xref: /openbmc/linux/drivers/net/ethernet/intel/i40e/i40e_txrx.c (revision 4f139972b489f8bc2c821aa25ac65018d92af3f7)
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 /**
44  * i40e_fdir - Generate a Flow Director descriptor based on fdata
45  * @tx_ring: Tx ring to send buffer on
46  * @fdata: Flow director filter data
47  * @add: Indicate if we are adding a rule or deleting one
48  *
49  **/
50 static void i40e_fdir(struct i40e_ring *tx_ring,
51 		      struct i40e_fdir_filter *fdata, bool add)
52 {
53 	struct i40e_filter_program_desc *fdir_desc;
54 	struct i40e_pf *pf = tx_ring->vsi->back;
55 	u32 flex_ptype, dtype_cmd;
56 	u16 i;
57 
58 	/* grab the next descriptor */
59 	i = tx_ring->next_to_use;
60 	fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
61 
62 	i++;
63 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
64 
65 	flex_ptype = I40E_TXD_FLTR_QW0_QINDEX_MASK &
66 		     (fdata->q_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT);
67 
68 	flex_ptype |= I40E_TXD_FLTR_QW0_FLEXOFF_MASK &
69 		      (fdata->flex_off << I40E_TXD_FLTR_QW0_FLEXOFF_SHIFT);
70 
71 	flex_ptype |= I40E_TXD_FLTR_QW0_PCTYPE_MASK &
72 		      (fdata->pctype << I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
73 
74 	flex_ptype |= I40E_TXD_FLTR_QW0_PCTYPE_MASK &
75 		      (fdata->flex_offset << I40E_TXD_FLTR_QW0_FLEXOFF_SHIFT);
76 
77 	/* Use LAN VSI Id if not programmed by user */
78 	flex_ptype |= I40E_TXD_FLTR_QW0_DEST_VSI_MASK &
79 		      ((u32)(fdata->dest_vsi ? : pf->vsi[pf->lan_vsi]->id) <<
80 		       I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT);
81 
82 	dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
83 
84 	dtype_cmd |= add ?
85 		     I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
86 		     I40E_TXD_FLTR_QW1_PCMD_SHIFT :
87 		     I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
88 		     I40E_TXD_FLTR_QW1_PCMD_SHIFT;
89 
90 	dtype_cmd |= I40E_TXD_FLTR_QW1_DEST_MASK &
91 		     (fdata->dest_ctl << I40E_TXD_FLTR_QW1_DEST_SHIFT);
92 
93 	dtype_cmd |= I40E_TXD_FLTR_QW1_FD_STATUS_MASK &
94 		     (fdata->fd_status << I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT);
95 
96 	if (fdata->cnt_index) {
97 		dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
98 		dtype_cmd |= I40E_TXD_FLTR_QW1_CNTINDEX_MASK &
99 			     ((u32)fdata->cnt_index <<
100 			      I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT);
101 	}
102 
103 	fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
104 	fdir_desc->rsvd = cpu_to_le32(0);
105 	fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
106 	fdir_desc->fd_id = cpu_to_le32(fdata->fd_id);
107 }
108 
109 #define I40E_FD_CLEAN_DELAY 10
110 /**
111  * i40e_program_fdir_filter - Program a Flow Director filter
112  * @fdir_data: Packet data that will be filter parameters
113  * @raw_packet: the pre-allocated packet buffer for FDir
114  * @pf: The PF pointer
115  * @add: True for add/update, False for remove
116  **/
117 static int i40e_program_fdir_filter(struct i40e_fdir_filter *fdir_data,
118 				    u8 *raw_packet, struct i40e_pf *pf,
119 				    bool add)
120 {
121 	struct i40e_tx_buffer *tx_buf, *first;
122 	struct i40e_tx_desc *tx_desc;
123 	struct i40e_ring *tx_ring;
124 	struct i40e_vsi *vsi;
125 	struct device *dev;
126 	dma_addr_t dma;
127 	u32 td_cmd = 0;
128 	u16 i;
129 
130 	/* find existing FDIR VSI */
131 	vsi = i40e_find_vsi_by_type(pf, I40E_VSI_FDIR);
132 	if (!vsi)
133 		return -ENOENT;
134 
135 	tx_ring = vsi->tx_rings[0];
136 	dev = tx_ring->dev;
137 
138 	/* we need two descriptors to add/del a filter and we can wait */
139 	for (i = I40E_FD_CLEAN_DELAY; I40E_DESC_UNUSED(tx_ring) < 2; i--) {
140 		if (!i)
141 			return -EAGAIN;
142 		msleep_interruptible(1);
143 	}
144 
145 	dma = dma_map_single(dev, raw_packet,
146 			     I40E_FDIR_MAX_RAW_PACKET_SIZE, DMA_TO_DEVICE);
147 	if (dma_mapping_error(dev, dma))
148 		goto dma_fail;
149 
150 	/* grab the next descriptor */
151 	i = tx_ring->next_to_use;
152 	first = &tx_ring->tx_bi[i];
153 	i40e_fdir(tx_ring, fdir_data, add);
154 
155 	/* Now program a dummy descriptor */
156 	i = tx_ring->next_to_use;
157 	tx_desc = I40E_TX_DESC(tx_ring, i);
158 	tx_buf = &tx_ring->tx_bi[i];
159 
160 	tx_ring->next_to_use = ((i + 1) < tx_ring->count) ? i + 1 : 0;
161 
162 	memset(tx_buf, 0, sizeof(struct i40e_tx_buffer));
163 
164 	/* record length, and DMA address */
165 	dma_unmap_len_set(tx_buf, len, I40E_FDIR_MAX_RAW_PACKET_SIZE);
166 	dma_unmap_addr_set(tx_buf, dma, dma);
167 
168 	tx_desc->buffer_addr = cpu_to_le64(dma);
169 	td_cmd = I40E_TXD_CMD | I40E_TX_DESC_CMD_DUMMY;
170 
171 	tx_buf->tx_flags = I40E_TX_FLAGS_FD_SB;
172 	tx_buf->raw_buf = (void *)raw_packet;
173 
174 	tx_desc->cmd_type_offset_bsz =
175 		build_ctob(td_cmd, 0, I40E_FDIR_MAX_RAW_PACKET_SIZE, 0);
176 
177 	/* Force memory writes to complete before letting h/w
178 	 * know there are new descriptors to fetch.
179 	 */
180 	wmb();
181 
182 	/* Mark the data descriptor to be watched */
183 	first->next_to_watch = tx_desc;
184 
185 	writel(tx_ring->next_to_use, tx_ring->tail);
186 	return 0;
187 
188 dma_fail:
189 	return -1;
190 }
191 
192 #define IP_HEADER_OFFSET 14
193 #define I40E_UDPIP_DUMMY_PACKET_LEN 42
194 /**
195  * i40e_add_del_fdir_udpv4 - Add/Remove UDPv4 filters
196  * @vsi: pointer to the targeted VSI
197  * @fd_data: the flow director data required for the FDir descriptor
198  * @add: true adds a filter, false removes it
199  *
200  * Returns 0 if the filters were successfully added or removed
201  **/
202 static int i40e_add_del_fdir_udpv4(struct i40e_vsi *vsi,
203 				   struct i40e_fdir_filter *fd_data,
204 				   bool add)
205 {
206 	struct i40e_pf *pf = vsi->back;
207 	struct udphdr *udp;
208 	struct iphdr *ip;
209 	u8 *raw_packet;
210 	int ret;
211 	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
212 		0x45, 0, 0, 0x1c, 0, 0, 0x40, 0, 0x40, 0x11, 0, 0, 0, 0, 0, 0,
213 		0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
214 
215 	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
216 	if (!raw_packet)
217 		return -ENOMEM;
218 	memcpy(raw_packet, packet, I40E_UDPIP_DUMMY_PACKET_LEN);
219 
220 	ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
221 	udp = (struct udphdr *)(raw_packet + IP_HEADER_OFFSET
222 	      + sizeof(struct iphdr));
223 
224 	ip->daddr = fd_data->dst_ip;
225 	udp->dest = fd_data->dst_port;
226 	ip->saddr = fd_data->src_ip;
227 	udp->source = fd_data->src_port;
228 
229 	if (fd_data->flex_filter) {
230 		u8 *payload = raw_packet + I40E_UDPIP_DUMMY_PACKET_LEN;
231 		__be16 pattern = fd_data->flex_word;
232 		u16 off = fd_data->flex_offset;
233 
234 		*((__force __be16 *)(payload + off)) = pattern;
235 	}
236 
237 	fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_UDP;
238 	ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
239 	if (ret) {
240 		dev_info(&pf->pdev->dev,
241 			 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
242 			 fd_data->pctype, fd_data->fd_id, ret);
243 		/* Free the packet buffer since it wasn't added to the ring */
244 		kfree(raw_packet);
245 		return -EOPNOTSUPP;
246 	} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
247 		if (add)
248 			dev_info(&pf->pdev->dev,
249 				 "Filter OK for PCTYPE %d loc = %d\n",
250 				 fd_data->pctype, fd_data->fd_id);
251 		else
252 			dev_info(&pf->pdev->dev,
253 				 "Filter deleted for PCTYPE %d loc = %d\n",
254 				 fd_data->pctype, fd_data->fd_id);
255 	}
256 
257 	if (add)
258 		pf->fd_udp4_filter_cnt++;
259 	else
260 		pf->fd_udp4_filter_cnt--;
261 
262 	return 0;
263 }
264 
265 #define I40E_TCPIP_DUMMY_PACKET_LEN 54
266 /**
267  * i40e_add_del_fdir_tcpv4 - Add/Remove TCPv4 filters
268  * @vsi: pointer to the targeted VSI
269  * @fd_data: the flow director data required for the FDir descriptor
270  * @add: true adds a filter, false removes it
271  *
272  * Returns 0 if the filters were successfully added or removed
273  **/
274 static int i40e_add_del_fdir_tcpv4(struct i40e_vsi *vsi,
275 				   struct i40e_fdir_filter *fd_data,
276 				   bool add)
277 {
278 	struct i40e_pf *pf = vsi->back;
279 	struct tcphdr *tcp;
280 	struct iphdr *ip;
281 	u8 *raw_packet;
282 	int ret;
283 	/* Dummy packet */
284 	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
285 		0x45, 0, 0, 0x28, 0, 0, 0x40, 0, 0x40, 0x6, 0, 0, 0, 0, 0, 0,
286 		0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x80, 0x11,
287 		0x0, 0x72, 0, 0, 0, 0};
288 
289 	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
290 	if (!raw_packet)
291 		return -ENOMEM;
292 	memcpy(raw_packet, packet, I40E_TCPIP_DUMMY_PACKET_LEN);
293 
294 	ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
295 	tcp = (struct tcphdr *)(raw_packet + IP_HEADER_OFFSET
296 	      + sizeof(struct iphdr));
297 
298 	ip->daddr = fd_data->dst_ip;
299 	tcp->dest = fd_data->dst_port;
300 	ip->saddr = fd_data->src_ip;
301 	tcp->source = fd_data->src_port;
302 
303 	if (fd_data->flex_filter) {
304 		u8 *payload = raw_packet + I40E_TCPIP_DUMMY_PACKET_LEN;
305 		__be16 pattern = fd_data->flex_word;
306 		u16 off = fd_data->flex_offset;
307 
308 		*((__force __be16 *)(payload + off)) = pattern;
309 	}
310 
311 	fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_TCP;
312 	ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
313 	if (ret) {
314 		dev_info(&pf->pdev->dev,
315 			 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
316 			 fd_data->pctype, fd_data->fd_id, ret);
317 		/* Free the packet buffer since it wasn't added to the ring */
318 		kfree(raw_packet);
319 		return -EOPNOTSUPP;
320 	} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
321 		if (add)
322 			dev_info(&pf->pdev->dev, "Filter OK for PCTYPE %d loc = %d)\n",
323 				 fd_data->pctype, fd_data->fd_id);
324 		else
325 			dev_info(&pf->pdev->dev,
326 				 "Filter deleted for PCTYPE %d loc = %d\n",
327 				 fd_data->pctype, fd_data->fd_id);
328 	}
329 
330 	if (add) {
331 		pf->fd_tcp4_filter_cnt++;
332 		if ((pf->flags & I40E_FLAG_FD_ATR_ENABLED) &&
333 		    I40E_DEBUG_FD & pf->hw.debug_mask)
334 			dev_info(&pf->pdev->dev, "Forcing ATR off, sideband rules for TCP/IPv4 flow being applied\n");
335 		pf->hw_disabled_flags |= I40E_FLAG_FD_ATR_ENABLED;
336 	} else {
337 		pf->fd_tcp4_filter_cnt--;
338 		if (pf->fd_tcp4_filter_cnt == 0) {
339 			if ((pf->flags & I40E_FLAG_FD_ATR_ENABLED) &&
340 			    I40E_DEBUG_FD & pf->hw.debug_mask)
341 				dev_info(&pf->pdev->dev, "ATR re-enabled due to no sideband TCP/IPv4 rules\n");
342 			pf->hw_disabled_flags &= ~I40E_FLAG_FD_ATR_ENABLED;
343 		}
344 	}
345 
346 	return 0;
347 }
348 
349 #define I40E_SCTPIP_DUMMY_PACKET_LEN 46
350 /**
351  * i40e_add_del_fdir_sctpv4 - Add/Remove SCTPv4 Flow Director filters for
352  * a specific flow spec
353  * @vsi: pointer to the targeted VSI
354  * @fd_data: the flow director data required for the FDir descriptor
355  * @add: true adds a filter, false removes it
356  *
357  * Returns 0 if the filters were successfully added or removed
358  **/
359 static int i40e_add_del_fdir_sctpv4(struct i40e_vsi *vsi,
360 				    struct i40e_fdir_filter *fd_data,
361 				    bool add)
362 {
363 	struct i40e_pf *pf = vsi->back;
364 	struct sctphdr *sctp;
365 	struct iphdr *ip;
366 	u8 *raw_packet;
367 	int ret;
368 	/* Dummy packet */
369 	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
370 		0x45, 0, 0, 0x20, 0, 0, 0x40, 0, 0x40, 0x84, 0, 0, 0, 0, 0, 0,
371 		0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
372 
373 	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
374 	if (!raw_packet)
375 		return -ENOMEM;
376 	memcpy(raw_packet, packet, I40E_SCTPIP_DUMMY_PACKET_LEN);
377 
378 	ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
379 	sctp = (struct sctphdr *)(raw_packet + IP_HEADER_OFFSET
380 	      + sizeof(struct iphdr));
381 
382 	ip->daddr = fd_data->dst_ip;
383 	sctp->dest = fd_data->dst_port;
384 	ip->saddr = fd_data->src_ip;
385 	sctp->source = fd_data->src_port;
386 
387 	if (fd_data->flex_filter) {
388 		u8 *payload = raw_packet + I40E_SCTPIP_DUMMY_PACKET_LEN;
389 		__be16 pattern = fd_data->flex_word;
390 		u16 off = fd_data->flex_offset;
391 
392 		*((__force __be16 *)(payload + off)) = pattern;
393 	}
394 
395 	fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_SCTP;
396 	ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
397 	if (ret) {
398 		dev_info(&pf->pdev->dev,
399 			 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
400 			 fd_data->pctype, fd_data->fd_id, ret);
401 		/* Free the packet buffer since it wasn't added to the ring */
402 		kfree(raw_packet);
403 		return -EOPNOTSUPP;
404 	} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
405 		if (add)
406 			dev_info(&pf->pdev->dev,
407 				 "Filter OK for PCTYPE %d loc = %d\n",
408 				 fd_data->pctype, fd_data->fd_id);
409 		else
410 			dev_info(&pf->pdev->dev,
411 				 "Filter deleted for PCTYPE %d loc = %d\n",
412 				 fd_data->pctype, fd_data->fd_id);
413 	}
414 
415 	if (add)
416 		pf->fd_sctp4_filter_cnt++;
417 	else
418 		pf->fd_sctp4_filter_cnt--;
419 
420 	return 0;
421 }
422 
423 #define I40E_IP_DUMMY_PACKET_LEN 34
424 /**
425  * i40e_add_del_fdir_ipv4 - Add/Remove IPv4 Flow Director filters for
426  * a specific flow spec
427  * @vsi: pointer to the targeted VSI
428  * @fd_data: the flow director data required for the FDir descriptor
429  * @add: true adds a filter, false removes it
430  *
431  * Returns 0 if the filters were successfully added or removed
432  **/
433 static int i40e_add_del_fdir_ipv4(struct i40e_vsi *vsi,
434 				  struct i40e_fdir_filter *fd_data,
435 				  bool add)
436 {
437 	struct i40e_pf *pf = vsi->back;
438 	struct iphdr *ip;
439 	u8 *raw_packet;
440 	int ret;
441 	int i;
442 	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
443 		0x45, 0, 0, 0x14, 0, 0, 0x40, 0, 0x40, 0x10, 0, 0, 0, 0, 0, 0,
444 		0, 0, 0, 0};
445 
446 	for (i = I40E_FILTER_PCTYPE_NONF_IPV4_OTHER;
447 	     i <= I40E_FILTER_PCTYPE_FRAG_IPV4;	i++) {
448 		raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
449 		if (!raw_packet)
450 			return -ENOMEM;
451 		memcpy(raw_packet, packet, I40E_IP_DUMMY_PACKET_LEN);
452 		ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
453 
454 		ip->saddr = fd_data->src_ip;
455 		ip->daddr = fd_data->dst_ip;
456 		ip->protocol = 0;
457 
458 		if (fd_data->flex_filter) {
459 			u8 *payload = raw_packet + I40E_IP_DUMMY_PACKET_LEN;
460 			__be16 pattern = fd_data->flex_word;
461 			u16 off = fd_data->flex_offset;
462 
463 			*((__force __be16 *)(payload + off)) = pattern;
464 		}
465 
466 		fd_data->pctype = i;
467 		ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
468 		if (ret) {
469 			dev_info(&pf->pdev->dev,
470 				 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
471 				 fd_data->pctype, fd_data->fd_id, ret);
472 			/* The packet buffer wasn't added to the ring so we
473 			 * need to free it now.
474 			 */
475 			kfree(raw_packet);
476 			return -EOPNOTSUPP;
477 		} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
478 			if (add)
479 				dev_info(&pf->pdev->dev,
480 					 "Filter OK for PCTYPE %d loc = %d\n",
481 					 fd_data->pctype, fd_data->fd_id);
482 			else
483 				dev_info(&pf->pdev->dev,
484 					 "Filter deleted for PCTYPE %d loc = %d\n",
485 					 fd_data->pctype, fd_data->fd_id);
486 		}
487 	}
488 
489 	if (add)
490 		pf->fd_ip4_filter_cnt++;
491 	else
492 		pf->fd_ip4_filter_cnt--;
493 
494 	return 0;
495 }
496 
497 /**
498  * i40e_add_del_fdir - Build raw packets to add/del fdir filter
499  * @vsi: pointer to the targeted VSI
500  * @cmd: command to get or set RX flow classification rules
501  * @add: true adds a filter, false removes it
502  *
503  **/
504 int i40e_add_del_fdir(struct i40e_vsi *vsi,
505 		      struct i40e_fdir_filter *input, bool add)
506 {
507 	struct i40e_pf *pf = vsi->back;
508 	int ret;
509 
510 	switch (input->flow_type & ~FLOW_EXT) {
511 	case TCP_V4_FLOW:
512 		ret = i40e_add_del_fdir_tcpv4(vsi, input, add);
513 		break;
514 	case UDP_V4_FLOW:
515 		ret = i40e_add_del_fdir_udpv4(vsi, input, add);
516 		break;
517 	case SCTP_V4_FLOW:
518 		ret = i40e_add_del_fdir_sctpv4(vsi, input, add);
519 		break;
520 	case IP_USER_FLOW:
521 		switch (input->ip4_proto) {
522 		case IPPROTO_TCP:
523 			ret = i40e_add_del_fdir_tcpv4(vsi, input, add);
524 			break;
525 		case IPPROTO_UDP:
526 			ret = i40e_add_del_fdir_udpv4(vsi, input, add);
527 			break;
528 		case IPPROTO_SCTP:
529 			ret = i40e_add_del_fdir_sctpv4(vsi, input, add);
530 			break;
531 		case IPPROTO_IP:
532 			ret = i40e_add_del_fdir_ipv4(vsi, input, add);
533 			break;
534 		default:
535 			/* We cannot support masking based on protocol */
536 			goto unsupported_flow;
537 		}
538 		break;
539 	default:
540 unsupported_flow:
541 		dev_info(&pf->pdev->dev, "Could not specify spec type %d\n",
542 			 input->flow_type);
543 		ret = -EINVAL;
544 	}
545 
546 	/* The buffer allocated here will be normally be freed by
547 	 * i40e_clean_fdir_tx_irq() as it reclaims resources after transmit
548 	 * completion. In the event of an error adding the buffer to the FDIR
549 	 * ring, it will immediately be freed. It may also be freed by
550 	 * i40e_clean_tx_ring() when closing the VSI.
551 	 */
552 	return ret;
553 }
554 
555 /**
556  * i40e_fd_handle_status - check the Programming Status for FD
557  * @rx_ring: the Rx ring for this descriptor
558  * @rx_desc: the Rx descriptor for programming Status, not a packet descriptor.
559  * @prog_id: the id originally used for programming
560  *
561  * This is used to verify if the FD programming or invalidation
562  * requested by SW to the HW is successful or not and take actions accordingly.
563  **/
564 static void i40e_fd_handle_status(struct i40e_ring *rx_ring,
565 				  union i40e_rx_desc *rx_desc, u8 prog_id)
566 {
567 	struct i40e_pf *pf = rx_ring->vsi->back;
568 	struct pci_dev *pdev = pf->pdev;
569 	u32 fcnt_prog, fcnt_avail;
570 	u32 error;
571 	u64 qw;
572 
573 	qw = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
574 	error = (qw & I40E_RX_PROG_STATUS_DESC_QW1_ERROR_MASK) >>
575 		I40E_RX_PROG_STATUS_DESC_QW1_ERROR_SHIFT;
576 
577 	if (error == BIT(I40E_RX_PROG_STATUS_DESC_FD_TBL_FULL_SHIFT)) {
578 		pf->fd_inv = le32_to_cpu(rx_desc->wb.qword0.hi_dword.fd_id);
579 		if ((rx_desc->wb.qword0.hi_dword.fd_id != 0) ||
580 		    (I40E_DEBUG_FD & pf->hw.debug_mask))
581 			dev_warn(&pdev->dev, "ntuple filter loc = %d, could not be added\n",
582 				 pf->fd_inv);
583 
584 		/* Check if the programming error is for ATR.
585 		 * If so, auto disable ATR and set a state for
586 		 * flush in progress. Next time we come here if flush is in
587 		 * progress do nothing, once flush is complete the state will
588 		 * be cleared.
589 		 */
590 		if (test_bit(__I40E_FD_FLUSH_REQUESTED, &pf->state))
591 			return;
592 
593 		pf->fd_add_err++;
594 		/* store the current atr filter count */
595 		pf->fd_atr_cnt = i40e_get_current_atr_cnt(pf);
596 
597 		if ((rx_desc->wb.qword0.hi_dword.fd_id == 0) &&
598 		    (pf->hw_disabled_flags & I40E_FLAG_FD_SB_ENABLED)) {
599 			pf->hw_disabled_flags |= I40E_FLAG_FD_ATR_ENABLED;
600 			set_bit(__I40E_FD_FLUSH_REQUESTED, &pf->state);
601 		}
602 
603 		/* filter programming failed most likely due to table full */
604 		fcnt_prog = i40e_get_global_fd_count(pf);
605 		fcnt_avail = pf->fdir_pf_filter_count;
606 		/* If ATR is running fcnt_prog can quickly change,
607 		 * if we are very close to full, it makes sense to disable
608 		 * FD ATR/SB and then re-enable it when there is room.
609 		 */
610 		if (fcnt_prog >= (fcnt_avail - I40E_FDIR_BUFFER_FULL_MARGIN)) {
611 			if ((pf->flags & I40E_FLAG_FD_SB_ENABLED) &&
612 			    !(pf->hw_disabled_flags &
613 				     I40E_FLAG_FD_SB_ENABLED)) {
614 				if (I40E_DEBUG_FD & pf->hw.debug_mask)
615 					dev_warn(&pdev->dev, "FD filter space full, new ntuple rules will not be added\n");
616 				pf->hw_disabled_flags |=
617 							I40E_FLAG_FD_SB_ENABLED;
618 			}
619 		}
620 	} else if (error == BIT(I40E_RX_PROG_STATUS_DESC_NO_FD_ENTRY_SHIFT)) {
621 		if (I40E_DEBUG_FD & pf->hw.debug_mask)
622 			dev_info(&pdev->dev, "ntuple filter fd_id = %d, could not be removed\n",
623 				 rx_desc->wb.qword0.hi_dword.fd_id);
624 	}
625 }
626 
627 /**
628  * i40e_unmap_and_free_tx_resource - Release a Tx buffer
629  * @ring:      the ring that owns the buffer
630  * @tx_buffer: the buffer to free
631  **/
632 static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring,
633 					    struct i40e_tx_buffer *tx_buffer)
634 {
635 	if (tx_buffer->skb) {
636 		if (tx_buffer->tx_flags & I40E_TX_FLAGS_FD_SB)
637 			kfree(tx_buffer->raw_buf);
638 		else
639 			dev_kfree_skb_any(tx_buffer->skb);
640 		if (dma_unmap_len(tx_buffer, len))
641 			dma_unmap_single(ring->dev,
642 					 dma_unmap_addr(tx_buffer, dma),
643 					 dma_unmap_len(tx_buffer, len),
644 					 DMA_TO_DEVICE);
645 	} else if (dma_unmap_len(tx_buffer, len)) {
646 		dma_unmap_page(ring->dev,
647 			       dma_unmap_addr(tx_buffer, dma),
648 			       dma_unmap_len(tx_buffer, len),
649 			       DMA_TO_DEVICE);
650 	}
651 
652 	tx_buffer->next_to_watch = NULL;
653 	tx_buffer->skb = NULL;
654 	dma_unmap_len_set(tx_buffer, len, 0);
655 	/* tx_buffer must be completely set up in the transmit path */
656 }
657 
658 /**
659  * i40e_clean_tx_ring - Free any empty Tx buffers
660  * @tx_ring: ring to be cleaned
661  **/
662 void i40e_clean_tx_ring(struct i40e_ring *tx_ring)
663 {
664 	unsigned long bi_size;
665 	u16 i;
666 
667 	/* ring already cleared, nothing to do */
668 	if (!tx_ring->tx_bi)
669 		return;
670 
671 	/* Free all the Tx ring sk_buffs */
672 	for (i = 0; i < tx_ring->count; i++)
673 		i40e_unmap_and_free_tx_resource(tx_ring, &tx_ring->tx_bi[i]);
674 
675 	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
676 	memset(tx_ring->tx_bi, 0, bi_size);
677 
678 	/* Zero out the descriptor ring */
679 	memset(tx_ring->desc, 0, tx_ring->size);
680 
681 	tx_ring->next_to_use = 0;
682 	tx_ring->next_to_clean = 0;
683 
684 	if (!tx_ring->netdev)
685 		return;
686 
687 	/* cleanup Tx queue statistics */
688 	netdev_tx_reset_queue(txring_txq(tx_ring));
689 }
690 
691 /**
692  * i40e_free_tx_resources - Free Tx resources per queue
693  * @tx_ring: Tx descriptor ring for a specific queue
694  *
695  * Free all transmit software resources
696  **/
697 void i40e_free_tx_resources(struct i40e_ring *tx_ring)
698 {
699 	i40e_clean_tx_ring(tx_ring);
700 	kfree(tx_ring->tx_bi);
701 	tx_ring->tx_bi = NULL;
702 
703 	if (tx_ring->desc) {
704 		dma_free_coherent(tx_ring->dev, tx_ring->size,
705 				  tx_ring->desc, tx_ring->dma);
706 		tx_ring->desc = NULL;
707 	}
708 }
709 
710 /**
711  * i40e_get_tx_pending - how many tx descriptors not processed
712  * @tx_ring: the ring of descriptors
713  * @in_sw: is tx_pending being checked in SW or HW
714  *
715  * Since there is no access to the ring head register
716  * in XL710, we need to use our local copies
717  **/
718 u32 i40e_get_tx_pending(struct i40e_ring *ring, bool in_sw)
719 {
720 	u32 head, tail;
721 
722 	if (!in_sw)
723 		head = i40e_get_head(ring);
724 	else
725 		head = ring->next_to_clean;
726 	tail = readl(ring->tail);
727 
728 	if (head != tail)
729 		return (head < tail) ?
730 			tail - head : (tail + ring->count - head);
731 
732 	return 0;
733 }
734 
735 #define WB_STRIDE 4
736 
737 /**
738  * i40e_clean_tx_irq - Reclaim resources after transmit completes
739  * @vsi: the VSI we care about
740  * @tx_ring: Tx ring to clean
741  * @napi_budget: Used to determine if we are in netpoll
742  *
743  * Returns true if there's any budget left (e.g. the clean is finished)
744  **/
745 static bool i40e_clean_tx_irq(struct i40e_vsi *vsi,
746 			      struct i40e_ring *tx_ring, int napi_budget)
747 {
748 	u16 i = tx_ring->next_to_clean;
749 	struct i40e_tx_buffer *tx_buf;
750 	struct i40e_tx_desc *tx_head;
751 	struct i40e_tx_desc *tx_desc;
752 	unsigned int total_bytes = 0, total_packets = 0;
753 	unsigned int budget = vsi->work_limit;
754 
755 	tx_buf = &tx_ring->tx_bi[i];
756 	tx_desc = I40E_TX_DESC(tx_ring, i);
757 	i -= tx_ring->count;
758 
759 	tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
760 
761 	do {
762 		struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
763 
764 		/* if next_to_watch is not set then there is no work pending */
765 		if (!eop_desc)
766 			break;
767 
768 		/* prevent any other reads prior to eop_desc */
769 		read_barrier_depends();
770 
771 		/* we have caught up to head, no work left to do */
772 		if (tx_head == tx_desc)
773 			break;
774 
775 		/* clear next_to_watch to prevent false hangs */
776 		tx_buf->next_to_watch = NULL;
777 
778 		/* update the statistics for this packet */
779 		total_bytes += tx_buf->bytecount;
780 		total_packets += tx_buf->gso_segs;
781 
782 		/* free the skb */
783 		napi_consume_skb(tx_buf->skb, napi_budget);
784 
785 		/* unmap skb header data */
786 		dma_unmap_single(tx_ring->dev,
787 				 dma_unmap_addr(tx_buf, dma),
788 				 dma_unmap_len(tx_buf, len),
789 				 DMA_TO_DEVICE);
790 
791 		/* clear tx_buffer data */
792 		tx_buf->skb = NULL;
793 		dma_unmap_len_set(tx_buf, len, 0);
794 
795 		/* unmap remaining buffers */
796 		while (tx_desc != eop_desc) {
797 
798 			tx_buf++;
799 			tx_desc++;
800 			i++;
801 			if (unlikely(!i)) {
802 				i -= tx_ring->count;
803 				tx_buf = tx_ring->tx_bi;
804 				tx_desc = I40E_TX_DESC(tx_ring, 0);
805 			}
806 
807 			/* unmap any remaining paged data */
808 			if (dma_unmap_len(tx_buf, len)) {
809 				dma_unmap_page(tx_ring->dev,
810 					       dma_unmap_addr(tx_buf, dma),
811 					       dma_unmap_len(tx_buf, len),
812 					       DMA_TO_DEVICE);
813 				dma_unmap_len_set(tx_buf, len, 0);
814 			}
815 		}
816 
817 		/* move us one more past the eop_desc for start of next pkt */
818 		tx_buf++;
819 		tx_desc++;
820 		i++;
821 		if (unlikely(!i)) {
822 			i -= tx_ring->count;
823 			tx_buf = tx_ring->tx_bi;
824 			tx_desc = I40E_TX_DESC(tx_ring, 0);
825 		}
826 
827 		prefetch(tx_desc);
828 
829 		/* update budget accounting */
830 		budget--;
831 	} while (likely(budget));
832 
833 	i += tx_ring->count;
834 	tx_ring->next_to_clean = i;
835 	u64_stats_update_begin(&tx_ring->syncp);
836 	tx_ring->stats.bytes += total_bytes;
837 	tx_ring->stats.packets += total_packets;
838 	u64_stats_update_end(&tx_ring->syncp);
839 	tx_ring->q_vector->tx.total_bytes += total_bytes;
840 	tx_ring->q_vector->tx.total_packets += total_packets;
841 
842 	if (tx_ring->flags & I40E_TXR_FLAGS_WB_ON_ITR) {
843 		/* check to see if there are < 4 descriptors
844 		 * waiting to be written back, then kick the hardware to force
845 		 * them to be written back in case we stay in NAPI.
846 		 * In this mode on X722 we do not enable Interrupt.
847 		 */
848 		unsigned int j = i40e_get_tx_pending(tx_ring, false);
849 
850 		if (budget &&
851 		    ((j / WB_STRIDE) == 0) && (j > 0) &&
852 		    !test_bit(__I40E_DOWN, &vsi->state) &&
853 		    (I40E_DESC_UNUSED(tx_ring) != tx_ring->count))
854 			tx_ring->arm_wb = true;
855 	}
856 
857 	/* notify netdev of completed buffers */
858 	netdev_tx_completed_queue(txring_txq(tx_ring),
859 				  total_packets, total_bytes);
860 
861 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
862 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
863 		     (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
864 		/* Make sure that anybody stopping the queue after this
865 		 * sees the new next_to_clean.
866 		 */
867 		smp_mb();
868 		if (__netif_subqueue_stopped(tx_ring->netdev,
869 					     tx_ring->queue_index) &&
870 		   !test_bit(__I40E_DOWN, &vsi->state)) {
871 			netif_wake_subqueue(tx_ring->netdev,
872 					    tx_ring->queue_index);
873 			++tx_ring->tx_stats.restart_queue;
874 		}
875 	}
876 
877 	return !!budget;
878 }
879 
880 /**
881  * i40e_enable_wb_on_itr - Arm hardware to do a wb, interrupts are not enabled
882  * @vsi: the VSI we care about
883  * @q_vector: the vector on which to enable writeback
884  *
885  **/
886 static void i40e_enable_wb_on_itr(struct i40e_vsi *vsi,
887 				  struct i40e_q_vector *q_vector)
888 {
889 	u16 flags = q_vector->tx.ring[0].flags;
890 	u32 val;
891 
892 	if (!(flags & I40E_TXR_FLAGS_WB_ON_ITR))
893 		return;
894 
895 	if (q_vector->arm_wb_state)
896 		return;
897 
898 	if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
899 		val = I40E_PFINT_DYN_CTLN_WB_ON_ITR_MASK |
900 		      I40E_PFINT_DYN_CTLN_ITR_INDX_MASK; /* set noitr */
901 
902 		wr32(&vsi->back->hw,
903 		     I40E_PFINT_DYN_CTLN(q_vector->v_idx + vsi->base_vector - 1),
904 		     val);
905 	} else {
906 		val = I40E_PFINT_DYN_CTL0_WB_ON_ITR_MASK |
907 		      I40E_PFINT_DYN_CTL0_ITR_INDX_MASK; /* set noitr */
908 
909 		wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
910 	}
911 	q_vector->arm_wb_state = true;
912 }
913 
914 /**
915  * i40e_force_wb - Issue SW Interrupt so HW does a wb
916  * @vsi: the VSI we care about
917  * @q_vector: the vector  on which to force writeback
918  *
919  **/
920 void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector)
921 {
922 	if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
923 		u32 val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
924 			  I40E_PFINT_DYN_CTLN_ITR_INDX_MASK | /* set noitr */
925 			  I40E_PFINT_DYN_CTLN_SWINT_TRIG_MASK |
926 			  I40E_PFINT_DYN_CTLN_SW_ITR_INDX_ENA_MASK;
927 			  /* allow 00 to be written to the index */
928 
929 		wr32(&vsi->back->hw,
930 		     I40E_PFINT_DYN_CTLN(q_vector->v_idx +
931 					 vsi->base_vector - 1), val);
932 	} else {
933 		u32 val = I40E_PFINT_DYN_CTL0_INTENA_MASK |
934 			  I40E_PFINT_DYN_CTL0_ITR_INDX_MASK | /* set noitr */
935 			  I40E_PFINT_DYN_CTL0_SWINT_TRIG_MASK |
936 			  I40E_PFINT_DYN_CTL0_SW_ITR_INDX_ENA_MASK;
937 			/* allow 00 to be written to the index */
938 
939 		wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
940 	}
941 }
942 
943 /**
944  * i40e_set_new_dynamic_itr - Find new ITR level
945  * @rc: structure containing ring performance data
946  *
947  * Returns true if ITR changed, false if not
948  *
949  * Stores a new ITR value based on packets and byte counts during
950  * the last interrupt.  The advantage of per interrupt computation
951  * is faster updates and more accurate ITR for the current traffic
952  * pattern.  Constants in this function were computed based on
953  * theoretical maximum wire speed and thresholds were set based on
954  * testing data as well as attempting to minimize response time
955  * while increasing bulk throughput.
956  **/
957 static bool i40e_set_new_dynamic_itr(struct i40e_ring_container *rc)
958 {
959 	enum i40e_latency_range new_latency_range = rc->latency_range;
960 	struct i40e_q_vector *qv = rc->ring->q_vector;
961 	u32 new_itr = rc->itr;
962 	int bytes_per_int;
963 	int usecs;
964 
965 	if (rc->total_packets == 0 || !rc->itr)
966 		return false;
967 
968 	/* simple throttlerate management
969 	 *   0-10MB/s   lowest (50000 ints/s)
970 	 *  10-20MB/s   low    (20000 ints/s)
971 	 *  20-1249MB/s bulk   (18000 ints/s)
972 	 *  > 40000 Rx packets per second (8000 ints/s)
973 	 *
974 	 * The math works out because the divisor is in 10^(-6) which
975 	 * turns the bytes/us input value into MB/s values, but
976 	 * make sure to use usecs, as the register values written
977 	 * are in 2 usec increments in the ITR registers, and make sure
978 	 * to use the smoothed values that the countdown timer gives us.
979 	 */
980 	usecs = (rc->itr << 1) * ITR_COUNTDOWN_START;
981 	bytes_per_int = rc->total_bytes / usecs;
982 
983 	switch (new_latency_range) {
984 	case I40E_LOWEST_LATENCY:
985 		if (bytes_per_int > 10)
986 			new_latency_range = I40E_LOW_LATENCY;
987 		break;
988 	case I40E_LOW_LATENCY:
989 		if (bytes_per_int > 20)
990 			new_latency_range = I40E_BULK_LATENCY;
991 		else if (bytes_per_int <= 10)
992 			new_latency_range = I40E_LOWEST_LATENCY;
993 		break;
994 	case I40E_BULK_LATENCY:
995 	case I40E_ULTRA_LATENCY:
996 	default:
997 		if (bytes_per_int <= 20)
998 			new_latency_range = I40E_LOW_LATENCY;
999 		break;
1000 	}
1001 
1002 	/* this is to adjust RX more aggressively when streaming small
1003 	 * packets.  The value of 40000 was picked as it is just beyond
1004 	 * what the hardware can receive per second if in low latency
1005 	 * mode.
1006 	 */
1007 #define RX_ULTRA_PACKET_RATE 40000
1008 
1009 	if ((((rc->total_packets * 1000000) / usecs) > RX_ULTRA_PACKET_RATE) &&
1010 	    (&qv->rx == rc))
1011 		new_latency_range = I40E_ULTRA_LATENCY;
1012 
1013 	rc->latency_range = new_latency_range;
1014 
1015 	switch (new_latency_range) {
1016 	case I40E_LOWEST_LATENCY:
1017 		new_itr = I40E_ITR_50K;
1018 		break;
1019 	case I40E_LOW_LATENCY:
1020 		new_itr = I40E_ITR_20K;
1021 		break;
1022 	case I40E_BULK_LATENCY:
1023 		new_itr = I40E_ITR_18K;
1024 		break;
1025 	case I40E_ULTRA_LATENCY:
1026 		new_itr = I40E_ITR_8K;
1027 		break;
1028 	default:
1029 		break;
1030 	}
1031 
1032 	rc->total_bytes = 0;
1033 	rc->total_packets = 0;
1034 
1035 	if (new_itr != rc->itr) {
1036 		rc->itr = new_itr;
1037 		return true;
1038 	}
1039 
1040 	return false;
1041 }
1042 
1043 /**
1044  * i40e_clean_programming_status - clean the programming status descriptor
1045  * @rx_ring: the rx ring that has this descriptor
1046  * @rx_desc: the rx descriptor written back by HW
1047  *
1048  * Flow director should handle FD_FILTER_STATUS to check its filter programming
1049  * status being successful or not and take actions accordingly. FCoE should
1050  * handle its context/filter programming/invalidation status and take actions.
1051  *
1052  **/
1053 static void i40e_clean_programming_status(struct i40e_ring *rx_ring,
1054 					  union i40e_rx_desc *rx_desc)
1055 {
1056 	u64 qw;
1057 	u8 id;
1058 
1059 	qw = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1060 	id = (qw & I40E_RX_PROG_STATUS_DESC_QW1_PROGID_MASK) >>
1061 		  I40E_RX_PROG_STATUS_DESC_QW1_PROGID_SHIFT;
1062 
1063 	if (id == I40E_RX_PROG_STATUS_DESC_FD_FILTER_STATUS)
1064 		i40e_fd_handle_status(rx_ring, rx_desc, id);
1065 }
1066 
1067 /**
1068  * i40e_setup_tx_descriptors - Allocate the Tx descriptors
1069  * @tx_ring: the tx ring to set up
1070  *
1071  * Return 0 on success, negative on error
1072  **/
1073 int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring)
1074 {
1075 	struct device *dev = tx_ring->dev;
1076 	int bi_size;
1077 
1078 	if (!dev)
1079 		return -ENOMEM;
1080 
1081 	/* warn if we are about to overwrite the pointer */
1082 	WARN_ON(tx_ring->tx_bi);
1083 	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
1084 	tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
1085 	if (!tx_ring->tx_bi)
1086 		goto err;
1087 
1088 	/* round up to nearest 4K */
1089 	tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
1090 	/* add u32 for head writeback, align after this takes care of
1091 	 * guaranteeing this is at least one cache line in size
1092 	 */
1093 	tx_ring->size += sizeof(u32);
1094 	tx_ring->size = ALIGN(tx_ring->size, 4096);
1095 	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
1096 					   &tx_ring->dma, GFP_KERNEL);
1097 	if (!tx_ring->desc) {
1098 		dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
1099 			 tx_ring->size);
1100 		goto err;
1101 	}
1102 
1103 	tx_ring->next_to_use = 0;
1104 	tx_ring->next_to_clean = 0;
1105 	return 0;
1106 
1107 err:
1108 	kfree(tx_ring->tx_bi);
1109 	tx_ring->tx_bi = NULL;
1110 	return -ENOMEM;
1111 }
1112 
1113 /**
1114  * i40e_clean_rx_ring - Free Rx buffers
1115  * @rx_ring: ring to be cleaned
1116  **/
1117 void i40e_clean_rx_ring(struct i40e_ring *rx_ring)
1118 {
1119 	unsigned long bi_size;
1120 	u16 i;
1121 
1122 	/* ring already cleared, nothing to do */
1123 	if (!rx_ring->rx_bi)
1124 		return;
1125 
1126 	if (rx_ring->skb) {
1127 		dev_kfree_skb(rx_ring->skb);
1128 		rx_ring->skb = NULL;
1129 	}
1130 
1131 	/* Free all the Rx ring sk_buffs */
1132 	for (i = 0; i < rx_ring->count; i++) {
1133 		struct i40e_rx_buffer *rx_bi = &rx_ring->rx_bi[i];
1134 
1135 		if (!rx_bi->page)
1136 			continue;
1137 
1138 		/* Invalidate cache lines that may have been written to by
1139 		 * device so that we avoid corrupting memory.
1140 		 */
1141 		dma_sync_single_range_for_cpu(rx_ring->dev,
1142 					      rx_bi->dma,
1143 					      rx_bi->page_offset,
1144 					      I40E_RXBUFFER_2048,
1145 					      DMA_FROM_DEVICE);
1146 
1147 		/* free resources associated with mapping */
1148 		dma_unmap_page_attrs(rx_ring->dev, rx_bi->dma,
1149 				     PAGE_SIZE,
1150 				     DMA_FROM_DEVICE,
1151 				     I40E_RX_DMA_ATTR);
1152 		__page_frag_cache_drain(rx_bi->page, rx_bi->pagecnt_bias);
1153 
1154 		rx_bi->page = NULL;
1155 		rx_bi->page_offset = 0;
1156 	}
1157 
1158 	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
1159 	memset(rx_ring->rx_bi, 0, bi_size);
1160 
1161 	/* Zero out the descriptor ring */
1162 	memset(rx_ring->desc, 0, rx_ring->size);
1163 
1164 	rx_ring->next_to_alloc = 0;
1165 	rx_ring->next_to_clean = 0;
1166 	rx_ring->next_to_use = 0;
1167 }
1168 
1169 /**
1170  * i40e_free_rx_resources - Free Rx resources
1171  * @rx_ring: ring to clean the resources from
1172  *
1173  * Free all receive software resources
1174  **/
1175 void i40e_free_rx_resources(struct i40e_ring *rx_ring)
1176 {
1177 	i40e_clean_rx_ring(rx_ring);
1178 	kfree(rx_ring->rx_bi);
1179 	rx_ring->rx_bi = NULL;
1180 
1181 	if (rx_ring->desc) {
1182 		dma_free_coherent(rx_ring->dev, rx_ring->size,
1183 				  rx_ring->desc, rx_ring->dma);
1184 		rx_ring->desc = NULL;
1185 	}
1186 }
1187 
1188 /**
1189  * i40e_setup_rx_descriptors - Allocate Rx descriptors
1190  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
1191  *
1192  * Returns 0 on success, negative on failure
1193  **/
1194 int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring)
1195 {
1196 	struct device *dev = rx_ring->dev;
1197 	int bi_size;
1198 
1199 	/* warn if we are about to overwrite the pointer */
1200 	WARN_ON(rx_ring->rx_bi);
1201 	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
1202 	rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
1203 	if (!rx_ring->rx_bi)
1204 		goto err;
1205 
1206 	u64_stats_init(&rx_ring->syncp);
1207 
1208 	/* Round up to nearest 4K */
1209 	rx_ring->size = rx_ring->count * sizeof(union i40e_32byte_rx_desc);
1210 	rx_ring->size = ALIGN(rx_ring->size, 4096);
1211 	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
1212 					   &rx_ring->dma, GFP_KERNEL);
1213 
1214 	if (!rx_ring->desc) {
1215 		dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
1216 			 rx_ring->size);
1217 		goto err;
1218 	}
1219 
1220 	rx_ring->next_to_alloc = 0;
1221 	rx_ring->next_to_clean = 0;
1222 	rx_ring->next_to_use = 0;
1223 
1224 	return 0;
1225 err:
1226 	kfree(rx_ring->rx_bi);
1227 	rx_ring->rx_bi = NULL;
1228 	return -ENOMEM;
1229 }
1230 
1231 /**
1232  * i40e_release_rx_desc - Store the new tail and head values
1233  * @rx_ring: ring to bump
1234  * @val: new head index
1235  **/
1236 static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
1237 {
1238 	rx_ring->next_to_use = val;
1239 
1240 	/* update next to alloc since we have filled the ring */
1241 	rx_ring->next_to_alloc = val;
1242 
1243 	/* Force memory writes to complete before letting h/w
1244 	 * know there are new descriptors to fetch.  (Only
1245 	 * applicable for weak-ordered memory model archs,
1246 	 * such as IA-64).
1247 	 */
1248 	wmb();
1249 	writel(val, rx_ring->tail);
1250 }
1251 
1252 /**
1253  * i40e_alloc_mapped_page - recycle or make a new page
1254  * @rx_ring: ring to use
1255  * @bi: rx_buffer struct to modify
1256  *
1257  * Returns true if the page was successfully allocated or
1258  * reused.
1259  **/
1260 static bool i40e_alloc_mapped_page(struct i40e_ring *rx_ring,
1261 				   struct i40e_rx_buffer *bi)
1262 {
1263 	struct page *page = bi->page;
1264 	dma_addr_t dma;
1265 
1266 	/* since we are recycling buffers we should seldom need to alloc */
1267 	if (likely(page)) {
1268 		rx_ring->rx_stats.page_reuse_count++;
1269 		return true;
1270 	}
1271 
1272 	/* alloc new page for storage */
1273 	page = dev_alloc_page();
1274 	if (unlikely(!page)) {
1275 		rx_ring->rx_stats.alloc_page_failed++;
1276 		return false;
1277 	}
1278 
1279 	/* map page for use */
1280 	dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1281 				 PAGE_SIZE,
1282 				 DMA_FROM_DEVICE,
1283 				 I40E_RX_DMA_ATTR);
1284 
1285 	/* if mapping failed free memory back to system since
1286 	 * there isn't much point in holding memory we can't use
1287 	 */
1288 	if (dma_mapping_error(rx_ring->dev, dma)) {
1289 		__free_pages(page, 0);
1290 		rx_ring->rx_stats.alloc_page_failed++;
1291 		return false;
1292 	}
1293 
1294 	bi->dma = dma;
1295 	bi->page = page;
1296 	bi->page_offset = 0;
1297 
1298 	/* initialize pagecnt_bias to 1 representing we fully own page */
1299 	bi->pagecnt_bias = 1;
1300 
1301 	return true;
1302 }
1303 
1304 /**
1305  * i40e_receive_skb - Send a completed packet up the stack
1306  * @rx_ring:  rx ring in play
1307  * @skb: packet to send up
1308  * @vlan_tag: vlan tag for packet
1309  **/
1310 static void i40e_receive_skb(struct i40e_ring *rx_ring,
1311 			     struct sk_buff *skb, u16 vlan_tag)
1312 {
1313 	struct i40e_q_vector *q_vector = rx_ring->q_vector;
1314 
1315 	if ((rx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1316 	    (vlan_tag & VLAN_VID_MASK))
1317 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
1318 
1319 	napi_gro_receive(&q_vector->napi, skb);
1320 }
1321 
1322 /**
1323  * i40e_alloc_rx_buffers - Replace used receive buffers
1324  * @rx_ring: ring to place buffers on
1325  * @cleaned_count: number of buffers to replace
1326  *
1327  * Returns false if all allocations were successful, true if any fail
1328  **/
1329 bool i40e_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count)
1330 {
1331 	u16 ntu = rx_ring->next_to_use;
1332 	union i40e_rx_desc *rx_desc;
1333 	struct i40e_rx_buffer *bi;
1334 
1335 	/* do nothing if no valid netdev defined */
1336 	if (!rx_ring->netdev || !cleaned_count)
1337 		return false;
1338 
1339 	rx_desc = I40E_RX_DESC(rx_ring, ntu);
1340 	bi = &rx_ring->rx_bi[ntu];
1341 
1342 	do {
1343 		if (!i40e_alloc_mapped_page(rx_ring, bi))
1344 			goto no_buffers;
1345 
1346 		/* sync the buffer for use by the device */
1347 		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1348 						 bi->page_offset,
1349 						 I40E_RXBUFFER_2048,
1350 						 DMA_FROM_DEVICE);
1351 
1352 		/* Refresh the desc even if buffer_addrs didn't change
1353 		 * because each write-back erases this info.
1354 		 */
1355 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1356 
1357 		rx_desc++;
1358 		bi++;
1359 		ntu++;
1360 		if (unlikely(ntu == rx_ring->count)) {
1361 			rx_desc = I40E_RX_DESC(rx_ring, 0);
1362 			bi = rx_ring->rx_bi;
1363 			ntu = 0;
1364 		}
1365 
1366 		/* clear the status bits for the next_to_use descriptor */
1367 		rx_desc->wb.qword1.status_error_len = 0;
1368 
1369 		cleaned_count--;
1370 	} while (cleaned_count);
1371 
1372 	if (rx_ring->next_to_use != ntu)
1373 		i40e_release_rx_desc(rx_ring, ntu);
1374 
1375 	return false;
1376 
1377 no_buffers:
1378 	if (rx_ring->next_to_use != ntu)
1379 		i40e_release_rx_desc(rx_ring, ntu);
1380 
1381 	/* make sure to come back via polling to try again after
1382 	 * allocation failure
1383 	 */
1384 	return true;
1385 }
1386 
1387 /**
1388  * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
1389  * @vsi: the VSI we care about
1390  * @skb: skb currently being received and modified
1391  * @rx_desc: the receive descriptor
1392  **/
1393 static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
1394 				    struct sk_buff *skb,
1395 				    union i40e_rx_desc *rx_desc)
1396 {
1397 	struct i40e_rx_ptype_decoded decoded;
1398 	u32 rx_error, rx_status;
1399 	bool ipv4, ipv6;
1400 	u8 ptype;
1401 	u64 qword;
1402 
1403 	qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1404 	ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >> I40E_RXD_QW1_PTYPE_SHIFT;
1405 	rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1406 		   I40E_RXD_QW1_ERROR_SHIFT;
1407 	rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1408 		    I40E_RXD_QW1_STATUS_SHIFT;
1409 	decoded = decode_rx_desc_ptype(ptype);
1410 
1411 	skb->ip_summed = CHECKSUM_NONE;
1412 
1413 	skb_checksum_none_assert(skb);
1414 
1415 	/* Rx csum enabled and ip headers found? */
1416 	if (!(vsi->netdev->features & NETIF_F_RXCSUM))
1417 		return;
1418 
1419 	/* did the hardware decode the packet and checksum? */
1420 	if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
1421 		return;
1422 
1423 	/* both known and outer_ip must be set for the below code to work */
1424 	if (!(decoded.known && decoded.outer_ip))
1425 		return;
1426 
1427 	ipv4 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1428 	       (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4);
1429 	ipv6 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1430 	       (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6);
1431 
1432 	if (ipv4 &&
1433 	    (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) |
1434 			 BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT))))
1435 		goto checksum_fail;
1436 
1437 	/* likely incorrect csum if alternate IP extension headers found */
1438 	if (ipv6 &&
1439 	    rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
1440 		/* don't increment checksum err here, non-fatal err */
1441 		return;
1442 
1443 	/* there was some L4 error, count error and punt packet to the stack */
1444 	if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT))
1445 		goto checksum_fail;
1446 
1447 	/* handle packets that were not able to be checksummed due
1448 	 * to arrival speed, in this case the stack can compute
1449 	 * the csum.
1450 	 */
1451 	if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT))
1452 		return;
1453 
1454 	/* If there is an outer header present that might contain a checksum
1455 	 * we need to bump the checksum level by 1 to reflect the fact that
1456 	 * we are indicating we validated the inner checksum.
1457 	 */
1458 	if (decoded.tunnel_type >= I40E_RX_PTYPE_TUNNEL_IP_GRENAT)
1459 		skb->csum_level = 1;
1460 
1461 	/* Only report checksum unnecessary for TCP, UDP, or SCTP */
1462 	switch (decoded.inner_prot) {
1463 	case I40E_RX_PTYPE_INNER_PROT_TCP:
1464 	case I40E_RX_PTYPE_INNER_PROT_UDP:
1465 	case I40E_RX_PTYPE_INNER_PROT_SCTP:
1466 		skb->ip_summed = CHECKSUM_UNNECESSARY;
1467 		/* fall though */
1468 	default:
1469 		break;
1470 	}
1471 
1472 	return;
1473 
1474 checksum_fail:
1475 	vsi->back->hw_csum_rx_error++;
1476 }
1477 
1478 /**
1479  * i40e_ptype_to_htype - get a hash type
1480  * @ptype: the ptype value from the descriptor
1481  *
1482  * Returns a hash type to be used by skb_set_hash
1483  **/
1484 static inline int i40e_ptype_to_htype(u8 ptype)
1485 {
1486 	struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
1487 
1488 	if (!decoded.known)
1489 		return PKT_HASH_TYPE_NONE;
1490 
1491 	if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1492 	    decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
1493 		return PKT_HASH_TYPE_L4;
1494 	else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1495 		 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
1496 		return PKT_HASH_TYPE_L3;
1497 	else
1498 		return PKT_HASH_TYPE_L2;
1499 }
1500 
1501 /**
1502  * i40e_rx_hash - set the hash value in the skb
1503  * @ring: descriptor ring
1504  * @rx_desc: specific descriptor
1505  **/
1506 static inline void i40e_rx_hash(struct i40e_ring *ring,
1507 				union i40e_rx_desc *rx_desc,
1508 				struct sk_buff *skb,
1509 				u8 rx_ptype)
1510 {
1511 	u32 hash;
1512 	const __le64 rss_mask =
1513 		cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
1514 			    I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
1515 
1516 	if (!(ring->netdev->features & NETIF_F_RXHASH))
1517 		return;
1518 
1519 	if ((rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask) {
1520 		hash = le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
1521 		skb_set_hash(skb, hash, i40e_ptype_to_htype(rx_ptype));
1522 	}
1523 }
1524 
1525 /**
1526  * i40e_process_skb_fields - Populate skb header fields from Rx descriptor
1527  * @rx_ring: rx descriptor ring packet is being transacted on
1528  * @rx_desc: pointer to the EOP Rx descriptor
1529  * @skb: pointer to current skb being populated
1530  * @rx_ptype: the packet type decoded by hardware
1531  *
1532  * This function checks the ring, descriptor, and packet information in
1533  * order to populate the hash, checksum, VLAN, protocol, and
1534  * other fields within the skb.
1535  **/
1536 static inline
1537 void i40e_process_skb_fields(struct i40e_ring *rx_ring,
1538 			     union i40e_rx_desc *rx_desc, struct sk_buff *skb,
1539 			     u8 rx_ptype)
1540 {
1541 	u64 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1542 	u32 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1543 			I40E_RXD_QW1_STATUS_SHIFT;
1544 	u32 tsynvalid = rx_status & I40E_RXD_QW1_STATUS_TSYNVALID_MASK;
1545 	u32 tsyn = (rx_status & I40E_RXD_QW1_STATUS_TSYNINDX_MASK) >>
1546 		   I40E_RXD_QW1_STATUS_TSYNINDX_SHIFT;
1547 
1548 	if (unlikely(tsynvalid))
1549 		i40e_ptp_rx_hwtstamp(rx_ring->vsi->back, skb, tsyn);
1550 
1551 	i40e_rx_hash(rx_ring, rx_desc, skb, rx_ptype);
1552 
1553 	i40e_rx_checksum(rx_ring->vsi, skb, rx_desc);
1554 
1555 	skb_record_rx_queue(skb, rx_ring->queue_index);
1556 
1557 	/* modifies the skb - consumes the enet header */
1558 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1559 }
1560 
1561 /**
1562  * i40e_cleanup_headers - Correct empty headers
1563  * @rx_ring: rx descriptor ring packet is being transacted on
1564  * @skb: pointer to current skb being fixed
1565  *
1566  * Also address the case where we are pulling data in on pages only
1567  * and as such no data is present in the skb header.
1568  *
1569  * In addition if skb is not at least 60 bytes we need to pad it so that
1570  * it is large enough to qualify as a valid Ethernet frame.
1571  *
1572  * Returns true if an error was encountered and skb was freed.
1573  **/
1574 static bool i40e_cleanup_headers(struct i40e_ring *rx_ring, struct sk_buff *skb)
1575 {
1576 	/* if eth_skb_pad returns an error the skb was freed */
1577 	if (eth_skb_pad(skb))
1578 		return true;
1579 
1580 	return false;
1581 }
1582 
1583 /**
1584  * i40e_reuse_rx_page - page flip buffer and store it back on the ring
1585  * @rx_ring: rx descriptor ring to store buffers on
1586  * @old_buff: donor buffer to have page reused
1587  *
1588  * Synchronizes page for reuse by the adapter
1589  **/
1590 static void i40e_reuse_rx_page(struct i40e_ring *rx_ring,
1591 			       struct i40e_rx_buffer *old_buff)
1592 {
1593 	struct i40e_rx_buffer *new_buff;
1594 	u16 nta = rx_ring->next_to_alloc;
1595 
1596 	new_buff = &rx_ring->rx_bi[nta];
1597 
1598 	/* update, and store next to alloc */
1599 	nta++;
1600 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1601 
1602 	/* transfer page from old buffer to new buffer */
1603 	new_buff->dma		= old_buff->dma;
1604 	new_buff->page		= old_buff->page;
1605 	new_buff->page_offset	= old_buff->page_offset;
1606 	new_buff->pagecnt_bias	= old_buff->pagecnt_bias;
1607 }
1608 
1609 /**
1610  * i40e_page_is_reusable - check if any reuse is possible
1611  * @page: page struct to check
1612  *
1613  * A page is not reusable if it was allocated under low memory
1614  * conditions, or it's not in the same NUMA node as this CPU.
1615  */
1616 static inline bool i40e_page_is_reusable(struct page *page)
1617 {
1618 	return (page_to_nid(page) == numa_mem_id()) &&
1619 		!page_is_pfmemalloc(page);
1620 }
1621 
1622 /**
1623  * i40e_can_reuse_rx_page - Determine if this page can be reused by
1624  * the adapter for another receive
1625  *
1626  * @rx_buffer: buffer containing the page
1627  *
1628  * If page is reusable, rx_buffer->page_offset is adjusted to point to
1629  * an unused region in the page.
1630  *
1631  * For small pages, @truesize will be a constant value, half the size
1632  * of the memory at page.  We'll attempt to alternate between high and
1633  * low halves of the page, with one half ready for use by the hardware
1634  * and the other half being consumed by the stack.  We use the page
1635  * ref count to determine whether the stack has finished consuming the
1636  * portion of this page that was passed up with a previous packet.  If
1637  * the page ref count is >1, we'll assume the "other" half page is
1638  * still busy, and this page cannot be reused.
1639  *
1640  * For larger pages, @truesize will be the actual space used by the
1641  * received packet (adjusted upward to an even multiple of the cache
1642  * line size).  This will advance through the page by the amount
1643  * actually consumed by the received packets while there is still
1644  * space for a buffer.  Each region of larger pages will be used at
1645  * most once, after which the page will not be reused.
1646  *
1647  * In either case, if the page is reusable its refcount is increased.
1648  **/
1649 static bool i40e_can_reuse_rx_page(struct i40e_rx_buffer *rx_buffer)
1650 {
1651 #if (PAGE_SIZE >= 8192)
1652 	unsigned int last_offset = PAGE_SIZE - I40E_RXBUFFER_2048;
1653 #endif
1654 	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1655 	struct page *page = rx_buffer->page;
1656 
1657 	/* Is any reuse possible? */
1658 	if (unlikely(!i40e_page_is_reusable(page)))
1659 		return false;
1660 
1661 #if (PAGE_SIZE < 8192)
1662 	/* if we are only owner of page we can reuse it */
1663 	if (unlikely((page_count(page) - pagecnt_bias) > 1))
1664 		return false;
1665 #else
1666 	if (rx_buffer->page_offset > last_offset)
1667 		return false;
1668 #endif
1669 
1670 	/* If we have drained the page fragment pool we need to update
1671 	 * the pagecnt_bias and page count so that we fully restock the
1672 	 * number of references the driver holds.
1673 	 */
1674 	if (unlikely(!pagecnt_bias)) {
1675 		page_ref_add(page, USHRT_MAX);
1676 		rx_buffer->pagecnt_bias = USHRT_MAX;
1677 	}
1678 
1679 	return true;
1680 }
1681 
1682 /**
1683  * i40e_add_rx_frag - Add contents of Rx buffer to sk_buff
1684  * @rx_ring: rx descriptor ring to transact packets on
1685  * @rx_buffer: buffer containing page to add
1686  * @skb: sk_buff to place the data into
1687  * @size: packet length from rx_desc
1688  *
1689  * This function will add the data contained in rx_buffer->page to the skb.
1690  * It will just attach the page as a frag to the skb.
1691  *
1692  * The function will then update the page offset.
1693  **/
1694 static void i40e_add_rx_frag(struct i40e_ring *rx_ring,
1695 			     struct i40e_rx_buffer *rx_buffer,
1696 			     struct sk_buff *skb,
1697 			     unsigned int size)
1698 {
1699 #if (PAGE_SIZE < 8192)
1700 	unsigned int truesize = I40E_RXBUFFER_2048;
1701 #else
1702 	unsigned int truesize = SKB_DATA_ALIGN(size);
1703 #endif
1704 
1705 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
1706 			rx_buffer->page_offset, size, truesize);
1707 
1708 	/* page is being used so we must update the page offset */
1709 #if (PAGE_SIZE < 8192)
1710 	rx_buffer->page_offset ^= truesize;
1711 #else
1712 	rx_buffer->page_offset += truesize;
1713 #endif
1714 }
1715 
1716 /**
1717  * i40e_get_rx_buffer - Fetch Rx buffer and synchronize data for use
1718  * @rx_ring: rx descriptor ring to transact packets on
1719  * @size: size of buffer to add to skb
1720  *
1721  * This function will pull an Rx buffer from the ring and synchronize it
1722  * for use by the CPU.
1723  */
1724 static struct i40e_rx_buffer *i40e_get_rx_buffer(struct i40e_ring *rx_ring,
1725 						 const unsigned int size)
1726 {
1727 	struct i40e_rx_buffer *rx_buffer;
1728 
1729 	rx_buffer = &rx_ring->rx_bi[rx_ring->next_to_clean];
1730 	prefetchw(rx_buffer->page);
1731 
1732 	/* we are reusing so sync this buffer for CPU use */
1733 	dma_sync_single_range_for_cpu(rx_ring->dev,
1734 				      rx_buffer->dma,
1735 				      rx_buffer->page_offset,
1736 				      size,
1737 				      DMA_FROM_DEVICE);
1738 
1739 	/* We have pulled a buffer for use, so decrement pagecnt_bias */
1740 	rx_buffer->pagecnt_bias--;
1741 
1742 	return rx_buffer;
1743 }
1744 
1745 /**
1746  * i40e_construct_skb - Allocate skb and populate it
1747  * @rx_ring: rx descriptor ring to transact packets on
1748  * @rx_buffer: rx buffer to pull data from
1749  * @size: size of buffer to add to skb
1750  *
1751  * This function allocates an skb.  It then populates it with the page
1752  * data from the current receive descriptor, taking care to set up the
1753  * skb correctly.
1754  */
1755 static struct sk_buff *i40e_construct_skb(struct i40e_ring *rx_ring,
1756 					  struct i40e_rx_buffer *rx_buffer,
1757 					  unsigned int size)
1758 {
1759 	void *va = page_address(rx_buffer->page) + rx_buffer->page_offset;
1760 #if (PAGE_SIZE < 8192)
1761 	unsigned int truesize = I40E_RXBUFFER_2048;
1762 #else
1763 	unsigned int truesize = SKB_DATA_ALIGN(size);
1764 #endif
1765 	unsigned int headlen;
1766 	struct sk_buff *skb;
1767 
1768 	/* prefetch first cache line of first page */
1769 	prefetch(va);
1770 #if L1_CACHE_BYTES < 128
1771 	prefetch(va + L1_CACHE_BYTES);
1772 #endif
1773 
1774 	/* allocate a skb to store the frags */
1775 	skb = __napi_alloc_skb(&rx_ring->q_vector->napi,
1776 			       I40E_RX_HDR_SIZE,
1777 			       GFP_ATOMIC | __GFP_NOWARN);
1778 	if (unlikely(!skb))
1779 		return NULL;
1780 
1781 	/* Determine available headroom for copy */
1782 	headlen = size;
1783 	if (headlen > I40E_RX_HDR_SIZE)
1784 		headlen = eth_get_headlen(va, I40E_RX_HDR_SIZE);
1785 
1786 	/* align pull length to size of long to optimize memcpy performance */
1787 	memcpy(__skb_put(skb, headlen), va, ALIGN(headlen, sizeof(long)));
1788 
1789 	/* update all of the pointers */
1790 	size -= headlen;
1791 	if (size) {
1792 		skb_add_rx_frag(skb, 0, rx_buffer->page,
1793 				rx_buffer->page_offset + headlen,
1794 				size, truesize);
1795 
1796 		/* buffer is used by skb, update page_offset */
1797 #if (PAGE_SIZE < 8192)
1798 		rx_buffer->page_offset ^= truesize;
1799 #else
1800 		rx_buffer->page_offset += truesize;
1801 #endif
1802 	} else {
1803 		/* buffer is unused, reset bias back to rx_buffer */
1804 		rx_buffer->pagecnt_bias++;
1805 	}
1806 
1807 	return skb;
1808 }
1809 
1810 /**
1811  * i40e_put_rx_buffer - Clean up used buffer and either recycle or free
1812  * @rx_ring: rx descriptor ring to transact packets on
1813  * @rx_buffer: rx buffer to pull data from
1814  *
1815  * This function will clean up the contents of the rx_buffer.  It will
1816  * either recycle the bufer or unmap it and free the associated resources.
1817  */
1818 static void i40e_put_rx_buffer(struct i40e_ring *rx_ring,
1819 			       struct i40e_rx_buffer *rx_buffer)
1820 {
1821 	if (i40e_can_reuse_rx_page(rx_buffer)) {
1822 		/* hand second half of page back to the ring */
1823 		i40e_reuse_rx_page(rx_ring, rx_buffer);
1824 		rx_ring->rx_stats.page_reuse_count++;
1825 	} else {
1826 		/* we are not reusing the buffer so unmap it */
1827 		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma, PAGE_SIZE,
1828 				     DMA_FROM_DEVICE, I40E_RX_DMA_ATTR);
1829 		__page_frag_cache_drain(rx_buffer->page,
1830 					rx_buffer->pagecnt_bias);
1831 	}
1832 
1833 	/* clear contents of buffer_info */
1834 	rx_buffer->page = NULL;
1835 }
1836 
1837 /**
1838  * i40e_is_non_eop - process handling of non-EOP buffers
1839  * @rx_ring: Rx ring being processed
1840  * @rx_desc: Rx descriptor for current buffer
1841  * @skb: Current socket buffer containing buffer in progress
1842  *
1843  * This function updates next to clean.  If the buffer is an EOP buffer
1844  * this function exits returning false, otherwise it will place the
1845  * sk_buff in the next buffer to be chained and return true indicating
1846  * that this is in fact a non-EOP buffer.
1847  **/
1848 static bool i40e_is_non_eop(struct i40e_ring *rx_ring,
1849 			    union i40e_rx_desc *rx_desc,
1850 			    struct sk_buff *skb)
1851 {
1852 	u32 ntc = rx_ring->next_to_clean + 1;
1853 
1854 	/* fetch, update, and store next to clean */
1855 	ntc = (ntc < rx_ring->count) ? ntc : 0;
1856 	rx_ring->next_to_clean = ntc;
1857 
1858 	prefetch(I40E_RX_DESC(rx_ring, ntc));
1859 
1860 #define staterrlen rx_desc->wb.qword1.status_error_len
1861 	if (unlikely(i40e_rx_is_programming_status(le64_to_cpu(staterrlen)))) {
1862 		i40e_clean_programming_status(rx_ring, rx_desc);
1863 		return true;
1864 	}
1865 	/* if we are the last buffer then there is nothing else to do */
1866 #define I40E_RXD_EOF BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)
1867 	if (likely(i40e_test_staterr(rx_desc, I40E_RXD_EOF)))
1868 		return false;
1869 
1870 	rx_ring->rx_stats.non_eop_descs++;
1871 
1872 	return true;
1873 }
1874 
1875 /**
1876  * i40e_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1877  * @rx_ring: rx descriptor ring to transact packets on
1878  * @budget: Total limit on number of packets to process
1879  *
1880  * This function provides a "bounce buffer" approach to Rx interrupt
1881  * processing.  The advantage to this is that on systems that have
1882  * expensive overhead for IOMMU access this provides a means of avoiding
1883  * it by maintaining the mapping of the page to the system.
1884  *
1885  * Returns amount of work completed
1886  **/
1887 static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget)
1888 {
1889 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1890 	struct sk_buff *skb = rx_ring->skb;
1891 	u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
1892 	bool failure = false;
1893 
1894 	while (likely(total_rx_packets < budget)) {
1895 		struct i40e_rx_buffer *rx_buffer;
1896 		union i40e_rx_desc *rx_desc;
1897 		unsigned int size;
1898 		u16 vlan_tag;
1899 		u8 rx_ptype;
1900 		u64 qword;
1901 
1902 		/* return some buffers to hardware, one at a time is too slow */
1903 		if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1904 			failure = failure ||
1905 				  i40e_alloc_rx_buffers(rx_ring, cleaned_count);
1906 			cleaned_count = 0;
1907 		}
1908 
1909 		rx_desc = I40E_RX_DESC(rx_ring, rx_ring->next_to_clean);
1910 
1911 		/* status_error_len will always be zero for unused descriptors
1912 		 * because it's cleared in cleanup, and overlaps with hdr_addr
1913 		 * which is always zero because packet split isn't used, if the
1914 		 * hardware wrote DD then the length will be non-zero
1915 		 */
1916 		qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1917 		size = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1918 		       I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1919 		if (!size)
1920 			break;
1921 
1922 		/* This memory barrier is needed to keep us from reading
1923 		 * any other fields out of the rx_desc until we have
1924 		 * verified the descriptor has been written back.
1925 		 */
1926 		dma_rmb();
1927 
1928 		rx_buffer = i40e_get_rx_buffer(rx_ring, size);
1929 
1930 		/* retrieve a buffer from the ring */
1931 		if (skb)
1932 			i40e_add_rx_frag(rx_ring, rx_buffer, skb, size);
1933 		else
1934 			skb = i40e_construct_skb(rx_ring, rx_buffer, size);
1935 
1936 		/* exit if we failed to retrieve a buffer */
1937 		if (!skb) {
1938 			rx_ring->rx_stats.alloc_buff_failed++;
1939 			rx_buffer->pagecnt_bias++;
1940 			break;
1941 		}
1942 
1943 		i40e_put_rx_buffer(rx_ring, rx_buffer);
1944 		cleaned_count++;
1945 
1946 		if (i40e_is_non_eop(rx_ring, rx_desc, skb))
1947 			continue;
1948 
1949 		/* ERR_MASK will only have valid bits if EOP set, and
1950 		 * what we are doing here is actually checking
1951 		 * I40E_RX_DESC_ERROR_RXE_SHIFT, since it is the zeroth bit in
1952 		 * the error field
1953 		 */
1954 		if (unlikely(i40e_test_staterr(rx_desc, BIT(I40E_RXD_QW1_ERROR_SHIFT)))) {
1955 			dev_kfree_skb_any(skb);
1956 			skb = NULL;
1957 			continue;
1958 		}
1959 
1960 		if (i40e_cleanup_headers(rx_ring, skb)) {
1961 			skb = NULL;
1962 			continue;
1963 		}
1964 
1965 		/* probably a little skewed due to removing CRC */
1966 		total_rx_bytes += skb->len;
1967 
1968 		qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1969 		rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1970 			   I40E_RXD_QW1_PTYPE_SHIFT;
1971 
1972 		/* populate checksum, VLAN, and protocol */
1973 		i40e_process_skb_fields(rx_ring, rx_desc, skb, rx_ptype);
1974 
1975 		vlan_tag = (qword & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)) ?
1976 			   le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1) : 0;
1977 
1978 		i40e_receive_skb(rx_ring, skb, vlan_tag);
1979 		skb = NULL;
1980 
1981 		/* update budget accounting */
1982 		total_rx_packets++;
1983 	}
1984 
1985 	rx_ring->skb = skb;
1986 
1987 	u64_stats_update_begin(&rx_ring->syncp);
1988 	rx_ring->stats.packets += total_rx_packets;
1989 	rx_ring->stats.bytes += total_rx_bytes;
1990 	u64_stats_update_end(&rx_ring->syncp);
1991 	rx_ring->q_vector->rx.total_packets += total_rx_packets;
1992 	rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1993 
1994 	/* guarantee a trip back through this routine if there was a failure */
1995 	return failure ? budget : total_rx_packets;
1996 }
1997 
1998 static u32 i40e_buildreg_itr(const int type, const u16 itr)
1999 {
2000 	u32 val;
2001 
2002 	val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
2003 	      /* Don't clear PBA because that can cause lost interrupts that
2004 	       * came in while we were cleaning/polling
2005 	       */
2006 	      (type << I40E_PFINT_DYN_CTLN_ITR_INDX_SHIFT) |
2007 	      (itr << I40E_PFINT_DYN_CTLN_INTERVAL_SHIFT);
2008 
2009 	return val;
2010 }
2011 
2012 /* a small macro to shorten up some long lines */
2013 #define INTREG I40E_PFINT_DYN_CTLN
2014 static inline int get_rx_itr(struct i40e_vsi *vsi, int idx)
2015 {
2016 	return vsi->rx_rings[idx]->rx_itr_setting;
2017 }
2018 
2019 static inline int get_tx_itr(struct i40e_vsi *vsi, int idx)
2020 {
2021 	return vsi->tx_rings[idx]->tx_itr_setting;
2022 }
2023 
2024 /**
2025  * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt
2026  * @vsi: the VSI we care about
2027  * @q_vector: q_vector for which itr is being updated and interrupt enabled
2028  *
2029  **/
2030 static inline void i40e_update_enable_itr(struct i40e_vsi *vsi,
2031 					  struct i40e_q_vector *q_vector)
2032 {
2033 	struct i40e_hw *hw = &vsi->back->hw;
2034 	bool rx = false, tx = false;
2035 	u32 rxval, txval;
2036 	int vector;
2037 	int idx = q_vector->v_idx;
2038 	int rx_itr_setting, tx_itr_setting;
2039 
2040 	vector = (q_vector->v_idx + vsi->base_vector);
2041 
2042 	/* avoid dynamic calculation if in countdown mode OR if
2043 	 * all dynamic is disabled
2044 	 */
2045 	rxval = txval = i40e_buildreg_itr(I40E_ITR_NONE, 0);
2046 
2047 	rx_itr_setting = get_rx_itr(vsi, idx);
2048 	tx_itr_setting = get_tx_itr(vsi, idx);
2049 
2050 	if (q_vector->itr_countdown > 0 ||
2051 	    (!ITR_IS_DYNAMIC(rx_itr_setting) &&
2052 	     !ITR_IS_DYNAMIC(tx_itr_setting))) {
2053 		goto enable_int;
2054 	}
2055 
2056 	if (ITR_IS_DYNAMIC(tx_itr_setting)) {
2057 		rx = i40e_set_new_dynamic_itr(&q_vector->rx);
2058 		rxval = i40e_buildreg_itr(I40E_RX_ITR, q_vector->rx.itr);
2059 	}
2060 
2061 	if (ITR_IS_DYNAMIC(tx_itr_setting)) {
2062 		tx = i40e_set_new_dynamic_itr(&q_vector->tx);
2063 		txval = i40e_buildreg_itr(I40E_TX_ITR, q_vector->tx.itr);
2064 	}
2065 
2066 	if (rx || tx) {
2067 		/* get the higher of the two ITR adjustments and
2068 		 * use the same value for both ITR registers
2069 		 * when in adaptive mode (Rx and/or Tx)
2070 		 */
2071 		u16 itr = max(q_vector->tx.itr, q_vector->rx.itr);
2072 
2073 		q_vector->tx.itr = q_vector->rx.itr = itr;
2074 		txval = i40e_buildreg_itr(I40E_TX_ITR, itr);
2075 		tx = true;
2076 		rxval = i40e_buildreg_itr(I40E_RX_ITR, itr);
2077 		rx = true;
2078 	}
2079 
2080 	/* only need to enable the interrupt once, but need
2081 	 * to possibly update both ITR values
2082 	 */
2083 	if (rx) {
2084 		/* set the INTENA_MSK_MASK so that this first write
2085 		 * won't actually enable the interrupt, instead just
2086 		 * updating the ITR (it's bit 31 PF and VF)
2087 		 */
2088 		rxval |= BIT(31);
2089 		/* don't check _DOWN because interrupt isn't being enabled */
2090 		wr32(hw, INTREG(vector - 1), rxval);
2091 	}
2092 
2093 enable_int:
2094 	if (!test_bit(__I40E_DOWN, &vsi->state))
2095 		wr32(hw, INTREG(vector - 1), txval);
2096 
2097 	if (q_vector->itr_countdown)
2098 		q_vector->itr_countdown--;
2099 	else
2100 		q_vector->itr_countdown = ITR_COUNTDOWN_START;
2101 }
2102 
2103 /**
2104  * i40e_napi_poll - NAPI polling Rx/Tx cleanup routine
2105  * @napi: napi struct with our devices info in it
2106  * @budget: amount of work driver is allowed to do this pass, in packets
2107  *
2108  * This function will clean all queues associated with a q_vector.
2109  *
2110  * Returns the amount of work done
2111  **/
2112 int i40e_napi_poll(struct napi_struct *napi, int budget)
2113 {
2114 	struct i40e_q_vector *q_vector =
2115 			       container_of(napi, struct i40e_q_vector, napi);
2116 	struct i40e_vsi *vsi = q_vector->vsi;
2117 	struct i40e_ring *ring;
2118 	bool clean_complete = true;
2119 	bool arm_wb = false;
2120 	int budget_per_ring;
2121 	int work_done = 0;
2122 
2123 	if (test_bit(__I40E_DOWN, &vsi->state)) {
2124 		napi_complete(napi);
2125 		return 0;
2126 	}
2127 
2128 	/* Clear hung_detected bit */
2129 	clear_bit(I40E_Q_VECTOR_HUNG_DETECT, &q_vector->hung_detected);
2130 	/* Since the actual Tx work is minimal, we can give the Tx a larger
2131 	 * budget and be more aggressive about cleaning up the Tx descriptors.
2132 	 */
2133 	i40e_for_each_ring(ring, q_vector->tx) {
2134 		if (!i40e_clean_tx_irq(vsi, ring, budget)) {
2135 			clean_complete = false;
2136 			continue;
2137 		}
2138 		arm_wb |= ring->arm_wb;
2139 		ring->arm_wb = false;
2140 	}
2141 
2142 	/* Handle case where we are called by netpoll with a budget of 0 */
2143 	if (budget <= 0)
2144 		goto tx_only;
2145 
2146 	/* We attempt to distribute budget to each Rx queue fairly, but don't
2147 	 * allow the budget to go below 1 because that would exit polling early.
2148 	 */
2149 	budget_per_ring = max(budget/q_vector->num_ringpairs, 1);
2150 
2151 	i40e_for_each_ring(ring, q_vector->rx) {
2152 		int cleaned = i40e_clean_rx_irq(ring, budget_per_ring);
2153 
2154 		work_done += cleaned;
2155 		/* if we clean as many as budgeted, we must not be done */
2156 		if (cleaned >= budget_per_ring)
2157 			clean_complete = false;
2158 	}
2159 
2160 	/* If work not completed, return budget and polling will return */
2161 	if (!clean_complete) {
2162 		const cpumask_t *aff_mask = &q_vector->affinity_mask;
2163 		int cpu_id = smp_processor_id();
2164 
2165 		/* It is possible that the interrupt affinity has changed but,
2166 		 * if the cpu is pegged at 100%, polling will never exit while
2167 		 * traffic continues and the interrupt will be stuck on this
2168 		 * cpu.  We check to make sure affinity is correct before we
2169 		 * continue to poll, otherwise we must stop polling so the
2170 		 * interrupt can move to the correct cpu.
2171 		 */
2172 		if (likely(cpumask_test_cpu(cpu_id, aff_mask) ||
2173 			   !(vsi->back->flags & I40E_FLAG_MSIX_ENABLED))) {
2174 tx_only:
2175 			if (arm_wb) {
2176 				q_vector->tx.ring[0].tx_stats.tx_force_wb++;
2177 				i40e_enable_wb_on_itr(vsi, q_vector);
2178 			}
2179 			return budget;
2180 		}
2181 	}
2182 
2183 	if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR)
2184 		q_vector->arm_wb_state = false;
2185 
2186 	/* Work is done so exit the polling mode and re-enable the interrupt */
2187 	napi_complete_done(napi, work_done);
2188 
2189 	/* If we're prematurely stopping polling to fix the interrupt
2190 	 * affinity we want to make sure polling starts back up so we
2191 	 * issue a call to i40e_force_wb which triggers a SW interrupt.
2192 	 */
2193 	if (!clean_complete)
2194 		i40e_force_wb(vsi, q_vector);
2195 	else if (!(vsi->back->flags & I40E_FLAG_MSIX_ENABLED))
2196 		i40e_irq_dynamic_enable_icr0(vsi->back, false);
2197 	else
2198 		i40e_update_enable_itr(vsi, q_vector);
2199 
2200 	return min(work_done, budget - 1);
2201 }
2202 
2203 /**
2204  * i40e_atr - Add a Flow Director ATR filter
2205  * @tx_ring:  ring to add programming descriptor to
2206  * @skb:      send buffer
2207  * @tx_flags: send tx flags
2208  **/
2209 static void i40e_atr(struct i40e_ring *tx_ring, struct sk_buff *skb,
2210 		     u32 tx_flags)
2211 {
2212 	struct i40e_filter_program_desc *fdir_desc;
2213 	struct i40e_pf *pf = tx_ring->vsi->back;
2214 	union {
2215 		unsigned char *network;
2216 		struct iphdr *ipv4;
2217 		struct ipv6hdr *ipv6;
2218 	} hdr;
2219 	struct tcphdr *th;
2220 	unsigned int hlen;
2221 	u32 flex_ptype, dtype_cmd;
2222 	int l4_proto;
2223 	u16 i;
2224 
2225 	/* make sure ATR is enabled */
2226 	if (!(pf->flags & I40E_FLAG_FD_ATR_ENABLED))
2227 		return;
2228 
2229 	if ((pf->hw_disabled_flags & I40E_FLAG_FD_ATR_ENABLED))
2230 		return;
2231 
2232 	/* if sampling is disabled do nothing */
2233 	if (!tx_ring->atr_sample_rate)
2234 		return;
2235 
2236 	/* Currently only IPv4/IPv6 with TCP is supported */
2237 	if (!(tx_flags & (I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6)))
2238 		return;
2239 
2240 	/* snag network header to get L4 type and address */
2241 	hdr.network = (tx_flags & I40E_TX_FLAGS_UDP_TUNNEL) ?
2242 		      skb_inner_network_header(skb) : skb_network_header(skb);
2243 
2244 	/* Note: tx_flags gets modified to reflect inner protocols in
2245 	 * tx_enable_csum function if encap is enabled.
2246 	 */
2247 	if (tx_flags & I40E_TX_FLAGS_IPV4) {
2248 		/* access ihl as u8 to avoid unaligned access on ia64 */
2249 		hlen = (hdr.network[0] & 0x0F) << 2;
2250 		l4_proto = hdr.ipv4->protocol;
2251 	} else {
2252 		hlen = hdr.network - skb->data;
2253 		l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
2254 		hlen -= hdr.network - skb->data;
2255 	}
2256 
2257 	if (l4_proto != IPPROTO_TCP)
2258 		return;
2259 
2260 	th = (struct tcphdr *)(hdr.network + hlen);
2261 
2262 	/* Due to lack of space, no more new filters can be programmed */
2263 	if (th->syn && (pf->hw_disabled_flags & I40E_FLAG_FD_ATR_ENABLED))
2264 		return;
2265 	if ((pf->flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE) &&
2266 	    (!(pf->hw_disabled_flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE))) {
2267 		/* HW ATR eviction will take care of removing filters on FIN
2268 		 * and RST packets.
2269 		 */
2270 		if (th->fin || th->rst)
2271 			return;
2272 	}
2273 
2274 	tx_ring->atr_count++;
2275 
2276 	/* sample on all syn/fin/rst packets or once every atr sample rate */
2277 	if (!th->fin &&
2278 	    !th->syn &&
2279 	    !th->rst &&
2280 	    (tx_ring->atr_count < tx_ring->atr_sample_rate))
2281 		return;
2282 
2283 	tx_ring->atr_count = 0;
2284 
2285 	/* grab the next descriptor */
2286 	i = tx_ring->next_to_use;
2287 	fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
2288 
2289 	i++;
2290 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2291 
2292 	flex_ptype = (tx_ring->queue_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) &
2293 		      I40E_TXD_FLTR_QW0_QINDEX_MASK;
2294 	flex_ptype |= (tx_flags & I40E_TX_FLAGS_IPV4) ?
2295 		      (I40E_FILTER_PCTYPE_NONF_IPV4_TCP <<
2296 		       I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) :
2297 		      (I40E_FILTER_PCTYPE_NONF_IPV6_TCP <<
2298 		       I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
2299 
2300 	flex_ptype |= tx_ring->vsi->id << I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT;
2301 
2302 	dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
2303 
2304 	dtype_cmd |= (th->fin || th->rst) ?
2305 		     (I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
2306 		      I40E_TXD_FLTR_QW1_PCMD_SHIFT) :
2307 		     (I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
2308 		      I40E_TXD_FLTR_QW1_PCMD_SHIFT);
2309 
2310 	dtype_cmd |= I40E_FILTER_PROGRAM_DESC_DEST_DIRECT_PACKET_QINDEX <<
2311 		     I40E_TXD_FLTR_QW1_DEST_SHIFT;
2312 
2313 	dtype_cmd |= I40E_FILTER_PROGRAM_DESC_FD_STATUS_FD_ID <<
2314 		     I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT;
2315 
2316 	dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
2317 	if (!(tx_flags & I40E_TX_FLAGS_UDP_TUNNEL))
2318 		dtype_cmd |=
2319 			((u32)I40E_FD_ATR_STAT_IDX(pf->hw.pf_id) <<
2320 			I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2321 			I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2322 	else
2323 		dtype_cmd |=
2324 			((u32)I40E_FD_ATR_TUNNEL_STAT_IDX(pf->hw.pf_id) <<
2325 			I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2326 			I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2327 
2328 	if ((pf->flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE) &&
2329 	    (!(pf->hw_disabled_flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE)))
2330 		dtype_cmd |= I40E_TXD_FLTR_QW1_ATR_MASK;
2331 
2332 	fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
2333 	fdir_desc->rsvd = cpu_to_le32(0);
2334 	fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
2335 	fdir_desc->fd_id = cpu_to_le32(0);
2336 }
2337 
2338 /**
2339  * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
2340  * @skb:     send buffer
2341  * @tx_ring: ring to send buffer on
2342  * @flags:   the tx flags to be set
2343  *
2344  * Checks the skb and set up correspondingly several generic transmit flags
2345  * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
2346  *
2347  * Returns error code indicate the frame should be dropped upon error and the
2348  * otherwise  returns 0 to indicate the flags has been set properly.
2349  **/
2350 static inline int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
2351 					     struct i40e_ring *tx_ring,
2352 					     u32 *flags)
2353 {
2354 	__be16 protocol = skb->protocol;
2355 	u32  tx_flags = 0;
2356 
2357 	if (protocol == htons(ETH_P_8021Q) &&
2358 	    !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
2359 		/* When HW VLAN acceleration is turned off by the user the
2360 		 * stack sets the protocol to 8021q so that the driver
2361 		 * can take any steps required to support the SW only
2362 		 * VLAN handling.  In our case the driver doesn't need
2363 		 * to take any further steps so just set the protocol
2364 		 * to the encapsulated ethertype.
2365 		 */
2366 		skb->protocol = vlan_get_protocol(skb);
2367 		goto out;
2368 	}
2369 
2370 	/* if we have a HW VLAN tag being added, default to the HW one */
2371 	if (skb_vlan_tag_present(skb)) {
2372 		tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
2373 		tx_flags |= I40E_TX_FLAGS_HW_VLAN;
2374 	/* else if it is a SW VLAN, check the next protocol and store the tag */
2375 	} else if (protocol == htons(ETH_P_8021Q)) {
2376 		struct vlan_hdr *vhdr, _vhdr;
2377 
2378 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
2379 		if (!vhdr)
2380 			return -EINVAL;
2381 
2382 		protocol = vhdr->h_vlan_encapsulated_proto;
2383 		tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
2384 		tx_flags |= I40E_TX_FLAGS_SW_VLAN;
2385 	}
2386 
2387 	if (!(tx_ring->vsi->back->flags & I40E_FLAG_DCB_ENABLED))
2388 		goto out;
2389 
2390 	/* Insert 802.1p priority into VLAN header */
2391 	if ((tx_flags & (I40E_TX_FLAGS_HW_VLAN | I40E_TX_FLAGS_SW_VLAN)) ||
2392 	    (skb->priority != TC_PRIO_CONTROL)) {
2393 		tx_flags &= ~I40E_TX_FLAGS_VLAN_PRIO_MASK;
2394 		tx_flags |= (skb->priority & 0x7) <<
2395 				I40E_TX_FLAGS_VLAN_PRIO_SHIFT;
2396 		if (tx_flags & I40E_TX_FLAGS_SW_VLAN) {
2397 			struct vlan_ethhdr *vhdr;
2398 			int rc;
2399 
2400 			rc = skb_cow_head(skb, 0);
2401 			if (rc < 0)
2402 				return rc;
2403 			vhdr = (struct vlan_ethhdr *)skb->data;
2404 			vhdr->h_vlan_TCI = htons(tx_flags >>
2405 						 I40E_TX_FLAGS_VLAN_SHIFT);
2406 		} else {
2407 			tx_flags |= I40E_TX_FLAGS_HW_VLAN;
2408 		}
2409 	}
2410 
2411 out:
2412 	*flags = tx_flags;
2413 	return 0;
2414 }
2415 
2416 /**
2417  * i40e_tso - set up the tso context descriptor
2418  * @first:    pointer to first Tx buffer for xmit
2419  * @hdr_len:  ptr to the size of the packet header
2420  * @cd_type_cmd_tso_mss: Quad Word 1
2421  *
2422  * Returns 0 if no TSO can happen, 1 if tso is going, or error
2423  **/
2424 static int i40e_tso(struct i40e_tx_buffer *first, u8 *hdr_len,
2425 		    u64 *cd_type_cmd_tso_mss)
2426 {
2427 	struct sk_buff *skb = first->skb;
2428 	u64 cd_cmd, cd_tso_len, cd_mss;
2429 	union {
2430 		struct iphdr *v4;
2431 		struct ipv6hdr *v6;
2432 		unsigned char *hdr;
2433 	} ip;
2434 	union {
2435 		struct tcphdr *tcp;
2436 		struct udphdr *udp;
2437 		unsigned char *hdr;
2438 	} l4;
2439 	u32 paylen, l4_offset;
2440 	u16 gso_segs, gso_size;
2441 	int err;
2442 
2443 	if (skb->ip_summed != CHECKSUM_PARTIAL)
2444 		return 0;
2445 
2446 	if (!skb_is_gso(skb))
2447 		return 0;
2448 
2449 	err = skb_cow_head(skb, 0);
2450 	if (err < 0)
2451 		return err;
2452 
2453 	ip.hdr = skb_network_header(skb);
2454 	l4.hdr = skb_transport_header(skb);
2455 
2456 	/* initialize outer IP header fields */
2457 	if (ip.v4->version == 4) {
2458 		ip.v4->tot_len = 0;
2459 		ip.v4->check = 0;
2460 	} else {
2461 		ip.v6->payload_len = 0;
2462 	}
2463 
2464 	if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
2465 					 SKB_GSO_GRE_CSUM |
2466 					 SKB_GSO_IPXIP4 |
2467 					 SKB_GSO_IPXIP6 |
2468 					 SKB_GSO_UDP_TUNNEL |
2469 					 SKB_GSO_UDP_TUNNEL_CSUM)) {
2470 		if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
2471 		    (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)) {
2472 			l4.udp->len = 0;
2473 
2474 			/* determine offset of outer transport header */
2475 			l4_offset = l4.hdr - skb->data;
2476 
2477 			/* remove payload length from outer checksum */
2478 			paylen = skb->len - l4_offset;
2479 			csum_replace_by_diff(&l4.udp->check,
2480 					     (__force __wsum)htonl(paylen));
2481 		}
2482 
2483 		/* reset pointers to inner headers */
2484 		ip.hdr = skb_inner_network_header(skb);
2485 		l4.hdr = skb_inner_transport_header(skb);
2486 
2487 		/* initialize inner IP header fields */
2488 		if (ip.v4->version == 4) {
2489 			ip.v4->tot_len = 0;
2490 			ip.v4->check = 0;
2491 		} else {
2492 			ip.v6->payload_len = 0;
2493 		}
2494 	}
2495 
2496 	/* determine offset of inner transport header */
2497 	l4_offset = l4.hdr - skb->data;
2498 
2499 	/* remove payload length from inner checksum */
2500 	paylen = skb->len - l4_offset;
2501 	csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen));
2502 
2503 	/* compute length of segmentation header */
2504 	*hdr_len = (l4.tcp->doff * 4) + l4_offset;
2505 
2506 	/* pull values out of skb_shinfo */
2507 	gso_size = skb_shinfo(skb)->gso_size;
2508 	gso_segs = skb_shinfo(skb)->gso_segs;
2509 
2510 	/* update GSO size and bytecount with header size */
2511 	first->gso_segs = gso_segs;
2512 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
2513 
2514 	/* find the field values */
2515 	cd_cmd = I40E_TX_CTX_DESC_TSO;
2516 	cd_tso_len = skb->len - *hdr_len;
2517 	cd_mss = gso_size;
2518 	*cd_type_cmd_tso_mss |= (cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
2519 				(cd_tso_len << I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
2520 				(cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
2521 	return 1;
2522 }
2523 
2524 /**
2525  * i40e_tsyn - set up the tsyn context descriptor
2526  * @tx_ring:  ptr to the ring to send
2527  * @skb:      ptr to the skb we're sending
2528  * @tx_flags: the collected send information
2529  * @cd_type_cmd_tso_mss: Quad Word 1
2530  *
2531  * Returns 0 if no Tx timestamp can happen and 1 if the timestamp will happen
2532  **/
2533 static int i40e_tsyn(struct i40e_ring *tx_ring, struct sk_buff *skb,
2534 		     u32 tx_flags, u64 *cd_type_cmd_tso_mss)
2535 {
2536 	struct i40e_pf *pf;
2537 
2538 	if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)))
2539 		return 0;
2540 
2541 	/* Tx timestamps cannot be sampled when doing TSO */
2542 	if (tx_flags & I40E_TX_FLAGS_TSO)
2543 		return 0;
2544 
2545 	/* only timestamp the outbound packet if the user has requested it and
2546 	 * we are not already transmitting a packet to be timestamped
2547 	 */
2548 	pf = i40e_netdev_to_pf(tx_ring->netdev);
2549 	if (!(pf->flags & I40E_FLAG_PTP))
2550 		return 0;
2551 
2552 	if (pf->ptp_tx &&
2553 	    !test_and_set_bit_lock(__I40E_PTP_TX_IN_PROGRESS, &pf->state)) {
2554 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
2555 		pf->ptp_tx_skb = skb_get(skb);
2556 	} else {
2557 		return 0;
2558 	}
2559 
2560 	*cd_type_cmd_tso_mss |= (u64)I40E_TX_CTX_DESC_TSYN <<
2561 				I40E_TXD_CTX_QW1_CMD_SHIFT;
2562 
2563 	return 1;
2564 }
2565 
2566 /**
2567  * i40e_tx_enable_csum - Enable Tx checksum offloads
2568  * @skb: send buffer
2569  * @tx_flags: pointer to Tx flags currently set
2570  * @td_cmd: Tx descriptor command bits to set
2571  * @td_offset: Tx descriptor header offsets to set
2572  * @tx_ring: Tx descriptor ring
2573  * @cd_tunneling: ptr to context desc bits
2574  **/
2575 static int i40e_tx_enable_csum(struct sk_buff *skb, u32 *tx_flags,
2576 			       u32 *td_cmd, u32 *td_offset,
2577 			       struct i40e_ring *tx_ring,
2578 			       u32 *cd_tunneling)
2579 {
2580 	union {
2581 		struct iphdr *v4;
2582 		struct ipv6hdr *v6;
2583 		unsigned char *hdr;
2584 	} ip;
2585 	union {
2586 		struct tcphdr *tcp;
2587 		struct udphdr *udp;
2588 		unsigned char *hdr;
2589 	} l4;
2590 	unsigned char *exthdr;
2591 	u32 offset, cmd = 0;
2592 	__be16 frag_off;
2593 	u8 l4_proto = 0;
2594 
2595 	if (skb->ip_summed != CHECKSUM_PARTIAL)
2596 		return 0;
2597 
2598 	ip.hdr = skb_network_header(skb);
2599 	l4.hdr = skb_transport_header(skb);
2600 
2601 	/* compute outer L2 header size */
2602 	offset = ((ip.hdr - skb->data) / 2) << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
2603 
2604 	if (skb->encapsulation) {
2605 		u32 tunnel = 0;
2606 		/* define outer network header type */
2607 		if (*tx_flags & I40E_TX_FLAGS_IPV4) {
2608 			tunnel |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
2609 				  I40E_TX_CTX_EXT_IP_IPV4 :
2610 				  I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
2611 
2612 			l4_proto = ip.v4->protocol;
2613 		} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
2614 			tunnel |= I40E_TX_CTX_EXT_IP_IPV6;
2615 
2616 			exthdr = ip.hdr + sizeof(*ip.v6);
2617 			l4_proto = ip.v6->nexthdr;
2618 			if (l4.hdr != exthdr)
2619 				ipv6_skip_exthdr(skb, exthdr - skb->data,
2620 						 &l4_proto, &frag_off);
2621 		}
2622 
2623 		/* define outer transport */
2624 		switch (l4_proto) {
2625 		case IPPROTO_UDP:
2626 			tunnel |= I40E_TXD_CTX_UDP_TUNNELING;
2627 			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
2628 			break;
2629 		case IPPROTO_GRE:
2630 			tunnel |= I40E_TXD_CTX_GRE_TUNNELING;
2631 			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
2632 			break;
2633 		case IPPROTO_IPIP:
2634 		case IPPROTO_IPV6:
2635 			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
2636 			l4.hdr = skb_inner_network_header(skb);
2637 			break;
2638 		default:
2639 			if (*tx_flags & I40E_TX_FLAGS_TSO)
2640 				return -1;
2641 
2642 			skb_checksum_help(skb);
2643 			return 0;
2644 		}
2645 
2646 		/* compute outer L3 header size */
2647 		tunnel |= ((l4.hdr - ip.hdr) / 4) <<
2648 			  I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT;
2649 
2650 		/* switch IP header pointer from outer to inner header */
2651 		ip.hdr = skb_inner_network_header(skb);
2652 
2653 		/* compute tunnel header size */
2654 		tunnel |= ((ip.hdr - l4.hdr) / 2) <<
2655 			  I40E_TXD_CTX_QW0_NATLEN_SHIFT;
2656 
2657 		/* indicate if we need to offload outer UDP header */
2658 		if ((*tx_flags & I40E_TX_FLAGS_TSO) &&
2659 		    !(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
2660 		    (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM))
2661 			tunnel |= I40E_TXD_CTX_QW0_L4T_CS_MASK;
2662 
2663 		/* record tunnel offload values */
2664 		*cd_tunneling |= tunnel;
2665 
2666 		/* switch L4 header pointer from outer to inner */
2667 		l4.hdr = skb_inner_transport_header(skb);
2668 		l4_proto = 0;
2669 
2670 		/* reset type as we transition from outer to inner headers */
2671 		*tx_flags &= ~(I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6);
2672 		if (ip.v4->version == 4)
2673 			*tx_flags |= I40E_TX_FLAGS_IPV4;
2674 		if (ip.v6->version == 6)
2675 			*tx_flags |= I40E_TX_FLAGS_IPV6;
2676 	}
2677 
2678 	/* Enable IP checksum offloads */
2679 	if (*tx_flags & I40E_TX_FLAGS_IPV4) {
2680 		l4_proto = ip.v4->protocol;
2681 		/* the stack computes the IP header already, the only time we
2682 		 * need the hardware to recompute it is in the case of TSO.
2683 		 */
2684 		cmd |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
2685 		       I40E_TX_DESC_CMD_IIPT_IPV4_CSUM :
2686 		       I40E_TX_DESC_CMD_IIPT_IPV4;
2687 	} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
2688 		cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
2689 
2690 		exthdr = ip.hdr + sizeof(*ip.v6);
2691 		l4_proto = ip.v6->nexthdr;
2692 		if (l4.hdr != exthdr)
2693 			ipv6_skip_exthdr(skb, exthdr - skb->data,
2694 					 &l4_proto, &frag_off);
2695 	}
2696 
2697 	/* compute inner L3 header size */
2698 	offset |= ((l4.hdr - ip.hdr) / 4) << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
2699 
2700 	/* Enable L4 checksum offloads */
2701 	switch (l4_proto) {
2702 	case IPPROTO_TCP:
2703 		/* enable checksum offloads */
2704 		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
2705 		offset |= l4.tcp->doff << I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
2706 		break;
2707 	case IPPROTO_SCTP:
2708 		/* enable SCTP checksum offload */
2709 		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
2710 		offset |= (sizeof(struct sctphdr) >> 2) <<
2711 			  I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
2712 		break;
2713 	case IPPROTO_UDP:
2714 		/* enable UDP checksum offload */
2715 		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
2716 		offset |= (sizeof(struct udphdr) >> 2) <<
2717 			  I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
2718 		break;
2719 	default:
2720 		if (*tx_flags & I40E_TX_FLAGS_TSO)
2721 			return -1;
2722 		skb_checksum_help(skb);
2723 		return 0;
2724 	}
2725 
2726 	*td_cmd |= cmd;
2727 	*td_offset |= offset;
2728 
2729 	return 1;
2730 }
2731 
2732 /**
2733  * i40e_create_tx_ctx Build the Tx context descriptor
2734  * @tx_ring:  ring to create the descriptor on
2735  * @cd_type_cmd_tso_mss: Quad Word 1
2736  * @cd_tunneling: Quad Word 0 - bits 0-31
2737  * @cd_l2tag2: Quad Word 0 - bits 32-63
2738  **/
2739 static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
2740 			       const u64 cd_type_cmd_tso_mss,
2741 			       const u32 cd_tunneling, const u32 cd_l2tag2)
2742 {
2743 	struct i40e_tx_context_desc *context_desc;
2744 	int i = tx_ring->next_to_use;
2745 
2746 	if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
2747 	    !cd_tunneling && !cd_l2tag2)
2748 		return;
2749 
2750 	/* grab the next descriptor */
2751 	context_desc = I40E_TX_CTXTDESC(tx_ring, i);
2752 
2753 	i++;
2754 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2755 
2756 	/* cpu_to_le32 and assign to struct fields */
2757 	context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
2758 	context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
2759 	context_desc->rsvd = cpu_to_le16(0);
2760 	context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
2761 }
2762 
2763 /**
2764  * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions
2765  * @tx_ring: the ring to be checked
2766  * @size:    the size buffer we want to assure is available
2767  *
2768  * Returns -EBUSY if a stop is needed, else 0
2769  **/
2770 int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
2771 {
2772 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
2773 	/* Memory barrier before checking head and tail */
2774 	smp_mb();
2775 
2776 	/* Check again in a case another CPU has just made room available. */
2777 	if (likely(I40E_DESC_UNUSED(tx_ring) < size))
2778 		return -EBUSY;
2779 
2780 	/* A reprieve! - use start_queue because it doesn't call schedule */
2781 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
2782 	++tx_ring->tx_stats.restart_queue;
2783 	return 0;
2784 }
2785 
2786 /**
2787  * __i40e_chk_linearize - Check if there are more than 8 buffers per packet
2788  * @skb:      send buffer
2789  *
2790  * Note: Our HW can't DMA more than 8 buffers to build a packet on the wire
2791  * and so we need to figure out the cases where we need to linearize the skb.
2792  *
2793  * For TSO we need to count the TSO header and segment payload separately.
2794  * As such we need to check cases where we have 7 fragments or more as we
2795  * can potentially require 9 DMA transactions, 1 for the TSO header, 1 for
2796  * the segment payload in the first descriptor, and another 7 for the
2797  * fragments.
2798  **/
2799 bool __i40e_chk_linearize(struct sk_buff *skb)
2800 {
2801 	const struct skb_frag_struct *frag, *stale;
2802 	int nr_frags, sum;
2803 
2804 	/* no need to check if number of frags is less than 7 */
2805 	nr_frags = skb_shinfo(skb)->nr_frags;
2806 	if (nr_frags < (I40E_MAX_BUFFER_TXD - 1))
2807 		return false;
2808 
2809 	/* We need to walk through the list and validate that each group
2810 	 * of 6 fragments totals at least gso_size.
2811 	 */
2812 	nr_frags -= I40E_MAX_BUFFER_TXD - 2;
2813 	frag = &skb_shinfo(skb)->frags[0];
2814 
2815 	/* Initialize size to the negative value of gso_size minus 1.  We
2816 	 * use this as the worst case scenerio in which the frag ahead
2817 	 * of us only provides one byte which is why we are limited to 6
2818 	 * descriptors for a single transmit as the header and previous
2819 	 * fragment are already consuming 2 descriptors.
2820 	 */
2821 	sum = 1 - skb_shinfo(skb)->gso_size;
2822 
2823 	/* Add size of frags 0 through 4 to create our initial sum */
2824 	sum += skb_frag_size(frag++);
2825 	sum += skb_frag_size(frag++);
2826 	sum += skb_frag_size(frag++);
2827 	sum += skb_frag_size(frag++);
2828 	sum += skb_frag_size(frag++);
2829 
2830 	/* Walk through fragments adding latest fragment, testing it, and
2831 	 * then removing stale fragments from the sum.
2832 	 */
2833 	stale = &skb_shinfo(skb)->frags[0];
2834 	for (;;) {
2835 		sum += skb_frag_size(frag++);
2836 
2837 		/* if sum is negative we failed to make sufficient progress */
2838 		if (sum < 0)
2839 			return true;
2840 
2841 		if (!nr_frags--)
2842 			break;
2843 
2844 		sum -= skb_frag_size(stale++);
2845 	}
2846 
2847 	return false;
2848 }
2849 
2850 /**
2851  * i40e_tx_map - Build the Tx descriptor
2852  * @tx_ring:  ring to send buffer on
2853  * @skb:      send buffer
2854  * @first:    first buffer info buffer to use
2855  * @tx_flags: collected send information
2856  * @hdr_len:  size of the packet header
2857  * @td_cmd:   the command field in the descriptor
2858  * @td_offset: offset for checksum or crc
2859  **/
2860 static inline void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
2861 			       struct i40e_tx_buffer *first, u32 tx_flags,
2862 			       const u8 hdr_len, u32 td_cmd, u32 td_offset)
2863 {
2864 	unsigned int data_len = skb->data_len;
2865 	unsigned int size = skb_headlen(skb);
2866 	struct skb_frag_struct *frag;
2867 	struct i40e_tx_buffer *tx_bi;
2868 	struct i40e_tx_desc *tx_desc;
2869 	u16 i = tx_ring->next_to_use;
2870 	u32 td_tag = 0;
2871 	dma_addr_t dma;
2872 	u16 desc_count = 1;
2873 
2874 	if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
2875 		td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
2876 		td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
2877 			 I40E_TX_FLAGS_VLAN_SHIFT;
2878 	}
2879 
2880 	first->tx_flags = tx_flags;
2881 
2882 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
2883 
2884 	tx_desc = I40E_TX_DESC(tx_ring, i);
2885 	tx_bi = first;
2886 
2887 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
2888 		unsigned int max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
2889 
2890 		if (dma_mapping_error(tx_ring->dev, dma))
2891 			goto dma_error;
2892 
2893 		/* record length, and DMA address */
2894 		dma_unmap_len_set(tx_bi, len, size);
2895 		dma_unmap_addr_set(tx_bi, dma, dma);
2896 
2897 		/* align size to end of page */
2898 		max_data += -dma & (I40E_MAX_READ_REQ_SIZE - 1);
2899 		tx_desc->buffer_addr = cpu_to_le64(dma);
2900 
2901 		while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
2902 			tx_desc->cmd_type_offset_bsz =
2903 				build_ctob(td_cmd, td_offset,
2904 					   max_data, td_tag);
2905 
2906 			tx_desc++;
2907 			i++;
2908 			desc_count++;
2909 
2910 			if (i == tx_ring->count) {
2911 				tx_desc = I40E_TX_DESC(tx_ring, 0);
2912 				i = 0;
2913 			}
2914 
2915 			dma += max_data;
2916 			size -= max_data;
2917 
2918 			max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
2919 			tx_desc->buffer_addr = cpu_to_le64(dma);
2920 		}
2921 
2922 		if (likely(!data_len))
2923 			break;
2924 
2925 		tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
2926 							  size, td_tag);
2927 
2928 		tx_desc++;
2929 		i++;
2930 		desc_count++;
2931 
2932 		if (i == tx_ring->count) {
2933 			tx_desc = I40E_TX_DESC(tx_ring, 0);
2934 			i = 0;
2935 		}
2936 
2937 		size = skb_frag_size(frag);
2938 		data_len -= size;
2939 
2940 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
2941 				       DMA_TO_DEVICE);
2942 
2943 		tx_bi = &tx_ring->tx_bi[i];
2944 	}
2945 
2946 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
2947 
2948 	i++;
2949 	if (i == tx_ring->count)
2950 		i = 0;
2951 
2952 	tx_ring->next_to_use = i;
2953 
2954 	i40e_maybe_stop_tx(tx_ring, DESC_NEEDED);
2955 
2956 	/* write last descriptor with EOP bit */
2957 	td_cmd |= I40E_TX_DESC_CMD_EOP;
2958 
2959 	/* We can OR these values together as they both are checked against
2960 	 * 4 below and at this point desc_count will be used as a boolean value
2961 	 * after this if/else block.
2962 	 */
2963 	desc_count |= ++tx_ring->packet_stride;
2964 
2965 	/* Algorithm to optimize tail and RS bit setting:
2966 	 * if queue is stopped
2967 	 *	mark RS bit
2968 	 *	reset packet counter
2969 	 * else if xmit_more is supported and is true
2970 	 *	advance packet counter to 4
2971 	 *	reset desc_count to 0
2972 	 *
2973 	 * if desc_count >= 4
2974 	 *	mark RS bit
2975 	 *	reset packet counter
2976 	 * if desc_count > 0
2977 	 *	update tail
2978 	 *
2979 	 * Note: If there are less than 4 descriptors
2980 	 * pending and interrupts were disabled the service task will
2981 	 * trigger a force WB.
2982 	 */
2983 	if (netif_xmit_stopped(txring_txq(tx_ring))) {
2984 		goto do_rs;
2985 	} else if (skb->xmit_more) {
2986 		/* set stride to arm on next packet and reset desc_count */
2987 		tx_ring->packet_stride = WB_STRIDE;
2988 		desc_count = 0;
2989 	} else if (desc_count >= WB_STRIDE) {
2990 do_rs:
2991 		/* write last descriptor with RS bit set */
2992 		td_cmd |= I40E_TX_DESC_CMD_RS;
2993 		tx_ring->packet_stride = 0;
2994 	}
2995 
2996 	tx_desc->cmd_type_offset_bsz =
2997 			build_ctob(td_cmd, td_offset, size, td_tag);
2998 
2999 	/* Force memory writes to complete before letting h/w know there
3000 	 * are new descriptors to fetch.
3001 	 *
3002 	 * We also use this memory barrier to make certain all of the
3003 	 * status bits have been updated before next_to_watch is written.
3004 	 */
3005 	wmb();
3006 
3007 	/* set next_to_watch value indicating a packet is present */
3008 	first->next_to_watch = tx_desc;
3009 
3010 	/* notify HW of packet */
3011 	if (desc_count) {
3012 		writel(i, tx_ring->tail);
3013 
3014 		/* we need this if more than one processor can write to our tail
3015 		 * at a time, it synchronizes IO on IA64/Altix systems
3016 		 */
3017 		mmiowb();
3018 	}
3019 
3020 	return;
3021 
3022 dma_error:
3023 	dev_info(tx_ring->dev, "TX DMA map failed\n");
3024 
3025 	/* clear dma mappings for failed tx_bi map */
3026 	for (;;) {
3027 		tx_bi = &tx_ring->tx_bi[i];
3028 		i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
3029 		if (tx_bi == first)
3030 			break;
3031 		if (i == 0)
3032 			i = tx_ring->count;
3033 		i--;
3034 	}
3035 
3036 	tx_ring->next_to_use = i;
3037 }
3038 
3039 /**
3040  * i40e_xmit_frame_ring - Sends buffer on Tx ring
3041  * @skb:     send buffer
3042  * @tx_ring: ring to send buffer on
3043  *
3044  * Returns NETDEV_TX_OK if sent, else an error code
3045  **/
3046 static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
3047 					struct i40e_ring *tx_ring)
3048 {
3049 	u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
3050 	u32 cd_tunneling = 0, cd_l2tag2 = 0;
3051 	struct i40e_tx_buffer *first;
3052 	u32 td_offset = 0;
3053 	u32 tx_flags = 0;
3054 	__be16 protocol;
3055 	u32 td_cmd = 0;
3056 	u8 hdr_len = 0;
3057 	int tso, count;
3058 	int tsyn;
3059 
3060 	/* prefetch the data, we'll need it later */
3061 	prefetch(skb->data);
3062 
3063 	count = i40e_xmit_descriptor_count(skb);
3064 	if (i40e_chk_linearize(skb, count)) {
3065 		if (__skb_linearize(skb)) {
3066 			dev_kfree_skb_any(skb);
3067 			return NETDEV_TX_OK;
3068 		}
3069 		count = i40e_txd_use_count(skb->len);
3070 		tx_ring->tx_stats.tx_linearize++;
3071 	}
3072 
3073 	/* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
3074 	 *       + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
3075 	 *       + 4 desc gap to avoid the cache line where head is,
3076 	 *       + 1 desc for context descriptor,
3077 	 * otherwise try next time
3078 	 */
3079 	if (i40e_maybe_stop_tx(tx_ring, count + 4 + 1)) {
3080 		tx_ring->tx_stats.tx_busy++;
3081 		return NETDEV_TX_BUSY;
3082 	}
3083 
3084 	/* record the location of the first descriptor for this packet */
3085 	first = &tx_ring->tx_bi[tx_ring->next_to_use];
3086 	first->skb = skb;
3087 	first->bytecount = skb->len;
3088 	first->gso_segs = 1;
3089 
3090 	/* prepare the xmit flags */
3091 	if (i40e_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
3092 		goto out_drop;
3093 
3094 	/* obtain protocol of skb */
3095 	protocol = vlan_get_protocol(skb);
3096 
3097 	/* setup IPv4/IPv6 offloads */
3098 	if (protocol == htons(ETH_P_IP))
3099 		tx_flags |= I40E_TX_FLAGS_IPV4;
3100 	else if (protocol == htons(ETH_P_IPV6))
3101 		tx_flags |= I40E_TX_FLAGS_IPV6;
3102 
3103 	tso = i40e_tso(first, &hdr_len, &cd_type_cmd_tso_mss);
3104 
3105 	if (tso < 0)
3106 		goto out_drop;
3107 	else if (tso)
3108 		tx_flags |= I40E_TX_FLAGS_TSO;
3109 
3110 	/* Always offload the checksum, since it's in the data descriptor */
3111 	tso = i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset,
3112 				  tx_ring, &cd_tunneling);
3113 	if (tso < 0)
3114 		goto out_drop;
3115 
3116 	tsyn = i40e_tsyn(tx_ring, skb, tx_flags, &cd_type_cmd_tso_mss);
3117 
3118 	if (tsyn)
3119 		tx_flags |= I40E_TX_FLAGS_TSYN;
3120 
3121 	skb_tx_timestamp(skb);
3122 
3123 	/* always enable CRC insertion offload */
3124 	td_cmd |= I40E_TX_DESC_CMD_ICRC;
3125 
3126 	i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
3127 			   cd_tunneling, cd_l2tag2);
3128 
3129 	/* Add Flow Director ATR if it's enabled.
3130 	 *
3131 	 * NOTE: this must always be directly before the data descriptor.
3132 	 */
3133 	i40e_atr(tx_ring, skb, tx_flags);
3134 
3135 	i40e_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
3136 		    td_cmd, td_offset);
3137 
3138 	return NETDEV_TX_OK;
3139 
3140 out_drop:
3141 	dev_kfree_skb_any(first->skb);
3142 	first->skb = NULL;
3143 	return NETDEV_TX_OK;
3144 }
3145 
3146 /**
3147  * i40e_lan_xmit_frame - Selects the correct VSI and Tx queue to send buffer
3148  * @skb:    send buffer
3149  * @netdev: network interface device structure
3150  *
3151  * Returns NETDEV_TX_OK if sent, else an error code
3152  **/
3153 netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3154 {
3155 	struct i40e_netdev_priv *np = netdev_priv(netdev);
3156 	struct i40e_vsi *vsi = np->vsi;
3157 	struct i40e_ring *tx_ring = vsi->tx_rings[skb->queue_mapping];
3158 
3159 	/* hardware can't handle really short frames, hardware padding works
3160 	 * beyond this point
3161 	 */
3162 	if (skb_put_padto(skb, I40E_MIN_TX_LEN))
3163 		return NETDEV_TX_OK;
3164 
3165 	return i40e_xmit_frame_ring(skb, tx_ring);
3166 }
3167