1 /*******************************************************************************
2  *
3  * Intel Ethernet Controller XL710 Family Linux Driver
4  * Copyright(c) 2013 - 2016 Intel Corporation.
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms and conditions of the GNU General Public License,
8  * version 2, as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope it will be useful, but WITHOUT
11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13  * more details.
14  *
15  * You should have received a copy of the GNU General Public License along
16  * with this program.  If not, see <http://www.gnu.org/licenses/>.
17  *
18  * The full GNU General Public License is included in this distribution in
19  * the file called "COPYING".
20  *
21  * Contact Information:
22  * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
23  * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
24  *
25  ******************************************************************************/
26 
27 #include <linux/prefetch.h>
28 #include <net/busy_poll.h>
29 #include "i40e.h"
30 #include "i40e_prototype.h"
31 
32 static inline __le64 build_ctob(u32 td_cmd, u32 td_offset, unsigned int size,
33 				u32 td_tag)
34 {
35 	return cpu_to_le64(I40E_TX_DESC_DTYPE_DATA |
36 			   ((u64)td_cmd  << I40E_TXD_QW1_CMD_SHIFT) |
37 			   ((u64)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) |
38 			   ((u64)size  << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) |
39 			   ((u64)td_tag  << I40E_TXD_QW1_L2TAG1_SHIFT));
40 }
41 
42 #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
43 #define I40E_FD_CLEAN_DELAY 10
44 /**
45  * i40e_program_fdir_filter - Program a Flow Director filter
46  * @fdir_data: Packet data that will be filter parameters
47  * @raw_packet: the pre-allocated packet buffer for FDir
48  * @pf: The PF pointer
49  * @add: True for add/update, False for remove
50  **/
51 int i40e_program_fdir_filter(struct i40e_fdir_filter *fdir_data, u8 *raw_packet,
52 			     struct i40e_pf *pf, bool add)
53 {
54 	struct i40e_filter_program_desc *fdir_desc;
55 	struct i40e_tx_buffer *tx_buf, *first;
56 	struct i40e_tx_desc *tx_desc;
57 	struct i40e_ring *tx_ring;
58 	unsigned int fpt, dcc;
59 	struct i40e_vsi *vsi;
60 	struct device *dev;
61 	dma_addr_t dma;
62 	u32 td_cmd = 0;
63 	u16 delay = 0;
64 	u16 i;
65 
66 	/* find existing FDIR VSI */
67 	vsi = NULL;
68 	for (i = 0; i < pf->num_alloc_vsi; i++)
69 		if (pf->vsi[i] && pf->vsi[i]->type == I40E_VSI_FDIR)
70 			vsi = pf->vsi[i];
71 	if (!vsi)
72 		return -ENOENT;
73 
74 	tx_ring = vsi->tx_rings[0];
75 	dev = tx_ring->dev;
76 
77 	/* we need two descriptors to add/del a filter and we can wait */
78 	do {
79 		if (I40E_DESC_UNUSED(tx_ring) > 1)
80 			break;
81 		msleep_interruptible(1);
82 		delay++;
83 	} while (delay < I40E_FD_CLEAN_DELAY);
84 
85 	if (!(I40E_DESC_UNUSED(tx_ring) > 1))
86 		return -EAGAIN;
87 
88 	dma = dma_map_single(dev, raw_packet,
89 			     I40E_FDIR_MAX_RAW_PACKET_SIZE, DMA_TO_DEVICE);
90 	if (dma_mapping_error(dev, dma))
91 		goto dma_fail;
92 
93 	/* grab the next descriptor */
94 	i = tx_ring->next_to_use;
95 	fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
96 	first = &tx_ring->tx_bi[i];
97 	memset(first, 0, sizeof(struct i40e_tx_buffer));
98 
99 	tx_ring->next_to_use = ((i + 1) < tx_ring->count) ? i + 1 : 0;
100 
101 	fpt = (fdir_data->q_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) &
102 	      I40E_TXD_FLTR_QW0_QINDEX_MASK;
103 
104 	fpt |= (fdir_data->flex_off << I40E_TXD_FLTR_QW0_FLEXOFF_SHIFT) &
105 	       I40E_TXD_FLTR_QW0_FLEXOFF_MASK;
106 
107 	fpt |= (fdir_data->pctype << I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) &
108 	       I40E_TXD_FLTR_QW0_PCTYPE_MASK;
109 
110 	/* Use LAN VSI Id if not programmed by user */
111 	if (fdir_data->dest_vsi == 0)
112 		fpt |= (pf->vsi[pf->lan_vsi]->id) <<
113 		       I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT;
114 	else
115 		fpt |= ((u32)fdir_data->dest_vsi <<
116 			I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT) &
117 		       I40E_TXD_FLTR_QW0_DEST_VSI_MASK;
118 
119 	dcc = I40E_TX_DESC_DTYPE_FILTER_PROG;
120 
121 	if (add)
122 		dcc |= I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
123 		       I40E_TXD_FLTR_QW1_PCMD_SHIFT;
124 	else
125 		dcc |= I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
126 		       I40E_TXD_FLTR_QW1_PCMD_SHIFT;
127 
128 	dcc |= (fdir_data->dest_ctl << I40E_TXD_FLTR_QW1_DEST_SHIFT) &
129 	       I40E_TXD_FLTR_QW1_DEST_MASK;
130 
131 	dcc |= (fdir_data->fd_status << I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT) &
132 	       I40E_TXD_FLTR_QW1_FD_STATUS_MASK;
133 
134 	if (fdir_data->cnt_index != 0) {
135 		dcc |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
136 		dcc |= ((u32)fdir_data->cnt_index <<
137 			I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
138 			I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
139 	}
140 
141 	fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(fpt);
142 	fdir_desc->rsvd = cpu_to_le32(0);
143 	fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dcc);
144 	fdir_desc->fd_id = cpu_to_le32(fdir_data->fd_id);
145 
146 	/* Now program a dummy descriptor */
147 	i = tx_ring->next_to_use;
148 	tx_desc = I40E_TX_DESC(tx_ring, i);
149 	tx_buf = &tx_ring->tx_bi[i];
150 
151 	tx_ring->next_to_use = ((i + 1) < tx_ring->count) ? i + 1 : 0;
152 
153 	memset(tx_buf, 0, sizeof(struct i40e_tx_buffer));
154 
155 	/* record length, and DMA address */
156 	dma_unmap_len_set(tx_buf, len, I40E_FDIR_MAX_RAW_PACKET_SIZE);
157 	dma_unmap_addr_set(tx_buf, dma, dma);
158 
159 	tx_desc->buffer_addr = cpu_to_le64(dma);
160 	td_cmd = I40E_TXD_CMD | I40E_TX_DESC_CMD_DUMMY;
161 
162 	tx_buf->tx_flags = I40E_TX_FLAGS_FD_SB;
163 	tx_buf->raw_buf = (void *)raw_packet;
164 
165 	tx_desc->cmd_type_offset_bsz =
166 		build_ctob(td_cmd, 0, I40E_FDIR_MAX_RAW_PACKET_SIZE, 0);
167 
168 	/* Force memory writes to complete before letting h/w
169 	 * know there are new descriptors to fetch.
170 	 */
171 	wmb();
172 
173 	/* Mark the data descriptor to be watched */
174 	first->next_to_watch = tx_desc;
175 
176 	writel(tx_ring->next_to_use, tx_ring->tail);
177 	return 0;
178 
179 dma_fail:
180 	return -1;
181 }
182 
183 #define IP_HEADER_OFFSET 14
184 #define I40E_UDPIP_DUMMY_PACKET_LEN 42
185 /**
186  * i40e_add_del_fdir_udpv4 - Add/Remove UDPv4 filters
187  * @vsi: pointer to the targeted VSI
188  * @fd_data: the flow director data required for the FDir descriptor
189  * @add: true adds a filter, false removes it
190  *
191  * Returns 0 if the filters were successfully added or removed
192  **/
193 static int i40e_add_del_fdir_udpv4(struct i40e_vsi *vsi,
194 				   struct i40e_fdir_filter *fd_data,
195 				   bool add)
196 {
197 	struct i40e_pf *pf = vsi->back;
198 	struct udphdr *udp;
199 	struct iphdr *ip;
200 	bool err = false;
201 	u8 *raw_packet;
202 	int ret;
203 	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
204 		0x45, 0, 0, 0x1c, 0, 0, 0x40, 0, 0x40, 0x11, 0, 0, 0, 0, 0, 0,
205 		0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
206 
207 	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
208 	if (!raw_packet)
209 		return -ENOMEM;
210 	memcpy(raw_packet, packet, I40E_UDPIP_DUMMY_PACKET_LEN);
211 
212 	ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
213 	udp = (struct udphdr *)(raw_packet + IP_HEADER_OFFSET
214 	      + sizeof(struct iphdr));
215 
216 	ip->daddr = fd_data->dst_ip[0];
217 	udp->dest = fd_data->dst_port;
218 	ip->saddr = fd_data->src_ip[0];
219 	udp->source = fd_data->src_port;
220 
221 	fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_UDP;
222 	ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
223 	if (ret) {
224 		dev_info(&pf->pdev->dev,
225 			 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
226 			 fd_data->pctype, fd_data->fd_id, ret);
227 		err = true;
228 	} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
229 		if (add)
230 			dev_info(&pf->pdev->dev,
231 				 "Filter OK for PCTYPE %d loc = %d\n",
232 				 fd_data->pctype, fd_data->fd_id);
233 		else
234 			dev_info(&pf->pdev->dev,
235 				 "Filter deleted for PCTYPE %d loc = %d\n",
236 				 fd_data->pctype, fd_data->fd_id);
237 	}
238 	if (err)
239 		kfree(raw_packet);
240 
241 	return err ? -EOPNOTSUPP : 0;
242 }
243 
244 #define I40E_TCPIP_DUMMY_PACKET_LEN 54
245 /**
246  * i40e_add_del_fdir_tcpv4 - Add/Remove TCPv4 filters
247  * @vsi: pointer to the targeted VSI
248  * @fd_data: the flow director data required for the FDir descriptor
249  * @add: true adds a filter, false removes it
250  *
251  * Returns 0 if the filters were successfully added or removed
252  **/
253 static int i40e_add_del_fdir_tcpv4(struct i40e_vsi *vsi,
254 				   struct i40e_fdir_filter *fd_data,
255 				   bool add)
256 {
257 	struct i40e_pf *pf = vsi->back;
258 	struct tcphdr *tcp;
259 	struct iphdr *ip;
260 	bool err = false;
261 	u8 *raw_packet;
262 	int ret;
263 	/* Dummy packet */
264 	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
265 		0x45, 0, 0, 0x28, 0, 0, 0x40, 0, 0x40, 0x6, 0, 0, 0, 0, 0, 0,
266 		0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x80, 0x11,
267 		0x0, 0x72, 0, 0, 0, 0};
268 
269 	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
270 	if (!raw_packet)
271 		return -ENOMEM;
272 	memcpy(raw_packet, packet, I40E_TCPIP_DUMMY_PACKET_LEN);
273 
274 	ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
275 	tcp = (struct tcphdr *)(raw_packet + IP_HEADER_OFFSET
276 	      + sizeof(struct iphdr));
277 
278 	ip->daddr = fd_data->dst_ip[0];
279 	tcp->dest = fd_data->dst_port;
280 	ip->saddr = fd_data->src_ip[0];
281 	tcp->source = fd_data->src_port;
282 
283 	if (add) {
284 		pf->fd_tcp_rule++;
285 		if (pf->flags & I40E_FLAG_FD_ATR_ENABLED) {
286 			if (I40E_DEBUG_FD & pf->hw.debug_mask)
287 				dev_info(&pf->pdev->dev, "Forcing ATR off, sideband rules for TCP/IPv4 flow being applied\n");
288 			pf->flags &= ~I40E_FLAG_FD_ATR_ENABLED;
289 		}
290 	} else {
291 		pf->fd_tcp_rule = (pf->fd_tcp_rule > 0) ?
292 				  (pf->fd_tcp_rule - 1) : 0;
293 		if (pf->fd_tcp_rule == 0) {
294 			pf->flags |= I40E_FLAG_FD_ATR_ENABLED;
295 			if (I40E_DEBUG_FD & pf->hw.debug_mask)
296 				dev_info(&pf->pdev->dev, "ATR re-enabled due to no sideband TCP/IPv4 rules\n");
297 		}
298 	}
299 
300 	fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_TCP;
301 	ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
302 
303 	if (ret) {
304 		dev_info(&pf->pdev->dev,
305 			 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
306 			 fd_data->pctype, fd_data->fd_id, ret);
307 		err = true;
308 	} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
309 		if (add)
310 			dev_info(&pf->pdev->dev, "Filter OK for PCTYPE %d loc = %d)\n",
311 				 fd_data->pctype, fd_data->fd_id);
312 		else
313 			dev_info(&pf->pdev->dev,
314 				 "Filter deleted for PCTYPE %d loc = %d\n",
315 				 fd_data->pctype, fd_data->fd_id);
316 	}
317 
318 	if (err)
319 		kfree(raw_packet);
320 
321 	return err ? -EOPNOTSUPP : 0;
322 }
323 
324 /**
325  * i40e_add_del_fdir_sctpv4 - Add/Remove SCTPv4 Flow Director filters for
326  * a specific flow spec
327  * @vsi: pointer to the targeted VSI
328  * @fd_data: the flow director data required for the FDir descriptor
329  * @add: true adds a filter, false removes it
330  *
331  * Returns 0 if the filters were successfully added or removed
332  **/
333 static int i40e_add_del_fdir_sctpv4(struct i40e_vsi *vsi,
334 				    struct i40e_fdir_filter *fd_data,
335 				    bool add)
336 {
337 	return -EOPNOTSUPP;
338 }
339 
340 #define I40E_IP_DUMMY_PACKET_LEN 34
341 /**
342  * i40e_add_del_fdir_ipv4 - Add/Remove IPv4 Flow Director filters for
343  * a specific flow spec
344  * @vsi: pointer to the targeted VSI
345  * @fd_data: the flow director data required for the FDir descriptor
346  * @add: true adds a filter, false removes it
347  *
348  * Returns 0 if the filters were successfully added or removed
349  **/
350 static int i40e_add_del_fdir_ipv4(struct i40e_vsi *vsi,
351 				  struct i40e_fdir_filter *fd_data,
352 				  bool add)
353 {
354 	struct i40e_pf *pf = vsi->back;
355 	struct iphdr *ip;
356 	bool err = false;
357 	u8 *raw_packet;
358 	int ret;
359 	int i;
360 	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
361 		0x45, 0, 0, 0x14, 0, 0, 0x40, 0, 0x40, 0x10, 0, 0, 0, 0, 0, 0,
362 		0, 0, 0, 0};
363 
364 	for (i = I40E_FILTER_PCTYPE_NONF_IPV4_OTHER;
365 	     i <= I40E_FILTER_PCTYPE_FRAG_IPV4;	i++) {
366 		raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
367 		if (!raw_packet)
368 			return -ENOMEM;
369 		memcpy(raw_packet, packet, I40E_IP_DUMMY_PACKET_LEN);
370 		ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
371 
372 		ip->saddr = fd_data->src_ip[0];
373 		ip->daddr = fd_data->dst_ip[0];
374 		ip->protocol = 0;
375 
376 		fd_data->pctype = i;
377 		ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
378 
379 		if (ret) {
380 			dev_info(&pf->pdev->dev,
381 				 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
382 				 fd_data->pctype, fd_data->fd_id, ret);
383 			err = true;
384 		} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
385 			if (add)
386 				dev_info(&pf->pdev->dev,
387 					 "Filter OK for PCTYPE %d loc = %d\n",
388 					 fd_data->pctype, fd_data->fd_id);
389 			else
390 				dev_info(&pf->pdev->dev,
391 					 "Filter deleted for PCTYPE %d loc = %d\n",
392 					 fd_data->pctype, fd_data->fd_id);
393 		}
394 	}
395 
396 	if (err)
397 		kfree(raw_packet);
398 
399 	return err ? -EOPNOTSUPP : 0;
400 }
401 
402 /**
403  * i40e_add_del_fdir - Build raw packets to add/del fdir filter
404  * @vsi: pointer to the targeted VSI
405  * @cmd: command to get or set RX flow classification rules
406  * @add: true adds a filter, false removes it
407  *
408  **/
409 int i40e_add_del_fdir(struct i40e_vsi *vsi,
410 		      struct i40e_fdir_filter *input, bool add)
411 {
412 	struct i40e_pf *pf = vsi->back;
413 	int ret;
414 
415 	switch (input->flow_type & ~FLOW_EXT) {
416 	case TCP_V4_FLOW:
417 		ret = i40e_add_del_fdir_tcpv4(vsi, input, add);
418 		break;
419 	case UDP_V4_FLOW:
420 		ret = i40e_add_del_fdir_udpv4(vsi, input, add);
421 		break;
422 	case SCTP_V4_FLOW:
423 		ret = i40e_add_del_fdir_sctpv4(vsi, input, add);
424 		break;
425 	case IPV4_FLOW:
426 		ret = i40e_add_del_fdir_ipv4(vsi, input, add);
427 		break;
428 	case IP_USER_FLOW:
429 		switch (input->ip4_proto) {
430 		case IPPROTO_TCP:
431 			ret = i40e_add_del_fdir_tcpv4(vsi, input, add);
432 			break;
433 		case IPPROTO_UDP:
434 			ret = i40e_add_del_fdir_udpv4(vsi, input, add);
435 			break;
436 		case IPPROTO_SCTP:
437 			ret = i40e_add_del_fdir_sctpv4(vsi, input, add);
438 			break;
439 		default:
440 			ret = i40e_add_del_fdir_ipv4(vsi, input, add);
441 			break;
442 		}
443 		break;
444 	default:
445 		dev_info(&pf->pdev->dev, "Could not specify spec type %d\n",
446 			 input->flow_type);
447 		ret = -EINVAL;
448 	}
449 
450 	/* The buffer allocated here is freed by the i40e_clean_tx_ring() */
451 	return ret;
452 }
453 
454 /**
455  * i40e_fd_handle_status - check the Programming Status for FD
456  * @rx_ring: the Rx ring for this descriptor
457  * @rx_desc: the Rx descriptor for programming Status, not a packet descriptor.
458  * @prog_id: the id originally used for programming
459  *
460  * This is used to verify if the FD programming or invalidation
461  * requested by SW to the HW is successful or not and take actions accordingly.
462  **/
463 static void i40e_fd_handle_status(struct i40e_ring *rx_ring,
464 				  union i40e_rx_desc *rx_desc, u8 prog_id)
465 {
466 	struct i40e_pf *pf = rx_ring->vsi->back;
467 	struct pci_dev *pdev = pf->pdev;
468 	u32 fcnt_prog, fcnt_avail;
469 	u32 error;
470 	u64 qw;
471 
472 	qw = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
473 	error = (qw & I40E_RX_PROG_STATUS_DESC_QW1_ERROR_MASK) >>
474 		I40E_RX_PROG_STATUS_DESC_QW1_ERROR_SHIFT;
475 
476 	if (error == BIT(I40E_RX_PROG_STATUS_DESC_FD_TBL_FULL_SHIFT)) {
477 		pf->fd_inv = le32_to_cpu(rx_desc->wb.qword0.hi_dword.fd_id);
478 		if ((rx_desc->wb.qword0.hi_dword.fd_id != 0) ||
479 		    (I40E_DEBUG_FD & pf->hw.debug_mask))
480 			dev_warn(&pdev->dev, "ntuple filter loc = %d, could not be added\n",
481 				 pf->fd_inv);
482 
483 		/* Check if the programming error is for ATR.
484 		 * If so, auto disable ATR and set a state for
485 		 * flush in progress. Next time we come here if flush is in
486 		 * progress do nothing, once flush is complete the state will
487 		 * be cleared.
488 		 */
489 		if (test_bit(__I40E_FD_FLUSH_REQUESTED, &pf->state))
490 			return;
491 
492 		pf->fd_add_err++;
493 		/* store the current atr filter count */
494 		pf->fd_atr_cnt = i40e_get_current_atr_cnt(pf);
495 
496 		if ((rx_desc->wb.qword0.hi_dword.fd_id == 0) &&
497 		    (pf->auto_disable_flags & I40E_FLAG_FD_SB_ENABLED)) {
498 			pf->auto_disable_flags |= I40E_FLAG_FD_ATR_ENABLED;
499 			set_bit(__I40E_FD_FLUSH_REQUESTED, &pf->state);
500 		}
501 
502 		/* filter programming failed most likely due to table full */
503 		fcnt_prog = i40e_get_global_fd_count(pf);
504 		fcnt_avail = pf->fdir_pf_filter_count;
505 		/* If ATR is running fcnt_prog can quickly change,
506 		 * if we are very close to full, it makes sense to disable
507 		 * FD ATR/SB and then re-enable it when there is room.
508 		 */
509 		if (fcnt_prog >= (fcnt_avail - I40E_FDIR_BUFFER_FULL_MARGIN)) {
510 			if ((pf->flags & I40E_FLAG_FD_SB_ENABLED) &&
511 			    !(pf->auto_disable_flags &
512 				     I40E_FLAG_FD_SB_ENABLED)) {
513 				if (I40E_DEBUG_FD & pf->hw.debug_mask)
514 					dev_warn(&pdev->dev, "FD filter space full, new ntuple rules will not be added\n");
515 				pf->auto_disable_flags |=
516 							I40E_FLAG_FD_SB_ENABLED;
517 			}
518 		}
519 	} else if (error == BIT(I40E_RX_PROG_STATUS_DESC_NO_FD_ENTRY_SHIFT)) {
520 		if (I40E_DEBUG_FD & pf->hw.debug_mask)
521 			dev_info(&pdev->dev, "ntuple filter fd_id = %d, could not be removed\n",
522 				 rx_desc->wb.qword0.hi_dword.fd_id);
523 	}
524 }
525 
526 /**
527  * i40e_unmap_and_free_tx_resource - Release a Tx buffer
528  * @ring:      the ring that owns the buffer
529  * @tx_buffer: the buffer to free
530  **/
531 static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring,
532 					    struct i40e_tx_buffer *tx_buffer)
533 {
534 	if (tx_buffer->skb) {
535 		dev_kfree_skb_any(tx_buffer->skb);
536 		if (dma_unmap_len(tx_buffer, len))
537 			dma_unmap_single(ring->dev,
538 					 dma_unmap_addr(tx_buffer, dma),
539 					 dma_unmap_len(tx_buffer, len),
540 					 DMA_TO_DEVICE);
541 	} else if (dma_unmap_len(tx_buffer, len)) {
542 		dma_unmap_page(ring->dev,
543 			       dma_unmap_addr(tx_buffer, dma),
544 			       dma_unmap_len(tx_buffer, len),
545 			       DMA_TO_DEVICE);
546 	}
547 
548 	if (tx_buffer->tx_flags & I40E_TX_FLAGS_FD_SB)
549 		kfree(tx_buffer->raw_buf);
550 
551 	tx_buffer->next_to_watch = NULL;
552 	tx_buffer->skb = NULL;
553 	dma_unmap_len_set(tx_buffer, len, 0);
554 	/* tx_buffer must be completely set up in the transmit path */
555 }
556 
557 /**
558  * i40e_clean_tx_ring - Free any empty Tx buffers
559  * @tx_ring: ring to be cleaned
560  **/
561 void i40e_clean_tx_ring(struct i40e_ring *tx_ring)
562 {
563 	unsigned long bi_size;
564 	u16 i;
565 
566 	/* ring already cleared, nothing to do */
567 	if (!tx_ring->tx_bi)
568 		return;
569 
570 	/* Free all the Tx ring sk_buffs */
571 	for (i = 0; i < tx_ring->count; i++)
572 		i40e_unmap_and_free_tx_resource(tx_ring, &tx_ring->tx_bi[i]);
573 
574 	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
575 	memset(tx_ring->tx_bi, 0, bi_size);
576 
577 	/* Zero out the descriptor ring */
578 	memset(tx_ring->desc, 0, tx_ring->size);
579 
580 	tx_ring->next_to_use = 0;
581 	tx_ring->next_to_clean = 0;
582 
583 	if (!tx_ring->netdev)
584 		return;
585 
586 	/* cleanup Tx queue statistics */
587 	netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
588 						  tx_ring->queue_index));
589 }
590 
591 /**
592  * i40e_free_tx_resources - Free Tx resources per queue
593  * @tx_ring: Tx descriptor ring for a specific queue
594  *
595  * Free all transmit software resources
596  **/
597 void i40e_free_tx_resources(struct i40e_ring *tx_ring)
598 {
599 	i40e_clean_tx_ring(tx_ring);
600 	kfree(tx_ring->tx_bi);
601 	tx_ring->tx_bi = NULL;
602 
603 	if (tx_ring->desc) {
604 		dma_free_coherent(tx_ring->dev, tx_ring->size,
605 				  tx_ring->desc, tx_ring->dma);
606 		tx_ring->desc = NULL;
607 	}
608 }
609 
610 /**
611  * i40e_get_tx_pending - how many tx descriptors not processed
612  * @tx_ring: the ring of descriptors
613  * @in_sw: is tx_pending being checked in SW or HW
614  *
615  * Since there is no access to the ring head register
616  * in XL710, we need to use our local copies
617  **/
618 u32 i40e_get_tx_pending(struct i40e_ring *ring, bool in_sw)
619 {
620 	u32 head, tail;
621 
622 	if (!in_sw)
623 		head = i40e_get_head(ring);
624 	else
625 		head = ring->next_to_clean;
626 	tail = readl(ring->tail);
627 
628 	if (head != tail)
629 		return (head < tail) ?
630 			tail - head : (tail + ring->count - head);
631 
632 	return 0;
633 }
634 
635 #define WB_STRIDE 0x3
636 
637 /**
638  * i40e_clean_tx_irq - Reclaim resources after transmit completes
639  * @vsi: the VSI we care about
640  * @tx_ring: Tx ring to clean
641  * @napi_budget: Used to determine if we are in netpoll
642  *
643  * Returns true if there's any budget left (e.g. the clean is finished)
644  **/
645 static bool i40e_clean_tx_irq(struct i40e_vsi *vsi,
646 			      struct i40e_ring *tx_ring, int napi_budget)
647 {
648 	u16 i = tx_ring->next_to_clean;
649 	struct i40e_tx_buffer *tx_buf;
650 	struct i40e_tx_desc *tx_head;
651 	struct i40e_tx_desc *tx_desc;
652 	unsigned int total_bytes = 0, total_packets = 0;
653 	unsigned int budget = vsi->work_limit;
654 
655 	tx_buf = &tx_ring->tx_bi[i];
656 	tx_desc = I40E_TX_DESC(tx_ring, i);
657 	i -= tx_ring->count;
658 
659 	tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
660 
661 	do {
662 		struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
663 
664 		/* if next_to_watch is not set then there is no work pending */
665 		if (!eop_desc)
666 			break;
667 
668 		/* prevent any other reads prior to eop_desc */
669 		read_barrier_depends();
670 
671 		/* we have caught up to head, no work left to do */
672 		if (tx_head == tx_desc)
673 			break;
674 
675 		/* clear next_to_watch to prevent false hangs */
676 		tx_buf->next_to_watch = NULL;
677 
678 		/* update the statistics for this packet */
679 		total_bytes += tx_buf->bytecount;
680 		total_packets += tx_buf->gso_segs;
681 
682 		/* free the skb */
683 		napi_consume_skb(tx_buf->skb, napi_budget);
684 
685 		/* unmap skb header data */
686 		dma_unmap_single(tx_ring->dev,
687 				 dma_unmap_addr(tx_buf, dma),
688 				 dma_unmap_len(tx_buf, len),
689 				 DMA_TO_DEVICE);
690 
691 		/* clear tx_buffer data */
692 		tx_buf->skb = NULL;
693 		dma_unmap_len_set(tx_buf, len, 0);
694 
695 		/* unmap remaining buffers */
696 		while (tx_desc != eop_desc) {
697 
698 			tx_buf++;
699 			tx_desc++;
700 			i++;
701 			if (unlikely(!i)) {
702 				i -= tx_ring->count;
703 				tx_buf = tx_ring->tx_bi;
704 				tx_desc = I40E_TX_DESC(tx_ring, 0);
705 			}
706 
707 			/* unmap any remaining paged data */
708 			if (dma_unmap_len(tx_buf, len)) {
709 				dma_unmap_page(tx_ring->dev,
710 					       dma_unmap_addr(tx_buf, dma),
711 					       dma_unmap_len(tx_buf, len),
712 					       DMA_TO_DEVICE);
713 				dma_unmap_len_set(tx_buf, len, 0);
714 			}
715 		}
716 
717 		/* move us one more past the eop_desc for start of next pkt */
718 		tx_buf++;
719 		tx_desc++;
720 		i++;
721 		if (unlikely(!i)) {
722 			i -= tx_ring->count;
723 			tx_buf = tx_ring->tx_bi;
724 			tx_desc = I40E_TX_DESC(tx_ring, 0);
725 		}
726 
727 		prefetch(tx_desc);
728 
729 		/* update budget accounting */
730 		budget--;
731 	} while (likely(budget));
732 
733 	i += tx_ring->count;
734 	tx_ring->next_to_clean = i;
735 	u64_stats_update_begin(&tx_ring->syncp);
736 	tx_ring->stats.bytes += total_bytes;
737 	tx_ring->stats.packets += total_packets;
738 	u64_stats_update_end(&tx_ring->syncp);
739 	tx_ring->q_vector->tx.total_bytes += total_bytes;
740 	tx_ring->q_vector->tx.total_packets += total_packets;
741 
742 	if (tx_ring->flags & I40E_TXR_FLAGS_WB_ON_ITR) {
743 		unsigned int j = 0;
744 
745 		/* check to see if there are < 4 descriptors
746 		 * waiting to be written back, then kick the hardware to force
747 		 * them to be written back in case we stay in NAPI.
748 		 * In this mode on X722 we do not enable Interrupt.
749 		 */
750 		j = i40e_get_tx_pending(tx_ring, false);
751 
752 		if (budget &&
753 		    ((j / (WB_STRIDE + 1)) == 0) && (j != 0) &&
754 		    !test_bit(__I40E_DOWN, &vsi->state) &&
755 		    (I40E_DESC_UNUSED(tx_ring) != tx_ring->count))
756 			tx_ring->arm_wb = true;
757 	}
758 
759 	netdev_tx_completed_queue(netdev_get_tx_queue(tx_ring->netdev,
760 						      tx_ring->queue_index),
761 				  total_packets, total_bytes);
762 
763 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
764 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
765 		     (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
766 		/* Make sure that anybody stopping the queue after this
767 		 * sees the new next_to_clean.
768 		 */
769 		smp_mb();
770 		if (__netif_subqueue_stopped(tx_ring->netdev,
771 					     tx_ring->queue_index) &&
772 		   !test_bit(__I40E_DOWN, &vsi->state)) {
773 			netif_wake_subqueue(tx_ring->netdev,
774 					    tx_ring->queue_index);
775 			++tx_ring->tx_stats.restart_queue;
776 		}
777 	}
778 
779 	return !!budget;
780 }
781 
782 /**
783  * i40e_enable_wb_on_itr - Arm hardware to do a wb, interrupts are not enabled
784  * @vsi: the VSI we care about
785  * @q_vector: the vector on which to enable writeback
786  *
787  **/
788 static void i40e_enable_wb_on_itr(struct i40e_vsi *vsi,
789 				  struct i40e_q_vector *q_vector)
790 {
791 	u16 flags = q_vector->tx.ring[0].flags;
792 	u32 val;
793 
794 	if (!(flags & I40E_TXR_FLAGS_WB_ON_ITR))
795 		return;
796 
797 	if (q_vector->arm_wb_state)
798 		return;
799 
800 	if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
801 		val = I40E_PFINT_DYN_CTLN_WB_ON_ITR_MASK |
802 		      I40E_PFINT_DYN_CTLN_ITR_INDX_MASK; /* set noitr */
803 
804 		wr32(&vsi->back->hw,
805 		     I40E_PFINT_DYN_CTLN(q_vector->v_idx + vsi->base_vector - 1),
806 		     val);
807 	} else {
808 		val = I40E_PFINT_DYN_CTL0_WB_ON_ITR_MASK |
809 		      I40E_PFINT_DYN_CTL0_ITR_INDX_MASK; /* set noitr */
810 
811 		wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
812 	}
813 	q_vector->arm_wb_state = true;
814 }
815 
816 /**
817  * i40e_force_wb - Issue SW Interrupt so HW does a wb
818  * @vsi: the VSI we care about
819  * @q_vector: the vector  on which to force writeback
820  *
821  **/
822 void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector)
823 {
824 	if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
825 		u32 val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
826 			  I40E_PFINT_DYN_CTLN_ITR_INDX_MASK | /* set noitr */
827 			  I40E_PFINT_DYN_CTLN_SWINT_TRIG_MASK |
828 			  I40E_PFINT_DYN_CTLN_SW_ITR_INDX_ENA_MASK;
829 			  /* allow 00 to be written to the index */
830 
831 		wr32(&vsi->back->hw,
832 		     I40E_PFINT_DYN_CTLN(q_vector->v_idx +
833 					 vsi->base_vector - 1), val);
834 	} else {
835 		u32 val = I40E_PFINT_DYN_CTL0_INTENA_MASK |
836 			  I40E_PFINT_DYN_CTL0_ITR_INDX_MASK | /* set noitr */
837 			  I40E_PFINT_DYN_CTL0_SWINT_TRIG_MASK |
838 			  I40E_PFINT_DYN_CTL0_SW_ITR_INDX_ENA_MASK;
839 			/* allow 00 to be written to the index */
840 
841 		wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
842 	}
843 }
844 
845 /**
846  * i40e_set_new_dynamic_itr - Find new ITR level
847  * @rc: structure containing ring performance data
848  *
849  * Returns true if ITR changed, false if not
850  *
851  * Stores a new ITR value based on packets and byte counts during
852  * the last interrupt.  The advantage of per interrupt computation
853  * is faster updates and more accurate ITR for the current traffic
854  * pattern.  Constants in this function were computed based on
855  * theoretical maximum wire speed and thresholds were set based on
856  * testing data as well as attempting to minimize response time
857  * while increasing bulk throughput.
858  **/
859 static bool i40e_set_new_dynamic_itr(struct i40e_ring_container *rc)
860 {
861 	enum i40e_latency_range new_latency_range = rc->latency_range;
862 	struct i40e_q_vector *qv = rc->ring->q_vector;
863 	u32 new_itr = rc->itr;
864 	int bytes_per_int;
865 	int usecs;
866 
867 	if (rc->total_packets == 0 || !rc->itr)
868 		return false;
869 
870 	/* simple throttlerate management
871 	 *   0-10MB/s   lowest (50000 ints/s)
872 	 *  10-20MB/s   low    (20000 ints/s)
873 	 *  20-1249MB/s bulk   (18000 ints/s)
874 	 *  > 40000 Rx packets per second (8000 ints/s)
875 	 *
876 	 * The math works out because the divisor is in 10^(-6) which
877 	 * turns the bytes/us input value into MB/s values, but
878 	 * make sure to use usecs, as the register values written
879 	 * are in 2 usec increments in the ITR registers, and make sure
880 	 * to use the smoothed values that the countdown timer gives us.
881 	 */
882 	usecs = (rc->itr << 1) * ITR_COUNTDOWN_START;
883 	bytes_per_int = rc->total_bytes / usecs;
884 
885 	switch (new_latency_range) {
886 	case I40E_LOWEST_LATENCY:
887 		if (bytes_per_int > 10)
888 			new_latency_range = I40E_LOW_LATENCY;
889 		break;
890 	case I40E_LOW_LATENCY:
891 		if (bytes_per_int > 20)
892 			new_latency_range = I40E_BULK_LATENCY;
893 		else if (bytes_per_int <= 10)
894 			new_latency_range = I40E_LOWEST_LATENCY;
895 		break;
896 	case I40E_BULK_LATENCY:
897 	case I40E_ULTRA_LATENCY:
898 	default:
899 		if (bytes_per_int <= 20)
900 			new_latency_range = I40E_LOW_LATENCY;
901 		break;
902 	}
903 
904 	/* this is to adjust RX more aggressively when streaming small
905 	 * packets.  The value of 40000 was picked as it is just beyond
906 	 * what the hardware can receive per second if in low latency
907 	 * mode.
908 	 */
909 #define RX_ULTRA_PACKET_RATE 40000
910 
911 	if ((((rc->total_packets * 1000000) / usecs) > RX_ULTRA_PACKET_RATE) &&
912 	    (&qv->rx == rc))
913 		new_latency_range = I40E_ULTRA_LATENCY;
914 
915 	rc->latency_range = new_latency_range;
916 
917 	switch (new_latency_range) {
918 	case I40E_LOWEST_LATENCY:
919 		new_itr = I40E_ITR_50K;
920 		break;
921 	case I40E_LOW_LATENCY:
922 		new_itr = I40E_ITR_20K;
923 		break;
924 	case I40E_BULK_LATENCY:
925 		new_itr = I40E_ITR_18K;
926 		break;
927 	case I40E_ULTRA_LATENCY:
928 		new_itr = I40E_ITR_8K;
929 		break;
930 	default:
931 		break;
932 	}
933 
934 	rc->total_bytes = 0;
935 	rc->total_packets = 0;
936 
937 	if (new_itr != rc->itr) {
938 		rc->itr = new_itr;
939 		return true;
940 	}
941 
942 	return false;
943 }
944 
945 /**
946  * i40e_clean_programming_status - clean the programming status descriptor
947  * @rx_ring: the rx ring that has this descriptor
948  * @rx_desc: the rx descriptor written back by HW
949  *
950  * Flow director should handle FD_FILTER_STATUS to check its filter programming
951  * status being successful or not and take actions accordingly. FCoE should
952  * handle its context/filter programming/invalidation status and take actions.
953  *
954  **/
955 static void i40e_clean_programming_status(struct i40e_ring *rx_ring,
956 					  union i40e_rx_desc *rx_desc)
957 {
958 	u64 qw;
959 	u8 id;
960 
961 	qw = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
962 	id = (qw & I40E_RX_PROG_STATUS_DESC_QW1_PROGID_MASK) >>
963 		  I40E_RX_PROG_STATUS_DESC_QW1_PROGID_SHIFT;
964 
965 	if (id == I40E_RX_PROG_STATUS_DESC_FD_FILTER_STATUS)
966 		i40e_fd_handle_status(rx_ring, rx_desc, id);
967 #ifdef I40E_FCOE
968 	else if ((id == I40E_RX_PROG_STATUS_DESC_FCOE_CTXT_PROG_STATUS) ||
969 		 (id == I40E_RX_PROG_STATUS_DESC_FCOE_CTXT_INVL_STATUS))
970 		i40e_fcoe_handle_status(rx_ring, rx_desc, id);
971 #endif
972 }
973 
974 /**
975  * i40e_setup_tx_descriptors - Allocate the Tx descriptors
976  * @tx_ring: the tx ring to set up
977  *
978  * Return 0 on success, negative on error
979  **/
980 int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring)
981 {
982 	struct device *dev = tx_ring->dev;
983 	int bi_size;
984 
985 	if (!dev)
986 		return -ENOMEM;
987 
988 	/* warn if we are about to overwrite the pointer */
989 	WARN_ON(tx_ring->tx_bi);
990 	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
991 	tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
992 	if (!tx_ring->tx_bi)
993 		goto err;
994 
995 	/* round up to nearest 4K */
996 	tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
997 	/* add u32 for head writeback, align after this takes care of
998 	 * guaranteeing this is at least one cache line in size
999 	 */
1000 	tx_ring->size += sizeof(u32);
1001 	tx_ring->size = ALIGN(tx_ring->size, 4096);
1002 	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
1003 					   &tx_ring->dma, GFP_KERNEL);
1004 	if (!tx_ring->desc) {
1005 		dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
1006 			 tx_ring->size);
1007 		goto err;
1008 	}
1009 
1010 	tx_ring->next_to_use = 0;
1011 	tx_ring->next_to_clean = 0;
1012 	return 0;
1013 
1014 err:
1015 	kfree(tx_ring->tx_bi);
1016 	tx_ring->tx_bi = NULL;
1017 	return -ENOMEM;
1018 }
1019 
1020 /**
1021  * i40e_clean_rx_ring - Free Rx buffers
1022  * @rx_ring: ring to be cleaned
1023  **/
1024 void i40e_clean_rx_ring(struct i40e_ring *rx_ring)
1025 {
1026 	struct device *dev = rx_ring->dev;
1027 	unsigned long bi_size;
1028 	u16 i;
1029 
1030 	/* ring already cleared, nothing to do */
1031 	if (!rx_ring->rx_bi)
1032 		return;
1033 
1034 	/* Free all the Rx ring sk_buffs */
1035 	for (i = 0; i < rx_ring->count; i++) {
1036 		struct i40e_rx_buffer *rx_bi = &rx_ring->rx_bi[i];
1037 
1038 		if (rx_bi->skb) {
1039 			dev_kfree_skb(rx_bi->skb);
1040 			rx_bi->skb = NULL;
1041 		}
1042 		if (!rx_bi->page)
1043 			continue;
1044 
1045 		dma_unmap_page(dev, rx_bi->dma, PAGE_SIZE, DMA_FROM_DEVICE);
1046 		__free_pages(rx_bi->page, 0);
1047 
1048 		rx_bi->page = NULL;
1049 		rx_bi->page_offset = 0;
1050 	}
1051 
1052 	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
1053 	memset(rx_ring->rx_bi, 0, bi_size);
1054 
1055 	/* Zero out the descriptor ring */
1056 	memset(rx_ring->desc, 0, rx_ring->size);
1057 
1058 	rx_ring->next_to_alloc = 0;
1059 	rx_ring->next_to_clean = 0;
1060 	rx_ring->next_to_use = 0;
1061 }
1062 
1063 /**
1064  * i40e_free_rx_resources - Free Rx resources
1065  * @rx_ring: ring to clean the resources from
1066  *
1067  * Free all receive software resources
1068  **/
1069 void i40e_free_rx_resources(struct i40e_ring *rx_ring)
1070 {
1071 	i40e_clean_rx_ring(rx_ring);
1072 	kfree(rx_ring->rx_bi);
1073 	rx_ring->rx_bi = NULL;
1074 
1075 	if (rx_ring->desc) {
1076 		dma_free_coherent(rx_ring->dev, rx_ring->size,
1077 				  rx_ring->desc, rx_ring->dma);
1078 		rx_ring->desc = NULL;
1079 	}
1080 }
1081 
1082 /**
1083  * i40e_setup_rx_descriptors - Allocate Rx descriptors
1084  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
1085  *
1086  * Returns 0 on success, negative on failure
1087  **/
1088 int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring)
1089 {
1090 	struct device *dev = rx_ring->dev;
1091 	int bi_size;
1092 
1093 	/* warn if we are about to overwrite the pointer */
1094 	WARN_ON(rx_ring->rx_bi);
1095 	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
1096 	rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
1097 	if (!rx_ring->rx_bi)
1098 		goto err;
1099 
1100 	u64_stats_init(&rx_ring->syncp);
1101 
1102 	/* Round up to nearest 4K */
1103 	rx_ring->size = rx_ring->count * sizeof(union i40e_32byte_rx_desc);
1104 	rx_ring->size = ALIGN(rx_ring->size, 4096);
1105 	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
1106 					   &rx_ring->dma, GFP_KERNEL);
1107 
1108 	if (!rx_ring->desc) {
1109 		dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
1110 			 rx_ring->size);
1111 		goto err;
1112 	}
1113 
1114 	rx_ring->next_to_alloc = 0;
1115 	rx_ring->next_to_clean = 0;
1116 	rx_ring->next_to_use = 0;
1117 
1118 	return 0;
1119 err:
1120 	kfree(rx_ring->rx_bi);
1121 	rx_ring->rx_bi = NULL;
1122 	return -ENOMEM;
1123 }
1124 
1125 /**
1126  * i40e_release_rx_desc - Store the new tail and head values
1127  * @rx_ring: ring to bump
1128  * @val: new head index
1129  **/
1130 static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
1131 {
1132 	rx_ring->next_to_use = val;
1133 
1134 	/* update next to alloc since we have filled the ring */
1135 	rx_ring->next_to_alloc = val;
1136 
1137 	/* Force memory writes to complete before letting h/w
1138 	 * know there are new descriptors to fetch.  (Only
1139 	 * applicable for weak-ordered memory model archs,
1140 	 * such as IA-64).
1141 	 */
1142 	wmb();
1143 	writel(val, rx_ring->tail);
1144 }
1145 
1146 /**
1147  * i40e_alloc_mapped_page - recycle or make a new page
1148  * @rx_ring: ring to use
1149  * @bi: rx_buffer struct to modify
1150  *
1151  * Returns true if the page was successfully allocated or
1152  * reused.
1153  **/
1154 static bool i40e_alloc_mapped_page(struct i40e_ring *rx_ring,
1155 				   struct i40e_rx_buffer *bi)
1156 {
1157 	struct page *page = bi->page;
1158 	dma_addr_t dma;
1159 
1160 	/* since we are recycling buffers we should seldom need to alloc */
1161 	if (likely(page)) {
1162 		rx_ring->rx_stats.page_reuse_count++;
1163 		return true;
1164 	}
1165 
1166 	/* alloc new page for storage */
1167 	page = dev_alloc_page();
1168 	if (unlikely(!page)) {
1169 		rx_ring->rx_stats.alloc_page_failed++;
1170 		return false;
1171 	}
1172 
1173 	/* map page for use */
1174 	dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
1175 
1176 	/* if mapping failed free memory back to system since
1177 	 * there isn't much point in holding memory we can't use
1178 	 */
1179 	if (dma_mapping_error(rx_ring->dev, dma)) {
1180 		__free_pages(page, 0);
1181 		rx_ring->rx_stats.alloc_page_failed++;
1182 		return false;
1183 	}
1184 
1185 	bi->dma = dma;
1186 	bi->page = page;
1187 	bi->page_offset = 0;
1188 
1189 	return true;
1190 }
1191 
1192 /**
1193  * i40e_receive_skb - Send a completed packet up the stack
1194  * @rx_ring:  rx ring in play
1195  * @skb: packet to send up
1196  * @vlan_tag: vlan tag for packet
1197  **/
1198 static void i40e_receive_skb(struct i40e_ring *rx_ring,
1199 			     struct sk_buff *skb, u16 vlan_tag)
1200 {
1201 	struct i40e_q_vector *q_vector = rx_ring->q_vector;
1202 
1203 	if ((rx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1204 	    (vlan_tag & VLAN_VID_MASK))
1205 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
1206 
1207 	napi_gro_receive(&q_vector->napi, skb);
1208 }
1209 
1210 /**
1211  * i40e_alloc_rx_buffers - Replace used receive buffers
1212  * @rx_ring: ring to place buffers on
1213  * @cleaned_count: number of buffers to replace
1214  *
1215  * Returns false if all allocations were successful, true if any fail
1216  **/
1217 bool i40e_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count)
1218 {
1219 	u16 ntu = rx_ring->next_to_use;
1220 	union i40e_rx_desc *rx_desc;
1221 	struct i40e_rx_buffer *bi;
1222 
1223 	/* do nothing if no valid netdev defined */
1224 	if (!rx_ring->netdev || !cleaned_count)
1225 		return false;
1226 
1227 	rx_desc = I40E_RX_DESC(rx_ring, ntu);
1228 	bi = &rx_ring->rx_bi[ntu];
1229 
1230 	do {
1231 		if (!i40e_alloc_mapped_page(rx_ring, bi))
1232 			goto no_buffers;
1233 
1234 		/* Refresh the desc even if buffer_addrs didn't change
1235 		 * because each write-back erases this info.
1236 		 */
1237 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1238 		rx_desc->read.hdr_addr = 0;
1239 
1240 		rx_desc++;
1241 		bi++;
1242 		ntu++;
1243 		if (unlikely(ntu == rx_ring->count)) {
1244 			rx_desc = I40E_RX_DESC(rx_ring, 0);
1245 			bi = rx_ring->rx_bi;
1246 			ntu = 0;
1247 		}
1248 
1249 		/* clear the status bits for the next_to_use descriptor */
1250 		rx_desc->wb.qword1.status_error_len = 0;
1251 
1252 		cleaned_count--;
1253 	} while (cleaned_count);
1254 
1255 	if (rx_ring->next_to_use != ntu)
1256 		i40e_release_rx_desc(rx_ring, ntu);
1257 
1258 	return false;
1259 
1260 no_buffers:
1261 	if (rx_ring->next_to_use != ntu)
1262 		i40e_release_rx_desc(rx_ring, ntu);
1263 
1264 	/* make sure to come back via polling to try again after
1265 	 * allocation failure
1266 	 */
1267 	return true;
1268 }
1269 
1270 /**
1271  * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
1272  * @vsi: the VSI we care about
1273  * @skb: skb currently being received and modified
1274  * @rx_desc: the receive descriptor
1275  *
1276  * skb->protocol must be set before this function is called
1277  **/
1278 static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
1279 				    struct sk_buff *skb,
1280 				    union i40e_rx_desc *rx_desc)
1281 {
1282 	struct i40e_rx_ptype_decoded decoded;
1283 	u32 rx_error, rx_status;
1284 	bool ipv4, ipv6;
1285 	u8 ptype;
1286 	u64 qword;
1287 
1288 	qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1289 	ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >> I40E_RXD_QW1_PTYPE_SHIFT;
1290 	rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1291 		   I40E_RXD_QW1_ERROR_SHIFT;
1292 	rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1293 		    I40E_RXD_QW1_STATUS_SHIFT;
1294 	decoded = decode_rx_desc_ptype(ptype);
1295 
1296 	skb->ip_summed = CHECKSUM_NONE;
1297 
1298 	skb_checksum_none_assert(skb);
1299 
1300 	/* Rx csum enabled and ip headers found? */
1301 	if (!(vsi->netdev->features & NETIF_F_RXCSUM))
1302 		return;
1303 
1304 	/* did the hardware decode the packet and checksum? */
1305 	if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
1306 		return;
1307 
1308 	/* both known and outer_ip must be set for the below code to work */
1309 	if (!(decoded.known && decoded.outer_ip))
1310 		return;
1311 
1312 	ipv4 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1313 	       (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4);
1314 	ipv6 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1315 	       (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6);
1316 
1317 	if (ipv4 &&
1318 	    (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) |
1319 			 BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT))))
1320 		goto checksum_fail;
1321 
1322 	/* likely incorrect csum if alternate IP extension headers found */
1323 	if (ipv6 &&
1324 	    rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
1325 		/* don't increment checksum err here, non-fatal err */
1326 		return;
1327 
1328 	/* there was some L4 error, count error and punt packet to the stack */
1329 	if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT))
1330 		goto checksum_fail;
1331 
1332 	/* handle packets that were not able to be checksummed due
1333 	 * to arrival speed, in this case the stack can compute
1334 	 * the csum.
1335 	 */
1336 	if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT))
1337 		return;
1338 
1339 	/* If there is an outer header present that might contain a checksum
1340 	 * we need to bump the checksum level by 1 to reflect the fact that
1341 	 * we are indicating we validated the inner checksum.
1342 	 */
1343 	if (decoded.tunnel_type >= I40E_RX_PTYPE_TUNNEL_IP_GRENAT)
1344 		skb->csum_level = 1;
1345 
1346 	/* Only report checksum unnecessary for TCP, UDP, or SCTP */
1347 	switch (decoded.inner_prot) {
1348 	case I40E_RX_PTYPE_INNER_PROT_TCP:
1349 	case I40E_RX_PTYPE_INNER_PROT_UDP:
1350 	case I40E_RX_PTYPE_INNER_PROT_SCTP:
1351 		skb->ip_summed = CHECKSUM_UNNECESSARY;
1352 		/* fall though */
1353 	default:
1354 		break;
1355 	}
1356 
1357 	return;
1358 
1359 checksum_fail:
1360 	vsi->back->hw_csum_rx_error++;
1361 }
1362 
1363 /**
1364  * i40e_ptype_to_htype - get a hash type
1365  * @ptype: the ptype value from the descriptor
1366  *
1367  * Returns a hash type to be used by skb_set_hash
1368  **/
1369 static inline int i40e_ptype_to_htype(u8 ptype)
1370 {
1371 	struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
1372 
1373 	if (!decoded.known)
1374 		return PKT_HASH_TYPE_NONE;
1375 
1376 	if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1377 	    decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
1378 		return PKT_HASH_TYPE_L4;
1379 	else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1380 		 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
1381 		return PKT_HASH_TYPE_L3;
1382 	else
1383 		return PKT_HASH_TYPE_L2;
1384 }
1385 
1386 /**
1387  * i40e_rx_hash - set the hash value in the skb
1388  * @ring: descriptor ring
1389  * @rx_desc: specific descriptor
1390  **/
1391 static inline void i40e_rx_hash(struct i40e_ring *ring,
1392 				union i40e_rx_desc *rx_desc,
1393 				struct sk_buff *skb,
1394 				u8 rx_ptype)
1395 {
1396 	u32 hash;
1397 	const __le64 rss_mask =
1398 		cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
1399 			    I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
1400 
1401 	if (!(ring->netdev->features & NETIF_F_RXHASH))
1402 		return;
1403 
1404 	if ((rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask) {
1405 		hash = le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
1406 		skb_set_hash(skb, hash, i40e_ptype_to_htype(rx_ptype));
1407 	}
1408 }
1409 
1410 /**
1411  * i40e_process_skb_fields - Populate skb header fields from Rx descriptor
1412  * @rx_ring: rx descriptor ring packet is being transacted on
1413  * @rx_desc: pointer to the EOP Rx descriptor
1414  * @skb: pointer to current skb being populated
1415  * @rx_ptype: the packet type decoded by hardware
1416  *
1417  * This function checks the ring, descriptor, and packet information in
1418  * order to populate the hash, checksum, VLAN, protocol, and
1419  * other fields within the skb.
1420  **/
1421 static inline
1422 void i40e_process_skb_fields(struct i40e_ring *rx_ring,
1423 			     union i40e_rx_desc *rx_desc, struct sk_buff *skb,
1424 			     u8 rx_ptype)
1425 {
1426 	u64 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1427 	u32 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1428 			I40E_RXD_QW1_STATUS_SHIFT;
1429 	u32 rsyn = (rx_status & I40E_RXD_QW1_STATUS_TSYNINDX_MASK) >>
1430 		   I40E_RXD_QW1_STATUS_TSYNINDX_SHIFT;
1431 
1432 	if (unlikely(rsyn)) {
1433 		i40e_ptp_rx_hwtstamp(rx_ring->vsi->back, skb, rsyn);
1434 		rx_ring->last_rx_timestamp = jiffies;
1435 	}
1436 
1437 	i40e_rx_hash(rx_ring, rx_desc, skb, rx_ptype);
1438 
1439 	/* modifies the skb - consumes the enet header */
1440 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1441 
1442 	i40e_rx_checksum(rx_ring->vsi, skb, rx_desc);
1443 
1444 	skb_record_rx_queue(skb, rx_ring->queue_index);
1445 }
1446 
1447 /**
1448  * i40e_pull_tail - i40e specific version of skb_pull_tail
1449  * @rx_ring: rx descriptor ring packet is being transacted on
1450  * @skb: pointer to current skb being adjusted
1451  *
1452  * This function is an i40e specific version of __pskb_pull_tail.  The
1453  * main difference between this version and the original function is that
1454  * this function can make several assumptions about the state of things
1455  * that allow for significant optimizations versus the standard function.
1456  * As a result we can do things like drop a frag and maintain an accurate
1457  * truesize for the skb.
1458  */
1459 static void i40e_pull_tail(struct i40e_ring *rx_ring, struct sk_buff *skb)
1460 {
1461 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1462 	unsigned char *va;
1463 	unsigned int pull_len;
1464 
1465 	/* it is valid to use page_address instead of kmap since we are
1466 	 * working with pages allocated out of the lomem pool per
1467 	 * alloc_page(GFP_ATOMIC)
1468 	 */
1469 	va = skb_frag_address(frag);
1470 
1471 	/* we need the header to contain the greater of either ETH_HLEN or
1472 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
1473 	 */
1474 	pull_len = eth_get_headlen(va, I40E_RX_HDR_SIZE);
1475 
1476 	/* align pull length to size of long to optimize memcpy performance */
1477 	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1478 
1479 	/* update all of the pointers */
1480 	skb_frag_size_sub(frag, pull_len);
1481 	frag->page_offset += pull_len;
1482 	skb->data_len -= pull_len;
1483 	skb->tail += pull_len;
1484 }
1485 
1486 /**
1487  * i40e_cleanup_headers - Correct empty headers
1488  * @rx_ring: rx descriptor ring packet is being transacted on
1489  * @skb: pointer to current skb being fixed
1490  *
1491  * Also address the case where we are pulling data in on pages only
1492  * and as such no data is present in the skb header.
1493  *
1494  * In addition if skb is not at least 60 bytes we need to pad it so that
1495  * it is large enough to qualify as a valid Ethernet frame.
1496  *
1497  * Returns true if an error was encountered and skb was freed.
1498  **/
1499 static bool i40e_cleanup_headers(struct i40e_ring *rx_ring, struct sk_buff *skb)
1500 {
1501 	/* place header in linear portion of buffer */
1502 	if (skb_is_nonlinear(skb))
1503 		i40e_pull_tail(rx_ring, skb);
1504 
1505 	/* if eth_skb_pad returns an error the skb was freed */
1506 	if (eth_skb_pad(skb))
1507 		return true;
1508 
1509 	return false;
1510 }
1511 
1512 /**
1513  * i40e_reuse_rx_page - page flip buffer and store it back on the ring
1514  * @rx_ring: rx descriptor ring to store buffers on
1515  * @old_buff: donor buffer to have page reused
1516  *
1517  * Synchronizes page for reuse by the adapter
1518  **/
1519 static void i40e_reuse_rx_page(struct i40e_ring *rx_ring,
1520 			       struct i40e_rx_buffer *old_buff)
1521 {
1522 	struct i40e_rx_buffer *new_buff;
1523 	u16 nta = rx_ring->next_to_alloc;
1524 
1525 	new_buff = &rx_ring->rx_bi[nta];
1526 
1527 	/* update, and store next to alloc */
1528 	nta++;
1529 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1530 
1531 	/* transfer page from old buffer to new buffer */
1532 	*new_buff = *old_buff;
1533 }
1534 
1535 /**
1536  * i40e_page_is_reserved - check if reuse is possible
1537  * @page: page struct to check
1538  */
1539 static inline bool i40e_page_is_reserved(struct page *page)
1540 {
1541 	return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1542 }
1543 
1544 /**
1545  * i40e_add_rx_frag - Add contents of Rx buffer to sk_buff
1546  * @rx_ring: rx descriptor ring to transact packets on
1547  * @rx_buffer: buffer containing page to add
1548  * @rx_desc: descriptor containing length of buffer written by hardware
1549  * @skb: sk_buff to place the data into
1550  *
1551  * This function will add the data contained in rx_buffer->page to the skb.
1552  * This is done either through a direct copy if the data in the buffer is
1553  * less than the skb header size, otherwise it will just attach the page as
1554  * a frag to the skb.
1555  *
1556  * The function will then update the page offset if necessary and return
1557  * true if the buffer can be reused by the adapter.
1558  **/
1559 static bool i40e_add_rx_frag(struct i40e_ring *rx_ring,
1560 			     struct i40e_rx_buffer *rx_buffer,
1561 			     union i40e_rx_desc *rx_desc,
1562 			     struct sk_buff *skb)
1563 {
1564 	struct page *page = rx_buffer->page;
1565 	u64 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1566 	unsigned int size = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1567 			    I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1568 #if (PAGE_SIZE < 8192)
1569 	unsigned int truesize = I40E_RXBUFFER_2048;
1570 #else
1571 	unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
1572 	unsigned int last_offset = PAGE_SIZE - I40E_RXBUFFER_2048;
1573 #endif
1574 
1575 	/* will the data fit in the skb we allocated? if so, just
1576 	 * copy it as it is pretty small anyway
1577 	 */
1578 	if ((size <= I40E_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
1579 		unsigned char *va = page_address(page) + rx_buffer->page_offset;
1580 
1581 		memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
1582 
1583 		/* page is not reserved, we can reuse buffer as-is */
1584 		if (likely(!i40e_page_is_reserved(page)))
1585 			return true;
1586 
1587 		/* this page cannot be reused so discard it */
1588 		__free_pages(page, 0);
1589 		return false;
1590 	}
1591 
1592 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1593 			rx_buffer->page_offset, size, truesize);
1594 
1595 	/* avoid re-using remote pages */
1596 	if (unlikely(i40e_page_is_reserved(page)))
1597 		return false;
1598 
1599 #if (PAGE_SIZE < 8192)
1600 	/* if we are only owner of page we can reuse it */
1601 	if (unlikely(page_count(page) != 1))
1602 		return false;
1603 
1604 	/* flip page offset to other buffer */
1605 	rx_buffer->page_offset ^= truesize;
1606 #else
1607 	/* move offset up to the next cache line */
1608 	rx_buffer->page_offset += truesize;
1609 
1610 	if (rx_buffer->page_offset > last_offset)
1611 		return false;
1612 #endif
1613 
1614 	/* Even if we own the page, we are not allowed to use atomic_set()
1615 	 * This would break get_page_unless_zero() users.
1616 	 */
1617 	get_page(rx_buffer->page);
1618 
1619 	return true;
1620 }
1621 
1622 /**
1623  * i40e_fetch_rx_buffer - Allocate skb and populate it
1624  * @rx_ring: rx descriptor ring to transact packets on
1625  * @rx_desc: descriptor containing info written by hardware
1626  *
1627  * This function allocates an skb on the fly, and populates it with the page
1628  * data from the current receive descriptor, taking care to set up the skb
1629  * correctly, as well as handling calling the page recycle function if
1630  * necessary.
1631  */
1632 static inline
1633 struct sk_buff *i40e_fetch_rx_buffer(struct i40e_ring *rx_ring,
1634 				     union i40e_rx_desc *rx_desc)
1635 {
1636 	struct i40e_rx_buffer *rx_buffer;
1637 	struct sk_buff *skb;
1638 	struct page *page;
1639 
1640 	rx_buffer = &rx_ring->rx_bi[rx_ring->next_to_clean];
1641 	page = rx_buffer->page;
1642 	prefetchw(page);
1643 
1644 	skb = rx_buffer->skb;
1645 
1646 	if (likely(!skb)) {
1647 		void *page_addr = page_address(page) + rx_buffer->page_offset;
1648 
1649 		/* prefetch first cache line of first page */
1650 		prefetch(page_addr);
1651 #if L1_CACHE_BYTES < 128
1652 		prefetch(page_addr + L1_CACHE_BYTES);
1653 #endif
1654 
1655 		/* allocate a skb to store the frags */
1656 		skb = __napi_alloc_skb(&rx_ring->q_vector->napi,
1657 				       I40E_RX_HDR_SIZE,
1658 				       GFP_ATOMIC | __GFP_NOWARN);
1659 		if (unlikely(!skb)) {
1660 			rx_ring->rx_stats.alloc_buff_failed++;
1661 			return NULL;
1662 		}
1663 
1664 		/* we will be copying header into skb->data in
1665 		 * pskb_may_pull so it is in our interest to prefetch
1666 		 * it now to avoid a possible cache miss
1667 		 */
1668 		prefetchw(skb->data);
1669 	} else {
1670 		rx_buffer->skb = NULL;
1671 	}
1672 
1673 	/* we are reusing so sync this buffer for CPU use */
1674 	dma_sync_single_range_for_cpu(rx_ring->dev,
1675 				      rx_buffer->dma,
1676 				      rx_buffer->page_offset,
1677 				      I40E_RXBUFFER_2048,
1678 				      DMA_FROM_DEVICE);
1679 
1680 	/* pull page into skb */
1681 	if (i40e_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
1682 		/* hand second half of page back to the ring */
1683 		i40e_reuse_rx_page(rx_ring, rx_buffer);
1684 		rx_ring->rx_stats.page_reuse_count++;
1685 	} else {
1686 		/* we are not reusing the buffer so unmap it */
1687 		dma_unmap_page(rx_ring->dev, rx_buffer->dma, PAGE_SIZE,
1688 			       DMA_FROM_DEVICE);
1689 	}
1690 
1691 	/* clear contents of buffer_info */
1692 	rx_buffer->page = NULL;
1693 
1694 	return skb;
1695 }
1696 
1697 /**
1698  * i40e_is_non_eop - process handling of non-EOP buffers
1699  * @rx_ring: Rx ring being processed
1700  * @rx_desc: Rx descriptor for current buffer
1701  * @skb: Current socket buffer containing buffer in progress
1702  *
1703  * This function updates next to clean.  If the buffer is an EOP buffer
1704  * this function exits returning false, otherwise it will place the
1705  * sk_buff in the next buffer to be chained and return true indicating
1706  * that this is in fact a non-EOP buffer.
1707  **/
1708 static bool i40e_is_non_eop(struct i40e_ring *rx_ring,
1709 			    union i40e_rx_desc *rx_desc,
1710 			    struct sk_buff *skb)
1711 {
1712 	u32 ntc = rx_ring->next_to_clean + 1;
1713 
1714 	/* fetch, update, and store next to clean */
1715 	ntc = (ntc < rx_ring->count) ? ntc : 0;
1716 	rx_ring->next_to_clean = ntc;
1717 
1718 	prefetch(I40E_RX_DESC(rx_ring, ntc));
1719 
1720 #define staterrlen rx_desc->wb.qword1.status_error_len
1721 	if (unlikely(i40e_rx_is_programming_status(le64_to_cpu(staterrlen)))) {
1722 		i40e_clean_programming_status(rx_ring, rx_desc);
1723 		rx_ring->rx_bi[ntc].skb = skb;
1724 		return true;
1725 	}
1726 	/* if we are the last buffer then there is nothing else to do */
1727 #define I40E_RXD_EOF BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)
1728 	if (likely(i40e_test_staterr(rx_desc, I40E_RXD_EOF)))
1729 		return false;
1730 
1731 	/* place skb in next buffer to be received */
1732 	rx_ring->rx_bi[ntc].skb = skb;
1733 	rx_ring->rx_stats.non_eop_descs++;
1734 
1735 	return true;
1736 }
1737 
1738 /**
1739  * i40e_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1740  * @rx_ring: rx descriptor ring to transact packets on
1741  * @budget: Total limit on number of packets to process
1742  *
1743  * This function provides a "bounce buffer" approach to Rx interrupt
1744  * processing.  The advantage to this is that on systems that have
1745  * expensive overhead for IOMMU access this provides a means of avoiding
1746  * it by maintaining the mapping of the page to the system.
1747  *
1748  * Returns amount of work completed
1749  **/
1750 static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget)
1751 {
1752 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1753 	u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
1754 	bool failure = false;
1755 
1756 	while (likely(total_rx_packets < budget)) {
1757 		union i40e_rx_desc *rx_desc;
1758 		struct sk_buff *skb;
1759 		u32 rx_status;
1760 		u16 vlan_tag;
1761 		u8 rx_ptype;
1762 		u64 qword;
1763 
1764 		/* return some buffers to hardware, one at a time is too slow */
1765 		if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1766 			failure = failure ||
1767 				  i40e_alloc_rx_buffers(rx_ring, cleaned_count);
1768 			cleaned_count = 0;
1769 		}
1770 
1771 		rx_desc = I40E_RX_DESC(rx_ring, rx_ring->next_to_clean);
1772 
1773 		qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1774 		rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1775 			   I40E_RXD_QW1_PTYPE_SHIFT;
1776 		rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1777 			    I40E_RXD_QW1_STATUS_SHIFT;
1778 
1779 		if (!(rx_status & BIT(I40E_RX_DESC_STATUS_DD_SHIFT)))
1780 			break;
1781 
1782 		/* status_error_len will always be zero for unused descriptors
1783 		 * because it's cleared in cleanup, and overlaps with hdr_addr
1784 		 * which is always zero because packet split isn't used, if the
1785 		 * hardware wrote DD then it will be non-zero
1786 		 */
1787 		if (!rx_desc->wb.qword1.status_error_len)
1788 			break;
1789 
1790 		/* This memory barrier is needed to keep us from reading
1791 		 * any other fields out of the rx_desc until we know the
1792 		 * DD bit is set.
1793 		 */
1794 		dma_rmb();
1795 
1796 		skb = i40e_fetch_rx_buffer(rx_ring, rx_desc);
1797 		if (!skb)
1798 			break;
1799 
1800 		cleaned_count++;
1801 
1802 		if (i40e_is_non_eop(rx_ring, rx_desc, skb))
1803 			continue;
1804 
1805 		/* ERR_MASK will only have valid bits if EOP set, and
1806 		 * what we are doing here is actually checking
1807 		 * I40E_RX_DESC_ERROR_RXE_SHIFT, since it is the zeroth bit in
1808 		 * the error field
1809 		 */
1810 		if (unlikely(i40e_test_staterr(rx_desc, BIT(I40E_RXD_QW1_ERROR_SHIFT)))) {
1811 			dev_kfree_skb_any(skb);
1812 			continue;
1813 		}
1814 
1815 		if (i40e_cleanup_headers(rx_ring, skb))
1816 			continue;
1817 
1818 		/* probably a little skewed due to removing CRC */
1819 		total_rx_bytes += skb->len;
1820 
1821 		/* populate checksum, VLAN, and protocol */
1822 		i40e_process_skb_fields(rx_ring, rx_desc, skb, rx_ptype);
1823 
1824 #ifdef I40E_FCOE
1825 		if (unlikely(
1826 		    i40e_rx_is_fcoe(rx_ptype) &&
1827 		    !i40e_fcoe_handle_offload(rx_ring, rx_desc, skb))) {
1828 			dev_kfree_skb_any(skb);
1829 			continue;
1830 		}
1831 #endif
1832 
1833 		vlan_tag = (qword & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)) ?
1834 			   le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1) : 0;
1835 
1836 		i40e_receive_skb(rx_ring, skb, vlan_tag);
1837 
1838 		/* update budget accounting */
1839 		total_rx_packets++;
1840 	}
1841 
1842 	u64_stats_update_begin(&rx_ring->syncp);
1843 	rx_ring->stats.packets += total_rx_packets;
1844 	rx_ring->stats.bytes += total_rx_bytes;
1845 	u64_stats_update_end(&rx_ring->syncp);
1846 	rx_ring->q_vector->rx.total_packets += total_rx_packets;
1847 	rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1848 
1849 	/* guarantee a trip back through this routine if there was a failure */
1850 	return failure ? budget : total_rx_packets;
1851 }
1852 
1853 static u32 i40e_buildreg_itr(const int type, const u16 itr)
1854 {
1855 	u32 val;
1856 
1857 	val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
1858 	      /* Don't clear PBA because that can cause lost interrupts that
1859 	       * came in while we were cleaning/polling
1860 	       */
1861 	      (type << I40E_PFINT_DYN_CTLN_ITR_INDX_SHIFT) |
1862 	      (itr << I40E_PFINT_DYN_CTLN_INTERVAL_SHIFT);
1863 
1864 	return val;
1865 }
1866 
1867 /* a small macro to shorten up some long lines */
1868 #define INTREG I40E_PFINT_DYN_CTLN
1869 
1870 /**
1871  * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt
1872  * @vsi: the VSI we care about
1873  * @q_vector: q_vector for which itr is being updated and interrupt enabled
1874  *
1875  **/
1876 static inline void i40e_update_enable_itr(struct i40e_vsi *vsi,
1877 					  struct i40e_q_vector *q_vector)
1878 {
1879 	struct i40e_hw *hw = &vsi->back->hw;
1880 	bool rx = false, tx = false;
1881 	u32 rxval, txval;
1882 	int vector;
1883 	int idx = q_vector->v_idx;
1884 
1885 	vector = (q_vector->v_idx + vsi->base_vector);
1886 
1887 	/* avoid dynamic calculation if in countdown mode OR if
1888 	 * all dynamic is disabled
1889 	 */
1890 	rxval = txval = i40e_buildreg_itr(I40E_ITR_NONE, 0);
1891 
1892 	if (q_vector->itr_countdown > 0 ||
1893 	    (!ITR_IS_DYNAMIC(vsi->rx_rings[idx]->rx_itr_setting) &&
1894 	     !ITR_IS_DYNAMIC(vsi->tx_rings[idx]->tx_itr_setting))) {
1895 		goto enable_int;
1896 	}
1897 
1898 	if (ITR_IS_DYNAMIC(vsi->rx_rings[idx]->rx_itr_setting)) {
1899 		rx = i40e_set_new_dynamic_itr(&q_vector->rx);
1900 		rxval = i40e_buildreg_itr(I40E_RX_ITR, q_vector->rx.itr);
1901 	}
1902 
1903 	if (ITR_IS_DYNAMIC(vsi->tx_rings[idx]->tx_itr_setting)) {
1904 		tx = i40e_set_new_dynamic_itr(&q_vector->tx);
1905 		txval = i40e_buildreg_itr(I40E_TX_ITR, q_vector->tx.itr);
1906 	}
1907 
1908 	if (rx || tx) {
1909 		/* get the higher of the two ITR adjustments and
1910 		 * use the same value for both ITR registers
1911 		 * when in adaptive mode (Rx and/or Tx)
1912 		 */
1913 		u16 itr = max(q_vector->tx.itr, q_vector->rx.itr);
1914 
1915 		q_vector->tx.itr = q_vector->rx.itr = itr;
1916 		txval = i40e_buildreg_itr(I40E_TX_ITR, itr);
1917 		tx = true;
1918 		rxval = i40e_buildreg_itr(I40E_RX_ITR, itr);
1919 		rx = true;
1920 	}
1921 
1922 	/* only need to enable the interrupt once, but need
1923 	 * to possibly update both ITR values
1924 	 */
1925 	if (rx) {
1926 		/* set the INTENA_MSK_MASK so that this first write
1927 		 * won't actually enable the interrupt, instead just
1928 		 * updating the ITR (it's bit 31 PF and VF)
1929 		 */
1930 		rxval |= BIT(31);
1931 		/* don't check _DOWN because interrupt isn't being enabled */
1932 		wr32(hw, INTREG(vector - 1), rxval);
1933 	}
1934 
1935 enable_int:
1936 	if (!test_bit(__I40E_DOWN, &vsi->state))
1937 		wr32(hw, INTREG(vector - 1), txval);
1938 
1939 	if (q_vector->itr_countdown)
1940 		q_vector->itr_countdown--;
1941 	else
1942 		q_vector->itr_countdown = ITR_COUNTDOWN_START;
1943 }
1944 
1945 /**
1946  * i40e_napi_poll - NAPI polling Rx/Tx cleanup routine
1947  * @napi: napi struct with our devices info in it
1948  * @budget: amount of work driver is allowed to do this pass, in packets
1949  *
1950  * This function will clean all queues associated with a q_vector.
1951  *
1952  * Returns the amount of work done
1953  **/
1954 int i40e_napi_poll(struct napi_struct *napi, int budget)
1955 {
1956 	struct i40e_q_vector *q_vector =
1957 			       container_of(napi, struct i40e_q_vector, napi);
1958 	struct i40e_vsi *vsi = q_vector->vsi;
1959 	struct i40e_ring *ring;
1960 	bool clean_complete = true;
1961 	bool arm_wb = false;
1962 	int budget_per_ring;
1963 	int work_done = 0;
1964 
1965 	if (test_bit(__I40E_DOWN, &vsi->state)) {
1966 		napi_complete(napi);
1967 		return 0;
1968 	}
1969 
1970 	/* Clear hung_detected bit */
1971 	clear_bit(I40E_Q_VECTOR_HUNG_DETECT, &q_vector->hung_detected);
1972 	/* Since the actual Tx work is minimal, we can give the Tx a larger
1973 	 * budget and be more aggressive about cleaning up the Tx descriptors.
1974 	 */
1975 	i40e_for_each_ring(ring, q_vector->tx) {
1976 		if (!i40e_clean_tx_irq(vsi, ring, budget)) {
1977 			clean_complete = false;
1978 			continue;
1979 		}
1980 		arm_wb |= ring->arm_wb;
1981 		ring->arm_wb = false;
1982 	}
1983 
1984 	/* Handle case where we are called by netpoll with a budget of 0 */
1985 	if (budget <= 0)
1986 		goto tx_only;
1987 
1988 	/* We attempt to distribute budget to each Rx queue fairly, but don't
1989 	 * allow the budget to go below 1 because that would exit polling early.
1990 	 */
1991 	budget_per_ring = max(budget/q_vector->num_ringpairs, 1);
1992 
1993 	i40e_for_each_ring(ring, q_vector->rx) {
1994 		int cleaned = i40e_clean_rx_irq(ring, budget_per_ring);
1995 
1996 		work_done += cleaned;
1997 		/* if we clean as many as budgeted, we must not be done */
1998 		if (cleaned >= budget_per_ring)
1999 			clean_complete = false;
2000 	}
2001 
2002 	/* If work not completed, return budget and polling will return */
2003 	if (!clean_complete) {
2004 tx_only:
2005 		if (arm_wb) {
2006 			q_vector->tx.ring[0].tx_stats.tx_force_wb++;
2007 			i40e_enable_wb_on_itr(vsi, q_vector);
2008 		}
2009 		return budget;
2010 	}
2011 
2012 	if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR)
2013 		q_vector->arm_wb_state = false;
2014 
2015 	/* Work is done so exit the polling mode and re-enable the interrupt */
2016 	napi_complete_done(napi, work_done);
2017 	if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
2018 		i40e_update_enable_itr(vsi, q_vector);
2019 	} else { /* Legacy mode */
2020 		i40e_irq_dynamic_enable_icr0(vsi->back, false);
2021 	}
2022 	return 0;
2023 }
2024 
2025 /**
2026  * i40e_atr - Add a Flow Director ATR filter
2027  * @tx_ring:  ring to add programming descriptor to
2028  * @skb:      send buffer
2029  * @tx_flags: send tx flags
2030  **/
2031 static void i40e_atr(struct i40e_ring *tx_ring, struct sk_buff *skb,
2032 		     u32 tx_flags)
2033 {
2034 	struct i40e_filter_program_desc *fdir_desc;
2035 	struct i40e_pf *pf = tx_ring->vsi->back;
2036 	union {
2037 		unsigned char *network;
2038 		struct iphdr *ipv4;
2039 		struct ipv6hdr *ipv6;
2040 	} hdr;
2041 	struct tcphdr *th;
2042 	unsigned int hlen;
2043 	u32 flex_ptype, dtype_cmd;
2044 	int l4_proto;
2045 	u16 i;
2046 
2047 	/* make sure ATR is enabled */
2048 	if (!(pf->flags & I40E_FLAG_FD_ATR_ENABLED))
2049 		return;
2050 
2051 	if ((pf->auto_disable_flags & I40E_FLAG_FD_ATR_ENABLED))
2052 		return;
2053 
2054 	/* if sampling is disabled do nothing */
2055 	if (!tx_ring->atr_sample_rate)
2056 		return;
2057 
2058 	/* Currently only IPv4/IPv6 with TCP is supported */
2059 	if (!(tx_flags & (I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6)))
2060 		return;
2061 
2062 	/* snag network header to get L4 type and address */
2063 	hdr.network = (tx_flags & I40E_TX_FLAGS_UDP_TUNNEL) ?
2064 		      skb_inner_network_header(skb) : skb_network_header(skb);
2065 
2066 	/* Note: tx_flags gets modified to reflect inner protocols in
2067 	 * tx_enable_csum function if encap is enabled.
2068 	 */
2069 	if (tx_flags & I40E_TX_FLAGS_IPV4) {
2070 		/* access ihl as u8 to avoid unaligned access on ia64 */
2071 		hlen = (hdr.network[0] & 0x0F) << 2;
2072 		l4_proto = hdr.ipv4->protocol;
2073 	} else {
2074 		hlen = hdr.network - skb->data;
2075 		l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
2076 		hlen -= hdr.network - skb->data;
2077 	}
2078 
2079 	if (l4_proto != IPPROTO_TCP)
2080 		return;
2081 
2082 	th = (struct tcphdr *)(hdr.network + hlen);
2083 
2084 	/* Due to lack of space, no more new filters can be programmed */
2085 	if (th->syn && (pf->auto_disable_flags & I40E_FLAG_FD_ATR_ENABLED))
2086 		return;
2087 	if ((pf->flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE) &&
2088 	    (!(pf->auto_disable_flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE))) {
2089 		/* HW ATR eviction will take care of removing filters on FIN
2090 		 * and RST packets.
2091 		 */
2092 		if (th->fin || th->rst)
2093 			return;
2094 	}
2095 
2096 	tx_ring->atr_count++;
2097 
2098 	/* sample on all syn/fin/rst packets or once every atr sample rate */
2099 	if (!th->fin &&
2100 	    !th->syn &&
2101 	    !th->rst &&
2102 	    (tx_ring->atr_count < tx_ring->atr_sample_rate))
2103 		return;
2104 
2105 	tx_ring->atr_count = 0;
2106 
2107 	/* grab the next descriptor */
2108 	i = tx_ring->next_to_use;
2109 	fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
2110 
2111 	i++;
2112 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2113 
2114 	flex_ptype = (tx_ring->queue_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) &
2115 		      I40E_TXD_FLTR_QW0_QINDEX_MASK;
2116 	flex_ptype |= (tx_flags & I40E_TX_FLAGS_IPV4) ?
2117 		      (I40E_FILTER_PCTYPE_NONF_IPV4_TCP <<
2118 		       I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) :
2119 		      (I40E_FILTER_PCTYPE_NONF_IPV6_TCP <<
2120 		       I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
2121 
2122 	flex_ptype |= tx_ring->vsi->id << I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT;
2123 
2124 	dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
2125 
2126 	dtype_cmd |= (th->fin || th->rst) ?
2127 		     (I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
2128 		      I40E_TXD_FLTR_QW1_PCMD_SHIFT) :
2129 		     (I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
2130 		      I40E_TXD_FLTR_QW1_PCMD_SHIFT);
2131 
2132 	dtype_cmd |= I40E_FILTER_PROGRAM_DESC_DEST_DIRECT_PACKET_QINDEX <<
2133 		     I40E_TXD_FLTR_QW1_DEST_SHIFT;
2134 
2135 	dtype_cmd |= I40E_FILTER_PROGRAM_DESC_FD_STATUS_FD_ID <<
2136 		     I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT;
2137 
2138 	dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
2139 	if (!(tx_flags & I40E_TX_FLAGS_UDP_TUNNEL))
2140 		dtype_cmd |=
2141 			((u32)I40E_FD_ATR_STAT_IDX(pf->hw.pf_id) <<
2142 			I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2143 			I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2144 	else
2145 		dtype_cmd |=
2146 			((u32)I40E_FD_ATR_TUNNEL_STAT_IDX(pf->hw.pf_id) <<
2147 			I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2148 			I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2149 
2150 	if ((pf->flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE) &&
2151 	    (!(pf->auto_disable_flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE)))
2152 		dtype_cmd |= I40E_TXD_FLTR_QW1_ATR_MASK;
2153 
2154 	fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
2155 	fdir_desc->rsvd = cpu_to_le32(0);
2156 	fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
2157 	fdir_desc->fd_id = cpu_to_le32(0);
2158 }
2159 
2160 /**
2161  * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
2162  * @skb:     send buffer
2163  * @tx_ring: ring to send buffer on
2164  * @flags:   the tx flags to be set
2165  *
2166  * Checks the skb and set up correspondingly several generic transmit flags
2167  * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
2168  *
2169  * Returns error code indicate the frame should be dropped upon error and the
2170  * otherwise  returns 0 to indicate the flags has been set properly.
2171  **/
2172 #ifdef I40E_FCOE
2173 inline int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
2174 				      struct i40e_ring *tx_ring,
2175 				      u32 *flags)
2176 #else
2177 static inline int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
2178 					     struct i40e_ring *tx_ring,
2179 					     u32 *flags)
2180 #endif
2181 {
2182 	__be16 protocol = skb->protocol;
2183 	u32  tx_flags = 0;
2184 
2185 	if (protocol == htons(ETH_P_8021Q) &&
2186 	    !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
2187 		/* When HW VLAN acceleration is turned off by the user the
2188 		 * stack sets the protocol to 8021q so that the driver
2189 		 * can take any steps required to support the SW only
2190 		 * VLAN handling.  In our case the driver doesn't need
2191 		 * to take any further steps so just set the protocol
2192 		 * to the encapsulated ethertype.
2193 		 */
2194 		skb->protocol = vlan_get_protocol(skb);
2195 		goto out;
2196 	}
2197 
2198 	/* if we have a HW VLAN tag being added, default to the HW one */
2199 	if (skb_vlan_tag_present(skb)) {
2200 		tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
2201 		tx_flags |= I40E_TX_FLAGS_HW_VLAN;
2202 	/* else if it is a SW VLAN, check the next protocol and store the tag */
2203 	} else if (protocol == htons(ETH_P_8021Q)) {
2204 		struct vlan_hdr *vhdr, _vhdr;
2205 
2206 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
2207 		if (!vhdr)
2208 			return -EINVAL;
2209 
2210 		protocol = vhdr->h_vlan_encapsulated_proto;
2211 		tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
2212 		tx_flags |= I40E_TX_FLAGS_SW_VLAN;
2213 	}
2214 
2215 	if (!(tx_ring->vsi->back->flags & I40E_FLAG_DCB_ENABLED))
2216 		goto out;
2217 
2218 	/* Insert 802.1p priority into VLAN header */
2219 	if ((tx_flags & (I40E_TX_FLAGS_HW_VLAN | I40E_TX_FLAGS_SW_VLAN)) ||
2220 	    (skb->priority != TC_PRIO_CONTROL)) {
2221 		tx_flags &= ~I40E_TX_FLAGS_VLAN_PRIO_MASK;
2222 		tx_flags |= (skb->priority & 0x7) <<
2223 				I40E_TX_FLAGS_VLAN_PRIO_SHIFT;
2224 		if (tx_flags & I40E_TX_FLAGS_SW_VLAN) {
2225 			struct vlan_ethhdr *vhdr;
2226 			int rc;
2227 
2228 			rc = skb_cow_head(skb, 0);
2229 			if (rc < 0)
2230 				return rc;
2231 			vhdr = (struct vlan_ethhdr *)skb->data;
2232 			vhdr->h_vlan_TCI = htons(tx_flags >>
2233 						 I40E_TX_FLAGS_VLAN_SHIFT);
2234 		} else {
2235 			tx_flags |= I40E_TX_FLAGS_HW_VLAN;
2236 		}
2237 	}
2238 
2239 out:
2240 	*flags = tx_flags;
2241 	return 0;
2242 }
2243 
2244 /**
2245  * i40e_tso - set up the tso context descriptor
2246  * @skb:      ptr to the skb we're sending
2247  * @hdr_len:  ptr to the size of the packet header
2248  * @cd_type_cmd_tso_mss: Quad Word 1
2249  *
2250  * Returns 0 if no TSO can happen, 1 if tso is going, or error
2251  **/
2252 static int i40e_tso(struct sk_buff *skb, u8 *hdr_len, u64 *cd_type_cmd_tso_mss)
2253 {
2254 	u64 cd_cmd, cd_tso_len, cd_mss;
2255 	union {
2256 		struct iphdr *v4;
2257 		struct ipv6hdr *v6;
2258 		unsigned char *hdr;
2259 	} ip;
2260 	union {
2261 		struct tcphdr *tcp;
2262 		struct udphdr *udp;
2263 		unsigned char *hdr;
2264 	} l4;
2265 	u32 paylen, l4_offset;
2266 	int err;
2267 
2268 	if (skb->ip_summed != CHECKSUM_PARTIAL)
2269 		return 0;
2270 
2271 	if (!skb_is_gso(skb))
2272 		return 0;
2273 
2274 	err = skb_cow_head(skb, 0);
2275 	if (err < 0)
2276 		return err;
2277 
2278 	ip.hdr = skb_network_header(skb);
2279 	l4.hdr = skb_transport_header(skb);
2280 
2281 	/* initialize outer IP header fields */
2282 	if (ip.v4->version == 4) {
2283 		ip.v4->tot_len = 0;
2284 		ip.v4->check = 0;
2285 	} else {
2286 		ip.v6->payload_len = 0;
2287 	}
2288 
2289 	if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
2290 					 SKB_GSO_GRE_CSUM |
2291 					 SKB_GSO_IPXIP4 |
2292 					 SKB_GSO_IPXIP6 |
2293 					 SKB_GSO_UDP_TUNNEL |
2294 					 SKB_GSO_UDP_TUNNEL_CSUM)) {
2295 		if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
2296 		    (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)) {
2297 			l4.udp->len = 0;
2298 
2299 			/* determine offset of outer transport header */
2300 			l4_offset = l4.hdr - skb->data;
2301 
2302 			/* remove payload length from outer checksum */
2303 			paylen = skb->len - l4_offset;
2304 			csum_replace_by_diff(&l4.udp->check, htonl(paylen));
2305 		}
2306 
2307 		/* reset pointers to inner headers */
2308 		ip.hdr = skb_inner_network_header(skb);
2309 		l4.hdr = skb_inner_transport_header(skb);
2310 
2311 		/* initialize inner IP header fields */
2312 		if (ip.v4->version == 4) {
2313 			ip.v4->tot_len = 0;
2314 			ip.v4->check = 0;
2315 		} else {
2316 			ip.v6->payload_len = 0;
2317 		}
2318 	}
2319 
2320 	/* determine offset of inner transport header */
2321 	l4_offset = l4.hdr - skb->data;
2322 
2323 	/* remove payload length from inner checksum */
2324 	paylen = skb->len - l4_offset;
2325 	csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
2326 
2327 	/* compute length of segmentation header */
2328 	*hdr_len = (l4.tcp->doff * 4) + l4_offset;
2329 
2330 	/* find the field values */
2331 	cd_cmd = I40E_TX_CTX_DESC_TSO;
2332 	cd_tso_len = skb->len - *hdr_len;
2333 	cd_mss = skb_shinfo(skb)->gso_size;
2334 	*cd_type_cmd_tso_mss |= (cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
2335 				(cd_tso_len << I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
2336 				(cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
2337 	return 1;
2338 }
2339 
2340 /**
2341  * i40e_tsyn - set up the tsyn context descriptor
2342  * @tx_ring:  ptr to the ring to send
2343  * @skb:      ptr to the skb we're sending
2344  * @tx_flags: the collected send information
2345  * @cd_type_cmd_tso_mss: Quad Word 1
2346  *
2347  * Returns 0 if no Tx timestamp can happen and 1 if the timestamp will happen
2348  **/
2349 static int i40e_tsyn(struct i40e_ring *tx_ring, struct sk_buff *skb,
2350 		     u32 tx_flags, u64 *cd_type_cmd_tso_mss)
2351 {
2352 	struct i40e_pf *pf;
2353 
2354 	if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)))
2355 		return 0;
2356 
2357 	/* Tx timestamps cannot be sampled when doing TSO */
2358 	if (tx_flags & I40E_TX_FLAGS_TSO)
2359 		return 0;
2360 
2361 	/* only timestamp the outbound packet if the user has requested it and
2362 	 * we are not already transmitting a packet to be timestamped
2363 	 */
2364 	pf = i40e_netdev_to_pf(tx_ring->netdev);
2365 	if (!(pf->flags & I40E_FLAG_PTP))
2366 		return 0;
2367 
2368 	if (pf->ptp_tx &&
2369 	    !test_and_set_bit_lock(__I40E_PTP_TX_IN_PROGRESS, &pf->state)) {
2370 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
2371 		pf->ptp_tx_skb = skb_get(skb);
2372 	} else {
2373 		return 0;
2374 	}
2375 
2376 	*cd_type_cmd_tso_mss |= (u64)I40E_TX_CTX_DESC_TSYN <<
2377 				I40E_TXD_CTX_QW1_CMD_SHIFT;
2378 
2379 	return 1;
2380 }
2381 
2382 /**
2383  * i40e_tx_enable_csum - Enable Tx checksum offloads
2384  * @skb: send buffer
2385  * @tx_flags: pointer to Tx flags currently set
2386  * @td_cmd: Tx descriptor command bits to set
2387  * @td_offset: Tx descriptor header offsets to set
2388  * @tx_ring: Tx descriptor ring
2389  * @cd_tunneling: ptr to context desc bits
2390  **/
2391 static int i40e_tx_enable_csum(struct sk_buff *skb, u32 *tx_flags,
2392 			       u32 *td_cmd, u32 *td_offset,
2393 			       struct i40e_ring *tx_ring,
2394 			       u32 *cd_tunneling)
2395 {
2396 	union {
2397 		struct iphdr *v4;
2398 		struct ipv6hdr *v6;
2399 		unsigned char *hdr;
2400 	} ip;
2401 	union {
2402 		struct tcphdr *tcp;
2403 		struct udphdr *udp;
2404 		unsigned char *hdr;
2405 	} l4;
2406 	unsigned char *exthdr;
2407 	u32 offset, cmd = 0;
2408 	__be16 frag_off;
2409 	u8 l4_proto = 0;
2410 
2411 	if (skb->ip_summed != CHECKSUM_PARTIAL)
2412 		return 0;
2413 
2414 	ip.hdr = skb_network_header(skb);
2415 	l4.hdr = skb_transport_header(skb);
2416 
2417 	/* compute outer L2 header size */
2418 	offset = ((ip.hdr - skb->data) / 2) << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
2419 
2420 	if (skb->encapsulation) {
2421 		u32 tunnel = 0;
2422 		/* define outer network header type */
2423 		if (*tx_flags & I40E_TX_FLAGS_IPV4) {
2424 			tunnel |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
2425 				  I40E_TX_CTX_EXT_IP_IPV4 :
2426 				  I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
2427 
2428 			l4_proto = ip.v4->protocol;
2429 		} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
2430 			tunnel |= I40E_TX_CTX_EXT_IP_IPV6;
2431 
2432 			exthdr = ip.hdr + sizeof(*ip.v6);
2433 			l4_proto = ip.v6->nexthdr;
2434 			if (l4.hdr != exthdr)
2435 				ipv6_skip_exthdr(skb, exthdr - skb->data,
2436 						 &l4_proto, &frag_off);
2437 		}
2438 
2439 		/* define outer transport */
2440 		switch (l4_proto) {
2441 		case IPPROTO_UDP:
2442 			tunnel |= I40E_TXD_CTX_UDP_TUNNELING;
2443 			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
2444 			break;
2445 		case IPPROTO_GRE:
2446 			tunnel |= I40E_TXD_CTX_GRE_TUNNELING;
2447 			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
2448 			break;
2449 		case IPPROTO_IPIP:
2450 		case IPPROTO_IPV6:
2451 			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
2452 			l4.hdr = skb_inner_network_header(skb);
2453 			break;
2454 		default:
2455 			if (*tx_flags & I40E_TX_FLAGS_TSO)
2456 				return -1;
2457 
2458 			skb_checksum_help(skb);
2459 			return 0;
2460 		}
2461 
2462 		/* compute outer L3 header size */
2463 		tunnel |= ((l4.hdr - ip.hdr) / 4) <<
2464 			  I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT;
2465 
2466 		/* switch IP header pointer from outer to inner header */
2467 		ip.hdr = skb_inner_network_header(skb);
2468 
2469 		/* compute tunnel header size */
2470 		tunnel |= ((ip.hdr - l4.hdr) / 2) <<
2471 			  I40E_TXD_CTX_QW0_NATLEN_SHIFT;
2472 
2473 		/* indicate if we need to offload outer UDP header */
2474 		if ((*tx_flags & I40E_TX_FLAGS_TSO) &&
2475 		    !(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
2476 		    (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM))
2477 			tunnel |= I40E_TXD_CTX_QW0_L4T_CS_MASK;
2478 
2479 		/* record tunnel offload values */
2480 		*cd_tunneling |= tunnel;
2481 
2482 		/* switch L4 header pointer from outer to inner */
2483 		l4.hdr = skb_inner_transport_header(skb);
2484 		l4_proto = 0;
2485 
2486 		/* reset type as we transition from outer to inner headers */
2487 		*tx_flags &= ~(I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6);
2488 		if (ip.v4->version == 4)
2489 			*tx_flags |= I40E_TX_FLAGS_IPV4;
2490 		if (ip.v6->version == 6)
2491 			*tx_flags |= I40E_TX_FLAGS_IPV6;
2492 	}
2493 
2494 	/* Enable IP checksum offloads */
2495 	if (*tx_flags & I40E_TX_FLAGS_IPV4) {
2496 		l4_proto = ip.v4->protocol;
2497 		/* the stack computes the IP header already, the only time we
2498 		 * need the hardware to recompute it is in the case of TSO.
2499 		 */
2500 		cmd |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
2501 		       I40E_TX_DESC_CMD_IIPT_IPV4_CSUM :
2502 		       I40E_TX_DESC_CMD_IIPT_IPV4;
2503 	} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
2504 		cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
2505 
2506 		exthdr = ip.hdr + sizeof(*ip.v6);
2507 		l4_proto = ip.v6->nexthdr;
2508 		if (l4.hdr != exthdr)
2509 			ipv6_skip_exthdr(skb, exthdr - skb->data,
2510 					 &l4_proto, &frag_off);
2511 	}
2512 
2513 	/* compute inner L3 header size */
2514 	offset |= ((l4.hdr - ip.hdr) / 4) << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
2515 
2516 	/* Enable L4 checksum offloads */
2517 	switch (l4_proto) {
2518 	case IPPROTO_TCP:
2519 		/* enable checksum offloads */
2520 		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
2521 		offset |= l4.tcp->doff << I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
2522 		break;
2523 	case IPPROTO_SCTP:
2524 		/* enable SCTP checksum offload */
2525 		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
2526 		offset |= (sizeof(struct sctphdr) >> 2) <<
2527 			  I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
2528 		break;
2529 	case IPPROTO_UDP:
2530 		/* enable UDP checksum offload */
2531 		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
2532 		offset |= (sizeof(struct udphdr) >> 2) <<
2533 			  I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
2534 		break;
2535 	default:
2536 		if (*tx_flags & I40E_TX_FLAGS_TSO)
2537 			return -1;
2538 		skb_checksum_help(skb);
2539 		return 0;
2540 	}
2541 
2542 	*td_cmd |= cmd;
2543 	*td_offset |= offset;
2544 
2545 	return 1;
2546 }
2547 
2548 /**
2549  * i40e_create_tx_ctx Build the Tx context descriptor
2550  * @tx_ring:  ring to create the descriptor on
2551  * @cd_type_cmd_tso_mss: Quad Word 1
2552  * @cd_tunneling: Quad Word 0 - bits 0-31
2553  * @cd_l2tag2: Quad Word 0 - bits 32-63
2554  **/
2555 static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
2556 			       const u64 cd_type_cmd_tso_mss,
2557 			       const u32 cd_tunneling, const u32 cd_l2tag2)
2558 {
2559 	struct i40e_tx_context_desc *context_desc;
2560 	int i = tx_ring->next_to_use;
2561 
2562 	if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
2563 	    !cd_tunneling && !cd_l2tag2)
2564 		return;
2565 
2566 	/* grab the next descriptor */
2567 	context_desc = I40E_TX_CTXTDESC(tx_ring, i);
2568 
2569 	i++;
2570 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2571 
2572 	/* cpu_to_le32 and assign to struct fields */
2573 	context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
2574 	context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
2575 	context_desc->rsvd = cpu_to_le16(0);
2576 	context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
2577 }
2578 
2579 /**
2580  * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions
2581  * @tx_ring: the ring to be checked
2582  * @size:    the size buffer we want to assure is available
2583  *
2584  * Returns -EBUSY if a stop is needed, else 0
2585  **/
2586 int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
2587 {
2588 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
2589 	/* Memory barrier before checking head and tail */
2590 	smp_mb();
2591 
2592 	/* Check again in a case another CPU has just made room available. */
2593 	if (likely(I40E_DESC_UNUSED(tx_ring) < size))
2594 		return -EBUSY;
2595 
2596 	/* A reprieve! - use start_queue because it doesn't call schedule */
2597 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
2598 	++tx_ring->tx_stats.restart_queue;
2599 	return 0;
2600 }
2601 
2602 /**
2603  * __i40e_chk_linearize - Check if there are more than 8 buffers per packet
2604  * @skb:      send buffer
2605  *
2606  * Note: Our HW can't DMA more than 8 buffers to build a packet on the wire
2607  * and so we need to figure out the cases where we need to linearize the skb.
2608  *
2609  * For TSO we need to count the TSO header and segment payload separately.
2610  * As such we need to check cases where we have 7 fragments or more as we
2611  * can potentially require 9 DMA transactions, 1 for the TSO header, 1 for
2612  * the segment payload in the first descriptor, and another 7 for the
2613  * fragments.
2614  **/
2615 bool __i40e_chk_linearize(struct sk_buff *skb)
2616 {
2617 	const struct skb_frag_struct *frag, *stale;
2618 	int nr_frags, sum;
2619 
2620 	/* no need to check if number of frags is less than 7 */
2621 	nr_frags = skb_shinfo(skb)->nr_frags;
2622 	if (nr_frags < (I40E_MAX_BUFFER_TXD - 1))
2623 		return false;
2624 
2625 	/* We need to walk through the list and validate that each group
2626 	 * of 6 fragments totals at least gso_size.  However we don't need
2627 	 * to perform such validation on the last 6 since the last 6 cannot
2628 	 * inherit any data from a descriptor after them.
2629 	 */
2630 	nr_frags -= I40E_MAX_BUFFER_TXD - 2;
2631 	frag = &skb_shinfo(skb)->frags[0];
2632 
2633 	/* Initialize size to the negative value of gso_size minus 1.  We
2634 	 * use this as the worst case scenerio in which the frag ahead
2635 	 * of us only provides one byte which is why we are limited to 6
2636 	 * descriptors for a single transmit as the header and previous
2637 	 * fragment are already consuming 2 descriptors.
2638 	 */
2639 	sum = 1 - skb_shinfo(skb)->gso_size;
2640 
2641 	/* Add size of frags 0 through 4 to create our initial sum */
2642 	sum += skb_frag_size(frag++);
2643 	sum += skb_frag_size(frag++);
2644 	sum += skb_frag_size(frag++);
2645 	sum += skb_frag_size(frag++);
2646 	sum += skb_frag_size(frag++);
2647 
2648 	/* Walk through fragments adding latest fragment, testing it, and
2649 	 * then removing stale fragments from the sum.
2650 	 */
2651 	stale = &skb_shinfo(skb)->frags[0];
2652 	for (;;) {
2653 		sum += skb_frag_size(frag++);
2654 
2655 		/* if sum is negative we failed to make sufficient progress */
2656 		if (sum < 0)
2657 			return true;
2658 
2659 		/* use pre-decrement to avoid processing last fragment */
2660 		if (!--nr_frags)
2661 			break;
2662 
2663 		sum -= skb_frag_size(stale++);
2664 	}
2665 
2666 	return false;
2667 }
2668 
2669 /**
2670  * i40e_tx_map - Build the Tx descriptor
2671  * @tx_ring:  ring to send buffer on
2672  * @skb:      send buffer
2673  * @first:    first buffer info buffer to use
2674  * @tx_flags: collected send information
2675  * @hdr_len:  size of the packet header
2676  * @td_cmd:   the command field in the descriptor
2677  * @td_offset: offset for checksum or crc
2678  **/
2679 #ifdef I40E_FCOE
2680 inline void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
2681 			struct i40e_tx_buffer *first, u32 tx_flags,
2682 			const u8 hdr_len, u32 td_cmd, u32 td_offset)
2683 #else
2684 static inline void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
2685 			       struct i40e_tx_buffer *first, u32 tx_flags,
2686 			       const u8 hdr_len, u32 td_cmd, u32 td_offset)
2687 #endif
2688 {
2689 	unsigned int data_len = skb->data_len;
2690 	unsigned int size = skb_headlen(skb);
2691 	struct skb_frag_struct *frag;
2692 	struct i40e_tx_buffer *tx_bi;
2693 	struct i40e_tx_desc *tx_desc;
2694 	u16 i = tx_ring->next_to_use;
2695 	u32 td_tag = 0;
2696 	dma_addr_t dma;
2697 	u16 gso_segs;
2698 	u16 desc_count = 0;
2699 	bool tail_bump = true;
2700 	bool do_rs = false;
2701 
2702 	if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
2703 		td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
2704 		td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
2705 			 I40E_TX_FLAGS_VLAN_SHIFT;
2706 	}
2707 
2708 	if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO))
2709 		gso_segs = skb_shinfo(skb)->gso_segs;
2710 	else
2711 		gso_segs = 1;
2712 
2713 	/* multiply data chunks by size of headers */
2714 	first->bytecount = skb->len - hdr_len + (gso_segs * hdr_len);
2715 	first->gso_segs = gso_segs;
2716 	first->skb = skb;
2717 	first->tx_flags = tx_flags;
2718 
2719 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
2720 
2721 	tx_desc = I40E_TX_DESC(tx_ring, i);
2722 	tx_bi = first;
2723 
2724 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
2725 		unsigned int max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
2726 
2727 		if (dma_mapping_error(tx_ring->dev, dma))
2728 			goto dma_error;
2729 
2730 		/* record length, and DMA address */
2731 		dma_unmap_len_set(tx_bi, len, size);
2732 		dma_unmap_addr_set(tx_bi, dma, dma);
2733 
2734 		/* align size to end of page */
2735 		max_data += -dma & (I40E_MAX_READ_REQ_SIZE - 1);
2736 		tx_desc->buffer_addr = cpu_to_le64(dma);
2737 
2738 		while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
2739 			tx_desc->cmd_type_offset_bsz =
2740 				build_ctob(td_cmd, td_offset,
2741 					   max_data, td_tag);
2742 
2743 			tx_desc++;
2744 			i++;
2745 			desc_count++;
2746 
2747 			if (i == tx_ring->count) {
2748 				tx_desc = I40E_TX_DESC(tx_ring, 0);
2749 				i = 0;
2750 			}
2751 
2752 			dma += max_data;
2753 			size -= max_data;
2754 
2755 			max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
2756 			tx_desc->buffer_addr = cpu_to_le64(dma);
2757 		}
2758 
2759 		if (likely(!data_len))
2760 			break;
2761 
2762 		tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
2763 							  size, td_tag);
2764 
2765 		tx_desc++;
2766 		i++;
2767 		desc_count++;
2768 
2769 		if (i == tx_ring->count) {
2770 			tx_desc = I40E_TX_DESC(tx_ring, 0);
2771 			i = 0;
2772 		}
2773 
2774 		size = skb_frag_size(frag);
2775 		data_len -= size;
2776 
2777 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
2778 				       DMA_TO_DEVICE);
2779 
2780 		tx_bi = &tx_ring->tx_bi[i];
2781 	}
2782 
2783 	/* set next_to_watch value indicating a packet is present */
2784 	first->next_to_watch = tx_desc;
2785 
2786 	i++;
2787 	if (i == tx_ring->count)
2788 		i = 0;
2789 
2790 	tx_ring->next_to_use = i;
2791 
2792 	netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev,
2793 						 tx_ring->queue_index),
2794 						 first->bytecount);
2795 	i40e_maybe_stop_tx(tx_ring, DESC_NEEDED);
2796 
2797 	/* Algorithm to optimize tail and RS bit setting:
2798 	 * if xmit_more is supported
2799 	 *	if xmit_more is true
2800 	 *		do not update tail and do not mark RS bit.
2801 	 *	if xmit_more is false and last xmit_more was false
2802 	 *		if every packet spanned less than 4 desc
2803 	 *			then set RS bit on 4th packet and update tail
2804 	 *			on every packet
2805 	 *		else
2806 	 *			update tail and set RS bit on every packet.
2807 	 *	if xmit_more is false and last_xmit_more was true
2808 	 *		update tail and set RS bit.
2809 	 *
2810 	 * Optimization: wmb to be issued only in case of tail update.
2811 	 * Also optimize the Descriptor WB path for RS bit with the same
2812 	 * algorithm.
2813 	 *
2814 	 * Note: If there are less than 4 packets
2815 	 * pending and interrupts were disabled the service task will
2816 	 * trigger a force WB.
2817 	 */
2818 	if (skb->xmit_more  &&
2819 	    !netif_xmit_stopped(netdev_get_tx_queue(tx_ring->netdev,
2820 						    tx_ring->queue_index))) {
2821 		tx_ring->flags |= I40E_TXR_FLAGS_LAST_XMIT_MORE_SET;
2822 		tail_bump = false;
2823 	} else if (!skb->xmit_more &&
2824 		   !netif_xmit_stopped(netdev_get_tx_queue(tx_ring->netdev,
2825 						       tx_ring->queue_index)) &&
2826 		   (!(tx_ring->flags & I40E_TXR_FLAGS_LAST_XMIT_MORE_SET)) &&
2827 		   (tx_ring->packet_stride < WB_STRIDE) &&
2828 		   (desc_count < WB_STRIDE)) {
2829 		tx_ring->packet_stride++;
2830 	} else {
2831 		tx_ring->packet_stride = 0;
2832 		tx_ring->flags &= ~I40E_TXR_FLAGS_LAST_XMIT_MORE_SET;
2833 		do_rs = true;
2834 	}
2835 	if (do_rs)
2836 		tx_ring->packet_stride = 0;
2837 
2838 	tx_desc->cmd_type_offset_bsz =
2839 			build_ctob(td_cmd, td_offset, size, td_tag) |
2840 			cpu_to_le64((u64)(do_rs ? I40E_TXD_CMD :
2841 						  I40E_TX_DESC_CMD_EOP) <<
2842 						  I40E_TXD_QW1_CMD_SHIFT);
2843 
2844 	/* notify HW of packet */
2845 	if (!tail_bump)
2846 		prefetchw(tx_desc + 1);
2847 
2848 	if (tail_bump) {
2849 		/* Force memory writes to complete before letting h/w
2850 		 * know there are new descriptors to fetch.  (Only
2851 		 * applicable for weak-ordered memory model archs,
2852 		 * such as IA-64).
2853 		 */
2854 		wmb();
2855 		writel(i, tx_ring->tail);
2856 	}
2857 
2858 	return;
2859 
2860 dma_error:
2861 	dev_info(tx_ring->dev, "TX DMA map failed\n");
2862 
2863 	/* clear dma mappings for failed tx_bi map */
2864 	for (;;) {
2865 		tx_bi = &tx_ring->tx_bi[i];
2866 		i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
2867 		if (tx_bi == first)
2868 			break;
2869 		if (i == 0)
2870 			i = tx_ring->count;
2871 		i--;
2872 	}
2873 
2874 	tx_ring->next_to_use = i;
2875 }
2876 
2877 /**
2878  * i40e_xmit_frame_ring - Sends buffer on Tx ring
2879  * @skb:     send buffer
2880  * @tx_ring: ring to send buffer on
2881  *
2882  * Returns NETDEV_TX_OK if sent, else an error code
2883  **/
2884 static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
2885 					struct i40e_ring *tx_ring)
2886 {
2887 	u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
2888 	u32 cd_tunneling = 0, cd_l2tag2 = 0;
2889 	struct i40e_tx_buffer *first;
2890 	u32 td_offset = 0;
2891 	u32 tx_flags = 0;
2892 	__be16 protocol;
2893 	u32 td_cmd = 0;
2894 	u8 hdr_len = 0;
2895 	int tso, count;
2896 	int tsyn;
2897 
2898 	/* prefetch the data, we'll need it later */
2899 	prefetch(skb->data);
2900 
2901 	count = i40e_xmit_descriptor_count(skb);
2902 	if (i40e_chk_linearize(skb, count)) {
2903 		if (__skb_linearize(skb))
2904 			goto out_drop;
2905 		count = i40e_txd_use_count(skb->len);
2906 		tx_ring->tx_stats.tx_linearize++;
2907 	}
2908 
2909 	/* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
2910 	 *       + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
2911 	 *       + 4 desc gap to avoid the cache line where head is,
2912 	 *       + 1 desc for context descriptor,
2913 	 * otherwise try next time
2914 	 */
2915 	if (i40e_maybe_stop_tx(tx_ring, count + 4 + 1)) {
2916 		tx_ring->tx_stats.tx_busy++;
2917 		return NETDEV_TX_BUSY;
2918 	}
2919 
2920 	/* prepare the xmit flags */
2921 	if (i40e_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
2922 		goto out_drop;
2923 
2924 	/* obtain protocol of skb */
2925 	protocol = vlan_get_protocol(skb);
2926 
2927 	/* record the location of the first descriptor for this packet */
2928 	first = &tx_ring->tx_bi[tx_ring->next_to_use];
2929 
2930 	/* setup IPv4/IPv6 offloads */
2931 	if (protocol == htons(ETH_P_IP))
2932 		tx_flags |= I40E_TX_FLAGS_IPV4;
2933 	else if (protocol == htons(ETH_P_IPV6))
2934 		tx_flags |= I40E_TX_FLAGS_IPV6;
2935 
2936 	tso = i40e_tso(skb, &hdr_len, &cd_type_cmd_tso_mss);
2937 
2938 	if (tso < 0)
2939 		goto out_drop;
2940 	else if (tso)
2941 		tx_flags |= I40E_TX_FLAGS_TSO;
2942 
2943 	/* Always offload the checksum, since it's in the data descriptor */
2944 	tso = i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset,
2945 				  tx_ring, &cd_tunneling);
2946 	if (tso < 0)
2947 		goto out_drop;
2948 
2949 	tsyn = i40e_tsyn(tx_ring, skb, tx_flags, &cd_type_cmd_tso_mss);
2950 
2951 	if (tsyn)
2952 		tx_flags |= I40E_TX_FLAGS_TSYN;
2953 
2954 	skb_tx_timestamp(skb);
2955 
2956 	/* always enable CRC insertion offload */
2957 	td_cmd |= I40E_TX_DESC_CMD_ICRC;
2958 
2959 	i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
2960 			   cd_tunneling, cd_l2tag2);
2961 
2962 	/* Add Flow Director ATR if it's enabled.
2963 	 *
2964 	 * NOTE: this must always be directly before the data descriptor.
2965 	 */
2966 	i40e_atr(tx_ring, skb, tx_flags);
2967 
2968 	i40e_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
2969 		    td_cmd, td_offset);
2970 
2971 	return NETDEV_TX_OK;
2972 
2973 out_drop:
2974 	dev_kfree_skb_any(skb);
2975 	return NETDEV_TX_OK;
2976 }
2977 
2978 /**
2979  * i40e_lan_xmit_frame - Selects the correct VSI and Tx queue to send buffer
2980  * @skb:    send buffer
2981  * @netdev: network interface device structure
2982  *
2983  * Returns NETDEV_TX_OK if sent, else an error code
2984  **/
2985 netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2986 {
2987 	struct i40e_netdev_priv *np = netdev_priv(netdev);
2988 	struct i40e_vsi *vsi = np->vsi;
2989 	struct i40e_ring *tx_ring = vsi->tx_rings[skb->queue_mapping];
2990 
2991 	/* hardware can't handle really short frames, hardware padding works
2992 	 * beyond this point
2993 	 */
2994 	if (skb_put_padto(skb, I40E_MIN_TX_LEN))
2995 		return NETDEV_TX_OK;
2996 
2997 	return i40e_xmit_frame_ring(skb, tx_ring);
2998 }
2999