320 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
3321 {
3322 	int size = ring->desc_num * sizeof(ring->desc[0]);
3323 
3324 	ring->desc = dma_alloc_coherent(ring_to_dev(ring), size,
3325 					&ring->desc_dma_addr, GFP_KERNEL);
3326 	if (!ring->desc)
3327 		return -ENOMEM;
3328 
3329 	return 0;
3330 }
3331 
3332 static int hns3_alloc_and_map_buffer(struct hns3_enet_ring *ring,
3333 				   struct hns3_desc_cb *cb)
3334 {
3335 	int ret;
3336 
3337 	ret = hns3_alloc_buffer(ring, cb);
3338 	if (ret || ring->page_pool)
3339 		goto out;
3340 
3341 	ret = hns3_map_buffer(ring, cb);
3342 	if (ret)
3343 		goto out_with_buf;
3344 
3345 	return 0;
3346 
3347 out_with_buf:
3348 	hns3_free_buffer(ring, cb, 0);
3349 out:
3350 	return ret;
3351 }
3352 
3353 static int hns3_alloc_and_attach_buffer(struct hns3_enet_ring *ring, int i)
3354 {
3355 	int ret = hns3_alloc_and_map_buffer(ring, &ring->desc_cb[i]);
3356 
3357 	if (ret)
3358 		return ret;
3359 
3360 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3361 					 ring->desc_cb[i].page_offset);
3362 
3363 	return 0;
3364 }
3365 
3366 /* Allocate memory for raw pkg, and map with dma */
3367 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
3368 {
3369 	int i, j, ret;
3370 
3371 	for (i = 0; i < ring->desc_num; i++) {
3372 		ret = hns3_alloc_and_attach_buffer(ring, i);
3373 		if (ret)
3374 			goto out_buffer_fail;
3375 	}
3376 
3377 	return 0;
3378 
3379 out_buffer_fail:
3380 	for (j = i - 1; j >= 0; j--)
3381 		hns3_free_buffer_detach(ring, j, 0);
3382 	return ret;
3383 }
3384 
3385 /* detach a in-used buffer and replace with a reserved one */
3386 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
3387 				struct hns3_desc_cb *res_cb)
3388 {
3389 	hns3_unmap_buffer(ring, &ring->desc_cb[i]);
3390 	ring->desc_cb[i] = *res_cb;
3391 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3392 					 ring->desc_cb[i].page_offset);
3393 	ring->desc[i].rx.bd_base_info = 0;
3394 }
3395 
3396 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
3397 {
3398 	ring->desc_cb[i].reuse_flag = 0;
3399 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3400 					 ring->desc_cb[i].page_offset);
3401 	ring->desc[i].rx.bd_base_info = 0;
3402 
3403 	dma_sync_single_for_device(ring_to_dev(ring),
3404 			ring->desc_cb[i].dma + ring->desc_cb[i].page_offset,
3405 			hns3_buf_size(ring),
3406 			DMA_FROM_DEVICE);
3407 }
3408 
3409 static bool hns3_nic_reclaim_desc(struct hns3_enet_ring *ring,
3410 				  int *bytes, int *pkts, int budget)
3411 {
3412 	/* pair with ring->last_to_use update in hns3_tx_doorbell(),
3413 	 * smp_store_release() is not used in hns3_tx_doorbell() because
3414 	 * the doorbell operation already have the needed barrier operation.
3415 	 */
3416 	int ltu = smp_load_acquire(&ring->last_to_use);
3417 	int ntc = ring->next_to_clean;
3418 	struct hns3_desc_cb *desc_cb;
3419 	bool reclaimed = false;
3420 	struct hns3_desc *desc;
3421 
3422 	while (ltu != ntc) {
3423 		desc = &ring->desc[ntc];
3424 
3425 		if (le16_to_cpu(desc->tx.bdtp_fe_sc_vld_ra_ri) &
3426 				BIT(HNS3_TXD_VLD_B))
3427 			break;
3428 
3429 		desc_cb = &ring->desc_cb[ntc];
3430 
3431 		if (desc_cb->type & (DESC_TYPE_SKB | DESC_TYPE_BOUNCE_ALL |
3432 				     DESC_TYPE_BOUNCE_HEAD |
3433 				     DESC_TYPE_SGL_SKB)) {
3434 			(*pkts)++;
3435 			(*bytes) += desc_cb->send_bytes;
3436 		}
3437 
3438 		/* desc_cb will be cleaned, after hnae3_free_buffer_detach */
3439 		hns3_free_buffer_detach(ring, ntc, budget);
3440 
3441 		if (++ntc == ring->desc_num)
3442 			ntc = 0;
3443 
3444 		/* Issue prefetch for next Tx descriptor */
3445 		prefetch(&ring->desc_cb[ntc]);
3446 		reclaimed = true;
3447 	}
3448 
3449 	if (unlikely(!reclaimed))
3450 		return false;
3451 
3452 	/* This smp_store_release() pairs with smp_load_acquire() in
3453 	 * ring_space called by hns3_nic_net_xmit.
3454 	 */
3455 	smp_store_release(&ring->next_to_clean, ntc);
3456 
3457 	hns3_tx_spare_update(ring);
3458 
3459 	return true;
3460 }
3461 
3462 void hns3_clean_tx_ring(struct hns3_enet_ring *ring, int budget)
3463 {
3464 	struct net_device *netdev = ring_to_netdev(ring);
3465 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3466 	struct netdev_queue *dev_queue;
3467 	int bytes, pkts;
3468 
3469 	bytes = 0;
3470 	pkts = 0;
3471 
3472 	if (unlikely(!hns3_nic_reclaim_desc(ring, &bytes, &pkts, budget)))
3473 		return;
3474 
3475 	ring->tqp_vector->tx_group.total_bytes += bytes;
3476 	ring->tqp_vector->tx_group.total_packets += pkts;
3477 
3478 	u64_stats_update_begin(&ring->syncp);
3479 	ring->stats.tx_bytes += bytes;
3480 	ring->stats.tx_pkts += pkts;
3481 	u64_stats_update_end(&ring->syncp);
3482 
3483 	dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
3484 	netdev_tx_completed_queue(dev_queue, pkts, bytes);
3485 
3486 	if (unlikely(netif_carrier_ok(netdev) &&
3487 		     ring_space(ring) > HNS3_MAX_TSO_BD_NUM)) {
3488 		/* Make sure that anybody stopping the queue after this
3489 		 * sees the new next_to_clean.
3490 		 */
3491 		smp_mb();
3492 		if (netif_tx_queue_stopped(dev_queue) &&
3493 		    !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
3494 			netif_tx_wake_queue(dev_queue);
3495 			ring->stats.restart_queue++;
3496 		}
3497 	}
3498 }
3499 
3500 static int hns3_desc_unused(struct hns3_enet_ring *ring)
3501 {
3502 	int ntc = ring->next_to_clean;
3503 	int ntu = ring->next_to_use;
3504 
3505 	return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
3506 }
3507 
3508 static void hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring,
3509 				      int cleand_count)
3510 {
3511 	struct hns3_desc_cb *desc_cb;
3512 	struct hns3_desc_cb res_cbs;
3513 	int i, ret;
3514 
3515 	for (i = 0; i < cleand_count; i++) {
3516 		desc_cb = &ring->desc_cb[ring->next_to_use];
3517 		if (desc_cb->reuse_flag) {
3518 			u64_stats_update_begin(&ring->syncp);
3519 			ring->stats.reuse_pg_cnt++;
3520 			u64_stats_update_end(&ring->syncp);
3521 
3522 			hns3_reuse_buffer(ring, ring->next_to_use);
3523 		} else {
3524 			ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
3525 			if (ret) {
3526 				u64_stats_update_begin(&ring->syncp);
3527 				ring->stats.sw_err_cnt++;
3528 				u64_stats_update_end(&ring->syncp);
3529 
3530 				hns3_rl_err(ring_to_netdev(ring),
3531 					    "alloc rx buffer failed: %d\n",
3532 					    ret);
3533 				break;
3534 			}
3535 			hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
3536 
3537 			u64_stats_update_begin(&ring->syncp);
3538 			ring->stats.non_reuse_pg++;
3539 			u64_stats_update_end(&ring->syncp);
3540 		}
3541 
3542 		ring_ptr_move_fw(ring, next_to_use);
3543 	}
3544 
3545 	writel(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
3546 }
3547 
3548 static bool hns3_can_reuse_page(struct hns3_desc_cb *cb)
3549 {
3550 	return page_count(cb->priv) == cb->pagecnt_bias;
3551 }
3552 
3553 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
3554 				struct hns3_enet_ring *ring, int pull_len,
3555 				struct hns3_desc_cb *desc_cb)
3556 {
3557 	struct hns3_desc *desc = &ring->desc[ring->next_to_clean];
3558 	u32 frag_offset = desc_cb->page_offset + pull_len;
3559 	int size = le16_to_cpu(desc->rx.size);
3560 	u32 truesize = hns3_buf_size(ring);
3561 	u32 frag_size = size - pull_len;
3562 	bool reused;
3563 
3564 	if (ring->page_pool) {
3565 		skb_add_rx_frag(skb, i, desc_cb->priv, frag_offset,
3566 				frag_size, truesize);
3567 		return;
3568 	}
3569 
3570 	/* Avoid re-using remote or pfmem page */
3571 	if (unlikely(!dev_page_is_reusable(desc_cb->priv)))
3572 		goto out;
3573 
3574 	reused = hns3_can_reuse_page(desc_cb);
3575 
3576 	/* Rx page can be reused when:
3577 	 * 1. Rx page is only owned by the driver when page_offset
3578 	 *    is zero, which means 0 @ truesize will be used by
3579 	 *    stack after skb_add_rx_frag() is called, and the rest
3580 	 *    of rx page can be reused by driver.
3581 	 * Or
3582 	 * 2. Rx page is only owned by the driver when page_offset
3583 	 *    is non-zero, which means page_offset @ truesize will
3584 	 *    be used by stack after skb_add_rx_frag() is called,
3585 	 *    and 0 @ truesize can be reused by driver.
3586 	 */
3587 	if ((!desc_cb->page_offset && reused) ||
3588 	    ((desc_cb->page_offset + truesize + truesize) <=
3589 	     hns3_page_size(ring) && desc_cb->page_offset)) {
3590 		desc_cb->page_offset += truesize;
3591 		desc_cb->reuse_flag = 1;
3592 	} else if (desc_cb->page_offset && reused) {
3593 		desc_cb->page_offset = 0;
3594 		desc_cb->reuse_flag = 1;
3595 	} else if (frag_size <= ring->rx_copybreak) {
3596 		void *frag = napi_alloc_frag(frag_size);
3597 
3598 		if (unlikely(!frag)) {
3599 			u64_stats_update_begin(&ring->syncp);
3600 			ring->stats.frag_alloc_err++;
3601 			u64_stats_update_end(&ring->syncp);
3602 
3603 			hns3_rl_err(ring_to_netdev(ring),
3604 				    "failed to allocate rx frag\n");
3605 			goto out;
3606 		}
3607 
3608 		desc_cb->reuse_flag = 1;
3609 		memcpy(frag, desc_cb->buf + frag_offset, frag_size);
3610 		skb_add_rx_frag(skb, i, virt_to_page(frag),
3611 				offset_in_page(frag), frag_size, frag_size);
3612 
3613 		u64_stats_update_begin(&ring->syncp);
3614 		ring->stats.frag_alloc++;
3615 		u64_stats_update_end(&ring->syncp);
3616 		return;
3617 	}
3618 
3619 out:
3620 	desc_cb->pagecnt_bias--;
3621 
3622 	if (unlikely(!desc_cb->pagecnt_bias)) {
3623 		page_ref_add(desc_cb->priv, USHRT_MAX);
3624 		desc_cb->pagecnt_bias = USHRT_MAX;
3625 	}
3626 
3627 	skb_add_rx_frag(skb, i, desc_cb->priv, frag_offset,
3628 			frag_size, truesize);
3629 
3630 	if (unlikely(!desc_cb->reuse_flag))
3631 		__page_frag_cache_drain(desc_cb->priv, desc_cb->pagecnt_bias);
3632 }
3633 
3634 static int hns3_gro_complete(struct sk_buff *skb, u32 l234info)
3635 {
3636 	__be16 type = skb->protocol;
3637 	struct tcphdr *th;
3638 	int depth = 0;
3639 
3640 	while (eth_type_vlan(type)) {
3641 		struct vlan_hdr *vh;
3642 
3643 		if ((depth + VLAN_HLEN) > skb_headlen(skb))
3644 			return -EFAULT;
3645 
3646 		vh = (struct vlan_hdr *)(skb->data + depth);
3647 		type = vh->h_vlan_encapsulated_proto;
3648 		depth += VLAN_HLEN;
3649 	}
3650 
3651 	skb_set_network_header(skb, depth);
3652 
3653 	if (type == htons(ETH_P_IP)) {
3654 		const struct iphdr *iph = ip_hdr(skb);
3655 
3656 		depth += sizeof(struct iphdr);
3657 		skb_set_transport_header(skb, depth);
3658 		th = tcp_hdr(skb);
3659 		th->check = ~tcp_v4_check(skb->len - depth, iph->saddr,
3660 					  iph->daddr, 0);
3661 	} else if (type == htons(ETH_P_IPV6)) {
3662 		const struct ipv6hdr *iph = ipv6_hdr(skb);
3663 
3664 		depth += sizeof(struct ipv6hdr);
3665 		skb_set_transport_header(skb, depth);
3666 		th = tcp_hdr(skb);
3667 		th->check = ~tcp_v6_check(skb->len - depth, &iph->saddr,
3668 					  &iph->daddr, 0);
3669 	} else {
3670 		hns3_rl_err(skb->dev,
3671 			    "Error: FW GRO supports only IPv4/IPv6, not 0x%04x, depth: %d\n",
3672 			    be16_to_cpu(type), depth);
3673 		return -EFAULT;
3674 	}
3675 
3676 	skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
3677 	if (th->cwr)
3678 		skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
3679 
3680 	if (l234info & BIT(HNS3_RXD_GRO_FIXID_B))
3681 		skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_FIXEDID;
3682 
3683 	skb->csum_start = (unsigned char *)th - skb->head;
3684 	skb->csum_offset = offsetof(struct tcphdr, check);
3685 	skb->ip_summed = CHECKSUM_PARTIAL;
3686 
3687 	trace_hns3_gro(skb);
3688 
3689 	return 0;
3690 }
3691 
3692 static bool hns3_checksum_complete(struct hns3_enet_ring *ring,
3693 				   struct sk_buff *skb, u32 ptype, u16 csum)
3694 {
3695 	if (ptype == HNS3_INVALID_PTYPE ||
3696 	    hns3_rx_ptype_tbl[ptype].ip_summed != CHECKSUM_COMPLETE)
3697 		return false;
3698 
3699 	u64_stats_update_begin(&ring->syncp);
3700 	ring->stats.csum_complete++;
3701 	u64_stats_update_end(&ring->syncp);
3702 	skb->ip_summed = CHECKSUM_COMPLETE;
3703 	skb->csum = csum_unfold((__force __sum16)csum);
3704 
3705 	return true;
3706 }
3707 
3708 static void hns3_rx_handle_csum(struct sk_buff *skb, u32 l234info,
3709 				u32 ol_info, u32 ptype)
3710 {
3711 	int l3_type, l4_type;
3712 	int ol4_type;
3713 
3714 	if (ptype != HNS3_INVALID_PTYPE) {
3715 		skb->csum_level = hns3_rx_ptype_tbl[ptype].csum_level;
3716 		skb->ip_summed = hns3_rx_ptype_tbl[ptype].ip_summed;
3717 
3718 		return;
3719 	}
3720 
3721 	ol4_type = hnae3_get_field(ol_info, HNS3_RXD_OL4ID_M,
3722 				   HNS3_RXD_OL4ID_S);
3723 	switch (ol4_type) {
3724 	case HNS3_OL4_TYPE_MAC_IN_UDP:
3725 	case HNS3_OL4_TYPE_NVGRE:
3726 		skb->csum_level = 1;
3727 		fallthrough;
3728 	case HNS3_OL4_TYPE_NO_TUN:
3729 		l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
3730 					  HNS3_RXD_L3ID_S);
3731 		l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
3732 					  HNS3_RXD_L4ID_S);
3733 		/* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
3734 		if ((l3_type == HNS3_L3_TYPE_IPV4 ||
3735 		     l3_type == HNS3_L3_TYPE_IPV6) &&
3736 		    (l4_type == HNS3_L4_TYPE_UDP ||
3737 		     l4_type == HNS3_L4_TYPE_TCP ||
3738 		     l4_type == HNS3_L4_TYPE_SCTP))
3739 			skb->ip_summed = CHECKSUM_UNNECESSARY;
3740 		break;
3741 	default:
3742 		break;
3743 	}
3744 }
3745 
3746 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
3747 			     u32 l234info, u32 bd_base_info, u32 ol_info,
3748 			     u16 csum)
3749 {
3750 	struct net_device *netdev = ring_to_netdev(ring);
3751 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3752 	u32 ptype = HNS3_INVALID_PTYPE;
3753 
3754 	skb->ip_summed = CHECKSUM_NONE;
3755 
3756 	skb_checksum_none_assert(skb);
3757 
3758 	if (!(netdev->features & NETIF_F_RXCSUM))
3759 		return;
3760 
3761 	if (test_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state))
3762 		ptype = hnae3_get_field(ol_info, HNS3_RXD_PTYPE_M,
3763 					HNS3_RXD_PTYPE_S);
3764 
3765 	if (hns3_checksum_complete(ring, skb, ptype, csum))
3766 		return;
3767 
3768 	/* check if hardware has done checksum */
3769 	if (!(bd_base_info & BIT(HNS3_RXD_L3L4P_B)))
3770 		return;
3771 
3772 	if (unlikely(l234info & (BIT(HNS3_RXD_L3E_B) | BIT(HNS3_RXD_L4E_B) |
3773 				 BIT(HNS3_RXD_OL3E_B) |
3774 				 BIT(HNS3_RXD_OL4E_B)))) {
3775 		u64_stats_update_begin(&ring->syncp);
3776 		ring->stats.l3l4_csum_err++;
3777 		u64_stats_update_end(&ring->syncp);
3778 
3779 		return;
3780 	}
3781 
3782 	hns3_rx_handle_csum(skb, l234info, ol_info, ptype);
3783 }
3784 
3785 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
3786 {
3787 	if (skb_has_frag_list(skb))
3788 		napi_gro_flush(&ring->tqp_vector->napi, false);
3789 
3790 	napi_gro_receive(&ring->tqp_vector->napi, skb);
3791 }
3792 
3793 static bool hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
3794 				struct hns3_desc *desc, u32 l234info,
3795 				u16 *vlan_tag)
3796 {
3797 	struct hnae3_handle *handle = ring->tqp->handle;
3798 	struct pci_dev *pdev = ring->tqp->handle->pdev;
3799 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3800 
3801 	if (unlikely(ae_dev->dev_version < HNAE3_DEVICE_VERSION_V2)) {
3802 		*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3803 		if (!(*vlan_tag & VLAN_VID_MASK))
3804 			*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3805 
3806 		return (*vlan_tag != 0);
3807 	}
3808 
3809 #define HNS3_STRP_OUTER_VLAN	0x1
3810 #define HNS3_STRP_INNER_VLAN	0x2
3811 #define HNS3_STRP_BOTH		0x3
3812 
3813 	/* Hardware always insert VLAN tag into RX descriptor when
3814 	 * remove the tag from packet, driver needs to determine
3815 	 * reporting which tag to stack.
3816 	 */
3817 	switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
3818 				HNS3_RXD_STRP_TAGP_S)) {
3819 	case HNS3_STRP_OUTER_VLAN:
3820 		if (handle->port_base_vlan_state !=
3821 				HNAE3_PORT_BASE_VLAN_DISABLE)
3822 			return false;
3823 
3824 		*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3825 		return true;
3826 	case HNS3_STRP_INNER_VLAN:
3827 		if (handle->port_base_vlan_state !=
3828 				HNAE3_PORT_BASE_VLAN_DISABLE)
3829 			return false;
3830 
3831 		*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3832 		return true;
3833 	case HNS3_STRP_BOTH:
3834 		if (handle->port_base_vlan_state ==
3835 				HNAE3_PORT_BASE_VLAN_DISABLE)
3836 			*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3837 		else
3838 			*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3839 
3840 		return true;
3841 	default:
3842 		return false;
3843 	}
3844 }
3845 
3846 static void hns3_rx_ring_move_fw(struct hns3_enet_ring *ring)
3847 {
3848 	ring->desc[ring->next_to_clean].rx.bd_base_info &=
3849 		cpu_to_le32(~BIT(HNS3_RXD_VLD_B));
3850 	ring->next_to_clean += 1;
3851 
3852 	if (unlikely(ring->next_to_clean == ring->desc_num))
3853 		ring->next_to_clean = 0;
3854 }
3855 
3856 static int hns3_alloc_skb(struct hns3_enet_ring *ring, unsigned int length,
3857 			  unsigned char *va)
3858 {
3859 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
3860 	struct net_device *netdev = ring_to_netdev(ring);
3861 	struct sk_buff *skb;
3862 
3863 	ring->skb = napi_alloc_skb(&ring->tqp_vector->napi, HNS3_RX_HEAD_SIZE);
3864 	skb = ring->skb;
3865 	if (unlikely(!skb)) {
3866 		hns3_rl_err(netdev, "alloc rx skb fail\n");
3867 
3868 		u64_stats_update_begin(&ring->syncp);
3869 		ring->stats.sw_err_cnt++;
3870 		u64_stats_update_end(&ring->syncp);
3871 
3872 		return -ENOMEM;
3873 	}
3874 
3875 	trace_hns3_rx_desc(ring);
3876 	prefetchw(skb->data);
3877 
3878 	ring->pending_buf = 1;
3879 	ring->frag_num = 0;
3880 	ring->tail_skb = NULL;
3881 	if (length <= HNS3_RX_HEAD_SIZE) {
3882 		memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
3883 
3884 		/* We can reuse buffer as-is, just make sure it is reusable */
3885 		if (dev_page_is_reusable(desc_cb->priv))
3886 			desc_cb->reuse_flag = 1;
3887 		else if (desc_cb->type & DESC_TYPE_PP_FRAG)
3888 			page_pool_put_full_page(ring->page_pool, desc_cb->priv,
3889 						false);
3890 		else /* This page cannot be reused so discard it */
3891 			__page_frag_cache_drain(desc_cb->priv,
3892 						desc_cb->pagecnt_bias);
3893 
3894 		hns3_rx_ring_move_fw(ring);
3895 		return 0;
3896 	}
3897 
3898 	if (ring->page_pool)
3899 		skb_mark_for_recycle(skb);
3900 
3901 	u64_stats_update_begin(&ring->syncp);
3902 	ring->stats.seg_pkt_cnt++;
3903 	u64_stats_update_end(&ring->syncp);
3904 
3905 	ring->pull_len = eth_get_headlen(netdev, va, HNS3_RX_HEAD_SIZE);
3906 	__skb_put(skb, ring->pull_len);
3907 	hns3_nic_reuse_page(skb, ring->frag_num++, ring, ring->pull_len,
3908 			    desc_cb);
3909 	hns3_rx_ring_move_fw(ring);
3910 
3911 	return 0;
3912 }
3913 
3914 static int hns3_add_frag(struct hns3_enet_ring *ring)
3915 {
3916 	struct sk_buff *skb = ring->skb;
3917 	struct sk_buff *head_skb = skb;
3918 	struct sk_buff *new_skb;
3919 	struct hns3_desc_cb *desc_cb;
3920 	struct hns3_desc *desc;
3921 	u32 bd_base_info;
3922 
3923 	do {
3924 		desc = &ring->desc[ring->next_to_clean];
3925 		desc_cb = &ring->desc_cb[ring->next_to_clean];
3926 		bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
3927 		/* make sure HW write desc complete */
3928 		dma_rmb();
3929 		if (!(bd_base_info & BIT(HNS3_RXD_VLD_B)))
3930 			return -ENXIO;
3931 
3932 		if (unlikely(ring->frag_num >= MAX_SKB_FRAGS)) {
3933 			new_skb = napi_alloc_skb(&ring->tqp_vector->napi, 0);
3934 			if (unlikely(!new_skb)) {
3935 				hns3_rl_err(ring_to_netdev(ring),
3936 					    "alloc rx fraglist skb fail\n");
3937 				return -ENXIO;
3938 			}
3939 
3940 			if (ring->page_pool)
3941 				skb_mark_for_recycle(new_skb);
3942 
3943 			ring->frag_num = 0;
3944 
3945 			if (ring->tail_skb) {
3946 				ring->tail_skb->next = new_skb;
3947 				ring->tail_skb = new_skb;
3948 			} else {
3949 				skb_shinfo(skb)->frag_list = new_skb;
3950 				ring->tail_skb = new_skb;
3951 			}
3952 		}
3953 
3954 		if (ring->tail_skb) {
3955 			head_skb->truesize += hns3_buf_size(ring);
3956 			head_skb->data_len += le16_to_cpu(desc->rx.size);
3957 			head_skb->len += le16_to_cpu(desc->rx.size);
3958 			skb = ring->tail_skb;
3959 		}
3960 
3961 		dma_sync_single_for_cpu(ring_to_dev(ring),
3962 				desc_cb->dma + desc_cb->page_offset,
3963 				hns3_buf_size(ring),
3964 				DMA_FROM_DEVICE);
3965 
3966 		hns3_nic_reuse_page(skb, ring->frag_num++, ring, 0, desc_cb);
3967 		trace_hns3_rx_desc(ring);
3968 		hns3_rx_ring_move_fw(ring);
3969 		ring->pending_buf++;
3970 	} while (!(bd_base_info & BIT(HNS3_RXD_FE_B)));
3971 
3972 	return 0;
3973 }
3974 
3975 static int hns3_set_gro_and_checksum(struct hns3_enet_ring *ring,
3976 				     struct sk_buff *skb, u32 l234info,
3977 				     u32 bd_base_info, u32 ol_info, u16 csum)
3978 {
3979 	struct net_device *netdev = ring_to_netdev(ring);
3980 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3981 	u32 l3_type;
3982 
3983 	skb_shinfo(skb)->gso_size = hnae3_get_field(bd_base_info,
3984 						    HNS3_RXD_GRO_SIZE_M,
3985 						    HNS3_RXD_GRO_SIZE_S);
3986 	/* if there is no HW GRO, do not set gro params */
3987 	if (!skb_shinfo(skb)->gso_size) {
3988 		hns3_rx_checksum(ring, skb, l234info, bd_base_info, ol_info,
3989 				 csum);
3990 		return 0;
3991 	}
3992 
3993 	NAPI_GRO_CB(skb)->count = hnae3_get_field(l234info,
3994 						  HNS3_RXD_GRO_COUNT_M,
3995 						  HNS3_RXD_GRO_COUNT_S);
3996 
3997 	if (test_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state)) {
3998 		u32 ptype = hnae3_get_field(ol_info, HNS3_RXD_PTYPE_M,
3999 					    HNS3_RXD_PTYPE_S);
4000 
4001 		l3_type = hns3_rx_ptype_tbl[ptype].l3_type;
4002 	} else {
4003 		l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
4004 					  HNS3_RXD_L3ID_S);
4005 	}
4006 
4007 	if (l3_type == HNS3_L3_TYPE_IPV4)
4008 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
4009 	else if (l3_type == HNS3_L3_TYPE_IPV6)
4010 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
4011 	else
4012 		return -EFAULT;
4013 
4014 	return  hns3_gro_complete(skb, l234info);
4015 }
4016 
4017 static void hns3_set_rx_skb_rss_type(struct hns3_enet_ring *ring,
4018 				     struct sk_buff *skb, u32 rss_hash)
4019 {
4020 	struct hnae3_handle *handle = ring->tqp->handle;
4021 	enum pkt_hash_types rss_type;
4022 
4023 	if (rss_hash)
4024 		rss_type = handle->kinfo.rss_type;
4025 	else
4026 		rss_type = PKT_HASH_TYPE_NONE;
4027 
4028 	skb_set_hash(skb, rss_hash, rss_type);
4029 }
4030 
4031 static int hns3_handle_bdinfo(struct hns3_enet_ring *ring, struct sk_buff *skb)
4032 {
4033 	struct net_device *netdev = ring_to_netdev(ring);
4034 	enum hns3_pkt_l2t_type l2_frame_type;
4035 	u32 bd_base_info, l234info, ol_info;
4036 	struct hns3_desc *desc;
4037 	unsigned int len;
4038 	int pre_ntc, ret;
4039 	u16 csum;
4040 
4041 	/* bdinfo handled below is only valid on the last BD of the
4042 	 * current packet, and ring->next_to_clean indicates the first
4043 	 * descriptor of next packet, so need - 1 below.
4044 	 */
4045 	pre_ntc = ring->next_to_clean ? (ring->next_to_clean - 1) :
4046 					(ring->desc_num - 1);
4047 	desc = &ring->desc[pre_ntc];
4048 	bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
4049 	l234info = le32_to_cpu(desc->rx.l234_info);
4050 	ol_info = le32_to_cpu(desc->rx.ol_info);
4051 	csum = le16_to_cpu(desc->csum);
4052 
4053 	if (unlikely(bd_base_info & BIT(HNS3_RXD_TS_VLD_B))) {
4054 		struct hnae3_handle *h = hns3_get_handle(netdev);
4055 		u32 nsec = le32_to_cpu(desc->ts_nsec);
4056 		u32 sec = le32_to_cpu(desc->ts_sec);
4057 
4058 		if (h->ae_algo->ops->get_rx_hwts)
4059 			h->ae_algo->ops->get_rx_hwts(h, skb, nsec, sec);
4060 	}
4061 
4062 	/* Based on hw strategy, the tag offloaded will be stored at
4063 	 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
4064 	 * in one layer tag case.
4065 	 */
4066 	if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
4067 		u16 vlan_tag;
4068 
4069 		if (hns3_parse_vlan_tag(ring, desc, l234info, &vlan_tag))
4070 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
4071 					       vlan_tag);
4072 	}
4073 
4074 	if (unlikely(!desc->rx.pkt_len || (l234info & (BIT(HNS3_RXD_TRUNCAT_B) |
4075 				  BIT(HNS3_RXD_L2E_B))))) {
4076 		u64_stats_update_begin(&ring->syncp);
4077 		if (l234info & BIT(HNS3_RXD_L2E_B))
4078 			ring->stats.l2_err++;
4079 		else
4080 			ring->stats.err_pkt_len++;
4081 		u64_stats_update_end(&ring->syncp);
4082 
4083 		return -EFAULT;
4084 	}
4085 
4086 	len = skb->len;
4087 
4088 	/* Do update ip stack process */
4089 	skb->protocol = eth_type_trans(skb, netdev);
4090 
4091 	/* This is needed in order to enable forwarding support */
4092 	ret = hns3_set_gro_and_checksum(ring, skb, l234info,
4093 					bd_base_info, ol_info, csum);
4094 	if (unlikely(ret)) {
4095 		u64_stats_update_begin(&ring->syncp);
4096 		ring->stats.rx_err_cnt++;
4097 		u64_stats_update_end(&ring->syncp);
4098 		return ret;
4099 	}
4100 
4101 	l2_frame_type = hnae3_get_field(l234info, HNS3_RXD_DMAC_M,
4102 					HNS3_RXD_DMAC_S);
4103 
4104 	u64_stats_update_begin(&ring->syncp);
4105 	ring->stats.rx_pkts++;
4106 	ring->stats.rx_bytes += len;
4107 
4108 	if (l2_frame_type == HNS3_L2_TYPE_MULTICAST)
4109 		ring->stats.rx_multicast++;
4110 
4111 	u64_stats_update_end(&ring->syncp);
4112 
4113 	ring->tqp_vector->rx_group.total_bytes += len;
4114 
4115 	hns3_set_rx_skb_rss_type(ring, skb, le32_to_cpu(desc->rx.rss_hash));
4116 	return 0;
4117 }
4118 
4119 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring)
4120 {
4121 	struct sk_buff *skb = ring->skb;
4122 	struct hns3_desc_cb *desc_cb;
4123 	struct hns3_desc *desc;
4124 	unsigned int length;
4125 	u32 bd_base_info;
4126 	int ret;
4127 
4128 	desc = &ring->desc[ring->next_to_clean];
4129 	desc_cb = &ring->desc_cb[ring->next_to_clean];
4130 
4131 	prefetch(desc);
4132 
4133 	if (!skb) {
4134 		bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
4135 		/* Check valid BD */
4136 		if (unlikely(!(bd_base_info & BIT(HNS3_RXD_VLD_B))))
4137 			return -ENXIO;
4138 
4139 		dma_rmb();
4140 		length = le16_to_cpu(desc->rx.size);
4141 
4142 		ring->va = desc_cb->buf + desc_cb->page_offset;
4143 
4144 		dma_sync_single_for_cpu(ring_to_dev(ring),
4145 				desc_cb->dma + desc_cb->page_offset,
4146 				hns3_buf_size(ring),
4147 				DMA_FROM_DEVICE);
4148 
4149 		/* Prefetch first cache line of first page.
4150 		 * Idea is to cache few bytes of the header of the packet.
4151 		 * Our L1 Cache line size is 64B so need to prefetch twice to make
4152 		 * it 128B. But in actual we can have greater size of caches with
4153 		 * 128B Level 1 cache lines. In such a case, single fetch would
4154 		 * suffice to cache in the relevant part of the header.
4155 		 */
4156 		net_prefetch(ring->va);
4157 
4158 		ret = hns3_alloc_skb(ring, length, ring->va);
4159 		skb = ring->skb;
4160 
4161 		if (ret < 0) /* alloc buffer fail */
4162 			return ret;
4163 		if (!(bd_base_info & BIT(HNS3_RXD_FE_B))) { /* need add frag */
4164 			ret = hns3_add_frag(ring);
4165 			if (ret)
4166 				return ret;
4167 		}
4168 	} else {
4169 		ret = hns3_add_frag(ring);
4170 		if (ret)
4171 			return ret;
4172 	}
4173 
4174 	/* As the head data may be changed when GRO enable, copy
4175 	 * the head data in after other data rx completed
4176 	 */
4177 	if (skb->len > HNS3_RX_HEAD_SIZE)
4178 		memcpy(skb->data, ring->va,
4179 		       ALIGN(ring->pull_len, sizeof(long)));
4180 
4181 	ret = hns3_handle_bdinfo(ring, skb);
4182 	if (unlikely(ret)) {
4183 		dev_kfree_skb_any(skb);
4184 		return ret;
4185 	}
4186 
4187 	skb_record_rx_queue(skb, ring->tqp->tqp_index);
4188 	return 0;
4189 }
4190 
4191 int hns3_clean_rx_ring(struct hns3_enet_ring *ring, int budget,
4192 		       void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
4193 {
4194 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
4195 	int unused_count = hns3_desc_unused(ring);
4196 	int recv_pkts = 0;
4197 	int err;
4198 
4199 	unused_count -= ring->pending_buf;
4200 
4201 	while (recv_pkts < budget) {
4202 		/* Reuse or realloc buffers */
4203 		if (unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
4204 			hns3_nic_alloc_rx_buffers(ring, unused_count);
4205 			unused_count = hns3_desc_unused(ring) -
4206 					ring->pending_buf;
4207 		}
4208 
4209 		/* Poll one pkt */
4210 		err = hns3_handle_rx_bd(ring);
4211 		/* Do not get FE for the packet or failed to alloc skb */
4212 		if (unlikely(!ring->skb || err == -ENXIO)) {
4213 			goto out;
4214 		} else if (likely(!err)) {
4215 			rx_fn(ring, ring->skb);
4216 			recv_pkts++;
4217 		}
4218 
4219 		unused_count += ring->pending_buf;
4220 		ring->skb = NULL;
4221 		ring->pending_buf = 0;
4222 	}
4223 
4224 out:
4225 	/* Make all data has been write before submit */
4226 	if (unused_count > 0)
4227 		hns3_nic_alloc_rx_buffers(ring, unused_count);
4228 
4229 	return recv_pkts;
4230 }
4231 
4232 static void hns3_update_rx_int_coalesce(struct hns3_enet_tqp_vector *tqp_vector)
4233 {
4234 	struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
4235 	struct dim_sample sample = {};
4236 
4237 	if (!rx_group->coal.adapt_enable)
4238 		return;
4239 
4240 	dim_update_sample(tqp_vector->event_cnt, rx_group->total_packets,
4241 			  rx_group->total_bytes, &sample);
4242 	net_dim(&rx_group->dim, sample);
4243 }
4244 
4245 static void hns3_update_tx_int_coalesce(struct hns3_enet_tqp_vector *tqp_vector)
4246 {
4247 	struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
4248 	struct dim_sample sample = {};
4249 
4250 	if (!tx_group->coal.adapt_enable)
4251 		return;
4252 
4253 	dim_update_sample(tqp_vector->event_cnt, tx_group->total_packets,
4254 			  tx_group->total_bytes, &sample);
4255 	net_dim(&tx_group->dim, sample);
4256 }
4257 
4258 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
4259 {
4260 	struct hns3_nic_priv *priv = netdev_priv(napi->dev);
4261 	struct hns3_enet_ring *ring;
4262 	int rx_pkt_total = 0;
4263 
4264 	struct hns3_enet_tqp_vector *tqp_vector =
4265 		container_of(napi, struct hns3_enet_tqp_vector, napi);
4266 	bool clean_complete = true;
4267 	int rx_budget = budget;
4268 
4269 	if (unlikely(test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
4270 		napi_complete(napi);
4271 		return 0;
4272 	}
4273 
4274 	/* Since the actual Tx work is minimal, we can give the Tx a larger
4275 	 * budget and be more aggressive about cleaning up the Tx descriptors.
4276 	 */
4277 	hns3_for_each_ring(ring, tqp_vector->tx_group)
4278 		hns3_clean_tx_ring(ring, budget);
4279 
4280 	/* make sure rx ring budget not smaller than 1 */
4281 	if (tqp_vector->num_tqps > 1)
4282 		rx_budget = max(budget / tqp_vector->num_tqps, 1);
4283 
4284 	hns3_for_each_ring(ring, tqp_vector->rx_group) {
4285 		int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
4286 						    hns3_rx_skb);
4287 		if (rx_cleaned >= rx_budget)
4288 			clean_complete = false;
4289 
4290 		rx_pkt_total += rx_cleaned;
4291 	}
4292 
4293 	tqp_vector->rx_group.total_packets += rx_pkt_total;
4294 
4295 	if (!clean_complete)
4296 		return budget;
4297 
4298 	if (napi_complete(napi) &&
4299 	    likely(!test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
4300 		hns3_update_rx_int_coalesce(tqp_vector);
4301 		hns3_update_tx_int_coalesce(tqp_vector);
4302 
4303 		hns3_mask_vector_irq(tqp_vector, 1);
4304 	}
4305 
4306 	return rx_pkt_total;
4307 }
4308 
4309 static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
4310 				      struct hnae3_ring_chain_node *head)
4311 {
4312 	struct pci_dev *pdev = tqp_vector->handle->pdev;
4313 	struct hnae3_ring_chain_node *cur_chain = head;
4314 	struct hnae3_ring_chain_node *chain;
4315 	struct hns3_enet_ring *tx_ring;
4316 	struct hns3_enet_ring *rx_ring;
4317 
4318 	tx_ring = tqp_vector->tx_group.ring;
4319 	if (tx_ring) {
4320 		cur_chain->tqp_index = tx_ring->tqp->tqp_index;
4321 		hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
4322 			      HNAE3_RING_TYPE_TX);
4323 		hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
4324 				HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_TX);
4325 
4326 		cur_chain->next = NULL;
4327 
4328 		while (tx_ring->next) {
4329 			tx_ring = tx_ring->next;
4330 
4331 			chain = devm_kzalloc(&pdev->dev, sizeof(*chain),
4332 					     GFP_KERNEL);
4333 			if (!chain)
4334 				goto err_free_chain;
4335 
4336 			cur_chain->next = chain;
4337 			chain->tqp_index = tx_ring->tqp->tqp_index;
4338 			hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
4339 				      HNAE3_RING_TYPE_TX);
4340 			hnae3_set_field(chain->int_gl_idx,
4341 					HNAE3_RING_GL_IDX_M,
4342 					HNAE3_RING_GL_IDX_S,
4343 					HNAE3_RING_GL_TX);
4344 
4345 			cur_chain = chain;
4346 		}
4347 	}
4348 
4349 	rx_ring = tqp_vector->rx_group.ring;
4350 	if (!tx_ring && rx_ring) {
4351 		cur_chain->next = NULL;
4352 		cur_chain->tqp_index = rx_ring->tqp->tqp_index;
4353 		hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
4354 			      HNAE3_RING_TYPE_RX);
4355 		hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
4356 				HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
4357 
4358 		rx_ring = rx_ring->next;
4359 	}
4360 
4361 	while (rx_ring) {
4362 		chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
4363 		if (!chain)
4364 			goto err_free_chain;
4365 
4366 		cur_chain->next = chain;
4367 		chain->tqp_index = rx_ring->tqp->tqp_index;
4368 		hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
4369 			      HNAE3_RING_TYPE_RX);
4370 		hnae3_set_field(chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
4371 				HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
4372 
4373 		cur_chain = chain;
4374 
4375 		rx_ring = rx_ring->next;
4376 	}
4377 
4378 	return 0;
4379 
4380 err_free_chain:
4381 	cur_chain = head->next;
4382 	while (cur_chain) {
4383 		chain = cur_chain->next;
4384 		devm_kfree(&pdev->dev, cur_chain);
4385 		cur_chain = chain;
4386 	}
4387 	head->next = NULL;
4388 
4389 	return -ENOMEM;
4390 }
4391 
4392 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
4393 					struct hnae3_ring_chain_node *head)
4394 {
4395 	struct pci_dev *pdev = tqp_vector->handle->pdev;
4396 	struct hnae3_ring_chain_node *chain_tmp, *chain;
4397 
4398 	chain = head->next;
4399 
4400 	while (chain) {
4401 		chain_tmp = chain->next;
4402 		devm_kfree(&pdev->dev, chain);
4403 		chain = chain_tmp;
4404 	}
4405 }
4406 
4407 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
4408 				   struct hns3_enet_ring *ring)
4409 {
4410 	ring->next = group->ring;
4411 	group->ring = ring;
4412 
4413 	group->count++;
4414 }
4415 
4416 static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
4417 {
4418 	struct pci_dev *pdev = priv->ae_handle->pdev;
4419 	struct hns3_enet_tqp_vector *tqp_vector;
4420 	int num_vectors = priv->vector_num;
4421 	int numa_node;
4422 	int vector_i;
4423 
4424 	numa_node = dev_to_node(&pdev->dev);
4425 
4426 	for (vector_i = 0; vector_i < num_vectors; vector_i++) {
4427 		tqp_vector = &priv->tqp_vector[vector_i];
4428 		cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
4429 				&tqp_vector->affinity_mask);
4430 	}
4431 }
4432 
4433 static void hns3_rx_dim_work(struct work_struct *work)
4434 {
4435 	struct dim *dim = container_of(work, struct dim, work);
4436 	struct hns3_enet_ring_group *group = container_of(dim,
4437 		struct hns3_enet_ring_group, dim);
4438 	struct hns3_enet_tqp_vector *tqp_vector = group->ring->tqp_vector;
4439 	struct dim_cq_moder cur_moder =
4440 		net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
4441 
4442 	hns3_set_vector_coalesce_rx_gl(group->ring->tqp_vector, cur_moder.usec);
4443 	tqp_vector->rx_group.coal.int_gl = cur_moder.usec;
4444 
4445 	if (cur_moder.pkts < tqp_vector->rx_group.coal.int_ql_max) {
4446 		hns3_set_vector_coalesce_rx_ql(tqp_vector, cur_moder.pkts);
4447 		tqp_vector->rx_group.coal.int_ql = cur_moder.pkts;
4448 	}
4449 
4450 	dim->state = DIM_START_MEASURE;
4451 }
4452 
4453 static void hns3_tx_dim_work(struct work_struct *work)
4454 {
4455 	struct dim *dim = container_of(work, struct dim, work);
4456 	struct hns3_enet_ring_group *group = container_of(dim,
4457 		struct hns3_enet_ring_group, dim);
4458 	struct hns3_enet_tqp_vector *tqp_vector = group->ring->tqp_vector;
4459 	struct dim_cq_moder cur_moder =
4460 		net_dim_get_tx_moderation(dim->mode, dim->profile_ix);
4461 
4462 	hns3_set_vector_coalesce_tx_gl(tqp_vector, cur_moder.usec);
4463 	tqp_vector->tx_group.coal.int_gl = cur_moder.usec;
4464 
4465 	if (cur_moder.pkts < tqp_vector->tx_group.coal.int_ql_max) {
4466 		hns3_set_vector_coalesce_tx_ql(tqp_vector, cur_moder.pkts);
4467 		tqp_vector->tx_group.coal.int_ql = cur_moder.pkts;
4468 	}
4469 
4470 	dim->state = DIM_START_MEASURE;
4471 }
4472 
4473 static void hns3_nic_init_dim(struct hns3_enet_tqp_vector *tqp_vector)
4474 {
4475 	INIT_WORK(&tqp_vector->rx_group.dim.work, hns3_rx_dim_work);
4476 	INIT_WORK(&tqp_vector->tx_group.dim.work, hns3_tx_dim_work);
4477 }
4478 
4479 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
4480 {
4481 	struct hnae3_handle *h = priv->ae_handle;
4482 	struct hns3_enet_tqp_vector *tqp_vector;
4483 	int ret;
4484 	int i;
4485 
4486 	hns3_nic_set_cpumask(priv);
4487 
4488 	for (i = 0; i < priv->vector_num; i++) {
4489 		tqp_vector = &priv->tqp_vector[i];
4490 		hns3_vector_coalesce_init_hw(tqp_vector, priv);
4491 		tqp_vector->num_tqps = 0;
4492 		hns3_nic_init_dim(tqp_vector);
4493 	}
4494 
4495 	for (i = 0; i < h->kinfo.num_tqps; i++) {
4496 		u16 vector_i = i % priv->vector_num;
4497 		u16 tqp_num = h->kinfo.num_tqps;
4498 
4499 		tqp_vector = &priv->tqp_vector[vector_i];
4500 
4501 		hns3_add_ring_to_group(&tqp_vector->tx_group,
4502 				       &priv->ring[i]);
4503 
4504 		hns3_add_ring_to_group(&tqp_vector->rx_group,
4505 				       &priv->ring[i + tqp_num]);
4506 
4507 		priv->ring[i].tqp_vector = tqp_vector;
4508 		priv->ring[i + tqp_num].tqp_vector = tqp_vector;
4509 		tqp_vector->num_tqps++;
4510 	}
4511 
4512 	for (i = 0; i < priv->vector_num; i++) {
4513 		struct hnae3_ring_chain_node vector_ring_chain;
4514 
4515 		tqp_vector = &priv->tqp_vector[i];
4516 
4517 		tqp_vector->rx_group.total_bytes = 0;
4518 		tqp_vector->rx_group.total_packets = 0;
4519 		tqp_vector->tx_group.total_bytes = 0;
4520 		tqp_vector->tx_group.total_packets = 0;
4521 		tqp_vector->handle = h;
4522 
4523 		ret = hns3_get_vector_ring_chain(tqp_vector,
4524 						 &vector_ring_chain);
4525 		if (ret)
4526 			goto map_ring_fail;
4527 
4528 		ret = h->ae_algo->ops->map_ring_to_vector(h,
4529 			tqp_vector->vector_irq, &vector_ring_chain);
4530 
4531 		hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
4532 
4533 		if (ret)
4534 			goto map_ring_fail;
4535 
4536 		netif_napi_add(priv->netdev, &tqp_vector->napi,
4537 			       hns3_nic_common_poll, NAPI_POLL_WEIGHT);
4538 	}
4539 
4540 	return 0;
4541 
4542 map_ring_fail:
4543 	while (i--)
4544 		netif_napi_del(&priv->tqp_vector[i].napi);
4545 
4546 	return ret;
4547 }
4548 
4549 static void hns3_nic_init_coal_cfg(struct hns3_nic_priv *priv)
4550 {
4551 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
4552 	struct hns3_enet_coalesce *tx_coal = &priv->tx_coal;
4553 	struct hns3_enet_coalesce *rx_coal = &priv->rx_coal;
4554 
4555 	/* initialize the configuration for interrupt coalescing.
4556 	 * 1. GL (Interrupt Gap Limiter)
4557 	 * 2. RL (Interrupt Rate Limiter)
4558 	 * 3. QL (Interrupt Quantity Limiter)
4559 	 *
4560 	 * Default: enable interrupt coalescing self-adaptive and GL
4561 	 */
4562 	tx_coal->adapt_enable = 1;
4563 	rx_coal->adapt_enable = 1;
4564 
4565 	tx_coal->int_gl = HNS3_INT_GL_50K;
4566 	rx_coal->int_gl = HNS3_INT_GL_50K;
4567 
4568 	rx_coal->flow_level = HNS3_FLOW_LOW;
4569 	tx_coal->flow_level = HNS3_FLOW_LOW;
4570 
4571 	if (ae_dev->dev_specs.int_ql_max) {
4572 		tx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
4573 		rx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
4574 	}
4575 }
4576 
4577 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
4578 {
4579 	struct hnae3_handle *h = priv->ae_handle;
4580 	struct hns3_enet_tqp_vector *tqp_vector;
4581 	struct hnae3_vector_info *vector;
4582 	struct pci_dev *pdev = h->pdev;
4583 	u16 tqp_num = h->kinfo.num_tqps;
4584 	u16 vector_num;
4585 	int ret = 0;
4586 	u16 i;
4587 
4588 	/* RSS size, cpu online and vector_num should be the same */
4589 	/* Should consider 2p/4p later */
4590 	vector_num = min_t(u16, num_online_cpus(), tqp_num);
4591 
4592 	vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
4593 			      GFP_KERNEL);
4594 	if (!vector)
4595 		return -ENOMEM;
4596 
4597 	/* save the actual available vector number */
4598 	vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
4599 
4600 	priv->vector_num = vector_num;
4601 	priv->tqp_vector = (struct hns3_enet_tqp_vector *)
4602 		devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
4603 			     GFP_KERNEL);
4604 	if (!priv->tqp_vector) {
4605 		ret = -ENOMEM;
4606 		goto out;
4607 	}
4608 
4609 	for (i = 0; i < priv->vector_num; i++) {
4610 		tqp_vector = &priv->tqp_vector[i];
4611 		tqp_vector->idx = i;
4612 		tqp_vector->mask_addr = vector[i].io_addr;
4613 		tqp_vector->vector_irq = vector[i].vector;
4614 		hns3_vector_coalesce_init(tqp_vector, priv);
4615 	}
4616 
4617 out:
4618 	devm_kfree(&pdev->dev, vector);
4619 	return ret;
4620 }
4621 
4622 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
4623 {
4624 	group->ring = NULL;
4625 	group->count = 0;
4626 }
4627 
4628 static void hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
4629 {
4630 	struct hnae3_ring_chain_node vector_ring_chain;
4631 	struct hnae3_handle *h = priv->ae_handle;
4632 	struct hns3_enet_tqp_vector *tqp_vector;
4633 	int i;
4634 
4635 	for (i = 0; i < priv->vector_num; i++) {
4636 		tqp_vector = &priv->tqp_vector[i];
4637 
4638 		if (!tqp_vector->rx_group.ring && !tqp_vector->tx_group.ring)
4639 			continue;
4640 
4641 		/* Since the mapping can be overwritten, when fail to get the
4642 		 * chain between vector and ring, we should go on to deal with
4643 		 * the remaining options.
4644 		 */
4645 		if (hns3_get_vector_ring_chain(tqp_vector, &vector_ring_chain))
4646 			dev_warn(priv->dev, "failed to get ring chain\n");
4647 
4648 		h->ae_algo->ops->unmap_ring_from_vector(h,
4649 			tqp_vector->vector_irq, &vector_ring_chain);
4650 
4651 		hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
4652 
4653 		hns3_clear_ring_group(&tqp_vector->rx_group);
4654 		hns3_clear_ring_group(&tqp_vector->tx_group);
4655 		netif_napi_del(&priv->tqp_vector[i].napi);
4656 	}
4657 }
4658 
4659 static void hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
4660 {
4661 	struct hnae3_handle *h = priv->ae_handle;
4662 	struct pci_dev *pdev = h->pdev;
4663 	int i, ret;
4664 
4665 	for (i = 0; i < priv->vector_num; i++) {
4666 		struct hns3_enet_tqp_vector *tqp_vector;
4667 
4668 		tqp_vector = &priv->tqp_vector[i];
4669 		ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
4670 		if (ret)
4671 			return;
4672 	}
4673 
4674 	devm_kfree(&pdev->dev, priv->tqp_vector);
4675 }
4676 
4677 static void hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
4678 			      unsigned int ring_type)
4679 {
4680 	int queue_num = priv->ae_handle->kinfo.num_tqps;
4681 	struct hns3_enet_ring *ring;
4682 	int desc_num;
4683 
4684 	if (ring_type == HNAE3_RING_TYPE_TX) {
4685 		ring = &priv->ring[q->tqp_index];
4686 		desc_num = priv->ae_handle->kinfo.num_tx_desc;
4687 		ring->queue_index = q->tqp_index;
4688 		ring->tx_copybreak = priv->tx_copybreak;
4689 		ring->last_to_use = 0;
4690 	} else {
4691 		ring = &priv->ring[q->tqp_index + queue_num];
4692 		desc_num = priv->ae_handle->kinfo.num_rx_desc;
4693 		ring->queue_index = q->tqp_index;
4694 		ring->rx_copybreak = priv->rx_copybreak;
4695 	}
4696 
4697 	hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
4698 
4699 	ring->tqp = q;
4700 	ring->desc = NULL;
4701 	ring->desc_cb = NULL;
4702 	ring->dev = priv->dev;
4703 	ring->desc_dma_addr = 0;
4704 	ring->buf_size = q->buf_size;
4705 	ring->desc_num = desc_num;
4706 	ring->next_to_use = 0;
4707 	ring->next_to_clean = 0;
4708 }
4709 
4710 static void hns3_queue_to_ring(struct hnae3_queue *tqp,
4711 			       struct hns3_nic_priv *priv)
4712 {
4713 	hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
4714 	hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
4715 }
4716 
4717 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
4718 {
4719 	struct hnae3_handle *h = priv->ae_handle;
4720 	struct pci_dev *pdev = h->pdev;
4721 	int i;
4722 
4723 	priv->ring = devm_kzalloc(&pdev->dev,
4724 				  array3_size(h->kinfo.num_tqps,
4725 					      sizeof(*priv->ring), 2),
4726 				  GFP_KERNEL);
4727 	if (!priv->ring)
4728 		return -ENOMEM;
4729 
4730 	for (i = 0; i < h->kinfo.num_tqps; i++)
4731 		hns3_queue_to_ring(h->kinfo.tqp[i], priv);
4732 
4733 	return 0;
4734 }
4735 
4736 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
4737 {
4738 	if (!priv->ring)
4739 		return;
4740 
4741 	devm_kfree(priv->dev, priv->ring);
4742 	priv->ring = NULL;
4743 }
4744 
4745 static void hns3_alloc_page_pool(struct hns3_enet_ring *ring)
4746 {
4747 	struct page_pool_params pp_params = {
4748 		.flags = PP_FLAG_DMA_MAP | PP_FLAG_PAGE_FRAG |
4749 				PP_FLAG_DMA_SYNC_DEV,
4750 		.order = hns3_page_order(ring),
4751 		.pool_size = ring->desc_num * hns3_buf_size(ring) /
4752 				(PAGE_SIZE << hns3_page_order(ring)),
4753 		.nid = dev_to_node(ring_to_dev(ring)),
4754 		.dev = ring_to_dev(ring),
4755 		.dma_dir = DMA_FROM_DEVICE,
4756 		.offset = 0,
4757 		.max_len = PAGE_SIZE << hns3_page_order(ring),
4758 	};
4759 
4760 	ring->page_pool = page_pool_create(&pp_params);
4761 	if (IS_ERR(ring->page_pool)) {
4762 		dev_warn(ring_to_dev(ring), "page pool creation failed: %ld\n",
4763 			 PTR_ERR(ring->page_pool));
4764 		ring->page_pool = NULL;
4765 	}
4766 }
4767 
4768 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
4769 {
4770 	int ret;
4771 
4772 	if (ring->desc_num <= 0 || ring->buf_size <= 0)
4773 		return -EINVAL;
4774 
4775 	ring->desc_cb = devm_kcalloc(ring_to_dev(ring), ring->desc_num,
4776 				     sizeof(ring->desc_cb[0]), GFP_KERNEL);
4777 	if (!ring->desc_cb) {
4778 		ret = -ENOMEM;
4779 		goto out;
4780 	}
4781 
4782 	ret = hns3_alloc_desc(ring);
4783 	if (ret)
4784 		goto out_with_desc_cb;
4785 
4786 	if (!HNAE3_IS_TX_RING(ring)) {
4787 		hns3_alloc_page_pool(ring);
4788 
4789 		ret = hns3_alloc_ring_buffers(ring);
4790 		if (ret)
4791 			goto out_with_desc;
4792 	} else {
4793 		hns3_init_tx_spare_buffer(ring);
4794 	}
4795 
4796 	return 0;
4797 
4798 out_with_desc:
4799 	hns3_free_desc(ring);
4800 out_with_desc_cb:
4801 	devm_kfree(ring_to_dev(ring), ring->desc_cb);
4802 	ring->desc_cb = NULL;
4803 out:
4804 	return ret;
4805 }
4806 
4807 void hns3_fini_ring(struct hns3_enet_ring *ring)
4808 {
4809 	hns3_free_desc(ring);
4810 	devm_kfree(ring_to_dev(ring), ring->desc_cb);
4811 	ring->desc_cb = NULL;
4812 	ring->next_to_clean = 0;
4813 	ring->next_to_use = 0;
4814 	ring->last_to_use = 0;
4815 	ring->pending_buf = 0;
4816 	if (!HNAE3_IS_TX_RING(ring) && ring->skb) {
4817 		dev_kfree_skb_any(ring->skb);
4818 		ring->skb = NULL;
4819 	} else if (HNAE3_IS_TX_RING(ring) && ring->tx_spare) {
4820 		struct hns3_tx_spare *tx_spare = ring->tx_spare;
4821 
4822 		dma_unmap_page(ring_to_dev(ring), tx_spare->dma, tx_spare->len,
4823 			       DMA_TO_DEVICE);
4824 		free_pages((unsigned long)tx_spare->buf,
4825 			   get_order(tx_spare->len));
4826 		devm_kfree(ring_to_dev(ring), tx_spare);
4827 		ring->tx_spare = NULL;
4828 	}
4829 
4830 	if (!HNAE3_IS_TX_RING(ring) && ring->page_pool) {
4831 		page_pool_destroy(ring->page_pool);
4832 		ring->page_pool = NULL;
4833 	}
4834 }
4835 
4836 static int hns3_buf_size2type(u32 buf_size)
4837 {
4838 	int bd_size_type;
4839 
4840 	switch (buf_size) {
4841 	case 512:
4842 		bd_size_type = HNS3_BD_SIZE_512_TYPE;
4843 		break;
4844 	case 1024:
4845 		bd_size_type = HNS3_BD_SIZE_1024_TYPE;
4846 		break;
4847 	case 2048:
4848 		bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4849 		break;
4850 	case 4096:
4851 		bd_size_type = HNS3_BD_SIZE_4096_TYPE;
4852 		break;
4853 	default:
4854 		bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4855 	}
4856 
4857 	return bd_size_type;
4858 }
4859 
4860 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
4861 {
4862 	dma_addr_t dma = ring->desc_dma_addr;
4863 	struct hnae3_queue *q = ring->tqp;
4864 
4865 	if (!HNAE3_IS_TX_RING(ring)) {
4866 		hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG, (u32)dma);
4867 		hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
4868 			       (u32)((dma >> 31) >> 1));
4869 
4870 		hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
4871 			       hns3_buf_size2type(ring->buf_size));
4872 		hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
4873 			       ring->desc_num / 8 - 1);
4874 	} else {
4875 		hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
4876 			       (u32)dma);
4877 		hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
4878 			       (u32)((dma >> 31) >> 1));
4879 
4880 		hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
4881 			       ring->desc_num / 8 - 1);
4882 	}
4883 }
4884 
4885 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
4886 {
4887 	struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
4888 	struct hnae3_tc_info *tc_info = &kinfo->tc_info;
4889 	int i;
4890 
4891 	for (i = 0; i < HNAE3_MAX_TC; i++) {
4892 		int j;
4893 
4894 		if (!test_bit(i, &tc_info->tc_en))
4895 			continue;
4896 
4897 		for (j = 0; j < tc_info->tqp_count[i]; j++) {
4898 			struct hnae3_queue *q;
4899 
4900 			q = priv->ring[tc_info->tqp_offset[i] + j].tqp;
4901 			hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG, i);
4902 		}
4903 	}
4904 }
4905 
4906 int hns3_init_all_ring(struct hns3_nic_priv *priv)
4907 {
4908 	struct hnae3_handle *h = priv->ae_handle;
4909 	int ring_num = h->kinfo.num_tqps * 2;
4910 	int i, j;
4911 	int ret;
4912 
4913 	for (i = 0; i < ring_num; i++) {
4914 		ret = hns3_alloc_ring_memory(&priv->ring[i]);
4915 		if (ret) {
4916 			dev_err(priv->dev,
4917 				"Alloc ring memory fail! ret=%d\n", ret);
4918 			goto out_when_alloc_ring_memory;
4919 		}
4920 
4921 		u64_stats_init(&priv->ring[i].syncp);
4922 	}
4923 
4924 	return 0;
4925 
4926 out_when_alloc_ring_memory:
4927 	for (j = i - 1; j >= 0; j--)
4928 		hns3_fini_ring(&priv->ring[j]);
4929 
4930 	return -ENOMEM;
4931 }
4932 
4933 static void hns3_uninit_all_ring(struct hns3_nic_priv *priv)
4934 {
4935 	struct hnae3_handle *h = priv->ae_handle;
4936 	int i;
4937 
4938 	for (i = 0; i < h->kinfo.num_tqps; i++) {
4939 		hns3_fini_ring(&priv->ring[i]);
4940 		hns3_fini_ring(&priv->ring[i + h->kinfo.num_tqps]);
4941 	}
4942 }
4943 
4944 /* Set mac addr if it is configured. or leave it to the AE driver */
4945 static int hns3_init_mac_addr(struct net_device *netdev)
4946 {
4947 	struct hns3_nic_priv *priv = netdev_priv(netdev);
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 		dev_warn(priv->dev, "using random MAC address %pM\n",
4959 			 netdev->dev_addr);
4960 	} else if (!ether_addr_equal(netdev->dev_addr, mac_addr_temp)) {
4961 		ether_addr_copy(netdev->dev_addr, mac_addr_temp);
4962 		ether_addr_copy(netdev->perm_addr, mac_addr_temp);
4963 	} else {
4964 		return 0;
4965 	}
4966 
4967 	if (h->ae_algo->ops->set_mac_addr)
4968 		ret = h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
4969 
4970 	return ret;
4971 }
4972 
4973 static int hns3_init_phy(struct net_device *netdev)
4974 {
4975 	struct hnae3_handle *h = hns3_get_handle(netdev);
4976 	int ret = 0;
4977 
4978 	if (h->ae_algo->ops->mac_connect_phy)
4979 		ret = h->ae_algo->ops->mac_connect_phy(h);
4980 
4981 	return ret;
4982 }
4983 
4984 static void hns3_uninit_phy(struct net_device *netdev)
4985 {
4986 	struct hnae3_handle *h = hns3_get_handle(netdev);
4987 
4988 	if (h->ae_algo->ops->mac_disconnect_phy)
4989 		h->ae_algo->ops->mac_disconnect_phy(h);
4990 }
4991 
4992 static int hns3_client_start(struct hnae3_handle *handle)
4993 {
4994 	if (!handle->ae_algo->ops->client_start)
4995 		return 0;
4996 
4997 	return handle->ae_algo->ops->client_start(handle);
4998 }
4999 
5000 static void hns3_client_stop(struct hnae3_handle *handle)
5001 {
5002 	if (!handle->ae_algo->ops->client_stop)
5003 		return;
5004 
5005 	handle->ae_algo->ops->client_stop(handle);
5006 }
5007 
5008 static void hns3_info_show(struct hns3_nic_priv *priv)
5009 {
5010 	struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
5011 
5012 	dev_info(priv->dev, "MAC address: %pM\n", priv->netdev->dev_addr);
5013 	dev_info(priv->dev, "Task queue pairs numbers: %u\n", kinfo->num_tqps);
5014 	dev_info(priv->dev, "RSS size: %u\n", kinfo->rss_size);
5015 	dev_info(priv->dev, "Allocated RSS size: %u\n", kinfo->req_rss_size);
5016 	dev_info(priv->dev, "RX buffer length: %u\n", kinfo->rx_buf_len);
5017 	dev_info(priv->dev, "Desc num per TX queue: %u\n", kinfo->num_tx_desc);
5018 	dev_info(priv->dev, "Desc num per RX queue: %u\n", kinfo->num_rx_desc);
5019 	dev_info(priv->dev, "Total number of enabled TCs: %u\n",
5020 		 kinfo->tc_info.num_tc);
5021 	dev_info(priv->dev, "Max mtu size: %u\n", priv->netdev->max_mtu);
5022 }
5023 
5024 static void hns3_set_cq_period_mode(struct hns3_nic_priv *priv,
5025 				    enum dim_cq_period_mode mode, bool is_tx)
5026 {
5027 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
5028 	struct hnae3_handle *handle = priv->ae_handle;
5029 	int i;
5030 
5031 	if (is_tx) {
5032 		priv->tx_cqe_mode = mode;
5033 
5034 		for (i = 0; i < priv->vector_num; i++)
5035 			priv->tqp_vector[i].tx_group.dim.mode = mode;
5036 	} else {
5037 		priv->rx_cqe_mode = mode;
5038 
5039 		for (i = 0; i < priv->vector_num; i++)
5040 			priv->tqp_vector[i].rx_group.dim.mode = mode;
5041 	}
5042 
5043 	/* only device version above V3(include V3), GL can switch CQ/EQ
5044 	 * period mode.
5045 	 */
5046 	if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3) {
5047 		u32 new_mode;
5048 		u64 reg;
5049 
5050 		new_mode = (mode == DIM_CQ_PERIOD_MODE_START_FROM_CQE) ?
5051 			HNS3_CQ_MODE_CQE : HNS3_CQ_MODE_EQE;
5052 		reg = is_tx ? HNS3_GL1_CQ_MODE_REG : HNS3_GL0_CQ_MODE_REG;
5053 
5054 		writel(new_mode, handle->kinfo.io_base + reg);
5055 	}
5056 }
5057 
5058 void hns3_cq_period_mode_init(struct hns3_nic_priv *priv,
5059 			      enum dim_cq_period_mode tx_mode,
5060 			      enum dim_cq_period_mode rx_mode)
5061 {
5062 	hns3_set_cq_period_mode(priv, tx_mode, true);
5063 	hns3_set_cq_period_mode(priv, rx_mode, false);
5064 }
5065 
5066 static int hns3_client_init(struct hnae3_handle *handle)
5067 {
5068 	struct pci_dev *pdev = handle->pdev;
5069 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
5070 	u16 alloc_tqps, max_rss_size;
5071 	struct hns3_nic_priv *priv;
5072 	struct net_device *netdev;
5073 	int ret;
5074 
5075 	handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
5076 						    &max_rss_size);
5077 	netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
5078 	if (!netdev)
5079 		return -ENOMEM;
5080 
5081 	priv = netdev_priv(netdev);
5082 	priv->dev = &pdev->dev;
5083 	priv->netdev = netdev;
5084 	priv->ae_handle = handle;
5085 	priv->tx_timeout_count = 0;
5086 	priv->max_non_tso_bd_num = ae_dev->dev_specs.max_non_tso_bd_num;
5087 	set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
5088 
5089 	handle->msg_enable = netif_msg_init(debug, DEFAULT_MSG_LEVEL);
5090 
5091 	handle->kinfo.netdev = netdev;
5092 	handle->priv = (void *)priv;
5093 
5094 	hns3_init_mac_addr(netdev);
5095 
5096 	hns3_set_default_feature(netdev);
5097 
5098 	netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
5099 	netdev->priv_flags |= IFF_UNICAST_FLT;
5100 	netdev->netdev_ops = &hns3_nic_netdev_ops;
5101 	SET_NETDEV_DEV(netdev, &pdev->dev);
5102 	hns3_ethtool_set_ops(netdev);
5103 
5104 	/* Carrier off reporting is important to ethtool even BEFORE open */
5105 	netif_carrier_off(netdev);
5106 
5107 	ret = hns3_get_ring_config(priv);
5108 	if (ret) {
5109 		ret = -ENOMEM;
5110 		goto out_get_ring_cfg;
5111 	}
5112 
5113 	hns3_nic_init_coal_cfg(priv);
5114 
5115 	ret = hns3_nic_alloc_vector_data(priv);
5116 	if (ret) {
5117 		ret = -ENOMEM;
5118 		goto out_alloc_vector_data;
5119 	}
5120 
5121 	ret = hns3_nic_init_vector_data(priv);
5122 	if (ret) {
5123 		ret = -ENOMEM;
5124 		goto out_init_vector_data;
5125 	}
5126 
5127 	ret = hns3_init_all_ring(priv);
5128 	if (ret) {
5129 		ret = -ENOMEM;
5130 		goto out_init_ring;
5131 	}
5132 
5133 	hns3_cq_period_mode_init(priv, DIM_CQ_PERIOD_MODE_START_FROM_EQE,
5134 				 DIM_CQ_PERIOD_MODE_START_FROM_EQE);
5135 
5136 	ret = hns3_init_phy(netdev);
5137 	if (ret)
5138 		goto out_init_phy;
5139 
5140 	/* the device can work without cpu rmap, only aRFS needs it */
5141 	ret = hns3_set_rx_cpu_rmap(netdev);
5142 	if (ret)
5143 		dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
5144 
5145 	ret = hns3_nic_init_irq(priv);
5146 	if (ret) {
5147 		dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
5148 		hns3_free_rx_cpu_rmap(netdev);
5149 		goto out_init_irq_fail;
5150 	}
5151 
5152 	ret = hns3_client_start(handle);
5153 	if (ret) {
5154 		dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
5155 		goto out_client_start;
5156 	}
5157 
5158 	hns3_dcbnl_setup(handle);
5159 
5160 	ret = hns3_dbg_init(handle);
5161 	if (ret) {
5162 		dev_err(priv->dev, "failed to init debugfs, ret = %d\n",
5163 			ret);
5164 		goto out_client_start;
5165 	}
5166 
5167 	netdev->max_mtu = HNS3_MAX_MTU(ae_dev->dev_specs.max_frm_size);
5168 
5169 	if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps))
5170 		set_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state);
5171 
5172 	if (hnae3_ae_dev_rxd_adv_layout_supported(ae_dev))
5173 		set_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state);
5174 
5175 	set_bit(HNS3_NIC_STATE_INITED, &priv->state);
5176 
5177 	if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
5178 		set_bit(HNAE3_PFLAG_LIMIT_PROMISC, &handle->supported_pflags);
5179 
5180 	ret = register_netdev(netdev);
5181 	if (ret) {
5182 		dev_err(priv->dev, "probe register netdev fail!\n");
5183 		goto out_reg_netdev_fail;
5184 	}
5185 
5186 	if (netif_msg_drv(handle))
5187 		hns3_info_show(priv);
5188 
5189 	return ret;
5190 
5191 out_reg_netdev_fail:
5192 	hns3_dbg_uninit(handle);
5193 out_client_start:
5194 	hns3_free_rx_cpu_rmap(netdev);
5195 	hns3_nic_uninit_irq(priv);
5196 out_init_irq_fail:
5197 	hns3_uninit_phy(netdev);
5198 out_init_phy:
5199 	hns3_uninit_all_ring(priv);
5200 out_init_ring:
5201 	hns3_nic_uninit_vector_data(priv);
5202 out_init_vector_data:
5203 	hns3_nic_dealloc_vector_data(priv);
5204 out_alloc_vector_data:
5205 	priv->ring = NULL;
5206 out_get_ring_cfg:
5207 	priv->ae_handle = NULL;
5208 	free_netdev(netdev);
5209 	return ret;
5210 }
5211 
5212 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
5213 {
5214 	struct net_device *netdev = handle->kinfo.netdev;
5215 	struct hns3_nic_priv *priv = netdev_priv(netdev);
5216 
5217 	if (netdev->reg_state != NETREG_UNINITIALIZED)
5218 		unregister_netdev(netdev);
5219 
5220 	hns3_client_stop(handle);
5221 
5222 	hns3_uninit_phy(netdev);
5223 
5224 	if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5225 		netdev_warn(netdev, "already uninitialized\n");
5226 		goto out_netdev_free;
5227 	}
5228 
5229 	hns3_free_rx_cpu_rmap(netdev);
5230 
5231 	hns3_nic_uninit_irq(priv);
5232 
5233 	hns3_clear_all_ring(handle, true);
5234 
5235 	hns3_nic_uninit_vector_data(priv);
5236 
5237 	hns3_nic_dealloc_vector_data(priv);
5238 
5239 	hns3_uninit_all_ring(priv);
5240 
5241 	hns3_put_ring_config(priv);
5242 
5243 out_netdev_free:
5244 	hns3_dbg_uninit(handle);
5245 	free_netdev(netdev);
5246 }
5247 
5248 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
5249 {
5250 	struct net_device *netdev = handle->kinfo.netdev;
5251 
5252 	if (!netdev)
5253 		return;
5254 
5255 	if (linkup) {
5256 		netif_tx_wake_all_queues(netdev);
5257 		netif_carrier_on(netdev);
5258 		if (netif_msg_link(handle))
5259 			netdev_info(netdev, "link up\n");
5260 	} else {
5261 		netif_carrier_off(netdev);
5262 		netif_tx_stop_all_queues(netdev);
5263 		if (netif_msg_link(handle))
5264 			netdev_info(netdev, "link down\n");
5265 	}
5266 }
5267 
5268 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
5269 {
5270 	while (ring->next_to_clean != ring->next_to_use) {
5271 		ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
5272 		hns3_free_buffer_detach(ring, ring->next_to_clean, 0);
5273 		ring_ptr_move_fw(ring, next_to_clean);
5274 	}
5275 
5276 	ring->pending_buf = 0;
5277 }
5278 
5279 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
5280 {
5281 	struct hns3_desc_cb res_cbs;
5282 	int ret;
5283 
5284 	while (ring->next_to_use != ring->next_to_clean) {
5285 		/* When a buffer is not reused, it's memory has been
5286 		 * freed in hns3_handle_rx_bd or will be freed by
5287 		 * stack, so we need to replace the buffer here.
5288 		 */
5289 		if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
5290 			ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
5291 			if (ret) {
5292 				u64_stats_update_begin(&ring->syncp);
5293 				ring->stats.sw_err_cnt++;
5294 				u64_stats_update_end(&ring->syncp);
5295 				/* if alloc new buffer fail, exit directly
5296 				 * and reclear in up flow.
5297 				 */
5298 				netdev_warn(ring_to_netdev(ring),
5299 					    "reserve buffer map failed, ret = %d\n",
5300 					    ret);
5301 				return ret;
5302 			}
5303 			hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
5304 		}
5305 		ring_ptr_move_fw(ring, next_to_use);
5306 	}
5307 
5308 	/* Free the pending skb in rx ring */
5309 	if (ring->skb) {
5310 		dev_kfree_skb_any(ring->skb);
5311 		ring->skb = NULL;
5312 		ring->pending_buf = 0;
5313 	}
5314 
5315 	return 0;
5316 }
5317 
5318 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
5319 {
5320 	while (ring->next_to_use != ring->next_to_clean) {
5321 		/* When a buffer is not reused, it's memory has been
5322 		 * freed in hns3_handle_rx_bd or will be freed by
5323 		 * stack, so only need to unmap the buffer here.
5324 		 */
5325 		if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
5326 			hns3_unmap_buffer(ring,
5327 					  &ring->desc_cb[ring->next_to_use]);
5328 			ring->desc_cb[ring->next_to_use].dma = 0;
5329 		}
5330 
5331 		ring_ptr_move_fw(ring, next_to_use);
5332 	}
5333 }
5334 
5335 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force)
5336 {
5337 	struct net_device *ndev = h->kinfo.netdev;
5338 	struct hns3_nic_priv *priv = netdev_priv(ndev);
5339 	u32 i;
5340 
5341 	for (i = 0; i < h->kinfo.num_tqps; i++) {
5342 		struct hns3_enet_ring *ring;
5343 
5344 		ring = &priv->ring[i];
5345 		hns3_clear_tx_ring(ring);
5346 
5347 		ring = &priv->ring[i + h->kinfo.num_tqps];
5348 		/* Continue to clear other rings even if clearing some
5349 		 * rings failed.
5350 		 */
5351 		if (force)
5352 			hns3_force_clear_rx_ring(ring);
5353 		else
5354 			hns3_clear_rx_ring(ring);
5355 	}
5356 }
5357 
5358 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
5359 {
5360 	struct net_device *ndev = h->kinfo.netdev;
5361 	struct hns3_nic_priv *priv = netdev_priv(ndev);
5362 	struct hns3_enet_ring *rx_ring;
5363 	int i, j;
5364 	int ret;
5365 
5366 	ret = h->ae_algo->ops->reset_queue(h);
5367 	if (ret)
5368 		return ret;
5369 
5370 	for (i = 0; i < h->kinfo.num_tqps; i++) {
5371 		hns3_init_ring_hw(&priv->ring[i]);
5372 
5373 		/* We need to clear tx ring here because self test will
5374 		 * use the ring and will not run down before up
5375 		 */
5376 		hns3_clear_tx_ring(&priv->ring[i]);
5377 		priv->ring[i].next_to_clean = 0;
5378 		priv->ring[i].next_to_use = 0;
5379 		priv->ring[i].last_to_use = 0;
5380 
5381 		rx_ring = &priv->ring[i + h->kinfo.num_tqps];
5382 		hns3_init_ring_hw(rx_ring);
5383 		ret = hns3_clear_rx_ring(rx_ring);
5384 		if (ret)
5385 			return ret;
5386 
5387 		/* We can not know the hardware head and tail when this
5388 		 * function is called in reset flow, so we reuse all desc.
5389 		 */
5390 		for (j = 0; j < rx_ring->desc_num; j++)
5391 			hns3_reuse_buffer(rx_ring, j);
5392 
5393 		rx_ring->next_to_clean = 0;
5394 		rx_ring->next_to_use = 0;
5395 	}
5396 
5397 	hns3_init_tx_ring_tc(priv);
5398 
5399 	return 0;
5400 }
5401 
5402 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
5403 {
5404 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
5405 	struct net_device *ndev = kinfo->netdev;
5406 	struct hns3_nic_priv *priv = netdev_priv(ndev);
5407 
5408 	if (test_and_set_bit(HNS3_NIC_STATE_RESETTING, &priv->state))
5409 		return 0;
5410 
5411 	if (!netif_running(ndev))
5412 		return 0;
5413 
5414 	return hns3_nic_net_stop(ndev);
5415 }
5416 
5417 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
5418 {
5419 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
5420 	struct hns3_nic_priv *priv = netdev_priv(kinfo->netdev);
5421 	int ret = 0;
5422 
5423 	if (!test_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5424 		netdev_err(kinfo->netdev, "device is not initialized yet\n");
5425 		return -EFAULT;
5426 	}
5427 
5428 	clear_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
5429 
5430 	if (netif_running(kinfo->netdev)) {
5431 		ret = hns3_nic_net_open(kinfo->netdev);
5432 		if (ret) {
5433 			set_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
5434 			netdev_err(kinfo->netdev,
5435 				   "net up fail, ret=%d!\n", ret);
5436 			return ret;
5437 		}
5438 	}
5439 
5440 	return ret;
5441 }
5442 
5443 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
5444 {
5445 	struct net_device *netdev = handle->kinfo.netdev;
5446 	struct hns3_nic_priv *priv = netdev_priv(netdev);
5447 	int ret;
5448 
5449 	/* Carrier off reporting is important to ethtool even BEFORE open */
5450 	netif_carrier_off(netdev);
5451 
5452 	ret = hns3_get_ring_config(priv);
5453 	if (ret)
5454 		return ret;
5455 
5456 	ret = hns3_nic_alloc_vector_data(priv);
5457 	if (ret)
5458 		goto err_put_ring;
5459 
5460 	ret = hns3_nic_init_vector_data(priv);
5461 	if (ret)
5462 		goto err_dealloc_vector;
5463 
5464 	ret = hns3_init_all_ring(priv);
5465 	if (ret)
5466 		goto err_uninit_vector;
5467 
5468 	hns3_cq_period_mode_init(priv, priv->tx_cqe_mode, priv->rx_cqe_mode);
5469 
5470 	/* the device can work without cpu rmap, only aRFS needs it */
5471 	ret = hns3_set_rx_cpu_rmap(netdev);
5472 	if (ret)
5473 		dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
5474 
5475 	ret = hns3_nic_init_irq(priv);
5476 	if (ret) {
5477 		dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
5478 		hns3_free_rx_cpu_rmap(netdev);
5479 		goto err_init_irq_fail;
5480 	}
5481 
5482 	if (!hns3_is_phys_func(handle->pdev))
5483 		hns3_init_mac_addr(netdev);
5484 
5485 	ret = hns3_client_start(handle);
5486 	if (ret) {
5487 		dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
5488 		goto err_client_start_fail;
5489 	}
5490 
5491 	set_bit(HNS3_NIC_STATE_INITED, &priv->state);
5492 
5493 	return ret;
5494 
5495 err_client_start_fail:
5496 	hns3_free_rx_cpu_rmap(netdev);
5497 	hns3_nic_uninit_irq(priv);
5498 err_init_irq_fail:
5499 	hns3_uninit_all_ring(priv);
5500 err_uninit_vector:
5501 	hns3_nic_uninit_vector_data(priv);
5502 err_dealloc_vector:
5503 	hns3_nic_dealloc_vector_data(priv);
5504 err_put_ring:
5505 	hns3_put_ring_config(priv);
5506 
5507 	return ret;
5508 }
5509 
5510 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
5511 {
5512 	struct net_device *netdev = handle->kinfo.netdev;
5513 	struct hns3_nic_priv *priv = netdev_priv(netdev);
5514 
5515 	if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5516 		netdev_warn(netdev, "already uninitialized\n");
5517 		return 0;
5518 	}
5519 
5520 	hns3_free_rx_cpu_rmap(netdev);
5521 	hns3_nic_uninit_irq(priv);
5522 	hns3_clear_all_ring(handle, true);
5523 	hns3_reset_tx_queue(priv->ae_handle);
5524 
5525 	hns3_nic_uninit_vector_data(priv);
5526 
5527 	hns3_nic_dealloc_vector_data(priv);
5528 
5529 	hns3_uninit_all_ring(priv);
5530 
5531 	hns3_put_ring_config(priv);
5532 
5533 	return 0;
5534 }
5535 
5536 static int hns3_reset_notify(struct hnae3_handle *handle,
5537 			     enum hnae3_reset_notify_type type)
5538 {
5539 	int ret = 0;
5540 
5541 	switch (type) {
5542 	case HNAE3_UP_CLIENT:
5543 		ret = hns3_reset_notify_up_enet(handle);
5544 		break;
5545 	case HNAE3_DOWN_CLIENT:
5546 		ret = hns3_reset_notify_down_enet(handle);
5547 		break;
5548 	case HNAE3_INIT_CLIENT:
5549 		ret = hns3_reset_notify_init_enet(handle);
5550 		break;
5551 	case HNAE3_UNINIT_CLIENT:
5552 		ret = hns3_reset_notify_uninit_enet(handle);
5553 		break;
5554 	default:
5555 		break;
5556 	}
5557 
5558 	return ret;
5559 }
5560 
5561 static int hns3_change_channels(struct hnae3_handle *handle, u32 new_tqp_num,
5562 				bool rxfh_configured)
5563 {
5564 	int ret;
5565 
5566 	ret = handle->ae_algo->ops->set_channels(handle, new_tqp_num,
5567 						 rxfh_configured);
5568 	if (ret) {
5569 		dev_err(&handle->pdev->dev,
5570 			"Change tqp num(%u) fail.\n", new_tqp_num);
5571 		return ret;
5572 	}
5573 
5574 	ret = hns3_reset_notify(handle, HNAE3_INIT_CLIENT);
5575 	if (ret)
5576 		return ret;
5577 
5578 	ret =  hns3_reset_notify(handle, HNAE3_UP_CLIENT);
5579 	if (ret)
5580 		hns3_reset_notify(handle, HNAE3_UNINIT_CLIENT);
5581 
5582 	return ret;
5583 }
5584 
5585 int hns3_set_channels(struct net_device *netdev,
5586 		      struct ethtool_channels *ch)
5587 {
5588 	struct hnae3_handle *h = hns3_get_handle(netdev);
5589 	struct hnae3_knic_private_info *kinfo = &h->kinfo;
5590 	bool rxfh_configured = netif_is_rxfh_configured(netdev);
5591 	u32 new_tqp_num = ch->combined_count;
5592 	u16 org_tqp_num;
5593 	int ret;
5594 
5595 	if (hns3_nic_resetting(netdev))
5596 		return -EBUSY;
5597 
5598 	if (ch->rx_count || ch->tx_count)
5599 		return -EINVAL;
5600 
5601 	if (kinfo->tc_info.mqprio_active) {
5602 		dev_err(&netdev->dev,
5603 			"it's not allowed to set channels via ethtool when MQPRIO mode is on\n");
5604 		return -EINVAL;
5605 	}
5606 
5607 	if (new_tqp_num > hns3_get_max_available_channels(h) ||
5608 	    new_tqp_num < 1) {
5609 		dev_err(&netdev->dev,
5610 			"Change tqps fail, the tqp range is from 1 to %u",
5611 			hns3_get_max_available_channels(h));
5612 		return -EINVAL;
5613 	}
5614 
5615 	if (kinfo->rss_size == new_tqp_num)
5616 		return 0;
5617 
5618 	netif_dbg(h, drv, netdev,
5619 		  "set channels: tqp_num=%u, rxfh=%d\n",
5620 		  new_tqp_num, rxfh_configured);
5621 
5622 	ret = hns3_reset_notify(h, HNAE3_DOWN_CLIENT);
5623 	if (ret)
5624 		return ret;
5625 
5626 	ret = hns3_reset_notify(h, HNAE3_UNINIT_CLIENT);
5627 	if (ret)
5628 		return ret;
5629 
5630 	org_tqp_num = h->kinfo.num_tqps;
5631 	ret = hns3_change_channels(h, new_tqp_num, rxfh_configured);
5632 	if (ret) {
5633 		int ret1;
5634 
5635 		netdev_warn(netdev,
5636 			    "Change channels fail, revert to old value\n");
5637 		ret1 = hns3_change_channels(h, org_tqp_num, rxfh_configured);
5638 		if (ret1) {
5639 			netdev_err(netdev,
5640 				   "revert to old channel fail\n");
5641 			return ret1;
5642 		}
5643 
5644 		return ret;
5645 	}
5646 
5647 	return 0;
5648 }
5649 
5650 static const struct hns3_hw_error_info hns3_hw_err[] = {
5651 	{ .type = HNAE3_PPU_POISON_ERROR,
5652 	  .msg = "PPU poison" },
5653 	{ .type = HNAE3_CMDQ_ECC_ERROR,
5654 	  .msg = "IMP CMDQ error" },
5655 	{ .type = HNAE3_IMP_RD_POISON_ERROR,
5656 	  .msg = "IMP RD poison" },
5657 	{ .type = HNAE3_ROCEE_AXI_RESP_ERROR,
5658 	  .msg = "ROCEE AXI RESP error" },
5659 };
5660 
5661 static void hns3_process_hw_error(struct hnae3_handle *handle,
5662 				  enum hnae3_hw_error_type type)
5663 {
5664 	int i;
5665 
5666 	for (i = 0; i < ARRAY_SIZE(hns3_hw_err); i++) {
5667 		if (hns3_hw_err[i].type == type) {
5668 			dev_err(&handle->pdev->dev, "Detected %s!\n",
5669 				hns3_hw_err[i].msg);
5670 			break;
5671 		}
5672 	}
5673 }
5674 
5675 static const struct hnae3_client_ops client_ops = {
5676 	.init_instance = hns3_client_init,
5677 	.uninit_instance = hns3_client_uninit,
5678 	.link_status_change = hns3_link_status_change,
5679 	.reset_notify = hns3_reset_notify,
5680 	.process_hw_error = hns3_process_hw_error,
5681 };
5682 
5683 /* hns3_init_module - Driver registration routine
5684  * hns3_init_module is the first routine called when the driver is
5685  * loaded. All it does is register with the PCI subsystem.
5686  */
5687 static int __init hns3_init_module(void)
5688 {
5689 	int ret;
5690 
5691 	pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
5692 	pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
5693 
5694 	client.type = HNAE3_CLIENT_KNIC;
5695 	snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH, "%s",
5696 		 hns3_driver_name);
5697 
5698 	client.ops = &client_ops;
5699 
5700 	INIT_LIST_HEAD(&client.node);
5701 
5702 	hns3_dbg_register_debugfs(hns3_driver_name);
5703 
5704 	ret = hnae3_register_client(&client);
5705 	if (ret)
5706 		goto err_reg_client;
5707 
5708 	ret = pci_register_driver(&hns3_driver);
5709 	if (ret)
5710 		goto err_reg_driver;
5711 
5712 	return ret;
5713 
5714 err_reg_driver:
5715 	hnae3_unregister_client(&client);
5716 err_reg_client:
5717 	hns3_dbg_unregister_debugfs();
5718 	return ret;
5719 }
5720 module_init(hns3_init_module);
5721 
5722 /* hns3_exit_module - Driver exit cleanup routine
5723  * hns3_exit_module is called just before the driver is removed
5724  * from memory.
5725  */
5726 static void __exit hns3_exit_module(void)
5727 {
5728 	pci_unregister_driver(&hns3_driver);
5729 	hnae3_unregister_client(&client);
5730 	hns3_dbg_unregister_debugfs();
5731 }
5732 module_exit(hns3_exit_module);
5733 
5734 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
5735 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
5736 MODULE_LICENSE("GPL");
5737 MODULE_ALIAS("pci:hns-nic");
5738