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 		/* check to see if there are < 4 descriptors
744 		 * waiting to be written back, then kick the hardware to force
745 		 * them to be written back in case we stay in NAPI.
746 		 * In this mode on X722 we do not enable Interrupt.
747 		 */
748 		unsigned int j = i40e_get_tx_pending(tx_ring, false);
749 
750 		if (budget &&
751 		    ((j / (WB_STRIDE + 1)) == 0) && (j != 0) &&
752 		    !test_bit(__I40E_DOWN, &vsi->state) &&
753 		    (I40E_DESC_UNUSED(tx_ring) != tx_ring->count))
754 			tx_ring->arm_wb = true;
755 	}
756 
757 	netdev_tx_completed_queue(netdev_get_tx_queue(tx_ring->netdev,
758 						      tx_ring->queue_index),
759 				  total_packets, total_bytes);
760 
761 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
762 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
763 		     (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
764 		/* Make sure that anybody stopping the queue after this
765 		 * sees the new next_to_clean.
766 		 */
767 		smp_mb();
768 		if (__netif_subqueue_stopped(tx_ring->netdev,
769 					     tx_ring->queue_index) &&
770 		   !test_bit(__I40E_DOWN, &vsi->state)) {
771 			netif_wake_subqueue(tx_ring->netdev,
772 					    tx_ring->queue_index);
773 			++tx_ring->tx_stats.restart_queue;
774 		}
775 	}
776 
777 	return !!budget;
778 }
779 
780 /**
781  * i40e_enable_wb_on_itr - Arm hardware to do a wb, interrupts are not enabled
782  * @vsi: the VSI we care about
783  * @q_vector: the vector on which to enable writeback
784  *
785  **/
786 static void i40e_enable_wb_on_itr(struct i40e_vsi *vsi,
787 				  struct i40e_q_vector *q_vector)
788 {
789 	u16 flags = q_vector->tx.ring[0].flags;
790 	u32 val;
791 
792 	if (!(flags & I40E_TXR_FLAGS_WB_ON_ITR))
793 		return;
794 
795 	if (q_vector->arm_wb_state)
796 		return;
797 
798 	if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
799 		val = I40E_PFINT_DYN_CTLN_WB_ON_ITR_MASK |
800 		      I40E_PFINT_DYN_CTLN_ITR_INDX_MASK; /* set noitr */
801 
802 		wr32(&vsi->back->hw,
803 		     I40E_PFINT_DYN_CTLN(q_vector->v_idx + vsi->base_vector - 1),
804 		     val);
805 	} else {
806 		val = I40E_PFINT_DYN_CTL0_WB_ON_ITR_MASK |
807 		      I40E_PFINT_DYN_CTL0_ITR_INDX_MASK; /* set noitr */
808 
809 		wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
810 	}
811 	q_vector->arm_wb_state = true;
812 }
813 
814 /**
815  * i40e_force_wb - Issue SW Interrupt so HW does a wb
816  * @vsi: the VSI we care about
817  * @q_vector: the vector  on which to force writeback
818  *
819  **/
820 void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector)
821 {
822 	if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
823 		u32 val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
824 			  I40E_PFINT_DYN_CTLN_ITR_INDX_MASK | /* set noitr */
825 			  I40E_PFINT_DYN_CTLN_SWINT_TRIG_MASK |
826 			  I40E_PFINT_DYN_CTLN_SW_ITR_INDX_ENA_MASK;
827 			  /* allow 00 to be written to the index */
828 
829 		wr32(&vsi->back->hw,
830 		     I40E_PFINT_DYN_CTLN(q_vector->v_idx +
831 					 vsi->base_vector - 1), val);
832 	} else {
833 		u32 val = I40E_PFINT_DYN_CTL0_INTENA_MASK |
834 			  I40E_PFINT_DYN_CTL0_ITR_INDX_MASK | /* set noitr */
835 			  I40E_PFINT_DYN_CTL0_SWINT_TRIG_MASK |
836 			  I40E_PFINT_DYN_CTL0_SW_ITR_INDX_ENA_MASK;
837 			/* allow 00 to be written to the index */
838 
839 		wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
840 	}
841 }
842 
843 /**
844  * i40e_set_new_dynamic_itr - Find new ITR level
845  * @rc: structure containing ring performance data
846  *
847  * Returns true if ITR changed, false if not
848  *
849  * Stores a new ITR value based on packets and byte counts during
850  * the last interrupt.  The advantage of per interrupt computation
851  * is faster updates and more accurate ITR for the current traffic
852  * pattern.  Constants in this function were computed based on
853  * theoretical maximum wire speed and thresholds were set based on
854  * testing data as well as attempting to minimize response time
855  * while increasing bulk throughput.
856  **/
857 static bool i40e_set_new_dynamic_itr(struct i40e_ring_container *rc)
858 {
859 	enum i40e_latency_range new_latency_range = rc->latency_range;
860 	struct i40e_q_vector *qv = rc->ring->q_vector;
861 	u32 new_itr = rc->itr;
862 	int bytes_per_int;
863 	int usecs;
864 
865 	if (rc->total_packets == 0 || !rc->itr)
866 		return false;
867 
868 	/* simple throttlerate management
869 	 *   0-10MB/s   lowest (50000 ints/s)
870 	 *  10-20MB/s   low    (20000 ints/s)
871 	 *  20-1249MB/s bulk   (18000 ints/s)
872 	 *  > 40000 Rx packets per second (8000 ints/s)
873 	 *
874 	 * The math works out because the divisor is in 10^(-6) which
875 	 * turns the bytes/us input value into MB/s values, but
876 	 * make sure to use usecs, as the register values written
877 	 * are in 2 usec increments in the ITR registers, and make sure
878 	 * to use the smoothed values that the countdown timer gives us.
879 	 */
880 	usecs = (rc->itr << 1) * ITR_COUNTDOWN_START;
881 	bytes_per_int = rc->total_bytes / usecs;
882 
883 	switch (new_latency_range) {
884 	case I40E_LOWEST_LATENCY:
885 		if (bytes_per_int > 10)
886 			new_latency_range = I40E_LOW_LATENCY;
887 		break;
888 	case I40E_LOW_LATENCY:
889 		if (bytes_per_int > 20)
890 			new_latency_range = I40E_BULK_LATENCY;
891 		else if (bytes_per_int <= 10)
892 			new_latency_range = I40E_LOWEST_LATENCY;
893 		break;
894 	case I40E_BULK_LATENCY:
895 	case I40E_ULTRA_LATENCY:
896 	default:
897 		if (bytes_per_int <= 20)
898 			new_latency_range = I40E_LOW_LATENCY;
899 		break;
900 	}
901 
902 	/* this is to adjust RX more aggressively when streaming small
903 	 * packets.  The value of 40000 was picked as it is just beyond
904 	 * what the hardware can receive per second if in low latency
905 	 * mode.
906 	 */
907 #define RX_ULTRA_PACKET_RATE 40000
908 
909 	if ((((rc->total_packets * 1000000) / usecs) > RX_ULTRA_PACKET_RATE) &&
910 	    (&qv->rx == rc))
911 		new_latency_range = I40E_ULTRA_LATENCY;
912 
913 	rc->latency_range = new_latency_range;
914 
915 	switch (new_latency_range) {
916 	case I40E_LOWEST_LATENCY:
917 		new_itr = I40E_ITR_50K;
918 		break;
919 	case I40E_LOW_LATENCY:
920 		new_itr = I40E_ITR_20K;
921 		break;
922 	case I40E_BULK_LATENCY:
923 		new_itr = I40E_ITR_18K;
924 		break;
925 	case I40E_ULTRA_LATENCY:
926 		new_itr = I40E_ITR_8K;
927 		break;
928 	default:
929 		break;
930 	}
931 
932 	rc->total_bytes = 0;
933 	rc->total_packets = 0;
934 
935 	if (new_itr != rc->itr) {
936 		rc->itr = new_itr;
937 		return true;
938 	}
939 
940 	return false;
941 }
942 
943 /**
944  * i40e_clean_programming_status - clean the programming status descriptor
945  * @rx_ring: the rx ring that has this descriptor
946  * @rx_desc: the rx descriptor written back by HW
947  *
948  * Flow director should handle FD_FILTER_STATUS to check its filter programming
949  * status being successful or not and take actions accordingly. FCoE should
950  * handle its context/filter programming/invalidation status and take actions.
951  *
952  **/
953 static void i40e_clean_programming_status(struct i40e_ring *rx_ring,
954 					  union i40e_rx_desc *rx_desc)
955 {
956 	u64 qw;
957 	u8 id;
958 
959 	qw = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
960 	id = (qw & I40E_RX_PROG_STATUS_DESC_QW1_PROGID_MASK) >>
961 		  I40E_RX_PROG_STATUS_DESC_QW1_PROGID_SHIFT;
962 
963 	if (id == I40E_RX_PROG_STATUS_DESC_FD_FILTER_STATUS)
964 		i40e_fd_handle_status(rx_ring, rx_desc, id);
965 #ifdef I40E_FCOE
966 	else if ((id == I40E_RX_PROG_STATUS_DESC_FCOE_CTXT_PROG_STATUS) ||
967 		 (id == I40E_RX_PROG_STATUS_DESC_FCOE_CTXT_INVL_STATUS))
968 		i40e_fcoe_handle_status(rx_ring, rx_desc, id);
969 #endif
970 }
971 
972 /**
973  * i40e_setup_tx_descriptors - Allocate the Tx descriptors
974  * @tx_ring: the tx ring to set up
975  *
976  * Return 0 on success, negative on error
977  **/
978 int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring)
979 {
980 	struct device *dev = tx_ring->dev;
981 	int bi_size;
982 
983 	if (!dev)
984 		return -ENOMEM;
985 
986 	/* warn if we are about to overwrite the pointer */
987 	WARN_ON(tx_ring->tx_bi);
988 	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
989 	tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
990 	if (!tx_ring->tx_bi)
991 		goto err;
992 
993 	/* round up to nearest 4K */
994 	tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
995 	/* add u32 for head writeback, align after this takes care of
996 	 * guaranteeing this is at least one cache line in size
997 	 */
998 	tx_ring->size += sizeof(u32);
999 	tx_ring->size = ALIGN(tx_ring->size, 4096);
1000 	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
1001 					   &tx_ring->dma, GFP_KERNEL);
1002 	if (!tx_ring->desc) {
1003 		dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
1004 			 tx_ring->size);
1005 		goto err;
1006 	}
1007 
1008 	tx_ring->next_to_use = 0;
1009 	tx_ring->next_to_clean = 0;
1010 	return 0;
1011 
1012 err:
1013 	kfree(tx_ring->tx_bi);
1014 	tx_ring->tx_bi = NULL;
1015 	return -ENOMEM;
1016 }
1017 
1018 /**
1019  * i40e_clean_rx_ring - Free Rx buffers
1020  * @rx_ring: ring to be cleaned
1021  **/
1022 void i40e_clean_rx_ring(struct i40e_ring *rx_ring)
1023 {
1024 	struct device *dev = rx_ring->dev;
1025 	unsigned long bi_size;
1026 	u16 i;
1027 
1028 	/* ring already cleared, nothing to do */
1029 	if (!rx_ring->rx_bi)
1030 		return;
1031 
1032 	/* Free all the Rx ring sk_buffs */
1033 	for (i = 0; i < rx_ring->count; i++) {
1034 		struct i40e_rx_buffer *rx_bi = &rx_ring->rx_bi[i];
1035 
1036 		if (rx_bi->skb) {
1037 			dev_kfree_skb(rx_bi->skb);
1038 			rx_bi->skb = NULL;
1039 		}
1040 		if (!rx_bi->page)
1041 			continue;
1042 
1043 		dma_unmap_page(dev, rx_bi->dma, PAGE_SIZE, DMA_FROM_DEVICE);
1044 		__free_pages(rx_bi->page, 0);
1045 
1046 		rx_bi->page = NULL;
1047 		rx_bi->page_offset = 0;
1048 	}
1049 
1050 	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
1051 	memset(rx_ring->rx_bi, 0, bi_size);
1052 
1053 	/* Zero out the descriptor ring */
1054 	memset(rx_ring->desc, 0, rx_ring->size);
1055 
1056 	rx_ring->next_to_alloc = 0;
1057 	rx_ring->next_to_clean = 0;
1058 	rx_ring->next_to_use = 0;
1059 }
1060 
1061 /**
1062  * i40e_free_rx_resources - Free Rx resources
1063  * @rx_ring: ring to clean the resources from
1064  *
1065  * Free all receive software resources
1066  **/
1067 void i40e_free_rx_resources(struct i40e_ring *rx_ring)
1068 {
1069 	i40e_clean_rx_ring(rx_ring);
1070 	kfree(rx_ring->rx_bi);
1071 	rx_ring->rx_bi = NULL;
1072 
1073 	if (rx_ring->desc) {
1074 		dma_free_coherent(rx_ring->dev, rx_ring->size,
1075 				  rx_ring->desc, rx_ring->dma);
1076 		rx_ring->desc = NULL;
1077 	}
1078 }
1079 
1080 /**
1081  * i40e_setup_rx_descriptors - Allocate Rx descriptors
1082  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
1083  *
1084  * Returns 0 on success, negative on failure
1085  **/
1086 int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring)
1087 {
1088 	struct device *dev = rx_ring->dev;
1089 	int bi_size;
1090 
1091 	/* warn if we are about to overwrite the pointer */
1092 	WARN_ON(rx_ring->rx_bi);
1093 	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
1094 	rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
1095 	if (!rx_ring->rx_bi)
1096 		goto err;
1097 
1098 	u64_stats_init(&rx_ring->syncp);
1099 
1100 	/* Round up to nearest 4K */
1101 	rx_ring->size = rx_ring->count * sizeof(union i40e_32byte_rx_desc);
1102 	rx_ring->size = ALIGN(rx_ring->size, 4096);
1103 	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
1104 					   &rx_ring->dma, GFP_KERNEL);
1105 
1106 	if (!rx_ring->desc) {
1107 		dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
1108 			 rx_ring->size);
1109 		goto err;
1110 	}
1111 
1112 	rx_ring->next_to_alloc = 0;
1113 	rx_ring->next_to_clean = 0;
1114 	rx_ring->next_to_use = 0;
1115 
1116 	return 0;
1117 err:
1118 	kfree(rx_ring->rx_bi);
1119 	rx_ring->rx_bi = NULL;
1120 	return -ENOMEM;
1121 }
1122 
1123 /**
1124  * i40e_release_rx_desc - Store the new tail and head values
1125  * @rx_ring: ring to bump
1126  * @val: new head index
1127  **/
1128 static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
1129 {
1130 	rx_ring->next_to_use = val;
1131 
1132 	/* update next to alloc since we have filled the ring */
1133 	rx_ring->next_to_alloc = val;
1134 
1135 	/* Force memory writes to complete before letting h/w
1136 	 * know there are new descriptors to fetch.  (Only
1137 	 * applicable for weak-ordered memory model archs,
1138 	 * such as IA-64).
1139 	 */
1140 	wmb();
1141 	writel(val, rx_ring->tail);
1142 }
1143 
1144 /**
1145  * i40e_alloc_mapped_page - recycle or make a new page
1146  * @rx_ring: ring to use
1147  * @bi: rx_buffer struct to modify
1148  *
1149  * Returns true if the page was successfully allocated or
1150  * reused.
1151  **/
1152 static bool i40e_alloc_mapped_page(struct i40e_ring *rx_ring,
1153 				   struct i40e_rx_buffer *bi)
1154 {
1155 	struct page *page = bi->page;
1156 	dma_addr_t dma;
1157 
1158 	/* since we are recycling buffers we should seldom need to alloc */
1159 	if (likely(page)) {
1160 		rx_ring->rx_stats.page_reuse_count++;
1161 		return true;
1162 	}
1163 
1164 	/* alloc new page for storage */
1165 	page = dev_alloc_page();
1166 	if (unlikely(!page)) {
1167 		rx_ring->rx_stats.alloc_page_failed++;
1168 		return false;
1169 	}
1170 
1171 	/* map page for use */
1172 	dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
1173 
1174 	/* if mapping failed free memory back to system since
1175 	 * there isn't much point in holding memory we can't use
1176 	 */
1177 	if (dma_mapping_error(rx_ring->dev, dma)) {
1178 		__free_pages(page, 0);
1179 		rx_ring->rx_stats.alloc_page_failed++;
1180 		return false;
1181 	}
1182 
1183 	bi->dma = dma;
1184 	bi->page = page;
1185 	bi->page_offset = 0;
1186 
1187 	return true;
1188 }
1189 
1190 /**
1191  * i40e_receive_skb - Send a completed packet up the stack
1192  * @rx_ring:  rx ring in play
1193  * @skb: packet to send up
1194  * @vlan_tag: vlan tag for packet
1195  **/
1196 static void i40e_receive_skb(struct i40e_ring *rx_ring,
1197 			     struct sk_buff *skb, u16 vlan_tag)
1198 {
1199 	struct i40e_q_vector *q_vector = rx_ring->q_vector;
1200 
1201 	if ((rx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1202 	    (vlan_tag & VLAN_VID_MASK))
1203 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
1204 
1205 	napi_gro_receive(&q_vector->napi, skb);
1206 }
1207 
1208 /**
1209  * i40e_alloc_rx_buffers - Replace used receive buffers
1210  * @rx_ring: ring to place buffers on
1211  * @cleaned_count: number of buffers to replace
1212  *
1213  * Returns false if all allocations were successful, true if any fail
1214  **/
1215 bool i40e_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count)
1216 {
1217 	u16 ntu = rx_ring->next_to_use;
1218 	union i40e_rx_desc *rx_desc;
1219 	struct i40e_rx_buffer *bi;
1220 
1221 	/* do nothing if no valid netdev defined */
1222 	if (!rx_ring->netdev || !cleaned_count)
1223 		return false;
1224 
1225 	rx_desc = I40E_RX_DESC(rx_ring, ntu);
1226 	bi = &rx_ring->rx_bi[ntu];
1227 
1228 	do {
1229 		if (!i40e_alloc_mapped_page(rx_ring, bi))
1230 			goto no_buffers;
1231 
1232 		/* Refresh the desc even if buffer_addrs didn't change
1233 		 * because each write-back erases this info.
1234 		 */
1235 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1236 		rx_desc->read.hdr_addr = 0;
1237 
1238 		rx_desc++;
1239 		bi++;
1240 		ntu++;
1241 		if (unlikely(ntu == rx_ring->count)) {
1242 			rx_desc = I40E_RX_DESC(rx_ring, 0);
1243 			bi = rx_ring->rx_bi;
1244 			ntu = 0;
1245 		}
1246 
1247 		/* clear the status bits for the next_to_use descriptor */
1248 		rx_desc->wb.qword1.status_error_len = 0;
1249 
1250 		cleaned_count--;
1251 	} while (cleaned_count);
1252 
1253 	if (rx_ring->next_to_use != ntu)
1254 		i40e_release_rx_desc(rx_ring, ntu);
1255 
1256 	return false;
1257 
1258 no_buffers:
1259 	if (rx_ring->next_to_use != ntu)
1260 		i40e_release_rx_desc(rx_ring, ntu);
1261 
1262 	/* make sure to come back via polling to try again after
1263 	 * allocation failure
1264 	 */
1265 	return true;
1266 }
1267 
1268 /**
1269  * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
1270  * @vsi: the VSI we care about
1271  * @skb: skb currently being received and modified
1272  * @rx_desc: the receive descriptor
1273  *
1274  * skb->protocol must be set before this function is called
1275  **/
1276 static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
1277 				    struct sk_buff *skb,
1278 				    union i40e_rx_desc *rx_desc)
1279 {
1280 	struct i40e_rx_ptype_decoded decoded;
1281 	u32 rx_error, rx_status;
1282 	bool ipv4, ipv6;
1283 	u8 ptype;
1284 	u64 qword;
1285 
1286 	qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1287 	ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >> I40E_RXD_QW1_PTYPE_SHIFT;
1288 	rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1289 		   I40E_RXD_QW1_ERROR_SHIFT;
1290 	rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1291 		    I40E_RXD_QW1_STATUS_SHIFT;
1292 	decoded = decode_rx_desc_ptype(ptype);
1293 
1294 	skb->ip_summed = CHECKSUM_NONE;
1295 
1296 	skb_checksum_none_assert(skb);
1297 
1298 	/* Rx csum enabled and ip headers found? */
1299 	if (!(vsi->netdev->features & NETIF_F_RXCSUM))
1300 		return;
1301 
1302 	/* did the hardware decode the packet and checksum? */
1303 	if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
1304 		return;
1305 
1306 	/* both known and outer_ip must be set for the below code to work */
1307 	if (!(decoded.known && decoded.outer_ip))
1308 		return;
1309 
1310 	ipv4 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1311 	       (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4);
1312 	ipv6 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1313 	       (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6);
1314 
1315 	if (ipv4 &&
1316 	    (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) |
1317 			 BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT))))
1318 		goto checksum_fail;
1319 
1320 	/* likely incorrect csum if alternate IP extension headers found */
1321 	if (ipv6 &&
1322 	    rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
1323 		/* don't increment checksum err here, non-fatal err */
1324 		return;
1325 
1326 	/* there was some L4 error, count error and punt packet to the stack */
1327 	if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT))
1328 		goto checksum_fail;
1329 
1330 	/* handle packets that were not able to be checksummed due
1331 	 * to arrival speed, in this case the stack can compute
1332 	 * the csum.
1333 	 */
1334 	if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT))
1335 		return;
1336 
1337 	/* If there is an outer header present that might contain a checksum
1338 	 * we need to bump the checksum level by 1 to reflect the fact that
1339 	 * we are indicating we validated the inner checksum.
1340 	 */
1341 	if (decoded.tunnel_type >= I40E_RX_PTYPE_TUNNEL_IP_GRENAT)
1342 		skb->csum_level = 1;
1343 
1344 	/* Only report checksum unnecessary for TCP, UDP, or SCTP */
1345 	switch (decoded.inner_prot) {
1346 	case I40E_RX_PTYPE_INNER_PROT_TCP:
1347 	case I40E_RX_PTYPE_INNER_PROT_UDP:
1348 	case I40E_RX_PTYPE_INNER_PROT_SCTP:
1349 		skb->ip_summed = CHECKSUM_UNNECESSARY;
1350 		/* fall though */
1351 	default:
1352 		break;
1353 	}
1354 
1355 	return;
1356 
1357 checksum_fail:
1358 	vsi->back->hw_csum_rx_error++;
1359 }
1360 
1361 /**
1362  * i40e_ptype_to_htype - get a hash type
1363  * @ptype: the ptype value from the descriptor
1364  *
1365  * Returns a hash type to be used by skb_set_hash
1366  **/
1367 static inline int i40e_ptype_to_htype(u8 ptype)
1368 {
1369 	struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
1370 
1371 	if (!decoded.known)
1372 		return PKT_HASH_TYPE_NONE;
1373 
1374 	if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1375 	    decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
1376 		return PKT_HASH_TYPE_L4;
1377 	else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1378 		 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
1379 		return PKT_HASH_TYPE_L3;
1380 	else
1381 		return PKT_HASH_TYPE_L2;
1382 }
1383 
1384 /**
1385  * i40e_rx_hash - set the hash value in the skb
1386  * @ring: descriptor ring
1387  * @rx_desc: specific descriptor
1388  **/
1389 static inline void i40e_rx_hash(struct i40e_ring *ring,
1390 				union i40e_rx_desc *rx_desc,
1391 				struct sk_buff *skb,
1392 				u8 rx_ptype)
1393 {
1394 	u32 hash;
1395 	const __le64 rss_mask =
1396 		cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
1397 			    I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
1398 
1399 	if (!(ring->netdev->features & NETIF_F_RXHASH))
1400 		return;
1401 
1402 	if ((rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask) {
1403 		hash = le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
1404 		skb_set_hash(skb, hash, i40e_ptype_to_htype(rx_ptype));
1405 	}
1406 }
1407 
1408 /**
1409  * i40e_process_skb_fields - Populate skb header fields from Rx descriptor
1410  * @rx_ring: rx descriptor ring packet is being transacted on
1411  * @rx_desc: pointer to the EOP Rx descriptor
1412  * @skb: pointer to current skb being populated
1413  * @rx_ptype: the packet type decoded by hardware
1414  *
1415  * This function checks the ring, descriptor, and packet information in
1416  * order to populate the hash, checksum, VLAN, protocol, and
1417  * other fields within the skb.
1418  **/
1419 static inline
1420 void i40e_process_skb_fields(struct i40e_ring *rx_ring,
1421 			     union i40e_rx_desc *rx_desc, struct sk_buff *skb,
1422 			     u8 rx_ptype)
1423 {
1424 	u64 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1425 	u32 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1426 			I40E_RXD_QW1_STATUS_SHIFT;
1427 	u32 rsyn = (rx_status & I40E_RXD_QW1_STATUS_TSYNINDX_MASK) >>
1428 		   I40E_RXD_QW1_STATUS_TSYNINDX_SHIFT;
1429 
1430 	if (unlikely(rsyn)) {
1431 		i40e_ptp_rx_hwtstamp(rx_ring->vsi->back, skb, rsyn);
1432 		rx_ring->last_rx_timestamp = jiffies;
1433 	}
1434 
1435 	i40e_rx_hash(rx_ring, rx_desc, skb, rx_ptype);
1436 
1437 	/* modifies the skb - consumes the enet header */
1438 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1439 
1440 	i40e_rx_checksum(rx_ring->vsi, skb, rx_desc);
1441 
1442 	skb_record_rx_queue(skb, rx_ring->queue_index);
1443 }
1444 
1445 /**
1446  * i40e_pull_tail - i40e specific version of skb_pull_tail
1447  * @rx_ring: rx descriptor ring packet is being transacted on
1448  * @skb: pointer to current skb being adjusted
1449  *
1450  * This function is an i40e specific version of __pskb_pull_tail.  The
1451  * main difference between this version and the original function is that
1452  * this function can make several assumptions about the state of things
1453  * that allow for significant optimizations versus the standard function.
1454  * As a result we can do things like drop a frag and maintain an accurate
1455  * truesize for the skb.
1456  */
1457 static void i40e_pull_tail(struct i40e_ring *rx_ring, struct sk_buff *skb)
1458 {
1459 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1460 	unsigned char *va;
1461 	unsigned int pull_len;
1462 
1463 	/* it is valid to use page_address instead of kmap since we are
1464 	 * working with pages allocated out of the lomem pool per
1465 	 * alloc_page(GFP_ATOMIC)
1466 	 */
1467 	va = skb_frag_address(frag);
1468 
1469 	/* we need the header to contain the greater of either ETH_HLEN or
1470 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
1471 	 */
1472 	pull_len = eth_get_headlen(va, I40E_RX_HDR_SIZE);
1473 
1474 	/* align pull length to size of long to optimize memcpy performance */
1475 	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1476 
1477 	/* update all of the pointers */
1478 	skb_frag_size_sub(frag, pull_len);
1479 	frag->page_offset += pull_len;
1480 	skb->data_len -= pull_len;
1481 	skb->tail += pull_len;
1482 }
1483 
1484 /**
1485  * i40e_cleanup_headers - Correct empty headers
1486  * @rx_ring: rx descriptor ring packet is being transacted on
1487  * @skb: pointer to current skb being fixed
1488  *
1489  * Also address the case where we are pulling data in on pages only
1490  * and as such no data is present in the skb header.
1491  *
1492  * In addition if skb is not at least 60 bytes we need to pad it so that
1493  * it is large enough to qualify as a valid Ethernet frame.
1494  *
1495  * Returns true if an error was encountered and skb was freed.
1496  **/
1497 static bool i40e_cleanup_headers(struct i40e_ring *rx_ring, struct sk_buff *skb)
1498 {
1499 	/* place header in linear portion of buffer */
1500 	if (skb_is_nonlinear(skb))
1501 		i40e_pull_tail(rx_ring, skb);
1502 
1503 	/* if eth_skb_pad returns an error the skb was freed */
1504 	if (eth_skb_pad(skb))
1505 		return true;
1506 
1507 	return false;
1508 }
1509 
1510 /**
1511  * i40e_reuse_rx_page - page flip buffer and store it back on the ring
1512  * @rx_ring: rx descriptor ring to store buffers on
1513  * @old_buff: donor buffer to have page reused
1514  *
1515  * Synchronizes page for reuse by the adapter
1516  **/
1517 static void i40e_reuse_rx_page(struct i40e_ring *rx_ring,
1518 			       struct i40e_rx_buffer *old_buff)
1519 {
1520 	struct i40e_rx_buffer *new_buff;
1521 	u16 nta = rx_ring->next_to_alloc;
1522 
1523 	new_buff = &rx_ring->rx_bi[nta];
1524 
1525 	/* update, and store next to alloc */
1526 	nta++;
1527 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1528 
1529 	/* transfer page from old buffer to new buffer */
1530 	*new_buff = *old_buff;
1531 }
1532 
1533 /**
1534  * i40e_page_is_reserved - check if reuse is possible
1535  * @page: page struct to check
1536  */
1537 static inline bool i40e_page_is_reserved(struct page *page)
1538 {
1539 	return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1540 }
1541 
1542 /**
1543  * i40e_add_rx_frag - Add contents of Rx buffer to sk_buff
1544  * @rx_ring: rx descriptor ring to transact packets on
1545  * @rx_buffer: buffer containing page to add
1546  * @rx_desc: descriptor containing length of buffer written by hardware
1547  * @skb: sk_buff to place the data into
1548  *
1549  * This function will add the data contained in rx_buffer->page to the skb.
1550  * This is done either through a direct copy if the data in the buffer is
1551  * less than the skb header size, otherwise it will just attach the page as
1552  * a frag to the skb.
1553  *
1554  * The function will then update the page offset if necessary and return
1555  * true if the buffer can be reused by the adapter.
1556  **/
1557 static bool i40e_add_rx_frag(struct i40e_ring *rx_ring,
1558 			     struct i40e_rx_buffer *rx_buffer,
1559 			     union i40e_rx_desc *rx_desc,
1560 			     struct sk_buff *skb)
1561 {
1562 	struct page *page = rx_buffer->page;
1563 	u64 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1564 	unsigned int size = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1565 			    I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1566 #if (PAGE_SIZE < 8192)
1567 	unsigned int truesize = I40E_RXBUFFER_2048;
1568 #else
1569 	unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
1570 	unsigned int last_offset = PAGE_SIZE - I40E_RXBUFFER_2048;
1571 #endif
1572 
1573 	/* will the data fit in the skb we allocated? if so, just
1574 	 * copy it as it is pretty small anyway
1575 	 */
1576 	if ((size <= I40E_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
1577 		unsigned char *va = page_address(page) + rx_buffer->page_offset;
1578 
1579 		memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
1580 
1581 		/* page is not reserved, we can reuse buffer as-is */
1582 		if (likely(!i40e_page_is_reserved(page)))
1583 			return true;
1584 
1585 		/* this page cannot be reused so discard it */
1586 		__free_pages(page, 0);
1587 		return false;
1588 	}
1589 
1590 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1591 			rx_buffer->page_offset, size, truesize);
1592 
1593 	/* avoid re-using remote pages */
1594 	if (unlikely(i40e_page_is_reserved(page)))
1595 		return false;
1596 
1597 #if (PAGE_SIZE < 8192)
1598 	/* if we are only owner of page we can reuse it */
1599 	if (unlikely(page_count(page) != 1))
1600 		return false;
1601 
1602 	/* flip page offset to other buffer */
1603 	rx_buffer->page_offset ^= truesize;
1604 #else
1605 	/* move offset up to the next cache line */
1606 	rx_buffer->page_offset += truesize;
1607 
1608 	if (rx_buffer->page_offset > last_offset)
1609 		return false;
1610 #endif
1611 
1612 	/* Even if we own the page, we are not allowed to use atomic_set()
1613 	 * This would break get_page_unless_zero() users.
1614 	 */
1615 	get_page(rx_buffer->page);
1616 
1617 	return true;
1618 }
1619 
1620 /**
1621  * i40e_fetch_rx_buffer - Allocate skb and populate it
1622  * @rx_ring: rx descriptor ring to transact packets on
1623  * @rx_desc: descriptor containing info written by hardware
1624  *
1625  * This function allocates an skb on the fly, and populates it with the page
1626  * data from the current receive descriptor, taking care to set up the skb
1627  * correctly, as well as handling calling the page recycle function if
1628  * necessary.
1629  */
1630 static inline
1631 struct sk_buff *i40e_fetch_rx_buffer(struct i40e_ring *rx_ring,
1632 				     union i40e_rx_desc *rx_desc)
1633 {
1634 	struct i40e_rx_buffer *rx_buffer;
1635 	struct sk_buff *skb;
1636 	struct page *page;
1637 
1638 	rx_buffer = &rx_ring->rx_bi[rx_ring->next_to_clean];
1639 	page = rx_buffer->page;
1640 	prefetchw(page);
1641 
1642 	skb = rx_buffer->skb;
1643 
1644 	if (likely(!skb)) {
1645 		void *page_addr = page_address(page) + rx_buffer->page_offset;
1646 
1647 		/* prefetch first cache line of first page */
1648 		prefetch(page_addr);
1649 #if L1_CACHE_BYTES < 128
1650 		prefetch(page_addr + L1_CACHE_BYTES);
1651 #endif
1652 
1653 		/* allocate a skb to store the frags */
1654 		skb = __napi_alloc_skb(&rx_ring->q_vector->napi,
1655 				       I40E_RX_HDR_SIZE,
1656 				       GFP_ATOMIC | __GFP_NOWARN);
1657 		if (unlikely(!skb)) {
1658 			rx_ring->rx_stats.alloc_buff_failed++;
1659 			return NULL;
1660 		}
1661 
1662 		/* we will be copying header into skb->data in
1663 		 * pskb_may_pull so it is in our interest to prefetch
1664 		 * it now to avoid a possible cache miss
1665 		 */
1666 		prefetchw(skb->data);
1667 	} else {
1668 		rx_buffer->skb = NULL;
1669 	}
1670 
1671 	/* we are reusing so sync this buffer for CPU use */
1672 	dma_sync_single_range_for_cpu(rx_ring->dev,
1673 				      rx_buffer->dma,
1674 				      rx_buffer->page_offset,
1675 				      I40E_RXBUFFER_2048,
1676 				      DMA_FROM_DEVICE);
1677 
1678 	/* pull page into skb */
1679 	if (i40e_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
1680 		/* hand second half of page back to the ring */
1681 		i40e_reuse_rx_page(rx_ring, rx_buffer);
1682 		rx_ring->rx_stats.page_reuse_count++;
1683 	} else {
1684 		/* we are not reusing the buffer so unmap it */
1685 		dma_unmap_page(rx_ring->dev, rx_buffer->dma, PAGE_SIZE,
1686 			       DMA_FROM_DEVICE);
1687 	}
1688 
1689 	/* clear contents of buffer_info */
1690 	rx_buffer->page = NULL;
1691 
1692 	return skb;
1693 }
1694 
1695 /**
1696  * i40e_is_non_eop - process handling of non-EOP buffers
1697  * @rx_ring: Rx ring being processed
1698  * @rx_desc: Rx descriptor for current buffer
1699  * @skb: Current socket buffer containing buffer in progress
1700  *
1701  * This function updates next to clean.  If the buffer is an EOP buffer
1702  * this function exits returning false, otherwise it will place the
1703  * sk_buff in the next buffer to be chained and return true indicating
1704  * that this is in fact a non-EOP buffer.
1705  **/
1706 static bool i40e_is_non_eop(struct i40e_ring *rx_ring,
1707 			    union i40e_rx_desc *rx_desc,
1708 			    struct sk_buff *skb)
1709 {
1710 	u32 ntc = rx_ring->next_to_clean + 1;
1711 
1712 	/* fetch, update, and store next to clean */
1713 	ntc = (ntc < rx_ring->count) ? ntc : 0;
1714 	rx_ring->next_to_clean = ntc;
1715 
1716 	prefetch(I40E_RX_DESC(rx_ring, ntc));
1717 
1718 #define staterrlen rx_desc->wb.qword1.status_error_len
1719 	if (unlikely(i40e_rx_is_programming_status(le64_to_cpu(staterrlen)))) {
1720 		i40e_clean_programming_status(rx_ring, rx_desc);
1721 		rx_ring->rx_bi[ntc].skb = skb;
1722 		return true;
1723 	}
1724 	/* if we are the last buffer then there is nothing else to do */
1725 #define I40E_RXD_EOF BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)
1726 	if (likely(i40e_test_staterr(rx_desc, I40E_RXD_EOF)))
1727 		return false;
1728 
1729 	/* place skb in next buffer to be received */
1730 	rx_ring->rx_bi[ntc].skb = skb;
1731 	rx_ring->rx_stats.non_eop_descs++;
1732 
1733 	return true;
1734 }
1735 
1736 /**
1737  * i40e_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1738  * @rx_ring: rx descriptor ring to transact packets on
1739  * @budget: Total limit on number of packets to process
1740  *
1741  * This function provides a "bounce buffer" approach to Rx interrupt
1742  * processing.  The advantage to this is that on systems that have
1743  * expensive overhead for IOMMU access this provides a means of avoiding
1744  * it by maintaining the mapping of the page to the system.
1745  *
1746  * Returns amount of work completed
1747  **/
1748 static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget)
1749 {
1750 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1751 	u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
1752 	bool failure = false;
1753 
1754 	while (likely(total_rx_packets < budget)) {
1755 		union i40e_rx_desc *rx_desc;
1756 		struct sk_buff *skb;
1757 		u32 rx_status;
1758 		u16 vlan_tag;
1759 		u8 rx_ptype;
1760 		u64 qword;
1761 
1762 		/* return some buffers to hardware, one at a time is too slow */
1763 		if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1764 			failure = failure ||
1765 				  i40e_alloc_rx_buffers(rx_ring, cleaned_count);
1766 			cleaned_count = 0;
1767 		}
1768 
1769 		rx_desc = I40E_RX_DESC(rx_ring, rx_ring->next_to_clean);
1770 
1771 		qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1772 		rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1773 			   I40E_RXD_QW1_PTYPE_SHIFT;
1774 		rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1775 			    I40E_RXD_QW1_STATUS_SHIFT;
1776 
1777 		if (!(rx_status & BIT(I40E_RX_DESC_STATUS_DD_SHIFT)))
1778 			break;
1779 
1780 		/* status_error_len will always be zero for unused descriptors
1781 		 * because it's cleared in cleanup, and overlaps with hdr_addr
1782 		 * which is always zero because packet split isn't used, if the
1783 		 * hardware wrote DD then it will be non-zero
1784 		 */
1785 		if (!rx_desc->wb.qword1.status_error_len)
1786 			break;
1787 
1788 		/* This memory barrier is needed to keep us from reading
1789 		 * any other fields out of the rx_desc until we know the
1790 		 * DD bit is set.
1791 		 */
1792 		dma_rmb();
1793 
1794 		skb = i40e_fetch_rx_buffer(rx_ring, rx_desc);
1795 		if (!skb)
1796 			break;
1797 
1798 		cleaned_count++;
1799 
1800 		if (i40e_is_non_eop(rx_ring, rx_desc, skb))
1801 			continue;
1802 
1803 		/* ERR_MASK will only have valid bits if EOP set, and
1804 		 * what we are doing here is actually checking
1805 		 * I40E_RX_DESC_ERROR_RXE_SHIFT, since it is the zeroth bit in
1806 		 * the error field
1807 		 */
1808 		if (unlikely(i40e_test_staterr(rx_desc, BIT(I40E_RXD_QW1_ERROR_SHIFT)))) {
1809 			dev_kfree_skb_any(skb);
1810 			continue;
1811 		}
1812 
1813 		if (i40e_cleanup_headers(rx_ring, skb))
1814 			continue;
1815 
1816 		/* probably a little skewed due to removing CRC */
1817 		total_rx_bytes += skb->len;
1818 
1819 		/* populate checksum, VLAN, and protocol */
1820 		i40e_process_skb_fields(rx_ring, rx_desc, skb, rx_ptype);
1821 
1822 #ifdef I40E_FCOE
1823 		if (unlikely(
1824 		    i40e_rx_is_fcoe(rx_ptype) &&
1825 		    !i40e_fcoe_handle_offload(rx_ring, rx_desc, skb))) {
1826 			dev_kfree_skb_any(skb);
1827 			continue;
1828 		}
1829 #endif
1830 
1831 		vlan_tag = (qword & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)) ?
1832 			   le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1) : 0;
1833 
1834 		i40e_receive_skb(rx_ring, skb, vlan_tag);
1835 
1836 		/* update budget accounting */
1837 		total_rx_packets++;
1838 	}
1839 
1840 	u64_stats_update_begin(&rx_ring->syncp);
1841 	rx_ring->stats.packets += total_rx_packets;
1842 	rx_ring->stats.bytes += total_rx_bytes;
1843 	u64_stats_update_end(&rx_ring->syncp);
1844 	rx_ring->q_vector->rx.total_packets += total_rx_packets;
1845 	rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1846 
1847 	/* guarantee a trip back through this routine if there was a failure */
1848 	return failure ? budget : total_rx_packets;
1849 }
1850 
1851 static u32 i40e_buildreg_itr(const int type, const u16 itr)
1852 {
1853 	u32 val;
1854 
1855 	val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
1856 	      /* Don't clear PBA because that can cause lost interrupts that
1857 	       * came in while we were cleaning/polling
1858 	       */
1859 	      (type << I40E_PFINT_DYN_CTLN_ITR_INDX_SHIFT) |
1860 	      (itr << I40E_PFINT_DYN_CTLN_INTERVAL_SHIFT);
1861 
1862 	return val;
1863 }
1864 
1865 /* a small macro to shorten up some long lines */
1866 #define INTREG I40E_PFINT_DYN_CTLN
1867 
1868 /**
1869  * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt
1870  * @vsi: the VSI we care about
1871  * @q_vector: q_vector for which itr is being updated and interrupt enabled
1872  *
1873  **/
1874 static inline void i40e_update_enable_itr(struct i40e_vsi *vsi,
1875 					  struct i40e_q_vector *q_vector)
1876 {
1877 	struct i40e_hw *hw = &vsi->back->hw;
1878 	bool rx = false, tx = false;
1879 	u32 rxval, txval;
1880 	int vector;
1881 	int idx = q_vector->v_idx;
1882 
1883 	vector = (q_vector->v_idx + vsi->base_vector);
1884 
1885 	/* avoid dynamic calculation if in countdown mode OR if
1886 	 * all dynamic is disabled
1887 	 */
1888 	rxval = txval = i40e_buildreg_itr(I40E_ITR_NONE, 0);
1889 
1890 	if (q_vector->itr_countdown > 0 ||
1891 	    (!ITR_IS_DYNAMIC(vsi->rx_rings[idx]->rx_itr_setting) &&
1892 	     !ITR_IS_DYNAMIC(vsi->tx_rings[idx]->tx_itr_setting))) {
1893 		goto enable_int;
1894 	}
1895 
1896 	if (ITR_IS_DYNAMIC(vsi->rx_rings[idx]->rx_itr_setting)) {
1897 		rx = i40e_set_new_dynamic_itr(&q_vector->rx);
1898 		rxval = i40e_buildreg_itr(I40E_RX_ITR, q_vector->rx.itr);
1899 	}
1900 
1901 	if (ITR_IS_DYNAMIC(vsi->tx_rings[idx]->tx_itr_setting)) {
1902 		tx = i40e_set_new_dynamic_itr(&q_vector->tx);
1903 		txval = i40e_buildreg_itr(I40E_TX_ITR, q_vector->tx.itr);
1904 	}
1905 
1906 	if (rx || tx) {
1907 		/* get the higher of the two ITR adjustments and
1908 		 * use the same value for both ITR registers
1909 		 * when in adaptive mode (Rx and/or Tx)
1910 		 */
1911 		u16 itr = max(q_vector->tx.itr, q_vector->rx.itr);
1912 
1913 		q_vector->tx.itr = q_vector->rx.itr = itr;
1914 		txval = i40e_buildreg_itr(I40E_TX_ITR, itr);
1915 		tx = true;
1916 		rxval = i40e_buildreg_itr(I40E_RX_ITR, itr);
1917 		rx = true;
1918 	}
1919 
1920 	/* only need to enable the interrupt once, but need
1921 	 * to possibly update both ITR values
1922 	 */
1923 	if (rx) {
1924 		/* set the INTENA_MSK_MASK so that this first write
1925 		 * won't actually enable the interrupt, instead just
1926 		 * updating the ITR (it's bit 31 PF and VF)
1927 		 */
1928 		rxval |= BIT(31);
1929 		/* don't check _DOWN because interrupt isn't being enabled */
1930 		wr32(hw, INTREG(vector - 1), rxval);
1931 	}
1932 
1933 enable_int:
1934 	if (!test_bit(__I40E_DOWN, &vsi->state))
1935 		wr32(hw, INTREG(vector - 1), txval);
1936 
1937 	if (q_vector->itr_countdown)
1938 		q_vector->itr_countdown--;
1939 	else
1940 		q_vector->itr_countdown = ITR_COUNTDOWN_START;
1941 }
1942 
1943 /**
1944  * i40e_napi_poll - NAPI polling Rx/Tx cleanup routine
1945  * @napi: napi struct with our devices info in it
1946  * @budget: amount of work driver is allowed to do this pass, in packets
1947  *
1948  * This function will clean all queues associated with a q_vector.
1949  *
1950  * Returns the amount of work done
1951  **/
1952 int i40e_napi_poll(struct napi_struct *napi, int budget)
1953 {
1954 	struct i40e_q_vector *q_vector =
1955 			       container_of(napi, struct i40e_q_vector, napi);
1956 	struct i40e_vsi *vsi = q_vector->vsi;
1957 	struct i40e_ring *ring;
1958 	bool clean_complete = true;
1959 	bool arm_wb = false;
1960 	int budget_per_ring;
1961 	int work_done = 0;
1962 
1963 	if (test_bit(__I40E_DOWN, &vsi->state)) {
1964 		napi_complete(napi);
1965 		return 0;
1966 	}
1967 
1968 	/* Clear hung_detected bit */
1969 	clear_bit(I40E_Q_VECTOR_HUNG_DETECT, &q_vector->hung_detected);
1970 	/* Since the actual Tx work is minimal, we can give the Tx a larger
1971 	 * budget and be more aggressive about cleaning up the Tx descriptors.
1972 	 */
1973 	i40e_for_each_ring(ring, q_vector->tx) {
1974 		if (!i40e_clean_tx_irq(vsi, ring, budget)) {
1975 			clean_complete = false;
1976 			continue;
1977 		}
1978 		arm_wb |= ring->arm_wb;
1979 		ring->arm_wb = false;
1980 	}
1981 
1982 	/* Handle case where we are called by netpoll with a budget of 0 */
1983 	if (budget <= 0)
1984 		goto tx_only;
1985 
1986 	/* We attempt to distribute budget to each Rx queue fairly, but don't
1987 	 * allow the budget to go below 1 because that would exit polling early.
1988 	 */
1989 	budget_per_ring = max(budget/q_vector->num_ringpairs, 1);
1990 
1991 	i40e_for_each_ring(ring, q_vector->rx) {
1992 		int cleaned = i40e_clean_rx_irq(ring, budget_per_ring);
1993 
1994 		work_done += cleaned;
1995 		/* if we clean as many as budgeted, we must not be done */
1996 		if (cleaned >= budget_per_ring)
1997 			clean_complete = false;
1998 	}
1999 
2000 	/* If work not completed, return budget and polling will return */
2001 	if (!clean_complete) {
2002 tx_only:
2003 		if (arm_wb) {
2004 			q_vector->tx.ring[0].tx_stats.tx_force_wb++;
2005 			i40e_enable_wb_on_itr(vsi, q_vector);
2006 		}
2007 		return budget;
2008 	}
2009 
2010 	if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR)
2011 		q_vector->arm_wb_state = false;
2012 
2013 	/* Work is done so exit the polling mode and re-enable the interrupt */
2014 	napi_complete_done(napi, work_done);
2015 	if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
2016 		i40e_update_enable_itr(vsi, q_vector);
2017 	} else { /* Legacy mode */
2018 		i40e_irq_dynamic_enable_icr0(vsi->back, false);
2019 	}
2020 	return 0;
2021 }
2022 
2023 /**
2024  * i40e_atr - Add a Flow Director ATR filter
2025  * @tx_ring:  ring to add programming descriptor to
2026  * @skb:      send buffer
2027  * @tx_flags: send tx flags
2028  **/
2029 static void i40e_atr(struct i40e_ring *tx_ring, struct sk_buff *skb,
2030 		     u32 tx_flags)
2031 {
2032 	struct i40e_filter_program_desc *fdir_desc;
2033 	struct i40e_pf *pf = tx_ring->vsi->back;
2034 	union {
2035 		unsigned char *network;
2036 		struct iphdr *ipv4;
2037 		struct ipv6hdr *ipv6;
2038 	} hdr;
2039 	struct tcphdr *th;
2040 	unsigned int hlen;
2041 	u32 flex_ptype, dtype_cmd;
2042 	int l4_proto;
2043 	u16 i;
2044 
2045 	/* make sure ATR is enabled */
2046 	if (!(pf->flags & I40E_FLAG_FD_ATR_ENABLED))
2047 		return;
2048 
2049 	if ((pf->auto_disable_flags & I40E_FLAG_FD_ATR_ENABLED))
2050 		return;
2051 
2052 	/* if sampling is disabled do nothing */
2053 	if (!tx_ring->atr_sample_rate)
2054 		return;
2055 
2056 	/* Currently only IPv4/IPv6 with TCP is supported */
2057 	if (!(tx_flags & (I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6)))
2058 		return;
2059 
2060 	/* snag network header to get L4 type and address */
2061 	hdr.network = (tx_flags & I40E_TX_FLAGS_UDP_TUNNEL) ?
2062 		      skb_inner_network_header(skb) : skb_network_header(skb);
2063 
2064 	/* Note: tx_flags gets modified to reflect inner protocols in
2065 	 * tx_enable_csum function if encap is enabled.
2066 	 */
2067 	if (tx_flags & I40E_TX_FLAGS_IPV4) {
2068 		/* access ihl as u8 to avoid unaligned access on ia64 */
2069 		hlen = (hdr.network[0] & 0x0F) << 2;
2070 		l4_proto = hdr.ipv4->protocol;
2071 	} else {
2072 		hlen = hdr.network - skb->data;
2073 		l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
2074 		hlen -= hdr.network - skb->data;
2075 	}
2076 
2077 	if (l4_proto != IPPROTO_TCP)
2078 		return;
2079 
2080 	th = (struct tcphdr *)(hdr.network + hlen);
2081 
2082 	/* Due to lack of space, no more new filters can be programmed */
2083 	if (th->syn && (pf->auto_disable_flags & I40E_FLAG_FD_ATR_ENABLED))
2084 		return;
2085 	if ((pf->flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE) &&
2086 	    (!(pf->auto_disable_flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE))) {
2087 		/* HW ATR eviction will take care of removing filters on FIN
2088 		 * and RST packets.
2089 		 */
2090 		if (th->fin || th->rst)
2091 			return;
2092 	}
2093 
2094 	tx_ring->atr_count++;
2095 
2096 	/* sample on all syn/fin/rst packets or once every atr sample rate */
2097 	if (!th->fin &&
2098 	    !th->syn &&
2099 	    !th->rst &&
2100 	    (tx_ring->atr_count < tx_ring->atr_sample_rate))
2101 		return;
2102 
2103 	tx_ring->atr_count = 0;
2104 
2105 	/* grab the next descriptor */
2106 	i = tx_ring->next_to_use;
2107 	fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
2108 
2109 	i++;
2110 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2111 
2112 	flex_ptype = (tx_ring->queue_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) &
2113 		      I40E_TXD_FLTR_QW0_QINDEX_MASK;
2114 	flex_ptype |= (tx_flags & I40E_TX_FLAGS_IPV4) ?
2115 		      (I40E_FILTER_PCTYPE_NONF_IPV4_TCP <<
2116 		       I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) :
2117 		      (I40E_FILTER_PCTYPE_NONF_IPV6_TCP <<
2118 		       I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
2119 
2120 	flex_ptype |= tx_ring->vsi->id << I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT;
2121 
2122 	dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
2123 
2124 	dtype_cmd |= (th->fin || th->rst) ?
2125 		     (I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
2126 		      I40E_TXD_FLTR_QW1_PCMD_SHIFT) :
2127 		     (I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
2128 		      I40E_TXD_FLTR_QW1_PCMD_SHIFT);
2129 
2130 	dtype_cmd |= I40E_FILTER_PROGRAM_DESC_DEST_DIRECT_PACKET_QINDEX <<
2131 		     I40E_TXD_FLTR_QW1_DEST_SHIFT;
2132 
2133 	dtype_cmd |= I40E_FILTER_PROGRAM_DESC_FD_STATUS_FD_ID <<
2134 		     I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT;
2135 
2136 	dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
2137 	if (!(tx_flags & I40E_TX_FLAGS_UDP_TUNNEL))
2138 		dtype_cmd |=
2139 			((u32)I40E_FD_ATR_STAT_IDX(pf->hw.pf_id) <<
2140 			I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2141 			I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2142 	else
2143 		dtype_cmd |=
2144 			((u32)I40E_FD_ATR_TUNNEL_STAT_IDX(pf->hw.pf_id) <<
2145 			I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2146 			I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2147 
2148 	if ((pf->flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE) &&
2149 	    (!(pf->auto_disable_flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE)))
2150 		dtype_cmd |= I40E_TXD_FLTR_QW1_ATR_MASK;
2151 
2152 	fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
2153 	fdir_desc->rsvd = cpu_to_le32(0);
2154 	fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
2155 	fdir_desc->fd_id = cpu_to_le32(0);
2156 }
2157 
2158 /**
2159  * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
2160  * @skb:     send buffer
2161  * @tx_ring: ring to send buffer on
2162  * @flags:   the tx flags to be set
2163  *
2164  * Checks the skb and set up correspondingly several generic transmit flags
2165  * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
2166  *
2167  * Returns error code indicate the frame should be dropped upon error and the
2168  * otherwise  returns 0 to indicate the flags has been set properly.
2169  **/
2170 #ifdef I40E_FCOE
2171 inline int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
2172 				      struct i40e_ring *tx_ring,
2173 				      u32 *flags)
2174 #else
2175 static inline int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
2176 					     struct i40e_ring *tx_ring,
2177 					     u32 *flags)
2178 #endif
2179 {
2180 	__be16 protocol = skb->protocol;
2181 	u32  tx_flags = 0;
2182 
2183 	if (protocol == htons(ETH_P_8021Q) &&
2184 	    !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
2185 		/* When HW VLAN acceleration is turned off by the user the
2186 		 * stack sets the protocol to 8021q so that the driver
2187 		 * can take any steps required to support the SW only
2188 		 * VLAN handling.  In our case the driver doesn't need
2189 		 * to take any further steps so just set the protocol
2190 		 * to the encapsulated ethertype.
2191 		 */
2192 		skb->protocol = vlan_get_protocol(skb);
2193 		goto out;
2194 	}
2195 
2196 	/* if we have a HW VLAN tag being added, default to the HW one */
2197 	if (skb_vlan_tag_present(skb)) {
2198 		tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
2199 		tx_flags |= I40E_TX_FLAGS_HW_VLAN;
2200 	/* else if it is a SW VLAN, check the next protocol and store the tag */
2201 	} else if (protocol == htons(ETH_P_8021Q)) {
2202 		struct vlan_hdr *vhdr, _vhdr;
2203 
2204 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
2205 		if (!vhdr)
2206 			return -EINVAL;
2207 
2208 		protocol = vhdr->h_vlan_encapsulated_proto;
2209 		tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
2210 		tx_flags |= I40E_TX_FLAGS_SW_VLAN;
2211 	}
2212 
2213 	if (!(tx_ring->vsi->back->flags & I40E_FLAG_DCB_ENABLED))
2214 		goto out;
2215 
2216 	/* Insert 802.1p priority into VLAN header */
2217 	if ((tx_flags & (I40E_TX_FLAGS_HW_VLAN | I40E_TX_FLAGS_SW_VLAN)) ||
2218 	    (skb->priority != TC_PRIO_CONTROL)) {
2219 		tx_flags &= ~I40E_TX_FLAGS_VLAN_PRIO_MASK;
2220 		tx_flags |= (skb->priority & 0x7) <<
2221 				I40E_TX_FLAGS_VLAN_PRIO_SHIFT;
2222 		if (tx_flags & I40E_TX_FLAGS_SW_VLAN) {
2223 			struct vlan_ethhdr *vhdr;
2224 			int rc;
2225 
2226 			rc = skb_cow_head(skb, 0);
2227 			if (rc < 0)
2228 				return rc;
2229 			vhdr = (struct vlan_ethhdr *)skb->data;
2230 			vhdr->h_vlan_TCI = htons(tx_flags >>
2231 						 I40E_TX_FLAGS_VLAN_SHIFT);
2232 		} else {
2233 			tx_flags |= I40E_TX_FLAGS_HW_VLAN;
2234 		}
2235 	}
2236 
2237 out:
2238 	*flags = tx_flags;
2239 	return 0;
2240 }
2241 
2242 /**
2243  * i40e_tso - set up the tso context descriptor
2244  * @skb:      ptr to the skb we're sending
2245  * @hdr_len:  ptr to the size of the packet header
2246  * @cd_type_cmd_tso_mss: Quad Word 1
2247  *
2248  * Returns 0 if no TSO can happen, 1 if tso is going, or error
2249  **/
2250 static int i40e_tso(struct sk_buff *skb, u8 *hdr_len, u64 *cd_type_cmd_tso_mss)
2251 {
2252 	u64 cd_cmd, cd_tso_len, cd_mss;
2253 	union {
2254 		struct iphdr *v4;
2255 		struct ipv6hdr *v6;
2256 		unsigned char *hdr;
2257 	} ip;
2258 	union {
2259 		struct tcphdr *tcp;
2260 		struct udphdr *udp;
2261 		unsigned char *hdr;
2262 	} l4;
2263 	u32 paylen, l4_offset;
2264 	int err;
2265 
2266 	if (skb->ip_summed != CHECKSUM_PARTIAL)
2267 		return 0;
2268 
2269 	if (!skb_is_gso(skb))
2270 		return 0;
2271 
2272 	err = skb_cow_head(skb, 0);
2273 	if (err < 0)
2274 		return err;
2275 
2276 	ip.hdr = skb_network_header(skb);
2277 	l4.hdr = skb_transport_header(skb);
2278 
2279 	/* initialize outer IP header fields */
2280 	if (ip.v4->version == 4) {
2281 		ip.v4->tot_len = 0;
2282 		ip.v4->check = 0;
2283 	} else {
2284 		ip.v6->payload_len = 0;
2285 	}
2286 
2287 	if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
2288 					 SKB_GSO_GRE_CSUM |
2289 					 SKB_GSO_IPXIP4 |
2290 					 SKB_GSO_IPXIP6 |
2291 					 SKB_GSO_UDP_TUNNEL |
2292 					 SKB_GSO_UDP_TUNNEL_CSUM)) {
2293 		if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
2294 		    (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)) {
2295 			l4.udp->len = 0;
2296 
2297 			/* determine offset of outer transport header */
2298 			l4_offset = l4.hdr - skb->data;
2299 
2300 			/* remove payload length from outer checksum */
2301 			paylen = skb->len - l4_offset;
2302 			csum_replace_by_diff(&l4.udp->check, htonl(paylen));
2303 		}
2304 
2305 		/* reset pointers to inner headers */
2306 		ip.hdr = skb_inner_network_header(skb);
2307 		l4.hdr = skb_inner_transport_header(skb);
2308 
2309 		/* initialize inner IP header fields */
2310 		if (ip.v4->version == 4) {
2311 			ip.v4->tot_len = 0;
2312 			ip.v4->check = 0;
2313 		} else {
2314 			ip.v6->payload_len = 0;
2315 		}
2316 	}
2317 
2318 	/* determine offset of inner transport header */
2319 	l4_offset = l4.hdr - skb->data;
2320 
2321 	/* remove payload length from inner checksum */
2322 	paylen = skb->len - l4_offset;
2323 	csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
2324 
2325 	/* compute length of segmentation header */
2326 	*hdr_len = (l4.tcp->doff * 4) + l4_offset;
2327 
2328 	/* find the field values */
2329 	cd_cmd = I40E_TX_CTX_DESC_TSO;
2330 	cd_tso_len = skb->len - *hdr_len;
2331 	cd_mss = skb_shinfo(skb)->gso_size;
2332 	*cd_type_cmd_tso_mss |= (cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
2333 				(cd_tso_len << I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
2334 				(cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
2335 	return 1;
2336 }
2337 
2338 /**
2339  * i40e_tsyn - set up the tsyn context descriptor
2340  * @tx_ring:  ptr to the ring to send
2341  * @skb:      ptr to the skb we're sending
2342  * @tx_flags: the collected send information
2343  * @cd_type_cmd_tso_mss: Quad Word 1
2344  *
2345  * Returns 0 if no Tx timestamp can happen and 1 if the timestamp will happen
2346  **/
2347 static int i40e_tsyn(struct i40e_ring *tx_ring, struct sk_buff *skb,
2348 		     u32 tx_flags, u64 *cd_type_cmd_tso_mss)
2349 {
2350 	struct i40e_pf *pf;
2351 
2352 	if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)))
2353 		return 0;
2354 
2355 	/* Tx timestamps cannot be sampled when doing TSO */
2356 	if (tx_flags & I40E_TX_FLAGS_TSO)
2357 		return 0;
2358 
2359 	/* only timestamp the outbound packet if the user has requested it and
2360 	 * we are not already transmitting a packet to be timestamped
2361 	 */
2362 	pf = i40e_netdev_to_pf(tx_ring->netdev);
2363 	if (!(pf->flags & I40E_FLAG_PTP))
2364 		return 0;
2365 
2366 	if (pf->ptp_tx &&
2367 	    !test_and_set_bit_lock(__I40E_PTP_TX_IN_PROGRESS, &pf->state)) {
2368 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
2369 		pf->ptp_tx_skb = skb_get(skb);
2370 	} else {
2371 		return 0;
2372 	}
2373 
2374 	*cd_type_cmd_tso_mss |= (u64)I40E_TX_CTX_DESC_TSYN <<
2375 				I40E_TXD_CTX_QW1_CMD_SHIFT;
2376 
2377 	return 1;
2378 }
2379 
2380 /**
2381  * i40e_tx_enable_csum - Enable Tx checksum offloads
2382  * @skb: send buffer
2383  * @tx_flags: pointer to Tx flags currently set
2384  * @td_cmd: Tx descriptor command bits to set
2385  * @td_offset: Tx descriptor header offsets to set
2386  * @tx_ring: Tx descriptor ring
2387  * @cd_tunneling: ptr to context desc bits
2388  **/
2389 static int i40e_tx_enable_csum(struct sk_buff *skb, u32 *tx_flags,
2390 			       u32 *td_cmd, u32 *td_offset,
2391 			       struct i40e_ring *tx_ring,
2392 			       u32 *cd_tunneling)
2393 {
2394 	union {
2395 		struct iphdr *v4;
2396 		struct ipv6hdr *v6;
2397 		unsigned char *hdr;
2398 	} ip;
2399 	union {
2400 		struct tcphdr *tcp;
2401 		struct udphdr *udp;
2402 		unsigned char *hdr;
2403 	} l4;
2404 	unsigned char *exthdr;
2405 	u32 offset, cmd = 0;
2406 	__be16 frag_off;
2407 	u8 l4_proto = 0;
2408 
2409 	if (skb->ip_summed != CHECKSUM_PARTIAL)
2410 		return 0;
2411 
2412 	ip.hdr = skb_network_header(skb);
2413 	l4.hdr = skb_transport_header(skb);
2414 
2415 	/* compute outer L2 header size */
2416 	offset = ((ip.hdr - skb->data) / 2) << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
2417 
2418 	if (skb->encapsulation) {
2419 		u32 tunnel = 0;
2420 		/* define outer network header type */
2421 		if (*tx_flags & I40E_TX_FLAGS_IPV4) {
2422 			tunnel |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
2423 				  I40E_TX_CTX_EXT_IP_IPV4 :
2424 				  I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
2425 
2426 			l4_proto = ip.v4->protocol;
2427 		} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
2428 			tunnel |= I40E_TX_CTX_EXT_IP_IPV6;
2429 
2430 			exthdr = ip.hdr + sizeof(*ip.v6);
2431 			l4_proto = ip.v6->nexthdr;
2432 			if (l4.hdr != exthdr)
2433 				ipv6_skip_exthdr(skb, exthdr - skb->data,
2434 						 &l4_proto, &frag_off);
2435 		}
2436 
2437 		/* define outer transport */
2438 		switch (l4_proto) {
2439 		case IPPROTO_UDP:
2440 			tunnel |= I40E_TXD_CTX_UDP_TUNNELING;
2441 			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
2442 			break;
2443 		case IPPROTO_GRE:
2444 			tunnel |= I40E_TXD_CTX_GRE_TUNNELING;
2445 			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
2446 			break;
2447 		case IPPROTO_IPIP:
2448 		case IPPROTO_IPV6:
2449 			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
2450 			l4.hdr = skb_inner_network_header(skb);
2451 			break;
2452 		default:
2453 			if (*tx_flags & I40E_TX_FLAGS_TSO)
2454 				return -1;
2455 
2456 			skb_checksum_help(skb);
2457 			return 0;
2458 		}
2459 
2460 		/* compute outer L3 header size */
2461 		tunnel |= ((l4.hdr - ip.hdr) / 4) <<
2462 			  I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT;
2463 
2464 		/* switch IP header pointer from outer to inner header */
2465 		ip.hdr = skb_inner_network_header(skb);
2466 
2467 		/* compute tunnel header size */
2468 		tunnel |= ((ip.hdr - l4.hdr) / 2) <<
2469 			  I40E_TXD_CTX_QW0_NATLEN_SHIFT;
2470 
2471 		/* indicate if we need to offload outer UDP header */
2472 		if ((*tx_flags & I40E_TX_FLAGS_TSO) &&
2473 		    !(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
2474 		    (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM))
2475 			tunnel |= I40E_TXD_CTX_QW0_L4T_CS_MASK;
2476 
2477 		/* record tunnel offload values */
2478 		*cd_tunneling |= tunnel;
2479 
2480 		/* switch L4 header pointer from outer to inner */
2481 		l4.hdr = skb_inner_transport_header(skb);
2482 		l4_proto = 0;
2483 
2484 		/* reset type as we transition from outer to inner headers */
2485 		*tx_flags &= ~(I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6);
2486 		if (ip.v4->version == 4)
2487 			*tx_flags |= I40E_TX_FLAGS_IPV4;
2488 		if (ip.v6->version == 6)
2489 			*tx_flags |= I40E_TX_FLAGS_IPV6;
2490 	}
2491 
2492 	/* Enable IP checksum offloads */
2493 	if (*tx_flags & I40E_TX_FLAGS_IPV4) {
2494 		l4_proto = ip.v4->protocol;
2495 		/* the stack computes the IP header already, the only time we
2496 		 * need the hardware to recompute it is in the case of TSO.
2497 		 */
2498 		cmd |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
2499 		       I40E_TX_DESC_CMD_IIPT_IPV4_CSUM :
2500 		       I40E_TX_DESC_CMD_IIPT_IPV4;
2501 	} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
2502 		cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
2503 
2504 		exthdr = ip.hdr + sizeof(*ip.v6);
2505 		l4_proto = ip.v6->nexthdr;
2506 		if (l4.hdr != exthdr)
2507 			ipv6_skip_exthdr(skb, exthdr - skb->data,
2508 					 &l4_proto, &frag_off);
2509 	}
2510 
2511 	/* compute inner L3 header size */
2512 	offset |= ((l4.hdr - ip.hdr) / 4) << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
2513 
2514 	/* Enable L4 checksum offloads */
2515 	switch (l4_proto) {
2516 	case IPPROTO_TCP:
2517 		/* enable checksum offloads */
2518 		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
2519 		offset |= l4.tcp->doff << I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
2520 		break;
2521 	case IPPROTO_SCTP:
2522 		/* enable SCTP checksum offload */
2523 		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
2524 		offset |= (sizeof(struct sctphdr) >> 2) <<
2525 			  I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
2526 		break;
2527 	case IPPROTO_UDP:
2528 		/* enable UDP checksum offload */
2529 		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
2530 		offset |= (sizeof(struct udphdr) >> 2) <<
2531 			  I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
2532 		break;
2533 	default:
2534 		if (*tx_flags & I40E_TX_FLAGS_TSO)
2535 			return -1;
2536 		skb_checksum_help(skb);
2537 		return 0;
2538 	}
2539 
2540 	*td_cmd |= cmd;
2541 	*td_offset |= offset;
2542 
2543 	return 1;
2544 }
2545 
2546 /**
2547  * i40e_create_tx_ctx Build the Tx context descriptor
2548  * @tx_ring:  ring to create the descriptor on
2549  * @cd_type_cmd_tso_mss: Quad Word 1
2550  * @cd_tunneling: Quad Word 0 - bits 0-31
2551  * @cd_l2tag2: Quad Word 0 - bits 32-63
2552  **/
2553 static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
2554 			       const u64 cd_type_cmd_tso_mss,
2555 			       const u32 cd_tunneling, const u32 cd_l2tag2)
2556 {
2557 	struct i40e_tx_context_desc *context_desc;
2558 	int i = tx_ring->next_to_use;
2559 
2560 	if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
2561 	    !cd_tunneling && !cd_l2tag2)
2562 		return;
2563 
2564 	/* grab the next descriptor */
2565 	context_desc = I40E_TX_CTXTDESC(tx_ring, i);
2566 
2567 	i++;
2568 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2569 
2570 	/* cpu_to_le32 and assign to struct fields */
2571 	context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
2572 	context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
2573 	context_desc->rsvd = cpu_to_le16(0);
2574 	context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
2575 }
2576 
2577 /**
2578  * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions
2579  * @tx_ring: the ring to be checked
2580  * @size:    the size buffer we want to assure is available
2581  *
2582  * Returns -EBUSY if a stop is needed, else 0
2583  **/
2584 int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
2585 {
2586 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
2587 	/* Memory barrier before checking head and tail */
2588 	smp_mb();
2589 
2590 	/* Check again in a case another CPU has just made room available. */
2591 	if (likely(I40E_DESC_UNUSED(tx_ring) < size))
2592 		return -EBUSY;
2593 
2594 	/* A reprieve! - use start_queue because it doesn't call schedule */
2595 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
2596 	++tx_ring->tx_stats.restart_queue;
2597 	return 0;
2598 }
2599 
2600 /**
2601  * __i40e_chk_linearize - Check if there are more than 8 buffers per packet
2602  * @skb:      send buffer
2603  *
2604  * Note: Our HW can't DMA more than 8 buffers to build a packet on the wire
2605  * and so we need to figure out the cases where we need to linearize the skb.
2606  *
2607  * For TSO we need to count the TSO header and segment payload separately.
2608  * As such we need to check cases where we have 7 fragments or more as we
2609  * can potentially require 9 DMA transactions, 1 for the TSO header, 1 for
2610  * the segment payload in the first descriptor, and another 7 for the
2611  * fragments.
2612  **/
2613 bool __i40e_chk_linearize(struct sk_buff *skb)
2614 {
2615 	const struct skb_frag_struct *frag, *stale;
2616 	int nr_frags, sum;
2617 
2618 	/* no need to check if number of frags is less than 7 */
2619 	nr_frags = skb_shinfo(skb)->nr_frags;
2620 	if (nr_frags < (I40E_MAX_BUFFER_TXD - 1))
2621 		return false;
2622 
2623 	/* We need to walk through the list and validate that each group
2624 	 * of 6 fragments totals at least gso_size.  However we don't need
2625 	 * to perform such validation on the last 6 since the last 6 cannot
2626 	 * inherit any data from a descriptor after them.
2627 	 */
2628 	nr_frags -= I40E_MAX_BUFFER_TXD - 2;
2629 	frag = &skb_shinfo(skb)->frags[0];
2630 
2631 	/* Initialize size to the negative value of gso_size minus 1.  We
2632 	 * use this as the worst case scenerio in which the frag ahead
2633 	 * of us only provides one byte which is why we are limited to 6
2634 	 * descriptors for a single transmit as the header and previous
2635 	 * fragment are already consuming 2 descriptors.
2636 	 */
2637 	sum = 1 - skb_shinfo(skb)->gso_size;
2638 
2639 	/* Add size of frags 0 through 4 to create our initial sum */
2640 	sum += skb_frag_size(frag++);
2641 	sum += skb_frag_size(frag++);
2642 	sum += skb_frag_size(frag++);
2643 	sum += skb_frag_size(frag++);
2644 	sum += skb_frag_size(frag++);
2645 
2646 	/* Walk through fragments adding latest fragment, testing it, and
2647 	 * then removing stale fragments from the sum.
2648 	 */
2649 	stale = &skb_shinfo(skb)->frags[0];
2650 	for (;;) {
2651 		sum += skb_frag_size(frag++);
2652 
2653 		/* if sum is negative we failed to make sufficient progress */
2654 		if (sum < 0)
2655 			return true;
2656 
2657 		/* use pre-decrement to avoid processing last fragment */
2658 		if (!--nr_frags)
2659 			break;
2660 
2661 		sum -= skb_frag_size(stale++);
2662 	}
2663 
2664 	return false;
2665 }
2666 
2667 /**
2668  * i40e_tx_map - Build the Tx descriptor
2669  * @tx_ring:  ring to send buffer on
2670  * @skb:      send buffer
2671  * @first:    first buffer info buffer to use
2672  * @tx_flags: collected send information
2673  * @hdr_len:  size of the packet header
2674  * @td_cmd:   the command field in the descriptor
2675  * @td_offset: offset for checksum or crc
2676  **/
2677 #ifdef I40E_FCOE
2678 inline void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
2679 			struct i40e_tx_buffer *first, u32 tx_flags,
2680 			const u8 hdr_len, u32 td_cmd, u32 td_offset)
2681 #else
2682 static inline void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
2683 			       struct i40e_tx_buffer *first, u32 tx_flags,
2684 			       const u8 hdr_len, u32 td_cmd, u32 td_offset)
2685 #endif
2686 {
2687 	unsigned int data_len = skb->data_len;
2688 	unsigned int size = skb_headlen(skb);
2689 	struct skb_frag_struct *frag;
2690 	struct i40e_tx_buffer *tx_bi;
2691 	struct i40e_tx_desc *tx_desc;
2692 	u16 i = tx_ring->next_to_use;
2693 	u32 td_tag = 0;
2694 	dma_addr_t dma;
2695 	u16 gso_segs;
2696 	u16 desc_count = 0;
2697 	bool tail_bump = true;
2698 	bool do_rs = false;
2699 
2700 	if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
2701 		td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
2702 		td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
2703 			 I40E_TX_FLAGS_VLAN_SHIFT;
2704 	}
2705 
2706 	if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO))
2707 		gso_segs = skb_shinfo(skb)->gso_segs;
2708 	else
2709 		gso_segs = 1;
2710 
2711 	/* multiply data chunks by size of headers */
2712 	first->bytecount = skb->len - hdr_len + (gso_segs * hdr_len);
2713 	first->gso_segs = gso_segs;
2714 	first->skb = skb;
2715 	first->tx_flags = tx_flags;
2716 
2717 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
2718 
2719 	tx_desc = I40E_TX_DESC(tx_ring, i);
2720 	tx_bi = first;
2721 
2722 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
2723 		unsigned int max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
2724 
2725 		if (dma_mapping_error(tx_ring->dev, dma))
2726 			goto dma_error;
2727 
2728 		/* record length, and DMA address */
2729 		dma_unmap_len_set(tx_bi, len, size);
2730 		dma_unmap_addr_set(tx_bi, dma, dma);
2731 
2732 		/* align size to end of page */
2733 		max_data += -dma & (I40E_MAX_READ_REQ_SIZE - 1);
2734 		tx_desc->buffer_addr = cpu_to_le64(dma);
2735 
2736 		while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
2737 			tx_desc->cmd_type_offset_bsz =
2738 				build_ctob(td_cmd, td_offset,
2739 					   max_data, td_tag);
2740 
2741 			tx_desc++;
2742 			i++;
2743 			desc_count++;
2744 
2745 			if (i == tx_ring->count) {
2746 				tx_desc = I40E_TX_DESC(tx_ring, 0);
2747 				i = 0;
2748 			}
2749 
2750 			dma += max_data;
2751 			size -= max_data;
2752 
2753 			max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
2754 			tx_desc->buffer_addr = cpu_to_le64(dma);
2755 		}
2756 
2757 		if (likely(!data_len))
2758 			break;
2759 
2760 		tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
2761 							  size, td_tag);
2762 
2763 		tx_desc++;
2764 		i++;
2765 		desc_count++;
2766 
2767 		if (i == tx_ring->count) {
2768 			tx_desc = I40E_TX_DESC(tx_ring, 0);
2769 			i = 0;
2770 		}
2771 
2772 		size = skb_frag_size(frag);
2773 		data_len -= size;
2774 
2775 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
2776 				       DMA_TO_DEVICE);
2777 
2778 		tx_bi = &tx_ring->tx_bi[i];
2779 	}
2780 
2781 	/* set next_to_watch value indicating a packet is present */
2782 	first->next_to_watch = tx_desc;
2783 
2784 	i++;
2785 	if (i == tx_ring->count)
2786 		i = 0;
2787 
2788 	tx_ring->next_to_use = i;
2789 
2790 	netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev,
2791 						 tx_ring->queue_index),
2792 						 first->bytecount);
2793 	i40e_maybe_stop_tx(tx_ring, DESC_NEEDED);
2794 
2795 	/* Algorithm to optimize tail and RS bit setting:
2796 	 * if xmit_more is supported
2797 	 *	if xmit_more is true
2798 	 *		do not update tail and do not mark RS bit.
2799 	 *	if xmit_more is false and last xmit_more was false
2800 	 *		if every packet spanned less than 4 desc
2801 	 *			then set RS bit on 4th packet and update tail
2802 	 *			on every packet
2803 	 *		else
2804 	 *			update tail and set RS bit on every packet.
2805 	 *	if xmit_more is false and last_xmit_more was true
2806 	 *		update tail and set RS bit.
2807 	 *
2808 	 * Optimization: wmb to be issued only in case of tail update.
2809 	 * Also optimize the Descriptor WB path for RS bit with the same
2810 	 * algorithm.
2811 	 *
2812 	 * Note: If there are less than 4 packets
2813 	 * pending and interrupts were disabled the service task will
2814 	 * trigger a force WB.
2815 	 */
2816 	if (skb->xmit_more  &&
2817 	    !netif_xmit_stopped(netdev_get_tx_queue(tx_ring->netdev,
2818 						    tx_ring->queue_index))) {
2819 		tx_ring->flags |= I40E_TXR_FLAGS_LAST_XMIT_MORE_SET;
2820 		tail_bump = false;
2821 	} else if (!skb->xmit_more &&
2822 		   !netif_xmit_stopped(netdev_get_tx_queue(tx_ring->netdev,
2823 						       tx_ring->queue_index)) &&
2824 		   (!(tx_ring->flags & I40E_TXR_FLAGS_LAST_XMIT_MORE_SET)) &&
2825 		   (tx_ring->packet_stride < WB_STRIDE) &&
2826 		   (desc_count < WB_STRIDE)) {
2827 		tx_ring->packet_stride++;
2828 	} else {
2829 		tx_ring->packet_stride = 0;
2830 		tx_ring->flags &= ~I40E_TXR_FLAGS_LAST_XMIT_MORE_SET;
2831 		do_rs = true;
2832 	}
2833 	if (do_rs)
2834 		tx_ring->packet_stride = 0;
2835 
2836 	tx_desc->cmd_type_offset_bsz =
2837 			build_ctob(td_cmd, td_offset, size, td_tag) |
2838 			cpu_to_le64((u64)(do_rs ? I40E_TXD_CMD :
2839 						  I40E_TX_DESC_CMD_EOP) <<
2840 						  I40E_TXD_QW1_CMD_SHIFT);
2841 
2842 	/* notify HW of packet */
2843 	if (!tail_bump)
2844 		prefetchw(tx_desc + 1);
2845 
2846 	if (tail_bump) {
2847 		/* Force memory writes to complete before letting h/w
2848 		 * know there are new descriptors to fetch.  (Only
2849 		 * applicable for weak-ordered memory model archs,
2850 		 * such as IA-64).
2851 		 */
2852 		wmb();
2853 		writel(i, tx_ring->tail);
2854 	}
2855 
2856 	return;
2857 
2858 dma_error:
2859 	dev_info(tx_ring->dev, "TX DMA map failed\n");
2860 
2861 	/* clear dma mappings for failed tx_bi map */
2862 	for (;;) {
2863 		tx_bi = &tx_ring->tx_bi[i];
2864 		i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
2865 		if (tx_bi == first)
2866 			break;
2867 		if (i == 0)
2868 			i = tx_ring->count;
2869 		i--;
2870 	}
2871 
2872 	tx_ring->next_to_use = i;
2873 }
2874 
2875 /**
2876  * i40e_xmit_frame_ring - Sends buffer on Tx ring
2877  * @skb:     send buffer
2878  * @tx_ring: ring to send buffer on
2879  *
2880  * Returns NETDEV_TX_OK if sent, else an error code
2881  **/
2882 static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
2883 					struct i40e_ring *tx_ring)
2884 {
2885 	u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
2886 	u32 cd_tunneling = 0, cd_l2tag2 = 0;
2887 	struct i40e_tx_buffer *first;
2888 	u32 td_offset = 0;
2889 	u32 tx_flags = 0;
2890 	__be16 protocol;
2891 	u32 td_cmd = 0;
2892 	u8 hdr_len = 0;
2893 	int tso, count;
2894 	int tsyn;
2895 
2896 	/* prefetch the data, we'll need it later */
2897 	prefetch(skb->data);
2898 
2899 	count = i40e_xmit_descriptor_count(skb);
2900 	if (i40e_chk_linearize(skb, count)) {
2901 		if (__skb_linearize(skb))
2902 			goto out_drop;
2903 		count = i40e_txd_use_count(skb->len);
2904 		tx_ring->tx_stats.tx_linearize++;
2905 	}
2906 
2907 	/* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
2908 	 *       + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
2909 	 *       + 4 desc gap to avoid the cache line where head is,
2910 	 *       + 1 desc for context descriptor,
2911 	 * otherwise try next time
2912 	 */
2913 	if (i40e_maybe_stop_tx(tx_ring, count + 4 + 1)) {
2914 		tx_ring->tx_stats.tx_busy++;
2915 		return NETDEV_TX_BUSY;
2916 	}
2917 
2918 	/* prepare the xmit flags */
2919 	if (i40e_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
2920 		goto out_drop;
2921 
2922 	/* obtain protocol of skb */
2923 	protocol = vlan_get_protocol(skb);
2924 
2925 	/* record the location of the first descriptor for this packet */
2926 	first = &tx_ring->tx_bi[tx_ring->next_to_use];
2927 
2928 	/* setup IPv4/IPv6 offloads */
2929 	if (protocol == htons(ETH_P_IP))
2930 		tx_flags |= I40E_TX_FLAGS_IPV4;
2931 	else if (protocol == htons(ETH_P_IPV6))
2932 		tx_flags |= I40E_TX_FLAGS_IPV6;
2933 
2934 	tso = i40e_tso(skb, &hdr_len, &cd_type_cmd_tso_mss);
2935 
2936 	if (tso < 0)
2937 		goto out_drop;
2938 	else if (tso)
2939 		tx_flags |= I40E_TX_FLAGS_TSO;
2940 
2941 	/* Always offload the checksum, since it's in the data descriptor */
2942 	tso = i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset,
2943 				  tx_ring, &cd_tunneling);
2944 	if (tso < 0)
2945 		goto out_drop;
2946 
2947 	tsyn = i40e_tsyn(tx_ring, skb, tx_flags, &cd_type_cmd_tso_mss);
2948 
2949 	if (tsyn)
2950 		tx_flags |= I40E_TX_FLAGS_TSYN;
2951 
2952 	skb_tx_timestamp(skb);
2953 
2954 	/* always enable CRC insertion offload */
2955 	td_cmd |= I40E_TX_DESC_CMD_ICRC;
2956 
2957 	i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
2958 			   cd_tunneling, cd_l2tag2);
2959 
2960 	/* Add Flow Director ATR if it's enabled.
2961 	 *
2962 	 * NOTE: this must always be directly before the data descriptor.
2963 	 */
2964 	i40e_atr(tx_ring, skb, tx_flags);
2965 
2966 	i40e_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
2967 		    td_cmd, td_offset);
2968 
2969 	return NETDEV_TX_OK;
2970 
2971 out_drop:
2972 	dev_kfree_skb_any(skb);
2973 	return NETDEV_TX_OK;
2974 }
2975 
2976 /**
2977  * i40e_lan_xmit_frame - Selects the correct VSI and Tx queue to send buffer
2978  * @skb:    send buffer
2979  * @netdev: network interface device structure
2980  *
2981  * Returns NETDEV_TX_OK if sent, else an error code
2982  **/
2983 netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2984 {
2985 	struct i40e_netdev_priv *np = netdev_priv(netdev);
2986 	struct i40e_vsi *vsi = np->vsi;
2987 	struct i40e_ring *tx_ring = vsi->tx_rings[skb->queue_mapping];
2988 
2989 	/* hardware can't handle really short frames, hardware padding works
2990 	 * beyond this point
2991 	 */
2992 	if (skb_put_padto(skb, I40E_MIN_TX_LEN))
2993 		return NETDEV_TX_OK;
2994 
2995 	return i40e_xmit_frame_ring(skb, tx_ring);
2996 }
2997