xref: /linux/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c (revision 061834624c87282c6d9d8c5395aaff4380e5e1fc)
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 	u32 alloc_size = ring->tqp->handle->kinfo.tx_spare_buf_size;
1032 	struct hns3_tx_spare *tx_spare;
1033 	struct page *page;
1034 	dma_addr_t dma;
1035 	int order;
1036 
1037 	if (!alloc_size)
1038 		return;
1039 
1040 	order = get_order(alloc_size);
1041 	if (order >= MAX_ORDER) {
1042 		if (net_ratelimit())
1043 			dev_warn(ring_to_dev(ring), "failed to allocate tx spare buffer, exceed to max order\n");
1044 		return;
1045 	}
1046 
1047 	tx_spare = devm_kzalloc(ring_to_dev(ring), sizeof(*tx_spare),
1048 				GFP_KERNEL);
1049 	if (!tx_spare) {
1050 		/* The driver still work without the tx spare buffer */
1051 		dev_warn(ring_to_dev(ring), "failed to allocate hns3_tx_spare\n");
1052 		goto devm_kzalloc_error;
1053 	}
1054 
1055 	page = alloc_pages_node(dev_to_node(ring_to_dev(ring)),
1056 				GFP_KERNEL, order);
1057 	if (!page) {
1058 		dev_warn(ring_to_dev(ring), "failed to allocate tx spare pages\n");
1059 		goto alloc_pages_error;
1060 	}
1061 
1062 	dma = dma_map_page(ring_to_dev(ring), page, 0,
1063 			   PAGE_SIZE << order, DMA_TO_DEVICE);
1064 	if (dma_mapping_error(ring_to_dev(ring), dma)) {
1065 		dev_warn(ring_to_dev(ring), "failed to map pages for tx spare\n");
1066 		goto dma_mapping_error;
1067 	}
1068 
1069 	tx_spare->dma = dma;
1070 	tx_spare->buf = page_address(page);
1071 	tx_spare->len = PAGE_SIZE << order;
1072 	ring->tx_spare = tx_spare;
1073 	return;
1074 
1075 dma_mapping_error:
1076 	put_page(page);
1077 alloc_pages_error:
1078 	devm_kfree(ring_to_dev(ring), tx_spare);
1079 devm_kzalloc_error:
1080 	ring->tqp->handle->kinfo.tx_spare_buf_size = 0;
1081 }
1082 
1083 /* Use hns3_tx_spare_space() to make sure there is enough buffer
1084  * before calling below function to allocate tx buffer.
1085  */
1086 static void *hns3_tx_spare_alloc(struct hns3_enet_ring *ring,
1087 				 unsigned int size, dma_addr_t *dma,
1088 				 u32 *cb_len)
1089 {
1090 	struct hns3_tx_spare *tx_spare = ring->tx_spare;
1091 	u32 ntu = tx_spare->next_to_use;
1092 
1093 	size = ALIGN(size, dma_get_cache_alignment());
1094 	*cb_len = size;
1095 
1096 	/* Tx spare buffer wraps back here because the end of
1097 	 * freed tx buffer is not enough.
1098 	 */
1099 	if (ntu + size > tx_spare->len) {
1100 		*cb_len += (tx_spare->len - ntu);
1101 		ntu = 0;
1102 	}
1103 
1104 	tx_spare->next_to_use = ntu + size;
1105 	if (tx_spare->next_to_use == tx_spare->len)
1106 		tx_spare->next_to_use = 0;
1107 
1108 	*dma = tx_spare->dma + ntu;
1109 
1110 	return tx_spare->buf + ntu;
1111 }
1112 
1113 static void hns3_tx_spare_rollback(struct hns3_enet_ring *ring, u32 len)
1114 {
1115 	struct hns3_tx_spare *tx_spare = ring->tx_spare;
1116 
1117 	if (len > tx_spare->next_to_use) {
1118 		len -= tx_spare->next_to_use;
1119 		tx_spare->next_to_use = tx_spare->len - len;
1120 	} else {
1121 		tx_spare->next_to_use -= len;
1122 	}
1123 }
1124 
1125 static void hns3_tx_spare_reclaim_cb(struct hns3_enet_ring *ring,
1126 				     struct hns3_desc_cb *cb)
1127 {
1128 	struct hns3_tx_spare *tx_spare = ring->tx_spare;
1129 	u32 ntc = tx_spare->next_to_clean;
1130 	u32 len = cb->length;
1131 
1132 	tx_spare->next_to_clean += len;
1133 
1134 	if (tx_spare->next_to_clean >= tx_spare->len) {
1135 		tx_spare->next_to_clean -= tx_spare->len;
1136 
1137 		if (tx_spare->next_to_clean) {
1138 			ntc = 0;
1139 			len = tx_spare->next_to_clean;
1140 		}
1141 	}
1142 
1143 	/* This tx spare buffer is only really reclaimed after calling
1144 	 * hns3_tx_spare_update(), so it is still safe to use the info in
1145 	 * the tx buffer to do the dma sync or sg unmapping after
1146 	 * tx_spare->next_to_clean is moved forword.
1147 	 */
1148 	if (cb->type & (DESC_TYPE_BOUNCE_HEAD | DESC_TYPE_BOUNCE_ALL)) {
1149 		dma_addr_t dma = tx_spare->dma + ntc;
1150 
1151 		dma_sync_single_for_cpu(ring_to_dev(ring), dma, len,
1152 					DMA_TO_DEVICE);
1153 	} else {
1154 		struct sg_table *sgt = tx_spare->buf + ntc;
1155 
1156 		dma_unmap_sg(ring_to_dev(ring), sgt->sgl, sgt->orig_nents,
1157 			     DMA_TO_DEVICE);
1158 	}
1159 }
1160 
1161 static int hns3_set_tso(struct sk_buff *skb, u32 *paylen_fdop_ol4cs,
1162 			u16 *mss, u32 *type_cs_vlan_tso, u32 *send_bytes)
1163 {
1164 	u32 l4_offset, hdr_len;
1165 	union l3_hdr_info l3;
1166 	union l4_hdr_info l4;
1167 	u32 l4_paylen;
1168 	int ret;
1169 
1170 	if (!skb_is_gso(skb))
1171 		return 0;
1172 
1173 	ret = skb_cow_head(skb, 0);
1174 	if (unlikely(ret < 0))
1175 		return ret;
1176 
1177 	l3.hdr = skb_network_header(skb);
1178 	l4.hdr = skb_transport_header(skb);
1179 
1180 	/* Software should clear the IPv4's checksum field when tso is
1181 	 * needed.
1182 	 */
1183 	if (l3.v4->version == 4)
1184 		l3.v4->check = 0;
1185 
1186 	/* tunnel packet */
1187 	if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
1188 					 SKB_GSO_GRE_CSUM |
1189 					 SKB_GSO_UDP_TUNNEL |
1190 					 SKB_GSO_UDP_TUNNEL_CSUM)) {
1191 		/* reset l3&l4 pointers from outer to inner headers */
1192 		l3.hdr = skb_inner_network_header(skb);
1193 		l4.hdr = skb_inner_transport_header(skb);
1194 
1195 		/* Software should clear the IPv4's checksum field when
1196 		 * tso is needed.
1197 		 */
1198 		if (l3.v4->version == 4)
1199 			l3.v4->check = 0;
1200 	}
1201 
1202 	/* normal or tunnel packet */
1203 	l4_offset = l4.hdr - skb->data;
1204 
1205 	/* remove payload length from inner pseudo checksum when tso */
1206 	l4_paylen = skb->len - l4_offset;
1207 
1208 	if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) {
1209 		hdr_len = sizeof(*l4.udp) + l4_offset;
1210 		csum_replace_by_diff(&l4.udp->check,
1211 				     (__force __wsum)htonl(l4_paylen));
1212 	} else {
1213 		hdr_len = (l4.tcp->doff << 2) + l4_offset;
1214 		csum_replace_by_diff(&l4.tcp->check,
1215 				     (__force __wsum)htonl(l4_paylen));
1216 	}
1217 
1218 	*send_bytes = (skb_shinfo(skb)->gso_segs - 1) * hdr_len + skb->len;
1219 
1220 	/* find the txbd field values */
1221 	*paylen_fdop_ol4cs = skb->len - hdr_len;
1222 	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_TSO_B, 1);
1223 
1224 	/* offload outer UDP header checksum */
1225 	if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)
1226 		hns3_set_field(*paylen_fdop_ol4cs, HNS3_TXD_OL4CS_B, 1);
1227 
1228 	/* get MSS for TSO */
1229 	*mss = skb_shinfo(skb)->gso_size;
1230 
1231 	trace_hns3_tso(skb);
1232 
1233 	return 0;
1234 }
1235 
1236 static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
1237 				u8 *il4_proto)
1238 {
1239 	union l3_hdr_info l3;
1240 	unsigned char *l4_hdr;
1241 	unsigned char *exthdr;
1242 	u8 l4_proto_tmp;
1243 	__be16 frag_off;
1244 
1245 	/* find outer header point */
1246 	l3.hdr = skb_network_header(skb);
1247 	l4_hdr = skb_transport_header(skb);
1248 
1249 	if (skb->protocol == htons(ETH_P_IPV6)) {
1250 		exthdr = l3.hdr + sizeof(*l3.v6);
1251 		l4_proto_tmp = l3.v6->nexthdr;
1252 		if (l4_hdr != exthdr)
1253 			ipv6_skip_exthdr(skb, exthdr - skb->data,
1254 					 &l4_proto_tmp, &frag_off);
1255 	} else if (skb->protocol == htons(ETH_P_IP)) {
1256 		l4_proto_tmp = l3.v4->protocol;
1257 	} else {
1258 		return -EINVAL;
1259 	}
1260 
1261 	*ol4_proto = l4_proto_tmp;
1262 
1263 	/* tunnel packet */
1264 	if (!skb->encapsulation) {
1265 		*il4_proto = 0;
1266 		return 0;
1267 	}
1268 
1269 	/* find inner header point */
1270 	l3.hdr = skb_inner_network_header(skb);
1271 	l4_hdr = skb_inner_transport_header(skb);
1272 
1273 	if (l3.v6->version == 6) {
1274 		exthdr = l3.hdr + sizeof(*l3.v6);
1275 		l4_proto_tmp = l3.v6->nexthdr;
1276 		if (l4_hdr != exthdr)
1277 			ipv6_skip_exthdr(skb, exthdr - skb->data,
1278 					 &l4_proto_tmp, &frag_off);
1279 	} else if (l3.v4->version == 4) {
1280 		l4_proto_tmp = l3.v4->protocol;
1281 	}
1282 
1283 	*il4_proto = l4_proto_tmp;
1284 
1285 	return 0;
1286 }
1287 
1288 /* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
1289  * and it is udp packet, which has a dest port as the IANA assigned.
1290  * the hardware is expected to do the checksum offload, but the
1291  * hardware will not do the checksum offload when udp dest port is
1292  * 4789, 4790 or 6081.
1293  */
1294 static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
1295 {
1296 	struct hns3_nic_priv *priv = netdev_priv(skb->dev);
1297 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
1298 	union l4_hdr_info l4;
1299 
1300 	/* device version above V3(include V3), the hardware can
1301 	 * do this checksum offload.
1302 	 */
1303 	if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
1304 		return false;
1305 
1306 	l4.hdr = skb_transport_header(skb);
1307 
1308 	if (!(!skb->encapsulation &&
1309 	      (l4.udp->dest == htons(IANA_VXLAN_UDP_PORT) ||
1310 	      l4.udp->dest == htons(GENEVE_UDP_PORT) ||
1311 	      l4.udp->dest == htons(IANA_VXLAN_GPE_UDP_PORT))))
1312 		return false;
1313 
1314 	return true;
1315 }
1316 
1317 static void hns3_set_outer_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
1318 				  u32 *ol_type_vlan_len_msec)
1319 {
1320 	u32 l2_len, l3_len, l4_len;
1321 	unsigned char *il2_hdr;
1322 	union l3_hdr_info l3;
1323 	union l4_hdr_info l4;
1324 
1325 	l3.hdr = skb_network_header(skb);
1326 	l4.hdr = skb_transport_header(skb);
1327 
1328 	/* compute OL2 header size, defined in 2 Bytes */
1329 	l2_len = l3.hdr - skb->data;
1330 	hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L2LEN_S, l2_len >> 1);
1331 
1332 	/* compute OL3 header size, defined in 4 Bytes */
1333 	l3_len = l4.hdr - l3.hdr;
1334 	hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_S, l3_len >> 2);
1335 
1336 	il2_hdr = skb_inner_mac_header(skb);
1337 	/* compute OL4 header size, defined in 4 Bytes */
1338 	l4_len = il2_hdr - l4.hdr;
1339 	hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L4LEN_S, l4_len >> 2);
1340 
1341 	/* define outer network header type */
1342 	if (skb->protocol == htons(ETH_P_IP)) {
1343 		if (skb_is_gso(skb))
1344 			hns3_set_field(*ol_type_vlan_len_msec,
1345 				       HNS3_TXD_OL3T_S,
1346 				       HNS3_OL3T_IPV4_CSUM);
1347 		else
1348 			hns3_set_field(*ol_type_vlan_len_msec,
1349 				       HNS3_TXD_OL3T_S,
1350 				       HNS3_OL3T_IPV4_NO_CSUM);
1351 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
1352 		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_S,
1353 			       HNS3_OL3T_IPV6);
1354 	}
1355 
1356 	if (ol4_proto == IPPROTO_UDP)
1357 		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
1358 			       HNS3_TUN_MAC_IN_UDP);
1359 	else if (ol4_proto == IPPROTO_GRE)
1360 		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
1361 			       HNS3_TUN_NVGRE);
1362 }
1363 
1364 static void hns3_set_l3_type(struct sk_buff *skb, union l3_hdr_info l3,
1365 			     u32 *type_cs_vlan_tso)
1366 {
1367 	if (l3.v4->version == 4) {
1368 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
1369 			       HNS3_L3T_IPV4);
1370 
1371 		/* the stack computes the IP header already, the only time we
1372 		 * need the hardware to recompute it is in the case of TSO.
1373 		 */
1374 		if (skb_is_gso(skb))
1375 			hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3CS_B, 1);
1376 	} else if (l3.v6->version == 6) {
1377 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
1378 			       HNS3_L3T_IPV6);
1379 	}
1380 }
1381 
1382 static int hns3_set_l4_csum_length(struct sk_buff *skb, union l4_hdr_info l4,
1383 				   u32 l4_proto, u32 *type_cs_vlan_tso)
1384 {
1385 	/* compute inner(/normal) L4 header size, defined in 4 Bytes */
1386 	switch (l4_proto) {
1387 	case IPPROTO_TCP:
1388 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1389 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1390 			       HNS3_L4T_TCP);
1391 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1392 			       l4.tcp->doff);
1393 		break;
1394 	case IPPROTO_UDP:
1395 		if (hns3_tunnel_csum_bug(skb)) {
1396 			int ret = skb_put_padto(skb, HNS3_MIN_TUN_PKT_LEN);
1397 
1398 			return ret ? ret : skb_checksum_help(skb);
1399 		}
1400 
1401 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1402 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1403 			       HNS3_L4T_UDP);
1404 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1405 			       (sizeof(struct udphdr) >> 2));
1406 		break;
1407 	case IPPROTO_SCTP:
1408 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
1409 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
1410 			       HNS3_L4T_SCTP);
1411 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
1412 			       (sizeof(struct sctphdr) >> 2));
1413 		break;
1414 	default:
1415 		/* drop the skb tunnel packet if hardware don't support,
1416 		 * because hardware can't calculate csum when TSO.
1417 		 */
1418 		if (skb_is_gso(skb))
1419 			return -EDOM;
1420 
1421 		/* the stack computes the IP header already,
1422 		 * driver calculate l4 checksum when not TSO.
1423 		 */
1424 		return skb_checksum_help(skb);
1425 	}
1426 
1427 	return 0;
1428 }
1429 
1430 static int hns3_set_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
1431 			   u8 il4_proto, u32 *type_cs_vlan_tso,
1432 			   u32 *ol_type_vlan_len_msec)
1433 {
1434 	unsigned char *l2_hdr = skb->data;
1435 	u32 l4_proto = ol4_proto;
1436 	union l4_hdr_info l4;
1437 	union l3_hdr_info l3;
1438 	u32 l2_len, l3_len;
1439 
1440 	l4.hdr = skb_transport_header(skb);
1441 	l3.hdr = skb_network_header(skb);
1442 
1443 	/* handle encapsulation skb */
1444 	if (skb->encapsulation) {
1445 		/* If this is a not UDP/GRE encapsulation skb */
1446 		if (!(ol4_proto == IPPROTO_UDP || ol4_proto == IPPROTO_GRE)) {
1447 			/* drop the skb tunnel packet if hardware don't support,
1448 			 * because hardware can't calculate csum when TSO.
1449 			 */
1450 			if (skb_is_gso(skb))
1451 				return -EDOM;
1452 
1453 			/* the stack computes the IP header already,
1454 			 * driver calculate l4 checksum when not TSO.
1455 			 */
1456 			return skb_checksum_help(skb);
1457 		}
1458 
1459 		hns3_set_outer_l2l3l4(skb, ol4_proto, ol_type_vlan_len_msec);
1460 
1461 		/* switch to inner header */
1462 		l2_hdr = skb_inner_mac_header(skb);
1463 		l3.hdr = skb_inner_network_header(skb);
1464 		l4.hdr = skb_inner_transport_header(skb);
1465 		l4_proto = il4_proto;
1466 	}
1467 
1468 	hns3_set_l3_type(skb, l3, type_cs_vlan_tso);
1469 
1470 	/* compute inner(/normal) L2 header size, defined in 2 Bytes */
1471 	l2_len = l3.hdr - l2_hdr;
1472 	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S, l2_len >> 1);
1473 
1474 	/* compute inner(/normal) L3 header size, defined in 4 Bytes */
1475 	l3_len = l4.hdr - l3.hdr;
1476 	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_S, l3_len >> 2);
1477 
1478 	return hns3_set_l4_csum_length(skb, l4, l4_proto, type_cs_vlan_tso);
1479 }
1480 
1481 static int hns3_handle_vtags(struct hns3_enet_ring *tx_ring,
1482 			     struct sk_buff *skb)
1483 {
1484 	struct hnae3_handle *handle = tx_ring->tqp->handle;
1485 	struct hnae3_ae_dev *ae_dev;
1486 	struct vlan_ethhdr *vhdr;
1487 	int rc;
1488 
1489 	if (!(skb->protocol == htons(ETH_P_8021Q) ||
1490 	      skb_vlan_tag_present(skb)))
1491 		return 0;
1492 
1493 	/* For HW limitation on HNAE3_DEVICE_VERSION_V2, if port based insert
1494 	 * VLAN enabled, only one VLAN header is allowed in skb, otherwise it
1495 	 * will cause RAS error.
1496 	 */
1497 	ae_dev = pci_get_drvdata(handle->pdev);
1498 	if (unlikely(skb_vlan_tagged_multi(skb) &&
1499 		     ae_dev->dev_version <= HNAE3_DEVICE_VERSION_V2 &&
1500 		     handle->port_base_vlan_state ==
1501 		     HNAE3_PORT_BASE_VLAN_ENABLE))
1502 		return -EINVAL;
1503 
1504 	if (skb->protocol == htons(ETH_P_8021Q) &&
1505 	    !(handle->kinfo.netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
1506 		/* When HW VLAN acceleration is turned off, and the stack
1507 		 * sets the protocol to 802.1q, the driver just need to
1508 		 * set the protocol to the encapsulated ethertype.
1509 		 */
1510 		skb->protocol = vlan_get_protocol(skb);
1511 		return 0;
1512 	}
1513 
1514 	if (skb_vlan_tag_present(skb)) {
1515 		/* Based on hw strategy, use out_vtag in two layer tag case,
1516 		 * and use inner_vtag in one tag case.
1517 		 */
1518 		if (skb->protocol == htons(ETH_P_8021Q) &&
1519 		    handle->port_base_vlan_state ==
1520 		    HNAE3_PORT_BASE_VLAN_DISABLE)
1521 			rc = HNS3_OUTER_VLAN_TAG;
1522 		else
1523 			rc = HNS3_INNER_VLAN_TAG;
1524 
1525 		skb->protocol = vlan_get_protocol(skb);
1526 		return rc;
1527 	}
1528 
1529 	rc = skb_cow_head(skb, 0);
1530 	if (unlikely(rc < 0))
1531 		return rc;
1532 
1533 	vhdr = (struct vlan_ethhdr *)skb->data;
1534 	vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority << VLAN_PRIO_SHIFT)
1535 					 & VLAN_PRIO_MASK);
1536 
1537 	skb->protocol = vlan_get_protocol(skb);
1538 	return 0;
1539 }
1540 
1541 /* check if the hardware is capable of checksum offloading */
1542 static bool hns3_check_hw_tx_csum(struct sk_buff *skb)
1543 {
1544 	struct hns3_nic_priv *priv = netdev_priv(skb->dev);
1545 
1546 	/* Kindly note, due to backward compatibility of the TX descriptor,
1547 	 * HW checksum of the non-IP packets and GSO packets is handled at
1548 	 * different place in the following code
1549 	 */
1550 	if (skb_csum_is_sctp(skb) || skb_is_gso(skb) ||
1551 	    !test_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state))
1552 		return false;
1553 
1554 	return true;
1555 }
1556 
1557 struct hns3_desc_param {
1558 	u32 paylen_ol4cs;
1559 	u32 ol_type_vlan_len_msec;
1560 	u32 type_cs_vlan_tso;
1561 	u16 mss_hw_csum;
1562 	u16 inner_vtag;
1563 	u16 out_vtag;
1564 };
1565 
1566 static void hns3_init_desc_data(struct sk_buff *skb, struct hns3_desc_param *pa)
1567 {
1568 	pa->paylen_ol4cs = skb->len;
1569 	pa->ol_type_vlan_len_msec = 0;
1570 	pa->type_cs_vlan_tso = 0;
1571 	pa->mss_hw_csum = 0;
1572 	pa->inner_vtag = 0;
1573 	pa->out_vtag = 0;
1574 }
1575 
1576 static int hns3_handle_vlan_info(struct hns3_enet_ring *ring,
1577 				 struct sk_buff *skb,
1578 				 struct hns3_desc_param *param)
1579 {
1580 	int ret;
1581 
1582 	ret = hns3_handle_vtags(ring, skb);
1583 	if (unlikely(ret < 0)) {
1584 		hns3_ring_stats_update(ring, tx_vlan_err);
1585 		return ret;
1586 	} else if (ret == HNS3_INNER_VLAN_TAG) {
1587 		param->inner_vtag = skb_vlan_tag_get(skb);
1588 		param->inner_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1589 				VLAN_PRIO_MASK;
1590 		hns3_set_field(param->type_cs_vlan_tso, HNS3_TXD_VLAN_B, 1);
1591 	} else if (ret == HNS3_OUTER_VLAN_TAG) {
1592 		param->out_vtag = skb_vlan_tag_get(skb);
1593 		param->out_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1594 				VLAN_PRIO_MASK;
1595 		hns3_set_field(param->ol_type_vlan_len_msec, HNS3_TXD_OVLAN_B,
1596 			       1);
1597 	}
1598 	return 0;
1599 }
1600 
1601 static int hns3_handle_csum_partial(struct hns3_enet_ring *ring,
1602 				    struct sk_buff *skb,
1603 				    struct hns3_desc_cb *desc_cb,
1604 				    struct hns3_desc_param *param)
1605 {
1606 	u8 ol4_proto, il4_proto;
1607 	int ret;
1608 
1609 	if (hns3_check_hw_tx_csum(skb)) {
1610 		/* set checksum start and offset, defined in 2 Bytes */
1611 		hns3_set_field(param->type_cs_vlan_tso, HNS3_TXD_CSUM_START_S,
1612 			       skb_checksum_start_offset(skb) >> 1);
1613 		hns3_set_field(param->ol_type_vlan_len_msec,
1614 			       HNS3_TXD_CSUM_OFFSET_S,
1615 			       skb->csum_offset >> 1);
1616 		param->mss_hw_csum |= BIT(HNS3_TXD_HW_CS_B);
1617 		return 0;
1618 	}
1619 
1620 	skb_reset_mac_len(skb);
1621 
1622 	ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
1623 	if (unlikely(ret < 0)) {
1624 		hns3_ring_stats_update(ring, tx_l4_proto_err);
1625 		return ret;
1626 	}
1627 
1628 	ret = hns3_set_l2l3l4(skb, ol4_proto, il4_proto,
1629 			      &param->type_cs_vlan_tso,
1630 			      &param->ol_type_vlan_len_msec);
1631 	if (unlikely(ret < 0)) {
1632 		hns3_ring_stats_update(ring, tx_l2l3l4_err);
1633 		return ret;
1634 	}
1635 
1636 	ret = hns3_set_tso(skb, &param->paylen_ol4cs, &param->mss_hw_csum,
1637 			   &param->type_cs_vlan_tso, &desc_cb->send_bytes);
1638 	if (unlikely(ret < 0)) {
1639 		hns3_ring_stats_update(ring, tx_tso_err);
1640 		return ret;
1641 	}
1642 	return 0;
1643 }
1644 
1645 static int hns3_fill_skb_desc(struct hns3_enet_ring *ring,
1646 			      struct sk_buff *skb, struct hns3_desc *desc,
1647 			      struct hns3_desc_cb *desc_cb)
1648 {
1649 	struct hns3_desc_param param;
1650 	int ret;
1651 
1652 	hns3_init_desc_data(skb, &param);
1653 	ret = hns3_handle_vlan_info(ring, skb, &param);
1654 	if (unlikely(ret < 0))
1655 		return ret;
1656 
1657 	desc_cb->send_bytes = skb->len;
1658 
1659 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1660 		ret = hns3_handle_csum_partial(ring, skb, desc_cb, &param);
1661 		if (ret)
1662 			return ret;
1663 	}
1664 
1665 	/* Set txbd */
1666 	desc->tx.ol_type_vlan_len_msec =
1667 		cpu_to_le32(param.ol_type_vlan_len_msec);
1668 	desc->tx.type_cs_vlan_tso_len = cpu_to_le32(param.type_cs_vlan_tso);
1669 	desc->tx.paylen_ol4cs = cpu_to_le32(param.paylen_ol4cs);
1670 	desc->tx.mss_hw_csum = cpu_to_le16(param.mss_hw_csum);
1671 	desc->tx.vlan_tag = cpu_to_le16(param.inner_vtag);
1672 	desc->tx.outer_vlan_tag = cpu_to_le16(param.out_vtag);
1673 
1674 	return 0;
1675 }
1676 
1677 static int hns3_fill_desc(struct hns3_enet_ring *ring, dma_addr_t dma,
1678 			  unsigned int size)
1679 {
1680 #define HNS3_LIKELY_BD_NUM	1
1681 
1682 	struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1683 	unsigned int frag_buf_num;
1684 	int k, sizeoflast;
1685 
1686 	if (likely(size <= HNS3_MAX_BD_SIZE)) {
1687 		desc->addr = cpu_to_le64(dma);
1688 		desc->tx.send_size = cpu_to_le16(size);
1689 		desc->tx.bdtp_fe_sc_vld_ra_ri =
1690 			cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1691 
1692 		trace_hns3_tx_desc(ring, ring->next_to_use);
1693 		ring_ptr_move_fw(ring, next_to_use);
1694 		return HNS3_LIKELY_BD_NUM;
1695 	}
1696 
1697 	frag_buf_num = hns3_tx_bd_count(size);
1698 	sizeoflast = size % HNS3_MAX_BD_SIZE;
1699 	sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1700 
1701 	/* When frag size is bigger than hardware limit, split this frag */
1702 	for (k = 0; k < frag_buf_num; k++) {
1703 		/* now, fill the descriptor */
1704 		desc->addr = cpu_to_le64(dma + HNS3_MAX_BD_SIZE * k);
1705 		desc->tx.send_size = cpu_to_le16((k == frag_buf_num - 1) ?
1706 				     (u16)sizeoflast : (u16)HNS3_MAX_BD_SIZE);
1707 		desc->tx.bdtp_fe_sc_vld_ra_ri =
1708 				cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1709 
1710 		trace_hns3_tx_desc(ring, ring->next_to_use);
1711 		/* move ring pointer to next */
1712 		ring_ptr_move_fw(ring, next_to_use);
1713 
1714 		desc = &ring->desc[ring->next_to_use];
1715 	}
1716 
1717 	return frag_buf_num;
1718 }
1719 
1720 static int hns3_map_and_fill_desc(struct hns3_enet_ring *ring, void *priv,
1721 				  unsigned int type)
1722 {
1723 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1724 	struct device *dev = ring_to_dev(ring);
1725 	unsigned int size;
1726 	dma_addr_t dma;
1727 
1728 	if (type & (DESC_TYPE_FRAGLIST_SKB | DESC_TYPE_SKB)) {
1729 		struct sk_buff *skb = (struct sk_buff *)priv;
1730 
1731 		size = skb_headlen(skb);
1732 		if (!size)
1733 			return 0;
1734 
1735 		dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1736 	} else if (type & DESC_TYPE_BOUNCE_HEAD) {
1737 		/* Head data has been filled in hns3_handle_tx_bounce(),
1738 		 * just return 0 here.
1739 		 */
1740 		return 0;
1741 	} else {
1742 		skb_frag_t *frag = (skb_frag_t *)priv;
1743 
1744 		size = skb_frag_size(frag);
1745 		if (!size)
1746 			return 0;
1747 
1748 		dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1749 	}
1750 
1751 	if (unlikely(dma_mapping_error(dev, dma))) {
1752 		hns3_ring_stats_update(ring, sw_err_cnt);
1753 		return -ENOMEM;
1754 	}
1755 
1756 	desc_cb->priv = priv;
1757 	desc_cb->length = size;
1758 	desc_cb->dma = dma;
1759 	desc_cb->type = type;
1760 
1761 	return hns3_fill_desc(ring, dma, size);
1762 }
1763 
1764 static unsigned int hns3_skb_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1765 				    unsigned int bd_num)
1766 {
1767 	unsigned int size;
1768 	int i;
1769 
1770 	size = skb_headlen(skb);
1771 	while (size > HNS3_MAX_BD_SIZE) {
1772 		bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1773 		size -= HNS3_MAX_BD_SIZE;
1774 
1775 		if (bd_num > HNS3_MAX_TSO_BD_NUM)
1776 			return bd_num;
1777 	}
1778 
1779 	if (size) {
1780 		bd_size[bd_num++] = size;
1781 		if (bd_num > HNS3_MAX_TSO_BD_NUM)
1782 			return bd_num;
1783 	}
1784 
1785 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1786 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1787 		size = skb_frag_size(frag);
1788 		if (!size)
1789 			continue;
1790 
1791 		while (size > HNS3_MAX_BD_SIZE) {
1792 			bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1793 			size -= HNS3_MAX_BD_SIZE;
1794 
1795 			if (bd_num > HNS3_MAX_TSO_BD_NUM)
1796 				return bd_num;
1797 		}
1798 
1799 		bd_size[bd_num++] = size;
1800 		if (bd_num > HNS3_MAX_TSO_BD_NUM)
1801 			return bd_num;
1802 	}
1803 
1804 	return bd_num;
1805 }
1806 
1807 static unsigned int hns3_tx_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1808 				   u8 max_non_tso_bd_num, unsigned int bd_num,
1809 				   unsigned int recursion_level)
1810 {
1811 #define HNS3_MAX_RECURSION_LEVEL	24
1812 
1813 	struct sk_buff *frag_skb;
1814 
1815 	/* If the total len is within the max bd limit */
1816 	if (likely(skb->len <= HNS3_MAX_BD_SIZE && !recursion_level &&
1817 		   !skb_has_frag_list(skb) &&
1818 		   skb_shinfo(skb)->nr_frags < max_non_tso_bd_num))
1819 		return skb_shinfo(skb)->nr_frags + 1U;
1820 
1821 	if (unlikely(recursion_level >= HNS3_MAX_RECURSION_LEVEL))
1822 		return UINT_MAX;
1823 
1824 	bd_num = hns3_skb_bd_num(skb, bd_size, bd_num);
1825 	if (!skb_has_frag_list(skb) || bd_num > HNS3_MAX_TSO_BD_NUM)
1826 		return bd_num;
1827 
1828 	skb_walk_frags(skb, frag_skb) {
1829 		bd_num = hns3_tx_bd_num(frag_skb, bd_size, max_non_tso_bd_num,
1830 					bd_num, recursion_level + 1);
1831 		if (bd_num > HNS3_MAX_TSO_BD_NUM)
1832 			return bd_num;
1833 	}
1834 
1835 	return bd_num;
1836 }
1837 
1838 static unsigned int hns3_gso_hdr_len(struct sk_buff *skb)
1839 {
1840 	if (!skb->encapsulation)
1841 		return skb_tcp_all_headers(skb);
1842 
1843 	return skb_inner_tcp_all_headers(skb);
1844 }
1845 
1846 /* HW need every continuous max_non_tso_bd_num buffer data to be larger
1847  * than MSS, we simplify it by ensuring skb_headlen + the first continuous
1848  * max_non_tso_bd_num - 1 frags to be larger than gso header len + mss,
1849  * and the remaining continuous max_non_tso_bd_num - 1 frags to be larger
1850  * than MSS except the last max_non_tso_bd_num - 1 frags.
1851  */
1852 static bool hns3_skb_need_linearized(struct sk_buff *skb, unsigned int *bd_size,
1853 				     unsigned int bd_num, u8 max_non_tso_bd_num)
1854 {
1855 	unsigned int tot_len = 0;
1856 	int i;
1857 
1858 	for (i = 0; i < max_non_tso_bd_num - 1U; i++)
1859 		tot_len += bd_size[i];
1860 
1861 	/* ensure the first max_non_tso_bd_num frags is greater than
1862 	 * mss + header
1863 	 */
1864 	if (tot_len + bd_size[max_non_tso_bd_num - 1U] <
1865 	    skb_shinfo(skb)->gso_size + hns3_gso_hdr_len(skb))
1866 		return true;
1867 
1868 	/* ensure every continuous max_non_tso_bd_num - 1 buffer is greater
1869 	 * than mss except the last one.
1870 	 */
1871 	for (i = 0; i < bd_num - max_non_tso_bd_num; i++) {
1872 		tot_len -= bd_size[i];
1873 		tot_len += bd_size[i + max_non_tso_bd_num - 1U];
1874 
1875 		if (tot_len < skb_shinfo(skb)->gso_size)
1876 			return true;
1877 	}
1878 
1879 	return false;
1880 }
1881 
1882 void hns3_shinfo_pack(struct skb_shared_info *shinfo, __u32 *size)
1883 {
1884 	int i;
1885 
1886 	for (i = 0; i < MAX_SKB_FRAGS; i++)
1887 		size[i] = skb_frag_size(&shinfo->frags[i]);
1888 }
1889 
1890 static int hns3_skb_linearize(struct hns3_enet_ring *ring,
1891 			      struct sk_buff *skb,
1892 			      unsigned int bd_num)
1893 {
1894 	/* 'bd_num == UINT_MAX' means the skb' fraglist has a
1895 	 * recursion level of over HNS3_MAX_RECURSION_LEVEL.
1896 	 */
1897 	if (bd_num == UINT_MAX) {
1898 		hns3_ring_stats_update(ring, over_max_recursion);
1899 		return -ENOMEM;
1900 	}
1901 
1902 	/* The skb->len has exceeded the hw limitation, linearization
1903 	 * will not help.
1904 	 */
1905 	if (skb->len > HNS3_MAX_TSO_SIZE ||
1906 	    (!skb_is_gso(skb) && skb->len > HNS3_MAX_NON_TSO_SIZE)) {
1907 		hns3_ring_stats_update(ring, hw_limitation);
1908 		return -ENOMEM;
1909 	}
1910 
1911 	if (__skb_linearize(skb)) {
1912 		hns3_ring_stats_update(ring, sw_err_cnt);
1913 		return -ENOMEM;
1914 	}
1915 
1916 	return 0;
1917 }
1918 
1919 static int hns3_nic_maybe_stop_tx(struct hns3_enet_ring *ring,
1920 				  struct net_device *netdev,
1921 				  struct sk_buff *skb)
1922 {
1923 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1924 	u8 max_non_tso_bd_num = priv->max_non_tso_bd_num;
1925 	unsigned int bd_size[HNS3_MAX_TSO_BD_NUM + 1U];
1926 	unsigned int bd_num;
1927 
1928 	bd_num = hns3_tx_bd_num(skb, bd_size, max_non_tso_bd_num, 0, 0);
1929 	if (unlikely(bd_num > max_non_tso_bd_num)) {
1930 		if (bd_num <= HNS3_MAX_TSO_BD_NUM && skb_is_gso(skb) &&
1931 		    !hns3_skb_need_linearized(skb, bd_size, bd_num,
1932 					      max_non_tso_bd_num)) {
1933 			trace_hns3_over_max_bd(skb);
1934 			goto out;
1935 		}
1936 
1937 		if (hns3_skb_linearize(ring, skb, bd_num))
1938 			return -ENOMEM;
1939 
1940 		bd_num = hns3_tx_bd_count(skb->len);
1941 
1942 		hns3_ring_stats_update(ring, tx_copy);
1943 	}
1944 
1945 out:
1946 	if (likely(ring_space(ring) >= bd_num))
1947 		return bd_num;
1948 
1949 	netif_stop_subqueue(netdev, ring->queue_index);
1950 	smp_mb(); /* Memory barrier before checking ring_space */
1951 
1952 	/* Start queue in case hns3_clean_tx_ring has just made room
1953 	 * available and has not seen the queue stopped state performed
1954 	 * by netif_stop_subqueue above.
1955 	 */
1956 	if (ring_space(ring) >= bd_num && netif_carrier_ok(netdev) &&
1957 	    !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
1958 		netif_start_subqueue(netdev, ring->queue_index);
1959 		return bd_num;
1960 	}
1961 
1962 	hns3_ring_stats_update(ring, tx_busy);
1963 
1964 	return -EBUSY;
1965 }
1966 
1967 static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
1968 {
1969 	struct device *dev = ring_to_dev(ring);
1970 	unsigned int i;
1971 
1972 	for (i = 0; i < ring->desc_num; i++) {
1973 		struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1974 		struct hns3_desc_cb *desc_cb;
1975 
1976 		memset(desc, 0, sizeof(*desc));
1977 
1978 		/* check if this is where we started */
1979 		if (ring->next_to_use == next_to_use_orig)
1980 			break;
1981 
1982 		/* rollback one */
1983 		ring_ptr_move_bw(ring, next_to_use);
1984 
1985 		desc_cb = &ring->desc_cb[ring->next_to_use];
1986 
1987 		if (!desc_cb->dma)
1988 			continue;
1989 
1990 		/* unmap the descriptor dma address */
1991 		if (desc_cb->type & (DESC_TYPE_SKB | DESC_TYPE_FRAGLIST_SKB))
1992 			dma_unmap_single(dev, desc_cb->dma, desc_cb->length,
1993 					 DMA_TO_DEVICE);
1994 		else if (desc_cb->type &
1995 			 (DESC_TYPE_BOUNCE_HEAD | DESC_TYPE_BOUNCE_ALL))
1996 			hns3_tx_spare_rollback(ring, desc_cb->length);
1997 		else if (desc_cb->length)
1998 			dma_unmap_page(dev, desc_cb->dma, desc_cb->length,
1999 				       DMA_TO_DEVICE);
2000 
2001 		desc_cb->length = 0;
2002 		desc_cb->dma = 0;
2003 		desc_cb->type = DESC_TYPE_UNKNOWN;
2004 	}
2005 }
2006 
2007 static int hns3_fill_skb_to_desc(struct hns3_enet_ring *ring,
2008 				 struct sk_buff *skb, unsigned int type)
2009 {
2010 	struct sk_buff *frag_skb;
2011 	int i, ret, bd_num = 0;
2012 
2013 	ret = hns3_map_and_fill_desc(ring, skb, type);
2014 	if (unlikely(ret < 0))
2015 		return ret;
2016 
2017 	bd_num += ret;
2018 
2019 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2020 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2021 
2022 		ret = hns3_map_and_fill_desc(ring, frag, DESC_TYPE_PAGE);
2023 		if (unlikely(ret < 0))
2024 			return ret;
2025 
2026 		bd_num += ret;
2027 	}
2028 
2029 	skb_walk_frags(skb, frag_skb) {
2030 		ret = hns3_fill_skb_to_desc(ring, frag_skb,
2031 					    DESC_TYPE_FRAGLIST_SKB);
2032 		if (unlikely(ret < 0))
2033 			return ret;
2034 
2035 		bd_num += ret;
2036 	}
2037 
2038 	return bd_num;
2039 }
2040 
2041 static void hns3_tx_push_bd(struct hns3_enet_ring *ring, int num)
2042 {
2043 #define HNS3_BYTES_PER_64BIT		8
2044 
2045 	struct hns3_desc desc[HNS3_MAX_PUSH_BD_NUM] = {};
2046 	int offset = 0;
2047 
2048 	/* make sure everything is visible to device before
2049 	 * excuting tx push or updating doorbell
2050 	 */
2051 	dma_wmb();
2052 
2053 	do {
2054 		int idx = (ring->next_to_use - num + ring->desc_num) %
2055 			  ring->desc_num;
2056 
2057 		u64_stats_update_begin(&ring->syncp);
2058 		ring->stats.tx_push++;
2059 		u64_stats_update_end(&ring->syncp);
2060 		memcpy(&desc[offset], &ring->desc[idx],
2061 		       sizeof(struct hns3_desc));
2062 		offset++;
2063 	} while (--num);
2064 
2065 	__iowrite64_copy(ring->tqp->mem_base, desc,
2066 			 (sizeof(struct hns3_desc) * HNS3_MAX_PUSH_BD_NUM) /
2067 			 HNS3_BYTES_PER_64BIT);
2068 
2069 	io_stop_wc();
2070 }
2071 
2072 static void hns3_tx_mem_doorbell(struct hns3_enet_ring *ring)
2073 {
2074 #define HNS3_MEM_DOORBELL_OFFSET	64
2075 
2076 	__le64 bd_num = cpu_to_le64((u64)ring->pending_buf);
2077 
2078 	/* make sure everything is visible to device before
2079 	 * excuting tx push or updating doorbell
2080 	 */
2081 	dma_wmb();
2082 
2083 	__iowrite64_copy(ring->tqp->mem_base + HNS3_MEM_DOORBELL_OFFSET,
2084 			 &bd_num, 1);
2085 	u64_stats_update_begin(&ring->syncp);
2086 	ring->stats.tx_mem_doorbell += ring->pending_buf;
2087 	u64_stats_update_end(&ring->syncp);
2088 
2089 	io_stop_wc();
2090 }
2091 
2092 static void hns3_tx_doorbell(struct hns3_enet_ring *ring, int num,
2093 			     bool doorbell)
2094 {
2095 	struct net_device *netdev = ring_to_netdev(ring);
2096 	struct hns3_nic_priv *priv = netdev_priv(netdev);
2097 
2098 	/* when tx push is enabled, the packet whose number of BD below
2099 	 * HNS3_MAX_PUSH_BD_NUM can be pushed directly.
2100 	 */
2101 	if (test_bit(HNS3_NIC_STATE_TX_PUSH_ENABLE, &priv->state) && num &&
2102 	    !ring->pending_buf && num <= HNS3_MAX_PUSH_BD_NUM && doorbell) {
2103 		hns3_tx_push_bd(ring, num);
2104 		WRITE_ONCE(ring->last_to_use, ring->next_to_use);
2105 		return;
2106 	}
2107 
2108 	ring->pending_buf += num;
2109 
2110 	if (!doorbell) {
2111 		hns3_ring_stats_update(ring, tx_more);
2112 		return;
2113 	}
2114 
2115 	if (ring->tqp->mem_base)
2116 		hns3_tx_mem_doorbell(ring);
2117 	else
2118 		writel(ring->pending_buf,
2119 		       ring->tqp->io_base + HNS3_RING_TX_RING_TAIL_REG);
2120 
2121 	ring->pending_buf = 0;
2122 	WRITE_ONCE(ring->last_to_use, ring->next_to_use);
2123 }
2124 
2125 static void hns3_tsyn(struct net_device *netdev, struct sk_buff *skb,
2126 		      struct hns3_desc *desc)
2127 {
2128 	struct hnae3_handle *h = hns3_get_handle(netdev);
2129 
2130 	if (!(h->ae_algo->ops->set_tx_hwts_info &&
2131 	      h->ae_algo->ops->set_tx_hwts_info(h, skb)))
2132 		return;
2133 
2134 	desc->tx.bdtp_fe_sc_vld_ra_ri |= cpu_to_le16(BIT(HNS3_TXD_TSYN_B));
2135 }
2136 
2137 static int hns3_handle_tx_bounce(struct hns3_enet_ring *ring,
2138 				 struct sk_buff *skb)
2139 {
2140 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
2141 	unsigned int type = DESC_TYPE_BOUNCE_HEAD;
2142 	unsigned int size = skb_headlen(skb);
2143 	dma_addr_t dma;
2144 	int bd_num = 0;
2145 	u32 cb_len;
2146 	void *buf;
2147 	int ret;
2148 
2149 	if (skb->len <= ring->tx_copybreak) {
2150 		size = skb->len;
2151 		type = DESC_TYPE_BOUNCE_ALL;
2152 	}
2153 
2154 	/* hns3_can_use_tx_bounce() is called to ensure the below
2155 	 * function can always return the tx buffer.
2156 	 */
2157 	buf = hns3_tx_spare_alloc(ring, size, &dma, &cb_len);
2158 
2159 	ret = skb_copy_bits(skb, 0, buf, size);
2160 	if (unlikely(ret < 0)) {
2161 		hns3_tx_spare_rollback(ring, cb_len);
2162 		hns3_ring_stats_update(ring, copy_bits_err);
2163 		return ret;
2164 	}
2165 
2166 	desc_cb->priv = skb;
2167 	desc_cb->length = cb_len;
2168 	desc_cb->dma = dma;
2169 	desc_cb->type = type;
2170 
2171 	bd_num += hns3_fill_desc(ring, dma, size);
2172 
2173 	if (type == DESC_TYPE_BOUNCE_HEAD) {
2174 		ret = hns3_fill_skb_to_desc(ring, skb,
2175 					    DESC_TYPE_BOUNCE_HEAD);
2176 		if (unlikely(ret < 0))
2177 			return ret;
2178 
2179 		bd_num += ret;
2180 	}
2181 
2182 	dma_sync_single_for_device(ring_to_dev(ring), dma, size,
2183 				   DMA_TO_DEVICE);
2184 
2185 	hns3_ring_stats_update(ring, tx_bounce);
2186 
2187 	return bd_num;
2188 }
2189 
2190 static int hns3_handle_tx_sgl(struct hns3_enet_ring *ring,
2191 			      struct sk_buff *skb)
2192 {
2193 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
2194 	u32 nfrag = skb_shinfo(skb)->nr_frags + 1;
2195 	struct sg_table *sgt;
2196 	int i, bd_num = 0;
2197 	dma_addr_t dma;
2198 	u32 cb_len;
2199 	int nents;
2200 
2201 	if (skb_has_frag_list(skb))
2202 		nfrag = HNS3_MAX_TSO_BD_NUM;
2203 
2204 	/* hns3_can_use_tx_sgl() is called to ensure the below
2205 	 * function can always return the tx buffer.
2206 	 */
2207 	sgt = hns3_tx_spare_alloc(ring, HNS3_SGL_SIZE(nfrag),
2208 				  &dma, &cb_len);
2209 
2210 	/* scatterlist follows by the sg table */
2211 	sgt->sgl = (struct scatterlist *)(sgt + 1);
2212 	sg_init_table(sgt->sgl, nfrag);
2213 	nents = skb_to_sgvec(skb, sgt->sgl, 0, skb->len);
2214 	if (unlikely(nents < 0)) {
2215 		hns3_tx_spare_rollback(ring, cb_len);
2216 		hns3_ring_stats_update(ring, skb2sgl_err);
2217 		return -ENOMEM;
2218 	}
2219 
2220 	sgt->orig_nents = nents;
2221 	sgt->nents = dma_map_sg(ring_to_dev(ring), sgt->sgl, sgt->orig_nents,
2222 				DMA_TO_DEVICE);
2223 	if (unlikely(!sgt->nents)) {
2224 		hns3_tx_spare_rollback(ring, cb_len);
2225 		hns3_ring_stats_update(ring, map_sg_err);
2226 		return -ENOMEM;
2227 	}
2228 
2229 	desc_cb->priv = skb;
2230 	desc_cb->length = cb_len;
2231 	desc_cb->dma = dma;
2232 	desc_cb->type = DESC_TYPE_SGL_SKB;
2233 
2234 	for (i = 0; i < sgt->nents; i++)
2235 		bd_num += hns3_fill_desc(ring, sg_dma_address(sgt->sgl + i),
2236 					 sg_dma_len(sgt->sgl + i));
2237 	hns3_ring_stats_update(ring, tx_sgl);
2238 
2239 	return bd_num;
2240 }
2241 
2242 static int hns3_handle_desc_filling(struct hns3_enet_ring *ring,
2243 				    struct sk_buff *skb)
2244 {
2245 	u32 space;
2246 
2247 	if (!ring->tx_spare)
2248 		goto out;
2249 
2250 	space = hns3_tx_spare_space(ring);
2251 
2252 	if (hns3_can_use_tx_sgl(ring, skb, space))
2253 		return hns3_handle_tx_sgl(ring, skb);
2254 
2255 	if (hns3_can_use_tx_bounce(ring, skb, space))
2256 		return hns3_handle_tx_bounce(ring, skb);
2257 
2258 out:
2259 	return hns3_fill_skb_to_desc(ring, skb, DESC_TYPE_SKB);
2260 }
2261 
2262 static int hns3_handle_skb_desc(struct hns3_enet_ring *ring,
2263 				struct sk_buff *skb,
2264 				struct hns3_desc_cb *desc_cb,
2265 				int next_to_use_head)
2266 {
2267 	int ret;
2268 
2269 	ret = hns3_fill_skb_desc(ring, skb, &ring->desc[ring->next_to_use],
2270 				 desc_cb);
2271 	if (unlikely(ret < 0))
2272 		goto fill_err;
2273 
2274 	/* 'ret < 0' means filling error, 'ret == 0' means skb->len is
2275 	 * zero, which is unlikely, and 'ret > 0' means how many tx desc
2276 	 * need to be notified to the hw.
2277 	 */
2278 	ret = hns3_handle_desc_filling(ring, skb);
2279 	if (likely(ret > 0))
2280 		return ret;
2281 
2282 fill_err:
2283 	hns3_clear_desc(ring, next_to_use_head);
2284 	return ret;
2285 }
2286 
2287 netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
2288 {
2289 	struct hns3_nic_priv *priv = netdev_priv(netdev);
2290 	struct hns3_enet_ring *ring = &priv->ring[skb->queue_mapping];
2291 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
2292 	struct netdev_queue *dev_queue;
2293 	int pre_ntu, ret;
2294 	bool doorbell;
2295 
2296 	/* Hardware can only handle short frames above 32 bytes */
2297 	if (skb_put_padto(skb, HNS3_MIN_TX_LEN)) {
2298 		hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
2299 
2300 		hns3_ring_stats_update(ring, sw_err_cnt);
2301 
2302 		return NETDEV_TX_OK;
2303 	}
2304 
2305 	/* Prefetch the data used later */
2306 	prefetch(skb->data);
2307 
2308 	ret = hns3_nic_maybe_stop_tx(ring, netdev, skb);
2309 	if (unlikely(ret <= 0)) {
2310 		if (ret == -EBUSY) {
2311 			hns3_tx_doorbell(ring, 0, true);
2312 			return NETDEV_TX_BUSY;
2313 		}
2314 
2315 		hns3_rl_err(netdev, "xmit error: %d!\n", ret);
2316 		goto out_err_tx_ok;
2317 	}
2318 
2319 	ret = hns3_handle_skb_desc(ring, skb, desc_cb, ring->next_to_use);
2320 	if (unlikely(ret <= 0))
2321 		goto out_err_tx_ok;
2322 
2323 	pre_ntu = ring->next_to_use ? (ring->next_to_use - 1) :
2324 					(ring->desc_num - 1);
2325 
2326 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP))
2327 		hns3_tsyn(netdev, skb, &ring->desc[pre_ntu]);
2328 
2329 	ring->desc[pre_ntu].tx.bdtp_fe_sc_vld_ra_ri |=
2330 				cpu_to_le16(BIT(HNS3_TXD_FE_B));
2331 	trace_hns3_tx_desc(ring, pre_ntu);
2332 
2333 	skb_tx_timestamp(skb);
2334 
2335 	/* Complete translate all packets */
2336 	dev_queue = netdev_get_tx_queue(netdev, ring->queue_index);
2337 	doorbell = __netdev_tx_sent_queue(dev_queue, desc_cb->send_bytes,
2338 					  netdev_xmit_more());
2339 	hns3_tx_doorbell(ring, ret, doorbell);
2340 
2341 	return NETDEV_TX_OK;
2342 
2343 out_err_tx_ok:
2344 	dev_kfree_skb_any(skb);
2345 	hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
2346 	return NETDEV_TX_OK;
2347 }
2348 
2349 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
2350 {
2351 	char format_mac_addr_perm[HNAE3_FORMAT_MAC_ADDR_LEN];
2352 	char format_mac_addr_sa[HNAE3_FORMAT_MAC_ADDR_LEN];
2353 	struct hnae3_handle *h = hns3_get_handle(netdev);
2354 	struct sockaddr *mac_addr = p;
2355 	int ret;
2356 
2357 	if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
2358 		return -EADDRNOTAVAIL;
2359 
2360 	if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
2361 		hnae3_format_mac_addr(format_mac_addr_sa, mac_addr->sa_data);
2362 		netdev_info(netdev, "already using mac address %s\n",
2363 			    format_mac_addr_sa);
2364 		return 0;
2365 	}
2366 
2367 	/* For VF device, if there is a perm_addr, then the user will not
2368 	 * be allowed to change the address.
2369 	 */
2370 	if (!hns3_is_phys_func(h->pdev) &&
2371 	    !is_zero_ether_addr(netdev->perm_addr)) {
2372 		hnae3_format_mac_addr(format_mac_addr_perm, netdev->perm_addr);
2373 		hnae3_format_mac_addr(format_mac_addr_sa, mac_addr->sa_data);
2374 		netdev_err(netdev, "has permanent MAC %s, user MAC %s not allow\n",
2375 			   format_mac_addr_perm, format_mac_addr_sa);
2376 		return -EPERM;
2377 	}
2378 
2379 	ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
2380 	if (ret) {
2381 		netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
2382 		return ret;
2383 	}
2384 
2385 	eth_hw_addr_set(netdev, mac_addr->sa_data);
2386 
2387 	return 0;
2388 }
2389 
2390 static int hns3_nic_do_ioctl(struct net_device *netdev,
2391 			     struct ifreq *ifr, int cmd)
2392 {
2393 	struct hnae3_handle *h = hns3_get_handle(netdev);
2394 
2395 	if (!netif_running(netdev))
2396 		return -EINVAL;
2397 
2398 	if (!h->ae_algo->ops->do_ioctl)
2399 		return -EOPNOTSUPP;
2400 
2401 	return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
2402 }
2403 
2404 static int hns3_nic_set_features(struct net_device *netdev,
2405 				 netdev_features_t features)
2406 {
2407 	netdev_features_t changed = netdev->features ^ features;
2408 	struct hns3_nic_priv *priv = netdev_priv(netdev);
2409 	struct hnae3_handle *h = priv->ae_handle;
2410 	bool enable;
2411 	int ret;
2412 
2413 	if (changed & (NETIF_F_GRO_HW) && h->ae_algo->ops->set_gro_en) {
2414 		enable = !!(features & NETIF_F_GRO_HW);
2415 		ret = h->ae_algo->ops->set_gro_en(h, enable);
2416 		if (ret)
2417 			return ret;
2418 	}
2419 
2420 	if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
2421 	    h->ae_algo->ops->enable_hw_strip_rxvtag) {
2422 		enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
2423 		ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, enable);
2424 		if (ret)
2425 			return ret;
2426 	}
2427 
2428 	if ((changed & NETIF_F_NTUPLE) && h->ae_algo->ops->enable_fd) {
2429 		enable = !!(features & NETIF_F_NTUPLE);
2430 		h->ae_algo->ops->enable_fd(h, enable);
2431 	}
2432 
2433 	if ((netdev->features & NETIF_F_HW_TC) > (features & NETIF_F_HW_TC) &&
2434 	    h->ae_algo->ops->cls_flower_active(h)) {
2435 		netdev_err(netdev,
2436 			   "there are offloaded TC filters active, cannot disable HW TC offload");
2437 		return -EINVAL;
2438 	}
2439 
2440 	if ((changed & NETIF_F_HW_VLAN_CTAG_FILTER) &&
2441 	    h->ae_algo->ops->enable_vlan_filter) {
2442 		enable = !!(features & NETIF_F_HW_VLAN_CTAG_FILTER);
2443 		ret = h->ae_algo->ops->enable_vlan_filter(h, enable);
2444 		if (ret)
2445 			return ret;
2446 	}
2447 
2448 	netdev->features = features;
2449 	return 0;
2450 }
2451 
2452 static netdev_features_t hns3_features_check(struct sk_buff *skb,
2453 					     struct net_device *dev,
2454 					     netdev_features_t features)
2455 {
2456 #define HNS3_MAX_HDR_LEN	480U
2457 #define HNS3_MAX_L4_HDR_LEN	60U
2458 
2459 	size_t len;
2460 
2461 	if (skb->ip_summed != CHECKSUM_PARTIAL)
2462 		return features;
2463 
2464 	if (skb->encapsulation)
2465 		len = skb_inner_transport_header(skb) - skb->data;
2466 	else
2467 		len = skb_transport_header(skb) - skb->data;
2468 
2469 	/* Assume L4 is 60 byte as TCP is the only protocol with a
2470 	 * a flexible value, and it's max len is 60 bytes.
2471 	 */
2472 	len += HNS3_MAX_L4_HDR_LEN;
2473 
2474 	/* Hardware only supports checksum on the skb with a max header
2475 	 * len of 480 bytes.
2476 	 */
2477 	if (len > HNS3_MAX_HDR_LEN)
2478 		features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2479 
2480 	return features;
2481 }
2482 
2483 static void hns3_fetch_stats(struct rtnl_link_stats64 *stats,
2484 			     struct hns3_enet_ring *ring, bool is_tx)
2485 {
2486 	unsigned int start;
2487 
2488 	do {
2489 		start = u64_stats_fetch_begin_irq(&ring->syncp);
2490 		if (is_tx) {
2491 			stats->tx_bytes += ring->stats.tx_bytes;
2492 			stats->tx_packets += ring->stats.tx_pkts;
2493 			stats->tx_dropped += ring->stats.sw_err_cnt;
2494 			stats->tx_dropped += ring->stats.tx_vlan_err;
2495 			stats->tx_dropped += ring->stats.tx_l4_proto_err;
2496 			stats->tx_dropped += ring->stats.tx_l2l3l4_err;
2497 			stats->tx_dropped += ring->stats.tx_tso_err;
2498 			stats->tx_dropped += ring->stats.over_max_recursion;
2499 			stats->tx_dropped += ring->stats.hw_limitation;
2500 			stats->tx_dropped += ring->stats.copy_bits_err;
2501 			stats->tx_dropped += ring->stats.skb2sgl_err;
2502 			stats->tx_dropped += ring->stats.map_sg_err;
2503 			stats->tx_errors += ring->stats.sw_err_cnt;
2504 			stats->tx_errors += ring->stats.tx_vlan_err;
2505 			stats->tx_errors += ring->stats.tx_l4_proto_err;
2506 			stats->tx_errors += ring->stats.tx_l2l3l4_err;
2507 			stats->tx_errors += ring->stats.tx_tso_err;
2508 			stats->tx_errors += ring->stats.over_max_recursion;
2509 			stats->tx_errors += ring->stats.hw_limitation;
2510 			stats->tx_errors += ring->stats.copy_bits_err;
2511 			stats->tx_errors += ring->stats.skb2sgl_err;
2512 			stats->tx_errors += ring->stats.map_sg_err;
2513 		} else {
2514 			stats->rx_bytes += ring->stats.rx_bytes;
2515 			stats->rx_packets += ring->stats.rx_pkts;
2516 			stats->rx_dropped += ring->stats.l2_err;
2517 			stats->rx_errors += ring->stats.l2_err;
2518 			stats->rx_errors += ring->stats.l3l4_csum_err;
2519 			stats->rx_crc_errors += ring->stats.l2_err;
2520 			stats->multicast += ring->stats.rx_multicast;
2521 			stats->rx_length_errors += ring->stats.err_pkt_len;
2522 		}
2523 	} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
2524 }
2525 
2526 static void hns3_nic_get_stats64(struct net_device *netdev,
2527 				 struct rtnl_link_stats64 *stats)
2528 {
2529 	struct hns3_nic_priv *priv = netdev_priv(netdev);
2530 	int queue_num = priv->ae_handle->kinfo.num_tqps;
2531 	struct hnae3_handle *handle = priv->ae_handle;
2532 	struct rtnl_link_stats64 ring_total_stats;
2533 	struct hns3_enet_ring *ring;
2534 	unsigned int idx;
2535 
2536 	if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
2537 		return;
2538 
2539 	handle->ae_algo->ops->update_stats(handle, &netdev->stats);
2540 
2541 	memset(&ring_total_stats, 0, sizeof(ring_total_stats));
2542 	for (idx = 0; idx < queue_num; idx++) {
2543 		/* fetch the tx stats */
2544 		ring = &priv->ring[idx];
2545 		hns3_fetch_stats(&ring_total_stats, ring, true);
2546 
2547 		/* fetch the rx stats */
2548 		ring = &priv->ring[idx + queue_num];
2549 		hns3_fetch_stats(&ring_total_stats, ring, false);
2550 	}
2551 
2552 	stats->tx_bytes = ring_total_stats.tx_bytes;
2553 	stats->tx_packets = ring_total_stats.tx_packets;
2554 	stats->rx_bytes = ring_total_stats.rx_bytes;
2555 	stats->rx_packets = ring_total_stats.rx_packets;
2556 
2557 	stats->rx_errors = ring_total_stats.rx_errors;
2558 	stats->multicast = ring_total_stats.multicast;
2559 	stats->rx_length_errors = ring_total_stats.rx_length_errors;
2560 	stats->rx_crc_errors = ring_total_stats.rx_crc_errors;
2561 	stats->rx_missed_errors = netdev->stats.rx_missed_errors;
2562 
2563 	stats->tx_errors = ring_total_stats.tx_errors;
2564 	stats->rx_dropped = ring_total_stats.rx_dropped;
2565 	stats->tx_dropped = ring_total_stats.tx_dropped;
2566 	stats->collisions = netdev->stats.collisions;
2567 	stats->rx_over_errors = netdev->stats.rx_over_errors;
2568 	stats->rx_frame_errors = netdev->stats.rx_frame_errors;
2569 	stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
2570 	stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
2571 	stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
2572 	stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
2573 	stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
2574 	stats->tx_window_errors = netdev->stats.tx_window_errors;
2575 	stats->rx_compressed = netdev->stats.rx_compressed;
2576 	stats->tx_compressed = netdev->stats.tx_compressed;
2577 }
2578 
2579 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
2580 {
2581 	struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
2582 	struct hnae3_knic_private_info *kinfo;
2583 	u8 tc = mqprio_qopt->qopt.num_tc;
2584 	u16 mode = mqprio_qopt->mode;
2585 	u8 hw = mqprio_qopt->qopt.hw;
2586 	struct hnae3_handle *h;
2587 
2588 	if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
2589 	       mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
2590 		return -EOPNOTSUPP;
2591 
2592 	if (tc > HNAE3_MAX_TC)
2593 		return -EINVAL;
2594 
2595 	if (!netdev)
2596 		return -EINVAL;
2597 
2598 	h = hns3_get_handle(netdev);
2599 	kinfo = &h->kinfo;
2600 
2601 	netif_dbg(h, drv, netdev, "setup tc: num_tc=%u\n", tc);
2602 
2603 	return (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
2604 		kinfo->dcb_ops->setup_tc(h, mqprio_qopt) : -EOPNOTSUPP;
2605 }
2606 
2607 static int hns3_setup_tc_cls_flower(struct hns3_nic_priv *priv,
2608 				    struct flow_cls_offload *flow)
2609 {
2610 	int tc = tc_classid_to_hwtc(priv->netdev, flow->classid);
2611 	struct hnae3_handle *h = hns3_get_handle(priv->netdev);
2612 
2613 	switch (flow->command) {
2614 	case FLOW_CLS_REPLACE:
2615 		if (h->ae_algo->ops->add_cls_flower)
2616 			return h->ae_algo->ops->add_cls_flower(h, flow, tc);
2617 		break;
2618 	case FLOW_CLS_DESTROY:
2619 		if (h->ae_algo->ops->del_cls_flower)
2620 			return h->ae_algo->ops->del_cls_flower(h, flow);
2621 		break;
2622 	default:
2623 		break;
2624 	}
2625 
2626 	return -EOPNOTSUPP;
2627 }
2628 
2629 static int hns3_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
2630 				  void *cb_priv)
2631 {
2632 	struct hns3_nic_priv *priv = cb_priv;
2633 
2634 	if (!tc_cls_can_offload_and_chain0(priv->netdev, type_data))
2635 		return -EOPNOTSUPP;
2636 
2637 	switch (type) {
2638 	case TC_SETUP_CLSFLOWER:
2639 		return hns3_setup_tc_cls_flower(priv, type_data);
2640 	default:
2641 		return -EOPNOTSUPP;
2642 	}
2643 }
2644 
2645 static LIST_HEAD(hns3_block_cb_list);
2646 
2647 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
2648 			     void *type_data)
2649 {
2650 	struct hns3_nic_priv *priv = netdev_priv(dev);
2651 	int ret;
2652 
2653 	switch (type) {
2654 	case TC_SETUP_QDISC_MQPRIO:
2655 		ret = hns3_setup_tc(dev, type_data);
2656 		break;
2657 	case TC_SETUP_BLOCK:
2658 		ret = flow_block_cb_setup_simple(type_data,
2659 						 &hns3_block_cb_list,
2660 						 hns3_setup_tc_block_cb,
2661 						 priv, priv, true);
2662 		break;
2663 	default:
2664 		return -EOPNOTSUPP;
2665 	}
2666 
2667 	return ret;
2668 }
2669 
2670 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
2671 				__be16 proto, u16 vid)
2672 {
2673 	struct hnae3_handle *h = hns3_get_handle(netdev);
2674 	int ret = -EIO;
2675 
2676 	if (h->ae_algo->ops->set_vlan_filter)
2677 		ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
2678 
2679 	return ret;
2680 }
2681 
2682 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
2683 				 __be16 proto, u16 vid)
2684 {
2685 	struct hnae3_handle *h = hns3_get_handle(netdev);
2686 	int ret = -EIO;
2687 
2688 	if (h->ae_algo->ops->set_vlan_filter)
2689 		ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
2690 
2691 	return ret;
2692 }
2693 
2694 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
2695 				u8 qos, __be16 vlan_proto)
2696 {
2697 	struct hnae3_handle *h = hns3_get_handle(netdev);
2698 	int ret = -EIO;
2699 
2700 	netif_dbg(h, drv, netdev,
2701 		  "set vf vlan: vf=%d, vlan=%u, qos=%u, vlan_proto=0x%x\n",
2702 		  vf, vlan, qos, ntohs(vlan_proto));
2703 
2704 	if (h->ae_algo->ops->set_vf_vlan_filter)
2705 		ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
2706 							  qos, vlan_proto);
2707 
2708 	return ret;
2709 }
2710 
2711 static int hns3_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
2712 {
2713 	struct hnae3_handle *handle = hns3_get_handle(netdev);
2714 
2715 	if (hns3_nic_resetting(netdev))
2716 		return -EBUSY;
2717 
2718 	if (!handle->ae_algo->ops->set_vf_spoofchk)
2719 		return -EOPNOTSUPP;
2720 
2721 	return handle->ae_algo->ops->set_vf_spoofchk(handle, vf, enable);
2722 }
2723 
2724 static int hns3_set_vf_trust(struct net_device *netdev, int vf, bool enable)
2725 {
2726 	struct hnae3_handle *handle = hns3_get_handle(netdev);
2727 
2728 	if (!handle->ae_algo->ops->set_vf_trust)
2729 		return -EOPNOTSUPP;
2730 
2731 	return handle->ae_algo->ops->set_vf_trust(handle, vf, enable);
2732 }
2733 
2734 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
2735 {
2736 	struct hnae3_handle *h = hns3_get_handle(netdev);
2737 	int ret;
2738 
2739 	if (hns3_nic_resetting(netdev))
2740 		return -EBUSY;
2741 
2742 	if (!h->ae_algo->ops->set_mtu)
2743 		return -EOPNOTSUPP;
2744 
2745 	netif_dbg(h, drv, netdev,
2746 		  "change mtu from %u to %d\n", netdev->mtu, new_mtu);
2747 
2748 	ret = h->ae_algo->ops->set_mtu(h, new_mtu);
2749 	if (ret)
2750 		netdev_err(netdev, "failed to change MTU in hardware %d\n",
2751 			   ret);
2752 	else
2753 		netdev->mtu = new_mtu;
2754 
2755 	return ret;
2756 }
2757 
2758 static int hns3_get_timeout_queue(struct net_device *ndev)
2759 {
2760 	int i;
2761 
2762 	/* Find the stopped queue the same way the stack does */
2763 	for (i = 0; i < ndev->num_tx_queues; i++) {
2764 		struct netdev_queue *q;
2765 		unsigned long trans_start;
2766 
2767 		q = netdev_get_tx_queue(ndev, i);
2768 		trans_start = READ_ONCE(q->trans_start);
2769 		if (netif_xmit_stopped(q) &&
2770 		    time_after(jiffies,
2771 			       (trans_start + ndev->watchdog_timeo))) {
2772 #ifdef CONFIG_BQL
2773 			struct dql *dql = &q->dql;
2774 
2775 			netdev_info(ndev, "DQL info last_cnt: %u, queued: %u, adj_limit: %u, completed: %u\n",
2776 				    dql->last_obj_cnt, dql->num_queued,
2777 				    dql->adj_limit, dql->num_completed);
2778 #endif
2779 			netdev_info(ndev, "queue state: 0x%lx, delta msecs: %u\n",
2780 				    q->state,
2781 				    jiffies_to_msecs(jiffies - trans_start));
2782 			break;
2783 		}
2784 	}
2785 
2786 	return i;
2787 }
2788 
2789 static void hns3_dump_queue_stats(struct net_device *ndev,
2790 				  struct hns3_enet_ring *tx_ring,
2791 				  int timeout_queue)
2792 {
2793 	struct napi_struct *napi = &tx_ring->tqp_vector->napi;
2794 	struct hns3_nic_priv *priv = netdev_priv(ndev);
2795 
2796 	netdev_info(ndev,
2797 		    "tx_timeout count: %llu, queue id: %d, SW_NTU: 0x%x, SW_NTC: 0x%x, napi state: %lu\n",
2798 		    priv->tx_timeout_count, timeout_queue, tx_ring->next_to_use,
2799 		    tx_ring->next_to_clean, napi->state);
2800 
2801 	netdev_info(ndev,
2802 		    "tx_pkts: %llu, tx_bytes: %llu, sw_err_cnt: %llu, tx_pending: %d\n",
2803 		    tx_ring->stats.tx_pkts, tx_ring->stats.tx_bytes,
2804 		    tx_ring->stats.sw_err_cnt, tx_ring->pending_buf);
2805 
2806 	netdev_info(ndev,
2807 		    "seg_pkt_cnt: %llu, tx_more: %llu, restart_queue: %llu, tx_busy: %llu\n",
2808 		    tx_ring->stats.seg_pkt_cnt, tx_ring->stats.tx_more,
2809 		    tx_ring->stats.restart_queue, tx_ring->stats.tx_busy);
2810 
2811 	netdev_info(ndev, "tx_push: %llu, tx_mem_doorbell: %llu\n",
2812 		    tx_ring->stats.tx_push, tx_ring->stats.tx_mem_doorbell);
2813 }
2814 
2815 static void hns3_dump_queue_reg(struct net_device *ndev,
2816 				struct hns3_enet_ring *tx_ring)
2817 {
2818 	netdev_info(ndev,
2819 		    "BD_NUM: 0x%x HW_HEAD: 0x%x, HW_TAIL: 0x%x, BD_ERR: 0x%x, INT: 0x%x\n",
2820 		    hns3_tqp_read_reg(tx_ring, HNS3_RING_TX_RING_BD_NUM_REG),
2821 		    hns3_tqp_read_reg(tx_ring, HNS3_RING_TX_RING_HEAD_REG),
2822 		    hns3_tqp_read_reg(tx_ring, HNS3_RING_TX_RING_TAIL_REG),
2823 		    hns3_tqp_read_reg(tx_ring, HNS3_RING_TX_RING_BD_ERR_REG),
2824 		    readl(tx_ring->tqp_vector->mask_addr));
2825 	netdev_info(ndev,
2826 		    "RING_EN: 0x%x, TC: 0x%x, FBD_NUM: 0x%x FBD_OFT: 0x%x, EBD_NUM: 0x%x, EBD_OFT: 0x%x\n",
2827 		    hns3_tqp_read_reg(tx_ring, HNS3_RING_EN_REG),
2828 		    hns3_tqp_read_reg(tx_ring, HNS3_RING_TX_RING_TC_REG),
2829 		    hns3_tqp_read_reg(tx_ring, HNS3_RING_TX_RING_FBDNUM_REG),
2830 		    hns3_tqp_read_reg(tx_ring, HNS3_RING_TX_RING_OFFSET_REG),
2831 		    hns3_tqp_read_reg(tx_ring, HNS3_RING_TX_RING_EBDNUM_REG),
2832 		    hns3_tqp_read_reg(tx_ring,
2833 				      HNS3_RING_TX_RING_EBD_OFFSET_REG));
2834 }
2835 
2836 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
2837 {
2838 	struct hns3_nic_priv *priv = netdev_priv(ndev);
2839 	struct hnae3_handle *h = hns3_get_handle(ndev);
2840 	struct hns3_enet_ring *tx_ring;
2841 	int timeout_queue;
2842 
2843 	timeout_queue = hns3_get_timeout_queue(ndev);
2844 	if (timeout_queue >= ndev->num_tx_queues) {
2845 		netdev_info(ndev,
2846 			    "no netdev TX timeout queue found, timeout count: %llu\n",
2847 			    priv->tx_timeout_count);
2848 		return false;
2849 	}
2850 
2851 	priv->tx_timeout_count++;
2852 
2853 	tx_ring = &priv->ring[timeout_queue];
2854 	hns3_dump_queue_stats(ndev, tx_ring, timeout_queue);
2855 
2856 	/* When mac received many pause frames continuous, it's unable to send
2857 	 * packets, which may cause tx timeout
2858 	 */
2859 	if (h->ae_algo->ops->get_mac_stats) {
2860 		struct hns3_mac_stats mac_stats;
2861 
2862 		h->ae_algo->ops->get_mac_stats(h, &mac_stats);
2863 		netdev_info(ndev, "tx_pause_cnt: %llu, rx_pause_cnt: %llu\n",
2864 			    mac_stats.tx_pause_cnt, mac_stats.rx_pause_cnt);
2865 	}
2866 
2867 	hns3_dump_queue_reg(ndev, tx_ring);
2868 
2869 	return true;
2870 }
2871 
2872 static void hns3_nic_net_timeout(struct net_device *ndev, unsigned int txqueue)
2873 {
2874 	struct hns3_nic_priv *priv = netdev_priv(ndev);
2875 	struct hnae3_handle *h = priv->ae_handle;
2876 
2877 	if (!hns3_get_tx_timeo_queue_info(ndev))
2878 		return;
2879 
2880 	/* request the reset, and let the hclge to determine
2881 	 * which reset level should be done
2882 	 */
2883 	if (h->ae_algo->ops->reset_event)
2884 		h->ae_algo->ops->reset_event(h->pdev, h);
2885 }
2886 
2887 #ifdef CONFIG_RFS_ACCEL
2888 static int hns3_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
2889 			      u16 rxq_index, u32 flow_id)
2890 {
2891 	struct hnae3_handle *h = hns3_get_handle(dev);
2892 	struct flow_keys fkeys;
2893 
2894 	if (!h->ae_algo->ops->add_arfs_entry)
2895 		return -EOPNOTSUPP;
2896 
2897 	if (skb->encapsulation)
2898 		return -EPROTONOSUPPORT;
2899 
2900 	if (!skb_flow_dissect_flow_keys(skb, &fkeys, 0))
2901 		return -EPROTONOSUPPORT;
2902 
2903 	if ((fkeys.basic.n_proto != htons(ETH_P_IP) &&
2904 	     fkeys.basic.n_proto != htons(ETH_P_IPV6)) ||
2905 	    (fkeys.basic.ip_proto != IPPROTO_TCP &&
2906 	     fkeys.basic.ip_proto != IPPROTO_UDP))
2907 		return -EPROTONOSUPPORT;
2908 
2909 	return h->ae_algo->ops->add_arfs_entry(h, rxq_index, flow_id, &fkeys);
2910 }
2911 #endif
2912 
2913 static int hns3_nic_get_vf_config(struct net_device *ndev, int vf,
2914 				  struct ifla_vf_info *ivf)
2915 {
2916 	struct hnae3_handle *h = hns3_get_handle(ndev);
2917 
2918 	if (!h->ae_algo->ops->get_vf_config)
2919 		return -EOPNOTSUPP;
2920 
2921 	return h->ae_algo->ops->get_vf_config(h, vf, ivf);
2922 }
2923 
2924 static int hns3_nic_set_vf_link_state(struct net_device *ndev, int vf,
2925 				      int link_state)
2926 {
2927 	struct hnae3_handle *h = hns3_get_handle(ndev);
2928 
2929 	if (!h->ae_algo->ops->set_vf_link_state)
2930 		return -EOPNOTSUPP;
2931 
2932 	return h->ae_algo->ops->set_vf_link_state(h, vf, link_state);
2933 }
2934 
2935 static int hns3_nic_set_vf_rate(struct net_device *ndev, int vf,
2936 				int min_tx_rate, int max_tx_rate)
2937 {
2938 	struct hnae3_handle *h = hns3_get_handle(ndev);
2939 
2940 	if (!h->ae_algo->ops->set_vf_rate)
2941 		return -EOPNOTSUPP;
2942 
2943 	return h->ae_algo->ops->set_vf_rate(h, vf, min_tx_rate, max_tx_rate,
2944 					    false);
2945 }
2946 
2947 static int hns3_nic_set_vf_mac(struct net_device *netdev, int vf_id, u8 *mac)
2948 {
2949 	struct hnae3_handle *h = hns3_get_handle(netdev);
2950 	char format_mac_addr[HNAE3_FORMAT_MAC_ADDR_LEN];
2951 
2952 	if (!h->ae_algo->ops->set_vf_mac)
2953 		return -EOPNOTSUPP;
2954 
2955 	if (is_multicast_ether_addr(mac)) {
2956 		hnae3_format_mac_addr(format_mac_addr, mac);
2957 		netdev_err(netdev,
2958 			   "Invalid MAC:%s specified. Could not set MAC\n",
2959 			   format_mac_addr);
2960 		return -EINVAL;
2961 	}
2962 
2963 	return h->ae_algo->ops->set_vf_mac(h, vf_id, mac);
2964 }
2965 
2966 static const struct net_device_ops hns3_nic_netdev_ops = {
2967 	.ndo_open		= hns3_nic_net_open,
2968 	.ndo_stop		= hns3_nic_net_stop,
2969 	.ndo_start_xmit		= hns3_nic_net_xmit,
2970 	.ndo_tx_timeout		= hns3_nic_net_timeout,
2971 	.ndo_set_mac_address	= hns3_nic_net_set_mac_address,
2972 	.ndo_eth_ioctl		= hns3_nic_do_ioctl,
2973 	.ndo_change_mtu		= hns3_nic_change_mtu,
2974 	.ndo_set_features	= hns3_nic_set_features,
2975 	.ndo_features_check	= hns3_features_check,
2976 	.ndo_get_stats64	= hns3_nic_get_stats64,
2977 	.ndo_setup_tc		= hns3_nic_setup_tc,
2978 	.ndo_set_rx_mode	= hns3_nic_set_rx_mode,
2979 	.ndo_vlan_rx_add_vid	= hns3_vlan_rx_add_vid,
2980 	.ndo_vlan_rx_kill_vid	= hns3_vlan_rx_kill_vid,
2981 	.ndo_set_vf_vlan	= hns3_ndo_set_vf_vlan,
2982 	.ndo_set_vf_spoofchk	= hns3_set_vf_spoofchk,
2983 	.ndo_set_vf_trust	= hns3_set_vf_trust,
2984 #ifdef CONFIG_RFS_ACCEL
2985 	.ndo_rx_flow_steer	= hns3_rx_flow_steer,
2986 #endif
2987 	.ndo_get_vf_config	= hns3_nic_get_vf_config,
2988 	.ndo_set_vf_link_state	= hns3_nic_set_vf_link_state,
2989 	.ndo_set_vf_rate	= hns3_nic_set_vf_rate,
2990 	.ndo_set_vf_mac		= hns3_nic_set_vf_mac,
2991 };
2992 
2993 bool hns3_is_phys_func(struct pci_dev *pdev)
2994 {
2995 	u32 dev_id = pdev->device;
2996 
2997 	switch (dev_id) {
2998 	case HNAE3_DEV_ID_GE:
2999 	case HNAE3_DEV_ID_25GE:
3000 	case HNAE3_DEV_ID_25GE_RDMA:
3001 	case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
3002 	case HNAE3_DEV_ID_50GE_RDMA:
3003 	case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
3004 	case HNAE3_DEV_ID_100G_RDMA_MACSEC:
3005 	case HNAE3_DEV_ID_200G_RDMA:
3006 		return true;
3007 	case HNAE3_DEV_ID_VF:
3008 	case HNAE3_DEV_ID_RDMA_DCB_PFC_VF:
3009 		return false;
3010 	default:
3011 		dev_warn(&pdev->dev, "un-recognized pci device-id %u",
3012 			 dev_id);
3013 	}
3014 
3015 	return false;
3016 }
3017 
3018 static void hns3_disable_sriov(struct pci_dev *pdev)
3019 {
3020 	/* If our VFs are assigned we cannot shut down SR-IOV
3021 	 * without causing issues, so just leave the hardware
3022 	 * available but disabled
3023 	 */
3024 	if (pci_vfs_assigned(pdev)) {
3025 		dev_warn(&pdev->dev,
3026 			 "disabling driver while VFs are assigned\n");
3027 		return;
3028 	}
3029 
3030 	pci_disable_sriov(pdev);
3031 }
3032 
3033 /* hns3_probe - Device initialization routine
3034  * @pdev: PCI device information struct
3035  * @ent: entry in hns3_pci_tbl
3036  *
3037  * hns3_probe initializes a PF identified by a pci_dev structure.
3038  * The OS initialization, configuring of the PF private structure,
3039  * and a hardware reset occur.
3040  *
3041  * Returns 0 on success, negative on failure
3042  */
3043 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3044 {
3045 	struct hnae3_ae_dev *ae_dev;
3046 	int ret;
3047 
3048 	ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev), GFP_KERNEL);
3049 	if (!ae_dev)
3050 		return -ENOMEM;
3051 
3052 	ae_dev->pdev = pdev;
3053 	ae_dev->flag = ent->driver_data;
3054 	pci_set_drvdata(pdev, ae_dev);
3055 
3056 	ret = hnae3_register_ae_dev(ae_dev);
3057 	if (ret)
3058 		pci_set_drvdata(pdev, NULL);
3059 
3060 	return ret;
3061 }
3062 
3063 /**
3064  * hns3_clean_vf_config
3065  * @pdev: pointer to a pci_dev structure
3066  * @num_vfs: number of VFs allocated
3067  *
3068  * Clean residual vf config after disable sriov
3069  **/
3070 static void hns3_clean_vf_config(struct pci_dev *pdev, int num_vfs)
3071 {
3072 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3073 
3074 	if (ae_dev->ops->clean_vf_config)
3075 		ae_dev->ops->clean_vf_config(ae_dev, num_vfs);
3076 }
3077 
3078 /* hns3_remove - Device removal routine
3079  * @pdev: PCI device information struct
3080  */
3081 static void hns3_remove(struct pci_dev *pdev)
3082 {
3083 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3084 
3085 	if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
3086 		hns3_disable_sriov(pdev);
3087 
3088 	hnae3_unregister_ae_dev(ae_dev);
3089 	pci_set_drvdata(pdev, NULL);
3090 }
3091 
3092 /**
3093  * hns3_pci_sriov_configure
3094  * @pdev: pointer to a pci_dev structure
3095  * @num_vfs: number of VFs to allocate
3096  *
3097  * Enable or change the number of VFs. Called when the user updates the number
3098  * of VFs in sysfs.
3099  **/
3100 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
3101 {
3102 	int ret;
3103 
3104 	if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
3105 		dev_warn(&pdev->dev, "Can not config SRIOV\n");
3106 		return -EINVAL;
3107 	}
3108 
3109 	if (num_vfs) {
3110 		ret = pci_enable_sriov(pdev, num_vfs);
3111 		if (ret)
3112 			dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
3113 		else
3114 			return num_vfs;
3115 	} else if (!pci_vfs_assigned(pdev)) {
3116 		int num_vfs_pre = pci_num_vf(pdev);
3117 
3118 		pci_disable_sriov(pdev);
3119 		hns3_clean_vf_config(pdev, num_vfs_pre);
3120 	} else {
3121 		dev_warn(&pdev->dev,
3122 			 "Unable to free VFs because some are assigned to VMs.\n");
3123 	}
3124 
3125 	return 0;
3126 }
3127 
3128 static void hns3_shutdown(struct pci_dev *pdev)
3129 {
3130 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3131 
3132 	hnae3_unregister_ae_dev(ae_dev);
3133 	pci_set_drvdata(pdev, NULL);
3134 
3135 	if (system_state == SYSTEM_POWER_OFF)
3136 		pci_set_power_state(pdev, PCI_D3hot);
3137 }
3138 
3139 static int __maybe_unused hns3_suspend(struct device *dev)
3140 {
3141 	struct hnae3_ae_dev *ae_dev = dev_get_drvdata(dev);
3142 
3143 	if (ae_dev && hns3_is_phys_func(ae_dev->pdev)) {
3144 		dev_info(dev, "Begin to suspend.\n");
3145 		if (ae_dev->ops && ae_dev->ops->reset_prepare)
3146 			ae_dev->ops->reset_prepare(ae_dev, HNAE3_FUNC_RESET);
3147 	}
3148 
3149 	return 0;
3150 }
3151 
3152 static int __maybe_unused hns3_resume(struct device *dev)
3153 {
3154 	struct hnae3_ae_dev *ae_dev = dev_get_drvdata(dev);
3155 
3156 	if (ae_dev && hns3_is_phys_func(ae_dev->pdev)) {
3157 		dev_info(dev, "Begin to resume.\n");
3158 		if (ae_dev->ops && ae_dev->ops->reset_done)
3159 			ae_dev->ops->reset_done(ae_dev);
3160 	}
3161 
3162 	return 0;
3163 }
3164 
3165 static pci_ers_result_t hns3_error_detected(struct pci_dev *pdev,
3166 					    pci_channel_state_t state)
3167 {
3168 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3169 	pci_ers_result_t ret;
3170 
3171 	dev_info(&pdev->dev, "PCI error detected, state(=%u)!!\n", state);
3172 
3173 	if (state == pci_channel_io_perm_failure)
3174 		return PCI_ERS_RESULT_DISCONNECT;
3175 
3176 	if (!ae_dev || !ae_dev->ops) {
3177 		dev_err(&pdev->dev,
3178 			"Can't recover - error happened before device initialized\n");
3179 		return PCI_ERS_RESULT_NONE;
3180 	}
3181 
3182 	if (ae_dev->ops->handle_hw_ras_error)
3183 		ret = ae_dev->ops->handle_hw_ras_error(ae_dev);
3184 	else
3185 		return PCI_ERS_RESULT_NONE;
3186 
3187 	return ret;
3188 }
3189 
3190 static pci_ers_result_t hns3_slot_reset(struct pci_dev *pdev)
3191 {
3192 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3193 	const struct hnae3_ae_ops *ops;
3194 	enum hnae3_reset_type reset_type;
3195 	struct device *dev = &pdev->dev;
3196 
3197 	if (!ae_dev || !ae_dev->ops)
3198 		return PCI_ERS_RESULT_NONE;
3199 
3200 	ops = ae_dev->ops;
3201 	/* request the reset */
3202 	if (ops->reset_event && ops->get_reset_level &&
3203 	    ops->set_default_reset_request) {
3204 		if (ae_dev->hw_err_reset_req) {
3205 			reset_type = ops->get_reset_level(ae_dev,
3206 						&ae_dev->hw_err_reset_req);
3207 			ops->set_default_reset_request(ae_dev, reset_type);
3208 			dev_info(dev, "requesting reset due to PCI error\n");
3209 			ops->reset_event(pdev, NULL);
3210 		}
3211 
3212 		return PCI_ERS_RESULT_RECOVERED;
3213 	}
3214 
3215 	return PCI_ERS_RESULT_DISCONNECT;
3216 }
3217 
3218 static void hns3_reset_prepare(struct pci_dev *pdev)
3219 {
3220 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3221 
3222 	dev_info(&pdev->dev, "FLR prepare\n");
3223 	if (ae_dev && ae_dev->ops && ae_dev->ops->reset_prepare)
3224 		ae_dev->ops->reset_prepare(ae_dev, HNAE3_FLR_RESET);
3225 }
3226 
3227 static void hns3_reset_done(struct pci_dev *pdev)
3228 {
3229 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3230 
3231 	dev_info(&pdev->dev, "FLR done\n");
3232 	if (ae_dev && ae_dev->ops && ae_dev->ops->reset_done)
3233 		ae_dev->ops->reset_done(ae_dev);
3234 }
3235 
3236 static const struct pci_error_handlers hns3_err_handler = {
3237 	.error_detected = hns3_error_detected,
3238 	.slot_reset     = hns3_slot_reset,
3239 	.reset_prepare	= hns3_reset_prepare,
3240 	.reset_done	= hns3_reset_done,
3241 };
3242 
3243 static SIMPLE_DEV_PM_OPS(hns3_pm_ops, hns3_suspend, hns3_resume);
3244 
3245 static struct pci_driver hns3_driver = {
3246 	.name     = hns3_driver_name,
3247 	.id_table = hns3_pci_tbl,
3248 	.probe    = hns3_probe,
3249 	.remove   = hns3_remove,
3250 	.shutdown = hns3_shutdown,
3251 	.driver.pm  = &hns3_pm_ops,
3252 	.sriov_configure = hns3_pci_sriov_configure,
3253 	.err_handler    = &hns3_err_handler,
3254 };
3255 
3256 /* set default feature to hns3 */
3257 static void hns3_set_default_feature(struct net_device *netdev)
3258 {
3259 	struct hnae3_handle *h = hns3_get_handle(netdev);
3260 	struct pci_dev *pdev = h->pdev;
3261 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3262 
3263 	netdev->priv_flags |= IFF_UNICAST_FLT;
3264 
3265 	netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
3266 
3267 	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
3268 		NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3269 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
3270 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
3271 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
3272 		NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
3273 
3274 	if (hnae3_ae_dev_gro_supported(ae_dev))
3275 		netdev->features |= NETIF_F_GRO_HW;
3276 
3277 	if (hnae3_ae_dev_fd_supported(ae_dev))
3278 		netdev->features |= NETIF_F_NTUPLE;
3279 
3280 	if (test_bit(HNAE3_DEV_SUPPORT_UDP_GSO_B, ae_dev->caps))
3281 		netdev->features |= NETIF_F_GSO_UDP_L4;
3282 
3283 	if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps))
3284 		netdev->features |= NETIF_F_HW_CSUM;
3285 	else
3286 		netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3287 
3288 	if (test_bit(HNAE3_DEV_SUPPORT_UDP_TUNNEL_CSUM_B, ae_dev->caps))
3289 		netdev->features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
3290 
3291 	if (test_bit(HNAE3_DEV_SUPPORT_FD_FORWARD_TC_B, ae_dev->caps))
3292 		netdev->features |= NETIF_F_HW_TC;
3293 
3294 	netdev->hw_features |= netdev->features;
3295 	if (!test_bit(HNAE3_DEV_SUPPORT_VLAN_FLTR_MDF_B, ae_dev->caps))
3296 		netdev->hw_features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
3297 
3298 	netdev->vlan_features |= netdev->features &
3299 		~(NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_CTAG_TX |
3300 		  NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_GRO_HW | NETIF_F_NTUPLE |
3301 		  NETIF_F_HW_TC);
3302 
3303 	netdev->hw_enc_features |= netdev->vlan_features | NETIF_F_TSO_MANGLEID;
3304 }
3305 
3306 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
3307 			     struct hns3_desc_cb *cb)
3308 {
3309 	unsigned int order = hns3_page_order(ring);
3310 	struct page *p;
3311 
3312 	if (ring->page_pool) {
3313 		p = page_pool_dev_alloc_frag(ring->page_pool,
3314 					     &cb->page_offset,
3315 					     hns3_buf_size(ring));
3316 		if (unlikely(!p))
3317 			return -ENOMEM;
3318 
3319 		cb->priv = p;
3320 		cb->buf = page_address(p);
3321 		cb->dma = page_pool_get_dma_addr(p);
3322 		cb->type = DESC_TYPE_PP_FRAG;
3323 		cb->reuse_flag = 0;
3324 		return 0;
3325 	}
3326 
3327 	p = dev_alloc_pages(order);
3328 	if (!p)
3329 		return -ENOMEM;
3330 
3331 	cb->priv = p;
3332 	cb->page_offset = 0;
3333 	cb->reuse_flag = 0;
3334 	cb->buf  = page_address(p);
3335 	cb->length = hns3_page_size(ring);
3336 	cb->type = DESC_TYPE_PAGE;
3337 	page_ref_add(p, USHRT_MAX - 1);
3338 	cb->pagecnt_bias = USHRT_MAX;
3339 
3340 	return 0;
3341 }
3342 
3343 static void hns3_free_buffer(struct hns3_enet_ring *ring,
3344 			     struct hns3_desc_cb *cb, int budget)
3345 {
3346 	if (cb->type & (DESC_TYPE_SKB | DESC_TYPE_BOUNCE_HEAD |
3347 			DESC_TYPE_BOUNCE_ALL | DESC_TYPE_SGL_SKB))
3348 		napi_consume_skb(cb->priv, budget);
3349 	else if (!HNAE3_IS_TX_RING(ring)) {
3350 		if (cb->type & DESC_TYPE_PAGE && cb->pagecnt_bias)
3351 			__page_frag_cache_drain(cb->priv, cb->pagecnt_bias);
3352 		else if (cb->type & DESC_TYPE_PP_FRAG)
3353 			page_pool_put_full_page(ring->page_pool, cb->priv,
3354 						false);
3355 	}
3356 	memset(cb, 0, sizeof(*cb));
3357 }
3358 
3359 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
3360 {
3361 	cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
3362 			       cb->length, ring_to_dma_dir(ring));
3363 
3364 	if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
3365 		return -EIO;
3366 
3367 	return 0;
3368 }
3369 
3370 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
3371 			      struct hns3_desc_cb *cb)
3372 {
3373 	if (cb->type & (DESC_TYPE_SKB | DESC_TYPE_FRAGLIST_SKB))
3374 		dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
3375 				 ring_to_dma_dir(ring));
3376 	else if ((cb->type & DESC_TYPE_PAGE) && cb->length)
3377 		dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
3378 			       ring_to_dma_dir(ring));
3379 	else if (cb->type & (DESC_TYPE_BOUNCE_ALL | DESC_TYPE_BOUNCE_HEAD |
3380 			     DESC_TYPE_SGL_SKB))
3381 		hns3_tx_spare_reclaim_cb(ring, cb);
3382 }
3383 
3384 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
3385 {
3386 	hns3_unmap_buffer(ring, &ring->desc_cb[i]);
3387 	ring->desc[i].addr = 0;
3388 	ring->desc_cb[i].refill = 0;
3389 }
3390 
3391 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i,
3392 				    int budget)
3393 {
3394 	struct hns3_desc_cb *cb = &ring->desc_cb[i];
3395 
3396 	if (!ring->desc_cb[i].dma)
3397 		return;
3398 
3399 	hns3_buffer_detach(ring, i);
3400 	hns3_free_buffer(ring, cb, budget);
3401 }
3402 
3403 static void hns3_free_buffers(struct hns3_enet_ring *ring)
3404 {
3405 	int i;
3406 
3407 	for (i = 0; i < ring->desc_num; i++)
3408 		hns3_free_buffer_detach(ring, i, 0);
3409 }
3410 
3411 /* free desc along with its attached buffer */
3412 static void hns3_free_desc(struct hns3_enet_ring *ring)
3413 {
3414 	int size = ring->desc_num * sizeof(ring->desc[0]);
3415 
3416 	hns3_free_buffers(ring);
3417 
3418 	if (ring->desc) {
3419 		dma_free_coherent(ring_to_dev(ring), size,
3420 				  ring->desc, ring->desc_dma_addr);
3421 		ring->desc = NULL;
3422 	}
3423 }
3424 
3425 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
3426 {
3427 	int size = ring->desc_num * sizeof(ring->desc[0]);
3428 
3429 	ring->desc = dma_alloc_coherent(ring_to_dev(ring), size,
3430 					&ring->desc_dma_addr, GFP_KERNEL);
3431 	if (!ring->desc)
3432 		return -ENOMEM;
3433 
3434 	return 0;
3435 }
3436 
3437 static int hns3_alloc_and_map_buffer(struct hns3_enet_ring *ring,
3438 				   struct hns3_desc_cb *cb)
3439 {
3440 	int ret;
3441 
3442 	ret = hns3_alloc_buffer(ring, cb);
3443 	if (ret || ring->page_pool)
3444 		goto out;
3445 
3446 	ret = hns3_map_buffer(ring, cb);
3447 	if (ret)
3448 		goto out_with_buf;
3449 
3450 	return 0;
3451 
3452 out_with_buf:
3453 	hns3_free_buffer(ring, cb, 0);
3454 out:
3455 	return ret;
3456 }
3457 
3458 static int hns3_alloc_and_attach_buffer(struct hns3_enet_ring *ring, int i)
3459 {
3460 	int ret = hns3_alloc_and_map_buffer(ring, &ring->desc_cb[i]);
3461 
3462 	if (ret)
3463 		return ret;
3464 
3465 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3466 					 ring->desc_cb[i].page_offset);
3467 	ring->desc_cb[i].refill = 1;
3468 
3469 	return 0;
3470 }
3471 
3472 /* Allocate memory for raw pkg, and map with dma */
3473 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
3474 {
3475 	int i, j, ret;
3476 
3477 	for (i = 0; i < ring->desc_num; i++) {
3478 		ret = hns3_alloc_and_attach_buffer(ring, i);
3479 		if (ret)
3480 			goto out_buffer_fail;
3481 	}
3482 
3483 	return 0;
3484 
3485 out_buffer_fail:
3486 	for (j = i - 1; j >= 0; j--)
3487 		hns3_free_buffer_detach(ring, j, 0);
3488 	return ret;
3489 }
3490 
3491 /* detach a in-used buffer and replace with a reserved one */
3492 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
3493 				struct hns3_desc_cb *res_cb)
3494 {
3495 	hns3_unmap_buffer(ring, &ring->desc_cb[i]);
3496 	ring->desc_cb[i] = *res_cb;
3497 	ring->desc_cb[i].refill = 1;
3498 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3499 					 ring->desc_cb[i].page_offset);
3500 	ring->desc[i].rx.bd_base_info = 0;
3501 }
3502 
3503 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
3504 {
3505 	ring->desc_cb[i].reuse_flag = 0;
3506 	ring->desc_cb[i].refill = 1;
3507 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
3508 					 ring->desc_cb[i].page_offset);
3509 	ring->desc[i].rx.bd_base_info = 0;
3510 
3511 	dma_sync_single_for_device(ring_to_dev(ring),
3512 			ring->desc_cb[i].dma + ring->desc_cb[i].page_offset,
3513 			hns3_buf_size(ring),
3514 			DMA_FROM_DEVICE);
3515 }
3516 
3517 static bool hns3_nic_reclaim_desc(struct hns3_enet_ring *ring,
3518 				  int *bytes, int *pkts, int budget)
3519 {
3520 	/* pair with ring->last_to_use update in hns3_tx_doorbell(),
3521 	 * smp_store_release() is not used in hns3_tx_doorbell() because
3522 	 * the doorbell operation already have the needed barrier operation.
3523 	 */
3524 	int ltu = smp_load_acquire(&ring->last_to_use);
3525 	int ntc = ring->next_to_clean;
3526 	struct hns3_desc_cb *desc_cb;
3527 	bool reclaimed = false;
3528 	struct hns3_desc *desc;
3529 
3530 	while (ltu != ntc) {
3531 		desc = &ring->desc[ntc];
3532 
3533 		if (le16_to_cpu(desc->tx.bdtp_fe_sc_vld_ra_ri) &
3534 				BIT(HNS3_TXD_VLD_B))
3535 			break;
3536 
3537 		desc_cb = &ring->desc_cb[ntc];
3538 
3539 		if (desc_cb->type & (DESC_TYPE_SKB | DESC_TYPE_BOUNCE_ALL |
3540 				     DESC_TYPE_BOUNCE_HEAD |
3541 				     DESC_TYPE_SGL_SKB)) {
3542 			(*pkts)++;
3543 			(*bytes) += desc_cb->send_bytes;
3544 		}
3545 
3546 		/* desc_cb will be cleaned, after hnae3_free_buffer_detach */
3547 		hns3_free_buffer_detach(ring, ntc, budget);
3548 
3549 		if (++ntc == ring->desc_num)
3550 			ntc = 0;
3551 
3552 		/* Issue prefetch for next Tx descriptor */
3553 		prefetch(&ring->desc_cb[ntc]);
3554 		reclaimed = true;
3555 	}
3556 
3557 	if (unlikely(!reclaimed))
3558 		return false;
3559 
3560 	/* This smp_store_release() pairs with smp_load_acquire() in
3561 	 * ring_space called by hns3_nic_net_xmit.
3562 	 */
3563 	smp_store_release(&ring->next_to_clean, ntc);
3564 
3565 	hns3_tx_spare_update(ring);
3566 
3567 	return true;
3568 }
3569 
3570 void hns3_clean_tx_ring(struct hns3_enet_ring *ring, int budget)
3571 {
3572 	struct net_device *netdev = ring_to_netdev(ring);
3573 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3574 	struct netdev_queue *dev_queue;
3575 	int bytes, pkts;
3576 
3577 	bytes = 0;
3578 	pkts = 0;
3579 
3580 	if (unlikely(!hns3_nic_reclaim_desc(ring, &bytes, &pkts, budget)))
3581 		return;
3582 
3583 	ring->tqp_vector->tx_group.total_bytes += bytes;
3584 	ring->tqp_vector->tx_group.total_packets += pkts;
3585 
3586 	u64_stats_update_begin(&ring->syncp);
3587 	ring->stats.tx_bytes += bytes;
3588 	ring->stats.tx_pkts += pkts;
3589 	u64_stats_update_end(&ring->syncp);
3590 
3591 	dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
3592 	netdev_tx_completed_queue(dev_queue, pkts, bytes);
3593 
3594 	if (unlikely(netif_carrier_ok(netdev) &&
3595 		     ring_space(ring) > HNS3_MAX_TSO_BD_NUM)) {
3596 		/* Make sure that anybody stopping the queue after this
3597 		 * sees the new next_to_clean.
3598 		 */
3599 		smp_mb();
3600 		if (netif_tx_queue_stopped(dev_queue) &&
3601 		    !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
3602 			netif_tx_wake_queue(dev_queue);
3603 			ring->stats.restart_queue++;
3604 		}
3605 	}
3606 }
3607 
3608 static int hns3_desc_unused(struct hns3_enet_ring *ring)
3609 {
3610 	int ntc = ring->next_to_clean;
3611 	int ntu = ring->next_to_use;
3612 
3613 	if (unlikely(ntc == ntu && !ring->desc_cb[ntc].refill))
3614 		return ring->desc_num;
3615 
3616 	return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
3617 }
3618 
3619 /* Return true if there is any allocation failure */
3620 static bool hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring,
3621 				      int cleand_count)
3622 {
3623 	struct hns3_desc_cb *desc_cb;
3624 	struct hns3_desc_cb res_cbs;
3625 	int i, ret;
3626 
3627 	for (i = 0; i < cleand_count; i++) {
3628 		desc_cb = &ring->desc_cb[ring->next_to_use];
3629 		if (desc_cb->reuse_flag) {
3630 			hns3_ring_stats_update(ring, reuse_pg_cnt);
3631 
3632 			hns3_reuse_buffer(ring, ring->next_to_use);
3633 		} else {
3634 			ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
3635 			if (ret) {
3636 				hns3_ring_stats_update(ring, sw_err_cnt);
3637 
3638 				hns3_rl_err(ring_to_netdev(ring),
3639 					    "alloc rx buffer failed: %d\n",
3640 					    ret);
3641 
3642 				writel(i, ring->tqp->io_base +
3643 				       HNS3_RING_RX_RING_HEAD_REG);
3644 				return true;
3645 			}
3646 			hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
3647 
3648 			hns3_ring_stats_update(ring, non_reuse_pg);
3649 		}
3650 
3651 		ring_ptr_move_fw(ring, next_to_use);
3652 	}
3653 
3654 	writel(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
3655 	return false;
3656 }
3657 
3658 static bool hns3_can_reuse_page(struct hns3_desc_cb *cb)
3659 {
3660 	return page_count(cb->priv) == cb->pagecnt_bias;
3661 }
3662 
3663 static int hns3_handle_rx_copybreak(struct sk_buff *skb, int i,
3664 				    struct hns3_enet_ring *ring,
3665 				    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 frag_size = size - pull_len;
3672 	void *frag = napi_alloc_frag(frag_size);
3673 
3674 	if (unlikely(!frag)) {
3675 		hns3_ring_stats_update(ring, frag_alloc_err);
3676 
3677 		hns3_rl_err(ring_to_netdev(ring),
3678 			    "failed to allocate rx frag\n");
3679 		return -ENOMEM;
3680 	}
3681 
3682 	desc_cb->reuse_flag = 1;
3683 	memcpy(frag, desc_cb->buf + frag_offset, frag_size);
3684 	skb_add_rx_frag(skb, i, virt_to_page(frag),
3685 			offset_in_page(frag), frag_size, frag_size);
3686 
3687 	hns3_ring_stats_update(ring, frag_alloc);
3688 	return 0;
3689 }
3690 
3691 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
3692 				struct hns3_enet_ring *ring, int pull_len,
3693 				struct hns3_desc_cb *desc_cb)
3694 {
3695 	struct hns3_desc *desc = &ring->desc[ring->next_to_clean];
3696 	u32 frag_offset = desc_cb->page_offset + pull_len;
3697 	int size = le16_to_cpu(desc->rx.size);
3698 	u32 truesize = hns3_buf_size(ring);
3699 	u32 frag_size = size - pull_len;
3700 	int ret = 0;
3701 	bool reused;
3702 
3703 	if (ring->page_pool) {
3704 		skb_add_rx_frag(skb, i, desc_cb->priv, frag_offset,
3705 				frag_size, truesize);
3706 		return;
3707 	}
3708 
3709 	/* Avoid re-using remote or pfmem page */
3710 	if (unlikely(!dev_page_is_reusable(desc_cb->priv)))
3711 		goto out;
3712 
3713 	reused = hns3_can_reuse_page(desc_cb);
3714 
3715 	/* Rx page can be reused when:
3716 	 * 1. Rx page is only owned by the driver when page_offset
3717 	 *    is zero, which means 0 @ truesize will be used by
3718 	 *    stack after skb_add_rx_frag() is called, and the rest
3719 	 *    of rx page can be reused by driver.
3720 	 * Or
3721 	 * 2. Rx page is only owned by the driver when page_offset
3722 	 *    is non-zero, which means page_offset @ truesize will
3723 	 *    be used by stack after skb_add_rx_frag() is called,
3724 	 *    and 0 @ truesize can be reused by driver.
3725 	 */
3726 	if ((!desc_cb->page_offset && reused) ||
3727 	    ((desc_cb->page_offset + truesize + truesize) <=
3728 	     hns3_page_size(ring) && desc_cb->page_offset)) {
3729 		desc_cb->page_offset += truesize;
3730 		desc_cb->reuse_flag = 1;
3731 	} else if (desc_cb->page_offset && reused) {
3732 		desc_cb->page_offset = 0;
3733 		desc_cb->reuse_flag = 1;
3734 	} else if (frag_size <= ring->rx_copybreak) {
3735 		ret = hns3_handle_rx_copybreak(skb, i, ring, pull_len, desc_cb);
3736 		if (ret)
3737 			goto out;
3738 	}
3739 
3740 out:
3741 	desc_cb->pagecnt_bias--;
3742 
3743 	if (unlikely(!desc_cb->pagecnt_bias)) {
3744 		page_ref_add(desc_cb->priv, USHRT_MAX);
3745 		desc_cb->pagecnt_bias = USHRT_MAX;
3746 	}
3747 
3748 	skb_add_rx_frag(skb, i, desc_cb->priv, frag_offset,
3749 			frag_size, truesize);
3750 
3751 	if (unlikely(!desc_cb->reuse_flag))
3752 		__page_frag_cache_drain(desc_cb->priv, desc_cb->pagecnt_bias);
3753 }
3754 
3755 static int hns3_gro_complete(struct sk_buff *skb, u32 l234info)
3756 {
3757 	__be16 type = skb->protocol;
3758 	struct tcphdr *th;
3759 	int depth = 0;
3760 
3761 	while (eth_type_vlan(type)) {
3762 		struct vlan_hdr *vh;
3763 
3764 		if ((depth + VLAN_HLEN) > skb_headlen(skb))
3765 			return -EFAULT;
3766 
3767 		vh = (struct vlan_hdr *)(skb->data + depth);
3768 		type = vh->h_vlan_encapsulated_proto;
3769 		depth += VLAN_HLEN;
3770 	}
3771 
3772 	skb_set_network_header(skb, depth);
3773 
3774 	if (type == htons(ETH_P_IP)) {
3775 		const struct iphdr *iph = ip_hdr(skb);
3776 
3777 		depth += sizeof(struct iphdr);
3778 		skb_set_transport_header(skb, depth);
3779 		th = tcp_hdr(skb);
3780 		th->check = ~tcp_v4_check(skb->len - depth, iph->saddr,
3781 					  iph->daddr, 0);
3782 	} else if (type == htons(ETH_P_IPV6)) {
3783 		const struct ipv6hdr *iph = ipv6_hdr(skb);
3784 
3785 		depth += sizeof(struct ipv6hdr);
3786 		skb_set_transport_header(skb, depth);
3787 		th = tcp_hdr(skb);
3788 		th->check = ~tcp_v6_check(skb->len - depth, &iph->saddr,
3789 					  &iph->daddr, 0);
3790 	} else {
3791 		hns3_rl_err(skb->dev,
3792 			    "Error: FW GRO supports only IPv4/IPv6, not 0x%04x, depth: %d\n",
3793 			    be16_to_cpu(type), depth);
3794 		return -EFAULT;
3795 	}
3796 
3797 	skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
3798 	if (th->cwr)
3799 		skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
3800 
3801 	if (l234info & BIT(HNS3_RXD_GRO_FIXID_B))
3802 		skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_FIXEDID;
3803 
3804 	skb->csum_start = (unsigned char *)th - skb->head;
3805 	skb->csum_offset = offsetof(struct tcphdr, check);
3806 	skb->ip_summed = CHECKSUM_PARTIAL;
3807 
3808 	trace_hns3_gro(skb);
3809 
3810 	return 0;
3811 }
3812 
3813 static bool hns3_checksum_complete(struct hns3_enet_ring *ring,
3814 				   struct sk_buff *skb, u32 ptype, u16 csum)
3815 {
3816 	if (ptype == HNS3_INVALID_PTYPE ||
3817 	    hns3_rx_ptype_tbl[ptype].ip_summed != CHECKSUM_COMPLETE)
3818 		return false;
3819 
3820 	hns3_ring_stats_update(ring, csum_complete);
3821 	skb->ip_summed = CHECKSUM_COMPLETE;
3822 	skb->csum = csum_unfold((__force __sum16)csum);
3823 
3824 	return true;
3825 }
3826 
3827 static void hns3_rx_handle_csum(struct sk_buff *skb, u32 l234info,
3828 				u32 ol_info, u32 ptype)
3829 {
3830 	int l3_type, l4_type;
3831 	int ol4_type;
3832 
3833 	if (ptype != HNS3_INVALID_PTYPE) {
3834 		skb->csum_level = hns3_rx_ptype_tbl[ptype].csum_level;
3835 		skb->ip_summed = hns3_rx_ptype_tbl[ptype].ip_summed;
3836 
3837 		return;
3838 	}
3839 
3840 	ol4_type = hnae3_get_field(ol_info, HNS3_RXD_OL4ID_M,
3841 				   HNS3_RXD_OL4ID_S);
3842 	switch (ol4_type) {
3843 	case HNS3_OL4_TYPE_MAC_IN_UDP:
3844 	case HNS3_OL4_TYPE_NVGRE:
3845 		skb->csum_level = 1;
3846 		fallthrough;
3847 	case HNS3_OL4_TYPE_NO_TUN:
3848 		l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
3849 					  HNS3_RXD_L3ID_S);
3850 		l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
3851 					  HNS3_RXD_L4ID_S);
3852 		/* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
3853 		if ((l3_type == HNS3_L3_TYPE_IPV4 ||
3854 		     l3_type == HNS3_L3_TYPE_IPV6) &&
3855 		    (l4_type == HNS3_L4_TYPE_UDP ||
3856 		     l4_type == HNS3_L4_TYPE_TCP ||
3857 		     l4_type == HNS3_L4_TYPE_SCTP))
3858 			skb->ip_summed = CHECKSUM_UNNECESSARY;
3859 		break;
3860 	default:
3861 		break;
3862 	}
3863 }
3864 
3865 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
3866 			     u32 l234info, u32 bd_base_info, u32 ol_info,
3867 			     u16 csum)
3868 {
3869 	struct net_device *netdev = ring_to_netdev(ring);
3870 	struct hns3_nic_priv *priv = netdev_priv(netdev);
3871 	u32 ptype = HNS3_INVALID_PTYPE;
3872 
3873 	skb->ip_summed = CHECKSUM_NONE;
3874 
3875 	skb_checksum_none_assert(skb);
3876 
3877 	if (!(netdev->features & NETIF_F_RXCSUM))
3878 		return;
3879 
3880 	if (test_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state))
3881 		ptype = hnae3_get_field(ol_info, HNS3_RXD_PTYPE_M,
3882 					HNS3_RXD_PTYPE_S);
3883 
3884 	if (hns3_checksum_complete(ring, skb, ptype, csum))
3885 		return;
3886 
3887 	/* check if hardware has done checksum */
3888 	if (!(bd_base_info & BIT(HNS3_RXD_L3L4P_B)))
3889 		return;
3890 
3891 	if (unlikely(l234info & (BIT(HNS3_RXD_L3E_B) | BIT(HNS3_RXD_L4E_B) |
3892 				 BIT(HNS3_RXD_OL3E_B) |
3893 				 BIT(HNS3_RXD_OL4E_B)))) {
3894 		hns3_ring_stats_update(ring, l3l4_csum_err);
3895 
3896 		return;
3897 	}
3898 
3899 	hns3_rx_handle_csum(skb, l234info, ol_info, ptype);
3900 }
3901 
3902 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
3903 {
3904 	if (skb_has_frag_list(skb))
3905 		napi_gro_flush(&ring->tqp_vector->napi, false);
3906 
3907 	napi_gro_receive(&ring->tqp_vector->napi, skb);
3908 }
3909 
3910 static bool hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
3911 				struct hns3_desc *desc, u32 l234info,
3912 				u16 *vlan_tag)
3913 {
3914 	struct hnae3_handle *handle = ring->tqp->handle;
3915 	struct pci_dev *pdev = ring->tqp->handle->pdev;
3916 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3917 
3918 	if (unlikely(ae_dev->dev_version < HNAE3_DEVICE_VERSION_V2)) {
3919 		*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3920 		if (!(*vlan_tag & VLAN_VID_MASK))
3921 			*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3922 
3923 		return (*vlan_tag != 0);
3924 	}
3925 
3926 #define HNS3_STRP_OUTER_VLAN	0x1
3927 #define HNS3_STRP_INNER_VLAN	0x2
3928 #define HNS3_STRP_BOTH		0x3
3929 
3930 	/* Hardware always insert VLAN tag into RX descriptor when
3931 	 * remove the tag from packet, driver needs to determine
3932 	 * reporting which tag to stack.
3933 	 */
3934 	switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
3935 				HNS3_RXD_STRP_TAGP_S)) {
3936 	case HNS3_STRP_OUTER_VLAN:
3937 		if (handle->port_base_vlan_state !=
3938 				HNAE3_PORT_BASE_VLAN_DISABLE)
3939 			return false;
3940 
3941 		*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3942 		return true;
3943 	case HNS3_STRP_INNER_VLAN:
3944 		if (handle->port_base_vlan_state !=
3945 				HNAE3_PORT_BASE_VLAN_DISABLE)
3946 			return false;
3947 
3948 		*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3949 		return true;
3950 	case HNS3_STRP_BOTH:
3951 		if (handle->port_base_vlan_state ==
3952 				HNAE3_PORT_BASE_VLAN_DISABLE)
3953 			*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3954 		else
3955 			*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3956 
3957 		return true;
3958 	default:
3959 		return false;
3960 	}
3961 }
3962 
3963 static void hns3_rx_ring_move_fw(struct hns3_enet_ring *ring)
3964 {
3965 	ring->desc[ring->next_to_clean].rx.bd_base_info &=
3966 		cpu_to_le32(~BIT(HNS3_RXD_VLD_B));
3967 	ring->desc_cb[ring->next_to_clean].refill = 0;
3968 	ring->next_to_clean += 1;
3969 
3970 	if (unlikely(ring->next_to_clean == ring->desc_num))
3971 		ring->next_to_clean = 0;
3972 }
3973 
3974 static int hns3_alloc_skb(struct hns3_enet_ring *ring, unsigned int length,
3975 			  unsigned char *va)
3976 {
3977 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
3978 	struct net_device *netdev = ring_to_netdev(ring);
3979 	struct sk_buff *skb;
3980 
3981 	ring->skb = napi_alloc_skb(&ring->tqp_vector->napi, HNS3_RX_HEAD_SIZE);
3982 	skb = ring->skb;
3983 	if (unlikely(!skb)) {
3984 		hns3_rl_err(netdev, "alloc rx skb fail\n");
3985 		hns3_ring_stats_update(ring, sw_err_cnt);
3986 
3987 		return -ENOMEM;
3988 	}
3989 
3990 	trace_hns3_rx_desc(ring);
3991 	prefetchw(skb->data);
3992 
3993 	ring->pending_buf = 1;
3994 	ring->frag_num = 0;
3995 	ring->tail_skb = NULL;
3996 	if (length <= HNS3_RX_HEAD_SIZE) {
3997 		memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
3998 
3999 		/* We can reuse buffer as-is, just make sure it is reusable */
4000 		if (dev_page_is_reusable(desc_cb->priv))
4001 			desc_cb->reuse_flag = 1;
4002 		else if (desc_cb->type & DESC_TYPE_PP_FRAG)
4003 			page_pool_put_full_page(ring->page_pool, desc_cb->priv,
4004 						false);
4005 		else /* This page cannot be reused so discard it */
4006 			__page_frag_cache_drain(desc_cb->priv,
4007 						desc_cb->pagecnt_bias);
4008 
4009 		hns3_rx_ring_move_fw(ring);
4010 		return 0;
4011 	}
4012 
4013 	if (ring->page_pool)
4014 		skb_mark_for_recycle(skb);
4015 
4016 	hns3_ring_stats_update(ring, seg_pkt_cnt);
4017 
4018 	ring->pull_len = eth_get_headlen(netdev, va, HNS3_RX_HEAD_SIZE);
4019 	__skb_put(skb, ring->pull_len);
4020 	hns3_nic_reuse_page(skb, ring->frag_num++, ring, ring->pull_len,
4021 			    desc_cb);
4022 	hns3_rx_ring_move_fw(ring);
4023 
4024 	return 0;
4025 }
4026 
4027 static int hns3_add_frag(struct hns3_enet_ring *ring)
4028 {
4029 	struct sk_buff *skb = ring->skb;
4030 	struct sk_buff *head_skb = skb;
4031 	struct sk_buff *new_skb;
4032 	struct hns3_desc_cb *desc_cb;
4033 	struct hns3_desc *desc;
4034 	u32 bd_base_info;
4035 
4036 	do {
4037 		desc = &ring->desc[ring->next_to_clean];
4038 		desc_cb = &ring->desc_cb[ring->next_to_clean];
4039 		bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
4040 		/* make sure HW write desc complete */
4041 		dma_rmb();
4042 		if (!(bd_base_info & BIT(HNS3_RXD_VLD_B)))
4043 			return -ENXIO;
4044 
4045 		if (unlikely(ring->frag_num >= MAX_SKB_FRAGS)) {
4046 			new_skb = napi_alloc_skb(&ring->tqp_vector->napi, 0);
4047 			if (unlikely(!new_skb)) {
4048 				hns3_rl_err(ring_to_netdev(ring),
4049 					    "alloc rx fraglist skb fail\n");
4050 				return -ENXIO;
4051 			}
4052 
4053 			if (ring->page_pool)
4054 				skb_mark_for_recycle(new_skb);
4055 
4056 			ring->frag_num = 0;
4057 
4058 			if (ring->tail_skb) {
4059 				ring->tail_skb->next = new_skb;
4060 				ring->tail_skb = new_skb;
4061 			} else {
4062 				skb_shinfo(skb)->frag_list = new_skb;
4063 				ring->tail_skb = new_skb;
4064 			}
4065 		}
4066 
4067 		if (ring->tail_skb) {
4068 			head_skb->truesize += hns3_buf_size(ring);
4069 			head_skb->data_len += le16_to_cpu(desc->rx.size);
4070 			head_skb->len += le16_to_cpu(desc->rx.size);
4071 			skb = ring->tail_skb;
4072 		}
4073 
4074 		dma_sync_single_for_cpu(ring_to_dev(ring),
4075 				desc_cb->dma + desc_cb->page_offset,
4076 				hns3_buf_size(ring),
4077 				DMA_FROM_DEVICE);
4078 
4079 		hns3_nic_reuse_page(skb, ring->frag_num++, ring, 0, desc_cb);
4080 		trace_hns3_rx_desc(ring);
4081 		hns3_rx_ring_move_fw(ring);
4082 		ring->pending_buf++;
4083 	} while (!(bd_base_info & BIT(HNS3_RXD_FE_B)));
4084 
4085 	return 0;
4086 }
4087 
4088 static int hns3_set_gro_and_checksum(struct hns3_enet_ring *ring,
4089 				     struct sk_buff *skb, u32 l234info,
4090 				     u32 bd_base_info, u32 ol_info, u16 csum)
4091 {
4092 	struct net_device *netdev = ring_to_netdev(ring);
4093 	struct hns3_nic_priv *priv = netdev_priv(netdev);
4094 	u32 l3_type;
4095 
4096 	skb_shinfo(skb)->gso_size = hnae3_get_field(bd_base_info,
4097 						    HNS3_RXD_GRO_SIZE_M,
4098 						    HNS3_RXD_GRO_SIZE_S);
4099 	/* if there is no HW GRO, do not set gro params */
4100 	if (!skb_shinfo(skb)->gso_size) {
4101 		hns3_rx_checksum(ring, skb, l234info, bd_base_info, ol_info,
4102 				 csum);
4103 		return 0;
4104 	}
4105 
4106 	NAPI_GRO_CB(skb)->count = hnae3_get_field(l234info,
4107 						  HNS3_RXD_GRO_COUNT_M,
4108 						  HNS3_RXD_GRO_COUNT_S);
4109 
4110 	if (test_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state)) {
4111 		u32 ptype = hnae3_get_field(ol_info, HNS3_RXD_PTYPE_M,
4112 					    HNS3_RXD_PTYPE_S);
4113 
4114 		l3_type = hns3_rx_ptype_tbl[ptype].l3_type;
4115 	} else {
4116 		l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
4117 					  HNS3_RXD_L3ID_S);
4118 	}
4119 
4120 	if (l3_type == HNS3_L3_TYPE_IPV4)
4121 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
4122 	else if (l3_type == HNS3_L3_TYPE_IPV6)
4123 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
4124 	else
4125 		return -EFAULT;
4126 
4127 	return  hns3_gro_complete(skb, l234info);
4128 }
4129 
4130 static void hns3_set_rx_skb_rss_type(struct hns3_enet_ring *ring,
4131 				     struct sk_buff *skb, u32 rss_hash)
4132 {
4133 	struct hnae3_handle *handle = ring->tqp->handle;
4134 	enum pkt_hash_types rss_type;
4135 
4136 	if (rss_hash)
4137 		rss_type = handle->kinfo.rss_type;
4138 	else
4139 		rss_type = PKT_HASH_TYPE_NONE;
4140 
4141 	skb_set_hash(skb, rss_hash, rss_type);
4142 }
4143 
4144 static void hns3_handle_rx_ts_info(struct net_device *netdev,
4145 				   struct hns3_desc *desc, struct sk_buff *skb,
4146 				   u32 bd_base_info)
4147 {
4148 	if (unlikely(bd_base_info & BIT(HNS3_RXD_TS_VLD_B))) {
4149 		struct hnae3_handle *h = hns3_get_handle(netdev);
4150 		u32 nsec = le32_to_cpu(desc->ts_nsec);
4151 		u32 sec = le32_to_cpu(desc->ts_sec);
4152 
4153 		if (h->ae_algo->ops->get_rx_hwts)
4154 			h->ae_algo->ops->get_rx_hwts(h, skb, nsec, sec);
4155 	}
4156 }
4157 
4158 static void hns3_handle_rx_vlan_tag(struct hns3_enet_ring *ring,
4159 				    struct hns3_desc *desc, struct sk_buff *skb,
4160 				    u32 l234info)
4161 {
4162 	struct net_device *netdev = ring_to_netdev(ring);
4163 
4164 	/* Based on hw strategy, the tag offloaded will be stored at
4165 	 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
4166 	 * in one layer tag case.
4167 	 */
4168 	if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
4169 		u16 vlan_tag;
4170 
4171 		if (hns3_parse_vlan_tag(ring, desc, l234info, &vlan_tag))
4172 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
4173 					       vlan_tag);
4174 	}
4175 }
4176 
4177 static int hns3_handle_bdinfo(struct hns3_enet_ring *ring, struct sk_buff *skb)
4178 {
4179 	struct net_device *netdev = ring_to_netdev(ring);
4180 	enum hns3_pkt_l2t_type l2_frame_type;
4181 	u32 bd_base_info, l234info, ol_info;
4182 	struct hns3_desc *desc;
4183 	unsigned int len;
4184 	int pre_ntc, ret;
4185 	u16 csum;
4186 
4187 	/* bdinfo handled below is only valid on the last BD of the
4188 	 * current packet, and ring->next_to_clean indicates the first
4189 	 * descriptor of next packet, so need - 1 below.
4190 	 */
4191 	pre_ntc = ring->next_to_clean ? (ring->next_to_clean - 1) :
4192 					(ring->desc_num - 1);
4193 	desc = &ring->desc[pre_ntc];
4194 	bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
4195 	l234info = le32_to_cpu(desc->rx.l234_info);
4196 	ol_info = le32_to_cpu(desc->rx.ol_info);
4197 	csum = le16_to_cpu(desc->csum);
4198 
4199 	hns3_handle_rx_ts_info(netdev, desc, skb, bd_base_info);
4200 
4201 	hns3_handle_rx_vlan_tag(ring, desc, skb, l234info);
4202 
4203 	if (unlikely(!desc->rx.pkt_len || (l234info & (BIT(HNS3_RXD_TRUNCAT_B) |
4204 				  BIT(HNS3_RXD_L2E_B))))) {
4205 		u64_stats_update_begin(&ring->syncp);
4206 		if (l234info & BIT(HNS3_RXD_L2E_B))
4207 			ring->stats.l2_err++;
4208 		else
4209 			ring->stats.err_pkt_len++;
4210 		u64_stats_update_end(&ring->syncp);
4211 
4212 		return -EFAULT;
4213 	}
4214 
4215 	len = skb->len;
4216 
4217 	/* Do update ip stack process */
4218 	skb->protocol = eth_type_trans(skb, netdev);
4219 
4220 	/* This is needed in order to enable forwarding support */
4221 	ret = hns3_set_gro_and_checksum(ring, skb, l234info,
4222 					bd_base_info, ol_info, csum);
4223 	if (unlikely(ret)) {
4224 		hns3_ring_stats_update(ring, rx_err_cnt);
4225 		return ret;
4226 	}
4227 
4228 	l2_frame_type = hnae3_get_field(l234info, HNS3_RXD_DMAC_M,
4229 					HNS3_RXD_DMAC_S);
4230 
4231 	u64_stats_update_begin(&ring->syncp);
4232 	ring->stats.rx_pkts++;
4233 	ring->stats.rx_bytes += len;
4234 
4235 	if (l2_frame_type == HNS3_L2_TYPE_MULTICAST)
4236 		ring->stats.rx_multicast++;
4237 
4238 	u64_stats_update_end(&ring->syncp);
4239 
4240 	ring->tqp_vector->rx_group.total_bytes += len;
4241 
4242 	hns3_set_rx_skb_rss_type(ring, skb, le32_to_cpu(desc->rx.rss_hash));
4243 	return 0;
4244 }
4245 
4246 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring)
4247 {
4248 	struct sk_buff *skb = ring->skb;
4249 	struct hns3_desc_cb *desc_cb;
4250 	struct hns3_desc *desc;
4251 	unsigned int length;
4252 	u32 bd_base_info;
4253 	int ret;
4254 
4255 	desc = &ring->desc[ring->next_to_clean];
4256 	desc_cb = &ring->desc_cb[ring->next_to_clean];
4257 
4258 	prefetch(desc);
4259 
4260 	if (!skb) {
4261 		bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
4262 		/* Check valid BD */
4263 		if (unlikely(!(bd_base_info & BIT(HNS3_RXD_VLD_B))))
4264 			return -ENXIO;
4265 
4266 		dma_rmb();
4267 		length = le16_to_cpu(desc->rx.size);
4268 
4269 		ring->va = desc_cb->buf + desc_cb->page_offset;
4270 
4271 		dma_sync_single_for_cpu(ring_to_dev(ring),
4272 				desc_cb->dma + desc_cb->page_offset,
4273 				hns3_buf_size(ring),
4274 				DMA_FROM_DEVICE);
4275 
4276 		/* Prefetch first cache line of first page.
4277 		 * Idea is to cache few bytes of the header of the packet.
4278 		 * Our L1 Cache line size is 64B so need to prefetch twice to make
4279 		 * it 128B. But in actual we can have greater size of caches with
4280 		 * 128B Level 1 cache lines. In such a case, single fetch would
4281 		 * suffice to cache in the relevant part of the header.
4282 		 */
4283 		net_prefetch(ring->va);
4284 
4285 		ret = hns3_alloc_skb(ring, length, ring->va);
4286 		skb = ring->skb;
4287 
4288 		if (ret < 0) /* alloc buffer fail */
4289 			return ret;
4290 		if (!(bd_base_info & BIT(HNS3_RXD_FE_B))) { /* need add frag */
4291 			ret = hns3_add_frag(ring);
4292 			if (ret)
4293 				return ret;
4294 		}
4295 	} else {
4296 		ret = hns3_add_frag(ring);
4297 		if (ret)
4298 			return ret;
4299 	}
4300 
4301 	/* As the head data may be changed when GRO enable, copy
4302 	 * the head data in after other data rx completed
4303 	 */
4304 	if (skb->len > HNS3_RX_HEAD_SIZE)
4305 		memcpy(skb->data, ring->va,
4306 		       ALIGN(ring->pull_len, sizeof(long)));
4307 
4308 	ret = hns3_handle_bdinfo(ring, skb);
4309 	if (unlikely(ret)) {
4310 		dev_kfree_skb_any(skb);
4311 		return ret;
4312 	}
4313 
4314 	skb_record_rx_queue(skb, ring->tqp->tqp_index);
4315 	return 0;
4316 }
4317 
4318 int hns3_clean_rx_ring(struct hns3_enet_ring *ring, int budget,
4319 		       void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
4320 {
4321 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
4322 	int unused_count = hns3_desc_unused(ring);
4323 	bool failure = false;
4324 	int recv_pkts = 0;
4325 	int err;
4326 
4327 	unused_count -= ring->pending_buf;
4328 
4329 	while (recv_pkts < budget) {
4330 		/* Reuse or realloc buffers */
4331 		if (unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
4332 			failure = failure ||
4333 				hns3_nic_alloc_rx_buffers(ring, unused_count);
4334 			unused_count = 0;
4335 		}
4336 
4337 		/* Poll one pkt */
4338 		err = hns3_handle_rx_bd(ring);
4339 		/* Do not get FE for the packet or failed to alloc skb */
4340 		if (unlikely(!ring->skb || err == -ENXIO)) {
4341 			goto out;
4342 		} else if (likely(!err)) {
4343 			rx_fn(ring, ring->skb);
4344 			recv_pkts++;
4345 		}
4346 
4347 		unused_count += ring->pending_buf;
4348 		ring->skb = NULL;
4349 		ring->pending_buf = 0;
4350 	}
4351 
4352 out:
4353 	/* sync head pointer before exiting, since hardware will calculate
4354 	 * FBD number with head pointer
4355 	 */
4356 	if (unused_count > 0)
4357 		failure = failure ||
4358 			  hns3_nic_alloc_rx_buffers(ring, unused_count);
4359 
4360 	return failure ? budget : recv_pkts;
4361 }
4362 
4363 static void hns3_update_rx_int_coalesce(struct hns3_enet_tqp_vector *tqp_vector)
4364 {
4365 	struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
4366 	struct dim_sample sample = {};
4367 
4368 	if (!rx_group->coal.adapt_enable)
4369 		return;
4370 
4371 	dim_update_sample(tqp_vector->event_cnt, rx_group->total_packets,
4372 			  rx_group->total_bytes, &sample);
4373 	net_dim(&rx_group->dim, sample);
4374 }
4375 
4376 static void hns3_update_tx_int_coalesce(struct hns3_enet_tqp_vector *tqp_vector)
4377 {
4378 	struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
4379 	struct dim_sample sample = {};
4380 
4381 	if (!tx_group->coal.adapt_enable)
4382 		return;
4383 
4384 	dim_update_sample(tqp_vector->event_cnt, tx_group->total_packets,
4385 			  tx_group->total_bytes, &sample);
4386 	net_dim(&tx_group->dim, sample);
4387 }
4388 
4389 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
4390 {
4391 	struct hns3_nic_priv *priv = netdev_priv(napi->dev);
4392 	struct hns3_enet_ring *ring;
4393 	int rx_pkt_total = 0;
4394 
4395 	struct hns3_enet_tqp_vector *tqp_vector =
4396 		container_of(napi, struct hns3_enet_tqp_vector, napi);
4397 	bool clean_complete = true;
4398 	int rx_budget = budget;
4399 
4400 	if (unlikely(test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
4401 		napi_complete(napi);
4402 		return 0;
4403 	}
4404 
4405 	/* Since the actual Tx work is minimal, we can give the Tx a larger
4406 	 * budget and be more aggressive about cleaning up the Tx descriptors.
4407 	 */
4408 	hns3_for_each_ring(ring, tqp_vector->tx_group)
4409 		hns3_clean_tx_ring(ring, budget);
4410 
4411 	/* make sure rx ring budget not smaller than 1 */
4412 	if (tqp_vector->num_tqps > 1)
4413 		rx_budget = max(budget / tqp_vector->num_tqps, 1);
4414 
4415 	hns3_for_each_ring(ring, tqp_vector->rx_group) {
4416 		int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
4417 						    hns3_rx_skb);
4418 		if (rx_cleaned >= rx_budget)
4419 			clean_complete = false;
4420 
4421 		rx_pkt_total += rx_cleaned;
4422 	}
4423 
4424 	tqp_vector->rx_group.total_packets += rx_pkt_total;
4425 
4426 	if (!clean_complete)
4427 		return budget;
4428 
4429 	if (napi_complete(napi) &&
4430 	    likely(!test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
4431 		hns3_update_rx_int_coalesce(tqp_vector);
4432 		hns3_update_tx_int_coalesce(tqp_vector);
4433 
4434 		hns3_mask_vector_irq(tqp_vector, 1);
4435 	}
4436 
4437 	return rx_pkt_total;
4438 }
4439 
4440 static int hns3_create_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
4441 				  struct hnae3_ring_chain_node **head,
4442 				  bool is_tx)
4443 {
4444 	u32 bit_value = is_tx ? HNAE3_RING_TYPE_TX : HNAE3_RING_TYPE_RX;
4445 	u32 field_value = is_tx ? HNAE3_RING_GL_TX : HNAE3_RING_GL_RX;
4446 	struct hnae3_ring_chain_node *cur_chain = *head;
4447 	struct pci_dev *pdev = tqp_vector->handle->pdev;
4448 	struct hnae3_ring_chain_node *chain;
4449 	struct hns3_enet_ring *ring;
4450 
4451 	ring = is_tx ? tqp_vector->tx_group.ring : tqp_vector->rx_group.ring;
4452 
4453 	if (cur_chain) {
4454 		while (cur_chain->next)
4455 			cur_chain = cur_chain->next;
4456 	}
4457 
4458 	while (ring) {
4459 		chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
4460 		if (!chain)
4461 			return -ENOMEM;
4462 		if (cur_chain)
4463 			cur_chain->next = chain;
4464 		else
4465 			*head = chain;
4466 		chain->tqp_index = ring->tqp->tqp_index;
4467 		hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
4468 				bit_value);
4469 		hnae3_set_field(chain->int_gl_idx,
4470 				HNAE3_RING_GL_IDX_M,
4471 				HNAE3_RING_GL_IDX_S, field_value);
4472 
4473 		cur_chain = chain;
4474 
4475 		ring = ring->next;
4476 	}
4477 
4478 	return 0;
4479 }
4480 
4481 static struct hnae3_ring_chain_node *
4482 hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector)
4483 {
4484 	struct pci_dev *pdev = tqp_vector->handle->pdev;
4485 	struct hnae3_ring_chain_node *cur_chain = NULL;
4486 	struct hnae3_ring_chain_node *chain;
4487 
4488 	if (hns3_create_ring_chain(tqp_vector, &cur_chain, true))
4489 		goto err_free_chain;
4490 
4491 	if (hns3_create_ring_chain(tqp_vector, &cur_chain, false))
4492 		goto err_free_chain;
4493 
4494 	return cur_chain;
4495 
4496 err_free_chain:
4497 	while (cur_chain) {
4498 		chain = cur_chain->next;
4499 		devm_kfree(&pdev->dev, cur_chain);
4500 		cur_chain = chain;
4501 	}
4502 
4503 	return NULL;
4504 }
4505 
4506 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
4507 					struct hnae3_ring_chain_node *head)
4508 {
4509 	struct pci_dev *pdev = tqp_vector->handle->pdev;
4510 	struct hnae3_ring_chain_node *chain_tmp, *chain;
4511 
4512 	chain = head;
4513 
4514 	while (chain) {
4515 		chain_tmp = chain->next;
4516 		devm_kfree(&pdev->dev, chain);
4517 		chain = chain_tmp;
4518 	}
4519 }
4520 
4521 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
4522 				   struct hns3_enet_ring *ring)
4523 {
4524 	ring->next = group->ring;
4525 	group->ring = ring;
4526 
4527 	group->count++;
4528 }
4529 
4530 static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
4531 {
4532 	struct pci_dev *pdev = priv->ae_handle->pdev;
4533 	struct hns3_enet_tqp_vector *tqp_vector;
4534 	int num_vectors = priv->vector_num;
4535 	int numa_node;
4536 	int vector_i;
4537 
4538 	numa_node = dev_to_node(&pdev->dev);
4539 
4540 	for (vector_i = 0; vector_i < num_vectors; vector_i++) {
4541 		tqp_vector = &priv->tqp_vector[vector_i];
4542 		cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
4543 				&tqp_vector->affinity_mask);
4544 	}
4545 }
4546 
4547 static void hns3_rx_dim_work(struct work_struct *work)
4548 {
4549 	struct dim *dim = container_of(work, struct dim, work);
4550 	struct hns3_enet_ring_group *group = container_of(dim,
4551 		struct hns3_enet_ring_group, dim);
4552 	struct hns3_enet_tqp_vector *tqp_vector = group->ring->tqp_vector;
4553 	struct dim_cq_moder cur_moder =
4554 		net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
4555 
4556 	hns3_set_vector_coalesce_rx_gl(group->ring->tqp_vector, cur_moder.usec);
4557 	tqp_vector->rx_group.coal.int_gl = cur_moder.usec;
4558 
4559 	if (cur_moder.pkts < tqp_vector->rx_group.coal.int_ql_max) {
4560 		hns3_set_vector_coalesce_rx_ql(tqp_vector, cur_moder.pkts);
4561 		tqp_vector->rx_group.coal.int_ql = cur_moder.pkts;
4562 	}
4563 
4564 	dim->state = DIM_START_MEASURE;
4565 }
4566 
4567 static void hns3_tx_dim_work(struct work_struct *work)
4568 {
4569 	struct dim *dim = container_of(work, struct dim, work);
4570 	struct hns3_enet_ring_group *group = container_of(dim,
4571 		struct hns3_enet_ring_group, dim);
4572 	struct hns3_enet_tqp_vector *tqp_vector = group->ring->tqp_vector;
4573 	struct dim_cq_moder cur_moder =
4574 		net_dim_get_tx_moderation(dim->mode, dim->profile_ix);
4575 
4576 	hns3_set_vector_coalesce_tx_gl(tqp_vector, cur_moder.usec);
4577 	tqp_vector->tx_group.coal.int_gl = cur_moder.usec;
4578 
4579 	if (cur_moder.pkts < tqp_vector->tx_group.coal.int_ql_max) {
4580 		hns3_set_vector_coalesce_tx_ql(tqp_vector, cur_moder.pkts);
4581 		tqp_vector->tx_group.coal.int_ql = cur_moder.pkts;
4582 	}
4583 
4584 	dim->state = DIM_START_MEASURE;
4585 }
4586 
4587 static void hns3_nic_init_dim(struct hns3_enet_tqp_vector *tqp_vector)
4588 {
4589 	INIT_WORK(&tqp_vector->rx_group.dim.work, hns3_rx_dim_work);
4590 	INIT_WORK(&tqp_vector->tx_group.dim.work, hns3_tx_dim_work);
4591 }
4592 
4593 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
4594 {
4595 	struct hnae3_handle *h = priv->ae_handle;
4596 	struct hns3_enet_tqp_vector *tqp_vector;
4597 	int ret;
4598 	int i;
4599 
4600 	hns3_nic_set_cpumask(priv);
4601 
4602 	for (i = 0; i < priv->vector_num; i++) {
4603 		tqp_vector = &priv->tqp_vector[i];
4604 		hns3_vector_coalesce_init_hw(tqp_vector, priv);
4605 		tqp_vector->num_tqps = 0;
4606 		hns3_nic_init_dim(tqp_vector);
4607 	}
4608 
4609 	for (i = 0; i < h->kinfo.num_tqps; i++) {
4610 		u16 vector_i = i % priv->vector_num;
4611 		u16 tqp_num = h->kinfo.num_tqps;
4612 
4613 		tqp_vector = &priv->tqp_vector[vector_i];
4614 
4615 		hns3_add_ring_to_group(&tqp_vector->tx_group,
4616 				       &priv->ring[i]);
4617 
4618 		hns3_add_ring_to_group(&tqp_vector->rx_group,
4619 				       &priv->ring[i + tqp_num]);
4620 
4621 		priv->ring[i].tqp_vector = tqp_vector;
4622 		priv->ring[i + tqp_num].tqp_vector = tqp_vector;
4623 		tqp_vector->num_tqps++;
4624 	}
4625 
4626 	for (i = 0; i < priv->vector_num; i++) {
4627 		struct hnae3_ring_chain_node *vector_ring_chain;
4628 
4629 		tqp_vector = &priv->tqp_vector[i];
4630 
4631 		tqp_vector->rx_group.total_bytes = 0;
4632 		tqp_vector->rx_group.total_packets = 0;
4633 		tqp_vector->tx_group.total_bytes = 0;
4634 		tqp_vector->tx_group.total_packets = 0;
4635 		tqp_vector->handle = h;
4636 
4637 		vector_ring_chain = hns3_get_vector_ring_chain(tqp_vector);
4638 		if (!vector_ring_chain) {
4639 			ret = -ENOMEM;
4640 			goto map_ring_fail;
4641 		}
4642 
4643 		ret = h->ae_algo->ops->map_ring_to_vector(h,
4644 			tqp_vector->vector_irq, vector_ring_chain);
4645 
4646 		hns3_free_vector_ring_chain(tqp_vector, vector_ring_chain);
4647 
4648 		if (ret)
4649 			goto map_ring_fail;
4650 
4651 		netif_napi_add(priv->netdev, &tqp_vector->napi,
4652 			       hns3_nic_common_poll, NAPI_POLL_WEIGHT);
4653 	}
4654 
4655 	return 0;
4656 
4657 map_ring_fail:
4658 	while (i--)
4659 		netif_napi_del(&priv->tqp_vector[i].napi);
4660 
4661 	return ret;
4662 }
4663 
4664 static void hns3_nic_init_coal_cfg(struct hns3_nic_priv *priv)
4665 {
4666 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
4667 	struct hns3_enet_coalesce *tx_coal = &priv->tx_coal;
4668 	struct hns3_enet_coalesce *rx_coal = &priv->rx_coal;
4669 
4670 	/* initialize the configuration for interrupt coalescing.
4671 	 * 1. GL (Interrupt Gap Limiter)
4672 	 * 2. RL (Interrupt Rate Limiter)
4673 	 * 3. QL (Interrupt Quantity Limiter)
4674 	 *
4675 	 * Default: enable interrupt coalescing self-adaptive and GL
4676 	 */
4677 	tx_coal->adapt_enable = 1;
4678 	rx_coal->adapt_enable = 1;
4679 
4680 	tx_coal->int_gl = HNS3_INT_GL_50K;
4681 	rx_coal->int_gl = HNS3_INT_GL_50K;
4682 
4683 	rx_coal->flow_level = HNS3_FLOW_LOW;
4684 	tx_coal->flow_level = HNS3_FLOW_LOW;
4685 
4686 	if (ae_dev->dev_specs.int_ql_max) {
4687 		tx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
4688 		rx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
4689 	}
4690 }
4691 
4692 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
4693 {
4694 	struct hnae3_handle *h = priv->ae_handle;
4695 	struct hns3_enet_tqp_vector *tqp_vector;
4696 	struct hnae3_vector_info *vector;
4697 	struct pci_dev *pdev = h->pdev;
4698 	u16 tqp_num = h->kinfo.num_tqps;
4699 	u16 vector_num;
4700 	int ret = 0;
4701 	u16 i;
4702 
4703 	/* RSS size, cpu online and vector_num should be the same */
4704 	/* Should consider 2p/4p later */
4705 	vector_num = min_t(u16, num_online_cpus(), tqp_num);
4706 
4707 	vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
4708 			      GFP_KERNEL);
4709 	if (!vector)
4710 		return -ENOMEM;
4711 
4712 	/* save the actual available vector number */
4713 	vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
4714 
4715 	priv->vector_num = vector_num;
4716 	priv->tqp_vector = (struct hns3_enet_tqp_vector *)
4717 		devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
4718 			     GFP_KERNEL);
4719 	if (!priv->tqp_vector) {
4720 		ret = -ENOMEM;
4721 		goto out;
4722 	}
4723 
4724 	for (i = 0; i < priv->vector_num; i++) {
4725 		tqp_vector = &priv->tqp_vector[i];
4726 		tqp_vector->idx = i;
4727 		tqp_vector->mask_addr = vector[i].io_addr;
4728 		tqp_vector->vector_irq = vector[i].vector;
4729 		hns3_vector_coalesce_init(tqp_vector, priv);
4730 	}
4731 
4732 out:
4733 	devm_kfree(&pdev->dev, vector);
4734 	return ret;
4735 }
4736 
4737 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
4738 {
4739 	group->ring = NULL;
4740 	group->count = 0;
4741 }
4742 
4743 static void hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
4744 {
4745 	struct hnae3_ring_chain_node *vector_ring_chain;
4746 	struct hnae3_handle *h = priv->ae_handle;
4747 	struct hns3_enet_tqp_vector *tqp_vector;
4748 	int i;
4749 
4750 	for (i = 0; i < priv->vector_num; i++) {
4751 		tqp_vector = &priv->tqp_vector[i];
4752 
4753 		if (!tqp_vector->rx_group.ring && !tqp_vector->tx_group.ring)
4754 			continue;
4755 
4756 		/* Since the mapping can be overwritten, when fail to get the
4757 		 * chain between vector and ring, we should go on to deal with
4758 		 * the remaining options.
4759 		 */
4760 		vector_ring_chain = hns3_get_vector_ring_chain(tqp_vector);
4761 		if (!vector_ring_chain)
4762 			dev_warn(priv->dev, "failed to get ring chain\n");
4763 
4764 		h->ae_algo->ops->unmap_ring_from_vector(h,
4765 			tqp_vector->vector_irq, vector_ring_chain);
4766 
4767 		hns3_free_vector_ring_chain(tqp_vector, vector_ring_chain);
4768 
4769 		hns3_clear_ring_group(&tqp_vector->rx_group);
4770 		hns3_clear_ring_group(&tqp_vector->tx_group);
4771 		netif_napi_del(&priv->tqp_vector[i].napi);
4772 	}
4773 }
4774 
4775 static void hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
4776 {
4777 	struct hnae3_handle *h = priv->ae_handle;
4778 	struct pci_dev *pdev = h->pdev;
4779 	int i, ret;
4780 
4781 	for (i = 0; i < priv->vector_num; i++) {
4782 		struct hns3_enet_tqp_vector *tqp_vector;
4783 
4784 		tqp_vector = &priv->tqp_vector[i];
4785 		ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
4786 		if (ret)
4787 			return;
4788 	}
4789 
4790 	devm_kfree(&pdev->dev, priv->tqp_vector);
4791 }
4792 
4793 static void hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
4794 			      unsigned int ring_type)
4795 {
4796 	int queue_num = priv->ae_handle->kinfo.num_tqps;
4797 	struct hns3_enet_ring *ring;
4798 	int desc_num;
4799 
4800 	if (ring_type == HNAE3_RING_TYPE_TX) {
4801 		ring = &priv->ring[q->tqp_index];
4802 		desc_num = priv->ae_handle->kinfo.num_tx_desc;
4803 		ring->queue_index = q->tqp_index;
4804 		ring->tx_copybreak = priv->tx_copybreak;
4805 		ring->last_to_use = 0;
4806 	} else {
4807 		ring = &priv->ring[q->tqp_index + queue_num];
4808 		desc_num = priv->ae_handle->kinfo.num_rx_desc;
4809 		ring->queue_index = q->tqp_index;
4810 		ring->rx_copybreak = priv->rx_copybreak;
4811 	}
4812 
4813 	hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
4814 
4815 	ring->tqp = q;
4816 	ring->desc = NULL;
4817 	ring->desc_cb = NULL;
4818 	ring->dev = priv->dev;
4819 	ring->desc_dma_addr = 0;
4820 	ring->buf_size = q->buf_size;
4821 	ring->desc_num = desc_num;
4822 	ring->next_to_use = 0;
4823 	ring->next_to_clean = 0;
4824 }
4825 
4826 static void hns3_queue_to_ring(struct hnae3_queue *tqp,
4827 			       struct hns3_nic_priv *priv)
4828 {
4829 	hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
4830 	hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
4831 }
4832 
4833 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
4834 {
4835 	struct hnae3_handle *h = priv->ae_handle;
4836 	struct pci_dev *pdev = h->pdev;
4837 	int i;
4838 
4839 	priv->ring = devm_kzalloc(&pdev->dev,
4840 				  array3_size(h->kinfo.num_tqps,
4841 					      sizeof(*priv->ring), 2),
4842 				  GFP_KERNEL);
4843 	if (!priv->ring)
4844 		return -ENOMEM;
4845 
4846 	for (i = 0; i < h->kinfo.num_tqps; i++)
4847 		hns3_queue_to_ring(h->kinfo.tqp[i], priv);
4848 
4849 	return 0;
4850 }
4851 
4852 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
4853 {
4854 	if (!priv->ring)
4855 		return;
4856 
4857 	devm_kfree(priv->dev, priv->ring);
4858 	priv->ring = NULL;
4859 }
4860 
4861 static void hns3_alloc_page_pool(struct hns3_enet_ring *ring)
4862 {
4863 	struct page_pool_params pp_params = {
4864 		.flags = PP_FLAG_DMA_MAP | PP_FLAG_PAGE_FRAG |
4865 				PP_FLAG_DMA_SYNC_DEV,
4866 		.order = hns3_page_order(ring),
4867 		.pool_size = ring->desc_num * hns3_buf_size(ring) /
4868 				(PAGE_SIZE << hns3_page_order(ring)),
4869 		.nid = dev_to_node(ring_to_dev(ring)),
4870 		.dev = ring_to_dev(ring),
4871 		.dma_dir = DMA_FROM_DEVICE,
4872 		.offset = 0,
4873 		.max_len = PAGE_SIZE << hns3_page_order(ring),
4874 	};
4875 
4876 	ring->page_pool = page_pool_create(&pp_params);
4877 	if (IS_ERR(ring->page_pool)) {
4878 		dev_warn(ring_to_dev(ring), "page pool creation failed: %ld\n",
4879 			 PTR_ERR(ring->page_pool));
4880 		ring->page_pool = NULL;
4881 	}
4882 }
4883 
4884 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
4885 {
4886 	int ret;
4887 
4888 	if (ring->desc_num <= 0 || ring->buf_size <= 0)
4889 		return -EINVAL;
4890 
4891 	ring->desc_cb = devm_kcalloc(ring_to_dev(ring), ring->desc_num,
4892 				     sizeof(ring->desc_cb[0]), GFP_KERNEL);
4893 	if (!ring->desc_cb) {
4894 		ret = -ENOMEM;
4895 		goto out;
4896 	}
4897 
4898 	ret = hns3_alloc_desc(ring);
4899 	if (ret)
4900 		goto out_with_desc_cb;
4901 
4902 	if (!HNAE3_IS_TX_RING(ring)) {
4903 		if (page_pool_enabled)
4904 			hns3_alloc_page_pool(ring);
4905 
4906 		ret = hns3_alloc_ring_buffers(ring);
4907 		if (ret)
4908 			goto out_with_desc;
4909 	} else {
4910 		hns3_init_tx_spare_buffer(ring);
4911 	}
4912 
4913 	return 0;
4914 
4915 out_with_desc:
4916 	hns3_free_desc(ring);
4917 out_with_desc_cb:
4918 	devm_kfree(ring_to_dev(ring), ring->desc_cb);
4919 	ring->desc_cb = NULL;
4920 out:
4921 	return ret;
4922 }
4923 
4924 void hns3_fini_ring(struct hns3_enet_ring *ring)
4925 {
4926 	hns3_free_desc(ring);
4927 	devm_kfree(ring_to_dev(ring), ring->desc_cb);
4928 	ring->desc_cb = NULL;
4929 	ring->next_to_clean = 0;
4930 	ring->next_to_use = 0;
4931 	ring->last_to_use = 0;
4932 	ring->pending_buf = 0;
4933 	if (!HNAE3_IS_TX_RING(ring) && ring->skb) {
4934 		dev_kfree_skb_any(ring->skb);
4935 		ring->skb = NULL;
4936 	} else if (HNAE3_IS_TX_RING(ring) && ring->tx_spare) {
4937 		struct hns3_tx_spare *tx_spare = ring->tx_spare;
4938 
4939 		dma_unmap_page(ring_to_dev(ring), tx_spare->dma, tx_spare->len,
4940 			       DMA_TO_DEVICE);
4941 		free_pages((unsigned long)tx_spare->buf,
4942 			   get_order(tx_spare->len));
4943 		devm_kfree(ring_to_dev(ring), tx_spare);
4944 		ring->tx_spare = NULL;
4945 	}
4946 
4947 	if (!HNAE3_IS_TX_RING(ring) && ring->page_pool) {
4948 		page_pool_destroy(ring->page_pool);
4949 		ring->page_pool = NULL;
4950 	}
4951 }
4952 
4953 static int hns3_buf_size2type(u32 buf_size)
4954 {
4955 	int bd_size_type;
4956 
4957 	switch (buf_size) {
4958 	case 512:
4959 		bd_size_type = HNS3_BD_SIZE_512_TYPE;
4960 		break;
4961 	case 1024:
4962 		bd_size_type = HNS3_BD_SIZE_1024_TYPE;
4963 		break;
4964 	case 2048:
4965 		bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4966 		break;
4967 	case 4096:
4968 		bd_size_type = HNS3_BD_SIZE_4096_TYPE;
4969 		break;
4970 	default:
4971 		bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4972 	}
4973 
4974 	return bd_size_type;
4975 }
4976 
4977 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
4978 {
4979 	dma_addr_t dma = ring->desc_dma_addr;
4980 	struct hnae3_queue *q = ring->tqp;
4981 
4982 	if (!HNAE3_IS_TX_RING(ring)) {
4983 		hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG, (u32)dma);
4984 		hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
4985 			       (u32)((dma >> 31) >> 1));
4986 
4987 		hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
4988 			       hns3_buf_size2type(ring->buf_size));
4989 		hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
4990 			       ring->desc_num / 8 - 1);
4991 	} else {
4992 		hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
4993 			       (u32)dma);
4994 		hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
4995 			       (u32)((dma >> 31) >> 1));
4996 
4997 		hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
4998 			       ring->desc_num / 8 - 1);
4999 	}
5000 }
5001 
5002 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
5003 {
5004 	struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
5005 	struct hnae3_tc_info *tc_info = &kinfo->tc_info;
5006 	int i;
5007 
5008 	for (i = 0; i < tc_info->num_tc; i++) {
5009 		int j;
5010 
5011 		for (j = 0; j < tc_info->tqp_count[i]; j++) {
5012 			struct hnae3_queue *q;
5013 
5014 			q = priv->ring[tc_info->tqp_offset[i] + j].tqp;
5015 			hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG, i);
5016 		}
5017 	}
5018 }
5019 
5020 int hns3_init_all_ring(struct hns3_nic_priv *priv)
5021 {
5022 	struct hnae3_handle *h = priv->ae_handle;
5023 	int ring_num = h->kinfo.num_tqps * 2;
5024 	int i, j;
5025 	int ret;
5026 
5027 	for (i = 0; i < ring_num; i++) {
5028 		ret = hns3_alloc_ring_memory(&priv->ring[i]);
5029 		if (ret) {
5030 			dev_err(priv->dev,
5031 				"Alloc ring memory fail! ret=%d\n", ret);
5032 			goto out_when_alloc_ring_memory;
5033 		}
5034 
5035 		u64_stats_init(&priv->ring[i].syncp);
5036 	}
5037 
5038 	return 0;
5039 
5040 out_when_alloc_ring_memory:
5041 	for (j = i - 1; j >= 0; j--)
5042 		hns3_fini_ring(&priv->ring[j]);
5043 
5044 	return -ENOMEM;
5045 }
5046 
5047 static void hns3_uninit_all_ring(struct hns3_nic_priv *priv)
5048 {
5049 	struct hnae3_handle *h = priv->ae_handle;
5050 	int i;
5051 
5052 	for (i = 0; i < h->kinfo.num_tqps; i++) {
5053 		hns3_fini_ring(&priv->ring[i]);
5054 		hns3_fini_ring(&priv->ring[i + h->kinfo.num_tqps]);
5055 	}
5056 }
5057 
5058 /* Set mac addr if it is configured. or leave it to the AE driver */
5059 static int hns3_init_mac_addr(struct net_device *netdev)
5060 {
5061 	struct hns3_nic_priv *priv = netdev_priv(netdev);
5062 	char format_mac_addr[HNAE3_FORMAT_MAC_ADDR_LEN];
5063 	struct hnae3_handle *h = priv->ae_handle;
5064 	u8 mac_addr_temp[ETH_ALEN];
5065 	int ret = 0;
5066 
5067 	if (h->ae_algo->ops->get_mac_addr)
5068 		h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
5069 
5070 	/* Check if the MAC address is valid, if not get a random one */
5071 	if (!is_valid_ether_addr(mac_addr_temp)) {
5072 		eth_hw_addr_random(netdev);
5073 		hnae3_format_mac_addr(format_mac_addr, netdev->dev_addr);
5074 		dev_warn(priv->dev, "using random MAC address %s\n",
5075 			 format_mac_addr);
5076 	} else if (!ether_addr_equal(netdev->dev_addr, mac_addr_temp)) {
5077 		eth_hw_addr_set(netdev, mac_addr_temp);
5078 		ether_addr_copy(netdev->perm_addr, mac_addr_temp);
5079 	} else {
5080 		return 0;
5081 	}
5082 
5083 	if (h->ae_algo->ops->set_mac_addr)
5084 		ret = h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
5085 
5086 	return ret;
5087 }
5088 
5089 static int hns3_init_phy(struct net_device *netdev)
5090 {
5091 	struct hnae3_handle *h = hns3_get_handle(netdev);
5092 	int ret = 0;
5093 
5094 	if (h->ae_algo->ops->mac_connect_phy)
5095 		ret = h->ae_algo->ops->mac_connect_phy(h);
5096 
5097 	return ret;
5098 }
5099 
5100 static void hns3_uninit_phy(struct net_device *netdev)
5101 {
5102 	struct hnae3_handle *h = hns3_get_handle(netdev);
5103 
5104 	if (h->ae_algo->ops->mac_disconnect_phy)
5105 		h->ae_algo->ops->mac_disconnect_phy(h);
5106 }
5107 
5108 static int hns3_client_start(struct hnae3_handle *handle)
5109 {
5110 	if (!handle->ae_algo->ops->client_start)
5111 		return 0;
5112 
5113 	return handle->ae_algo->ops->client_start(handle);
5114 }
5115 
5116 static void hns3_client_stop(struct hnae3_handle *handle)
5117 {
5118 	if (!handle->ae_algo->ops->client_stop)
5119 		return;
5120 
5121 	handle->ae_algo->ops->client_stop(handle);
5122 }
5123 
5124 static void hns3_info_show(struct hns3_nic_priv *priv)
5125 {
5126 	struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
5127 	char format_mac_addr[HNAE3_FORMAT_MAC_ADDR_LEN];
5128 
5129 	hnae3_format_mac_addr(format_mac_addr, priv->netdev->dev_addr);
5130 	dev_info(priv->dev, "MAC address: %s\n", format_mac_addr);
5131 	dev_info(priv->dev, "Task queue pairs numbers: %u\n", kinfo->num_tqps);
5132 	dev_info(priv->dev, "RSS size: %u\n", kinfo->rss_size);
5133 	dev_info(priv->dev, "Allocated RSS size: %u\n", kinfo->req_rss_size);
5134 	dev_info(priv->dev, "RX buffer length: %u\n", kinfo->rx_buf_len);
5135 	dev_info(priv->dev, "Desc num per TX queue: %u\n", kinfo->num_tx_desc);
5136 	dev_info(priv->dev, "Desc num per RX queue: %u\n", kinfo->num_rx_desc);
5137 	dev_info(priv->dev, "Total number of enabled TCs: %u\n",
5138 		 kinfo->tc_info.num_tc);
5139 	dev_info(priv->dev, "Max mtu size: %u\n", priv->netdev->max_mtu);
5140 }
5141 
5142 static void hns3_set_cq_period_mode(struct hns3_nic_priv *priv,
5143 				    enum dim_cq_period_mode mode, bool is_tx)
5144 {
5145 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
5146 	struct hnae3_handle *handle = priv->ae_handle;
5147 	int i;
5148 
5149 	if (is_tx) {
5150 		priv->tx_cqe_mode = mode;
5151 
5152 		for (i = 0; i < priv->vector_num; i++)
5153 			priv->tqp_vector[i].tx_group.dim.mode = mode;
5154 	} else {
5155 		priv->rx_cqe_mode = mode;
5156 
5157 		for (i = 0; i < priv->vector_num; i++)
5158 			priv->tqp_vector[i].rx_group.dim.mode = mode;
5159 	}
5160 
5161 	if (hnae3_ae_dev_cq_supported(ae_dev)) {
5162 		u32 new_mode;
5163 		u64 reg;
5164 
5165 		new_mode = (mode == DIM_CQ_PERIOD_MODE_START_FROM_CQE) ?
5166 			HNS3_CQ_MODE_CQE : HNS3_CQ_MODE_EQE;
5167 		reg = is_tx ? HNS3_GL1_CQ_MODE_REG : HNS3_GL0_CQ_MODE_REG;
5168 
5169 		writel(new_mode, handle->kinfo.io_base + reg);
5170 	}
5171 }
5172 
5173 void hns3_cq_period_mode_init(struct hns3_nic_priv *priv,
5174 			      enum dim_cq_period_mode tx_mode,
5175 			      enum dim_cq_period_mode rx_mode)
5176 {
5177 	hns3_set_cq_period_mode(priv, tx_mode, true);
5178 	hns3_set_cq_period_mode(priv, rx_mode, false);
5179 }
5180 
5181 static void hns3_state_init(struct hnae3_handle *handle)
5182 {
5183 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(handle->pdev);
5184 	struct net_device *netdev = handle->kinfo.netdev;
5185 	struct hns3_nic_priv *priv = netdev_priv(netdev);
5186 
5187 	set_bit(HNS3_NIC_STATE_INITED, &priv->state);
5188 
5189 	if (test_bit(HNAE3_DEV_SUPPORT_TX_PUSH_B, ae_dev->caps))
5190 		set_bit(HNS3_NIC_STATE_TX_PUSH_ENABLE, &priv->state);
5191 
5192 	if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
5193 		set_bit(HNAE3_PFLAG_LIMIT_PROMISC, &handle->supported_pflags);
5194 
5195 	if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps))
5196 		set_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state);
5197 
5198 	if (hnae3_ae_dev_rxd_adv_layout_supported(ae_dev))
5199 		set_bit(HNS3_NIC_STATE_RXD_ADV_LAYOUT_ENABLE, &priv->state);
5200 }
5201 
5202 static void hns3_state_uninit(struct hnae3_handle *handle)
5203 {
5204 	struct hns3_nic_priv *priv  = handle->priv;
5205 
5206 	clear_bit(HNS3_NIC_STATE_INITED, &priv->state);
5207 }
5208 
5209 static int hns3_client_init(struct hnae3_handle *handle)
5210 {
5211 	struct pci_dev *pdev = handle->pdev;
5212 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
5213 	u16 alloc_tqps, max_rss_size;
5214 	struct hns3_nic_priv *priv;
5215 	struct net_device *netdev;
5216 	int ret;
5217 
5218 	handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
5219 						    &max_rss_size);
5220 	netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
5221 	if (!netdev)
5222 		return -ENOMEM;
5223 
5224 	priv = netdev_priv(netdev);
5225 	priv->dev = &pdev->dev;
5226 	priv->netdev = netdev;
5227 	priv->ae_handle = handle;
5228 	priv->tx_timeout_count = 0;
5229 	priv->max_non_tso_bd_num = ae_dev->dev_specs.max_non_tso_bd_num;
5230 	set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
5231 
5232 	handle->msg_enable = netif_msg_init(debug, DEFAULT_MSG_LEVEL);
5233 
5234 	handle->kinfo.netdev = netdev;
5235 	handle->priv = (void *)priv;
5236 
5237 	hns3_init_mac_addr(netdev);
5238 
5239 	hns3_set_default_feature(netdev);
5240 
5241 	netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
5242 	netdev->priv_flags |= IFF_UNICAST_FLT;
5243 	netdev->netdev_ops = &hns3_nic_netdev_ops;
5244 	SET_NETDEV_DEV(netdev, &pdev->dev);
5245 	hns3_ethtool_set_ops(netdev);
5246 
5247 	/* Carrier off reporting is important to ethtool even BEFORE open */
5248 	netif_carrier_off(netdev);
5249 
5250 	ret = hns3_get_ring_config(priv);
5251 	if (ret) {
5252 		ret = -ENOMEM;
5253 		goto out_get_ring_cfg;
5254 	}
5255 
5256 	hns3_nic_init_coal_cfg(priv);
5257 
5258 	ret = hns3_nic_alloc_vector_data(priv);
5259 	if (ret) {
5260 		ret = -ENOMEM;
5261 		goto out_alloc_vector_data;
5262 	}
5263 
5264 	ret = hns3_nic_init_vector_data(priv);
5265 	if (ret) {
5266 		ret = -ENOMEM;
5267 		goto out_init_vector_data;
5268 	}
5269 
5270 	ret = hns3_init_all_ring(priv);
5271 	if (ret) {
5272 		ret = -ENOMEM;
5273 		goto out_init_ring;
5274 	}
5275 
5276 	hns3_cq_period_mode_init(priv, DIM_CQ_PERIOD_MODE_START_FROM_EQE,
5277 				 DIM_CQ_PERIOD_MODE_START_FROM_EQE);
5278 
5279 	ret = hns3_init_phy(netdev);
5280 	if (ret)
5281 		goto out_init_phy;
5282 
5283 	/* the device can work without cpu rmap, only aRFS needs it */
5284 	ret = hns3_set_rx_cpu_rmap(netdev);
5285 	if (ret)
5286 		dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
5287 
5288 	ret = hns3_nic_init_irq(priv);
5289 	if (ret) {
5290 		dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
5291 		hns3_free_rx_cpu_rmap(netdev);
5292 		goto out_init_irq_fail;
5293 	}
5294 
5295 	ret = hns3_client_start(handle);
5296 	if (ret) {
5297 		dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
5298 		goto out_client_start;
5299 	}
5300 
5301 	hns3_dcbnl_setup(handle);
5302 
5303 	ret = hns3_dbg_init(handle);
5304 	if (ret) {
5305 		dev_err(priv->dev, "failed to init debugfs, ret = %d\n",
5306 			ret);
5307 		goto out_client_start;
5308 	}
5309 
5310 	netdev->max_mtu = HNS3_MAX_MTU(ae_dev->dev_specs.max_frm_size);
5311 
5312 	hns3_state_init(handle);
5313 
5314 	ret = register_netdev(netdev);
5315 	if (ret) {
5316 		dev_err(priv->dev, "probe register netdev fail!\n");
5317 		goto out_reg_netdev_fail;
5318 	}
5319 
5320 	if (netif_msg_drv(handle))
5321 		hns3_info_show(priv);
5322 
5323 	return ret;
5324 
5325 out_reg_netdev_fail:
5326 	hns3_state_uninit(handle);
5327 	hns3_dbg_uninit(handle);
5328 	hns3_client_stop(handle);
5329 out_client_start:
5330 	hns3_free_rx_cpu_rmap(netdev);
5331 	hns3_nic_uninit_irq(priv);
5332 out_init_irq_fail:
5333 	hns3_uninit_phy(netdev);
5334 out_init_phy:
5335 	hns3_uninit_all_ring(priv);
5336 out_init_ring:
5337 	hns3_nic_uninit_vector_data(priv);
5338 out_init_vector_data:
5339 	hns3_nic_dealloc_vector_data(priv);
5340 out_alloc_vector_data:
5341 	priv->ring = NULL;
5342 out_get_ring_cfg:
5343 	priv->ae_handle = NULL;
5344 	free_netdev(netdev);
5345 	return ret;
5346 }
5347 
5348 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
5349 {
5350 	struct net_device *netdev = handle->kinfo.netdev;
5351 	struct hns3_nic_priv *priv = netdev_priv(netdev);
5352 
5353 	if (netdev->reg_state != NETREG_UNINITIALIZED)
5354 		unregister_netdev(netdev);
5355 
5356 	hns3_client_stop(handle);
5357 
5358 	hns3_uninit_phy(netdev);
5359 
5360 	if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5361 		netdev_warn(netdev, "already uninitialized\n");
5362 		goto out_netdev_free;
5363 	}
5364 
5365 	hns3_free_rx_cpu_rmap(netdev);
5366 
5367 	hns3_nic_uninit_irq(priv);
5368 
5369 	hns3_clear_all_ring(handle, true);
5370 
5371 	hns3_nic_uninit_vector_data(priv);
5372 
5373 	hns3_nic_dealloc_vector_data(priv);
5374 
5375 	hns3_uninit_all_ring(priv);
5376 
5377 	hns3_put_ring_config(priv);
5378 
5379 out_netdev_free:
5380 	hns3_dbg_uninit(handle);
5381 	free_netdev(netdev);
5382 }
5383 
5384 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
5385 {
5386 	struct net_device *netdev = handle->kinfo.netdev;
5387 
5388 	if (!netdev)
5389 		return;
5390 
5391 	if (linkup) {
5392 		netif_tx_wake_all_queues(netdev);
5393 		netif_carrier_on(netdev);
5394 		if (netif_msg_link(handle))
5395 			netdev_info(netdev, "link up\n");
5396 	} else {
5397 		netif_carrier_off(netdev);
5398 		netif_tx_stop_all_queues(netdev);
5399 		if (netif_msg_link(handle))
5400 			netdev_info(netdev, "link down\n");
5401 	}
5402 }
5403 
5404 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
5405 {
5406 	while (ring->next_to_clean != ring->next_to_use) {
5407 		ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
5408 		hns3_free_buffer_detach(ring, ring->next_to_clean, 0);
5409 		ring_ptr_move_fw(ring, next_to_clean);
5410 	}
5411 
5412 	ring->pending_buf = 0;
5413 }
5414 
5415 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
5416 {
5417 	struct hns3_desc_cb res_cbs;
5418 	int ret;
5419 
5420 	while (ring->next_to_use != ring->next_to_clean) {
5421 		/* When a buffer is not reused, it's memory has been
5422 		 * freed in hns3_handle_rx_bd or will be freed by
5423 		 * stack, so we need to replace the buffer here.
5424 		 */
5425 		if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
5426 			ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
5427 			if (ret) {
5428 				hns3_ring_stats_update(ring, sw_err_cnt);
5429 				/* if alloc new buffer fail, exit directly
5430 				 * and reclear in up flow.
5431 				 */
5432 				netdev_warn(ring_to_netdev(ring),
5433 					    "reserve buffer map failed, ret = %d\n",
5434 					    ret);
5435 				return ret;
5436 			}
5437 			hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
5438 		}
5439 		ring_ptr_move_fw(ring, next_to_use);
5440 	}
5441 
5442 	/* Free the pending skb in rx ring */
5443 	if (ring->skb) {
5444 		dev_kfree_skb_any(ring->skb);
5445 		ring->skb = NULL;
5446 		ring->pending_buf = 0;
5447 	}
5448 
5449 	return 0;
5450 }
5451 
5452 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
5453 {
5454 	while (ring->next_to_use != ring->next_to_clean) {
5455 		/* When a buffer is not reused, it's memory has been
5456 		 * freed in hns3_handle_rx_bd or will be freed by
5457 		 * stack, so only need to unmap the buffer here.
5458 		 */
5459 		if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
5460 			hns3_unmap_buffer(ring,
5461 					  &ring->desc_cb[ring->next_to_use]);
5462 			ring->desc_cb[ring->next_to_use].dma = 0;
5463 		}
5464 
5465 		ring_ptr_move_fw(ring, next_to_use);
5466 	}
5467 }
5468 
5469 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force)
5470 {
5471 	struct net_device *ndev = h->kinfo.netdev;
5472 	struct hns3_nic_priv *priv = netdev_priv(ndev);
5473 	u32 i;
5474 
5475 	for (i = 0; i < h->kinfo.num_tqps; i++) {
5476 		struct hns3_enet_ring *ring;
5477 
5478 		ring = &priv->ring[i];
5479 		hns3_clear_tx_ring(ring);
5480 
5481 		ring = &priv->ring[i + h->kinfo.num_tqps];
5482 		/* Continue to clear other rings even if clearing some
5483 		 * rings failed.
5484 		 */
5485 		if (force)
5486 			hns3_force_clear_rx_ring(ring);
5487 		else
5488 			hns3_clear_rx_ring(ring);
5489 	}
5490 }
5491 
5492 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
5493 {
5494 	struct net_device *ndev = h->kinfo.netdev;
5495 	struct hns3_nic_priv *priv = netdev_priv(ndev);
5496 	struct hns3_enet_ring *rx_ring;
5497 	int i, j;
5498 	int ret;
5499 
5500 	ret = h->ae_algo->ops->reset_queue(h);
5501 	if (ret)
5502 		return ret;
5503 
5504 	for (i = 0; i < h->kinfo.num_tqps; i++) {
5505 		hns3_init_ring_hw(&priv->ring[i]);
5506 
5507 		/* We need to clear tx ring here because self test will
5508 		 * use the ring and will not run down before up
5509 		 */
5510 		hns3_clear_tx_ring(&priv->ring[i]);
5511 		priv->ring[i].next_to_clean = 0;
5512 		priv->ring[i].next_to_use = 0;
5513 		priv->ring[i].last_to_use = 0;
5514 
5515 		rx_ring = &priv->ring[i + h->kinfo.num_tqps];
5516 		hns3_init_ring_hw(rx_ring);
5517 		ret = hns3_clear_rx_ring(rx_ring);
5518 		if (ret)
5519 			return ret;
5520 
5521 		/* We can not know the hardware head and tail when this
5522 		 * function is called in reset flow, so we reuse all desc.
5523 		 */
5524 		for (j = 0; j < rx_ring->desc_num; j++)
5525 			hns3_reuse_buffer(rx_ring, j);
5526 
5527 		rx_ring->next_to_clean = 0;
5528 		rx_ring->next_to_use = 0;
5529 	}
5530 
5531 	hns3_init_tx_ring_tc(priv);
5532 
5533 	return 0;
5534 }
5535 
5536 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
5537 {
5538 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
5539 	struct net_device *ndev = kinfo->netdev;
5540 	struct hns3_nic_priv *priv = netdev_priv(ndev);
5541 
5542 	if (test_and_set_bit(HNS3_NIC_STATE_RESETTING, &priv->state))
5543 		return 0;
5544 
5545 	if (!netif_running(ndev))
5546 		return 0;
5547 
5548 	return hns3_nic_net_stop(ndev);
5549 }
5550 
5551 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
5552 {
5553 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
5554 	struct hns3_nic_priv *priv = netdev_priv(kinfo->netdev);
5555 	int ret = 0;
5556 
5557 	if (!test_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5558 		netdev_err(kinfo->netdev, "device is not initialized yet\n");
5559 		return -EFAULT;
5560 	}
5561 
5562 	clear_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
5563 
5564 	if (netif_running(kinfo->netdev)) {
5565 		ret = hns3_nic_net_open(kinfo->netdev);
5566 		if (ret) {
5567 			set_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
5568 			netdev_err(kinfo->netdev,
5569 				   "net up fail, ret=%d!\n", ret);
5570 			return ret;
5571 		}
5572 	}
5573 
5574 	return ret;
5575 }
5576 
5577 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
5578 {
5579 	struct net_device *netdev = handle->kinfo.netdev;
5580 	struct hns3_nic_priv *priv = netdev_priv(netdev);
5581 	int ret;
5582 
5583 	/* Carrier off reporting is important to ethtool even BEFORE open */
5584 	netif_carrier_off(netdev);
5585 
5586 	ret = hns3_get_ring_config(priv);
5587 	if (ret)
5588 		return ret;
5589 
5590 	ret = hns3_nic_alloc_vector_data(priv);
5591 	if (ret)
5592 		goto err_put_ring;
5593 
5594 	ret = hns3_nic_init_vector_data(priv);
5595 	if (ret)
5596 		goto err_dealloc_vector;
5597 
5598 	ret = hns3_init_all_ring(priv);
5599 	if (ret)
5600 		goto err_uninit_vector;
5601 
5602 	hns3_cq_period_mode_init(priv, priv->tx_cqe_mode, priv->rx_cqe_mode);
5603 
5604 	/* the device can work without cpu rmap, only aRFS needs it */
5605 	ret = hns3_set_rx_cpu_rmap(netdev);
5606 	if (ret)
5607 		dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
5608 
5609 	ret = hns3_nic_init_irq(priv);
5610 	if (ret) {
5611 		dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
5612 		hns3_free_rx_cpu_rmap(netdev);
5613 		goto err_init_irq_fail;
5614 	}
5615 
5616 	if (!hns3_is_phys_func(handle->pdev))
5617 		hns3_init_mac_addr(netdev);
5618 
5619 	ret = hns3_client_start(handle);
5620 	if (ret) {
5621 		dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
5622 		goto err_client_start_fail;
5623 	}
5624 
5625 	set_bit(HNS3_NIC_STATE_INITED, &priv->state);
5626 
5627 	return ret;
5628 
5629 err_client_start_fail:
5630 	hns3_free_rx_cpu_rmap(netdev);
5631 	hns3_nic_uninit_irq(priv);
5632 err_init_irq_fail:
5633 	hns3_uninit_all_ring(priv);
5634 err_uninit_vector:
5635 	hns3_nic_uninit_vector_data(priv);
5636 err_dealloc_vector:
5637 	hns3_nic_dealloc_vector_data(priv);
5638 err_put_ring:
5639 	hns3_put_ring_config(priv);
5640 
5641 	return ret;
5642 }
5643 
5644 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
5645 {
5646 	struct net_device *netdev = handle->kinfo.netdev;
5647 	struct hns3_nic_priv *priv = netdev_priv(netdev);
5648 
5649 	if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
5650 		netdev_warn(netdev, "already uninitialized\n");
5651 		return 0;
5652 	}
5653 
5654 	hns3_free_rx_cpu_rmap(netdev);
5655 	hns3_nic_uninit_irq(priv);
5656 	hns3_clear_all_ring(handle, true);
5657 	hns3_reset_tx_queue(priv->ae_handle);
5658 
5659 	hns3_nic_uninit_vector_data(priv);
5660 
5661 	hns3_nic_dealloc_vector_data(priv);
5662 
5663 	hns3_uninit_all_ring(priv);
5664 
5665 	hns3_put_ring_config(priv);
5666 
5667 	return 0;
5668 }
5669 
5670 int hns3_reset_notify(struct hnae3_handle *handle,
5671 		      enum hnae3_reset_notify_type type)
5672 {
5673 	int ret = 0;
5674 
5675 	switch (type) {
5676 	case HNAE3_UP_CLIENT:
5677 		ret = hns3_reset_notify_up_enet(handle);
5678 		break;
5679 	case HNAE3_DOWN_CLIENT:
5680 		ret = hns3_reset_notify_down_enet(handle);
5681 		break;
5682 	case HNAE3_INIT_CLIENT:
5683 		ret = hns3_reset_notify_init_enet(handle);
5684 		break;
5685 	case HNAE3_UNINIT_CLIENT:
5686 		ret = hns3_reset_notify_uninit_enet(handle);
5687 		break;
5688 	default:
5689 		break;
5690 	}
5691 
5692 	return ret;
5693 }
5694 
5695 static int hns3_change_channels(struct hnae3_handle *handle, u32 new_tqp_num,
5696 				bool rxfh_configured)
5697 {
5698 	int ret;
5699 
5700 	ret = handle->ae_algo->ops->set_channels(handle, new_tqp_num,
5701 						 rxfh_configured);
5702 	if (ret) {
5703 		dev_err(&handle->pdev->dev,
5704 			"Change tqp num(%u) fail.\n", new_tqp_num);
5705 		return ret;
5706 	}
5707 
5708 	ret = hns3_reset_notify(handle, HNAE3_INIT_CLIENT);
5709 	if (ret)
5710 		return ret;
5711 
5712 	ret =  hns3_reset_notify(handle, HNAE3_UP_CLIENT);
5713 	if (ret)
5714 		hns3_reset_notify(handle, HNAE3_UNINIT_CLIENT);
5715 
5716 	return ret;
5717 }
5718 
5719 int hns3_set_channels(struct net_device *netdev,
5720 		      struct ethtool_channels *ch)
5721 {
5722 	struct hnae3_handle *h = hns3_get_handle(netdev);
5723 	struct hnae3_knic_private_info *kinfo = &h->kinfo;
5724 	bool rxfh_configured = netif_is_rxfh_configured(netdev);
5725 	u32 new_tqp_num = ch->combined_count;
5726 	u16 org_tqp_num;
5727 	int ret;
5728 
5729 	if (hns3_nic_resetting(netdev))
5730 		return -EBUSY;
5731 
5732 	if (ch->rx_count || ch->tx_count)
5733 		return -EINVAL;
5734 
5735 	if (kinfo->tc_info.mqprio_active) {
5736 		dev_err(&netdev->dev,
5737 			"it's not allowed to set channels via ethtool when MQPRIO mode is on\n");
5738 		return -EINVAL;
5739 	}
5740 
5741 	if (new_tqp_num > hns3_get_max_available_channels(h) ||
5742 	    new_tqp_num < 1) {
5743 		dev_err(&netdev->dev,
5744 			"Change tqps fail, the tqp range is from 1 to %u",
5745 			hns3_get_max_available_channels(h));
5746 		return -EINVAL;
5747 	}
5748 
5749 	if (kinfo->rss_size == new_tqp_num)
5750 		return 0;
5751 
5752 	netif_dbg(h, drv, netdev,
5753 		  "set channels: tqp_num=%u, rxfh=%d\n",
5754 		  new_tqp_num, rxfh_configured);
5755 
5756 	ret = hns3_reset_notify(h, HNAE3_DOWN_CLIENT);
5757 	if (ret)
5758 		return ret;
5759 
5760 	ret = hns3_reset_notify(h, HNAE3_UNINIT_CLIENT);
5761 	if (ret)
5762 		return ret;
5763 
5764 	org_tqp_num = h->kinfo.num_tqps;
5765 	ret = hns3_change_channels(h, new_tqp_num, rxfh_configured);
5766 	if (ret) {
5767 		int ret1;
5768 
5769 		netdev_warn(netdev,
5770 			    "Change channels fail, revert to old value\n");
5771 		ret1 = hns3_change_channels(h, org_tqp_num, rxfh_configured);
5772 		if (ret1) {
5773 			netdev_err(netdev,
5774 				   "revert to old channel fail\n");
5775 			return ret1;
5776 		}
5777 
5778 		return ret;
5779 	}
5780 
5781 	return 0;
5782 }
5783 
5784 static const struct hns3_hw_error_info hns3_hw_err[] = {
5785 	{ .type = HNAE3_PPU_POISON_ERROR,
5786 	  .msg = "PPU poison" },
5787 	{ .type = HNAE3_CMDQ_ECC_ERROR,
5788 	  .msg = "IMP CMDQ error" },
5789 	{ .type = HNAE3_IMP_RD_POISON_ERROR,
5790 	  .msg = "IMP RD poison" },
5791 	{ .type = HNAE3_ROCEE_AXI_RESP_ERROR,
5792 	  .msg = "ROCEE AXI RESP error" },
5793 };
5794 
5795 static void hns3_process_hw_error(struct hnae3_handle *handle,
5796 				  enum hnae3_hw_error_type type)
5797 {
5798 	int i;
5799 
5800 	for (i = 0; i < ARRAY_SIZE(hns3_hw_err); i++) {
5801 		if (hns3_hw_err[i].type == type) {
5802 			dev_err(&handle->pdev->dev, "Detected %s!\n",
5803 				hns3_hw_err[i].msg);
5804 			break;
5805 		}
5806 	}
5807 }
5808 
5809 static const struct hnae3_client_ops client_ops = {
5810 	.init_instance = hns3_client_init,
5811 	.uninit_instance = hns3_client_uninit,
5812 	.link_status_change = hns3_link_status_change,
5813 	.reset_notify = hns3_reset_notify,
5814 	.process_hw_error = hns3_process_hw_error,
5815 };
5816 
5817 /* hns3_init_module - Driver registration routine
5818  * hns3_init_module is the first routine called when the driver is
5819  * loaded. All it does is register with the PCI subsystem.
5820  */
5821 static int __init hns3_init_module(void)
5822 {
5823 	int ret;
5824 
5825 	pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
5826 	pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
5827 
5828 	client.type = HNAE3_CLIENT_KNIC;
5829 	snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH, "%s",
5830 		 hns3_driver_name);
5831 
5832 	client.ops = &client_ops;
5833 
5834 	INIT_LIST_HEAD(&client.node);
5835 
5836 	hns3_dbg_register_debugfs(hns3_driver_name);
5837 
5838 	ret = hnae3_register_client(&client);
5839 	if (ret)
5840 		goto err_reg_client;
5841 
5842 	ret = pci_register_driver(&hns3_driver);
5843 	if (ret)
5844 		goto err_reg_driver;
5845 
5846 	return ret;
5847 
5848 err_reg_driver:
5849 	hnae3_unregister_client(&client);
5850 err_reg_client:
5851 	hns3_dbg_unregister_debugfs();
5852 	return ret;
5853 }
5854 module_init(hns3_init_module);
5855 
5856 /* hns3_exit_module - Driver exit cleanup routine
5857  * hns3_exit_module is called just before the driver is removed
5858  * from memory.
5859  */
5860 static void __exit hns3_exit_module(void)
5861 {
5862 	pci_unregister_driver(&hns3_driver);
5863 	hnae3_unregister_client(&client);
5864 	hns3_dbg_unregister_debugfs();
5865 }
5866 module_exit(hns3_exit_module);
5867 
5868 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
5869 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
5870 MODULE_LICENSE("GPL");
5871 MODULE_ALIAS("pci:hns-nic");
5872