xref: /linux/drivers/net/ethernet/wangxun/libwx/wx_lib.c (revision eed4edda910fe34dfae8c6bfbcf57f4593a54295)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2019 - 2022 Beijing WangXun Technology Co., Ltd. */
3 
4 #include <linux/etherdevice.h>
5 #include <net/ip6_checksum.h>
6 #include <net/page_pool/helpers.h>
7 #include <net/inet_ecn.h>
8 #include <linux/iopoll.h>
9 #include <linux/sctp.h>
10 #include <linux/pci.h>
11 #include <net/tcp.h>
12 #include <net/ip.h>
13 
14 #include "wx_type.h"
15 #include "wx_lib.h"
16 #include "wx_hw.h"
17 
18 /* Lookup table mapping the HW PTYPE to the bit field for decoding */
19 static struct wx_dec_ptype wx_ptype_lookup[256] = {
20 	/* L2: mac */
21 	[0x11] = WX_PTT(L2, NONE, NONE, NONE, NONE, PAY2),
22 	[0x12] = WX_PTT(L2, NONE, NONE, NONE, TS,   PAY2),
23 	[0x13] = WX_PTT(L2, NONE, NONE, NONE, NONE, PAY2),
24 	[0x14] = WX_PTT(L2, NONE, NONE, NONE, NONE, PAY2),
25 	[0x15] = WX_PTT(L2, NONE, NONE, NONE, NONE, NONE),
26 	[0x16] = WX_PTT(L2, NONE, NONE, NONE, NONE, PAY2),
27 	[0x17] = WX_PTT(L2, NONE, NONE, NONE, NONE, NONE),
28 
29 	/* L2: ethertype filter */
30 	[0x18 ... 0x1F] = WX_PTT(L2, NONE, NONE, NONE, NONE, NONE),
31 
32 	/* L3: ip non-tunnel */
33 	[0x21] = WX_PTT(IP, FGV4, NONE, NONE, NONE, PAY3),
34 	[0x22] = WX_PTT(IP, IPV4, NONE, NONE, NONE, PAY3),
35 	[0x23] = WX_PTT(IP, IPV4, NONE, NONE, UDP,  PAY4),
36 	[0x24] = WX_PTT(IP, IPV4, NONE, NONE, TCP,  PAY4),
37 	[0x25] = WX_PTT(IP, IPV4, NONE, NONE, SCTP, PAY4),
38 	[0x29] = WX_PTT(IP, FGV6, NONE, NONE, NONE, PAY3),
39 	[0x2A] = WX_PTT(IP, IPV6, NONE, NONE, NONE, PAY3),
40 	[0x2B] = WX_PTT(IP, IPV6, NONE, NONE, UDP,  PAY3),
41 	[0x2C] = WX_PTT(IP, IPV6, NONE, NONE, TCP,  PAY4),
42 	[0x2D] = WX_PTT(IP, IPV6, NONE, NONE, SCTP, PAY4),
43 
44 	/* L2: fcoe */
45 	[0x30 ... 0x34] = WX_PTT(FCOE, NONE, NONE, NONE, NONE, PAY3),
46 	[0x38 ... 0x3C] = WX_PTT(FCOE, NONE, NONE, NONE, NONE, PAY3),
47 
48 	/* IPv4 --> IPv4/IPv6 */
49 	[0x81] = WX_PTT(IP, IPV4, IPIP, FGV4, NONE, PAY3),
50 	[0x82] = WX_PTT(IP, IPV4, IPIP, IPV4, NONE, PAY3),
51 	[0x83] = WX_PTT(IP, IPV4, IPIP, IPV4, UDP,  PAY4),
52 	[0x84] = WX_PTT(IP, IPV4, IPIP, IPV4, TCP,  PAY4),
53 	[0x85] = WX_PTT(IP, IPV4, IPIP, IPV4, SCTP, PAY4),
54 	[0x89] = WX_PTT(IP, IPV4, IPIP, FGV6, NONE, PAY3),
55 	[0x8A] = WX_PTT(IP, IPV4, IPIP, IPV6, NONE, PAY3),
56 	[0x8B] = WX_PTT(IP, IPV4, IPIP, IPV6, UDP,  PAY4),
57 	[0x8C] = WX_PTT(IP, IPV4, IPIP, IPV6, TCP,  PAY4),
58 	[0x8D] = WX_PTT(IP, IPV4, IPIP, IPV6, SCTP, PAY4),
59 
60 	/* IPv4 --> GRE/NAT --> NONE/IPv4/IPv6 */
61 	[0x90] = WX_PTT(IP, IPV4, IG, NONE, NONE, PAY3),
62 	[0x91] = WX_PTT(IP, IPV4, IG, FGV4, NONE, PAY3),
63 	[0x92] = WX_PTT(IP, IPV4, IG, IPV4, NONE, PAY3),
64 	[0x93] = WX_PTT(IP, IPV4, IG, IPV4, UDP,  PAY4),
65 	[0x94] = WX_PTT(IP, IPV4, IG, IPV4, TCP,  PAY4),
66 	[0x95] = WX_PTT(IP, IPV4, IG, IPV4, SCTP, PAY4),
67 	[0x99] = WX_PTT(IP, IPV4, IG, FGV6, NONE, PAY3),
68 	[0x9A] = WX_PTT(IP, IPV4, IG, IPV6, NONE, PAY3),
69 	[0x9B] = WX_PTT(IP, IPV4, IG, IPV6, UDP,  PAY4),
70 	[0x9C] = WX_PTT(IP, IPV4, IG, IPV6, TCP,  PAY4),
71 	[0x9D] = WX_PTT(IP, IPV4, IG, IPV6, SCTP, PAY4),
72 
73 	/* IPv4 --> GRE/NAT --> MAC --> NONE/IPv4/IPv6 */
74 	[0xA0] = WX_PTT(IP, IPV4, IGM, NONE, NONE, PAY3),
75 	[0xA1] = WX_PTT(IP, IPV4, IGM, FGV4, NONE, PAY3),
76 	[0xA2] = WX_PTT(IP, IPV4, IGM, IPV4, NONE, PAY3),
77 	[0xA3] = WX_PTT(IP, IPV4, IGM, IPV4, UDP,  PAY4),
78 	[0xA4] = WX_PTT(IP, IPV4, IGM, IPV4, TCP,  PAY4),
79 	[0xA5] = WX_PTT(IP, IPV4, IGM, IPV4, SCTP, PAY4),
80 	[0xA9] = WX_PTT(IP, IPV4, IGM, FGV6, NONE, PAY3),
81 	[0xAA] = WX_PTT(IP, IPV4, IGM, IPV6, NONE, PAY3),
82 	[0xAB] = WX_PTT(IP, IPV4, IGM, IPV6, UDP,  PAY4),
83 	[0xAC] = WX_PTT(IP, IPV4, IGM, IPV6, TCP,  PAY4),
84 	[0xAD] = WX_PTT(IP, IPV4, IGM, IPV6, SCTP, PAY4),
85 
86 	/* IPv4 --> GRE/NAT --> MAC+VLAN --> NONE/IPv4/IPv6 */
87 	[0xB0] = WX_PTT(IP, IPV4, IGMV, NONE, NONE, PAY3),
88 	[0xB1] = WX_PTT(IP, IPV4, IGMV, FGV4, NONE, PAY3),
89 	[0xB2] = WX_PTT(IP, IPV4, IGMV, IPV4, NONE, PAY3),
90 	[0xB3] = WX_PTT(IP, IPV4, IGMV, IPV4, UDP,  PAY4),
91 	[0xB4] = WX_PTT(IP, IPV4, IGMV, IPV4, TCP,  PAY4),
92 	[0xB5] = WX_PTT(IP, IPV4, IGMV, IPV4, SCTP, PAY4),
93 	[0xB9] = WX_PTT(IP, IPV4, IGMV, FGV6, NONE, PAY3),
94 	[0xBA] = WX_PTT(IP, IPV4, IGMV, IPV6, NONE, PAY3),
95 	[0xBB] = WX_PTT(IP, IPV4, IGMV, IPV6, UDP,  PAY4),
96 	[0xBC] = WX_PTT(IP, IPV4, IGMV, IPV6, TCP,  PAY4),
97 	[0xBD] = WX_PTT(IP, IPV4, IGMV, IPV6, SCTP, PAY4),
98 
99 	/* IPv6 --> IPv4/IPv6 */
100 	[0xC1] = WX_PTT(IP, IPV6, IPIP, FGV4, NONE, PAY3),
101 	[0xC2] = WX_PTT(IP, IPV6, IPIP, IPV4, NONE, PAY3),
102 	[0xC3] = WX_PTT(IP, IPV6, IPIP, IPV4, UDP,  PAY4),
103 	[0xC4] = WX_PTT(IP, IPV6, IPIP, IPV4, TCP,  PAY4),
104 	[0xC5] = WX_PTT(IP, IPV6, IPIP, IPV4, SCTP, PAY4),
105 	[0xC9] = WX_PTT(IP, IPV6, IPIP, FGV6, NONE, PAY3),
106 	[0xCA] = WX_PTT(IP, IPV6, IPIP, IPV6, NONE, PAY3),
107 	[0xCB] = WX_PTT(IP, IPV6, IPIP, IPV6, UDP,  PAY4),
108 	[0xCC] = WX_PTT(IP, IPV6, IPIP, IPV6, TCP,  PAY4),
109 	[0xCD] = WX_PTT(IP, IPV6, IPIP, IPV6, SCTP, PAY4),
110 
111 	/* IPv6 --> GRE/NAT -> NONE/IPv4/IPv6 */
112 	[0xD0] = WX_PTT(IP, IPV6, IG, NONE, NONE, PAY3),
113 	[0xD1] = WX_PTT(IP, IPV6, IG, FGV4, NONE, PAY3),
114 	[0xD2] = WX_PTT(IP, IPV6, IG, IPV4, NONE, PAY3),
115 	[0xD3] = WX_PTT(IP, IPV6, IG, IPV4, UDP,  PAY4),
116 	[0xD4] = WX_PTT(IP, IPV6, IG, IPV4, TCP,  PAY4),
117 	[0xD5] = WX_PTT(IP, IPV6, IG, IPV4, SCTP, PAY4),
118 	[0xD9] = WX_PTT(IP, IPV6, IG, FGV6, NONE, PAY3),
119 	[0xDA] = WX_PTT(IP, IPV6, IG, IPV6, NONE, PAY3),
120 	[0xDB] = WX_PTT(IP, IPV6, IG, IPV6, UDP,  PAY4),
121 	[0xDC] = WX_PTT(IP, IPV6, IG, IPV6, TCP,  PAY4),
122 	[0xDD] = WX_PTT(IP, IPV6, IG, IPV6, SCTP, PAY4),
123 
124 	/* IPv6 --> GRE/NAT -> MAC -> NONE/IPv4/IPv6 */
125 	[0xE0] = WX_PTT(IP, IPV6, IGM, NONE, NONE, PAY3),
126 	[0xE1] = WX_PTT(IP, IPV6, IGM, FGV4, NONE, PAY3),
127 	[0xE2] = WX_PTT(IP, IPV6, IGM, IPV4, NONE, PAY3),
128 	[0xE3] = WX_PTT(IP, IPV6, IGM, IPV4, UDP,  PAY4),
129 	[0xE4] = WX_PTT(IP, IPV6, IGM, IPV4, TCP,  PAY4),
130 	[0xE5] = WX_PTT(IP, IPV6, IGM, IPV4, SCTP, PAY4),
131 	[0xE9] = WX_PTT(IP, IPV6, IGM, FGV6, NONE, PAY3),
132 	[0xEA] = WX_PTT(IP, IPV6, IGM, IPV6, NONE, PAY3),
133 	[0xEB] = WX_PTT(IP, IPV6, IGM, IPV6, UDP,  PAY4),
134 	[0xEC] = WX_PTT(IP, IPV6, IGM, IPV6, TCP,  PAY4),
135 	[0xED] = WX_PTT(IP, IPV6, IGM, IPV6, SCTP, PAY4),
136 
137 	/* IPv6 --> GRE/NAT -> MAC--> NONE/IPv */
138 	[0xF0] = WX_PTT(IP, IPV6, IGMV, NONE, NONE, PAY3),
139 	[0xF1] = WX_PTT(IP, IPV6, IGMV, FGV4, NONE, PAY3),
140 	[0xF2] = WX_PTT(IP, IPV6, IGMV, IPV4, NONE, PAY3),
141 	[0xF3] = WX_PTT(IP, IPV6, IGMV, IPV4, UDP,  PAY4),
142 	[0xF4] = WX_PTT(IP, IPV6, IGMV, IPV4, TCP,  PAY4),
143 	[0xF5] = WX_PTT(IP, IPV6, IGMV, IPV4, SCTP, PAY4),
144 	[0xF9] = WX_PTT(IP, IPV6, IGMV, FGV6, NONE, PAY3),
145 	[0xFA] = WX_PTT(IP, IPV6, IGMV, IPV6, NONE, PAY3),
146 	[0xFB] = WX_PTT(IP, IPV6, IGMV, IPV6, UDP,  PAY4),
147 	[0xFC] = WX_PTT(IP, IPV6, IGMV, IPV6, TCP,  PAY4),
148 	[0xFD] = WX_PTT(IP, IPV6, IGMV, IPV6, SCTP, PAY4),
149 };
150 
151 static struct wx_dec_ptype wx_decode_ptype(const u8 ptype)
152 {
153 	return wx_ptype_lookup[ptype];
154 }
155 
156 /* wx_test_staterr - tests bits in Rx descriptor status and error fields */
157 static __le32 wx_test_staterr(union wx_rx_desc *rx_desc,
158 			      const u32 stat_err_bits)
159 {
160 	return rx_desc->wb.upper.status_error & cpu_to_le32(stat_err_bits);
161 }
162 
163 static void wx_dma_sync_frag(struct wx_ring *rx_ring,
164 			     struct wx_rx_buffer *rx_buffer)
165 {
166 	struct sk_buff *skb = rx_buffer->skb;
167 	skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
168 
169 	dma_sync_single_range_for_cpu(rx_ring->dev,
170 				      WX_CB(skb)->dma,
171 				      skb_frag_off(frag),
172 				      skb_frag_size(frag),
173 				      DMA_FROM_DEVICE);
174 
175 	/* If the page was released, just unmap it. */
176 	if (unlikely(WX_CB(skb)->page_released))
177 		page_pool_put_full_page(rx_ring->page_pool, rx_buffer->page, false);
178 }
179 
180 static struct wx_rx_buffer *wx_get_rx_buffer(struct wx_ring *rx_ring,
181 					     union wx_rx_desc *rx_desc,
182 					     struct sk_buff **skb,
183 					     int *rx_buffer_pgcnt)
184 {
185 	struct wx_rx_buffer *rx_buffer;
186 	unsigned int size;
187 
188 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
189 	size = le16_to_cpu(rx_desc->wb.upper.length);
190 
191 #if (PAGE_SIZE < 8192)
192 	*rx_buffer_pgcnt = page_count(rx_buffer->page);
193 #else
194 	*rx_buffer_pgcnt = 0;
195 #endif
196 
197 	prefetchw(rx_buffer->page);
198 	*skb = rx_buffer->skb;
199 
200 	/* Delay unmapping of the first packet. It carries the header
201 	 * information, HW may still access the header after the writeback.
202 	 * Only unmap it when EOP is reached
203 	 */
204 	if (!wx_test_staterr(rx_desc, WX_RXD_STAT_EOP)) {
205 		if (!*skb)
206 			goto skip_sync;
207 	} else {
208 		if (*skb)
209 			wx_dma_sync_frag(rx_ring, rx_buffer);
210 	}
211 
212 	/* we are reusing so sync this buffer for CPU use */
213 	dma_sync_single_range_for_cpu(rx_ring->dev,
214 				      rx_buffer->dma,
215 				      rx_buffer->page_offset,
216 				      size,
217 				      DMA_FROM_DEVICE);
218 skip_sync:
219 	return rx_buffer;
220 }
221 
222 static void wx_put_rx_buffer(struct wx_ring *rx_ring,
223 			     struct wx_rx_buffer *rx_buffer,
224 			     struct sk_buff *skb,
225 			     int rx_buffer_pgcnt)
226 {
227 	if (!IS_ERR(skb) && WX_CB(skb)->dma == rx_buffer->dma)
228 		/* the page has been released from the ring */
229 		WX_CB(skb)->page_released = true;
230 
231 	/* clear contents of rx_buffer */
232 	rx_buffer->page = NULL;
233 	rx_buffer->skb = NULL;
234 }
235 
236 static struct sk_buff *wx_build_skb(struct wx_ring *rx_ring,
237 				    struct wx_rx_buffer *rx_buffer,
238 				    union wx_rx_desc *rx_desc)
239 {
240 	unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
241 #if (PAGE_SIZE < 8192)
242 	unsigned int truesize = WX_RX_BUFSZ;
243 #else
244 	unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
245 #endif
246 	struct sk_buff *skb = rx_buffer->skb;
247 
248 	if (!skb) {
249 		void *page_addr = page_address(rx_buffer->page) +
250 				  rx_buffer->page_offset;
251 
252 		/* prefetch first cache line of first page */
253 		prefetch(page_addr);
254 #if L1_CACHE_BYTES < 128
255 		prefetch(page_addr + L1_CACHE_BYTES);
256 #endif
257 
258 		/* allocate a skb to store the frags */
259 		skb = napi_alloc_skb(&rx_ring->q_vector->napi, WX_RXBUFFER_256);
260 		if (unlikely(!skb))
261 			return NULL;
262 
263 		/* we will be copying header into skb->data in
264 		 * pskb_may_pull so it is in our interest to prefetch
265 		 * it now to avoid a possible cache miss
266 		 */
267 		prefetchw(skb->data);
268 
269 		if (size <= WX_RXBUFFER_256) {
270 			memcpy(__skb_put(skb, size), page_addr,
271 			       ALIGN(size, sizeof(long)));
272 			page_pool_put_full_page(rx_ring->page_pool, rx_buffer->page, true);
273 			return skb;
274 		}
275 
276 		skb_mark_for_recycle(skb);
277 
278 		if (!wx_test_staterr(rx_desc, WX_RXD_STAT_EOP))
279 			WX_CB(skb)->dma = rx_buffer->dma;
280 
281 		skb_add_rx_frag(skb, 0, rx_buffer->page,
282 				rx_buffer->page_offset,
283 				size, truesize);
284 		goto out;
285 
286 	} else {
287 		skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
288 				rx_buffer->page_offset, size, truesize);
289 	}
290 
291 out:
292 #if (PAGE_SIZE < 8192)
293 	/* flip page offset to other buffer */
294 	rx_buffer->page_offset ^= truesize;
295 #else
296 	/* move offset up to the next cache line */
297 	rx_buffer->page_offset += truesize;
298 #endif
299 
300 	return skb;
301 }
302 
303 static bool wx_alloc_mapped_page(struct wx_ring *rx_ring,
304 				 struct wx_rx_buffer *bi)
305 {
306 	struct page *page = bi->page;
307 	dma_addr_t dma;
308 
309 	/* since we are recycling buffers we should seldom need to alloc */
310 	if (likely(page))
311 		return true;
312 
313 	page = page_pool_dev_alloc_pages(rx_ring->page_pool);
314 	WARN_ON(!page);
315 	dma = page_pool_get_dma_addr(page);
316 
317 	bi->page_dma = dma;
318 	bi->page = page;
319 	bi->page_offset = 0;
320 
321 	return true;
322 }
323 
324 /**
325  * wx_alloc_rx_buffers - Replace used receive buffers
326  * @rx_ring: ring to place buffers on
327  * @cleaned_count: number of buffers to replace
328  **/
329 void wx_alloc_rx_buffers(struct wx_ring *rx_ring, u16 cleaned_count)
330 {
331 	u16 i = rx_ring->next_to_use;
332 	union wx_rx_desc *rx_desc;
333 	struct wx_rx_buffer *bi;
334 
335 	/* nothing to do */
336 	if (!cleaned_count)
337 		return;
338 
339 	rx_desc = WX_RX_DESC(rx_ring, i);
340 	bi = &rx_ring->rx_buffer_info[i];
341 	i -= rx_ring->count;
342 
343 	do {
344 		if (!wx_alloc_mapped_page(rx_ring, bi))
345 			break;
346 
347 		/* sync the buffer for use by the device */
348 		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
349 						 bi->page_offset,
350 						 WX_RX_BUFSZ,
351 						 DMA_FROM_DEVICE);
352 
353 		rx_desc->read.pkt_addr =
354 			cpu_to_le64(bi->page_dma + bi->page_offset);
355 
356 		rx_desc++;
357 		bi++;
358 		i++;
359 		if (unlikely(!i)) {
360 			rx_desc = WX_RX_DESC(rx_ring, 0);
361 			bi = rx_ring->rx_buffer_info;
362 			i -= rx_ring->count;
363 		}
364 
365 		/* clear the status bits for the next_to_use descriptor */
366 		rx_desc->wb.upper.status_error = 0;
367 
368 		cleaned_count--;
369 	} while (cleaned_count);
370 
371 	i += rx_ring->count;
372 
373 	if (rx_ring->next_to_use != i) {
374 		rx_ring->next_to_use = i;
375 		/* update next to alloc since we have filled the ring */
376 		rx_ring->next_to_alloc = i;
377 
378 		/* Force memory writes to complete before letting h/w
379 		 * know there are new descriptors to fetch.  (Only
380 		 * applicable for weak-ordered memory model archs,
381 		 * such as IA-64).
382 		 */
383 		wmb();
384 		writel(i, rx_ring->tail);
385 	}
386 }
387 
388 u16 wx_desc_unused(struct wx_ring *ring)
389 {
390 	u16 ntc = ring->next_to_clean;
391 	u16 ntu = ring->next_to_use;
392 
393 	return ((ntc > ntu) ? 0 : ring->count) + ntc - ntu - 1;
394 }
395 
396 /**
397  * wx_is_non_eop - process handling of non-EOP buffers
398  * @rx_ring: Rx ring being processed
399  * @rx_desc: Rx descriptor for current buffer
400  * @skb: Current socket buffer containing buffer in progress
401  *
402  * This function updates next to clean. If the buffer is an EOP buffer
403  * this function exits returning false, otherwise it will place the
404  * sk_buff in the next buffer to be chained and return true indicating
405  * that this is in fact a non-EOP buffer.
406  **/
407 static bool wx_is_non_eop(struct wx_ring *rx_ring,
408 			  union wx_rx_desc *rx_desc,
409 			  struct sk_buff *skb)
410 {
411 	u32 ntc = rx_ring->next_to_clean + 1;
412 
413 	/* fetch, update, and store next to clean */
414 	ntc = (ntc < rx_ring->count) ? ntc : 0;
415 	rx_ring->next_to_clean = ntc;
416 
417 	prefetch(WX_RX_DESC(rx_ring, ntc));
418 
419 	/* if we are the last buffer then there is nothing else to do */
420 	if (likely(wx_test_staterr(rx_desc, WX_RXD_STAT_EOP)))
421 		return false;
422 
423 	rx_ring->rx_buffer_info[ntc].skb = skb;
424 	rx_ring->rx_stats.non_eop_descs++;
425 
426 	return true;
427 }
428 
429 static void wx_pull_tail(struct sk_buff *skb)
430 {
431 	skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
432 	unsigned int pull_len;
433 	unsigned char *va;
434 
435 	/* it is valid to use page_address instead of kmap since we are
436 	 * working with pages allocated out of the lomem pool per
437 	 * alloc_page(GFP_ATOMIC)
438 	 */
439 	va = skb_frag_address(frag);
440 
441 	/* we need the header to contain the greater of either ETH_HLEN or
442 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
443 	 */
444 	pull_len = eth_get_headlen(skb->dev, va, WX_RXBUFFER_256);
445 
446 	/* align pull length to size of long to optimize memcpy performance */
447 	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
448 
449 	/* update all of the pointers */
450 	skb_frag_size_sub(frag, pull_len);
451 	skb_frag_off_add(frag, pull_len);
452 	skb->data_len -= pull_len;
453 	skb->tail += pull_len;
454 }
455 
456 /**
457  * wx_cleanup_headers - Correct corrupted or empty headers
458  * @rx_ring: rx descriptor ring packet is being transacted on
459  * @rx_desc: pointer to the EOP Rx descriptor
460  * @skb: pointer to current skb being fixed
461  *
462  * Check for corrupted packet headers caused by senders on the local L2
463  * embedded NIC switch not setting up their Tx Descriptors right.  These
464  * should be very rare.
465  *
466  * Also address the case where we are pulling data in on pages only
467  * and as such no data is present in the skb header.
468  *
469  * In addition if skb is not at least 60 bytes we need to pad it so that
470  * it is large enough to qualify as a valid Ethernet frame.
471  *
472  * Returns true if an error was encountered and skb was freed.
473  **/
474 static bool wx_cleanup_headers(struct wx_ring *rx_ring,
475 			       union wx_rx_desc *rx_desc,
476 			       struct sk_buff *skb)
477 {
478 	struct net_device *netdev = rx_ring->netdev;
479 
480 	/* verify that the packet does not have any known errors */
481 	if (!netdev ||
482 	    unlikely(wx_test_staterr(rx_desc, WX_RXD_ERR_RXE) &&
483 		     !(netdev->features & NETIF_F_RXALL))) {
484 		dev_kfree_skb_any(skb);
485 		return true;
486 	}
487 
488 	/* place header in linear portion of buffer */
489 	if (!skb_headlen(skb))
490 		wx_pull_tail(skb);
491 
492 	/* if eth_skb_pad returns an error the skb was freed */
493 	if (eth_skb_pad(skb))
494 		return true;
495 
496 	return false;
497 }
498 
499 static void wx_rx_hash(struct wx_ring *ring,
500 		       union wx_rx_desc *rx_desc,
501 		       struct sk_buff *skb)
502 {
503 	u16 rss_type;
504 
505 	if (!(ring->netdev->features & NETIF_F_RXHASH))
506 		return;
507 
508 	rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
509 			       WX_RXD_RSSTYPE_MASK;
510 
511 	if (!rss_type)
512 		return;
513 
514 	skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
515 		     (WX_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
516 		     PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
517 }
518 
519 /**
520  * wx_rx_checksum - indicate in skb if hw indicated a good cksum
521  * @ring: structure containing ring specific data
522  * @rx_desc: current Rx descriptor being processed
523  * @skb: skb currently being received and modified
524  **/
525 static void wx_rx_checksum(struct wx_ring *ring,
526 			   union wx_rx_desc *rx_desc,
527 			   struct sk_buff *skb)
528 {
529 	struct wx_dec_ptype dptype = wx_decode_ptype(WX_RXD_PKTTYPE(rx_desc));
530 
531 	skb_checksum_none_assert(skb);
532 	/* Rx csum disabled */
533 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
534 		return;
535 
536 	/* if IPv4 header checksum error */
537 	if ((wx_test_staterr(rx_desc, WX_RXD_STAT_IPCS) &&
538 	     wx_test_staterr(rx_desc, WX_RXD_ERR_IPE)) ||
539 	    (wx_test_staterr(rx_desc, WX_RXD_STAT_OUTERIPCS) &&
540 	     wx_test_staterr(rx_desc, WX_RXD_ERR_OUTERIPER))) {
541 		ring->rx_stats.csum_err++;
542 		return;
543 	}
544 
545 	/* L4 checksum offload flag must set for the below code to work */
546 	if (!wx_test_staterr(rx_desc, WX_RXD_STAT_L4CS))
547 		return;
548 
549 	/* Hardware can't guarantee csum if IPv6 Dest Header found */
550 	if (dptype.prot != WX_DEC_PTYPE_PROT_SCTP && WX_RXD_IPV6EX(rx_desc))
551 		return;
552 
553 	/* if L4 checksum error */
554 	if (wx_test_staterr(rx_desc, WX_RXD_ERR_TCPE)) {
555 		ring->rx_stats.csum_err++;
556 		return;
557 	}
558 
559 	/* It must be a TCP or UDP or SCTP packet with a valid checksum */
560 	skb->ip_summed = CHECKSUM_UNNECESSARY;
561 
562 	/* If there is an outer header present that might contain a checksum
563 	 * we need to bump the checksum level by 1 to reflect the fact that
564 	 * we are indicating we validated the inner checksum.
565 	 */
566 	if (dptype.etype >= WX_DEC_PTYPE_ETYPE_IG)
567 		__skb_incr_checksum_unnecessary(skb);
568 	ring->rx_stats.csum_good_cnt++;
569 }
570 
571 static void wx_rx_vlan(struct wx_ring *ring, union wx_rx_desc *rx_desc,
572 		       struct sk_buff *skb)
573 {
574 	u16 ethertype;
575 	u8 idx = 0;
576 
577 	if ((ring->netdev->features &
578 	     (NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_STAG_RX)) &&
579 	    wx_test_staterr(rx_desc, WX_RXD_STAT_VP)) {
580 		idx = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
581 		       0x1c0) >> 6;
582 		ethertype = ring->q_vector->wx->tpid[idx];
583 		__vlan_hwaccel_put_tag(skb, htons(ethertype),
584 				       le16_to_cpu(rx_desc->wb.upper.vlan));
585 	}
586 }
587 
588 /**
589  * wx_process_skb_fields - Populate skb header fields from Rx descriptor
590  * @rx_ring: rx descriptor ring packet is being transacted on
591  * @rx_desc: pointer to the EOP Rx descriptor
592  * @skb: pointer to current skb being populated
593  *
594  * This function checks the ring, descriptor, and packet information in
595  * order to populate the hash, checksum, protocol, and
596  * other fields within the skb.
597  **/
598 static void wx_process_skb_fields(struct wx_ring *rx_ring,
599 				  union wx_rx_desc *rx_desc,
600 				  struct sk_buff *skb)
601 {
602 	wx_rx_hash(rx_ring, rx_desc, skb);
603 	wx_rx_checksum(rx_ring, rx_desc, skb);
604 	wx_rx_vlan(rx_ring, rx_desc, skb);
605 	skb_record_rx_queue(skb, rx_ring->queue_index);
606 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
607 }
608 
609 /**
610  * wx_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
611  * @q_vector: structure containing interrupt and ring information
612  * @rx_ring: rx descriptor ring to transact packets on
613  * @budget: Total limit on number of packets to process
614  *
615  * This function provides a "bounce buffer" approach to Rx interrupt
616  * processing.  The advantage to this is that on systems that have
617  * expensive overhead for IOMMU access this provides a means of avoiding
618  * it by maintaining the mapping of the page to the system.
619  *
620  * Returns amount of work completed.
621  **/
622 static int wx_clean_rx_irq(struct wx_q_vector *q_vector,
623 			   struct wx_ring *rx_ring,
624 			   int budget)
625 {
626 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
627 	u16 cleaned_count = wx_desc_unused(rx_ring);
628 
629 	do {
630 		struct wx_rx_buffer *rx_buffer;
631 		union wx_rx_desc *rx_desc;
632 		struct sk_buff *skb;
633 		int rx_buffer_pgcnt;
634 
635 		/* return some buffers to hardware, one at a time is too slow */
636 		if (cleaned_count >= WX_RX_BUFFER_WRITE) {
637 			wx_alloc_rx_buffers(rx_ring, cleaned_count);
638 			cleaned_count = 0;
639 		}
640 
641 		rx_desc = WX_RX_DESC(rx_ring, rx_ring->next_to_clean);
642 		if (!wx_test_staterr(rx_desc, WX_RXD_STAT_DD))
643 			break;
644 
645 		/* This memory barrier is needed to keep us from reading
646 		 * any other fields out of the rx_desc until we know the
647 		 * descriptor has been written back
648 		 */
649 		dma_rmb();
650 
651 		rx_buffer = wx_get_rx_buffer(rx_ring, rx_desc, &skb, &rx_buffer_pgcnt);
652 
653 		/* retrieve a buffer from the ring */
654 		skb = wx_build_skb(rx_ring, rx_buffer, rx_desc);
655 
656 		/* exit if we failed to retrieve a buffer */
657 		if (!skb) {
658 			rx_ring->rx_stats.alloc_rx_buff_failed++;
659 			break;
660 		}
661 
662 		wx_put_rx_buffer(rx_ring, rx_buffer, skb, rx_buffer_pgcnt);
663 		cleaned_count++;
664 
665 		/* place incomplete frames back on ring for completion */
666 		if (wx_is_non_eop(rx_ring, rx_desc, skb))
667 			continue;
668 
669 		/* verify the packet layout is correct */
670 		if (wx_cleanup_headers(rx_ring, rx_desc, skb))
671 			continue;
672 
673 		/* probably a little skewed due to removing CRC */
674 		total_rx_bytes += skb->len;
675 
676 		/* populate checksum, timestamp, VLAN, and protocol */
677 		wx_process_skb_fields(rx_ring, rx_desc, skb);
678 		napi_gro_receive(&q_vector->napi, skb);
679 
680 		/* update budget accounting */
681 		total_rx_packets++;
682 	} while (likely(total_rx_packets < budget));
683 
684 	u64_stats_update_begin(&rx_ring->syncp);
685 	rx_ring->stats.packets += total_rx_packets;
686 	rx_ring->stats.bytes += total_rx_bytes;
687 	u64_stats_update_end(&rx_ring->syncp);
688 	q_vector->rx.total_packets += total_rx_packets;
689 	q_vector->rx.total_bytes += total_rx_bytes;
690 
691 	return total_rx_packets;
692 }
693 
694 static struct netdev_queue *wx_txring_txq(const struct wx_ring *ring)
695 {
696 	return netdev_get_tx_queue(ring->netdev, ring->queue_index);
697 }
698 
699 /**
700  * wx_clean_tx_irq - Reclaim resources after transmit completes
701  * @q_vector: structure containing interrupt and ring information
702  * @tx_ring: tx ring to clean
703  * @napi_budget: Used to determine if we are in netpoll
704  **/
705 static bool wx_clean_tx_irq(struct wx_q_vector *q_vector,
706 			    struct wx_ring *tx_ring, int napi_budget)
707 {
708 	unsigned int budget = q_vector->wx->tx_work_limit;
709 	unsigned int total_bytes = 0, total_packets = 0;
710 	unsigned int i = tx_ring->next_to_clean;
711 	struct wx_tx_buffer *tx_buffer;
712 	union wx_tx_desc *tx_desc;
713 
714 	if (!netif_carrier_ok(tx_ring->netdev))
715 		return true;
716 
717 	tx_buffer = &tx_ring->tx_buffer_info[i];
718 	tx_desc = WX_TX_DESC(tx_ring, i);
719 	i -= tx_ring->count;
720 
721 	do {
722 		union wx_tx_desc *eop_desc = tx_buffer->next_to_watch;
723 
724 		/* if next_to_watch is not set then there is no work pending */
725 		if (!eop_desc)
726 			break;
727 
728 		/* prevent any other reads prior to eop_desc */
729 		smp_rmb();
730 
731 		/* if DD is not set pending work has not been completed */
732 		if (!(eop_desc->wb.status & cpu_to_le32(WX_TXD_STAT_DD)))
733 			break;
734 
735 		/* clear next_to_watch to prevent false hangs */
736 		tx_buffer->next_to_watch = NULL;
737 
738 		/* update the statistics for this packet */
739 		total_bytes += tx_buffer->bytecount;
740 		total_packets += tx_buffer->gso_segs;
741 
742 		/* free the skb */
743 		napi_consume_skb(tx_buffer->skb, napi_budget);
744 
745 		/* unmap skb header data */
746 		dma_unmap_single(tx_ring->dev,
747 				 dma_unmap_addr(tx_buffer, dma),
748 				 dma_unmap_len(tx_buffer, len),
749 				 DMA_TO_DEVICE);
750 
751 		/* clear tx_buffer data */
752 		dma_unmap_len_set(tx_buffer, len, 0);
753 
754 		/* unmap remaining buffers */
755 		while (tx_desc != eop_desc) {
756 			tx_buffer++;
757 			tx_desc++;
758 			i++;
759 			if (unlikely(!i)) {
760 				i -= tx_ring->count;
761 				tx_buffer = tx_ring->tx_buffer_info;
762 				tx_desc = WX_TX_DESC(tx_ring, 0);
763 			}
764 
765 			/* unmap any remaining paged data */
766 			if (dma_unmap_len(tx_buffer, len)) {
767 				dma_unmap_page(tx_ring->dev,
768 					       dma_unmap_addr(tx_buffer, dma),
769 					       dma_unmap_len(tx_buffer, len),
770 					       DMA_TO_DEVICE);
771 				dma_unmap_len_set(tx_buffer, len, 0);
772 			}
773 		}
774 
775 		/* move us one more past the eop_desc for start of next pkt */
776 		tx_buffer++;
777 		tx_desc++;
778 		i++;
779 		if (unlikely(!i)) {
780 			i -= tx_ring->count;
781 			tx_buffer = tx_ring->tx_buffer_info;
782 			tx_desc = WX_TX_DESC(tx_ring, 0);
783 		}
784 
785 		/* issue prefetch for next Tx descriptor */
786 		prefetch(tx_desc);
787 
788 		/* update budget accounting */
789 		budget--;
790 	} while (likely(budget));
791 
792 	i += tx_ring->count;
793 	tx_ring->next_to_clean = i;
794 	u64_stats_update_begin(&tx_ring->syncp);
795 	tx_ring->stats.bytes += total_bytes;
796 	tx_ring->stats.packets += total_packets;
797 	u64_stats_update_end(&tx_ring->syncp);
798 	q_vector->tx.total_bytes += total_bytes;
799 	q_vector->tx.total_packets += total_packets;
800 
801 	netdev_tx_completed_queue(wx_txring_txq(tx_ring),
802 				  total_packets, total_bytes);
803 
804 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
805 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
806 		     (wx_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
807 		/* Make sure that anybody stopping the queue after this
808 		 * sees the new next_to_clean.
809 		 */
810 		smp_mb();
811 
812 		if (__netif_subqueue_stopped(tx_ring->netdev,
813 					     tx_ring->queue_index) &&
814 		    netif_running(tx_ring->netdev)) {
815 			netif_wake_subqueue(tx_ring->netdev,
816 					    tx_ring->queue_index);
817 			++tx_ring->tx_stats.restart_queue;
818 		}
819 	}
820 
821 	return !!budget;
822 }
823 
824 /**
825  * wx_poll - NAPI polling RX/TX cleanup routine
826  * @napi: napi struct with our devices info in it
827  * @budget: amount of work driver is allowed to do this pass, in packets
828  *
829  * This function will clean all queues associated with a q_vector.
830  **/
831 static int wx_poll(struct napi_struct *napi, int budget)
832 {
833 	struct wx_q_vector *q_vector = container_of(napi, struct wx_q_vector, napi);
834 	int per_ring_budget, work_done = 0;
835 	struct wx *wx = q_vector->wx;
836 	bool clean_complete = true;
837 	struct wx_ring *ring;
838 
839 	wx_for_each_ring(ring, q_vector->tx) {
840 		if (!wx_clean_tx_irq(q_vector, ring, budget))
841 			clean_complete = false;
842 	}
843 
844 	/* Exit if we are called by netpoll */
845 	if (budget <= 0)
846 		return budget;
847 
848 	/* attempt to distribute budget to each queue fairly, but don't allow
849 	 * the budget to go below 1 because we'll exit polling
850 	 */
851 	if (q_vector->rx.count > 1)
852 		per_ring_budget = max(budget / q_vector->rx.count, 1);
853 	else
854 		per_ring_budget = budget;
855 
856 	wx_for_each_ring(ring, q_vector->rx) {
857 		int cleaned = wx_clean_rx_irq(q_vector, ring, per_ring_budget);
858 
859 		work_done += cleaned;
860 		if (cleaned >= per_ring_budget)
861 			clean_complete = false;
862 	}
863 
864 	/* If all work not completed, return budget and keep polling */
865 	if (!clean_complete)
866 		return budget;
867 
868 	/* all work done, exit the polling mode */
869 	if (likely(napi_complete_done(napi, work_done))) {
870 		if (netif_running(wx->netdev))
871 			wx_intr_enable(wx, WX_INTR_Q(q_vector->v_idx));
872 	}
873 
874 	return min(work_done, budget - 1);
875 }
876 
877 static int wx_maybe_stop_tx(struct wx_ring *tx_ring, u16 size)
878 {
879 	if (likely(wx_desc_unused(tx_ring) >= size))
880 		return 0;
881 
882 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
883 
884 	/* For the next check */
885 	smp_mb();
886 
887 	/* We need to check again in a case another CPU has just
888 	 * made room available.
889 	 */
890 	if (likely(wx_desc_unused(tx_ring) < size))
891 		return -EBUSY;
892 
893 	/* A reprieve! - use start_queue because it doesn't call schedule */
894 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
895 	++tx_ring->tx_stats.restart_queue;
896 
897 	return 0;
898 }
899 
900 static u32 wx_tx_cmd_type(u32 tx_flags)
901 {
902 	/* set type for advanced descriptor with frame checksum insertion */
903 	u32 cmd_type = WX_TXD_DTYP_DATA | WX_TXD_IFCS;
904 
905 	/* set HW vlan bit if vlan is present */
906 	cmd_type |= WX_SET_FLAG(tx_flags, WX_TX_FLAGS_HW_VLAN, WX_TXD_VLE);
907 	/* set segmentation enable bits for TSO/FSO */
908 	cmd_type |= WX_SET_FLAG(tx_flags, WX_TX_FLAGS_TSO, WX_TXD_TSE);
909 	/* set timestamp bit if present */
910 	cmd_type |= WX_SET_FLAG(tx_flags, WX_TX_FLAGS_TSTAMP, WX_TXD_MAC_TSTAMP);
911 	cmd_type |= WX_SET_FLAG(tx_flags, WX_TX_FLAGS_LINKSEC, WX_TXD_LINKSEC);
912 
913 	return cmd_type;
914 }
915 
916 static void wx_tx_olinfo_status(union wx_tx_desc *tx_desc,
917 				u32 tx_flags, unsigned int paylen)
918 {
919 	u32 olinfo_status = paylen << WX_TXD_PAYLEN_SHIFT;
920 
921 	/* enable L4 checksum for TSO and TX checksum offload */
922 	olinfo_status |= WX_SET_FLAG(tx_flags, WX_TX_FLAGS_CSUM, WX_TXD_L4CS);
923 	/* enable IPv4 checksum for TSO */
924 	olinfo_status |= WX_SET_FLAG(tx_flags, WX_TX_FLAGS_IPV4, WX_TXD_IIPCS);
925 	/* enable outer IPv4 checksum for TSO */
926 	olinfo_status |= WX_SET_FLAG(tx_flags, WX_TX_FLAGS_OUTER_IPV4,
927 				     WX_TXD_EIPCS);
928 	/* Check Context must be set if Tx switch is enabled, which it
929 	 * always is for case where virtual functions are running
930 	 */
931 	olinfo_status |= WX_SET_FLAG(tx_flags, WX_TX_FLAGS_CC, WX_TXD_CC);
932 	olinfo_status |= WX_SET_FLAG(tx_flags, WX_TX_FLAGS_IPSEC,
933 				     WX_TXD_IPSEC);
934 	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
935 }
936 
937 static void wx_tx_map(struct wx_ring *tx_ring,
938 		      struct wx_tx_buffer *first,
939 		      const u8 hdr_len)
940 {
941 	struct sk_buff *skb = first->skb;
942 	struct wx_tx_buffer *tx_buffer;
943 	u32 tx_flags = first->tx_flags;
944 	u16 i = tx_ring->next_to_use;
945 	unsigned int data_len, size;
946 	union wx_tx_desc *tx_desc;
947 	skb_frag_t *frag;
948 	dma_addr_t dma;
949 	u32 cmd_type;
950 
951 	cmd_type = wx_tx_cmd_type(tx_flags);
952 	tx_desc = WX_TX_DESC(tx_ring, i);
953 	wx_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
954 
955 	size = skb_headlen(skb);
956 	data_len = skb->data_len;
957 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
958 
959 	tx_buffer = first;
960 
961 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
962 		if (dma_mapping_error(tx_ring->dev, dma))
963 			goto dma_error;
964 
965 		/* record length, and DMA address */
966 		dma_unmap_len_set(tx_buffer, len, size);
967 		dma_unmap_addr_set(tx_buffer, dma, dma);
968 
969 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
970 
971 		while (unlikely(size > WX_MAX_DATA_PER_TXD)) {
972 			tx_desc->read.cmd_type_len =
973 				cpu_to_le32(cmd_type ^ WX_MAX_DATA_PER_TXD);
974 
975 			i++;
976 			tx_desc++;
977 			if (i == tx_ring->count) {
978 				tx_desc = WX_TX_DESC(tx_ring, 0);
979 				i = 0;
980 			}
981 			tx_desc->read.olinfo_status = 0;
982 
983 			dma += WX_MAX_DATA_PER_TXD;
984 			size -= WX_MAX_DATA_PER_TXD;
985 
986 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
987 		}
988 
989 		if (likely(!data_len))
990 			break;
991 
992 		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
993 
994 		i++;
995 		tx_desc++;
996 		if (i == tx_ring->count) {
997 			tx_desc = WX_TX_DESC(tx_ring, 0);
998 			i = 0;
999 		}
1000 		tx_desc->read.olinfo_status = 0;
1001 
1002 		size = skb_frag_size(frag);
1003 
1004 		data_len -= size;
1005 
1006 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
1007 				       DMA_TO_DEVICE);
1008 
1009 		tx_buffer = &tx_ring->tx_buffer_info[i];
1010 	}
1011 
1012 	/* write last descriptor with RS and EOP bits */
1013 	cmd_type |= size | WX_TXD_EOP | WX_TXD_RS;
1014 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
1015 
1016 	netdev_tx_sent_queue(wx_txring_txq(tx_ring), first->bytecount);
1017 
1018 	skb_tx_timestamp(skb);
1019 
1020 	/* Force memory writes to complete before letting h/w know there
1021 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
1022 	 * memory model archs, such as IA-64).
1023 	 *
1024 	 * We also need this memory barrier to make certain all of the
1025 	 * status bits have been updated before next_to_watch is written.
1026 	 */
1027 	wmb();
1028 
1029 	/* set next_to_watch value indicating a packet is present */
1030 	first->next_to_watch = tx_desc;
1031 
1032 	i++;
1033 	if (i == tx_ring->count)
1034 		i = 0;
1035 
1036 	tx_ring->next_to_use = i;
1037 
1038 	wx_maybe_stop_tx(tx_ring, DESC_NEEDED);
1039 
1040 	if (netif_xmit_stopped(wx_txring_txq(tx_ring)) || !netdev_xmit_more())
1041 		writel(i, tx_ring->tail);
1042 
1043 	return;
1044 dma_error:
1045 	dev_err(tx_ring->dev, "TX DMA map failed\n");
1046 
1047 	/* clear dma mappings for failed tx_buffer_info map */
1048 	for (;;) {
1049 		tx_buffer = &tx_ring->tx_buffer_info[i];
1050 		if (dma_unmap_len(tx_buffer, len))
1051 			dma_unmap_page(tx_ring->dev,
1052 				       dma_unmap_addr(tx_buffer, dma),
1053 				       dma_unmap_len(tx_buffer, len),
1054 				       DMA_TO_DEVICE);
1055 		dma_unmap_len_set(tx_buffer, len, 0);
1056 		if (tx_buffer == first)
1057 			break;
1058 		if (i == 0)
1059 			i += tx_ring->count;
1060 		i--;
1061 	}
1062 
1063 	dev_kfree_skb_any(first->skb);
1064 	first->skb = NULL;
1065 
1066 	tx_ring->next_to_use = i;
1067 }
1068 
1069 static void wx_tx_ctxtdesc(struct wx_ring *tx_ring, u32 vlan_macip_lens,
1070 			   u32 fcoe_sof_eof, u32 type_tucmd, u32 mss_l4len_idx)
1071 {
1072 	struct wx_tx_context_desc *context_desc;
1073 	u16 i = tx_ring->next_to_use;
1074 
1075 	context_desc = WX_TX_CTXTDESC(tx_ring, i);
1076 	i++;
1077 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1078 
1079 	/* set bits to identify this as an advanced context descriptor */
1080 	type_tucmd |= WX_TXD_DTYP_CTXT;
1081 	context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
1082 	context_desc->seqnum_seed       = cpu_to_le32(fcoe_sof_eof);
1083 	context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
1084 	context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
1085 }
1086 
1087 static void wx_get_ipv6_proto(struct sk_buff *skb, int offset, u8 *nexthdr)
1088 {
1089 	struct ipv6hdr *hdr = (struct ipv6hdr *)(skb->data + offset);
1090 
1091 	*nexthdr = hdr->nexthdr;
1092 	offset += sizeof(struct ipv6hdr);
1093 	while (ipv6_ext_hdr(*nexthdr)) {
1094 		struct ipv6_opt_hdr _hdr, *hp;
1095 
1096 		if (*nexthdr == NEXTHDR_NONE)
1097 			return;
1098 		hp = skb_header_pointer(skb, offset, sizeof(_hdr), &_hdr);
1099 		if (!hp)
1100 			return;
1101 		if (*nexthdr == NEXTHDR_FRAGMENT)
1102 			break;
1103 		*nexthdr = hp->nexthdr;
1104 	}
1105 }
1106 
1107 union network_header {
1108 	struct iphdr *ipv4;
1109 	struct ipv6hdr *ipv6;
1110 	void *raw;
1111 };
1112 
1113 static u8 wx_encode_tx_desc_ptype(const struct wx_tx_buffer *first)
1114 {
1115 	u8 tun_prot = 0, l4_prot = 0, ptype = 0;
1116 	struct sk_buff *skb = first->skb;
1117 
1118 	if (skb->encapsulation) {
1119 		union network_header hdr;
1120 
1121 		switch (first->protocol) {
1122 		case htons(ETH_P_IP):
1123 			tun_prot = ip_hdr(skb)->protocol;
1124 			ptype = WX_PTYPE_TUN_IPV4;
1125 			break;
1126 		case htons(ETH_P_IPV6):
1127 			wx_get_ipv6_proto(skb, skb_network_offset(skb), &tun_prot);
1128 			ptype = WX_PTYPE_TUN_IPV6;
1129 			break;
1130 		default:
1131 			return ptype;
1132 		}
1133 
1134 		if (tun_prot == IPPROTO_IPIP) {
1135 			hdr.raw = (void *)inner_ip_hdr(skb);
1136 			ptype |= WX_PTYPE_PKT_IPIP;
1137 		} else if (tun_prot == IPPROTO_UDP) {
1138 			hdr.raw = (void *)inner_ip_hdr(skb);
1139 			if (skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
1140 			    skb->inner_protocol != htons(ETH_P_TEB)) {
1141 				ptype |= WX_PTYPE_PKT_IG;
1142 			} else {
1143 				if (((struct ethhdr *)skb_inner_mac_header(skb))->h_proto
1144 				     == htons(ETH_P_8021Q))
1145 					ptype |= WX_PTYPE_PKT_IGMV;
1146 				else
1147 					ptype |= WX_PTYPE_PKT_IGM;
1148 			}
1149 
1150 		} else if (tun_prot == IPPROTO_GRE) {
1151 			hdr.raw = (void *)inner_ip_hdr(skb);
1152 			if (skb->inner_protocol ==  htons(ETH_P_IP) ||
1153 			    skb->inner_protocol ==  htons(ETH_P_IPV6)) {
1154 				ptype |= WX_PTYPE_PKT_IG;
1155 			} else {
1156 				if (((struct ethhdr *)skb_inner_mac_header(skb))->h_proto
1157 				    == htons(ETH_P_8021Q))
1158 					ptype |= WX_PTYPE_PKT_IGMV;
1159 				else
1160 					ptype |= WX_PTYPE_PKT_IGM;
1161 			}
1162 		} else {
1163 			return ptype;
1164 		}
1165 
1166 		switch (hdr.ipv4->version) {
1167 		case IPVERSION:
1168 			l4_prot = hdr.ipv4->protocol;
1169 			break;
1170 		case 6:
1171 			wx_get_ipv6_proto(skb, skb_inner_network_offset(skb), &l4_prot);
1172 			ptype |= WX_PTYPE_PKT_IPV6;
1173 			break;
1174 		default:
1175 			return ptype;
1176 		}
1177 	} else {
1178 		switch (first->protocol) {
1179 		case htons(ETH_P_IP):
1180 			l4_prot = ip_hdr(skb)->protocol;
1181 			ptype = WX_PTYPE_PKT_IP;
1182 			break;
1183 		case htons(ETH_P_IPV6):
1184 			wx_get_ipv6_proto(skb, skb_network_offset(skb), &l4_prot);
1185 			ptype = WX_PTYPE_PKT_IP | WX_PTYPE_PKT_IPV6;
1186 			break;
1187 		default:
1188 			return WX_PTYPE_PKT_MAC | WX_PTYPE_TYP_MAC;
1189 		}
1190 	}
1191 	switch (l4_prot) {
1192 	case IPPROTO_TCP:
1193 		ptype |= WX_PTYPE_TYP_TCP;
1194 		break;
1195 	case IPPROTO_UDP:
1196 		ptype |= WX_PTYPE_TYP_UDP;
1197 		break;
1198 	case IPPROTO_SCTP:
1199 		ptype |= WX_PTYPE_TYP_SCTP;
1200 		break;
1201 	default:
1202 		ptype |= WX_PTYPE_TYP_IP;
1203 		break;
1204 	}
1205 
1206 	return ptype;
1207 }
1208 
1209 static int wx_tso(struct wx_ring *tx_ring, struct wx_tx_buffer *first,
1210 		  u8 *hdr_len, u8 ptype)
1211 {
1212 	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
1213 	struct net_device *netdev = tx_ring->netdev;
1214 	u32 l4len, tunhdr_eiplen_tunlen = 0;
1215 	struct sk_buff *skb = first->skb;
1216 	bool enc = skb->encapsulation;
1217 	struct ipv6hdr *ipv6h;
1218 	struct tcphdr *tcph;
1219 	struct iphdr *iph;
1220 	u8 tun_prot = 0;
1221 	int err;
1222 
1223 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1224 		return 0;
1225 
1226 	if (!skb_is_gso(skb))
1227 		return 0;
1228 
1229 	err = skb_cow_head(skb, 0);
1230 	if (err < 0)
1231 		return err;
1232 
1233 	/* indicates the inner headers in the skbuff are valid. */
1234 	iph = enc ? inner_ip_hdr(skb) : ip_hdr(skb);
1235 	if (iph->version == 4) {
1236 		tcph = enc ? inner_tcp_hdr(skb) : tcp_hdr(skb);
1237 		iph->tot_len = 0;
1238 		iph->check = 0;
1239 		tcph->check = ~csum_tcpudp_magic(iph->saddr,
1240 						 iph->daddr, 0,
1241 						 IPPROTO_TCP, 0);
1242 		first->tx_flags |= WX_TX_FLAGS_TSO |
1243 				   WX_TX_FLAGS_CSUM |
1244 				   WX_TX_FLAGS_IPV4 |
1245 				   WX_TX_FLAGS_CC;
1246 	} else if (iph->version == 6 && skb_is_gso_v6(skb)) {
1247 		ipv6h = enc ? inner_ipv6_hdr(skb) : ipv6_hdr(skb);
1248 		tcph = enc ? inner_tcp_hdr(skb) : tcp_hdr(skb);
1249 		ipv6h->payload_len = 0;
1250 		tcph->check = ~csum_ipv6_magic(&ipv6h->saddr,
1251 					       &ipv6h->daddr, 0,
1252 					       IPPROTO_TCP, 0);
1253 		first->tx_flags |= WX_TX_FLAGS_TSO |
1254 				   WX_TX_FLAGS_CSUM |
1255 				   WX_TX_FLAGS_CC;
1256 	}
1257 
1258 	/* compute header lengths */
1259 	l4len = enc ? inner_tcp_hdrlen(skb) : tcp_hdrlen(skb);
1260 	*hdr_len = enc ? (skb_inner_transport_header(skb) - skb->data) :
1261 			 skb_transport_offset(skb);
1262 	*hdr_len += l4len;
1263 
1264 	/* update gso size and bytecount with header size */
1265 	first->gso_segs = skb_shinfo(skb)->gso_segs;
1266 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
1267 
1268 	/* mss_l4len_id: use 0 as index for TSO */
1269 	mss_l4len_idx = l4len << WX_TXD_L4LEN_SHIFT;
1270 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << WX_TXD_MSS_SHIFT;
1271 
1272 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
1273 	if (enc) {
1274 		switch (first->protocol) {
1275 		case htons(ETH_P_IP):
1276 			tun_prot = ip_hdr(skb)->protocol;
1277 			first->tx_flags |= WX_TX_FLAGS_OUTER_IPV4;
1278 			break;
1279 		case htons(ETH_P_IPV6):
1280 			tun_prot = ipv6_hdr(skb)->nexthdr;
1281 			break;
1282 		default:
1283 			break;
1284 		}
1285 		switch (tun_prot) {
1286 		case IPPROTO_UDP:
1287 			tunhdr_eiplen_tunlen = WX_TXD_TUNNEL_UDP;
1288 			tunhdr_eiplen_tunlen |= ((skb_network_header_len(skb) >> 2) <<
1289 						 WX_TXD_OUTER_IPLEN_SHIFT) |
1290 						(((skb_inner_mac_header(skb) -
1291 						skb_transport_header(skb)) >> 1) <<
1292 						WX_TXD_TUNNEL_LEN_SHIFT);
1293 			break;
1294 		case IPPROTO_GRE:
1295 			tunhdr_eiplen_tunlen = WX_TXD_TUNNEL_GRE;
1296 			tunhdr_eiplen_tunlen |= ((skb_network_header_len(skb) >> 2) <<
1297 						 WX_TXD_OUTER_IPLEN_SHIFT) |
1298 						(((skb_inner_mac_header(skb) -
1299 						skb_transport_header(skb)) >> 1) <<
1300 						WX_TXD_TUNNEL_LEN_SHIFT);
1301 			break;
1302 		case IPPROTO_IPIP:
1303 			tunhdr_eiplen_tunlen = (((char *)inner_ip_hdr(skb) -
1304 						(char *)ip_hdr(skb)) >> 2) <<
1305 						WX_TXD_OUTER_IPLEN_SHIFT;
1306 			break;
1307 		default:
1308 			break;
1309 		}
1310 		vlan_macip_lens = skb_inner_network_header_len(skb) >> 1;
1311 	} else {
1312 		vlan_macip_lens = skb_network_header_len(skb) >> 1;
1313 	}
1314 
1315 	vlan_macip_lens |= skb_network_offset(skb) << WX_TXD_MACLEN_SHIFT;
1316 	vlan_macip_lens |= first->tx_flags & WX_TX_FLAGS_VLAN_MASK;
1317 
1318 	type_tucmd = ptype << 24;
1319 	if (skb->vlan_proto == htons(ETH_P_8021AD) &&
1320 	    netdev->features & NETIF_F_HW_VLAN_STAG_TX)
1321 		type_tucmd |= WX_SET_FLAG(first->tx_flags,
1322 					  WX_TX_FLAGS_HW_VLAN,
1323 					  0x1 << WX_TXD_TAG_TPID_SEL_SHIFT);
1324 	wx_tx_ctxtdesc(tx_ring, vlan_macip_lens, tunhdr_eiplen_tunlen,
1325 		       type_tucmd, mss_l4len_idx);
1326 
1327 	return 1;
1328 }
1329 
1330 static void wx_tx_csum(struct wx_ring *tx_ring, struct wx_tx_buffer *first,
1331 		       u8 ptype)
1332 {
1333 	u32 tunhdr_eiplen_tunlen = 0, vlan_macip_lens = 0;
1334 	struct net_device *netdev = tx_ring->netdev;
1335 	u32 mss_l4len_idx = 0, type_tucmd;
1336 	struct sk_buff *skb = first->skb;
1337 	u8 tun_prot = 0;
1338 
1339 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
1340 		if (!(first->tx_flags & WX_TX_FLAGS_HW_VLAN) &&
1341 		    !(first->tx_flags & WX_TX_FLAGS_CC))
1342 			return;
1343 		vlan_macip_lens = skb_network_offset(skb) <<
1344 				  WX_TXD_MACLEN_SHIFT;
1345 	} else {
1346 		u8 l4_prot = 0;
1347 		union {
1348 			struct iphdr *ipv4;
1349 			struct ipv6hdr *ipv6;
1350 			u8 *raw;
1351 		} network_hdr;
1352 		union {
1353 			struct tcphdr *tcphdr;
1354 			u8 *raw;
1355 		} transport_hdr;
1356 
1357 		if (skb->encapsulation) {
1358 			network_hdr.raw = skb_inner_network_header(skb);
1359 			transport_hdr.raw = skb_inner_transport_header(skb);
1360 			vlan_macip_lens = skb_network_offset(skb) <<
1361 					  WX_TXD_MACLEN_SHIFT;
1362 			switch (first->protocol) {
1363 			case htons(ETH_P_IP):
1364 				tun_prot = ip_hdr(skb)->protocol;
1365 				break;
1366 			case htons(ETH_P_IPV6):
1367 				tun_prot = ipv6_hdr(skb)->nexthdr;
1368 				break;
1369 			default:
1370 				return;
1371 			}
1372 			switch (tun_prot) {
1373 			case IPPROTO_UDP:
1374 				tunhdr_eiplen_tunlen = WX_TXD_TUNNEL_UDP;
1375 				tunhdr_eiplen_tunlen |=
1376 					((skb_network_header_len(skb) >> 2) <<
1377 					WX_TXD_OUTER_IPLEN_SHIFT) |
1378 					(((skb_inner_mac_header(skb) -
1379 					skb_transport_header(skb)) >> 1) <<
1380 					WX_TXD_TUNNEL_LEN_SHIFT);
1381 				break;
1382 			case IPPROTO_GRE:
1383 				tunhdr_eiplen_tunlen = WX_TXD_TUNNEL_GRE;
1384 				tunhdr_eiplen_tunlen |= ((skb_network_header_len(skb) >> 2) <<
1385 							 WX_TXD_OUTER_IPLEN_SHIFT) |
1386 							 (((skb_inner_mac_header(skb) -
1387 							    skb_transport_header(skb)) >> 1) <<
1388 							  WX_TXD_TUNNEL_LEN_SHIFT);
1389 				break;
1390 			case IPPROTO_IPIP:
1391 				tunhdr_eiplen_tunlen = (((char *)inner_ip_hdr(skb) -
1392 							(char *)ip_hdr(skb)) >> 2) <<
1393 							WX_TXD_OUTER_IPLEN_SHIFT;
1394 				break;
1395 			default:
1396 				break;
1397 			}
1398 
1399 		} else {
1400 			network_hdr.raw = skb_network_header(skb);
1401 			transport_hdr.raw = skb_transport_header(skb);
1402 			vlan_macip_lens = skb_network_offset(skb) <<
1403 					  WX_TXD_MACLEN_SHIFT;
1404 		}
1405 
1406 		switch (network_hdr.ipv4->version) {
1407 		case IPVERSION:
1408 			vlan_macip_lens |= (transport_hdr.raw - network_hdr.raw) >> 1;
1409 			l4_prot = network_hdr.ipv4->protocol;
1410 			break;
1411 		case 6:
1412 			vlan_macip_lens |= (transport_hdr.raw - network_hdr.raw) >> 1;
1413 			l4_prot = network_hdr.ipv6->nexthdr;
1414 			break;
1415 		default:
1416 			break;
1417 		}
1418 
1419 		switch (l4_prot) {
1420 		case IPPROTO_TCP:
1421 		mss_l4len_idx = (transport_hdr.tcphdr->doff * 4) <<
1422 				WX_TXD_L4LEN_SHIFT;
1423 			break;
1424 		case IPPROTO_SCTP:
1425 			mss_l4len_idx = sizeof(struct sctphdr) <<
1426 					WX_TXD_L4LEN_SHIFT;
1427 			break;
1428 		case IPPROTO_UDP:
1429 			mss_l4len_idx = sizeof(struct udphdr) <<
1430 					WX_TXD_L4LEN_SHIFT;
1431 			break;
1432 		default:
1433 			break;
1434 		}
1435 
1436 		/* update TX checksum flag */
1437 		first->tx_flags |= WX_TX_FLAGS_CSUM;
1438 	}
1439 	first->tx_flags |= WX_TX_FLAGS_CC;
1440 	/* vlan_macip_lens: MACLEN, VLAN tag */
1441 	vlan_macip_lens |= first->tx_flags & WX_TX_FLAGS_VLAN_MASK;
1442 
1443 	type_tucmd = ptype << 24;
1444 	if (skb->vlan_proto == htons(ETH_P_8021AD) &&
1445 	    netdev->features & NETIF_F_HW_VLAN_STAG_TX)
1446 		type_tucmd |= WX_SET_FLAG(first->tx_flags,
1447 					  WX_TX_FLAGS_HW_VLAN,
1448 					  0x1 << WX_TXD_TAG_TPID_SEL_SHIFT);
1449 	wx_tx_ctxtdesc(tx_ring, vlan_macip_lens, tunhdr_eiplen_tunlen,
1450 		       type_tucmd, mss_l4len_idx);
1451 }
1452 
1453 static netdev_tx_t wx_xmit_frame_ring(struct sk_buff *skb,
1454 				      struct wx_ring *tx_ring)
1455 {
1456 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
1457 	struct wx_tx_buffer *first;
1458 	u8 hdr_len = 0, ptype;
1459 	unsigned short f;
1460 	u32 tx_flags = 0;
1461 	int tso;
1462 
1463 	/* need: 1 descriptor per page * PAGE_SIZE/WX_MAX_DATA_PER_TXD,
1464 	 *       + 1 desc for skb_headlen/WX_MAX_DATA_PER_TXD,
1465 	 *       + 2 desc gap to keep tail from touching head,
1466 	 *       + 1 desc for context descriptor,
1467 	 * otherwise try next time
1468 	 */
1469 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1470 		count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->
1471 						     frags[f]));
1472 
1473 	if (wx_maybe_stop_tx(tx_ring, count + 3)) {
1474 		tx_ring->tx_stats.tx_busy++;
1475 		return NETDEV_TX_BUSY;
1476 	}
1477 
1478 	/* record the location of the first descriptor for this packet */
1479 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
1480 	first->skb = skb;
1481 	first->bytecount = skb->len;
1482 	first->gso_segs = 1;
1483 
1484 	/* if we have a HW VLAN tag being added default to the HW one */
1485 	if (skb_vlan_tag_present(skb)) {
1486 		tx_flags |= skb_vlan_tag_get(skb) << WX_TX_FLAGS_VLAN_SHIFT;
1487 		tx_flags |= WX_TX_FLAGS_HW_VLAN;
1488 	}
1489 
1490 	/* record initial flags and protocol */
1491 	first->tx_flags = tx_flags;
1492 	first->protocol = vlan_get_protocol(skb);
1493 
1494 	ptype = wx_encode_tx_desc_ptype(first);
1495 
1496 	tso = wx_tso(tx_ring, first, &hdr_len, ptype);
1497 	if (tso < 0)
1498 		goto out_drop;
1499 	else if (!tso)
1500 		wx_tx_csum(tx_ring, first, ptype);
1501 	wx_tx_map(tx_ring, first, hdr_len);
1502 
1503 	return NETDEV_TX_OK;
1504 out_drop:
1505 	dev_kfree_skb_any(first->skb);
1506 	first->skb = NULL;
1507 
1508 	return NETDEV_TX_OK;
1509 }
1510 
1511 netdev_tx_t wx_xmit_frame(struct sk_buff *skb,
1512 			  struct net_device *netdev)
1513 {
1514 	unsigned int r_idx = skb->queue_mapping;
1515 	struct wx *wx = netdev_priv(netdev);
1516 	struct wx_ring *tx_ring;
1517 
1518 	if (!netif_carrier_ok(netdev)) {
1519 		dev_kfree_skb_any(skb);
1520 		return NETDEV_TX_OK;
1521 	}
1522 
1523 	/* The minimum packet size for olinfo paylen is 17 so pad the skb
1524 	 * in order to meet this minimum size requirement.
1525 	 */
1526 	if (skb_put_padto(skb, 17))
1527 		return NETDEV_TX_OK;
1528 
1529 	if (r_idx >= wx->num_tx_queues)
1530 		r_idx = r_idx % wx->num_tx_queues;
1531 	tx_ring = wx->tx_ring[r_idx];
1532 
1533 	return wx_xmit_frame_ring(skb, tx_ring);
1534 }
1535 EXPORT_SYMBOL(wx_xmit_frame);
1536 
1537 void wx_napi_enable_all(struct wx *wx)
1538 {
1539 	struct wx_q_vector *q_vector;
1540 	int q_idx;
1541 
1542 	for (q_idx = 0; q_idx < wx->num_q_vectors; q_idx++) {
1543 		q_vector = wx->q_vector[q_idx];
1544 		napi_enable(&q_vector->napi);
1545 	}
1546 }
1547 EXPORT_SYMBOL(wx_napi_enable_all);
1548 
1549 void wx_napi_disable_all(struct wx *wx)
1550 {
1551 	struct wx_q_vector *q_vector;
1552 	int q_idx;
1553 
1554 	for (q_idx = 0; q_idx < wx->num_q_vectors; q_idx++) {
1555 		q_vector = wx->q_vector[q_idx];
1556 		napi_disable(&q_vector->napi);
1557 	}
1558 }
1559 EXPORT_SYMBOL(wx_napi_disable_all);
1560 
1561 /**
1562  * wx_set_rss_queues: Allocate queues for RSS
1563  * @wx: board private structure to initialize
1564  *
1565  * This is our "base" multiqueue mode.  RSS (Receive Side Scaling) will try
1566  * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
1567  *
1568  **/
1569 static void wx_set_rss_queues(struct wx *wx)
1570 {
1571 	struct wx_ring_feature *f;
1572 
1573 	/* set mask for 16 queue limit of RSS */
1574 	f = &wx->ring_feature[RING_F_RSS];
1575 	f->indices = f->limit;
1576 
1577 	wx->num_rx_queues = f->limit;
1578 	wx->num_tx_queues = f->limit;
1579 }
1580 
1581 static void wx_set_num_queues(struct wx *wx)
1582 {
1583 	/* Start with base case */
1584 	wx->num_rx_queues = 1;
1585 	wx->num_tx_queues = 1;
1586 	wx->queues_per_pool = 1;
1587 
1588 	wx_set_rss_queues(wx);
1589 }
1590 
1591 /**
1592  * wx_acquire_msix_vectors - acquire MSI-X vectors
1593  * @wx: board private structure
1594  *
1595  * Attempts to acquire a suitable range of MSI-X vector interrupts. Will
1596  * return a negative error code if unable to acquire MSI-X vectors for any
1597  * reason.
1598  */
1599 static int wx_acquire_msix_vectors(struct wx *wx)
1600 {
1601 	struct irq_affinity affd = {0, };
1602 	int nvecs, i;
1603 
1604 	/* We start by asking for one vector per queue pair */
1605 	nvecs = max(wx->num_rx_queues, wx->num_tx_queues);
1606 	nvecs = min_t(int, nvecs, num_online_cpus());
1607 	nvecs = min_t(int, nvecs, wx->mac.max_msix_vectors);
1608 
1609 	wx->msix_q_entries = kcalloc(nvecs, sizeof(struct msix_entry),
1610 				     GFP_KERNEL);
1611 	if (!wx->msix_q_entries)
1612 		return -ENOMEM;
1613 
1614 	/* One for non-queue interrupts */
1615 	nvecs += 1;
1616 
1617 	if (!wx->msix_in_use) {
1618 		wx->msix_entry = kcalloc(1, sizeof(struct msix_entry),
1619 					 GFP_KERNEL);
1620 		if (!wx->msix_entry) {
1621 			kfree(wx->msix_q_entries);
1622 			wx->msix_q_entries = NULL;
1623 			return -ENOMEM;
1624 		}
1625 	}
1626 
1627 	nvecs = pci_alloc_irq_vectors_affinity(wx->pdev, nvecs,
1628 					       nvecs,
1629 					       PCI_IRQ_MSIX | PCI_IRQ_AFFINITY,
1630 					       &affd);
1631 	if (nvecs < 0) {
1632 		wx_err(wx, "Failed to allocate MSI-X interrupts. Err: %d\n", nvecs);
1633 		kfree(wx->msix_q_entries);
1634 		wx->msix_q_entries = NULL;
1635 		kfree(wx->msix_entry);
1636 		wx->msix_entry = NULL;
1637 		return nvecs;
1638 	}
1639 
1640 	wx->msix_entry->entry = 0;
1641 	wx->msix_entry->vector = pci_irq_vector(wx->pdev, 0);
1642 	nvecs -= 1;
1643 	for (i = 0; i < nvecs; i++) {
1644 		wx->msix_q_entries[i].entry = i;
1645 		wx->msix_q_entries[i].vector = pci_irq_vector(wx->pdev, i + 1);
1646 	}
1647 
1648 	wx->num_q_vectors = nvecs;
1649 
1650 	return 0;
1651 }
1652 
1653 /**
1654  * wx_set_interrupt_capability - set MSI-X or MSI if supported
1655  * @wx: board private structure to initialize
1656  *
1657  * Attempt to configure the interrupts using the best available
1658  * capabilities of the hardware and the kernel.
1659  **/
1660 static int wx_set_interrupt_capability(struct wx *wx)
1661 {
1662 	struct pci_dev *pdev = wx->pdev;
1663 	int nvecs, ret;
1664 
1665 	/* We will try to get MSI-X interrupts first */
1666 	ret = wx_acquire_msix_vectors(wx);
1667 	if (ret == 0 || (ret == -ENOMEM))
1668 		return ret;
1669 
1670 	/* Disable RSS */
1671 	dev_warn(&wx->pdev->dev, "Disabling RSS support\n");
1672 	wx->ring_feature[RING_F_RSS].limit = 1;
1673 
1674 	wx_set_num_queues(wx);
1675 
1676 	/* minmum one for queue, one for misc*/
1677 	nvecs = 1;
1678 	nvecs = pci_alloc_irq_vectors(pdev, nvecs,
1679 				      nvecs, PCI_IRQ_MSI | PCI_IRQ_LEGACY);
1680 	if (nvecs == 1) {
1681 		if (pdev->msi_enabled)
1682 			wx_err(wx, "Fallback to MSI.\n");
1683 		else
1684 			wx_err(wx, "Fallback to LEGACY.\n");
1685 	} else {
1686 		wx_err(wx, "Failed to allocate MSI/LEGACY interrupts. Error: %d\n", nvecs);
1687 		return nvecs;
1688 	}
1689 
1690 	pdev->irq = pci_irq_vector(pdev, 0);
1691 
1692 	return 0;
1693 }
1694 
1695 /**
1696  * wx_cache_ring_rss - Descriptor ring to register mapping for RSS
1697  * @wx: board private structure to initialize
1698  *
1699  * Cache the descriptor ring offsets for RSS, ATR, FCoE, and SR-IOV.
1700  *
1701  **/
1702 static void wx_cache_ring_rss(struct wx *wx)
1703 {
1704 	u16 i;
1705 
1706 	for (i = 0; i < wx->num_rx_queues; i++)
1707 		wx->rx_ring[i]->reg_idx = i;
1708 
1709 	for (i = 0; i < wx->num_tx_queues; i++)
1710 		wx->tx_ring[i]->reg_idx = i;
1711 }
1712 
1713 static void wx_add_ring(struct wx_ring *ring, struct wx_ring_container *head)
1714 {
1715 	ring->next = head->ring;
1716 	head->ring = ring;
1717 	head->count++;
1718 }
1719 
1720 /**
1721  * wx_alloc_q_vector - Allocate memory for a single interrupt vector
1722  * @wx: board private structure to initialize
1723  * @v_count: q_vectors allocated on wx, used for ring interleaving
1724  * @v_idx: index of vector in wx struct
1725  * @txr_count: total number of Tx rings to allocate
1726  * @txr_idx: index of first Tx ring to allocate
1727  * @rxr_count: total number of Rx rings to allocate
1728  * @rxr_idx: index of first Rx ring to allocate
1729  *
1730  * We allocate one q_vector.  If allocation fails we return -ENOMEM.
1731  **/
1732 static int wx_alloc_q_vector(struct wx *wx,
1733 			     unsigned int v_count, unsigned int v_idx,
1734 			     unsigned int txr_count, unsigned int txr_idx,
1735 			     unsigned int rxr_count, unsigned int rxr_idx)
1736 {
1737 	struct wx_q_vector *q_vector;
1738 	int ring_count, default_itr;
1739 	struct wx_ring *ring;
1740 
1741 	/* note this will allocate space for the ring structure as well! */
1742 	ring_count = txr_count + rxr_count;
1743 
1744 	q_vector = kzalloc(struct_size(q_vector, ring, ring_count),
1745 			   GFP_KERNEL);
1746 	if (!q_vector)
1747 		return -ENOMEM;
1748 
1749 	/* initialize NAPI */
1750 	netif_napi_add(wx->netdev, &q_vector->napi,
1751 		       wx_poll);
1752 
1753 	/* tie q_vector and wx together */
1754 	wx->q_vector[v_idx] = q_vector;
1755 	q_vector->wx = wx;
1756 	q_vector->v_idx = v_idx;
1757 	if (cpu_online(v_idx))
1758 		q_vector->numa_node = cpu_to_node(v_idx);
1759 
1760 	/* initialize pointer to rings */
1761 	ring = q_vector->ring;
1762 
1763 	if (wx->mac.type == wx_mac_sp)
1764 		default_itr = WX_12K_ITR;
1765 	else
1766 		default_itr = WX_7K_ITR;
1767 	/* initialize ITR */
1768 	if (txr_count && !rxr_count)
1769 		/* tx only vector */
1770 		q_vector->itr = wx->tx_itr_setting ?
1771 				default_itr : wx->tx_itr_setting;
1772 	else
1773 		/* rx or rx/tx vector */
1774 		q_vector->itr = wx->rx_itr_setting ?
1775 				default_itr : wx->rx_itr_setting;
1776 
1777 	while (txr_count) {
1778 		/* assign generic ring traits */
1779 		ring->dev = &wx->pdev->dev;
1780 		ring->netdev = wx->netdev;
1781 
1782 		/* configure backlink on ring */
1783 		ring->q_vector = q_vector;
1784 
1785 		/* update q_vector Tx values */
1786 		wx_add_ring(ring, &q_vector->tx);
1787 
1788 		/* apply Tx specific ring traits */
1789 		ring->count = wx->tx_ring_count;
1790 
1791 		ring->queue_index = txr_idx;
1792 
1793 		/* assign ring to wx */
1794 		wx->tx_ring[txr_idx] = ring;
1795 
1796 		/* update count and index */
1797 		txr_count--;
1798 		txr_idx += v_count;
1799 
1800 		/* push pointer to next ring */
1801 		ring++;
1802 	}
1803 
1804 	while (rxr_count) {
1805 		/* assign generic ring traits */
1806 		ring->dev = &wx->pdev->dev;
1807 		ring->netdev = wx->netdev;
1808 
1809 		/* configure backlink on ring */
1810 		ring->q_vector = q_vector;
1811 
1812 		/* update q_vector Rx values */
1813 		wx_add_ring(ring, &q_vector->rx);
1814 
1815 		/* apply Rx specific ring traits */
1816 		ring->count = wx->rx_ring_count;
1817 		ring->queue_index = rxr_idx;
1818 
1819 		/* assign ring to wx */
1820 		wx->rx_ring[rxr_idx] = ring;
1821 
1822 		/* update count and index */
1823 		rxr_count--;
1824 		rxr_idx += v_count;
1825 
1826 		/* push pointer to next ring */
1827 		ring++;
1828 	}
1829 
1830 	return 0;
1831 }
1832 
1833 /**
1834  * wx_free_q_vector - Free memory allocated for specific interrupt vector
1835  * @wx: board private structure to initialize
1836  * @v_idx: Index of vector to be freed
1837  *
1838  * This function frees the memory allocated to the q_vector.  In addition if
1839  * NAPI is enabled it will delete any references to the NAPI struct prior
1840  * to freeing the q_vector.
1841  **/
1842 static void wx_free_q_vector(struct wx *wx, int v_idx)
1843 {
1844 	struct wx_q_vector *q_vector = wx->q_vector[v_idx];
1845 	struct wx_ring *ring;
1846 
1847 	wx_for_each_ring(ring, q_vector->tx)
1848 		wx->tx_ring[ring->queue_index] = NULL;
1849 
1850 	wx_for_each_ring(ring, q_vector->rx)
1851 		wx->rx_ring[ring->queue_index] = NULL;
1852 
1853 	wx->q_vector[v_idx] = NULL;
1854 	netif_napi_del(&q_vector->napi);
1855 	kfree_rcu(q_vector, rcu);
1856 }
1857 
1858 /**
1859  * wx_alloc_q_vectors - Allocate memory for interrupt vectors
1860  * @wx: board private structure to initialize
1861  *
1862  * We allocate one q_vector per queue interrupt.  If allocation fails we
1863  * return -ENOMEM.
1864  **/
1865 static int wx_alloc_q_vectors(struct wx *wx)
1866 {
1867 	unsigned int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1868 	unsigned int rxr_remaining = wx->num_rx_queues;
1869 	unsigned int txr_remaining = wx->num_tx_queues;
1870 	unsigned int q_vectors = wx->num_q_vectors;
1871 	int rqpv, tqpv;
1872 	int err;
1873 
1874 	for (; v_idx < q_vectors; v_idx++) {
1875 		rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1876 		tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1877 		err = wx_alloc_q_vector(wx, q_vectors, v_idx,
1878 					tqpv, txr_idx,
1879 					rqpv, rxr_idx);
1880 
1881 		if (err)
1882 			goto err_out;
1883 
1884 		/* update counts and index */
1885 		rxr_remaining -= rqpv;
1886 		txr_remaining -= tqpv;
1887 		rxr_idx++;
1888 		txr_idx++;
1889 	}
1890 
1891 	return 0;
1892 
1893 err_out:
1894 	wx->num_tx_queues = 0;
1895 	wx->num_rx_queues = 0;
1896 	wx->num_q_vectors = 0;
1897 
1898 	while (v_idx--)
1899 		wx_free_q_vector(wx, v_idx);
1900 
1901 	return -ENOMEM;
1902 }
1903 
1904 /**
1905  * wx_free_q_vectors - Free memory allocated for interrupt vectors
1906  * @wx: board private structure to initialize
1907  *
1908  * This function frees the memory allocated to the q_vectors.  In addition if
1909  * NAPI is enabled it will delete any references to the NAPI struct prior
1910  * to freeing the q_vector.
1911  **/
1912 static void wx_free_q_vectors(struct wx *wx)
1913 {
1914 	int v_idx = wx->num_q_vectors;
1915 
1916 	wx->num_tx_queues = 0;
1917 	wx->num_rx_queues = 0;
1918 	wx->num_q_vectors = 0;
1919 
1920 	while (v_idx--)
1921 		wx_free_q_vector(wx, v_idx);
1922 }
1923 
1924 void wx_reset_interrupt_capability(struct wx *wx)
1925 {
1926 	struct pci_dev *pdev = wx->pdev;
1927 
1928 	if (!pdev->msi_enabled && !pdev->msix_enabled)
1929 		return;
1930 
1931 	if (pdev->msix_enabled) {
1932 		kfree(wx->msix_q_entries);
1933 		wx->msix_q_entries = NULL;
1934 		if (!wx->msix_in_use) {
1935 			kfree(wx->msix_entry);
1936 			wx->msix_entry = NULL;
1937 		}
1938 	}
1939 	pci_free_irq_vectors(wx->pdev);
1940 }
1941 EXPORT_SYMBOL(wx_reset_interrupt_capability);
1942 
1943 /**
1944  * wx_clear_interrupt_scheme - Clear the current interrupt scheme settings
1945  * @wx: board private structure to clear interrupt scheme on
1946  *
1947  * We go through and clear interrupt specific resources and reset the structure
1948  * to pre-load conditions
1949  **/
1950 void wx_clear_interrupt_scheme(struct wx *wx)
1951 {
1952 	wx_free_q_vectors(wx);
1953 	wx_reset_interrupt_capability(wx);
1954 }
1955 EXPORT_SYMBOL(wx_clear_interrupt_scheme);
1956 
1957 int wx_init_interrupt_scheme(struct wx *wx)
1958 {
1959 	int ret;
1960 
1961 	/* Number of supported queues */
1962 	wx_set_num_queues(wx);
1963 
1964 	/* Set interrupt mode */
1965 	ret = wx_set_interrupt_capability(wx);
1966 	if (ret) {
1967 		wx_err(wx, "Allocate irq vectors for failed.\n");
1968 		return ret;
1969 	}
1970 
1971 	/* Allocate memory for queues */
1972 	ret = wx_alloc_q_vectors(wx);
1973 	if (ret) {
1974 		wx_err(wx, "Unable to allocate memory for queue vectors.\n");
1975 		wx_reset_interrupt_capability(wx);
1976 		return ret;
1977 	}
1978 
1979 	wx_cache_ring_rss(wx);
1980 
1981 	return 0;
1982 }
1983 EXPORT_SYMBOL(wx_init_interrupt_scheme);
1984 
1985 irqreturn_t wx_msix_clean_rings(int __always_unused irq, void *data)
1986 {
1987 	struct wx_q_vector *q_vector = data;
1988 
1989 	/* EIAM disabled interrupts (on this vector) for us */
1990 	if (q_vector->rx.ring || q_vector->tx.ring)
1991 		napi_schedule_irqoff(&q_vector->napi);
1992 
1993 	return IRQ_HANDLED;
1994 }
1995 EXPORT_SYMBOL(wx_msix_clean_rings);
1996 
1997 void wx_free_irq(struct wx *wx)
1998 {
1999 	struct pci_dev *pdev = wx->pdev;
2000 	int vector;
2001 
2002 	if (!(pdev->msix_enabled)) {
2003 		free_irq(pdev->irq, wx);
2004 		return;
2005 	}
2006 
2007 	for (vector = 0; vector < wx->num_q_vectors; vector++) {
2008 		struct wx_q_vector *q_vector = wx->q_vector[vector];
2009 		struct msix_entry *entry = &wx->msix_q_entries[vector];
2010 
2011 		/* free only the irqs that were actually requested */
2012 		if (!q_vector->rx.ring && !q_vector->tx.ring)
2013 			continue;
2014 
2015 		free_irq(entry->vector, q_vector);
2016 	}
2017 
2018 	if (wx->mac.type == wx_mac_em)
2019 		free_irq(wx->msix_entry->vector, wx);
2020 }
2021 EXPORT_SYMBOL(wx_free_irq);
2022 
2023 /**
2024  * wx_setup_isb_resources - allocate interrupt status resources
2025  * @wx: board private structure
2026  *
2027  * Return 0 on success, negative on failure
2028  **/
2029 int wx_setup_isb_resources(struct wx *wx)
2030 {
2031 	struct pci_dev *pdev = wx->pdev;
2032 
2033 	wx->isb_mem = dma_alloc_coherent(&pdev->dev,
2034 					 sizeof(u32) * 4,
2035 					 &wx->isb_dma,
2036 					 GFP_KERNEL);
2037 	if (!wx->isb_mem) {
2038 		wx_err(wx, "Alloc isb_mem failed\n");
2039 		return -ENOMEM;
2040 	}
2041 
2042 	return 0;
2043 }
2044 EXPORT_SYMBOL(wx_setup_isb_resources);
2045 
2046 /**
2047  * wx_free_isb_resources - allocate all queues Rx resources
2048  * @wx: board private structure
2049  *
2050  * Return 0 on success, negative on failure
2051  **/
2052 void wx_free_isb_resources(struct wx *wx)
2053 {
2054 	struct pci_dev *pdev = wx->pdev;
2055 
2056 	dma_free_coherent(&pdev->dev, sizeof(u32) * 4,
2057 			  wx->isb_mem, wx->isb_dma);
2058 	wx->isb_mem = NULL;
2059 }
2060 EXPORT_SYMBOL(wx_free_isb_resources);
2061 
2062 u32 wx_misc_isb(struct wx *wx, enum wx_isb_idx idx)
2063 {
2064 	u32 cur_tag = 0;
2065 
2066 	cur_tag = wx->isb_mem[WX_ISB_HEADER];
2067 	wx->isb_tag[idx] = cur_tag;
2068 
2069 	return (__force u32)cpu_to_le32(wx->isb_mem[idx]);
2070 }
2071 EXPORT_SYMBOL(wx_misc_isb);
2072 
2073 /**
2074  * wx_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
2075  * @wx: pointer to wx struct
2076  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
2077  * @queue: queue to map the corresponding interrupt to
2078  * @msix_vector: the vector to map to the corresponding queue
2079  *
2080  **/
2081 static void wx_set_ivar(struct wx *wx, s8 direction,
2082 			u16 queue, u16 msix_vector)
2083 {
2084 	u32 ivar, index;
2085 
2086 	if (direction == -1) {
2087 		/* other causes */
2088 		msix_vector |= WX_PX_IVAR_ALLOC_VAL;
2089 		index = 0;
2090 		ivar = rd32(wx, WX_PX_MISC_IVAR);
2091 		ivar &= ~(0xFF << index);
2092 		ivar |= (msix_vector << index);
2093 		wr32(wx, WX_PX_MISC_IVAR, ivar);
2094 	} else {
2095 		/* tx or rx causes */
2096 		msix_vector += 1; /* offset for queue vectors */
2097 		msix_vector |= WX_PX_IVAR_ALLOC_VAL;
2098 		index = ((16 * (queue & 1)) + (8 * direction));
2099 		ivar = rd32(wx, WX_PX_IVAR(queue >> 1));
2100 		ivar &= ~(0xFF << index);
2101 		ivar |= (msix_vector << index);
2102 		wr32(wx, WX_PX_IVAR(queue >> 1), ivar);
2103 	}
2104 }
2105 
2106 /**
2107  * wx_write_eitr - write EITR register in hardware specific way
2108  * @q_vector: structure containing interrupt and ring information
2109  *
2110  * This function is made to be called by ethtool and by the driver
2111  * when it needs to update EITR registers at runtime.  Hardware
2112  * specific quirks/differences are taken care of here.
2113  */
2114 void wx_write_eitr(struct wx_q_vector *q_vector)
2115 {
2116 	struct wx *wx = q_vector->wx;
2117 	int v_idx = q_vector->v_idx;
2118 	u32 itr_reg;
2119 
2120 	if (wx->mac.type == wx_mac_sp)
2121 		itr_reg = q_vector->itr & WX_SP_MAX_EITR;
2122 	else
2123 		itr_reg = q_vector->itr & WX_EM_MAX_EITR;
2124 
2125 	itr_reg |= WX_PX_ITR_CNT_WDIS;
2126 
2127 	wr32(wx, WX_PX_ITR(v_idx + 1), itr_reg);
2128 }
2129 
2130 /**
2131  * wx_configure_vectors - Configure vectors for hardware
2132  * @wx: board private structure
2133  *
2134  * wx_configure_vectors sets up the hardware to properly generate MSI-X/MSI/LEGACY
2135  * interrupts.
2136  **/
2137 void wx_configure_vectors(struct wx *wx)
2138 {
2139 	struct pci_dev *pdev = wx->pdev;
2140 	u32 eitrsel = 0;
2141 	u16 v_idx;
2142 
2143 	if (pdev->msix_enabled) {
2144 		/* Populate MSIX to EITR Select */
2145 		wr32(wx, WX_PX_ITRSEL, eitrsel);
2146 		/* use EIAM to auto-mask when MSI-X interrupt is asserted
2147 		 * this saves a register write for every interrupt
2148 		 */
2149 		wr32(wx, WX_PX_GPIE, WX_PX_GPIE_MODEL);
2150 	} else {
2151 		/* legacy interrupts, use EIAM to auto-mask when reading EICR,
2152 		 * specifically only auto mask tx and rx interrupts.
2153 		 */
2154 		wr32(wx, WX_PX_GPIE, 0);
2155 	}
2156 
2157 	/* Populate the IVAR table and set the ITR values to the
2158 	 * corresponding register.
2159 	 */
2160 	for (v_idx = 0; v_idx < wx->num_q_vectors; v_idx++) {
2161 		struct wx_q_vector *q_vector = wx->q_vector[v_idx];
2162 		struct wx_ring *ring;
2163 
2164 		wx_for_each_ring(ring, q_vector->rx)
2165 			wx_set_ivar(wx, 0, ring->reg_idx, v_idx);
2166 
2167 		wx_for_each_ring(ring, q_vector->tx)
2168 			wx_set_ivar(wx, 1, ring->reg_idx, v_idx);
2169 
2170 		wx_write_eitr(q_vector);
2171 	}
2172 
2173 	wx_set_ivar(wx, -1, 0, 0);
2174 	if (pdev->msix_enabled)
2175 		wr32(wx, WX_PX_ITR(0), 1950);
2176 }
2177 EXPORT_SYMBOL(wx_configure_vectors);
2178 
2179 /**
2180  * wx_clean_rx_ring - Free Rx Buffers per Queue
2181  * @rx_ring: ring to free buffers from
2182  **/
2183 static void wx_clean_rx_ring(struct wx_ring *rx_ring)
2184 {
2185 	struct wx_rx_buffer *rx_buffer;
2186 	u16 i = rx_ring->next_to_clean;
2187 
2188 	rx_buffer = &rx_ring->rx_buffer_info[i];
2189 
2190 	/* Free all the Rx ring sk_buffs */
2191 	while (i != rx_ring->next_to_alloc) {
2192 		if (rx_buffer->skb) {
2193 			struct sk_buff *skb = rx_buffer->skb;
2194 
2195 			if (WX_CB(skb)->page_released)
2196 				page_pool_put_full_page(rx_ring->page_pool, rx_buffer->page, false);
2197 
2198 			dev_kfree_skb(skb);
2199 		}
2200 
2201 		/* Invalidate cache lines that may have been written to by
2202 		 * device so that we avoid corrupting memory.
2203 		 */
2204 		dma_sync_single_range_for_cpu(rx_ring->dev,
2205 					      rx_buffer->dma,
2206 					      rx_buffer->page_offset,
2207 					      WX_RX_BUFSZ,
2208 					      DMA_FROM_DEVICE);
2209 
2210 		/* free resources associated with mapping */
2211 		page_pool_put_full_page(rx_ring->page_pool, rx_buffer->page, false);
2212 
2213 		i++;
2214 		rx_buffer++;
2215 		if (i == rx_ring->count) {
2216 			i = 0;
2217 			rx_buffer = rx_ring->rx_buffer_info;
2218 		}
2219 	}
2220 
2221 	rx_ring->next_to_alloc = 0;
2222 	rx_ring->next_to_clean = 0;
2223 	rx_ring->next_to_use = 0;
2224 }
2225 
2226 /**
2227  * wx_clean_all_rx_rings - Free Rx Buffers for all queues
2228  * @wx: board private structure
2229  **/
2230 void wx_clean_all_rx_rings(struct wx *wx)
2231 {
2232 	int i;
2233 
2234 	for (i = 0; i < wx->num_rx_queues; i++)
2235 		wx_clean_rx_ring(wx->rx_ring[i]);
2236 }
2237 EXPORT_SYMBOL(wx_clean_all_rx_rings);
2238 
2239 /**
2240  * wx_free_rx_resources - Free Rx Resources
2241  * @rx_ring: ring to clean the resources from
2242  *
2243  * Free all receive software resources
2244  **/
2245 static void wx_free_rx_resources(struct wx_ring *rx_ring)
2246 {
2247 	wx_clean_rx_ring(rx_ring);
2248 	kvfree(rx_ring->rx_buffer_info);
2249 	rx_ring->rx_buffer_info = NULL;
2250 
2251 	/* if not set, then don't free */
2252 	if (!rx_ring->desc)
2253 		return;
2254 
2255 	dma_free_coherent(rx_ring->dev, rx_ring->size,
2256 			  rx_ring->desc, rx_ring->dma);
2257 
2258 	rx_ring->desc = NULL;
2259 
2260 	if (rx_ring->page_pool) {
2261 		page_pool_destroy(rx_ring->page_pool);
2262 		rx_ring->page_pool = NULL;
2263 	}
2264 }
2265 
2266 /**
2267  * wx_free_all_rx_resources - Free Rx Resources for All Queues
2268  * @wx: pointer to hardware structure
2269  *
2270  * Free all receive software resources
2271  **/
2272 static void wx_free_all_rx_resources(struct wx *wx)
2273 {
2274 	int i;
2275 
2276 	for (i = 0; i < wx->num_rx_queues; i++)
2277 		wx_free_rx_resources(wx->rx_ring[i]);
2278 }
2279 
2280 /**
2281  * wx_clean_tx_ring - Free Tx Buffers
2282  * @tx_ring: ring to be cleaned
2283  **/
2284 static void wx_clean_tx_ring(struct wx_ring *tx_ring)
2285 {
2286 	struct wx_tx_buffer *tx_buffer;
2287 	u16 i = tx_ring->next_to_clean;
2288 
2289 	tx_buffer = &tx_ring->tx_buffer_info[i];
2290 
2291 	while (i != tx_ring->next_to_use) {
2292 		union wx_tx_desc *eop_desc, *tx_desc;
2293 
2294 		/* Free all the Tx ring sk_buffs */
2295 		dev_kfree_skb_any(tx_buffer->skb);
2296 
2297 		/* unmap skb header data */
2298 		dma_unmap_single(tx_ring->dev,
2299 				 dma_unmap_addr(tx_buffer, dma),
2300 				 dma_unmap_len(tx_buffer, len),
2301 				 DMA_TO_DEVICE);
2302 
2303 		/* check for eop_desc to determine the end of the packet */
2304 		eop_desc = tx_buffer->next_to_watch;
2305 		tx_desc = WX_TX_DESC(tx_ring, i);
2306 
2307 		/* unmap remaining buffers */
2308 		while (tx_desc != eop_desc) {
2309 			tx_buffer++;
2310 			tx_desc++;
2311 			i++;
2312 			if (unlikely(i == tx_ring->count)) {
2313 				i = 0;
2314 				tx_buffer = tx_ring->tx_buffer_info;
2315 				tx_desc = WX_TX_DESC(tx_ring, 0);
2316 			}
2317 
2318 			/* unmap any remaining paged data */
2319 			if (dma_unmap_len(tx_buffer, len))
2320 				dma_unmap_page(tx_ring->dev,
2321 					       dma_unmap_addr(tx_buffer, dma),
2322 					       dma_unmap_len(tx_buffer, len),
2323 					       DMA_TO_DEVICE);
2324 		}
2325 
2326 		/* move us one more past the eop_desc for start of next pkt */
2327 		tx_buffer++;
2328 		i++;
2329 		if (unlikely(i == tx_ring->count)) {
2330 			i = 0;
2331 			tx_buffer = tx_ring->tx_buffer_info;
2332 		}
2333 	}
2334 
2335 	netdev_tx_reset_queue(wx_txring_txq(tx_ring));
2336 
2337 	/* reset next_to_use and next_to_clean */
2338 	tx_ring->next_to_use = 0;
2339 	tx_ring->next_to_clean = 0;
2340 }
2341 
2342 /**
2343  * wx_clean_all_tx_rings - Free Tx Buffers for all queues
2344  * @wx: board private structure
2345  **/
2346 void wx_clean_all_tx_rings(struct wx *wx)
2347 {
2348 	int i;
2349 
2350 	for (i = 0; i < wx->num_tx_queues; i++)
2351 		wx_clean_tx_ring(wx->tx_ring[i]);
2352 }
2353 EXPORT_SYMBOL(wx_clean_all_tx_rings);
2354 
2355 /**
2356  * wx_free_tx_resources - Free Tx Resources per Queue
2357  * @tx_ring: Tx descriptor ring for a specific queue
2358  *
2359  * Free all transmit software resources
2360  **/
2361 static void wx_free_tx_resources(struct wx_ring *tx_ring)
2362 {
2363 	wx_clean_tx_ring(tx_ring);
2364 	kvfree(tx_ring->tx_buffer_info);
2365 	tx_ring->tx_buffer_info = NULL;
2366 
2367 	/* if not set, then don't free */
2368 	if (!tx_ring->desc)
2369 		return;
2370 
2371 	dma_free_coherent(tx_ring->dev, tx_ring->size,
2372 			  tx_ring->desc, tx_ring->dma);
2373 	tx_ring->desc = NULL;
2374 }
2375 
2376 /**
2377  * wx_free_all_tx_resources - Free Tx Resources for All Queues
2378  * @wx: pointer to hardware structure
2379  *
2380  * Free all transmit software resources
2381  **/
2382 static void wx_free_all_tx_resources(struct wx *wx)
2383 {
2384 	int i;
2385 
2386 	for (i = 0; i < wx->num_tx_queues; i++)
2387 		wx_free_tx_resources(wx->tx_ring[i]);
2388 }
2389 
2390 void wx_free_resources(struct wx *wx)
2391 {
2392 	wx_free_isb_resources(wx);
2393 	wx_free_all_rx_resources(wx);
2394 	wx_free_all_tx_resources(wx);
2395 }
2396 EXPORT_SYMBOL(wx_free_resources);
2397 
2398 static int wx_alloc_page_pool(struct wx_ring *rx_ring)
2399 {
2400 	int ret = 0;
2401 
2402 	struct page_pool_params pp_params = {
2403 		.flags = PP_FLAG_DMA_MAP | PP_FLAG_DMA_SYNC_DEV,
2404 		.order = 0,
2405 		.pool_size = rx_ring->size,
2406 		.nid = dev_to_node(rx_ring->dev),
2407 		.dev = rx_ring->dev,
2408 		.dma_dir = DMA_FROM_DEVICE,
2409 		.offset = 0,
2410 		.max_len = PAGE_SIZE,
2411 	};
2412 
2413 	rx_ring->page_pool = page_pool_create(&pp_params);
2414 	if (IS_ERR(rx_ring->page_pool)) {
2415 		ret = PTR_ERR(rx_ring->page_pool);
2416 		rx_ring->page_pool = NULL;
2417 	}
2418 
2419 	return ret;
2420 }
2421 
2422 /**
2423  * wx_setup_rx_resources - allocate Rx resources (Descriptors)
2424  * @rx_ring: rx descriptor ring (for a specific queue) to setup
2425  *
2426  * Returns 0 on success, negative on failure
2427  **/
2428 static int wx_setup_rx_resources(struct wx_ring *rx_ring)
2429 {
2430 	struct device *dev = rx_ring->dev;
2431 	int orig_node = dev_to_node(dev);
2432 	int numa_node = NUMA_NO_NODE;
2433 	int size, ret;
2434 
2435 	size = sizeof(struct wx_rx_buffer) * rx_ring->count;
2436 
2437 	if (rx_ring->q_vector)
2438 		numa_node = rx_ring->q_vector->numa_node;
2439 
2440 	rx_ring->rx_buffer_info = kvmalloc_node(size, GFP_KERNEL, numa_node);
2441 	if (!rx_ring->rx_buffer_info)
2442 		rx_ring->rx_buffer_info = kvmalloc(size, GFP_KERNEL);
2443 	if (!rx_ring->rx_buffer_info)
2444 		goto err;
2445 
2446 	/* Round up to nearest 4K */
2447 	rx_ring->size = rx_ring->count * sizeof(union wx_rx_desc);
2448 	rx_ring->size = ALIGN(rx_ring->size, 4096);
2449 
2450 	set_dev_node(dev, numa_node);
2451 	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
2452 					   &rx_ring->dma, GFP_KERNEL);
2453 	if (!rx_ring->desc) {
2454 		set_dev_node(dev, orig_node);
2455 		rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
2456 						   &rx_ring->dma, GFP_KERNEL);
2457 	}
2458 
2459 	if (!rx_ring->desc)
2460 		goto err;
2461 
2462 	rx_ring->next_to_clean = 0;
2463 	rx_ring->next_to_use = 0;
2464 
2465 	ret = wx_alloc_page_pool(rx_ring);
2466 	if (ret < 0) {
2467 		dev_err(rx_ring->dev, "Page pool creation failed: %d\n", ret);
2468 		goto err_desc;
2469 	}
2470 
2471 	return 0;
2472 
2473 err_desc:
2474 	dma_free_coherent(dev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2475 err:
2476 	kvfree(rx_ring->rx_buffer_info);
2477 	rx_ring->rx_buffer_info = NULL;
2478 	dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
2479 	return -ENOMEM;
2480 }
2481 
2482 /**
2483  * wx_setup_all_rx_resources - allocate all queues Rx resources
2484  * @wx: pointer to hardware structure
2485  *
2486  * If this function returns with an error, then it's possible one or
2487  * more of the rings is populated (while the rest are not).  It is the
2488  * callers duty to clean those orphaned rings.
2489  *
2490  * Return 0 on success, negative on failure
2491  **/
2492 static int wx_setup_all_rx_resources(struct wx *wx)
2493 {
2494 	int i, err = 0;
2495 
2496 	for (i = 0; i < wx->num_rx_queues; i++) {
2497 		err = wx_setup_rx_resources(wx->rx_ring[i]);
2498 		if (!err)
2499 			continue;
2500 
2501 		wx_err(wx, "Allocation for Rx Queue %u failed\n", i);
2502 		goto err_setup_rx;
2503 	}
2504 
2505 	return 0;
2506 err_setup_rx:
2507 	/* rewind the index freeing the rings as we go */
2508 	while (i--)
2509 		wx_free_rx_resources(wx->rx_ring[i]);
2510 	return err;
2511 }
2512 
2513 /**
2514  * wx_setup_tx_resources - allocate Tx resources (Descriptors)
2515  * @tx_ring: tx descriptor ring (for a specific queue) to setup
2516  *
2517  * Return 0 on success, negative on failure
2518  **/
2519 static int wx_setup_tx_resources(struct wx_ring *tx_ring)
2520 {
2521 	struct device *dev = tx_ring->dev;
2522 	int orig_node = dev_to_node(dev);
2523 	int numa_node = NUMA_NO_NODE;
2524 	int size;
2525 
2526 	size = sizeof(struct wx_tx_buffer) * tx_ring->count;
2527 
2528 	if (tx_ring->q_vector)
2529 		numa_node = tx_ring->q_vector->numa_node;
2530 
2531 	tx_ring->tx_buffer_info = kvmalloc_node(size, GFP_KERNEL, numa_node);
2532 	if (!tx_ring->tx_buffer_info)
2533 		tx_ring->tx_buffer_info = kvmalloc(size, GFP_KERNEL);
2534 	if (!tx_ring->tx_buffer_info)
2535 		goto err;
2536 
2537 	/* round up to nearest 4K */
2538 	tx_ring->size = tx_ring->count * sizeof(union wx_tx_desc);
2539 	tx_ring->size = ALIGN(tx_ring->size, 4096);
2540 
2541 	set_dev_node(dev, numa_node);
2542 	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
2543 					   &tx_ring->dma, GFP_KERNEL);
2544 	if (!tx_ring->desc) {
2545 		set_dev_node(dev, orig_node);
2546 		tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
2547 						   &tx_ring->dma, GFP_KERNEL);
2548 	}
2549 
2550 	if (!tx_ring->desc)
2551 		goto err;
2552 
2553 	tx_ring->next_to_use = 0;
2554 	tx_ring->next_to_clean = 0;
2555 
2556 	return 0;
2557 
2558 err:
2559 	kvfree(tx_ring->tx_buffer_info);
2560 	tx_ring->tx_buffer_info = NULL;
2561 	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
2562 	return -ENOMEM;
2563 }
2564 
2565 /**
2566  * wx_setup_all_tx_resources - allocate all queues Tx resources
2567  * @wx: pointer to private structure
2568  *
2569  * If this function returns with an error, then it's possible one or
2570  * more of the rings is populated (while the rest are not).  It is the
2571  * callers duty to clean those orphaned rings.
2572  *
2573  * Return 0 on success, negative on failure
2574  **/
2575 static int wx_setup_all_tx_resources(struct wx *wx)
2576 {
2577 	int i, err = 0;
2578 
2579 	for (i = 0; i < wx->num_tx_queues; i++) {
2580 		err = wx_setup_tx_resources(wx->tx_ring[i]);
2581 		if (!err)
2582 			continue;
2583 
2584 		wx_err(wx, "Allocation for Tx Queue %u failed\n", i);
2585 		goto err_setup_tx;
2586 	}
2587 
2588 	return 0;
2589 err_setup_tx:
2590 	/* rewind the index freeing the rings as we go */
2591 	while (i--)
2592 		wx_free_tx_resources(wx->tx_ring[i]);
2593 	return err;
2594 }
2595 
2596 int wx_setup_resources(struct wx *wx)
2597 {
2598 	int err;
2599 
2600 	/* allocate transmit descriptors */
2601 	err = wx_setup_all_tx_resources(wx);
2602 	if (err)
2603 		return err;
2604 
2605 	/* allocate receive descriptors */
2606 	err = wx_setup_all_rx_resources(wx);
2607 	if (err)
2608 		goto err_free_tx;
2609 
2610 	err = wx_setup_isb_resources(wx);
2611 	if (err)
2612 		goto err_free_rx;
2613 
2614 	return 0;
2615 
2616 err_free_rx:
2617 	wx_free_all_rx_resources(wx);
2618 err_free_tx:
2619 	wx_free_all_tx_resources(wx);
2620 
2621 	return err;
2622 }
2623 EXPORT_SYMBOL(wx_setup_resources);
2624 
2625 /**
2626  * wx_get_stats64 - Get System Network Statistics
2627  * @netdev: network interface device structure
2628  * @stats: storage space for 64bit statistics
2629  */
2630 void wx_get_stats64(struct net_device *netdev,
2631 		    struct rtnl_link_stats64 *stats)
2632 {
2633 	struct wx *wx = netdev_priv(netdev);
2634 	struct wx_hw_stats *hwstats;
2635 	int i;
2636 
2637 	wx_update_stats(wx);
2638 
2639 	rcu_read_lock();
2640 	for (i = 0; i < wx->num_rx_queues; i++) {
2641 		struct wx_ring *ring = READ_ONCE(wx->rx_ring[i]);
2642 		u64 bytes, packets;
2643 		unsigned int start;
2644 
2645 		if (ring) {
2646 			do {
2647 				start = u64_stats_fetch_begin(&ring->syncp);
2648 				packets = ring->stats.packets;
2649 				bytes   = ring->stats.bytes;
2650 			} while (u64_stats_fetch_retry(&ring->syncp, start));
2651 			stats->rx_packets += packets;
2652 			stats->rx_bytes   += bytes;
2653 		}
2654 	}
2655 
2656 	for (i = 0; i < wx->num_tx_queues; i++) {
2657 		struct wx_ring *ring = READ_ONCE(wx->tx_ring[i]);
2658 		u64 bytes, packets;
2659 		unsigned int start;
2660 
2661 		if (ring) {
2662 			do {
2663 				start = u64_stats_fetch_begin(&ring->syncp);
2664 				packets = ring->stats.packets;
2665 				bytes   = ring->stats.bytes;
2666 			} while (u64_stats_fetch_retry(&ring->syncp,
2667 							   start));
2668 			stats->tx_packets += packets;
2669 			stats->tx_bytes   += bytes;
2670 		}
2671 	}
2672 
2673 	rcu_read_unlock();
2674 
2675 	hwstats = &wx->stats;
2676 	stats->rx_errors = hwstats->crcerrs + hwstats->rlec;
2677 	stats->multicast = hwstats->qmprc;
2678 	stats->rx_length_errors = hwstats->rlec;
2679 	stats->rx_crc_errors = hwstats->crcerrs;
2680 }
2681 EXPORT_SYMBOL(wx_get_stats64);
2682 
2683 int wx_set_features(struct net_device *netdev, netdev_features_t features)
2684 {
2685 	netdev_features_t changed = netdev->features ^ features;
2686 	struct wx *wx = netdev_priv(netdev);
2687 
2688 	if (features & NETIF_F_RXHASH) {
2689 		wr32m(wx, WX_RDB_RA_CTL, WX_RDB_RA_CTL_RSS_EN,
2690 		      WX_RDB_RA_CTL_RSS_EN);
2691 		wx->rss_enabled = true;
2692 	} else {
2693 		wr32m(wx, WX_RDB_RA_CTL, WX_RDB_RA_CTL_RSS_EN, 0);
2694 		wx->rss_enabled = false;
2695 	}
2696 
2697 	if (changed &
2698 	    (NETIF_F_HW_VLAN_CTAG_RX |
2699 	     NETIF_F_HW_VLAN_STAG_RX))
2700 		wx_set_rx_mode(netdev);
2701 
2702 	return 1;
2703 }
2704 EXPORT_SYMBOL(wx_set_features);
2705 
2706 void wx_set_ring(struct wx *wx, u32 new_tx_count,
2707 		 u32 new_rx_count, struct wx_ring *temp_ring)
2708 {
2709 	int i, err = 0;
2710 
2711 	/* Setup new Tx resources and free the old Tx resources in that order.
2712 	 * We can then assign the new resources to the rings via a memcpy.
2713 	 * The advantage to this approach is that we are guaranteed to still
2714 	 * have resources even in the case of an allocation failure.
2715 	 */
2716 	if (new_tx_count != wx->tx_ring_count) {
2717 		for (i = 0; i < wx->num_tx_queues; i++) {
2718 			memcpy(&temp_ring[i], wx->tx_ring[i],
2719 			       sizeof(struct wx_ring));
2720 
2721 			temp_ring[i].count = new_tx_count;
2722 			err = wx_setup_tx_resources(&temp_ring[i]);
2723 			if (err) {
2724 				wx_err(wx, "setup new tx resources failed, keep using the old config\n");
2725 				while (i) {
2726 					i--;
2727 					wx_free_tx_resources(&temp_ring[i]);
2728 				}
2729 				return;
2730 			}
2731 		}
2732 
2733 		for (i = 0; i < wx->num_tx_queues; i++) {
2734 			wx_free_tx_resources(wx->tx_ring[i]);
2735 
2736 			memcpy(wx->tx_ring[i], &temp_ring[i],
2737 			       sizeof(struct wx_ring));
2738 		}
2739 
2740 		wx->tx_ring_count = new_tx_count;
2741 	}
2742 
2743 	/* Repeat the process for the Rx rings if needed */
2744 	if (new_rx_count != wx->rx_ring_count) {
2745 		for (i = 0; i < wx->num_rx_queues; i++) {
2746 			memcpy(&temp_ring[i], wx->rx_ring[i],
2747 			       sizeof(struct wx_ring));
2748 
2749 			temp_ring[i].count = new_rx_count;
2750 			err = wx_setup_rx_resources(&temp_ring[i]);
2751 			if (err) {
2752 				wx_err(wx, "setup new rx resources failed, keep using the old config\n");
2753 				while (i) {
2754 					i--;
2755 					wx_free_rx_resources(&temp_ring[i]);
2756 				}
2757 				return;
2758 			}
2759 		}
2760 
2761 		for (i = 0; i < wx->num_rx_queues; i++) {
2762 			wx_free_rx_resources(wx->rx_ring[i]);
2763 			memcpy(wx->rx_ring[i], &temp_ring[i],
2764 			       sizeof(struct wx_ring));
2765 		}
2766 
2767 		wx->rx_ring_count = new_rx_count;
2768 	}
2769 }
2770 EXPORT_SYMBOL(wx_set_ring);
2771 
2772 MODULE_DESCRIPTION("Common library for Wangxun(R) Ethernet drivers.");
2773 MODULE_LICENSE("GPL");
2774