xref: /linux/drivers/net/ethernet/intel/i40e/i40e_txrx.c (revision 0883c2c06fb5bcf5b9e008270827e63c09a88c1e)
1 /*******************************************************************************
2  *
3  * Intel Ethernet Controller XL710 Family Linux Driver
4  * Copyright(c) 2013 - 2016 Intel Corporation.
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms and conditions of the GNU General Public License,
8  * version 2, as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope it will be useful, but WITHOUT
11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13  * more details.
14  *
15  * You should have received a copy of the GNU General Public License along
16  * with this program.  If not, see <http://www.gnu.org/licenses/>.
17  *
18  * The full GNU General Public License is included in this distribution in
19  * the file called "COPYING".
20  *
21  * Contact Information:
22  * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
23  * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
24  *
25  ******************************************************************************/
26 
27 #include <linux/prefetch.h>
28 #include <net/busy_poll.h>
29 #include "i40e.h"
30 #include "i40e_prototype.h"
31 
32 static inline __le64 build_ctob(u32 td_cmd, u32 td_offset, unsigned int size,
33 				u32 td_tag)
34 {
35 	return cpu_to_le64(I40E_TX_DESC_DTYPE_DATA |
36 			   ((u64)td_cmd  << I40E_TXD_QW1_CMD_SHIFT) |
37 			   ((u64)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) |
38 			   ((u64)size  << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) |
39 			   ((u64)td_tag  << I40E_TXD_QW1_L2TAG1_SHIFT));
40 }
41 
42 #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
43 #define I40E_FD_CLEAN_DELAY 10
44 /**
45  * i40e_program_fdir_filter - Program a Flow Director filter
46  * @fdir_data: Packet data that will be filter parameters
47  * @raw_packet: the pre-allocated packet buffer for FDir
48  * @pf: The PF pointer
49  * @add: True for add/update, False for remove
50  **/
51 int i40e_program_fdir_filter(struct i40e_fdir_filter *fdir_data, u8 *raw_packet,
52 			     struct i40e_pf *pf, bool add)
53 {
54 	struct i40e_filter_program_desc *fdir_desc;
55 	struct i40e_tx_buffer *tx_buf, *first;
56 	struct i40e_tx_desc *tx_desc;
57 	struct i40e_ring *tx_ring;
58 	unsigned int fpt, dcc;
59 	struct i40e_vsi *vsi;
60 	struct device *dev;
61 	dma_addr_t dma;
62 	u32 td_cmd = 0;
63 	u16 delay = 0;
64 	u16 i;
65 
66 	/* find existing FDIR VSI */
67 	vsi = NULL;
68 	for (i = 0; i < pf->num_alloc_vsi; i++)
69 		if (pf->vsi[i] && pf->vsi[i]->type == I40E_VSI_FDIR)
70 			vsi = pf->vsi[i];
71 	if (!vsi)
72 		return -ENOENT;
73 
74 	tx_ring = vsi->tx_rings[0];
75 	dev = tx_ring->dev;
76 
77 	/* we need two descriptors to add/del a filter and we can wait */
78 	do {
79 		if (I40E_DESC_UNUSED(tx_ring) > 1)
80 			break;
81 		msleep_interruptible(1);
82 		delay++;
83 	} while (delay < I40E_FD_CLEAN_DELAY);
84 
85 	if (!(I40E_DESC_UNUSED(tx_ring) > 1))
86 		return -EAGAIN;
87 
88 	dma = dma_map_single(dev, raw_packet,
89 			     I40E_FDIR_MAX_RAW_PACKET_SIZE, DMA_TO_DEVICE);
90 	if (dma_mapping_error(dev, dma))
91 		goto dma_fail;
92 
93 	/* grab the next descriptor */
94 	i = tx_ring->next_to_use;
95 	fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
96 	first = &tx_ring->tx_bi[i];
97 	memset(first, 0, sizeof(struct i40e_tx_buffer));
98 
99 	tx_ring->next_to_use = ((i + 1) < tx_ring->count) ? i + 1 : 0;
100 
101 	fpt = (fdir_data->q_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) &
102 	      I40E_TXD_FLTR_QW0_QINDEX_MASK;
103 
104 	fpt |= (fdir_data->flex_off << I40E_TXD_FLTR_QW0_FLEXOFF_SHIFT) &
105 	       I40E_TXD_FLTR_QW0_FLEXOFF_MASK;
106 
107 	fpt |= (fdir_data->pctype << I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) &
108 	       I40E_TXD_FLTR_QW0_PCTYPE_MASK;
109 
110 	/* Use LAN VSI Id if not programmed by user */
111 	if (fdir_data->dest_vsi == 0)
112 		fpt |= (pf->vsi[pf->lan_vsi]->id) <<
113 		       I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT;
114 	else
115 		fpt |= ((u32)fdir_data->dest_vsi <<
116 			I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT) &
117 		       I40E_TXD_FLTR_QW0_DEST_VSI_MASK;
118 
119 	dcc = I40E_TX_DESC_DTYPE_FILTER_PROG;
120 
121 	if (add)
122 		dcc |= I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
123 		       I40E_TXD_FLTR_QW1_PCMD_SHIFT;
124 	else
125 		dcc |= I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
126 		       I40E_TXD_FLTR_QW1_PCMD_SHIFT;
127 
128 	dcc |= (fdir_data->dest_ctl << I40E_TXD_FLTR_QW1_DEST_SHIFT) &
129 	       I40E_TXD_FLTR_QW1_DEST_MASK;
130 
131 	dcc |= (fdir_data->fd_status << I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT) &
132 	       I40E_TXD_FLTR_QW1_FD_STATUS_MASK;
133 
134 	if (fdir_data->cnt_index != 0) {
135 		dcc |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
136 		dcc |= ((u32)fdir_data->cnt_index <<
137 			I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
138 			I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
139 	}
140 
141 	fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(fpt);
142 	fdir_desc->rsvd = cpu_to_le32(0);
143 	fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dcc);
144 	fdir_desc->fd_id = cpu_to_le32(fdir_data->fd_id);
145 
146 	/* Now program a dummy descriptor */
147 	i = tx_ring->next_to_use;
148 	tx_desc = I40E_TX_DESC(tx_ring, i);
149 	tx_buf = &tx_ring->tx_bi[i];
150 
151 	tx_ring->next_to_use = ((i + 1) < tx_ring->count) ? i + 1 : 0;
152 
153 	memset(tx_buf, 0, sizeof(struct i40e_tx_buffer));
154 
155 	/* record length, and DMA address */
156 	dma_unmap_len_set(tx_buf, len, I40E_FDIR_MAX_RAW_PACKET_SIZE);
157 	dma_unmap_addr_set(tx_buf, dma, dma);
158 
159 	tx_desc->buffer_addr = cpu_to_le64(dma);
160 	td_cmd = I40E_TXD_CMD | I40E_TX_DESC_CMD_DUMMY;
161 
162 	tx_buf->tx_flags = I40E_TX_FLAGS_FD_SB;
163 	tx_buf->raw_buf = (void *)raw_packet;
164 
165 	tx_desc->cmd_type_offset_bsz =
166 		build_ctob(td_cmd, 0, I40E_FDIR_MAX_RAW_PACKET_SIZE, 0);
167 
168 	/* Force memory writes to complete before letting h/w
169 	 * know there are new descriptors to fetch.
170 	 */
171 	wmb();
172 
173 	/* Mark the data descriptor to be watched */
174 	first->next_to_watch = tx_desc;
175 
176 	writel(tx_ring->next_to_use, tx_ring->tail);
177 	return 0;
178 
179 dma_fail:
180 	return -1;
181 }
182 
183 #define IP_HEADER_OFFSET 14
184 #define I40E_UDPIP_DUMMY_PACKET_LEN 42
185 /**
186  * i40e_add_del_fdir_udpv4 - Add/Remove UDPv4 filters
187  * @vsi: pointer to the targeted VSI
188  * @fd_data: the flow director data required for the FDir descriptor
189  * @add: true adds a filter, false removes it
190  *
191  * Returns 0 if the filters were successfully added or removed
192  **/
193 static int i40e_add_del_fdir_udpv4(struct i40e_vsi *vsi,
194 				   struct i40e_fdir_filter *fd_data,
195 				   bool add)
196 {
197 	struct i40e_pf *pf = vsi->back;
198 	struct udphdr *udp;
199 	struct iphdr *ip;
200 	bool err = false;
201 	u8 *raw_packet;
202 	int ret;
203 	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
204 		0x45, 0, 0, 0x1c, 0, 0, 0x40, 0, 0x40, 0x11, 0, 0, 0, 0, 0, 0,
205 		0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
206 
207 	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
208 	if (!raw_packet)
209 		return -ENOMEM;
210 	memcpy(raw_packet, packet, I40E_UDPIP_DUMMY_PACKET_LEN);
211 
212 	ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
213 	udp = (struct udphdr *)(raw_packet + IP_HEADER_OFFSET
214 	      + sizeof(struct iphdr));
215 
216 	ip->daddr = fd_data->dst_ip[0];
217 	udp->dest = fd_data->dst_port;
218 	ip->saddr = fd_data->src_ip[0];
219 	udp->source = fd_data->src_port;
220 
221 	fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_UDP;
222 	ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
223 	if (ret) {
224 		dev_info(&pf->pdev->dev,
225 			 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
226 			 fd_data->pctype, fd_data->fd_id, ret);
227 		err = true;
228 	} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
229 		if (add)
230 			dev_info(&pf->pdev->dev,
231 				 "Filter OK for PCTYPE %d loc = %d\n",
232 				 fd_data->pctype, fd_data->fd_id);
233 		else
234 			dev_info(&pf->pdev->dev,
235 				 "Filter deleted for PCTYPE %d loc = %d\n",
236 				 fd_data->pctype, fd_data->fd_id);
237 	}
238 	if (err)
239 		kfree(raw_packet);
240 
241 	return err ? -EOPNOTSUPP : 0;
242 }
243 
244 #define I40E_TCPIP_DUMMY_PACKET_LEN 54
245 /**
246  * i40e_add_del_fdir_tcpv4 - Add/Remove TCPv4 filters
247  * @vsi: pointer to the targeted VSI
248  * @fd_data: the flow director data required for the FDir descriptor
249  * @add: true adds a filter, false removes it
250  *
251  * Returns 0 if the filters were successfully added or removed
252  **/
253 static int i40e_add_del_fdir_tcpv4(struct i40e_vsi *vsi,
254 				   struct i40e_fdir_filter *fd_data,
255 				   bool add)
256 {
257 	struct i40e_pf *pf = vsi->back;
258 	struct tcphdr *tcp;
259 	struct iphdr *ip;
260 	bool err = false;
261 	u8 *raw_packet;
262 	int ret;
263 	/* Dummy packet */
264 	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
265 		0x45, 0, 0, 0x28, 0, 0, 0x40, 0, 0x40, 0x6, 0, 0, 0, 0, 0, 0,
266 		0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x80, 0x11,
267 		0x0, 0x72, 0, 0, 0, 0};
268 
269 	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
270 	if (!raw_packet)
271 		return -ENOMEM;
272 	memcpy(raw_packet, packet, I40E_TCPIP_DUMMY_PACKET_LEN);
273 
274 	ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
275 	tcp = (struct tcphdr *)(raw_packet + IP_HEADER_OFFSET
276 	      + sizeof(struct iphdr));
277 
278 	ip->daddr = fd_data->dst_ip[0];
279 	tcp->dest = fd_data->dst_port;
280 	ip->saddr = fd_data->src_ip[0];
281 	tcp->source = fd_data->src_port;
282 
283 	if (add) {
284 		pf->fd_tcp_rule++;
285 		if (pf->flags & I40E_FLAG_FD_ATR_ENABLED) {
286 			if (I40E_DEBUG_FD & pf->hw.debug_mask)
287 				dev_info(&pf->pdev->dev, "Forcing ATR off, sideband rules for TCP/IPv4 flow being applied\n");
288 			pf->flags &= ~I40E_FLAG_FD_ATR_ENABLED;
289 		}
290 	} else {
291 		pf->fd_tcp_rule = (pf->fd_tcp_rule > 0) ?
292 				  (pf->fd_tcp_rule - 1) : 0;
293 		if (pf->fd_tcp_rule == 0) {
294 			pf->flags |= I40E_FLAG_FD_ATR_ENABLED;
295 			if (I40E_DEBUG_FD & pf->hw.debug_mask)
296 				dev_info(&pf->pdev->dev, "ATR re-enabled due to no sideband TCP/IPv4 rules\n");
297 		}
298 	}
299 
300 	fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_TCP;
301 	ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
302 
303 	if (ret) {
304 		dev_info(&pf->pdev->dev,
305 			 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
306 			 fd_data->pctype, fd_data->fd_id, ret);
307 		err = true;
308 	} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
309 		if (add)
310 			dev_info(&pf->pdev->dev, "Filter OK for PCTYPE %d loc = %d)\n",
311 				 fd_data->pctype, fd_data->fd_id);
312 		else
313 			dev_info(&pf->pdev->dev,
314 				 "Filter deleted for PCTYPE %d loc = %d\n",
315 				 fd_data->pctype, fd_data->fd_id);
316 	}
317 
318 	if (err)
319 		kfree(raw_packet);
320 
321 	return err ? -EOPNOTSUPP : 0;
322 }
323 
324 /**
325  * i40e_add_del_fdir_sctpv4 - Add/Remove SCTPv4 Flow Director filters for
326  * a specific flow spec
327  * @vsi: pointer to the targeted VSI
328  * @fd_data: the flow director data required for the FDir descriptor
329  * @add: true adds a filter, false removes it
330  *
331  * Returns 0 if the filters were successfully added or removed
332  **/
333 static int i40e_add_del_fdir_sctpv4(struct i40e_vsi *vsi,
334 				    struct i40e_fdir_filter *fd_data,
335 				    bool add)
336 {
337 	return -EOPNOTSUPP;
338 }
339 
340 #define I40E_IP_DUMMY_PACKET_LEN 34
341 /**
342  * i40e_add_del_fdir_ipv4 - Add/Remove IPv4 Flow Director filters for
343  * a specific flow spec
344  * @vsi: pointer to the targeted VSI
345  * @fd_data: the flow director data required for the FDir descriptor
346  * @add: true adds a filter, false removes it
347  *
348  * Returns 0 if the filters were successfully added or removed
349  **/
350 static int i40e_add_del_fdir_ipv4(struct i40e_vsi *vsi,
351 				  struct i40e_fdir_filter *fd_data,
352 				  bool add)
353 {
354 	struct i40e_pf *pf = vsi->back;
355 	struct iphdr *ip;
356 	bool err = false;
357 	u8 *raw_packet;
358 	int ret;
359 	int i;
360 	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
361 		0x45, 0, 0, 0x14, 0, 0, 0x40, 0, 0x40, 0x10, 0, 0, 0, 0, 0, 0,
362 		0, 0, 0, 0};
363 
364 	for (i = I40E_FILTER_PCTYPE_NONF_IPV4_OTHER;
365 	     i <= I40E_FILTER_PCTYPE_FRAG_IPV4;	i++) {
366 		raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
367 		if (!raw_packet)
368 			return -ENOMEM;
369 		memcpy(raw_packet, packet, I40E_IP_DUMMY_PACKET_LEN);
370 		ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
371 
372 		ip->saddr = fd_data->src_ip[0];
373 		ip->daddr = fd_data->dst_ip[0];
374 		ip->protocol = 0;
375 
376 		fd_data->pctype = i;
377 		ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
378 
379 		if (ret) {
380 			dev_info(&pf->pdev->dev,
381 				 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
382 				 fd_data->pctype, fd_data->fd_id, ret);
383 			err = true;
384 		} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
385 			if (add)
386 				dev_info(&pf->pdev->dev,
387 					 "Filter OK for PCTYPE %d loc = %d\n",
388 					 fd_data->pctype, fd_data->fd_id);
389 			else
390 				dev_info(&pf->pdev->dev,
391 					 "Filter deleted for PCTYPE %d loc = %d\n",
392 					 fd_data->pctype, fd_data->fd_id);
393 		}
394 	}
395 
396 	if (err)
397 		kfree(raw_packet);
398 
399 	return err ? -EOPNOTSUPP : 0;
400 }
401 
402 /**
403  * i40e_add_del_fdir - Build raw packets to add/del fdir filter
404  * @vsi: pointer to the targeted VSI
405  * @cmd: command to get or set RX flow classification rules
406  * @add: true adds a filter, false removes it
407  *
408  **/
409 int i40e_add_del_fdir(struct i40e_vsi *vsi,
410 		      struct i40e_fdir_filter *input, bool add)
411 {
412 	struct i40e_pf *pf = vsi->back;
413 	int ret;
414 
415 	switch (input->flow_type & ~FLOW_EXT) {
416 	case TCP_V4_FLOW:
417 		ret = i40e_add_del_fdir_tcpv4(vsi, input, add);
418 		break;
419 	case UDP_V4_FLOW:
420 		ret = i40e_add_del_fdir_udpv4(vsi, input, add);
421 		break;
422 	case SCTP_V4_FLOW:
423 		ret = i40e_add_del_fdir_sctpv4(vsi, input, add);
424 		break;
425 	case IPV4_FLOW:
426 		ret = i40e_add_del_fdir_ipv4(vsi, input, add);
427 		break;
428 	case IP_USER_FLOW:
429 		switch (input->ip4_proto) {
430 		case IPPROTO_TCP:
431 			ret = i40e_add_del_fdir_tcpv4(vsi, input, add);
432 			break;
433 		case IPPROTO_UDP:
434 			ret = i40e_add_del_fdir_udpv4(vsi, input, add);
435 			break;
436 		case IPPROTO_SCTP:
437 			ret = i40e_add_del_fdir_sctpv4(vsi, input, add);
438 			break;
439 		default:
440 			ret = i40e_add_del_fdir_ipv4(vsi, input, add);
441 			break;
442 		}
443 		break;
444 	default:
445 		dev_info(&pf->pdev->dev, "Could not specify spec type %d\n",
446 			 input->flow_type);
447 		ret = -EINVAL;
448 	}
449 
450 	/* The buffer allocated here is freed by the i40e_clean_tx_ring() */
451 	return ret;
452 }
453 
454 /**
455  * i40e_fd_handle_status - check the Programming Status for FD
456  * @rx_ring: the Rx ring for this descriptor
457  * @rx_desc: the Rx descriptor for programming Status, not a packet descriptor.
458  * @prog_id: the id originally used for programming
459  *
460  * This is used to verify if the FD programming or invalidation
461  * requested by SW to the HW is successful or not and take actions accordingly.
462  **/
463 static void i40e_fd_handle_status(struct i40e_ring *rx_ring,
464 				  union i40e_rx_desc *rx_desc, u8 prog_id)
465 {
466 	struct i40e_pf *pf = rx_ring->vsi->back;
467 	struct pci_dev *pdev = pf->pdev;
468 	u32 fcnt_prog, fcnt_avail;
469 	u32 error;
470 	u64 qw;
471 
472 	qw = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
473 	error = (qw & I40E_RX_PROG_STATUS_DESC_QW1_ERROR_MASK) >>
474 		I40E_RX_PROG_STATUS_DESC_QW1_ERROR_SHIFT;
475 
476 	if (error == BIT(I40E_RX_PROG_STATUS_DESC_FD_TBL_FULL_SHIFT)) {
477 		pf->fd_inv = le32_to_cpu(rx_desc->wb.qword0.hi_dword.fd_id);
478 		if ((rx_desc->wb.qword0.hi_dword.fd_id != 0) ||
479 		    (I40E_DEBUG_FD & pf->hw.debug_mask))
480 			dev_warn(&pdev->dev, "ntuple filter loc = %d, could not be added\n",
481 				 pf->fd_inv);
482 
483 		/* Check if the programming error is for ATR.
484 		 * If so, auto disable ATR and set a state for
485 		 * flush in progress. Next time we come here if flush is in
486 		 * progress do nothing, once flush is complete the state will
487 		 * be cleared.
488 		 */
489 		if (test_bit(__I40E_FD_FLUSH_REQUESTED, &pf->state))
490 			return;
491 
492 		pf->fd_add_err++;
493 		/* store the current atr filter count */
494 		pf->fd_atr_cnt = i40e_get_current_atr_cnt(pf);
495 
496 		if ((rx_desc->wb.qword0.hi_dword.fd_id == 0) &&
497 		    (pf->auto_disable_flags & I40E_FLAG_FD_SB_ENABLED)) {
498 			pf->auto_disable_flags |= I40E_FLAG_FD_ATR_ENABLED;
499 			set_bit(__I40E_FD_FLUSH_REQUESTED, &pf->state);
500 		}
501 
502 		/* filter programming failed most likely due to table full */
503 		fcnt_prog = i40e_get_global_fd_count(pf);
504 		fcnt_avail = pf->fdir_pf_filter_count;
505 		/* If ATR is running fcnt_prog can quickly change,
506 		 * if we are very close to full, it makes sense to disable
507 		 * FD ATR/SB and then re-enable it when there is room.
508 		 */
509 		if (fcnt_prog >= (fcnt_avail - I40E_FDIR_BUFFER_FULL_MARGIN)) {
510 			if ((pf->flags & I40E_FLAG_FD_SB_ENABLED) &&
511 			    !(pf->auto_disable_flags &
512 				     I40E_FLAG_FD_SB_ENABLED)) {
513 				if (I40E_DEBUG_FD & pf->hw.debug_mask)
514 					dev_warn(&pdev->dev, "FD filter space full, new ntuple rules will not be added\n");
515 				pf->auto_disable_flags |=
516 							I40E_FLAG_FD_SB_ENABLED;
517 			}
518 		}
519 	} else if (error == BIT(I40E_RX_PROG_STATUS_DESC_NO_FD_ENTRY_SHIFT)) {
520 		if (I40E_DEBUG_FD & pf->hw.debug_mask)
521 			dev_info(&pdev->dev, "ntuple filter fd_id = %d, could not be removed\n",
522 				 rx_desc->wb.qword0.hi_dword.fd_id);
523 	}
524 }
525 
526 /**
527  * i40e_unmap_and_free_tx_resource - Release a Tx buffer
528  * @ring:      the ring that owns the buffer
529  * @tx_buffer: the buffer to free
530  **/
531 static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring,
532 					    struct i40e_tx_buffer *tx_buffer)
533 {
534 	if (tx_buffer->skb) {
535 		dev_kfree_skb_any(tx_buffer->skb);
536 		if (dma_unmap_len(tx_buffer, len))
537 			dma_unmap_single(ring->dev,
538 					 dma_unmap_addr(tx_buffer, dma),
539 					 dma_unmap_len(tx_buffer, len),
540 					 DMA_TO_DEVICE);
541 	} else if (dma_unmap_len(tx_buffer, len)) {
542 		dma_unmap_page(ring->dev,
543 			       dma_unmap_addr(tx_buffer, dma),
544 			       dma_unmap_len(tx_buffer, len),
545 			       DMA_TO_DEVICE);
546 	}
547 
548 	if (tx_buffer->tx_flags & I40E_TX_FLAGS_FD_SB)
549 		kfree(tx_buffer->raw_buf);
550 
551 	tx_buffer->next_to_watch = NULL;
552 	tx_buffer->skb = NULL;
553 	dma_unmap_len_set(tx_buffer, len, 0);
554 	/* tx_buffer must be completely set up in the transmit path */
555 }
556 
557 /**
558  * i40e_clean_tx_ring - Free any empty Tx buffers
559  * @tx_ring: ring to be cleaned
560  **/
561 void i40e_clean_tx_ring(struct i40e_ring *tx_ring)
562 {
563 	unsigned long bi_size;
564 	u16 i;
565 
566 	/* ring already cleared, nothing to do */
567 	if (!tx_ring->tx_bi)
568 		return;
569 
570 	/* Free all the Tx ring sk_buffs */
571 	for (i = 0; i < tx_ring->count; i++)
572 		i40e_unmap_and_free_tx_resource(tx_ring, &tx_ring->tx_bi[i]);
573 
574 	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
575 	memset(tx_ring->tx_bi, 0, bi_size);
576 
577 	/* Zero out the descriptor ring */
578 	memset(tx_ring->desc, 0, tx_ring->size);
579 
580 	tx_ring->next_to_use = 0;
581 	tx_ring->next_to_clean = 0;
582 
583 	if (!tx_ring->netdev)
584 		return;
585 
586 	/* cleanup Tx queue statistics */
587 	netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
588 						  tx_ring->queue_index));
589 }
590 
591 /**
592  * i40e_free_tx_resources - Free Tx resources per queue
593  * @tx_ring: Tx descriptor ring for a specific queue
594  *
595  * Free all transmit software resources
596  **/
597 void i40e_free_tx_resources(struct i40e_ring *tx_ring)
598 {
599 	i40e_clean_tx_ring(tx_ring);
600 	kfree(tx_ring->tx_bi);
601 	tx_ring->tx_bi = NULL;
602 
603 	if (tx_ring->desc) {
604 		dma_free_coherent(tx_ring->dev, tx_ring->size,
605 				  tx_ring->desc, tx_ring->dma);
606 		tx_ring->desc = NULL;
607 	}
608 }
609 
610 /**
611  * i40e_get_tx_pending - how many tx descriptors not processed
612  * @tx_ring: the ring of descriptors
613  * @in_sw: is tx_pending being checked in SW or HW
614  *
615  * Since there is no access to the ring head register
616  * in XL710, we need to use our local copies
617  **/
618 u32 i40e_get_tx_pending(struct i40e_ring *ring, bool in_sw)
619 {
620 	u32 head, tail;
621 
622 	if (!in_sw)
623 		head = i40e_get_head(ring);
624 	else
625 		head = ring->next_to_clean;
626 	tail = readl(ring->tail);
627 
628 	if (head != tail)
629 		return (head < tail) ?
630 			tail - head : (tail + ring->count - head);
631 
632 	return 0;
633 }
634 
635 #define WB_STRIDE 0x3
636 
637 /**
638  * i40e_clean_tx_irq - Reclaim resources after transmit completes
639  * @vsi: the VSI we care about
640  * @tx_ring: Tx ring to clean
641  * @napi_budget: Used to determine if we are in netpoll
642  *
643  * Returns true if there's any budget left (e.g. the clean is finished)
644  **/
645 static bool i40e_clean_tx_irq(struct i40e_vsi *vsi,
646 			      struct i40e_ring *tx_ring, int napi_budget)
647 {
648 	u16 i = tx_ring->next_to_clean;
649 	struct i40e_tx_buffer *tx_buf;
650 	struct i40e_tx_desc *tx_head;
651 	struct i40e_tx_desc *tx_desc;
652 	unsigned int total_bytes = 0, total_packets = 0;
653 	unsigned int budget = vsi->work_limit;
654 
655 	tx_buf = &tx_ring->tx_bi[i];
656 	tx_desc = I40E_TX_DESC(tx_ring, i);
657 	i -= tx_ring->count;
658 
659 	tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
660 
661 	do {
662 		struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
663 
664 		/* if next_to_watch is not set then there is no work pending */
665 		if (!eop_desc)
666 			break;
667 
668 		/* prevent any other reads prior to eop_desc */
669 		read_barrier_depends();
670 
671 		/* we have caught up to head, no work left to do */
672 		if (tx_head == tx_desc)
673 			break;
674 
675 		/* clear next_to_watch to prevent false hangs */
676 		tx_buf->next_to_watch = NULL;
677 
678 		/* update the statistics for this packet */
679 		total_bytes += tx_buf->bytecount;
680 		total_packets += tx_buf->gso_segs;
681 
682 		/* free the skb */
683 		napi_consume_skb(tx_buf->skb, napi_budget);
684 
685 		/* unmap skb header data */
686 		dma_unmap_single(tx_ring->dev,
687 				 dma_unmap_addr(tx_buf, dma),
688 				 dma_unmap_len(tx_buf, len),
689 				 DMA_TO_DEVICE);
690 
691 		/* clear tx_buffer data */
692 		tx_buf->skb = NULL;
693 		dma_unmap_len_set(tx_buf, len, 0);
694 
695 		/* unmap remaining buffers */
696 		while (tx_desc != eop_desc) {
697 
698 			tx_buf++;
699 			tx_desc++;
700 			i++;
701 			if (unlikely(!i)) {
702 				i -= tx_ring->count;
703 				tx_buf = tx_ring->tx_bi;
704 				tx_desc = I40E_TX_DESC(tx_ring, 0);
705 			}
706 
707 			/* unmap any remaining paged data */
708 			if (dma_unmap_len(tx_buf, len)) {
709 				dma_unmap_page(tx_ring->dev,
710 					       dma_unmap_addr(tx_buf, dma),
711 					       dma_unmap_len(tx_buf, len),
712 					       DMA_TO_DEVICE);
713 				dma_unmap_len_set(tx_buf, len, 0);
714 			}
715 		}
716 
717 		/* move us one more past the eop_desc for start of next pkt */
718 		tx_buf++;
719 		tx_desc++;
720 		i++;
721 		if (unlikely(!i)) {
722 			i -= tx_ring->count;
723 			tx_buf = tx_ring->tx_bi;
724 			tx_desc = I40E_TX_DESC(tx_ring, 0);
725 		}
726 
727 		prefetch(tx_desc);
728 
729 		/* update budget accounting */
730 		budget--;
731 	} while (likely(budget));
732 
733 	i += tx_ring->count;
734 	tx_ring->next_to_clean = i;
735 	u64_stats_update_begin(&tx_ring->syncp);
736 	tx_ring->stats.bytes += total_bytes;
737 	tx_ring->stats.packets += total_packets;
738 	u64_stats_update_end(&tx_ring->syncp);
739 	tx_ring->q_vector->tx.total_bytes += total_bytes;
740 	tx_ring->q_vector->tx.total_packets += total_packets;
741 
742 	if (tx_ring->flags & I40E_TXR_FLAGS_WB_ON_ITR) {
743 		unsigned int j = 0;
744 
745 		/* check to see if there are < 4 descriptors
746 		 * waiting to be written back, then kick the hardware to force
747 		 * them to be written back in case we stay in NAPI.
748 		 * In this mode on X722 we do not enable Interrupt.
749 		 */
750 		j = i40e_get_tx_pending(tx_ring, false);
751 
752 		if (budget &&
753 		    ((j / (WB_STRIDE + 1)) == 0) && (j != 0) &&
754 		    !test_bit(__I40E_DOWN, &vsi->state) &&
755 		    (I40E_DESC_UNUSED(tx_ring) != tx_ring->count))
756 			tx_ring->arm_wb = true;
757 	}
758 
759 	netdev_tx_completed_queue(netdev_get_tx_queue(tx_ring->netdev,
760 						      tx_ring->queue_index),
761 				  total_packets, total_bytes);
762 
763 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
764 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
765 		     (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
766 		/* Make sure that anybody stopping the queue after this
767 		 * sees the new next_to_clean.
768 		 */
769 		smp_mb();
770 		if (__netif_subqueue_stopped(tx_ring->netdev,
771 					     tx_ring->queue_index) &&
772 		   !test_bit(__I40E_DOWN, &vsi->state)) {
773 			netif_wake_subqueue(tx_ring->netdev,
774 					    tx_ring->queue_index);
775 			++tx_ring->tx_stats.restart_queue;
776 		}
777 	}
778 
779 	return !!budget;
780 }
781 
782 /**
783  * i40e_enable_wb_on_itr - Arm hardware to do a wb, interrupts are not enabled
784  * @vsi: the VSI we care about
785  * @q_vector: the vector on which to enable writeback
786  *
787  **/
788 static void i40e_enable_wb_on_itr(struct i40e_vsi *vsi,
789 				  struct i40e_q_vector *q_vector)
790 {
791 	u16 flags = q_vector->tx.ring[0].flags;
792 	u32 val;
793 
794 	if (!(flags & I40E_TXR_FLAGS_WB_ON_ITR))
795 		return;
796 
797 	if (q_vector->arm_wb_state)
798 		return;
799 
800 	if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
801 		val = I40E_PFINT_DYN_CTLN_WB_ON_ITR_MASK |
802 		      I40E_PFINT_DYN_CTLN_ITR_INDX_MASK; /* set noitr */
803 
804 		wr32(&vsi->back->hw,
805 		     I40E_PFINT_DYN_CTLN(q_vector->v_idx + vsi->base_vector - 1),
806 		     val);
807 	} else {
808 		val = I40E_PFINT_DYN_CTL0_WB_ON_ITR_MASK |
809 		      I40E_PFINT_DYN_CTL0_ITR_INDX_MASK; /* set noitr */
810 
811 		wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
812 	}
813 	q_vector->arm_wb_state = true;
814 }
815 
816 /**
817  * i40e_force_wb - Issue SW Interrupt so HW does a wb
818  * @vsi: the VSI we care about
819  * @q_vector: the vector  on which to force writeback
820  *
821  **/
822 void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector)
823 {
824 	if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
825 		u32 val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
826 			  I40E_PFINT_DYN_CTLN_ITR_INDX_MASK | /* set noitr */
827 			  I40E_PFINT_DYN_CTLN_SWINT_TRIG_MASK |
828 			  I40E_PFINT_DYN_CTLN_SW_ITR_INDX_ENA_MASK;
829 			  /* allow 00 to be written to the index */
830 
831 		wr32(&vsi->back->hw,
832 		     I40E_PFINT_DYN_CTLN(q_vector->v_idx +
833 					 vsi->base_vector - 1), val);
834 	} else {
835 		u32 val = I40E_PFINT_DYN_CTL0_INTENA_MASK |
836 			  I40E_PFINT_DYN_CTL0_ITR_INDX_MASK | /* set noitr */
837 			  I40E_PFINT_DYN_CTL0_SWINT_TRIG_MASK |
838 			  I40E_PFINT_DYN_CTL0_SW_ITR_INDX_ENA_MASK;
839 			/* allow 00 to be written to the index */
840 
841 		wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
842 	}
843 }
844 
845 /**
846  * i40e_set_new_dynamic_itr - Find new ITR level
847  * @rc: structure containing ring performance data
848  *
849  * Returns true if ITR changed, false if not
850  *
851  * Stores a new ITR value based on packets and byte counts during
852  * the last interrupt.  The advantage of per interrupt computation
853  * is faster updates and more accurate ITR for the current traffic
854  * pattern.  Constants in this function were computed based on
855  * theoretical maximum wire speed and thresholds were set based on
856  * testing data as well as attempting to minimize response time
857  * while increasing bulk throughput.
858  **/
859 static bool i40e_set_new_dynamic_itr(struct i40e_ring_container *rc)
860 {
861 	enum i40e_latency_range new_latency_range = rc->latency_range;
862 	struct i40e_q_vector *qv = rc->ring->q_vector;
863 	u32 new_itr = rc->itr;
864 	int bytes_per_int;
865 	int usecs;
866 
867 	if (rc->total_packets == 0 || !rc->itr)
868 		return false;
869 
870 	/* simple throttlerate management
871 	 *   0-10MB/s   lowest (50000 ints/s)
872 	 *  10-20MB/s   low    (20000 ints/s)
873 	 *  20-1249MB/s bulk   (18000 ints/s)
874 	 *  > 40000 Rx packets per second (8000 ints/s)
875 	 *
876 	 * The math works out because the divisor is in 10^(-6) which
877 	 * turns the bytes/us input value into MB/s values, but
878 	 * make sure to use usecs, as the register values written
879 	 * are in 2 usec increments in the ITR registers, and make sure
880 	 * to use the smoothed values that the countdown timer gives us.
881 	 */
882 	usecs = (rc->itr << 1) * ITR_COUNTDOWN_START;
883 	bytes_per_int = rc->total_bytes / usecs;
884 
885 	switch (new_latency_range) {
886 	case I40E_LOWEST_LATENCY:
887 		if (bytes_per_int > 10)
888 			new_latency_range = I40E_LOW_LATENCY;
889 		break;
890 	case I40E_LOW_LATENCY:
891 		if (bytes_per_int > 20)
892 			new_latency_range = I40E_BULK_LATENCY;
893 		else if (bytes_per_int <= 10)
894 			new_latency_range = I40E_LOWEST_LATENCY;
895 		break;
896 	case I40E_BULK_LATENCY:
897 	case I40E_ULTRA_LATENCY:
898 	default:
899 		if (bytes_per_int <= 20)
900 			new_latency_range = I40E_LOW_LATENCY;
901 		break;
902 	}
903 
904 	/* this is to adjust RX more aggressively when streaming small
905 	 * packets.  The value of 40000 was picked as it is just beyond
906 	 * what the hardware can receive per second if in low latency
907 	 * mode.
908 	 */
909 #define RX_ULTRA_PACKET_RATE 40000
910 
911 	if ((((rc->total_packets * 1000000) / usecs) > RX_ULTRA_PACKET_RATE) &&
912 	    (&qv->rx == rc))
913 		new_latency_range = I40E_ULTRA_LATENCY;
914 
915 	rc->latency_range = new_latency_range;
916 
917 	switch (new_latency_range) {
918 	case I40E_LOWEST_LATENCY:
919 		new_itr = I40E_ITR_50K;
920 		break;
921 	case I40E_LOW_LATENCY:
922 		new_itr = I40E_ITR_20K;
923 		break;
924 	case I40E_BULK_LATENCY:
925 		new_itr = I40E_ITR_18K;
926 		break;
927 	case I40E_ULTRA_LATENCY:
928 		new_itr = I40E_ITR_8K;
929 		break;
930 	default:
931 		break;
932 	}
933 
934 	rc->total_bytes = 0;
935 	rc->total_packets = 0;
936 
937 	if (new_itr != rc->itr) {
938 		rc->itr = new_itr;
939 		return true;
940 	}
941 
942 	return false;
943 }
944 
945 /**
946  * i40e_clean_programming_status - clean the programming status descriptor
947  * @rx_ring: the rx ring that has this descriptor
948  * @rx_desc: the rx descriptor written back by HW
949  *
950  * Flow director should handle FD_FILTER_STATUS to check its filter programming
951  * status being successful or not and take actions accordingly. FCoE should
952  * handle its context/filter programming/invalidation status and take actions.
953  *
954  **/
955 static void i40e_clean_programming_status(struct i40e_ring *rx_ring,
956 					  union i40e_rx_desc *rx_desc)
957 {
958 	u64 qw;
959 	u8 id;
960 
961 	qw = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
962 	id = (qw & I40E_RX_PROG_STATUS_DESC_QW1_PROGID_MASK) >>
963 		  I40E_RX_PROG_STATUS_DESC_QW1_PROGID_SHIFT;
964 
965 	if (id == I40E_RX_PROG_STATUS_DESC_FD_FILTER_STATUS)
966 		i40e_fd_handle_status(rx_ring, rx_desc, id);
967 #ifdef I40E_FCOE
968 	else if ((id == I40E_RX_PROG_STATUS_DESC_FCOE_CTXT_PROG_STATUS) ||
969 		 (id == I40E_RX_PROG_STATUS_DESC_FCOE_CTXT_INVL_STATUS))
970 		i40e_fcoe_handle_status(rx_ring, rx_desc, id);
971 #endif
972 }
973 
974 /**
975  * i40e_setup_tx_descriptors - Allocate the Tx descriptors
976  * @tx_ring: the tx ring to set up
977  *
978  * Return 0 on success, negative on error
979  **/
980 int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring)
981 {
982 	struct device *dev = tx_ring->dev;
983 	int bi_size;
984 
985 	if (!dev)
986 		return -ENOMEM;
987 
988 	/* warn if we are about to overwrite the pointer */
989 	WARN_ON(tx_ring->tx_bi);
990 	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
991 	tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
992 	if (!tx_ring->tx_bi)
993 		goto err;
994 
995 	/* round up to nearest 4K */
996 	tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
997 	/* add u32 for head writeback, align after this takes care of
998 	 * guaranteeing this is at least one cache line in size
999 	 */
1000 	tx_ring->size += sizeof(u32);
1001 	tx_ring->size = ALIGN(tx_ring->size, 4096);
1002 	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
1003 					   &tx_ring->dma, GFP_KERNEL);
1004 	if (!tx_ring->desc) {
1005 		dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
1006 			 tx_ring->size);
1007 		goto err;
1008 	}
1009 
1010 	tx_ring->next_to_use = 0;
1011 	tx_ring->next_to_clean = 0;
1012 	return 0;
1013 
1014 err:
1015 	kfree(tx_ring->tx_bi);
1016 	tx_ring->tx_bi = NULL;
1017 	return -ENOMEM;
1018 }
1019 
1020 /**
1021  * i40e_clean_rx_ring - Free Rx buffers
1022  * @rx_ring: ring to be cleaned
1023  **/
1024 void i40e_clean_rx_ring(struct i40e_ring *rx_ring)
1025 {
1026 	struct device *dev = rx_ring->dev;
1027 	unsigned long bi_size;
1028 	u16 i;
1029 
1030 	/* ring already cleared, nothing to do */
1031 	if (!rx_ring->rx_bi)
1032 		return;
1033 
1034 	/* Free all the Rx ring sk_buffs */
1035 	for (i = 0; i < rx_ring->count; i++) {
1036 		struct i40e_rx_buffer *rx_bi = &rx_ring->rx_bi[i];
1037 
1038 		if (rx_bi->skb) {
1039 			dev_kfree_skb(rx_bi->skb);
1040 			rx_bi->skb = NULL;
1041 		}
1042 		if (!rx_bi->page)
1043 			continue;
1044 
1045 		dma_unmap_page(dev, rx_bi->dma, PAGE_SIZE, DMA_FROM_DEVICE);
1046 		__free_pages(rx_bi->page, 0);
1047 
1048 		rx_bi->page = NULL;
1049 		rx_bi->page_offset = 0;
1050 	}
1051 
1052 	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
1053 	memset(rx_ring->rx_bi, 0, bi_size);
1054 
1055 	/* Zero out the descriptor ring */
1056 	memset(rx_ring->desc, 0, rx_ring->size);
1057 
1058 	rx_ring->next_to_alloc = 0;
1059 	rx_ring->next_to_clean = 0;
1060 	rx_ring->next_to_use = 0;
1061 }
1062 
1063 /**
1064  * i40e_free_rx_resources - Free Rx resources
1065  * @rx_ring: ring to clean the resources from
1066  *
1067  * Free all receive software resources
1068  **/
1069 void i40e_free_rx_resources(struct i40e_ring *rx_ring)
1070 {
1071 	i40e_clean_rx_ring(rx_ring);
1072 	kfree(rx_ring->rx_bi);
1073 	rx_ring->rx_bi = NULL;
1074 
1075 	if (rx_ring->desc) {
1076 		dma_free_coherent(rx_ring->dev, rx_ring->size,
1077 				  rx_ring->desc, rx_ring->dma);
1078 		rx_ring->desc = NULL;
1079 	}
1080 }
1081 
1082 /**
1083  * i40e_setup_rx_descriptors - Allocate Rx descriptors
1084  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
1085  *
1086  * Returns 0 on success, negative on failure
1087  **/
1088 int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring)
1089 {
1090 	struct device *dev = rx_ring->dev;
1091 	int bi_size;
1092 
1093 	/* warn if we are about to overwrite the pointer */
1094 	WARN_ON(rx_ring->rx_bi);
1095 	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
1096 	rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
1097 	if (!rx_ring->rx_bi)
1098 		goto err;
1099 
1100 	u64_stats_init(&rx_ring->syncp);
1101 
1102 	/* Round up to nearest 4K */
1103 	rx_ring->size = rx_ring->count * sizeof(union i40e_32byte_rx_desc);
1104 	rx_ring->size = ALIGN(rx_ring->size, 4096);
1105 	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
1106 					   &rx_ring->dma, GFP_KERNEL);
1107 
1108 	if (!rx_ring->desc) {
1109 		dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
1110 			 rx_ring->size);
1111 		goto err;
1112 	}
1113 
1114 	rx_ring->next_to_alloc = 0;
1115 	rx_ring->next_to_clean = 0;
1116 	rx_ring->next_to_use = 0;
1117 
1118 	return 0;
1119 err:
1120 	kfree(rx_ring->rx_bi);
1121 	rx_ring->rx_bi = NULL;
1122 	return -ENOMEM;
1123 }
1124 
1125 /**
1126  * i40e_release_rx_desc - Store the new tail and head values
1127  * @rx_ring: ring to bump
1128  * @val: new head index
1129  **/
1130 static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
1131 {
1132 	rx_ring->next_to_use = val;
1133 
1134 	/* update next to alloc since we have filled the ring */
1135 	rx_ring->next_to_alloc = val;
1136 
1137 	/* Force memory writes to complete before letting h/w
1138 	 * know there are new descriptors to fetch.  (Only
1139 	 * applicable for weak-ordered memory model archs,
1140 	 * such as IA-64).
1141 	 */
1142 	wmb();
1143 	writel(val, rx_ring->tail);
1144 }
1145 
1146 /**
1147  * i40e_alloc_mapped_page - recycle or make a new page
1148  * @rx_ring: ring to use
1149  * @bi: rx_buffer struct to modify
1150  *
1151  * Returns true if the page was successfully allocated or
1152  * reused.
1153  **/
1154 static bool i40e_alloc_mapped_page(struct i40e_ring *rx_ring,
1155 				   struct i40e_rx_buffer *bi)
1156 {
1157 	struct page *page = bi->page;
1158 	dma_addr_t dma;
1159 
1160 	/* since we are recycling buffers we should seldom need to alloc */
1161 	if (likely(page)) {
1162 		rx_ring->rx_stats.page_reuse_count++;
1163 		return true;
1164 	}
1165 
1166 	/* alloc new page for storage */
1167 	page = dev_alloc_page();
1168 	if (unlikely(!page)) {
1169 		rx_ring->rx_stats.alloc_page_failed++;
1170 		return false;
1171 	}
1172 
1173 	/* map page for use */
1174 	dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
1175 
1176 	/* if mapping failed free memory back to system since
1177 	 * there isn't much point in holding memory we can't use
1178 	 */
1179 	if (dma_mapping_error(rx_ring->dev, dma)) {
1180 		__free_pages(page, 0);
1181 		rx_ring->rx_stats.alloc_page_failed++;
1182 		return false;
1183 	}
1184 
1185 	bi->dma = dma;
1186 	bi->page = page;
1187 	bi->page_offset = 0;
1188 
1189 	return true;
1190 }
1191 
1192 /**
1193  * i40e_receive_skb - Send a completed packet up the stack
1194  * @rx_ring:  rx ring in play
1195  * @skb: packet to send up
1196  * @vlan_tag: vlan tag for packet
1197  **/
1198 static void i40e_receive_skb(struct i40e_ring *rx_ring,
1199 			     struct sk_buff *skb, u16 vlan_tag)
1200 {
1201 	struct i40e_q_vector *q_vector = rx_ring->q_vector;
1202 
1203 	if ((rx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1204 	    (vlan_tag & VLAN_VID_MASK))
1205 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
1206 
1207 	napi_gro_receive(&q_vector->napi, skb);
1208 }
1209 
1210 /**
1211  * i40e_alloc_rx_buffers - Replace used receive buffers
1212  * @rx_ring: ring to place buffers on
1213  * @cleaned_count: number of buffers to replace
1214  *
1215  * Returns false if all allocations were successful, true if any fail
1216  **/
1217 bool i40e_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count)
1218 {
1219 	u16 ntu = rx_ring->next_to_use;
1220 	union i40e_rx_desc *rx_desc;
1221 	struct i40e_rx_buffer *bi;
1222 
1223 	/* do nothing if no valid netdev defined */
1224 	if (!rx_ring->netdev || !cleaned_count)
1225 		return false;
1226 
1227 	rx_desc = I40E_RX_DESC(rx_ring, ntu);
1228 	bi = &rx_ring->rx_bi[ntu];
1229 
1230 	do {
1231 		if (!i40e_alloc_mapped_page(rx_ring, bi))
1232 			goto no_buffers;
1233 
1234 		/* Refresh the desc even if buffer_addrs didn't change
1235 		 * because each write-back erases this info.
1236 		 */
1237 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1238 		rx_desc->read.hdr_addr = 0;
1239 
1240 		rx_desc++;
1241 		bi++;
1242 		ntu++;
1243 		if (unlikely(ntu == rx_ring->count)) {
1244 			rx_desc = I40E_RX_DESC(rx_ring, 0);
1245 			bi = rx_ring->rx_bi;
1246 			ntu = 0;
1247 		}
1248 
1249 		/* clear the status bits for the next_to_use descriptor */
1250 		rx_desc->wb.qword1.status_error_len = 0;
1251 
1252 		cleaned_count--;
1253 	} while (cleaned_count);
1254 
1255 	if (rx_ring->next_to_use != ntu)
1256 		i40e_release_rx_desc(rx_ring, ntu);
1257 
1258 	return false;
1259 
1260 no_buffers:
1261 	if (rx_ring->next_to_use != ntu)
1262 		i40e_release_rx_desc(rx_ring, ntu);
1263 
1264 	/* make sure to come back via polling to try again after
1265 	 * allocation failure
1266 	 */
1267 	return true;
1268 }
1269 
1270 /**
1271  * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
1272  * @vsi: the VSI we care about
1273  * @skb: skb currently being received and modified
1274  * @rx_desc: the receive descriptor
1275  *
1276  * skb->protocol must be set before this function is called
1277  **/
1278 static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
1279 				    struct sk_buff *skb,
1280 				    union i40e_rx_desc *rx_desc)
1281 {
1282 	struct i40e_rx_ptype_decoded decoded;
1283 	bool ipv4, ipv6, tunnel = false;
1284 	u32 rx_error, rx_status;
1285 	u8 ptype;
1286 	u64 qword;
1287 
1288 	qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1289 	ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >> I40E_RXD_QW1_PTYPE_SHIFT;
1290 	rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1291 		   I40E_RXD_QW1_ERROR_SHIFT;
1292 	rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1293 		    I40E_RXD_QW1_STATUS_SHIFT;
1294 	decoded = decode_rx_desc_ptype(ptype);
1295 
1296 	skb->ip_summed = CHECKSUM_NONE;
1297 
1298 	skb_checksum_none_assert(skb);
1299 
1300 	/* Rx csum enabled and ip headers found? */
1301 	if (!(vsi->netdev->features & NETIF_F_RXCSUM))
1302 		return;
1303 
1304 	/* did the hardware decode the packet and checksum? */
1305 	if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
1306 		return;
1307 
1308 	/* both known and outer_ip must be set for the below code to work */
1309 	if (!(decoded.known && decoded.outer_ip))
1310 		return;
1311 
1312 	ipv4 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1313 	       (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4);
1314 	ipv6 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1315 	       (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6);
1316 
1317 	if (ipv4 &&
1318 	    (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) |
1319 			 BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT))))
1320 		goto checksum_fail;
1321 
1322 	/* likely incorrect csum if alternate IP extension headers found */
1323 	if (ipv6 &&
1324 	    rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
1325 		/* don't increment checksum err here, non-fatal err */
1326 		return;
1327 
1328 	/* there was some L4 error, count error and punt packet to the stack */
1329 	if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT))
1330 		goto checksum_fail;
1331 
1332 	/* handle packets that were not able to be checksummed due
1333 	 * to arrival speed, in this case the stack can compute
1334 	 * the csum.
1335 	 */
1336 	if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT))
1337 		return;
1338 
1339 	/* The hardware supported by this driver does not validate outer
1340 	 * checksums for tunneled VXLAN or GENEVE frames.  I don't agree
1341 	 * with it but the specification states that you "MAY validate", it
1342 	 * doesn't make it a hard requirement so if we have validated the
1343 	 * inner checksum report CHECKSUM_UNNECESSARY.
1344 	 */
1345 	if (decoded.inner_prot & (I40E_RX_PTYPE_INNER_PROT_TCP |
1346 				  I40E_RX_PTYPE_INNER_PROT_UDP |
1347 				  I40E_RX_PTYPE_INNER_PROT_SCTP))
1348 		tunnel = true;
1349 
1350 	skb->ip_summed = CHECKSUM_UNNECESSARY;
1351 	skb->csum_level = tunnel ? 1 : 0;
1352 
1353 	return;
1354 
1355 checksum_fail:
1356 	vsi->back->hw_csum_rx_error++;
1357 }
1358 
1359 /**
1360  * i40e_ptype_to_htype - get a hash type
1361  * @ptype: the ptype value from the descriptor
1362  *
1363  * Returns a hash type to be used by skb_set_hash
1364  **/
1365 static inline int i40e_ptype_to_htype(u8 ptype)
1366 {
1367 	struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
1368 
1369 	if (!decoded.known)
1370 		return PKT_HASH_TYPE_NONE;
1371 
1372 	if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1373 	    decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
1374 		return PKT_HASH_TYPE_L4;
1375 	else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1376 		 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
1377 		return PKT_HASH_TYPE_L3;
1378 	else
1379 		return PKT_HASH_TYPE_L2;
1380 }
1381 
1382 /**
1383  * i40e_rx_hash - set the hash value in the skb
1384  * @ring: descriptor ring
1385  * @rx_desc: specific descriptor
1386  **/
1387 static inline void i40e_rx_hash(struct i40e_ring *ring,
1388 				union i40e_rx_desc *rx_desc,
1389 				struct sk_buff *skb,
1390 				u8 rx_ptype)
1391 {
1392 	u32 hash;
1393 	const __le64 rss_mask =
1394 		cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
1395 			    I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
1396 
1397 	if (!(ring->netdev->features & NETIF_F_RXHASH))
1398 		return;
1399 
1400 	if ((rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask) {
1401 		hash = le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
1402 		skb_set_hash(skb, hash, i40e_ptype_to_htype(rx_ptype));
1403 	}
1404 }
1405 
1406 /**
1407  * i40e_process_skb_fields - Populate skb header fields from Rx descriptor
1408  * @rx_ring: rx descriptor ring packet is being transacted on
1409  * @rx_desc: pointer to the EOP Rx descriptor
1410  * @skb: pointer to current skb being populated
1411  * @rx_ptype: the packet type decoded by hardware
1412  *
1413  * This function checks the ring, descriptor, and packet information in
1414  * order to populate the hash, checksum, VLAN, protocol, and
1415  * other fields within the skb.
1416  **/
1417 static inline
1418 void i40e_process_skb_fields(struct i40e_ring *rx_ring,
1419 			     union i40e_rx_desc *rx_desc, struct sk_buff *skb,
1420 			     u8 rx_ptype)
1421 {
1422 	u64 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1423 	u32 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1424 			I40E_RXD_QW1_STATUS_SHIFT;
1425 	u32 rsyn = (rx_status & I40E_RXD_QW1_STATUS_TSYNINDX_MASK) >>
1426 		   I40E_RXD_QW1_STATUS_TSYNINDX_SHIFT;
1427 
1428 	if (unlikely(rsyn)) {
1429 		i40e_ptp_rx_hwtstamp(rx_ring->vsi->back, skb, rsyn);
1430 		rx_ring->last_rx_timestamp = jiffies;
1431 	}
1432 
1433 	i40e_rx_hash(rx_ring, rx_desc, skb, rx_ptype);
1434 
1435 	/* modifies the skb - consumes the enet header */
1436 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1437 
1438 	i40e_rx_checksum(rx_ring->vsi, skb, rx_desc);
1439 
1440 	skb_record_rx_queue(skb, rx_ring->queue_index);
1441 }
1442 
1443 /**
1444  * i40e_pull_tail - i40e specific version of skb_pull_tail
1445  * @rx_ring: rx descriptor ring packet is being transacted on
1446  * @skb: pointer to current skb being adjusted
1447  *
1448  * This function is an i40e specific version of __pskb_pull_tail.  The
1449  * main difference between this version and the original function is that
1450  * this function can make several assumptions about the state of things
1451  * that allow for significant optimizations versus the standard function.
1452  * As a result we can do things like drop a frag and maintain an accurate
1453  * truesize for the skb.
1454  */
1455 static void i40e_pull_tail(struct i40e_ring *rx_ring, struct sk_buff *skb)
1456 {
1457 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1458 	unsigned char *va;
1459 	unsigned int pull_len;
1460 
1461 	/* it is valid to use page_address instead of kmap since we are
1462 	 * working with pages allocated out of the lomem pool per
1463 	 * alloc_page(GFP_ATOMIC)
1464 	 */
1465 	va = skb_frag_address(frag);
1466 
1467 	/* we need the header to contain the greater of either ETH_HLEN or
1468 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
1469 	 */
1470 	pull_len = eth_get_headlen(va, I40E_RX_HDR_SIZE);
1471 
1472 	/* align pull length to size of long to optimize memcpy performance */
1473 	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1474 
1475 	/* update all of the pointers */
1476 	skb_frag_size_sub(frag, pull_len);
1477 	frag->page_offset += pull_len;
1478 	skb->data_len -= pull_len;
1479 	skb->tail += pull_len;
1480 }
1481 
1482 /**
1483  * i40e_cleanup_headers - Correct empty headers
1484  * @rx_ring: rx descriptor ring packet is being transacted on
1485  * @skb: pointer to current skb being fixed
1486  *
1487  * Also address the case where we are pulling data in on pages only
1488  * and as such no data is present in the skb header.
1489  *
1490  * In addition if skb is not at least 60 bytes we need to pad it so that
1491  * it is large enough to qualify as a valid Ethernet frame.
1492  *
1493  * Returns true if an error was encountered and skb was freed.
1494  **/
1495 static bool i40e_cleanup_headers(struct i40e_ring *rx_ring, struct sk_buff *skb)
1496 {
1497 	/* place header in linear portion of buffer */
1498 	if (skb_is_nonlinear(skb))
1499 		i40e_pull_tail(rx_ring, skb);
1500 
1501 	/* if eth_skb_pad returns an error the skb was freed */
1502 	if (eth_skb_pad(skb))
1503 		return true;
1504 
1505 	return false;
1506 }
1507 
1508 /**
1509  * i40e_reuse_rx_page - page flip buffer and store it back on the ring
1510  * @rx_ring: rx descriptor ring to store buffers on
1511  * @old_buff: donor buffer to have page reused
1512  *
1513  * Synchronizes page for reuse by the adapter
1514  **/
1515 static void i40e_reuse_rx_page(struct i40e_ring *rx_ring,
1516 			       struct i40e_rx_buffer *old_buff)
1517 {
1518 	struct i40e_rx_buffer *new_buff;
1519 	u16 nta = rx_ring->next_to_alloc;
1520 
1521 	new_buff = &rx_ring->rx_bi[nta];
1522 
1523 	/* update, and store next to alloc */
1524 	nta++;
1525 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1526 
1527 	/* transfer page from old buffer to new buffer */
1528 	*new_buff = *old_buff;
1529 }
1530 
1531 /**
1532  * i40e_page_is_reserved - check if reuse is possible
1533  * @page: page struct to check
1534  */
1535 static inline bool i40e_page_is_reserved(struct page *page)
1536 {
1537 	return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1538 }
1539 
1540 /**
1541  * i40e_add_rx_frag - Add contents of Rx buffer to sk_buff
1542  * @rx_ring: rx descriptor ring to transact packets on
1543  * @rx_buffer: buffer containing page to add
1544  * @rx_desc: descriptor containing length of buffer written by hardware
1545  * @skb: sk_buff to place the data into
1546  *
1547  * This function will add the data contained in rx_buffer->page to the skb.
1548  * This is done either through a direct copy if the data in the buffer is
1549  * less than the skb header size, otherwise it will just attach the page as
1550  * a frag to the skb.
1551  *
1552  * The function will then update the page offset if necessary and return
1553  * true if the buffer can be reused by the adapter.
1554  **/
1555 static bool i40e_add_rx_frag(struct i40e_ring *rx_ring,
1556 			     struct i40e_rx_buffer *rx_buffer,
1557 			     union i40e_rx_desc *rx_desc,
1558 			     struct sk_buff *skb)
1559 {
1560 	struct page *page = rx_buffer->page;
1561 	u64 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1562 	unsigned int size = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1563 			    I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1564 #if (PAGE_SIZE < 8192)
1565 	unsigned int truesize = I40E_RXBUFFER_2048;
1566 #else
1567 	unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
1568 	unsigned int last_offset = PAGE_SIZE - I40E_RXBUFFER_2048;
1569 #endif
1570 
1571 	/* will the data fit in the skb we allocated? if so, just
1572 	 * copy it as it is pretty small anyway
1573 	 */
1574 	if ((size <= I40E_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
1575 		unsigned char *va = page_address(page) + rx_buffer->page_offset;
1576 
1577 		memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
1578 
1579 		/* page is not reserved, we can reuse buffer as-is */
1580 		if (likely(!i40e_page_is_reserved(page)))
1581 			return true;
1582 
1583 		/* this page cannot be reused so discard it */
1584 		__free_pages(page, 0);
1585 		return false;
1586 	}
1587 
1588 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1589 			rx_buffer->page_offset, size, truesize);
1590 
1591 	/* avoid re-using remote pages */
1592 	if (unlikely(i40e_page_is_reserved(page)))
1593 		return false;
1594 
1595 #if (PAGE_SIZE < 8192)
1596 	/* if we are only owner of page we can reuse it */
1597 	if (unlikely(page_count(page) != 1))
1598 		return false;
1599 
1600 	/* flip page offset to other buffer */
1601 	rx_buffer->page_offset ^= truesize;
1602 #else
1603 	/* move offset up to the next cache line */
1604 	rx_buffer->page_offset += truesize;
1605 
1606 	if (rx_buffer->page_offset > last_offset)
1607 		return false;
1608 #endif
1609 
1610 	/* Even if we own the page, we are not allowed to use atomic_set()
1611 	 * This would break get_page_unless_zero() users.
1612 	 */
1613 	get_page(rx_buffer->page);
1614 
1615 	return true;
1616 }
1617 
1618 /**
1619  * i40e_fetch_rx_buffer - Allocate skb and populate it
1620  * @rx_ring: rx descriptor ring to transact packets on
1621  * @rx_desc: descriptor containing info written by hardware
1622  *
1623  * This function allocates an skb on the fly, and populates it with the page
1624  * data from the current receive descriptor, taking care to set up the skb
1625  * correctly, as well as handling calling the page recycle function if
1626  * necessary.
1627  */
1628 static inline
1629 struct sk_buff *i40e_fetch_rx_buffer(struct i40e_ring *rx_ring,
1630 				     union i40e_rx_desc *rx_desc)
1631 {
1632 	struct i40e_rx_buffer *rx_buffer;
1633 	struct sk_buff *skb;
1634 	struct page *page;
1635 
1636 	rx_buffer = &rx_ring->rx_bi[rx_ring->next_to_clean];
1637 	page = rx_buffer->page;
1638 	prefetchw(page);
1639 
1640 	skb = rx_buffer->skb;
1641 
1642 	if (likely(!skb)) {
1643 		void *page_addr = page_address(page) + rx_buffer->page_offset;
1644 
1645 		/* prefetch first cache line of first page */
1646 		prefetch(page_addr);
1647 #if L1_CACHE_BYTES < 128
1648 		prefetch(page_addr + L1_CACHE_BYTES);
1649 #endif
1650 
1651 		/* allocate a skb to store the frags */
1652 		skb = __napi_alloc_skb(&rx_ring->q_vector->napi,
1653 				       I40E_RX_HDR_SIZE,
1654 				       GFP_ATOMIC | __GFP_NOWARN);
1655 		if (unlikely(!skb)) {
1656 			rx_ring->rx_stats.alloc_buff_failed++;
1657 			return NULL;
1658 		}
1659 
1660 		/* we will be copying header into skb->data in
1661 		 * pskb_may_pull so it is in our interest to prefetch
1662 		 * it now to avoid a possible cache miss
1663 		 */
1664 		prefetchw(skb->data);
1665 	} else {
1666 		rx_buffer->skb = NULL;
1667 	}
1668 
1669 	/* we are reusing so sync this buffer for CPU use */
1670 	dma_sync_single_range_for_cpu(rx_ring->dev,
1671 				      rx_buffer->dma,
1672 				      rx_buffer->page_offset,
1673 				      I40E_RXBUFFER_2048,
1674 				      DMA_FROM_DEVICE);
1675 
1676 	/* pull page into skb */
1677 	if (i40e_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
1678 		/* hand second half of page back to the ring */
1679 		i40e_reuse_rx_page(rx_ring, rx_buffer);
1680 		rx_ring->rx_stats.page_reuse_count++;
1681 	} else {
1682 		/* we are not reusing the buffer so unmap it */
1683 		dma_unmap_page(rx_ring->dev, rx_buffer->dma, PAGE_SIZE,
1684 			       DMA_FROM_DEVICE);
1685 	}
1686 
1687 	/* clear contents of buffer_info */
1688 	rx_buffer->page = NULL;
1689 
1690 	return skb;
1691 }
1692 
1693 /**
1694  * i40e_is_non_eop - process handling of non-EOP buffers
1695  * @rx_ring: Rx ring being processed
1696  * @rx_desc: Rx descriptor for current buffer
1697  * @skb: Current socket buffer containing buffer in progress
1698  *
1699  * This function updates next to clean.  If the buffer is an EOP buffer
1700  * this function exits returning false, otherwise it will place the
1701  * sk_buff in the next buffer to be chained and return true indicating
1702  * that this is in fact a non-EOP buffer.
1703  **/
1704 static bool i40e_is_non_eop(struct i40e_ring *rx_ring,
1705 			    union i40e_rx_desc *rx_desc,
1706 			    struct sk_buff *skb)
1707 {
1708 	u32 ntc = rx_ring->next_to_clean + 1;
1709 
1710 	/* fetch, update, and store next to clean */
1711 	ntc = (ntc < rx_ring->count) ? ntc : 0;
1712 	rx_ring->next_to_clean = ntc;
1713 
1714 	prefetch(I40E_RX_DESC(rx_ring, ntc));
1715 
1716 #define staterrlen rx_desc->wb.qword1.status_error_len
1717 	if (unlikely(i40e_rx_is_programming_status(le64_to_cpu(staterrlen)))) {
1718 		i40e_clean_programming_status(rx_ring, rx_desc);
1719 		rx_ring->rx_bi[ntc].skb = skb;
1720 		return true;
1721 	}
1722 	/* if we are the last buffer then there is nothing else to do */
1723 #define I40E_RXD_EOF BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)
1724 	if (likely(i40e_test_staterr(rx_desc, I40E_RXD_EOF)))
1725 		return false;
1726 
1727 	/* place skb in next buffer to be received */
1728 	rx_ring->rx_bi[ntc].skb = skb;
1729 	rx_ring->rx_stats.non_eop_descs++;
1730 
1731 	return true;
1732 }
1733 
1734 /**
1735  * i40e_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1736  * @rx_ring: rx descriptor ring to transact packets on
1737  * @budget: Total limit on number of packets to process
1738  *
1739  * This function provides a "bounce buffer" approach to Rx interrupt
1740  * processing.  The advantage to this is that on systems that have
1741  * expensive overhead for IOMMU access this provides a means of avoiding
1742  * it by maintaining the mapping of the page to the system.
1743  *
1744  * Returns amount of work completed
1745  **/
1746 static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget)
1747 {
1748 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1749 	u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
1750 	bool failure = false;
1751 
1752 	while (likely(total_rx_packets < budget)) {
1753 		union i40e_rx_desc *rx_desc;
1754 		struct sk_buff *skb;
1755 		u32 rx_status;
1756 		u16 vlan_tag;
1757 		u8 rx_ptype;
1758 		u64 qword;
1759 
1760 		/* return some buffers to hardware, one at a time is too slow */
1761 		if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1762 			failure = failure ||
1763 				  i40e_alloc_rx_buffers(rx_ring, cleaned_count);
1764 			cleaned_count = 0;
1765 		}
1766 
1767 		rx_desc = I40E_RX_DESC(rx_ring, rx_ring->next_to_clean);
1768 
1769 		qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1770 		rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1771 			   I40E_RXD_QW1_PTYPE_SHIFT;
1772 		rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1773 			    I40E_RXD_QW1_STATUS_SHIFT;
1774 
1775 		if (!(rx_status & BIT(I40E_RX_DESC_STATUS_DD_SHIFT)))
1776 			break;
1777 
1778 		/* status_error_len will always be zero for unused descriptors
1779 		 * because it's cleared in cleanup, and overlaps with hdr_addr
1780 		 * which is always zero because packet split isn't used, if the
1781 		 * hardware wrote DD then it will be non-zero
1782 		 */
1783 		if (!rx_desc->wb.qword1.status_error_len)
1784 			break;
1785 
1786 		/* This memory barrier is needed to keep us from reading
1787 		 * any other fields out of the rx_desc until we know the
1788 		 * DD bit is set.
1789 		 */
1790 		dma_rmb();
1791 
1792 		skb = i40e_fetch_rx_buffer(rx_ring, rx_desc);
1793 		if (!skb)
1794 			break;
1795 
1796 		cleaned_count++;
1797 
1798 		if (i40e_is_non_eop(rx_ring, rx_desc, skb))
1799 			continue;
1800 
1801 		/* ERR_MASK will only have valid bits if EOP set, and
1802 		 * what we are doing here is actually checking
1803 		 * I40E_RX_DESC_ERROR_RXE_SHIFT, since it is the zeroth bit in
1804 		 * the error field
1805 		 */
1806 		if (unlikely(i40e_test_staterr(rx_desc, BIT(I40E_RXD_QW1_ERROR_SHIFT)))) {
1807 			dev_kfree_skb_any(skb);
1808 			continue;
1809 		}
1810 
1811 		if (i40e_cleanup_headers(rx_ring, skb))
1812 			continue;
1813 
1814 		/* probably a little skewed due to removing CRC */
1815 		total_rx_bytes += skb->len;
1816 
1817 		/* populate checksum, VLAN, and protocol */
1818 		i40e_process_skb_fields(rx_ring, rx_desc, skb, rx_ptype);
1819 
1820 #ifdef I40E_FCOE
1821 		if (unlikely(
1822 		    i40e_rx_is_fcoe(rx_ptype) &&
1823 		    !i40e_fcoe_handle_offload(rx_ring, rx_desc, skb))) {
1824 			dev_kfree_skb_any(skb);
1825 			continue;
1826 		}
1827 #endif
1828 
1829 		vlan_tag = (qword & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)) ?
1830 			   le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1) : 0;
1831 
1832 		i40e_receive_skb(rx_ring, skb, vlan_tag);
1833 
1834 		/* update budget accounting */
1835 		total_rx_packets++;
1836 	}
1837 
1838 	u64_stats_update_begin(&rx_ring->syncp);
1839 	rx_ring->stats.packets += total_rx_packets;
1840 	rx_ring->stats.bytes += total_rx_bytes;
1841 	u64_stats_update_end(&rx_ring->syncp);
1842 	rx_ring->q_vector->rx.total_packets += total_rx_packets;
1843 	rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1844 
1845 	/* guarantee a trip back through this routine if there was a failure */
1846 	return failure ? budget : total_rx_packets;
1847 }
1848 
1849 static u32 i40e_buildreg_itr(const int type, const u16 itr)
1850 {
1851 	u32 val;
1852 
1853 	val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
1854 	      /* Don't clear PBA because that can cause lost interrupts that
1855 	       * came in while we were cleaning/polling
1856 	       */
1857 	      (type << I40E_PFINT_DYN_CTLN_ITR_INDX_SHIFT) |
1858 	      (itr << I40E_PFINT_DYN_CTLN_INTERVAL_SHIFT);
1859 
1860 	return val;
1861 }
1862 
1863 /* a small macro to shorten up some long lines */
1864 #define INTREG I40E_PFINT_DYN_CTLN
1865 
1866 /**
1867  * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt
1868  * @vsi: the VSI we care about
1869  * @q_vector: q_vector for which itr is being updated and interrupt enabled
1870  *
1871  **/
1872 static inline void i40e_update_enable_itr(struct i40e_vsi *vsi,
1873 					  struct i40e_q_vector *q_vector)
1874 {
1875 	struct i40e_hw *hw = &vsi->back->hw;
1876 	bool rx = false, tx = false;
1877 	u32 rxval, txval;
1878 	int vector;
1879 	int idx = q_vector->v_idx;
1880 
1881 	vector = (q_vector->v_idx + vsi->base_vector);
1882 
1883 	/* avoid dynamic calculation if in countdown mode OR if
1884 	 * all dynamic is disabled
1885 	 */
1886 	rxval = txval = i40e_buildreg_itr(I40E_ITR_NONE, 0);
1887 
1888 	if (q_vector->itr_countdown > 0 ||
1889 	    (!ITR_IS_DYNAMIC(vsi->rx_rings[idx]->rx_itr_setting) &&
1890 	     !ITR_IS_DYNAMIC(vsi->tx_rings[idx]->tx_itr_setting))) {
1891 		goto enable_int;
1892 	}
1893 
1894 	if (ITR_IS_DYNAMIC(vsi->rx_rings[idx]->rx_itr_setting)) {
1895 		rx = i40e_set_new_dynamic_itr(&q_vector->rx);
1896 		rxval = i40e_buildreg_itr(I40E_RX_ITR, q_vector->rx.itr);
1897 	}
1898 
1899 	if (ITR_IS_DYNAMIC(vsi->tx_rings[idx]->tx_itr_setting)) {
1900 		tx = i40e_set_new_dynamic_itr(&q_vector->tx);
1901 		txval = i40e_buildreg_itr(I40E_TX_ITR, q_vector->tx.itr);
1902 	}
1903 
1904 	if (rx || tx) {
1905 		/* get the higher of the two ITR adjustments and
1906 		 * use the same value for both ITR registers
1907 		 * when in adaptive mode (Rx and/or Tx)
1908 		 */
1909 		u16 itr = max(q_vector->tx.itr, q_vector->rx.itr);
1910 
1911 		q_vector->tx.itr = q_vector->rx.itr = itr;
1912 		txval = i40e_buildreg_itr(I40E_TX_ITR, itr);
1913 		tx = true;
1914 		rxval = i40e_buildreg_itr(I40E_RX_ITR, itr);
1915 		rx = true;
1916 	}
1917 
1918 	/* only need to enable the interrupt once, but need
1919 	 * to possibly update both ITR values
1920 	 */
1921 	if (rx) {
1922 		/* set the INTENA_MSK_MASK so that this first write
1923 		 * won't actually enable the interrupt, instead just
1924 		 * updating the ITR (it's bit 31 PF and VF)
1925 		 */
1926 		rxval |= BIT(31);
1927 		/* don't check _DOWN because interrupt isn't being enabled */
1928 		wr32(hw, INTREG(vector - 1), rxval);
1929 	}
1930 
1931 enable_int:
1932 	if (!test_bit(__I40E_DOWN, &vsi->state))
1933 		wr32(hw, INTREG(vector - 1), txval);
1934 
1935 	if (q_vector->itr_countdown)
1936 		q_vector->itr_countdown--;
1937 	else
1938 		q_vector->itr_countdown = ITR_COUNTDOWN_START;
1939 }
1940 
1941 /**
1942  * i40e_napi_poll - NAPI polling Rx/Tx cleanup routine
1943  * @napi: napi struct with our devices info in it
1944  * @budget: amount of work driver is allowed to do this pass, in packets
1945  *
1946  * This function will clean all queues associated with a q_vector.
1947  *
1948  * Returns the amount of work done
1949  **/
1950 int i40e_napi_poll(struct napi_struct *napi, int budget)
1951 {
1952 	struct i40e_q_vector *q_vector =
1953 			       container_of(napi, struct i40e_q_vector, napi);
1954 	struct i40e_vsi *vsi = q_vector->vsi;
1955 	struct i40e_ring *ring;
1956 	bool clean_complete = true;
1957 	bool arm_wb = false;
1958 	int budget_per_ring;
1959 	int work_done = 0;
1960 
1961 	if (test_bit(__I40E_DOWN, &vsi->state)) {
1962 		napi_complete(napi);
1963 		return 0;
1964 	}
1965 
1966 	/* Clear hung_detected bit */
1967 	clear_bit(I40E_Q_VECTOR_HUNG_DETECT, &q_vector->hung_detected);
1968 	/* Since the actual Tx work is minimal, we can give the Tx a larger
1969 	 * budget and be more aggressive about cleaning up the Tx descriptors.
1970 	 */
1971 	i40e_for_each_ring(ring, q_vector->tx) {
1972 		if (!i40e_clean_tx_irq(vsi, ring, budget)) {
1973 			clean_complete = false;
1974 			continue;
1975 		}
1976 		arm_wb |= ring->arm_wb;
1977 		ring->arm_wb = false;
1978 	}
1979 
1980 	/* Handle case where we are called by netpoll with a budget of 0 */
1981 	if (budget <= 0)
1982 		goto tx_only;
1983 
1984 	/* We attempt to distribute budget to each Rx queue fairly, but don't
1985 	 * allow the budget to go below 1 because that would exit polling early.
1986 	 */
1987 	budget_per_ring = max(budget/q_vector->num_ringpairs, 1);
1988 
1989 	i40e_for_each_ring(ring, q_vector->rx) {
1990 		int cleaned = i40e_clean_rx_irq(ring, budget_per_ring);
1991 
1992 		work_done += cleaned;
1993 		/* if we clean as many as budgeted, we must not be done */
1994 		if (cleaned >= budget_per_ring)
1995 			clean_complete = false;
1996 	}
1997 
1998 	/* If work not completed, return budget and polling will return */
1999 	if (!clean_complete) {
2000 tx_only:
2001 		if (arm_wb) {
2002 			q_vector->tx.ring[0].tx_stats.tx_force_wb++;
2003 			i40e_enable_wb_on_itr(vsi, q_vector);
2004 		}
2005 		return budget;
2006 	}
2007 
2008 	if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR)
2009 		q_vector->arm_wb_state = false;
2010 
2011 	/* Work is done so exit the polling mode and re-enable the interrupt */
2012 	napi_complete_done(napi, work_done);
2013 	if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
2014 		i40e_update_enable_itr(vsi, q_vector);
2015 	} else { /* Legacy mode */
2016 		i40e_irq_dynamic_enable_icr0(vsi->back, false);
2017 	}
2018 	return 0;
2019 }
2020 
2021 /**
2022  * i40e_atr - Add a Flow Director ATR filter
2023  * @tx_ring:  ring to add programming descriptor to
2024  * @skb:      send buffer
2025  * @tx_flags: send tx flags
2026  **/
2027 static void i40e_atr(struct i40e_ring *tx_ring, struct sk_buff *skb,
2028 		     u32 tx_flags)
2029 {
2030 	struct i40e_filter_program_desc *fdir_desc;
2031 	struct i40e_pf *pf = tx_ring->vsi->back;
2032 	union {
2033 		unsigned char *network;
2034 		struct iphdr *ipv4;
2035 		struct ipv6hdr *ipv6;
2036 	} hdr;
2037 	struct tcphdr *th;
2038 	unsigned int hlen;
2039 	u32 flex_ptype, dtype_cmd;
2040 	int l4_proto;
2041 	u16 i;
2042 
2043 	/* make sure ATR is enabled */
2044 	if (!(pf->flags & I40E_FLAG_FD_ATR_ENABLED))
2045 		return;
2046 
2047 	if ((pf->auto_disable_flags & I40E_FLAG_FD_ATR_ENABLED))
2048 		return;
2049 
2050 	/* if sampling is disabled do nothing */
2051 	if (!tx_ring->atr_sample_rate)
2052 		return;
2053 
2054 	/* Currently only IPv4/IPv6 with TCP is supported */
2055 	if (!(tx_flags & (I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6)))
2056 		return;
2057 
2058 	/* snag network header to get L4 type and address */
2059 	hdr.network = (tx_flags & I40E_TX_FLAGS_UDP_TUNNEL) ?
2060 		      skb_inner_network_header(skb) : skb_network_header(skb);
2061 
2062 	/* Note: tx_flags gets modified to reflect inner protocols in
2063 	 * tx_enable_csum function if encap is enabled.
2064 	 */
2065 	if (tx_flags & I40E_TX_FLAGS_IPV4) {
2066 		/* access ihl as u8 to avoid unaligned access on ia64 */
2067 		hlen = (hdr.network[0] & 0x0F) << 2;
2068 		l4_proto = hdr.ipv4->protocol;
2069 	} else {
2070 		hlen = hdr.network - skb->data;
2071 		l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
2072 		hlen -= hdr.network - skb->data;
2073 	}
2074 
2075 	if (l4_proto != IPPROTO_TCP)
2076 		return;
2077 
2078 	th = (struct tcphdr *)(hdr.network + hlen);
2079 
2080 	/* Due to lack of space, no more new filters can be programmed */
2081 	if (th->syn && (pf->auto_disable_flags & I40E_FLAG_FD_ATR_ENABLED))
2082 		return;
2083 	if ((pf->flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE) &&
2084 	    (!(pf->auto_disable_flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE))) {
2085 		/* HW ATR eviction will take care of removing filters on FIN
2086 		 * and RST packets.
2087 		 */
2088 		if (th->fin || th->rst)
2089 			return;
2090 	}
2091 
2092 	tx_ring->atr_count++;
2093 
2094 	/* sample on all syn/fin/rst packets or once every atr sample rate */
2095 	if (!th->fin &&
2096 	    !th->syn &&
2097 	    !th->rst &&
2098 	    (tx_ring->atr_count < tx_ring->atr_sample_rate))
2099 		return;
2100 
2101 	tx_ring->atr_count = 0;
2102 
2103 	/* grab the next descriptor */
2104 	i = tx_ring->next_to_use;
2105 	fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
2106 
2107 	i++;
2108 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2109 
2110 	flex_ptype = (tx_ring->queue_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) &
2111 		      I40E_TXD_FLTR_QW0_QINDEX_MASK;
2112 	flex_ptype |= (tx_flags & I40E_TX_FLAGS_IPV4) ?
2113 		      (I40E_FILTER_PCTYPE_NONF_IPV4_TCP <<
2114 		       I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) :
2115 		      (I40E_FILTER_PCTYPE_NONF_IPV6_TCP <<
2116 		       I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
2117 
2118 	flex_ptype |= tx_ring->vsi->id << I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT;
2119 
2120 	dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
2121 
2122 	dtype_cmd |= (th->fin || th->rst) ?
2123 		     (I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
2124 		      I40E_TXD_FLTR_QW1_PCMD_SHIFT) :
2125 		     (I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
2126 		      I40E_TXD_FLTR_QW1_PCMD_SHIFT);
2127 
2128 	dtype_cmd |= I40E_FILTER_PROGRAM_DESC_DEST_DIRECT_PACKET_QINDEX <<
2129 		     I40E_TXD_FLTR_QW1_DEST_SHIFT;
2130 
2131 	dtype_cmd |= I40E_FILTER_PROGRAM_DESC_FD_STATUS_FD_ID <<
2132 		     I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT;
2133 
2134 	dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
2135 	if (!(tx_flags & I40E_TX_FLAGS_UDP_TUNNEL))
2136 		dtype_cmd |=
2137 			((u32)I40E_FD_ATR_STAT_IDX(pf->hw.pf_id) <<
2138 			I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2139 			I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2140 	else
2141 		dtype_cmd |=
2142 			((u32)I40E_FD_ATR_TUNNEL_STAT_IDX(pf->hw.pf_id) <<
2143 			I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2144 			I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2145 
2146 	if ((pf->flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE) &&
2147 	    (!(pf->auto_disable_flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE)))
2148 		dtype_cmd |= I40E_TXD_FLTR_QW1_ATR_MASK;
2149 
2150 	fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
2151 	fdir_desc->rsvd = cpu_to_le32(0);
2152 	fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
2153 	fdir_desc->fd_id = cpu_to_le32(0);
2154 }
2155 
2156 /**
2157  * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
2158  * @skb:     send buffer
2159  * @tx_ring: ring to send buffer on
2160  * @flags:   the tx flags to be set
2161  *
2162  * Checks the skb and set up correspondingly several generic transmit flags
2163  * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
2164  *
2165  * Returns error code indicate the frame should be dropped upon error and the
2166  * otherwise  returns 0 to indicate the flags has been set properly.
2167  **/
2168 #ifdef I40E_FCOE
2169 inline int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
2170 				      struct i40e_ring *tx_ring,
2171 				      u32 *flags)
2172 #else
2173 static inline int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
2174 					     struct i40e_ring *tx_ring,
2175 					     u32 *flags)
2176 #endif
2177 {
2178 	__be16 protocol = skb->protocol;
2179 	u32  tx_flags = 0;
2180 
2181 	if (protocol == htons(ETH_P_8021Q) &&
2182 	    !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
2183 		/* When HW VLAN acceleration is turned off by the user the
2184 		 * stack sets the protocol to 8021q so that the driver
2185 		 * can take any steps required to support the SW only
2186 		 * VLAN handling.  In our case the driver doesn't need
2187 		 * to take any further steps so just set the protocol
2188 		 * to the encapsulated ethertype.
2189 		 */
2190 		skb->protocol = vlan_get_protocol(skb);
2191 		goto out;
2192 	}
2193 
2194 	/* if we have a HW VLAN tag being added, default to the HW one */
2195 	if (skb_vlan_tag_present(skb)) {
2196 		tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
2197 		tx_flags |= I40E_TX_FLAGS_HW_VLAN;
2198 	/* else if it is a SW VLAN, check the next protocol and store the tag */
2199 	} else if (protocol == htons(ETH_P_8021Q)) {
2200 		struct vlan_hdr *vhdr, _vhdr;
2201 
2202 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
2203 		if (!vhdr)
2204 			return -EINVAL;
2205 
2206 		protocol = vhdr->h_vlan_encapsulated_proto;
2207 		tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
2208 		tx_flags |= I40E_TX_FLAGS_SW_VLAN;
2209 	}
2210 
2211 	if (!(tx_ring->vsi->back->flags & I40E_FLAG_DCB_ENABLED))
2212 		goto out;
2213 
2214 	/* Insert 802.1p priority into VLAN header */
2215 	if ((tx_flags & (I40E_TX_FLAGS_HW_VLAN | I40E_TX_FLAGS_SW_VLAN)) ||
2216 	    (skb->priority != TC_PRIO_CONTROL)) {
2217 		tx_flags &= ~I40E_TX_FLAGS_VLAN_PRIO_MASK;
2218 		tx_flags |= (skb->priority & 0x7) <<
2219 				I40E_TX_FLAGS_VLAN_PRIO_SHIFT;
2220 		if (tx_flags & I40E_TX_FLAGS_SW_VLAN) {
2221 			struct vlan_ethhdr *vhdr;
2222 			int rc;
2223 
2224 			rc = skb_cow_head(skb, 0);
2225 			if (rc < 0)
2226 				return rc;
2227 			vhdr = (struct vlan_ethhdr *)skb->data;
2228 			vhdr->h_vlan_TCI = htons(tx_flags >>
2229 						 I40E_TX_FLAGS_VLAN_SHIFT);
2230 		} else {
2231 			tx_flags |= I40E_TX_FLAGS_HW_VLAN;
2232 		}
2233 	}
2234 
2235 out:
2236 	*flags = tx_flags;
2237 	return 0;
2238 }
2239 
2240 /**
2241  * i40e_tso - set up the tso context descriptor
2242  * @skb:      ptr to the skb we're sending
2243  * @hdr_len:  ptr to the size of the packet header
2244  * @cd_type_cmd_tso_mss: Quad Word 1
2245  *
2246  * Returns 0 if no TSO can happen, 1 if tso is going, or error
2247  **/
2248 static int i40e_tso(struct sk_buff *skb, u8 *hdr_len, u64 *cd_type_cmd_tso_mss)
2249 {
2250 	u64 cd_cmd, cd_tso_len, cd_mss;
2251 	union {
2252 		struct iphdr *v4;
2253 		struct ipv6hdr *v6;
2254 		unsigned char *hdr;
2255 	} ip;
2256 	union {
2257 		struct tcphdr *tcp;
2258 		struct udphdr *udp;
2259 		unsigned char *hdr;
2260 	} l4;
2261 	u32 paylen, l4_offset;
2262 	int err;
2263 
2264 	if (skb->ip_summed != CHECKSUM_PARTIAL)
2265 		return 0;
2266 
2267 	if (!skb_is_gso(skb))
2268 		return 0;
2269 
2270 	err = skb_cow_head(skb, 0);
2271 	if (err < 0)
2272 		return err;
2273 
2274 	ip.hdr = skb_network_header(skb);
2275 	l4.hdr = skb_transport_header(skb);
2276 
2277 	/* initialize outer IP header fields */
2278 	if (ip.v4->version == 4) {
2279 		ip.v4->tot_len = 0;
2280 		ip.v4->check = 0;
2281 	} else {
2282 		ip.v6->payload_len = 0;
2283 	}
2284 
2285 	if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
2286 					 SKB_GSO_GRE_CSUM |
2287 					 SKB_GSO_IPXIP4 |
2288 					 SKB_GSO_IPXIP6 |
2289 					 SKB_GSO_UDP_TUNNEL |
2290 					 SKB_GSO_UDP_TUNNEL_CSUM)) {
2291 		if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
2292 		    (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)) {
2293 			l4.udp->len = 0;
2294 
2295 			/* determine offset of outer transport header */
2296 			l4_offset = l4.hdr - skb->data;
2297 
2298 			/* remove payload length from outer checksum */
2299 			paylen = skb->len - l4_offset;
2300 			csum_replace_by_diff(&l4.udp->check, htonl(paylen));
2301 		}
2302 
2303 		/* reset pointers to inner headers */
2304 		ip.hdr = skb_inner_network_header(skb);
2305 		l4.hdr = skb_inner_transport_header(skb);
2306 
2307 		/* initialize inner IP header fields */
2308 		if (ip.v4->version == 4) {
2309 			ip.v4->tot_len = 0;
2310 			ip.v4->check = 0;
2311 		} else {
2312 			ip.v6->payload_len = 0;
2313 		}
2314 	}
2315 
2316 	/* determine offset of inner transport header */
2317 	l4_offset = l4.hdr - skb->data;
2318 
2319 	/* remove payload length from inner checksum */
2320 	paylen = skb->len - l4_offset;
2321 	csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
2322 
2323 	/* compute length of segmentation header */
2324 	*hdr_len = (l4.tcp->doff * 4) + l4_offset;
2325 
2326 	/* find the field values */
2327 	cd_cmd = I40E_TX_CTX_DESC_TSO;
2328 	cd_tso_len = skb->len - *hdr_len;
2329 	cd_mss = skb_shinfo(skb)->gso_size;
2330 	*cd_type_cmd_tso_mss |= (cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
2331 				(cd_tso_len << I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
2332 				(cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
2333 	return 1;
2334 }
2335 
2336 /**
2337  * i40e_tsyn - set up the tsyn context descriptor
2338  * @tx_ring:  ptr to the ring to send
2339  * @skb:      ptr to the skb we're sending
2340  * @tx_flags: the collected send information
2341  * @cd_type_cmd_tso_mss: Quad Word 1
2342  *
2343  * Returns 0 if no Tx timestamp can happen and 1 if the timestamp will happen
2344  **/
2345 static int i40e_tsyn(struct i40e_ring *tx_ring, struct sk_buff *skb,
2346 		     u32 tx_flags, u64 *cd_type_cmd_tso_mss)
2347 {
2348 	struct i40e_pf *pf;
2349 
2350 	if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)))
2351 		return 0;
2352 
2353 	/* Tx timestamps cannot be sampled when doing TSO */
2354 	if (tx_flags & I40E_TX_FLAGS_TSO)
2355 		return 0;
2356 
2357 	/* only timestamp the outbound packet if the user has requested it and
2358 	 * we are not already transmitting a packet to be timestamped
2359 	 */
2360 	pf = i40e_netdev_to_pf(tx_ring->netdev);
2361 	if (!(pf->flags & I40E_FLAG_PTP))
2362 		return 0;
2363 
2364 	if (pf->ptp_tx &&
2365 	    !test_and_set_bit_lock(__I40E_PTP_TX_IN_PROGRESS, &pf->state)) {
2366 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
2367 		pf->ptp_tx_skb = skb_get(skb);
2368 	} else {
2369 		return 0;
2370 	}
2371 
2372 	*cd_type_cmd_tso_mss |= (u64)I40E_TX_CTX_DESC_TSYN <<
2373 				I40E_TXD_CTX_QW1_CMD_SHIFT;
2374 
2375 	return 1;
2376 }
2377 
2378 /**
2379  * i40e_tx_enable_csum - Enable Tx checksum offloads
2380  * @skb: send buffer
2381  * @tx_flags: pointer to Tx flags currently set
2382  * @td_cmd: Tx descriptor command bits to set
2383  * @td_offset: Tx descriptor header offsets to set
2384  * @tx_ring: Tx descriptor ring
2385  * @cd_tunneling: ptr to context desc bits
2386  **/
2387 static int i40e_tx_enable_csum(struct sk_buff *skb, u32 *tx_flags,
2388 			       u32 *td_cmd, u32 *td_offset,
2389 			       struct i40e_ring *tx_ring,
2390 			       u32 *cd_tunneling)
2391 {
2392 	union {
2393 		struct iphdr *v4;
2394 		struct ipv6hdr *v6;
2395 		unsigned char *hdr;
2396 	} ip;
2397 	union {
2398 		struct tcphdr *tcp;
2399 		struct udphdr *udp;
2400 		unsigned char *hdr;
2401 	} l4;
2402 	unsigned char *exthdr;
2403 	u32 offset, cmd = 0;
2404 	__be16 frag_off;
2405 	u8 l4_proto = 0;
2406 
2407 	if (skb->ip_summed != CHECKSUM_PARTIAL)
2408 		return 0;
2409 
2410 	ip.hdr = skb_network_header(skb);
2411 	l4.hdr = skb_transport_header(skb);
2412 
2413 	/* compute outer L2 header size */
2414 	offset = ((ip.hdr - skb->data) / 2) << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
2415 
2416 	if (skb->encapsulation) {
2417 		u32 tunnel = 0;
2418 		/* define outer network header type */
2419 		if (*tx_flags & I40E_TX_FLAGS_IPV4) {
2420 			tunnel |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
2421 				  I40E_TX_CTX_EXT_IP_IPV4 :
2422 				  I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
2423 
2424 			l4_proto = ip.v4->protocol;
2425 		} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
2426 			tunnel |= I40E_TX_CTX_EXT_IP_IPV6;
2427 
2428 			exthdr = ip.hdr + sizeof(*ip.v6);
2429 			l4_proto = ip.v6->nexthdr;
2430 			if (l4.hdr != exthdr)
2431 				ipv6_skip_exthdr(skb, exthdr - skb->data,
2432 						 &l4_proto, &frag_off);
2433 		}
2434 
2435 		/* define outer transport */
2436 		switch (l4_proto) {
2437 		case IPPROTO_UDP:
2438 			tunnel |= I40E_TXD_CTX_UDP_TUNNELING;
2439 			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
2440 			break;
2441 		case IPPROTO_GRE:
2442 			tunnel |= I40E_TXD_CTX_GRE_TUNNELING;
2443 			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
2444 			break;
2445 		case IPPROTO_IPIP:
2446 		case IPPROTO_IPV6:
2447 			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
2448 			l4.hdr = skb_inner_network_header(skb);
2449 			break;
2450 		default:
2451 			if (*tx_flags & I40E_TX_FLAGS_TSO)
2452 				return -1;
2453 
2454 			skb_checksum_help(skb);
2455 			return 0;
2456 		}
2457 
2458 		/* compute outer L3 header size */
2459 		tunnel |= ((l4.hdr - ip.hdr) / 4) <<
2460 			  I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT;
2461 
2462 		/* switch IP header pointer from outer to inner header */
2463 		ip.hdr = skb_inner_network_header(skb);
2464 
2465 		/* compute tunnel header size */
2466 		tunnel |= ((ip.hdr - l4.hdr) / 2) <<
2467 			  I40E_TXD_CTX_QW0_NATLEN_SHIFT;
2468 
2469 		/* indicate if we need to offload outer UDP header */
2470 		if ((*tx_flags & I40E_TX_FLAGS_TSO) &&
2471 		    !(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
2472 		    (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM))
2473 			tunnel |= I40E_TXD_CTX_QW0_L4T_CS_MASK;
2474 
2475 		/* record tunnel offload values */
2476 		*cd_tunneling |= tunnel;
2477 
2478 		/* switch L4 header pointer from outer to inner */
2479 		l4.hdr = skb_inner_transport_header(skb);
2480 		l4_proto = 0;
2481 
2482 		/* reset type as we transition from outer to inner headers */
2483 		*tx_flags &= ~(I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6);
2484 		if (ip.v4->version == 4)
2485 			*tx_flags |= I40E_TX_FLAGS_IPV4;
2486 		if (ip.v6->version == 6)
2487 			*tx_flags |= I40E_TX_FLAGS_IPV6;
2488 	}
2489 
2490 	/* Enable IP checksum offloads */
2491 	if (*tx_flags & I40E_TX_FLAGS_IPV4) {
2492 		l4_proto = ip.v4->protocol;
2493 		/* the stack computes the IP header already, the only time we
2494 		 * need the hardware to recompute it is in the case of TSO.
2495 		 */
2496 		cmd |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
2497 		       I40E_TX_DESC_CMD_IIPT_IPV4_CSUM :
2498 		       I40E_TX_DESC_CMD_IIPT_IPV4;
2499 	} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
2500 		cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
2501 
2502 		exthdr = ip.hdr + sizeof(*ip.v6);
2503 		l4_proto = ip.v6->nexthdr;
2504 		if (l4.hdr != exthdr)
2505 			ipv6_skip_exthdr(skb, exthdr - skb->data,
2506 					 &l4_proto, &frag_off);
2507 	}
2508 
2509 	/* compute inner L3 header size */
2510 	offset |= ((l4.hdr - ip.hdr) / 4) << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
2511 
2512 	/* Enable L4 checksum offloads */
2513 	switch (l4_proto) {
2514 	case IPPROTO_TCP:
2515 		/* enable checksum offloads */
2516 		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
2517 		offset |= l4.tcp->doff << I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
2518 		break;
2519 	case IPPROTO_SCTP:
2520 		/* enable SCTP checksum offload */
2521 		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
2522 		offset |= (sizeof(struct sctphdr) >> 2) <<
2523 			  I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
2524 		break;
2525 	case IPPROTO_UDP:
2526 		/* enable UDP checksum offload */
2527 		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
2528 		offset |= (sizeof(struct udphdr) >> 2) <<
2529 			  I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
2530 		break;
2531 	default:
2532 		if (*tx_flags & I40E_TX_FLAGS_TSO)
2533 			return -1;
2534 		skb_checksum_help(skb);
2535 		return 0;
2536 	}
2537 
2538 	*td_cmd |= cmd;
2539 	*td_offset |= offset;
2540 
2541 	return 1;
2542 }
2543 
2544 /**
2545  * i40e_create_tx_ctx Build the Tx context descriptor
2546  * @tx_ring:  ring to create the descriptor on
2547  * @cd_type_cmd_tso_mss: Quad Word 1
2548  * @cd_tunneling: Quad Word 0 - bits 0-31
2549  * @cd_l2tag2: Quad Word 0 - bits 32-63
2550  **/
2551 static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
2552 			       const u64 cd_type_cmd_tso_mss,
2553 			       const u32 cd_tunneling, const u32 cd_l2tag2)
2554 {
2555 	struct i40e_tx_context_desc *context_desc;
2556 	int i = tx_ring->next_to_use;
2557 
2558 	if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
2559 	    !cd_tunneling && !cd_l2tag2)
2560 		return;
2561 
2562 	/* grab the next descriptor */
2563 	context_desc = I40E_TX_CTXTDESC(tx_ring, i);
2564 
2565 	i++;
2566 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2567 
2568 	/* cpu_to_le32 and assign to struct fields */
2569 	context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
2570 	context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
2571 	context_desc->rsvd = cpu_to_le16(0);
2572 	context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
2573 }
2574 
2575 /**
2576  * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions
2577  * @tx_ring: the ring to be checked
2578  * @size:    the size buffer we want to assure is available
2579  *
2580  * Returns -EBUSY if a stop is needed, else 0
2581  **/
2582 int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
2583 {
2584 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
2585 	/* Memory barrier before checking head and tail */
2586 	smp_mb();
2587 
2588 	/* Check again in a case another CPU has just made room available. */
2589 	if (likely(I40E_DESC_UNUSED(tx_ring) < size))
2590 		return -EBUSY;
2591 
2592 	/* A reprieve! - use start_queue because it doesn't call schedule */
2593 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
2594 	++tx_ring->tx_stats.restart_queue;
2595 	return 0;
2596 }
2597 
2598 /**
2599  * __i40e_chk_linearize - Check if there are more than 8 buffers per packet
2600  * @skb:      send buffer
2601  *
2602  * Note: Our HW can't DMA more than 8 buffers to build a packet on the wire
2603  * and so we need to figure out the cases where we need to linearize the skb.
2604  *
2605  * For TSO we need to count the TSO header and segment payload separately.
2606  * As such we need to check cases where we have 7 fragments or more as we
2607  * can potentially require 9 DMA transactions, 1 for the TSO header, 1 for
2608  * the segment payload in the first descriptor, and another 7 for the
2609  * fragments.
2610  **/
2611 bool __i40e_chk_linearize(struct sk_buff *skb)
2612 {
2613 	const struct skb_frag_struct *frag, *stale;
2614 	int nr_frags, sum;
2615 
2616 	/* no need to check if number of frags is less than 7 */
2617 	nr_frags = skb_shinfo(skb)->nr_frags;
2618 	if (nr_frags < (I40E_MAX_BUFFER_TXD - 1))
2619 		return false;
2620 
2621 	/* We need to walk through the list and validate that each group
2622 	 * of 6 fragments totals at least gso_size.  However we don't need
2623 	 * to perform such validation on the last 6 since the last 6 cannot
2624 	 * inherit any data from a descriptor after them.
2625 	 */
2626 	nr_frags -= I40E_MAX_BUFFER_TXD - 2;
2627 	frag = &skb_shinfo(skb)->frags[0];
2628 
2629 	/* Initialize size to the negative value of gso_size minus 1.  We
2630 	 * use this as the worst case scenerio in which the frag ahead
2631 	 * of us only provides one byte which is why we are limited to 6
2632 	 * descriptors for a single transmit as the header and previous
2633 	 * fragment are already consuming 2 descriptors.
2634 	 */
2635 	sum = 1 - skb_shinfo(skb)->gso_size;
2636 
2637 	/* Add size of frags 0 through 4 to create our initial sum */
2638 	sum += skb_frag_size(frag++);
2639 	sum += skb_frag_size(frag++);
2640 	sum += skb_frag_size(frag++);
2641 	sum += skb_frag_size(frag++);
2642 	sum += skb_frag_size(frag++);
2643 
2644 	/* Walk through fragments adding latest fragment, testing it, and
2645 	 * then removing stale fragments from the sum.
2646 	 */
2647 	stale = &skb_shinfo(skb)->frags[0];
2648 	for (;;) {
2649 		sum += skb_frag_size(frag++);
2650 
2651 		/* if sum is negative we failed to make sufficient progress */
2652 		if (sum < 0)
2653 			return true;
2654 
2655 		/* use pre-decrement to avoid processing last fragment */
2656 		if (!--nr_frags)
2657 			break;
2658 
2659 		sum -= skb_frag_size(stale++);
2660 	}
2661 
2662 	return false;
2663 }
2664 
2665 /**
2666  * i40e_tx_map - Build the Tx descriptor
2667  * @tx_ring:  ring to send buffer on
2668  * @skb:      send buffer
2669  * @first:    first buffer info buffer to use
2670  * @tx_flags: collected send information
2671  * @hdr_len:  size of the packet header
2672  * @td_cmd:   the command field in the descriptor
2673  * @td_offset: offset for checksum or crc
2674  **/
2675 #ifdef I40E_FCOE
2676 inline void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
2677 			struct i40e_tx_buffer *first, u32 tx_flags,
2678 			const u8 hdr_len, u32 td_cmd, u32 td_offset)
2679 #else
2680 static inline void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
2681 			       struct i40e_tx_buffer *first, u32 tx_flags,
2682 			       const u8 hdr_len, u32 td_cmd, u32 td_offset)
2683 #endif
2684 {
2685 	unsigned int data_len = skb->data_len;
2686 	unsigned int size = skb_headlen(skb);
2687 	struct skb_frag_struct *frag;
2688 	struct i40e_tx_buffer *tx_bi;
2689 	struct i40e_tx_desc *tx_desc;
2690 	u16 i = tx_ring->next_to_use;
2691 	u32 td_tag = 0;
2692 	dma_addr_t dma;
2693 	u16 gso_segs;
2694 	u16 desc_count = 0;
2695 	bool tail_bump = true;
2696 	bool do_rs = false;
2697 
2698 	if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
2699 		td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
2700 		td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
2701 			 I40E_TX_FLAGS_VLAN_SHIFT;
2702 	}
2703 
2704 	if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO))
2705 		gso_segs = skb_shinfo(skb)->gso_segs;
2706 	else
2707 		gso_segs = 1;
2708 
2709 	/* multiply data chunks by size of headers */
2710 	first->bytecount = skb->len - hdr_len + (gso_segs * hdr_len);
2711 	first->gso_segs = gso_segs;
2712 	first->skb = skb;
2713 	first->tx_flags = tx_flags;
2714 
2715 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
2716 
2717 	tx_desc = I40E_TX_DESC(tx_ring, i);
2718 	tx_bi = first;
2719 
2720 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
2721 		unsigned int max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
2722 
2723 		if (dma_mapping_error(tx_ring->dev, dma))
2724 			goto dma_error;
2725 
2726 		/* record length, and DMA address */
2727 		dma_unmap_len_set(tx_bi, len, size);
2728 		dma_unmap_addr_set(tx_bi, dma, dma);
2729 
2730 		/* align size to end of page */
2731 		max_data += -dma & (I40E_MAX_READ_REQ_SIZE - 1);
2732 		tx_desc->buffer_addr = cpu_to_le64(dma);
2733 
2734 		while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
2735 			tx_desc->cmd_type_offset_bsz =
2736 				build_ctob(td_cmd, td_offset,
2737 					   max_data, td_tag);
2738 
2739 			tx_desc++;
2740 			i++;
2741 			desc_count++;
2742 
2743 			if (i == tx_ring->count) {
2744 				tx_desc = I40E_TX_DESC(tx_ring, 0);
2745 				i = 0;
2746 			}
2747 
2748 			dma += max_data;
2749 			size -= max_data;
2750 
2751 			max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
2752 			tx_desc->buffer_addr = cpu_to_le64(dma);
2753 		}
2754 
2755 		if (likely(!data_len))
2756 			break;
2757 
2758 		tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
2759 							  size, td_tag);
2760 
2761 		tx_desc++;
2762 		i++;
2763 		desc_count++;
2764 
2765 		if (i == tx_ring->count) {
2766 			tx_desc = I40E_TX_DESC(tx_ring, 0);
2767 			i = 0;
2768 		}
2769 
2770 		size = skb_frag_size(frag);
2771 		data_len -= size;
2772 
2773 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
2774 				       DMA_TO_DEVICE);
2775 
2776 		tx_bi = &tx_ring->tx_bi[i];
2777 	}
2778 
2779 	/* set next_to_watch value indicating a packet is present */
2780 	first->next_to_watch = tx_desc;
2781 
2782 	i++;
2783 	if (i == tx_ring->count)
2784 		i = 0;
2785 
2786 	tx_ring->next_to_use = i;
2787 
2788 	netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev,
2789 						 tx_ring->queue_index),
2790 						 first->bytecount);
2791 	i40e_maybe_stop_tx(tx_ring, DESC_NEEDED);
2792 
2793 	/* Algorithm to optimize tail and RS bit setting:
2794 	 * if xmit_more is supported
2795 	 *	if xmit_more is true
2796 	 *		do not update tail and do not mark RS bit.
2797 	 *	if xmit_more is false and last xmit_more was false
2798 	 *		if every packet spanned less than 4 desc
2799 	 *			then set RS bit on 4th packet and update tail
2800 	 *			on every packet
2801 	 *		else
2802 	 *			update tail and set RS bit on every packet.
2803 	 *	if xmit_more is false and last_xmit_more was true
2804 	 *		update tail and set RS bit.
2805 	 *
2806 	 * Optimization: wmb to be issued only in case of tail update.
2807 	 * Also optimize the Descriptor WB path for RS bit with the same
2808 	 * algorithm.
2809 	 *
2810 	 * Note: If there are less than 4 packets
2811 	 * pending and interrupts were disabled the service task will
2812 	 * trigger a force WB.
2813 	 */
2814 	if (skb->xmit_more  &&
2815 	    !netif_xmit_stopped(netdev_get_tx_queue(tx_ring->netdev,
2816 						    tx_ring->queue_index))) {
2817 		tx_ring->flags |= I40E_TXR_FLAGS_LAST_XMIT_MORE_SET;
2818 		tail_bump = false;
2819 	} else if (!skb->xmit_more &&
2820 		   !netif_xmit_stopped(netdev_get_tx_queue(tx_ring->netdev,
2821 						       tx_ring->queue_index)) &&
2822 		   (!(tx_ring->flags & I40E_TXR_FLAGS_LAST_XMIT_MORE_SET)) &&
2823 		   (tx_ring->packet_stride < WB_STRIDE) &&
2824 		   (desc_count < WB_STRIDE)) {
2825 		tx_ring->packet_stride++;
2826 	} else {
2827 		tx_ring->packet_stride = 0;
2828 		tx_ring->flags &= ~I40E_TXR_FLAGS_LAST_XMIT_MORE_SET;
2829 		do_rs = true;
2830 	}
2831 	if (do_rs)
2832 		tx_ring->packet_stride = 0;
2833 
2834 	tx_desc->cmd_type_offset_bsz =
2835 			build_ctob(td_cmd, td_offset, size, td_tag) |
2836 			cpu_to_le64((u64)(do_rs ? I40E_TXD_CMD :
2837 						  I40E_TX_DESC_CMD_EOP) <<
2838 						  I40E_TXD_QW1_CMD_SHIFT);
2839 
2840 	/* notify HW of packet */
2841 	if (!tail_bump)
2842 		prefetchw(tx_desc + 1);
2843 
2844 	if (tail_bump) {
2845 		/* Force memory writes to complete before letting h/w
2846 		 * know there are new descriptors to fetch.  (Only
2847 		 * applicable for weak-ordered memory model archs,
2848 		 * such as IA-64).
2849 		 */
2850 		wmb();
2851 		writel(i, tx_ring->tail);
2852 	}
2853 
2854 	return;
2855 
2856 dma_error:
2857 	dev_info(tx_ring->dev, "TX DMA map failed\n");
2858 
2859 	/* clear dma mappings for failed tx_bi map */
2860 	for (;;) {
2861 		tx_bi = &tx_ring->tx_bi[i];
2862 		i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
2863 		if (tx_bi == first)
2864 			break;
2865 		if (i == 0)
2866 			i = tx_ring->count;
2867 		i--;
2868 	}
2869 
2870 	tx_ring->next_to_use = i;
2871 }
2872 
2873 /**
2874  * i40e_xmit_frame_ring - Sends buffer on Tx ring
2875  * @skb:     send buffer
2876  * @tx_ring: ring to send buffer on
2877  *
2878  * Returns NETDEV_TX_OK if sent, else an error code
2879  **/
2880 static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
2881 					struct i40e_ring *tx_ring)
2882 {
2883 	u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
2884 	u32 cd_tunneling = 0, cd_l2tag2 = 0;
2885 	struct i40e_tx_buffer *first;
2886 	u32 td_offset = 0;
2887 	u32 tx_flags = 0;
2888 	__be16 protocol;
2889 	u32 td_cmd = 0;
2890 	u8 hdr_len = 0;
2891 	int tso, count;
2892 	int tsyn;
2893 
2894 	/* prefetch the data, we'll need it later */
2895 	prefetch(skb->data);
2896 
2897 	count = i40e_xmit_descriptor_count(skb);
2898 	if (i40e_chk_linearize(skb, count)) {
2899 		if (__skb_linearize(skb))
2900 			goto out_drop;
2901 		count = i40e_txd_use_count(skb->len);
2902 		tx_ring->tx_stats.tx_linearize++;
2903 	}
2904 
2905 	/* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
2906 	 *       + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
2907 	 *       + 4 desc gap to avoid the cache line where head is,
2908 	 *       + 1 desc for context descriptor,
2909 	 * otherwise try next time
2910 	 */
2911 	if (i40e_maybe_stop_tx(tx_ring, count + 4 + 1)) {
2912 		tx_ring->tx_stats.tx_busy++;
2913 		return NETDEV_TX_BUSY;
2914 	}
2915 
2916 	/* prepare the xmit flags */
2917 	if (i40e_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
2918 		goto out_drop;
2919 
2920 	/* obtain protocol of skb */
2921 	protocol = vlan_get_protocol(skb);
2922 
2923 	/* record the location of the first descriptor for this packet */
2924 	first = &tx_ring->tx_bi[tx_ring->next_to_use];
2925 
2926 	/* setup IPv4/IPv6 offloads */
2927 	if (protocol == htons(ETH_P_IP))
2928 		tx_flags |= I40E_TX_FLAGS_IPV4;
2929 	else if (protocol == htons(ETH_P_IPV6))
2930 		tx_flags |= I40E_TX_FLAGS_IPV6;
2931 
2932 	tso = i40e_tso(skb, &hdr_len, &cd_type_cmd_tso_mss);
2933 
2934 	if (tso < 0)
2935 		goto out_drop;
2936 	else if (tso)
2937 		tx_flags |= I40E_TX_FLAGS_TSO;
2938 
2939 	/* Always offload the checksum, since it's in the data descriptor */
2940 	tso = i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset,
2941 				  tx_ring, &cd_tunneling);
2942 	if (tso < 0)
2943 		goto out_drop;
2944 
2945 	tsyn = i40e_tsyn(tx_ring, skb, tx_flags, &cd_type_cmd_tso_mss);
2946 
2947 	if (tsyn)
2948 		tx_flags |= I40E_TX_FLAGS_TSYN;
2949 
2950 	skb_tx_timestamp(skb);
2951 
2952 	/* always enable CRC insertion offload */
2953 	td_cmd |= I40E_TX_DESC_CMD_ICRC;
2954 
2955 	i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
2956 			   cd_tunneling, cd_l2tag2);
2957 
2958 	/* Add Flow Director ATR if it's enabled.
2959 	 *
2960 	 * NOTE: this must always be directly before the data descriptor.
2961 	 */
2962 	i40e_atr(tx_ring, skb, tx_flags);
2963 
2964 	i40e_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
2965 		    td_cmd, td_offset);
2966 
2967 	return NETDEV_TX_OK;
2968 
2969 out_drop:
2970 	dev_kfree_skb_any(skb);
2971 	return NETDEV_TX_OK;
2972 }
2973 
2974 /**
2975  * i40e_lan_xmit_frame - Selects the correct VSI and Tx queue to send buffer
2976  * @skb:    send buffer
2977  * @netdev: network interface device structure
2978  *
2979  * Returns NETDEV_TX_OK if sent, else an error code
2980  **/
2981 netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2982 {
2983 	struct i40e_netdev_priv *np = netdev_priv(netdev);
2984 	struct i40e_vsi *vsi = np->vsi;
2985 	struct i40e_ring *tx_ring = vsi->tx_rings[skb->queue_mapping];
2986 
2987 	/* hardware can't handle really short frames, hardware padding works
2988 	 * beyond this point
2989 	 */
2990 	if (skb_put_padto(skb, I40E_MIN_TX_LEN))
2991 		return NETDEV_TX_OK;
2992 
2993 	return i40e_xmit_frame_ring(skb, tx_ring);
2994 }
2995