]>
Commit | Line | Data |
---|---|---|
fd0a05ce JB |
1 | /******************************************************************************* |
2 | * | |
3 | * Intel Ethernet Controller XL710 Family Linux Driver | |
4 | * Copyright(c) 2013 Intel Corporation. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify it | |
7 | * under the terms and conditions of the GNU General Public License, | |
8 | * version 2, as published by the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope it will be useful, but WITHOUT | |
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
13 | * more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License along with | |
16 | * this program; if not, write to the Free Software Foundation, Inc., | |
17 | * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. | |
18 | * | |
19 | * The full GNU General Public License is included in this distribution in | |
20 | * the file called "COPYING". | |
21 | * | |
22 | * Contact Information: | |
23 | * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net> | |
24 | * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497 | |
25 | * | |
26 | ******************************************************************************/ | |
27 | ||
28 | #include "i40e.h" | |
29 | ||
30 | static inline __le64 build_ctob(u32 td_cmd, u32 td_offset, unsigned int size, | |
31 | u32 td_tag) | |
32 | { | |
33 | return cpu_to_le64(I40E_TX_DESC_DTYPE_DATA | | |
34 | ((u64)td_cmd << I40E_TXD_QW1_CMD_SHIFT) | | |
35 | ((u64)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) | | |
36 | ((u64)size << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) | | |
37 | ((u64)td_tag << I40E_TXD_QW1_L2TAG1_SHIFT)); | |
38 | } | |
39 | ||
40 | /** | |
41 | * i40e_program_fdir_filter - Program a Flow Director filter | |
42 | * @fdir_input: Packet data that will be filter parameters | |
43 | * @pf: The pf pointer | |
44 | * @add: True for add/update, False for remove | |
45 | **/ | |
46 | int i40e_program_fdir_filter(struct i40e_fdir_data *fdir_data, | |
47 | struct i40e_pf *pf, bool add) | |
48 | { | |
49 | struct i40e_filter_program_desc *fdir_desc; | |
50 | struct i40e_tx_buffer *tx_buf; | |
51 | struct i40e_tx_desc *tx_desc; | |
52 | struct i40e_ring *tx_ring; | |
53 | struct i40e_vsi *vsi; | |
54 | struct device *dev; | |
55 | dma_addr_t dma; | |
56 | u32 td_cmd = 0; | |
57 | u16 i; | |
58 | ||
59 | /* find existing FDIR VSI */ | |
60 | vsi = NULL; | |
61 | for (i = 0; i < pf->hw.func_caps.num_vsis; i++) | |
62 | if (pf->vsi[i] && pf->vsi[i]->type == I40E_VSI_FDIR) | |
63 | vsi = pf->vsi[i]; | |
64 | if (!vsi) | |
65 | return -ENOENT; | |
66 | ||
9f65e15b | 67 | tx_ring = vsi->tx_rings[0]; |
fd0a05ce JB |
68 | dev = tx_ring->dev; |
69 | ||
70 | dma = dma_map_single(dev, fdir_data->raw_packet, | |
71 | I40E_FDIR_MAX_RAW_PACKET_LOOKUP, DMA_TO_DEVICE); | |
72 | if (dma_mapping_error(dev, dma)) | |
73 | goto dma_fail; | |
74 | ||
75 | /* grab the next descriptor */ | |
fc4ac67b AD |
76 | i = tx_ring->next_to_use; |
77 | fdir_desc = I40E_TX_FDIRDESC(tx_ring, i); | |
78 | tx_buf = &tx_ring->tx_bi[i]; | |
79 | ||
80 | i++; | |
81 | tx_ring->next_to_use = (i < tx_ring->count) ? i : 0; | |
fd0a05ce JB |
82 | |
83 | fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32((fdir_data->q_index | |
84 | << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) | |
85 | & I40E_TXD_FLTR_QW0_QINDEX_MASK); | |
86 | ||
87 | fdir_desc->qindex_flex_ptype_vsi |= cpu_to_le32((fdir_data->flex_off | |
88 | << I40E_TXD_FLTR_QW0_FLEXOFF_SHIFT) | |
89 | & I40E_TXD_FLTR_QW0_FLEXOFF_MASK); | |
90 | ||
91 | fdir_desc->qindex_flex_ptype_vsi |= cpu_to_le32((fdir_data->pctype | |
92 | << I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) | |
93 | & I40E_TXD_FLTR_QW0_PCTYPE_MASK); | |
94 | ||
95 | /* Use LAN VSI Id if not programmed by user */ | |
96 | if (fdir_data->dest_vsi == 0) | |
97 | fdir_desc->qindex_flex_ptype_vsi |= | |
98 | cpu_to_le32((pf->vsi[pf->lan_vsi]->id) | |
99 | << I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT); | |
100 | else | |
101 | fdir_desc->qindex_flex_ptype_vsi |= | |
102 | cpu_to_le32((fdir_data->dest_vsi | |
103 | << I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT) | |
104 | & I40E_TXD_FLTR_QW0_DEST_VSI_MASK); | |
105 | ||
106 | fdir_desc->dtype_cmd_cntindex = | |
107 | cpu_to_le32(I40E_TX_DESC_DTYPE_FILTER_PROG); | |
108 | ||
109 | if (add) | |
110 | fdir_desc->dtype_cmd_cntindex |= cpu_to_le32( | |
111 | I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE | |
112 | << I40E_TXD_FLTR_QW1_PCMD_SHIFT); | |
113 | else | |
114 | fdir_desc->dtype_cmd_cntindex |= cpu_to_le32( | |
115 | I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE | |
116 | << I40E_TXD_FLTR_QW1_PCMD_SHIFT); | |
117 | ||
118 | fdir_desc->dtype_cmd_cntindex |= cpu_to_le32((fdir_data->dest_ctl | |
119 | << I40E_TXD_FLTR_QW1_DEST_SHIFT) | |
120 | & I40E_TXD_FLTR_QW1_DEST_MASK); | |
121 | ||
122 | fdir_desc->dtype_cmd_cntindex |= cpu_to_le32( | |
123 | (fdir_data->fd_status << I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT) | |
124 | & I40E_TXD_FLTR_QW1_FD_STATUS_MASK); | |
125 | ||
126 | if (fdir_data->cnt_index != 0) { | |
127 | fdir_desc->dtype_cmd_cntindex |= | |
128 | cpu_to_le32(I40E_TXD_FLTR_QW1_CNT_ENA_MASK); | |
129 | fdir_desc->dtype_cmd_cntindex |= | |
130 | cpu_to_le32((fdir_data->cnt_index | |
131 | << I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) | |
132 | & I40E_TXD_FLTR_QW1_CNTINDEX_MASK); | |
133 | } | |
134 | ||
135 | fdir_desc->fd_id = cpu_to_le32(fdir_data->fd_id); | |
136 | ||
137 | /* Now program a dummy descriptor */ | |
fc4ac67b AD |
138 | i = tx_ring->next_to_use; |
139 | tx_desc = I40E_TX_DESC(tx_ring, i); | |
140 | ||
141 | i++; | |
142 | tx_ring->next_to_use = (i < tx_ring->count) ? i : 0; | |
fd0a05ce JB |
143 | |
144 | tx_desc->buffer_addr = cpu_to_le64(dma); | |
145 | td_cmd = I40E_TX_DESC_CMD_EOP | | |
146 | I40E_TX_DESC_CMD_RS | | |
147 | I40E_TX_DESC_CMD_DUMMY; | |
148 | ||
149 | tx_desc->cmd_type_offset_bsz = | |
150 | build_ctob(td_cmd, 0, I40E_FDIR_MAX_RAW_PACKET_LOOKUP, 0); | |
151 | ||
fd0a05ce JB |
152 | /* Force memory writes to complete before letting h/w |
153 | * know there are new descriptors to fetch. (Only | |
154 | * applicable for weak-ordered memory model archs, | |
155 | * such as IA-64). | |
156 | */ | |
157 | wmb(); | |
158 | ||
fc4ac67b AD |
159 | /* Mark the data descriptor to be watched */ |
160 | tx_buf->next_to_watch = tx_desc; | |
161 | ||
fd0a05ce JB |
162 | writel(tx_ring->next_to_use, tx_ring->tail); |
163 | return 0; | |
164 | ||
165 | dma_fail: | |
166 | return -1; | |
167 | } | |
168 | ||
169 | /** | |
170 | * i40e_fd_handle_status - check the Programming Status for FD | |
171 | * @rx_ring: the Rx ring for this descriptor | |
172 | * @qw: the descriptor data | |
173 | * @prog_id: the id originally used for programming | |
174 | * | |
175 | * This is used to verify if the FD programming or invalidation | |
176 | * requested by SW to the HW is successful or not and take actions accordingly. | |
177 | **/ | |
178 | static void i40e_fd_handle_status(struct i40e_ring *rx_ring, u32 qw, u8 prog_id) | |
179 | { | |
180 | struct pci_dev *pdev = rx_ring->vsi->back->pdev; | |
181 | u32 error; | |
182 | ||
183 | error = (qw & I40E_RX_PROG_STATUS_DESC_QW1_ERROR_MASK) >> | |
184 | I40E_RX_PROG_STATUS_DESC_QW1_ERROR_SHIFT; | |
185 | ||
186 | /* for now just print the Status */ | |
187 | dev_info(&pdev->dev, "FD programming id %02x, Status %08x\n", | |
188 | prog_id, error); | |
189 | } | |
190 | ||
191 | /** | |
a5e9c572 | 192 | * i40e_unmap_and_free_tx_resource - Release a Tx buffer |
fd0a05ce JB |
193 | * @ring: the ring that owns the buffer |
194 | * @tx_buffer: the buffer to free | |
195 | **/ | |
a5e9c572 AD |
196 | static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring, |
197 | struct i40e_tx_buffer *tx_buffer) | |
fd0a05ce | 198 | { |
a5e9c572 AD |
199 | if (tx_buffer->skb) { |
200 | dev_kfree_skb_any(tx_buffer->skb); | |
201 | if (dma_unmap_len(tx_buffer, len)) | |
fd0a05ce | 202 | dma_unmap_single(ring->dev, |
35a1e2ad AD |
203 | dma_unmap_addr(tx_buffer, dma), |
204 | dma_unmap_len(tx_buffer, len), | |
fd0a05ce | 205 | DMA_TO_DEVICE); |
a5e9c572 AD |
206 | } else if (dma_unmap_len(tx_buffer, len)) { |
207 | dma_unmap_page(ring->dev, | |
208 | dma_unmap_addr(tx_buffer, dma), | |
209 | dma_unmap_len(tx_buffer, len), | |
210 | DMA_TO_DEVICE); | |
fd0a05ce | 211 | } |
a5e9c572 AD |
212 | tx_buffer->next_to_watch = NULL; |
213 | tx_buffer->skb = NULL; | |
35a1e2ad | 214 | dma_unmap_len_set(tx_buffer, len, 0); |
a5e9c572 | 215 | /* tx_buffer must be completely set up in the transmit path */ |
fd0a05ce JB |
216 | } |
217 | ||
218 | /** | |
219 | * i40e_clean_tx_ring - Free any empty Tx buffers | |
220 | * @tx_ring: ring to be cleaned | |
221 | **/ | |
222 | void i40e_clean_tx_ring(struct i40e_ring *tx_ring) | |
223 | { | |
fd0a05ce JB |
224 | unsigned long bi_size; |
225 | u16 i; | |
226 | ||
227 | /* ring already cleared, nothing to do */ | |
228 | if (!tx_ring->tx_bi) | |
229 | return; | |
230 | ||
231 | /* Free all the Tx ring sk_buffs */ | |
a5e9c572 AD |
232 | for (i = 0; i < tx_ring->count; i++) |
233 | i40e_unmap_and_free_tx_resource(tx_ring, &tx_ring->tx_bi[i]); | |
fd0a05ce JB |
234 | |
235 | bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count; | |
236 | memset(tx_ring->tx_bi, 0, bi_size); | |
237 | ||
238 | /* Zero out the descriptor ring */ | |
239 | memset(tx_ring->desc, 0, tx_ring->size); | |
240 | ||
241 | tx_ring->next_to_use = 0; | |
242 | tx_ring->next_to_clean = 0; | |
7070ce0a AD |
243 | |
244 | if (!tx_ring->netdev) | |
245 | return; | |
246 | ||
247 | /* cleanup Tx queue statistics */ | |
248 | netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev, | |
249 | tx_ring->queue_index)); | |
fd0a05ce JB |
250 | } |
251 | ||
252 | /** | |
253 | * i40e_free_tx_resources - Free Tx resources per queue | |
254 | * @tx_ring: Tx descriptor ring for a specific queue | |
255 | * | |
256 | * Free all transmit software resources | |
257 | **/ | |
258 | void i40e_free_tx_resources(struct i40e_ring *tx_ring) | |
259 | { | |
260 | i40e_clean_tx_ring(tx_ring); | |
261 | kfree(tx_ring->tx_bi); | |
262 | tx_ring->tx_bi = NULL; | |
263 | ||
264 | if (tx_ring->desc) { | |
265 | dma_free_coherent(tx_ring->dev, tx_ring->size, | |
266 | tx_ring->desc, tx_ring->dma); | |
267 | tx_ring->desc = NULL; | |
268 | } | |
269 | } | |
270 | ||
271 | /** | |
272 | * i40e_get_tx_pending - how many tx descriptors not processed | |
273 | * @tx_ring: the ring of descriptors | |
274 | * | |
275 | * Since there is no access to the ring head register | |
276 | * in XL710, we need to use our local copies | |
277 | **/ | |
278 | static u32 i40e_get_tx_pending(struct i40e_ring *ring) | |
279 | { | |
280 | u32 ntu = ((ring->next_to_clean <= ring->next_to_use) | |
281 | ? ring->next_to_use | |
282 | : ring->next_to_use + ring->count); | |
283 | return ntu - ring->next_to_clean; | |
284 | } | |
285 | ||
286 | /** | |
287 | * i40e_check_tx_hang - Is there a hang in the Tx queue | |
288 | * @tx_ring: the ring of descriptors | |
289 | **/ | |
290 | static bool i40e_check_tx_hang(struct i40e_ring *tx_ring) | |
291 | { | |
292 | u32 tx_pending = i40e_get_tx_pending(tx_ring); | |
293 | bool ret = false; | |
294 | ||
295 | clear_check_for_tx_hang(tx_ring); | |
296 | ||
297 | /* Check for a hung queue, but be thorough. This verifies | |
298 | * that a transmit has been completed since the previous | |
299 | * check AND there is at least one packet pending. The | |
300 | * ARMED bit is set to indicate a potential hang. The | |
301 | * bit is cleared if a pause frame is received to remove | |
302 | * false hang detection due to PFC or 802.3x frames. By | |
303 | * requiring this to fail twice we avoid races with | |
304 | * PFC clearing the ARMED bit and conditions where we | |
305 | * run the check_tx_hang logic with a transmit completion | |
306 | * pending but without time to complete it yet. | |
307 | */ | |
a114d0a6 | 308 | if ((tx_ring->tx_stats.tx_done_old == tx_ring->stats.packets) && |
fd0a05ce JB |
309 | tx_pending) { |
310 | /* make sure it is true for two checks in a row */ | |
311 | ret = test_and_set_bit(__I40E_HANG_CHECK_ARMED, | |
312 | &tx_ring->state); | |
313 | } else { | |
314 | /* update completed stats and disarm the hang check */ | |
a114d0a6 | 315 | tx_ring->tx_stats.tx_done_old = tx_ring->stats.packets; |
fd0a05ce JB |
316 | clear_bit(__I40E_HANG_CHECK_ARMED, &tx_ring->state); |
317 | } | |
318 | ||
319 | return ret; | |
320 | } | |
321 | ||
322 | /** | |
323 | * i40e_clean_tx_irq - Reclaim resources after transmit completes | |
324 | * @tx_ring: tx ring to clean | |
325 | * @budget: how many cleans we're allowed | |
326 | * | |
327 | * Returns true if there's any budget left (e.g. the clean is finished) | |
328 | **/ | |
329 | static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget) | |
330 | { | |
331 | u16 i = tx_ring->next_to_clean; | |
332 | struct i40e_tx_buffer *tx_buf; | |
333 | struct i40e_tx_desc *tx_desc; | |
334 | unsigned int total_packets = 0; | |
335 | unsigned int total_bytes = 0; | |
336 | ||
337 | tx_buf = &tx_ring->tx_bi[i]; | |
338 | tx_desc = I40E_TX_DESC(tx_ring, i); | |
a5e9c572 | 339 | i -= tx_ring->count; |
fd0a05ce | 340 | |
a5e9c572 AD |
341 | do { |
342 | struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch; | |
fd0a05ce JB |
343 | |
344 | /* if next_to_watch is not set then there is no work pending */ | |
345 | if (!eop_desc) | |
346 | break; | |
347 | ||
a5e9c572 AD |
348 | /* prevent any other reads prior to eop_desc */ |
349 | read_barrier_depends(); | |
350 | ||
fd0a05ce JB |
351 | /* if the descriptor isn't done, no work yet to do */ |
352 | if (!(eop_desc->cmd_type_offset_bsz & | |
353 | cpu_to_le64(I40E_TX_DESC_DTYPE_DESC_DONE))) | |
354 | break; | |
355 | ||
c304fdac | 356 | /* clear next_to_watch to prevent false hangs */ |
fd0a05ce | 357 | tx_buf->next_to_watch = NULL; |
fd0a05ce | 358 | |
a5e9c572 AD |
359 | /* update the statistics for this packet */ |
360 | total_bytes += tx_buf->bytecount; | |
361 | total_packets += tx_buf->gso_segs; | |
fd0a05ce | 362 | |
a5e9c572 AD |
363 | /* free the skb */ |
364 | dev_kfree_skb_any(tx_buf->skb); | |
fd0a05ce | 365 | |
a5e9c572 AD |
366 | /* unmap skb header data */ |
367 | dma_unmap_single(tx_ring->dev, | |
368 | dma_unmap_addr(tx_buf, dma), | |
369 | dma_unmap_len(tx_buf, len), | |
370 | DMA_TO_DEVICE); | |
fd0a05ce | 371 | |
a5e9c572 AD |
372 | /* clear tx_buffer data */ |
373 | tx_buf->skb = NULL; | |
374 | dma_unmap_len_set(tx_buf, len, 0); | |
fd0a05ce | 375 | |
a5e9c572 AD |
376 | /* unmap remaining buffers */ |
377 | while (tx_desc != eop_desc) { | |
fd0a05ce JB |
378 | |
379 | tx_buf++; | |
380 | tx_desc++; | |
381 | i++; | |
a5e9c572 AD |
382 | if (unlikely(!i)) { |
383 | i -= tx_ring->count; | |
fd0a05ce JB |
384 | tx_buf = tx_ring->tx_bi; |
385 | tx_desc = I40E_TX_DESC(tx_ring, 0); | |
386 | } | |
fd0a05ce | 387 | |
a5e9c572 AD |
388 | /* unmap any remaining paged data */ |
389 | if (dma_unmap_len(tx_buf, len)) { | |
390 | dma_unmap_page(tx_ring->dev, | |
391 | dma_unmap_addr(tx_buf, dma), | |
392 | dma_unmap_len(tx_buf, len), | |
393 | DMA_TO_DEVICE); | |
394 | dma_unmap_len_set(tx_buf, len, 0); | |
395 | } | |
396 | } | |
397 | ||
398 | /* move us one more past the eop_desc for start of next pkt */ | |
399 | tx_buf++; | |
400 | tx_desc++; | |
401 | i++; | |
402 | if (unlikely(!i)) { | |
403 | i -= tx_ring->count; | |
404 | tx_buf = tx_ring->tx_bi; | |
405 | tx_desc = I40E_TX_DESC(tx_ring, 0); | |
406 | } | |
407 | ||
408 | /* update budget accounting */ | |
409 | budget--; | |
410 | } while (likely(budget)); | |
411 | ||
412 | i += tx_ring->count; | |
fd0a05ce | 413 | tx_ring->next_to_clean = i; |
a114d0a6 AD |
414 | tx_ring->stats.bytes += total_bytes; |
415 | tx_ring->stats.packets += total_packets; | |
fd0a05ce JB |
416 | tx_ring->q_vector->tx.total_bytes += total_bytes; |
417 | tx_ring->q_vector->tx.total_packets += total_packets; | |
a5e9c572 | 418 | |
fd0a05ce JB |
419 | if (check_for_tx_hang(tx_ring) && i40e_check_tx_hang(tx_ring)) { |
420 | /* schedule immediate reset if we believe we hung */ | |
421 | dev_info(tx_ring->dev, "Detected Tx Unit Hang\n" | |
422 | " VSI <%d>\n" | |
423 | " Tx Queue <%d>\n" | |
424 | " next_to_use <%x>\n" | |
425 | " next_to_clean <%x>\n", | |
426 | tx_ring->vsi->seid, | |
427 | tx_ring->queue_index, | |
428 | tx_ring->next_to_use, i); | |
429 | dev_info(tx_ring->dev, "tx_bi[next_to_clean]\n" | |
430 | " time_stamp <%lx>\n" | |
431 | " jiffies <%lx>\n", | |
432 | tx_ring->tx_bi[i].time_stamp, jiffies); | |
433 | ||
434 | netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index); | |
435 | ||
436 | dev_info(tx_ring->dev, | |
437 | "tx hang detected on queue %d, resetting adapter\n", | |
438 | tx_ring->queue_index); | |
439 | ||
440 | tx_ring->netdev->netdev_ops->ndo_tx_timeout(tx_ring->netdev); | |
441 | ||
442 | /* the adapter is about to reset, no point in enabling stuff */ | |
443 | return true; | |
444 | } | |
445 | ||
7070ce0a AD |
446 | netdev_tx_completed_queue(netdev_get_tx_queue(tx_ring->netdev, |
447 | tx_ring->queue_index), | |
448 | total_packets, total_bytes); | |
449 | ||
fd0a05ce JB |
450 | #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2) |
451 | if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) && | |
452 | (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) { | |
453 | /* Make sure that anybody stopping the queue after this | |
454 | * sees the new next_to_clean. | |
455 | */ | |
456 | smp_mb(); | |
457 | if (__netif_subqueue_stopped(tx_ring->netdev, | |
458 | tx_ring->queue_index) && | |
459 | !test_bit(__I40E_DOWN, &tx_ring->vsi->state)) { | |
460 | netif_wake_subqueue(tx_ring->netdev, | |
461 | tx_ring->queue_index); | |
462 | ++tx_ring->tx_stats.restart_queue; | |
463 | } | |
464 | } | |
465 | ||
466 | return budget > 0; | |
467 | } | |
468 | ||
469 | /** | |
470 | * i40e_set_new_dynamic_itr - Find new ITR level | |
471 | * @rc: structure containing ring performance data | |
472 | * | |
473 | * Stores a new ITR value based on packets and byte counts during | |
474 | * the last interrupt. The advantage of per interrupt computation | |
475 | * is faster updates and more accurate ITR for the current traffic | |
476 | * pattern. Constants in this function were computed based on | |
477 | * theoretical maximum wire speed and thresholds were set based on | |
478 | * testing data as well as attempting to minimize response time | |
479 | * while increasing bulk throughput. | |
480 | **/ | |
481 | static void i40e_set_new_dynamic_itr(struct i40e_ring_container *rc) | |
482 | { | |
483 | enum i40e_latency_range new_latency_range = rc->latency_range; | |
484 | u32 new_itr = rc->itr; | |
485 | int bytes_per_int; | |
486 | ||
487 | if (rc->total_packets == 0 || !rc->itr) | |
488 | return; | |
489 | ||
490 | /* simple throttlerate management | |
491 | * 0-10MB/s lowest (100000 ints/s) | |
492 | * 10-20MB/s low (20000 ints/s) | |
493 | * 20-1249MB/s bulk (8000 ints/s) | |
494 | */ | |
495 | bytes_per_int = rc->total_bytes / rc->itr; | |
496 | switch (rc->itr) { | |
497 | case I40E_LOWEST_LATENCY: | |
498 | if (bytes_per_int > 10) | |
499 | new_latency_range = I40E_LOW_LATENCY; | |
500 | break; | |
501 | case I40E_LOW_LATENCY: | |
502 | if (bytes_per_int > 20) | |
503 | new_latency_range = I40E_BULK_LATENCY; | |
504 | else if (bytes_per_int <= 10) | |
505 | new_latency_range = I40E_LOWEST_LATENCY; | |
506 | break; | |
507 | case I40E_BULK_LATENCY: | |
508 | if (bytes_per_int <= 20) | |
509 | rc->latency_range = I40E_LOW_LATENCY; | |
510 | break; | |
511 | } | |
512 | ||
513 | switch (new_latency_range) { | |
514 | case I40E_LOWEST_LATENCY: | |
515 | new_itr = I40E_ITR_100K; | |
516 | break; | |
517 | case I40E_LOW_LATENCY: | |
518 | new_itr = I40E_ITR_20K; | |
519 | break; | |
520 | case I40E_BULK_LATENCY: | |
521 | new_itr = I40E_ITR_8K; | |
522 | break; | |
523 | default: | |
524 | break; | |
525 | } | |
526 | ||
527 | if (new_itr != rc->itr) { | |
528 | /* do an exponential smoothing */ | |
529 | new_itr = (10 * new_itr * rc->itr) / | |
530 | ((9 * new_itr) + rc->itr); | |
531 | rc->itr = new_itr & I40E_MAX_ITR; | |
532 | } | |
533 | ||
534 | rc->total_bytes = 0; | |
535 | rc->total_packets = 0; | |
536 | } | |
537 | ||
538 | /** | |
539 | * i40e_update_dynamic_itr - Adjust ITR based on bytes per int | |
540 | * @q_vector: the vector to adjust | |
541 | **/ | |
542 | static void i40e_update_dynamic_itr(struct i40e_q_vector *q_vector) | |
543 | { | |
544 | u16 vector = q_vector->vsi->base_vector + q_vector->v_idx; | |
545 | struct i40e_hw *hw = &q_vector->vsi->back->hw; | |
546 | u32 reg_addr; | |
547 | u16 old_itr; | |
548 | ||
549 | reg_addr = I40E_PFINT_ITRN(I40E_RX_ITR, vector - 1); | |
550 | old_itr = q_vector->rx.itr; | |
551 | i40e_set_new_dynamic_itr(&q_vector->rx); | |
552 | if (old_itr != q_vector->rx.itr) | |
553 | wr32(hw, reg_addr, q_vector->rx.itr); | |
554 | ||
555 | reg_addr = I40E_PFINT_ITRN(I40E_TX_ITR, vector - 1); | |
556 | old_itr = q_vector->tx.itr; | |
557 | i40e_set_new_dynamic_itr(&q_vector->tx); | |
558 | if (old_itr != q_vector->tx.itr) | |
559 | wr32(hw, reg_addr, q_vector->tx.itr); | |
560 | ||
561 | i40e_flush(hw); | |
562 | } | |
563 | ||
564 | /** | |
565 | * i40e_clean_programming_status - clean the programming status descriptor | |
566 | * @rx_ring: the rx ring that has this descriptor | |
567 | * @rx_desc: the rx descriptor written back by HW | |
568 | * | |
569 | * Flow director should handle FD_FILTER_STATUS to check its filter programming | |
570 | * status being successful or not and take actions accordingly. FCoE should | |
571 | * handle its context/filter programming/invalidation status and take actions. | |
572 | * | |
573 | **/ | |
574 | static void i40e_clean_programming_status(struct i40e_ring *rx_ring, | |
575 | union i40e_rx_desc *rx_desc) | |
576 | { | |
577 | u64 qw; | |
578 | u8 id; | |
579 | ||
580 | qw = le64_to_cpu(rx_desc->wb.qword1.status_error_len); | |
581 | id = (qw & I40E_RX_PROG_STATUS_DESC_QW1_PROGID_MASK) >> | |
582 | I40E_RX_PROG_STATUS_DESC_QW1_PROGID_SHIFT; | |
583 | ||
584 | if (id == I40E_RX_PROG_STATUS_DESC_FD_FILTER_STATUS) | |
585 | i40e_fd_handle_status(rx_ring, qw, id); | |
586 | } | |
587 | ||
588 | /** | |
589 | * i40e_setup_tx_descriptors - Allocate the Tx descriptors | |
590 | * @tx_ring: the tx ring to set up | |
591 | * | |
592 | * Return 0 on success, negative on error | |
593 | **/ | |
594 | int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring) | |
595 | { | |
596 | struct device *dev = tx_ring->dev; | |
597 | int bi_size; | |
598 | ||
599 | if (!dev) | |
600 | return -ENOMEM; | |
601 | ||
602 | bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count; | |
603 | tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL); | |
604 | if (!tx_ring->tx_bi) | |
605 | goto err; | |
606 | ||
607 | /* round up to nearest 4K */ | |
608 | tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc); | |
609 | tx_ring->size = ALIGN(tx_ring->size, 4096); | |
610 | tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size, | |
611 | &tx_ring->dma, GFP_KERNEL); | |
612 | if (!tx_ring->desc) { | |
613 | dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n", | |
614 | tx_ring->size); | |
615 | goto err; | |
616 | } | |
617 | ||
618 | tx_ring->next_to_use = 0; | |
619 | tx_ring->next_to_clean = 0; | |
620 | return 0; | |
621 | ||
622 | err: | |
623 | kfree(tx_ring->tx_bi); | |
624 | tx_ring->tx_bi = NULL; | |
625 | return -ENOMEM; | |
626 | } | |
627 | ||
628 | /** | |
629 | * i40e_clean_rx_ring - Free Rx buffers | |
630 | * @rx_ring: ring to be cleaned | |
631 | **/ | |
632 | void i40e_clean_rx_ring(struct i40e_ring *rx_ring) | |
633 | { | |
634 | struct device *dev = rx_ring->dev; | |
635 | struct i40e_rx_buffer *rx_bi; | |
636 | unsigned long bi_size; | |
637 | u16 i; | |
638 | ||
639 | /* ring already cleared, nothing to do */ | |
640 | if (!rx_ring->rx_bi) | |
641 | return; | |
642 | ||
643 | /* Free all the Rx ring sk_buffs */ | |
644 | for (i = 0; i < rx_ring->count; i++) { | |
645 | rx_bi = &rx_ring->rx_bi[i]; | |
646 | if (rx_bi->dma) { | |
647 | dma_unmap_single(dev, | |
648 | rx_bi->dma, | |
649 | rx_ring->rx_buf_len, | |
650 | DMA_FROM_DEVICE); | |
651 | rx_bi->dma = 0; | |
652 | } | |
653 | if (rx_bi->skb) { | |
654 | dev_kfree_skb(rx_bi->skb); | |
655 | rx_bi->skb = NULL; | |
656 | } | |
657 | if (rx_bi->page) { | |
658 | if (rx_bi->page_dma) { | |
659 | dma_unmap_page(dev, | |
660 | rx_bi->page_dma, | |
661 | PAGE_SIZE / 2, | |
662 | DMA_FROM_DEVICE); | |
663 | rx_bi->page_dma = 0; | |
664 | } | |
665 | __free_page(rx_bi->page); | |
666 | rx_bi->page = NULL; | |
667 | rx_bi->page_offset = 0; | |
668 | } | |
669 | } | |
670 | ||
671 | bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count; | |
672 | memset(rx_ring->rx_bi, 0, bi_size); | |
673 | ||
674 | /* Zero out the descriptor ring */ | |
675 | memset(rx_ring->desc, 0, rx_ring->size); | |
676 | ||
677 | rx_ring->next_to_clean = 0; | |
678 | rx_ring->next_to_use = 0; | |
679 | } | |
680 | ||
681 | /** | |
682 | * i40e_free_rx_resources - Free Rx resources | |
683 | * @rx_ring: ring to clean the resources from | |
684 | * | |
685 | * Free all receive software resources | |
686 | **/ | |
687 | void i40e_free_rx_resources(struct i40e_ring *rx_ring) | |
688 | { | |
689 | i40e_clean_rx_ring(rx_ring); | |
690 | kfree(rx_ring->rx_bi); | |
691 | rx_ring->rx_bi = NULL; | |
692 | ||
693 | if (rx_ring->desc) { | |
694 | dma_free_coherent(rx_ring->dev, rx_ring->size, | |
695 | rx_ring->desc, rx_ring->dma); | |
696 | rx_ring->desc = NULL; | |
697 | } | |
698 | } | |
699 | ||
700 | /** | |
701 | * i40e_setup_rx_descriptors - Allocate Rx descriptors | |
702 | * @rx_ring: Rx descriptor ring (for a specific queue) to setup | |
703 | * | |
704 | * Returns 0 on success, negative on failure | |
705 | **/ | |
706 | int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring) | |
707 | { | |
708 | struct device *dev = rx_ring->dev; | |
709 | int bi_size; | |
710 | ||
711 | bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count; | |
712 | rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL); | |
713 | if (!rx_ring->rx_bi) | |
714 | goto err; | |
715 | ||
716 | /* Round up to nearest 4K */ | |
717 | rx_ring->size = ring_is_16byte_desc_enabled(rx_ring) | |
718 | ? rx_ring->count * sizeof(union i40e_16byte_rx_desc) | |
719 | : rx_ring->count * sizeof(union i40e_32byte_rx_desc); | |
720 | rx_ring->size = ALIGN(rx_ring->size, 4096); | |
721 | rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size, | |
722 | &rx_ring->dma, GFP_KERNEL); | |
723 | ||
724 | if (!rx_ring->desc) { | |
725 | dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n", | |
726 | rx_ring->size); | |
727 | goto err; | |
728 | } | |
729 | ||
730 | rx_ring->next_to_clean = 0; | |
731 | rx_ring->next_to_use = 0; | |
732 | ||
733 | return 0; | |
734 | err: | |
735 | kfree(rx_ring->rx_bi); | |
736 | rx_ring->rx_bi = NULL; | |
737 | return -ENOMEM; | |
738 | } | |
739 | ||
740 | /** | |
741 | * i40e_release_rx_desc - Store the new tail and head values | |
742 | * @rx_ring: ring to bump | |
743 | * @val: new head index | |
744 | **/ | |
745 | static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val) | |
746 | { | |
747 | rx_ring->next_to_use = val; | |
748 | /* Force memory writes to complete before letting h/w | |
749 | * know there are new descriptors to fetch. (Only | |
750 | * applicable for weak-ordered memory model archs, | |
751 | * such as IA-64). | |
752 | */ | |
753 | wmb(); | |
754 | writel(val, rx_ring->tail); | |
755 | } | |
756 | ||
757 | /** | |
758 | * i40e_alloc_rx_buffers - Replace used receive buffers; packet split | |
759 | * @rx_ring: ring to place buffers on | |
760 | * @cleaned_count: number of buffers to replace | |
761 | **/ | |
762 | void i40e_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count) | |
763 | { | |
764 | u16 i = rx_ring->next_to_use; | |
765 | union i40e_rx_desc *rx_desc; | |
766 | struct i40e_rx_buffer *bi; | |
767 | struct sk_buff *skb; | |
768 | ||
769 | /* do nothing if no valid netdev defined */ | |
770 | if (!rx_ring->netdev || !cleaned_count) | |
771 | return; | |
772 | ||
773 | while (cleaned_count--) { | |
774 | rx_desc = I40E_RX_DESC(rx_ring, i); | |
775 | bi = &rx_ring->rx_bi[i]; | |
776 | skb = bi->skb; | |
777 | ||
778 | if (!skb) { | |
779 | skb = netdev_alloc_skb_ip_align(rx_ring->netdev, | |
780 | rx_ring->rx_buf_len); | |
781 | if (!skb) { | |
782 | rx_ring->rx_stats.alloc_rx_buff_failed++; | |
783 | goto no_buffers; | |
784 | } | |
785 | /* initialize queue mapping */ | |
786 | skb_record_rx_queue(skb, rx_ring->queue_index); | |
787 | bi->skb = skb; | |
788 | } | |
789 | ||
790 | if (!bi->dma) { | |
791 | bi->dma = dma_map_single(rx_ring->dev, | |
792 | skb->data, | |
793 | rx_ring->rx_buf_len, | |
794 | DMA_FROM_DEVICE); | |
795 | if (dma_mapping_error(rx_ring->dev, bi->dma)) { | |
796 | rx_ring->rx_stats.alloc_rx_buff_failed++; | |
797 | bi->dma = 0; | |
798 | goto no_buffers; | |
799 | } | |
800 | } | |
801 | ||
802 | if (ring_is_ps_enabled(rx_ring)) { | |
803 | if (!bi->page) { | |
804 | bi->page = alloc_page(GFP_ATOMIC); | |
805 | if (!bi->page) { | |
806 | rx_ring->rx_stats.alloc_rx_page_failed++; | |
807 | goto no_buffers; | |
808 | } | |
809 | } | |
810 | ||
811 | if (!bi->page_dma) { | |
812 | /* use a half page if we're re-using */ | |
813 | bi->page_offset ^= PAGE_SIZE / 2; | |
814 | bi->page_dma = dma_map_page(rx_ring->dev, | |
815 | bi->page, | |
816 | bi->page_offset, | |
817 | PAGE_SIZE / 2, | |
818 | DMA_FROM_DEVICE); | |
819 | if (dma_mapping_error(rx_ring->dev, | |
820 | bi->page_dma)) { | |
821 | rx_ring->rx_stats.alloc_rx_page_failed++; | |
822 | bi->page_dma = 0; | |
823 | goto no_buffers; | |
824 | } | |
825 | } | |
826 | ||
827 | /* Refresh the desc even if buffer_addrs didn't change | |
828 | * because each write-back erases this info. | |
829 | */ | |
830 | rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma); | |
831 | rx_desc->read.hdr_addr = cpu_to_le64(bi->dma); | |
832 | } else { | |
833 | rx_desc->read.pkt_addr = cpu_to_le64(bi->dma); | |
834 | rx_desc->read.hdr_addr = 0; | |
835 | } | |
836 | i++; | |
837 | if (i == rx_ring->count) | |
838 | i = 0; | |
839 | } | |
840 | ||
841 | no_buffers: | |
842 | if (rx_ring->next_to_use != i) | |
843 | i40e_release_rx_desc(rx_ring, i); | |
844 | } | |
845 | ||
846 | /** | |
847 | * i40e_receive_skb - Send a completed packet up the stack | |
848 | * @rx_ring: rx ring in play | |
849 | * @skb: packet to send up | |
850 | * @vlan_tag: vlan tag for packet | |
851 | **/ | |
852 | static void i40e_receive_skb(struct i40e_ring *rx_ring, | |
853 | struct sk_buff *skb, u16 vlan_tag) | |
854 | { | |
855 | struct i40e_q_vector *q_vector = rx_ring->q_vector; | |
856 | struct i40e_vsi *vsi = rx_ring->vsi; | |
857 | u64 flags = vsi->back->flags; | |
858 | ||
859 | if (vlan_tag & VLAN_VID_MASK) | |
860 | __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag); | |
861 | ||
862 | if (flags & I40E_FLAG_IN_NETPOLL) | |
863 | netif_rx(skb); | |
864 | else | |
865 | napi_gro_receive(&q_vector->napi, skb); | |
866 | } | |
867 | ||
868 | /** | |
869 | * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum | |
870 | * @vsi: the VSI we care about | |
871 | * @skb: skb currently being received and modified | |
872 | * @rx_status: status value of last descriptor in packet | |
873 | * @rx_error: error value of last descriptor in packet | |
874 | **/ | |
875 | static inline void i40e_rx_checksum(struct i40e_vsi *vsi, | |
876 | struct sk_buff *skb, | |
877 | u32 rx_status, | |
878 | u32 rx_error) | |
879 | { | |
880 | skb->ip_summed = CHECKSUM_NONE; | |
881 | ||
882 | /* Rx csum enabled and ip headers found? */ | |
883 | if (!(vsi->netdev->features & NETIF_F_RXCSUM && | |
884 | rx_status & (1 << I40E_RX_DESC_STATUS_L3L4P_SHIFT))) | |
885 | return; | |
886 | ||
887 | /* IP or L4 checksum error */ | |
888 | if (rx_error & ((1 << I40E_RX_DESC_ERROR_IPE_SHIFT) | | |
889 | (1 << I40E_RX_DESC_ERROR_L4E_SHIFT))) { | |
890 | vsi->back->hw_csum_rx_error++; | |
891 | return; | |
892 | } | |
893 | ||
894 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
895 | } | |
896 | ||
897 | /** | |
898 | * i40e_rx_hash - returns the hash value from the Rx descriptor | |
899 | * @ring: descriptor ring | |
900 | * @rx_desc: specific descriptor | |
901 | **/ | |
902 | static inline u32 i40e_rx_hash(struct i40e_ring *ring, | |
903 | union i40e_rx_desc *rx_desc) | |
904 | { | |
905 | if (ring->netdev->features & NETIF_F_RXHASH) { | |
906 | if ((le64_to_cpu(rx_desc->wb.qword1.status_error_len) >> | |
907 | I40E_RX_DESC_STATUS_FLTSTAT_SHIFT) & | |
908 | I40E_RX_DESC_FLTSTAT_RSS_HASH) | |
909 | return le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss); | |
910 | } | |
911 | return 0; | |
912 | } | |
913 | ||
914 | /** | |
915 | * i40e_clean_rx_irq - Reclaim resources after receive completes | |
916 | * @rx_ring: rx ring to clean | |
917 | * @budget: how many cleans we're allowed | |
918 | * | |
919 | * Returns true if there's any budget left (e.g. the clean is finished) | |
920 | **/ | |
921 | static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget) | |
922 | { | |
923 | unsigned int total_rx_bytes = 0, total_rx_packets = 0; | |
924 | u16 rx_packet_len, rx_header_len, rx_sph, rx_hbo; | |
925 | u16 cleaned_count = I40E_DESC_UNUSED(rx_ring); | |
926 | const int current_node = numa_node_id(); | |
927 | struct i40e_vsi *vsi = rx_ring->vsi; | |
928 | u16 i = rx_ring->next_to_clean; | |
929 | union i40e_rx_desc *rx_desc; | |
930 | u32 rx_error, rx_status; | |
931 | u64 qword; | |
932 | ||
933 | rx_desc = I40E_RX_DESC(rx_ring, i); | |
934 | qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len); | |
935 | rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) | |
936 | >> I40E_RXD_QW1_STATUS_SHIFT; | |
937 | ||
938 | while (rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)) { | |
939 | union i40e_rx_desc *next_rxd; | |
940 | struct i40e_rx_buffer *rx_bi; | |
941 | struct sk_buff *skb; | |
942 | u16 vlan_tag; | |
943 | if (i40e_rx_is_programming_status(qword)) { | |
944 | i40e_clean_programming_status(rx_ring, rx_desc); | |
945 | I40E_RX_NEXT_DESC_PREFETCH(rx_ring, i, next_rxd); | |
946 | goto next_desc; | |
947 | } | |
948 | rx_bi = &rx_ring->rx_bi[i]; | |
949 | skb = rx_bi->skb; | |
950 | prefetch(skb->data); | |
951 | ||
952 | rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) | |
953 | >> I40E_RXD_QW1_LENGTH_PBUF_SHIFT; | |
954 | rx_header_len = (qword & I40E_RXD_QW1_LENGTH_HBUF_MASK) | |
955 | >> I40E_RXD_QW1_LENGTH_HBUF_SHIFT; | |
956 | rx_sph = (qword & I40E_RXD_QW1_LENGTH_SPH_MASK) | |
957 | >> I40E_RXD_QW1_LENGTH_SPH_SHIFT; | |
958 | ||
959 | rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) | |
960 | >> I40E_RXD_QW1_ERROR_SHIFT; | |
961 | rx_hbo = rx_error & (1 << I40E_RX_DESC_ERROR_HBO_SHIFT); | |
962 | rx_error &= ~(1 << I40E_RX_DESC_ERROR_HBO_SHIFT); | |
963 | ||
964 | rx_bi->skb = NULL; | |
965 | ||
966 | /* This memory barrier is needed to keep us from reading | |
967 | * any other fields out of the rx_desc until we know the | |
968 | * STATUS_DD bit is set | |
969 | */ | |
970 | rmb(); | |
971 | ||
972 | /* Get the header and possibly the whole packet | |
973 | * If this is an skb from previous receive dma will be 0 | |
974 | */ | |
975 | if (rx_bi->dma) { | |
976 | u16 len; | |
977 | ||
978 | if (rx_hbo) | |
979 | len = I40E_RX_HDR_SIZE; | |
980 | else if (rx_sph) | |
981 | len = rx_header_len; | |
982 | else if (rx_packet_len) | |
983 | len = rx_packet_len; /* 1buf/no split found */ | |
984 | else | |
985 | len = rx_header_len; /* split always mode */ | |
986 | ||
987 | skb_put(skb, len); | |
988 | dma_unmap_single(rx_ring->dev, | |
989 | rx_bi->dma, | |
990 | rx_ring->rx_buf_len, | |
991 | DMA_FROM_DEVICE); | |
992 | rx_bi->dma = 0; | |
993 | } | |
994 | ||
995 | /* Get the rest of the data if this was a header split */ | |
996 | if (ring_is_ps_enabled(rx_ring) && rx_packet_len) { | |
997 | ||
998 | skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags, | |
999 | rx_bi->page, | |
1000 | rx_bi->page_offset, | |
1001 | rx_packet_len); | |
1002 | ||
1003 | skb->len += rx_packet_len; | |
1004 | skb->data_len += rx_packet_len; | |
1005 | skb->truesize += rx_packet_len; | |
1006 | ||
1007 | if ((page_count(rx_bi->page) == 1) && | |
1008 | (page_to_nid(rx_bi->page) == current_node)) | |
1009 | get_page(rx_bi->page); | |
1010 | else | |
1011 | rx_bi->page = NULL; | |
1012 | ||
1013 | dma_unmap_page(rx_ring->dev, | |
1014 | rx_bi->page_dma, | |
1015 | PAGE_SIZE / 2, | |
1016 | DMA_FROM_DEVICE); | |
1017 | rx_bi->page_dma = 0; | |
1018 | } | |
1019 | I40E_RX_NEXT_DESC_PREFETCH(rx_ring, i, next_rxd); | |
1020 | ||
1021 | if (unlikely( | |
1022 | !(rx_status & (1 << I40E_RX_DESC_STATUS_EOF_SHIFT)))) { | |
1023 | struct i40e_rx_buffer *next_buffer; | |
1024 | ||
1025 | next_buffer = &rx_ring->rx_bi[i]; | |
1026 | ||
1027 | if (ring_is_ps_enabled(rx_ring)) { | |
1028 | rx_bi->skb = next_buffer->skb; | |
1029 | rx_bi->dma = next_buffer->dma; | |
1030 | next_buffer->skb = skb; | |
1031 | next_buffer->dma = 0; | |
1032 | } | |
1033 | rx_ring->rx_stats.non_eop_descs++; | |
1034 | goto next_desc; | |
1035 | } | |
1036 | ||
1037 | /* ERR_MASK will only have valid bits if EOP set */ | |
1038 | if (unlikely(rx_error & (1 << I40E_RX_DESC_ERROR_RXE_SHIFT))) { | |
1039 | dev_kfree_skb_any(skb); | |
1040 | goto next_desc; | |
1041 | } | |
1042 | ||
1043 | skb->rxhash = i40e_rx_hash(rx_ring, rx_desc); | |
1044 | i40e_rx_checksum(vsi, skb, rx_status, rx_error); | |
1045 | ||
1046 | /* probably a little skewed due to removing CRC */ | |
1047 | total_rx_bytes += skb->len; | |
1048 | total_rx_packets++; | |
1049 | ||
1050 | skb->protocol = eth_type_trans(skb, rx_ring->netdev); | |
1051 | vlan_tag = rx_status & (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT) | |
1052 | ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1) | |
1053 | : 0; | |
1054 | i40e_receive_skb(rx_ring, skb, vlan_tag); | |
1055 | ||
1056 | rx_ring->netdev->last_rx = jiffies; | |
1057 | budget--; | |
1058 | next_desc: | |
1059 | rx_desc->wb.qword1.status_error_len = 0; | |
1060 | if (!budget) | |
1061 | break; | |
1062 | ||
1063 | cleaned_count++; | |
1064 | /* return some buffers to hardware, one at a time is too slow */ | |
1065 | if (cleaned_count >= I40E_RX_BUFFER_WRITE) { | |
1066 | i40e_alloc_rx_buffers(rx_ring, cleaned_count); | |
1067 | cleaned_count = 0; | |
1068 | } | |
1069 | ||
1070 | /* use prefetched values */ | |
1071 | rx_desc = next_rxd; | |
1072 | qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len); | |
1073 | rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) | |
1074 | >> I40E_RXD_QW1_STATUS_SHIFT; | |
1075 | } | |
1076 | ||
1077 | rx_ring->next_to_clean = i; | |
a114d0a6 AD |
1078 | rx_ring->stats.packets += total_rx_packets; |
1079 | rx_ring->stats.bytes += total_rx_bytes; | |
fd0a05ce JB |
1080 | rx_ring->q_vector->rx.total_packets += total_rx_packets; |
1081 | rx_ring->q_vector->rx.total_bytes += total_rx_bytes; | |
1082 | ||
1083 | if (cleaned_count) | |
1084 | i40e_alloc_rx_buffers(rx_ring, cleaned_count); | |
1085 | ||
1086 | return budget > 0; | |
1087 | } | |
1088 | ||
1089 | /** | |
1090 | * i40e_napi_poll - NAPI polling Rx/Tx cleanup routine | |
1091 | * @napi: napi struct with our devices info in it | |
1092 | * @budget: amount of work driver is allowed to do this pass, in packets | |
1093 | * | |
1094 | * This function will clean all queues associated with a q_vector. | |
1095 | * | |
1096 | * Returns the amount of work done | |
1097 | **/ | |
1098 | int i40e_napi_poll(struct napi_struct *napi, int budget) | |
1099 | { | |
1100 | struct i40e_q_vector *q_vector = | |
1101 | container_of(napi, struct i40e_q_vector, napi); | |
1102 | struct i40e_vsi *vsi = q_vector->vsi; | |
cd0b6fa6 | 1103 | struct i40e_ring *ring; |
fd0a05ce JB |
1104 | bool clean_complete = true; |
1105 | int budget_per_ring; | |
fd0a05ce JB |
1106 | |
1107 | if (test_bit(__I40E_DOWN, &vsi->state)) { | |
1108 | napi_complete(napi); | |
1109 | return 0; | |
1110 | } | |
1111 | ||
cd0b6fa6 AD |
1112 | /* Since the actual Tx work is minimal, we can give the Tx a larger |
1113 | * budget and be more aggressive about cleaning up the Tx descriptors. | |
1114 | */ | |
1115 | i40e_for_each_ring(ring, q_vector->tx) | |
1116 | clean_complete &= i40e_clean_tx_irq(ring, vsi->work_limit); | |
1117 | ||
fd0a05ce JB |
1118 | /* We attempt to distribute budget to each Rx queue fairly, but don't |
1119 | * allow the budget to go below 1 because that would exit polling early. | |
fd0a05ce JB |
1120 | */ |
1121 | budget_per_ring = max(budget/q_vector->num_ringpairs, 1); | |
cd0b6fa6 AD |
1122 | |
1123 | i40e_for_each_ring(ring, q_vector->rx) | |
1124 | clean_complete &= i40e_clean_rx_irq(ring, budget_per_ring); | |
fd0a05ce JB |
1125 | |
1126 | /* If work not completed, return budget and polling will return */ | |
1127 | if (!clean_complete) | |
1128 | return budget; | |
1129 | ||
1130 | /* Work is done so exit the polling mode and re-enable the interrupt */ | |
1131 | napi_complete(napi); | |
1132 | if (ITR_IS_DYNAMIC(vsi->rx_itr_setting) || | |
1133 | ITR_IS_DYNAMIC(vsi->tx_itr_setting)) | |
1134 | i40e_update_dynamic_itr(q_vector); | |
1135 | ||
1136 | if (!test_bit(__I40E_DOWN, &vsi->state)) { | |
1137 | if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) { | |
1138 | i40e_irq_dynamic_enable(vsi, | |
1139 | q_vector->v_idx + vsi->base_vector); | |
1140 | } else { | |
1141 | struct i40e_hw *hw = &vsi->back->hw; | |
1142 | /* We re-enable the queue 0 cause, but | |
1143 | * don't worry about dynamic_enable | |
1144 | * because we left it on for the other | |
1145 | * possible interrupts during napi | |
1146 | */ | |
1147 | u32 qval = rd32(hw, I40E_QINT_RQCTL(0)); | |
1148 | qval |= I40E_QINT_RQCTL_CAUSE_ENA_MASK; | |
1149 | wr32(hw, I40E_QINT_RQCTL(0), qval); | |
1150 | ||
1151 | qval = rd32(hw, I40E_QINT_TQCTL(0)); | |
1152 | qval |= I40E_QINT_TQCTL_CAUSE_ENA_MASK; | |
1153 | wr32(hw, I40E_QINT_TQCTL(0), qval); | |
1154 | i40e_flush(hw); | |
1155 | } | |
1156 | } | |
1157 | ||
1158 | return 0; | |
1159 | } | |
1160 | ||
1161 | /** | |
1162 | * i40e_atr - Add a Flow Director ATR filter | |
1163 | * @tx_ring: ring to add programming descriptor to | |
1164 | * @skb: send buffer | |
1165 | * @flags: send flags | |
1166 | * @protocol: wire protocol | |
1167 | **/ | |
1168 | static void i40e_atr(struct i40e_ring *tx_ring, struct sk_buff *skb, | |
1169 | u32 flags, __be16 protocol) | |
1170 | { | |
1171 | struct i40e_filter_program_desc *fdir_desc; | |
1172 | struct i40e_pf *pf = tx_ring->vsi->back; | |
1173 | union { | |
1174 | unsigned char *network; | |
1175 | struct iphdr *ipv4; | |
1176 | struct ipv6hdr *ipv6; | |
1177 | } hdr; | |
1178 | struct tcphdr *th; | |
1179 | unsigned int hlen; | |
1180 | u32 flex_ptype, dtype_cmd; | |
fc4ac67b | 1181 | u16 i; |
fd0a05ce JB |
1182 | |
1183 | /* make sure ATR is enabled */ | |
1184 | if (!(pf->flags & I40E_FLAG_FDIR_ATR_ENABLED)) | |
1185 | return; | |
1186 | ||
1187 | /* if sampling is disabled do nothing */ | |
1188 | if (!tx_ring->atr_sample_rate) | |
1189 | return; | |
1190 | ||
1191 | tx_ring->atr_count++; | |
1192 | ||
1193 | /* snag network header to get L4 type and address */ | |
1194 | hdr.network = skb_network_header(skb); | |
1195 | ||
1196 | /* Currently only IPv4/IPv6 with TCP is supported */ | |
1197 | if (protocol == htons(ETH_P_IP)) { | |
1198 | if (hdr.ipv4->protocol != IPPROTO_TCP) | |
1199 | return; | |
1200 | ||
1201 | /* access ihl as a u8 to avoid unaligned access on ia64 */ | |
1202 | hlen = (hdr.network[0] & 0x0F) << 2; | |
1203 | } else if (protocol == htons(ETH_P_IPV6)) { | |
1204 | if (hdr.ipv6->nexthdr != IPPROTO_TCP) | |
1205 | return; | |
1206 | ||
1207 | hlen = sizeof(struct ipv6hdr); | |
1208 | } else { | |
1209 | return; | |
1210 | } | |
1211 | ||
1212 | th = (struct tcphdr *)(hdr.network + hlen); | |
1213 | ||
1214 | /* sample on all syn/fin packets or once every atr sample rate */ | |
1215 | if (!th->fin && !th->syn && (tx_ring->atr_count < tx_ring->atr_sample_rate)) | |
1216 | return; | |
1217 | ||
1218 | tx_ring->atr_count = 0; | |
1219 | ||
1220 | /* grab the next descriptor */ | |
fc4ac67b AD |
1221 | i = tx_ring->next_to_use; |
1222 | fdir_desc = I40E_TX_FDIRDESC(tx_ring, i); | |
1223 | ||
1224 | i++; | |
1225 | tx_ring->next_to_use = (i < tx_ring->count) ? i : 0; | |
fd0a05ce JB |
1226 | |
1227 | flex_ptype = (tx_ring->queue_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) & | |
1228 | I40E_TXD_FLTR_QW0_QINDEX_MASK; | |
1229 | flex_ptype |= (protocol == htons(ETH_P_IP)) ? | |
1230 | (I40E_FILTER_PCTYPE_NONF_IPV4_TCP << | |
1231 | I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) : | |
1232 | (I40E_FILTER_PCTYPE_NONF_IPV6_TCP << | |
1233 | I40E_TXD_FLTR_QW0_PCTYPE_SHIFT); | |
1234 | ||
1235 | flex_ptype |= tx_ring->vsi->id << I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT; | |
1236 | ||
1237 | dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG; | |
1238 | ||
1239 | dtype_cmd |= th->fin ? | |
1240 | (I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE << | |
1241 | I40E_TXD_FLTR_QW1_PCMD_SHIFT) : | |
1242 | (I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE << | |
1243 | I40E_TXD_FLTR_QW1_PCMD_SHIFT); | |
1244 | ||
1245 | dtype_cmd |= I40E_FILTER_PROGRAM_DESC_DEST_DIRECT_PACKET_QINDEX << | |
1246 | I40E_TXD_FLTR_QW1_DEST_SHIFT; | |
1247 | ||
1248 | dtype_cmd |= I40E_FILTER_PROGRAM_DESC_FD_STATUS_FD_ID << | |
1249 | I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT; | |
1250 | ||
1251 | fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype); | |
1252 | fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd); | |
1253 | } | |
1254 | ||
1255 | #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS) | |
1256 | /** | |
1257 | * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW | |
1258 | * @skb: send buffer | |
1259 | * @tx_ring: ring to send buffer on | |
1260 | * @flags: the tx flags to be set | |
1261 | * | |
1262 | * Checks the skb and set up correspondingly several generic transmit flags | |
1263 | * related to VLAN tagging for the HW, such as VLAN, DCB, etc. | |
1264 | * | |
1265 | * Returns error code indicate the frame should be dropped upon error and the | |
1266 | * otherwise returns 0 to indicate the flags has been set properly. | |
1267 | **/ | |
1268 | static int i40e_tx_prepare_vlan_flags(struct sk_buff *skb, | |
1269 | struct i40e_ring *tx_ring, | |
1270 | u32 *flags) | |
1271 | { | |
1272 | __be16 protocol = skb->protocol; | |
1273 | u32 tx_flags = 0; | |
1274 | ||
1275 | /* if we have a HW VLAN tag being added, default to the HW one */ | |
1276 | if (vlan_tx_tag_present(skb)) { | |
1277 | tx_flags |= vlan_tx_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT; | |
1278 | tx_flags |= I40E_TX_FLAGS_HW_VLAN; | |
1279 | /* else if it is a SW VLAN, check the next protocol and store the tag */ | |
1280 | } else if (protocol == __constant_htons(ETH_P_8021Q)) { | |
1281 | struct vlan_hdr *vhdr, _vhdr; | |
1282 | vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr); | |
1283 | if (!vhdr) | |
1284 | return -EINVAL; | |
1285 | ||
1286 | protocol = vhdr->h_vlan_encapsulated_proto; | |
1287 | tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT; | |
1288 | tx_flags |= I40E_TX_FLAGS_SW_VLAN; | |
1289 | } | |
1290 | ||
1291 | /* Insert 802.1p priority into VLAN header */ | |
1292 | if ((tx_ring->vsi->back->flags & I40E_FLAG_DCB_ENABLED) && | |
1293 | ((tx_flags & (I40E_TX_FLAGS_HW_VLAN | I40E_TX_FLAGS_SW_VLAN)) || | |
1294 | (skb->priority != TC_PRIO_CONTROL))) { | |
1295 | tx_flags &= ~I40E_TX_FLAGS_VLAN_PRIO_MASK; | |
1296 | tx_flags |= (skb->priority & 0x7) << | |
1297 | I40E_TX_FLAGS_VLAN_PRIO_SHIFT; | |
1298 | if (tx_flags & I40E_TX_FLAGS_SW_VLAN) { | |
1299 | struct vlan_ethhdr *vhdr; | |
1300 | if (skb_header_cloned(skb) && | |
1301 | pskb_expand_head(skb, 0, 0, GFP_ATOMIC)) | |
1302 | return -ENOMEM; | |
1303 | vhdr = (struct vlan_ethhdr *)skb->data; | |
1304 | vhdr->h_vlan_TCI = htons(tx_flags >> | |
1305 | I40E_TX_FLAGS_VLAN_SHIFT); | |
1306 | } else { | |
1307 | tx_flags |= I40E_TX_FLAGS_HW_VLAN; | |
1308 | } | |
1309 | } | |
1310 | *flags = tx_flags; | |
1311 | return 0; | |
1312 | } | |
1313 | ||
fd0a05ce JB |
1314 | /** |
1315 | * i40e_tso - set up the tso context descriptor | |
1316 | * @tx_ring: ptr to the ring to send | |
1317 | * @skb: ptr to the skb we're sending | |
1318 | * @tx_flags: the collected send information | |
1319 | * @protocol: the send protocol | |
1320 | * @hdr_len: ptr to the size of the packet header | |
1321 | * @cd_tunneling: ptr to context descriptor bits | |
1322 | * | |
1323 | * Returns 0 if no TSO can happen, 1 if tso is going, or error | |
1324 | **/ | |
1325 | static int i40e_tso(struct i40e_ring *tx_ring, struct sk_buff *skb, | |
1326 | u32 tx_flags, __be16 protocol, u8 *hdr_len, | |
1327 | u64 *cd_type_cmd_tso_mss, u32 *cd_tunneling) | |
1328 | { | |
1329 | u32 cd_cmd, cd_tso_len, cd_mss; | |
1330 | struct tcphdr *tcph; | |
1331 | struct iphdr *iph; | |
1332 | u32 l4len; | |
1333 | int err; | |
1334 | struct ipv6hdr *ipv6h; | |
1335 | ||
1336 | if (!skb_is_gso(skb)) | |
1337 | return 0; | |
1338 | ||
1339 | if (skb_header_cloned(skb)) { | |
1340 | err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); | |
1341 | if (err) | |
1342 | return err; | |
1343 | } | |
1344 | ||
1345 | if (protocol == __constant_htons(ETH_P_IP)) { | |
1346 | iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb); | |
1347 | tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb); | |
1348 | iph->tot_len = 0; | |
1349 | iph->check = 0; | |
1350 | tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, | |
1351 | 0, IPPROTO_TCP, 0); | |
1352 | } else if (skb_is_gso_v6(skb)) { | |
1353 | ||
1354 | ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) | |
1355 | : ipv6_hdr(skb); | |
1356 | tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb); | |
1357 | ipv6h->payload_len = 0; | |
1358 | tcph->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, | |
1359 | 0, IPPROTO_TCP, 0); | |
1360 | } | |
1361 | ||
1362 | l4len = skb->encapsulation ? inner_tcp_hdrlen(skb) : tcp_hdrlen(skb); | |
1363 | *hdr_len = (skb->encapsulation | |
1364 | ? (skb_inner_transport_header(skb) - skb->data) | |
1365 | : skb_transport_offset(skb)) + l4len; | |
1366 | ||
1367 | /* find the field values */ | |
1368 | cd_cmd = I40E_TX_CTX_DESC_TSO; | |
1369 | cd_tso_len = skb->len - *hdr_len; | |
1370 | cd_mss = skb_shinfo(skb)->gso_size; | |
1371 | *cd_type_cmd_tso_mss |= ((u64)cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) | |
1372 | | ((u64)cd_tso_len | |
1373 | << I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) | |
1374 | | ((u64)cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT); | |
1375 | return 1; | |
1376 | } | |
1377 | ||
1378 | /** | |
1379 | * i40e_tx_enable_csum - Enable Tx checksum offloads | |
1380 | * @skb: send buffer | |
1381 | * @tx_flags: Tx flags currently set | |
1382 | * @td_cmd: Tx descriptor command bits to set | |
1383 | * @td_offset: Tx descriptor header offsets to set | |
1384 | * @cd_tunneling: ptr to context desc bits | |
1385 | **/ | |
1386 | static void i40e_tx_enable_csum(struct sk_buff *skb, u32 tx_flags, | |
1387 | u32 *td_cmd, u32 *td_offset, | |
1388 | struct i40e_ring *tx_ring, | |
1389 | u32 *cd_tunneling) | |
1390 | { | |
1391 | struct ipv6hdr *this_ipv6_hdr; | |
1392 | unsigned int this_tcp_hdrlen; | |
1393 | struct iphdr *this_ip_hdr; | |
1394 | u32 network_hdr_len; | |
1395 | u8 l4_hdr = 0; | |
1396 | ||
1397 | if (skb->encapsulation) { | |
1398 | network_hdr_len = skb_inner_network_header_len(skb); | |
1399 | this_ip_hdr = inner_ip_hdr(skb); | |
1400 | this_ipv6_hdr = inner_ipv6_hdr(skb); | |
1401 | this_tcp_hdrlen = inner_tcp_hdrlen(skb); | |
1402 | ||
1403 | if (tx_flags & I40E_TX_FLAGS_IPV4) { | |
1404 | ||
1405 | if (tx_flags & I40E_TX_FLAGS_TSO) { | |
1406 | *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4; | |
1407 | ip_hdr(skb)->check = 0; | |
1408 | } else { | |
1409 | *cd_tunneling |= | |
1410 | I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM; | |
1411 | } | |
1412 | } else if (tx_flags & I40E_TX_FLAGS_IPV6) { | |
1413 | if (tx_flags & I40E_TX_FLAGS_TSO) { | |
1414 | *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV6; | |
1415 | ip_hdr(skb)->check = 0; | |
1416 | } else { | |
1417 | *cd_tunneling |= | |
1418 | I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM; | |
1419 | } | |
1420 | } | |
1421 | ||
1422 | /* Now set the ctx descriptor fields */ | |
1423 | *cd_tunneling |= (skb_network_header_len(skb) >> 2) << | |
1424 | I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT | | |
1425 | I40E_TXD_CTX_UDP_TUNNELING | | |
1426 | ((skb_inner_network_offset(skb) - | |
1427 | skb_transport_offset(skb)) >> 1) << | |
1428 | I40E_TXD_CTX_QW0_NATLEN_SHIFT; | |
1429 | ||
1430 | } else { | |
1431 | network_hdr_len = skb_network_header_len(skb); | |
1432 | this_ip_hdr = ip_hdr(skb); | |
1433 | this_ipv6_hdr = ipv6_hdr(skb); | |
1434 | this_tcp_hdrlen = tcp_hdrlen(skb); | |
1435 | } | |
1436 | ||
1437 | /* Enable IP checksum offloads */ | |
1438 | if (tx_flags & I40E_TX_FLAGS_IPV4) { | |
1439 | l4_hdr = this_ip_hdr->protocol; | |
1440 | /* the stack computes the IP header already, the only time we | |
1441 | * need the hardware to recompute it is in the case of TSO. | |
1442 | */ | |
1443 | if (tx_flags & I40E_TX_FLAGS_TSO) { | |
1444 | *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM; | |
1445 | this_ip_hdr->check = 0; | |
1446 | } else { | |
1447 | *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4; | |
1448 | } | |
1449 | /* Now set the td_offset for IP header length */ | |
1450 | *td_offset = (network_hdr_len >> 2) << | |
1451 | I40E_TX_DESC_LENGTH_IPLEN_SHIFT; | |
1452 | } else if (tx_flags & I40E_TX_FLAGS_IPV6) { | |
1453 | l4_hdr = this_ipv6_hdr->nexthdr; | |
1454 | *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6; | |
1455 | /* Now set the td_offset for IP header length */ | |
1456 | *td_offset = (network_hdr_len >> 2) << | |
1457 | I40E_TX_DESC_LENGTH_IPLEN_SHIFT; | |
1458 | } | |
1459 | /* words in MACLEN + dwords in IPLEN + dwords in L4Len */ | |
1460 | *td_offset |= (skb_network_offset(skb) >> 1) << | |
1461 | I40E_TX_DESC_LENGTH_MACLEN_SHIFT; | |
1462 | ||
1463 | /* Enable L4 checksum offloads */ | |
1464 | switch (l4_hdr) { | |
1465 | case IPPROTO_TCP: | |
1466 | /* enable checksum offloads */ | |
1467 | *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP; | |
1468 | *td_offset |= (this_tcp_hdrlen >> 2) << | |
1469 | I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT; | |
1470 | break; | |
1471 | case IPPROTO_SCTP: | |
1472 | /* enable SCTP checksum offload */ | |
1473 | *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP; | |
1474 | *td_offset |= (sizeof(struct sctphdr) >> 2) << | |
1475 | I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT; | |
1476 | break; | |
1477 | case IPPROTO_UDP: | |
1478 | /* enable UDP checksum offload */ | |
1479 | *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP; | |
1480 | *td_offset |= (sizeof(struct udphdr) >> 2) << | |
1481 | I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT; | |
1482 | break; | |
1483 | default: | |
1484 | break; | |
1485 | } | |
1486 | } | |
1487 | ||
1488 | /** | |
1489 | * i40e_create_tx_ctx Build the Tx context descriptor | |
1490 | * @tx_ring: ring to create the descriptor on | |
1491 | * @cd_type_cmd_tso_mss: Quad Word 1 | |
1492 | * @cd_tunneling: Quad Word 0 - bits 0-31 | |
1493 | * @cd_l2tag2: Quad Word 0 - bits 32-63 | |
1494 | **/ | |
1495 | static void i40e_create_tx_ctx(struct i40e_ring *tx_ring, | |
1496 | const u64 cd_type_cmd_tso_mss, | |
1497 | const u32 cd_tunneling, const u32 cd_l2tag2) | |
1498 | { | |
1499 | struct i40e_tx_context_desc *context_desc; | |
fc4ac67b | 1500 | int i = tx_ring->next_to_use; |
fd0a05ce JB |
1501 | |
1502 | if (!cd_type_cmd_tso_mss && !cd_tunneling && !cd_l2tag2) | |
1503 | return; | |
1504 | ||
1505 | /* grab the next descriptor */ | |
fc4ac67b AD |
1506 | context_desc = I40E_TX_CTXTDESC(tx_ring, i); |
1507 | ||
1508 | i++; | |
1509 | tx_ring->next_to_use = (i < tx_ring->count) ? i : 0; | |
fd0a05ce JB |
1510 | |
1511 | /* cpu_to_le32 and assign to struct fields */ | |
1512 | context_desc->tunneling_params = cpu_to_le32(cd_tunneling); | |
1513 | context_desc->l2tag2 = cpu_to_le16(cd_l2tag2); | |
1514 | context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss); | |
1515 | } | |
1516 | ||
1517 | /** | |
1518 | * i40e_tx_map - Build the Tx descriptor | |
1519 | * @tx_ring: ring to send buffer on | |
1520 | * @skb: send buffer | |
1521 | * @first: first buffer info buffer to use | |
1522 | * @tx_flags: collected send information | |
1523 | * @hdr_len: size of the packet header | |
1524 | * @td_cmd: the command field in the descriptor | |
1525 | * @td_offset: offset for checksum or crc | |
1526 | **/ | |
1527 | static void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb, | |
1528 | struct i40e_tx_buffer *first, u32 tx_flags, | |
1529 | const u8 hdr_len, u32 td_cmd, u32 td_offset) | |
1530 | { | |
fd0a05ce JB |
1531 | unsigned int data_len = skb->data_len; |
1532 | unsigned int size = skb_headlen(skb); | |
a5e9c572 | 1533 | struct skb_frag_struct *frag; |
fd0a05ce JB |
1534 | struct i40e_tx_buffer *tx_bi; |
1535 | struct i40e_tx_desc *tx_desc; | |
a5e9c572 | 1536 | u16 i = tx_ring->next_to_use; |
fd0a05ce JB |
1537 | u32 td_tag = 0; |
1538 | dma_addr_t dma; | |
1539 | u16 gso_segs; | |
1540 | ||
fd0a05ce JB |
1541 | if (tx_flags & I40E_TX_FLAGS_HW_VLAN) { |
1542 | td_cmd |= I40E_TX_DESC_CMD_IL2TAG1; | |
1543 | td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >> | |
1544 | I40E_TX_FLAGS_VLAN_SHIFT; | |
1545 | } | |
1546 | ||
a5e9c572 AD |
1547 | if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO)) |
1548 | gso_segs = skb_shinfo(skb)->gso_segs; | |
1549 | else | |
1550 | gso_segs = 1; | |
1551 | ||
1552 | /* multiply data chunks by size of headers */ | |
1553 | first->bytecount = skb->len - hdr_len + (gso_segs * hdr_len); | |
1554 | first->gso_segs = gso_segs; | |
1555 | first->skb = skb; | |
1556 | first->tx_flags = tx_flags; | |
1557 | ||
1558 | dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE); | |
1559 | ||
fd0a05ce | 1560 | tx_desc = I40E_TX_DESC(tx_ring, i); |
a5e9c572 AD |
1561 | tx_bi = first; |
1562 | ||
1563 | for (frag = &skb_shinfo(skb)->frags[0];; frag++) { | |
1564 | if (dma_mapping_error(tx_ring->dev, dma)) | |
1565 | goto dma_error; | |
1566 | ||
1567 | /* record length, and DMA address */ | |
1568 | dma_unmap_len_set(tx_bi, len, size); | |
1569 | dma_unmap_addr_set(tx_bi, dma, dma); | |
1570 | ||
1571 | tx_desc->buffer_addr = cpu_to_le64(dma); | |
1572 | ||
1573 | while (unlikely(size > I40E_MAX_DATA_PER_TXD)) { | |
fd0a05ce JB |
1574 | tx_desc->cmd_type_offset_bsz = |
1575 | build_ctob(td_cmd, td_offset, | |
1576 | I40E_MAX_DATA_PER_TXD, td_tag); | |
1577 | ||
fd0a05ce JB |
1578 | tx_desc++; |
1579 | i++; | |
1580 | if (i == tx_ring->count) { | |
1581 | tx_desc = I40E_TX_DESC(tx_ring, 0); | |
1582 | i = 0; | |
1583 | } | |
fd0a05ce | 1584 | |
a5e9c572 AD |
1585 | dma += I40E_MAX_DATA_PER_TXD; |
1586 | size -= I40E_MAX_DATA_PER_TXD; | |
fd0a05ce | 1587 | |
a5e9c572 AD |
1588 | tx_desc->buffer_addr = cpu_to_le64(dma); |
1589 | } | |
fd0a05ce JB |
1590 | |
1591 | if (likely(!data_len)) | |
1592 | break; | |
1593 | ||
a5e9c572 AD |
1594 | tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset, |
1595 | size, td_tag); | |
fd0a05ce JB |
1596 | |
1597 | tx_desc++; | |
1598 | i++; | |
1599 | if (i == tx_ring->count) { | |
1600 | tx_desc = I40E_TX_DESC(tx_ring, 0); | |
1601 | i = 0; | |
1602 | } | |
1603 | ||
a5e9c572 AD |
1604 | size = skb_frag_size(frag); |
1605 | data_len -= size; | |
fd0a05ce | 1606 | |
a5e9c572 AD |
1607 | dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size, |
1608 | DMA_TO_DEVICE); | |
fd0a05ce | 1609 | |
a5e9c572 AD |
1610 | tx_bi = &tx_ring->tx_bi[i]; |
1611 | } | |
fd0a05ce | 1612 | |
a5e9c572 AD |
1613 | tx_desc->cmd_type_offset_bsz = |
1614 | build_ctob(td_cmd, td_offset, size, td_tag) | | |
1615 | cpu_to_le64((u64)I40E_TXD_CMD << I40E_TXD_QW1_CMD_SHIFT); | |
fd0a05ce | 1616 | |
7070ce0a AD |
1617 | netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev, |
1618 | tx_ring->queue_index), | |
1619 | first->bytecount); | |
1620 | ||
a5e9c572 | 1621 | /* set the timestamp */ |
fd0a05ce | 1622 | first->time_stamp = jiffies; |
fd0a05ce JB |
1623 | |
1624 | /* Force memory writes to complete before letting h/w | |
1625 | * know there are new descriptors to fetch. (Only | |
1626 | * applicable for weak-ordered memory model archs, | |
1627 | * such as IA-64). | |
1628 | */ | |
1629 | wmb(); | |
1630 | ||
a5e9c572 AD |
1631 | /* set next_to_watch value indicating a packet is present */ |
1632 | first->next_to_watch = tx_desc; | |
1633 | ||
1634 | i++; | |
1635 | if (i == tx_ring->count) | |
1636 | i = 0; | |
1637 | ||
1638 | tx_ring->next_to_use = i; | |
1639 | ||
1640 | /* notify HW of packet */ | |
fd0a05ce | 1641 | writel(i, tx_ring->tail); |
a5e9c572 | 1642 | |
fd0a05ce JB |
1643 | return; |
1644 | ||
1645 | dma_error: | |
a5e9c572 | 1646 | dev_info(tx_ring->dev, "TX DMA map failed\n"); |
fd0a05ce JB |
1647 | |
1648 | /* clear dma mappings for failed tx_bi map */ | |
1649 | for (;;) { | |
1650 | tx_bi = &tx_ring->tx_bi[i]; | |
a5e9c572 | 1651 | i40e_unmap_and_free_tx_resource(tx_ring, tx_bi); |
fd0a05ce JB |
1652 | if (tx_bi == first) |
1653 | break; | |
1654 | if (i == 0) | |
1655 | i = tx_ring->count; | |
1656 | i--; | |
1657 | } | |
1658 | ||
fd0a05ce JB |
1659 | tx_ring->next_to_use = i; |
1660 | } | |
1661 | ||
1662 | /** | |
1663 | * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions | |
1664 | * @tx_ring: the ring to be checked | |
1665 | * @size: the size buffer we want to assure is available | |
1666 | * | |
1667 | * Returns -EBUSY if a stop is needed, else 0 | |
1668 | **/ | |
1669 | static inline int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size) | |
1670 | { | |
1671 | netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index); | |
1672 | smp_mb(); | |
1673 | ||
1674 | /* Check again in a case another CPU has just made room available. */ | |
1675 | if (likely(I40E_DESC_UNUSED(tx_ring) < size)) | |
1676 | return -EBUSY; | |
1677 | ||
1678 | /* A reprieve! - use start_queue because it doesn't call schedule */ | |
1679 | netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index); | |
1680 | ++tx_ring->tx_stats.restart_queue; | |
1681 | return 0; | |
1682 | } | |
1683 | ||
1684 | /** | |
1685 | * i40e_maybe_stop_tx - 1st level check for tx stop conditions | |
1686 | * @tx_ring: the ring to be checked | |
1687 | * @size: the size buffer we want to assure is available | |
1688 | * | |
1689 | * Returns 0 if stop is not needed | |
1690 | **/ | |
1691 | static int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size) | |
1692 | { | |
1693 | if (likely(I40E_DESC_UNUSED(tx_ring) >= size)) | |
1694 | return 0; | |
1695 | return __i40e_maybe_stop_tx(tx_ring, size); | |
1696 | } | |
1697 | ||
1698 | /** | |
1699 | * i40e_xmit_descriptor_count - calculate number of tx descriptors needed | |
1700 | * @skb: send buffer | |
1701 | * @tx_ring: ring to send buffer on | |
1702 | * | |
1703 | * Returns number of data descriptors needed for this skb. Returns 0 to indicate | |
1704 | * there is not enough descriptors available in this ring since we need at least | |
1705 | * one descriptor. | |
1706 | **/ | |
1707 | static int i40e_xmit_descriptor_count(struct sk_buff *skb, | |
1708 | struct i40e_ring *tx_ring) | |
1709 | { | |
1710 | #if PAGE_SIZE > I40E_MAX_DATA_PER_TXD | |
1711 | unsigned int f; | |
1712 | #endif | |
1713 | int count = 0; | |
1714 | ||
1715 | /* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD, | |
1716 | * + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD, | |
1717 | * + 2 desc gap to keep tail from touching head, | |
1718 | * + 1 desc for context descriptor, | |
1719 | * otherwise try next time | |
1720 | */ | |
1721 | #if PAGE_SIZE > I40E_MAX_DATA_PER_TXD | |
1722 | for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) | |
1723 | count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size); | |
1724 | #else | |
1725 | count += skb_shinfo(skb)->nr_frags; | |
1726 | #endif | |
1727 | count += TXD_USE_COUNT(skb_headlen(skb)); | |
1728 | if (i40e_maybe_stop_tx(tx_ring, count + 3)) { | |
1729 | tx_ring->tx_stats.tx_busy++; | |
1730 | return 0; | |
1731 | } | |
1732 | return count; | |
1733 | } | |
1734 | ||
1735 | /** | |
1736 | * i40e_xmit_frame_ring - Sends buffer on Tx ring | |
1737 | * @skb: send buffer | |
1738 | * @tx_ring: ring to send buffer on | |
1739 | * | |
1740 | * Returns NETDEV_TX_OK if sent, else an error code | |
1741 | **/ | |
1742 | static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb, | |
1743 | struct i40e_ring *tx_ring) | |
1744 | { | |
1745 | u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT; | |
1746 | u32 cd_tunneling = 0, cd_l2tag2 = 0; | |
1747 | struct i40e_tx_buffer *first; | |
1748 | u32 td_offset = 0; | |
1749 | u32 tx_flags = 0; | |
1750 | __be16 protocol; | |
1751 | u32 td_cmd = 0; | |
1752 | u8 hdr_len = 0; | |
1753 | int tso; | |
1754 | if (0 == i40e_xmit_descriptor_count(skb, tx_ring)) | |
1755 | return NETDEV_TX_BUSY; | |
1756 | ||
1757 | /* prepare the xmit flags */ | |
1758 | if (i40e_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags)) | |
1759 | goto out_drop; | |
1760 | ||
1761 | /* obtain protocol of skb */ | |
1762 | protocol = skb->protocol; | |
1763 | ||
1764 | /* record the location of the first descriptor for this packet */ | |
1765 | first = &tx_ring->tx_bi[tx_ring->next_to_use]; | |
1766 | ||
1767 | /* setup IPv4/IPv6 offloads */ | |
1768 | if (protocol == __constant_htons(ETH_P_IP)) | |
1769 | tx_flags |= I40E_TX_FLAGS_IPV4; | |
1770 | else if (protocol == __constant_htons(ETH_P_IPV6)) | |
1771 | tx_flags |= I40E_TX_FLAGS_IPV6; | |
1772 | ||
1773 | tso = i40e_tso(tx_ring, skb, tx_flags, protocol, &hdr_len, | |
1774 | &cd_type_cmd_tso_mss, &cd_tunneling); | |
1775 | ||
1776 | if (tso < 0) | |
1777 | goto out_drop; | |
1778 | else if (tso) | |
1779 | tx_flags |= I40E_TX_FLAGS_TSO; | |
1780 | ||
1781 | skb_tx_timestamp(skb); | |
1782 | ||
b1941306 AD |
1783 | /* always enable CRC insertion offload */ |
1784 | td_cmd |= I40E_TX_DESC_CMD_ICRC; | |
1785 | ||
fd0a05ce | 1786 | /* Always offload the checksum, since it's in the data descriptor */ |
b1941306 | 1787 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
fd0a05ce JB |
1788 | tx_flags |= I40E_TX_FLAGS_CSUM; |
1789 | ||
fd0a05ce JB |
1790 | i40e_tx_enable_csum(skb, tx_flags, &td_cmd, &td_offset, |
1791 | tx_ring, &cd_tunneling); | |
b1941306 | 1792 | } |
fd0a05ce JB |
1793 | |
1794 | i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss, | |
1795 | cd_tunneling, cd_l2tag2); | |
1796 | ||
1797 | /* Add Flow Director ATR if it's enabled. | |
1798 | * | |
1799 | * NOTE: this must always be directly before the data descriptor. | |
1800 | */ | |
1801 | i40e_atr(tx_ring, skb, tx_flags, protocol); | |
1802 | ||
1803 | i40e_tx_map(tx_ring, skb, first, tx_flags, hdr_len, | |
1804 | td_cmd, td_offset); | |
1805 | ||
1806 | i40e_maybe_stop_tx(tx_ring, DESC_NEEDED); | |
1807 | ||
1808 | return NETDEV_TX_OK; | |
1809 | ||
1810 | out_drop: | |
1811 | dev_kfree_skb_any(skb); | |
1812 | return NETDEV_TX_OK; | |
1813 | } | |
1814 | ||
1815 | /** | |
1816 | * i40e_lan_xmit_frame - Selects the correct VSI and Tx queue to send buffer | |
1817 | * @skb: send buffer | |
1818 | * @netdev: network interface device structure | |
1819 | * | |
1820 | * Returns NETDEV_TX_OK if sent, else an error code | |
1821 | **/ | |
1822 | netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev) | |
1823 | { | |
1824 | struct i40e_netdev_priv *np = netdev_priv(netdev); | |
1825 | struct i40e_vsi *vsi = np->vsi; | |
9f65e15b | 1826 | struct i40e_ring *tx_ring = vsi->tx_rings[skb->queue_mapping]; |
fd0a05ce JB |
1827 | |
1828 | /* hardware can't handle really short frames, hardware padding works | |
1829 | * beyond this point | |
1830 | */ | |
1831 | if (unlikely(skb->len < I40E_MIN_TX_LEN)) { | |
1832 | if (skb_pad(skb, I40E_MIN_TX_LEN - skb->len)) | |
1833 | return NETDEV_TX_OK; | |
1834 | skb->len = I40E_MIN_TX_LEN; | |
1835 | skb_set_tail_pointer(skb, I40E_MIN_TX_LEN); | |
1836 | } | |
1837 | ||
1838 | return i40e_xmit_frame_ring(skb, tx_ring); | |
1839 | } |