1 /*******************************************************************************
2  *
3  * Intel Ethernet Controller XL710 Family Linux Driver
4  * Copyright(c) 2013 - 2014 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 "i40e.h"
28 #include "i40e_prototype.h"
29 
30 static inline __le64 build_ctob(u32 td_cmd, u32 td_offset, unsigned int size,
31 				u32 td_tag)
32 {
33 	return cpu_to_le64(I40E_TX_DESC_DTYPE_DATA |
34 			   ((u64)td_cmd  << I40E_TXD_QW1_CMD_SHIFT) |
35 			   ((u64)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) |
36 			   ((u64)size  << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) |
37 			   ((u64)td_tag  << I40E_TXD_QW1_L2TAG1_SHIFT));
38 }
39 
40 #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
41 /**
42  * i40e_program_fdir_filter - Program a Flow Director filter
43  * @fdir_data: Packet data that will be filter parameters
44  * @raw_packet: the pre-allocated packet buffer for FDir
45  * @pf: The pf pointer
46  * @add: True for add/update, False for remove
47  **/
48 int i40e_program_fdir_filter(struct i40e_fdir_filter *fdir_data, u8 *raw_packet,
49 			     struct i40e_pf *pf, bool add)
50 {
51 	struct i40e_filter_program_desc *fdir_desc;
52 	struct i40e_tx_buffer *tx_buf;
53 	struct i40e_tx_desc *tx_desc;
54 	struct i40e_ring *tx_ring;
55 	unsigned int fpt, dcc;
56 	struct i40e_vsi *vsi;
57 	struct device *dev;
58 	dma_addr_t dma;
59 	u32 td_cmd = 0;
60 	u16 i;
61 
62 	/* find existing FDIR VSI */
63 	vsi = NULL;
64 	for (i = 0; i < pf->hw.func_caps.num_vsis; i++)
65 		if (pf->vsi[i] && pf->vsi[i]->type == I40E_VSI_FDIR)
66 			vsi = pf->vsi[i];
67 	if (!vsi)
68 		return -ENOENT;
69 
70 	tx_ring = vsi->tx_rings[0];
71 	dev = tx_ring->dev;
72 
73 	dma = dma_map_single(dev, raw_packet,
74 			     I40E_FDIR_MAX_RAW_PACKET_SIZE, DMA_TO_DEVICE);
75 	if (dma_mapping_error(dev, dma))
76 		goto dma_fail;
77 
78 	/* grab the next descriptor */
79 	i = tx_ring->next_to_use;
80 	fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
81 
82 	tx_ring->next_to_use = (i + 1 < tx_ring->count) ? i + 1 : 0;
83 
84 	fpt = (fdir_data->q_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) &
85 	      I40E_TXD_FLTR_QW0_QINDEX_MASK;
86 
87 	fpt |= (fdir_data->flex_off << I40E_TXD_FLTR_QW0_FLEXOFF_SHIFT) &
88 	       I40E_TXD_FLTR_QW0_FLEXOFF_MASK;
89 
90 	fpt |= (fdir_data->pctype << I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) &
91 	       I40E_TXD_FLTR_QW0_PCTYPE_MASK;
92 
93 	/* Use LAN VSI Id if not programmed by user */
94 	if (fdir_data->dest_vsi == 0)
95 		fpt |= (pf->vsi[pf->lan_vsi]->id) <<
96 		       I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT;
97 	else
98 		fpt |= ((u32)fdir_data->dest_vsi <<
99 			I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT) &
100 		       I40E_TXD_FLTR_QW0_DEST_VSI_MASK;
101 
102 	fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(fpt);
103 
104 	dcc = I40E_TX_DESC_DTYPE_FILTER_PROG;
105 
106 	if (add)
107 		dcc |= I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
108 		       I40E_TXD_FLTR_QW1_PCMD_SHIFT;
109 	else
110 		dcc |= I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
111 		       I40E_TXD_FLTR_QW1_PCMD_SHIFT;
112 
113 	dcc |= (fdir_data->dest_ctl << I40E_TXD_FLTR_QW1_DEST_SHIFT) &
114 	       I40E_TXD_FLTR_QW1_DEST_MASK;
115 
116 	dcc |= (fdir_data->fd_status << I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT) &
117 	       I40E_TXD_FLTR_QW1_FD_STATUS_MASK;
118 
119 	if (fdir_data->cnt_index != 0) {
120 		dcc |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
121 		dcc |= ((u32)fdir_data->cnt_index <<
122 			I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
123 		       I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
124 	}
125 
126 	fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dcc);
127 	fdir_desc->fd_id = cpu_to_le32(fdir_data->fd_id);
128 
129 	/* Now program a dummy descriptor */
130 	i = tx_ring->next_to_use;
131 	tx_desc = I40E_TX_DESC(tx_ring, i);
132 	tx_buf = &tx_ring->tx_bi[i];
133 
134 	tx_ring->next_to_use = (i + 1 < tx_ring->count) ? i + 1 : 0;
135 
136 	/* record length, and DMA address */
137 	dma_unmap_len_set(tx_buf, len, I40E_FDIR_MAX_RAW_PACKET_SIZE);
138 	dma_unmap_addr_set(tx_buf, dma, dma);
139 
140 	tx_desc->buffer_addr = cpu_to_le64(dma);
141 	td_cmd = I40E_TXD_CMD | I40E_TX_DESC_CMD_DUMMY;
142 
143 	tx_desc->cmd_type_offset_bsz =
144 		build_ctob(td_cmd, 0, I40E_FDIR_MAX_RAW_PACKET_SIZE, 0);
145 
146 	/* set the timestamp */
147 	tx_buf->time_stamp = jiffies;
148 
149 	/* Force memory writes to complete before letting h/w
150 	 * know there are new descriptors to fetch.  (Only
151 	 * applicable for weak-ordered memory model archs,
152 	 * such as IA-64).
153 	 */
154 	wmb();
155 
156 	/* Mark the data descriptor to be watched */
157 	tx_buf->next_to_watch = tx_desc;
158 
159 	writel(tx_ring->next_to_use, tx_ring->tail);
160 	return 0;
161 
162 dma_fail:
163 	return -1;
164 }
165 
166 #define IP_HEADER_OFFSET 14
167 #define I40E_UDPIP_DUMMY_PACKET_LEN 42
168 /**
169  * i40e_add_del_fdir_udpv4 - Add/Remove UDPv4 filters
170  * @vsi: pointer to the targeted VSI
171  * @fd_data: the flow director data required for the FDir descriptor
172  * @raw_packet: the pre-allocated packet buffer for FDir
173  * @add: true adds a filter, false removes it
174  *
175  * Returns 0 if the filters were successfully added or removed
176  **/
177 static int i40e_add_del_fdir_udpv4(struct i40e_vsi *vsi,
178 				   struct i40e_fdir_filter *fd_data,
179 				   u8 *raw_packet, bool add)
180 {
181 	struct i40e_pf *pf = vsi->back;
182 	struct udphdr *udp;
183 	struct iphdr *ip;
184 	bool err = false;
185 	int ret;
186 	int i;
187 	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
188 		0x45, 0, 0, 0x1c, 0, 0, 0x40, 0, 0x40, 0x11, 0, 0, 0, 0, 0, 0,
189 		0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
190 
191 	memcpy(raw_packet, packet, I40E_UDPIP_DUMMY_PACKET_LEN);
192 
193 	ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
194 	udp = (struct udphdr *)(raw_packet + IP_HEADER_OFFSET
195 	      + sizeof(struct iphdr));
196 
197 	ip->daddr = fd_data->dst_ip[0];
198 	udp->dest = fd_data->dst_port;
199 	ip->saddr = fd_data->src_ip[0];
200 	udp->source = fd_data->src_port;
201 
202 	for (i = I40E_FILTER_PCTYPE_NONF_UNICAST_IPV4_UDP;
203 	     i <= I40E_FILTER_PCTYPE_NONF_IPV4_UDP; i++) {
204 		fd_data->pctype = i;
205 		ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
206 
207 		if (ret) {
208 			dev_info(&pf->pdev->dev,
209 				 "Filter command send failed for PCTYPE %d (ret = %d)\n",
210 				 fd_data->pctype, ret);
211 			err = true;
212 		} else {
213 			dev_info(&pf->pdev->dev,
214 				 "Filter OK for PCTYPE %d (ret = %d)\n",
215 				 fd_data->pctype, ret);
216 		}
217 	}
218 
219 	return err ? -EOPNOTSUPP : 0;
220 }
221 
222 #define I40E_TCPIP_DUMMY_PACKET_LEN 54
223 /**
224  * i40e_add_del_fdir_tcpv4 - Add/Remove TCPv4 filters
225  * @vsi: pointer to the targeted VSI
226  * @fd_data: the flow director data required for the FDir descriptor
227  * @raw_packet: the pre-allocated packet buffer for FDir
228  * @add: true adds a filter, false removes it
229  *
230  * Returns 0 if the filters were successfully added or removed
231  **/
232 static int i40e_add_del_fdir_tcpv4(struct i40e_vsi *vsi,
233 				   struct i40e_fdir_filter *fd_data,
234 				   u8 *raw_packet, bool add)
235 {
236 	struct i40e_pf *pf = vsi->back;
237 	struct tcphdr *tcp;
238 	struct iphdr *ip;
239 	bool err = false;
240 	int ret;
241 	/* Dummy packet */
242 	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
243 		0x45, 0, 0, 0x28, 0, 0, 0x40, 0, 0x40, 0x6, 0, 0, 0, 0, 0, 0,
244 		0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x80, 0x11,
245 		0x0, 0x72, 0, 0, 0, 0};
246 
247 	memcpy(raw_packet, packet, I40E_TCPIP_DUMMY_PACKET_LEN);
248 
249 	ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
250 	tcp = (struct tcphdr *)(raw_packet + IP_HEADER_OFFSET
251 	      + sizeof(struct iphdr));
252 
253 	ip->daddr = fd_data->dst_ip[0];
254 	tcp->dest = fd_data->dst_port;
255 	ip->saddr = fd_data->src_ip[0];
256 	tcp->source = fd_data->src_port;
257 
258 	if (add) {
259 		if (pf->flags & I40E_FLAG_FD_ATR_ENABLED) {
260 			dev_info(&pf->pdev->dev, "Forcing ATR off, sideband rules for TCP/IPv4 flow being applied\n");
261 			pf->flags &= ~I40E_FLAG_FD_ATR_ENABLED;
262 		}
263 	}
264 
265 	fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_TCP_SYN;
266 	ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
267 
268 	if (ret) {
269 		dev_info(&pf->pdev->dev,
270 			 "Filter command send failed for PCTYPE %d (ret = %d)\n",
271 			 fd_data->pctype, ret);
272 		err = true;
273 	} else {
274 		dev_info(&pf->pdev->dev, "Filter OK for PCTYPE %d (ret = %d)\n",
275 			 fd_data->pctype, ret);
276 	}
277 
278 	fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_TCP;
279 
280 	ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
281 	if (ret) {
282 		dev_info(&pf->pdev->dev,
283 			 "Filter command send failed for PCTYPE %d (ret = %d)\n",
284 			 fd_data->pctype, ret);
285 		err = true;
286 	} else {
287 		dev_info(&pf->pdev->dev, "Filter OK for PCTYPE %d (ret = %d)\n",
288 			  fd_data->pctype, ret);
289 	}
290 
291 	return err ? -EOPNOTSUPP : 0;
292 }
293 
294 /**
295  * i40e_add_del_fdir_sctpv4 - Add/Remove SCTPv4 Flow Director filters for
296  * a specific flow spec
297  * @vsi: pointer to the targeted VSI
298  * @fd_data: the flow director data required for the FDir descriptor
299  * @raw_packet: the pre-allocated packet buffer for FDir
300  * @add: true adds a filter, false removes it
301  *
302  * Always returns -EOPNOTSUPP
303  **/
304 static int i40e_add_del_fdir_sctpv4(struct i40e_vsi *vsi,
305 				    struct i40e_fdir_filter *fd_data,
306 				    u8 *raw_packet, bool add)
307 {
308 	return -EOPNOTSUPP;
309 }
310 
311 #define I40E_IP_DUMMY_PACKET_LEN 34
312 /**
313  * i40e_add_del_fdir_ipv4 - Add/Remove IPv4 Flow Director filters for
314  * a specific flow spec
315  * @vsi: pointer to the targeted VSI
316  * @fd_data: the flow director data required for the FDir descriptor
317  * @raw_packet: the pre-allocated packet buffer for FDir
318  * @add: true adds a filter, false removes it
319  *
320  * Returns 0 if the filters were successfully added or removed
321  **/
322 static int i40e_add_del_fdir_ipv4(struct i40e_vsi *vsi,
323 				  struct i40e_fdir_filter *fd_data,
324 				  u8 *raw_packet, bool add)
325 {
326 	struct i40e_pf *pf = vsi->back;
327 	struct iphdr *ip;
328 	bool err = false;
329 	int ret;
330 	int i;
331 	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
332 		0x45, 0, 0, 0x14, 0, 0, 0x40, 0, 0x40, 0x10, 0, 0, 0, 0, 0, 0,
333 		0, 0, 0, 0};
334 
335 	memcpy(raw_packet, packet, I40E_IP_DUMMY_PACKET_LEN);
336 	ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
337 
338 	ip->saddr = fd_data->src_ip[0];
339 	ip->daddr = fd_data->dst_ip[0];
340 	ip->protocol = 0;
341 
342 	for (i = I40E_FILTER_PCTYPE_NONF_IPV4_OTHER;
343 	     i <= I40E_FILTER_PCTYPE_FRAG_IPV4;	i++) {
344 		fd_data->pctype = i;
345 		ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
346 
347 		if (ret) {
348 			dev_info(&pf->pdev->dev,
349 				 "Filter command send failed for PCTYPE %d (ret = %d)\n",
350 				 fd_data->pctype, ret);
351 			err = true;
352 		} else {
353 			dev_info(&pf->pdev->dev,
354 				 "Filter OK for PCTYPE %d (ret = %d)\n",
355 				 fd_data->pctype, ret);
356 		}
357 	}
358 
359 	return err ? -EOPNOTSUPP : 0;
360 }
361 
362 /**
363  * i40e_add_del_fdir - Build raw packets to add/del fdir filter
364  * @vsi: pointer to the targeted VSI
365  * @cmd: command to get or set RX flow classification rules
366  * @add: true adds a filter, false removes it
367  *
368  **/
369 int i40e_add_del_fdir(struct i40e_vsi *vsi,
370 		      struct i40e_fdir_filter *input, bool add)
371 {
372 	struct i40e_pf *pf = vsi->back;
373 	u8 *raw_packet;
374 	int ret;
375 
376 	/* Populate the Flow Director that we have at the moment
377 	 * and allocate the raw packet buffer for the calling functions
378 	 */
379 	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
380 	if (!raw_packet)
381 		return -ENOMEM;
382 
383 	switch (input->flow_type & ~FLOW_EXT) {
384 	case TCP_V4_FLOW:
385 		ret = i40e_add_del_fdir_tcpv4(vsi, input, raw_packet,
386 					      add);
387 		break;
388 	case UDP_V4_FLOW:
389 		ret = i40e_add_del_fdir_udpv4(vsi, input, raw_packet,
390 					      add);
391 		break;
392 	case SCTP_V4_FLOW:
393 		ret = i40e_add_del_fdir_sctpv4(vsi, input, raw_packet,
394 					       add);
395 		break;
396 	case IPV4_FLOW:
397 		ret = i40e_add_del_fdir_ipv4(vsi, input, raw_packet,
398 					     add);
399 		break;
400 	case IP_USER_FLOW:
401 		switch (input->ip4_proto) {
402 		case IPPROTO_TCP:
403 			ret = i40e_add_del_fdir_tcpv4(vsi, input,
404 						      raw_packet, add);
405 			break;
406 		case IPPROTO_UDP:
407 			ret = i40e_add_del_fdir_udpv4(vsi, input,
408 						      raw_packet, add);
409 			break;
410 		case IPPROTO_SCTP:
411 			ret = i40e_add_del_fdir_sctpv4(vsi, input,
412 						       raw_packet, add);
413 			break;
414 		default:
415 			ret = i40e_add_del_fdir_ipv4(vsi, input,
416 						     raw_packet, add);
417 			break;
418 		}
419 		break;
420 	default:
421 		dev_info(&pf->pdev->dev, "Could not specify spec type %d",
422 			 input->flow_type);
423 		ret = -EINVAL;
424 	}
425 
426 	kfree(raw_packet);
427 	return ret;
428 }
429 
430 /**
431  * i40e_fd_handle_status - check the Programming Status for FD
432  * @rx_ring: the Rx ring for this descriptor
433  * @rx_desc: the Rx descriptor for programming Status, not a packet descriptor.
434  * @prog_id: the id originally used for programming
435  *
436  * This is used to verify if the FD programming or invalidation
437  * requested by SW to the HW is successful or not and take actions accordingly.
438  **/
439 static void i40e_fd_handle_status(struct i40e_ring *rx_ring,
440 				  union i40e_rx_desc *rx_desc, u8 prog_id)
441 {
442 	struct i40e_pf *pf = rx_ring->vsi->back;
443 	struct pci_dev *pdev = pf->pdev;
444 	u32 fcnt_prog, fcnt_avail;
445 	u32 error;
446 	u64 qw;
447 
448 	qw = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
449 	error = (qw & I40E_RX_PROG_STATUS_DESC_QW1_ERROR_MASK) >>
450 		I40E_RX_PROG_STATUS_DESC_QW1_ERROR_SHIFT;
451 
452 	if (error == (0x1 << I40E_RX_PROG_STATUS_DESC_FD_TBL_FULL_SHIFT)) {
453 		dev_warn(&pdev->dev, "ntuple filter loc = %d, could not be added\n",
454 			 rx_desc->wb.qword0.hi_dword.fd_id);
455 
456 		/* filter programming failed most likely due to table full */
457 		fcnt_prog = i40e_get_current_fd_count(pf);
458 		fcnt_avail = pf->hw.fdir_shared_filter_count +
459 						       pf->fdir_pf_filter_count;
460 
461 		/* If ATR is running fcnt_prog can quickly change,
462 		 * if we are very close to full, it makes sense to disable
463 		 * FD ATR/SB and then re-enable it when there is room.
464 		 */
465 		if (fcnt_prog >= (fcnt_avail - I40E_FDIR_BUFFER_FULL_MARGIN)) {
466 			/* Turn off ATR first */
467 			if (pf->flags | I40E_FLAG_FD_ATR_ENABLED) {
468 				pf->flags &= ~I40E_FLAG_FD_ATR_ENABLED;
469 				dev_warn(&pdev->dev, "FD filter space full, ATR for further flows will be turned off\n");
470 				pf->auto_disable_flags |=
471 						       I40E_FLAG_FD_ATR_ENABLED;
472 				pf->flags |= I40E_FLAG_FDIR_REQUIRES_REINIT;
473 			} else if (pf->flags | I40E_FLAG_FD_SB_ENABLED) {
474 				pf->flags &= ~I40E_FLAG_FD_SB_ENABLED;
475 				dev_warn(&pdev->dev, "FD filter space full, new ntuple rules will not be added\n");
476 				pf->auto_disable_flags |=
477 							I40E_FLAG_FD_SB_ENABLED;
478 				pf->flags |= I40E_FLAG_FDIR_REQUIRES_REINIT;
479 			}
480 		} else {
481 			dev_info(&pdev->dev, "FD filter programming error");
482 		}
483 	} else if (error ==
484 			  (0x1 << I40E_RX_PROG_STATUS_DESC_NO_FD_ENTRY_SHIFT)) {
485 		if (I40E_DEBUG_FD & pf->hw.debug_mask)
486 			dev_info(&pdev->dev, "ntuple filter loc = %d, could not be removed\n",
487 				 rx_desc->wb.qword0.hi_dword.fd_id);
488 	}
489 }
490 
491 /**
492  * i40e_unmap_and_free_tx_resource - Release a Tx buffer
493  * @ring:      the ring that owns the buffer
494  * @tx_buffer: the buffer to free
495  **/
496 static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring,
497 					    struct i40e_tx_buffer *tx_buffer)
498 {
499 	if (tx_buffer->skb) {
500 		dev_kfree_skb_any(tx_buffer->skb);
501 		if (dma_unmap_len(tx_buffer, len))
502 			dma_unmap_single(ring->dev,
503 					 dma_unmap_addr(tx_buffer, dma),
504 					 dma_unmap_len(tx_buffer, len),
505 					 DMA_TO_DEVICE);
506 	} else if (dma_unmap_len(tx_buffer, len)) {
507 		dma_unmap_page(ring->dev,
508 			       dma_unmap_addr(tx_buffer, dma),
509 			       dma_unmap_len(tx_buffer, len),
510 			       DMA_TO_DEVICE);
511 	}
512 	tx_buffer->next_to_watch = NULL;
513 	tx_buffer->skb = NULL;
514 	dma_unmap_len_set(tx_buffer, len, 0);
515 	/* tx_buffer must be completely set up in the transmit path */
516 }
517 
518 /**
519  * i40e_clean_tx_ring - Free any empty Tx buffers
520  * @tx_ring: ring to be cleaned
521  **/
522 void i40e_clean_tx_ring(struct i40e_ring *tx_ring)
523 {
524 	unsigned long bi_size;
525 	u16 i;
526 
527 	/* ring already cleared, nothing to do */
528 	if (!tx_ring->tx_bi)
529 		return;
530 
531 	/* Free all the Tx ring sk_buffs */
532 	for (i = 0; i < tx_ring->count; i++)
533 		i40e_unmap_and_free_tx_resource(tx_ring, &tx_ring->tx_bi[i]);
534 
535 	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
536 	memset(tx_ring->tx_bi, 0, bi_size);
537 
538 	/* Zero out the descriptor ring */
539 	memset(tx_ring->desc, 0, tx_ring->size);
540 
541 	tx_ring->next_to_use = 0;
542 	tx_ring->next_to_clean = 0;
543 
544 	if (!tx_ring->netdev)
545 		return;
546 
547 	/* cleanup Tx queue statistics */
548 	netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
549 						  tx_ring->queue_index));
550 }
551 
552 /**
553  * i40e_free_tx_resources - Free Tx resources per queue
554  * @tx_ring: Tx descriptor ring for a specific queue
555  *
556  * Free all transmit software resources
557  **/
558 void i40e_free_tx_resources(struct i40e_ring *tx_ring)
559 {
560 	i40e_clean_tx_ring(tx_ring);
561 	kfree(tx_ring->tx_bi);
562 	tx_ring->tx_bi = NULL;
563 
564 	if (tx_ring->desc) {
565 		dma_free_coherent(tx_ring->dev, tx_ring->size,
566 				  tx_ring->desc, tx_ring->dma);
567 		tx_ring->desc = NULL;
568 	}
569 }
570 
571 /**
572  * i40e_get_tx_pending - how many tx descriptors not processed
573  * @tx_ring: the ring of descriptors
574  *
575  * Since there is no access to the ring head register
576  * in XL710, we need to use our local copies
577  **/
578 static u32 i40e_get_tx_pending(struct i40e_ring *ring)
579 {
580 	u32 ntu = ((ring->next_to_clean <= ring->next_to_use)
581 			? ring->next_to_use
582 			: ring->next_to_use + ring->count);
583 	return ntu - ring->next_to_clean;
584 }
585 
586 /**
587  * i40e_check_tx_hang - Is there a hang in the Tx queue
588  * @tx_ring: the ring of descriptors
589  **/
590 static bool i40e_check_tx_hang(struct i40e_ring *tx_ring)
591 {
592 	u32 tx_pending = i40e_get_tx_pending(tx_ring);
593 	bool ret = false;
594 
595 	clear_check_for_tx_hang(tx_ring);
596 
597 	/* Check for a hung queue, but be thorough. This verifies
598 	 * that a transmit has been completed since the previous
599 	 * check AND there is at least one packet pending. The
600 	 * ARMED bit is set to indicate a potential hang. The
601 	 * bit is cleared if a pause frame is received to remove
602 	 * false hang detection due to PFC or 802.3x frames. By
603 	 * requiring this to fail twice we avoid races with
604 	 * PFC clearing the ARMED bit and conditions where we
605 	 * run the check_tx_hang logic with a transmit completion
606 	 * pending but without time to complete it yet.
607 	 */
608 	if ((tx_ring->tx_stats.tx_done_old == tx_ring->stats.packets) &&
609 	    tx_pending) {
610 		/* make sure it is true for two checks in a row */
611 		ret = test_and_set_bit(__I40E_HANG_CHECK_ARMED,
612 				       &tx_ring->state);
613 	} else {
614 		/* update completed stats and disarm the hang check */
615 		tx_ring->tx_stats.tx_done_old = tx_ring->stats.packets;
616 		clear_bit(__I40E_HANG_CHECK_ARMED, &tx_ring->state);
617 	}
618 
619 	return ret;
620 }
621 
622 /**
623  * i40e_get_head - Retrieve head from head writeback
624  * @tx_ring:  tx ring to fetch head of
625  *
626  * Returns value of Tx ring head based on value stored
627  * in head write-back location
628  **/
629 static inline u32 i40e_get_head(struct i40e_ring *tx_ring)
630 {
631 	void *head = (struct i40e_tx_desc *)tx_ring->desc + tx_ring->count;
632 
633 	return le32_to_cpu(*(volatile __le32 *)head);
634 }
635 
636 /**
637  * i40e_clean_tx_irq - Reclaim resources after transmit completes
638  * @tx_ring:  tx ring to clean
639  * @budget:   how many cleans we're allowed
640  *
641  * Returns true if there's any budget left (e.g. the clean is finished)
642  **/
643 static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget)
644 {
645 	u16 i = tx_ring->next_to_clean;
646 	struct i40e_tx_buffer *tx_buf;
647 	struct i40e_tx_desc *tx_head;
648 	struct i40e_tx_desc *tx_desc;
649 	unsigned int total_packets = 0;
650 	unsigned int total_bytes = 0;
651 
652 	tx_buf = &tx_ring->tx_bi[i];
653 	tx_desc = I40E_TX_DESC(tx_ring, i);
654 	i -= tx_ring->count;
655 
656 	tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
657 
658 	do {
659 		struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
660 
661 		/* if next_to_watch is not set then there is no work pending */
662 		if (!eop_desc)
663 			break;
664 
665 		/* prevent any other reads prior to eop_desc */
666 		read_barrier_depends();
667 
668 		/* we have caught up to head, no work left to do */
669 		if (tx_head == tx_desc)
670 			break;
671 
672 		/* clear next_to_watch to prevent false hangs */
673 		tx_buf->next_to_watch = NULL;
674 
675 		/* update the statistics for this packet */
676 		total_bytes += tx_buf->bytecount;
677 		total_packets += tx_buf->gso_segs;
678 
679 		/* free the skb */
680 		dev_kfree_skb_any(tx_buf->skb);
681 
682 		/* unmap skb header data */
683 		dma_unmap_single(tx_ring->dev,
684 				 dma_unmap_addr(tx_buf, dma),
685 				 dma_unmap_len(tx_buf, len),
686 				 DMA_TO_DEVICE);
687 
688 		/* clear tx_buffer data */
689 		tx_buf->skb = NULL;
690 		dma_unmap_len_set(tx_buf, len, 0);
691 
692 		/* unmap remaining buffers */
693 		while (tx_desc != eop_desc) {
694 
695 			tx_buf++;
696 			tx_desc++;
697 			i++;
698 			if (unlikely(!i)) {
699 				i -= tx_ring->count;
700 				tx_buf = tx_ring->tx_bi;
701 				tx_desc = I40E_TX_DESC(tx_ring, 0);
702 			}
703 
704 			/* unmap any remaining paged data */
705 			if (dma_unmap_len(tx_buf, len)) {
706 				dma_unmap_page(tx_ring->dev,
707 					       dma_unmap_addr(tx_buf, dma),
708 					       dma_unmap_len(tx_buf, len),
709 					       DMA_TO_DEVICE);
710 				dma_unmap_len_set(tx_buf, len, 0);
711 			}
712 		}
713 
714 		/* move us one more past the eop_desc for start of next pkt */
715 		tx_buf++;
716 		tx_desc++;
717 		i++;
718 		if (unlikely(!i)) {
719 			i -= tx_ring->count;
720 			tx_buf = tx_ring->tx_bi;
721 			tx_desc = I40E_TX_DESC(tx_ring, 0);
722 		}
723 
724 		/* update budget accounting */
725 		budget--;
726 	} while (likely(budget));
727 
728 	i += tx_ring->count;
729 	tx_ring->next_to_clean = i;
730 	u64_stats_update_begin(&tx_ring->syncp);
731 	tx_ring->stats.bytes += total_bytes;
732 	tx_ring->stats.packets += total_packets;
733 	u64_stats_update_end(&tx_ring->syncp);
734 	tx_ring->q_vector->tx.total_bytes += total_bytes;
735 	tx_ring->q_vector->tx.total_packets += total_packets;
736 
737 	if (check_for_tx_hang(tx_ring) && i40e_check_tx_hang(tx_ring)) {
738 		/* schedule immediate reset if we believe we hung */
739 		dev_info(tx_ring->dev, "Detected Tx Unit Hang\n"
740 			 "  VSI                  <%d>\n"
741 			 "  Tx Queue             <%d>\n"
742 			 "  next_to_use          <%x>\n"
743 			 "  next_to_clean        <%x>\n",
744 			 tx_ring->vsi->seid,
745 			 tx_ring->queue_index,
746 			 tx_ring->next_to_use, i);
747 		dev_info(tx_ring->dev, "tx_bi[next_to_clean]\n"
748 			 "  time_stamp           <%lx>\n"
749 			 "  jiffies              <%lx>\n",
750 			 tx_ring->tx_bi[i].time_stamp, jiffies);
751 
752 		netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
753 
754 		dev_info(tx_ring->dev,
755 			 "tx hang detected on queue %d, resetting adapter\n",
756 			 tx_ring->queue_index);
757 
758 		tx_ring->netdev->netdev_ops->ndo_tx_timeout(tx_ring->netdev);
759 
760 		/* the adapter is about to reset, no point in enabling stuff */
761 		return true;
762 	}
763 
764 	netdev_tx_completed_queue(netdev_get_tx_queue(tx_ring->netdev,
765 						      tx_ring->queue_index),
766 				  total_packets, total_bytes);
767 
768 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
769 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
770 		     (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
771 		/* Make sure that anybody stopping the queue after this
772 		 * sees the new next_to_clean.
773 		 */
774 		smp_mb();
775 		if (__netif_subqueue_stopped(tx_ring->netdev,
776 					     tx_ring->queue_index) &&
777 		   !test_bit(__I40E_DOWN, &tx_ring->vsi->state)) {
778 			netif_wake_subqueue(tx_ring->netdev,
779 					    tx_ring->queue_index);
780 			++tx_ring->tx_stats.restart_queue;
781 		}
782 	}
783 
784 	return budget > 0;
785 }
786 
787 /**
788  * i40e_set_new_dynamic_itr - Find new ITR level
789  * @rc: structure containing ring performance data
790  *
791  * Stores a new ITR value based on packets and byte counts during
792  * the last interrupt.  The advantage of per interrupt computation
793  * is faster updates and more accurate ITR for the current traffic
794  * pattern.  Constants in this function were computed based on
795  * theoretical maximum wire speed and thresholds were set based on
796  * testing data as well as attempting to minimize response time
797  * while increasing bulk throughput.
798  **/
799 static void i40e_set_new_dynamic_itr(struct i40e_ring_container *rc)
800 {
801 	enum i40e_latency_range new_latency_range = rc->latency_range;
802 	u32 new_itr = rc->itr;
803 	int bytes_per_int;
804 
805 	if (rc->total_packets == 0 || !rc->itr)
806 		return;
807 
808 	/* simple throttlerate management
809 	 *   0-10MB/s   lowest (100000 ints/s)
810 	 *  10-20MB/s   low    (20000 ints/s)
811 	 *  20-1249MB/s bulk   (8000 ints/s)
812 	 */
813 	bytes_per_int = rc->total_bytes / rc->itr;
814 	switch (rc->itr) {
815 	case I40E_LOWEST_LATENCY:
816 		if (bytes_per_int > 10)
817 			new_latency_range = I40E_LOW_LATENCY;
818 		break;
819 	case I40E_LOW_LATENCY:
820 		if (bytes_per_int > 20)
821 			new_latency_range = I40E_BULK_LATENCY;
822 		else if (bytes_per_int <= 10)
823 			new_latency_range = I40E_LOWEST_LATENCY;
824 		break;
825 	case I40E_BULK_LATENCY:
826 		if (bytes_per_int <= 20)
827 			rc->latency_range = I40E_LOW_LATENCY;
828 		break;
829 	}
830 
831 	switch (new_latency_range) {
832 	case I40E_LOWEST_LATENCY:
833 		new_itr = I40E_ITR_100K;
834 		break;
835 	case I40E_LOW_LATENCY:
836 		new_itr = I40E_ITR_20K;
837 		break;
838 	case I40E_BULK_LATENCY:
839 		new_itr = I40E_ITR_8K;
840 		break;
841 	default:
842 		break;
843 	}
844 
845 	if (new_itr != rc->itr) {
846 		/* do an exponential smoothing */
847 		new_itr = (10 * new_itr * rc->itr) /
848 			  ((9 * new_itr) + rc->itr);
849 		rc->itr = new_itr & I40E_MAX_ITR;
850 	}
851 
852 	rc->total_bytes = 0;
853 	rc->total_packets = 0;
854 }
855 
856 /**
857  * i40e_update_dynamic_itr - Adjust ITR based on bytes per int
858  * @q_vector: the vector to adjust
859  **/
860 static void i40e_update_dynamic_itr(struct i40e_q_vector *q_vector)
861 {
862 	u16 vector = q_vector->vsi->base_vector + q_vector->v_idx;
863 	struct i40e_hw *hw = &q_vector->vsi->back->hw;
864 	u32 reg_addr;
865 	u16 old_itr;
866 
867 	reg_addr = I40E_PFINT_ITRN(I40E_RX_ITR, vector - 1);
868 	old_itr = q_vector->rx.itr;
869 	i40e_set_new_dynamic_itr(&q_vector->rx);
870 	if (old_itr != q_vector->rx.itr)
871 		wr32(hw, reg_addr, q_vector->rx.itr);
872 
873 	reg_addr = I40E_PFINT_ITRN(I40E_TX_ITR, vector - 1);
874 	old_itr = q_vector->tx.itr;
875 	i40e_set_new_dynamic_itr(&q_vector->tx);
876 	if (old_itr != q_vector->tx.itr)
877 		wr32(hw, reg_addr, q_vector->tx.itr);
878 }
879 
880 /**
881  * i40e_clean_programming_status - clean the programming status descriptor
882  * @rx_ring: the rx ring that has this descriptor
883  * @rx_desc: the rx descriptor written back by HW
884  *
885  * Flow director should handle FD_FILTER_STATUS to check its filter programming
886  * status being successful or not and take actions accordingly. FCoE should
887  * handle its context/filter programming/invalidation status and take actions.
888  *
889  **/
890 static void i40e_clean_programming_status(struct i40e_ring *rx_ring,
891 					  union i40e_rx_desc *rx_desc)
892 {
893 	u64 qw;
894 	u8 id;
895 
896 	qw = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
897 	id = (qw & I40E_RX_PROG_STATUS_DESC_QW1_PROGID_MASK) >>
898 		  I40E_RX_PROG_STATUS_DESC_QW1_PROGID_SHIFT;
899 
900 	if (id == I40E_RX_PROG_STATUS_DESC_FD_FILTER_STATUS)
901 		i40e_fd_handle_status(rx_ring, rx_desc, id);
902 }
903 
904 /**
905  * i40e_setup_tx_descriptors - Allocate the Tx descriptors
906  * @tx_ring: the tx ring to set up
907  *
908  * Return 0 on success, negative on error
909  **/
910 int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring)
911 {
912 	struct device *dev = tx_ring->dev;
913 	int bi_size;
914 
915 	if (!dev)
916 		return -ENOMEM;
917 
918 	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
919 	tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
920 	if (!tx_ring->tx_bi)
921 		goto err;
922 
923 	/* round up to nearest 4K */
924 	tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
925 	/* add u32 for head writeback, align after this takes care of
926 	 * guaranteeing this is at least one cache line in size
927 	 */
928 	tx_ring->size += sizeof(u32);
929 	tx_ring->size = ALIGN(tx_ring->size, 4096);
930 	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
931 					   &tx_ring->dma, GFP_KERNEL);
932 	if (!tx_ring->desc) {
933 		dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
934 			 tx_ring->size);
935 		goto err;
936 	}
937 
938 	tx_ring->next_to_use = 0;
939 	tx_ring->next_to_clean = 0;
940 	return 0;
941 
942 err:
943 	kfree(tx_ring->tx_bi);
944 	tx_ring->tx_bi = NULL;
945 	return -ENOMEM;
946 }
947 
948 /**
949  * i40e_clean_rx_ring - Free Rx buffers
950  * @rx_ring: ring to be cleaned
951  **/
952 void i40e_clean_rx_ring(struct i40e_ring *rx_ring)
953 {
954 	struct device *dev = rx_ring->dev;
955 	struct i40e_rx_buffer *rx_bi;
956 	unsigned long bi_size;
957 	u16 i;
958 
959 	/* ring already cleared, nothing to do */
960 	if (!rx_ring->rx_bi)
961 		return;
962 
963 	/* Free all the Rx ring sk_buffs */
964 	for (i = 0; i < rx_ring->count; i++) {
965 		rx_bi = &rx_ring->rx_bi[i];
966 		if (rx_bi->dma) {
967 			dma_unmap_single(dev,
968 					 rx_bi->dma,
969 					 rx_ring->rx_buf_len,
970 					 DMA_FROM_DEVICE);
971 			rx_bi->dma = 0;
972 		}
973 		if (rx_bi->skb) {
974 			dev_kfree_skb(rx_bi->skb);
975 			rx_bi->skb = NULL;
976 		}
977 		if (rx_bi->page) {
978 			if (rx_bi->page_dma) {
979 				dma_unmap_page(dev,
980 					       rx_bi->page_dma,
981 					       PAGE_SIZE / 2,
982 					       DMA_FROM_DEVICE);
983 				rx_bi->page_dma = 0;
984 			}
985 			__free_page(rx_bi->page);
986 			rx_bi->page = NULL;
987 			rx_bi->page_offset = 0;
988 		}
989 	}
990 
991 	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
992 	memset(rx_ring->rx_bi, 0, bi_size);
993 
994 	/* Zero out the descriptor ring */
995 	memset(rx_ring->desc, 0, rx_ring->size);
996 
997 	rx_ring->next_to_clean = 0;
998 	rx_ring->next_to_use = 0;
999 }
1000 
1001 /**
1002  * i40e_free_rx_resources - Free Rx resources
1003  * @rx_ring: ring to clean the resources from
1004  *
1005  * Free all receive software resources
1006  **/
1007 void i40e_free_rx_resources(struct i40e_ring *rx_ring)
1008 {
1009 	i40e_clean_rx_ring(rx_ring);
1010 	kfree(rx_ring->rx_bi);
1011 	rx_ring->rx_bi = NULL;
1012 
1013 	if (rx_ring->desc) {
1014 		dma_free_coherent(rx_ring->dev, rx_ring->size,
1015 				  rx_ring->desc, rx_ring->dma);
1016 		rx_ring->desc = NULL;
1017 	}
1018 }
1019 
1020 /**
1021  * i40e_setup_rx_descriptors - Allocate Rx descriptors
1022  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
1023  *
1024  * Returns 0 on success, negative on failure
1025  **/
1026 int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring)
1027 {
1028 	struct device *dev = rx_ring->dev;
1029 	int bi_size;
1030 
1031 	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
1032 	rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
1033 	if (!rx_ring->rx_bi)
1034 		goto err;
1035 
1036 	/* Round up to nearest 4K */
1037 	rx_ring->size = ring_is_16byte_desc_enabled(rx_ring)
1038 		? rx_ring->count * sizeof(union i40e_16byte_rx_desc)
1039 		: rx_ring->count * sizeof(union i40e_32byte_rx_desc);
1040 	rx_ring->size = ALIGN(rx_ring->size, 4096);
1041 	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
1042 					   &rx_ring->dma, GFP_KERNEL);
1043 
1044 	if (!rx_ring->desc) {
1045 		dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
1046 			 rx_ring->size);
1047 		goto err;
1048 	}
1049 
1050 	rx_ring->next_to_clean = 0;
1051 	rx_ring->next_to_use = 0;
1052 
1053 	return 0;
1054 err:
1055 	kfree(rx_ring->rx_bi);
1056 	rx_ring->rx_bi = NULL;
1057 	return -ENOMEM;
1058 }
1059 
1060 /**
1061  * i40e_release_rx_desc - Store the new tail and head values
1062  * @rx_ring: ring to bump
1063  * @val: new head index
1064  **/
1065 static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
1066 {
1067 	rx_ring->next_to_use = val;
1068 	/* Force memory writes to complete before letting h/w
1069 	 * know there are new descriptors to fetch.  (Only
1070 	 * applicable for weak-ordered memory model archs,
1071 	 * such as IA-64).
1072 	 */
1073 	wmb();
1074 	writel(val, rx_ring->tail);
1075 }
1076 
1077 /**
1078  * i40e_alloc_rx_buffers - Replace used receive buffers; packet split
1079  * @rx_ring: ring to place buffers on
1080  * @cleaned_count: number of buffers to replace
1081  **/
1082 void i40e_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count)
1083 {
1084 	u16 i = rx_ring->next_to_use;
1085 	union i40e_rx_desc *rx_desc;
1086 	struct i40e_rx_buffer *bi;
1087 	struct sk_buff *skb;
1088 
1089 	/* do nothing if no valid netdev defined */
1090 	if (!rx_ring->netdev || !cleaned_count)
1091 		return;
1092 
1093 	while (cleaned_count--) {
1094 		rx_desc = I40E_RX_DESC(rx_ring, i);
1095 		bi = &rx_ring->rx_bi[i];
1096 		skb = bi->skb;
1097 
1098 		if (!skb) {
1099 			skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1100 							rx_ring->rx_buf_len);
1101 			if (!skb) {
1102 				rx_ring->rx_stats.alloc_buff_failed++;
1103 				goto no_buffers;
1104 			}
1105 			/* initialize queue mapping */
1106 			skb_record_rx_queue(skb, rx_ring->queue_index);
1107 			bi->skb = skb;
1108 		}
1109 
1110 		if (!bi->dma) {
1111 			bi->dma = dma_map_single(rx_ring->dev,
1112 						 skb->data,
1113 						 rx_ring->rx_buf_len,
1114 						 DMA_FROM_DEVICE);
1115 			if (dma_mapping_error(rx_ring->dev, bi->dma)) {
1116 				rx_ring->rx_stats.alloc_buff_failed++;
1117 				bi->dma = 0;
1118 				goto no_buffers;
1119 			}
1120 		}
1121 
1122 		if (ring_is_ps_enabled(rx_ring)) {
1123 			if (!bi->page) {
1124 				bi->page = alloc_page(GFP_ATOMIC);
1125 				if (!bi->page) {
1126 					rx_ring->rx_stats.alloc_page_failed++;
1127 					goto no_buffers;
1128 				}
1129 			}
1130 
1131 			if (!bi->page_dma) {
1132 				/* use a half page if we're re-using */
1133 				bi->page_offset ^= PAGE_SIZE / 2;
1134 				bi->page_dma = dma_map_page(rx_ring->dev,
1135 							    bi->page,
1136 							    bi->page_offset,
1137 							    PAGE_SIZE / 2,
1138 							    DMA_FROM_DEVICE);
1139 				if (dma_mapping_error(rx_ring->dev,
1140 						      bi->page_dma)) {
1141 					rx_ring->rx_stats.alloc_page_failed++;
1142 					bi->page_dma = 0;
1143 					goto no_buffers;
1144 				}
1145 			}
1146 
1147 			/* Refresh the desc even if buffer_addrs didn't change
1148 			 * because each write-back erases this info.
1149 			 */
1150 			rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1151 			rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1152 		} else {
1153 			rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1154 			rx_desc->read.hdr_addr = 0;
1155 		}
1156 		i++;
1157 		if (i == rx_ring->count)
1158 			i = 0;
1159 	}
1160 
1161 no_buffers:
1162 	if (rx_ring->next_to_use != i)
1163 		i40e_release_rx_desc(rx_ring, i);
1164 }
1165 
1166 /**
1167  * i40e_receive_skb - Send a completed packet up the stack
1168  * @rx_ring:  rx ring in play
1169  * @skb: packet to send up
1170  * @vlan_tag: vlan tag for packet
1171  **/
1172 static void i40e_receive_skb(struct i40e_ring *rx_ring,
1173 			     struct sk_buff *skb, u16 vlan_tag)
1174 {
1175 	struct i40e_q_vector *q_vector = rx_ring->q_vector;
1176 	struct i40e_vsi *vsi = rx_ring->vsi;
1177 	u64 flags = vsi->back->flags;
1178 
1179 	if (vlan_tag & VLAN_VID_MASK)
1180 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
1181 
1182 	if (flags & I40E_FLAG_IN_NETPOLL)
1183 		netif_rx(skb);
1184 	else
1185 		napi_gro_receive(&q_vector->napi, skb);
1186 }
1187 
1188 /**
1189  * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
1190  * @vsi: the VSI we care about
1191  * @skb: skb currently being received and modified
1192  * @rx_status: status value of last descriptor in packet
1193  * @rx_error: error value of last descriptor in packet
1194  * @rx_ptype: ptype value of last descriptor in packet
1195  **/
1196 static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
1197 				    struct sk_buff *skb,
1198 				    u32 rx_status,
1199 				    u32 rx_error,
1200 				    u16 rx_ptype)
1201 {
1202 	bool ipv4_tunnel, ipv6_tunnel;
1203 	__wsum rx_udp_csum;
1204 	__sum16 csum;
1205 	struct iphdr *iph;
1206 
1207 	ipv4_tunnel = (rx_ptype > I40E_RX_PTYPE_GRENAT4_MAC_PAY3) &&
1208 		      (rx_ptype < I40E_RX_PTYPE_GRENAT4_MACVLAN_IPV6_ICMP_PAY4);
1209 	ipv6_tunnel = (rx_ptype > I40E_RX_PTYPE_GRENAT6_MAC_PAY3) &&
1210 		      (rx_ptype < I40E_RX_PTYPE_GRENAT6_MACVLAN_IPV6_ICMP_PAY4);
1211 
1212 	skb->encapsulation = ipv4_tunnel || ipv6_tunnel;
1213 	skb->ip_summed = CHECKSUM_NONE;
1214 
1215 	/* Rx csum enabled and ip headers found? */
1216 	if (!(vsi->netdev->features & NETIF_F_RXCSUM &&
1217 	      rx_status & (1 << I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
1218 		return;
1219 
1220 	/* likely incorrect csum if alternate IP extension headers found */
1221 	if (rx_status & (1 << I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
1222 		return;
1223 
1224 	/* IP or L4 or outmost IP checksum error */
1225 	if (rx_error & ((1 << I40E_RX_DESC_ERROR_IPE_SHIFT) |
1226 			(1 << I40E_RX_DESC_ERROR_L4E_SHIFT) |
1227 			(1 << I40E_RX_DESC_ERROR_EIPE_SHIFT))) {
1228 		vsi->back->hw_csum_rx_error++;
1229 		return;
1230 	}
1231 
1232 	if (ipv4_tunnel &&
1233 	    !(rx_status & (1 << I40E_RX_DESC_STATUS_UDP_0_SHIFT))) {
1234 		/* If VXLAN traffic has an outer UDPv4 checksum we need to check
1235 		 * it in the driver, hardware does not do it for us.
1236 		 * Since L3L4P bit was set we assume a valid IHL value (>=5)
1237 		 * so the total length of IPv4 header is IHL*4 bytes
1238 		 */
1239 		skb->transport_header = skb->mac_header +
1240 					sizeof(struct ethhdr) +
1241 					(ip_hdr(skb)->ihl * 4);
1242 
1243 		/* Add 4 bytes for VLAN tagged packets */
1244 		skb->transport_header += (skb->protocol == htons(ETH_P_8021Q) ||
1245 					  skb->protocol == htons(ETH_P_8021AD))
1246 					  ? VLAN_HLEN : 0;
1247 
1248 		rx_udp_csum = udp_csum(skb);
1249 		iph = ip_hdr(skb);
1250 		csum = csum_tcpudp_magic(
1251 				iph->saddr, iph->daddr,
1252 				(skb->len - skb_transport_offset(skb)),
1253 				IPPROTO_UDP, rx_udp_csum);
1254 
1255 		if (udp_hdr(skb)->check != csum) {
1256 			vsi->back->hw_csum_rx_error++;
1257 			return;
1258 		}
1259 	}
1260 
1261 	skb->ip_summed = CHECKSUM_UNNECESSARY;
1262 }
1263 
1264 /**
1265  * i40e_rx_hash - returns the hash value from the Rx descriptor
1266  * @ring: descriptor ring
1267  * @rx_desc: specific descriptor
1268  **/
1269 static inline u32 i40e_rx_hash(struct i40e_ring *ring,
1270 			       union i40e_rx_desc *rx_desc)
1271 {
1272 	const __le64 rss_mask =
1273 		cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
1274 			    I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
1275 
1276 	if ((ring->netdev->features & NETIF_F_RXHASH) &&
1277 	    (rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask)
1278 		return le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
1279 	else
1280 		return 0;
1281 }
1282 
1283 /**
1284  * i40e_ptype_to_hash - get a hash type
1285  * @ptype: the ptype value from the descriptor
1286  *
1287  * Returns a hash type to be used by skb_set_hash
1288  **/
1289 static inline enum pkt_hash_types i40e_ptype_to_hash(u8 ptype)
1290 {
1291 	struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
1292 
1293 	if (!decoded.known)
1294 		return PKT_HASH_TYPE_NONE;
1295 
1296 	if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1297 	    decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
1298 		return PKT_HASH_TYPE_L4;
1299 	else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1300 		 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
1301 		return PKT_HASH_TYPE_L3;
1302 	else
1303 		return PKT_HASH_TYPE_L2;
1304 }
1305 
1306 /**
1307  * i40e_clean_rx_irq - Reclaim resources after receive completes
1308  * @rx_ring:  rx ring to clean
1309  * @budget:   how many cleans we're allowed
1310  *
1311  * Returns true if there's any budget left (e.g. the clean is finished)
1312  **/
1313 static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget)
1314 {
1315 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1316 	u16 rx_packet_len, rx_header_len, rx_sph, rx_hbo;
1317 	u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
1318 	const int current_node = numa_node_id();
1319 	struct i40e_vsi *vsi = rx_ring->vsi;
1320 	u16 i = rx_ring->next_to_clean;
1321 	union i40e_rx_desc *rx_desc;
1322 	u32 rx_error, rx_status;
1323 	u8 rx_ptype;
1324 	u64 qword;
1325 
1326 	if (budget <= 0)
1327 		return 0;
1328 
1329 	rx_desc = I40E_RX_DESC(rx_ring, i);
1330 	qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1331 	rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1332 		    I40E_RXD_QW1_STATUS_SHIFT;
1333 
1334 	while (rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)) {
1335 		union i40e_rx_desc *next_rxd;
1336 		struct i40e_rx_buffer *rx_bi;
1337 		struct sk_buff *skb;
1338 		u16 vlan_tag;
1339 		if (i40e_rx_is_programming_status(qword)) {
1340 			i40e_clean_programming_status(rx_ring, rx_desc);
1341 			I40E_RX_NEXT_DESC_PREFETCH(rx_ring, i, next_rxd);
1342 			goto next_desc;
1343 		}
1344 		rx_bi = &rx_ring->rx_bi[i];
1345 		skb = rx_bi->skb;
1346 		prefetch(skb->data);
1347 
1348 		rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1349 				I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1350 		rx_header_len = (qword & I40E_RXD_QW1_LENGTH_HBUF_MASK) >>
1351 				I40E_RXD_QW1_LENGTH_HBUF_SHIFT;
1352 		rx_sph = (qword & I40E_RXD_QW1_LENGTH_SPH_MASK) >>
1353 			 I40E_RXD_QW1_LENGTH_SPH_SHIFT;
1354 
1355 		rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1356 			   I40E_RXD_QW1_ERROR_SHIFT;
1357 		rx_hbo = rx_error & (1 << I40E_RX_DESC_ERROR_HBO_SHIFT);
1358 		rx_error &= ~(1 << I40E_RX_DESC_ERROR_HBO_SHIFT);
1359 
1360 		rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1361 			   I40E_RXD_QW1_PTYPE_SHIFT;
1362 		rx_bi->skb = NULL;
1363 
1364 		/* This memory barrier is needed to keep us from reading
1365 		 * any other fields out of the rx_desc until we know the
1366 		 * STATUS_DD bit is set
1367 		 */
1368 		rmb();
1369 
1370 		/* Get the header and possibly the whole packet
1371 		 * If this is an skb from previous receive dma will be 0
1372 		 */
1373 		if (rx_bi->dma) {
1374 			u16 len;
1375 
1376 			if (rx_hbo)
1377 				len = I40E_RX_HDR_SIZE;
1378 			else if (rx_sph)
1379 				len = rx_header_len;
1380 			else if (rx_packet_len)
1381 				len = rx_packet_len;   /* 1buf/no split found */
1382 			else
1383 				len = rx_header_len;   /* split always mode */
1384 
1385 			skb_put(skb, len);
1386 			dma_unmap_single(rx_ring->dev,
1387 					 rx_bi->dma,
1388 					 rx_ring->rx_buf_len,
1389 					 DMA_FROM_DEVICE);
1390 			rx_bi->dma = 0;
1391 		}
1392 
1393 		/* Get the rest of the data if this was a header split */
1394 		if (ring_is_ps_enabled(rx_ring) && rx_packet_len) {
1395 
1396 			skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1397 					   rx_bi->page,
1398 					   rx_bi->page_offset,
1399 					   rx_packet_len);
1400 
1401 			skb->len += rx_packet_len;
1402 			skb->data_len += rx_packet_len;
1403 			skb->truesize += rx_packet_len;
1404 
1405 			if ((page_count(rx_bi->page) == 1) &&
1406 			    (page_to_nid(rx_bi->page) == current_node))
1407 				get_page(rx_bi->page);
1408 			else
1409 				rx_bi->page = NULL;
1410 
1411 			dma_unmap_page(rx_ring->dev,
1412 				       rx_bi->page_dma,
1413 				       PAGE_SIZE / 2,
1414 				       DMA_FROM_DEVICE);
1415 			rx_bi->page_dma = 0;
1416 		}
1417 		I40E_RX_NEXT_DESC_PREFETCH(rx_ring, i, next_rxd);
1418 
1419 		if (unlikely(
1420 		    !(rx_status & (1 << I40E_RX_DESC_STATUS_EOF_SHIFT)))) {
1421 			struct i40e_rx_buffer *next_buffer;
1422 
1423 			next_buffer = &rx_ring->rx_bi[i];
1424 
1425 			if (ring_is_ps_enabled(rx_ring)) {
1426 				rx_bi->skb = next_buffer->skb;
1427 				rx_bi->dma = next_buffer->dma;
1428 				next_buffer->skb = skb;
1429 				next_buffer->dma = 0;
1430 			}
1431 			rx_ring->rx_stats.non_eop_descs++;
1432 			goto next_desc;
1433 		}
1434 
1435 		/* ERR_MASK will only have valid bits if EOP set */
1436 		if (unlikely(rx_error & (1 << I40E_RX_DESC_ERROR_RXE_SHIFT))) {
1437 			dev_kfree_skb_any(skb);
1438 			goto next_desc;
1439 		}
1440 
1441 		skb_set_hash(skb, i40e_rx_hash(rx_ring, rx_desc),
1442 			     i40e_ptype_to_hash(rx_ptype));
1443 		if (unlikely(rx_status & I40E_RXD_QW1_STATUS_TSYNVALID_MASK)) {
1444 			i40e_ptp_rx_hwtstamp(vsi->back, skb, (rx_status &
1445 					   I40E_RXD_QW1_STATUS_TSYNINDX_MASK) >>
1446 					   I40E_RXD_QW1_STATUS_TSYNINDX_SHIFT);
1447 			rx_ring->last_rx_timestamp = jiffies;
1448 		}
1449 
1450 		/* probably a little skewed due to removing CRC */
1451 		total_rx_bytes += skb->len;
1452 		total_rx_packets++;
1453 
1454 		skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1455 
1456 		i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype);
1457 
1458 		vlan_tag = rx_status & (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)
1459 			 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1)
1460 			 : 0;
1461 		i40e_receive_skb(rx_ring, skb, vlan_tag);
1462 
1463 		rx_ring->netdev->last_rx = jiffies;
1464 		budget--;
1465 next_desc:
1466 		rx_desc->wb.qword1.status_error_len = 0;
1467 		if (!budget)
1468 			break;
1469 
1470 		cleaned_count++;
1471 		/* return some buffers to hardware, one at a time is too slow */
1472 		if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1473 			i40e_alloc_rx_buffers(rx_ring, cleaned_count);
1474 			cleaned_count = 0;
1475 		}
1476 
1477 		/* use prefetched values */
1478 		rx_desc = next_rxd;
1479 		qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1480 		rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1481 			    I40E_RXD_QW1_STATUS_SHIFT;
1482 	}
1483 
1484 	rx_ring->next_to_clean = i;
1485 	u64_stats_update_begin(&rx_ring->syncp);
1486 	rx_ring->stats.packets += total_rx_packets;
1487 	rx_ring->stats.bytes += total_rx_bytes;
1488 	u64_stats_update_end(&rx_ring->syncp);
1489 	rx_ring->q_vector->rx.total_packets += total_rx_packets;
1490 	rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1491 
1492 	if (cleaned_count)
1493 		i40e_alloc_rx_buffers(rx_ring, cleaned_count);
1494 
1495 	return budget > 0;
1496 }
1497 
1498 /**
1499  * i40e_napi_poll - NAPI polling Rx/Tx cleanup routine
1500  * @napi: napi struct with our devices info in it
1501  * @budget: amount of work driver is allowed to do this pass, in packets
1502  *
1503  * This function will clean all queues associated with a q_vector.
1504  *
1505  * Returns the amount of work done
1506  **/
1507 int i40e_napi_poll(struct napi_struct *napi, int budget)
1508 {
1509 	struct i40e_q_vector *q_vector =
1510 			       container_of(napi, struct i40e_q_vector, napi);
1511 	struct i40e_vsi *vsi = q_vector->vsi;
1512 	struct i40e_ring *ring;
1513 	bool clean_complete = true;
1514 	int budget_per_ring;
1515 
1516 	if (test_bit(__I40E_DOWN, &vsi->state)) {
1517 		napi_complete(napi);
1518 		return 0;
1519 	}
1520 
1521 	/* Since the actual Tx work is minimal, we can give the Tx a larger
1522 	 * budget and be more aggressive about cleaning up the Tx descriptors.
1523 	 */
1524 	i40e_for_each_ring(ring, q_vector->tx)
1525 		clean_complete &= i40e_clean_tx_irq(ring, vsi->work_limit);
1526 
1527 	/* We attempt to distribute budget to each Rx queue fairly, but don't
1528 	 * allow the budget to go below 1 because that would exit polling early.
1529 	 */
1530 	budget_per_ring = max(budget/q_vector->num_ringpairs, 1);
1531 
1532 	i40e_for_each_ring(ring, q_vector->rx)
1533 		clean_complete &= i40e_clean_rx_irq(ring, budget_per_ring);
1534 
1535 	/* If work not completed, return budget and polling will return */
1536 	if (!clean_complete)
1537 		return budget;
1538 
1539 	/* Work is done so exit the polling mode and re-enable the interrupt */
1540 	napi_complete(napi);
1541 	if (ITR_IS_DYNAMIC(vsi->rx_itr_setting) ||
1542 	    ITR_IS_DYNAMIC(vsi->tx_itr_setting))
1543 		i40e_update_dynamic_itr(q_vector);
1544 
1545 	if (!test_bit(__I40E_DOWN, &vsi->state)) {
1546 		if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
1547 			i40e_irq_dynamic_enable(vsi,
1548 					q_vector->v_idx + vsi->base_vector);
1549 		} else {
1550 			struct i40e_hw *hw = &vsi->back->hw;
1551 			/* We re-enable the queue 0 cause, but
1552 			 * don't worry about dynamic_enable
1553 			 * because we left it on for the other
1554 			 * possible interrupts during napi
1555 			 */
1556 			u32 qval = rd32(hw, I40E_QINT_RQCTL(0));
1557 			qval |= I40E_QINT_RQCTL_CAUSE_ENA_MASK;
1558 			wr32(hw, I40E_QINT_RQCTL(0), qval);
1559 
1560 			qval = rd32(hw, I40E_QINT_TQCTL(0));
1561 			qval |= I40E_QINT_TQCTL_CAUSE_ENA_MASK;
1562 			wr32(hw, I40E_QINT_TQCTL(0), qval);
1563 
1564 			i40e_irq_dynamic_enable_icr0(vsi->back);
1565 		}
1566 	}
1567 
1568 	return 0;
1569 }
1570 
1571 /**
1572  * i40e_atr - Add a Flow Director ATR filter
1573  * @tx_ring:  ring to add programming descriptor to
1574  * @skb:      send buffer
1575  * @flags:    send flags
1576  * @protocol: wire protocol
1577  **/
1578 static void i40e_atr(struct i40e_ring *tx_ring, struct sk_buff *skb,
1579 		     u32 flags, __be16 protocol)
1580 {
1581 	struct i40e_filter_program_desc *fdir_desc;
1582 	struct i40e_pf *pf = tx_ring->vsi->back;
1583 	union {
1584 		unsigned char *network;
1585 		struct iphdr *ipv4;
1586 		struct ipv6hdr *ipv6;
1587 	} hdr;
1588 	struct tcphdr *th;
1589 	unsigned int hlen;
1590 	u32 flex_ptype, dtype_cmd;
1591 	u16 i;
1592 
1593 	/* make sure ATR is enabled */
1594 	if (!(pf->flags & I40E_FLAG_FD_ATR_ENABLED))
1595 		return;
1596 
1597 	/* if sampling is disabled do nothing */
1598 	if (!tx_ring->atr_sample_rate)
1599 		return;
1600 
1601 	/* snag network header to get L4 type and address */
1602 	hdr.network = skb_network_header(skb);
1603 
1604 	/* Currently only IPv4/IPv6 with TCP is supported */
1605 	if (protocol == htons(ETH_P_IP)) {
1606 		if (hdr.ipv4->protocol != IPPROTO_TCP)
1607 			return;
1608 
1609 		/* access ihl as a u8 to avoid unaligned access on ia64 */
1610 		hlen = (hdr.network[0] & 0x0F) << 2;
1611 	} else if (protocol == htons(ETH_P_IPV6)) {
1612 		if (hdr.ipv6->nexthdr != IPPROTO_TCP)
1613 			return;
1614 
1615 		hlen = sizeof(struct ipv6hdr);
1616 	} else {
1617 		return;
1618 	}
1619 
1620 	th = (struct tcphdr *)(hdr.network + hlen);
1621 
1622 	/* Due to lack of space, no more new filters can be programmed */
1623 	if (th->syn && (pf->auto_disable_flags & I40E_FLAG_FD_ATR_ENABLED))
1624 		return;
1625 
1626 	tx_ring->atr_count++;
1627 
1628 	/* sample on all syn/fin/rst packets or once every atr sample rate */
1629 	if (!th->fin &&
1630 	    !th->syn &&
1631 	    !th->rst &&
1632 	    (tx_ring->atr_count < tx_ring->atr_sample_rate))
1633 		return;
1634 
1635 	tx_ring->atr_count = 0;
1636 
1637 	/* grab the next descriptor */
1638 	i = tx_ring->next_to_use;
1639 	fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
1640 
1641 	i++;
1642 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1643 
1644 	flex_ptype = (tx_ring->queue_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) &
1645 		      I40E_TXD_FLTR_QW0_QINDEX_MASK;
1646 	flex_ptype |= (protocol == htons(ETH_P_IP)) ?
1647 		      (I40E_FILTER_PCTYPE_NONF_IPV4_TCP <<
1648 		       I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) :
1649 		      (I40E_FILTER_PCTYPE_NONF_IPV6_TCP <<
1650 		       I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
1651 
1652 	flex_ptype |= tx_ring->vsi->id << I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT;
1653 
1654 	dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
1655 
1656 	dtype_cmd |= (th->fin || th->rst) ?
1657 		     (I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
1658 		      I40E_TXD_FLTR_QW1_PCMD_SHIFT) :
1659 		     (I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
1660 		      I40E_TXD_FLTR_QW1_PCMD_SHIFT);
1661 
1662 	dtype_cmd |= I40E_FILTER_PROGRAM_DESC_DEST_DIRECT_PACKET_QINDEX <<
1663 		     I40E_TXD_FLTR_QW1_DEST_SHIFT;
1664 
1665 	dtype_cmd |= I40E_FILTER_PROGRAM_DESC_FD_STATUS_FD_ID <<
1666 		     I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT;
1667 
1668 	fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
1669 	fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
1670 }
1671 
1672 /**
1673  * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
1674  * @skb:     send buffer
1675  * @tx_ring: ring to send buffer on
1676  * @flags:   the tx flags to be set
1677  *
1678  * Checks the skb and set up correspondingly several generic transmit flags
1679  * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
1680  *
1681  * Returns error code indicate the frame should be dropped upon error and the
1682  * otherwise  returns 0 to indicate the flags has been set properly.
1683  **/
1684 static int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
1685 				      struct i40e_ring *tx_ring,
1686 				      u32 *flags)
1687 {
1688 	__be16 protocol = skb->protocol;
1689 	u32  tx_flags = 0;
1690 
1691 	/* if we have a HW VLAN tag being added, default to the HW one */
1692 	if (vlan_tx_tag_present(skb)) {
1693 		tx_flags |= vlan_tx_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
1694 		tx_flags |= I40E_TX_FLAGS_HW_VLAN;
1695 	/* else if it is a SW VLAN, check the next protocol and store the tag */
1696 	} else if (protocol == htons(ETH_P_8021Q)) {
1697 		struct vlan_hdr *vhdr, _vhdr;
1698 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
1699 		if (!vhdr)
1700 			return -EINVAL;
1701 
1702 		protocol = vhdr->h_vlan_encapsulated_proto;
1703 		tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
1704 		tx_flags |= I40E_TX_FLAGS_SW_VLAN;
1705 	}
1706 
1707 	/* Insert 802.1p priority into VLAN header */
1708 	if ((tx_ring->vsi->back->flags & I40E_FLAG_DCB_ENABLED) &&
1709 	    ((tx_flags & (I40E_TX_FLAGS_HW_VLAN | I40E_TX_FLAGS_SW_VLAN)) ||
1710 	     (skb->priority != TC_PRIO_CONTROL))) {
1711 		tx_flags &= ~I40E_TX_FLAGS_VLAN_PRIO_MASK;
1712 		tx_flags |= (skb->priority & 0x7) <<
1713 				I40E_TX_FLAGS_VLAN_PRIO_SHIFT;
1714 		if (tx_flags & I40E_TX_FLAGS_SW_VLAN) {
1715 			struct vlan_ethhdr *vhdr;
1716 			if (skb_header_cloned(skb) &&
1717 			    pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
1718 				return -ENOMEM;
1719 			vhdr = (struct vlan_ethhdr *)skb->data;
1720 			vhdr->h_vlan_TCI = htons(tx_flags >>
1721 						 I40E_TX_FLAGS_VLAN_SHIFT);
1722 		} else {
1723 			tx_flags |= I40E_TX_FLAGS_HW_VLAN;
1724 		}
1725 	}
1726 	*flags = tx_flags;
1727 	return 0;
1728 }
1729 
1730 /**
1731  * i40e_tso - set up the tso context descriptor
1732  * @tx_ring:  ptr to the ring to send
1733  * @skb:      ptr to the skb we're sending
1734  * @tx_flags: the collected send information
1735  * @protocol: the send protocol
1736  * @hdr_len:  ptr to the size of the packet header
1737  * @cd_tunneling: ptr to context descriptor bits
1738  *
1739  * Returns 0 if no TSO can happen, 1 if tso is going, or error
1740  **/
1741 static int i40e_tso(struct i40e_ring *tx_ring, struct sk_buff *skb,
1742 		    u32 tx_flags, __be16 protocol, u8 *hdr_len,
1743 		    u64 *cd_type_cmd_tso_mss, u32 *cd_tunneling)
1744 {
1745 	u32 cd_cmd, cd_tso_len, cd_mss;
1746 	struct tcphdr *tcph;
1747 	struct iphdr *iph;
1748 	u32 l4len;
1749 	int err;
1750 	struct ipv6hdr *ipv6h;
1751 
1752 	if (!skb_is_gso(skb))
1753 		return 0;
1754 
1755 	if (skb_header_cloned(skb)) {
1756 		err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1757 		if (err)
1758 			return err;
1759 	}
1760 
1761 	if (protocol == htons(ETH_P_IP)) {
1762 		iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb);
1763 		tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb);
1764 		iph->tot_len = 0;
1765 		iph->check = 0;
1766 		tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1767 						 0, IPPROTO_TCP, 0);
1768 	} else if (skb_is_gso_v6(skb)) {
1769 
1770 		ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb)
1771 					   : ipv6_hdr(skb);
1772 		tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb);
1773 		ipv6h->payload_len = 0;
1774 		tcph->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
1775 					       0, IPPROTO_TCP, 0);
1776 	}
1777 
1778 	l4len = skb->encapsulation ? inner_tcp_hdrlen(skb) : tcp_hdrlen(skb);
1779 	*hdr_len = (skb->encapsulation
1780 		    ? (skb_inner_transport_header(skb) - skb->data)
1781 		    : skb_transport_offset(skb)) + l4len;
1782 
1783 	/* find the field values */
1784 	cd_cmd = I40E_TX_CTX_DESC_TSO;
1785 	cd_tso_len = skb->len - *hdr_len;
1786 	cd_mss = skb_shinfo(skb)->gso_size;
1787 	*cd_type_cmd_tso_mss |= ((u64)cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
1788 				((u64)cd_tso_len <<
1789 				 I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
1790 				((u64)cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
1791 	return 1;
1792 }
1793 
1794 /**
1795  * i40e_tsyn - set up the tsyn context descriptor
1796  * @tx_ring:  ptr to the ring to send
1797  * @skb:      ptr to the skb we're sending
1798  * @tx_flags: the collected send information
1799  *
1800  * Returns 0 if no Tx timestamp can happen and 1 if the timestamp will happen
1801  **/
1802 static int i40e_tsyn(struct i40e_ring *tx_ring, struct sk_buff *skb,
1803 		     u32 tx_flags, u64 *cd_type_cmd_tso_mss)
1804 {
1805 	struct i40e_pf *pf;
1806 
1807 	if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)))
1808 		return 0;
1809 
1810 	/* Tx timestamps cannot be sampled when doing TSO */
1811 	if (tx_flags & I40E_TX_FLAGS_TSO)
1812 		return 0;
1813 
1814 	/* only timestamp the outbound packet if the user has requested it and
1815 	 * we are not already transmitting a packet to be timestamped
1816 	 */
1817 	pf = i40e_netdev_to_pf(tx_ring->netdev);
1818 	if (pf->ptp_tx && !pf->ptp_tx_skb) {
1819 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
1820 		pf->ptp_tx_skb = skb_get(skb);
1821 	} else {
1822 		return 0;
1823 	}
1824 
1825 	*cd_type_cmd_tso_mss |= (u64)I40E_TX_CTX_DESC_TSYN <<
1826 				I40E_TXD_CTX_QW1_CMD_SHIFT;
1827 
1828 	pf->ptp_tx_start = jiffies;
1829 	schedule_work(&pf->ptp_tx_work);
1830 
1831 	return 1;
1832 }
1833 
1834 /**
1835  * i40e_tx_enable_csum - Enable Tx checksum offloads
1836  * @skb: send buffer
1837  * @tx_flags: Tx flags currently set
1838  * @td_cmd: Tx descriptor command bits to set
1839  * @td_offset: Tx descriptor header offsets to set
1840  * @cd_tunneling: ptr to context desc bits
1841  **/
1842 static void i40e_tx_enable_csum(struct sk_buff *skb, u32 tx_flags,
1843 				u32 *td_cmd, u32 *td_offset,
1844 				struct i40e_ring *tx_ring,
1845 				u32 *cd_tunneling)
1846 {
1847 	struct ipv6hdr *this_ipv6_hdr;
1848 	unsigned int this_tcp_hdrlen;
1849 	struct iphdr *this_ip_hdr;
1850 	u32 network_hdr_len;
1851 	u8 l4_hdr = 0;
1852 
1853 	if (skb->encapsulation) {
1854 		network_hdr_len = skb_inner_network_header_len(skb);
1855 		this_ip_hdr = inner_ip_hdr(skb);
1856 		this_ipv6_hdr = inner_ipv6_hdr(skb);
1857 		this_tcp_hdrlen = inner_tcp_hdrlen(skb);
1858 
1859 		if (tx_flags & I40E_TX_FLAGS_IPV4) {
1860 
1861 			if (tx_flags & I40E_TX_FLAGS_TSO) {
1862 				*cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4;
1863 				ip_hdr(skb)->check = 0;
1864 			} else {
1865 				*cd_tunneling |=
1866 					 I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
1867 			}
1868 		} else if (tx_flags & I40E_TX_FLAGS_IPV6) {
1869 			if (tx_flags & I40E_TX_FLAGS_TSO) {
1870 				*cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV6;
1871 				ip_hdr(skb)->check = 0;
1872 			} else {
1873 				*cd_tunneling |=
1874 					 I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
1875 			}
1876 		}
1877 
1878 		/* Now set the ctx descriptor fields */
1879 		*cd_tunneling |= (skb_network_header_len(skb) >> 2) <<
1880 					I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT |
1881 				   I40E_TXD_CTX_UDP_TUNNELING            |
1882 				   ((skb_inner_network_offset(skb) -
1883 					skb_transport_offset(skb)) >> 1) <<
1884 				   I40E_TXD_CTX_QW0_NATLEN_SHIFT;
1885 
1886 	} else {
1887 		network_hdr_len = skb_network_header_len(skb);
1888 		this_ip_hdr = ip_hdr(skb);
1889 		this_ipv6_hdr = ipv6_hdr(skb);
1890 		this_tcp_hdrlen = tcp_hdrlen(skb);
1891 	}
1892 
1893 	/* Enable IP checksum offloads */
1894 	if (tx_flags & I40E_TX_FLAGS_IPV4) {
1895 		l4_hdr = this_ip_hdr->protocol;
1896 		/* the stack computes the IP header already, the only time we
1897 		 * need the hardware to recompute it is in the case of TSO.
1898 		 */
1899 		if (tx_flags & I40E_TX_FLAGS_TSO) {
1900 			*td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM;
1901 			this_ip_hdr->check = 0;
1902 		} else {
1903 			*td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4;
1904 		}
1905 		/* Now set the td_offset for IP header length */
1906 		*td_offset = (network_hdr_len >> 2) <<
1907 			      I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1908 	} else if (tx_flags & I40E_TX_FLAGS_IPV6) {
1909 		l4_hdr = this_ipv6_hdr->nexthdr;
1910 		*td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
1911 		/* Now set the td_offset for IP header length */
1912 		*td_offset = (network_hdr_len >> 2) <<
1913 			      I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1914 	}
1915 	/* words in MACLEN + dwords in IPLEN + dwords in L4Len */
1916 	*td_offset |= (skb_network_offset(skb) >> 1) <<
1917 		       I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
1918 
1919 	/* Enable L4 checksum offloads */
1920 	switch (l4_hdr) {
1921 	case IPPROTO_TCP:
1922 		/* enable checksum offloads */
1923 		*td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
1924 		*td_offset |= (this_tcp_hdrlen >> 2) <<
1925 			       I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1926 		break;
1927 	case IPPROTO_SCTP:
1928 		/* enable SCTP checksum offload */
1929 		*td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
1930 		*td_offset |= (sizeof(struct sctphdr) >> 2) <<
1931 			       I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1932 		break;
1933 	case IPPROTO_UDP:
1934 		/* enable UDP checksum offload */
1935 		*td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
1936 		*td_offset |= (sizeof(struct udphdr) >> 2) <<
1937 			       I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1938 		break;
1939 	default:
1940 		break;
1941 	}
1942 }
1943 
1944 /**
1945  * i40e_create_tx_ctx Build the Tx context descriptor
1946  * @tx_ring:  ring to create the descriptor on
1947  * @cd_type_cmd_tso_mss: Quad Word 1
1948  * @cd_tunneling: Quad Word 0 - bits 0-31
1949  * @cd_l2tag2: Quad Word 0 - bits 32-63
1950  **/
1951 static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
1952 			       const u64 cd_type_cmd_tso_mss,
1953 			       const u32 cd_tunneling, const u32 cd_l2tag2)
1954 {
1955 	struct i40e_tx_context_desc *context_desc;
1956 	int i = tx_ring->next_to_use;
1957 
1958 	if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
1959 	    !cd_tunneling && !cd_l2tag2)
1960 		return;
1961 
1962 	/* grab the next descriptor */
1963 	context_desc = I40E_TX_CTXTDESC(tx_ring, i);
1964 
1965 	i++;
1966 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1967 
1968 	/* cpu_to_le32 and assign to struct fields */
1969 	context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
1970 	context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
1971 	context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
1972 }
1973 
1974 /**
1975  * i40e_tx_map - Build the Tx descriptor
1976  * @tx_ring:  ring to send buffer on
1977  * @skb:      send buffer
1978  * @first:    first buffer info buffer to use
1979  * @tx_flags: collected send information
1980  * @hdr_len:  size of the packet header
1981  * @td_cmd:   the command field in the descriptor
1982  * @td_offset: offset for checksum or crc
1983  **/
1984 static void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
1985 			struct i40e_tx_buffer *first, u32 tx_flags,
1986 			const u8 hdr_len, u32 td_cmd, u32 td_offset)
1987 {
1988 	unsigned int data_len = skb->data_len;
1989 	unsigned int size = skb_headlen(skb);
1990 	struct skb_frag_struct *frag;
1991 	struct i40e_tx_buffer *tx_bi;
1992 	struct i40e_tx_desc *tx_desc;
1993 	u16 i = tx_ring->next_to_use;
1994 	u32 td_tag = 0;
1995 	dma_addr_t dma;
1996 	u16 gso_segs;
1997 
1998 	if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
1999 		td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
2000 		td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
2001 			 I40E_TX_FLAGS_VLAN_SHIFT;
2002 	}
2003 
2004 	if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO))
2005 		gso_segs = skb_shinfo(skb)->gso_segs;
2006 	else
2007 		gso_segs = 1;
2008 
2009 	/* multiply data chunks by size of headers */
2010 	first->bytecount = skb->len - hdr_len + (gso_segs * hdr_len);
2011 	first->gso_segs = gso_segs;
2012 	first->skb = skb;
2013 	first->tx_flags = tx_flags;
2014 
2015 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
2016 
2017 	tx_desc = I40E_TX_DESC(tx_ring, i);
2018 	tx_bi = first;
2019 
2020 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
2021 		if (dma_mapping_error(tx_ring->dev, dma))
2022 			goto dma_error;
2023 
2024 		/* record length, and DMA address */
2025 		dma_unmap_len_set(tx_bi, len, size);
2026 		dma_unmap_addr_set(tx_bi, dma, dma);
2027 
2028 		tx_desc->buffer_addr = cpu_to_le64(dma);
2029 
2030 		while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
2031 			tx_desc->cmd_type_offset_bsz =
2032 				build_ctob(td_cmd, td_offset,
2033 					   I40E_MAX_DATA_PER_TXD, td_tag);
2034 
2035 			tx_desc++;
2036 			i++;
2037 			if (i == tx_ring->count) {
2038 				tx_desc = I40E_TX_DESC(tx_ring, 0);
2039 				i = 0;
2040 			}
2041 
2042 			dma += I40E_MAX_DATA_PER_TXD;
2043 			size -= I40E_MAX_DATA_PER_TXD;
2044 
2045 			tx_desc->buffer_addr = cpu_to_le64(dma);
2046 		}
2047 
2048 		if (likely(!data_len))
2049 			break;
2050 
2051 		tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
2052 							  size, td_tag);
2053 
2054 		tx_desc++;
2055 		i++;
2056 		if (i == tx_ring->count) {
2057 			tx_desc = I40E_TX_DESC(tx_ring, 0);
2058 			i = 0;
2059 		}
2060 
2061 		size = skb_frag_size(frag);
2062 		data_len -= size;
2063 
2064 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
2065 				       DMA_TO_DEVICE);
2066 
2067 		tx_bi = &tx_ring->tx_bi[i];
2068 	}
2069 
2070 	/* Place RS bit on last descriptor of any packet that spans across the
2071 	 * 4th descriptor (WB_STRIDE aka 0x3) in a 64B cacheline.
2072 	 */
2073 #define WB_STRIDE 0x3
2074 	if (((i & WB_STRIDE) != WB_STRIDE) &&
2075 	    (first <= &tx_ring->tx_bi[i]) &&
2076 	    (first >= &tx_ring->tx_bi[i & ~WB_STRIDE])) {
2077 		tx_desc->cmd_type_offset_bsz =
2078 			build_ctob(td_cmd, td_offset, size, td_tag) |
2079 			cpu_to_le64((u64)I40E_TX_DESC_CMD_EOP <<
2080 					 I40E_TXD_QW1_CMD_SHIFT);
2081 	} else {
2082 		tx_desc->cmd_type_offset_bsz =
2083 			build_ctob(td_cmd, td_offset, size, td_tag) |
2084 			cpu_to_le64((u64)I40E_TXD_CMD <<
2085 					 I40E_TXD_QW1_CMD_SHIFT);
2086 	}
2087 
2088 	netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev,
2089 						 tx_ring->queue_index),
2090 			     first->bytecount);
2091 
2092 	/* set the timestamp */
2093 	first->time_stamp = jiffies;
2094 
2095 	/* Force memory writes to complete before letting h/w
2096 	 * know there are new descriptors to fetch.  (Only
2097 	 * applicable for weak-ordered memory model archs,
2098 	 * such as IA-64).
2099 	 */
2100 	wmb();
2101 
2102 	/* set next_to_watch value indicating a packet is present */
2103 	first->next_to_watch = tx_desc;
2104 
2105 	i++;
2106 	if (i == tx_ring->count)
2107 		i = 0;
2108 
2109 	tx_ring->next_to_use = i;
2110 
2111 	/* notify HW of packet */
2112 	writel(i, tx_ring->tail);
2113 
2114 	return;
2115 
2116 dma_error:
2117 	dev_info(tx_ring->dev, "TX DMA map failed\n");
2118 
2119 	/* clear dma mappings for failed tx_bi map */
2120 	for (;;) {
2121 		tx_bi = &tx_ring->tx_bi[i];
2122 		i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
2123 		if (tx_bi == first)
2124 			break;
2125 		if (i == 0)
2126 			i = tx_ring->count;
2127 		i--;
2128 	}
2129 
2130 	tx_ring->next_to_use = i;
2131 }
2132 
2133 /**
2134  * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions
2135  * @tx_ring: the ring to be checked
2136  * @size:    the size buffer we want to assure is available
2137  *
2138  * Returns -EBUSY if a stop is needed, else 0
2139  **/
2140 static inline int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
2141 {
2142 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
2143 	/* Memory barrier before checking head and tail */
2144 	smp_mb();
2145 
2146 	/* Check again in a case another CPU has just made room available. */
2147 	if (likely(I40E_DESC_UNUSED(tx_ring) < size))
2148 		return -EBUSY;
2149 
2150 	/* A reprieve! - use start_queue because it doesn't call schedule */
2151 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
2152 	++tx_ring->tx_stats.restart_queue;
2153 	return 0;
2154 }
2155 
2156 /**
2157  * i40e_maybe_stop_tx - 1st level check for tx stop conditions
2158  * @tx_ring: the ring to be checked
2159  * @size:    the size buffer we want to assure is available
2160  *
2161  * Returns 0 if stop is not needed
2162  **/
2163 static int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
2164 {
2165 	if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
2166 		return 0;
2167 	return __i40e_maybe_stop_tx(tx_ring, size);
2168 }
2169 
2170 /**
2171  * i40e_xmit_descriptor_count - calculate number of tx descriptors needed
2172  * @skb:     send buffer
2173  * @tx_ring: ring to send buffer on
2174  *
2175  * Returns number of data descriptors needed for this skb. Returns 0 to indicate
2176  * there is not enough descriptors available in this ring since we need at least
2177  * one descriptor.
2178  **/
2179 static int i40e_xmit_descriptor_count(struct sk_buff *skb,
2180 				      struct i40e_ring *tx_ring)
2181 {
2182 #if PAGE_SIZE > I40E_MAX_DATA_PER_TXD
2183 	unsigned int f;
2184 #endif
2185 	int count = 0;
2186 
2187 	/* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
2188 	 *       + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
2189 	 *       + 4 desc gap to avoid the cache line where head is,
2190 	 *       + 1 desc for context descriptor,
2191 	 * otherwise try next time
2192 	 */
2193 #if PAGE_SIZE > I40E_MAX_DATA_PER_TXD
2194 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
2195 		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
2196 #else
2197 	count += skb_shinfo(skb)->nr_frags;
2198 #endif
2199 	count += TXD_USE_COUNT(skb_headlen(skb));
2200 	if (i40e_maybe_stop_tx(tx_ring, count + 4 + 1)) {
2201 		tx_ring->tx_stats.tx_busy++;
2202 		return 0;
2203 	}
2204 	return count;
2205 }
2206 
2207 /**
2208  * i40e_xmit_frame_ring - Sends buffer on Tx ring
2209  * @skb:     send buffer
2210  * @tx_ring: ring to send buffer on
2211  *
2212  * Returns NETDEV_TX_OK if sent, else an error code
2213  **/
2214 static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
2215 					struct i40e_ring *tx_ring)
2216 {
2217 	u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
2218 	u32 cd_tunneling = 0, cd_l2tag2 = 0;
2219 	struct i40e_tx_buffer *first;
2220 	u32 td_offset = 0;
2221 	u32 tx_flags = 0;
2222 	__be16 protocol;
2223 	u32 td_cmd = 0;
2224 	u8 hdr_len = 0;
2225 	int tsyn;
2226 	int tso;
2227 	if (0 == i40e_xmit_descriptor_count(skb, tx_ring))
2228 		return NETDEV_TX_BUSY;
2229 
2230 	/* prepare the xmit flags */
2231 	if (i40e_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
2232 		goto out_drop;
2233 
2234 	/* obtain protocol of skb */
2235 	protocol = skb->protocol;
2236 
2237 	/* record the location of the first descriptor for this packet */
2238 	first = &tx_ring->tx_bi[tx_ring->next_to_use];
2239 
2240 	/* setup IPv4/IPv6 offloads */
2241 	if (protocol == htons(ETH_P_IP))
2242 		tx_flags |= I40E_TX_FLAGS_IPV4;
2243 	else if (protocol == htons(ETH_P_IPV6))
2244 		tx_flags |= I40E_TX_FLAGS_IPV6;
2245 
2246 	tso = i40e_tso(tx_ring, skb, tx_flags, protocol, &hdr_len,
2247 		       &cd_type_cmd_tso_mss, &cd_tunneling);
2248 
2249 	if (tso < 0)
2250 		goto out_drop;
2251 	else if (tso)
2252 		tx_flags |= I40E_TX_FLAGS_TSO;
2253 
2254 	skb_tx_timestamp(skb);
2255 
2256 	tsyn = i40e_tsyn(tx_ring, skb, tx_flags, &cd_type_cmd_tso_mss);
2257 
2258 	if (tsyn)
2259 		tx_flags |= I40E_TX_FLAGS_TSYN;
2260 
2261 	/* always enable CRC insertion offload */
2262 	td_cmd |= I40E_TX_DESC_CMD_ICRC;
2263 
2264 	/* Always offload the checksum, since it's in the data descriptor */
2265 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
2266 		tx_flags |= I40E_TX_FLAGS_CSUM;
2267 
2268 		i40e_tx_enable_csum(skb, tx_flags, &td_cmd, &td_offset,
2269 				    tx_ring, &cd_tunneling);
2270 	}
2271 
2272 	i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
2273 			   cd_tunneling, cd_l2tag2);
2274 
2275 	/* Add Flow Director ATR if it's enabled.
2276 	 *
2277 	 * NOTE: this must always be directly before the data descriptor.
2278 	 */
2279 	i40e_atr(tx_ring, skb, tx_flags, protocol);
2280 
2281 	i40e_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
2282 		    td_cmd, td_offset);
2283 
2284 	i40e_maybe_stop_tx(tx_ring, DESC_NEEDED);
2285 
2286 	return NETDEV_TX_OK;
2287 
2288 out_drop:
2289 	dev_kfree_skb_any(skb);
2290 	return NETDEV_TX_OK;
2291 }
2292 
2293 /**
2294  * i40e_lan_xmit_frame - Selects the correct VSI and Tx queue to send buffer
2295  * @skb:    send buffer
2296  * @netdev: network interface device structure
2297  *
2298  * Returns NETDEV_TX_OK if sent, else an error code
2299  **/
2300 netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2301 {
2302 	struct i40e_netdev_priv *np = netdev_priv(netdev);
2303 	struct i40e_vsi *vsi = np->vsi;
2304 	struct i40e_ring *tx_ring = vsi->tx_rings[skb->queue_mapping];
2305 
2306 	/* hardware can't handle really short frames, hardware padding works
2307 	 * beyond this point
2308 	 */
2309 	if (unlikely(skb->len < I40E_MIN_TX_LEN)) {
2310 		if (skb_pad(skb, I40E_MIN_TX_LEN - skb->len))
2311 			return NETDEV_TX_OK;
2312 		skb->len = I40E_MIN_TX_LEN;
2313 		skb_set_tail_pointer(skb, I40E_MIN_TX_LEN);
2314 	}
2315 
2316 	return i40e_xmit_frame_ring(skb, tx_ring);
2317 }
2318