xref: /openbmc/linux/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c (revision 22a41e9a5044bf3519f05b4a00e99af34bfeb40c)
1 // SPDX-License-Identifier: GPL-2.0+
2 // Copyright (c) 2016-2017 Hisilicon Limited.
3 
4 #include <linux/dma-mapping.h>
5 #include <linux/etherdevice.h>
6 #include <linux/interrupt.h>
7 #ifdef CONFIG_RFS_ACCEL
8 #include <linux/cpu_rmap.h>
9 #endif
10 #include <linux/if_vlan.h>
11 #include <linux/irq.h>
12 #include <linux/ip.h>
13 #include <linux/ipv6.h>
14 #include <linux/module.h>
15 #include <linux/pci.h>
16 #include <linux/aer.h>
17 #include <linux/skbuff.h>
18 #include <linux/sctp.h>
19 #include <net/gre.h>
20 #include <net/gro.h>
21 #include <net/ip6_checksum.h>
22 #include <net/pkt_cls.h>
23 #include <net/tcp.h>
24 #include <net/vxlan.h>
25 #include <net/geneve.h>
26 
27 #include "hnae3.h"
28 #include "hns3_enet.h"
29 /* All hns3 tracepoints are defined by the include below, which
30  * must be included exactly once across the whole kernel with
31  * CREATE_TRACE_POINTS defined
32  */
33 #define CREATE_TRACE_POINTS
34 #include "hns3_trace.h"
35 
36 #define hns3_set_field(origin, shift, val)	((origin) |= (val) << (shift))
37 #define hns3_tx_bd_count(S)	DIV_ROUND_UP(S, HNS3_MAX_BD_SIZE)
38 
39 #define hns3_rl_err(fmt, ...)						\
40 	do {								\
41 		if (net_ratelimit())					\
42 			netdev_err(fmt, ##__VA_ARGS__);			\
43 	} while (0)
44 
45 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force);
46 
47 static const char hns3_driver_name[] = "hns3";
48 static const char hns3_driver_string[] =
49 			"Hisilicon Ethernet Network Driver for Hip08 Family";
50 static const char hns3_copyright[] = "Copyright (c) 2017 Huawei Corporation.";
51 static struct hnae3_client client;
52 
53 static int debug = -1;
54 module_param(debug, int, 0);
55 MODULE_PARM_DESC(debug, " Network interface message level setting");
56 
57 static unsigned int tx_sgl = 1;
58 module_param(tx_sgl, uint, 0600);
59 MODULE_PARM_DESC(tx_sgl, "Minimum number of frags when using dma_map_sg() to optimize the IOMMU mapping");
60 
61 static bool page_pool_enabled = true;
62 module_param(page_pool_enabled, bool, 0400);
63 
64 #define HNS3_SGL_SIZE(nfrag)	(sizeof(struct scatterlist) * (nfrag) +	\
65 				 sizeof(struct sg_table))
66 #define HNS3_MAX_SGL_SIZE	ALIGN(HNS3_SGL_SIZE(HNS3_MAX_TSO_BD_NUM), \
67 				      dma_get_cache_alignment())
68 
69 #define DEFAULT_MSG_LEVEL (NETIF_MSG_PROBE | NETIF_MSG_LINK | \
70 			   NETIF_MSG_IFDOWN | NETIF_MSG_IFUP)
71 
72 #define HNS3_INNER_VLAN_TAG	1
73 #define HNS3_OUTER_VLAN_TAG	2
74 
75 #define HNS3_MIN_TX_LEN		33U
76 #define HNS3_MIN_TUN_PKT_LEN	65U
77 
78 /* hns3_pci_tbl - PCI Device ID Table
79  *
80  * Last entry must be all 0s
81  *
82  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
83  *   Class, Class Mask, private data (not used) }
84  */
85 static const struct pci_device_id hns3_pci_tbl[] = {
86 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
87 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
88 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA),
89 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
90 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC),
91 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
92 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA),
93 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
94 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC),
95 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
96 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC),
97 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
98 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_200G_RDMA),
99 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
100 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_VF), 0},
101 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_RDMA_DCB_PFC_VF),
102 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
103 	/* required last entry */
104 	{0,}
105 };
106 MODULE_DEVICE_TABLE(pci, hns3_pci_tbl);
107 
108 #define HNS3_RX_PTYPE_ENTRY(ptype, l, s, t) \
109 	{	ptype, \
110 		l, \
111 		CHECKSUM_##s, \
112 		HNS3_L3_TYPE_##t, \
113 		1 }
114 
115 #define HNS3_RX_PTYPE_UNUSED_ENTRY(ptype) \
116 		{ ptype, 0, CHECKSUM_NONE, HNS3_L3_TYPE_PARSE_FAIL, 0 }
117 
118 static const struct hns3_rx_ptype hns3_rx_ptype_tbl[] = {
119 	HNS3_RX_PTYPE_UNUSED_ENTRY(0),
120 	HNS3_RX_PTYPE_ENTRY(1, 0, COMPLETE, ARP),
121 	HNS3_RX_PTYPE_ENTRY(2, 0, COMPLETE, RARP),
122 	HNS3_RX_PTYPE_ENTRY(3, 0, COMPLETE, LLDP),
123 	HNS3_RX_PTYPE_ENTRY(4, 0, COMPLETE, PARSE_FAIL),
124 	HNS3_RX_PTYPE_ENTRY(5, 0, COMPLETE, PARSE_FAIL),
125 	HNS3_RX_PTYPE_ENTRY(6, 0, COMPLETE, PARSE_FAIL),
126 	HNS3_RX_PTYPE_ENTRY(7, 0, COMPLETE, CNM),
127 	HNS3_RX_PTYPE_ENTRY(8, 0, NONE, PARSE_FAIL),
128 	HNS3_RX_PTYPE_UNUSED_ENTRY(9),
129 	HNS3_RX_PTYPE_UNUSED_ENTRY(10),
130 	HNS3_RX_PTYPE_UNUSED_ENTRY(11),
131 	HNS3_RX_PTYPE_UNUSED_ENTRY(12),
132 	HNS3_RX_PTYPE_UNUSED_ENTRY(13),
133 	HNS3_RX_PTYPE_UNUSED_ENTRY(14),
134 	HNS3_RX_PTYPE_UNUSED_ENTRY(15),
135 	HNS3_RX_PTYPE_ENTRY(16, 0, COMPLETE, PARSE_FAIL),
136 	HNS3_RX_PTYPE_ENTRY(17, 0, COMPLETE, IPV4),
137 	HNS3_RX_PTYPE_ENTRY(18, 0, COMPLETE, IPV4),
138 	HNS3_RX_PTYPE_ENTRY(19, 0, UNNECESSARY, IPV4),
139 	HNS3_RX_PTYPE_ENTRY(20, 0, UNNECESSARY, IPV4),
140 	HNS3_RX_PTYPE_ENTRY(21, 0, NONE, IPV4),
141 	HNS3_RX_PTYPE_ENTRY(22, 0, UNNECESSARY, IPV4),
142 	HNS3_RX_PTYPE_ENTRY(23, 0, NONE, IPV4),
143 	HNS3_RX_PTYPE_ENTRY(24, 0, NONE, IPV4),
144 	HNS3_RX_PTYPE_ENTRY(25, 0, UNNECESSARY, IPV4),
145 	HNS3_RX_PTYPE_UNUSED_ENTRY(26),
146 	HNS3_RX_PTYPE_UNUSED_ENTRY(27),
147 	HNS3_RX_PTYPE_UNUSED_ENTRY(28),
148 	HNS3_RX_PTYPE_ENTRY(29, 0, COMPLETE, PARSE_FAIL),
149 	HNS3_RX_PTYPE_ENTRY(30, 0, COMPLETE, PARSE_FAIL),
150 	HNS3_RX_PTYPE_ENTRY(31, 0, COMPLETE, IPV4),
151 	HNS3_RX_PTYPE_ENTRY(32, 0, COMPLETE, IPV4),
152 	HNS3_RX_PTYPE_ENTRY(33, 1, UNNECESSARY, IPV4),
153 	HNS3_RX_PTYPE_ENTRY(34, 1, UNNECESSARY, IPV4),
154 	HNS3_RX_PTYPE_ENTRY(35, 1, UNNECESSARY, IPV4),
155 	HNS3_RX_PTYPE_ENTRY(36, 0, COMPLETE, IPV4),
156 	HNS3_RX_PTYPE_ENTRY(37, 0, COMPLETE, IPV4),
157 	HNS3_RX_PTYPE_UNUSED_ENTRY(38),
158 	HNS3_RX_PTYPE_ENTRY(39, 0, COMPLETE, IPV6),
159 	HNS3_RX_PTYPE_ENTRY(40, 0, COMPLETE, IPV6),
160 	HNS3_RX_PTYPE_ENTRY(41, 1, UNNECESSARY, IPV6),
161 	HNS3_RX_PTYPE_ENTRY(42, 1, UNNECESSARY, IPV6),
162 	HNS3_RX_PTYPE_ENTRY(43, 1, UNNECESSARY, IPV6),
163 	HNS3_RX_PTYPE_ENTRY(44, 0, COMPLETE, IPV6),
164 	HNS3_RX_PTYPE_ENTRY(45, 0, COMPLETE, IPV6),
165 	HNS3_RX_PTYPE_UNUSED_ENTRY(46),
166 	HNS3_RX_PTYPE_UNUSED_ENTRY(47),
167 	HNS3_RX_PTYPE_UNUSED_ENTRY(48),
168 	HNS3_RX_PTYPE_UNUSED_ENTRY(49),
169 	HNS3_RX_PTYPE_UNUSED_ENTRY(50),
170 	HNS3_RX_PTYPE_UNUSED_ENTRY(51),
171 	HNS3_RX_PTYPE_UNUSED_ENTRY(52),
172 	HNS3_RX_PTYPE_UNUSED_ENTRY(53),
173 	HNS3_RX_PTYPE_UNUSED_ENTRY(54),
174 	HNS3_RX_PTYPE_UNUSED_ENTRY(55),
175 	HNS3_RX_PTYPE_UNUSED_ENTRY(56),
176 	HNS3_RX_PTYPE_UNUSED_ENTRY(57),
177 	HNS3_RX_PTYPE_UNUSED_ENTRY(58),
178 	HNS3_RX_PTYPE_UNUSED_ENTRY(59),
179 	HNS3_RX_PTYPE_UNUSED_ENTRY(60),
180 	HNS3_RX_PTYPE_UNUSED_ENTRY(61),
181 	HNS3_RX_PTYPE_UNUSED_ENTRY(62),
182 	HNS3_RX_PTYPE_UNUSED_ENTRY(63),
183 	HNS3_RX_PTYPE_UNUSED_ENTRY(64),
184 	HNS3_RX_PTYPE_UNUSED_ENTRY(65),
185 	HNS3_RX_PTYPE_UNUSED_ENTRY(66),
186 	HNS3_RX_PTYPE_UNUSED_ENTRY(67),
187 	HNS3_RX_PTYPE_UNUSED_ENTRY(68),
188 	HNS3_RX_PTYPE_UNUSED_ENTRY(69),
189 	HNS3_RX_PTYPE_UNUSED_ENTRY(70),
190 	HNS3_RX_PTYPE_UNUSED_ENTRY(71),
191 	HNS3_RX_PTYPE_UNUSED_ENTRY(72),
192 	HNS3_RX_PTYPE_UNUSED_ENTRY(73),
193 	HNS3_RX_PTYPE_UNUSED_ENTRY(74),
194 	HNS3_RX_PTYPE_UNUSED_ENTRY(75),
195 	HNS3_RX_PTYPE_UNUSED_ENTRY(76),
196 	HNS3_RX_PTYPE_UNUSED_ENTRY(77),
197 	HNS3_RX_PTYPE_UNUSED_ENTRY(78),
198 	HNS3_RX_PTYPE_UNUSED_ENTRY(79),
199 	HNS3_RX_PTYPE_UNUSED_ENTRY(80),
200 	HNS3_RX_PTYPE_UNUSED_ENTRY(81),
201 	HNS3_RX_PTYPE_UNUSED_ENTRY(82),
202 	HNS3_RX_PTYPE_UNUSED_ENTRY(83),
203 	HNS3_RX_PTYPE_UNUSED_ENTRY(84),
204 	HNS3_RX_PTYPE_UNUSED_ENTRY(85),
205 	HNS3_RX_PTYPE_UNUSED_ENTRY(86),
206 	HNS3_RX_PTYPE_UNUSED_ENTRY(87),
207 	HNS3_RX_PTYPE_UNUSED_ENTRY(88),
208 	HNS3_RX_PTYPE_UNUSED_ENTRY(89),
209 	HNS3_RX_PTYPE_UNUSED_ENTRY(90),
210 	HNS3_RX_PTYPE_UNUSED_ENTRY(91),
211 	HNS3_RX_PTYPE_UNUSED_ENTRY(92),
212 	HNS3_RX_PTYPE_UNUSED_ENTRY(93),
213 	HNS3_RX_PTYPE_UNUSED_ENTRY(94),
214 	HNS3_RX_PTYPE_UNUSED_ENTRY(95),
215 	HNS3_RX_PTYPE_UNUSED_ENTRY(96),
216 	HNS3_RX_PTYPE_UNUSED_ENTRY(97),
217 	HNS3_RX_PTYPE_UNUSED_ENTRY(98),
218 	HNS3_RX_PTYPE_UNUSED_ENTRY(99),
219 	HNS3_RX_PTYPE_UNUSED_ENTRY(100),
220 	HNS3_RX_PTYPE_UNUSED_ENTRY(101),
221 	HNS3_RX_PTYPE_UNUSED_ENTRY(102),
222 	HNS3_RX_PTYPE_UNUSED_ENTRY(103),
223 	HNS3_RX_PTYPE_UNUSED_ENTRY(104),
224 	HNS3_RX_PTYPE_UNUSED_ENTRY(105),
225 	HNS3_RX_PTYPE_UNUSED_ENTRY(106),
226 	HNS3_RX_PTYPE_UNUSED_ENTRY(107),
227 	HNS3_RX_PTYPE_UNUSED_ENTRY(108),
228 	HNS3_RX_PTYPE_UNUSED_ENTRY(109),
229 	HNS3_RX_PTYPE_UNUSED_ENTRY(110),
230 	HNS3_RX_PTYPE_ENTRY(111, 0, COMPLETE, IPV6),
231 	HNS3_RX_PTYPE_ENTRY(112, 0, COMPLETE, IPV6),
232 	HNS3_RX_PTYPE_ENTRY(113, 0, UNNECESSARY, IPV6),
233 	HNS3_RX_PTYPE_ENTRY(114, 0, UNNECESSARY, IPV6),
234 	HNS3_RX_PTYPE_ENTRY(115, 0, NONE, IPV6),
235 	HNS3_RX_PTYPE_ENTRY(116, 0, UNNECESSARY, IPV6),
236 	HNS3_RX_PTYPE_ENTRY(117, 0, NONE, IPV6),
237 	HNS3_RX_PTYPE_ENTRY(118, 0, NONE, IPV6),
238 	HNS3_RX_PTYPE_ENTRY(119, 0, UNNECESSARY, IPV6),
239 	HNS3_RX_PTYPE_UNUSED_ENTRY(120),
240 	HNS3_RX_PTYPE_UNUSED_ENTRY(121),
241 	HNS3_RX_PTYPE_UNUSED_ENTRY(122),
242 	HNS3_RX_PTYPE_ENTRY(123, 0, COMPLETE, PARSE_FAIL),
243 	HNS3_RX_PTYPE_ENTRY(124, 0, COMPLETE, PARSE_FAIL),
244 	HNS3_RX_PTYPE_ENTRY(125, 0, COMPLETE, IPV4),
245 	HNS3_RX_PTYPE_ENTRY(126, 0, COMPLETE, IPV4),
246 	HNS3_RX_PTYPE_ENTRY(127, 1, UNNECESSARY, IPV4),
247 	HNS3_RX_PTYPE_ENTRY(128, 1, UNNECESSARY, IPV4),
248 	HNS3_RX_PTYPE_ENTRY(129, 1, UNNECESSARY, IPV4),
249 	HNS3_RX_PTYPE_ENTRY(130, 0, COMPLETE, IPV4),
250 	HNS3_RX_PTYPE_ENTRY(131, 0, COMPLETE, IPV4),
251 	HNS3_RX_PTYPE_UNUSED_ENTRY(132),
252 	HNS3_RX_PTYPE_ENTRY(133, 0, COMPLETE, IPV6),
253 	HNS3_RX_PTYPE_ENTRY(134, 0, COMPLETE, IPV6),
254 	HNS3_RX_PTYPE_ENTRY(135, 1, UNNECESSARY, IPV6),
255 	HNS3_RX_PTYPE_ENTRY(136, 1, UNNECESSARY, IPV6),
256 	HNS3_RX_PTYPE_ENTRY(137, 1, UNNECESSARY, IPV6),
257 	HNS3_RX_PTYPE_ENTRY(138, 0, COMPLETE, IPV6),
258 	HNS3_RX_PTYPE_ENTRY(139, 0, COMPLETE, IPV6),
259 	HNS3_RX_PTYPE_UNUSED_ENTRY(140),
260 	HNS3_RX_PTYPE_UNUSED_ENTRY(141),
261 	HNS3_RX_PTYPE_UNUSED_ENTRY(142),
262 	HNS3_RX_PTYPE_UNUSED_ENTRY(143),
263 	HNS3_RX_PTYPE_UNUSED_ENTRY(144),
264 	HNS3_RX_PTYPE_UNUSED_ENTRY(145),
265 	HNS3_RX_PTYPE_UNUSED_ENTRY(146),
266 	HNS3_RX_PTYPE_UNUSED_ENTRY(147),
267 	HNS3_RX_PTYPE_UNUSED_ENTRY(148),
268 	HNS3_RX_PTYPE_UNUSED_ENTRY(149),
269 	HNS3_RX_PTYPE_UNUSED_ENTRY(150),
270 	HNS3_RX_PTYPE_UNUSED_ENTRY(151),
271 	HNS3_RX_PTYPE_UNUSED_ENTRY(152),
272 	HNS3_RX_PTYPE_UNUSED_ENTRY(153),
273 	HNS3_RX_PTYPE_UNUSED_ENTRY(154),
274 	HNS3_RX_PTYPE_UNUSED_ENTRY(155),
275 	HNS3_RX_PTYPE_UNUSED_ENTRY(156),
276 	HNS3_RX_PTYPE_UNUSED_ENTRY(157),
277 	HNS3_RX_PTYPE_UNUSED_ENTRY(158),
278 	HNS3_RX_PTYPE_UNUSED_ENTRY(159),
279 	HNS3_RX_PTYPE_UNUSED_ENTRY(160),
280 	HNS3_RX_PTYPE_UNUSED_ENTRY(161),
281 	HNS3_RX_PTYPE_UNUSED_ENTRY(162),
282 	HNS3_RX_PTYPE_UNUSED_ENTRY(163),
283 	HNS3_RX_PTYPE_UNUSED_ENTRY(164),
284 	HNS3_RX_PTYPE_UNUSED_ENTRY(165),
285 	HNS3_RX_PTYPE_UNUSED_ENTRY(166),
286 	HNS3_RX_PTYPE_UNUSED_ENTRY(167),
287 	HNS3_RX_PTYPE_UNUSED_ENTRY(168),
288 	HNS3_RX_PTYPE_UNUSED_ENTRY(169),
289 	HNS3_RX_PTYPE_UNUSED_ENTRY(170),
290 	HNS3_RX_PTYPE_UNUSED_ENTRY(171),
291 	HNS3_RX_PTYPE_UNUSED_ENTRY(172),
292 	HNS3_RX_PTYPE_UNUSED_ENTRY(173),
293 	HNS3_RX_PTYPE_UNUSED_ENTRY(174),
294 	HNS3_RX_PTYPE_UNUSED_ENTRY(175),
295 	HNS3_RX_PTYPE_UNUSED_ENTRY(176),
296 	HNS3_RX_PTYPE_UNUSED_ENTRY(177),
297 	HNS3_RX_PTYPE_UNUSED_ENTRY(178),
298 	HNS3_RX_PTYPE_UNUSED_ENTRY(179),
299 	HNS3_RX_PTYPE_UNUSED_ENTRY(180),
300 	HNS3_RX_PTYPE_UNUSED_ENTRY(181),
301 	HNS3_RX_PTYPE_UNUSED_ENTRY(182),
302 	HNS3_RX_PTYPE_UNUSED_ENTRY(183),
303 	HNS3_RX_PTYPE_UNUSED_ENTRY(184),
304 	HNS3_RX_PTYPE_UNUSED_ENTRY(185),
305 	HNS3_RX_PTYPE_UNUSED_ENTRY(186),
306 	HNS3_RX_PTYPE_UNUSED_ENTRY(187),
307 	HNS3_RX_PTYPE_UNUSED_ENTRY(188),
308 	HNS3_RX_PTYPE_UNUSED_ENTRY(189),
309 	HNS3_RX_PTYPE_UNUSED_ENTRY(190),
310 	HNS3_RX_PTYPE_UNUSED_ENTRY(191),
311 	HNS3_RX_PTYPE_UNUSED_ENTRY(192),
312 	HNS3_RX_PTYPE_UNUSED_ENTRY(193),
313 	HNS3_RX_PTYPE_UNUSED_ENTRY(194),
314 	HNS3_RX_PTYPE_UNUSED_ENTRY(195),
315 	HNS3_RX_PTYPE_UNUSED_ENTRY(196),
316 	HNS3_RX_PTYPE_UNUSED_ENTRY(197),
317 	HNS3_RX_PTYPE_UNUSED_ENTRY(198),
318 	HNS3_RX_PTYPE_UNUSED_ENTRY(199),
319 	HNS3_RX_PTYPE_UNUSED_ENTRY(200),
320 	HNS3_RX_PTYPE_UNUSED_ENTRY(201),
321 	HNS3_RX_PTYPE_UNUSED_ENTRY(202),
322 	HNS3_RX_PTYPE_UNUSED_ENTRY(203),
323 	HNS3_RX_PTYPE_UNUSED_ENTRY(204),
324 	HNS3_RX_PTYPE_UNUSED_ENTRY(205),
325 	HNS3_RX_PTYPE_UNUSED_ENTRY(206),
326 	HNS3_RX_PTYPE_UNUSED_ENTRY(207),
327 	HNS3_RX_PTYPE_UNUSED_ENTRY(208),
328 	HNS3_RX_PTYPE_UNUSED_ENTRY(209),
329 	HNS3_RX_PTYPE_UNUSED_ENTRY(210),
330 	HNS3_RX_PTYPE_UNUSED_ENTRY(211),
331 	HNS3_RX_PTYPE_UNUSED_ENTRY(212),
332 	HNS3_RX_PTYPE_UNUSED_ENTRY(213),
333 	HNS3_RX_PTYPE_UNUSED_ENTRY(214),
334 	HNS3_RX_PTYPE_UNUSED_ENTRY(215),
335 	HNS3_RX_PTYPE_UNUSED_ENTRY(216),
336 	HNS3_RX_PTYPE_UNUSED_ENTRY(217),
337 	HNS3_RX_PTYPE_UNUSED_ENTRY(218),
338 	HNS3_RX_PTYPE_UNUSED_ENTRY(219),
339 	HNS3_RX_PTYPE_UNUSED_ENTRY(220),
340 	HNS3_RX_PTYPE_UNUSED_ENTRY(221),
341 	HNS3_RX_PTYPE_UNUSED_ENTRY(222),
342 	HNS3_RX_PTYPE_UNUSED_ENTRY(223),
343 	HNS3_RX_PTYPE_UNUSED_ENTRY(224),
344 	HNS3_RX_PTYPE_UNUSED_ENTRY(225),
345 	HNS3_RX_PTYPE_UNUSED_ENTRY(226),
346 	HNS3_RX_PTYPE_UNUSED_ENTRY(227),
347 	HNS3_RX_PTYPE_UNUSED_ENTRY(228),
348 	HNS3_RX_PTYPE_UNUSED_ENTRY(229),
349 	HNS3_RX_PTYPE_UNUSED_ENTRY(230),
350 	HNS3_RX_PTYPE_UNUSED_ENTRY(231),
351 	HNS3_RX_PTYPE_UNUSED_ENTRY(232),
352 	HNS3_RX_PTYPE_UNUSED_ENTRY(233),
353 	HNS3_RX_PTYPE_UNUSED_ENTRY(234),
354 	HNS3_RX_PTYPE_UNUSED_ENTRY(235),
355 	HNS3_RX_PTYPE_UNUSED_ENTRY(236),
356 	HNS3_RX_PTYPE_UNUSED_ENTRY(237),
357 	HNS3_RX_PTYPE_UNUSED_ENTRY(238),
358 	HNS3_RX_PTYPE_UNUSED_ENTRY(239),
359 	HNS3_RX_PTYPE_UNUSED_ENTRY(240),
360 	HNS3_RX_PTYPE_UNUSED_ENTRY(241),
361 	HNS3_RX_PTYPE_UNUSED_ENTRY(242),
362 	HNS3_RX_PTYPE_UNUSED_ENTRY(243),
363 	HNS3_RX_PTYPE_UNUSED_ENTRY(244),
364 	HNS3_RX_PTYPE_UNUSED_ENTRY(245),
365 	HNS3_RX_PTYPE_UNUSED_ENTRY(246),
366 	HNS3_RX_PTYPE_UNUSED_ENTRY(247),
367 	HNS3_RX_PTYPE_UNUSED_ENTRY(248),
368 	HNS3_RX_PTYPE_UNUSED_ENTRY(249),
369 	HNS3_RX_PTYPE_UNUSED_ENTRY(250),
370 	HNS3_RX_PTYPE_UNUSED_ENTRY(251),
371 	HNS3_RX_PTYPE_UNUSED_ENTRY(252),
372 	HNS3_RX_PTYPE_UNUSED_ENTRY(253),
373 	HNS3_RX_PTYPE_UNUSED_ENTRY(254),
374 	HNS3_RX_PTYPE_UNUSED_ENTRY(255),
375 };
376 
377 #define HNS3_INVALID_PTYPE \
378 		ARRAY_SIZE(hns3_rx_ptype_tbl)
379 
380 static irqreturn_t hns3_irq_handle(int irq, void *vector)
381 {
382 	struct hns3_enet_tqp_vector *tqp_vector = vector;
383 
384 	napi_schedule_irqoff(&tqp_vector->napi);
385 	tqp_vector->event_cnt++;
386 
387 	return IRQ_HANDLED;
388 }
389 
390 static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv)
391 {
392 	struct hns3_enet_tqp_vector *tqp_vectors;
393 	unsigned int i;
394 
395 	for (i = 0; i < priv->vector_num; i++) {
396 		tqp_vectors = &priv->tqp_vector[i];
397 
398 		if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED)
399 			continue;
400 
401 		/* clear the affinity mask */
402 		irq_set_affinity_hint(tqp_vectors->vector_irq, NULL);
403 
404 		/* release the irq resource */
405 		free_irq(tqp_vectors->vector_irq, tqp_vectors);
406 		tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED;
407 	}
408 }
409 
410 static int hns3_nic_init_irq(struct hns3_nic_priv *priv)
411 {
412 	struct hns3_enet_tqp_vector *tqp_vectors;
413 	int txrx_int_idx = 0;
414 	int rx_int_idx = 0;
415 	int tx_int_idx = 0;
416 	unsigned int i;
417 	int ret;
418 
419 	for (i = 0; i < priv->vector_num; i++) {
420 		tqp_vectors = &priv->tqp_vector[i];
421 
422 		if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED)
423 			continue;
424 
425 		if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) {
426 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
427 				 "%s-%s-%s-%d", hns3_driver_name,
428 				 pci_name(priv->ae_handle->pdev),
429 				 "TxRx", txrx_int_idx++);
430 			txrx_int_idx++;
431 		} else if (tqp_vectors->rx_group.ring) {
432 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
433 				 "%s-%s-%s-%d", hns3_driver_name,
434 				 pci_name(priv->ae_handle->pdev),
435 				 "Rx", rx_int_idx++);
436 		} else if (tqp_vectors->tx_group.ring) {
437 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
438 				 "%s-%s-%s-%d", hns3_driver_name,
439 				 pci_name(priv->ae_handle->pdev),
440 				 "Tx", tx_int_idx++);
441 		} else {
442 			/* Skip this unused q_vector */
443 			continue;
444 		}
445 
446 		tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0';
447 
448 		irq_set_status_flags(tqp_vectors->vector_irq, IRQ_NOAUTOEN);
449 		ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0,
450 				  tqp_vectors->name, tqp_vectors);
451 		if (ret) {
452 			netdev_err(priv->netdev, "request irq(%d) fail\n",
453 				   tqp_vectors->vector_irq);
454 			hns3_nic_uninit_irq(priv);
455 			return ret;
456 		}
457 
458 		irq_set_affinity_hint(tqp_vectors->vector_irq,
459 				      &tqp_vectors->affinity_mask);
460 
461 		tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED;
462 	}
463 
464 	return 0;
465 }
466 
467 static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector,
468 				 u32 mask_en)
469 {
470 	writel(mask_en, tqp_vector->mask_addr);
471 }
472 
473 static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector)
474 {
475 	napi_enable(&tqp_vector->napi);
476 	enable_irq(tqp_vector->vector_irq);
477 
478 	/* enable vector */
479 	hns3_mask_vector_irq(tqp_vector, 1);
480 }
481 
482 static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector)
483 {
484 	/* disable vector */
485 	hns3_mask_vector_irq(tqp_vector, 0);
486 
487 	disable_irq(tqp_vector->vector_irq);
488 	napi_disable(&tqp_vector->napi);
489 	cancel_work_sync(&tqp_vector->rx_group.dim.work);
490 	cancel_work_sync(&tqp_vector->tx_group.dim.work);
491 }
492 
493 void hns3_set_vector_coalesce_rl(struct hns3_enet_tqp_vector *tqp_vector,
494 				 u32 rl_value)
495 {
496 	u32 rl_reg = hns3_rl_usec_to_reg(rl_value);
497 
498 	/* this defines the configuration for RL (Interrupt Rate Limiter).
499 	 * Rl defines rate of interrupts i.e. number of interrupts-per-second
500 	 * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing
501 	 */
502 	if (rl_reg > 0 && !tqp_vector->tx_group.coal.adapt_enable &&
503 	    !tqp_vector->rx_group.coal.adapt_enable)
504 		/* According to the hardware, the range of rl_reg is
505 		 * 0-59 and the unit is 4.
506 		 */
507 		rl_reg |=  HNS3_INT_RL_ENABLE_MASK;
508 
509 	writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
510 }
511 
512 void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
513 				    u32 gl_value)
514 {
515 	u32 new_val;
516 
517 	if (tqp_vector->rx_group.coal.unit_1us)
518 		new_val = gl_value | HNS3_INT_GL_1US;
519 	else
520 		new_val = hns3_gl_usec_to_reg(gl_value);
521 
522 	writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
523 }
524 
525 void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
526 				    u32 gl_value)
527 {
528 	u32 new_val;
529 
530 	if (tqp_vector->tx_group.coal.unit_1us)
531 		new_val = gl_value | HNS3_INT_GL_1US;
532 	else
533 		new_val = hns3_gl_usec_to_reg(gl_value);
534 
535 	writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
536 }
537 
538 void hns3_set_vector_coalesce_tx_ql(struct hns3_enet_tqp_vector *tqp_vector,
539 				    u32 ql_value)
540 {
541 	writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_TX_QL_OFFSET);
542 }
543 
544 void hns3_set_vector_coalesce_rx_ql(struct hns3_enet_tqp_vector *tqp_vector,
545 				    u32 ql_value)
546 {
547 	writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_RX_QL_OFFSET);
548 }
549 
550 static void hns3_vector_coalesce_init(struct hns3_enet_tqp_vector *tqp_vector,
551 				      struct hns3_nic_priv *priv)
552 {
553 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
554 	struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
555 	struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
556 	struct hns3_enet_coalesce *ptx_coal = &priv->tx_coal;
557 	struct hns3_enet_coalesce *prx_coal = &priv->rx_coal;
558 
559 	tx_coal->adapt_enable = ptx_coal->adapt_enable;
560 	rx_coal->adapt_enable = prx_coal->adapt_enable;
561 
562 	tx_coal->int_gl = ptx_coal->int_gl;
563 	rx_coal->int_gl = prx_coal->int_gl;
564 
565 	rx_coal->flow_level = prx_coal->flow_level;
566 	tx_coal->flow_level = ptx_coal->flow_level;
567 
568 	/* device version above V3(include V3), GL can configure 1us
569 	 * unit, so uses 1us unit.
570 	 */
571 	if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3) {
572 		tx_coal->unit_1us = 1;
573 		rx_coal->unit_1us = 1;
574 	}
575 
576 	if (ae_dev->dev_specs.int_ql_max) {
577 		tx_coal->ql_enable = 1;
578 		rx_coal->ql_enable = 1;
579 		tx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
580 		rx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
581 		tx_coal->int_ql = ptx_coal->int_ql;
582 		rx_coal->int_ql = prx_coal->int_ql;
583 	}
584 }
585 
586 static void
587 hns3_vector_coalesce_init_hw(struct hns3_enet_tqp_vector *tqp_vector,
588 			     struct hns3_nic_priv *priv)
589 {
590 	struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
591 	struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
592 	struct hnae3_handle *h = priv->ae_handle;
593 
594 	hns3_set_vector_coalesce_tx_gl(tqp_vector, tx_coal->int_gl);
595 	hns3_set_vector_coalesce_rx_gl(tqp_vector, rx_coal->int_gl);
596 	hns3_set_vector_coalesce_rl(tqp_vector, h->kinfo.int_rl_setting);
597 
598 	if (tx_coal->ql_enable)
599 		hns3_set_vector_coalesce_tx_ql(tqp_vector, tx_coal->int_ql);
600 
601 	if (rx_coal->ql_enable)
602 		hns3_set_vector_coalesce_rx_ql(tqp_vector, rx_coal->int_ql);
603 }
604 
605 static int hns3_nic_set_real_num_queue(struct net_device *netdev)
606 {
607 	struct hnae3_handle *h = hns3_get_handle(netdev);
608 	struct hnae3_knic_private_info *kinfo = &h->kinfo;
609 	struct hnae3_tc_info *tc_info = &kinfo->tc_info;
610 	unsigned int queue_size = kinfo->num_tqps;
611 	int i, ret;
612 
613 	if (tc_info->num_tc <= 1 && !tc_info->mqprio_active) {
614 		netdev_reset_tc(netdev);
615 	} else {
616 		ret = netdev_set_num_tc(netdev, tc_info->num_tc);
617 		if (ret) {
618 			netdev_err(netdev,
619 				   "netdev_set_num_tc fail, ret=%d!\n", ret);
620 			return ret;
621 		}
622 
623 		for (i = 0; i < tc_info->num_tc; i++)
624 			netdev_set_tc_queue(netdev, i, tc_info->tqp_count[i],
625 					    tc_info->tqp_offset[i]);
626 	}
627 
628 	ret = netif_set_real_num_tx_queues(netdev, queue_size);
629 	if (ret) {
630 		netdev_err(netdev,
631 			   "netif_set_real_num_tx_queues fail, ret=%d!\n", ret);
632 		return ret;
633 	}
634 
635 	ret = netif_set_real_num_rx_queues(netdev, queue_size);
636 	if (ret) {
637 		netdev_err(netdev,
638 			   "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
639 		return ret;
640 	}
641 
642 	return 0;
643 }
644 
645 u16 hns3_get_max_available_channels(struct hnae3_handle *h)
646 {
647 	u16 alloc_tqps, max_rss_size, rss_size;
648 
649 	h->ae_algo->ops->get_tqps_and_rss_info(h, &alloc_tqps, &max_rss_size);
650 	rss_size = alloc_tqps / h->kinfo.tc_info.num_tc;
651 
652 	return min_t(u16, rss_size, max_rss_size);
653 }
654 
655 static void hns3_tqp_enable(struct hnae3_queue *tqp)
656 {
657 	u32 rcb_reg;
658 
659 	rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
660 	rcb_reg |= BIT(HNS3_RING_EN_B);
661 	hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
662 }
663 
664 static void hns3_tqp_disable(struct hnae3_queue *tqp)
665 {
666 	u32 rcb_reg;
667 
668 	rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
669 	rcb_reg &= ~BIT(HNS3_RING_EN_B);
670 	hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
671 }
672 
673 static void hns3_free_rx_cpu_rmap(struct net_device *netdev)
674 {
675 #ifdef CONFIG_RFS_ACCEL
676 	free_irq_cpu_rmap(netdev->rx_cpu_rmap);
677 	netdev->rx_cpu_rmap = NULL;
678 #endif
679 }
680 
681 static int hns3_set_rx_cpu_rmap(struct net_device *netdev)
682 {
683 #ifdef CONFIG_RFS_ACCEL
684 	struct hns3_nic_priv *priv = netdev_priv(netdev);
685 	struct hns3_enet_tqp_vector *tqp_vector;
686 	int i, ret;
687 
688 	if (!netdev->rx_cpu_rmap) {
689 		netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->vector_num);
690 		if (!netdev->rx_cpu_rmap)
691 			return -ENOMEM;
692 	}
693 
694 	for (i = 0; i < priv->vector_num; i++) {
695 		tqp_vector = &priv->tqp_vector[i];
696 		ret = irq_cpu_rmap_add(netdev->rx_cpu_rmap,
697 				       tqp_vector->vector_irq);
698 		if (ret) {
699 			hns3_free_rx_cpu_rmap(netdev);
700 			return ret;
701 		}
702 	}
703 #endif
704 	return 0;
705 }
706 
707 static int hns3_nic_net_up(struct net_device *netdev)
708 {
709 	struct hns3_nic_priv *priv = netdev_priv(netdev);
710 	struct hnae3_handle *h = priv->ae_handle;
711 	int i, j;
712 	int ret;
713 
714 	ret = hns3_nic_reset_all_ring(h);
715 	if (ret)
716 		return ret;
717 
718 	clear_bit(HNS3_NIC_STATE_DOWN, &priv->state);
719 
720 	/* enable the vectors */
721 	for (i = 0; i < priv->vector_num; i++)
722 		hns3_vector_enable(&priv->tqp_vector[i]);
723 
724 	/* enable rcb */
725 	for (j = 0; j < h->kinfo.num_tqps; j++)
726 		hns3_tqp_enable(h->kinfo.tqp[j]);
727 
728 	/* start the ae_dev */
729 	ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0;
730 	if (ret) {
731 		set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
732 		while (j--)
733 			hns3_tqp_disable(h->kinfo.tqp[j]);
734 
735 		for (j = i - 1; j >= 0; j--)
736 			hns3_vector_disable(&priv->tqp_vector[j]);
737 	}
738 
739 	return ret;
740 }
741 
742 static void hns3_config_xps(struct hns3_nic_priv *priv)
743 {
744 	int i;
745 
746 	for (i = 0; i < priv->vector_num; i++) {
747 		struct hns3_enet_tqp_vector *tqp_vector = &priv->tqp_vector[i];
748 		struct hns3_enet_ring *ring = tqp_vector->tx_group.ring;
749 
750 		while (ring) {
751 			int ret;
752 
753 			ret = netif_set_xps_queue(priv->netdev,
754 						  &tqp_vector->affinity_mask,
755 						  ring->tqp->tqp_index);
756 			if (ret)
757 				netdev_warn(priv->netdev,
758 					    "set xps queue failed: %d", ret);
759 
760 			ring = ring->next;
761 		}
762 	}
763 }
764 
765 static int hns3_nic_net_open(struct net_device *netdev)
766 {
767 	struct hns3_nic_priv *priv = netdev_priv(netdev);
768 	struct hnae3_handle *h = hns3_get_handle(netdev);
769 	struct hnae3_knic_private_info *kinfo;
770 	int i, ret;
771 
772 	if (hns3_nic_resetting(netdev))
773 		return -EBUSY;
774 
775 	if (!test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
776 		netdev_warn(netdev, "net open repeatedly!\n");
777 		return 0;
778 	}
779 
780 	netif_carrier_off(netdev);
781 
782 	ret = hns3_nic_set_real_num_queue(netdev);
783 	if (ret)
784 		return ret;
785 
786 	ret = hns3_nic_net_up(netdev);
787 	if (ret) {
788 		netdev_err(netdev, "net up fail, ret=%d!\n", ret);
789 		return ret;
790 	}
791 
792 	kinfo = &h->kinfo;
793 	for (i = 0; i < HNAE3_MAX_USER_PRIO; i++)
794 		netdev_set_prio_tc_map(netdev, i, kinfo->tc_info.prio_tc[i]);
795 
796 	if (h->ae_algo->ops->set_timer_task)
797 		h->ae_algo->ops->set_timer_task(priv->ae_handle, true);
798 
799 	hns3_config_xps(priv);
800 
801 	netif_dbg(h, drv, netdev, "net open\n");
802 
803 	return 0;
804 }
805 
806 static void hns3_reset_tx_queue(struct hnae3_handle *h)
807 {
808 	struct net_device *ndev = h->kinfo.netdev;
809 	struct hns3_nic_priv *priv = netdev_priv(ndev);
810 	struct netdev_queue *dev_queue;
811 	u32 i;
812 
813 	for (i = 0; i < h->kinfo.num_tqps; i++) {
814 		dev_queue = netdev_get_tx_queue(ndev,
815 						priv->ring[i].queue_index);
816 		netdev_tx_reset_queue(dev_queue);
817 	}
818 }
819 
820 static void hns3_nic_net_down(struct net_device *netdev)
821 {
822 	struct hns3_nic_priv *priv = netdev_priv(netdev);
823 	struct hnae3_handle *h = hns3_get_handle(netdev);
824 	const struct hnae3_ae_ops *ops;
825 	int i;
826 
827 	/* disable vectors */
828 	for (i = 0; i < priv->vector_num; i++)
829 		hns3_vector_disable(&priv->tqp_vector[i]);
830 
831 	/* disable rcb */
832 	for (i = 0; i < h->kinfo.num_tqps; i++)
833 		hns3_tqp_disable(h->kinfo.tqp[i]);
834 
835 	/* stop ae_dev */
836 	ops = priv->ae_handle->ae_algo->ops;
837 	if (ops->stop)
838 		ops->stop(priv->ae_handle);
839 
840 	/* delay ring buffer clearing to hns3_reset_notify_uninit_enet
841 	 * during reset process, because driver may not be able
842 	 * to disable the ring through firmware when downing the netdev.
843 	 */
844 	if (!hns3_nic_resetting(netdev))
845 		hns3_clear_all_ring(priv->ae_handle, false);
846 
847 	hns3_reset_tx_queue(priv->ae_handle);
848 }
849 
850 static int hns3_nic_net_stop(struct net_device *netdev)
851 {
852 	struct hns3_nic_priv *priv = netdev_priv(netdev);
853 	struct hnae3_handle *h = hns3_get_handle(netdev);
854 
855 	if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
856 		return 0;
857 
858 	netif_dbg(h, drv, netdev, "net stop\n");
859 
860 	if (h->ae_algo->ops->set_timer_task)
861 		h->ae_algo->ops->set_timer_task(priv->ae_handle, false);
862 
863 	netif_carrier_off(netdev);
864 	netif_tx_disable(netdev);
865 
866 	hns3_nic_net_down(netdev);
867 
868 	return 0;
869 }
870 
871 static int hns3_nic_uc_sync(struct net_device *netdev,
872 			    const unsigned char *addr)
873 {
874 	struct hnae3_handle *h = hns3_get_handle(netdev);
875 
876 	if (h->ae_algo->ops->add_uc_addr)
877 		return h->ae_algo->ops->add_uc_addr(h, addr);
878 
879 	return 0;
880 }
881 
882 static int hns3_nic_uc_unsync(struct net_device *netdev,
883 			      const unsigned char *addr)
884 {
885 	struct hnae3_handle *h = hns3_get_handle(netdev);
886 
887 	/* need ignore the request of removing device address, because
888 	 * we store the device address and other addresses of uc list
889 	 * in the function's mac filter list.
890 	 */
891 	if (ether_addr_equal(addr, netdev->dev_addr))
892 		return 0;
893 
894 	if (h->ae_algo->ops->rm_uc_addr)
895 		return h->ae_algo->ops->rm_uc_addr(h, addr);
896 
897 	return 0;
898 }
899 
900 static int hns3_nic_mc_sync(struct net_device *netdev,
901 			    const unsigned char *addr)
902 {
903 	struct hnae3_handle *h = hns3_get_handle(netdev);
904 
905 	if (h->ae_algo->ops->add_mc_addr)
906 		return h->ae_algo->ops->add_mc_addr(h, addr);
907 
908 	return 0;
909 }
910 
911 static int hns3_nic_mc_unsync(struct net_device *netdev,
912 			      const unsigned char *addr)
913 {
914 	struct hnae3_handle *h = hns3_get_handle(netdev);
915 
916 	if (h->ae_algo->ops->rm_mc_addr)
917 		return h->ae_algo->ops->rm_mc_addr(h, addr);
918 
919 	return 0;
920 }
921 
922 static u8 hns3_get_netdev_flags(struct net_device *netdev)
923 {
924 	u8 flags = 0;
925 
926 	if (netdev->flags & IFF_PROMISC)
927 		flags = HNAE3_USER_UPE | HNAE3_USER_MPE | HNAE3_BPE;
928 	else if (netdev->flags & IFF_ALLMULTI)
929 		flags = HNAE3_USER_MPE;
930 
931 	return flags;
932 }
933 
934 static void hns3_nic_set_rx_mode(struct net_device *netdev)
935 {
936 	struct hnae3_handle *h = hns3_get_handle(netdev);
937 	u8 new_flags;
938 
939 	new_flags = hns3_get_netdev_flags(netdev);
940 
941 	__dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync);
942 	__dev_mc_sync(netdev, hns3_nic_mc_sync, hns3_nic_mc_unsync);
943 
944 	/* User mode Promisc mode enable and vlan filtering is disabled to
945 	 * let all packets in.
946 	 */
947 	h->netdev_flags = new_flags;
948 	hns3_request_update_promisc_mode(h);
949 }
950 
951 void hns3_request_update_promisc_mode(struct hnae3_handle *handle)
952 {
953 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
954 
955 	if (ops->request_update_promisc_mode)
956 		ops->request_update_promisc_mode(handle);
957 }
958 
959 static u32 hns3_tx_spare_space(struct hns3_enet_ring *ring)
960 {
961 	struct hns3_tx_spare *tx_spare = ring->tx_spare;
962 	u32 ntc, ntu;
963 
964 	/* This smp_load_acquire() pairs with smp_store_release() in
965 	 * hns3_tx_spare_update() called in tx desc cleaning process.
966 	 */
967 	ntc = smp_load_acquire(&tx_spare->last_to_clean);
968 	ntu = tx_spare->next_to_use;
969 
970 	if (ntc > ntu)
971 		return ntc - ntu - 1;
972 
973 	/* The free tx buffer is divided into two part, so pick the
974 	 * larger one.
975 	 */
976 	return max(ntc, tx_spare->len - ntu) - 1;
977 }
978 
979 static void hns3_tx_spare_update(struct hns3_enet_ring *ring)
980 {
981 	struct hns3_tx_spare *tx_spare = ring->tx_spare;
982 
983 	if (!tx_spare ||
984 	    tx_spare->last_to_clean == tx_spare->next_to_clean)
985 		return;
986 
987 	/* This smp_store_release() pairs with smp_load_acquire() in
988 	 * hns3_tx_spare_space() called in xmit process.
989 	 */
990 	smp_store_release(&tx_spare->last_to_clean,
991 			  tx_spare->next_to_clean);
992 }
993 
994 static bool hns3_can_use_tx_bounce(struct hns3_enet_ring *ring,
995 				   struct sk_buff *skb,
996 				   u32 space)
997 {
998 	u32 len = skb->len <= ring->tx_copybreak ? skb->len :
999 				skb_headlen(skb);
1000 
1001 	if (len > ring->tx_copybreak)
1002 		return false;
1003 
1004 	if (ALIGN(len, dma_get_cache_alignment()) > space) {
1005 		hns3_ring_stats_update(ring, tx_spare_full);
1006 		return false;
1007 	}
1008 
1009 	return true;
1010 }
1011 
1012 static bool hns3_can_use_tx_sgl(struct hns3_enet_ring *ring,
1013 				struct sk_buff *skb,
1014 				u32 space)
1015 {
1016 	if (skb->len <= ring->tx_copybreak || !tx_sgl ||
1017 	    (!skb_has_frag_list(skb) &&
1018 	     skb_shinfo(skb)->nr_frags < tx_sgl))
1019 		return false;
1020 
1021 	if (space < HNS3_MAX_SGL_SIZE) {
1022 		hns3_ring_stats_update(ring, tx_spare_full);
1023 		return false;
1024 	}
1025 
1026 	return true;
1027 }
1028 
1029 static void hns3_init_tx_spare_buffer(struct hns3_enet_ring *ring)
1030 {
1031 	struct hns3_tx_spare *tx_spare;
1032 	struct page *page;
1033 	u32 alloc_size;
1034 	dma_addr_t dma;
1035 	int order;
1036 
1037 	alloc_size = ring->tqp->handle->kinfo.tx_spare_buf_size;
1038 	if (!alloc_size)
1039 		return;
1040 
1041 	order = get_order(alloc_size);
1042 	tx_spare = devm_kzalloc(ring_to_dev(ring), sizeof(*tx_spare),
1043 				GFP_KERNEL);
1044 	if (!tx_spare) {
1045 		/* The driver still work without the tx spare buffer */
1046 		dev_warn(ring_to_dev(ring), "failed to allocate hns3_tx_spare\n");
1047 		return;
1048 	}
1049 
1050 	page = alloc_pages_node(dev_to_node(ring_to_dev(ring)),
1051 				GFP_KERNEL, order);
1052 	if (!page) {
1053 		dev_warn(ring_to_dev(ring), "failed to allocate tx spare pages\n");
1054 		devm_kfree(ring_to_dev(ring), tx_spare);
1055 		return;
1056 	}
1057 
1058 	dma = dma_map_page(ring_to_dev(ring), page, 0,
1059 			   PAGE_SIZE << order, DMA_TO_DEVICE);
1060 	if (dma_mapping_error(ring_to_dev(ring), dma)) {
1061 		dev_warn(ring_to_dev(ring), "failed to map pages for tx spare\n");
1062 		put_page(page);
1063 		devm_kfree(ring_to_dev(ring), tx_spare);
1064 		return;
1065 	}
1066 
1067 	tx_spare->dma = dma;
1068 	tx_spare->buf = page_address(page);
1069 	tx_spare->len = PAGE_SIZE << order;
1070 	ring->tx_spare = tx_spare;
1071 }
1072 
1073 /* Use hns3_tx_spare_space() to make sure there is enough buffer
1074  * before calling below function to allocate tx buffer.
1075  */
1076 static void *hns3_tx_spare_alloc(struct hns3_enet_ring *ring,
1077 				 unsigned int size, dma_addr_t *dma,
1078 				 u32 *cb_len)
1079 {
1080 	struct hns3_tx_spare *tx_spare = ring->tx_spare;
1081 	u32 ntu = tx_spare->next_to_use;
1082 
1083 	size = ALIGN(size, dma_get_cache_alignment());
1084 	*cb_len = size;
1085 
1086 	/* Tx spare buffer wraps back here because the end of
1087 	 * freed tx buffer is not enough.
1088 	 */
1089 	if (ntu + size > tx_spare->len) {
1090 		*cb_len += (tx_spare->len - ntu);
1091 		ntu = 0;
1092 	}
1093 
1094 	tx_spare->next_to_use = ntu + size;
1095 	if (tx_spare->next_to_use == tx_spare->len)
1096 		tx_spare->next_to_use = 0;
1097 
1098 	*dma = tx_spare->dma + ntu;
1099 
1100 	return tx_spare->buf + ntu;
1101 }
1102 
1103 static void hns3_tx_spare_rollback(struct hns3_enet_ring *ring, u32 len)
1104 {
1105 	struct hns3_tx_spare *tx_spare = ring->tx_spare;
1106 
1107 	if (len > tx_spare->next_to_use) {
1108 		len -= tx_spare->next_to_use;
1109 		tx_spare->next_to_use = tx_spare->len - len;
1110 	} else {
1111 		tx_spare->next_to_use -= len;
1112 	}
1113 }
1114 
1115 static void hns3_tx_spare_reclaim_cb(struct hns3_enet_ring *ring,
1116 				     struct hns3_desc_cb *cb)
1117 {
1118 	struct hns3_tx_spare *tx_spare = ring->tx_spare;
1119 	u32 ntc = tx_spare->next_to_clean;
1120 	u32 len = cb->length;
1121 
1122 	tx_spare->next_to_clean += len;
1123 
1124 	if (tx_spare->next_to_clean >= tx_spare->len) {
1125 		tx_spare->next_to_clean -= tx_spare->len;
1126 
1127 		if (tx_spare->next_to_clean) {
1128 			ntc = 0;
1129 			len = tx_spare->next_to_clean;
1130 		}
1131 	}
1132 
1133 	/* This tx spare buffer is only really reclaimed after calling
1134 	 * hns3_tx_spare_update(), so it is still safe to use the info in
1135 	 * the tx buffer to do the dma sync or sg unmapping after
1136 	 * tx_spare->next_to_clean is moved forword.
1137 	 */
1138 	if (cb->type & (DESC_TYPE_BOUNCE_HEAD | DESC_TYPE_BOUNCE_ALL)) {
1139 		dma_addr_t dma = tx_spare->dma + ntc;
1140 
1141 		dma_sync_single_for_cpu(ring_to_dev(ring), dma, len,
1142 					DMA_TO_DEVICE);
1143 	} else {
1144 		struct sg_table *sgt = tx_spare->buf + ntc;
1145 
1146 		dma_unmap_sg(ring_to_dev(ring), sgt->sgl, sgt->orig_nents,
1147 			     DMA_TO_DEVICE);
1148 	}
1149 }
1150 
1151 static int hns3_set_tso(struct sk_buff *skb, u32 *paylen_fdop_ol4cs,
1152 			u16 *mss, u32 *type_cs_vlan_tso, u32 *send_bytes)
1153 {
1154 	u32 l4_offset, hdr_len;
1155 	union l3_hdr_info l3;
1156 	union l4_hdr_info l4;
1157 	u32 l4_paylen;
1158 	int ret;
1159 
1160 	if (!skb_is_gso(skb))
1161 		return 0;
1162 
1163 	ret = skb_cow_head(skb, 0);
1164 	if (unlikely(ret < 0))
1165 		return ret;
1166 
1167 	l3.hdr = skb_network_header(skb);
1168 	l4.hdr = skb_transport_header(skb);
1169 
1170 	/* Software should clear the IPv4's checksum field when tso is
1171 	 * needed.
1172 	 */
1173 	if (l3.v4->version == 4)
1174 		l3.v4->check = 0;
1175 
1176 	/* tunnel packet */
1177 	if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
1178 					 SKB_GSO_GRE_CSUM |
1179 					 SKB_GSO_UDP_TUNNEL |
1180 					 SKB_GSO_UDP_TUNNEL_CSUM)) {
1181 		/* reset l3&l4 pointers from outer to inner headers */
1182 		l3.hdr = skb_inner_network_header(skb);
1183 		l4.hdr = skb_inner_transport_header(skb);
1184 
1185 		/* Software should clear the IPv4's checksum field when
1186 		 * tso is needed.
1187 		 */
1188 		if (l3.v4->version == 4)
1189 			l3.v4->check = 0;
1190 	}
1191 
1192 	/* normal or tunnel packet */
1193 	l4_offset = l4.hdr - skb->data;
1194 
1195 	/* remove payload length from inner pseudo checksum when tso */
1196 	l4_paylen = skb->len - l4_offset;
1197 
1198 	if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) {
1199 		hdr_len = sizeof(*l4.udp) + l4_offset;
1200 		csum_replace_by_diff(&l4.udp->check,
1201 				     (__force __wsum)htonl(l4_paylen));
1202 	} else {
1203 		hdr_len = (l4.tcp->doff << 2) + l4_offset;
1204 		csum_replace_by_diff(&l4.tcp->check,
1205 				     (__force __wsum)htonl(l4_paylen));
1206 	}
1207 
1208 	*send_bytes = (skb_shinfo(skb)->gso_segs - 1) * hdr_len + skb->len;
1209 
1210 	/* find the txbd field values */
1211 	*paylen_fdop_ol4cs = skb->len - hdr_len;
1212 	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_TSO_B, 1);
1213 
1214 	/* offload outer UDP header checksum */
1215 	if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)
1216 		hns3_set_field(*paylen_fdop_ol4cs, HNS3_TXD_OL4CS_B, 1);
1217 
1218 	/* get MSS for TSO */
1219 	*mss = skb_shinfo(skb)->gso_size;
1220 
1221 	trace_hns3_tso(skb);
1222 
1223 	return 0;
1224 }
1225 
1226 static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
1227 				u8 *il4_proto)
1228 {
1229 	union l3_hdr_info l3;
1230 	unsigned char *l4_hdr;
1231 	unsigned char *exthdr;
1232 	u8 l4_proto_tmp;
1233 	__be16 frag_off;
1234 
1235 	/* find outer header point */
1236 	l3.hdr = skb_network_header(skb);
1237 	l4_hdr = skb_transport_header(skb);
1238 
1239 	if (skb->protocol == htons(ETH_P_IPV6)) {
1240 		exthdr = l3.hdr + sizeof(*l3.v6);
1241 		l4_proto_tmp = l3.v6->nexthdr;
1242 		if (l4_hdr != exthdr)
1243 			ipv6_skip_exthdr(skb, exthdr - skb->data,
1244 					 &l4_proto_tmp, &frag_off);
1245 	} else if (skb->protocol == htons(ETH_P_IP)) {
1246 		l4_proto_tmp = l3.v4->protocol;
1247 	} else {
1248 		return -EINVAL;
1249 	}
1250 
1251 	*ol4_proto = l4_proto_tmp;
1252 
1253 	/* tunnel packet */
1254 	if (!skb->encapsulation) {
1255 		*il4_proto = 0;
1256 		return 0;
1257 	}
1258 
1259 	/* find inner header point */
1260 	l3.hdr = skb_inner_network_header(skb);
1261 	l4_hdr = skb_inner_transport_header(skb);
1262 
1263 	if (l3.v6->version == 6) {
1264 		exthdr = l3.hdr + sizeof(*l3.v6);
1265 		l4_proto_tmp = l3.v6->nexthdr;
1266 		if (l4_hdr != exthdr)
1267 			ipv6_skip_exthdr(skb, exthdr - skb->data,
1268 					 &l4_proto_tmp, &frag_off);
1269 	} else if (l3.v4->version == 4) {
1270 		l4_proto_tmp = l3.v4->protocol;
1271 	}
1272 
1273 	*il4_proto = l4_proto_tmp;
1274 
1275 	return 0;
1276 }
1277 
1278 /* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
1279  * and it is udp packet, which has a dest port as the IANA assigned.
1280  * the hardware is expected to do the checksum offload, but the
1281  * hardware will not do the checksum offload when udp dest port is
1282  * 4789, 4790 or 6081.
1283  */
1284 static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
1285 {
1286 	struct hns3_nic_priv *priv = netdev_priv(skb->dev);
1287 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
1288 	union l4_hdr_info l4;
1289 
1290 	/* device version above V3(include V3), the hardware can
1291 	 * do this checksum offload.
1292 	 */
1293 	if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
1294 		return false;
1295 
1296 	l4.hdr = skb_transport_header(skb);
1297 
1298 	if (!(!skb->encapsulation &&
1299 	      (l4.udp->dest == htons(IANA_VXLAN_UDP_PORT) ||
1300 	      l4.udp->dest == htons(GENEVE_UDP_PORT) ||
1301 	      l4.udp->dest == htons(IANA_VXLAN_GPE_UDP_PORT))))
1302 		return false;
1303 
1304 	return true;
1305 }
1306 
1307 static void hns3_set_outer_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
1308 				  u32 *ol_type_vlan_len_msec)
1309 {
1310 	u32 l2_len, l3_len, l4_len;
1311 	unsigned char *il2_hdr;
1312 	union l3_hdr_info l3;
1313 	union l4_hdr_info l4;
1314 
1315 	l3.hdr = skb_network_header(skb);
1316 	l4.hdr = skb_transport_header(skb);
1317 
1318 	/* compute OL2 header size, defined in 2 Bytes */
1319 	l2_len = l3.hdr - skb->data;
1320 	hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L2LEN_S, l2_len >> 1);
1321 
1322 	/* compute OL3 header size, defined in 4 Bytes */
1323 	l3_len = l4.hdr - l3.hdr;
1324 	hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_S, l3_len >> 2);
1325 
1326 	il2_hdr = skb_inner_mac_header(skb);
1327 	/* compute OL4 header size, defined in 4 Bytes */
1328 	l4_len = il2_hdr - l4.hdr;
1329 	hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L4LEN_S, l4_len >> 2);
1330 
1331 	/* define outer network header type */
1332 	if (skb->protocol == htons(ETH_P_IP)) {
1333 		if (skb_is_gso(skb))
1334 			hns3_set_field(*ol_type_vlan_len_msec,
1335 				       HNS3_TXD_OL3T_S,
1336 				       HNS3_OL3T_IPV4_CSUM);
1337 		else
1338 			hns3_set_field(*ol_type_vlan_len_msec,
1339 				       HNS3_TXD_OL3T_S,
1340 				       HNS3_OL3T_IPV4_NO_CSUM);
1341 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
1342 		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_S,
1343 			       HNS3_OL3T_IPV6);
1344 	}
1345 
1346 	if (ol4_proto == IPPROTO_UDP)
1347 		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
1348 			       HNS3_TUN_MAC_IN_UDP);
1349 	else if (ol4_proto == IPPROTO_GRE)
1350 		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
1351 			       HNS3_TUN_NVGRE);
1352 }
1353 
1354 static void hns3_set_l3_type(struct sk_buff *skb, union l3_hdr_info l3,
1355 			     u32 *type_cs_vlan_tso)
1356 {
1357 	if (l3.v4->version == 4) {
1358 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
1359 			       HNS3_L3T_IPV4);
1360 
1361 		/* the stack computes the IP header already, the only time we
1362 		 * need the hardware to recompute it is in the case of TSO.
1363 		 */
1364 		if (skb_is_gso(skb))
1365 			hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3CS_B, 1);
1366 	} else if (l3.v6->version == 6) {
1367 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
1368 			       HNS3_L3T_IPV6);
1369 	}
1370 }
1371 
1372 static int hns3_set_l4_csum_length(struct sk_buff *skb, union l4_hdr_info l4,
1373 				   u32 l4_proto, u32 *type_cs_vlan_tso)
1374 {
1375 	/* compute inner(/normal) L4 header size, defined in 4 Bytes */
1376 	switch (l4_proto) {
1377 	case IPPROTO_TCP:
1378 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1379 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1380 			       HNS3_L4T_TCP);
1381 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1382 			       l4.tcp->doff);
1383 		break;
1384 	case IPPROTO_UDP:
1385 		if (hns3_tunnel_csum_bug(skb)) {
1386 			int ret = skb_put_padto(skb, HNS3_MIN_TUN_PKT_LEN);
1387 
1388 			return ret ? ret : skb_checksum_help(skb);
1389 		}
1390 
1391 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1392 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1393 			       HNS3_L4T_UDP);
1394 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1395 			       (sizeof(struct udphdr) >> 2));
1396 		break;
1397 	case IPPROTO_SCTP:
1398 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1399 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1400 			       HNS3_L4T_SCTP);
1401 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1402 			       (sizeof(struct sctphdr) >> 2));
1403 		break;
1404 	default:
1405 		/* drop the skb tunnel packet if hardware don't support,
1406 		 * because hardware can't calculate csum when TSO.
1407 		 */
1408 		if (skb_is_gso(skb))
1409 			return -EDOM;
1410 
1411 		/* the stack computes the IP header already,
1412 		 * driver calculate l4 checksum when not TSO.
1413 		 */
1414 		return skb_checksum_help(skb);
1415 	}
1416 
1417 	return 0;
1418 }
1419 
1420 static int hns3_set_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
1421 			   u8 il4_proto, u32 *type_cs_vlan_tso,
1422 			   u32 *ol_type_vlan_len_msec)
1423 {
1424 	unsigned char *l2_hdr = skb->data;
1425 	u32 l4_proto = ol4_proto;
1426 	union l4_hdr_info l4;
1427 	union l3_hdr_info l3;
1428 	u32 l2_len, l3_len;
1429 
1430 	l4.hdr = skb_transport_header(skb);
1431 	l3.hdr = skb_network_header(skb);
1432 
1433 	/* handle encapsulation skb */
1434 	if (skb->encapsulation) {
1435 		/* If this is a not UDP/GRE encapsulation skb */
1436 		if (!(ol4_proto == IPPROTO_UDP || ol4_proto == IPPROTO_GRE)) {
1437 			/* drop the skb tunnel packet if hardware don't support,
1438 			 * because hardware can't calculate csum when TSO.
1439 			 */
1440 			if (skb_is_gso(skb))
1441 				return -EDOM;
1442 
1443 			/* the stack computes the IP header already,
1444 			 * driver calculate l4 checksum when not TSO.
1445 			 */
1446 			return skb_checksum_help(skb);
1447 		}
1448 
1449 		hns3_set_outer_l2l3l4(skb, ol4_proto, ol_type_vlan_len_msec);
1450 
1451 		/* switch to inner header */
1452 		l2_hdr = skb_inner_mac_header(skb);
1453 		l3.hdr = skb_inner_network_header(skb);
1454 		l4.hdr = skb_inner_transport_header(skb);
1455 		l4_proto = il4_proto;
1456 	}
1457 
1458 	hns3_set_l3_type(skb, l3, type_cs_vlan_tso);
1459 
1460 	/* compute inner(/normal) L2 header size, defined in 2 Bytes */
1461 	l2_len = l3.hdr - l2_hdr;
1462 	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S, l2_len >> 1);
1463 
1464 	/* compute inner(/normal) L3 header size, defined in 4 Bytes */
1465 	l3_len = l4.hdr - l3.hdr;
1466 	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_S, l3_len >> 2);
1467 
1468 	return hns3_set_l4_csum_length(skb, l4, l4_proto, type_cs_vlan_tso);
1469 }
1470 
1471 static int hns3_handle_vtags(struct hns3_enet_ring *tx_ring,
1472 			     struct sk_buff *skb)
1473 {
1474 	struct hnae3_handle *handle = tx_ring->tqp->handle;
1475 	struct hnae3_ae_dev *ae_dev;
1476 	struct vlan_ethhdr *vhdr;
1477 	int rc;
1478 
1479 	if (!(skb->protocol == htons(ETH_P_8021Q) ||
1480 	      skb_vlan_tag_present(skb)))
1481 		return 0;
1482 
1483 	/* For HW limitation on HNAE3_DEVICE_VERSION_V2, if port based insert
1484 	 * VLAN enabled, only one VLAN header is allowed in skb, otherwise it
1485 	 * will cause RAS error.
1486 	 */
1487 	ae_dev = pci_get_drvdata(handle->pdev);
1488 	if (unlikely(skb_vlan_tagged_multi(skb) &&
1489 		     ae_dev->dev_version <= HNAE3_DEVICE_VERSION_V2 &&
1490 		     handle->port_base_vlan_state ==
1491 		     HNAE3_PORT_BASE_VLAN_ENABLE))
1492 		return -EINVAL;
1493 
1494 	if (skb->protocol == htons(ETH_P_8021Q) &&
1495 	    !(handle->kinfo.netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
1496 		/* When HW VLAN acceleration is turned off, and the stack
1497 		 * sets the protocol to 802.1q, the driver just need to
1498 		 * set the protocol to the encapsulated ethertype.
1499 		 */
1500 		skb->protocol = vlan_get_protocol(skb);
1501 		return 0;
1502 	}
1503 
1504 	if (skb_vlan_tag_present(skb)) {
1505 		/* Based on hw strategy, use out_vtag in two layer tag case,
1506 		 * and use inner_vtag in one tag case.
1507 		 */
1508 		if (skb->protocol == htons(ETH_P_8021Q) &&
1509 		    handle->port_base_vlan_state ==
1510 		    HNAE3_PORT_BASE_VLAN_DISABLE)
1511 			rc = HNS3_OUTER_VLAN_TAG;
1512 		else
1513 			rc = HNS3_INNER_VLAN_TAG;
1514 
1515 		skb->protocol = vlan_get_protocol(skb);
1516 		return rc;
1517 	}
1518 
1519 	rc = skb_cow_head(skb, 0);
1520 	if (unlikely(rc < 0))
1521 		return rc;
1522 
1523 	vhdr = (struct vlan_ethhdr *)skb->data;
1524 	vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority << VLAN_PRIO_SHIFT)
1525 					 & VLAN_PRIO_MASK);
1526 
1527 	skb->protocol = vlan_get_protocol(skb);
1528 	return 0;
1529 }
1530 
1531 /* check if the hardware is capable of checksum offloading */
1532 static bool hns3_check_hw_tx_csum(struct sk_buff *skb)
1533 {
1534 	struct hns3_nic_priv *priv = netdev_priv(skb->dev);
1535 
1536 	/* Kindly note, due to backward compatibility of the TX descriptor,
1537 	 * HW checksum of the non-IP packets and GSO packets is handled at
1538 	 * different place in the following code
1539 	 */
1540 	if (skb_csum_is_sctp(skb) || skb_is_gso(skb) ||
1541 	    !test_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state))
1542 		return false;
1543 
1544 	return true;
1545 }
1546 
1547 struct hns3_desc_param {
1548 	u32 paylen_ol4cs;
1549 	u32 ol_type_vlan_len_msec;
1550 	u32 type_cs_vlan_tso;
1551 	u16 mss_hw_csum;
1552 	u16 inner_vtag;
1553 	u16 out_vtag;
1554 };
1555 
1556 static void hns3_init_desc_data(struct sk_buff *skb, struct hns3_desc_param *pa)
1557 {
1558 	pa->paylen_ol4cs = skb->len;
1559 	pa->ol_type_vlan_len_msec = 0;
1560 	pa->type_cs_vlan_tso = 0;
1561 	pa->mss_hw_csum = 0;
1562 	pa->inner_vtag = 0;
1563 	pa->out_vtag = 0;
1564 }
1565 
1566 static int hns3_handle_vlan_info(struct hns3_enet_ring *ring,
1567 				 struct sk_buff *skb,
1568 				 struct hns3_desc_param *param)
1569 {
1570 	int ret;
1571 
1572 	ret = hns3_handle_vtags(ring, skb);
1573 	if (unlikely(ret < 0)) {
1574 		hns3_ring_stats_update(ring, tx_vlan_err);
1575 		return ret;
1576 	} else if (ret == HNS3_INNER_VLAN_TAG) {
1577 		param->inner_vtag = skb_vlan_tag_get(skb);
1578 		param->inner_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1579 				VLAN_PRIO_MASK;
1580 		hns3_set_field(param->type_cs_vlan_tso, HNS3_TXD_VLAN_B, 1);
1581 	} else if (ret == HNS3_OUTER_VLAN_TAG) {
1582 		param->out_vtag = skb_vlan_tag_get(skb);
1583 		param->out_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1584 				VLAN_PRIO_MASK;
1585 		hns3_set_field(param->ol_type_vlan_len_msec, HNS3_TXD_OVLAN_B,
1586 			       1);
1587 	}
1588 	return 0;
1589 }
1590 
1591 static int hns3_handle_csum_partial(struct hns3_enet_ring *ring,
1592 				    struct sk_buff *skb,
1593 				    struct hns3_desc_cb *desc_cb,
1594 				    struct hns3_desc_param *param)
1595 {
1596 	u8 ol4_proto, il4_proto;
1597 	int ret;
1598 
1599 	if (hns3_check_hw_tx_csum(skb)) {
1600 		/* set checksum start and offset, defined in 2 Bytes */
1601 		hns3_set_field(param->type_cs_vlan_tso, HNS3_TXD_CSUM_START_S,
1602 			       skb_checksum_start_offset(skb) >> 1);
1603 		hns3_set_field(param->ol_type_vlan_len_msec,
1604 			       HNS3_TXD_CSUM_OFFSET_S,
1605 			       skb->csum_offset >> 1);
1606 		param->mss_hw_csum |= BIT(HNS3_TXD_HW_CS_B);
1607 		return 0;
1608 	}
1609 
1610 	skb_reset_mac_len(skb);
1611 
1612 	ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
1613 	if (unlikely(ret < 0)) {
1614 		hns3_ring_stats_update(ring, tx_l4_proto_err);
1615 		return ret;
1616 	}
1617 
1618 	ret = hns3_set_l2l3l4(skb, ol4_proto, il4_proto,
1619 			      &param->type_cs_vlan_tso,
1620 			      &param->ol_type_vlan_len_msec);
1621 	if (unlikely(ret < 0)) {
1622 		hns3_ring_stats_update(ring, tx_l2l3l4_err);
1623 		return ret;
1624 	}
1625 
1626 	ret = hns3_set_tso(skb, &param->paylen_ol4cs, &param->mss_hw_csum,
1627 			   &param->type_cs_vlan_tso, &desc_cb->send_bytes);
1628 	if (unlikely(ret < 0)) {
1629 		hns3_ring_stats_update(ring, tx_tso_err);
1630 		return ret;
1631 	}
1632 	return 0;
1633 }
1634 
1635 static int hns3_fill_skb_desc(struct hns3_enet_ring *ring,
1636 			      struct sk_buff *skb, struct hns3_desc *desc,
1637 			      struct hns3_desc_cb *desc_cb)
1638 {
1639 	struct hns3_desc_param param;
1640 	int ret;
1641 
1642 	hns3_init_desc_data(skb, &param);
1643 	ret = hns3_handle_vlan_info(ring, skb, &param);
1644 	if (unlikely(ret < 0))
1645 		return ret;
1646 
1647 	desc_cb->send_bytes = skb->len;
1648 
1649 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1650 		ret = hns3_handle_csum_partial(ring, skb, desc_cb, &param);
1651 		if (ret)
1652 			return ret;
1653 	}
1654 
1655 	/* Set txbd */
1656 	desc->tx.ol_type_vlan_len_msec =
1657 		cpu_to_le32(param.ol_type_vlan_len_msec);
1658 	desc->tx.type_cs_vlan_tso_len = cpu_to_le32(param.type_cs_vlan_tso);
1659 	desc->tx.paylen_ol4cs = cpu_to_le32(param.paylen_ol4cs);
1660 	desc->tx.mss_hw_csum = cpu_to_le16(param.mss_hw_csum);
1661 	desc->tx.vlan_tag = cpu_to_le16(param.inner_vtag);
1662 	desc->tx.outer_vlan_tag = cpu_to_le16(param.out_vtag);
1663 
1664 	return 0;
1665 }
1666 
1667 static int hns3_fill_desc(struct hns3_enet_ring *ring, dma_addr_t dma,
1668 			  unsigned int size)
1669 {
1670 #define HNS3_LIKELY_BD_NUM	1
1671 
1672 	struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1673 	unsigned int frag_buf_num;
1674 	int k, sizeoflast;
1675 
1676 	if (likely(size <= HNS3_MAX_BD_SIZE)) {
1677 		desc->addr = cpu_to_le64(dma);
1678 		desc->tx.send_size = cpu_to_le16(size);
1679 		desc->tx.bdtp_fe_sc_vld_ra_ri =
1680 			cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1681 
1682 		trace_hns3_tx_desc(ring, ring->next_to_use);
1683 		ring_ptr_move_fw(ring, next_to_use);
1684 		return HNS3_LIKELY_BD_NUM;
1685 	}
1686 
1687 	frag_buf_num = hns3_tx_bd_count(size);
1688 	sizeoflast = size % HNS3_MAX_BD_SIZE;
1689 	sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1690 
1691 	/* When frag size is bigger than hardware limit, split this frag */
1692 	for (k = 0; k < frag_buf_num; k++) {
1693 		/* now, fill the descriptor */
1694 		desc->addr = cpu_to_le64(dma + HNS3_MAX_BD_SIZE * k);
1695 		desc->tx.send_size = cpu_to_le16((k == frag_buf_num - 1) ?
1696 				     (u16)sizeoflast : (u16)HNS3_MAX_BD_SIZE);
1697 		desc->tx.bdtp_fe_sc_vld_ra_ri =
1698 				cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1699 
1700 		trace_hns3_tx_desc(ring, ring->next_to_use);
1701 		/* move ring pointer to next */
1702 		ring_ptr_move_fw(ring, next_to_use);
1703 
1704 		desc = &ring->desc[ring->next_to_use];
1705 	}
1706 
1707 	return frag_buf_num;
1708 }
1709 
1710 static int hns3_map_and_fill_desc(struct hns3_enet_ring *ring, void *priv,
1711 				  unsigned int type)
1712 {
1713 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1714 	struct device *dev = ring_to_dev(ring);
1715 	unsigned int size;
1716 	dma_addr_t dma;
1717 
1718 	if (type & (DESC_TYPE_FRAGLIST_SKB | DESC_TYPE_SKB)) {
1719 		struct sk_buff *skb = (struct sk_buff *)priv;
1720 
1721 		size = skb_headlen(skb);
1722 		if (!size)
1723 			return 0;
1724 
1725 		dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1726 	} else if (type & DESC_TYPE_BOUNCE_HEAD) {
1727 		/* Head data has been filled in hns3_handle_tx_bounce(),
1728 		 * just return 0 here.
1729 		 */
1730 		return 0;
1731 	} else {
1732 		skb_frag_t *frag = (skb_frag_t *)priv;
1733 
1734 		size = skb_frag_size(frag);
1735 		if (!size)
1736 			return 0;
1737 
1738 		dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1739 	}
1740 
1741 	if (unlikely(dma_mapping_error(dev, dma))) {
1742 		hns3_ring_stats_update(ring, sw_err_cnt);
1743 		return -ENOMEM;
1744 	}
1745 
1746 	desc_cb->priv = priv;
1747 	desc_cb->length = size;
1748 	desc_cb->dma = dma;
1749 	desc_cb->type = type;
1750 
1751 	return hns3_fill_desc(ring, dma, size);
1752 }
1753 
1754 static unsigned int hns3_skb_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1755 				    unsigned int bd_num)
1756 {
1757 	unsigned int size;
1758 	int i;
1759 
1760 	size = skb_headlen(skb);
1761 	while (size > HNS3_MAX_BD_SIZE) {
1762 		bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1763 		size -= HNS3_MAX_BD_SIZE;
1764 
1765 		if (bd_num > HNS3_MAX_TSO_BD_NUM)
1766 			return bd_num;
1767 	}
1768 
1769 	if (size) {
1770 		bd_size[bd_num++] = size;
1771 		if (bd_num > HNS3_MAX_TSO_BD_NUM)
1772 			return bd_num;
1773 	}
1774 
1775 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1776 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1777 		size = skb_frag_size(frag);
1778 		if (!size)
1779 			continue;
1780 
1781 		while (size > HNS3_MAX_BD_SIZE) {
1782 			bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1783 			size -= HNS3_MAX_BD_SIZE;
1784 
1785 			if (bd_num > HNS3_MAX_TSO_BD_NUM)
1786 				return bd_num;
1787 		}
1788 
1789 		bd_size[bd_num++] = size;
1790 		if (bd_num > HNS3_MAX_TSO_BD_NUM)
1791 			return bd_num;
1792 	}
1793 
1794 	return bd_num;
1795 }
1796 
1797 static unsigned int hns3_tx_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1798 				   u8 max_non_tso_bd_num, unsigned int bd_num,
1799 				   unsigned int recursion_level)
1800 {
1801 #define HNS3_MAX_RECURSION_LEVEL	24
1802 
1803 	struct sk_buff *frag_skb;
1804 
1805 	/* If the total len is within the max bd limit */
1806 	if (likely(skb->len <= HNS3_MAX_BD_SIZE && !recursion_level &&
1807 		   !skb_has_frag_list(skb) &&
1808 		   skb_shinfo(skb)->nr_frags < max_non_tso_bd_num))
1809 		return skb_shinfo(skb)->nr_frags + 1U;
1810 
1811 	if (unlikely(recursion_level >= HNS3_MAX_RECURSION_LEVEL))
1812 		return UINT_MAX;
1813 
1814 	bd_num = hns3_skb_bd_num(skb, bd_size, bd_num);
1815 	if (!skb_has_frag_list(skb) || bd_num > HNS3_MAX_TSO_BD_NUM)
1816 		return bd_num;
1817 
1818 	skb_walk_frags(skb, frag_skb) {
1819 		bd_num = hns3_tx_bd_num(frag_skb, bd_size, max_non_tso_bd_num,
1820 					bd_num, recursion_level + 1);
1821 		if (bd_num > HNS3_MAX_TSO_BD_NUM)
1822 			return bd_num;
1823 	}
1824 
1825 	return bd_num;
1826 }
1827 
1828 static unsigned int hns3_gso_hdr_len(struct sk_buff *skb)
1829 {
1830 	if (!skb->encapsulation)
1831 		return skb_transport_offset(skb) + tcp_hdrlen(skb);
1832 
1833 	return skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb);
1834 }
1835 
1836 /* HW need every continuous max_non_tso_bd_num buffer data to be larger
1837  * than MSS, we simplify it by ensuring skb_headlen + the first continuous
1838  * max_non_tso_bd_num - 1 frags to be larger than gso header len + mss,
1839  * and the remaining continuous max_non_tso_bd_num - 1 frags to be larger
1840  * than MSS except the last max_non_tso_bd_num - 1 frags.
1841  */
1842 static bool hns3_skb_need_linearized(struct sk_buff *skb, unsigned int *bd_size,
1843 				     unsigned int bd_num, u8 max_non_tso_bd_num)
1844 {
1845 	unsigned int tot_len = 0;
1846 	int i;
1847 
1848 	for (i = 0; i < max_non_tso_bd_num - 1U; i++)
1849 		tot_len += bd_size[i];
1850 
1851 	/* ensure the first max_non_tso_bd_num frags is greater than
1852 	 * mss + header
1853 	 */
1854 	if (tot_len + bd_size[max_non_tso_bd_num - 1U] <
1855 	    skb_shinfo(skb)->gso_size + hns3_gso_hdr_len(skb))
1856 		return true;
1857 
1858 	/* ensure every continuous max_non_tso_bd_num - 1 buffer is greater
1859 	 * than mss except the last one.
1860 	 */
1861 	for (i = 0; i < bd_num - max_non_tso_bd_num; i++) {
1862 		tot_len -= bd_size[i];
1863 		tot_len += bd_size[i + max_non_tso_bd_num - 1U];
1864 
1865 		if (tot_len < skb_shinfo(skb)->gso_size)
1866 			return true;
1867 	}
1868 
1869 	return false;
1870 }
1871 
1872 void hns3_shinfo_pack(struct skb_shared_info *shinfo, __u32 *size)
1873 {
1874 	int i;
1875 
1876 	for (i = 0; i < MAX_SKB_FRAGS; i++)
1877 		size[i] = skb_frag_size(&shinfo->frags[i]);
1878 }
1879 
1880 static int hns3_skb_linearize(struct hns3_enet_ring *ring,
1881 			      struct sk_buff *skb,
1882 			      unsigned int bd_num)
1883 {
1884 	/* 'bd_num == UINT_MAX' means the skb' fraglist has a
1885 	 * recursion level of over HNS3_MAX_RECURSION_LEVEL.
1886 	 */
1887 	if (bd_num == UINT_MAX) {
1888 		hns3_ring_stats_update(ring, over_max_recursion);
1889 		return -ENOMEM;
1890 	}
1891 
1892 	/* The skb->len has exceeded the hw limitation, linearization
1893 	 * will not help.
1894 	 */
1895 	if (skb->len > HNS3_MAX_TSO_SIZE ||
1896 	    (!skb_is_gso(skb) && skb->len > HNS3_MAX_NON_TSO_SIZE)) {
1897 		hns3_ring_stats_update(ring, hw_limitation);
1898 		return -ENOMEM;
1899 	}
1900 
1901 	if (__skb_linearize(skb)) {
1902 		hns3_ring_stats_update(ring, sw_err_cnt);
1903 		return -ENOMEM;
1904 	}
1905 
1906 	return 0;
1907 }
1908 
1909 static int hns3_nic_maybe_stop_tx(struct hns3_enet_ring *ring,
1910 				  struct net_device *netdev,
1911 				  struct sk_buff *skb)
1912 {
1913 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1914 	u8 max_non_tso_bd_num = priv->max_non_tso_bd_num;
1915 	unsigned int bd_size[HNS3_MAX_TSO_BD_NUM + 1U];
1916 	unsigned int bd_num;
1917 
1918 	bd_num = hns3_tx_bd_num(skb, bd_size, max_non_tso_bd_num, 0, 0);
1919 	if (unlikely(bd_num > max_non_tso_bd_num)) {
1920 		if (bd_num <= HNS3_MAX_TSO_BD_NUM && skb_is_gso(skb) &&
1921 		    !hns3_skb_need_linearized(skb, bd_size, bd_num,
1922 					      max_non_tso_bd_num)) {
1923 			trace_hns3_over_max_bd(skb);
1924 			goto out;
1925 		}
1926 
1927 		if (hns3_skb_linearize(ring, skb, bd_num))
1928 			return -ENOMEM;
1929 
1930 		bd_num = hns3_tx_bd_count(skb->len);
1931 
1932 		hns3_ring_stats_update(ring, tx_copy);
1933 	}
1934 
1935 out:
1936 	if (likely(ring_space(ring) >= bd_num))
1937 		return bd_num;
1938 
1939 	netif_stop_subqueue(netdev, ring->queue_index);
1940 	smp_mb(); /* Memory barrier before checking ring_space */
1941 
1942 	/* Start queue in case hns3_clean_tx_ring has just made room
1943 	 * available and has not seen the queue stopped state performed
1944 	 * by netif_stop_subqueue above.
1945 	 */
1946 	if (ring_space(ring) >= bd_num && netif_carrier_ok(netdev) &&
1947 	    !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
1948 		netif_start_subqueue(netdev, ring->queue_index);
1949 		return bd_num;
1950 	}
1951 
1952 	hns3_ring_stats_update(ring, tx_busy);
1953 
1954 	return -EBUSY;
1955 }
1956 
1957 static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
1958 {
1959 	struct device *dev = ring_to_dev(ring);
1960 	unsigned int i;
1961 
1962 	for (i = 0; i < ring->desc_num; i++) {
1963 		struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1964 		struct hns3_desc_cb *desc_cb;
1965 
1966 		memset(desc, 0, sizeof(*desc));
1967 
1968 		/* check if this is where we started */
1969 		if (ring->next_to_use == next_to_use_orig)
1970 			break;
1971 
1972 		/* rollback one */
1973 		ring_ptr_move_bw(ring, next_to_use);
1974 
1975 		desc_cb = &ring->desc_cb[ring->next_to_use];
1976 
1977 		if (!desc_cb->dma)
1978 			continue;
1979 
1980 		/* unmap the descriptor dma address */
1981 		if (desc_cb->type & (DESC_TYPE_SKB | DESC_TYPE_FRAGLIST_SKB))
1982 			dma_unmap_single(dev, desc_cb->dma, desc_cb->length,
1983 					 DMA_TO_DEVICE);
1984 		else if (desc_cb->type &
1985 			 (DESC_TYPE_BOUNCE_HEAD | DESC_TYPE_BOUNCE_ALL))
1986 			hns3_tx_spare_rollback(ring, desc_cb->length);
1987 		else if (desc_cb->length)
1988 			dma_unmap_page(dev, desc_cb->dma, desc_cb->length,
1989 				       DMA_TO_DEVICE);
1990 
1991 		desc_cb->length = 0;
1992 		desc_cb->dma = 0;
1993 		desc_cb->type = DESC_TYPE_UNKNOWN;
1994 	}
1995 }
1996 
1997 static int hns3_fill_skb_to_desc(struct hns3_enet_ring *ring,
1998 				 struct sk_buff *skb, unsigned int type)
1999 {
2000 	struct sk_buff *frag_skb;
2001 	int i, ret, bd_num = 0;
2002 
2003 	ret = hns3_map_and_fill_desc(ring, skb, type);
2004 	if (unlikely(ret < 0))
2005 		return ret;
2006 
2007 	bd_num += ret;
2008 
2009 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2010 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2011 
2012 		ret = hns3_map_and_fill_desc(ring, frag, DESC_TYPE_PAGE);
2013 		if (unlikely(ret < 0))
2014 			return ret;
2015 
2016 		bd_num += ret;
2017 	}
2018 
2019 	skb_walk_frags(skb, frag_skb) {
2020 		ret = hns3_fill_skb_to_desc(ring, frag_skb,
2021 					    DESC_TYPE_FRAGLIST_SKB);
2022 		if (unlikely(ret < 0))
2023 			return ret;
2024 
2025 		bd_num += ret;
2026 	}
2027 
2028 	return bd_num;
2029 }
2030 
2031 static void hns3_tx_push_bd(struct hns3_enet_ring *ring, int num)
2032 {
2033 #define HNS3_BYTES_PER_64BIT		8
2034 
2035 	struct hns3_desc desc[HNS3_MAX_PUSH_BD_NUM] = {};
2036 	int offset = 0;
2037 
2038 	/* make sure everything is visible to device before
2039 	 * excuting tx push or updating doorbell
2040 	 */
2041 	dma_wmb();
2042 
2043 	do {
2044 		int idx = (ring->next_to_use - num + ring->desc_num) %
2045 			  ring->desc_num;
2046 
2047 		u64_stats_update_begin(&ring->syncp);
2048 		ring->stats.tx_push++;
2049 		u64_stats_update_end(&ring->syncp);
2050 		memcpy(&desc[offset], &ring->desc[idx],
2051 		       sizeof(struct hns3_desc));
2052 		offset++;
2053 	} while (--num);
2054 
2055 	__iowrite64_copy(ring->tqp->mem_base, desc,
2056 			 (sizeof(struct hns3_desc) * HNS3_MAX_PUSH_BD_NUM) /
2057 			 HNS3_BYTES_PER_64BIT);
2058 
2059 	io_stop_wc();
2060 }
2061 
2062 static void hns3_tx_mem_doorbell(struct hns3_enet_ring *ring)
2063 {
2064 #define HNS3_MEM_DOORBELL_OFFSET	64
2065 
2066 	__le64 bd_num = cpu_to_le64((u64)ring->pending_buf);
2067 
2068 	/* make sure everything is visible to device before
2069 	 * excuting tx push or updating doorbell
2070 	 */
2071 	dma_wmb();
2072 
2073 	__iowrite64_copy(ring->tqp->mem_base + HNS3_MEM_DOORBELL_OFFSET,
2074 			 &bd_num, 1);
2075 	u64_stats_update_begin(&ring->syncp);
2076 	ring->stats.tx_mem_doorbell += ring->pending_buf;
2077 	u64_stats_update_end(&ring->syncp);
2078 
2079 	io_stop_wc();
2080 }
2081 
2082 static void hns3_tx_doorbell(struct hns3_enet_ring *ring, int num,
2083 			     bool doorbell)
2084 {
2085 	struct net_device *netdev = ring_to_netdev(ring);
2086 	struct hns3_nic_priv *priv = netdev_priv(netdev);
2087 
2088 	/* when tx push is enabled, the packet whose number of BD below
2089 	 * HNS3_MAX_PUSH_BD_NUM can be pushed directly.
2090 	 */
2091 	if (test_bit(HNS3_NIC_STATE_TX_PUSH_ENABLE, &priv->state) && num &&
2092 	    !ring->pending_buf && num <= HNS3_MAX_PUSH_BD_NUM && doorbell) {
2093 		hns3_tx_push_bd(ring, num);
2094 		WRITE_ONCE(ring->last_to_use, ring->next_to_use);
2095 		return;
2096 	}
2097 
2098 	ring->pending_buf += num;
2099 
2100 	if (!doorbell) {
2101 		hns3_ring_stats_update(ring, tx_more);
2102 		return;
2103 	}
2104 
2105 	if (ring->tqp->mem_base)
2106 		hns3_tx_mem_doorbell(ring);
2107 	else
2108 		writel(ring->pending_buf,
2109 		       ring->tqp->io_base + HNS3_RING_TX_RING_TAIL_REG);
2110 
2111 	ring->pending_buf = 0;
2112 	WRITE_ONCE(ring->last_to_use, ring->next_to_use);
2113 }
2114 
2115 static void hns3_tsyn(struct net_device *netdev, struct sk_buff *skb,
2116 		      struct hns3_desc *desc)
2117 {
2118 	struct hnae3_handle *h = hns3_get_handle(netdev);
2119 
2120 	if (!(h->ae_algo->ops->set_tx_hwts_info &&
2121 	      h->ae_algo->ops->set_tx_hwts_info(h, skb)))
2122 		return;
2123 
2124 	desc->tx.bdtp_fe_sc_vld_ra_ri |= cpu_to_le16(BIT(HNS3_TXD_TSYN_B));
2125 }
2126 
2127 static int hns3_handle_tx_bounce(struct hns3_enet_ring *ring,
2128 				 struct sk_buff *skb)
2129 {
2130 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
2131 	unsigned int type = DESC_TYPE_BOUNCE_HEAD;
2132 	unsigned int size = skb_headlen(skb);
2133 	dma_addr_t dma;
2134 	int bd_num = 0;
2135 	u32 cb_len;
2136 	void *buf;
2137 	int ret;
2138 
2139 	if (skb->len <= ring->tx_copybreak) {
2140 		size = skb->len;
2141 		type = DESC_TYPE_BOUNCE_ALL;
2142 	}
2143 
2144 	/* hns3_can_use_tx_bounce() is called to ensure the below
2145 	 * function can always return the tx buffer.
2146 	 */
2147 	buf = hns3_tx_spare_alloc(ring, size, &dma, &cb_len);
2148 
2149 	ret = skb_copy_bits(skb, 0, buf, size);
2150 	if (unlikely(ret < 0)) {
2151 		hns3_tx_spare_rollback(ring, cb_len);
2152 		hns3_ring_stats_update(ring, copy_bits_err);
2153 		return ret;
2154 	}
2155 
2156 	desc_cb->priv = skb;
2157 	desc_cb->length = cb_len;
2158 	desc_cb->dma = dma;
2159 	desc_cb->type = type;
2160 
2161 	bd_num += hns3_fill_desc(ring, dma, size);
2162 
2163 	if (type == DESC_TYPE_BOUNCE_HEAD) {
2164 		ret = hns3_fill_skb_to_desc(ring, skb,
2165 					    DESC_TYPE_BOUNCE_HEAD);
2166 		if (unlikely(ret < 0))
2167 			return ret;
2168 
2169 		bd_num += ret;
2170 	}
2171 
2172 	dma_sync_single_for_device(ring_to_dev(ring), dma, size,
2173 				   DMA_TO_DEVICE);
2174 
2175 	hns3_ring_stats_update(ring, tx_bounce);
2176 
2177 	return bd_num;
2178 }
2179 
2180 static int hns3_handle_tx_sgl(struct hns3_enet_ring *ring,
2181 			      struct sk_buff *skb)
2182 {
2183 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
2184 	u32 nfrag = skb_shinfo(skb)->nr_frags + 1;
2185 	struct sg_table *sgt;
2186 	int i, bd_num = 0;
2187 	dma_addr_t dma;
2188 	u32 cb_len;
2189 	int nents;
2190 
2191 	if (skb_has_frag_list(skb))
2192 		nfrag = HNS3_MAX_TSO_BD_NUM;
2193 
2194 	/* hns3_can_use_tx_sgl() is called to ensure the below
2195 	 * function can always return the tx buffer.
2196 	 */
2197 	sgt = hns3_tx_spare_alloc(ring, HNS3_SGL_SIZE(nfrag),
2198 				  &dma, &cb_len);
2199 
2200 	/* scatterlist follows by the sg table */
2201 	sgt->sgl = (struct scatterlist *)(sgt + 1);
2202 	sg_init_table(sgt->sgl, nfrag);
2203 	nents = skb_to_sgvec(skb, sgt->sgl, 0, skb->len);
2204 	if (unlikely(nents < 0)) {
2205 		hns3_tx_spare_rollback(ring, cb_len);
2206 		hns3_ring_stats_update(ring, skb2sgl_err);
2207 		return -ENOMEM;
2208 	}
2209 
2210 	sgt->orig_nents = nents;
2211 	sgt->nents = dma_map_sg(ring_to_dev(ring), sgt->sgl, sgt->orig_nents,
2212 				DMA_TO_DEVICE);
2213 	if (unlikely(!sgt->nents)) {
2214 		hns3_tx_spare_rollback(ring, cb_len);
2215 		hns3_ring_stats_update(ring, map_sg_err);
2216 		return -ENOMEM;
2217 	}
2218 
2219 	desc_cb->priv = skb;
2220 	desc_cb->length = cb_len;
2221 	desc_cb->dma = dma;
2222 	desc_cb->type = DESC_TYPE_SGL_SKB;
2223 
2224 	for (i = 0; i < sgt->nents; i++)
2225 		bd_num += hns3_fill_desc(ring, sg_dma_address(sgt->sgl + i),
2226 					 sg_dma_len(sgt->sgl + i));
2227 	hns3_ring_stats_update(ring, tx_sgl);
2228 
2229 	return bd_num;
2230 }
2231 
2232 static int hns3_handle_desc_filling(struct hns3_enet_ring *ring,
2233 				    struct sk_buff *skb)
2234 {
2235 	u32 space;
2236 
2237 	if (!ring->tx_spare)
2238 		goto out;
2239 
2240 	space = hns3_tx_spare_space(ring);
2241 
2242 	if (hns3_can_use_tx_sgl(ring, skb, space))
2243 		return hns3_handle_tx_sgl(ring, skb);
2244 
2245 	if (hns3_can_use_tx_bounce(ring, skb, space))
2246 		return hns3_handle_tx_bounce(ring, skb);
2247 
2248 out:
2249 	return hns3_fill_skb_to_desc(ring, skb, DESC_TYPE_SKB);
2250 }
2251 
2252 static int hns3_handle_skb_desc(struct hns3_enet_ring *ring,
2253 				struct sk_buff *skb,
2254 				struct hns3_desc_cb *desc_cb,
2255 				int next_to_use_head)
2256 {
2257 	int ret;
2258 
2259 	ret = hns3_fill_skb_desc(ring, skb, &ring->desc[ring->next_to_use],
2260 				 desc_cb);
2261 	if (unlikely(ret < 0))
2262 		goto fill_err;
2263 
2264 	/* 'ret < 0' means filling error, 'ret == 0' means skb->len is
2265 	 * zero, which is unlikely, and 'ret > 0' means how many tx desc
2266 	 * need to be notified to the hw.
2267 	 */
2268 	ret = hns3_handle_desc_filling(ring, skb);
2269 	if (likely(ret > 0))
2270 		return ret;
2271 
2272 fill_err:
2273 	hns3_clear_desc(ring, next_to_use_head);
2274 	return ret;
2275 }
2276 
2277 netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
2278 {
2279 	struct hns3_nic_priv *priv = netdev_priv(netdev);
2280 	struct hns3_enet_ring *ring = &priv->ring[skb->queue_mapping];
2281 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
2282 	struct netdev_queue *dev_queue;
2283 	int pre_ntu, ret;
2284 	bool doorbell;
2285 
2286 	/* Hardware can only handle short frames above 32 bytes */
2287 	if (skb_put_padto(skb, HNS3_MIN_TX_LEN)) {
2288 		hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
2289 
2290 		hns3_ring_stats_update(ring, sw_err_cnt);
2291 
2292 		return NETDEV_TX_OK;
2293 	}
2294 
2295 	/* Prefetch the data used later */
2296 	prefetch(skb->data);
2297 
2298 	ret = hns3_nic_maybe_stop_tx(ring, netdev, skb);
2299 	if (unlikely(ret <= 0)) {
2300 		if (ret == -EBUSY) {
2301 			hns3_tx_doorbell(ring, 0, true);
2302 			return NETDEV_TX_BUSY;
2303 		}
2304 
2305 		hns3_rl_err(netdev, "xmit error: %d!\n", ret);
2306 		goto out_err_tx_ok;
2307 	}
2308 
2309 	ret = hns3_handle_skb_desc(ring, skb, desc_cb, ring->next_to_use);
2310 	if (unlikely(ret <= 0))
2311 		goto out_err_tx_ok;
2312 
2313 	pre_ntu = ring->next_to_use ? (ring->next_to_use - 1) :
2314 					(ring->desc_num - 1);
2315 
2316 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP))
2317 		hns3_tsyn(netdev, skb, &ring->desc[pre_ntu]);
2318 
2319 	ring->desc[pre_ntu].tx.bdtp_fe_sc_vld_ra_ri |=
2320 				cpu_to_le16(BIT(HNS3_TXD_FE_B));
2321 	trace_hns3_tx_desc(ring, pre_ntu);
2322 
2323 	skb_tx_timestamp(skb);
2324 
2325 	/* Complete translate all packets */
2326 	dev_queue = netdev_get_tx_queue(netdev, ring->queue_index);
2327 	doorbell = __netdev_tx_sent_queue(dev_queue, desc_cb->send_bytes,
2328 					  netdev_xmit_more());
2329 	hns3_tx_doorbell(ring, ret, doorbell);
2330 
2331 	return NETDEV_TX_OK;
2332 
2333 out_err_tx_ok:
2334 	dev_kfree_skb_any(skb);
2335 	hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
2336 	return NETDEV_TX_OK;
2337 }
2338 
2339 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
2340 {
2341 	char format_mac_addr_perm[HNAE3_FORMAT_MAC_ADDR_LEN];
2342 	char format_mac_addr_sa[HNAE3_FORMAT_MAC_ADDR_LEN];
2343 	struct hnae3_handle *h = hns3_get_handle(netdev);
2344 	struct sockaddr *mac_addr = p;
2345 	int ret;
2346 
2347 	if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
2348 		return -EADDRNOTAVAIL;
2349 
2350 	if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
2351 		hnae3_format_mac_addr(format_mac_addr_sa, mac_addr->sa_data);
2352 		netdev_info(netdev, "already using mac address %s\n",
2353 			    format_mac_addr_sa);
2354 		return 0;
2355 	}
2356 
2357 	/* For VF device, if there is a perm_addr, then the user will not
2358 	 * be allowed to change the address.
2359 	 */
2360 	if (!hns3_is_phys_func(h->pdev) &&
2361 	    !is_zero_ether_addr(netdev->perm_addr)) {
2362 		hnae3_format_mac_addr(format_mac_addr_perm, netdev->perm_addr);
2363 		hnae3_format_mac_addr(format_mac_addr_sa, mac_addr->sa_data);
2364 		netdev_err(netdev, "has permanent MAC %s, user MAC %s not allow\n",
2365 			   format_mac_addr_perm, format_mac_addr_sa);
2366 		return -EPERM;
2367 	}
2368 
2369 	ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
2370 	if (ret) {
2371 		netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
2372 		return ret;
2373 	}
2374 
2375 	eth_hw_addr_set(netdev, mac_addr->sa_data);
2376 
2377 	return 0;
2378 }
2379 
2380 static int hns3_nic_do_ioctl(struct net_device *netdev,
2381 			     struct ifreq *ifr, int cmd)
2382 {
2383 	struct hnae3_handle *h = hns3_get_handle(netdev);
2384 
2385 	if (!netif_running(netdev))
2386 		return -EINVAL;
2387 
2388 	if (!h->ae_algo->ops->do_ioctl)
2389 		return -EOPNOTSUPP;
2390 
2391 	return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
2392 }
2393 
2394 static int hns3_nic_set_features(struct net_device *netdev,
2395 				 netdev_features_t features)
2396 {
2397 	netdev_features_t changed = netdev->features ^ features;
2398 	struct hns3_nic_priv *priv = netdev_priv(netdev);
2399 	struct hnae3_handle *h = priv->ae_handle;
2400 	bool enable;
2401 	int ret;
2402 
2403 	if (changed & (NETIF_F_GRO_HW) && h->ae_algo->ops->set_gro_en) {
2404 		enable = !!(features & NETIF_F_GRO_HW);
2405 		ret = h->ae_algo->ops->set_gro_en(h, enable);
2406 		if (ret)
2407 			return ret;
2408 	}
2409 
2410 	if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
2411 	    h->ae_algo->ops->enable_hw_strip_rxvtag) {
2412 		enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
2413 		ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, enable);
2414 		if (ret)
2415 			return ret;
2416 	}
2417 
2418 	if ((changed & NETIF_F_NTUPLE) && h->ae_algo->ops->enable_fd) {
2419 		enable = !!(features & NETIF_F_NTUPLE);
2420 		h->ae_algo->ops->enable_fd(h, enable);
2421 	}
2422 
2423 	if ((netdev->features & NETIF_F_HW_TC) > (features & NETIF_F_HW_TC) &&
2424 	    h->ae_algo->ops->cls_flower_active(h)) {
2425 		netdev_err(netdev,
2426 			   "there are offloaded TC filters active, cannot disable HW TC offload");
2427 		return -EINVAL;
2428 	}
2429 
2430 	if ((changed & NETIF_F_HW_VLAN_CTAG_FILTER) &&
2431 	    h->ae_algo->ops->enable_vlan_filter) {
2432 		enable = !!(features & NETIF_F_HW_VLAN_CTAG_FILTER);
2433 		ret = h->ae_algo->ops->enable_vlan_filter(h, enable);
2434 		if (ret)
2435 			return ret;
2436 	}
2437 
2438 	netdev->features = features;
2439 	return 0;
2440 }
2441 
2442 static netdev_features_t hns3_features_check(struct sk_buff *skb,
2443 					     struct net_device *dev,
2444 					     netdev_features_t features)
2445 {
2446 #define HNS3_MAX_HDR_LEN	480U
2447 #define HNS3_MAX_L4_HDR_LEN	60U
2448 
2449 	size_t len;
2450 
2451 	if (skb->ip_summed != CHECKSUM_PARTIAL)
2452 		return features;
2453 
2454 	if (skb->encapsulation)
2455 		len = skb_inner_transport_header(skb) - skb->data;
2456 	else
2457 		len = skb_transport_header(skb) - skb->data;
2458 
2459 	/* Assume L4 is 60 byte as TCP is the only protocol with a
2460 	 * a flexible value, and it's max len is 60 bytes.
2461 	 */
2462 	len += HNS3_MAX_L4_HDR_LEN;
2463 
2464 	/* Hardware only supports checksum on the skb with a max header
2465 	 * len of 480 bytes.
2466 	 */
2467 	if (len > HNS3_MAX_HDR_LEN)
2468 		features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2469 
2470 	return features;
2471 }
2472 
2473 static void hns3_fetch_stats(struct rtnl_link_stats64 *stats,
2474 			     struct hns3_enet_ring *ring, bool is_tx)
2475 {
2476 	unsigned int start;
2477 
2478 	do {
2479 		start = u64_stats_fetch_begin_irq(&ring->syncp);
2480 		if (is_tx) {
2481 			stats->tx_bytes += ring->stats.tx_bytes;
2482 			stats->tx_packets += ring->stats.tx_pkts;
2483 			stats->tx_dropped += ring->stats.sw_err_cnt;
2484 			stats->tx_dropped += ring->stats.tx_vlan_err;
2485 			stats->tx_dropped += ring->stats.tx_l4_proto_err;
2486 			stats->tx_dropped += ring->stats.tx_l2l3l4_err;
2487 			stats->tx_dropped += ring->stats.tx_tso_err;
2488 			stats->tx_dropped += ring->stats.over_max_recursion;
2489 			stats->tx_dropped += ring->stats.hw_limitation;
2490 			stats->tx_dropped += ring->stats.copy_bits_err;
2491 			stats->tx_dropped += ring->stats.skb2sgl_err;
2492 			stats->tx_dropped += ring->stats.map_sg_err;
2493 			stats->tx_errors += ring->stats.sw_err_cnt;
2494 			stats->tx_errors += ring->stats.tx_vlan_err;
2495 			stats->tx_errors += ring->stats.tx_l4_proto_err;
2496 			stats->tx_errors += ring->stats.tx_l2l3l4_err;
2497 			stats->tx_errors += ring->stats.tx_tso_err;
2498 			stats->tx_errors += ring->stats.over_max_recursion;
2499 			stats->tx_errors += ring->stats.hw_limitation;
2500 			stats->tx_errors += ring->stats.copy_bits_err;
2501 			stats->tx_errors += ring->stats.skb2sgl_err;
2502 			stats->tx_errors += ring->stats.map_sg_err;
2503 		} else {
2504 			stats->rx_bytes += ring->stats.rx_bytes;
2505 			stats->rx_packets += ring->stats.rx_pkts;
2506 			stats->rx_dropped += ring->stats.l2_err;
2507 			stats->rx_errors += ring->stats.l2_err;
2508 			stats->rx_errors += ring->stats.l3l4_csum_err;
2509 			stats->rx_crc_errors += ring->stats.l2_err;
2510 			stats->multicast += ring->stats.rx_multicast;
2511 			stats->rx_length_errors += ring->stats.err_pkt_len;
2512 		}
2513 	} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
2514 }
2515 
2516 static void hns3_nic_get_stats64(struct net_device *netdev,
2517 				 struct rtnl_link_stats64 *stats)
2518 {
2519 	struct hns3_nic_priv *priv = netdev_priv(netdev);
2520 	int queue_num = priv->ae_handle->kinfo.num_tqps;
2521 	struct hnae3_handle *handle = priv->ae_handle;
2522 	struct rtnl_link_stats64 ring_total_stats;
2523 	struct hns3_enet_ring *ring;
2524 	unsigned int idx;
2525 
2526 	if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
2527 		return;
2528 
2529 	handle->ae_algo->ops->update_stats(handle, &netdev->stats);
2530 
2531 	memset(&ring_total_stats, 0, sizeof(ring_total_stats));
2532 	for (idx = 0; idx < queue_num; idx++) {
2533 		/* fetch the tx stats */
2534 		ring = &priv->ring[idx];
2535 		hns3_fetch_stats(&ring_total_stats, ring, true);
2536 
2537 		/* fetch the rx stats */
2538 		ring = &priv->ring[idx + queue_num];
2539 		hns3_fetch_stats(&ring_total_stats, ring, false);
2540 	}
2541 
2542 	stats->tx_bytes = ring_total_stats.tx_bytes;
2543 	stats->tx_packets = ring_total_stats.tx_packets;
2544 	stats->rx_bytes = ring_total_stats.rx_bytes;
2545 	stats->rx_packets = ring_total_stats.rx_packets;
2546 
2547 	stats->rx_errors = ring_total_stats.rx_errors;
2548 	stats->multicast = ring_total_stats.multicast;
2549 	stats->rx_length_errors = ring_total_stats.rx_length_errors;
2550 	stats->rx_crc_errors = ring_total_stats.rx_crc_errors;
2551 	stats->rx_missed_errors = netdev->stats.rx_missed_errors;
2552 
2553 	stats->tx_errors = ring_total_stats.tx_errors;
2554 	stats->rx_dropped = ring_total_stats.rx_dropped;
2555 	stats->tx_dropped = ring_total_stats.tx_dropped;
2556 	stats->collisions = netdev->stats.collisions;
2557 	stats->rx_over_errors = netdev->stats.rx_over_errors;
2558 	stats->rx_frame_errors = netdev->stats.rx_frame_errors;
2559 	stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
2560 	stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
2561 	stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
2562 	stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
2563 	stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
2564 	stats->tx_window_errors = netdev->stats.tx_window_errors;
2565 	stats->rx_compressed = netdev->stats.rx_compressed;
2566 	stats->tx_compressed = netdev->stats.tx_compressed;
2567 }
2568 
2569 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
2570 {
2571 	struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
2572 	struct hnae3_knic_private_info *kinfo;
2573 	u8 tc = mqprio_qopt->qopt.num_tc;
2574 	u16 mode = mqprio_qopt->mode;
2575 	u8 hw = mqprio_qopt->qopt.hw;
2576 	struct hnae3_handle *h;
2577 
2578 	if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
2579 	       mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
2580 		return -EOPNOTSUPP;
2581 
2582 	if (tc > HNAE3_MAX_TC)
2583 		return -EINVAL;
2584 
2585 	if (!netdev)
2586 		return -EINVAL;
2587 
2588 	h = hns3_get_handle(netdev);
2589 	kinfo = &h->kinfo;
2590 
2591 	netif_dbg(h, drv, netdev, "setup tc: num_tc=%u\n", tc);
2592 
2593 	return (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
2594 		kinfo->dcb_ops->setup_tc(h, mqprio_qopt) : -EOPNOTSUPP;
2595 }
2596 
2597 static int hns3_setup_tc_cls_flower(struct hns3_nic_priv *priv,
2598 				    struct flow_cls_offload *flow)
2599 {
2600 	int tc = tc_classid_to_hwtc(priv->netdev, flow->classid);
2601 	struct hnae3_handle *h = hns3_get_handle(priv->netdev);
2602 
2603 	switch (flow->command) {
2604 	case FLOW_CLS_REPLACE:
2605 		if (h->ae_algo->ops->add_cls_flower)
2606 			return h->ae_algo->ops->add_cls_flower(h, flow, tc);
2607 		break;
2608 	case FLOW_CLS_DESTROY:
2609 		if (h->ae_algo->ops->del_cls_flower)
2610 			return h->ae_algo->ops->del_cls_flower(h, flow);
2611 		break;
2612 	default:
2613 		break;
2614 	}
2615 
2616 	return -EOPNOTSUPP;
2617 }
2618 
2619 static int hns3_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
2620 				  void *cb_priv)
2621 {
2622 	struct hns3_nic_priv *priv = cb_priv;
2623 
2624 	if (!tc_cls_can_offload_and_chain0(priv->netdev, type_data))
2625 		return -EOPNOTSUPP;
2626 
2627 	switch (type) {
2628 	case TC_SETUP_CLSFLOWER:
2629 		return hns3_setup_tc_cls_flower(priv, type_data);
2630 	default:
2631 		return -EOPNOTSUPP;
2632 	}
2633 }
2634 
2635 static LIST_HEAD(hns3_block_cb_list);
2636 
2637 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
2638 			     void *type_data)
2639 {
2640 	struct hns3_nic_priv *priv = netdev_priv(dev);
2641 	int ret;
2642 
2643 	switch (type) {
2644 	case TC_SETUP_QDISC_MQPRIO:
2645 		ret = hns3_setup_tc(dev, type_data);
2646 		break;
2647 	case TC_SETUP_BLOCK:
2648 		ret = flow_block_cb_setup_simple(type_data,
2649 						 &hns3_block_cb_list,
2650 						 hns3_setup_tc_block_cb,
2651 						 priv, priv, true);
2652 		break;
2653 	default:
2654 		return -EOPNOTSUPP;
2655 	}
2656 
2657 	return ret;
2658 }
2659 
2660 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
2661 				__be16 proto, u16 vid)
2662 {
2663 	struct hnae3_handle *h = hns3_get_handle(netdev);
2664 	int ret = -EIO;
2665 
2666 	if (h->ae_algo->ops->set_vlan_filter)
2667 		ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
2668 
2669 	return ret;
2670 }
2671 
2672 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
2673 				 __be16 proto, u16 vid)
2674 {
2675 	struct hnae3_handle *h = hns3_get_handle(netdev);
2676 	int ret = -EIO;
2677 
2678 	if (h->ae_algo->ops->set_vlan_filter)
2679 		ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
2680 
2681 	return ret;
2682 }
2683 
2684 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
2685 				u8 qos, __be16 vlan_proto)
2686 {
2687 	struct hnae3_handle *h = hns3_get_handle(netdev);
2688 	int ret = -EIO;
2689 
2690 	netif_dbg(h, drv, netdev,
2691 		  "set vf vlan: vf=%d, vlan=%u, qos=%u, vlan_proto=0x%x\n",
2692 		  vf, vlan, qos, ntohs(vlan_proto));
2693 
2694 	if (h->ae_algo->ops->set_vf_vlan_filter)
2695 		ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
2696 							  qos, vlan_proto);
2697 
2698 	return ret;
2699 }
2700 
2701 static int hns3_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
2702 {
2703 	struct hnae3_handle *handle = hns3_get_handle(netdev);
2704 
2705 	if (hns3_nic_resetting(netdev))
2706 		return -EBUSY;
2707 
2708 	if (!handle->ae_algo->ops->set_vf_spoofchk)
2709 		return -EOPNOTSUPP;
2710 
2711 	return handle->ae_algo->ops->set_vf_spoofchk(handle, vf, enable);
2712 }
2713 
2714 static int hns3_set_vf_trust(struct net_device *netdev, int vf, bool enable)
2715 {
2716 	struct hnae3_handle *handle = hns3_get_handle(netdev);
2717 
2718 	if (!handle->ae_algo->ops->set_vf_trust)
2719 		return -EOPNOTSUPP;
2720 
2721 	return handle->ae_algo->ops->set_vf_trust(handle, vf, enable);
2722 }
2723 
2724 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
2725 {
2726 	struct hnae3_handle *h = hns3_get_handle(netdev);
2727 	int ret;
2728 
2729 	if (hns3_nic_resetting(netdev))
2730 		return -EBUSY;
2731 
2732 	if (!h->ae_algo->ops->set_mtu)
2733 		return -EOPNOTSUPP;
2734 
2735 	netif_dbg(h, drv, netdev,
2736 		  "change mtu from %u to %d\n", netdev->mtu, new_mtu);
2737 
2738 	ret = h->ae_algo->ops->set_mtu(h, new_mtu);
2739 	if (ret)
2740 		netdev_err(netdev, "failed to change MTU in hardware %d\n",
2741 			   ret);
2742 	else
2743 		netdev->mtu = new_mtu;
2744 
2745 	return ret;
2746 }
2747 
2748 static int hns3_get_timeout_queue(struct net_device *ndev)
2749 {
2750 	int i;
2751 
2752 	/* Find the stopped queue the same way the stack does */
2753 	for (i = 0; i < ndev->num_tx_queues; i++) {
2754 		struct netdev_queue *q;
2755 		unsigned long trans_start;
2756 
2757 		q = netdev_get_tx_queue(ndev, i);
2758 		trans_start = READ_ONCE(q->trans_start);
2759 		if (netif_xmit_stopped(q) &&
2760 		    time_after(jiffies,
2761 			       (trans_start + ndev->watchdog_timeo))) {
2762 #ifdef CONFIG_BQL
2763 			struct dql *dql = &q->dql;
2764 
2765 			netdev_info(ndev, "DQL info last_cnt: %u, queued: %u, adj_limit: %u, completed: %u\n",
2766 				    dql->last_obj_cnt, dql->num_queued,
2767 				    dql->adj_limit, dql->num_completed);
2768 #endif
2769 			netdev_info(ndev, "queue state: 0x%lx, delta msecs: %u\n",
2770 				    q->state,
2771 				    jiffies_to_msecs(jiffies - trans_start));
2772 			break;
2773 		}
2774 	}
2775 
2776 	return i;
2777 }
2778 
2779 static void hns3_dump_queue_stats(struct net_device *ndev,
2780 				  struct hns3_enet_ring *tx_ring,
2781 				  int timeout_queue)
2782 {
2783 	struct napi_struct *napi = &tx_ring->tqp_vector->napi;
2784 	struct hns3_nic_priv *priv = netdev_priv(ndev);
2785 
2786 	netdev_info(ndev,
2787 		    "tx_timeout count: %llu, queue id: %d, SW_NTU: 0x%x, SW_NTC: 0x%x, napi state: %lu\n",
2788 		    priv->tx_timeout_count, timeout_queue, tx_ring->next_to_use,
2789 		    tx_ring->next_to_clean, napi->state);
2790 
2791 	netdev_info(ndev,
2792 		    "tx_pkts: %llu, tx_bytes: %llu, sw_err_cnt: %llu, tx_pending: %d\n",
2793 		    tx_ring->stats.tx_pkts, tx_ring->stats.tx_bytes,
2794 		    tx_ring->stats.sw_err_cnt, tx_ring->pending_buf);
2795 
2796 	netdev_info(ndev,
2797 		    "seg_pkt_cnt: %llu, tx_more: %llu, restart_queue: %llu, tx_busy: %llu\n",
2798 		    tx_ring->stats.seg_pkt_cnt, tx_ring->stats.tx_more,
2799 		    tx_ring->stats.restart_queue, tx_ring->stats.tx_busy);
2800 
2801 	netdev_info(ndev, "tx_push: %llu, tx_mem_doorbell: %llu\n",
2802 		    tx_ring->stats.tx_push, tx_ring->stats.tx_mem_doorbell);
2803 }
2804 
2805 static void hns3_dump_queue_reg(struct net_device *ndev,
2806 				struct hns3_enet_ring *tx_ring)
2807 {
2808 	netdev_info(ndev,
2809 		    "BD_NUM: 0x%x HW_HEAD: 0x%x, HW_TAIL: 0x%x, BD_ERR: 0x%x, INT: 0x%x\n",
2810 		    hns3_tqp_read_reg(tx_ring, HNS3_RING_TX_RING_BD_NUM_REG),
2811 		    hns3_tqp_read_reg(tx_ring, HNS3_RING_TX_RING_HEAD_REG),
2812 		    hns3_tqp_read_reg(tx_ring, HNS3_RING_TX_RING_TAIL_REG),
2813 		    hns3_tqp_read_reg(tx_ring, HNS3_RING_TX_RING_BD_ERR_REG),
2814 		    readl(tx_ring->tqp_vector->mask_addr));
2815 	netdev_info(ndev,
2816 		    "RING_EN: 0x%x, TC: 0x%x, FBD_NUM: 0x%x FBD_OFT: 0x%x, EBD_NUM: 0x%x, EBD_OFT: 0x%x\n",
2817 		    hns3_tqp_read_reg(tx_ring, HNS3_RING_EN_REG),
2818 		    hns3_tqp_read_reg(tx_ring, HNS3_RING_TX_RING_TC_REG),
2819 		    hns3_tqp_read_reg(tx_ring, HNS3_RING_TX_RING_FBDNUM_REG),
2820 		    hns3_tqp_read_reg(tx_ring, HNS3_RING_TX_RING_OFFSET_REG),
2821 		    hns3_tqp_read_reg(tx_ring, HNS3_RING_TX_RING_EBDNUM_REG),
2822 		    hns3_tqp_read_reg(tx_ring,
2823 				      HNS3_RING_TX_RING_EBD_OFFSET_REG));
2824 }
2825 
2826 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
2827 {
2828 	struct hns3_nic_priv *priv = netdev_priv(ndev);
2829 	struct hnae3_handle *h = hns3_get_handle(ndev);
2830 	struct hns3_enet_ring *tx_ring;
2831 	int timeout_queue;
2832 
2833 	timeout_queue = hns3_get_timeout_queue(ndev);
2834 	if (timeout_queue >= ndev->num_tx_queues) {
2835 		netdev_info(ndev,
2836 			    "no netdev TX timeout queue found, timeout count: %llu\n",
2837 			    priv->tx_timeout_count);
2838 		return false;
2839 	}
2840 
2841 	priv->tx_timeout_count++;
2842 
2843 	tx_ring = &priv->ring[timeout_queue];
2844 	hns3_dump_queue_stats(ndev, tx_ring, timeout_queue);
2845 
2846 	/* When mac received many pause frames continuous, it's unable to send
2847 	 * packets, which may cause tx timeout
2848 	 */
2849 	if (h->ae_algo->ops->get_mac_stats) {
2850 		struct hns3_mac_stats mac_stats;
2851 
2852 		h->ae_algo->ops->get_mac_stats(h, &mac_stats);
2853 		netdev_info(ndev, "tx_pause_cnt: %llu, rx_pause_cnt: %llu\n",
2854 			    mac_stats.tx_pause_cnt, mac_stats.rx_pause_cnt);
2855 	}
2856 
2857 	hns3_dump_queue_reg(ndev, tx_ring);
2858 
2859 	return true;
2860 }
2861 
2862 static void hns3_nic_net_timeout(struct net_device *ndev, unsigned int txqueue)
2863 {
2864 	struct hns3_nic_priv *priv = netdev_priv(ndev);
2865 	struct hnae3_handle *h = priv->ae_handle;
2866 
2867 	if (!hns3_get_tx_timeo_queue_info(ndev))
2868 		return;
2869 
2870 	/* request the reset, and let the hclge to determine
2871 	 * which reset level should be done
2872 	 */
2873 	if (h->ae_algo->ops->reset_event)
2874 		h->ae_algo->ops->reset_event(h->pdev, h);
2875 }
2876 
2877 #ifdef CONFIG_RFS_ACCEL
2878 static int hns3_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
2879 			      u16 rxq_index, u32 flow_id)
2880 {
2881 	struct hnae3_handle *h = hns3_get_handle(dev);
2882 	struct flow_keys fkeys;
2883 
2884 	if (!h->ae_algo->ops->add_arfs_entry)
2885 		return -EOPNOTSUPP;
2886 
2887 	if (skb->encapsulation)
2888 		return -EPROTONOSUPPORT;
2889 
2890 	if (!skb_flow_dissect_flow_keys(skb, &fkeys, 0))
2891 		return -EPROTONOSUPPORT;
2892 
2893 	if ((fkeys.basic.n_proto != htons(ETH_P_IP) &&
2894 	     fkeys.basic.n_proto != htons(ETH_P_IPV6)) ||
2895 	    (fkeys.basic.ip_proto != IPPROTO_TCP &&
2896 	     fkeys.basic.ip_proto != IPPROTO_UDP))
2897 		return -EPROTONOSUPPORT;
2898 
2899 	return h->ae_algo->ops->add_arfs_entry(h, rxq_index, flow_id, &fkeys);
2900 }
2901 #endif
2902 
2903 static int hns3_nic_get_vf_config(struct net_device *ndev, int vf,
2904 				  struct ifla_vf_info *ivf)
2905 {
2906 	struct hnae3_handle *h = hns3_get_handle(ndev);
2907 
2908 	if (!h->ae_algo->ops->get_vf_config)
2909 		return -EOPNOTSUPP;
2910 
2911 	return h->ae_algo->ops->get_vf_config(h, vf, ivf);
2912 }
2913 
2914 static int hns3_nic_set_vf_link_state(struct net_device *ndev, int vf,
2915 				      int link_state)
2916 {
2917 	struct hnae3_handle *h = hns3_get_handle(ndev);
2918 
2919 	if (!h->ae_algo->ops->set_vf_link_state)
2920 		return -EOPNOTSUPP;
2921 
2922 	return h->ae_algo->ops->set_vf_link_state(h, vf, link_state);
2923 }
2924 
2925 static int hns3_nic_set_vf_rate(struct net_device *ndev, int vf,
2926 				int min_tx_rate, int max_tx_rate)
2927 {
2928 	struct hnae3_handle *h = hns3_get_handle(ndev);
2929 
2930 	if (!h->ae_algo->ops->set_vf_rate)
2931 		return -EOPNOTSUPP;
2932 
2933 	return h->ae_algo->ops->set_vf_rate(h, vf, min_tx_rate, max_tx_rate,
2934 					    false);
2935 }
2936 
2937 static int hns3_nic_set_vf_mac(struct net_device *netdev, int vf_id, u8 *mac)
2938 {
2939 	struct hnae3_handle *h = hns3_get_handle(netdev);
2940 	char format_mac_addr[HNAE3_FORMAT_MAC_ADDR_LEN];
2941 
2942 	if (!h->ae_algo->ops->set_vf_mac)
2943 		return -EOPNOTSUPP;
2944 
2945 	if (is_multicast_ether_addr(mac)) {
2946 		hnae3_format_mac_addr(format_mac_addr, mac);
2947 		netdev_err(netdev,
2948 			   "Invalid MAC:%s specified. Could not set MAC\n",
2949 			   format_mac_addr);
2950 		return -EINVAL;
2951 	}
2952 
2953 	return h->ae_algo->ops->set_vf_mac(h, vf_id, mac);
2954 }
2955 
2956 static const struct net_device_ops hns3_nic_netdev_ops = {
2957 	.ndo_open		= hns3_nic_net_open,
2958 	.ndo_stop		= hns3_nic_net_stop,
2959 	.ndo_start_xmit		= hns3_nic_net_xmit,
2960 	.ndo_tx_timeout		= hns3_nic_net_timeout,
2961 	.ndo_set_mac_address	= hns3_nic_net_set_mac_address,
2962 	.ndo_eth_ioctl		= hns3_nic_do_ioctl,
2963 	.ndo_change_mtu		= hns3_nic_change_mtu,
2964 	.ndo_set_features	= hns3_nic_set_features,
2965 	.ndo_features_check	= hns3_features_check,
2966 	.ndo_get_stats64	= hns3_nic_get_stats64,
2967 	.ndo_setup_tc		= hns3_nic_setup_tc,
2968 	.ndo_set_rx_mode	= hns3_nic_set_rx_mode,
2969 	.ndo_vlan_rx_add_vid	= hns3_vlan_rx_add_vid,
2970 	.ndo_vlan_rx_kill_vid	= hns3_vlan_rx_kill_vid,
2971 	.ndo_set_vf_vlan	= hns3_ndo_set_vf_vlan,
2972 	.ndo_set_vf_spoofchk	= hns3_set_vf_spoofchk,
2973 	.ndo_set_vf_trust	= hns3_set_vf_trust,
2974 #ifdef CONFIG_RFS_ACCEL
2975 	.ndo_rx_flow_steer	= hns3_rx_flow_steer,
2976 #endif
2977 	.ndo_get_vf_config	= hns3_nic_get_vf_config,
2978 	.ndo_set_vf_link_state	= hns3_nic_set_vf_link_state,
2979 	.ndo_set_vf_rate	= hns3_nic_set_vf_rate,
2980 	.ndo_set_vf_mac		= hns3_nic_set_vf_mac,
2981 };
2982 
2983 bool hns3_is_phys_func(struct pci_dev *pdev)
2984 {
2985 	u32 dev_id = pdev->device;
2986 
2987 	switch (dev_id) {
2988 	case HNAE3_DEV_ID_GE:
2989 	case HNAE3_DEV_ID_25GE:
2990 	case HNAE3_DEV_ID_25GE_RDMA:
2991 	case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
2992 	case HNAE3_DEV_ID_50GE_RDMA:
2993 	case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
2994 	case HNAE3_DEV_ID_100G_RDMA_MACSEC:
2995 	case HNAE3_DEV_ID_200G_RDMA:
2996 		return true;
2997 	case HNAE3_DEV_ID_VF:
2998 	case HNAE3_DEV_ID_RDMA_DCB_PFC_VF:
2999 		return false;
3000 	default:
3001 		dev_warn(&pdev->dev, "un-recognized pci device-id %u",
3002 			 dev_id);
3003 	}
3004 
3005 	return false;
3006 }
3007 
3008 static void hns3_disable_sriov(struct pci_dev *pdev)
3009 {
3010 	/* If our VFs are assigned we cannot shut down SR-IOV
3011 	 * without causing issues, so just leave the hardware
3012 	 * available but disabled
3013 	 */
3014 	if (pci_vfs_assigned(pdev)) {
3015 		dev_warn(&pdev->dev,
3016 			 "disabling driver while VFs are assigned\n");
3017 		return;
3018 	}
3019 
3020 	pci_disable_sriov(pdev);
3021 }
3022 
3023 /* hns3_probe - Device initialization routine
3024  * @pdev: PCI device information struct
3025  * @ent: entry in hns3_pci_tbl
3026  *
3027  * hns3_probe initializes a PF identified by a pci_dev structure.
3028  * The OS initialization, configuring of the PF private structure,
3029  * and a hardware reset occur.
3030  *
3031  * Returns 0 on success, negative on failure
3032  */
3033 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3034 {
3035 	struct hnae3_ae_dev *ae_dev;
3036 	int ret;
3037 
3038 	ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev), GFP_KERNEL);
3039 	if (!ae_dev)
3040 		return -ENOMEM;
3041 
3042 	ae_dev->pdev = pdev;
3043 	ae_dev->flag = ent->driver_data;
3044 	pci_set_drvdata(pdev, ae_dev);
3045 
3046 	ret = hnae3_register_ae_dev(ae_dev);
3047 	if (ret)
3048 		pci_set_drvdata(pdev, NULL);
3049 
3050 	return ret;
3051 }
3052 
3053 /* hns3_remove - Device removal routine
3054  * @pdev: PCI device information struct
3055  */
3056 static void hns3_remove(struct pci_dev *pdev)
3057 {
3058 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3059 
3060 	if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
3061 		hns3_disable_sriov(pdev);
3062 
3063 	hnae3_unregister_ae_dev(ae_dev);
3064 	pci_set_drvdata(pdev, NULL);
3065 }
3066 
3067 /**
3068  * hns3_pci_sriov_configure
3069  * @pdev: pointer to a pci_dev structure
3070  * @num_vfs: number of VFs to allocate
3071  *
3072  * Enable or change the number of VFs. Called when the user updates the number
3073  * of VFs in sysfs.
3074  **/
3075 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
3076 {
3077 	int ret;
3078 
3079 	if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
3080 		dev_warn(&pdev->dev, "Can not config SRIOV\n");
3081 		return -EINVAL;
3082 	}
3083 
3084 	if (num_vfs) {
3085 		ret = pci_enable_sriov(pdev, num_vfs);
3086 		if (ret)
3087 			dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
3088 		else
3089 			return num_vfs;
3090 	} else if (!pci_vfs_assigned(pdev)) {
3091 		pci_disable_sriov(pdev);
3092 	} else {
3093 		dev_warn(&pdev->dev,
3094 			 "Unable to free VFs because some are assigned to VMs.\n");
3095 	}
3096 
3097 	return 0;
3098 }
3099 
3100 static void hns3_shutdown(struct pci_dev *pdev)
3101 {
3102 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3103 
3104 	hnae3_unregister_ae_dev(ae_dev);
3105 	pci_set_drvdata(pdev, NULL);
3106 
3107 	if (system_state == SYSTEM_POWER_OFF)
3108 		pci_set_power_state(pdev, PCI_D3hot);
3109 }
3110 
3111 static int __maybe_unused hns3_suspend(struct device *dev)
3112 {
3113 	struct hnae3_ae_dev *ae_dev = dev_get_drvdata(dev);
3114 
3115 	if (ae_dev && hns3_is_phys_func(ae_dev->pdev)) {
3116 		dev_info(dev, "Begin to suspend.\n");
3117 		if (ae_dev->ops && ae_dev->ops->reset_prepare)
3118 			ae_dev->ops->reset_prepare(ae_dev, HNAE3_FUNC_RESET);
3119 	}
3120 
3121 	return 0;
3122 }
3123 
3124 static int __maybe_unused hns3_resume(struct device *dev)
3125 {
3126 	struct hnae3_ae_dev *ae_dev = dev_get_drvdata(dev);
3127 
3128 	if (ae_dev && hns3_is_phys_func(ae_dev->pdev)) {
3129 		dev_info(dev, "Begin to resume.\n");
3130 		if (ae_dev->ops && ae_dev->ops->reset_done)
3131 			ae_dev->ops->reset_done(ae_dev);
3132 	}
3133 
3134 	return 0;
3135 }
3136 
3137 static pci_ers_result_t hns3_error_detected(struct pci_dev *pdev,
3138 					    pci_channel_state_t state)
3139 {
3140 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3141 	pci_ers_result_t ret;
3142 
3143 	dev_info(&pdev->dev, "PCI error detected, state(=%u)!!\n", state);
3144 
3145 	if (state == pci_channel_io_perm_failure)
3146 		return PCI_ERS_RESULT_DISCONNECT;
3147 
3148 	if (!ae_dev || !ae_dev->ops) {
3149 		dev_err(&pdev->dev,
3150 			"Can't recover - error happened before device initialized\n");
3151 		return PCI_ERS_RESULT_NONE;
3152 	}
3153 
3154 	if (ae_dev->ops->handle_hw_ras_error)
3155 		ret = ae_dev->ops->handle_hw_ras_error(ae_dev);
3156 	else
3157 		return PCI_ERS_RESULT_NONE;
3158 
3159 	return ret;
3160 }
3161 
3162 static pci_ers_result_t hns3_slot_reset(struct pci_dev *pdev)
3163 {
3164 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3165 	const struct hnae3_ae_ops *ops;
3166 	enum hnae3_reset_type reset_type;
3167 	struct device *dev = &pdev->dev;
3168 
3169 	if (!ae_dev || !ae_dev->ops)
3170 		return PCI_ERS_RESULT_NONE;
3171 
3172 	ops = ae_dev->ops;
3173 	/* request the reset */
3174 	if (ops->reset_event && ops->get_reset_level &&
3175 	    ops->set_default_reset_request) {
3176 		if (ae_dev->hw_err_reset_req) {
3177 			reset_type = ops->get_reset_level(ae_dev,
3178 						&ae_dev->hw_err_reset_req);
3179 			ops->set_default_reset_request(ae_dev, reset_type);
3180 			dev_info(dev, "requesting reset due to PCI error\n");
3181 			ops->reset_event(pdev, NULL);
3182 		}
3183 
3184 		return PCI_ERS_RESULT_RECOVERED;
3185 	}
3186 
3187 	return PCI_ERS_RESULT_DISCONNECT;
3188 }
3189 
3190 static void hns3_reset_prepare(struct pci_dev *pdev)
3191 {
3192 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3193 
3194 	dev_info(&pdev->dev, "FLR prepare\n");
3195 	if (ae_dev && ae_dev->ops && ae_dev->ops->reset_prepare)
3196 		ae_dev->ops->reset_prepare(ae_dev, HNAE3_FLR_RESET);
3197 }
3198 
3199 static void hns3_reset_done(struct pci_dev *pdev)
3200 {
3201 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3202 
3203 	dev_info(&pdev->dev, "FLR done\n");
3204 	if (ae_dev && ae_dev->ops && ae_dev->ops->reset_done)
3205 		ae_dev->ops->reset_done(ae_dev);
3206 }
3207 
3208 static const struct pci_error_handlers hns3_err_handler = {
3209 	.error_detected = hns3_error_detected,
3210 	.slot_reset     = hns3_slot_reset,
3211 	.reset_prepare	= hns3_reset_prepare,
3212 	.reset_done	= hns3_reset_done,
3213 };
3214 
3215 static SIMPLE_DEV_PM_OPS(hns3_pm_ops, hns3_suspend, hns3_resume);
3216 
3217 static struct pci_driver hns3_driver = {
3218 	.name     = hns3_driver_name,
3219 	.id_table = hns3_pci_tbl,
3220 	.probe    = hns3_probe,
3221 	.remove   = hns3_remove,
3222 	.shutdown = hns3_shutdown,
3223 	.driver.pm  = &hns3_pm_ops,
3224 	.sriov_configure = hns3_pci_sriov_configure,
3225 	.err_handler    = &hns3_err_handler,
3226 };
3227 
3228 /* set default feature to hns3 */
3229 static void hns3_set_default_feature(struct net_device *netdev)
3230 {
3231 	struct hnae3_handle *h = hns3_get_handle(netdev);
3232 	struct pci_dev *pdev = h->pdev;
3233 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3234 
3235 	netdev->priv_flags |= IFF_UNICAST_FLT;
3236 
3237 	netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
3238 
3239 	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
3240 		NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3241 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
3242 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
3243 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
3244 		NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
3245 
3246 	if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V2) {
3247 		netdev->features |= NETIF_F_GRO_HW;
3248 
3249 		if (!(h->flags & HNAE3_SUPPORT_VF))
3250 			netdev->features |= NETIF_F_NTUPLE;
3251 	}
3252 
3253 	if (test_bit(HNAE3_DEV_SUPPORT_UDP_GSO_B, ae_dev->caps))
3254 		netdev->features |= NETIF_F_GSO_UDP_L4;
3255 
3256 	if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps))
3257 		netdev->features |= NETIF_F_HW_CSUM;
3258 	else
3259 		netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3260 
3261 	if (test_bit(HNAE3_DEV_SUPPORT_UDP_TUNNEL_CSUM_B, ae_dev->caps))
3262 		netdev->features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
3263 
3264 	if (test_bit(HNAE3_DEV_SUPPORT_FD_FORWARD_TC_B, ae_dev->caps))
3265 		netdev->features |= NETIF_F_HW_TC;
3266 
3267 	netdev->hw_features |= netdev->features;
3268 	if (!test_bit(HNAE3_DEV_SUPPORT_VLAN_FLTR_MDF_B, ae_dev->caps))
3269 		netdev->hw_features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
3270 
3271 	netdev->vlan_features |= netdev->features &
3272 		~(NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_CTAG_TX |
3273 		  NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_GRO_HW | NETIF_F_NTUPLE |
3274 		  NETIF_F_HW_TC);
3275 
3276 	netdev->hw_enc_features |= netdev->vlan_features | NETIF_F_TSO_MANGLEID;
3277 }
3278 
3279 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
3280 			     struct hns3_desc_cb *cb)
3281 {
3282 	unsigned int order = hns3_page_order(ring);
3283 	struct page *p;
3284 
3285 	if (ring->page_pool) {
3286 		p = page_pool_dev_alloc_frag(ring->page_pool,
3287 					     &cb->page_offset,
3288 					     hns3_buf_size(ring));
3289 		if (unlikely(!p))
3290 			return -ENOMEM;
3291 
3292 		cb->priv = p;
3293 		cb->buf = page_address(p);
3294 		cb->dma = page_pool_get_dma_addr(p);
3295 		cb->type = DESC_TYPE_PP_FRAG;
3296 		cb->reuse_flag = 0;
3297 		return 0;
3298 	}
3299 
3300 	p = dev_alloc_pages(order);
3301 	if (!p)
3302 		return -ENOMEM;
3303 
3304 	cb->priv = p;
3305 	cb->page_offset = 0;
3306 	cb->reuse_flag = 0;
3307 	cb->buf  = page_address(p);
3308 	cb->length = hns3_page_size(ring);
3309 	cb->type = DESC_TYPE_PAGE;
3310 	page_ref_add(p, USHRT_MAX - 1);
3311 	cb->pagecnt_bias = USHRT_MAX;
3312 
3313 	return 0;
3314 }
3315 
3316 static void hns3_free_buffer(struct hns3_enet_ring *ring,
3317 			     struct hns3_desc_cb *cb, int budget)
3318 {
3319 	if (cb->type & (DESC_TYPE_SKB | DESC_TYPE_BOUNCE_HEAD |
3320 			DESC_TYPE_BOUNCE_ALL | DESC_TYPE_SGL_SKB))
3321 		napi_consume_skb(cb->priv, budget);
3322 	else if (!HNAE3_IS_TX_RING(ring)) {
3323 		if (cb->type & DESC_TYPE_PAGE && cb->pagecnt_bias)
3324 			__page_frag_cache_drain(cb->priv, cb->pagecnt_bias);
3325 		else if (cb->type & DESC_TYPE_PP_FRAG)
3326 			page_pool_put_full_page(ring->page_pool, cb->priv,
3327 						false);
3328 	}
3329 	memset(cb, 0, sizeof(*cb));
3330 }
3331 
3332 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
3333 {
3334 	cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
3335 			       cb->length, ring_to_dma_dir(ring));
3336 
3337 	if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
3338 		return -EIO;
3339 
3340 	return 0;
3341 }
3342 
3343 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
3344 			      struct hns3_desc_cb *cb)
3345 {
3346 	if (cb->type & (DESC_TYPE_SKB | DESC_TYPE_FRAGLIST_SKB))
3347 		dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
3348 				 ring_to_dma_dir(ring));
3349 	else if ((cb->type & DESC_TYPE_PAGE) && cb->length)
3350 		dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
3351 			       ring_to_dma_dir(ring));
3352 	else if (cb->type & (DESC_TYPE_BOUNCE_ALL | DESC_TYPE_BOUNCE_HEAD |
3353 			     DESC_TYPE_SGL_SKB))
3354 		hns3_tx_spare_reclaim_cb(ring, cb);
3355 }
3356 
3357 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
3358 {
3359 	hns3_unmap_buffer(ring, &ring->desc_cb[i]);
3360 	ring->desc[i].addr = 0;
3361 	ring->desc_cb[i].refill = 0;
3362 }
3363 
3364 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i,
3365 				    int budget)
3366 {
3367 	struct hns3_desc_cb *cb = &ring->desc_cb[i];
3368 
3369 	if (!ring->desc_cb[i].dma)
3370 		return;
3371 
3372 	hns3_buffer_detach(ring, i);
3373 	hns3_free_buffer(ring, cb, budget);
3374 }
3375 
3376 static void hns3_free_buffers(struct hns3_enet_ring *ring)
3377 {
3378 	int i;
3379 
3380 	for (i = 0; i < ring->desc_num; i++)
3381 		hns3_free_buffer_detach(ring, i, 0);
3382 }
3383 
3384 /* free desc along with its attached buffer */
3385 static void hns3_free_desc(struct hns3_enet_ring *ring)
3386 {
3387 	int size = ring->desc_num * sizeof(ring->desc[0]);
3388 
3389 	hns3_free_buffers(ring);
3390 
3391 	if (ring->desc) {
3392 		dma_free_coherent(ring_to_dev(ring), size,
3393 				  ring->desc, ring->desc_dma_addr);
3394 		ring->desc = NULL;
3395 	}
3396 }
3397 
3398 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
3399 {
3400 	int size = ring->desc_num * sizeof(ring->desc[0]);
3401 
3402 	ring->desc = dma_alloc_coherent(ring_to_dev(ring), size,
3403 					&ring->desc_dma_addr, GFP_KERNEL);
3404 	if (!ring->desc)
3405 		return -ENOMEM;
3406 
3407 	return 0;
3408 }
3409 
3410 static int hns3_alloc_and_map_buffer(struct hns3_enet_ring *ring,
3411 				   struct hns3_desc_cb *cb)
3412 {
3413 	int ret;
3414 
3415 	ret = hns3_alloc_buffer(ring, cb);
3416 	if (ret || ring->page_pool)
3417 		goto out;
3418 
3419 	ret = hns3_map_buffer(ring, cb);
3420 	if (ret)
3421 		goto out_with_buf;
3422 
3423 	return 0;
3424 
3425 out_with_buf:
3426 	hns3_free_buffer(ring, cb, 0);
3427 out:
3428 	return ret;
3429 }
3430 
3431 static int hns3_alloc_and_attach_buffer(struct hns3_enet_ring *ring, int i)
3432 {
3433 	int ret = hns3_alloc_and_map_buffer(ring, &ring->desc_cb[i]);
3434 
3435 	if (ret)
3436 		return ret;
3437 
3438 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3439 					 ring->desc_cb[i].page_offset);
3440 	ring->desc_cb[i].refill = 1;
3441 
3442 	return 0;
3443 }
3444 
3445 /* Allocate memory for raw pkg, and map with dma */
3446 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
3447 {
3448 	int i, j, ret;
3449 
3450 	for (i = 0; i < ring->desc_num; i++) {
3451 		ret = hns3_alloc_and_attach_buffer(ring, i);
3452 		if (ret)
3453 			goto out_buffer_fail;
3454 	}
3455 
3456 	return 0;
3457 
3458 out_buffer_fail:
3459 	for (j = i - 1; j >= 0; j--)
3460 		hns3_free_buffer_detach(ring, j, 0);
3461 	return ret;
3462 }
3463 
3464 /* detach a in-used buffer and replace with a reserved one */
3465 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
3466 				struct hns3_desc_cb *res_cb)
3467 {
3468 	hns3_unmap_buffer(ring, &ring->desc_cb[i]);
3469 	ring->desc_cb[i] = *res_cb;
3470 	ring->desc_cb[i].refill = 1;
3471 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3472 					 ring->desc_cb[i].page_offset);
3473 	ring->desc[i].rx.bd_base_info = 0;
3474 }
3475 
3476 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
3477 {
3478 	ring->desc_cb[i].reuse_flag = 0;
3479 	ring->desc_cb[i].refill = 1;
3480 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3481 					 ring->desc_cb[i].page_offset);
3482 	ring->desc[i].rx.bd_base_info = 0;
3483 
3484 	dma_sync_single_for_device(ring_to_dev(ring),
3485 			ring->desc_cb[i].dma + ring->desc_cb[i].page_offset,
3486 			hns3_buf_size(ring),
3487 			DMA_FROM_DEVICE);
3488 }
3489 
3490 static bool hns3_nic_reclaim_desc(struct hns3_enet_ring *ring,
3491 				  int *bytes, int *pkts, int budget)
3492 {
3493 	/* pair with ring->last_to_use update in hns3_tx_doorbell(),
3494 	 * smp_store_release() is not used in hns3_tx_doorbell() because
3495 	 * the doorbell operation already have the needed barrier operation.
3496 	 */
3497 	int ltu = smp_load_acquire(&ring->last_to_use);
3498 	int ntc = ring->next_to_clean;
3499 	struct hns3_desc_cb *desc_cb;
3500 	bool reclaimed = false;
3501 	struct hns3_desc *desc;
3502 
3503 	while (ltu != ntc) {
3504 		desc = &ring->desc[ntc];
3505 
3506 		if (le16_to_cpu(desc->tx.bdtp_fe_sc_vld_ra_ri) &
3507 				BIT(HNS3_TXD_VLD_B))
3508 			break;
3509 
3510 		desc_cb = &ring->desc_cb[ntc];
3511 
3512 		if (desc_cb->type & (DESC_TYPE_SKB | DESC_TYPE_BOUNCE_ALL |
3513 				     DESC_TYPE_BOUNCE_HEAD |
3514 				     DESC_TYPE_SGL_SKB)) {
3515 			(*pkts)++;
3516 			(*bytes) += desc_cb->send_bytes;
3517 		}
3518 
3519 		/* desc_cb will be cleaned, after hnae3_free_buffer_detach */
3520 		hns3_free_buffer_detach(ring, ntc, budget);
3521 
3522 		if (++ntc == ring->desc_num)
3523 			ntc = 0;
3524 
3525 		/* Issue prefetch for next Tx descriptor */
3526 		prefetch(&ring->desc_cb[ntc]);
3527 		reclaimed = true;
3528 	}
3529 
3530 	if (unlikely(!reclaimed))
3531 		return false;
3532 
3533 	/* This smp_store_release() pairs with smp_load_acquire() in
3534 	 * ring_space called by hns3_nic_net_xmit.
3535 	 */
3536 	smp_store_release(&ring->next_to_clean, ntc);
3537 
3538 	hns3_tx_spare_update(ring);
3539 
3540 	return true;
3541 }
3542 
3543 void hns3_clean_tx_ring(struct hns3_enet_ring *ring, int budget)
3544 {
3545 	struct net_device *netdev = ring_to_netdev(ring);
3546 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3547 	struct netdev_queue *dev_queue;
3548 	int bytes, pkts;
3549 
3550 	bytes = 0;
3551 	pkts = 0;
3552 
3553 	if (unlikely(!hns3_nic_reclaim_desc(ring, &bytes, &pkts, budget)))
3554 		return;
3555 
3556 	ring->tqp_vector->tx_group.total_bytes += bytes;
3557 	ring->tqp_vector->tx_group.total_packets += pkts;
3558 
3559 	u64_stats_update_begin(&ring->syncp);
3560 	ring->stats.tx_bytes += bytes;
3561 	ring->stats.tx_pkts += pkts;
3562 	u64_stats_update_end(&ring->syncp);
3563 
3564 	dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
3565 	netdev_tx_completed_queue(dev_queue, pkts, bytes);
3566 
3567 	if (unlikely(netif_carrier_ok(netdev) &&
3568 		     ring_space(ring) > HNS3_MAX_TSO_BD_NUM)) {
3569 		/* Make sure that anybody stopping the queue after this
3570 		 * sees the new next_to_clean.
3571 		 */
3572 		smp_mb();
3573 		if (netif_tx_queue_stopped(dev_queue) &&
3574 		    !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
3575 			netif_tx_wake_queue(dev_queue);
3576 			ring->stats.restart_queue++;
3577 		}
3578 	}
3579 }
3580 
3581 static int hns3_desc_unused(struct hns3_enet_ring *ring)
3582 {
3583 	int ntc = ring->next_to_clean;
3584 	int ntu = ring->next_to_use;
3585 
3586 	if (unlikely(ntc == ntu && !ring->desc_cb[ntc].refill))
3587 		return ring->desc_num;
3588 
3589 	return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
3590 }
3591 
3592 /* Return true if there is any allocation failure */
3593 static bool hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring,
3594 				      int cleand_count)
3595 {
3596 	struct hns3_desc_cb *desc_cb;
3597 	struct hns3_desc_cb res_cbs;
3598 	int i, ret;
3599 
3600 	for (i = 0; i < cleand_count; i++) {
3601 		desc_cb = &ring->desc_cb[ring->next_to_use];
3602 		if (desc_cb->reuse_flag) {
3603 			hns3_ring_stats_update(ring, reuse_pg_cnt);
3604 
3605 			hns3_reuse_buffer(ring, ring->next_to_use);
3606 		} else {
3607 			ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
3608 			if (ret) {
3609 				hns3_ring_stats_update(ring, sw_err_cnt);
3610 
3611 				hns3_rl_err(ring_to_netdev(ring),
3612 					    "alloc rx buffer failed: %d\n",
3613 					    ret);
3614 
3615 				writel(i, ring->tqp->io_base +
3616 				       HNS3_RING_RX_RING_HEAD_REG);
3617 				return true;
3618 			}
3619 			hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
3620 
3621 			hns3_ring_stats_update(ring, non_reuse_pg);
3622 		}
3623 
3624 		ring_ptr_move_fw(ring, next_to_use);
3625 	}
3626 
3627 	writel(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
3628 	return false;
3629 }
3630 
3631 static bool hns3_can_reuse_page(struct hns3_desc_cb *cb)
3632 {
3633 	return page_count(cb->priv) == cb->pagecnt_bias;
3634 }
3635 
3636 static int hns3_handle_rx_copybreak(struct sk_buff *skb, int i,
3637 				    struct hns3_enet_ring *ring,
3638 				    int pull_len,
3639 				    struct hns3_desc_cb *desc_cb)
3640 {
3641 	struct hns3_desc *desc = &ring->desc[ring->next_to_clean];
3642 	u32 frag_offset = desc_cb->page_offset + pull_len;
3643 	int size = le16_to_cpu(desc->rx.size);
3644 	u32 frag_size = size - pull_len;
3645 	void *frag = napi_alloc_frag(frag_size);
3646 
3647 	if (unlikely(!frag)) {
3648 		hns3_ring_stats_update(ring, frag_alloc_err);
3649 
3650 		hns3_rl_err(ring_to_netdev(ring),
3651 			    "failed to allocate rx frag\n");
3652 		return -ENOMEM;
3653 	}
3654 
3655 	desc_cb->reuse_flag = 1;
3656 	memcpy(frag, desc_cb->buf + frag_offset, frag_size);
3657 	skb_add_rx_frag(skb, i, virt_to_page(frag),
3658 			offset_in_page(frag), frag_size, frag_size);
3659 
3660 	hns3_ring_stats_update(ring, frag_alloc);
3661 	return 0;
3662 }
3663 
3664 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
3665 				struct hns3_enet_ring *ring, int pull_len,
3666 				struct hns3_desc_cb *desc_cb)
3667 {
3668 	struct hns3_desc *desc = &ring->desc[ring->next_to_clean];
3669 	u32 frag_offset = desc_cb->page_offset + pull_len;
3670 	int size = le16_to_cpu(desc->rx.size);
3671 	u32 truesize = hns3_buf_size(ring);
3672 	u32 frag_size = size - pull_len;
3673 	int ret = 0;
3674 	bool reused;
3675 
3676 	if (ring->page_pool) {
3677 		skb_add_rx_frag(skb, i, desc_cb->priv, frag_offset,
3678 				frag_size, truesize);
3679 		return;
3680 	}
3681 
3682 	/* Avoid re-using remote or pfmem page */
3683 	if (unlikely(!dev_page_is_reusable(desc_cb->priv)))
3684 		goto out;
3685 
3686 	reused = hns3_can_reuse_page(desc_cb);
3687 
3688 	/* Rx page can be reused when:
3689 	 * 1. Rx page is only owned by the driver when page_offset
3690 	 *    is zero, which means 0 @ truesize will be used by
3691 	 *    stack after skb_add_rx_frag() is called, and the rest
3692 	 *    of rx page can be reused by driver.
3693 	 * Or
3694 	 * 2. Rx page is only owned by the driver when page_offset
3695 	 *    is non-zero, which means page_offset @ truesize will
3696 	 *    be used by stack after skb_add_rx_frag() is called,
3697 	 *    and 0 @ truesize can be reused by driver.
3698 	 */
3699 	if ((!desc_cb->page_offset && reused) ||
3700 	    ((desc_cb->page_offset + truesize + truesize) <=
3701 	     hns3_page_size(ring) && desc_cb->page_offset)) {
3702 		desc_cb->page_offset += truesize;
3703 		desc_cb->reuse_flag = 1;
3704 	} else if (desc_cb->page_offset && reused) {
3705 		desc_cb->page_offset = 0;
3706 		desc_cb->reuse_flag = 1;
3707 	} else if (frag_size <= ring->rx_copybreak) {
3708 		ret = hns3_handle_rx_copybreak(skb, i, ring, pull_len, desc_cb);
3709 		if (ret)
3710 			goto out;
3711 	}
3712 
3713 out:
3714 	desc_cb->pagecnt_bias--;
3715 
3716 	if (unlikely(!desc_cb->pagecnt_bias)) {
3717 		page_ref_add(desc_cb->priv, USHRT_MAX);
3718 		desc_cb->pagecnt_bias = USHRT_MAX;
3719 	}
3720 
3721 	skb_add_rx_frag(skb, i, desc_cb->priv, frag_offset,
3722 			frag_size, truesize);
3723 
3724 	if (unlikely(!desc_cb->reuse_flag))
3725 		__page_frag_cache_drain(desc_cb->priv, desc_cb->pagecnt_bias);
3726 }
3727 
3728 static int hns3_gro_complete(struct sk_buff *skb, u32 l234info)
3729 {
3730 	__be16 type = skb->protocol;
3731 	struct tcphdr *th;
3732 	int depth = 0;
3733 
3734 	while (eth_type_vlan(type)) {
3735 		struct vlan_hdr *vh;
3736 
3737 		if ((depth + VLAN_HLEN) > skb_headlen(skb))
3738 			return -EFAULT;
3739 
3740 		vh = (struct vlan_hdr *)(skb->data + depth);
3741 		type = vh->h_vlan_encapsulated_proto;
3742 		depth += VLAN_HLEN;
3743 	}
3744 
3745 	skb_set_network_header(skb, depth);
3746 
3747 	if (type == htons(ETH_P_IP)) {
3748 		const struct iphdr *iph = ip_hdr(skb);
3749 
3750 		depth += sizeof(struct iphdr);
3751 		skb_set_transport_header(skb, depth);
3752 		th = tcp_hdr(skb);
3753 		th->check = ~tcp_v4_check(skb->len - depth, iph->saddr,
3754 					  iph->daddr, 0);
3755 	} else if (type == htons(ETH_P_IPV6)) {
3756 		const struct ipv6hdr *iph = ipv6_hdr(skb);
3757 
3758 		depth += sizeof(struct ipv6hdr);
3759 		skb_set_transport_header(skb, depth);
3760 		th = tcp_hdr(skb);
3761 		th->check = ~tcp_v6_check(skb->len - depth, &iph->saddr,
3762 					  &iph->daddr, 0);
3763 	} else {
3764 		hns3_rl_err(skb->dev,
3765 			    "Error: FW GRO supports only IPv4/IPv6, not 0x%04x, depth: %d\n",
3766 			    be16_to_cpu(type), depth);
3767 		return -EFAULT;
3768 	}
3769 
3770 	skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
3771 	if (th->cwr)
3772 		skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
3773 
3774 	if (l234info & BIT(HNS3_RXD_GRO_FIXID_B))
3775 		skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_FIXEDID;
3776 
3777 	skb->csum_start = (unsigned char *)th - skb->head;
3778 	skb->csum_offset = offsetof(struct tcphdr, check);
3779 	skb->ip_summed = CHECKSUM_PARTIAL;
3780 
3781 	trace_hns3_gro(skb);
3782 
3783 	return 0;
3784 }
3785 
3786 static bool hns3_checksum_complete(struct hns3_enet_ring *ring,
3787 				   struct sk_buff *skb, u32 ptype, u16 csum)
3788 {
3789 	if (ptype == HNS3_INVALID_PTYPE ||
3790 	    hns3_rx_ptype_tbl[ptype].ip_summed != CHECKSUM_COMPLETE)
3791 		return false;
3792 
3793 	hns3_ring_stats_update(ring, csum_complete);
3794 	skb->ip_summed = CHECKSUM_COMPLETE;
3795 	skb->csum = csum_unfold((__force __sum16)csum);
3796 
3797 	return true;
3798 }
3799 
3800 static void hns3_rx_handle_csum(struct sk_buff *skb, u32 l234info,
3801 				u32 ol_info, u32 ptype)
3802 {
3803 	int l3_type, l4_type;
3804 	int ol4_type;
3805 
3806 	if (ptype != HNS3_INVALID_PTYPE) {
3807 		skb->csum_level = hns3_rx_ptype_tbl[ptype].csum_level;
3808 		skb->ip_summed = hns3_rx_ptype_tbl[ptype].ip_summed;
3809 
3810 		return;
3811 	}
3812 
3813 	ol4_type = hnae3_get_field(ol_info, HNS3_RXD_OL4ID_M,
3814 				   HNS3_RXD_OL4ID_S);
3815 	switch (ol4_type) {
3816 	case HNS3_OL4_TYPE_MAC_IN_UDP:
3817 	case HNS3_OL4_TYPE_NVGRE:
3818 		skb->csum_level = 1;
3819 		fallthrough;
3820 	case HNS3_OL4_TYPE_NO_TUN:
3821 		l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
3822 					  HNS3_RXD_L3ID_S);
3823 		l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
3824 					  HNS3_RXD_L4ID_S);
3825 		/* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
3826 		if ((l3_type == HNS3_L3_TYPE_IPV4 ||
3827 		     l3_type == HNS3_L3_TYPE_IPV6) &&
3828 		    (l4_type == HNS3_L4_TYPE_UDP ||
3829 		     l4_type == HNS3_L4_TYPE_TCP ||
3830 		     l4_type == HNS3_L4_TYPE_SCTP))
3831 			skb->ip_summed = CHECKSUM_UNNECESSARY;
3832 		break;
3833 	default:
3834 		break;
3835 	}
3836 }
3837 
3838 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
3839 			     u32 l234info, u32 bd_base_info, u32 ol_info,
3840 			     u16 csum)
3841 {
3842 	struct net_device *netdev = ring_to_netdev(ring);
3843 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3844 	u32 ptype = HNS3_INVALID_PTYPE;
3845 
3846 	skb->ip_summed = CHECKSUM_NONE;
3847 
3848 	skb_checksum_none_assert(skb);
3849 
3850 	if (!(netdev->features & NETIF_F_RXCSUM))
3851 		return;
3852 
3853 	if (test_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state))
3854 		ptype = hnae3_get_field(ol_info, HNS3_RXD_PTYPE_M,
3855 					HNS3_RXD_PTYPE_S);
3856 
3857 	if (hns3_checksum_complete(ring, skb, ptype, csum))
3858 		return;
3859 
3860 	/* check if hardware has done checksum */
3861 	if (!(bd_base_info & BIT(HNS3_RXD_L3L4P_B)))
3862 		return;
3863 
3864 	if (unlikely(l234info & (BIT(HNS3_RXD_L3E_B) | BIT(HNS3_RXD_L4E_B) |
3865 				 BIT(HNS3_RXD_OL3E_B) |
3866 				 BIT(HNS3_RXD_OL4E_B)))) {
3867 		hns3_ring_stats_update(ring, l3l4_csum_err);
3868 
3869 		return;
3870 	}
3871 
3872 	hns3_rx_handle_csum(skb, l234info, ol_info, ptype);
3873 }
3874 
3875 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
3876 {
3877 	if (skb_has_frag_list(skb))
3878 		napi_gro_flush(&ring->tqp_vector->napi, false);
3879 
3880 	napi_gro_receive(&ring->tqp_vector->napi, skb);
3881 }
3882 
3883 static bool hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
3884 				struct hns3_desc *desc, u32 l234info,
3885 				u16 *vlan_tag)
3886 {
3887 	struct hnae3_handle *handle = ring->tqp->handle;
3888 	struct pci_dev *pdev = ring->tqp->handle->pdev;
3889 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3890 
3891 	if (unlikely(ae_dev->dev_version < HNAE3_DEVICE_VERSION_V2)) {
3892 		*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3893 		if (!(*vlan_tag & VLAN_VID_MASK))
3894 			*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3895 
3896 		return (*vlan_tag != 0);
3897 	}
3898 
3899 #define HNS3_STRP_OUTER_VLAN	0x1
3900 #define HNS3_STRP_INNER_VLAN	0x2
3901 #define HNS3_STRP_BOTH		0x3
3902 
3903 	/* Hardware always insert VLAN tag into RX descriptor when
3904 	 * remove the tag from packet, driver needs to determine
3905 	 * reporting which tag to stack.
3906 	 */
3907 	switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
3908 				HNS3_RXD_STRP_TAGP_S)) {
3909 	case HNS3_STRP_OUTER_VLAN:
3910 		if (handle->port_base_vlan_state !=
3911 				HNAE3_PORT_BASE_VLAN_DISABLE)
3912 			return false;
3913 
3914 		*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3915 		return true;
3916 	case HNS3_STRP_INNER_VLAN:
3917 		if (handle->port_base_vlan_state !=
3918 				HNAE3_PORT_BASE_VLAN_DISABLE)
3919 			return false;
3920 
3921 		*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3922 		return true;
3923 	case HNS3_STRP_BOTH:
3924 		if (handle->port_base_vlan_state ==
3925 				HNAE3_PORT_BASE_VLAN_DISABLE)
3926 			*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3927 		else
3928 			*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3929 
3930 		return true;
3931 	default:
3932 		return false;
3933 	}
3934 }
3935 
3936 static void hns3_rx_ring_move_fw(struct hns3_enet_ring *ring)
3937 {
3938 	ring->desc[ring->next_to_clean].rx.bd_base_info &=
3939 		cpu_to_le32(~BIT(HNS3_RXD_VLD_B));
3940 	ring->desc_cb[ring->next_to_clean].refill = 0;
3941 	ring->next_to_clean += 1;
3942 
3943 	if (unlikely(ring->next_to_clean == ring->desc_num))
3944 		ring->next_to_clean = 0;
3945 }
3946 
3947 static int hns3_alloc_skb(struct hns3_enet_ring *ring, unsigned int length,
3948 			  unsigned char *va)
3949 {
3950 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
3951 	struct net_device *netdev = ring_to_netdev(ring);
3952 	struct sk_buff *skb;
3953 
3954 	ring->skb = napi_alloc_skb(&ring->tqp_vector->napi, HNS3_RX_HEAD_SIZE);
3955 	skb = ring->skb;
3956 	if (unlikely(!skb)) {
3957 		hns3_rl_err(netdev, "alloc rx skb fail\n");
3958 		hns3_ring_stats_update(ring, sw_err_cnt);
3959 
3960 		return -ENOMEM;
3961 	}
3962 
3963 	trace_hns3_rx_desc(ring);
3964 	prefetchw(skb->data);
3965 
3966 	ring->pending_buf = 1;
3967 	ring->frag_num = 0;
3968 	ring->tail_skb = NULL;
3969 	if (length <= HNS3_RX_HEAD_SIZE) {
3970 		memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
3971 
3972 		/* We can reuse buffer as-is, just make sure it is reusable */
3973 		if (dev_page_is_reusable(desc_cb->priv))
3974 			desc_cb->reuse_flag = 1;
3975 		else if (desc_cb->type & DESC_TYPE_PP_FRAG)
3976 			page_pool_put_full_page(ring->page_pool, desc_cb->priv,
3977 						false);
3978 		else /* This page cannot be reused so discard it */
3979 			__page_frag_cache_drain(desc_cb->priv,
3980 						desc_cb->pagecnt_bias);
3981 
3982 		hns3_rx_ring_move_fw(ring);
3983 		return 0;
3984 	}
3985 
3986 	if (ring->page_pool)
3987 		skb_mark_for_recycle(skb);
3988 
3989 	hns3_ring_stats_update(ring, seg_pkt_cnt);
3990 
3991 	ring->pull_len = eth_get_headlen(netdev, va, HNS3_RX_HEAD_SIZE);
3992 	__skb_put(skb, ring->pull_len);
3993 	hns3_nic_reuse_page(skb, ring->frag_num++, ring, ring->pull_len,
3994 			    desc_cb);
3995 	hns3_rx_ring_move_fw(ring);
3996 
3997 	return 0;
3998 }
3999 
4000 static int hns3_add_frag(struct hns3_enet_ring *ring)
4001 {
4002 	struct sk_buff *skb = ring->skb;
4003 	struct sk_buff *head_skb = skb;
4004 	struct sk_buff *new_skb;
4005 	struct hns3_desc_cb *desc_cb;
4006 	struct hns3_desc *desc;
4007 	u32 bd_base_info;
4008 
4009 	do {
4010 		desc = &ring->desc[ring->next_to_clean];
4011 		desc_cb = &ring->desc_cb[ring->next_to_clean];
4012 		bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
4013 		/* make sure HW write desc complete */
4014 		dma_rmb();
4015 		if (!(bd_base_info & BIT(HNS3_RXD_VLD_B)))
4016 			return -ENXIO;
4017 
4018 		if (unlikely(ring->frag_num >= MAX_SKB_FRAGS)) {
4019 			new_skb = napi_alloc_skb(&ring->tqp_vector->napi, 0);
4020 			if (unlikely(!new_skb)) {
4021 				hns3_rl_err(ring_to_netdev(ring),
4022 					    "alloc rx fraglist skb fail\n");
4023 				return -ENXIO;
4024 			}
4025 
4026 			if (ring->page_pool)
4027 				skb_mark_for_recycle(new_skb);
4028 
4029 			ring->frag_num = 0;
4030 
4031 			if (ring->tail_skb) {
4032 				ring->tail_skb->next = new_skb;
4033 				ring->tail_skb = new_skb;
4034 			} else {
4035 				skb_shinfo(skb)->frag_list = new_skb;
4036 				ring->tail_skb = new_skb;
4037 			}
4038 		}
4039 
4040 		if (ring->tail_skb) {
4041 			head_skb->truesize += hns3_buf_size(ring);
4042 			head_skb->data_len += le16_to_cpu(desc->rx.size);
4043 			head_skb->len += le16_to_cpu(desc->rx.size);
4044 			skb = ring->tail_skb;
4045 		}
4046 
4047 		dma_sync_single_for_cpu(ring_to_dev(ring),
4048 				desc_cb->dma + desc_cb->page_offset,
4049 				hns3_buf_size(ring),
4050 				DMA_FROM_DEVICE);
4051 
4052 		hns3_nic_reuse_page(skb, ring->frag_num++, ring, 0, desc_cb);
4053 		trace_hns3_rx_desc(ring);
4054 		hns3_rx_ring_move_fw(ring);
4055 		ring->pending_buf++;
4056 	} while (!(bd_base_info & BIT(HNS3_RXD_FE_B)));
4057 
4058 	return 0;
4059 }
4060 
4061 static int hns3_set_gro_and_checksum(struct hns3_enet_ring *ring,
4062 				     struct sk_buff *skb, u32 l234info,
4063 				     u32 bd_base_info, u32 ol_info, u16 csum)
4064 {
4065 	struct net_device *netdev = ring_to_netdev(ring);
4066 	struct hns3_nic_priv *priv = netdev_priv(netdev);
4067 	u32 l3_type;
4068 
4069 	skb_shinfo(skb)->gso_size = hnae3_get_field(bd_base_info,
4070 						    HNS3_RXD_GRO_SIZE_M,
4071 						    HNS3_RXD_GRO_SIZE_S);
4072 	/* if there is no HW GRO, do not set gro params */
4073 	if (!skb_shinfo(skb)->gso_size) {
4074 		hns3_rx_checksum(ring, skb, l234info, bd_base_info, ol_info,
4075 				 csum);
4076 		return 0;
4077 	}
4078 
4079 	NAPI_GRO_CB(skb)->count = hnae3_get_field(l234info,
4080 						  HNS3_RXD_GRO_COUNT_M,
4081 						  HNS3_RXD_GRO_COUNT_S);
4082 
4083 	if (test_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state)) {
4084 		u32 ptype = hnae3_get_field(ol_info, HNS3_RXD_PTYPE_M,
4085 					    HNS3_RXD_PTYPE_S);
4086 
4087 		l3_type = hns3_rx_ptype_tbl[ptype].l3_type;
4088 	} else {
4089 		l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
4090 					  HNS3_RXD_L3ID_S);
4091 	}
4092 
4093 	if (l3_type == HNS3_L3_TYPE_IPV4)
4094 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
4095 	else if (l3_type == HNS3_L3_TYPE_IPV6)
4096 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
4097 	else
4098 		return -EFAULT;
4099 
4100 	return  hns3_gro_complete(skb, l234info);
4101 }
4102 
4103 static void hns3_set_rx_skb_rss_type(struct hns3_enet_ring *ring,
4104 				     struct sk_buff *skb, u32 rss_hash)
4105 {
4106 	struct hnae3_handle *handle = ring->tqp->handle;
4107 	enum pkt_hash_types rss_type;
4108 
4109 	if (rss_hash)
4110 		rss_type = handle->kinfo.rss_type;
4111 	else
4112 		rss_type = PKT_HASH_TYPE_NONE;
4113 
4114 	skb_set_hash(skb, rss_hash, rss_type);
4115 }
4116 
4117 static void hns3_handle_rx_ts_info(struct net_device *netdev,
4118 				   struct hns3_desc *desc, struct sk_buff *skb,
4119 				   u32 bd_base_info)
4120 {
4121 	if (unlikely(bd_base_info & BIT(HNS3_RXD_TS_VLD_B))) {
4122 		struct hnae3_handle *h = hns3_get_handle(netdev);
4123 		u32 nsec = le32_to_cpu(desc->ts_nsec);
4124 		u32 sec = le32_to_cpu(desc->ts_sec);
4125 
4126 		if (h->ae_algo->ops->get_rx_hwts)
4127 			h->ae_algo->ops->get_rx_hwts(h, skb, nsec, sec);
4128 	}
4129 }
4130 
4131 static void hns3_handle_rx_vlan_tag(struct hns3_enet_ring *ring,
4132 				    struct hns3_desc *desc, struct sk_buff *skb,
4133 				    u32 l234info)
4134 {
4135 	struct net_device *netdev = ring_to_netdev(ring);
4136 
4137 	/* Based on hw strategy, the tag offloaded will be stored at
4138 	 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
4139 	 * in one layer tag case.
4140 	 */
4141 	if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
4142 		u16 vlan_tag;
4143 
4144 		if (hns3_parse_vlan_tag(ring, desc, l234info, &vlan_tag))
4145 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
4146 					       vlan_tag);
4147 	}
4148 }
4149 
4150 static int hns3_handle_bdinfo(struct hns3_enet_ring *ring, struct sk_buff *skb)
4151 {
4152 	struct net_device *netdev = ring_to_netdev(ring);
4153 	enum hns3_pkt_l2t_type l2_frame_type;
4154 	u32 bd_base_info, l234info, ol_info;
4155 	struct hns3_desc *desc;
4156 	unsigned int len;
4157 	int pre_ntc, ret;
4158 	u16 csum;
4159 
4160 	/* bdinfo handled below is only valid on the last BD of the
4161 	 * current packet, and ring->next_to_clean indicates the first
4162 	 * descriptor of next packet, so need - 1 below.
4163 	 */
4164 	pre_ntc = ring->next_to_clean ? (ring->next_to_clean - 1) :
4165 					(ring->desc_num - 1);
4166 	desc = &ring->desc[pre_ntc];
4167 	bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
4168 	l234info = le32_to_cpu(desc->rx.l234_info);
4169 	ol_info = le32_to_cpu(desc->rx.ol_info);
4170 	csum = le16_to_cpu(desc->csum);
4171 
4172 	hns3_handle_rx_ts_info(netdev, desc, skb, bd_base_info);
4173 
4174 	hns3_handle_rx_vlan_tag(ring, desc, skb, l234info);
4175 
4176 	if (unlikely(!desc->rx.pkt_len || (l234info & (BIT(HNS3_RXD_TRUNCAT_B) |
4177 				  BIT(HNS3_RXD_L2E_B))))) {
4178 		u64_stats_update_begin(&ring->syncp);
4179 		if (l234info & BIT(HNS3_RXD_L2E_B))
4180 			ring->stats.l2_err++;
4181 		else
4182 			ring->stats.err_pkt_len++;
4183 		u64_stats_update_end(&ring->syncp);
4184 
4185 		return -EFAULT;
4186 	}
4187 
4188 	len = skb->len;
4189 
4190 	/* Do update ip stack process */
4191 	skb->protocol = eth_type_trans(skb, netdev);
4192 
4193 	/* This is needed in order to enable forwarding support */
4194 	ret = hns3_set_gro_and_checksum(ring, skb, l234info,
4195 					bd_base_info, ol_info, csum);
4196 	if (unlikely(ret)) {
4197 		hns3_ring_stats_update(ring, rx_err_cnt);
4198 		return ret;
4199 	}
4200 
4201 	l2_frame_type = hnae3_get_field(l234info, HNS3_RXD_DMAC_M,
4202 					HNS3_RXD_DMAC_S);
4203 
4204 	u64_stats_update_begin(&ring->syncp);
4205 	ring->stats.rx_pkts++;
4206 	ring->stats.rx_bytes += len;
4207 
4208 	if (l2_frame_type == HNS3_L2_TYPE_MULTICAST)
4209 		ring->stats.rx_multicast++;
4210 
4211 	u64_stats_update_end(&ring->syncp);
4212 
4213 	ring->tqp_vector->rx_group.total_bytes += len;
4214 
4215 	hns3_set_rx_skb_rss_type(ring, skb, le32_to_cpu(desc->rx.rss_hash));
4216 	return 0;
4217 }
4218 
4219 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring)
4220 {
4221 	struct sk_buff *skb = ring->skb;
4222 	struct hns3_desc_cb *desc_cb;
4223 	struct hns3_desc *desc;
4224 	unsigned int length;
4225 	u32 bd_base_info;
4226 	int ret;
4227 
4228 	desc = &ring->desc[ring->next_to_clean];
4229 	desc_cb = &ring->desc_cb[ring->next_to_clean];
4230 
4231 	prefetch(desc);
4232 
4233 	if (!skb) {
4234 		bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
4235 		/* Check valid BD */
4236 		if (unlikely(!(bd_base_info & BIT(HNS3_RXD_VLD_B))))
4237 			return -ENXIO;
4238 
4239 		dma_rmb();
4240 		length = le16_to_cpu(desc->rx.size);
4241 
4242 		ring->va = desc_cb->buf + desc_cb->page_offset;
4243 
4244 		dma_sync_single_for_cpu(ring_to_dev(ring),
4245 				desc_cb->dma + desc_cb->page_offset,
4246 				hns3_buf_size(ring),
4247 				DMA_FROM_DEVICE);
4248 
4249 		/* Prefetch first cache line of first page.
4250 		 * Idea is to cache few bytes of the header of the packet.
4251 		 * Our L1 Cache line size is 64B so need to prefetch twice to make
4252 		 * it 128B. But in actual we can have greater size of caches with
4253 		 * 128B Level 1 cache lines. In such a case, single fetch would
4254 		 * suffice to cache in the relevant part of the header.
4255 		 */
4256 		net_prefetch(ring->va);
4257 
4258 		ret = hns3_alloc_skb(ring, length, ring->va);
4259 		skb = ring->skb;
4260 
4261 		if (ret < 0) /* alloc buffer fail */
4262 			return ret;
4263 		if (!(bd_base_info & BIT(HNS3_RXD_FE_B))) { /* need add frag */
4264 			ret = hns3_add_frag(ring);
4265 			if (ret)
4266 				return ret;
4267 		}
4268 	} else {
4269 		ret = hns3_add_frag(ring);
4270 		if (ret)
4271 			return ret;
4272 	}
4273 
4274 	/* As the head data may be changed when GRO enable, copy
4275 	 * the head data in after other data rx completed
4276 	 */
4277 	if (skb->len > HNS3_RX_HEAD_SIZE)
4278 		memcpy(skb->data, ring->va,
4279 		       ALIGN(ring->pull_len, sizeof(long)));
4280 
4281 	ret = hns3_handle_bdinfo(ring, skb);
4282 	if (unlikely(ret)) {
4283 		dev_kfree_skb_any(skb);
4284 		return ret;
4285 	}
4286 
4287 	skb_record_rx_queue(skb, ring->tqp->tqp_index);
4288 	return 0;
4289 }
4290 
4291 int hns3_clean_rx_ring(struct hns3_enet_ring *ring, int budget,
4292 		       void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
4293 {
4294 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
4295 	int unused_count = hns3_desc_unused(ring);
4296 	bool failure = false;
4297 	int recv_pkts = 0;
4298 	int err;
4299 
4300 	unused_count -= ring->pending_buf;
4301 
4302 	while (recv_pkts < budget) {
4303 		/* Reuse or realloc buffers */
4304 		if (unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
4305 			failure = failure ||
4306 				hns3_nic_alloc_rx_buffers(ring, unused_count);
4307 			unused_count = 0;
4308 		}
4309 
4310 		/* Poll one pkt */
4311 		err = hns3_handle_rx_bd(ring);
4312 		/* Do not get FE for the packet or failed to alloc skb */
4313 		if (unlikely(!ring->skb || err == -ENXIO)) {
4314 			goto out;
4315 		} else if (likely(!err)) {
4316 			rx_fn(ring, ring->skb);
4317 			recv_pkts++;
4318 		}
4319 
4320 		unused_count += ring->pending_buf;
4321 		ring->skb = NULL;
4322 		ring->pending_buf = 0;
4323 	}
4324 
4325 out:
4326 	/* sync head pointer before exiting, since hardware will calculate
4327 	 * FBD number with head pointer
4328 	 */
4329 	if (unused_count > 0)
4330 		failure = failure ||
4331 			  hns3_nic_alloc_rx_buffers(ring, unused_count);
4332 
4333 	return failure ? budget : recv_pkts;
4334 }
4335 
4336 static void hns3_update_rx_int_coalesce(struct hns3_enet_tqp_vector *tqp_vector)
4337 {
4338 	struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
4339 	struct dim_sample sample = {};
4340 
4341 	if (!rx_group->coal.adapt_enable)
4342 		return;
4343 
4344 	dim_update_sample(tqp_vector->event_cnt, rx_group->total_packets,
4345 			  rx_group->total_bytes, &sample);
4346 	net_dim(&rx_group->dim, sample);
4347 }
4348 
4349 static void hns3_update_tx_int_coalesce(struct hns3_enet_tqp_vector *tqp_vector)
4350 {
4351 	struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
4352 	struct dim_sample sample = {};
4353 
4354 	if (!tx_group->coal.adapt_enable)
4355 		return;
4356 
4357 	dim_update_sample(tqp_vector->event_cnt, tx_group->total_packets,
4358 			  tx_group->total_bytes, &sample);
4359 	net_dim(&tx_group->dim, sample);
4360 }
4361 
4362 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
4363 {
4364 	struct hns3_nic_priv *priv = netdev_priv(napi->dev);
4365 	struct hns3_enet_ring *ring;
4366 	int rx_pkt_total = 0;
4367 
4368 	struct hns3_enet_tqp_vector *tqp_vector =
4369 		container_of(napi, struct hns3_enet_tqp_vector, napi);
4370 	bool clean_complete = true;
4371 	int rx_budget = budget;
4372 
4373 	if (unlikely(test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
4374 		napi_complete(napi);
4375 		return 0;
4376 	}
4377 
4378 	/* Since the actual Tx work is minimal, we can give the Tx a larger
4379 	 * budget and be more aggressive about cleaning up the Tx descriptors.
4380 	 */
4381 	hns3_for_each_ring(ring, tqp_vector->tx_group)
4382 		hns3_clean_tx_ring(ring, budget);
4383 
4384 	/* make sure rx ring budget not smaller than 1 */
4385 	if (tqp_vector->num_tqps > 1)
4386 		rx_budget = max(budget / tqp_vector->num_tqps, 1);
4387 
4388 	hns3_for_each_ring(ring, tqp_vector->rx_group) {
4389 		int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
4390 						    hns3_rx_skb);
4391 		if (rx_cleaned >= rx_budget)
4392 			clean_complete = false;
4393 
4394 		rx_pkt_total += rx_cleaned;
4395 	}
4396 
4397 	tqp_vector->rx_group.total_packets += rx_pkt_total;
4398 
4399 	if (!clean_complete)
4400 		return budget;
4401 
4402 	if (napi_complete(napi) &&
4403 	    likely(!test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
4404 		hns3_update_rx_int_coalesce(tqp_vector);
4405 		hns3_update_tx_int_coalesce(tqp_vector);
4406 
4407 		hns3_mask_vector_irq(tqp_vector, 1);
4408 	}
4409 
4410 	return rx_pkt_total;
4411 }
4412 
4413 static int hns3_create_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
4414 				  struct hnae3_ring_chain_node **head,
4415 				  bool is_tx)
4416 {
4417 	u32 bit_value = is_tx ? HNAE3_RING_TYPE_TX : HNAE3_RING_TYPE_RX;
4418 	u32 field_value = is_tx ? HNAE3_RING_GL_TX : HNAE3_RING_GL_RX;
4419 	struct hnae3_ring_chain_node *cur_chain = *head;
4420 	struct pci_dev *pdev = tqp_vector->handle->pdev;
4421 	struct hnae3_ring_chain_node *chain;
4422 	struct hns3_enet_ring *ring;
4423 
4424 	ring = is_tx ? tqp_vector->tx_group.ring : tqp_vector->rx_group.ring;
4425 
4426 	if (cur_chain) {
4427 		while (cur_chain->next)
4428 			cur_chain = cur_chain->next;
4429 	}
4430 
4431 	while (ring) {
4432 		chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
4433 		if (!chain)
4434 			return -ENOMEM;
4435 		if (cur_chain)
4436 			cur_chain->next = chain;
4437 		else
4438 			*head = chain;
4439 		chain->tqp_index = ring->tqp->tqp_index;
4440 		hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
4441 				bit_value);
4442 		hnae3_set_field(chain->int_gl_idx,
4443 				HNAE3_RING_GL_IDX_M,
4444 				HNAE3_RING_GL_IDX_S, field_value);
4445 
4446 		cur_chain = chain;
4447 
4448 		ring = ring->next;
4449 	}
4450 
4451 	return 0;
4452 }
4453 
4454 static struct hnae3_ring_chain_node *
4455 hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector)
4456 {
4457 	struct pci_dev *pdev = tqp_vector->handle->pdev;
4458 	struct hnae3_ring_chain_node *cur_chain = NULL;
4459 	struct hnae3_ring_chain_node *chain;
4460 
4461 	if (hns3_create_ring_chain(tqp_vector, &cur_chain, true))
4462 		goto err_free_chain;
4463 
4464 	if (hns3_create_ring_chain(tqp_vector, &cur_chain, false))
4465 		goto err_free_chain;
4466 
4467 	return cur_chain;
4468 
4469 err_free_chain:
4470 	while (cur_chain) {
4471 		chain = cur_chain->next;
4472 		devm_kfree(&pdev->dev, cur_chain);
4473 		cur_chain = chain;
4474 	}
4475 
4476 	return NULL;
4477 }
4478 
4479 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
4480 					struct hnae3_ring_chain_node *head)
4481 {
4482 	struct pci_dev *pdev = tqp_vector->handle->pdev;
4483 	struct hnae3_ring_chain_node *chain_tmp, *chain;
4484 
4485 	chain = head;
4486 
4487 	while (chain) {
4488 		chain_tmp = chain->next;
4489 		devm_kfree(&pdev->dev, chain);
4490 		chain = chain_tmp;
4491 	}
4492 }
4493 
4494 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
4495 				   struct hns3_enet_ring *ring)
4496 {
4497 	ring->next = group->ring;
4498 	group->ring = ring;
4499 
4500 	group->count++;
4501 }
4502 
4503 static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
4504 {
4505 	struct pci_dev *pdev = priv->ae_handle->pdev;
4506 	struct hns3_enet_tqp_vector *tqp_vector;
4507 	int num_vectors = priv->vector_num;
4508 	int numa_node;
4509 	int vector_i;
4510 
4511 	numa_node = dev_to_node(&pdev->dev);
4512 
4513 	for (vector_i = 0; vector_i < num_vectors; vector_i++) {
4514 		tqp_vector = &priv->tqp_vector[vector_i];
4515 		cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
4516 				&tqp_vector->affinity_mask);
4517 	}
4518 }
4519 
4520 static void hns3_rx_dim_work(struct work_struct *work)
4521 {
4522 	struct dim *dim = container_of(work, struct dim, work);
4523 	struct hns3_enet_ring_group *group = container_of(dim,
4524 		struct hns3_enet_ring_group, dim);
4525 	struct hns3_enet_tqp_vector *tqp_vector = group->ring->tqp_vector;
4526 	struct dim_cq_moder cur_moder =
4527 		net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
4528 
4529 	hns3_set_vector_coalesce_rx_gl(group->ring->tqp_vector, cur_moder.usec);
4530 	tqp_vector->rx_group.coal.int_gl = cur_moder.usec;
4531 
4532 	if (cur_moder.pkts < tqp_vector->rx_group.coal.int_ql_max) {
4533 		hns3_set_vector_coalesce_rx_ql(tqp_vector, cur_moder.pkts);
4534 		tqp_vector->rx_group.coal.int_ql = cur_moder.pkts;
4535 	}
4536 
4537 	dim->state = DIM_START_MEASURE;
4538 }
4539 
4540 static void hns3_tx_dim_work(struct work_struct *work)
4541 {
4542 	struct dim *dim = container_of(work, struct dim, work);
4543 	struct hns3_enet_ring_group *group = container_of(dim,
4544 		struct hns3_enet_ring_group, dim);
4545 	struct hns3_enet_tqp_vector *tqp_vector = group->ring->tqp_vector;
4546 	struct dim_cq_moder cur_moder =
4547 		net_dim_get_tx_moderation(dim->mode, dim->profile_ix);
4548 
4549 	hns3_set_vector_coalesce_tx_gl(tqp_vector, cur_moder.usec);
4550 	tqp_vector->tx_group.coal.int_gl = cur_moder.usec;
4551 
4552 	if (cur_moder.pkts < tqp_vector->tx_group.coal.int_ql_max) {
4553 		hns3_set_vector_coalesce_tx_ql(tqp_vector, cur_moder.pkts);
4554 		tqp_vector->tx_group.coal.int_ql = cur_moder.pkts;
4555 	}
4556 
4557 	dim->state = DIM_START_MEASURE;
4558 }
4559 
4560 static void hns3_nic_init_dim(struct hns3_enet_tqp_vector *tqp_vector)
4561 {
4562 	INIT_WORK(&tqp_vector->rx_group.dim.work, hns3_rx_dim_work);
4563 	INIT_WORK(&tqp_vector->tx_group.dim.work, hns3_tx_dim_work);
4564 }
4565 
4566 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
4567 {
4568 	struct hnae3_handle *h = priv->ae_handle;
4569 	struct hns3_enet_tqp_vector *tqp_vector;
4570 	int ret;
4571 	int i;
4572 
4573 	hns3_nic_set_cpumask(priv);
4574 
4575 	for (i = 0; i < priv->vector_num; i++) {
4576 		tqp_vector = &priv->tqp_vector[i];
4577 		hns3_vector_coalesce_init_hw(tqp_vector, priv);
4578 		tqp_vector->num_tqps = 0;
4579 		hns3_nic_init_dim(tqp_vector);
4580 	}
4581 
4582 	for (i = 0; i < h->kinfo.num_tqps; i++) {
4583 		u16 vector_i = i % priv->vector_num;
4584 		u16 tqp_num = h->kinfo.num_tqps;
4585 
4586 		tqp_vector = &priv->tqp_vector[vector_i];
4587 
4588 		hns3_add_ring_to_group(&tqp_vector->tx_group,
4589 				       &priv->ring[i]);
4590 
4591 		hns3_add_ring_to_group(&tqp_vector->rx_group,
4592 				       &priv->ring[i + tqp_num]);
4593 
4594 		priv->ring[i].tqp_vector = tqp_vector;
4595 		priv->ring[i + tqp_num].tqp_vector = tqp_vector;
4596 		tqp_vector->num_tqps++;
4597 	}
4598 
4599 	for (i = 0; i < priv->vector_num; i++) {
4600 		struct hnae3_ring_chain_node *vector_ring_chain;
4601 
4602 		tqp_vector = &priv->tqp_vector[i];
4603 
4604 		tqp_vector->rx_group.total_bytes = 0;
4605 		tqp_vector->rx_group.total_packets = 0;
4606 		tqp_vector->tx_group.total_bytes = 0;
4607 		tqp_vector->tx_group.total_packets = 0;
4608 		tqp_vector->handle = h;
4609 
4610 		vector_ring_chain = hns3_get_vector_ring_chain(tqp_vector);
4611 		if (!vector_ring_chain) {
4612 			ret = -ENOMEM;
4613 			goto map_ring_fail;
4614 		}
4615 
4616 		ret = h->ae_algo->ops->map_ring_to_vector(h,
4617 			tqp_vector->vector_irq, vector_ring_chain);
4618 
4619 		hns3_free_vector_ring_chain(tqp_vector, vector_ring_chain);
4620 
4621 		if (ret)
4622 			goto map_ring_fail;
4623 
4624 		netif_napi_add(priv->netdev, &tqp_vector->napi,
4625 			       hns3_nic_common_poll, NAPI_POLL_WEIGHT);
4626 	}
4627 
4628 	return 0;
4629 
4630 map_ring_fail:
4631 	while (i--)
4632 		netif_napi_del(&priv->tqp_vector[i].napi);
4633 
4634 	return ret;
4635 }
4636 
4637 static void hns3_nic_init_coal_cfg(struct hns3_nic_priv *priv)
4638 {
4639 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
4640 	struct hns3_enet_coalesce *tx_coal = &priv->tx_coal;
4641 	struct hns3_enet_coalesce *rx_coal = &priv->rx_coal;
4642 
4643 	/* initialize the configuration for interrupt coalescing.
4644 	 * 1. GL (Interrupt Gap Limiter)
4645 	 * 2. RL (Interrupt Rate Limiter)
4646 	 * 3. QL (Interrupt Quantity Limiter)
4647 	 *
4648 	 * Default: enable interrupt coalescing self-adaptive and GL
4649 	 */
4650 	tx_coal->adapt_enable = 1;
4651 	rx_coal->adapt_enable = 1;
4652 
4653 	tx_coal->int_gl = HNS3_INT_GL_50K;
4654 	rx_coal->int_gl = HNS3_INT_GL_50K;
4655 
4656 	rx_coal->flow_level = HNS3_FLOW_LOW;
4657 	tx_coal->flow_level = HNS3_FLOW_LOW;
4658 
4659 	if (ae_dev->dev_specs.int_ql_max) {
4660 		tx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
4661 		rx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
4662 	}
4663 }
4664 
4665 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
4666 {
4667 	struct hnae3_handle *h = priv->ae_handle;
4668 	struct hns3_enet_tqp_vector *tqp_vector;
4669 	struct hnae3_vector_info *vector;
4670 	struct pci_dev *pdev = h->pdev;
4671 	u16 tqp_num = h->kinfo.num_tqps;
4672 	u16 vector_num;
4673 	int ret = 0;
4674 	u16 i;
4675 
4676 	/* RSS size, cpu online and vector_num should be the same */
4677 	/* Should consider 2p/4p later */
4678 	vector_num = min_t(u16, num_online_cpus(), tqp_num);
4679 
4680 	vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
4681 			      GFP_KERNEL);
4682 	if (!vector)
4683 		return -ENOMEM;
4684 
4685 	/* save the actual available vector number */
4686 	vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
4687 
4688 	priv->vector_num = vector_num;
4689 	priv->tqp_vector = (struct hns3_enet_tqp_vector *)
4690 		devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
4691 			     GFP_KERNEL);
4692 	if (!priv->tqp_vector) {
4693 		ret = -ENOMEM;
4694 		goto out;
4695 	}
4696 
4697 	for (i = 0; i < priv->vector_num; i++) {
4698 		tqp_vector = &priv->tqp_vector[i];
4699 		tqp_vector->idx = i;
4700 		tqp_vector->mask_addr = vector[i].io_addr;
4701 		tqp_vector->vector_irq = vector[i].vector;
4702 		hns3_vector_coalesce_init(tqp_vector, priv);
4703 	}
4704 
4705 out:
4706 	devm_kfree(&pdev->dev, vector);
4707 	return ret;
4708 }
4709 
4710 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
4711 {
4712 	group->ring = NULL;
4713 	group->count = 0;
4714 }
4715 
4716 static void hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
4717 {
4718 	struct hnae3_ring_chain_node *vector_ring_chain;
4719 	struct hnae3_handle *h = priv->ae_handle;
4720 	struct hns3_enet_tqp_vector *tqp_vector;
4721 	int i;
4722 
4723 	for (i = 0; i < priv->vector_num; i++) {
4724 		tqp_vector = &priv->tqp_vector[i];
4725 
4726 		if (!tqp_vector->rx_group.ring && !tqp_vector->tx_group.ring)
4727 			continue;
4728 
4729 		/* Since the mapping can be overwritten, when fail to get the
4730 		 * chain between vector and ring, we should go on to deal with
4731 		 * the remaining options.
4732 		 */
4733 		vector_ring_chain = hns3_get_vector_ring_chain(tqp_vector);
4734 		if (!vector_ring_chain)
4735 			dev_warn(priv->dev, "failed to get ring chain\n");
4736 
4737 		h->ae_algo->ops->unmap_ring_from_vector(h,
4738 			tqp_vector->vector_irq, vector_ring_chain);
4739 
4740 		hns3_free_vector_ring_chain(tqp_vector, vector_ring_chain);
4741 
4742 		hns3_clear_ring_group(&tqp_vector->rx_group);
4743 		hns3_clear_ring_group(&tqp_vector->tx_group);
4744 		netif_napi_del(&priv->tqp_vector[i].napi);
4745 	}
4746 }
4747 
4748 static void hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
4749 {
4750 	struct hnae3_handle *h = priv->ae_handle;
4751 	struct pci_dev *pdev = h->pdev;
4752 	int i, ret;
4753 
4754 	for (i = 0; i < priv->vector_num; i++) {
4755 		struct hns3_enet_tqp_vector *tqp_vector;
4756 
4757 		tqp_vector = &priv->tqp_vector[i];
4758 		ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
4759 		if (ret)
4760 			return;
4761 	}
4762 
4763 	devm_kfree(&pdev->dev, priv->tqp_vector);
4764 }
4765 
4766 static void hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
4767 			      unsigned int ring_type)
4768 {
4769 	int queue_num = priv->ae_handle->kinfo.num_tqps;
4770 	struct hns3_enet_ring *ring;
4771 	int desc_num;
4772 
4773 	if (ring_type == HNAE3_RING_TYPE_TX) {
4774 		ring = &priv->ring[q->tqp_index];
4775 		desc_num = priv->ae_handle->kinfo.num_tx_desc;
4776 		ring->queue_index = q->tqp_index;
4777 		ring->tx_copybreak = priv->tx_copybreak;
4778 		ring->last_to_use = 0;
4779 	} else {
4780 		ring = &priv->ring[q->tqp_index + queue_num];
4781 		desc_num = priv->ae_handle->kinfo.num_rx_desc;
4782 		ring->queue_index = q->tqp_index;
4783 		ring->rx_copybreak = priv->rx_copybreak;
4784 	}
4785 
4786 	hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
4787 
4788 	ring->tqp = q;
4789 	ring->desc = NULL;
4790 	ring->desc_cb = NULL;
4791 	ring->dev = priv->dev;
4792 	ring->desc_dma_addr = 0;
4793 	ring->buf_size = q->buf_size;
4794 	ring->desc_num = desc_num;
4795 	ring->next_to_use = 0;
4796 	ring->next_to_clean = 0;
4797 }
4798 
4799 static void hns3_queue_to_ring(struct hnae3_queue *tqp,
4800 			       struct hns3_nic_priv *priv)
4801 {
4802 	hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
4803 	hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
4804 }
4805 
4806 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
4807 {
4808 	struct hnae3_handle *h = priv->ae_handle;
4809 	struct pci_dev *pdev = h->pdev;
4810 	int i;
4811 
4812 	priv->ring = devm_kzalloc(&pdev->dev,
4813 				  array3_size(h->kinfo.num_tqps,
4814 					      sizeof(*priv->ring), 2),
4815 				  GFP_KERNEL);
4816 	if (!priv->ring)
4817 		return -ENOMEM;
4818 
4819 	for (i = 0; i < h->kinfo.num_tqps; i++)
4820 		hns3_queue_to_ring(h->kinfo.tqp[i], priv);
4821 
4822 	return 0;
4823 }
4824 
4825 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
4826 {
4827 	if (!priv->ring)
4828 		return;
4829 
4830 	devm_kfree(priv->dev, priv->ring);
4831 	priv->ring = NULL;
4832 }
4833 
4834 static void hns3_alloc_page_pool(struct hns3_enet_ring *ring)
4835 {
4836 	struct page_pool_params pp_params = {
4837 		.flags = PP_FLAG_DMA_MAP | PP_FLAG_PAGE_FRAG |
4838 				PP_FLAG_DMA_SYNC_DEV,
4839 		.order = hns3_page_order(ring),
4840 		.pool_size = ring->desc_num * hns3_buf_size(ring) /
4841 				(PAGE_SIZE << hns3_page_order(ring)),
4842 		.nid = dev_to_node(ring_to_dev(ring)),
4843 		.dev = ring_to_dev(ring),
4844 		.dma_dir = DMA_FROM_DEVICE,
4845 		.offset = 0,
4846 		.max_len = PAGE_SIZE << hns3_page_order(ring),
4847 	};
4848 
4849 	ring->page_pool = page_pool_create(&pp_params);
4850 	if (IS_ERR(ring->page_pool)) {
4851 		dev_warn(ring_to_dev(ring), "page pool creation failed: %ld\n",
4852 			 PTR_ERR(ring->page_pool));
4853 		ring->page_pool = NULL;
4854 	}
4855 }
4856 
4857 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
4858 {
4859 	int ret;
4860 
4861 	if (ring->desc_num <= 0 || ring->buf_size <= 0)
4862 		return -EINVAL;
4863 
4864 	ring->desc_cb = devm_kcalloc(ring_to_dev(ring), ring->desc_num,
4865 				     sizeof(ring->desc_cb[0]), GFP_KERNEL);
4866 	if (!ring->desc_cb) {
4867 		ret = -ENOMEM;
4868 		goto out;
4869 	}
4870 
4871 	ret = hns3_alloc_desc(ring);
4872 	if (ret)
4873 		goto out_with_desc_cb;
4874 
4875 	if (!HNAE3_IS_TX_RING(ring)) {
4876 		if (page_pool_enabled)
4877 			hns3_alloc_page_pool(ring);
4878 
4879 		ret = hns3_alloc_ring_buffers(ring);
4880 		if (ret)
4881 			goto out_with_desc;
4882 	} else {
4883 		hns3_init_tx_spare_buffer(ring);
4884 	}
4885 
4886 	return 0;
4887 
4888 out_with_desc:
4889 	hns3_free_desc(ring);
4890 out_with_desc_cb:
4891 	devm_kfree(ring_to_dev(ring), ring->desc_cb);
4892 	ring->desc_cb = NULL;
4893 out:
4894 	return ret;
4895 }
4896 
4897 void hns3_fini_ring(struct hns3_enet_ring *ring)
4898 {
4899 	hns3_free_desc(ring);
4900 	devm_kfree(ring_to_dev(ring), ring->desc_cb);
4901 	ring->desc_cb = NULL;
4902 	ring->next_to_clean = 0;
4903 	ring->next_to_use = 0;
4904 	ring->last_to_use = 0;
4905 	ring->pending_buf = 0;
4906 	if (!HNAE3_IS_TX_RING(ring) && ring->skb) {
4907 		dev_kfree_skb_any(ring->skb);
4908 		ring->skb = NULL;
4909 	} else if (HNAE3_IS_TX_RING(ring) && ring->tx_spare) {
4910 		struct hns3_tx_spare *tx_spare = ring->tx_spare;
4911 
4912 		dma_unmap_page(ring_to_dev(ring), tx_spare->dma, tx_spare->len,
4913 			       DMA_TO_DEVICE);
4914 		free_pages((unsigned long)tx_spare->buf,
4915 			   get_order(tx_spare->len));
4916 		devm_kfree(ring_to_dev(ring), tx_spare);
4917 		ring->tx_spare = NULL;
4918 	}
4919 
4920 	if (!HNAE3_IS_TX_RING(ring) && ring->page_pool) {
4921 		page_pool_destroy(ring->page_pool);
4922 		ring->page_pool = NULL;
4923 	}
4924 }
4925 
4926 static int hns3_buf_size2type(u32 buf_size)
4927 {
4928 	int bd_size_type;
4929 
4930 	switch (buf_size) {
4931 	case 512:
4932 		bd_size_type = HNS3_BD_SIZE_512_TYPE;
4933 		break;
4934 	case 1024:
4935 		bd_size_type = HNS3_BD_SIZE_1024_TYPE;
4936 		break;
4937 	case 2048:
4938 		bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4939 		break;
4940 	case 4096:
4941 		bd_size_type = HNS3_BD_SIZE_4096_TYPE;
4942 		break;
4943 	default:
4944 		bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4945 	}
4946 
4947 	return bd_size_type;
4948 }
4949 
4950 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
4951 {
4952 	dma_addr_t dma = ring->desc_dma_addr;
4953 	struct hnae3_queue *q = ring->tqp;
4954 
4955 	if (!HNAE3_IS_TX_RING(ring)) {
4956 		hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG, (u32)dma);
4957 		hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
4958 			       (u32)((dma >> 31) >> 1));
4959 
4960 		hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
4961 			       hns3_buf_size2type(ring->buf_size));
4962 		hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
4963 			       ring->desc_num / 8 - 1);
4964 	} else {
4965 		hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
4966 			       (u32)dma);
4967 		hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
4968 			       (u32)((dma >> 31) >> 1));
4969 
4970 		hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
4971 			       ring->desc_num / 8 - 1);
4972 	}
4973 }
4974 
4975 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
4976 {
4977 	struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
4978 	struct hnae3_tc_info *tc_info = &kinfo->tc_info;
4979 	int i;
4980 
4981 	for (i = 0; i < tc_info->num_tc; i++) {
4982 		int j;
4983 
4984 		for (j = 0; j < tc_info->tqp_count[i]; j++) {
4985 			struct hnae3_queue *q;
4986 
4987 			q = priv->ring[tc_info->tqp_offset[i] + j].tqp;
4988 			hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG, i);
4989 		}
4990 	}
4991 }
4992 
4993 int hns3_init_all_ring(struct hns3_nic_priv *priv)
4994 {
4995 	struct hnae3_handle *h = priv->ae_handle;
4996 	int ring_num = h->kinfo.num_tqps * 2;
4997 	int i, j;
4998 	int ret;
4999 
5000 	for (i = 0; i < ring_num; i++) {
5001 		ret = hns3_alloc_ring_memory(&priv->ring[i]);
5002 		if (ret) {
5003 			dev_err(priv->dev,
5004 				"Alloc ring memory fail! ret=%d\n", ret);
5005 			goto out_when_alloc_ring_memory;
5006 		}
5007 
5008 		u64_stats_init(&priv->ring[i].syncp);
5009 	}
5010 
5011 	return 0;
5012 
5013 out_when_alloc_ring_memory:
5014 	for (j = i - 1; j >= 0; j--)
5015 		hns3_fini_ring(&priv->ring[j]);
5016 
5017 	return -ENOMEM;
5018 }
5019 
5020 static void hns3_uninit_all_ring(struct hns3_nic_priv *priv)
5021 {
5022 	struct hnae3_handle *h = priv->ae_handle;
5023 	int i;
5024 
5025 	for (i = 0; i < h->kinfo.num_tqps; i++) {
5026 		hns3_fini_ring(&priv->ring[i]);
5027 		hns3_fini_ring(&priv->ring[i + h->kinfo.num_tqps]);
5028 	}
5029 }
5030 
5031 /* Set mac addr if it is configured. or leave it to the AE driver */
5032 static int hns3_init_mac_addr(struct net_device *netdev)
5033 {
5034 	struct hns3_nic_priv *priv = netdev_priv(netdev);
5035 	char format_mac_addr[HNAE3_FORMAT_MAC_ADDR_LEN];
5036 	struct hnae3_handle *h = priv->ae_handle;
5037 	u8 mac_addr_temp[ETH_ALEN];
5038 	int ret = 0;
5039 
5040 	if (h->ae_algo->ops->get_mac_addr)
5041 		h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
5042 
5043 	/* Check if the MAC address is valid, if not get a random one */
5044 	if (!is_valid_ether_addr(mac_addr_temp)) {
5045 		eth_hw_addr_random(netdev);
5046 		hnae3_format_mac_addr(format_mac_addr, netdev->dev_addr);
5047 		dev_warn(priv->dev, "using random MAC address %s\n",
5048 			 format_mac_addr);
5049 	} else if (!ether_addr_equal(netdev->dev_addr, mac_addr_temp)) {
5050 		eth_hw_addr_set(netdev, mac_addr_temp);
5051 		ether_addr_copy(netdev->perm_addr, mac_addr_temp);
5052 	} else {
5053 		return 0;
5054 	}
5055 
5056 	if (h->ae_algo->ops->set_mac_addr)
5057 		ret = h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
5058 
5059 	return ret;
5060 }
5061 
5062 static int hns3_init_phy(struct net_device *netdev)
5063 {
5064 	struct hnae3_handle *h = hns3_get_handle(netdev);
5065 	int ret = 0;
5066 
5067 	if (h->ae_algo->ops->mac_connect_phy)
5068 		ret = h->ae_algo->ops->mac_connect_phy(h);
5069 
5070 	return ret;
5071 }
5072 
5073 static void hns3_uninit_phy(struct net_device *netdev)
5074 {
5075 	struct hnae3_handle *h = hns3_get_handle(netdev);
5076 
5077 	if (h->ae_algo->ops->mac_disconnect_phy)
5078 		h->ae_algo->ops->mac_disconnect_phy(h);
5079 }
5080 
5081 static int hns3_client_start(struct hnae3_handle *handle)
5082 {
5083 	if (!handle->ae_algo->ops->client_start)
5084 		return 0;
5085 
5086 	return handle->ae_algo->ops->client_start(handle);
5087 }
5088 
5089 static void hns3_client_stop(struct hnae3_handle *handle)
5090 {
5091 	if (!handle->ae_algo->ops->client_stop)
5092 		return;
5093 
5094 	handle->ae_algo->ops->client_stop(handle);
5095 }
5096 
5097 static void hns3_info_show(struct hns3_nic_priv *priv)
5098 {
5099 	struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
5100 	char format_mac_addr[HNAE3_FORMAT_MAC_ADDR_LEN];
5101 
5102 	hnae3_format_mac_addr(format_mac_addr, priv->netdev->dev_addr);
5103 	dev_info(priv->dev, "MAC address: %s\n", format_mac_addr);
5104 	dev_info(priv->dev, "Task queue pairs numbers: %u\n", kinfo->num_tqps);
5105 	dev_info(priv->dev, "RSS size: %u\n", kinfo->rss_size);
5106 	dev_info(priv->dev, "Allocated RSS size: %u\n", kinfo->req_rss_size);
5107 	dev_info(priv->dev, "RX buffer length: %u\n", kinfo->rx_buf_len);
5108 	dev_info(priv->dev, "Desc num per TX queue: %u\n", kinfo->num_tx_desc);
5109 	dev_info(priv->dev, "Desc num per RX queue: %u\n", kinfo->num_rx_desc);
5110 	dev_info(priv->dev, "Total number of enabled TCs: %u\n",
5111 		 kinfo->tc_info.num_tc);
5112 	dev_info(priv->dev, "Max mtu size: %u\n", priv->netdev->max_mtu);
5113 }
5114 
5115 static void hns3_set_cq_period_mode(struct hns3_nic_priv *priv,
5116 				    enum dim_cq_period_mode mode, bool is_tx)
5117 {
5118 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
5119 	struct hnae3_handle *handle = priv->ae_handle;
5120 	int i;
5121 
5122 	if (is_tx) {
5123 		priv->tx_cqe_mode = mode;
5124 
5125 		for (i = 0; i < priv->vector_num; i++)
5126 			priv->tqp_vector[i].tx_group.dim.mode = mode;
5127 	} else {
5128 		priv->rx_cqe_mode = mode;
5129 
5130 		for (i = 0; i < priv->vector_num; i++)
5131 			priv->tqp_vector[i].rx_group.dim.mode = mode;
5132 	}
5133 
5134 	/* only device version above V3(include V3), GL can switch CQ/EQ
5135 	 * period mode.
5136 	 */
5137 	if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3) {
5138 		u32 new_mode;
5139 		u64 reg;
5140 
5141 		new_mode = (mode == DIM_CQ_PERIOD_MODE_START_FROM_CQE) ?
5142 			HNS3_CQ_MODE_CQE : HNS3_CQ_MODE_EQE;
5143 		reg = is_tx ? HNS3_GL1_CQ_MODE_REG : HNS3_GL0_CQ_MODE_REG;
5144 
5145 		writel(new_mode, handle->kinfo.io_base + reg);
5146 	}
5147 }
5148 
5149 void hns3_cq_period_mode_init(struct hns3_nic_priv *priv,
5150 			      enum dim_cq_period_mode tx_mode,
5151 			      enum dim_cq_period_mode rx_mode)
5152 {
5153 	hns3_set_cq_period_mode(priv, tx_mode, true);
5154 	hns3_set_cq_period_mode(priv, rx_mode, false);
5155 }
5156 
5157 static void hns3_state_init(struct hnae3_handle *handle)
5158 {
5159 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(handle->pdev);
5160 	struct net_device *netdev = handle->kinfo.netdev;
5161 	struct hns3_nic_priv *priv = netdev_priv(netdev);
5162 
5163 	set_bit(HNS3_NIC_STATE_INITED, &priv->state);
5164 
5165 	if (test_bit(HNAE3_DEV_SUPPORT_TX_PUSH_B, ae_dev->caps))
5166 		set_bit(HNS3_NIC_STATE_TX_PUSH_ENABLE, &priv->state);
5167 
5168 	if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
5169 		set_bit(HNAE3_PFLAG_LIMIT_PROMISC, &handle->supported_pflags);
5170 
5171 	if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps))
5172 		set_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state);
5173 
5174 	if (hnae3_ae_dev_rxd_adv_layout_supported(ae_dev))
5175 		set_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state);
5176 }
5177 
5178 static int hns3_client_init(struct hnae3_handle *handle)
5179 {
5180 	struct pci_dev *pdev = handle->pdev;
5181 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
5182 	u16 alloc_tqps, max_rss_size;
5183 	struct hns3_nic_priv *priv;
5184 	struct net_device *netdev;
5185 	int ret;
5186 
5187 	handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
5188 						    &max_rss_size);
5189 	netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
5190 	if (!netdev)
5191 		return -ENOMEM;
5192 
5193 	priv = netdev_priv(netdev);
5194 	priv->dev = &pdev->dev;
5195 	priv->netdev = netdev;
5196 	priv->ae_handle = handle;
5197 	priv->tx_timeout_count = 0;
5198 	priv->max_non_tso_bd_num = ae_dev->dev_specs.max_non_tso_bd_num;
5199 	set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
5200 
5201 	handle->msg_enable = netif_msg_init(debug, DEFAULT_MSG_LEVEL);
5202 
5203 	handle->kinfo.netdev = netdev;
5204 	handle->priv = (void *)priv;
5205 
5206 	hns3_init_mac_addr(netdev);
5207 
5208 	hns3_set_default_feature(netdev);
5209 
5210 	netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
5211 	netdev->priv_flags |= IFF_UNICAST_FLT;
5212 	netdev->netdev_ops = &hns3_nic_netdev_ops;
5213 	SET_NETDEV_DEV(netdev, &pdev->dev);
5214 	hns3_ethtool_set_ops(netdev);
5215 
5216 	/* Carrier off reporting is important to ethtool even BEFORE open */
5217 	netif_carrier_off(netdev);
5218 
5219 	ret = hns3_get_ring_config(priv);
5220 	if (ret) {
5221 		ret = -ENOMEM;
5222 		goto out_get_ring_cfg;
5223 	}
5224 
5225 	hns3_nic_init_coal_cfg(priv);
5226 
5227 	ret = hns3_nic_alloc_vector_data(priv);
5228 	if (ret) {
5229 		ret = -ENOMEM;
5230 		goto out_alloc_vector_data;
5231 	}
5232 
5233 	ret = hns3_nic_init_vector_data(priv);
5234 	if (ret) {
5235 		ret = -ENOMEM;
5236 		goto out_init_vector_data;
5237 	}
5238 
5239 	ret = hns3_init_all_ring(priv);
5240 	if (ret) {
5241 		ret = -ENOMEM;
5242 		goto out_init_ring;
5243 	}
5244 
5245 	hns3_cq_period_mode_init(priv, DIM_CQ_PERIOD_MODE_START_FROM_EQE,
5246 				 DIM_CQ_PERIOD_MODE_START_FROM_EQE);
5247 
5248 	ret = hns3_init_phy(netdev);
5249 	if (ret)
5250 		goto out_init_phy;
5251 
5252 	/* the device can work without cpu rmap, only aRFS needs it */
5253 	ret = hns3_set_rx_cpu_rmap(netdev);
5254 	if (ret)
5255 		dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
5256 
5257 	ret = hns3_nic_init_irq(priv);
5258 	if (ret) {
5259 		dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
5260 		hns3_free_rx_cpu_rmap(netdev);
5261 		goto out_init_irq_fail;
5262 	}
5263 
5264 	ret = hns3_client_start(handle);
5265 	if (ret) {
5266 		dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
5267 		goto out_client_start;
5268 	}
5269 
5270 	hns3_dcbnl_setup(handle);
5271 
5272 	ret = hns3_dbg_init(handle);
5273 	if (ret) {
5274 		dev_err(priv->dev, "failed to init debugfs, ret = %d\n",
5275 			ret);
5276 		goto out_client_start;
5277 	}
5278 
5279 	netdev->max_mtu = HNS3_MAX_MTU(ae_dev->dev_specs.max_frm_size);
5280 
5281 	hns3_state_init(handle);
5282 
5283 	ret = register_netdev(netdev);
5284 	if (ret) {
5285 		dev_err(priv->dev, "probe register netdev fail!\n");
5286 		goto out_reg_netdev_fail;
5287 	}
5288 
5289 	if (netif_msg_drv(handle))
5290 		hns3_info_show(priv);
5291 
5292 	return ret;
5293 
5294 out_reg_netdev_fail:
5295 	hns3_dbg_uninit(handle);
5296 out_client_start:
5297 	hns3_free_rx_cpu_rmap(netdev);
5298 	hns3_nic_uninit_irq(priv);
5299 out_init_irq_fail:
5300 	hns3_uninit_phy(netdev);
5301 out_init_phy:
5302 	hns3_uninit_all_ring(priv);
5303 out_init_ring:
5304 	hns3_nic_uninit_vector_data(priv);
5305 out_init_vector_data:
5306 	hns3_nic_dealloc_vector_data(priv);
5307 out_alloc_vector_data:
5308 	priv->ring = NULL;
5309 out_get_ring_cfg:
5310 	priv->ae_handle = NULL;
5311 	free_netdev(netdev);
5312 	return ret;
5313 }
5314 
5315 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
5316 {
5317 	struct net_device *netdev = handle->kinfo.netdev;
5318 	struct hns3_nic_priv *priv = netdev_priv(netdev);
5319 
5320 	if (netdev->reg_state != NETREG_UNINITIALIZED)
5321 		unregister_netdev(netdev);
5322 
5323 	hns3_client_stop(handle);
5324 
5325 	hns3_uninit_phy(netdev);
5326 
5327 	if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5328 		netdev_warn(netdev, "already uninitialized\n");
5329 		goto out_netdev_free;
5330 	}
5331 
5332 	hns3_free_rx_cpu_rmap(netdev);
5333 
5334 	hns3_nic_uninit_irq(priv);
5335 
5336 	hns3_clear_all_ring(handle, true);
5337 
5338 	hns3_nic_uninit_vector_data(priv);
5339 
5340 	hns3_nic_dealloc_vector_data(priv);
5341 
5342 	hns3_uninit_all_ring(priv);
5343 
5344 	hns3_put_ring_config(priv);
5345 
5346 out_netdev_free:
5347 	hns3_dbg_uninit(handle);
5348 	free_netdev(netdev);
5349 }
5350 
5351 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
5352 {
5353 	struct net_device *netdev = handle->kinfo.netdev;
5354 
5355 	if (!netdev)
5356 		return;
5357 
5358 	if (linkup) {
5359 		netif_tx_wake_all_queues(netdev);
5360 		netif_carrier_on(netdev);
5361 		if (netif_msg_link(handle))
5362 			netdev_info(netdev, "link up\n");
5363 	} else {
5364 		netif_carrier_off(netdev);
5365 		netif_tx_stop_all_queues(netdev);
5366 		if (netif_msg_link(handle))
5367 			netdev_info(netdev, "link down\n");
5368 	}
5369 }
5370 
5371 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
5372 {
5373 	while (ring->next_to_clean != ring->next_to_use) {
5374 		ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
5375 		hns3_free_buffer_detach(ring, ring->next_to_clean, 0);
5376 		ring_ptr_move_fw(ring, next_to_clean);
5377 	}
5378 
5379 	ring->pending_buf = 0;
5380 }
5381 
5382 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
5383 {
5384 	struct hns3_desc_cb res_cbs;
5385 	int ret;
5386 
5387 	while (ring->next_to_use != ring->next_to_clean) {
5388 		/* When a buffer is not reused, it's memory has been
5389 		 * freed in hns3_handle_rx_bd or will be freed by
5390 		 * stack, so we need to replace the buffer here.
5391 		 */
5392 		if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
5393 			ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
5394 			if (ret) {
5395 				hns3_ring_stats_update(ring, sw_err_cnt);
5396 				/* if alloc new buffer fail, exit directly
5397 				 * and reclear in up flow.
5398 				 */
5399 				netdev_warn(ring_to_netdev(ring),
5400 					    "reserve buffer map failed, ret = %d\n",
5401 					    ret);
5402 				return ret;
5403 			}
5404 			hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
5405 		}
5406 		ring_ptr_move_fw(ring, next_to_use);
5407 	}
5408 
5409 	/* Free the pending skb in rx ring */
5410 	if (ring->skb) {
5411 		dev_kfree_skb_any(ring->skb);
5412 		ring->skb = NULL;
5413 		ring->pending_buf = 0;
5414 	}
5415 
5416 	return 0;
5417 }
5418 
5419 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
5420 {
5421 	while (ring->next_to_use != ring->next_to_clean) {
5422 		/* When a buffer is not reused, it's memory has been
5423 		 * freed in hns3_handle_rx_bd or will be freed by
5424 		 * stack, so only need to unmap the buffer here.
5425 		 */
5426 		if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
5427 			hns3_unmap_buffer(ring,
5428 					  &ring->desc_cb[ring->next_to_use]);
5429 			ring->desc_cb[ring->next_to_use].dma = 0;
5430 		}
5431 
5432 		ring_ptr_move_fw(ring, next_to_use);
5433 	}
5434 }
5435 
5436 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force)
5437 {
5438 	struct net_device *ndev = h->kinfo.netdev;
5439 	struct hns3_nic_priv *priv = netdev_priv(ndev);
5440 	u32 i;
5441 
5442 	for (i = 0; i < h->kinfo.num_tqps; i++) {
5443 		struct hns3_enet_ring *ring;
5444 
5445 		ring = &priv->ring[i];
5446 		hns3_clear_tx_ring(ring);
5447 
5448 		ring = &priv->ring[i + h->kinfo.num_tqps];
5449 		/* Continue to clear other rings even if clearing some
5450 		 * rings failed.
5451 		 */
5452 		if (force)
5453 			hns3_force_clear_rx_ring(ring);
5454 		else
5455 			hns3_clear_rx_ring(ring);
5456 	}
5457 }
5458 
5459 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
5460 {
5461 	struct net_device *ndev = h->kinfo.netdev;
5462 	struct hns3_nic_priv *priv = netdev_priv(ndev);
5463 	struct hns3_enet_ring *rx_ring;
5464 	int i, j;
5465 	int ret;
5466 
5467 	ret = h->ae_algo->ops->reset_queue(h);
5468 	if (ret)
5469 		return ret;
5470 
5471 	for (i = 0; i < h->kinfo.num_tqps; i++) {
5472 		hns3_init_ring_hw(&priv->ring[i]);
5473 
5474 		/* We need to clear tx ring here because self test will
5475 		 * use the ring and will not run down before up
5476 		 */
5477 		hns3_clear_tx_ring(&priv->ring[i]);
5478 		priv->ring[i].next_to_clean = 0;
5479 		priv->ring[i].next_to_use = 0;
5480 		priv->ring[i].last_to_use = 0;
5481 
5482 		rx_ring = &priv->ring[i + h->kinfo.num_tqps];
5483 		hns3_init_ring_hw(rx_ring);
5484 		ret = hns3_clear_rx_ring(rx_ring);
5485 		if (ret)
5486 			return ret;
5487 
5488 		/* We can not know the hardware head and tail when this
5489 		 * function is called in reset flow, so we reuse all desc.
5490 		 */
5491 		for (j = 0; j < rx_ring->desc_num; j++)
5492 			hns3_reuse_buffer(rx_ring, j);
5493 
5494 		rx_ring->next_to_clean = 0;
5495 		rx_ring->next_to_use = 0;
5496 	}
5497 
5498 	hns3_init_tx_ring_tc(priv);
5499 
5500 	return 0;
5501 }
5502 
5503 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
5504 {
5505 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
5506 	struct net_device *ndev = kinfo->netdev;
5507 	struct hns3_nic_priv *priv = netdev_priv(ndev);
5508 
5509 	if (test_and_set_bit(HNS3_NIC_STATE_RESETTING, &priv->state))
5510 		return 0;
5511 
5512 	if (!netif_running(ndev))
5513 		return 0;
5514 
5515 	return hns3_nic_net_stop(ndev);
5516 }
5517 
5518 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
5519 {
5520 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
5521 	struct hns3_nic_priv *priv = netdev_priv(kinfo->netdev);
5522 	int ret = 0;
5523 
5524 	if (!test_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5525 		netdev_err(kinfo->netdev, "device is not initialized yet\n");
5526 		return -EFAULT;
5527 	}
5528 
5529 	clear_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
5530 
5531 	if (netif_running(kinfo->netdev)) {
5532 		ret = hns3_nic_net_open(kinfo->netdev);
5533 		if (ret) {
5534 			set_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
5535 			netdev_err(kinfo->netdev,
5536 				   "net up fail, ret=%d!\n", ret);
5537 			return ret;
5538 		}
5539 	}
5540 
5541 	return ret;
5542 }
5543 
5544 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
5545 {
5546 	struct net_device *netdev = handle->kinfo.netdev;
5547 	struct hns3_nic_priv *priv = netdev_priv(netdev);
5548 	int ret;
5549 
5550 	/* Carrier off reporting is important to ethtool even BEFORE open */
5551 	netif_carrier_off(netdev);
5552 
5553 	ret = hns3_get_ring_config(priv);
5554 	if (ret)
5555 		return ret;
5556 
5557 	ret = hns3_nic_alloc_vector_data(priv);
5558 	if (ret)
5559 		goto err_put_ring;
5560 
5561 	ret = hns3_nic_init_vector_data(priv);
5562 	if (ret)
5563 		goto err_dealloc_vector;
5564 
5565 	ret = hns3_init_all_ring(priv);
5566 	if (ret)
5567 		goto err_uninit_vector;
5568 
5569 	hns3_cq_period_mode_init(priv, priv->tx_cqe_mode, priv->rx_cqe_mode);
5570 
5571 	/* the device can work without cpu rmap, only aRFS needs it */
5572 	ret = hns3_set_rx_cpu_rmap(netdev);
5573 	if (ret)
5574 		dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
5575 
5576 	ret = hns3_nic_init_irq(priv);
5577 	if (ret) {
5578 		dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
5579 		hns3_free_rx_cpu_rmap(netdev);
5580 		goto err_init_irq_fail;
5581 	}
5582 
5583 	if (!hns3_is_phys_func(handle->pdev))
5584 		hns3_init_mac_addr(netdev);
5585 
5586 	ret = hns3_client_start(handle);
5587 	if (ret) {
5588 		dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
5589 		goto err_client_start_fail;
5590 	}
5591 
5592 	set_bit(HNS3_NIC_STATE_INITED, &priv->state);
5593 
5594 	return ret;
5595 
5596 err_client_start_fail:
5597 	hns3_free_rx_cpu_rmap(netdev);
5598 	hns3_nic_uninit_irq(priv);
5599 err_init_irq_fail:
5600 	hns3_uninit_all_ring(priv);
5601 err_uninit_vector:
5602 	hns3_nic_uninit_vector_data(priv);
5603 err_dealloc_vector:
5604 	hns3_nic_dealloc_vector_data(priv);
5605 err_put_ring:
5606 	hns3_put_ring_config(priv);
5607 
5608 	return ret;
5609 }
5610 
5611 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
5612 {
5613 	struct net_device *netdev = handle->kinfo.netdev;
5614 	struct hns3_nic_priv *priv = netdev_priv(netdev);
5615 
5616 	if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5617 		netdev_warn(netdev, "already uninitialized\n");
5618 		return 0;
5619 	}
5620 
5621 	hns3_free_rx_cpu_rmap(netdev);
5622 	hns3_nic_uninit_irq(priv);
5623 	hns3_clear_all_ring(handle, true);
5624 	hns3_reset_tx_queue(priv->ae_handle);
5625 
5626 	hns3_nic_uninit_vector_data(priv);
5627 
5628 	hns3_nic_dealloc_vector_data(priv);
5629 
5630 	hns3_uninit_all_ring(priv);
5631 
5632 	hns3_put_ring_config(priv);
5633 
5634 	return 0;
5635 }
5636 
5637 int hns3_reset_notify(struct hnae3_handle *handle,
5638 		      enum hnae3_reset_notify_type type)
5639 {
5640 	int ret = 0;
5641 
5642 	switch (type) {
5643 	case HNAE3_UP_CLIENT:
5644 		ret = hns3_reset_notify_up_enet(handle);
5645 		break;
5646 	case HNAE3_DOWN_CLIENT:
5647 		ret = hns3_reset_notify_down_enet(handle);
5648 		break;
5649 	case HNAE3_INIT_CLIENT:
5650 		ret = hns3_reset_notify_init_enet(handle);
5651 		break;
5652 	case HNAE3_UNINIT_CLIENT:
5653 		ret = hns3_reset_notify_uninit_enet(handle);
5654 		break;
5655 	default:
5656 		break;
5657 	}
5658 
5659 	return ret;
5660 }
5661 
5662 static int hns3_change_channels(struct hnae3_handle *handle, u32 new_tqp_num,
5663 				bool rxfh_configured)
5664 {
5665 	int ret;
5666 
5667 	ret = handle->ae_algo->ops->set_channels(handle, new_tqp_num,
5668 						 rxfh_configured);
5669 	if (ret) {
5670 		dev_err(&handle->pdev->dev,
5671 			"Change tqp num(%u) fail.\n", new_tqp_num);
5672 		return ret;
5673 	}
5674 
5675 	ret = hns3_reset_notify(handle, HNAE3_INIT_CLIENT);
5676 	if (ret)
5677 		return ret;
5678 
5679 	ret =  hns3_reset_notify(handle, HNAE3_UP_CLIENT);
5680 	if (ret)
5681 		hns3_reset_notify(handle, HNAE3_UNINIT_CLIENT);
5682 
5683 	return ret;
5684 }
5685 
5686 int hns3_set_channels(struct net_device *netdev,
5687 		      struct ethtool_channels *ch)
5688 {
5689 	struct hnae3_handle *h = hns3_get_handle(netdev);
5690 	struct hnae3_knic_private_info *kinfo = &h->kinfo;
5691 	bool rxfh_configured = netif_is_rxfh_configured(netdev);
5692 	u32 new_tqp_num = ch->combined_count;
5693 	u16 org_tqp_num;
5694 	int ret;
5695 
5696 	if (hns3_nic_resetting(netdev))
5697 		return -EBUSY;
5698 
5699 	if (ch->rx_count || ch->tx_count)
5700 		return -EINVAL;
5701 
5702 	if (kinfo->tc_info.mqprio_active) {
5703 		dev_err(&netdev->dev,
5704 			"it's not allowed to set channels via ethtool when MQPRIO mode is on\n");
5705 		return -EINVAL;
5706 	}
5707 
5708 	if (new_tqp_num > hns3_get_max_available_channels(h) ||
5709 	    new_tqp_num < 1) {
5710 		dev_err(&netdev->dev,
5711 			"Change tqps fail, the tqp range is from 1 to %u",
5712 			hns3_get_max_available_channels(h));
5713 		return -EINVAL;
5714 	}
5715 
5716 	if (kinfo->rss_size == new_tqp_num)
5717 		return 0;
5718 
5719 	netif_dbg(h, drv, netdev,
5720 		  "set channels: tqp_num=%u, rxfh=%d\n",
5721 		  new_tqp_num, rxfh_configured);
5722 
5723 	ret = hns3_reset_notify(h, HNAE3_DOWN_CLIENT);
5724 	if (ret)
5725 		return ret;
5726 
5727 	ret = hns3_reset_notify(h, HNAE3_UNINIT_CLIENT);
5728 	if (ret)
5729 		return ret;
5730 
5731 	org_tqp_num = h->kinfo.num_tqps;
5732 	ret = hns3_change_channels(h, new_tqp_num, rxfh_configured);
5733 	if (ret) {
5734 		int ret1;
5735 
5736 		netdev_warn(netdev,
5737 			    "Change channels fail, revert to old value\n");
5738 		ret1 = hns3_change_channels(h, org_tqp_num, rxfh_configured);
5739 		if (ret1) {
5740 			netdev_err(netdev,
5741 				   "revert to old channel fail\n");
5742 			return ret1;
5743 		}
5744 
5745 		return ret;
5746 	}
5747 
5748 	return 0;
5749 }
5750 
5751 static const struct hns3_hw_error_info hns3_hw_err[] = {
5752 	{ .type = HNAE3_PPU_POISON_ERROR,
5753 	  .msg = "PPU poison" },
5754 	{ .type = HNAE3_CMDQ_ECC_ERROR,
5755 	  .msg = "IMP CMDQ error" },
5756 	{ .type = HNAE3_IMP_RD_POISON_ERROR,
5757 	  .msg = "IMP RD poison" },
5758 	{ .type = HNAE3_ROCEE_AXI_RESP_ERROR,
5759 	  .msg = "ROCEE AXI RESP error" },
5760 };
5761 
5762 static void hns3_process_hw_error(struct hnae3_handle *handle,
5763 				  enum hnae3_hw_error_type type)
5764 {
5765 	int i;
5766 
5767 	for (i = 0; i < ARRAY_SIZE(hns3_hw_err); i++) {
5768 		if (hns3_hw_err[i].type == type) {
5769 			dev_err(&handle->pdev->dev, "Detected %s!\n",
5770 				hns3_hw_err[i].msg);
5771 			break;
5772 		}
5773 	}
5774 }
5775 
5776 static const struct hnae3_client_ops client_ops = {
5777 	.init_instance = hns3_client_init,
5778 	.uninit_instance = hns3_client_uninit,
5779 	.link_status_change = hns3_link_status_change,
5780 	.reset_notify = hns3_reset_notify,
5781 	.process_hw_error = hns3_process_hw_error,
5782 };
5783 
5784 /* hns3_init_module - Driver registration routine
5785  * hns3_init_module is the first routine called when the driver is
5786  * loaded. All it does is register with the PCI subsystem.
5787  */
5788 static int __init hns3_init_module(void)
5789 {
5790 	int ret;
5791 
5792 	pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
5793 	pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
5794 
5795 	client.type = HNAE3_CLIENT_KNIC;
5796 	snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH, "%s",
5797 		 hns3_driver_name);
5798 
5799 	client.ops = &client_ops;
5800 
5801 	INIT_LIST_HEAD(&client.node);
5802 
5803 	hns3_dbg_register_debugfs(hns3_driver_name);
5804 
5805 	ret = hnae3_register_client(&client);
5806 	if (ret)
5807 		goto err_reg_client;
5808 
5809 	ret = pci_register_driver(&hns3_driver);
5810 	if (ret)
5811 		goto err_reg_driver;
5812 
5813 	return ret;
5814 
5815 err_reg_driver:
5816 	hnae3_unregister_client(&client);
5817 err_reg_client:
5818 	hns3_dbg_unregister_debugfs();
5819 	return ret;
5820 }
5821 module_init(hns3_init_module);
5822 
5823 /* hns3_exit_module - Driver exit cleanup routine
5824  * hns3_exit_module is called just before the driver is removed
5825  * from memory.
5826  */
5827 static void __exit hns3_exit_module(void)
5828 {
5829 	pci_unregister_driver(&hns3_driver);
5830 	hnae3_unregister_client(&client);
5831 	hns3_dbg_unregister_debugfs();
5832 }
5833 module_exit(hns3_exit_module);
5834 
5835 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
5836 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
5837 MODULE_LICENSE("GPL");
5838 MODULE_ALIAS("pci:hns-nic");
5839