Commit | Line | Data |
---|---|---|
fd0a05ce JB |
1 | /******************************************************************************* |
2 | * | |
3 | * Intel Ethernet Controller XL710 Family Linux Driver | |
dc641b73 | 4 | * Copyright(c) 2013 - 2014 Intel Corporation. |
fd0a05ce JB |
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 | * | |
dc641b73 GR |
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/>. | |
fd0a05ce JB |
17 | * |
18 | * The full GNU General Public License is included in this distribution in | |
19 | * the file called "COPYING". | |
20 | * | |
21 | * Contact Information: | |
22 | * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net> | |
23 | * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497 | |
24 | * | |
25 | ******************************************************************************/ | |
26 | ||
27 | #include "i40e.h" | |
206812b5 | 28 | #include "i40e_prototype.h" |
fd0a05ce JB |
29 | |
30 | static inline __le64 build_ctob(u32 td_cmd, u32 td_offset, unsigned int size, | |
31 | u32 td_tag) | |
32 | { | |
33 | return cpu_to_le64(I40E_TX_DESC_DTYPE_DATA | | |
34 | ((u64)td_cmd << I40E_TXD_QW1_CMD_SHIFT) | | |
35 | ((u64)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) | | |
36 | ((u64)size << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) | | |
37 | ((u64)td_tag << I40E_TXD_QW1_L2TAG1_SHIFT)); | |
38 | } | |
39 | ||
eaefbd06 | 40 | #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS) |
fd0a05ce JB |
41 | /** |
42 | * i40e_program_fdir_filter - Program a Flow Director filter | |
17a73f6b JG |
43 | * @fdir_data: Packet data that will be filter parameters |
44 | * @raw_packet: the pre-allocated packet buffer for FDir | |
fd0a05ce JB |
45 | * @pf: The pf pointer |
46 | * @add: True for add/update, False for remove | |
47 | **/ | |
17a73f6b | 48 | int i40e_program_fdir_filter(struct i40e_fdir_filter *fdir_data, u8 *raw_packet, |
fd0a05ce JB |
49 | struct i40e_pf *pf, bool add) |
50 | { | |
51 | struct i40e_filter_program_desc *fdir_desc; | |
52 | struct i40e_tx_buffer *tx_buf; | |
53 | struct i40e_tx_desc *tx_desc; | |
54 | struct i40e_ring *tx_ring; | |
eaefbd06 | 55 | unsigned int fpt, dcc; |
fd0a05ce JB |
56 | struct i40e_vsi *vsi; |
57 | struct device *dev; | |
58 | dma_addr_t dma; | |
59 | u32 td_cmd = 0; | |
60 | u16 i; | |
61 | ||
62 | /* find existing FDIR VSI */ | |
63 | vsi = NULL; | |
64 | for (i = 0; i < pf->hw.func_caps.num_vsis; i++) | |
65 | if (pf->vsi[i] && pf->vsi[i]->type == I40E_VSI_FDIR) | |
66 | vsi = pf->vsi[i]; | |
67 | if (!vsi) | |
68 | return -ENOENT; | |
69 | ||
9f65e15b | 70 | tx_ring = vsi->tx_rings[0]; |
fd0a05ce JB |
71 | dev = tx_ring->dev; |
72 | ||
17a73f6b JG |
73 | dma = dma_map_single(dev, raw_packet, |
74 | I40E_FDIR_MAX_RAW_PACKET_SIZE, DMA_TO_DEVICE); | |
fd0a05ce JB |
75 | if (dma_mapping_error(dev, dma)) |
76 | goto dma_fail; | |
77 | ||
78 | /* grab the next descriptor */ | |
fc4ac67b AD |
79 | i = tx_ring->next_to_use; |
80 | fdir_desc = I40E_TX_FDIRDESC(tx_ring, i); | |
fc4ac67b | 81 | |
eaefbd06 | 82 | tx_ring->next_to_use = (i + 1 < tx_ring->count) ? i + 1 : 0; |
fd0a05ce | 83 | |
eaefbd06 JB |
84 | fpt = (fdir_data->q_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) & |
85 | I40E_TXD_FLTR_QW0_QINDEX_MASK; | |
fd0a05ce | 86 | |
eaefbd06 JB |
87 | fpt |= (fdir_data->flex_off << I40E_TXD_FLTR_QW0_FLEXOFF_SHIFT) & |
88 | I40E_TXD_FLTR_QW0_FLEXOFF_MASK; | |
fd0a05ce | 89 | |
eaefbd06 JB |
90 | fpt |= (fdir_data->pctype << I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) & |
91 | I40E_TXD_FLTR_QW0_PCTYPE_MASK; | |
fd0a05ce JB |
92 | |
93 | /* Use LAN VSI Id if not programmed by user */ | |
94 | if (fdir_data->dest_vsi == 0) | |
eaefbd06 JB |
95 | fpt |= (pf->vsi[pf->lan_vsi]->id) << |
96 | I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT; | |
fd0a05ce | 97 | else |
eaefbd06 JB |
98 | fpt |= ((u32)fdir_data->dest_vsi << |
99 | I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT) & | |
100 | I40E_TXD_FLTR_QW0_DEST_VSI_MASK; | |
101 | ||
102 | fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(fpt); | |
fd0a05ce | 103 | |
eaefbd06 | 104 | dcc = I40E_TX_DESC_DTYPE_FILTER_PROG; |
fd0a05ce JB |
105 | |
106 | if (add) | |
eaefbd06 JB |
107 | dcc |= I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE << |
108 | I40E_TXD_FLTR_QW1_PCMD_SHIFT; | |
fd0a05ce | 109 | else |
eaefbd06 JB |
110 | dcc |= I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE << |
111 | I40E_TXD_FLTR_QW1_PCMD_SHIFT; | |
fd0a05ce | 112 | |
eaefbd06 JB |
113 | dcc |= (fdir_data->dest_ctl << I40E_TXD_FLTR_QW1_DEST_SHIFT) & |
114 | I40E_TXD_FLTR_QW1_DEST_MASK; | |
fd0a05ce | 115 | |
eaefbd06 JB |
116 | dcc |= (fdir_data->fd_status << I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT) & |
117 | I40E_TXD_FLTR_QW1_FD_STATUS_MASK; | |
fd0a05ce JB |
118 | |
119 | if (fdir_data->cnt_index != 0) { | |
eaefbd06 JB |
120 | dcc |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK; |
121 | dcc |= ((u32)fdir_data->cnt_index << | |
122 | I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) & | |
123 | I40E_TXD_FLTR_QW1_CNTINDEX_MASK; | |
fd0a05ce JB |
124 | } |
125 | ||
eaefbd06 | 126 | fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dcc); |
fd0a05ce JB |
127 | fdir_desc->fd_id = cpu_to_le32(fdir_data->fd_id); |
128 | ||
129 | /* Now program a dummy descriptor */ | |
fc4ac67b AD |
130 | i = tx_ring->next_to_use; |
131 | tx_desc = I40E_TX_DESC(tx_ring, i); | |
298deef1 | 132 | tx_buf = &tx_ring->tx_bi[i]; |
fc4ac67b | 133 | |
eaefbd06 | 134 | tx_ring->next_to_use = (i + 1 < tx_ring->count) ? i + 1 : 0; |
fd0a05ce | 135 | |
298deef1 | 136 | /* record length, and DMA address */ |
17a73f6b | 137 | dma_unmap_len_set(tx_buf, len, I40E_FDIR_MAX_RAW_PACKET_SIZE); |
298deef1 ASJ |
138 | dma_unmap_addr_set(tx_buf, dma, dma); |
139 | ||
fd0a05ce | 140 | tx_desc->buffer_addr = cpu_to_le64(dma); |
eaefbd06 | 141 | td_cmd = I40E_TXD_CMD | I40E_TX_DESC_CMD_DUMMY; |
fd0a05ce JB |
142 | |
143 | tx_desc->cmd_type_offset_bsz = | |
17a73f6b | 144 | build_ctob(td_cmd, 0, I40E_FDIR_MAX_RAW_PACKET_SIZE, 0); |
fd0a05ce | 145 | |
298deef1 ASJ |
146 | /* set the timestamp */ |
147 | tx_buf->time_stamp = jiffies; | |
148 | ||
fd0a05ce JB |
149 | /* Force memory writes to complete before letting h/w |
150 | * know there are new descriptors to fetch. (Only | |
151 | * applicable for weak-ordered memory model archs, | |
152 | * such as IA-64). | |
153 | */ | |
154 | wmb(); | |
155 | ||
fc4ac67b AD |
156 | /* Mark the data descriptor to be watched */ |
157 | tx_buf->next_to_watch = tx_desc; | |
158 | ||
fd0a05ce JB |
159 | writel(tx_ring->next_to_use, tx_ring->tail); |
160 | return 0; | |
161 | ||
162 | dma_fail: | |
163 | return -1; | |
164 | } | |
165 | ||
17a73f6b JG |
166 | #define IP_HEADER_OFFSET 14 |
167 | #define I40E_UDPIP_DUMMY_PACKET_LEN 42 | |
168 | /** | |
169 | * i40e_add_del_fdir_udpv4 - Add/Remove UDPv4 filters | |
170 | * @vsi: pointer to the targeted VSI | |
171 | * @fd_data: the flow director data required for the FDir descriptor | |
172 | * @raw_packet: the pre-allocated packet buffer for FDir | |
173 | * @add: true adds a filter, false removes it | |
174 | * | |
175 | * Returns 0 if the filters were successfully added or removed | |
176 | **/ | |
177 | static int i40e_add_del_fdir_udpv4(struct i40e_vsi *vsi, | |
178 | struct i40e_fdir_filter *fd_data, | |
179 | u8 *raw_packet, bool add) | |
180 | { | |
181 | struct i40e_pf *pf = vsi->back; | |
182 | struct udphdr *udp; | |
183 | struct iphdr *ip; | |
184 | bool err = false; | |
185 | int ret; | |
186 | int i; | |
187 | static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0, | |
188 | 0x45, 0, 0, 0x1c, 0, 0, 0x40, 0, 0x40, 0x11, 0, 0, 0, 0, 0, 0, | |
189 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}; | |
190 | ||
191 | memcpy(raw_packet, packet, I40E_UDPIP_DUMMY_PACKET_LEN); | |
192 | ||
193 | ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET); | |
194 | udp = (struct udphdr *)(raw_packet + IP_HEADER_OFFSET | |
195 | + sizeof(struct iphdr)); | |
196 | ||
197 | ip->daddr = fd_data->dst_ip[0]; | |
198 | udp->dest = fd_data->dst_port; | |
199 | ip->saddr = fd_data->src_ip[0]; | |
200 | udp->source = fd_data->src_port; | |
201 | ||
202 | for (i = I40E_FILTER_PCTYPE_NONF_UNICAST_IPV4_UDP; | |
203 | i <= I40E_FILTER_PCTYPE_NONF_IPV4_UDP; i++) { | |
204 | fd_data->pctype = i; | |
205 | ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add); | |
206 | ||
207 | if (ret) { | |
208 | dev_info(&pf->pdev->dev, | |
209 | "Filter command send failed for PCTYPE %d (ret = %d)\n", | |
210 | fd_data->pctype, ret); | |
211 | err = true; | |
212 | } else { | |
213 | dev_info(&pf->pdev->dev, | |
214 | "Filter OK for PCTYPE %d (ret = %d)\n", | |
215 | fd_data->pctype, ret); | |
216 | } | |
217 | } | |
218 | ||
219 | return err ? -EOPNOTSUPP : 0; | |
220 | } | |
221 | ||
222 | #define I40E_TCPIP_DUMMY_PACKET_LEN 54 | |
223 | /** | |
224 | * i40e_add_del_fdir_tcpv4 - Add/Remove TCPv4 filters | |
225 | * @vsi: pointer to the targeted VSI | |
226 | * @fd_data: the flow director data required for the FDir descriptor | |
227 | * @raw_packet: the pre-allocated packet buffer for FDir | |
228 | * @add: true adds a filter, false removes it | |
229 | * | |
230 | * Returns 0 if the filters were successfully added or removed | |
231 | **/ | |
232 | static int i40e_add_del_fdir_tcpv4(struct i40e_vsi *vsi, | |
233 | struct i40e_fdir_filter *fd_data, | |
234 | u8 *raw_packet, bool add) | |
235 | { | |
236 | struct i40e_pf *pf = vsi->back; | |
237 | struct tcphdr *tcp; | |
238 | struct iphdr *ip; | |
239 | bool err = false; | |
240 | int ret; | |
241 | /* Dummy packet */ | |
242 | static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0, | |
243 | 0x45, 0, 0, 0x28, 0, 0, 0x40, 0, 0x40, 0x6, 0, 0, 0, 0, 0, 0, | |
244 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x80, 0x11, | |
245 | 0x0, 0x72, 0, 0, 0, 0}; | |
246 | ||
247 | memcpy(raw_packet, packet, I40E_TCPIP_DUMMY_PACKET_LEN); | |
248 | ||
249 | ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET); | |
250 | tcp = (struct tcphdr *)(raw_packet + IP_HEADER_OFFSET | |
251 | + sizeof(struct iphdr)); | |
252 | ||
253 | ip->daddr = fd_data->dst_ip[0]; | |
254 | tcp->dest = fd_data->dst_port; | |
255 | ip->saddr = fd_data->src_ip[0]; | |
256 | tcp->source = fd_data->src_port; | |
257 | ||
258 | if (add) { | |
259 | if (pf->flags & I40E_FLAG_FD_ATR_ENABLED) { | |
260 | dev_info(&pf->pdev->dev, "Forcing ATR off, sideband rules for TCP/IPv4 flow being applied\n"); | |
261 | pf->flags &= ~I40E_FLAG_FD_ATR_ENABLED; | |
262 | } | |
263 | } | |
264 | ||
265 | fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_TCP_SYN; | |
266 | ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add); | |
267 | ||
268 | if (ret) { | |
269 | dev_info(&pf->pdev->dev, | |
270 | "Filter command send failed for PCTYPE %d (ret = %d)\n", | |
271 | fd_data->pctype, ret); | |
272 | err = true; | |
273 | } else { | |
274 | dev_info(&pf->pdev->dev, "Filter OK for PCTYPE %d (ret = %d)\n", | |
275 | fd_data->pctype, ret); | |
276 | } | |
277 | ||
278 | fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_TCP; | |
279 | ||
280 | ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add); | |
281 | if (ret) { | |
282 | dev_info(&pf->pdev->dev, | |
283 | "Filter command send failed for PCTYPE %d (ret = %d)\n", | |
284 | fd_data->pctype, ret); | |
285 | err = true; | |
286 | } else { | |
287 | dev_info(&pf->pdev->dev, "Filter OK for PCTYPE %d (ret = %d)\n", | |
288 | fd_data->pctype, ret); | |
289 | } | |
290 | ||
291 | return err ? -EOPNOTSUPP : 0; | |
292 | } | |
293 | ||
294 | /** | |
295 | * i40e_add_del_fdir_sctpv4 - Add/Remove SCTPv4 Flow Director filters for | |
296 | * a specific flow spec | |
297 | * @vsi: pointer to the targeted VSI | |
298 | * @fd_data: the flow director data required for the FDir descriptor | |
299 | * @raw_packet: the pre-allocated packet buffer for FDir | |
300 | * @add: true adds a filter, false removes it | |
301 | * | |
302 | * Returns 0 if the filters were successfully added or removed | |
303 | **/ | |
304 | static int i40e_add_del_fdir_sctpv4(struct i40e_vsi *vsi, | |
305 | struct i40e_fdir_filter *fd_data, | |
306 | u8 *raw_packet, bool add) | |
307 | { | |
308 | return -EOPNOTSUPP; | |
309 | } | |
310 | ||
311 | #define I40E_IP_DUMMY_PACKET_LEN 34 | |
312 | /** | |
313 | * i40e_add_del_fdir_ipv4 - Add/Remove IPv4 Flow Director filters for | |
314 | * a specific flow spec | |
315 | * @vsi: pointer to the targeted VSI | |
316 | * @fd_data: the flow director data required for the FDir descriptor | |
317 | * @raw_packet: the pre-allocated packet buffer for FDir | |
318 | * @add: true adds a filter, false removes it | |
319 | * | |
320 | * Returns 0 if the filters were successfully added or removed | |
321 | **/ | |
322 | static int i40e_add_del_fdir_ipv4(struct i40e_vsi *vsi, | |
323 | struct i40e_fdir_filter *fd_data, | |
324 | u8 *raw_packet, bool add) | |
325 | { | |
326 | struct i40e_pf *pf = vsi->back; | |
327 | struct iphdr *ip; | |
328 | bool err = false; | |
329 | int ret; | |
330 | int i; | |
331 | static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0, | |
332 | 0x45, 0, 0, 0x14, 0, 0, 0x40, 0, 0x40, 0x10, 0, 0, 0, 0, 0, 0, | |
333 | 0, 0, 0, 0}; | |
334 | ||
335 | memcpy(raw_packet, packet, I40E_IP_DUMMY_PACKET_LEN); | |
336 | ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET); | |
337 | ||
338 | ip->saddr = fd_data->src_ip[0]; | |
339 | ip->daddr = fd_data->dst_ip[0]; | |
340 | ip->protocol = 0; | |
341 | ||
342 | for (i = I40E_FILTER_PCTYPE_NONF_IPV4_OTHER; | |
343 | i <= I40E_FILTER_PCTYPE_FRAG_IPV4; i++) { | |
344 | fd_data->pctype = i; | |
345 | ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add); | |
346 | ||
347 | if (ret) { | |
348 | dev_info(&pf->pdev->dev, | |
349 | "Filter command send failed for PCTYPE %d (ret = %d)\n", | |
350 | fd_data->pctype, ret); | |
351 | err = true; | |
352 | } else { | |
353 | dev_info(&pf->pdev->dev, | |
354 | "Filter OK for PCTYPE %d (ret = %d)\n", | |
355 | fd_data->pctype, ret); | |
356 | } | |
357 | } | |
358 | ||
359 | return err ? -EOPNOTSUPP : 0; | |
360 | } | |
361 | ||
362 | /** | |
363 | * i40e_add_del_fdir - Build raw packets to add/del fdir filter | |
364 | * @vsi: pointer to the targeted VSI | |
365 | * @cmd: command to get or set RX flow classification rules | |
366 | * @add: true adds a filter, false removes it | |
367 | * | |
368 | **/ | |
369 | int i40e_add_del_fdir(struct i40e_vsi *vsi, | |
370 | struct i40e_fdir_filter *input, bool add) | |
371 | { | |
372 | struct i40e_pf *pf = vsi->back; | |
373 | u8 *raw_packet; | |
374 | int ret; | |
375 | ||
376 | /* Populate the Flow Director that we have at the moment | |
377 | * and allocate the raw packet buffer for the calling functions | |
378 | */ | |
379 | raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL); | |
380 | if (!raw_packet) | |
381 | return -ENOMEM; | |
382 | ||
383 | switch (input->flow_type & ~FLOW_EXT) { | |
384 | case TCP_V4_FLOW: | |
385 | ret = i40e_add_del_fdir_tcpv4(vsi, input, raw_packet, | |
386 | add); | |
387 | break; | |
388 | case UDP_V4_FLOW: | |
389 | ret = i40e_add_del_fdir_udpv4(vsi, input, raw_packet, | |
390 | add); | |
391 | break; | |
392 | case SCTP_V4_FLOW: | |
393 | ret = i40e_add_del_fdir_sctpv4(vsi, input, raw_packet, | |
394 | add); | |
395 | break; | |
396 | case IPV4_FLOW: | |
397 | ret = i40e_add_del_fdir_ipv4(vsi, input, raw_packet, | |
398 | add); | |
399 | break; | |
400 | case IP_USER_FLOW: | |
401 | switch (input->ip4_proto) { | |
402 | case IPPROTO_TCP: | |
403 | ret = i40e_add_del_fdir_tcpv4(vsi, input, | |
404 | raw_packet, add); | |
405 | break; | |
406 | case IPPROTO_UDP: | |
407 | ret = i40e_add_del_fdir_udpv4(vsi, input, | |
408 | raw_packet, add); | |
409 | break; | |
410 | case IPPROTO_SCTP: | |
411 | ret = i40e_add_del_fdir_sctpv4(vsi, input, | |
412 | raw_packet, add); | |
413 | break; | |
414 | default: | |
415 | ret = i40e_add_del_fdir_ipv4(vsi, input, | |
416 | raw_packet, add); | |
417 | break; | |
418 | } | |
419 | break; | |
420 | default: | |
421 | dev_info(&pf->pdev->dev, "Could not specify spec type %d", | |
422 | input->flow_type); | |
423 | ret = -EINVAL; | |
424 | } | |
425 | ||
426 | kfree(raw_packet); | |
427 | return ret; | |
428 | } | |
429 | ||
fd0a05ce JB |
430 | /** |
431 | * i40e_fd_handle_status - check the Programming Status for FD | |
432 | * @rx_ring: the Rx ring for this descriptor | |
433 | * @qw: the descriptor data | |
434 | * @prog_id: the id originally used for programming | |
435 | * | |
436 | * This is used to verify if the FD programming or invalidation | |
437 | * requested by SW to the HW is successful or not and take actions accordingly. | |
438 | **/ | |
439 | static void i40e_fd_handle_status(struct i40e_ring *rx_ring, u32 qw, u8 prog_id) | |
440 | { | |
441 | struct pci_dev *pdev = rx_ring->vsi->back->pdev; | |
442 | u32 error; | |
443 | ||
444 | error = (qw & I40E_RX_PROG_STATUS_DESC_QW1_ERROR_MASK) >> | |
445 | I40E_RX_PROG_STATUS_DESC_QW1_ERROR_SHIFT; | |
446 | ||
447 | /* for now just print the Status */ | |
448 | dev_info(&pdev->dev, "FD programming id %02x, Status %08x\n", | |
449 | prog_id, error); | |
450 | } | |
451 | ||
452 | /** | |
a5e9c572 | 453 | * i40e_unmap_and_free_tx_resource - Release a Tx buffer |
fd0a05ce JB |
454 | * @ring: the ring that owns the buffer |
455 | * @tx_buffer: the buffer to free | |
456 | **/ | |
a5e9c572 AD |
457 | static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring, |
458 | struct i40e_tx_buffer *tx_buffer) | |
fd0a05ce | 459 | { |
a5e9c572 AD |
460 | if (tx_buffer->skb) { |
461 | dev_kfree_skb_any(tx_buffer->skb); | |
462 | if (dma_unmap_len(tx_buffer, len)) | |
fd0a05ce | 463 | dma_unmap_single(ring->dev, |
35a1e2ad AD |
464 | dma_unmap_addr(tx_buffer, dma), |
465 | dma_unmap_len(tx_buffer, len), | |
fd0a05ce | 466 | DMA_TO_DEVICE); |
a5e9c572 AD |
467 | } else if (dma_unmap_len(tx_buffer, len)) { |
468 | dma_unmap_page(ring->dev, | |
469 | dma_unmap_addr(tx_buffer, dma), | |
470 | dma_unmap_len(tx_buffer, len), | |
471 | DMA_TO_DEVICE); | |
fd0a05ce | 472 | } |
a5e9c572 AD |
473 | tx_buffer->next_to_watch = NULL; |
474 | tx_buffer->skb = NULL; | |
35a1e2ad | 475 | dma_unmap_len_set(tx_buffer, len, 0); |
a5e9c572 | 476 | /* tx_buffer must be completely set up in the transmit path */ |
fd0a05ce JB |
477 | } |
478 | ||
479 | /** | |
480 | * i40e_clean_tx_ring - Free any empty Tx buffers | |
481 | * @tx_ring: ring to be cleaned | |
482 | **/ | |
483 | void i40e_clean_tx_ring(struct i40e_ring *tx_ring) | |
484 | { | |
fd0a05ce JB |
485 | unsigned long bi_size; |
486 | u16 i; | |
487 | ||
488 | /* ring already cleared, nothing to do */ | |
489 | if (!tx_ring->tx_bi) | |
490 | return; | |
491 | ||
492 | /* Free all the Tx ring sk_buffs */ | |
a5e9c572 AD |
493 | for (i = 0; i < tx_ring->count; i++) |
494 | i40e_unmap_and_free_tx_resource(tx_ring, &tx_ring->tx_bi[i]); | |
fd0a05ce JB |
495 | |
496 | bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count; | |
497 | memset(tx_ring->tx_bi, 0, bi_size); | |
498 | ||
499 | /* Zero out the descriptor ring */ | |
500 | memset(tx_ring->desc, 0, tx_ring->size); | |
501 | ||
502 | tx_ring->next_to_use = 0; | |
503 | tx_ring->next_to_clean = 0; | |
7070ce0a AD |
504 | |
505 | if (!tx_ring->netdev) | |
506 | return; | |
507 | ||
508 | /* cleanup Tx queue statistics */ | |
509 | netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev, | |
510 | tx_ring->queue_index)); | |
fd0a05ce JB |
511 | } |
512 | ||
513 | /** | |
514 | * i40e_free_tx_resources - Free Tx resources per queue | |
515 | * @tx_ring: Tx descriptor ring for a specific queue | |
516 | * | |
517 | * Free all transmit software resources | |
518 | **/ | |
519 | void i40e_free_tx_resources(struct i40e_ring *tx_ring) | |
520 | { | |
521 | i40e_clean_tx_ring(tx_ring); | |
522 | kfree(tx_ring->tx_bi); | |
523 | tx_ring->tx_bi = NULL; | |
524 | ||
525 | if (tx_ring->desc) { | |
526 | dma_free_coherent(tx_ring->dev, tx_ring->size, | |
527 | tx_ring->desc, tx_ring->dma); | |
528 | tx_ring->desc = NULL; | |
529 | } | |
530 | } | |
531 | ||
532 | /** | |
533 | * i40e_get_tx_pending - how many tx descriptors not processed | |
534 | * @tx_ring: the ring of descriptors | |
535 | * | |
536 | * Since there is no access to the ring head register | |
537 | * in XL710, we need to use our local copies | |
538 | **/ | |
539 | static u32 i40e_get_tx_pending(struct i40e_ring *ring) | |
540 | { | |
541 | u32 ntu = ((ring->next_to_clean <= ring->next_to_use) | |
542 | ? ring->next_to_use | |
543 | : ring->next_to_use + ring->count); | |
544 | return ntu - ring->next_to_clean; | |
545 | } | |
546 | ||
547 | /** | |
548 | * i40e_check_tx_hang - Is there a hang in the Tx queue | |
549 | * @tx_ring: the ring of descriptors | |
550 | **/ | |
551 | static bool i40e_check_tx_hang(struct i40e_ring *tx_ring) | |
552 | { | |
553 | u32 tx_pending = i40e_get_tx_pending(tx_ring); | |
554 | bool ret = false; | |
555 | ||
556 | clear_check_for_tx_hang(tx_ring); | |
557 | ||
558 | /* Check for a hung queue, but be thorough. This verifies | |
559 | * that a transmit has been completed since the previous | |
560 | * check AND there is at least one packet pending. The | |
561 | * ARMED bit is set to indicate a potential hang. The | |
562 | * bit is cleared if a pause frame is received to remove | |
563 | * false hang detection due to PFC or 802.3x frames. By | |
564 | * requiring this to fail twice we avoid races with | |
565 | * PFC clearing the ARMED bit and conditions where we | |
566 | * run the check_tx_hang logic with a transmit completion | |
567 | * pending but without time to complete it yet. | |
568 | */ | |
a114d0a6 | 569 | if ((tx_ring->tx_stats.tx_done_old == tx_ring->stats.packets) && |
fd0a05ce JB |
570 | tx_pending) { |
571 | /* make sure it is true for two checks in a row */ | |
572 | ret = test_and_set_bit(__I40E_HANG_CHECK_ARMED, | |
573 | &tx_ring->state); | |
574 | } else { | |
575 | /* update completed stats and disarm the hang check */ | |
a114d0a6 | 576 | tx_ring->tx_stats.tx_done_old = tx_ring->stats.packets; |
fd0a05ce JB |
577 | clear_bit(__I40E_HANG_CHECK_ARMED, &tx_ring->state); |
578 | } | |
579 | ||
580 | return ret; | |
581 | } | |
582 | ||
583 | /** | |
584 | * i40e_clean_tx_irq - Reclaim resources after transmit completes | |
585 | * @tx_ring: tx ring to clean | |
586 | * @budget: how many cleans we're allowed | |
587 | * | |
588 | * Returns true if there's any budget left (e.g. the clean is finished) | |
589 | **/ | |
590 | static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget) | |
591 | { | |
592 | u16 i = tx_ring->next_to_clean; | |
593 | struct i40e_tx_buffer *tx_buf; | |
594 | struct i40e_tx_desc *tx_desc; | |
595 | unsigned int total_packets = 0; | |
596 | unsigned int total_bytes = 0; | |
597 | ||
598 | tx_buf = &tx_ring->tx_bi[i]; | |
599 | tx_desc = I40E_TX_DESC(tx_ring, i); | |
a5e9c572 | 600 | i -= tx_ring->count; |
fd0a05ce | 601 | |
a5e9c572 AD |
602 | do { |
603 | struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch; | |
fd0a05ce JB |
604 | |
605 | /* if next_to_watch is not set then there is no work pending */ | |
606 | if (!eop_desc) | |
607 | break; | |
608 | ||
a5e9c572 AD |
609 | /* prevent any other reads prior to eop_desc */ |
610 | read_barrier_depends(); | |
611 | ||
fd0a05ce JB |
612 | /* if the descriptor isn't done, no work yet to do */ |
613 | if (!(eop_desc->cmd_type_offset_bsz & | |
614 | cpu_to_le64(I40E_TX_DESC_DTYPE_DESC_DONE))) | |
615 | break; | |
616 | ||
c304fdac | 617 | /* clear next_to_watch to prevent false hangs */ |
fd0a05ce | 618 | tx_buf->next_to_watch = NULL; |
fd0a05ce | 619 | |
a5e9c572 AD |
620 | /* update the statistics for this packet */ |
621 | total_bytes += tx_buf->bytecount; | |
622 | total_packets += tx_buf->gso_segs; | |
fd0a05ce | 623 | |
a5e9c572 AD |
624 | /* free the skb */ |
625 | dev_kfree_skb_any(tx_buf->skb); | |
fd0a05ce | 626 | |
a5e9c572 AD |
627 | /* unmap skb header data */ |
628 | dma_unmap_single(tx_ring->dev, | |
629 | dma_unmap_addr(tx_buf, dma), | |
630 | dma_unmap_len(tx_buf, len), | |
631 | DMA_TO_DEVICE); | |
fd0a05ce | 632 | |
a5e9c572 AD |
633 | /* clear tx_buffer data */ |
634 | tx_buf->skb = NULL; | |
635 | dma_unmap_len_set(tx_buf, len, 0); | |
fd0a05ce | 636 | |
a5e9c572 AD |
637 | /* unmap remaining buffers */ |
638 | while (tx_desc != eop_desc) { | |
fd0a05ce JB |
639 | |
640 | tx_buf++; | |
641 | tx_desc++; | |
642 | i++; | |
a5e9c572 AD |
643 | if (unlikely(!i)) { |
644 | i -= tx_ring->count; | |
fd0a05ce JB |
645 | tx_buf = tx_ring->tx_bi; |
646 | tx_desc = I40E_TX_DESC(tx_ring, 0); | |
647 | } | |
fd0a05ce | 648 | |
a5e9c572 AD |
649 | /* unmap any remaining paged data */ |
650 | if (dma_unmap_len(tx_buf, len)) { | |
651 | dma_unmap_page(tx_ring->dev, | |
652 | dma_unmap_addr(tx_buf, dma), | |
653 | dma_unmap_len(tx_buf, len), | |
654 | DMA_TO_DEVICE); | |
655 | dma_unmap_len_set(tx_buf, len, 0); | |
656 | } | |
657 | } | |
658 | ||
659 | /* move us one more past the eop_desc for start of next pkt */ | |
660 | tx_buf++; | |
661 | tx_desc++; | |
662 | i++; | |
663 | if (unlikely(!i)) { | |
664 | i -= tx_ring->count; | |
665 | tx_buf = tx_ring->tx_bi; | |
666 | tx_desc = I40E_TX_DESC(tx_ring, 0); | |
667 | } | |
668 | ||
669 | /* update budget accounting */ | |
670 | budget--; | |
671 | } while (likely(budget)); | |
672 | ||
673 | i += tx_ring->count; | |
fd0a05ce | 674 | tx_ring->next_to_clean = i; |
980e9b11 | 675 | u64_stats_update_begin(&tx_ring->syncp); |
a114d0a6 AD |
676 | tx_ring->stats.bytes += total_bytes; |
677 | tx_ring->stats.packets += total_packets; | |
980e9b11 | 678 | u64_stats_update_end(&tx_ring->syncp); |
fd0a05ce JB |
679 | tx_ring->q_vector->tx.total_bytes += total_bytes; |
680 | tx_ring->q_vector->tx.total_packets += total_packets; | |
a5e9c572 | 681 | |
fd0a05ce JB |
682 | if (check_for_tx_hang(tx_ring) && i40e_check_tx_hang(tx_ring)) { |
683 | /* schedule immediate reset if we believe we hung */ | |
684 | dev_info(tx_ring->dev, "Detected Tx Unit Hang\n" | |
685 | " VSI <%d>\n" | |
686 | " Tx Queue <%d>\n" | |
687 | " next_to_use <%x>\n" | |
688 | " next_to_clean <%x>\n", | |
689 | tx_ring->vsi->seid, | |
690 | tx_ring->queue_index, | |
691 | tx_ring->next_to_use, i); | |
692 | dev_info(tx_ring->dev, "tx_bi[next_to_clean]\n" | |
693 | " time_stamp <%lx>\n" | |
694 | " jiffies <%lx>\n", | |
695 | tx_ring->tx_bi[i].time_stamp, jiffies); | |
696 | ||
697 | netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index); | |
698 | ||
699 | dev_info(tx_ring->dev, | |
700 | "tx hang detected on queue %d, resetting adapter\n", | |
701 | tx_ring->queue_index); | |
702 | ||
703 | tx_ring->netdev->netdev_ops->ndo_tx_timeout(tx_ring->netdev); | |
704 | ||
705 | /* the adapter is about to reset, no point in enabling stuff */ | |
706 | return true; | |
707 | } | |
708 | ||
7070ce0a AD |
709 | netdev_tx_completed_queue(netdev_get_tx_queue(tx_ring->netdev, |
710 | tx_ring->queue_index), | |
711 | total_packets, total_bytes); | |
712 | ||
fd0a05ce JB |
713 | #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2) |
714 | if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) && | |
715 | (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) { | |
716 | /* Make sure that anybody stopping the queue after this | |
717 | * sees the new next_to_clean. | |
718 | */ | |
719 | smp_mb(); | |
720 | if (__netif_subqueue_stopped(tx_ring->netdev, | |
721 | tx_ring->queue_index) && | |
722 | !test_bit(__I40E_DOWN, &tx_ring->vsi->state)) { | |
723 | netif_wake_subqueue(tx_ring->netdev, | |
724 | tx_ring->queue_index); | |
725 | ++tx_ring->tx_stats.restart_queue; | |
726 | } | |
727 | } | |
728 | ||
729 | return budget > 0; | |
730 | } | |
731 | ||
732 | /** | |
733 | * i40e_set_new_dynamic_itr - Find new ITR level | |
734 | * @rc: structure containing ring performance data | |
735 | * | |
736 | * Stores a new ITR value based on packets and byte counts during | |
737 | * the last interrupt. The advantage of per interrupt computation | |
738 | * is faster updates and more accurate ITR for the current traffic | |
739 | * pattern. Constants in this function were computed based on | |
740 | * theoretical maximum wire speed and thresholds were set based on | |
741 | * testing data as well as attempting to minimize response time | |
742 | * while increasing bulk throughput. | |
743 | **/ | |
744 | static void i40e_set_new_dynamic_itr(struct i40e_ring_container *rc) | |
745 | { | |
746 | enum i40e_latency_range new_latency_range = rc->latency_range; | |
747 | u32 new_itr = rc->itr; | |
748 | int bytes_per_int; | |
749 | ||
750 | if (rc->total_packets == 0 || !rc->itr) | |
751 | return; | |
752 | ||
753 | /* simple throttlerate management | |
754 | * 0-10MB/s lowest (100000 ints/s) | |
755 | * 10-20MB/s low (20000 ints/s) | |
756 | * 20-1249MB/s bulk (8000 ints/s) | |
757 | */ | |
758 | bytes_per_int = rc->total_bytes / rc->itr; | |
759 | switch (rc->itr) { | |
760 | case I40E_LOWEST_LATENCY: | |
761 | if (bytes_per_int > 10) | |
762 | new_latency_range = I40E_LOW_LATENCY; | |
763 | break; | |
764 | case I40E_LOW_LATENCY: | |
765 | if (bytes_per_int > 20) | |
766 | new_latency_range = I40E_BULK_LATENCY; | |
767 | else if (bytes_per_int <= 10) | |
768 | new_latency_range = I40E_LOWEST_LATENCY; | |
769 | break; | |
770 | case I40E_BULK_LATENCY: | |
771 | if (bytes_per_int <= 20) | |
772 | rc->latency_range = I40E_LOW_LATENCY; | |
773 | break; | |
774 | } | |
775 | ||
776 | switch (new_latency_range) { | |
777 | case I40E_LOWEST_LATENCY: | |
778 | new_itr = I40E_ITR_100K; | |
779 | break; | |
780 | case I40E_LOW_LATENCY: | |
781 | new_itr = I40E_ITR_20K; | |
782 | break; | |
783 | case I40E_BULK_LATENCY: | |
784 | new_itr = I40E_ITR_8K; | |
785 | break; | |
786 | default: | |
787 | break; | |
788 | } | |
789 | ||
790 | if (new_itr != rc->itr) { | |
791 | /* do an exponential smoothing */ | |
792 | new_itr = (10 * new_itr * rc->itr) / | |
793 | ((9 * new_itr) + rc->itr); | |
794 | rc->itr = new_itr & I40E_MAX_ITR; | |
795 | } | |
796 | ||
797 | rc->total_bytes = 0; | |
798 | rc->total_packets = 0; | |
799 | } | |
800 | ||
801 | /** | |
802 | * i40e_update_dynamic_itr - Adjust ITR based on bytes per int | |
803 | * @q_vector: the vector to adjust | |
804 | **/ | |
805 | static void i40e_update_dynamic_itr(struct i40e_q_vector *q_vector) | |
806 | { | |
807 | u16 vector = q_vector->vsi->base_vector + q_vector->v_idx; | |
808 | struct i40e_hw *hw = &q_vector->vsi->back->hw; | |
809 | u32 reg_addr; | |
810 | u16 old_itr; | |
811 | ||
812 | reg_addr = I40E_PFINT_ITRN(I40E_RX_ITR, vector - 1); | |
813 | old_itr = q_vector->rx.itr; | |
814 | i40e_set_new_dynamic_itr(&q_vector->rx); | |
815 | if (old_itr != q_vector->rx.itr) | |
816 | wr32(hw, reg_addr, q_vector->rx.itr); | |
817 | ||
818 | reg_addr = I40E_PFINT_ITRN(I40E_TX_ITR, vector - 1); | |
819 | old_itr = q_vector->tx.itr; | |
820 | i40e_set_new_dynamic_itr(&q_vector->tx); | |
821 | if (old_itr != q_vector->tx.itr) | |
822 | wr32(hw, reg_addr, q_vector->tx.itr); | |
fd0a05ce JB |
823 | } |
824 | ||
825 | /** | |
826 | * i40e_clean_programming_status - clean the programming status descriptor | |
827 | * @rx_ring: the rx ring that has this descriptor | |
828 | * @rx_desc: the rx descriptor written back by HW | |
829 | * | |
830 | * Flow director should handle FD_FILTER_STATUS to check its filter programming | |
831 | * status being successful or not and take actions accordingly. FCoE should | |
832 | * handle its context/filter programming/invalidation status and take actions. | |
833 | * | |
834 | **/ | |
835 | static void i40e_clean_programming_status(struct i40e_ring *rx_ring, | |
836 | union i40e_rx_desc *rx_desc) | |
837 | { | |
838 | u64 qw; | |
839 | u8 id; | |
840 | ||
841 | qw = le64_to_cpu(rx_desc->wb.qword1.status_error_len); | |
842 | id = (qw & I40E_RX_PROG_STATUS_DESC_QW1_PROGID_MASK) >> | |
843 | I40E_RX_PROG_STATUS_DESC_QW1_PROGID_SHIFT; | |
844 | ||
845 | if (id == I40E_RX_PROG_STATUS_DESC_FD_FILTER_STATUS) | |
846 | i40e_fd_handle_status(rx_ring, qw, id); | |
847 | } | |
848 | ||
849 | /** | |
850 | * i40e_setup_tx_descriptors - Allocate the Tx descriptors | |
851 | * @tx_ring: the tx ring to set up | |
852 | * | |
853 | * Return 0 on success, negative on error | |
854 | **/ | |
855 | int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring) | |
856 | { | |
857 | struct device *dev = tx_ring->dev; | |
858 | int bi_size; | |
859 | ||
860 | if (!dev) | |
861 | return -ENOMEM; | |
862 | ||
863 | bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count; | |
864 | tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL); | |
865 | if (!tx_ring->tx_bi) | |
866 | goto err; | |
867 | ||
868 | /* round up to nearest 4K */ | |
869 | tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc); | |
870 | tx_ring->size = ALIGN(tx_ring->size, 4096); | |
871 | tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size, | |
872 | &tx_ring->dma, GFP_KERNEL); | |
873 | if (!tx_ring->desc) { | |
874 | dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n", | |
875 | tx_ring->size); | |
876 | goto err; | |
877 | } | |
878 | ||
879 | tx_ring->next_to_use = 0; | |
880 | tx_ring->next_to_clean = 0; | |
881 | return 0; | |
882 | ||
883 | err: | |
884 | kfree(tx_ring->tx_bi); | |
885 | tx_ring->tx_bi = NULL; | |
886 | return -ENOMEM; | |
887 | } | |
888 | ||
889 | /** | |
890 | * i40e_clean_rx_ring - Free Rx buffers | |
891 | * @rx_ring: ring to be cleaned | |
892 | **/ | |
893 | void i40e_clean_rx_ring(struct i40e_ring *rx_ring) | |
894 | { | |
895 | struct device *dev = rx_ring->dev; | |
896 | struct i40e_rx_buffer *rx_bi; | |
897 | unsigned long bi_size; | |
898 | u16 i; | |
899 | ||
900 | /* ring already cleared, nothing to do */ | |
901 | if (!rx_ring->rx_bi) | |
902 | return; | |
903 | ||
904 | /* Free all the Rx ring sk_buffs */ | |
905 | for (i = 0; i < rx_ring->count; i++) { | |
906 | rx_bi = &rx_ring->rx_bi[i]; | |
907 | if (rx_bi->dma) { | |
908 | dma_unmap_single(dev, | |
909 | rx_bi->dma, | |
910 | rx_ring->rx_buf_len, | |
911 | DMA_FROM_DEVICE); | |
912 | rx_bi->dma = 0; | |
913 | } | |
914 | if (rx_bi->skb) { | |
915 | dev_kfree_skb(rx_bi->skb); | |
916 | rx_bi->skb = NULL; | |
917 | } | |
918 | if (rx_bi->page) { | |
919 | if (rx_bi->page_dma) { | |
920 | dma_unmap_page(dev, | |
921 | rx_bi->page_dma, | |
922 | PAGE_SIZE / 2, | |
923 | DMA_FROM_DEVICE); | |
924 | rx_bi->page_dma = 0; | |
925 | } | |
926 | __free_page(rx_bi->page); | |
927 | rx_bi->page = NULL; | |
928 | rx_bi->page_offset = 0; | |
929 | } | |
930 | } | |
931 | ||
932 | bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count; | |
933 | memset(rx_ring->rx_bi, 0, bi_size); | |
934 | ||
935 | /* Zero out the descriptor ring */ | |
936 | memset(rx_ring->desc, 0, rx_ring->size); | |
937 | ||
938 | rx_ring->next_to_clean = 0; | |
939 | rx_ring->next_to_use = 0; | |
940 | } | |
941 | ||
942 | /** | |
943 | * i40e_free_rx_resources - Free Rx resources | |
944 | * @rx_ring: ring to clean the resources from | |
945 | * | |
946 | * Free all receive software resources | |
947 | **/ | |
948 | void i40e_free_rx_resources(struct i40e_ring *rx_ring) | |
949 | { | |
950 | i40e_clean_rx_ring(rx_ring); | |
951 | kfree(rx_ring->rx_bi); | |
952 | rx_ring->rx_bi = NULL; | |
953 | ||
954 | if (rx_ring->desc) { | |
955 | dma_free_coherent(rx_ring->dev, rx_ring->size, | |
956 | rx_ring->desc, rx_ring->dma); | |
957 | rx_ring->desc = NULL; | |
958 | } | |
959 | } | |
960 | ||
961 | /** | |
962 | * i40e_setup_rx_descriptors - Allocate Rx descriptors | |
963 | * @rx_ring: Rx descriptor ring (for a specific queue) to setup | |
964 | * | |
965 | * Returns 0 on success, negative on failure | |
966 | **/ | |
967 | int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring) | |
968 | { | |
969 | struct device *dev = rx_ring->dev; | |
970 | int bi_size; | |
971 | ||
972 | bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count; | |
973 | rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL); | |
974 | if (!rx_ring->rx_bi) | |
975 | goto err; | |
976 | ||
977 | /* Round up to nearest 4K */ | |
978 | rx_ring->size = ring_is_16byte_desc_enabled(rx_ring) | |
979 | ? rx_ring->count * sizeof(union i40e_16byte_rx_desc) | |
980 | : rx_ring->count * sizeof(union i40e_32byte_rx_desc); | |
981 | rx_ring->size = ALIGN(rx_ring->size, 4096); | |
982 | rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size, | |
983 | &rx_ring->dma, GFP_KERNEL); | |
984 | ||
985 | if (!rx_ring->desc) { | |
986 | dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n", | |
987 | rx_ring->size); | |
988 | goto err; | |
989 | } | |
990 | ||
991 | rx_ring->next_to_clean = 0; | |
992 | rx_ring->next_to_use = 0; | |
993 | ||
994 | return 0; | |
995 | err: | |
996 | kfree(rx_ring->rx_bi); | |
997 | rx_ring->rx_bi = NULL; | |
998 | return -ENOMEM; | |
999 | } | |
1000 | ||
1001 | /** | |
1002 | * i40e_release_rx_desc - Store the new tail and head values | |
1003 | * @rx_ring: ring to bump | |
1004 | * @val: new head index | |
1005 | **/ | |
1006 | static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val) | |
1007 | { | |
1008 | rx_ring->next_to_use = val; | |
1009 | /* Force memory writes to complete before letting h/w | |
1010 | * know there are new descriptors to fetch. (Only | |
1011 | * applicable for weak-ordered memory model archs, | |
1012 | * such as IA-64). | |
1013 | */ | |
1014 | wmb(); | |
1015 | writel(val, rx_ring->tail); | |
1016 | } | |
1017 | ||
1018 | /** | |
1019 | * i40e_alloc_rx_buffers - Replace used receive buffers; packet split | |
1020 | * @rx_ring: ring to place buffers on | |
1021 | * @cleaned_count: number of buffers to replace | |
1022 | **/ | |
1023 | void i40e_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count) | |
1024 | { | |
1025 | u16 i = rx_ring->next_to_use; | |
1026 | union i40e_rx_desc *rx_desc; | |
1027 | struct i40e_rx_buffer *bi; | |
1028 | struct sk_buff *skb; | |
1029 | ||
1030 | /* do nothing if no valid netdev defined */ | |
1031 | if (!rx_ring->netdev || !cleaned_count) | |
1032 | return; | |
1033 | ||
1034 | while (cleaned_count--) { | |
1035 | rx_desc = I40E_RX_DESC(rx_ring, i); | |
1036 | bi = &rx_ring->rx_bi[i]; | |
1037 | skb = bi->skb; | |
1038 | ||
1039 | if (!skb) { | |
1040 | skb = netdev_alloc_skb_ip_align(rx_ring->netdev, | |
1041 | rx_ring->rx_buf_len); | |
1042 | if (!skb) { | |
420136cc | 1043 | rx_ring->rx_stats.alloc_buff_failed++; |
fd0a05ce JB |
1044 | goto no_buffers; |
1045 | } | |
1046 | /* initialize queue mapping */ | |
1047 | skb_record_rx_queue(skb, rx_ring->queue_index); | |
1048 | bi->skb = skb; | |
1049 | } | |
1050 | ||
1051 | if (!bi->dma) { | |
1052 | bi->dma = dma_map_single(rx_ring->dev, | |
1053 | skb->data, | |
1054 | rx_ring->rx_buf_len, | |
1055 | DMA_FROM_DEVICE); | |
1056 | if (dma_mapping_error(rx_ring->dev, bi->dma)) { | |
420136cc | 1057 | rx_ring->rx_stats.alloc_buff_failed++; |
fd0a05ce JB |
1058 | bi->dma = 0; |
1059 | goto no_buffers; | |
1060 | } | |
1061 | } | |
1062 | ||
1063 | if (ring_is_ps_enabled(rx_ring)) { | |
1064 | if (!bi->page) { | |
1065 | bi->page = alloc_page(GFP_ATOMIC); | |
1066 | if (!bi->page) { | |
420136cc | 1067 | rx_ring->rx_stats.alloc_page_failed++; |
fd0a05ce JB |
1068 | goto no_buffers; |
1069 | } | |
1070 | } | |
1071 | ||
1072 | if (!bi->page_dma) { | |
1073 | /* use a half page if we're re-using */ | |
1074 | bi->page_offset ^= PAGE_SIZE / 2; | |
1075 | bi->page_dma = dma_map_page(rx_ring->dev, | |
1076 | bi->page, | |
1077 | bi->page_offset, | |
1078 | PAGE_SIZE / 2, | |
1079 | DMA_FROM_DEVICE); | |
1080 | if (dma_mapping_error(rx_ring->dev, | |
1081 | bi->page_dma)) { | |
420136cc | 1082 | rx_ring->rx_stats.alloc_page_failed++; |
fd0a05ce JB |
1083 | bi->page_dma = 0; |
1084 | goto no_buffers; | |
1085 | } | |
1086 | } | |
1087 | ||
1088 | /* Refresh the desc even if buffer_addrs didn't change | |
1089 | * because each write-back erases this info. | |
1090 | */ | |
1091 | rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma); | |
1092 | rx_desc->read.hdr_addr = cpu_to_le64(bi->dma); | |
1093 | } else { | |
1094 | rx_desc->read.pkt_addr = cpu_to_le64(bi->dma); | |
1095 | rx_desc->read.hdr_addr = 0; | |
1096 | } | |
1097 | i++; | |
1098 | if (i == rx_ring->count) | |
1099 | i = 0; | |
1100 | } | |
1101 | ||
1102 | no_buffers: | |
1103 | if (rx_ring->next_to_use != i) | |
1104 | i40e_release_rx_desc(rx_ring, i); | |
1105 | } | |
1106 | ||
1107 | /** | |
1108 | * i40e_receive_skb - Send a completed packet up the stack | |
1109 | * @rx_ring: rx ring in play | |
1110 | * @skb: packet to send up | |
1111 | * @vlan_tag: vlan tag for packet | |
1112 | **/ | |
1113 | static void i40e_receive_skb(struct i40e_ring *rx_ring, | |
1114 | struct sk_buff *skb, u16 vlan_tag) | |
1115 | { | |
1116 | struct i40e_q_vector *q_vector = rx_ring->q_vector; | |
1117 | struct i40e_vsi *vsi = rx_ring->vsi; | |
1118 | u64 flags = vsi->back->flags; | |
1119 | ||
1120 | if (vlan_tag & VLAN_VID_MASK) | |
1121 | __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag); | |
1122 | ||
1123 | if (flags & I40E_FLAG_IN_NETPOLL) | |
1124 | netif_rx(skb); | |
1125 | else | |
1126 | napi_gro_receive(&q_vector->napi, skb); | |
1127 | } | |
1128 | ||
1129 | /** | |
1130 | * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum | |
1131 | * @vsi: the VSI we care about | |
1132 | * @skb: skb currently being received and modified | |
1133 | * @rx_status: status value of last descriptor in packet | |
1134 | * @rx_error: error value of last descriptor in packet | |
8144f0f7 | 1135 | * @rx_ptype: ptype value of last descriptor in packet |
fd0a05ce JB |
1136 | **/ |
1137 | static inline void i40e_rx_checksum(struct i40e_vsi *vsi, | |
1138 | struct sk_buff *skb, | |
1139 | u32 rx_status, | |
8144f0f7 JG |
1140 | u32 rx_error, |
1141 | u16 rx_ptype) | |
fd0a05ce | 1142 | { |
8144f0f7 JG |
1143 | bool ipv4_tunnel, ipv6_tunnel; |
1144 | __wsum rx_udp_csum; | |
1145 | __sum16 csum; | |
1146 | struct iphdr *iph; | |
1147 | ||
1148 | ipv4_tunnel = (rx_ptype > I40E_RX_PTYPE_GRENAT4_MAC_PAY3) && | |
1149 | (rx_ptype < I40E_RX_PTYPE_GRENAT4_MACVLAN_IPV6_ICMP_PAY4); | |
1150 | ipv6_tunnel = (rx_ptype > I40E_RX_PTYPE_GRENAT6_MAC_PAY3) && | |
1151 | (rx_ptype < I40E_RX_PTYPE_GRENAT6_MACVLAN_IPV6_ICMP_PAY4); | |
1152 | ||
1153 | skb->encapsulation = ipv4_tunnel || ipv6_tunnel; | |
fd0a05ce JB |
1154 | skb->ip_summed = CHECKSUM_NONE; |
1155 | ||
1156 | /* Rx csum enabled and ip headers found? */ | |
1157 | if (!(vsi->netdev->features & NETIF_F_RXCSUM && | |
1158 | rx_status & (1 << I40E_RX_DESC_STATUS_L3L4P_SHIFT))) | |
1159 | return; | |
1160 | ||
ddf1d0d7 | 1161 | /* likely incorrect csum if alternate IP extension headers found */ |
8ee75a8e SN |
1162 | if (rx_status & (1 << I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT)) |
1163 | return; | |
1164 | ||
8144f0f7 | 1165 | /* IP or L4 or outmost IP checksum error */ |
fd0a05ce | 1166 | if (rx_error & ((1 << I40E_RX_DESC_ERROR_IPE_SHIFT) | |
8144f0f7 JG |
1167 | (1 << I40E_RX_DESC_ERROR_L4E_SHIFT) | |
1168 | (1 << I40E_RX_DESC_ERROR_EIPE_SHIFT))) { | |
fd0a05ce JB |
1169 | vsi->back->hw_csum_rx_error++; |
1170 | return; | |
1171 | } | |
1172 | ||
8144f0f7 JG |
1173 | if (ipv4_tunnel && |
1174 | !(rx_status & (1 << I40E_RX_DESC_STATUS_UDP_0_SHIFT))) { | |
1175 | /* If VXLAN traffic has an outer UDPv4 checksum we need to check | |
1176 | * it in the driver, hardware does not do it for us. | |
1177 | * Since L3L4P bit was set we assume a valid IHL value (>=5) | |
1178 | * so the total length of IPv4 header is IHL*4 bytes | |
1179 | */ | |
1180 | skb->transport_header = skb->mac_header + | |
1181 | sizeof(struct ethhdr) + | |
1182 | (ip_hdr(skb)->ihl * 4); | |
1183 | ||
1184 | /* Add 4 bytes for VLAN tagged packets */ | |
1185 | skb->transport_header += (skb->protocol == htons(ETH_P_8021Q) || | |
1186 | skb->protocol == htons(ETH_P_8021AD)) | |
1187 | ? VLAN_HLEN : 0; | |
1188 | ||
1189 | rx_udp_csum = udp_csum(skb); | |
1190 | iph = ip_hdr(skb); | |
1191 | csum = csum_tcpudp_magic( | |
1192 | iph->saddr, iph->daddr, | |
1193 | (skb->len - skb_transport_offset(skb)), | |
1194 | IPPROTO_UDP, rx_udp_csum); | |
1195 | ||
1196 | if (udp_hdr(skb)->check != csum) { | |
1197 | vsi->back->hw_csum_rx_error++; | |
1198 | return; | |
1199 | } | |
1200 | } | |
1201 | ||
fd0a05ce JB |
1202 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
1203 | } | |
1204 | ||
1205 | /** | |
1206 | * i40e_rx_hash - returns the hash value from the Rx descriptor | |
1207 | * @ring: descriptor ring | |
1208 | * @rx_desc: specific descriptor | |
1209 | **/ | |
1210 | static inline u32 i40e_rx_hash(struct i40e_ring *ring, | |
1211 | union i40e_rx_desc *rx_desc) | |
1212 | { | |
8a494920 JB |
1213 | const __le64 rss_mask = |
1214 | cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH << | |
1215 | I40E_RX_DESC_STATUS_FLTSTAT_SHIFT); | |
1216 | ||
1217 | if ((ring->netdev->features & NETIF_F_RXHASH) && | |
1218 | (rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask) | |
1219 | return le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss); | |
1220 | else | |
1221 | return 0; | |
fd0a05ce JB |
1222 | } |
1223 | ||
206812b5 JB |
1224 | /** |
1225 | * i40e_ptype_to_hash - get a hash type | |
1226 | * @ptype: the ptype value from the descriptor | |
1227 | * | |
1228 | * Returns a hash type to be used by skb_set_hash | |
1229 | **/ | |
1230 | static inline enum pkt_hash_types i40e_ptype_to_hash(u8 ptype) | |
1231 | { | |
1232 | struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype); | |
1233 | ||
1234 | if (!decoded.known) | |
1235 | return PKT_HASH_TYPE_NONE; | |
1236 | ||
1237 | if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP && | |
1238 | decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4) | |
1239 | return PKT_HASH_TYPE_L4; | |
1240 | else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP && | |
1241 | decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3) | |
1242 | return PKT_HASH_TYPE_L3; | |
1243 | else | |
1244 | return PKT_HASH_TYPE_L2; | |
1245 | } | |
1246 | ||
fd0a05ce JB |
1247 | /** |
1248 | * i40e_clean_rx_irq - Reclaim resources after receive completes | |
1249 | * @rx_ring: rx ring to clean | |
1250 | * @budget: how many cleans we're allowed | |
1251 | * | |
1252 | * Returns true if there's any budget left (e.g. the clean is finished) | |
1253 | **/ | |
1254 | static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget) | |
1255 | { | |
1256 | unsigned int total_rx_bytes = 0, total_rx_packets = 0; | |
1257 | u16 rx_packet_len, rx_header_len, rx_sph, rx_hbo; | |
1258 | u16 cleaned_count = I40E_DESC_UNUSED(rx_ring); | |
1259 | const int current_node = numa_node_id(); | |
1260 | struct i40e_vsi *vsi = rx_ring->vsi; | |
1261 | u16 i = rx_ring->next_to_clean; | |
1262 | union i40e_rx_desc *rx_desc; | |
1263 | u32 rx_error, rx_status; | |
206812b5 | 1264 | u8 rx_ptype; |
fd0a05ce JB |
1265 | u64 qword; |
1266 | ||
1267 | rx_desc = I40E_RX_DESC(rx_ring, i); | |
1268 | qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len); | |
6838b535 JB |
1269 | rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >> |
1270 | I40E_RXD_QW1_STATUS_SHIFT; | |
fd0a05ce JB |
1271 | |
1272 | while (rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)) { | |
1273 | union i40e_rx_desc *next_rxd; | |
1274 | struct i40e_rx_buffer *rx_bi; | |
1275 | struct sk_buff *skb; | |
1276 | u16 vlan_tag; | |
1277 | if (i40e_rx_is_programming_status(qword)) { | |
1278 | i40e_clean_programming_status(rx_ring, rx_desc); | |
1279 | I40E_RX_NEXT_DESC_PREFETCH(rx_ring, i, next_rxd); | |
1280 | goto next_desc; | |
1281 | } | |
1282 | rx_bi = &rx_ring->rx_bi[i]; | |
1283 | skb = rx_bi->skb; | |
1284 | prefetch(skb->data); | |
1285 | ||
829af3ac MW |
1286 | rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >> |
1287 | I40E_RXD_QW1_LENGTH_PBUF_SHIFT; | |
1288 | rx_header_len = (qword & I40E_RXD_QW1_LENGTH_HBUF_MASK) >> | |
1289 | I40E_RXD_QW1_LENGTH_HBUF_SHIFT; | |
1290 | rx_sph = (qword & I40E_RXD_QW1_LENGTH_SPH_MASK) >> | |
1291 | I40E_RXD_QW1_LENGTH_SPH_SHIFT; | |
1292 | ||
1293 | rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >> | |
1294 | I40E_RXD_QW1_ERROR_SHIFT; | |
fd0a05ce JB |
1295 | rx_hbo = rx_error & (1 << I40E_RX_DESC_ERROR_HBO_SHIFT); |
1296 | rx_error &= ~(1 << I40E_RX_DESC_ERROR_HBO_SHIFT); | |
1297 | ||
8144f0f7 JG |
1298 | rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >> |
1299 | I40E_RXD_QW1_PTYPE_SHIFT; | |
fd0a05ce JB |
1300 | rx_bi->skb = NULL; |
1301 | ||
1302 | /* This memory barrier is needed to keep us from reading | |
1303 | * any other fields out of the rx_desc until we know the | |
1304 | * STATUS_DD bit is set | |
1305 | */ | |
1306 | rmb(); | |
1307 | ||
1308 | /* Get the header and possibly the whole packet | |
1309 | * If this is an skb from previous receive dma will be 0 | |
1310 | */ | |
1311 | if (rx_bi->dma) { | |
1312 | u16 len; | |
1313 | ||
1314 | if (rx_hbo) | |
1315 | len = I40E_RX_HDR_SIZE; | |
1316 | else if (rx_sph) | |
1317 | len = rx_header_len; | |
1318 | else if (rx_packet_len) | |
1319 | len = rx_packet_len; /* 1buf/no split found */ | |
1320 | else | |
1321 | len = rx_header_len; /* split always mode */ | |
1322 | ||
1323 | skb_put(skb, len); | |
1324 | dma_unmap_single(rx_ring->dev, | |
1325 | rx_bi->dma, | |
1326 | rx_ring->rx_buf_len, | |
1327 | DMA_FROM_DEVICE); | |
1328 | rx_bi->dma = 0; | |
1329 | } | |
1330 | ||
1331 | /* Get the rest of the data if this was a header split */ | |
1332 | if (ring_is_ps_enabled(rx_ring) && rx_packet_len) { | |
1333 | ||
1334 | skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags, | |
1335 | rx_bi->page, | |
1336 | rx_bi->page_offset, | |
1337 | rx_packet_len); | |
1338 | ||
1339 | skb->len += rx_packet_len; | |
1340 | skb->data_len += rx_packet_len; | |
1341 | skb->truesize += rx_packet_len; | |
1342 | ||
1343 | if ((page_count(rx_bi->page) == 1) && | |
1344 | (page_to_nid(rx_bi->page) == current_node)) | |
1345 | get_page(rx_bi->page); | |
1346 | else | |
1347 | rx_bi->page = NULL; | |
1348 | ||
1349 | dma_unmap_page(rx_ring->dev, | |
1350 | rx_bi->page_dma, | |
1351 | PAGE_SIZE / 2, | |
1352 | DMA_FROM_DEVICE); | |
1353 | rx_bi->page_dma = 0; | |
1354 | } | |
1355 | I40E_RX_NEXT_DESC_PREFETCH(rx_ring, i, next_rxd); | |
1356 | ||
1357 | if (unlikely( | |
1358 | !(rx_status & (1 << I40E_RX_DESC_STATUS_EOF_SHIFT)))) { | |
1359 | struct i40e_rx_buffer *next_buffer; | |
1360 | ||
1361 | next_buffer = &rx_ring->rx_bi[i]; | |
1362 | ||
1363 | if (ring_is_ps_enabled(rx_ring)) { | |
1364 | rx_bi->skb = next_buffer->skb; | |
1365 | rx_bi->dma = next_buffer->dma; | |
1366 | next_buffer->skb = skb; | |
1367 | next_buffer->dma = 0; | |
1368 | } | |
1369 | rx_ring->rx_stats.non_eop_descs++; | |
1370 | goto next_desc; | |
1371 | } | |
1372 | ||
1373 | /* ERR_MASK will only have valid bits if EOP set */ | |
1374 | if (unlikely(rx_error & (1 << I40E_RX_DESC_ERROR_RXE_SHIFT))) { | |
1375 | dev_kfree_skb_any(skb); | |
1376 | goto next_desc; | |
1377 | } | |
1378 | ||
206812b5 JB |
1379 | skb_set_hash(skb, i40e_rx_hash(rx_ring, rx_desc), |
1380 | i40e_ptype_to_hash(rx_ptype)); | |
beb0dff1 JK |
1381 | if (unlikely(rx_status & I40E_RXD_QW1_STATUS_TSYNVALID_MASK)) { |
1382 | i40e_ptp_rx_hwtstamp(vsi->back, skb, (rx_status & | |
1383 | I40E_RXD_QW1_STATUS_TSYNINDX_MASK) >> | |
1384 | I40E_RXD_QW1_STATUS_TSYNINDX_SHIFT); | |
1385 | rx_ring->last_rx_timestamp = jiffies; | |
1386 | } | |
1387 | ||
fd0a05ce JB |
1388 | /* probably a little skewed due to removing CRC */ |
1389 | total_rx_bytes += skb->len; | |
1390 | total_rx_packets++; | |
1391 | ||
1392 | skb->protocol = eth_type_trans(skb, rx_ring->netdev); | |
8144f0f7 JG |
1393 | |
1394 | i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype); | |
1395 | ||
fd0a05ce JB |
1396 | vlan_tag = rx_status & (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT) |
1397 | ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1) | |
1398 | : 0; | |
1399 | i40e_receive_skb(rx_ring, skb, vlan_tag); | |
1400 | ||
1401 | rx_ring->netdev->last_rx = jiffies; | |
1402 | budget--; | |
1403 | next_desc: | |
1404 | rx_desc->wb.qword1.status_error_len = 0; | |
1405 | if (!budget) | |
1406 | break; | |
1407 | ||
1408 | cleaned_count++; | |
1409 | /* return some buffers to hardware, one at a time is too slow */ | |
1410 | if (cleaned_count >= I40E_RX_BUFFER_WRITE) { | |
1411 | i40e_alloc_rx_buffers(rx_ring, cleaned_count); | |
1412 | cleaned_count = 0; | |
1413 | } | |
1414 | ||
1415 | /* use prefetched values */ | |
1416 | rx_desc = next_rxd; | |
1417 | qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len); | |
829af3ac MW |
1418 | rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >> |
1419 | I40E_RXD_QW1_STATUS_SHIFT; | |
fd0a05ce JB |
1420 | } |
1421 | ||
1422 | rx_ring->next_to_clean = i; | |
980e9b11 | 1423 | u64_stats_update_begin(&rx_ring->syncp); |
a114d0a6 AD |
1424 | rx_ring->stats.packets += total_rx_packets; |
1425 | rx_ring->stats.bytes += total_rx_bytes; | |
980e9b11 | 1426 | u64_stats_update_end(&rx_ring->syncp); |
fd0a05ce JB |
1427 | rx_ring->q_vector->rx.total_packets += total_rx_packets; |
1428 | rx_ring->q_vector->rx.total_bytes += total_rx_bytes; | |
1429 | ||
1430 | if (cleaned_count) | |
1431 | i40e_alloc_rx_buffers(rx_ring, cleaned_count); | |
1432 | ||
1433 | return budget > 0; | |
1434 | } | |
1435 | ||
1436 | /** | |
1437 | * i40e_napi_poll - NAPI polling Rx/Tx cleanup routine | |
1438 | * @napi: napi struct with our devices info in it | |
1439 | * @budget: amount of work driver is allowed to do this pass, in packets | |
1440 | * | |
1441 | * This function will clean all queues associated with a q_vector. | |
1442 | * | |
1443 | * Returns the amount of work done | |
1444 | **/ | |
1445 | int i40e_napi_poll(struct napi_struct *napi, int budget) | |
1446 | { | |
1447 | struct i40e_q_vector *q_vector = | |
1448 | container_of(napi, struct i40e_q_vector, napi); | |
1449 | struct i40e_vsi *vsi = q_vector->vsi; | |
cd0b6fa6 | 1450 | struct i40e_ring *ring; |
fd0a05ce JB |
1451 | bool clean_complete = true; |
1452 | int budget_per_ring; | |
fd0a05ce JB |
1453 | |
1454 | if (test_bit(__I40E_DOWN, &vsi->state)) { | |
1455 | napi_complete(napi); | |
1456 | return 0; | |
1457 | } | |
1458 | ||
cd0b6fa6 AD |
1459 | /* Since the actual Tx work is minimal, we can give the Tx a larger |
1460 | * budget and be more aggressive about cleaning up the Tx descriptors. | |
1461 | */ | |
1462 | i40e_for_each_ring(ring, q_vector->tx) | |
1463 | clean_complete &= i40e_clean_tx_irq(ring, vsi->work_limit); | |
1464 | ||
fd0a05ce JB |
1465 | /* We attempt to distribute budget to each Rx queue fairly, but don't |
1466 | * allow the budget to go below 1 because that would exit polling early. | |
fd0a05ce JB |
1467 | */ |
1468 | budget_per_ring = max(budget/q_vector->num_ringpairs, 1); | |
cd0b6fa6 AD |
1469 | |
1470 | i40e_for_each_ring(ring, q_vector->rx) | |
1471 | clean_complete &= i40e_clean_rx_irq(ring, budget_per_ring); | |
fd0a05ce JB |
1472 | |
1473 | /* If work not completed, return budget and polling will return */ | |
1474 | if (!clean_complete) | |
1475 | return budget; | |
1476 | ||
1477 | /* Work is done so exit the polling mode and re-enable the interrupt */ | |
1478 | napi_complete(napi); | |
1479 | if (ITR_IS_DYNAMIC(vsi->rx_itr_setting) || | |
1480 | ITR_IS_DYNAMIC(vsi->tx_itr_setting)) | |
1481 | i40e_update_dynamic_itr(q_vector); | |
1482 | ||
1483 | if (!test_bit(__I40E_DOWN, &vsi->state)) { | |
1484 | if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) { | |
1485 | i40e_irq_dynamic_enable(vsi, | |
1486 | q_vector->v_idx + vsi->base_vector); | |
1487 | } else { | |
1488 | struct i40e_hw *hw = &vsi->back->hw; | |
1489 | /* We re-enable the queue 0 cause, but | |
1490 | * don't worry about dynamic_enable | |
1491 | * because we left it on for the other | |
1492 | * possible interrupts during napi | |
1493 | */ | |
1494 | u32 qval = rd32(hw, I40E_QINT_RQCTL(0)); | |
1495 | qval |= I40E_QINT_RQCTL_CAUSE_ENA_MASK; | |
1496 | wr32(hw, I40E_QINT_RQCTL(0), qval); | |
1497 | ||
1498 | qval = rd32(hw, I40E_QINT_TQCTL(0)); | |
1499 | qval |= I40E_QINT_TQCTL_CAUSE_ENA_MASK; | |
1500 | wr32(hw, I40E_QINT_TQCTL(0), qval); | |
116a57d4 SN |
1501 | |
1502 | i40e_irq_dynamic_enable_icr0(vsi->back); | |
fd0a05ce JB |
1503 | } |
1504 | } | |
1505 | ||
1506 | return 0; | |
1507 | } | |
1508 | ||
1509 | /** | |
1510 | * i40e_atr - Add a Flow Director ATR filter | |
1511 | * @tx_ring: ring to add programming descriptor to | |
1512 | * @skb: send buffer | |
1513 | * @flags: send flags | |
1514 | * @protocol: wire protocol | |
1515 | **/ | |
1516 | static void i40e_atr(struct i40e_ring *tx_ring, struct sk_buff *skb, | |
1517 | u32 flags, __be16 protocol) | |
1518 | { | |
1519 | struct i40e_filter_program_desc *fdir_desc; | |
1520 | struct i40e_pf *pf = tx_ring->vsi->back; | |
1521 | union { | |
1522 | unsigned char *network; | |
1523 | struct iphdr *ipv4; | |
1524 | struct ipv6hdr *ipv6; | |
1525 | } hdr; | |
1526 | struct tcphdr *th; | |
1527 | unsigned int hlen; | |
1528 | u32 flex_ptype, dtype_cmd; | |
fc4ac67b | 1529 | u16 i; |
fd0a05ce JB |
1530 | |
1531 | /* make sure ATR is enabled */ | |
60ea5f83 | 1532 | if (!(pf->flags & I40E_FLAG_FD_ATR_ENABLED)) |
fd0a05ce JB |
1533 | return; |
1534 | ||
1535 | /* if sampling is disabled do nothing */ | |
1536 | if (!tx_ring->atr_sample_rate) | |
1537 | return; | |
1538 | ||
1539 | tx_ring->atr_count++; | |
1540 | ||
1541 | /* snag network header to get L4 type and address */ | |
1542 | hdr.network = skb_network_header(skb); | |
1543 | ||
1544 | /* Currently only IPv4/IPv6 with TCP is supported */ | |
1545 | if (protocol == htons(ETH_P_IP)) { | |
1546 | if (hdr.ipv4->protocol != IPPROTO_TCP) | |
1547 | return; | |
1548 | ||
1549 | /* access ihl as a u8 to avoid unaligned access on ia64 */ | |
1550 | hlen = (hdr.network[0] & 0x0F) << 2; | |
1551 | } else if (protocol == htons(ETH_P_IPV6)) { | |
1552 | if (hdr.ipv6->nexthdr != IPPROTO_TCP) | |
1553 | return; | |
1554 | ||
1555 | hlen = sizeof(struct ipv6hdr); | |
1556 | } else { | |
1557 | return; | |
1558 | } | |
1559 | ||
1560 | th = (struct tcphdr *)(hdr.network + hlen); | |
1561 | ||
1562 | /* sample on all syn/fin packets or once every atr sample rate */ | |
1563 | if (!th->fin && !th->syn && (tx_ring->atr_count < tx_ring->atr_sample_rate)) | |
1564 | return; | |
1565 | ||
1566 | tx_ring->atr_count = 0; | |
1567 | ||
1568 | /* grab the next descriptor */ | |
fc4ac67b AD |
1569 | i = tx_ring->next_to_use; |
1570 | fdir_desc = I40E_TX_FDIRDESC(tx_ring, i); | |
1571 | ||
1572 | i++; | |
1573 | tx_ring->next_to_use = (i < tx_ring->count) ? i : 0; | |
fd0a05ce JB |
1574 | |
1575 | flex_ptype = (tx_ring->queue_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) & | |
1576 | I40E_TXD_FLTR_QW0_QINDEX_MASK; | |
1577 | flex_ptype |= (protocol == htons(ETH_P_IP)) ? | |
1578 | (I40E_FILTER_PCTYPE_NONF_IPV4_TCP << | |
1579 | I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) : | |
1580 | (I40E_FILTER_PCTYPE_NONF_IPV6_TCP << | |
1581 | I40E_TXD_FLTR_QW0_PCTYPE_SHIFT); | |
1582 | ||
1583 | flex_ptype |= tx_ring->vsi->id << I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT; | |
1584 | ||
1585 | dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG; | |
1586 | ||
1587 | dtype_cmd |= th->fin ? | |
1588 | (I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE << | |
1589 | I40E_TXD_FLTR_QW1_PCMD_SHIFT) : | |
1590 | (I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE << | |
1591 | I40E_TXD_FLTR_QW1_PCMD_SHIFT); | |
1592 | ||
1593 | dtype_cmd |= I40E_FILTER_PROGRAM_DESC_DEST_DIRECT_PACKET_QINDEX << | |
1594 | I40E_TXD_FLTR_QW1_DEST_SHIFT; | |
1595 | ||
1596 | dtype_cmd |= I40E_FILTER_PROGRAM_DESC_FD_STATUS_FD_ID << | |
1597 | I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT; | |
1598 | ||
1599 | fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype); | |
1600 | fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd); | |
1601 | } | |
1602 | ||
fd0a05ce JB |
1603 | /** |
1604 | * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW | |
1605 | * @skb: send buffer | |
1606 | * @tx_ring: ring to send buffer on | |
1607 | * @flags: the tx flags to be set | |
1608 | * | |
1609 | * Checks the skb and set up correspondingly several generic transmit flags | |
1610 | * related to VLAN tagging for the HW, such as VLAN, DCB, etc. | |
1611 | * | |
1612 | * Returns error code indicate the frame should be dropped upon error and the | |
1613 | * otherwise returns 0 to indicate the flags has been set properly. | |
1614 | **/ | |
1615 | static int i40e_tx_prepare_vlan_flags(struct sk_buff *skb, | |
1616 | struct i40e_ring *tx_ring, | |
1617 | u32 *flags) | |
1618 | { | |
1619 | __be16 protocol = skb->protocol; | |
1620 | u32 tx_flags = 0; | |
1621 | ||
1622 | /* if we have a HW VLAN tag being added, default to the HW one */ | |
1623 | if (vlan_tx_tag_present(skb)) { | |
1624 | tx_flags |= vlan_tx_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT; | |
1625 | tx_flags |= I40E_TX_FLAGS_HW_VLAN; | |
1626 | /* else if it is a SW VLAN, check the next protocol and store the tag */ | |
0e2fe46c | 1627 | } else if (protocol == htons(ETH_P_8021Q)) { |
fd0a05ce JB |
1628 | struct vlan_hdr *vhdr, _vhdr; |
1629 | vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr); | |
1630 | if (!vhdr) | |
1631 | return -EINVAL; | |
1632 | ||
1633 | protocol = vhdr->h_vlan_encapsulated_proto; | |
1634 | tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT; | |
1635 | tx_flags |= I40E_TX_FLAGS_SW_VLAN; | |
1636 | } | |
1637 | ||
1638 | /* Insert 802.1p priority into VLAN header */ | |
1639 | if ((tx_ring->vsi->back->flags & I40E_FLAG_DCB_ENABLED) && | |
1640 | ((tx_flags & (I40E_TX_FLAGS_HW_VLAN | I40E_TX_FLAGS_SW_VLAN)) || | |
1641 | (skb->priority != TC_PRIO_CONTROL))) { | |
1642 | tx_flags &= ~I40E_TX_FLAGS_VLAN_PRIO_MASK; | |
1643 | tx_flags |= (skb->priority & 0x7) << | |
1644 | I40E_TX_FLAGS_VLAN_PRIO_SHIFT; | |
1645 | if (tx_flags & I40E_TX_FLAGS_SW_VLAN) { | |
1646 | struct vlan_ethhdr *vhdr; | |
1647 | if (skb_header_cloned(skb) && | |
1648 | pskb_expand_head(skb, 0, 0, GFP_ATOMIC)) | |
1649 | return -ENOMEM; | |
1650 | vhdr = (struct vlan_ethhdr *)skb->data; | |
1651 | vhdr->h_vlan_TCI = htons(tx_flags >> | |
1652 | I40E_TX_FLAGS_VLAN_SHIFT); | |
1653 | } else { | |
1654 | tx_flags |= I40E_TX_FLAGS_HW_VLAN; | |
1655 | } | |
1656 | } | |
1657 | *flags = tx_flags; | |
1658 | return 0; | |
1659 | } | |
1660 | ||
fd0a05ce JB |
1661 | /** |
1662 | * i40e_tso - set up the tso context descriptor | |
1663 | * @tx_ring: ptr to the ring to send | |
1664 | * @skb: ptr to the skb we're sending | |
1665 | * @tx_flags: the collected send information | |
1666 | * @protocol: the send protocol | |
1667 | * @hdr_len: ptr to the size of the packet header | |
1668 | * @cd_tunneling: ptr to context descriptor bits | |
1669 | * | |
1670 | * Returns 0 if no TSO can happen, 1 if tso is going, or error | |
1671 | **/ | |
1672 | static int i40e_tso(struct i40e_ring *tx_ring, struct sk_buff *skb, | |
1673 | u32 tx_flags, __be16 protocol, u8 *hdr_len, | |
1674 | u64 *cd_type_cmd_tso_mss, u32 *cd_tunneling) | |
1675 | { | |
1676 | u32 cd_cmd, cd_tso_len, cd_mss; | |
1677 | struct tcphdr *tcph; | |
1678 | struct iphdr *iph; | |
1679 | u32 l4len; | |
1680 | int err; | |
1681 | struct ipv6hdr *ipv6h; | |
1682 | ||
1683 | if (!skb_is_gso(skb)) | |
1684 | return 0; | |
1685 | ||
1686 | if (skb_header_cloned(skb)) { | |
1687 | err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); | |
1688 | if (err) | |
1689 | return err; | |
1690 | } | |
1691 | ||
0e2fe46c | 1692 | if (protocol == htons(ETH_P_IP)) { |
fd0a05ce JB |
1693 | iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb); |
1694 | tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb); | |
1695 | iph->tot_len = 0; | |
1696 | iph->check = 0; | |
1697 | tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, | |
1698 | 0, IPPROTO_TCP, 0); | |
1699 | } else if (skb_is_gso_v6(skb)) { | |
1700 | ||
1701 | ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) | |
1702 | : ipv6_hdr(skb); | |
1703 | tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb); | |
1704 | ipv6h->payload_len = 0; | |
1705 | tcph->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, | |
1706 | 0, IPPROTO_TCP, 0); | |
1707 | } | |
1708 | ||
1709 | l4len = skb->encapsulation ? inner_tcp_hdrlen(skb) : tcp_hdrlen(skb); | |
1710 | *hdr_len = (skb->encapsulation | |
1711 | ? (skb_inner_transport_header(skb) - skb->data) | |
1712 | : skb_transport_offset(skb)) + l4len; | |
1713 | ||
1714 | /* find the field values */ | |
1715 | cd_cmd = I40E_TX_CTX_DESC_TSO; | |
1716 | cd_tso_len = skb->len - *hdr_len; | |
1717 | cd_mss = skb_shinfo(skb)->gso_size; | |
829af3ac MW |
1718 | *cd_type_cmd_tso_mss |= ((u64)cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) | |
1719 | ((u64)cd_tso_len << | |
1720 | I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) | | |
1721 | ((u64)cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT); | |
fd0a05ce JB |
1722 | return 1; |
1723 | } | |
1724 | ||
beb0dff1 JK |
1725 | /** |
1726 | * i40e_tsyn - set up the tsyn context descriptor | |
1727 | * @tx_ring: ptr to the ring to send | |
1728 | * @skb: ptr to the skb we're sending | |
1729 | * @tx_flags: the collected send information | |
1730 | * | |
1731 | * Returns 0 if no Tx timestamp can happen and 1 if the timestamp will happen | |
1732 | **/ | |
1733 | static int i40e_tsyn(struct i40e_ring *tx_ring, struct sk_buff *skb, | |
1734 | u32 tx_flags, u64 *cd_type_cmd_tso_mss) | |
1735 | { | |
1736 | struct i40e_pf *pf; | |
1737 | ||
1738 | if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP))) | |
1739 | return 0; | |
1740 | ||
1741 | /* Tx timestamps cannot be sampled when doing TSO */ | |
1742 | if (tx_flags & I40E_TX_FLAGS_TSO) | |
1743 | return 0; | |
1744 | ||
1745 | /* only timestamp the outbound packet if the user has requested it and | |
1746 | * we are not already transmitting a packet to be timestamped | |
1747 | */ | |
1748 | pf = i40e_netdev_to_pf(tx_ring->netdev); | |
1749 | if (pf->ptp_tx && !pf->ptp_tx_skb) { | |
1750 | skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS; | |
1751 | pf->ptp_tx_skb = skb_get(skb); | |
1752 | } else { | |
1753 | return 0; | |
1754 | } | |
1755 | ||
1756 | *cd_type_cmd_tso_mss |= (u64)I40E_TX_CTX_DESC_TSYN << | |
1757 | I40E_TXD_CTX_QW1_CMD_SHIFT; | |
1758 | ||
1759 | pf->ptp_tx_start = jiffies; | |
1760 | schedule_work(&pf->ptp_tx_work); | |
1761 | ||
1762 | return 1; | |
1763 | } | |
1764 | ||
fd0a05ce JB |
1765 | /** |
1766 | * i40e_tx_enable_csum - Enable Tx checksum offloads | |
1767 | * @skb: send buffer | |
1768 | * @tx_flags: Tx flags currently set | |
1769 | * @td_cmd: Tx descriptor command bits to set | |
1770 | * @td_offset: Tx descriptor header offsets to set | |
1771 | * @cd_tunneling: ptr to context desc bits | |
1772 | **/ | |
1773 | static void i40e_tx_enable_csum(struct sk_buff *skb, u32 tx_flags, | |
1774 | u32 *td_cmd, u32 *td_offset, | |
1775 | struct i40e_ring *tx_ring, | |
1776 | u32 *cd_tunneling) | |
1777 | { | |
1778 | struct ipv6hdr *this_ipv6_hdr; | |
1779 | unsigned int this_tcp_hdrlen; | |
1780 | struct iphdr *this_ip_hdr; | |
1781 | u32 network_hdr_len; | |
1782 | u8 l4_hdr = 0; | |
1783 | ||
1784 | if (skb->encapsulation) { | |
1785 | network_hdr_len = skb_inner_network_header_len(skb); | |
1786 | this_ip_hdr = inner_ip_hdr(skb); | |
1787 | this_ipv6_hdr = inner_ipv6_hdr(skb); | |
1788 | this_tcp_hdrlen = inner_tcp_hdrlen(skb); | |
1789 | ||
1790 | if (tx_flags & I40E_TX_FLAGS_IPV4) { | |
1791 | ||
1792 | if (tx_flags & I40E_TX_FLAGS_TSO) { | |
1793 | *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4; | |
1794 | ip_hdr(skb)->check = 0; | |
1795 | } else { | |
1796 | *cd_tunneling |= | |
1797 | I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM; | |
1798 | } | |
1799 | } else if (tx_flags & I40E_TX_FLAGS_IPV6) { | |
1800 | if (tx_flags & I40E_TX_FLAGS_TSO) { | |
1801 | *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV6; | |
1802 | ip_hdr(skb)->check = 0; | |
1803 | } else { | |
1804 | *cd_tunneling |= | |
1805 | I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM; | |
1806 | } | |
1807 | } | |
1808 | ||
1809 | /* Now set the ctx descriptor fields */ | |
1810 | *cd_tunneling |= (skb_network_header_len(skb) >> 2) << | |
1811 | I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT | | |
1812 | I40E_TXD_CTX_UDP_TUNNELING | | |
1813 | ((skb_inner_network_offset(skb) - | |
1814 | skb_transport_offset(skb)) >> 1) << | |
1815 | I40E_TXD_CTX_QW0_NATLEN_SHIFT; | |
1816 | ||
1817 | } else { | |
1818 | network_hdr_len = skb_network_header_len(skb); | |
1819 | this_ip_hdr = ip_hdr(skb); | |
1820 | this_ipv6_hdr = ipv6_hdr(skb); | |
1821 | this_tcp_hdrlen = tcp_hdrlen(skb); | |
1822 | } | |
1823 | ||
1824 | /* Enable IP checksum offloads */ | |
1825 | if (tx_flags & I40E_TX_FLAGS_IPV4) { | |
1826 | l4_hdr = this_ip_hdr->protocol; | |
1827 | /* the stack computes the IP header already, the only time we | |
1828 | * need the hardware to recompute it is in the case of TSO. | |
1829 | */ | |
1830 | if (tx_flags & I40E_TX_FLAGS_TSO) { | |
1831 | *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM; | |
1832 | this_ip_hdr->check = 0; | |
1833 | } else { | |
1834 | *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4; | |
1835 | } | |
1836 | /* Now set the td_offset for IP header length */ | |
1837 | *td_offset = (network_hdr_len >> 2) << | |
1838 | I40E_TX_DESC_LENGTH_IPLEN_SHIFT; | |
1839 | } else if (tx_flags & I40E_TX_FLAGS_IPV6) { | |
1840 | l4_hdr = this_ipv6_hdr->nexthdr; | |
1841 | *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6; | |
1842 | /* Now set the td_offset for IP header length */ | |
1843 | *td_offset = (network_hdr_len >> 2) << | |
1844 | I40E_TX_DESC_LENGTH_IPLEN_SHIFT; | |
1845 | } | |
1846 | /* words in MACLEN + dwords in IPLEN + dwords in L4Len */ | |
1847 | *td_offset |= (skb_network_offset(skb) >> 1) << | |
1848 | I40E_TX_DESC_LENGTH_MACLEN_SHIFT; | |
1849 | ||
1850 | /* Enable L4 checksum offloads */ | |
1851 | switch (l4_hdr) { | |
1852 | case IPPROTO_TCP: | |
1853 | /* enable checksum offloads */ | |
1854 | *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP; | |
1855 | *td_offset |= (this_tcp_hdrlen >> 2) << | |
1856 | I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT; | |
1857 | break; | |
1858 | case IPPROTO_SCTP: | |
1859 | /* enable SCTP checksum offload */ | |
1860 | *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP; | |
1861 | *td_offset |= (sizeof(struct sctphdr) >> 2) << | |
1862 | I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT; | |
1863 | break; | |
1864 | case IPPROTO_UDP: | |
1865 | /* enable UDP checksum offload */ | |
1866 | *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP; | |
1867 | *td_offset |= (sizeof(struct udphdr) >> 2) << | |
1868 | I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT; | |
1869 | break; | |
1870 | default: | |
1871 | break; | |
1872 | } | |
1873 | } | |
1874 | ||
1875 | /** | |
1876 | * i40e_create_tx_ctx Build the Tx context descriptor | |
1877 | * @tx_ring: ring to create the descriptor on | |
1878 | * @cd_type_cmd_tso_mss: Quad Word 1 | |
1879 | * @cd_tunneling: Quad Word 0 - bits 0-31 | |
1880 | * @cd_l2tag2: Quad Word 0 - bits 32-63 | |
1881 | **/ | |
1882 | static void i40e_create_tx_ctx(struct i40e_ring *tx_ring, | |
1883 | const u64 cd_type_cmd_tso_mss, | |
1884 | const u32 cd_tunneling, const u32 cd_l2tag2) | |
1885 | { | |
1886 | struct i40e_tx_context_desc *context_desc; | |
fc4ac67b | 1887 | int i = tx_ring->next_to_use; |
fd0a05ce JB |
1888 | |
1889 | if (!cd_type_cmd_tso_mss && !cd_tunneling && !cd_l2tag2) | |
1890 | return; | |
1891 | ||
1892 | /* grab the next descriptor */ | |
fc4ac67b AD |
1893 | context_desc = I40E_TX_CTXTDESC(tx_ring, i); |
1894 | ||
1895 | i++; | |
1896 | tx_ring->next_to_use = (i < tx_ring->count) ? i : 0; | |
fd0a05ce JB |
1897 | |
1898 | /* cpu_to_le32 and assign to struct fields */ | |
1899 | context_desc->tunneling_params = cpu_to_le32(cd_tunneling); | |
1900 | context_desc->l2tag2 = cpu_to_le16(cd_l2tag2); | |
1901 | context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss); | |
1902 | } | |
1903 | ||
1904 | /** | |
1905 | * i40e_tx_map - Build the Tx descriptor | |
1906 | * @tx_ring: ring to send buffer on | |
1907 | * @skb: send buffer | |
1908 | * @first: first buffer info buffer to use | |
1909 | * @tx_flags: collected send information | |
1910 | * @hdr_len: size of the packet header | |
1911 | * @td_cmd: the command field in the descriptor | |
1912 | * @td_offset: offset for checksum or crc | |
1913 | **/ | |
1914 | static void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb, | |
1915 | struct i40e_tx_buffer *first, u32 tx_flags, | |
1916 | const u8 hdr_len, u32 td_cmd, u32 td_offset) | |
1917 | { | |
fd0a05ce JB |
1918 | unsigned int data_len = skb->data_len; |
1919 | unsigned int size = skb_headlen(skb); | |
a5e9c572 | 1920 | struct skb_frag_struct *frag; |
fd0a05ce JB |
1921 | struct i40e_tx_buffer *tx_bi; |
1922 | struct i40e_tx_desc *tx_desc; | |
a5e9c572 | 1923 | u16 i = tx_ring->next_to_use; |
fd0a05ce JB |
1924 | u32 td_tag = 0; |
1925 | dma_addr_t dma; | |
1926 | u16 gso_segs; | |
1927 | ||
fd0a05ce JB |
1928 | if (tx_flags & I40E_TX_FLAGS_HW_VLAN) { |
1929 | td_cmd |= I40E_TX_DESC_CMD_IL2TAG1; | |
1930 | td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >> | |
1931 | I40E_TX_FLAGS_VLAN_SHIFT; | |
1932 | } | |
1933 | ||
a5e9c572 AD |
1934 | if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO)) |
1935 | gso_segs = skb_shinfo(skb)->gso_segs; | |
1936 | else | |
1937 | gso_segs = 1; | |
1938 | ||
1939 | /* multiply data chunks by size of headers */ | |
1940 | first->bytecount = skb->len - hdr_len + (gso_segs * hdr_len); | |
1941 | first->gso_segs = gso_segs; | |
1942 | first->skb = skb; | |
1943 | first->tx_flags = tx_flags; | |
1944 | ||
1945 | dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE); | |
1946 | ||
fd0a05ce | 1947 | tx_desc = I40E_TX_DESC(tx_ring, i); |
a5e9c572 AD |
1948 | tx_bi = first; |
1949 | ||
1950 | for (frag = &skb_shinfo(skb)->frags[0];; frag++) { | |
1951 | if (dma_mapping_error(tx_ring->dev, dma)) | |
1952 | goto dma_error; | |
1953 | ||
1954 | /* record length, and DMA address */ | |
1955 | dma_unmap_len_set(tx_bi, len, size); | |
1956 | dma_unmap_addr_set(tx_bi, dma, dma); | |
1957 | ||
1958 | tx_desc->buffer_addr = cpu_to_le64(dma); | |
1959 | ||
1960 | while (unlikely(size > I40E_MAX_DATA_PER_TXD)) { | |
fd0a05ce JB |
1961 | tx_desc->cmd_type_offset_bsz = |
1962 | build_ctob(td_cmd, td_offset, | |
1963 | I40E_MAX_DATA_PER_TXD, td_tag); | |
1964 | ||
fd0a05ce JB |
1965 | tx_desc++; |
1966 | i++; | |
1967 | if (i == tx_ring->count) { | |
1968 | tx_desc = I40E_TX_DESC(tx_ring, 0); | |
1969 | i = 0; | |
1970 | } | |
fd0a05ce | 1971 | |
a5e9c572 AD |
1972 | dma += I40E_MAX_DATA_PER_TXD; |
1973 | size -= I40E_MAX_DATA_PER_TXD; | |
fd0a05ce | 1974 | |
a5e9c572 AD |
1975 | tx_desc->buffer_addr = cpu_to_le64(dma); |
1976 | } | |
fd0a05ce JB |
1977 | |
1978 | if (likely(!data_len)) | |
1979 | break; | |
1980 | ||
a5e9c572 AD |
1981 | tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset, |
1982 | size, td_tag); | |
fd0a05ce JB |
1983 | |
1984 | tx_desc++; | |
1985 | i++; | |
1986 | if (i == tx_ring->count) { | |
1987 | tx_desc = I40E_TX_DESC(tx_ring, 0); | |
1988 | i = 0; | |
1989 | } | |
1990 | ||
a5e9c572 AD |
1991 | size = skb_frag_size(frag); |
1992 | data_len -= size; | |
fd0a05ce | 1993 | |
a5e9c572 AD |
1994 | dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size, |
1995 | DMA_TO_DEVICE); | |
fd0a05ce | 1996 | |
a5e9c572 AD |
1997 | tx_bi = &tx_ring->tx_bi[i]; |
1998 | } | |
fd0a05ce | 1999 | |
a5e9c572 AD |
2000 | tx_desc->cmd_type_offset_bsz = |
2001 | build_ctob(td_cmd, td_offset, size, td_tag) | | |
2002 | cpu_to_le64((u64)I40E_TXD_CMD << I40E_TXD_QW1_CMD_SHIFT); | |
fd0a05ce | 2003 | |
7070ce0a AD |
2004 | netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev, |
2005 | tx_ring->queue_index), | |
2006 | first->bytecount); | |
2007 | ||
a5e9c572 | 2008 | /* set the timestamp */ |
fd0a05ce | 2009 | first->time_stamp = jiffies; |
fd0a05ce JB |
2010 | |
2011 | /* Force memory writes to complete before letting h/w | |
2012 | * know there are new descriptors to fetch. (Only | |
2013 | * applicable for weak-ordered memory model archs, | |
2014 | * such as IA-64). | |
2015 | */ | |
2016 | wmb(); | |
2017 | ||
a5e9c572 AD |
2018 | /* set next_to_watch value indicating a packet is present */ |
2019 | first->next_to_watch = tx_desc; | |
2020 | ||
2021 | i++; | |
2022 | if (i == tx_ring->count) | |
2023 | i = 0; | |
2024 | ||
2025 | tx_ring->next_to_use = i; | |
2026 | ||
2027 | /* notify HW of packet */ | |
fd0a05ce | 2028 | writel(i, tx_ring->tail); |
a5e9c572 | 2029 | |
fd0a05ce JB |
2030 | return; |
2031 | ||
2032 | dma_error: | |
a5e9c572 | 2033 | dev_info(tx_ring->dev, "TX DMA map failed\n"); |
fd0a05ce JB |
2034 | |
2035 | /* clear dma mappings for failed tx_bi map */ | |
2036 | for (;;) { | |
2037 | tx_bi = &tx_ring->tx_bi[i]; | |
a5e9c572 | 2038 | i40e_unmap_and_free_tx_resource(tx_ring, tx_bi); |
fd0a05ce JB |
2039 | if (tx_bi == first) |
2040 | break; | |
2041 | if (i == 0) | |
2042 | i = tx_ring->count; | |
2043 | i--; | |
2044 | } | |
2045 | ||
fd0a05ce JB |
2046 | tx_ring->next_to_use = i; |
2047 | } | |
2048 | ||
2049 | /** | |
2050 | * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions | |
2051 | * @tx_ring: the ring to be checked | |
2052 | * @size: the size buffer we want to assure is available | |
2053 | * | |
2054 | * Returns -EBUSY if a stop is needed, else 0 | |
2055 | **/ | |
2056 | static inline int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size) | |
2057 | { | |
2058 | netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index); | |
8e9dca53 | 2059 | /* Memory barrier before checking head and tail */ |
fd0a05ce JB |
2060 | smp_mb(); |
2061 | ||
2062 | /* Check again in a case another CPU has just made room available. */ | |
2063 | if (likely(I40E_DESC_UNUSED(tx_ring) < size)) | |
2064 | return -EBUSY; | |
2065 | ||
2066 | /* A reprieve! - use start_queue because it doesn't call schedule */ | |
2067 | netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index); | |
2068 | ++tx_ring->tx_stats.restart_queue; | |
2069 | return 0; | |
2070 | } | |
2071 | ||
2072 | /** | |
2073 | * i40e_maybe_stop_tx - 1st level check for tx stop conditions | |
2074 | * @tx_ring: the ring to be checked | |
2075 | * @size: the size buffer we want to assure is available | |
2076 | * | |
2077 | * Returns 0 if stop is not needed | |
2078 | **/ | |
2079 | static int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size) | |
2080 | { | |
2081 | if (likely(I40E_DESC_UNUSED(tx_ring) >= size)) | |
2082 | return 0; | |
2083 | return __i40e_maybe_stop_tx(tx_ring, size); | |
2084 | } | |
2085 | ||
2086 | /** | |
2087 | * i40e_xmit_descriptor_count - calculate number of tx descriptors needed | |
2088 | * @skb: send buffer | |
2089 | * @tx_ring: ring to send buffer on | |
2090 | * | |
2091 | * Returns number of data descriptors needed for this skb. Returns 0 to indicate | |
2092 | * there is not enough descriptors available in this ring since we need at least | |
2093 | * one descriptor. | |
2094 | **/ | |
2095 | static int i40e_xmit_descriptor_count(struct sk_buff *skb, | |
2096 | struct i40e_ring *tx_ring) | |
2097 | { | |
2098 | #if PAGE_SIZE > I40E_MAX_DATA_PER_TXD | |
2099 | unsigned int f; | |
2100 | #endif | |
2101 | int count = 0; | |
2102 | ||
2103 | /* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD, | |
2104 | * + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD, | |
2105 | * + 2 desc gap to keep tail from touching head, | |
2106 | * + 1 desc for context descriptor, | |
2107 | * otherwise try next time | |
2108 | */ | |
2109 | #if PAGE_SIZE > I40E_MAX_DATA_PER_TXD | |
2110 | for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) | |
2111 | count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size); | |
2112 | #else | |
2113 | count += skb_shinfo(skb)->nr_frags; | |
2114 | #endif | |
2115 | count += TXD_USE_COUNT(skb_headlen(skb)); | |
2116 | if (i40e_maybe_stop_tx(tx_ring, count + 3)) { | |
2117 | tx_ring->tx_stats.tx_busy++; | |
2118 | return 0; | |
2119 | } | |
2120 | return count; | |
2121 | } | |
2122 | ||
2123 | /** | |
2124 | * i40e_xmit_frame_ring - Sends buffer on Tx ring | |
2125 | * @skb: send buffer | |
2126 | * @tx_ring: ring to send buffer on | |
2127 | * | |
2128 | * Returns NETDEV_TX_OK if sent, else an error code | |
2129 | **/ | |
2130 | static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb, | |
2131 | struct i40e_ring *tx_ring) | |
2132 | { | |
2133 | u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT; | |
2134 | u32 cd_tunneling = 0, cd_l2tag2 = 0; | |
2135 | struct i40e_tx_buffer *first; | |
2136 | u32 td_offset = 0; | |
2137 | u32 tx_flags = 0; | |
2138 | __be16 protocol; | |
2139 | u32 td_cmd = 0; | |
2140 | u8 hdr_len = 0; | |
beb0dff1 | 2141 | int tsyn; |
fd0a05ce JB |
2142 | int tso; |
2143 | if (0 == i40e_xmit_descriptor_count(skb, tx_ring)) | |
2144 | return NETDEV_TX_BUSY; | |
2145 | ||
2146 | /* prepare the xmit flags */ | |
2147 | if (i40e_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags)) | |
2148 | goto out_drop; | |
2149 | ||
2150 | /* obtain protocol of skb */ | |
2151 | protocol = skb->protocol; | |
2152 | ||
2153 | /* record the location of the first descriptor for this packet */ | |
2154 | first = &tx_ring->tx_bi[tx_ring->next_to_use]; | |
2155 | ||
2156 | /* setup IPv4/IPv6 offloads */ | |
0e2fe46c | 2157 | if (protocol == htons(ETH_P_IP)) |
fd0a05ce | 2158 | tx_flags |= I40E_TX_FLAGS_IPV4; |
0e2fe46c | 2159 | else if (protocol == htons(ETH_P_IPV6)) |
fd0a05ce JB |
2160 | tx_flags |= I40E_TX_FLAGS_IPV6; |
2161 | ||
2162 | tso = i40e_tso(tx_ring, skb, tx_flags, protocol, &hdr_len, | |
2163 | &cd_type_cmd_tso_mss, &cd_tunneling); | |
2164 | ||
2165 | if (tso < 0) | |
2166 | goto out_drop; | |
2167 | else if (tso) | |
2168 | tx_flags |= I40E_TX_FLAGS_TSO; | |
2169 | ||
2170 | skb_tx_timestamp(skb); | |
2171 | ||
beb0dff1 JK |
2172 | tsyn = i40e_tsyn(tx_ring, skb, tx_flags, &cd_type_cmd_tso_mss); |
2173 | ||
2174 | if (tsyn) | |
2175 | tx_flags |= I40E_TX_FLAGS_TSYN; | |
2176 | ||
b1941306 AD |
2177 | /* always enable CRC insertion offload */ |
2178 | td_cmd |= I40E_TX_DESC_CMD_ICRC; | |
2179 | ||
fd0a05ce | 2180 | /* Always offload the checksum, since it's in the data descriptor */ |
b1941306 | 2181 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
fd0a05ce JB |
2182 | tx_flags |= I40E_TX_FLAGS_CSUM; |
2183 | ||
fd0a05ce JB |
2184 | i40e_tx_enable_csum(skb, tx_flags, &td_cmd, &td_offset, |
2185 | tx_ring, &cd_tunneling); | |
b1941306 | 2186 | } |
fd0a05ce JB |
2187 | |
2188 | i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss, | |
2189 | cd_tunneling, cd_l2tag2); | |
2190 | ||
2191 | /* Add Flow Director ATR if it's enabled. | |
2192 | * | |
2193 | * NOTE: this must always be directly before the data descriptor. | |
2194 | */ | |
2195 | i40e_atr(tx_ring, skb, tx_flags, protocol); | |
2196 | ||
2197 | i40e_tx_map(tx_ring, skb, first, tx_flags, hdr_len, | |
2198 | td_cmd, td_offset); | |
2199 | ||
2200 | i40e_maybe_stop_tx(tx_ring, DESC_NEEDED); | |
2201 | ||
2202 | return NETDEV_TX_OK; | |
2203 | ||
2204 | out_drop: | |
2205 | dev_kfree_skb_any(skb); | |
2206 | return NETDEV_TX_OK; | |
2207 | } | |
2208 | ||
2209 | /** | |
2210 | * i40e_lan_xmit_frame - Selects the correct VSI and Tx queue to send buffer | |
2211 | * @skb: send buffer | |
2212 | * @netdev: network interface device structure | |
2213 | * | |
2214 | * Returns NETDEV_TX_OK if sent, else an error code | |
2215 | **/ | |
2216 | netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev) | |
2217 | { | |
2218 | struct i40e_netdev_priv *np = netdev_priv(netdev); | |
2219 | struct i40e_vsi *vsi = np->vsi; | |
9f65e15b | 2220 | struct i40e_ring *tx_ring = vsi->tx_rings[skb->queue_mapping]; |
fd0a05ce JB |
2221 | |
2222 | /* hardware can't handle really short frames, hardware padding works | |
2223 | * beyond this point | |
2224 | */ | |
2225 | if (unlikely(skb->len < I40E_MIN_TX_LEN)) { | |
2226 | if (skb_pad(skb, I40E_MIN_TX_LEN - skb->len)) | |
2227 | return NETDEV_TX_OK; | |
2228 | skb->len = I40E_MIN_TX_LEN; | |
2229 | skb_set_tail_pointer(skb, I40E_MIN_TX_LEN); | |
2230 | } | |
2231 | ||
2232 | return i40e_xmit_frame_ring(skb, tx_ring); | |
2233 | } |