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