xref: /openbmc/linux/drivers/net/ethernet/intel/i40e/i40e_txrx.c (revision 05cf4fe738242183f1237f1b3a28b4479348c0a1)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 2013 - 2018 Intel Corporation. */
3 
4 #include <linux/prefetch.h>
5 #include <linux/bpf_trace.h>
6 #include <net/xdp.h>
7 #include "i40e.h"
8 #include "i40e_trace.h"
9 #include "i40e_prototype.h"
10 #include "i40e_txrx_common.h"
11 #include "i40e_xsk.h"
12 
13 #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
14 /**
15  * i40e_fdir - Generate a Flow Director descriptor based on fdata
16  * @tx_ring: Tx ring to send buffer on
17  * @fdata: Flow director filter data
18  * @add: Indicate if we are adding a rule or deleting one
19  *
20  **/
21 static void i40e_fdir(struct i40e_ring *tx_ring,
22 		      struct i40e_fdir_filter *fdata, bool add)
23 {
24 	struct i40e_filter_program_desc *fdir_desc;
25 	struct i40e_pf *pf = tx_ring->vsi->back;
26 	u32 flex_ptype, dtype_cmd;
27 	u16 i;
28 
29 	/* grab the next descriptor */
30 	i = tx_ring->next_to_use;
31 	fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
32 
33 	i++;
34 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
35 
36 	flex_ptype = I40E_TXD_FLTR_QW0_QINDEX_MASK &
37 		     (fdata->q_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT);
38 
39 	flex_ptype |= I40E_TXD_FLTR_QW0_FLEXOFF_MASK &
40 		      (fdata->flex_off << I40E_TXD_FLTR_QW0_FLEXOFF_SHIFT);
41 
42 	flex_ptype |= I40E_TXD_FLTR_QW0_PCTYPE_MASK &
43 		      (fdata->pctype << I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
44 
45 	flex_ptype |= I40E_TXD_FLTR_QW0_PCTYPE_MASK &
46 		      (fdata->flex_offset << I40E_TXD_FLTR_QW0_FLEXOFF_SHIFT);
47 
48 	/* Use LAN VSI Id if not programmed by user */
49 	flex_ptype |= I40E_TXD_FLTR_QW0_DEST_VSI_MASK &
50 		      ((u32)(fdata->dest_vsi ? : pf->vsi[pf->lan_vsi]->id) <<
51 		       I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT);
52 
53 	dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
54 
55 	dtype_cmd |= add ?
56 		     I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
57 		     I40E_TXD_FLTR_QW1_PCMD_SHIFT :
58 		     I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
59 		     I40E_TXD_FLTR_QW1_PCMD_SHIFT;
60 
61 	dtype_cmd |= I40E_TXD_FLTR_QW1_DEST_MASK &
62 		     (fdata->dest_ctl << I40E_TXD_FLTR_QW1_DEST_SHIFT);
63 
64 	dtype_cmd |= I40E_TXD_FLTR_QW1_FD_STATUS_MASK &
65 		     (fdata->fd_status << I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT);
66 
67 	if (fdata->cnt_index) {
68 		dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
69 		dtype_cmd |= I40E_TXD_FLTR_QW1_CNTINDEX_MASK &
70 			     ((u32)fdata->cnt_index <<
71 			      I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT);
72 	}
73 
74 	fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
75 	fdir_desc->rsvd = cpu_to_le32(0);
76 	fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
77 	fdir_desc->fd_id = cpu_to_le32(fdata->fd_id);
78 }
79 
80 #define I40E_FD_CLEAN_DELAY 10
81 /**
82  * i40e_program_fdir_filter - Program a Flow Director filter
83  * @fdir_data: Packet data that will be filter parameters
84  * @raw_packet: the pre-allocated packet buffer for FDir
85  * @pf: The PF pointer
86  * @add: True for add/update, False for remove
87  **/
88 static int i40e_program_fdir_filter(struct i40e_fdir_filter *fdir_data,
89 				    u8 *raw_packet, struct i40e_pf *pf,
90 				    bool add)
91 {
92 	struct i40e_tx_buffer *tx_buf, *first;
93 	struct i40e_tx_desc *tx_desc;
94 	struct i40e_ring *tx_ring;
95 	struct i40e_vsi *vsi;
96 	struct device *dev;
97 	dma_addr_t dma;
98 	u32 td_cmd = 0;
99 	u16 i;
100 
101 	/* find existing FDIR VSI */
102 	vsi = i40e_find_vsi_by_type(pf, I40E_VSI_FDIR);
103 	if (!vsi)
104 		return -ENOENT;
105 
106 	tx_ring = vsi->tx_rings[0];
107 	dev = tx_ring->dev;
108 
109 	/* we need two descriptors to add/del a filter and we can wait */
110 	for (i = I40E_FD_CLEAN_DELAY; I40E_DESC_UNUSED(tx_ring) < 2; i--) {
111 		if (!i)
112 			return -EAGAIN;
113 		msleep_interruptible(1);
114 	}
115 
116 	dma = dma_map_single(dev, raw_packet,
117 			     I40E_FDIR_MAX_RAW_PACKET_SIZE, DMA_TO_DEVICE);
118 	if (dma_mapping_error(dev, dma))
119 		goto dma_fail;
120 
121 	/* grab the next descriptor */
122 	i = tx_ring->next_to_use;
123 	first = &tx_ring->tx_bi[i];
124 	i40e_fdir(tx_ring, fdir_data, add);
125 
126 	/* Now program a dummy descriptor */
127 	i = tx_ring->next_to_use;
128 	tx_desc = I40E_TX_DESC(tx_ring, i);
129 	tx_buf = &tx_ring->tx_bi[i];
130 
131 	tx_ring->next_to_use = ((i + 1) < tx_ring->count) ? i + 1 : 0;
132 
133 	memset(tx_buf, 0, sizeof(struct i40e_tx_buffer));
134 
135 	/* record length, and DMA address */
136 	dma_unmap_len_set(tx_buf, len, I40E_FDIR_MAX_RAW_PACKET_SIZE);
137 	dma_unmap_addr_set(tx_buf, dma, dma);
138 
139 	tx_desc->buffer_addr = cpu_to_le64(dma);
140 	td_cmd = I40E_TXD_CMD | I40E_TX_DESC_CMD_DUMMY;
141 
142 	tx_buf->tx_flags = I40E_TX_FLAGS_FD_SB;
143 	tx_buf->raw_buf = (void *)raw_packet;
144 
145 	tx_desc->cmd_type_offset_bsz =
146 		build_ctob(td_cmd, 0, I40E_FDIR_MAX_RAW_PACKET_SIZE, 0);
147 
148 	/* Force memory writes to complete before letting h/w
149 	 * know there are new descriptors to fetch.
150 	 */
151 	wmb();
152 
153 	/* Mark the data descriptor to be watched */
154 	first->next_to_watch = tx_desc;
155 
156 	writel(tx_ring->next_to_use, tx_ring->tail);
157 	return 0;
158 
159 dma_fail:
160 	return -1;
161 }
162 
163 #define IP_HEADER_OFFSET 14
164 #define I40E_UDPIP_DUMMY_PACKET_LEN 42
165 /**
166  * i40e_add_del_fdir_udpv4 - Add/Remove UDPv4 filters
167  * @vsi: pointer to the targeted VSI
168  * @fd_data: the flow director data required for the FDir descriptor
169  * @add: true adds a filter, false removes it
170  *
171  * Returns 0 if the filters were successfully added or removed
172  **/
173 static int i40e_add_del_fdir_udpv4(struct i40e_vsi *vsi,
174 				   struct i40e_fdir_filter *fd_data,
175 				   bool add)
176 {
177 	struct i40e_pf *pf = vsi->back;
178 	struct udphdr *udp;
179 	struct iphdr *ip;
180 	u8 *raw_packet;
181 	int ret;
182 	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
183 		0x45, 0, 0, 0x1c, 0, 0, 0x40, 0, 0x40, 0x11, 0, 0, 0, 0, 0, 0,
184 		0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
185 
186 	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
187 	if (!raw_packet)
188 		return -ENOMEM;
189 	memcpy(raw_packet, packet, I40E_UDPIP_DUMMY_PACKET_LEN);
190 
191 	ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
192 	udp = (struct udphdr *)(raw_packet + IP_HEADER_OFFSET
193 	      + sizeof(struct iphdr));
194 
195 	ip->daddr = fd_data->dst_ip;
196 	udp->dest = fd_data->dst_port;
197 	ip->saddr = fd_data->src_ip;
198 	udp->source = fd_data->src_port;
199 
200 	if (fd_data->flex_filter) {
201 		u8 *payload = raw_packet + I40E_UDPIP_DUMMY_PACKET_LEN;
202 		__be16 pattern = fd_data->flex_word;
203 		u16 off = fd_data->flex_offset;
204 
205 		*((__force __be16 *)(payload + off)) = pattern;
206 	}
207 
208 	fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_UDP;
209 	ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
210 	if (ret) {
211 		dev_info(&pf->pdev->dev,
212 			 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
213 			 fd_data->pctype, fd_data->fd_id, ret);
214 		/* Free the packet buffer since it wasn't added to the ring */
215 		kfree(raw_packet);
216 		return -EOPNOTSUPP;
217 	} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
218 		if (add)
219 			dev_info(&pf->pdev->dev,
220 				 "Filter OK for PCTYPE %d loc = %d\n",
221 				 fd_data->pctype, fd_data->fd_id);
222 		else
223 			dev_info(&pf->pdev->dev,
224 				 "Filter deleted for PCTYPE %d loc = %d\n",
225 				 fd_data->pctype, fd_data->fd_id);
226 	}
227 
228 	if (add)
229 		pf->fd_udp4_filter_cnt++;
230 	else
231 		pf->fd_udp4_filter_cnt--;
232 
233 	return 0;
234 }
235 
236 #define I40E_TCPIP_DUMMY_PACKET_LEN 54
237 /**
238  * i40e_add_del_fdir_tcpv4 - Add/Remove TCPv4 filters
239  * @vsi: pointer to the targeted VSI
240  * @fd_data: the flow director data required for the FDir descriptor
241  * @add: true adds a filter, false removes it
242  *
243  * Returns 0 if the filters were successfully added or removed
244  **/
245 static int i40e_add_del_fdir_tcpv4(struct i40e_vsi *vsi,
246 				   struct i40e_fdir_filter *fd_data,
247 				   bool add)
248 {
249 	struct i40e_pf *pf = vsi->back;
250 	struct tcphdr *tcp;
251 	struct iphdr *ip;
252 	u8 *raw_packet;
253 	int ret;
254 	/* Dummy packet */
255 	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
256 		0x45, 0, 0, 0x28, 0, 0, 0x40, 0, 0x40, 0x6, 0, 0, 0, 0, 0, 0,
257 		0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x80, 0x11,
258 		0x0, 0x72, 0, 0, 0, 0};
259 
260 	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
261 	if (!raw_packet)
262 		return -ENOMEM;
263 	memcpy(raw_packet, packet, I40E_TCPIP_DUMMY_PACKET_LEN);
264 
265 	ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
266 	tcp = (struct tcphdr *)(raw_packet + IP_HEADER_OFFSET
267 	      + sizeof(struct iphdr));
268 
269 	ip->daddr = fd_data->dst_ip;
270 	tcp->dest = fd_data->dst_port;
271 	ip->saddr = fd_data->src_ip;
272 	tcp->source = fd_data->src_port;
273 
274 	if (fd_data->flex_filter) {
275 		u8 *payload = raw_packet + I40E_TCPIP_DUMMY_PACKET_LEN;
276 		__be16 pattern = fd_data->flex_word;
277 		u16 off = fd_data->flex_offset;
278 
279 		*((__force __be16 *)(payload + off)) = pattern;
280 	}
281 
282 	fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_TCP;
283 	ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
284 	if (ret) {
285 		dev_info(&pf->pdev->dev,
286 			 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
287 			 fd_data->pctype, fd_data->fd_id, ret);
288 		/* Free the packet buffer since it wasn't added to the ring */
289 		kfree(raw_packet);
290 		return -EOPNOTSUPP;
291 	} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
292 		if (add)
293 			dev_info(&pf->pdev->dev, "Filter OK for PCTYPE %d loc = %d)\n",
294 				 fd_data->pctype, fd_data->fd_id);
295 		else
296 			dev_info(&pf->pdev->dev,
297 				 "Filter deleted for PCTYPE %d loc = %d\n",
298 				 fd_data->pctype, fd_data->fd_id);
299 	}
300 
301 	if (add) {
302 		pf->fd_tcp4_filter_cnt++;
303 		if ((pf->flags & I40E_FLAG_FD_ATR_ENABLED) &&
304 		    I40E_DEBUG_FD & pf->hw.debug_mask)
305 			dev_info(&pf->pdev->dev, "Forcing ATR off, sideband rules for TCP/IPv4 flow being applied\n");
306 		set_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state);
307 	} else {
308 		pf->fd_tcp4_filter_cnt--;
309 	}
310 
311 	return 0;
312 }
313 
314 #define I40E_SCTPIP_DUMMY_PACKET_LEN 46
315 /**
316  * i40e_add_del_fdir_sctpv4 - Add/Remove SCTPv4 Flow Director filters for
317  * a specific flow spec
318  * @vsi: pointer to the targeted VSI
319  * @fd_data: the flow director data required for the FDir descriptor
320  * @add: true adds a filter, false removes it
321  *
322  * Returns 0 if the filters were successfully added or removed
323  **/
324 static int i40e_add_del_fdir_sctpv4(struct i40e_vsi *vsi,
325 				    struct i40e_fdir_filter *fd_data,
326 				    bool add)
327 {
328 	struct i40e_pf *pf = vsi->back;
329 	struct sctphdr *sctp;
330 	struct iphdr *ip;
331 	u8 *raw_packet;
332 	int ret;
333 	/* Dummy packet */
334 	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
335 		0x45, 0, 0, 0x20, 0, 0, 0x40, 0, 0x40, 0x84, 0, 0, 0, 0, 0, 0,
336 		0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
337 
338 	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
339 	if (!raw_packet)
340 		return -ENOMEM;
341 	memcpy(raw_packet, packet, I40E_SCTPIP_DUMMY_PACKET_LEN);
342 
343 	ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
344 	sctp = (struct sctphdr *)(raw_packet + IP_HEADER_OFFSET
345 	      + sizeof(struct iphdr));
346 
347 	ip->daddr = fd_data->dst_ip;
348 	sctp->dest = fd_data->dst_port;
349 	ip->saddr = fd_data->src_ip;
350 	sctp->source = fd_data->src_port;
351 
352 	if (fd_data->flex_filter) {
353 		u8 *payload = raw_packet + I40E_SCTPIP_DUMMY_PACKET_LEN;
354 		__be16 pattern = fd_data->flex_word;
355 		u16 off = fd_data->flex_offset;
356 
357 		*((__force __be16 *)(payload + off)) = pattern;
358 	}
359 
360 	fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_SCTP;
361 	ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
362 	if (ret) {
363 		dev_info(&pf->pdev->dev,
364 			 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
365 			 fd_data->pctype, fd_data->fd_id, ret);
366 		/* Free the packet buffer since it wasn't added to the ring */
367 		kfree(raw_packet);
368 		return -EOPNOTSUPP;
369 	} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
370 		if (add)
371 			dev_info(&pf->pdev->dev,
372 				 "Filter OK for PCTYPE %d loc = %d\n",
373 				 fd_data->pctype, fd_data->fd_id);
374 		else
375 			dev_info(&pf->pdev->dev,
376 				 "Filter deleted for PCTYPE %d loc = %d\n",
377 				 fd_data->pctype, fd_data->fd_id);
378 	}
379 
380 	if (add)
381 		pf->fd_sctp4_filter_cnt++;
382 	else
383 		pf->fd_sctp4_filter_cnt--;
384 
385 	return 0;
386 }
387 
388 #define I40E_IP_DUMMY_PACKET_LEN 34
389 /**
390  * i40e_add_del_fdir_ipv4 - Add/Remove IPv4 Flow Director filters for
391  * a specific flow spec
392  * @vsi: pointer to the targeted VSI
393  * @fd_data: the flow director data required for the FDir descriptor
394  * @add: true adds a filter, false removes it
395  *
396  * Returns 0 if the filters were successfully added or removed
397  **/
398 static int i40e_add_del_fdir_ipv4(struct i40e_vsi *vsi,
399 				  struct i40e_fdir_filter *fd_data,
400 				  bool add)
401 {
402 	struct i40e_pf *pf = vsi->back;
403 	struct iphdr *ip;
404 	u8 *raw_packet;
405 	int ret;
406 	int i;
407 	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
408 		0x45, 0, 0, 0x14, 0, 0, 0x40, 0, 0x40, 0x10, 0, 0, 0, 0, 0, 0,
409 		0, 0, 0, 0};
410 
411 	for (i = I40E_FILTER_PCTYPE_NONF_IPV4_OTHER;
412 	     i <= I40E_FILTER_PCTYPE_FRAG_IPV4;	i++) {
413 		raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
414 		if (!raw_packet)
415 			return -ENOMEM;
416 		memcpy(raw_packet, packet, I40E_IP_DUMMY_PACKET_LEN);
417 		ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
418 
419 		ip->saddr = fd_data->src_ip;
420 		ip->daddr = fd_data->dst_ip;
421 		ip->protocol = 0;
422 
423 		if (fd_data->flex_filter) {
424 			u8 *payload = raw_packet + I40E_IP_DUMMY_PACKET_LEN;
425 			__be16 pattern = fd_data->flex_word;
426 			u16 off = fd_data->flex_offset;
427 
428 			*((__force __be16 *)(payload + off)) = pattern;
429 		}
430 
431 		fd_data->pctype = i;
432 		ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
433 		if (ret) {
434 			dev_info(&pf->pdev->dev,
435 				 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
436 				 fd_data->pctype, fd_data->fd_id, ret);
437 			/* The packet buffer wasn't added to the ring so we
438 			 * need to free it now.
439 			 */
440 			kfree(raw_packet);
441 			return -EOPNOTSUPP;
442 		} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
443 			if (add)
444 				dev_info(&pf->pdev->dev,
445 					 "Filter OK for PCTYPE %d loc = %d\n",
446 					 fd_data->pctype, fd_data->fd_id);
447 			else
448 				dev_info(&pf->pdev->dev,
449 					 "Filter deleted for PCTYPE %d loc = %d\n",
450 					 fd_data->pctype, fd_data->fd_id);
451 		}
452 	}
453 
454 	if (add)
455 		pf->fd_ip4_filter_cnt++;
456 	else
457 		pf->fd_ip4_filter_cnt--;
458 
459 	return 0;
460 }
461 
462 /**
463  * i40e_add_del_fdir - Build raw packets to add/del fdir filter
464  * @vsi: pointer to the targeted VSI
465  * @input: filter to add or delete
466  * @add: true adds a filter, false removes it
467  *
468  **/
469 int i40e_add_del_fdir(struct i40e_vsi *vsi,
470 		      struct i40e_fdir_filter *input, bool add)
471 {
472 	struct i40e_pf *pf = vsi->back;
473 	int ret;
474 
475 	switch (input->flow_type & ~FLOW_EXT) {
476 	case TCP_V4_FLOW:
477 		ret = i40e_add_del_fdir_tcpv4(vsi, input, add);
478 		break;
479 	case UDP_V4_FLOW:
480 		ret = i40e_add_del_fdir_udpv4(vsi, input, add);
481 		break;
482 	case SCTP_V4_FLOW:
483 		ret = i40e_add_del_fdir_sctpv4(vsi, input, add);
484 		break;
485 	case IP_USER_FLOW:
486 		switch (input->ip4_proto) {
487 		case IPPROTO_TCP:
488 			ret = i40e_add_del_fdir_tcpv4(vsi, input, add);
489 			break;
490 		case IPPROTO_UDP:
491 			ret = i40e_add_del_fdir_udpv4(vsi, input, add);
492 			break;
493 		case IPPROTO_SCTP:
494 			ret = i40e_add_del_fdir_sctpv4(vsi, input, add);
495 			break;
496 		case IPPROTO_IP:
497 			ret = i40e_add_del_fdir_ipv4(vsi, input, add);
498 			break;
499 		default:
500 			/* We cannot support masking based on protocol */
501 			dev_info(&pf->pdev->dev, "Unsupported IPv4 protocol 0x%02x\n",
502 				 input->ip4_proto);
503 			return -EINVAL;
504 		}
505 		break;
506 	default:
507 		dev_info(&pf->pdev->dev, "Unsupported flow type 0x%02x\n",
508 			 input->flow_type);
509 		return -EINVAL;
510 	}
511 
512 	/* The buffer allocated here will be normally be freed by
513 	 * i40e_clean_fdir_tx_irq() as it reclaims resources after transmit
514 	 * completion. In the event of an error adding the buffer to the FDIR
515 	 * ring, it will immediately be freed. It may also be freed by
516 	 * i40e_clean_tx_ring() when closing the VSI.
517 	 */
518 	return ret;
519 }
520 
521 /**
522  * i40e_fd_handle_status - check the Programming Status for FD
523  * @rx_ring: the Rx ring for this descriptor
524  * @rx_desc: the Rx descriptor for programming Status, not a packet descriptor.
525  * @prog_id: the id originally used for programming
526  *
527  * This is used to verify if the FD programming or invalidation
528  * requested by SW to the HW is successful or not and take actions accordingly.
529  **/
530 void i40e_fd_handle_status(struct i40e_ring *rx_ring,
531 			   union i40e_rx_desc *rx_desc, u8 prog_id)
532 {
533 	struct i40e_pf *pf = rx_ring->vsi->back;
534 	struct pci_dev *pdev = pf->pdev;
535 	u32 fcnt_prog, fcnt_avail;
536 	u32 error;
537 	u64 qw;
538 
539 	qw = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
540 	error = (qw & I40E_RX_PROG_STATUS_DESC_QW1_ERROR_MASK) >>
541 		I40E_RX_PROG_STATUS_DESC_QW1_ERROR_SHIFT;
542 
543 	if (error == BIT(I40E_RX_PROG_STATUS_DESC_FD_TBL_FULL_SHIFT)) {
544 		pf->fd_inv = le32_to_cpu(rx_desc->wb.qword0.hi_dword.fd_id);
545 		if ((rx_desc->wb.qword0.hi_dword.fd_id != 0) ||
546 		    (I40E_DEBUG_FD & pf->hw.debug_mask))
547 			dev_warn(&pdev->dev, "ntuple filter loc = %d, could not be added\n",
548 				 pf->fd_inv);
549 
550 		/* Check if the programming error is for ATR.
551 		 * If so, auto disable ATR and set a state for
552 		 * flush in progress. Next time we come here if flush is in
553 		 * progress do nothing, once flush is complete the state will
554 		 * be cleared.
555 		 */
556 		if (test_bit(__I40E_FD_FLUSH_REQUESTED, pf->state))
557 			return;
558 
559 		pf->fd_add_err++;
560 		/* store the current atr filter count */
561 		pf->fd_atr_cnt = i40e_get_current_atr_cnt(pf);
562 
563 		if ((rx_desc->wb.qword0.hi_dword.fd_id == 0) &&
564 		    test_bit(__I40E_FD_SB_AUTO_DISABLED, pf->state)) {
565 			/* These set_bit() calls aren't atomic with the
566 			 * test_bit() here, but worse case we potentially
567 			 * disable ATR and queue a flush right after SB
568 			 * support is re-enabled. That shouldn't cause an
569 			 * issue in practice
570 			 */
571 			set_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state);
572 			set_bit(__I40E_FD_FLUSH_REQUESTED, pf->state);
573 		}
574 
575 		/* filter programming failed most likely due to table full */
576 		fcnt_prog = i40e_get_global_fd_count(pf);
577 		fcnt_avail = pf->fdir_pf_filter_count;
578 		/* If ATR is running fcnt_prog can quickly change,
579 		 * if we are very close to full, it makes sense to disable
580 		 * FD ATR/SB and then re-enable it when there is room.
581 		 */
582 		if (fcnt_prog >= (fcnt_avail - I40E_FDIR_BUFFER_FULL_MARGIN)) {
583 			if ((pf->flags & I40E_FLAG_FD_SB_ENABLED) &&
584 			    !test_and_set_bit(__I40E_FD_SB_AUTO_DISABLED,
585 					      pf->state))
586 				if (I40E_DEBUG_FD & pf->hw.debug_mask)
587 					dev_warn(&pdev->dev, "FD filter space full, new ntuple rules will not be added\n");
588 		}
589 	} else if (error == BIT(I40E_RX_PROG_STATUS_DESC_NO_FD_ENTRY_SHIFT)) {
590 		if (I40E_DEBUG_FD & pf->hw.debug_mask)
591 			dev_info(&pdev->dev, "ntuple filter fd_id = %d, could not be removed\n",
592 				 rx_desc->wb.qword0.hi_dword.fd_id);
593 	}
594 }
595 
596 /**
597  * i40e_unmap_and_free_tx_resource - Release a Tx buffer
598  * @ring:      the ring that owns the buffer
599  * @tx_buffer: the buffer to free
600  **/
601 static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring,
602 					    struct i40e_tx_buffer *tx_buffer)
603 {
604 	if (tx_buffer->skb) {
605 		if (tx_buffer->tx_flags & I40E_TX_FLAGS_FD_SB)
606 			kfree(tx_buffer->raw_buf);
607 		else if (ring_is_xdp(ring))
608 			xdp_return_frame(tx_buffer->xdpf);
609 		else
610 			dev_kfree_skb_any(tx_buffer->skb);
611 		if (dma_unmap_len(tx_buffer, len))
612 			dma_unmap_single(ring->dev,
613 					 dma_unmap_addr(tx_buffer, dma),
614 					 dma_unmap_len(tx_buffer, len),
615 					 DMA_TO_DEVICE);
616 	} else if (dma_unmap_len(tx_buffer, len)) {
617 		dma_unmap_page(ring->dev,
618 			       dma_unmap_addr(tx_buffer, dma),
619 			       dma_unmap_len(tx_buffer, len),
620 			       DMA_TO_DEVICE);
621 	}
622 
623 	tx_buffer->next_to_watch = NULL;
624 	tx_buffer->skb = NULL;
625 	dma_unmap_len_set(tx_buffer, len, 0);
626 	/* tx_buffer must be completely set up in the transmit path */
627 }
628 
629 /**
630  * i40e_clean_tx_ring - Free any empty Tx buffers
631  * @tx_ring: ring to be cleaned
632  **/
633 void i40e_clean_tx_ring(struct i40e_ring *tx_ring)
634 {
635 	unsigned long bi_size;
636 	u16 i;
637 
638 	if (ring_is_xdp(tx_ring) && tx_ring->xsk_umem) {
639 		i40e_xsk_clean_tx_ring(tx_ring);
640 	} else {
641 		/* ring already cleared, nothing to do */
642 		if (!tx_ring->tx_bi)
643 			return;
644 
645 		/* Free all the Tx ring sk_buffs */
646 		for (i = 0; i < tx_ring->count; i++)
647 			i40e_unmap_and_free_tx_resource(tx_ring,
648 							&tx_ring->tx_bi[i]);
649 	}
650 
651 	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
652 	memset(tx_ring->tx_bi, 0, bi_size);
653 
654 	/* Zero out the descriptor ring */
655 	memset(tx_ring->desc, 0, tx_ring->size);
656 
657 	tx_ring->next_to_use = 0;
658 	tx_ring->next_to_clean = 0;
659 
660 	if (!tx_ring->netdev)
661 		return;
662 
663 	/* cleanup Tx queue statistics */
664 	netdev_tx_reset_queue(txring_txq(tx_ring));
665 }
666 
667 /**
668  * i40e_free_tx_resources - Free Tx resources per queue
669  * @tx_ring: Tx descriptor ring for a specific queue
670  *
671  * Free all transmit software resources
672  **/
673 void i40e_free_tx_resources(struct i40e_ring *tx_ring)
674 {
675 	i40e_clean_tx_ring(tx_ring);
676 	kfree(tx_ring->tx_bi);
677 	tx_ring->tx_bi = NULL;
678 
679 	if (tx_ring->desc) {
680 		dma_free_coherent(tx_ring->dev, tx_ring->size,
681 				  tx_ring->desc, tx_ring->dma);
682 		tx_ring->desc = NULL;
683 	}
684 }
685 
686 /**
687  * i40e_get_tx_pending - how many tx descriptors not processed
688  * @ring: the ring of descriptors
689  * @in_sw: use SW variables
690  *
691  * Since there is no access to the ring head register
692  * in XL710, we need to use our local copies
693  **/
694 u32 i40e_get_tx_pending(struct i40e_ring *ring, bool in_sw)
695 {
696 	u32 head, tail;
697 
698 	if (!in_sw) {
699 		head = i40e_get_head(ring);
700 		tail = readl(ring->tail);
701 	} else {
702 		head = ring->next_to_clean;
703 		tail = ring->next_to_use;
704 	}
705 
706 	if (head != tail)
707 		return (head < tail) ?
708 			tail - head : (tail + ring->count - head);
709 
710 	return 0;
711 }
712 
713 /**
714  * i40e_detect_recover_hung - Function to detect and recover hung_queues
715  * @vsi:  pointer to vsi struct with tx queues
716  *
717  * VSI has netdev and netdev has TX queues. This function is to check each of
718  * those TX queues if they are hung, trigger recovery by issuing SW interrupt.
719  **/
720 void i40e_detect_recover_hung(struct i40e_vsi *vsi)
721 {
722 	struct i40e_ring *tx_ring = NULL;
723 	struct net_device *netdev;
724 	unsigned int i;
725 	int packets;
726 
727 	if (!vsi)
728 		return;
729 
730 	if (test_bit(__I40E_VSI_DOWN, vsi->state))
731 		return;
732 
733 	netdev = vsi->netdev;
734 	if (!netdev)
735 		return;
736 
737 	if (!netif_carrier_ok(netdev))
738 		return;
739 
740 	for (i = 0; i < vsi->num_queue_pairs; i++) {
741 		tx_ring = vsi->tx_rings[i];
742 		if (tx_ring && tx_ring->desc) {
743 			/* If packet counter has not changed the queue is
744 			 * likely stalled, so force an interrupt for this
745 			 * queue.
746 			 *
747 			 * prev_pkt_ctr would be negative if there was no
748 			 * pending work.
749 			 */
750 			packets = tx_ring->stats.packets & INT_MAX;
751 			if (tx_ring->tx_stats.prev_pkt_ctr == packets) {
752 				i40e_force_wb(vsi, tx_ring->q_vector);
753 				continue;
754 			}
755 
756 			/* Memory barrier between read of packet count and call
757 			 * to i40e_get_tx_pending()
758 			 */
759 			smp_rmb();
760 			tx_ring->tx_stats.prev_pkt_ctr =
761 			    i40e_get_tx_pending(tx_ring, true) ? packets : -1;
762 		}
763 	}
764 }
765 
766 /**
767  * i40e_clean_tx_irq - Reclaim resources after transmit completes
768  * @vsi: the VSI we care about
769  * @tx_ring: Tx ring to clean
770  * @napi_budget: Used to determine if we are in netpoll
771  *
772  * Returns true if there's any budget left (e.g. the clean is finished)
773  **/
774 static bool i40e_clean_tx_irq(struct i40e_vsi *vsi,
775 			      struct i40e_ring *tx_ring, int napi_budget)
776 {
777 	u16 i = tx_ring->next_to_clean;
778 	struct i40e_tx_buffer *tx_buf;
779 	struct i40e_tx_desc *tx_head;
780 	struct i40e_tx_desc *tx_desc;
781 	unsigned int total_bytes = 0, total_packets = 0;
782 	unsigned int budget = vsi->work_limit;
783 
784 	tx_buf = &tx_ring->tx_bi[i];
785 	tx_desc = I40E_TX_DESC(tx_ring, i);
786 	i -= tx_ring->count;
787 
788 	tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
789 
790 	do {
791 		struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
792 
793 		/* if next_to_watch is not set then there is no work pending */
794 		if (!eop_desc)
795 			break;
796 
797 		/* prevent any other reads prior to eop_desc */
798 		smp_rmb();
799 
800 		i40e_trace(clean_tx_irq, tx_ring, tx_desc, tx_buf);
801 		/* we have caught up to head, no work left to do */
802 		if (tx_head == tx_desc)
803 			break;
804 
805 		/* clear next_to_watch to prevent false hangs */
806 		tx_buf->next_to_watch = NULL;
807 
808 		/* update the statistics for this packet */
809 		total_bytes += tx_buf->bytecount;
810 		total_packets += tx_buf->gso_segs;
811 
812 		/* free the skb/XDP data */
813 		if (ring_is_xdp(tx_ring))
814 			xdp_return_frame(tx_buf->xdpf);
815 		else
816 			napi_consume_skb(tx_buf->skb, napi_budget);
817 
818 		/* unmap skb header data */
819 		dma_unmap_single(tx_ring->dev,
820 				 dma_unmap_addr(tx_buf, dma),
821 				 dma_unmap_len(tx_buf, len),
822 				 DMA_TO_DEVICE);
823 
824 		/* clear tx_buffer data */
825 		tx_buf->skb = NULL;
826 		dma_unmap_len_set(tx_buf, len, 0);
827 
828 		/* unmap remaining buffers */
829 		while (tx_desc != eop_desc) {
830 			i40e_trace(clean_tx_irq_unmap,
831 				   tx_ring, tx_desc, tx_buf);
832 
833 			tx_buf++;
834 			tx_desc++;
835 			i++;
836 			if (unlikely(!i)) {
837 				i -= tx_ring->count;
838 				tx_buf = tx_ring->tx_bi;
839 				tx_desc = I40E_TX_DESC(tx_ring, 0);
840 			}
841 
842 			/* unmap any remaining paged data */
843 			if (dma_unmap_len(tx_buf, len)) {
844 				dma_unmap_page(tx_ring->dev,
845 					       dma_unmap_addr(tx_buf, dma),
846 					       dma_unmap_len(tx_buf, len),
847 					       DMA_TO_DEVICE);
848 				dma_unmap_len_set(tx_buf, len, 0);
849 			}
850 		}
851 
852 		/* move us one more past the eop_desc for start of next pkt */
853 		tx_buf++;
854 		tx_desc++;
855 		i++;
856 		if (unlikely(!i)) {
857 			i -= tx_ring->count;
858 			tx_buf = tx_ring->tx_bi;
859 			tx_desc = I40E_TX_DESC(tx_ring, 0);
860 		}
861 
862 		prefetch(tx_desc);
863 
864 		/* update budget accounting */
865 		budget--;
866 	} while (likely(budget));
867 
868 	i += tx_ring->count;
869 	tx_ring->next_to_clean = i;
870 	i40e_update_tx_stats(tx_ring, total_packets, total_bytes);
871 	i40e_arm_wb(tx_ring, vsi, budget);
872 
873 	if (ring_is_xdp(tx_ring))
874 		return !!budget;
875 
876 	/* notify netdev of completed buffers */
877 	netdev_tx_completed_queue(txring_txq(tx_ring),
878 				  total_packets, total_bytes);
879 
880 #define TX_WAKE_THRESHOLD ((s16)(DESC_NEEDED * 2))
881 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
882 		     (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
883 		/* Make sure that anybody stopping the queue after this
884 		 * sees the new next_to_clean.
885 		 */
886 		smp_mb();
887 		if (__netif_subqueue_stopped(tx_ring->netdev,
888 					     tx_ring->queue_index) &&
889 		   !test_bit(__I40E_VSI_DOWN, vsi->state)) {
890 			netif_wake_subqueue(tx_ring->netdev,
891 					    tx_ring->queue_index);
892 			++tx_ring->tx_stats.restart_queue;
893 		}
894 	}
895 
896 	return !!budget;
897 }
898 
899 /**
900  * i40e_enable_wb_on_itr - Arm hardware to do a wb, interrupts are not enabled
901  * @vsi: the VSI we care about
902  * @q_vector: the vector on which to enable writeback
903  *
904  **/
905 static void i40e_enable_wb_on_itr(struct i40e_vsi *vsi,
906 				  struct i40e_q_vector *q_vector)
907 {
908 	u16 flags = q_vector->tx.ring[0].flags;
909 	u32 val;
910 
911 	if (!(flags & I40E_TXR_FLAGS_WB_ON_ITR))
912 		return;
913 
914 	if (q_vector->arm_wb_state)
915 		return;
916 
917 	if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
918 		val = I40E_PFINT_DYN_CTLN_WB_ON_ITR_MASK |
919 		      I40E_PFINT_DYN_CTLN_ITR_INDX_MASK; /* set noitr */
920 
921 		wr32(&vsi->back->hw,
922 		     I40E_PFINT_DYN_CTLN(q_vector->reg_idx),
923 		     val);
924 	} else {
925 		val = I40E_PFINT_DYN_CTL0_WB_ON_ITR_MASK |
926 		      I40E_PFINT_DYN_CTL0_ITR_INDX_MASK; /* set noitr */
927 
928 		wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
929 	}
930 	q_vector->arm_wb_state = true;
931 }
932 
933 /**
934  * i40e_force_wb - Issue SW Interrupt so HW does a wb
935  * @vsi: the VSI we care about
936  * @q_vector: the vector  on which to force writeback
937  *
938  **/
939 void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector)
940 {
941 	if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
942 		u32 val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
943 			  I40E_PFINT_DYN_CTLN_ITR_INDX_MASK | /* set noitr */
944 			  I40E_PFINT_DYN_CTLN_SWINT_TRIG_MASK |
945 			  I40E_PFINT_DYN_CTLN_SW_ITR_INDX_ENA_MASK;
946 			  /* allow 00 to be written to the index */
947 
948 		wr32(&vsi->back->hw,
949 		     I40E_PFINT_DYN_CTLN(q_vector->reg_idx), val);
950 	} else {
951 		u32 val = I40E_PFINT_DYN_CTL0_INTENA_MASK |
952 			  I40E_PFINT_DYN_CTL0_ITR_INDX_MASK | /* set noitr */
953 			  I40E_PFINT_DYN_CTL0_SWINT_TRIG_MASK |
954 			  I40E_PFINT_DYN_CTL0_SW_ITR_INDX_ENA_MASK;
955 			/* allow 00 to be written to the index */
956 
957 		wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
958 	}
959 }
960 
961 static inline bool i40e_container_is_rx(struct i40e_q_vector *q_vector,
962 					struct i40e_ring_container *rc)
963 {
964 	return &q_vector->rx == rc;
965 }
966 
967 static inline unsigned int i40e_itr_divisor(struct i40e_q_vector *q_vector)
968 {
969 	unsigned int divisor;
970 
971 	switch (q_vector->vsi->back->hw.phy.link_info.link_speed) {
972 	case I40E_LINK_SPEED_40GB:
973 		divisor = I40E_ITR_ADAPTIVE_MIN_INC * 1024;
974 		break;
975 	case I40E_LINK_SPEED_25GB:
976 	case I40E_LINK_SPEED_20GB:
977 		divisor = I40E_ITR_ADAPTIVE_MIN_INC * 512;
978 		break;
979 	default:
980 	case I40E_LINK_SPEED_10GB:
981 		divisor = I40E_ITR_ADAPTIVE_MIN_INC * 256;
982 		break;
983 	case I40E_LINK_SPEED_1GB:
984 	case I40E_LINK_SPEED_100MB:
985 		divisor = I40E_ITR_ADAPTIVE_MIN_INC * 32;
986 		break;
987 	}
988 
989 	return divisor;
990 }
991 
992 /**
993  * i40e_update_itr - update the dynamic ITR value based on statistics
994  * @q_vector: structure containing interrupt and ring information
995  * @rc: structure containing ring performance data
996  *
997  * Stores a new ITR value based on packets and byte
998  * counts during the last interrupt.  The advantage of per interrupt
999  * computation is faster updates and more accurate ITR for the current
1000  * traffic pattern.  Constants in this function were computed
1001  * based on theoretical maximum wire speed and thresholds were set based
1002  * on testing data as well as attempting to minimize response time
1003  * while increasing bulk throughput.
1004  **/
1005 static void i40e_update_itr(struct i40e_q_vector *q_vector,
1006 			    struct i40e_ring_container *rc)
1007 {
1008 	unsigned int avg_wire_size, packets, bytes, itr;
1009 	unsigned long next_update = jiffies;
1010 
1011 	/* If we don't have any rings just leave ourselves set for maximum
1012 	 * possible latency so we take ourselves out of the equation.
1013 	 */
1014 	if (!rc->ring || !ITR_IS_DYNAMIC(rc->ring->itr_setting))
1015 		return;
1016 
1017 	/* For Rx we want to push the delay up and default to low latency.
1018 	 * for Tx we want to pull the delay down and default to high latency.
1019 	 */
1020 	itr = i40e_container_is_rx(q_vector, rc) ?
1021 	      I40E_ITR_ADAPTIVE_MIN_USECS | I40E_ITR_ADAPTIVE_LATENCY :
1022 	      I40E_ITR_ADAPTIVE_MAX_USECS | I40E_ITR_ADAPTIVE_LATENCY;
1023 
1024 	/* If we didn't update within up to 1 - 2 jiffies we can assume
1025 	 * that either packets are coming in so slow there hasn't been
1026 	 * any work, or that there is so much work that NAPI is dealing
1027 	 * with interrupt moderation and we don't need to do anything.
1028 	 */
1029 	if (time_after(next_update, rc->next_update))
1030 		goto clear_counts;
1031 
1032 	/* If itr_countdown is set it means we programmed an ITR within
1033 	 * the last 4 interrupt cycles. This has a side effect of us
1034 	 * potentially firing an early interrupt. In order to work around
1035 	 * this we need to throw out any data received for a few
1036 	 * interrupts following the update.
1037 	 */
1038 	if (q_vector->itr_countdown) {
1039 		itr = rc->target_itr;
1040 		goto clear_counts;
1041 	}
1042 
1043 	packets = rc->total_packets;
1044 	bytes = rc->total_bytes;
1045 
1046 	if (i40e_container_is_rx(q_vector, rc)) {
1047 		/* If Rx there are 1 to 4 packets and bytes are less than
1048 		 * 9000 assume insufficient data to use bulk rate limiting
1049 		 * approach unless Tx is already in bulk rate limiting. We
1050 		 * are likely latency driven.
1051 		 */
1052 		if (packets && packets < 4 && bytes < 9000 &&
1053 		    (q_vector->tx.target_itr & I40E_ITR_ADAPTIVE_LATENCY)) {
1054 			itr = I40E_ITR_ADAPTIVE_LATENCY;
1055 			goto adjust_by_size;
1056 		}
1057 	} else if (packets < 4) {
1058 		/* If we have Tx and Rx ITR maxed and Tx ITR is running in
1059 		 * bulk mode and we are receiving 4 or fewer packets just
1060 		 * reset the ITR_ADAPTIVE_LATENCY bit for latency mode so
1061 		 * that the Rx can relax.
1062 		 */
1063 		if (rc->target_itr == I40E_ITR_ADAPTIVE_MAX_USECS &&
1064 		    (q_vector->rx.target_itr & I40E_ITR_MASK) ==
1065 		     I40E_ITR_ADAPTIVE_MAX_USECS)
1066 			goto clear_counts;
1067 	} else if (packets > 32) {
1068 		/* If we have processed over 32 packets in a single interrupt
1069 		 * for Tx assume we need to switch over to "bulk" mode.
1070 		 */
1071 		rc->target_itr &= ~I40E_ITR_ADAPTIVE_LATENCY;
1072 	}
1073 
1074 	/* We have no packets to actually measure against. This means
1075 	 * either one of the other queues on this vector is active or
1076 	 * we are a Tx queue doing TSO with too high of an interrupt rate.
1077 	 *
1078 	 * Between 4 and 56 we can assume that our current interrupt delay
1079 	 * is only slightly too low. As such we should increase it by a small
1080 	 * fixed amount.
1081 	 */
1082 	if (packets < 56) {
1083 		itr = rc->target_itr + I40E_ITR_ADAPTIVE_MIN_INC;
1084 		if ((itr & I40E_ITR_MASK) > I40E_ITR_ADAPTIVE_MAX_USECS) {
1085 			itr &= I40E_ITR_ADAPTIVE_LATENCY;
1086 			itr += I40E_ITR_ADAPTIVE_MAX_USECS;
1087 		}
1088 		goto clear_counts;
1089 	}
1090 
1091 	if (packets <= 256) {
1092 		itr = min(q_vector->tx.current_itr, q_vector->rx.current_itr);
1093 		itr &= I40E_ITR_MASK;
1094 
1095 		/* Between 56 and 112 is our "goldilocks" zone where we are
1096 		 * working out "just right". Just report that our current
1097 		 * ITR is good for us.
1098 		 */
1099 		if (packets <= 112)
1100 			goto clear_counts;
1101 
1102 		/* If packet count is 128 or greater we are likely looking
1103 		 * at a slight overrun of the delay we want. Try halving
1104 		 * our delay to see if that will cut the number of packets
1105 		 * in half per interrupt.
1106 		 */
1107 		itr /= 2;
1108 		itr &= I40E_ITR_MASK;
1109 		if (itr < I40E_ITR_ADAPTIVE_MIN_USECS)
1110 			itr = I40E_ITR_ADAPTIVE_MIN_USECS;
1111 
1112 		goto clear_counts;
1113 	}
1114 
1115 	/* The paths below assume we are dealing with a bulk ITR since
1116 	 * number of packets is greater than 256. We are just going to have
1117 	 * to compute a value and try to bring the count under control,
1118 	 * though for smaller packet sizes there isn't much we can do as
1119 	 * NAPI polling will likely be kicking in sooner rather than later.
1120 	 */
1121 	itr = I40E_ITR_ADAPTIVE_BULK;
1122 
1123 adjust_by_size:
1124 	/* If packet counts are 256 or greater we can assume we have a gross
1125 	 * overestimation of what the rate should be. Instead of trying to fine
1126 	 * tune it just use the formula below to try and dial in an exact value
1127 	 * give the current packet size of the frame.
1128 	 */
1129 	avg_wire_size = bytes / packets;
1130 
1131 	/* The following is a crude approximation of:
1132 	 *  wmem_default / (size + overhead) = desired_pkts_per_int
1133 	 *  rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
1134 	 *  (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
1135 	 *
1136 	 * Assuming wmem_default is 212992 and overhead is 640 bytes per
1137 	 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
1138 	 * formula down to
1139 	 *
1140 	 *  (170 * (size + 24)) / (size + 640) = ITR
1141 	 *
1142 	 * We first do some math on the packet size and then finally bitshift
1143 	 * by 8 after rounding up. We also have to account for PCIe link speed
1144 	 * difference as ITR scales based on this.
1145 	 */
1146 	if (avg_wire_size <= 60) {
1147 		/* Start at 250k ints/sec */
1148 		avg_wire_size = 4096;
1149 	} else if (avg_wire_size <= 380) {
1150 		/* 250K ints/sec to 60K ints/sec */
1151 		avg_wire_size *= 40;
1152 		avg_wire_size += 1696;
1153 	} else if (avg_wire_size <= 1084) {
1154 		/* 60K ints/sec to 36K ints/sec */
1155 		avg_wire_size *= 15;
1156 		avg_wire_size += 11452;
1157 	} else if (avg_wire_size <= 1980) {
1158 		/* 36K ints/sec to 30K ints/sec */
1159 		avg_wire_size *= 5;
1160 		avg_wire_size += 22420;
1161 	} else {
1162 		/* plateau at a limit of 30K ints/sec */
1163 		avg_wire_size = 32256;
1164 	}
1165 
1166 	/* If we are in low latency mode halve our delay which doubles the
1167 	 * rate to somewhere between 100K to 16K ints/sec
1168 	 */
1169 	if (itr & I40E_ITR_ADAPTIVE_LATENCY)
1170 		avg_wire_size /= 2;
1171 
1172 	/* Resultant value is 256 times larger than it needs to be. This
1173 	 * gives us room to adjust the value as needed to either increase
1174 	 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
1175 	 *
1176 	 * Use addition as we have already recorded the new latency flag
1177 	 * for the ITR value.
1178 	 */
1179 	itr += DIV_ROUND_UP(avg_wire_size, i40e_itr_divisor(q_vector)) *
1180 	       I40E_ITR_ADAPTIVE_MIN_INC;
1181 
1182 	if ((itr & I40E_ITR_MASK) > I40E_ITR_ADAPTIVE_MAX_USECS) {
1183 		itr &= I40E_ITR_ADAPTIVE_LATENCY;
1184 		itr += I40E_ITR_ADAPTIVE_MAX_USECS;
1185 	}
1186 
1187 clear_counts:
1188 	/* write back value */
1189 	rc->target_itr = itr;
1190 
1191 	/* next update should occur within next jiffy */
1192 	rc->next_update = next_update + 1;
1193 
1194 	rc->total_bytes = 0;
1195 	rc->total_packets = 0;
1196 }
1197 
1198 /**
1199  * i40e_reuse_rx_page - page flip buffer and store it back on the ring
1200  * @rx_ring: rx descriptor ring to store buffers on
1201  * @old_buff: donor buffer to have page reused
1202  *
1203  * Synchronizes page for reuse by the adapter
1204  **/
1205 static void i40e_reuse_rx_page(struct i40e_ring *rx_ring,
1206 			       struct i40e_rx_buffer *old_buff)
1207 {
1208 	struct i40e_rx_buffer *new_buff;
1209 	u16 nta = rx_ring->next_to_alloc;
1210 
1211 	new_buff = &rx_ring->rx_bi[nta];
1212 
1213 	/* update, and store next to alloc */
1214 	nta++;
1215 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1216 
1217 	/* transfer page from old buffer to new buffer */
1218 	new_buff->dma		= old_buff->dma;
1219 	new_buff->page		= old_buff->page;
1220 	new_buff->page_offset	= old_buff->page_offset;
1221 	new_buff->pagecnt_bias	= old_buff->pagecnt_bias;
1222 
1223 	rx_ring->rx_stats.page_reuse_count++;
1224 
1225 	/* clear contents of buffer_info */
1226 	old_buff->page = NULL;
1227 }
1228 
1229 /**
1230  * i40e_rx_is_programming_status - check for programming status descriptor
1231  * @qw: qword representing status_error_len in CPU ordering
1232  *
1233  * The value of in the descriptor length field indicate if this
1234  * is a programming status descriptor for flow director or FCoE
1235  * by the value of I40E_RX_PROG_STATUS_DESC_LENGTH, otherwise
1236  * it is a packet descriptor.
1237  **/
1238 static inline bool i40e_rx_is_programming_status(u64 qw)
1239 {
1240 	/* The Rx filter programming status and SPH bit occupy the same
1241 	 * spot in the descriptor. Since we don't support packet split we
1242 	 * can just reuse the bit as an indication that this is a
1243 	 * programming status descriptor.
1244 	 */
1245 	return qw & I40E_RXD_QW1_LENGTH_SPH_MASK;
1246 }
1247 
1248 /**
1249  * i40e_clean_programming_status - try clean the programming status descriptor
1250  * @rx_ring: the rx ring that has this descriptor
1251  * @rx_desc: the rx descriptor written back by HW
1252  * @qw: qword representing status_error_len in CPU ordering
1253  *
1254  * Flow director should handle FD_FILTER_STATUS to check its filter programming
1255  * status being successful or not and take actions accordingly. FCoE should
1256  * handle its context/filter programming/invalidation status and take actions.
1257  *
1258  * Returns an i40e_rx_buffer to reuse if the cleanup occurred, otherwise NULL.
1259  **/
1260 struct i40e_rx_buffer *i40e_clean_programming_status(
1261 	struct i40e_ring *rx_ring,
1262 	union i40e_rx_desc *rx_desc,
1263 	u64 qw)
1264 {
1265 	struct i40e_rx_buffer *rx_buffer;
1266 	u32 ntc;
1267 	u8 id;
1268 
1269 	if (!i40e_rx_is_programming_status(qw))
1270 		return NULL;
1271 
1272 	ntc = rx_ring->next_to_clean;
1273 
1274 	/* fetch, update, and store next to clean */
1275 	rx_buffer = &rx_ring->rx_bi[ntc++];
1276 	ntc = (ntc < rx_ring->count) ? ntc : 0;
1277 	rx_ring->next_to_clean = ntc;
1278 
1279 	prefetch(I40E_RX_DESC(rx_ring, ntc));
1280 
1281 	id = (qw & I40E_RX_PROG_STATUS_DESC_QW1_PROGID_MASK) >>
1282 		  I40E_RX_PROG_STATUS_DESC_QW1_PROGID_SHIFT;
1283 
1284 	if (id == I40E_RX_PROG_STATUS_DESC_FD_FILTER_STATUS)
1285 		i40e_fd_handle_status(rx_ring, rx_desc, id);
1286 
1287 	return rx_buffer;
1288 }
1289 
1290 /**
1291  * i40e_setup_tx_descriptors - Allocate the Tx descriptors
1292  * @tx_ring: the tx ring to set up
1293  *
1294  * Return 0 on success, negative on error
1295  **/
1296 int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring)
1297 {
1298 	struct device *dev = tx_ring->dev;
1299 	int bi_size;
1300 
1301 	if (!dev)
1302 		return -ENOMEM;
1303 
1304 	/* warn if we are about to overwrite the pointer */
1305 	WARN_ON(tx_ring->tx_bi);
1306 	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
1307 	tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
1308 	if (!tx_ring->tx_bi)
1309 		goto err;
1310 
1311 	u64_stats_init(&tx_ring->syncp);
1312 
1313 	/* round up to nearest 4K */
1314 	tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
1315 	/* add u32 for head writeback, align after this takes care of
1316 	 * guaranteeing this is at least one cache line in size
1317 	 */
1318 	tx_ring->size += sizeof(u32);
1319 	tx_ring->size = ALIGN(tx_ring->size, 4096);
1320 	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
1321 					   &tx_ring->dma, GFP_KERNEL);
1322 	if (!tx_ring->desc) {
1323 		dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
1324 			 tx_ring->size);
1325 		goto err;
1326 	}
1327 
1328 	tx_ring->next_to_use = 0;
1329 	tx_ring->next_to_clean = 0;
1330 	tx_ring->tx_stats.prev_pkt_ctr = -1;
1331 	return 0;
1332 
1333 err:
1334 	kfree(tx_ring->tx_bi);
1335 	tx_ring->tx_bi = NULL;
1336 	return -ENOMEM;
1337 }
1338 
1339 /**
1340  * i40e_clean_rx_ring - Free Rx buffers
1341  * @rx_ring: ring to be cleaned
1342  **/
1343 void i40e_clean_rx_ring(struct i40e_ring *rx_ring)
1344 {
1345 	unsigned long bi_size;
1346 	u16 i;
1347 
1348 	/* ring already cleared, nothing to do */
1349 	if (!rx_ring->rx_bi)
1350 		return;
1351 
1352 	if (rx_ring->skb) {
1353 		dev_kfree_skb(rx_ring->skb);
1354 		rx_ring->skb = NULL;
1355 	}
1356 
1357 	if (rx_ring->xsk_umem) {
1358 		i40e_xsk_clean_rx_ring(rx_ring);
1359 		goto skip_free;
1360 	}
1361 
1362 	/* Free all the Rx ring sk_buffs */
1363 	for (i = 0; i < rx_ring->count; i++) {
1364 		struct i40e_rx_buffer *rx_bi = &rx_ring->rx_bi[i];
1365 
1366 		if (!rx_bi->page)
1367 			continue;
1368 
1369 		/* Invalidate cache lines that may have been written to by
1370 		 * device so that we avoid corrupting memory.
1371 		 */
1372 		dma_sync_single_range_for_cpu(rx_ring->dev,
1373 					      rx_bi->dma,
1374 					      rx_bi->page_offset,
1375 					      rx_ring->rx_buf_len,
1376 					      DMA_FROM_DEVICE);
1377 
1378 		/* free resources associated with mapping */
1379 		dma_unmap_page_attrs(rx_ring->dev, rx_bi->dma,
1380 				     i40e_rx_pg_size(rx_ring),
1381 				     DMA_FROM_DEVICE,
1382 				     I40E_RX_DMA_ATTR);
1383 
1384 		__page_frag_cache_drain(rx_bi->page, rx_bi->pagecnt_bias);
1385 
1386 		rx_bi->page = NULL;
1387 		rx_bi->page_offset = 0;
1388 	}
1389 
1390 skip_free:
1391 	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
1392 	memset(rx_ring->rx_bi, 0, bi_size);
1393 
1394 	/* Zero out the descriptor ring */
1395 	memset(rx_ring->desc, 0, rx_ring->size);
1396 
1397 	rx_ring->next_to_alloc = 0;
1398 	rx_ring->next_to_clean = 0;
1399 	rx_ring->next_to_use = 0;
1400 }
1401 
1402 /**
1403  * i40e_free_rx_resources - Free Rx resources
1404  * @rx_ring: ring to clean the resources from
1405  *
1406  * Free all receive software resources
1407  **/
1408 void i40e_free_rx_resources(struct i40e_ring *rx_ring)
1409 {
1410 	i40e_clean_rx_ring(rx_ring);
1411 	if (rx_ring->vsi->type == I40E_VSI_MAIN)
1412 		xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
1413 	rx_ring->xdp_prog = NULL;
1414 	kfree(rx_ring->rx_bi);
1415 	rx_ring->rx_bi = NULL;
1416 
1417 	if (rx_ring->desc) {
1418 		dma_free_coherent(rx_ring->dev, rx_ring->size,
1419 				  rx_ring->desc, rx_ring->dma);
1420 		rx_ring->desc = NULL;
1421 	}
1422 }
1423 
1424 /**
1425  * i40e_setup_rx_descriptors - Allocate Rx descriptors
1426  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
1427  *
1428  * Returns 0 on success, negative on failure
1429  **/
1430 int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring)
1431 {
1432 	struct device *dev = rx_ring->dev;
1433 	int err = -ENOMEM;
1434 	int bi_size;
1435 
1436 	/* warn if we are about to overwrite the pointer */
1437 	WARN_ON(rx_ring->rx_bi);
1438 	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
1439 	rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
1440 	if (!rx_ring->rx_bi)
1441 		goto err;
1442 
1443 	u64_stats_init(&rx_ring->syncp);
1444 
1445 	/* Round up to nearest 4K */
1446 	rx_ring->size = rx_ring->count * sizeof(union i40e_32byte_rx_desc);
1447 	rx_ring->size = ALIGN(rx_ring->size, 4096);
1448 	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
1449 					   &rx_ring->dma, GFP_KERNEL);
1450 
1451 	if (!rx_ring->desc) {
1452 		dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
1453 			 rx_ring->size);
1454 		goto err;
1455 	}
1456 
1457 	rx_ring->next_to_alloc = 0;
1458 	rx_ring->next_to_clean = 0;
1459 	rx_ring->next_to_use = 0;
1460 
1461 	/* XDP RX-queue info only needed for RX rings exposed to XDP */
1462 	if (rx_ring->vsi->type == I40E_VSI_MAIN) {
1463 		err = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev,
1464 				       rx_ring->queue_index);
1465 		if (err < 0)
1466 			goto err;
1467 	}
1468 
1469 	rx_ring->xdp_prog = rx_ring->vsi->xdp_prog;
1470 
1471 	return 0;
1472 err:
1473 	kfree(rx_ring->rx_bi);
1474 	rx_ring->rx_bi = NULL;
1475 	return err;
1476 }
1477 
1478 /**
1479  * i40e_release_rx_desc - Store the new tail and head values
1480  * @rx_ring: ring to bump
1481  * @val: new head index
1482  **/
1483 void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
1484 {
1485 	rx_ring->next_to_use = val;
1486 
1487 	/* update next to alloc since we have filled the ring */
1488 	rx_ring->next_to_alloc = val;
1489 
1490 	/* Force memory writes to complete before letting h/w
1491 	 * know there are new descriptors to fetch.  (Only
1492 	 * applicable for weak-ordered memory model archs,
1493 	 * such as IA-64).
1494 	 */
1495 	wmb();
1496 	writel(val, rx_ring->tail);
1497 }
1498 
1499 /**
1500  * i40e_rx_offset - Return expected offset into page to access data
1501  * @rx_ring: Ring we are requesting offset of
1502  *
1503  * Returns the offset value for ring into the data buffer.
1504  */
1505 static inline unsigned int i40e_rx_offset(struct i40e_ring *rx_ring)
1506 {
1507 	return ring_uses_build_skb(rx_ring) ? I40E_SKB_PAD : 0;
1508 }
1509 
1510 /**
1511  * i40e_alloc_mapped_page - recycle or make a new page
1512  * @rx_ring: ring to use
1513  * @bi: rx_buffer struct to modify
1514  *
1515  * Returns true if the page was successfully allocated or
1516  * reused.
1517  **/
1518 static bool i40e_alloc_mapped_page(struct i40e_ring *rx_ring,
1519 				   struct i40e_rx_buffer *bi)
1520 {
1521 	struct page *page = bi->page;
1522 	dma_addr_t dma;
1523 
1524 	/* since we are recycling buffers we should seldom need to alloc */
1525 	if (likely(page)) {
1526 		rx_ring->rx_stats.page_reuse_count++;
1527 		return true;
1528 	}
1529 
1530 	/* alloc new page for storage */
1531 	page = dev_alloc_pages(i40e_rx_pg_order(rx_ring));
1532 	if (unlikely(!page)) {
1533 		rx_ring->rx_stats.alloc_page_failed++;
1534 		return false;
1535 	}
1536 
1537 	/* map page for use */
1538 	dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1539 				 i40e_rx_pg_size(rx_ring),
1540 				 DMA_FROM_DEVICE,
1541 				 I40E_RX_DMA_ATTR);
1542 
1543 	/* if mapping failed free memory back to system since
1544 	 * there isn't much point in holding memory we can't use
1545 	 */
1546 	if (dma_mapping_error(rx_ring->dev, dma)) {
1547 		__free_pages(page, i40e_rx_pg_order(rx_ring));
1548 		rx_ring->rx_stats.alloc_page_failed++;
1549 		return false;
1550 	}
1551 
1552 	bi->dma = dma;
1553 	bi->page = page;
1554 	bi->page_offset = i40e_rx_offset(rx_ring);
1555 	page_ref_add(page, USHRT_MAX - 1);
1556 	bi->pagecnt_bias = USHRT_MAX;
1557 
1558 	return true;
1559 }
1560 
1561 /**
1562  * i40e_receive_skb - Send a completed packet up the stack
1563  * @rx_ring:  rx ring in play
1564  * @skb: packet to send up
1565  * @vlan_tag: vlan tag for packet
1566  **/
1567 void i40e_receive_skb(struct i40e_ring *rx_ring,
1568 		      struct sk_buff *skb, u16 vlan_tag)
1569 {
1570 	struct i40e_q_vector *q_vector = rx_ring->q_vector;
1571 
1572 	if ((rx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1573 	    (vlan_tag & VLAN_VID_MASK))
1574 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
1575 
1576 	napi_gro_receive(&q_vector->napi, skb);
1577 }
1578 
1579 /**
1580  * i40e_alloc_rx_buffers - Replace used receive buffers
1581  * @rx_ring: ring to place buffers on
1582  * @cleaned_count: number of buffers to replace
1583  *
1584  * Returns false if all allocations were successful, true if any fail
1585  **/
1586 bool i40e_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count)
1587 {
1588 	u16 ntu = rx_ring->next_to_use;
1589 	union i40e_rx_desc *rx_desc;
1590 	struct i40e_rx_buffer *bi;
1591 
1592 	/* do nothing if no valid netdev defined */
1593 	if (!rx_ring->netdev || !cleaned_count)
1594 		return false;
1595 
1596 	rx_desc = I40E_RX_DESC(rx_ring, ntu);
1597 	bi = &rx_ring->rx_bi[ntu];
1598 
1599 	do {
1600 		if (!i40e_alloc_mapped_page(rx_ring, bi))
1601 			goto no_buffers;
1602 
1603 		/* sync the buffer for use by the device */
1604 		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1605 						 bi->page_offset,
1606 						 rx_ring->rx_buf_len,
1607 						 DMA_FROM_DEVICE);
1608 
1609 		/* Refresh the desc even if buffer_addrs didn't change
1610 		 * because each write-back erases this info.
1611 		 */
1612 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1613 
1614 		rx_desc++;
1615 		bi++;
1616 		ntu++;
1617 		if (unlikely(ntu == rx_ring->count)) {
1618 			rx_desc = I40E_RX_DESC(rx_ring, 0);
1619 			bi = rx_ring->rx_bi;
1620 			ntu = 0;
1621 		}
1622 
1623 		/* clear the status bits for the next_to_use descriptor */
1624 		rx_desc->wb.qword1.status_error_len = 0;
1625 
1626 		cleaned_count--;
1627 	} while (cleaned_count);
1628 
1629 	if (rx_ring->next_to_use != ntu)
1630 		i40e_release_rx_desc(rx_ring, ntu);
1631 
1632 	return false;
1633 
1634 no_buffers:
1635 	if (rx_ring->next_to_use != ntu)
1636 		i40e_release_rx_desc(rx_ring, ntu);
1637 
1638 	/* make sure to come back via polling to try again after
1639 	 * allocation failure
1640 	 */
1641 	return true;
1642 }
1643 
1644 /**
1645  * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
1646  * @vsi: the VSI we care about
1647  * @skb: skb currently being received and modified
1648  * @rx_desc: the receive descriptor
1649  **/
1650 static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
1651 				    struct sk_buff *skb,
1652 				    union i40e_rx_desc *rx_desc)
1653 {
1654 	struct i40e_rx_ptype_decoded decoded;
1655 	u32 rx_error, rx_status;
1656 	bool ipv4, ipv6;
1657 	u8 ptype;
1658 	u64 qword;
1659 
1660 	qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1661 	ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >> I40E_RXD_QW1_PTYPE_SHIFT;
1662 	rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1663 		   I40E_RXD_QW1_ERROR_SHIFT;
1664 	rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1665 		    I40E_RXD_QW1_STATUS_SHIFT;
1666 	decoded = decode_rx_desc_ptype(ptype);
1667 
1668 	skb->ip_summed = CHECKSUM_NONE;
1669 
1670 	skb_checksum_none_assert(skb);
1671 
1672 	/* Rx csum enabled and ip headers found? */
1673 	if (!(vsi->netdev->features & NETIF_F_RXCSUM))
1674 		return;
1675 
1676 	/* did the hardware decode the packet and checksum? */
1677 	if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
1678 		return;
1679 
1680 	/* both known and outer_ip must be set for the below code to work */
1681 	if (!(decoded.known && decoded.outer_ip))
1682 		return;
1683 
1684 	ipv4 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1685 	       (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4);
1686 	ipv6 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1687 	       (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6);
1688 
1689 	if (ipv4 &&
1690 	    (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) |
1691 			 BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT))))
1692 		goto checksum_fail;
1693 
1694 	/* likely incorrect csum if alternate IP extension headers found */
1695 	if (ipv6 &&
1696 	    rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
1697 		/* don't increment checksum err here, non-fatal err */
1698 		return;
1699 
1700 	/* there was some L4 error, count error and punt packet to the stack */
1701 	if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT))
1702 		goto checksum_fail;
1703 
1704 	/* handle packets that were not able to be checksummed due
1705 	 * to arrival speed, in this case the stack can compute
1706 	 * the csum.
1707 	 */
1708 	if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT))
1709 		return;
1710 
1711 	/* If there is an outer header present that might contain a checksum
1712 	 * we need to bump the checksum level by 1 to reflect the fact that
1713 	 * we are indicating we validated the inner checksum.
1714 	 */
1715 	if (decoded.tunnel_type >= I40E_RX_PTYPE_TUNNEL_IP_GRENAT)
1716 		skb->csum_level = 1;
1717 
1718 	/* Only report checksum unnecessary for TCP, UDP, or SCTP */
1719 	switch (decoded.inner_prot) {
1720 	case I40E_RX_PTYPE_INNER_PROT_TCP:
1721 	case I40E_RX_PTYPE_INNER_PROT_UDP:
1722 	case I40E_RX_PTYPE_INNER_PROT_SCTP:
1723 		skb->ip_summed = CHECKSUM_UNNECESSARY;
1724 		/* fall though */
1725 	default:
1726 		break;
1727 	}
1728 
1729 	return;
1730 
1731 checksum_fail:
1732 	vsi->back->hw_csum_rx_error++;
1733 }
1734 
1735 /**
1736  * i40e_ptype_to_htype - get a hash type
1737  * @ptype: the ptype value from the descriptor
1738  *
1739  * Returns a hash type to be used by skb_set_hash
1740  **/
1741 static inline int i40e_ptype_to_htype(u8 ptype)
1742 {
1743 	struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
1744 
1745 	if (!decoded.known)
1746 		return PKT_HASH_TYPE_NONE;
1747 
1748 	if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1749 	    decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
1750 		return PKT_HASH_TYPE_L4;
1751 	else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1752 		 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
1753 		return PKT_HASH_TYPE_L3;
1754 	else
1755 		return PKT_HASH_TYPE_L2;
1756 }
1757 
1758 /**
1759  * i40e_rx_hash - set the hash value in the skb
1760  * @ring: descriptor ring
1761  * @rx_desc: specific descriptor
1762  * @skb: skb currently being received and modified
1763  * @rx_ptype: Rx packet type
1764  **/
1765 static inline void i40e_rx_hash(struct i40e_ring *ring,
1766 				union i40e_rx_desc *rx_desc,
1767 				struct sk_buff *skb,
1768 				u8 rx_ptype)
1769 {
1770 	u32 hash;
1771 	const __le64 rss_mask =
1772 		cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
1773 			    I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
1774 
1775 	if (!(ring->netdev->features & NETIF_F_RXHASH))
1776 		return;
1777 
1778 	if ((rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask) {
1779 		hash = le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
1780 		skb_set_hash(skb, hash, i40e_ptype_to_htype(rx_ptype));
1781 	}
1782 }
1783 
1784 /**
1785  * i40e_process_skb_fields - Populate skb header fields from Rx descriptor
1786  * @rx_ring: rx descriptor ring packet is being transacted on
1787  * @rx_desc: pointer to the EOP Rx descriptor
1788  * @skb: pointer to current skb being populated
1789  * @rx_ptype: the packet type decoded by hardware
1790  *
1791  * This function checks the ring, descriptor, and packet information in
1792  * order to populate the hash, checksum, VLAN, protocol, and
1793  * other fields within the skb.
1794  **/
1795 void i40e_process_skb_fields(struct i40e_ring *rx_ring,
1796 			     union i40e_rx_desc *rx_desc, struct sk_buff *skb,
1797 			     u8 rx_ptype)
1798 {
1799 	u64 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1800 	u32 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1801 			I40E_RXD_QW1_STATUS_SHIFT;
1802 	u32 tsynvalid = rx_status & I40E_RXD_QW1_STATUS_TSYNVALID_MASK;
1803 	u32 tsyn = (rx_status & I40E_RXD_QW1_STATUS_TSYNINDX_MASK) >>
1804 		   I40E_RXD_QW1_STATUS_TSYNINDX_SHIFT;
1805 
1806 	if (unlikely(tsynvalid))
1807 		i40e_ptp_rx_hwtstamp(rx_ring->vsi->back, skb, tsyn);
1808 
1809 	i40e_rx_hash(rx_ring, rx_desc, skb, rx_ptype);
1810 
1811 	i40e_rx_checksum(rx_ring->vsi, skb, rx_desc);
1812 
1813 	skb_record_rx_queue(skb, rx_ring->queue_index);
1814 
1815 	/* modifies the skb - consumes the enet header */
1816 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1817 }
1818 
1819 /**
1820  * i40e_cleanup_headers - Correct empty headers
1821  * @rx_ring: rx descriptor ring packet is being transacted on
1822  * @skb: pointer to current skb being fixed
1823  * @rx_desc: pointer to the EOP Rx descriptor
1824  *
1825  * Also address the case where we are pulling data in on pages only
1826  * and as such no data is present in the skb header.
1827  *
1828  * In addition if skb is not at least 60 bytes we need to pad it so that
1829  * it is large enough to qualify as a valid Ethernet frame.
1830  *
1831  * Returns true if an error was encountered and skb was freed.
1832  **/
1833 static bool i40e_cleanup_headers(struct i40e_ring *rx_ring, struct sk_buff *skb,
1834 				 union i40e_rx_desc *rx_desc)
1835 
1836 {
1837 	/* XDP packets use error pointer so abort at this point */
1838 	if (IS_ERR(skb))
1839 		return true;
1840 
1841 	/* ERR_MASK will only have valid bits if EOP set, and
1842 	 * what we are doing here is actually checking
1843 	 * I40E_RX_DESC_ERROR_RXE_SHIFT, since it is the zeroth bit in
1844 	 * the error field
1845 	 */
1846 	if (unlikely(i40e_test_staterr(rx_desc,
1847 				       BIT(I40E_RXD_QW1_ERROR_SHIFT)))) {
1848 		dev_kfree_skb_any(skb);
1849 		return true;
1850 	}
1851 
1852 	/* if eth_skb_pad returns an error the skb was freed */
1853 	if (eth_skb_pad(skb))
1854 		return true;
1855 
1856 	return false;
1857 }
1858 
1859 /**
1860  * i40e_page_is_reusable - check if any reuse is possible
1861  * @page: page struct to check
1862  *
1863  * A page is not reusable if it was allocated under low memory
1864  * conditions, or it's not in the same NUMA node as this CPU.
1865  */
1866 static inline bool i40e_page_is_reusable(struct page *page)
1867 {
1868 	return (page_to_nid(page) == numa_mem_id()) &&
1869 		!page_is_pfmemalloc(page);
1870 }
1871 
1872 /**
1873  * i40e_can_reuse_rx_page - Determine if this page can be reused by
1874  * the adapter for another receive
1875  *
1876  * @rx_buffer: buffer containing the page
1877  *
1878  * If page is reusable, rx_buffer->page_offset is adjusted to point to
1879  * an unused region in the page.
1880  *
1881  * For small pages, @truesize will be a constant value, half the size
1882  * of the memory at page.  We'll attempt to alternate between high and
1883  * low halves of the page, with one half ready for use by the hardware
1884  * and the other half being consumed by the stack.  We use the page
1885  * ref count to determine whether the stack has finished consuming the
1886  * portion of this page that was passed up with a previous packet.  If
1887  * the page ref count is >1, we'll assume the "other" half page is
1888  * still busy, and this page cannot be reused.
1889  *
1890  * For larger pages, @truesize will be the actual space used by the
1891  * received packet (adjusted upward to an even multiple of the cache
1892  * line size).  This will advance through the page by the amount
1893  * actually consumed by the received packets while there is still
1894  * space for a buffer.  Each region of larger pages will be used at
1895  * most once, after which the page will not be reused.
1896  *
1897  * In either case, if the page is reusable its refcount is increased.
1898  **/
1899 static bool i40e_can_reuse_rx_page(struct i40e_rx_buffer *rx_buffer)
1900 {
1901 	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1902 	struct page *page = rx_buffer->page;
1903 
1904 	/* Is any reuse possible? */
1905 	if (unlikely(!i40e_page_is_reusable(page)))
1906 		return false;
1907 
1908 #if (PAGE_SIZE < 8192)
1909 	/* if we are only owner of page we can reuse it */
1910 	if (unlikely((page_count(page) - pagecnt_bias) > 1))
1911 		return false;
1912 #else
1913 #define I40E_LAST_OFFSET \
1914 	(SKB_WITH_OVERHEAD(PAGE_SIZE) - I40E_RXBUFFER_2048)
1915 	if (rx_buffer->page_offset > I40E_LAST_OFFSET)
1916 		return false;
1917 #endif
1918 
1919 	/* If we have drained the page fragment pool we need to update
1920 	 * the pagecnt_bias and page count so that we fully restock the
1921 	 * number of references the driver holds.
1922 	 */
1923 	if (unlikely(pagecnt_bias == 1)) {
1924 		page_ref_add(page, USHRT_MAX - 1);
1925 		rx_buffer->pagecnt_bias = USHRT_MAX;
1926 	}
1927 
1928 	return true;
1929 }
1930 
1931 /**
1932  * i40e_add_rx_frag - Add contents of Rx buffer to sk_buff
1933  * @rx_ring: rx descriptor ring to transact packets on
1934  * @rx_buffer: buffer containing page to add
1935  * @skb: sk_buff to place the data into
1936  * @size: packet length from rx_desc
1937  *
1938  * This function will add the data contained in rx_buffer->page to the skb.
1939  * It will just attach the page as a frag to the skb.
1940  *
1941  * The function will then update the page offset.
1942  **/
1943 static void i40e_add_rx_frag(struct i40e_ring *rx_ring,
1944 			     struct i40e_rx_buffer *rx_buffer,
1945 			     struct sk_buff *skb,
1946 			     unsigned int size)
1947 {
1948 #if (PAGE_SIZE < 8192)
1949 	unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
1950 #else
1951 	unsigned int truesize = SKB_DATA_ALIGN(size + i40e_rx_offset(rx_ring));
1952 #endif
1953 
1954 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
1955 			rx_buffer->page_offset, size, truesize);
1956 
1957 	/* page is being used so we must update the page offset */
1958 #if (PAGE_SIZE < 8192)
1959 	rx_buffer->page_offset ^= truesize;
1960 #else
1961 	rx_buffer->page_offset += truesize;
1962 #endif
1963 }
1964 
1965 /**
1966  * i40e_get_rx_buffer - Fetch Rx buffer and synchronize data for use
1967  * @rx_ring: rx descriptor ring to transact packets on
1968  * @size: size of buffer to add to skb
1969  *
1970  * This function will pull an Rx buffer from the ring and synchronize it
1971  * for use by the CPU.
1972  */
1973 static struct i40e_rx_buffer *i40e_get_rx_buffer(struct i40e_ring *rx_ring,
1974 						 const unsigned int size)
1975 {
1976 	struct i40e_rx_buffer *rx_buffer;
1977 
1978 	rx_buffer = &rx_ring->rx_bi[rx_ring->next_to_clean];
1979 	prefetchw(rx_buffer->page);
1980 
1981 	/* we are reusing so sync this buffer for CPU use */
1982 	dma_sync_single_range_for_cpu(rx_ring->dev,
1983 				      rx_buffer->dma,
1984 				      rx_buffer->page_offset,
1985 				      size,
1986 				      DMA_FROM_DEVICE);
1987 
1988 	/* We have pulled a buffer for use, so decrement pagecnt_bias */
1989 	rx_buffer->pagecnt_bias--;
1990 
1991 	return rx_buffer;
1992 }
1993 
1994 /**
1995  * i40e_construct_skb - Allocate skb and populate it
1996  * @rx_ring: rx descriptor ring to transact packets on
1997  * @rx_buffer: rx buffer to pull data from
1998  * @xdp: xdp_buff pointing to the data
1999  *
2000  * This function allocates an skb.  It then populates it with the page
2001  * data from the current receive descriptor, taking care to set up the
2002  * skb correctly.
2003  */
2004 static struct sk_buff *i40e_construct_skb(struct i40e_ring *rx_ring,
2005 					  struct i40e_rx_buffer *rx_buffer,
2006 					  struct xdp_buff *xdp)
2007 {
2008 	unsigned int size = xdp->data_end - xdp->data;
2009 #if (PAGE_SIZE < 8192)
2010 	unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
2011 #else
2012 	unsigned int truesize = SKB_DATA_ALIGN(size);
2013 #endif
2014 	unsigned int headlen;
2015 	struct sk_buff *skb;
2016 
2017 	/* prefetch first cache line of first page */
2018 	prefetch(xdp->data);
2019 #if L1_CACHE_BYTES < 128
2020 	prefetch(xdp->data + L1_CACHE_BYTES);
2021 #endif
2022 	/* Note, we get here by enabling legacy-rx via:
2023 	 *
2024 	 *    ethtool --set-priv-flags <dev> legacy-rx on
2025 	 *
2026 	 * In this mode, we currently get 0 extra XDP headroom as
2027 	 * opposed to having legacy-rx off, where we process XDP
2028 	 * packets going to stack via i40e_build_skb(). The latter
2029 	 * provides us currently with 192 bytes of headroom.
2030 	 *
2031 	 * For i40e_construct_skb() mode it means that the
2032 	 * xdp->data_meta will always point to xdp->data, since
2033 	 * the helper cannot expand the head. Should this ever
2034 	 * change in future for legacy-rx mode on, then lets also
2035 	 * add xdp->data_meta handling here.
2036 	 */
2037 
2038 	/* allocate a skb to store the frags */
2039 	skb = __napi_alloc_skb(&rx_ring->q_vector->napi,
2040 			       I40E_RX_HDR_SIZE,
2041 			       GFP_ATOMIC | __GFP_NOWARN);
2042 	if (unlikely(!skb))
2043 		return NULL;
2044 
2045 	/* Determine available headroom for copy */
2046 	headlen = size;
2047 	if (headlen > I40E_RX_HDR_SIZE)
2048 		headlen = eth_get_headlen(xdp->data, I40E_RX_HDR_SIZE);
2049 
2050 	/* align pull length to size of long to optimize memcpy performance */
2051 	memcpy(__skb_put(skb, headlen), xdp->data,
2052 	       ALIGN(headlen, sizeof(long)));
2053 
2054 	/* update all of the pointers */
2055 	size -= headlen;
2056 	if (size) {
2057 		skb_add_rx_frag(skb, 0, rx_buffer->page,
2058 				rx_buffer->page_offset + headlen,
2059 				size, truesize);
2060 
2061 		/* buffer is used by skb, update page_offset */
2062 #if (PAGE_SIZE < 8192)
2063 		rx_buffer->page_offset ^= truesize;
2064 #else
2065 		rx_buffer->page_offset += truesize;
2066 #endif
2067 	} else {
2068 		/* buffer is unused, reset bias back to rx_buffer */
2069 		rx_buffer->pagecnt_bias++;
2070 	}
2071 
2072 	return skb;
2073 }
2074 
2075 /**
2076  * i40e_build_skb - Build skb around an existing buffer
2077  * @rx_ring: Rx descriptor ring to transact packets on
2078  * @rx_buffer: Rx buffer to pull data from
2079  * @xdp: xdp_buff pointing to the data
2080  *
2081  * This function builds an skb around an existing Rx buffer, taking care
2082  * to set up the skb correctly and avoid any memcpy overhead.
2083  */
2084 static struct sk_buff *i40e_build_skb(struct i40e_ring *rx_ring,
2085 				      struct i40e_rx_buffer *rx_buffer,
2086 				      struct xdp_buff *xdp)
2087 {
2088 	unsigned int metasize = xdp->data - xdp->data_meta;
2089 #if (PAGE_SIZE < 8192)
2090 	unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
2091 #else
2092 	unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
2093 				SKB_DATA_ALIGN(xdp->data_end -
2094 					       xdp->data_hard_start);
2095 #endif
2096 	struct sk_buff *skb;
2097 
2098 	/* Prefetch first cache line of first page. If xdp->data_meta
2099 	 * is unused, this points exactly as xdp->data, otherwise we
2100 	 * likely have a consumer accessing first few bytes of meta
2101 	 * data, and then actual data.
2102 	 */
2103 	prefetch(xdp->data_meta);
2104 #if L1_CACHE_BYTES < 128
2105 	prefetch(xdp->data_meta + L1_CACHE_BYTES);
2106 #endif
2107 	/* build an skb around the page buffer */
2108 	skb = build_skb(xdp->data_hard_start, truesize);
2109 	if (unlikely(!skb))
2110 		return NULL;
2111 
2112 	/* update pointers within the skb to store the data */
2113 	skb_reserve(skb, xdp->data - xdp->data_hard_start);
2114 	__skb_put(skb, xdp->data_end - xdp->data);
2115 	if (metasize)
2116 		skb_metadata_set(skb, metasize);
2117 
2118 	/* buffer is used by skb, update page_offset */
2119 #if (PAGE_SIZE < 8192)
2120 	rx_buffer->page_offset ^= truesize;
2121 #else
2122 	rx_buffer->page_offset += truesize;
2123 #endif
2124 
2125 	return skb;
2126 }
2127 
2128 /**
2129  * i40e_put_rx_buffer - Clean up used buffer and either recycle or free
2130  * @rx_ring: rx descriptor ring to transact packets on
2131  * @rx_buffer: rx buffer to pull data from
2132  *
2133  * This function will clean up the contents of the rx_buffer.  It will
2134  * either recycle the buffer or unmap it and free the associated resources.
2135  */
2136 static void i40e_put_rx_buffer(struct i40e_ring *rx_ring,
2137 			       struct i40e_rx_buffer *rx_buffer)
2138 {
2139 	if (i40e_can_reuse_rx_page(rx_buffer)) {
2140 		/* hand second half of page back to the ring */
2141 		i40e_reuse_rx_page(rx_ring, rx_buffer);
2142 	} else {
2143 		/* we are not reusing the buffer so unmap it */
2144 		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2145 				     i40e_rx_pg_size(rx_ring),
2146 				     DMA_FROM_DEVICE, I40E_RX_DMA_ATTR);
2147 		__page_frag_cache_drain(rx_buffer->page,
2148 					rx_buffer->pagecnt_bias);
2149 		/* clear contents of buffer_info */
2150 		rx_buffer->page = NULL;
2151 	}
2152 }
2153 
2154 /**
2155  * i40e_is_non_eop - process handling of non-EOP buffers
2156  * @rx_ring: Rx ring being processed
2157  * @rx_desc: Rx descriptor for current buffer
2158  * @skb: Current socket buffer containing buffer in progress
2159  *
2160  * This function updates next to clean.  If the buffer is an EOP buffer
2161  * this function exits returning false, otherwise it will place the
2162  * sk_buff in the next buffer to be chained and return true indicating
2163  * that this is in fact a non-EOP buffer.
2164  **/
2165 static bool i40e_is_non_eop(struct i40e_ring *rx_ring,
2166 			    union i40e_rx_desc *rx_desc,
2167 			    struct sk_buff *skb)
2168 {
2169 	u32 ntc = rx_ring->next_to_clean + 1;
2170 
2171 	/* fetch, update, and store next to clean */
2172 	ntc = (ntc < rx_ring->count) ? ntc : 0;
2173 	rx_ring->next_to_clean = ntc;
2174 
2175 	prefetch(I40E_RX_DESC(rx_ring, ntc));
2176 
2177 	/* if we are the last buffer then there is nothing else to do */
2178 #define I40E_RXD_EOF BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)
2179 	if (likely(i40e_test_staterr(rx_desc, I40E_RXD_EOF)))
2180 		return false;
2181 
2182 	rx_ring->rx_stats.non_eop_descs++;
2183 
2184 	return true;
2185 }
2186 
2187 static int i40e_xmit_xdp_ring(struct xdp_frame *xdpf,
2188 			      struct i40e_ring *xdp_ring);
2189 
2190 int i40e_xmit_xdp_tx_ring(struct xdp_buff *xdp, struct i40e_ring *xdp_ring)
2191 {
2192 	struct xdp_frame *xdpf = convert_to_xdp_frame(xdp);
2193 
2194 	if (unlikely(!xdpf))
2195 		return I40E_XDP_CONSUMED;
2196 
2197 	return i40e_xmit_xdp_ring(xdpf, xdp_ring);
2198 }
2199 
2200 /**
2201  * i40e_run_xdp - run an XDP program
2202  * @rx_ring: Rx ring being processed
2203  * @xdp: XDP buffer containing the frame
2204  **/
2205 static struct sk_buff *i40e_run_xdp(struct i40e_ring *rx_ring,
2206 				    struct xdp_buff *xdp)
2207 {
2208 	int err, result = I40E_XDP_PASS;
2209 	struct i40e_ring *xdp_ring;
2210 	struct bpf_prog *xdp_prog;
2211 	u32 act;
2212 
2213 	rcu_read_lock();
2214 	xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2215 
2216 	if (!xdp_prog)
2217 		goto xdp_out;
2218 
2219 	prefetchw(xdp->data_hard_start); /* xdp_frame write */
2220 
2221 	act = bpf_prog_run_xdp(xdp_prog, xdp);
2222 	switch (act) {
2223 	case XDP_PASS:
2224 		break;
2225 	case XDP_TX:
2226 		xdp_ring = rx_ring->vsi->xdp_rings[rx_ring->queue_index];
2227 		result = i40e_xmit_xdp_tx_ring(xdp, xdp_ring);
2228 		break;
2229 	case XDP_REDIRECT:
2230 		err = xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog);
2231 		result = !err ? I40E_XDP_REDIR : I40E_XDP_CONSUMED;
2232 		break;
2233 	default:
2234 		bpf_warn_invalid_xdp_action(act);
2235 		/* fall through */
2236 	case XDP_ABORTED:
2237 		trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2238 		/* fall through -- handle aborts by dropping packet */
2239 	case XDP_DROP:
2240 		result = I40E_XDP_CONSUMED;
2241 		break;
2242 	}
2243 xdp_out:
2244 	rcu_read_unlock();
2245 	return ERR_PTR(-result);
2246 }
2247 
2248 /**
2249  * i40e_rx_buffer_flip - adjusted rx_buffer to point to an unused region
2250  * @rx_ring: Rx ring
2251  * @rx_buffer: Rx buffer to adjust
2252  * @size: Size of adjustment
2253  **/
2254 static void i40e_rx_buffer_flip(struct i40e_ring *rx_ring,
2255 				struct i40e_rx_buffer *rx_buffer,
2256 				unsigned int size)
2257 {
2258 #if (PAGE_SIZE < 8192)
2259 	unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
2260 
2261 	rx_buffer->page_offset ^= truesize;
2262 #else
2263 	unsigned int truesize = SKB_DATA_ALIGN(i40e_rx_offset(rx_ring) + size);
2264 
2265 	rx_buffer->page_offset += truesize;
2266 #endif
2267 }
2268 
2269 /**
2270  * i40e_xdp_ring_update_tail - Updates the XDP Tx ring tail register
2271  * @xdp_ring: XDP Tx ring
2272  *
2273  * This function updates the XDP Tx ring tail register.
2274  **/
2275 void i40e_xdp_ring_update_tail(struct i40e_ring *xdp_ring)
2276 {
2277 	/* Force memory writes to complete before letting h/w
2278 	 * know there are new descriptors to fetch.
2279 	 */
2280 	wmb();
2281 	writel_relaxed(xdp_ring->next_to_use, xdp_ring->tail);
2282 }
2283 
2284 /**
2285  * i40e_update_rx_stats - Update Rx ring statistics
2286  * @rx_ring: rx descriptor ring
2287  * @total_rx_bytes: number of bytes received
2288  * @total_rx_packets: number of packets received
2289  *
2290  * This function updates the Rx ring statistics.
2291  **/
2292 void i40e_update_rx_stats(struct i40e_ring *rx_ring,
2293 			  unsigned int total_rx_bytes,
2294 			  unsigned int total_rx_packets)
2295 {
2296 	u64_stats_update_begin(&rx_ring->syncp);
2297 	rx_ring->stats.packets += total_rx_packets;
2298 	rx_ring->stats.bytes += total_rx_bytes;
2299 	u64_stats_update_end(&rx_ring->syncp);
2300 	rx_ring->q_vector->rx.total_packets += total_rx_packets;
2301 	rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
2302 }
2303 
2304 /**
2305  * i40e_finalize_xdp_rx - Bump XDP Tx tail and/or flush redirect map
2306  * @rx_ring: Rx ring
2307  * @xdp_res: Result of the receive batch
2308  *
2309  * This function bumps XDP Tx tail and/or flush redirect map, and
2310  * should be called when a batch of packets has been processed in the
2311  * napi loop.
2312  **/
2313 void i40e_finalize_xdp_rx(struct i40e_ring *rx_ring, unsigned int xdp_res)
2314 {
2315 	if (xdp_res & I40E_XDP_REDIR)
2316 		xdp_do_flush_map();
2317 
2318 	if (xdp_res & I40E_XDP_TX) {
2319 		struct i40e_ring *xdp_ring =
2320 			rx_ring->vsi->xdp_rings[rx_ring->queue_index];
2321 
2322 		i40e_xdp_ring_update_tail(xdp_ring);
2323 	}
2324 }
2325 
2326 /**
2327  * i40e_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2328  * @rx_ring: rx descriptor ring to transact packets on
2329  * @budget: Total limit on number of packets to process
2330  *
2331  * This function provides a "bounce buffer" approach to Rx interrupt
2332  * processing.  The advantage to this is that on systems that have
2333  * expensive overhead for IOMMU access this provides a means of avoiding
2334  * it by maintaining the mapping of the page to the system.
2335  *
2336  * Returns amount of work completed
2337  **/
2338 static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget)
2339 {
2340 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2341 	struct sk_buff *skb = rx_ring->skb;
2342 	u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
2343 	unsigned int xdp_xmit = 0;
2344 	bool failure = false;
2345 	struct xdp_buff xdp;
2346 
2347 	xdp.rxq = &rx_ring->xdp_rxq;
2348 
2349 	while (likely(total_rx_packets < (unsigned int)budget)) {
2350 		struct i40e_rx_buffer *rx_buffer;
2351 		union i40e_rx_desc *rx_desc;
2352 		unsigned int size;
2353 		u16 vlan_tag;
2354 		u8 rx_ptype;
2355 		u64 qword;
2356 
2357 		/* return some buffers to hardware, one at a time is too slow */
2358 		if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
2359 			failure = failure ||
2360 				  i40e_alloc_rx_buffers(rx_ring, cleaned_count);
2361 			cleaned_count = 0;
2362 		}
2363 
2364 		rx_desc = I40E_RX_DESC(rx_ring, rx_ring->next_to_clean);
2365 
2366 		/* status_error_len will always be zero for unused descriptors
2367 		 * because it's cleared in cleanup, and overlaps with hdr_addr
2368 		 * which is always zero because packet split isn't used, if the
2369 		 * hardware wrote DD then the length will be non-zero
2370 		 */
2371 		qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
2372 
2373 		/* This memory barrier is needed to keep us from reading
2374 		 * any other fields out of the rx_desc until we have
2375 		 * verified the descriptor has been written back.
2376 		 */
2377 		dma_rmb();
2378 
2379 		rx_buffer = i40e_clean_programming_status(rx_ring, rx_desc,
2380 							  qword);
2381 		if (unlikely(rx_buffer)) {
2382 			i40e_reuse_rx_page(rx_ring, rx_buffer);
2383 			cleaned_count++;
2384 			continue;
2385 		}
2386 
2387 		size = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
2388 		       I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
2389 		if (!size)
2390 			break;
2391 
2392 		i40e_trace(clean_rx_irq, rx_ring, rx_desc, skb);
2393 		rx_buffer = i40e_get_rx_buffer(rx_ring, size);
2394 
2395 		/* retrieve a buffer from the ring */
2396 		if (!skb) {
2397 			xdp.data = page_address(rx_buffer->page) +
2398 				   rx_buffer->page_offset;
2399 			xdp.data_meta = xdp.data;
2400 			xdp.data_hard_start = xdp.data -
2401 					      i40e_rx_offset(rx_ring);
2402 			xdp.data_end = xdp.data + size;
2403 
2404 			skb = i40e_run_xdp(rx_ring, &xdp);
2405 		}
2406 
2407 		if (IS_ERR(skb)) {
2408 			unsigned int xdp_res = -PTR_ERR(skb);
2409 
2410 			if (xdp_res & (I40E_XDP_TX | I40E_XDP_REDIR)) {
2411 				xdp_xmit |= xdp_res;
2412 				i40e_rx_buffer_flip(rx_ring, rx_buffer, size);
2413 			} else {
2414 				rx_buffer->pagecnt_bias++;
2415 			}
2416 			total_rx_bytes += size;
2417 			total_rx_packets++;
2418 		} else if (skb) {
2419 			i40e_add_rx_frag(rx_ring, rx_buffer, skb, size);
2420 		} else if (ring_uses_build_skb(rx_ring)) {
2421 			skb = i40e_build_skb(rx_ring, rx_buffer, &xdp);
2422 		} else {
2423 			skb = i40e_construct_skb(rx_ring, rx_buffer, &xdp);
2424 		}
2425 
2426 		/* exit if we failed to retrieve a buffer */
2427 		if (!skb) {
2428 			rx_ring->rx_stats.alloc_buff_failed++;
2429 			rx_buffer->pagecnt_bias++;
2430 			break;
2431 		}
2432 
2433 		i40e_put_rx_buffer(rx_ring, rx_buffer);
2434 		cleaned_count++;
2435 
2436 		if (i40e_is_non_eop(rx_ring, rx_desc, skb))
2437 			continue;
2438 
2439 		if (i40e_cleanup_headers(rx_ring, skb, rx_desc)) {
2440 			skb = NULL;
2441 			continue;
2442 		}
2443 
2444 		/* probably a little skewed due to removing CRC */
2445 		total_rx_bytes += skb->len;
2446 
2447 		qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
2448 		rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
2449 			   I40E_RXD_QW1_PTYPE_SHIFT;
2450 
2451 		/* populate checksum, VLAN, and protocol */
2452 		i40e_process_skb_fields(rx_ring, rx_desc, skb, rx_ptype);
2453 
2454 		vlan_tag = (qword & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)) ?
2455 			   le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1) : 0;
2456 
2457 		i40e_trace(clean_rx_irq_rx, rx_ring, rx_desc, skb);
2458 		i40e_receive_skb(rx_ring, skb, vlan_tag);
2459 		skb = NULL;
2460 
2461 		/* update budget accounting */
2462 		total_rx_packets++;
2463 	}
2464 
2465 	i40e_finalize_xdp_rx(rx_ring, xdp_xmit);
2466 	rx_ring->skb = skb;
2467 
2468 	i40e_update_rx_stats(rx_ring, total_rx_bytes, total_rx_packets);
2469 
2470 	/* guarantee a trip back through this routine if there was a failure */
2471 	return failure ? budget : (int)total_rx_packets;
2472 }
2473 
2474 static inline u32 i40e_buildreg_itr(const int type, u16 itr)
2475 {
2476 	u32 val;
2477 
2478 	/* We don't bother with setting the CLEARPBA bit as the data sheet
2479 	 * points out doing so is "meaningless since it was already
2480 	 * auto-cleared". The auto-clearing happens when the interrupt is
2481 	 * asserted.
2482 	 *
2483 	 * Hardware errata 28 for also indicates that writing to a
2484 	 * xxINT_DYN_CTLx CSR with INTENA_MSK (bit 31) set to 0 will clear
2485 	 * an event in the PBA anyway so we need to rely on the automask
2486 	 * to hold pending events for us until the interrupt is re-enabled
2487 	 *
2488 	 * The itr value is reported in microseconds, and the register
2489 	 * value is recorded in 2 microsecond units. For this reason we
2490 	 * only need to shift by the interval shift - 1 instead of the
2491 	 * full value.
2492 	 */
2493 	itr &= I40E_ITR_MASK;
2494 
2495 	val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
2496 	      (type << I40E_PFINT_DYN_CTLN_ITR_INDX_SHIFT) |
2497 	      (itr << (I40E_PFINT_DYN_CTLN_INTERVAL_SHIFT - 1));
2498 
2499 	return val;
2500 }
2501 
2502 /* a small macro to shorten up some long lines */
2503 #define INTREG I40E_PFINT_DYN_CTLN
2504 
2505 /* The act of updating the ITR will cause it to immediately trigger. In order
2506  * to prevent this from throwing off adaptive update statistics we defer the
2507  * update so that it can only happen so often. So after either Tx or Rx are
2508  * updated we make the adaptive scheme wait until either the ITR completely
2509  * expires via the next_update expiration or we have been through at least
2510  * 3 interrupts.
2511  */
2512 #define ITR_COUNTDOWN_START 3
2513 
2514 /**
2515  * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt
2516  * @vsi: the VSI we care about
2517  * @q_vector: q_vector for which itr is being updated and interrupt enabled
2518  *
2519  **/
2520 static inline void i40e_update_enable_itr(struct i40e_vsi *vsi,
2521 					  struct i40e_q_vector *q_vector)
2522 {
2523 	struct i40e_hw *hw = &vsi->back->hw;
2524 	u32 intval;
2525 
2526 	/* If we don't have MSIX, then we only need to re-enable icr0 */
2527 	if (!(vsi->back->flags & I40E_FLAG_MSIX_ENABLED)) {
2528 		i40e_irq_dynamic_enable_icr0(vsi->back);
2529 		return;
2530 	}
2531 
2532 	/* These will do nothing if dynamic updates are not enabled */
2533 	i40e_update_itr(q_vector, &q_vector->tx);
2534 	i40e_update_itr(q_vector, &q_vector->rx);
2535 
2536 	/* This block of logic allows us to get away with only updating
2537 	 * one ITR value with each interrupt. The idea is to perform a
2538 	 * pseudo-lazy update with the following criteria.
2539 	 *
2540 	 * 1. Rx is given higher priority than Tx if both are in same state
2541 	 * 2. If we must reduce an ITR that is given highest priority.
2542 	 * 3. We then give priority to increasing ITR based on amount.
2543 	 */
2544 	if (q_vector->rx.target_itr < q_vector->rx.current_itr) {
2545 		/* Rx ITR needs to be reduced, this is highest priority */
2546 		intval = i40e_buildreg_itr(I40E_RX_ITR,
2547 					   q_vector->rx.target_itr);
2548 		q_vector->rx.current_itr = q_vector->rx.target_itr;
2549 		q_vector->itr_countdown = ITR_COUNTDOWN_START;
2550 	} else if ((q_vector->tx.target_itr < q_vector->tx.current_itr) ||
2551 		   ((q_vector->rx.target_itr - q_vector->rx.current_itr) <
2552 		    (q_vector->tx.target_itr - q_vector->tx.current_itr))) {
2553 		/* Tx ITR needs to be reduced, this is second priority
2554 		 * Tx ITR needs to be increased more than Rx, fourth priority
2555 		 */
2556 		intval = i40e_buildreg_itr(I40E_TX_ITR,
2557 					   q_vector->tx.target_itr);
2558 		q_vector->tx.current_itr = q_vector->tx.target_itr;
2559 		q_vector->itr_countdown = ITR_COUNTDOWN_START;
2560 	} else if (q_vector->rx.current_itr != q_vector->rx.target_itr) {
2561 		/* Rx ITR needs to be increased, third priority */
2562 		intval = i40e_buildreg_itr(I40E_RX_ITR,
2563 					   q_vector->rx.target_itr);
2564 		q_vector->rx.current_itr = q_vector->rx.target_itr;
2565 		q_vector->itr_countdown = ITR_COUNTDOWN_START;
2566 	} else {
2567 		/* No ITR update, lowest priority */
2568 		intval = i40e_buildreg_itr(I40E_ITR_NONE, 0);
2569 		if (q_vector->itr_countdown)
2570 			q_vector->itr_countdown--;
2571 	}
2572 
2573 	if (!test_bit(__I40E_VSI_DOWN, vsi->state))
2574 		wr32(hw, INTREG(q_vector->reg_idx), intval);
2575 }
2576 
2577 /**
2578  * i40e_napi_poll - NAPI polling Rx/Tx cleanup routine
2579  * @napi: napi struct with our devices info in it
2580  * @budget: amount of work driver is allowed to do this pass, in packets
2581  *
2582  * This function will clean all queues associated with a q_vector.
2583  *
2584  * Returns the amount of work done
2585  **/
2586 int i40e_napi_poll(struct napi_struct *napi, int budget)
2587 {
2588 	struct i40e_q_vector *q_vector =
2589 			       container_of(napi, struct i40e_q_vector, napi);
2590 	struct i40e_vsi *vsi = q_vector->vsi;
2591 	struct i40e_ring *ring;
2592 	bool clean_complete = true;
2593 	bool arm_wb = false;
2594 	int budget_per_ring;
2595 	int work_done = 0;
2596 
2597 	if (test_bit(__I40E_VSI_DOWN, vsi->state)) {
2598 		napi_complete(napi);
2599 		return 0;
2600 	}
2601 
2602 	/* Since the actual Tx work is minimal, we can give the Tx a larger
2603 	 * budget and be more aggressive about cleaning up the Tx descriptors.
2604 	 */
2605 	i40e_for_each_ring(ring, q_vector->tx) {
2606 		bool wd = ring->xsk_umem ?
2607 			  i40e_clean_xdp_tx_irq(vsi, ring, budget) :
2608 			  i40e_clean_tx_irq(vsi, ring, budget);
2609 
2610 		if (!wd) {
2611 			clean_complete = false;
2612 			continue;
2613 		}
2614 		arm_wb |= ring->arm_wb;
2615 		ring->arm_wb = false;
2616 	}
2617 
2618 	/* Handle case where we are called by netpoll with a budget of 0 */
2619 	if (budget <= 0)
2620 		goto tx_only;
2621 
2622 	/* We attempt to distribute budget to each Rx queue fairly, but don't
2623 	 * allow the budget to go below 1 because that would exit polling early.
2624 	 */
2625 	budget_per_ring = max(budget/q_vector->num_ringpairs, 1);
2626 
2627 	i40e_for_each_ring(ring, q_vector->rx) {
2628 		int cleaned = ring->xsk_umem ?
2629 			      i40e_clean_rx_irq_zc(ring, budget_per_ring) :
2630 			      i40e_clean_rx_irq(ring, budget_per_ring);
2631 
2632 		work_done += cleaned;
2633 		/* if we clean as many as budgeted, we must not be done */
2634 		if (cleaned >= budget_per_ring)
2635 			clean_complete = false;
2636 	}
2637 
2638 	/* If work not completed, return budget and polling will return */
2639 	if (!clean_complete) {
2640 		int cpu_id = smp_processor_id();
2641 
2642 		/* It is possible that the interrupt affinity has changed but,
2643 		 * if the cpu is pegged at 100%, polling will never exit while
2644 		 * traffic continues and the interrupt will be stuck on this
2645 		 * cpu.  We check to make sure affinity is correct before we
2646 		 * continue to poll, otherwise we must stop polling so the
2647 		 * interrupt can move to the correct cpu.
2648 		 */
2649 		if (!cpumask_test_cpu(cpu_id, &q_vector->affinity_mask)) {
2650 			/* Tell napi that we are done polling */
2651 			napi_complete_done(napi, work_done);
2652 
2653 			/* Force an interrupt */
2654 			i40e_force_wb(vsi, q_vector);
2655 
2656 			/* Return budget-1 so that polling stops */
2657 			return budget - 1;
2658 		}
2659 tx_only:
2660 		if (arm_wb) {
2661 			q_vector->tx.ring[0].tx_stats.tx_force_wb++;
2662 			i40e_enable_wb_on_itr(vsi, q_vector);
2663 		}
2664 		return budget;
2665 	}
2666 
2667 	if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR)
2668 		q_vector->arm_wb_state = false;
2669 
2670 	/* Work is done so exit the polling mode and re-enable the interrupt */
2671 	napi_complete_done(napi, work_done);
2672 
2673 	i40e_update_enable_itr(vsi, q_vector);
2674 
2675 	return min(work_done, budget - 1);
2676 }
2677 
2678 /**
2679  * i40e_atr - Add a Flow Director ATR filter
2680  * @tx_ring:  ring to add programming descriptor to
2681  * @skb:      send buffer
2682  * @tx_flags: send tx flags
2683  **/
2684 static void i40e_atr(struct i40e_ring *tx_ring, struct sk_buff *skb,
2685 		     u32 tx_flags)
2686 {
2687 	struct i40e_filter_program_desc *fdir_desc;
2688 	struct i40e_pf *pf = tx_ring->vsi->back;
2689 	union {
2690 		unsigned char *network;
2691 		struct iphdr *ipv4;
2692 		struct ipv6hdr *ipv6;
2693 	} hdr;
2694 	struct tcphdr *th;
2695 	unsigned int hlen;
2696 	u32 flex_ptype, dtype_cmd;
2697 	int l4_proto;
2698 	u16 i;
2699 
2700 	/* make sure ATR is enabled */
2701 	if (!(pf->flags & I40E_FLAG_FD_ATR_ENABLED))
2702 		return;
2703 
2704 	if (test_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state))
2705 		return;
2706 
2707 	/* if sampling is disabled do nothing */
2708 	if (!tx_ring->atr_sample_rate)
2709 		return;
2710 
2711 	/* Currently only IPv4/IPv6 with TCP is supported */
2712 	if (!(tx_flags & (I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6)))
2713 		return;
2714 
2715 	/* snag network header to get L4 type and address */
2716 	hdr.network = (tx_flags & I40E_TX_FLAGS_UDP_TUNNEL) ?
2717 		      skb_inner_network_header(skb) : skb_network_header(skb);
2718 
2719 	/* Note: tx_flags gets modified to reflect inner protocols in
2720 	 * tx_enable_csum function if encap is enabled.
2721 	 */
2722 	if (tx_flags & I40E_TX_FLAGS_IPV4) {
2723 		/* access ihl as u8 to avoid unaligned access on ia64 */
2724 		hlen = (hdr.network[0] & 0x0F) << 2;
2725 		l4_proto = hdr.ipv4->protocol;
2726 	} else {
2727 		/* find the start of the innermost ipv6 header */
2728 		unsigned int inner_hlen = hdr.network - skb->data;
2729 		unsigned int h_offset = inner_hlen;
2730 
2731 		/* this function updates h_offset to the end of the header */
2732 		l4_proto =
2733 		  ipv6_find_hdr(skb, &h_offset, IPPROTO_TCP, NULL, NULL);
2734 		/* hlen will contain our best estimate of the tcp header */
2735 		hlen = h_offset - inner_hlen;
2736 	}
2737 
2738 	if (l4_proto != IPPROTO_TCP)
2739 		return;
2740 
2741 	th = (struct tcphdr *)(hdr.network + hlen);
2742 
2743 	/* Due to lack of space, no more new filters can be programmed */
2744 	if (th->syn && test_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state))
2745 		return;
2746 	if (pf->flags & I40E_FLAG_HW_ATR_EVICT_ENABLED) {
2747 		/* HW ATR eviction will take care of removing filters on FIN
2748 		 * and RST packets.
2749 		 */
2750 		if (th->fin || th->rst)
2751 			return;
2752 	}
2753 
2754 	tx_ring->atr_count++;
2755 
2756 	/* sample on all syn/fin/rst packets or once every atr sample rate */
2757 	if (!th->fin &&
2758 	    !th->syn &&
2759 	    !th->rst &&
2760 	    (tx_ring->atr_count < tx_ring->atr_sample_rate))
2761 		return;
2762 
2763 	tx_ring->atr_count = 0;
2764 
2765 	/* grab the next descriptor */
2766 	i = tx_ring->next_to_use;
2767 	fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
2768 
2769 	i++;
2770 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2771 
2772 	flex_ptype = (tx_ring->queue_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) &
2773 		      I40E_TXD_FLTR_QW0_QINDEX_MASK;
2774 	flex_ptype |= (tx_flags & I40E_TX_FLAGS_IPV4) ?
2775 		      (I40E_FILTER_PCTYPE_NONF_IPV4_TCP <<
2776 		       I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) :
2777 		      (I40E_FILTER_PCTYPE_NONF_IPV6_TCP <<
2778 		       I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
2779 
2780 	flex_ptype |= tx_ring->vsi->id << I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT;
2781 
2782 	dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
2783 
2784 	dtype_cmd |= (th->fin || th->rst) ?
2785 		     (I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
2786 		      I40E_TXD_FLTR_QW1_PCMD_SHIFT) :
2787 		     (I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
2788 		      I40E_TXD_FLTR_QW1_PCMD_SHIFT);
2789 
2790 	dtype_cmd |= I40E_FILTER_PROGRAM_DESC_DEST_DIRECT_PACKET_QINDEX <<
2791 		     I40E_TXD_FLTR_QW1_DEST_SHIFT;
2792 
2793 	dtype_cmd |= I40E_FILTER_PROGRAM_DESC_FD_STATUS_FD_ID <<
2794 		     I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT;
2795 
2796 	dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
2797 	if (!(tx_flags & I40E_TX_FLAGS_UDP_TUNNEL))
2798 		dtype_cmd |=
2799 			((u32)I40E_FD_ATR_STAT_IDX(pf->hw.pf_id) <<
2800 			I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2801 			I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2802 	else
2803 		dtype_cmd |=
2804 			((u32)I40E_FD_ATR_TUNNEL_STAT_IDX(pf->hw.pf_id) <<
2805 			I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2806 			I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2807 
2808 	if (pf->flags & I40E_FLAG_HW_ATR_EVICT_ENABLED)
2809 		dtype_cmd |= I40E_TXD_FLTR_QW1_ATR_MASK;
2810 
2811 	fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
2812 	fdir_desc->rsvd = cpu_to_le32(0);
2813 	fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
2814 	fdir_desc->fd_id = cpu_to_le32(0);
2815 }
2816 
2817 /**
2818  * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
2819  * @skb:     send buffer
2820  * @tx_ring: ring to send buffer on
2821  * @flags:   the tx flags to be set
2822  *
2823  * Checks the skb and set up correspondingly several generic transmit flags
2824  * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
2825  *
2826  * Returns error code indicate the frame should be dropped upon error and the
2827  * otherwise  returns 0 to indicate the flags has been set properly.
2828  **/
2829 static inline int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
2830 					     struct i40e_ring *tx_ring,
2831 					     u32 *flags)
2832 {
2833 	__be16 protocol = skb->protocol;
2834 	u32  tx_flags = 0;
2835 
2836 	if (protocol == htons(ETH_P_8021Q) &&
2837 	    !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
2838 		/* When HW VLAN acceleration is turned off by the user the
2839 		 * stack sets the protocol to 8021q so that the driver
2840 		 * can take any steps required to support the SW only
2841 		 * VLAN handling.  In our case the driver doesn't need
2842 		 * to take any further steps so just set the protocol
2843 		 * to the encapsulated ethertype.
2844 		 */
2845 		skb->protocol = vlan_get_protocol(skb);
2846 		goto out;
2847 	}
2848 
2849 	/* if we have a HW VLAN tag being added, default to the HW one */
2850 	if (skb_vlan_tag_present(skb)) {
2851 		tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
2852 		tx_flags |= I40E_TX_FLAGS_HW_VLAN;
2853 	/* else if it is a SW VLAN, check the next protocol and store the tag */
2854 	} else if (protocol == htons(ETH_P_8021Q)) {
2855 		struct vlan_hdr *vhdr, _vhdr;
2856 
2857 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
2858 		if (!vhdr)
2859 			return -EINVAL;
2860 
2861 		protocol = vhdr->h_vlan_encapsulated_proto;
2862 		tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
2863 		tx_flags |= I40E_TX_FLAGS_SW_VLAN;
2864 	}
2865 
2866 	if (!(tx_ring->vsi->back->flags & I40E_FLAG_DCB_ENABLED))
2867 		goto out;
2868 
2869 	/* Insert 802.1p priority into VLAN header */
2870 	if ((tx_flags & (I40E_TX_FLAGS_HW_VLAN | I40E_TX_FLAGS_SW_VLAN)) ||
2871 	    (skb->priority != TC_PRIO_CONTROL)) {
2872 		tx_flags &= ~I40E_TX_FLAGS_VLAN_PRIO_MASK;
2873 		tx_flags |= (skb->priority & 0x7) <<
2874 				I40E_TX_FLAGS_VLAN_PRIO_SHIFT;
2875 		if (tx_flags & I40E_TX_FLAGS_SW_VLAN) {
2876 			struct vlan_ethhdr *vhdr;
2877 			int rc;
2878 
2879 			rc = skb_cow_head(skb, 0);
2880 			if (rc < 0)
2881 				return rc;
2882 			vhdr = (struct vlan_ethhdr *)skb->data;
2883 			vhdr->h_vlan_TCI = htons(tx_flags >>
2884 						 I40E_TX_FLAGS_VLAN_SHIFT);
2885 		} else {
2886 			tx_flags |= I40E_TX_FLAGS_HW_VLAN;
2887 		}
2888 	}
2889 
2890 out:
2891 	*flags = tx_flags;
2892 	return 0;
2893 }
2894 
2895 /**
2896  * i40e_tso - set up the tso context descriptor
2897  * @first:    pointer to first Tx buffer for xmit
2898  * @hdr_len:  ptr to the size of the packet header
2899  * @cd_type_cmd_tso_mss: Quad Word 1
2900  *
2901  * Returns 0 if no TSO can happen, 1 if tso is going, or error
2902  **/
2903 static int i40e_tso(struct i40e_tx_buffer *first, u8 *hdr_len,
2904 		    u64 *cd_type_cmd_tso_mss)
2905 {
2906 	struct sk_buff *skb = first->skb;
2907 	u64 cd_cmd, cd_tso_len, cd_mss;
2908 	union {
2909 		struct iphdr *v4;
2910 		struct ipv6hdr *v6;
2911 		unsigned char *hdr;
2912 	} ip;
2913 	union {
2914 		struct tcphdr *tcp;
2915 		struct udphdr *udp;
2916 		unsigned char *hdr;
2917 	} l4;
2918 	u32 paylen, l4_offset;
2919 	u16 gso_segs, gso_size;
2920 	int err;
2921 
2922 	if (skb->ip_summed != CHECKSUM_PARTIAL)
2923 		return 0;
2924 
2925 	if (!skb_is_gso(skb))
2926 		return 0;
2927 
2928 	err = skb_cow_head(skb, 0);
2929 	if (err < 0)
2930 		return err;
2931 
2932 	ip.hdr = skb_network_header(skb);
2933 	l4.hdr = skb_transport_header(skb);
2934 
2935 	/* initialize outer IP header fields */
2936 	if (ip.v4->version == 4) {
2937 		ip.v4->tot_len = 0;
2938 		ip.v4->check = 0;
2939 	} else {
2940 		ip.v6->payload_len = 0;
2941 	}
2942 
2943 	if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
2944 					 SKB_GSO_GRE_CSUM |
2945 					 SKB_GSO_IPXIP4 |
2946 					 SKB_GSO_IPXIP6 |
2947 					 SKB_GSO_UDP_TUNNEL |
2948 					 SKB_GSO_UDP_TUNNEL_CSUM)) {
2949 		if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
2950 		    (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)) {
2951 			l4.udp->len = 0;
2952 
2953 			/* determine offset of outer transport header */
2954 			l4_offset = l4.hdr - skb->data;
2955 
2956 			/* remove payload length from outer checksum */
2957 			paylen = skb->len - l4_offset;
2958 			csum_replace_by_diff(&l4.udp->check,
2959 					     (__force __wsum)htonl(paylen));
2960 		}
2961 
2962 		/* reset pointers to inner headers */
2963 		ip.hdr = skb_inner_network_header(skb);
2964 		l4.hdr = skb_inner_transport_header(skb);
2965 
2966 		/* initialize inner IP header fields */
2967 		if (ip.v4->version == 4) {
2968 			ip.v4->tot_len = 0;
2969 			ip.v4->check = 0;
2970 		} else {
2971 			ip.v6->payload_len = 0;
2972 		}
2973 	}
2974 
2975 	/* determine offset of inner transport header */
2976 	l4_offset = l4.hdr - skb->data;
2977 
2978 	/* remove payload length from inner checksum */
2979 	paylen = skb->len - l4_offset;
2980 	csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen));
2981 
2982 	/* compute length of segmentation header */
2983 	*hdr_len = (l4.tcp->doff * 4) + l4_offset;
2984 
2985 	/* pull values out of skb_shinfo */
2986 	gso_size = skb_shinfo(skb)->gso_size;
2987 	gso_segs = skb_shinfo(skb)->gso_segs;
2988 
2989 	/* update GSO size and bytecount with header size */
2990 	first->gso_segs = gso_segs;
2991 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
2992 
2993 	/* find the field values */
2994 	cd_cmd = I40E_TX_CTX_DESC_TSO;
2995 	cd_tso_len = skb->len - *hdr_len;
2996 	cd_mss = gso_size;
2997 	*cd_type_cmd_tso_mss |= (cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
2998 				(cd_tso_len << I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
2999 				(cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
3000 	return 1;
3001 }
3002 
3003 /**
3004  * i40e_tsyn - set up the tsyn context descriptor
3005  * @tx_ring:  ptr to the ring to send
3006  * @skb:      ptr to the skb we're sending
3007  * @tx_flags: the collected send information
3008  * @cd_type_cmd_tso_mss: Quad Word 1
3009  *
3010  * Returns 0 if no Tx timestamp can happen and 1 if the timestamp will happen
3011  **/
3012 static int i40e_tsyn(struct i40e_ring *tx_ring, struct sk_buff *skb,
3013 		     u32 tx_flags, u64 *cd_type_cmd_tso_mss)
3014 {
3015 	struct i40e_pf *pf;
3016 
3017 	if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)))
3018 		return 0;
3019 
3020 	/* Tx timestamps cannot be sampled when doing TSO */
3021 	if (tx_flags & I40E_TX_FLAGS_TSO)
3022 		return 0;
3023 
3024 	/* only timestamp the outbound packet if the user has requested it and
3025 	 * we are not already transmitting a packet to be timestamped
3026 	 */
3027 	pf = i40e_netdev_to_pf(tx_ring->netdev);
3028 	if (!(pf->flags & I40E_FLAG_PTP))
3029 		return 0;
3030 
3031 	if (pf->ptp_tx &&
3032 	    !test_and_set_bit_lock(__I40E_PTP_TX_IN_PROGRESS, pf->state)) {
3033 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
3034 		pf->ptp_tx_start = jiffies;
3035 		pf->ptp_tx_skb = skb_get(skb);
3036 	} else {
3037 		pf->tx_hwtstamp_skipped++;
3038 		return 0;
3039 	}
3040 
3041 	*cd_type_cmd_tso_mss |= (u64)I40E_TX_CTX_DESC_TSYN <<
3042 				I40E_TXD_CTX_QW1_CMD_SHIFT;
3043 
3044 	return 1;
3045 }
3046 
3047 /**
3048  * i40e_tx_enable_csum - Enable Tx checksum offloads
3049  * @skb: send buffer
3050  * @tx_flags: pointer to Tx flags currently set
3051  * @td_cmd: Tx descriptor command bits to set
3052  * @td_offset: Tx descriptor header offsets to set
3053  * @tx_ring: Tx descriptor ring
3054  * @cd_tunneling: ptr to context desc bits
3055  **/
3056 static int i40e_tx_enable_csum(struct sk_buff *skb, u32 *tx_flags,
3057 			       u32 *td_cmd, u32 *td_offset,
3058 			       struct i40e_ring *tx_ring,
3059 			       u32 *cd_tunneling)
3060 {
3061 	union {
3062 		struct iphdr *v4;
3063 		struct ipv6hdr *v6;
3064 		unsigned char *hdr;
3065 	} ip;
3066 	union {
3067 		struct tcphdr *tcp;
3068 		struct udphdr *udp;
3069 		unsigned char *hdr;
3070 	} l4;
3071 	unsigned char *exthdr;
3072 	u32 offset, cmd = 0;
3073 	__be16 frag_off;
3074 	u8 l4_proto = 0;
3075 
3076 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3077 		return 0;
3078 
3079 	ip.hdr = skb_network_header(skb);
3080 	l4.hdr = skb_transport_header(skb);
3081 
3082 	/* compute outer L2 header size */
3083 	offset = ((ip.hdr - skb->data) / 2) << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
3084 
3085 	if (skb->encapsulation) {
3086 		u32 tunnel = 0;
3087 		/* define outer network header type */
3088 		if (*tx_flags & I40E_TX_FLAGS_IPV4) {
3089 			tunnel |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
3090 				  I40E_TX_CTX_EXT_IP_IPV4 :
3091 				  I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
3092 
3093 			l4_proto = ip.v4->protocol;
3094 		} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
3095 			tunnel |= I40E_TX_CTX_EXT_IP_IPV6;
3096 
3097 			exthdr = ip.hdr + sizeof(*ip.v6);
3098 			l4_proto = ip.v6->nexthdr;
3099 			if (l4.hdr != exthdr)
3100 				ipv6_skip_exthdr(skb, exthdr - skb->data,
3101 						 &l4_proto, &frag_off);
3102 		}
3103 
3104 		/* define outer transport */
3105 		switch (l4_proto) {
3106 		case IPPROTO_UDP:
3107 			tunnel |= I40E_TXD_CTX_UDP_TUNNELING;
3108 			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
3109 			break;
3110 		case IPPROTO_GRE:
3111 			tunnel |= I40E_TXD_CTX_GRE_TUNNELING;
3112 			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
3113 			break;
3114 		case IPPROTO_IPIP:
3115 		case IPPROTO_IPV6:
3116 			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
3117 			l4.hdr = skb_inner_network_header(skb);
3118 			break;
3119 		default:
3120 			if (*tx_flags & I40E_TX_FLAGS_TSO)
3121 				return -1;
3122 
3123 			skb_checksum_help(skb);
3124 			return 0;
3125 		}
3126 
3127 		/* compute outer L3 header size */
3128 		tunnel |= ((l4.hdr - ip.hdr) / 4) <<
3129 			  I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT;
3130 
3131 		/* switch IP header pointer from outer to inner header */
3132 		ip.hdr = skb_inner_network_header(skb);
3133 
3134 		/* compute tunnel header size */
3135 		tunnel |= ((ip.hdr - l4.hdr) / 2) <<
3136 			  I40E_TXD_CTX_QW0_NATLEN_SHIFT;
3137 
3138 		/* indicate if we need to offload outer UDP header */
3139 		if ((*tx_flags & I40E_TX_FLAGS_TSO) &&
3140 		    !(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
3141 		    (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM))
3142 			tunnel |= I40E_TXD_CTX_QW0_L4T_CS_MASK;
3143 
3144 		/* record tunnel offload values */
3145 		*cd_tunneling |= tunnel;
3146 
3147 		/* switch L4 header pointer from outer to inner */
3148 		l4.hdr = skb_inner_transport_header(skb);
3149 		l4_proto = 0;
3150 
3151 		/* reset type as we transition from outer to inner headers */
3152 		*tx_flags &= ~(I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6);
3153 		if (ip.v4->version == 4)
3154 			*tx_flags |= I40E_TX_FLAGS_IPV4;
3155 		if (ip.v6->version == 6)
3156 			*tx_flags |= I40E_TX_FLAGS_IPV6;
3157 	}
3158 
3159 	/* Enable IP checksum offloads */
3160 	if (*tx_flags & I40E_TX_FLAGS_IPV4) {
3161 		l4_proto = ip.v4->protocol;
3162 		/* the stack computes the IP header already, the only time we
3163 		 * need the hardware to recompute it is in the case of TSO.
3164 		 */
3165 		cmd |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
3166 		       I40E_TX_DESC_CMD_IIPT_IPV4_CSUM :
3167 		       I40E_TX_DESC_CMD_IIPT_IPV4;
3168 	} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
3169 		cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
3170 
3171 		exthdr = ip.hdr + sizeof(*ip.v6);
3172 		l4_proto = ip.v6->nexthdr;
3173 		if (l4.hdr != exthdr)
3174 			ipv6_skip_exthdr(skb, exthdr - skb->data,
3175 					 &l4_proto, &frag_off);
3176 	}
3177 
3178 	/* compute inner L3 header size */
3179 	offset |= ((l4.hdr - ip.hdr) / 4) << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
3180 
3181 	/* Enable L4 checksum offloads */
3182 	switch (l4_proto) {
3183 	case IPPROTO_TCP:
3184 		/* enable checksum offloads */
3185 		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
3186 		offset |= l4.tcp->doff << I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
3187 		break;
3188 	case IPPROTO_SCTP:
3189 		/* enable SCTP checksum offload */
3190 		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
3191 		offset |= (sizeof(struct sctphdr) >> 2) <<
3192 			  I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
3193 		break;
3194 	case IPPROTO_UDP:
3195 		/* enable UDP checksum offload */
3196 		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
3197 		offset |= (sizeof(struct udphdr) >> 2) <<
3198 			  I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
3199 		break;
3200 	default:
3201 		if (*tx_flags & I40E_TX_FLAGS_TSO)
3202 			return -1;
3203 		skb_checksum_help(skb);
3204 		return 0;
3205 	}
3206 
3207 	*td_cmd |= cmd;
3208 	*td_offset |= offset;
3209 
3210 	return 1;
3211 }
3212 
3213 /**
3214  * i40e_create_tx_ctx Build the Tx context descriptor
3215  * @tx_ring:  ring to create the descriptor on
3216  * @cd_type_cmd_tso_mss: Quad Word 1
3217  * @cd_tunneling: Quad Word 0 - bits 0-31
3218  * @cd_l2tag2: Quad Word 0 - bits 32-63
3219  **/
3220 static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
3221 			       const u64 cd_type_cmd_tso_mss,
3222 			       const u32 cd_tunneling, const u32 cd_l2tag2)
3223 {
3224 	struct i40e_tx_context_desc *context_desc;
3225 	int i = tx_ring->next_to_use;
3226 
3227 	if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
3228 	    !cd_tunneling && !cd_l2tag2)
3229 		return;
3230 
3231 	/* grab the next descriptor */
3232 	context_desc = I40E_TX_CTXTDESC(tx_ring, i);
3233 
3234 	i++;
3235 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3236 
3237 	/* cpu_to_le32 and assign to struct fields */
3238 	context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
3239 	context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
3240 	context_desc->rsvd = cpu_to_le16(0);
3241 	context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
3242 }
3243 
3244 /**
3245  * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions
3246  * @tx_ring: the ring to be checked
3247  * @size:    the size buffer we want to assure is available
3248  *
3249  * Returns -EBUSY if a stop is needed, else 0
3250  **/
3251 int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
3252 {
3253 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3254 	/* Memory barrier before checking head and tail */
3255 	smp_mb();
3256 
3257 	/* Check again in a case another CPU has just made room available. */
3258 	if (likely(I40E_DESC_UNUSED(tx_ring) < size))
3259 		return -EBUSY;
3260 
3261 	/* A reprieve! - use start_queue because it doesn't call schedule */
3262 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3263 	++tx_ring->tx_stats.restart_queue;
3264 	return 0;
3265 }
3266 
3267 /**
3268  * __i40e_chk_linearize - Check if there are more than 8 buffers per packet
3269  * @skb:      send buffer
3270  *
3271  * Note: Our HW can't DMA more than 8 buffers to build a packet on the wire
3272  * and so we need to figure out the cases where we need to linearize the skb.
3273  *
3274  * For TSO we need to count the TSO header and segment payload separately.
3275  * As such we need to check cases where we have 7 fragments or more as we
3276  * can potentially require 9 DMA transactions, 1 for the TSO header, 1 for
3277  * the segment payload in the first descriptor, and another 7 for the
3278  * fragments.
3279  **/
3280 bool __i40e_chk_linearize(struct sk_buff *skb)
3281 {
3282 	const struct skb_frag_struct *frag, *stale;
3283 	int nr_frags, sum;
3284 
3285 	/* no need to check if number of frags is less than 7 */
3286 	nr_frags = skb_shinfo(skb)->nr_frags;
3287 	if (nr_frags < (I40E_MAX_BUFFER_TXD - 1))
3288 		return false;
3289 
3290 	/* We need to walk through the list and validate that each group
3291 	 * of 6 fragments totals at least gso_size.
3292 	 */
3293 	nr_frags -= I40E_MAX_BUFFER_TXD - 2;
3294 	frag = &skb_shinfo(skb)->frags[0];
3295 
3296 	/* Initialize size to the negative value of gso_size minus 1.  We
3297 	 * use this as the worst case scenerio in which the frag ahead
3298 	 * of us only provides one byte which is why we are limited to 6
3299 	 * descriptors for a single transmit as the header and previous
3300 	 * fragment are already consuming 2 descriptors.
3301 	 */
3302 	sum = 1 - skb_shinfo(skb)->gso_size;
3303 
3304 	/* Add size of frags 0 through 4 to create our initial sum */
3305 	sum += skb_frag_size(frag++);
3306 	sum += skb_frag_size(frag++);
3307 	sum += skb_frag_size(frag++);
3308 	sum += skb_frag_size(frag++);
3309 	sum += skb_frag_size(frag++);
3310 
3311 	/* Walk through fragments adding latest fragment, testing it, and
3312 	 * then removing stale fragments from the sum.
3313 	 */
3314 	for (stale = &skb_shinfo(skb)->frags[0];; stale++) {
3315 		int stale_size = skb_frag_size(stale);
3316 
3317 		sum += skb_frag_size(frag++);
3318 
3319 		/* The stale fragment may present us with a smaller
3320 		 * descriptor than the actual fragment size. To account
3321 		 * for that we need to remove all the data on the front and
3322 		 * figure out what the remainder would be in the last
3323 		 * descriptor associated with the fragment.
3324 		 */
3325 		if (stale_size > I40E_MAX_DATA_PER_TXD) {
3326 			int align_pad = -(stale->page_offset) &
3327 					(I40E_MAX_READ_REQ_SIZE - 1);
3328 
3329 			sum -= align_pad;
3330 			stale_size -= align_pad;
3331 
3332 			do {
3333 				sum -= I40E_MAX_DATA_PER_TXD_ALIGNED;
3334 				stale_size -= I40E_MAX_DATA_PER_TXD_ALIGNED;
3335 			} while (stale_size > I40E_MAX_DATA_PER_TXD);
3336 		}
3337 
3338 		/* if sum is negative we failed to make sufficient progress */
3339 		if (sum < 0)
3340 			return true;
3341 
3342 		if (!nr_frags--)
3343 			break;
3344 
3345 		sum -= stale_size;
3346 	}
3347 
3348 	return false;
3349 }
3350 
3351 /**
3352  * i40e_tx_map - Build the Tx descriptor
3353  * @tx_ring:  ring to send buffer on
3354  * @skb:      send buffer
3355  * @first:    first buffer info buffer to use
3356  * @tx_flags: collected send information
3357  * @hdr_len:  size of the packet header
3358  * @td_cmd:   the command field in the descriptor
3359  * @td_offset: offset for checksum or crc
3360  *
3361  * Returns 0 on success, -1 on failure to DMA
3362  **/
3363 static inline int i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
3364 			      struct i40e_tx_buffer *first, u32 tx_flags,
3365 			      const u8 hdr_len, u32 td_cmd, u32 td_offset)
3366 {
3367 	unsigned int data_len = skb->data_len;
3368 	unsigned int size = skb_headlen(skb);
3369 	struct skb_frag_struct *frag;
3370 	struct i40e_tx_buffer *tx_bi;
3371 	struct i40e_tx_desc *tx_desc;
3372 	u16 i = tx_ring->next_to_use;
3373 	u32 td_tag = 0;
3374 	dma_addr_t dma;
3375 	u16 desc_count = 1;
3376 
3377 	if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
3378 		td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
3379 		td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
3380 			 I40E_TX_FLAGS_VLAN_SHIFT;
3381 	}
3382 
3383 	first->tx_flags = tx_flags;
3384 
3385 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3386 
3387 	tx_desc = I40E_TX_DESC(tx_ring, i);
3388 	tx_bi = first;
3389 
3390 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
3391 		unsigned int max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
3392 
3393 		if (dma_mapping_error(tx_ring->dev, dma))
3394 			goto dma_error;
3395 
3396 		/* record length, and DMA address */
3397 		dma_unmap_len_set(tx_bi, len, size);
3398 		dma_unmap_addr_set(tx_bi, dma, dma);
3399 
3400 		/* align size to end of page */
3401 		max_data += -dma & (I40E_MAX_READ_REQ_SIZE - 1);
3402 		tx_desc->buffer_addr = cpu_to_le64(dma);
3403 
3404 		while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
3405 			tx_desc->cmd_type_offset_bsz =
3406 				build_ctob(td_cmd, td_offset,
3407 					   max_data, td_tag);
3408 
3409 			tx_desc++;
3410 			i++;
3411 			desc_count++;
3412 
3413 			if (i == tx_ring->count) {
3414 				tx_desc = I40E_TX_DESC(tx_ring, 0);
3415 				i = 0;
3416 			}
3417 
3418 			dma += max_data;
3419 			size -= max_data;
3420 
3421 			max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
3422 			tx_desc->buffer_addr = cpu_to_le64(dma);
3423 		}
3424 
3425 		if (likely(!data_len))
3426 			break;
3427 
3428 		tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
3429 							  size, td_tag);
3430 
3431 		tx_desc++;
3432 		i++;
3433 		desc_count++;
3434 
3435 		if (i == tx_ring->count) {
3436 			tx_desc = I40E_TX_DESC(tx_ring, 0);
3437 			i = 0;
3438 		}
3439 
3440 		size = skb_frag_size(frag);
3441 		data_len -= size;
3442 
3443 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3444 				       DMA_TO_DEVICE);
3445 
3446 		tx_bi = &tx_ring->tx_bi[i];
3447 	}
3448 
3449 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
3450 
3451 	i++;
3452 	if (i == tx_ring->count)
3453 		i = 0;
3454 
3455 	tx_ring->next_to_use = i;
3456 
3457 	i40e_maybe_stop_tx(tx_ring, DESC_NEEDED);
3458 
3459 	/* write last descriptor with EOP bit */
3460 	td_cmd |= I40E_TX_DESC_CMD_EOP;
3461 
3462 	/* We OR these values together to check both against 4 (WB_STRIDE)
3463 	 * below. This is safe since we don't re-use desc_count afterwards.
3464 	 */
3465 	desc_count |= ++tx_ring->packet_stride;
3466 
3467 	if (desc_count >= WB_STRIDE) {
3468 		/* write last descriptor with RS bit set */
3469 		td_cmd |= I40E_TX_DESC_CMD_RS;
3470 		tx_ring->packet_stride = 0;
3471 	}
3472 
3473 	tx_desc->cmd_type_offset_bsz =
3474 			build_ctob(td_cmd, td_offset, size, td_tag);
3475 
3476 	/* Force memory writes to complete before letting h/w know there
3477 	 * are new descriptors to fetch.
3478 	 *
3479 	 * We also use this memory barrier to make certain all of the
3480 	 * status bits have been updated before next_to_watch is written.
3481 	 */
3482 	wmb();
3483 
3484 	/* set next_to_watch value indicating a packet is present */
3485 	first->next_to_watch = tx_desc;
3486 
3487 	/* notify HW of packet */
3488 	if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
3489 		writel(i, tx_ring->tail);
3490 
3491 		/* we need this if more than one processor can write to our tail
3492 		 * at a time, it synchronizes IO on IA64/Altix systems
3493 		 */
3494 		mmiowb();
3495 	}
3496 
3497 	return 0;
3498 
3499 dma_error:
3500 	dev_info(tx_ring->dev, "TX DMA map failed\n");
3501 
3502 	/* clear dma mappings for failed tx_bi map */
3503 	for (;;) {
3504 		tx_bi = &tx_ring->tx_bi[i];
3505 		i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
3506 		if (tx_bi == first)
3507 			break;
3508 		if (i == 0)
3509 			i = tx_ring->count;
3510 		i--;
3511 	}
3512 
3513 	tx_ring->next_to_use = i;
3514 
3515 	return -1;
3516 }
3517 
3518 /**
3519  * i40e_xmit_xdp_ring - transmits an XDP buffer to an XDP Tx ring
3520  * @xdp: data to transmit
3521  * @xdp_ring: XDP Tx ring
3522  **/
3523 static int i40e_xmit_xdp_ring(struct xdp_frame *xdpf,
3524 			      struct i40e_ring *xdp_ring)
3525 {
3526 	u16 i = xdp_ring->next_to_use;
3527 	struct i40e_tx_buffer *tx_bi;
3528 	struct i40e_tx_desc *tx_desc;
3529 	u32 size = xdpf->len;
3530 	dma_addr_t dma;
3531 
3532 	if (!unlikely(I40E_DESC_UNUSED(xdp_ring))) {
3533 		xdp_ring->tx_stats.tx_busy++;
3534 		return I40E_XDP_CONSUMED;
3535 	}
3536 
3537 	dma = dma_map_single(xdp_ring->dev, xdpf->data, size, DMA_TO_DEVICE);
3538 	if (dma_mapping_error(xdp_ring->dev, dma))
3539 		return I40E_XDP_CONSUMED;
3540 
3541 	tx_bi = &xdp_ring->tx_bi[i];
3542 	tx_bi->bytecount = size;
3543 	tx_bi->gso_segs = 1;
3544 	tx_bi->xdpf = xdpf;
3545 
3546 	/* record length, and DMA address */
3547 	dma_unmap_len_set(tx_bi, len, size);
3548 	dma_unmap_addr_set(tx_bi, dma, dma);
3549 
3550 	tx_desc = I40E_TX_DESC(xdp_ring, i);
3551 	tx_desc->buffer_addr = cpu_to_le64(dma);
3552 	tx_desc->cmd_type_offset_bsz = build_ctob(I40E_TX_DESC_CMD_ICRC
3553 						  | I40E_TXD_CMD,
3554 						  0, size, 0);
3555 
3556 	/* Make certain all of the status bits have been updated
3557 	 * before next_to_watch is written.
3558 	 */
3559 	smp_wmb();
3560 
3561 	i++;
3562 	if (i == xdp_ring->count)
3563 		i = 0;
3564 
3565 	tx_bi->next_to_watch = tx_desc;
3566 	xdp_ring->next_to_use = i;
3567 
3568 	return I40E_XDP_TX;
3569 }
3570 
3571 /**
3572  * i40e_xmit_frame_ring - Sends buffer on Tx ring
3573  * @skb:     send buffer
3574  * @tx_ring: ring to send buffer on
3575  *
3576  * Returns NETDEV_TX_OK if sent, else an error code
3577  **/
3578 static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
3579 					struct i40e_ring *tx_ring)
3580 {
3581 	u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
3582 	u32 cd_tunneling = 0, cd_l2tag2 = 0;
3583 	struct i40e_tx_buffer *first;
3584 	u32 td_offset = 0;
3585 	u32 tx_flags = 0;
3586 	__be16 protocol;
3587 	u32 td_cmd = 0;
3588 	u8 hdr_len = 0;
3589 	int tso, count;
3590 	int tsyn;
3591 
3592 	/* prefetch the data, we'll need it later */
3593 	prefetch(skb->data);
3594 
3595 	i40e_trace(xmit_frame_ring, skb, tx_ring);
3596 
3597 	count = i40e_xmit_descriptor_count(skb);
3598 	if (i40e_chk_linearize(skb, count)) {
3599 		if (__skb_linearize(skb)) {
3600 			dev_kfree_skb_any(skb);
3601 			return NETDEV_TX_OK;
3602 		}
3603 		count = i40e_txd_use_count(skb->len);
3604 		tx_ring->tx_stats.tx_linearize++;
3605 	}
3606 
3607 	/* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
3608 	 *       + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
3609 	 *       + 4 desc gap to avoid the cache line where head is,
3610 	 *       + 1 desc for context descriptor,
3611 	 * otherwise try next time
3612 	 */
3613 	if (i40e_maybe_stop_tx(tx_ring, count + 4 + 1)) {
3614 		tx_ring->tx_stats.tx_busy++;
3615 		return NETDEV_TX_BUSY;
3616 	}
3617 
3618 	/* record the location of the first descriptor for this packet */
3619 	first = &tx_ring->tx_bi[tx_ring->next_to_use];
3620 	first->skb = skb;
3621 	first->bytecount = skb->len;
3622 	first->gso_segs = 1;
3623 
3624 	/* prepare the xmit flags */
3625 	if (i40e_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
3626 		goto out_drop;
3627 
3628 	/* obtain protocol of skb */
3629 	protocol = vlan_get_protocol(skb);
3630 
3631 	/* setup IPv4/IPv6 offloads */
3632 	if (protocol == htons(ETH_P_IP))
3633 		tx_flags |= I40E_TX_FLAGS_IPV4;
3634 	else if (protocol == htons(ETH_P_IPV6))
3635 		tx_flags |= I40E_TX_FLAGS_IPV6;
3636 
3637 	tso = i40e_tso(first, &hdr_len, &cd_type_cmd_tso_mss);
3638 
3639 	if (tso < 0)
3640 		goto out_drop;
3641 	else if (tso)
3642 		tx_flags |= I40E_TX_FLAGS_TSO;
3643 
3644 	/* Always offload the checksum, since it's in the data descriptor */
3645 	tso = i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset,
3646 				  tx_ring, &cd_tunneling);
3647 	if (tso < 0)
3648 		goto out_drop;
3649 
3650 	tsyn = i40e_tsyn(tx_ring, skb, tx_flags, &cd_type_cmd_tso_mss);
3651 
3652 	if (tsyn)
3653 		tx_flags |= I40E_TX_FLAGS_TSYN;
3654 
3655 	skb_tx_timestamp(skb);
3656 
3657 	/* always enable CRC insertion offload */
3658 	td_cmd |= I40E_TX_DESC_CMD_ICRC;
3659 
3660 	i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
3661 			   cd_tunneling, cd_l2tag2);
3662 
3663 	/* Add Flow Director ATR if it's enabled.
3664 	 *
3665 	 * NOTE: this must always be directly before the data descriptor.
3666 	 */
3667 	i40e_atr(tx_ring, skb, tx_flags);
3668 
3669 	if (i40e_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
3670 			td_cmd, td_offset))
3671 		goto cleanup_tx_tstamp;
3672 
3673 	return NETDEV_TX_OK;
3674 
3675 out_drop:
3676 	i40e_trace(xmit_frame_ring_drop, first->skb, tx_ring);
3677 	dev_kfree_skb_any(first->skb);
3678 	first->skb = NULL;
3679 cleanup_tx_tstamp:
3680 	if (unlikely(tx_flags & I40E_TX_FLAGS_TSYN)) {
3681 		struct i40e_pf *pf = i40e_netdev_to_pf(tx_ring->netdev);
3682 
3683 		dev_kfree_skb_any(pf->ptp_tx_skb);
3684 		pf->ptp_tx_skb = NULL;
3685 		clear_bit_unlock(__I40E_PTP_TX_IN_PROGRESS, pf->state);
3686 	}
3687 
3688 	return NETDEV_TX_OK;
3689 }
3690 
3691 /**
3692  * i40e_lan_xmit_frame - Selects the correct VSI and Tx queue to send buffer
3693  * @skb:    send buffer
3694  * @netdev: network interface device structure
3695  *
3696  * Returns NETDEV_TX_OK if sent, else an error code
3697  **/
3698 netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3699 {
3700 	struct i40e_netdev_priv *np = netdev_priv(netdev);
3701 	struct i40e_vsi *vsi = np->vsi;
3702 	struct i40e_ring *tx_ring = vsi->tx_rings[skb->queue_mapping];
3703 
3704 	/* hardware can't handle really short frames, hardware padding works
3705 	 * beyond this point
3706 	 */
3707 	if (skb_put_padto(skb, I40E_MIN_TX_LEN))
3708 		return NETDEV_TX_OK;
3709 
3710 	return i40e_xmit_frame_ring(skb, tx_ring);
3711 }
3712 
3713 /**
3714  * i40e_xdp_xmit - Implements ndo_xdp_xmit
3715  * @dev: netdev
3716  * @xdp: XDP buffer
3717  *
3718  * Returns number of frames successfully sent. Frames that fail are
3719  * free'ed via XDP return API.
3720  *
3721  * For error cases, a negative errno code is returned and no-frames
3722  * are transmitted (caller must handle freeing frames).
3723  **/
3724 int i40e_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **frames,
3725 		  u32 flags)
3726 {
3727 	struct i40e_netdev_priv *np = netdev_priv(dev);
3728 	unsigned int queue_index = smp_processor_id();
3729 	struct i40e_vsi *vsi = np->vsi;
3730 	struct i40e_ring *xdp_ring;
3731 	int drops = 0;
3732 	int i;
3733 
3734 	if (test_bit(__I40E_VSI_DOWN, vsi->state))
3735 		return -ENETDOWN;
3736 
3737 	if (!i40e_enabled_xdp_vsi(vsi) || queue_index >= vsi->num_queue_pairs)
3738 		return -ENXIO;
3739 
3740 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
3741 		return -EINVAL;
3742 
3743 	xdp_ring = vsi->xdp_rings[queue_index];
3744 
3745 	for (i = 0; i < n; i++) {
3746 		struct xdp_frame *xdpf = frames[i];
3747 		int err;
3748 
3749 		err = i40e_xmit_xdp_ring(xdpf, xdp_ring);
3750 		if (err != I40E_XDP_TX) {
3751 			xdp_return_frame_rx_napi(xdpf);
3752 			drops++;
3753 		}
3754 	}
3755 
3756 	if (unlikely(flags & XDP_XMIT_FLUSH))
3757 		i40e_xdp_ring_update_tail(xdp_ring);
3758 
3759 	return n - drops;
3760 }
3761