Commit | Line | Data |
---|---|---|
7f12ad74 GR |
1 | /******************************************************************************* |
2 | * | |
3 | * Intel Ethernet Controller XL710 Family Linux Virtual Function Driver | |
ecc6a239 | 4 | * Copyright(c) 2013 - 2016 Intel Corporation. |
7f12ad74 GR |
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 | * | |
b831607d JB |
15 | * You should have received a copy of the GNU General Public License along |
16 | * with this program. If not, see <http://www.gnu.org/licenses/>. | |
17 | * | |
7f12ad74 GR |
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 | ||
7ed3f5f0 | 27 | #include <linux/prefetch.h> |
a132af24 | 28 | #include <net/busy_poll.h> |
7ed3f5f0 | 29 | |
7f12ad74 | 30 | #include "i40evf.h" |
206812b5 | 31 | #include "i40e_prototype.h" |
7f12ad74 GR |
32 | |
33 | static inline __le64 build_ctob(u32 td_cmd, u32 td_offset, unsigned int size, | |
34 | u32 td_tag) | |
35 | { | |
36 | return cpu_to_le64(I40E_TX_DESC_DTYPE_DATA | | |
37 | ((u64)td_cmd << I40E_TXD_QW1_CMD_SHIFT) | | |
38 | ((u64)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) | | |
39 | ((u64)size << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) | | |
40 | ((u64)td_tag << I40E_TXD_QW1_L2TAG1_SHIFT)); | |
41 | } | |
42 | ||
43 | #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS) | |
44 | ||
45 | /** | |
46 | * i40e_unmap_and_free_tx_resource - Release a Tx buffer | |
47 | * @ring: the ring that owns the buffer | |
48 | * @tx_buffer: the buffer to free | |
49 | **/ | |
50 | static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring, | |
51 | struct i40e_tx_buffer *tx_buffer) | |
52 | { | |
53 | if (tx_buffer->skb) { | |
a42e7a36 | 54 | dev_kfree_skb_any(tx_buffer->skb); |
7f12ad74 GR |
55 | if (dma_unmap_len(tx_buffer, len)) |
56 | dma_unmap_single(ring->dev, | |
57 | dma_unmap_addr(tx_buffer, dma), | |
58 | dma_unmap_len(tx_buffer, len), | |
59 | DMA_TO_DEVICE); | |
60 | } else if (dma_unmap_len(tx_buffer, len)) { | |
61 | dma_unmap_page(ring->dev, | |
62 | dma_unmap_addr(tx_buffer, dma), | |
63 | dma_unmap_len(tx_buffer, len), | |
64 | DMA_TO_DEVICE); | |
65 | } | |
a42e7a36 KP |
66 | |
67 | if (tx_buffer->tx_flags & I40E_TX_FLAGS_FD_SB) | |
68 | kfree(tx_buffer->raw_buf); | |
69 | ||
7f12ad74 GR |
70 | tx_buffer->next_to_watch = NULL; |
71 | tx_buffer->skb = NULL; | |
72 | dma_unmap_len_set(tx_buffer, len, 0); | |
73 | /* tx_buffer must be completely set up in the transmit path */ | |
74 | } | |
75 | ||
76 | /** | |
77 | * i40evf_clean_tx_ring - Free any empty Tx buffers | |
78 | * @tx_ring: ring to be cleaned | |
79 | **/ | |
80 | void i40evf_clean_tx_ring(struct i40e_ring *tx_ring) | |
81 | { | |
82 | unsigned long bi_size; | |
83 | u16 i; | |
84 | ||
85 | /* ring already cleared, nothing to do */ | |
86 | if (!tx_ring->tx_bi) | |
87 | return; | |
88 | ||
89 | /* Free all the Tx ring sk_buffs */ | |
90 | for (i = 0; i < tx_ring->count; i++) | |
91 | i40e_unmap_and_free_tx_resource(tx_ring, &tx_ring->tx_bi[i]); | |
92 | ||
93 | bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count; | |
94 | memset(tx_ring->tx_bi, 0, bi_size); | |
95 | ||
96 | /* Zero out the descriptor ring */ | |
97 | memset(tx_ring->desc, 0, tx_ring->size); | |
98 | ||
99 | tx_ring->next_to_use = 0; | |
100 | tx_ring->next_to_clean = 0; | |
101 | ||
102 | if (!tx_ring->netdev) | |
103 | return; | |
104 | ||
105 | /* cleanup Tx queue statistics */ | |
106 | netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev, | |
107 | tx_ring->queue_index)); | |
108 | } | |
109 | ||
110 | /** | |
111 | * i40evf_free_tx_resources - Free Tx resources per queue | |
112 | * @tx_ring: Tx descriptor ring for a specific queue | |
113 | * | |
114 | * Free all transmit software resources | |
115 | **/ | |
116 | void i40evf_free_tx_resources(struct i40e_ring *tx_ring) | |
117 | { | |
118 | i40evf_clean_tx_ring(tx_ring); | |
119 | kfree(tx_ring->tx_bi); | |
120 | tx_ring->tx_bi = NULL; | |
121 | ||
122 | if (tx_ring->desc) { | |
123 | dma_free_coherent(tx_ring->dev, tx_ring->size, | |
124 | tx_ring->desc, tx_ring->dma); | |
125 | tx_ring->desc = NULL; | |
126 | } | |
127 | } | |
128 | ||
a68de58d | 129 | /** |
9c6c1259 KP |
130 | * i40evf_get_tx_pending - how many Tx descriptors not processed |
131 | * @tx_ring: the ring of descriptors | |
a68de58d | 132 | * |
9c6c1259 KP |
133 | * Since there is no access to the ring head register |
134 | * in XL710, we need to use our local copies | |
a68de58d | 135 | **/ |
9c6c1259 | 136 | u32 i40evf_get_tx_pending(struct i40e_ring *ring) |
a68de58d | 137 | { |
9c6c1259 | 138 | u32 head, tail; |
a68de58d | 139 | |
9c6c1259 KP |
140 | head = i40e_get_head(ring); |
141 | tail = readl(ring->tail); | |
142 | ||
143 | if (head != tail) | |
144 | return (head < tail) ? | |
145 | tail - head : (tail + ring->count - head); | |
146 | ||
147 | return 0; | |
a68de58d JB |
148 | } |
149 | ||
c29af37f ASJ |
150 | #define WB_STRIDE 0x3 |
151 | ||
7f12ad74 GR |
152 | /** |
153 | * i40e_clean_tx_irq - Reclaim resources after transmit completes | |
154 | * @tx_ring: tx ring to clean | |
155 | * @budget: how many cleans we're allowed | |
156 | * | |
157 | * Returns true if there's any budget left (e.g. the clean is finished) | |
158 | **/ | |
159 | static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget) | |
160 | { | |
161 | u16 i = tx_ring->next_to_clean; | |
162 | struct i40e_tx_buffer *tx_buf; | |
1943d8ba | 163 | struct i40e_tx_desc *tx_head; |
7f12ad74 GR |
164 | struct i40e_tx_desc *tx_desc; |
165 | unsigned int total_packets = 0; | |
166 | unsigned int total_bytes = 0; | |
167 | ||
168 | tx_buf = &tx_ring->tx_bi[i]; | |
169 | tx_desc = I40E_TX_DESC(tx_ring, i); | |
170 | i -= tx_ring->count; | |
171 | ||
1943d8ba JB |
172 | tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring)); |
173 | ||
7f12ad74 GR |
174 | do { |
175 | struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch; | |
176 | ||
177 | /* if next_to_watch is not set then there is no work pending */ | |
178 | if (!eop_desc) | |
179 | break; | |
180 | ||
181 | /* prevent any other reads prior to eop_desc */ | |
182 | read_barrier_depends(); | |
183 | ||
1943d8ba JB |
184 | /* we have caught up to head, no work left to do */ |
185 | if (tx_head == tx_desc) | |
7f12ad74 GR |
186 | break; |
187 | ||
188 | /* clear next_to_watch to prevent false hangs */ | |
189 | tx_buf->next_to_watch = NULL; | |
190 | ||
191 | /* update the statistics for this packet */ | |
192 | total_bytes += tx_buf->bytecount; | |
193 | total_packets += tx_buf->gso_segs; | |
194 | ||
195 | /* free the skb */ | |
196 | dev_kfree_skb_any(tx_buf->skb); | |
197 | ||
198 | /* unmap skb header data */ | |
199 | dma_unmap_single(tx_ring->dev, | |
200 | dma_unmap_addr(tx_buf, dma), | |
201 | dma_unmap_len(tx_buf, len), | |
202 | DMA_TO_DEVICE); | |
203 | ||
204 | /* clear tx_buffer data */ | |
205 | tx_buf->skb = NULL; | |
206 | dma_unmap_len_set(tx_buf, len, 0); | |
207 | ||
208 | /* unmap remaining buffers */ | |
209 | while (tx_desc != eop_desc) { | |
210 | ||
211 | tx_buf++; | |
212 | tx_desc++; | |
213 | i++; | |
214 | if (unlikely(!i)) { | |
215 | i -= tx_ring->count; | |
216 | tx_buf = tx_ring->tx_bi; | |
217 | tx_desc = I40E_TX_DESC(tx_ring, 0); | |
218 | } | |
219 | ||
220 | /* unmap any remaining paged data */ | |
221 | if (dma_unmap_len(tx_buf, len)) { | |
222 | dma_unmap_page(tx_ring->dev, | |
223 | dma_unmap_addr(tx_buf, dma), | |
224 | dma_unmap_len(tx_buf, len), | |
225 | DMA_TO_DEVICE); | |
226 | dma_unmap_len_set(tx_buf, len, 0); | |
227 | } | |
228 | } | |
229 | ||
230 | /* move us one more past the eop_desc for start of next pkt */ | |
231 | tx_buf++; | |
232 | tx_desc++; | |
233 | i++; | |
234 | if (unlikely(!i)) { | |
235 | i -= tx_ring->count; | |
236 | tx_buf = tx_ring->tx_bi; | |
237 | tx_desc = I40E_TX_DESC(tx_ring, 0); | |
238 | } | |
239 | ||
016890b9 JB |
240 | prefetch(tx_desc); |
241 | ||
7f12ad74 GR |
242 | /* update budget accounting */ |
243 | budget--; | |
244 | } while (likely(budget)); | |
245 | ||
246 | i += tx_ring->count; | |
247 | tx_ring->next_to_clean = i; | |
248 | u64_stats_update_begin(&tx_ring->syncp); | |
249 | tx_ring->stats.bytes += total_bytes; | |
250 | tx_ring->stats.packets += total_packets; | |
251 | u64_stats_update_end(&tx_ring->syncp); | |
252 | tx_ring->q_vector->tx.total_bytes += total_bytes; | |
253 | tx_ring->q_vector->tx.total_packets += total_packets; | |
254 | ||
f6d83d13 ASJ |
255 | if (tx_ring->flags & I40E_TXR_FLAGS_WB_ON_ITR) { |
256 | unsigned int j = 0; | |
257 | /* check to see if there are < 4 descriptors | |
258 | * waiting to be written back, then kick the hardware to force | |
259 | * them to be written back in case we stay in NAPI. | |
260 | * In this mode on X722 we do not enable Interrupt. | |
261 | */ | |
262 | j = i40evf_get_tx_pending(tx_ring); | |
263 | ||
264 | if (budget && | |
265 | ((j / (WB_STRIDE + 1)) == 0) && (j > 0) && | |
266 | !test_bit(__I40E_DOWN, &tx_ring->vsi->state) && | |
267 | (I40E_DESC_UNUSED(tx_ring) != tx_ring->count)) | |
268 | tx_ring->arm_wb = true; | |
269 | } | |
270 | ||
7f12ad74 GR |
271 | netdev_tx_completed_queue(netdev_get_tx_queue(tx_ring->netdev, |
272 | tx_ring->queue_index), | |
273 | total_packets, total_bytes); | |
274 | ||
275 | #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2) | |
276 | if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) && | |
277 | (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) { | |
278 | /* Make sure that anybody stopping the queue after this | |
279 | * sees the new next_to_clean. | |
280 | */ | |
281 | smp_mb(); | |
282 | if (__netif_subqueue_stopped(tx_ring->netdev, | |
283 | tx_ring->queue_index) && | |
284 | !test_bit(__I40E_DOWN, &tx_ring->vsi->state)) { | |
285 | netif_wake_subqueue(tx_ring->netdev, | |
286 | tx_ring->queue_index); | |
287 | ++tx_ring->tx_stats.restart_queue; | |
288 | } | |
289 | } | |
290 | ||
b03a8c1f | 291 | return !!budget; |
7f12ad74 GR |
292 | } |
293 | ||
c29af37f | 294 | /** |
ecc6a239 | 295 | * i40evf_enable_wb_on_itr - Arm hardware to do a wb, interrupts are not enabled |
c29af37f | 296 | * @vsi: the VSI we care about |
ecc6a239 | 297 | * @q_vector: the vector on which to enable writeback |
c29af37f ASJ |
298 | * |
299 | **/ | |
ecc6a239 ASJ |
300 | static void i40e_enable_wb_on_itr(struct i40e_vsi *vsi, |
301 | struct i40e_q_vector *q_vector) | |
c29af37f | 302 | { |
8e0764b4 | 303 | u16 flags = q_vector->tx.ring[0].flags; |
ecc6a239 | 304 | u32 val; |
8e0764b4 | 305 | |
ecc6a239 ASJ |
306 | if (!(flags & I40E_TXR_FLAGS_WB_ON_ITR)) |
307 | return; | |
308 | ||
309 | if (q_vector->arm_wb_state) | |
310 | return; | |
311 | ||
312 | val = I40E_VFINT_DYN_CTLN1_WB_ON_ITR_MASK | | |
313 | I40E_VFINT_DYN_CTLN1_ITR_INDX_MASK; /* set noitr */ | |
314 | ||
315 | wr32(&vsi->back->hw, | |
316 | I40E_VFINT_DYN_CTLN1(q_vector->v_idx + | |
317 | vsi->base_vector - 1), val); | |
318 | q_vector->arm_wb_state = true; | |
319 | } | |
320 | ||
321 | /** | |
322 | * i40evf_force_wb - Issue SW Interrupt so HW does a wb | |
323 | * @vsi: the VSI we care about | |
324 | * @q_vector: the vector on which to force writeback | |
325 | * | |
326 | **/ | |
327 | void i40evf_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector) | |
328 | { | |
329 | u32 val = I40E_VFINT_DYN_CTLN1_INTENA_MASK | | |
330 | I40E_VFINT_DYN_CTLN1_ITR_INDX_MASK | /* set noitr */ | |
331 | I40E_VFINT_DYN_CTLN1_SWINT_TRIG_MASK | | |
332 | I40E_VFINT_DYN_CTLN1_SW_ITR_INDX_ENA_MASK | |
333 | /* allow 00 to be written to the index */; | |
334 | ||
335 | wr32(&vsi->back->hw, | |
336 | I40E_VFINT_DYN_CTLN1(q_vector->v_idx + vsi->base_vector - 1), | |
337 | val); | |
c29af37f ASJ |
338 | } |
339 | ||
7f12ad74 GR |
340 | /** |
341 | * i40e_set_new_dynamic_itr - Find new ITR level | |
342 | * @rc: structure containing ring performance data | |
343 | * | |
8f5e39ce JB |
344 | * Returns true if ITR changed, false if not |
345 | * | |
7f12ad74 GR |
346 | * Stores a new ITR value based on packets and byte counts during |
347 | * the last interrupt. The advantage of per interrupt computation | |
348 | * is faster updates and more accurate ITR for the current traffic | |
349 | * pattern. Constants in this function were computed based on | |
350 | * theoretical maximum wire speed and thresholds were set based on | |
351 | * testing data as well as attempting to minimize response time | |
352 | * while increasing bulk throughput. | |
353 | **/ | |
8f5e39ce | 354 | static bool i40e_set_new_dynamic_itr(struct i40e_ring_container *rc) |
7f12ad74 GR |
355 | { |
356 | enum i40e_latency_range new_latency_range = rc->latency_range; | |
c56625d5 | 357 | struct i40e_q_vector *qv = rc->ring->q_vector; |
7f12ad74 GR |
358 | u32 new_itr = rc->itr; |
359 | int bytes_per_int; | |
51cc6d9f | 360 | int usecs; |
7f12ad74 GR |
361 | |
362 | if (rc->total_packets == 0 || !rc->itr) | |
8f5e39ce | 363 | return false; |
7f12ad74 GR |
364 | |
365 | /* simple throttlerate management | |
c56625d5 | 366 | * 0-10MB/s lowest (50000 ints/s) |
7f12ad74 | 367 | * 10-20MB/s low (20000 ints/s) |
c56625d5 JB |
368 | * 20-1249MB/s bulk (18000 ints/s) |
369 | * > 40000 Rx packets per second (8000 ints/s) | |
51cc6d9f JB |
370 | * |
371 | * The math works out because the divisor is in 10^(-6) which | |
372 | * turns the bytes/us input value into MB/s values, but | |
373 | * make sure to use usecs, as the register values written | |
ee2319cf JB |
374 | * are in 2 usec increments in the ITR registers, and make sure |
375 | * to use the smoothed values that the countdown timer gives us. | |
7f12ad74 | 376 | */ |
ee2319cf | 377 | usecs = (rc->itr << 1) * ITR_COUNTDOWN_START; |
51cc6d9f | 378 | bytes_per_int = rc->total_bytes / usecs; |
ee2319cf | 379 | |
de32e3ef | 380 | switch (new_latency_range) { |
7f12ad74 GR |
381 | case I40E_LOWEST_LATENCY: |
382 | if (bytes_per_int > 10) | |
383 | new_latency_range = I40E_LOW_LATENCY; | |
384 | break; | |
385 | case I40E_LOW_LATENCY: | |
386 | if (bytes_per_int > 20) | |
387 | new_latency_range = I40E_BULK_LATENCY; | |
388 | else if (bytes_per_int <= 10) | |
389 | new_latency_range = I40E_LOWEST_LATENCY; | |
390 | break; | |
391 | case I40E_BULK_LATENCY: | |
c56625d5 | 392 | case I40E_ULTRA_LATENCY: |
de32e3ef CW |
393 | default: |
394 | if (bytes_per_int <= 20) | |
395 | new_latency_range = I40E_LOW_LATENCY; | |
7f12ad74 GR |
396 | break; |
397 | } | |
c56625d5 JB |
398 | |
399 | /* this is to adjust RX more aggressively when streaming small | |
400 | * packets. The value of 40000 was picked as it is just beyond | |
401 | * what the hardware can receive per second if in low latency | |
402 | * mode. | |
403 | */ | |
404 | #define RX_ULTRA_PACKET_RATE 40000 | |
405 | ||
406 | if ((((rc->total_packets * 1000000) / usecs) > RX_ULTRA_PACKET_RATE) && | |
407 | (&qv->rx == rc)) | |
408 | new_latency_range = I40E_ULTRA_LATENCY; | |
409 | ||
de32e3ef | 410 | rc->latency_range = new_latency_range; |
7f12ad74 GR |
411 | |
412 | switch (new_latency_range) { | |
413 | case I40E_LOWEST_LATENCY: | |
c56625d5 | 414 | new_itr = I40E_ITR_50K; |
7f12ad74 GR |
415 | break; |
416 | case I40E_LOW_LATENCY: | |
417 | new_itr = I40E_ITR_20K; | |
418 | break; | |
419 | case I40E_BULK_LATENCY: | |
c56625d5 JB |
420 | new_itr = I40E_ITR_18K; |
421 | break; | |
422 | case I40E_ULTRA_LATENCY: | |
7f12ad74 GR |
423 | new_itr = I40E_ITR_8K; |
424 | break; | |
425 | default: | |
426 | break; | |
427 | } | |
428 | ||
7f12ad74 GR |
429 | rc->total_bytes = 0; |
430 | rc->total_packets = 0; | |
8f5e39ce JB |
431 | |
432 | if (new_itr != rc->itr) { | |
433 | rc->itr = new_itr; | |
434 | return true; | |
435 | } | |
436 | ||
437 | return false; | |
7f12ad74 GR |
438 | } |
439 | ||
4eeb1fff | 440 | /** |
7f12ad74 GR |
441 | * i40evf_setup_tx_descriptors - Allocate the Tx descriptors |
442 | * @tx_ring: the tx ring to set up | |
443 | * | |
444 | * Return 0 on success, negative on error | |
445 | **/ | |
446 | int i40evf_setup_tx_descriptors(struct i40e_ring *tx_ring) | |
447 | { | |
448 | struct device *dev = tx_ring->dev; | |
449 | int bi_size; | |
450 | ||
451 | if (!dev) | |
452 | return -ENOMEM; | |
453 | ||
67c818a1 MW |
454 | /* warn if we are about to overwrite the pointer */ |
455 | WARN_ON(tx_ring->tx_bi); | |
7f12ad74 GR |
456 | bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count; |
457 | tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL); | |
458 | if (!tx_ring->tx_bi) | |
459 | goto err; | |
460 | ||
461 | /* round up to nearest 4K */ | |
462 | tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc); | |
1943d8ba JB |
463 | /* add u32 for head writeback, align after this takes care of |
464 | * guaranteeing this is at least one cache line in size | |
465 | */ | |
466 | tx_ring->size += sizeof(u32); | |
7f12ad74 GR |
467 | tx_ring->size = ALIGN(tx_ring->size, 4096); |
468 | tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size, | |
469 | &tx_ring->dma, GFP_KERNEL); | |
470 | if (!tx_ring->desc) { | |
471 | dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n", | |
472 | tx_ring->size); | |
473 | goto err; | |
474 | } | |
475 | ||
476 | tx_ring->next_to_use = 0; | |
477 | tx_ring->next_to_clean = 0; | |
478 | return 0; | |
479 | ||
480 | err: | |
481 | kfree(tx_ring->tx_bi); | |
482 | tx_ring->tx_bi = NULL; | |
483 | return -ENOMEM; | |
484 | } | |
485 | ||
486 | /** | |
487 | * i40evf_clean_rx_ring - Free Rx buffers | |
488 | * @rx_ring: ring to be cleaned | |
489 | **/ | |
490 | void i40evf_clean_rx_ring(struct i40e_ring *rx_ring) | |
491 | { | |
492 | struct device *dev = rx_ring->dev; | |
493 | struct i40e_rx_buffer *rx_bi; | |
494 | unsigned long bi_size; | |
495 | u16 i; | |
496 | ||
497 | /* ring already cleared, nothing to do */ | |
498 | if (!rx_ring->rx_bi) | |
499 | return; | |
500 | ||
a132af24 MW |
501 | if (ring_is_ps_enabled(rx_ring)) { |
502 | int bufsz = ALIGN(rx_ring->rx_hdr_len, 256) * rx_ring->count; | |
503 | ||
504 | rx_bi = &rx_ring->rx_bi[0]; | |
505 | if (rx_bi->hdr_buf) { | |
506 | dma_free_coherent(dev, | |
507 | bufsz, | |
508 | rx_bi->hdr_buf, | |
509 | rx_bi->dma); | |
510 | for (i = 0; i < rx_ring->count; i++) { | |
511 | rx_bi = &rx_ring->rx_bi[i]; | |
512 | rx_bi->dma = 0; | |
37a2973a | 513 | rx_bi->hdr_buf = NULL; |
a132af24 MW |
514 | } |
515 | } | |
516 | } | |
7f12ad74 GR |
517 | /* Free all the Rx ring sk_buffs */ |
518 | for (i = 0; i < rx_ring->count; i++) { | |
519 | rx_bi = &rx_ring->rx_bi[i]; | |
520 | if (rx_bi->dma) { | |
521 | dma_unmap_single(dev, | |
522 | rx_bi->dma, | |
523 | rx_ring->rx_buf_len, | |
524 | DMA_FROM_DEVICE); | |
525 | rx_bi->dma = 0; | |
526 | } | |
527 | if (rx_bi->skb) { | |
528 | dev_kfree_skb(rx_bi->skb); | |
529 | rx_bi->skb = NULL; | |
530 | } | |
531 | if (rx_bi->page) { | |
532 | if (rx_bi->page_dma) { | |
533 | dma_unmap_page(dev, | |
534 | rx_bi->page_dma, | |
f16704e5 | 535 | PAGE_SIZE, |
7f12ad74 GR |
536 | DMA_FROM_DEVICE); |
537 | rx_bi->page_dma = 0; | |
538 | } | |
539 | __free_page(rx_bi->page); | |
540 | rx_bi->page = NULL; | |
541 | rx_bi->page_offset = 0; | |
542 | } | |
543 | } | |
544 | ||
545 | bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count; | |
546 | memset(rx_ring->rx_bi, 0, bi_size); | |
547 | ||
548 | /* Zero out the descriptor ring */ | |
549 | memset(rx_ring->desc, 0, rx_ring->size); | |
550 | ||
551 | rx_ring->next_to_clean = 0; | |
552 | rx_ring->next_to_use = 0; | |
553 | } | |
554 | ||
555 | /** | |
556 | * i40evf_free_rx_resources - Free Rx resources | |
557 | * @rx_ring: ring to clean the resources from | |
558 | * | |
559 | * Free all receive software resources | |
560 | **/ | |
561 | void i40evf_free_rx_resources(struct i40e_ring *rx_ring) | |
562 | { | |
563 | i40evf_clean_rx_ring(rx_ring); | |
564 | kfree(rx_ring->rx_bi); | |
565 | rx_ring->rx_bi = NULL; | |
566 | ||
567 | if (rx_ring->desc) { | |
568 | dma_free_coherent(rx_ring->dev, rx_ring->size, | |
569 | rx_ring->desc, rx_ring->dma); | |
570 | rx_ring->desc = NULL; | |
571 | } | |
572 | } | |
573 | ||
a132af24 MW |
574 | /** |
575 | * i40evf_alloc_rx_headers - allocate rx header buffers | |
576 | * @rx_ring: ring to alloc buffers | |
577 | * | |
578 | * Allocate rx header buffers for the entire ring. As these are static, | |
579 | * this is only called when setting up a new ring. | |
580 | **/ | |
581 | void i40evf_alloc_rx_headers(struct i40e_ring *rx_ring) | |
582 | { | |
583 | struct device *dev = rx_ring->dev; | |
584 | struct i40e_rx_buffer *rx_bi; | |
585 | dma_addr_t dma; | |
586 | void *buffer; | |
587 | int buf_size; | |
588 | int i; | |
589 | ||
590 | if (rx_ring->rx_bi[0].hdr_buf) | |
591 | return; | |
592 | /* Make sure the buffers don't cross cache line boundaries. */ | |
593 | buf_size = ALIGN(rx_ring->rx_hdr_len, 256); | |
594 | buffer = dma_alloc_coherent(dev, buf_size * rx_ring->count, | |
595 | &dma, GFP_KERNEL); | |
596 | if (!buffer) | |
597 | return; | |
598 | for (i = 0; i < rx_ring->count; i++) { | |
599 | rx_bi = &rx_ring->rx_bi[i]; | |
600 | rx_bi->dma = dma + (i * buf_size); | |
601 | rx_bi->hdr_buf = buffer + (i * buf_size); | |
602 | } | |
603 | } | |
604 | ||
7f12ad74 GR |
605 | /** |
606 | * i40evf_setup_rx_descriptors - Allocate Rx descriptors | |
607 | * @rx_ring: Rx descriptor ring (for a specific queue) to setup | |
608 | * | |
609 | * Returns 0 on success, negative on failure | |
610 | **/ | |
611 | int i40evf_setup_rx_descriptors(struct i40e_ring *rx_ring) | |
612 | { | |
613 | struct device *dev = rx_ring->dev; | |
614 | int bi_size; | |
615 | ||
67c818a1 MW |
616 | /* warn if we are about to overwrite the pointer */ |
617 | WARN_ON(rx_ring->rx_bi); | |
7f12ad74 GR |
618 | bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count; |
619 | rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL); | |
620 | if (!rx_ring->rx_bi) | |
621 | goto err; | |
622 | ||
f217d6ca | 623 | u64_stats_init(&rx_ring->syncp); |
638702bd | 624 | |
7f12ad74 GR |
625 | /* Round up to nearest 4K */ |
626 | rx_ring->size = ring_is_16byte_desc_enabled(rx_ring) | |
627 | ? rx_ring->count * sizeof(union i40e_16byte_rx_desc) | |
628 | : rx_ring->count * sizeof(union i40e_32byte_rx_desc); | |
629 | rx_ring->size = ALIGN(rx_ring->size, 4096); | |
630 | rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size, | |
631 | &rx_ring->dma, GFP_KERNEL); | |
632 | ||
633 | if (!rx_ring->desc) { | |
634 | dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n", | |
635 | rx_ring->size); | |
636 | goto err; | |
637 | } | |
638 | ||
639 | rx_ring->next_to_clean = 0; | |
640 | rx_ring->next_to_use = 0; | |
641 | ||
642 | return 0; | |
643 | err: | |
644 | kfree(rx_ring->rx_bi); | |
645 | rx_ring->rx_bi = NULL; | |
646 | return -ENOMEM; | |
647 | } | |
648 | ||
649 | /** | |
650 | * i40e_release_rx_desc - Store the new tail and head values | |
651 | * @rx_ring: ring to bump | |
652 | * @val: new head index | |
653 | **/ | |
654 | static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val) | |
655 | { | |
656 | rx_ring->next_to_use = val; | |
657 | /* Force memory writes to complete before letting h/w | |
658 | * know there are new descriptors to fetch. (Only | |
659 | * applicable for weak-ordered memory model archs, | |
660 | * such as IA-64). | |
661 | */ | |
662 | wmb(); | |
663 | writel(val, rx_ring->tail); | |
664 | } | |
665 | ||
666 | /** | |
a132af24 MW |
667 | * i40evf_alloc_rx_buffers_ps - Replace used receive buffers; packet split |
668 | * @rx_ring: ring to place buffers on | |
669 | * @cleaned_count: number of buffers to replace | |
c2e245ab JB |
670 | * |
671 | * Returns true if any errors on allocation | |
a132af24 | 672 | **/ |
c2e245ab | 673 | bool i40evf_alloc_rx_buffers_ps(struct i40e_ring *rx_ring, u16 cleaned_count) |
a132af24 MW |
674 | { |
675 | u16 i = rx_ring->next_to_use; | |
676 | union i40e_rx_desc *rx_desc; | |
677 | struct i40e_rx_buffer *bi; | |
f16704e5 | 678 | const int current_node = numa_node_id(); |
a132af24 MW |
679 | |
680 | /* do nothing if no valid netdev defined */ | |
681 | if (!rx_ring->netdev || !cleaned_count) | |
c2e245ab | 682 | return false; |
a132af24 MW |
683 | |
684 | while (cleaned_count--) { | |
685 | rx_desc = I40E_RX_DESC(rx_ring, i); | |
686 | bi = &rx_ring->rx_bi[i]; | |
687 | ||
688 | if (bi->skb) /* desc is in use */ | |
689 | goto no_buffers; | |
f16704e5 MW |
690 | |
691 | /* If we've been moved to a different NUMA node, release the | |
692 | * page so we can get a new one on the current node. | |
693 | */ | |
694 | if (bi->page && page_to_nid(bi->page) != current_node) { | |
695 | dma_unmap_page(rx_ring->dev, | |
696 | bi->page_dma, | |
697 | PAGE_SIZE, | |
698 | DMA_FROM_DEVICE); | |
699 | __free_page(bi->page); | |
700 | bi->page = NULL; | |
701 | bi->page_dma = 0; | |
702 | rx_ring->rx_stats.realloc_count++; | |
703 | } else if (bi->page) { | |
704 | rx_ring->rx_stats.page_reuse_count++; | |
705 | } | |
706 | ||
a132af24 MW |
707 | if (!bi->page) { |
708 | bi->page = alloc_page(GFP_ATOMIC); | |
709 | if (!bi->page) { | |
710 | rx_ring->rx_stats.alloc_page_failed++; | |
711 | goto no_buffers; | |
712 | } | |
a132af24 MW |
713 | bi->page_dma = dma_map_page(rx_ring->dev, |
714 | bi->page, | |
f16704e5 MW |
715 | 0, |
716 | PAGE_SIZE, | |
a132af24 | 717 | DMA_FROM_DEVICE); |
f16704e5 | 718 | if (dma_mapping_error(rx_ring->dev, bi->page_dma)) { |
a132af24 | 719 | rx_ring->rx_stats.alloc_page_failed++; |
f16704e5 MW |
720 | __free_page(bi->page); |
721 | bi->page = NULL; | |
a132af24 | 722 | bi->page_dma = 0; |
f16704e5 | 723 | bi->page_offset = 0; |
a132af24 MW |
724 | goto no_buffers; |
725 | } | |
f16704e5 | 726 | bi->page_offset = 0; |
a132af24 MW |
727 | } |
728 | ||
a132af24 MW |
729 | /* Refresh the desc even if buffer_addrs didn't change |
730 | * because each write-back erases this info. | |
731 | */ | |
f16704e5 MW |
732 | rx_desc->read.pkt_addr = |
733 | cpu_to_le64(bi->page_dma + bi->page_offset); | |
a132af24 MW |
734 | rx_desc->read.hdr_addr = cpu_to_le64(bi->dma); |
735 | i++; | |
736 | if (i == rx_ring->count) | |
737 | i = 0; | |
738 | } | |
739 | ||
c2e245ab JB |
740 | if (rx_ring->next_to_use != i) |
741 | i40e_release_rx_desc(rx_ring, i); | |
742 | ||
743 | return false; | |
744 | ||
a132af24 MW |
745 | no_buffers: |
746 | if (rx_ring->next_to_use != i) | |
747 | i40e_release_rx_desc(rx_ring, i); | |
c2e245ab JB |
748 | |
749 | /* make sure to come back via polling to try again after | |
750 | * allocation failure | |
751 | */ | |
752 | return true; | |
a132af24 MW |
753 | } |
754 | ||
755 | /** | |
756 | * i40evf_alloc_rx_buffers_1buf - Replace used receive buffers; single buffer | |
7f12ad74 GR |
757 | * @rx_ring: ring to place buffers on |
758 | * @cleaned_count: number of buffers to replace | |
c2e245ab JB |
759 | * |
760 | * Returns true if any errors on allocation | |
7f12ad74 | 761 | **/ |
c2e245ab | 762 | bool i40evf_alloc_rx_buffers_1buf(struct i40e_ring *rx_ring, u16 cleaned_count) |
7f12ad74 GR |
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) | |
c2e245ab | 771 | return false; |
7f12ad74 GR |
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) { | |
dd1a5df8 JB |
779 | skb = __netdev_alloc_skb_ip_align(rx_ring->netdev, |
780 | rx_ring->rx_buf_len, | |
781 | GFP_ATOMIC | | |
782 | __GFP_NOWARN); | |
7f12ad74 GR |
783 | if (!skb) { |
784 | rx_ring->rx_stats.alloc_buff_failed++; | |
785 | goto no_buffers; | |
786 | } | |
787 | /* initialize queue mapping */ | |
788 | skb_record_rx_queue(skb, rx_ring->queue_index); | |
789 | bi->skb = skb; | |
790 | } | |
791 | ||
792 | if (!bi->dma) { | |
793 | bi->dma = dma_map_single(rx_ring->dev, | |
794 | skb->data, | |
795 | rx_ring->rx_buf_len, | |
796 | DMA_FROM_DEVICE); | |
797 | if (dma_mapping_error(rx_ring->dev, bi->dma)) { | |
798 | rx_ring->rx_stats.alloc_buff_failed++; | |
799 | bi->dma = 0; | |
c2e245ab JB |
800 | dev_kfree_skb(bi->skb); |
801 | bi->skb = NULL; | |
7f12ad74 GR |
802 | goto no_buffers; |
803 | } | |
804 | } | |
805 | ||
a132af24 MW |
806 | rx_desc->read.pkt_addr = cpu_to_le64(bi->dma); |
807 | rx_desc->read.hdr_addr = 0; | |
7f12ad74 GR |
808 | i++; |
809 | if (i == rx_ring->count) | |
810 | i = 0; | |
811 | } | |
812 | ||
c2e245ab JB |
813 | if (rx_ring->next_to_use != i) |
814 | i40e_release_rx_desc(rx_ring, i); | |
815 | ||
816 | return false; | |
817 | ||
7f12ad74 GR |
818 | no_buffers: |
819 | if (rx_ring->next_to_use != i) | |
820 | i40e_release_rx_desc(rx_ring, i); | |
c2e245ab JB |
821 | |
822 | /* make sure to come back via polling to try again after | |
823 | * allocation failure | |
824 | */ | |
825 | return true; | |
7f12ad74 GR |
826 | } |
827 | ||
828 | /** | |
829 | * i40e_receive_skb - Send a completed packet up the stack | |
830 | * @rx_ring: rx ring in play | |
831 | * @skb: packet to send up | |
832 | * @vlan_tag: vlan tag for packet | |
833 | **/ | |
834 | static void i40e_receive_skb(struct i40e_ring *rx_ring, | |
835 | struct sk_buff *skb, u16 vlan_tag) | |
836 | { | |
837 | struct i40e_q_vector *q_vector = rx_ring->q_vector; | |
7f12ad74 GR |
838 | |
839 | if (vlan_tag & VLAN_VID_MASK) | |
840 | __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag); | |
841 | ||
8b650359 | 842 | napi_gro_receive(&q_vector->napi, skb); |
7f12ad74 GR |
843 | } |
844 | ||
845 | /** | |
846 | * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum | |
847 | * @vsi: the VSI we care about | |
848 | * @skb: skb currently being received and modified | |
849 | * @rx_status: status value of last descriptor in packet | |
850 | * @rx_error: error value of last descriptor in packet | |
851 | * @rx_ptype: ptype value of last descriptor in packet | |
852 | **/ | |
853 | static inline void i40e_rx_checksum(struct i40e_vsi *vsi, | |
854 | struct sk_buff *skb, | |
855 | u32 rx_status, | |
856 | u32 rx_error, | |
857 | u16 rx_ptype) | |
858 | { | |
8a3c91cc JB |
859 | struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(rx_ptype); |
860 | bool ipv4 = false, ipv6 = false; | |
7f12ad74 GR |
861 | bool ipv4_tunnel, ipv6_tunnel; |
862 | __wsum rx_udp_csum; | |
7f12ad74 | 863 | struct iphdr *iph; |
8a3c91cc | 864 | __sum16 csum; |
7f12ad74 | 865 | |
f8faaa40 ASJ |
866 | ipv4_tunnel = (rx_ptype >= I40E_RX_PTYPE_GRENAT4_MAC_PAY3) && |
867 | (rx_ptype <= I40E_RX_PTYPE_GRENAT4_MACVLAN_IPV6_ICMP_PAY4); | |
868 | ipv6_tunnel = (rx_ptype >= I40E_RX_PTYPE_GRENAT6_MAC_PAY3) && | |
869 | (rx_ptype <= I40E_RX_PTYPE_GRENAT6_MACVLAN_IPV6_ICMP_PAY4); | |
7f12ad74 | 870 | |
7f12ad74 GR |
871 | skb->ip_summed = CHECKSUM_NONE; |
872 | ||
873 | /* Rx csum enabled and ip headers found? */ | |
8a3c91cc JB |
874 | if (!(vsi->netdev->features & NETIF_F_RXCSUM)) |
875 | return; | |
876 | ||
877 | /* did the hardware decode the packet and checksum? */ | |
41a1d04b | 878 | if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT))) |
8a3c91cc JB |
879 | return; |
880 | ||
881 | /* both known and outer_ip must be set for the below code to work */ | |
882 | if (!(decoded.known && decoded.outer_ip)) | |
7f12ad74 GR |
883 | return; |
884 | ||
8a3c91cc JB |
885 | if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP && |
886 | decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4) | |
887 | ipv4 = true; | |
888 | else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP && | |
889 | decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6) | |
890 | ipv6 = true; | |
891 | ||
892 | if (ipv4 && | |
41a1d04b JB |
893 | (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) | |
894 | BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT)))) | |
8a3c91cc JB |
895 | goto checksum_fail; |
896 | ||
ddf1d0d7 | 897 | /* likely incorrect csum if alternate IP extension headers found */ |
8a3c91cc | 898 | if (ipv6 && |
41a1d04b | 899 | rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT)) |
8a3c91cc | 900 | /* don't increment checksum err here, non-fatal err */ |
7f12ad74 GR |
901 | return; |
902 | ||
8a3c91cc | 903 | /* there was some L4 error, count error and punt packet to the stack */ |
41a1d04b | 904 | if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT)) |
8a3c91cc JB |
905 | goto checksum_fail; |
906 | ||
907 | /* handle packets that were not able to be checksummed due | |
908 | * to arrival speed, in this case the stack can compute | |
909 | * the csum. | |
910 | */ | |
41a1d04b | 911 | if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT)) |
7f12ad74 | 912 | return; |
7f12ad74 | 913 | |
8a3c91cc JB |
914 | /* If VXLAN traffic has an outer UDPv4 checksum we need to check |
915 | * it in the driver, hardware does not do it for us. | |
916 | * Since L3L4P bit was set we assume a valid IHL value (>=5) | |
917 | * so the total length of IPv4 header is IHL*4 bytes | |
918 | * The UDP_0 bit *may* bet set if the *inner* header is UDP | |
919 | */ | |
818f2e7b | 920 | if (ipv4_tunnel) { |
7f12ad74 GR |
921 | skb->transport_header = skb->mac_header + |
922 | sizeof(struct ethhdr) + | |
923 | (ip_hdr(skb)->ihl * 4); | |
924 | ||
925 | /* Add 4 bytes for VLAN tagged packets */ | |
926 | skb->transport_header += (skb->protocol == htons(ETH_P_8021Q) || | |
927 | skb->protocol == htons(ETH_P_8021AD)) | |
928 | ? VLAN_HLEN : 0; | |
929 | ||
818f2e7b ASJ |
930 | if ((ip_hdr(skb)->protocol == IPPROTO_UDP) && |
931 | (udp_hdr(skb)->check != 0)) { | |
932 | rx_udp_csum = udp_csum(skb); | |
933 | iph = ip_hdr(skb); | |
934 | csum = csum_tcpudp_magic(iph->saddr, iph->daddr, | |
935 | (skb->len - | |
936 | skb_transport_offset(skb)), | |
937 | IPPROTO_UDP, rx_udp_csum); | |
7f12ad74 | 938 | |
818f2e7b ASJ |
939 | if (udp_hdr(skb)->check != csum) |
940 | goto checksum_fail; | |
941 | ||
942 | } /* else its GRE and so no outer UDP header */ | |
7f12ad74 GR |
943 | } |
944 | ||
945 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
407fa085 | 946 | skb->csum_level = ipv4_tunnel || ipv6_tunnel; |
8a3c91cc JB |
947 | |
948 | return; | |
949 | ||
950 | checksum_fail: | |
951 | vsi->back->hw_csum_rx_error++; | |
7f12ad74 GR |
952 | } |
953 | ||
954 | /** | |
857942fd | 955 | * i40e_ptype_to_htype - get a hash type |
206812b5 JB |
956 | * @ptype: the ptype value from the descriptor |
957 | * | |
958 | * Returns a hash type to be used by skb_set_hash | |
959 | **/ | |
857942fd | 960 | static inline enum pkt_hash_types i40e_ptype_to_htype(u8 ptype) |
206812b5 JB |
961 | { |
962 | struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype); | |
963 | ||
964 | if (!decoded.known) | |
965 | return PKT_HASH_TYPE_NONE; | |
966 | ||
967 | if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP && | |
968 | decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4) | |
969 | return PKT_HASH_TYPE_L4; | |
970 | else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP && | |
971 | decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3) | |
972 | return PKT_HASH_TYPE_L3; | |
973 | else | |
974 | return PKT_HASH_TYPE_L2; | |
975 | } | |
976 | ||
857942fd ASJ |
977 | /** |
978 | * i40e_rx_hash - set the hash value in the skb | |
979 | * @ring: descriptor ring | |
980 | * @rx_desc: specific descriptor | |
981 | **/ | |
982 | static inline void i40e_rx_hash(struct i40e_ring *ring, | |
983 | union i40e_rx_desc *rx_desc, | |
984 | struct sk_buff *skb, | |
985 | u8 rx_ptype) | |
986 | { | |
987 | u32 hash; | |
988 | const __le64 rss_mask = | |
989 | cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH << | |
990 | I40E_RX_DESC_STATUS_FLTSTAT_SHIFT); | |
991 | ||
992 | if (ring->netdev->features & NETIF_F_RXHASH) | |
993 | return; | |
994 | ||
995 | if ((rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask) { | |
996 | hash = le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss); | |
997 | skb_set_hash(skb, hash, i40e_ptype_to_htype(rx_ptype)); | |
998 | } | |
999 | } | |
1000 | ||
7f12ad74 | 1001 | /** |
a132af24 | 1002 | * i40e_clean_rx_irq_ps - Reclaim resources after receive; packet split |
7f12ad74 GR |
1003 | * @rx_ring: rx ring to clean |
1004 | * @budget: how many cleans we're allowed | |
1005 | * | |
1006 | * Returns true if there's any budget left (e.g. the clean is finished) | |
1007 | **/ | |
c2e245ab | 1008 | static int i40e_clean_rx_irq_ps(struct i40e_ring *rx_ring, const int budget) |
7f12ad74 GR |
1009 | { |
1010 | unsigned int total_rx_bytes = 0, total_rx_packets = 0; | |
1011 | u16 rx_packet_len, rx_header_len, rx_sph, rx_hbo; | |
1012 | u16 cleaned_count = I40E_DESC_UNUSED(rx_ring); | |
7f12ad74 GR |
1013 | struct i40e_vsi *vsi = rx_ring->vsi; |
1014 | u16 i = rx_ring->next_to_clean; | |
1015 | union i40e_rx_desc *rx_desc; | |
1016 | u32 rx_error, rx_status; | |
c2e245ab | 1017 | bool failure = false; |
206812b5 | 1018 | u8 rx_ptype; |
7f12ad74 | 1019 | u64 qword; |
f16704e5 | 1020 | u32 copysize; |
7f12ad74 | 1021 | |
a132af24 | 1022 | do { |
7f12ad74 GR |
1023 | struct i40e_rx_buffer *rx_bi; |
1024 | struct sk_buff *skb; | |
1025 | u16 vlan_tag; | |
a132af24 MW |
1026 | /* return some buffers to hardware, one at a time is too slow */ |
1027 | if (cleaned_count >= I40E_RX_BUFFER_WRITE) { | |
c2e245ab JB |
1028 | failure = failure || |
1029 | i40evf_alloc_rx_buffers_ps(rx_ring, | |
1030 | cleaned_count); | |
a132af24 MW |
1031 | cleaned_count = 0; |
1032 | } | |
1033 | ||
1034 | i = rx_ring->next_to_clean; | |
1035 | rx_desc = I40E_RX_DESC(rx_ring, i); | |
1036 | qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len); | |
1037 | rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >> | |
1038 | I40E_RXD_QW1_STATUS_SHIFT; | |
1039 | ||
41a1d04b | 1040 | if (!(rx_status & BIT(I40E_RX_DESC_STATUS_DD_SHIFT))) |
a132af24 MW |
1041 | break; |
1042 | ||
1043 | /* This memory barrier is needed to keep us from reading | |
1044 | * any other fields out of the rx_desc until we know the | |
1045 | * DD bit is set. | |
1046 | */ | |
67317166 | 1047 | dma_rmb(); |
f16704e5 MW |
1048 | /* sync header buffer for reading */ |
1049 | dma_sync_single_range_for_cpu(rx_ring->dev, | |
1050 | rx_ring->rx_bi[0].dma, | |
1051 | i * rx_ring->rx_hdr_len, | |
1052 | rx_ring->rx_hdr_len, | |
1053 | DMA_FROM_DEVICE); | |
7f12ad74 GR |
1054 | rx_bi = &rx_ring->rx_bi[i]; |
1055 | skb = rx_bi->skb; | |
a132af24 | 1056 | if (likely(!skb)) { |
dd1a5df8 JB |
1057 | skb = __netdev_alloc_skb_ip_align(rx_ring->netdev, |
1058 | rx_ring->rx_hdr_len, | |
1059 | GFP_ATOMIC | | |
1060 | __GFP_NOWARN); | |
8b6ed9c2 | 1061 | if (!skb) { |
a132af24 | 1062 | rx_ring->rx_stats.alloc_buff_failed++; |
c2e245ab | 1063 | failure = true; |
8b6ed9c2 JB |
1064 | break; |
1065 | } | |
1066 | ||
a132af24 MW |
1067 | /* initialize queue mapping */ |
1068 | skb_record_rx_queue(skb, rx_ring->queue_index); | |
1069 | /* we are reusing so sync this buffer for CPU use */ | |
1070 | dma_sync_single_range_for_cpu(rx_ring->dev, | |
3578fa0a JB |
1071 | rx_ring->rx_bi[0].dma, |
1072 | i * rx_ring->rx_hdr_len, | |
a132af24 MW |
1073 | rx_ring->rx_hdr_len, |
1074 | DMA_FROM_DEVICE); | |
1075 | } | |
7f12ad74 GR |
1076 | rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >> |
1077 | I40E_RXD_QW1_LENGTH_PBUF_SHIFT; | |
1078 | rx_header_len = (qword & I40E_RXD_QW1_LENGTH_HBUF_MASK) >> | |
1079 | I40E_RXD_QW1_LENGTH_HBUF_SHIFT; | |
1080 | rx_sph = (qword & I40E_RXD_QW1_LENGTH_SPH_MASK) >> | |
1081 | I40E_RXD_QW1_LENGTH_SPH_SHIFT; | |
1082 | ||
1083 | rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >> | |
1084 | I40E_RXD_QW1_ERROR_SHIFT; | |
41a1d04b JB |
1085 | rx_hbo = rx_error & BIT(I40E_RX_DESC_ERROR_HBO_SHIFT); |
1086 | rx_error &= ~BIT(I40E_RX_DESC_ERROR_HBO_SHIFT); | |
7f12ad74 GR |
1087 | |
1088 | rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >> | |
1089 | I40E_RXD_QW1_PTYPE_SHIFT; | |
f16704e5 MW |
1090 | /* sync half-page for reading */ |
1091 | dma_sync_single_range_for_cpu(rx_ring->dev, | |
1092 | rx_bi->page_dma, | |
1093 | rx_bi->page_offset, | |
1094 | PAGE_SIZE / 2, | |
1095 | DMA_FROM_DEVICE); | |
1096 | prefetch(page_address(rx_bi->page) + rx_bi->page_offset); | |
7f12ad74 | 1097 | rx_bi->skb = NULL; |
a132af24 | 1098 | cleaned_count++; |
f16704e5 | 1099 | copysize = 0; |
a132af24 MW |
1100 | if (rx_hbo || rx_sph) { |
1101 | int len; | |
6995b36c | 1102 | |
7f12ad74 GR |
1103 | if (rx_hbo) |
1104 | len = I40E_RX_HDR_SIZE; | |
7f12ad74 | 1105 | else |
a132af24 MW |
1106 | len = rx_header_len; |
1107 | memcpy(__skb_put(skb, len), rx_bi->hdr_buf, len); | |
1108 | } else if (skb->len == 0) { | |
1109 | int len; | |
f16704e5 MW |
1110 | unsigned char *va = page_address(rx_bi->page) + |
1111 | rx_bi->page_offset; | |
a132af24 | 1112 | |
f16704e5 MW |
1113 | len = min(rx_packet_len, rx_ring->rx_hdr_len); |
1114 | memcpy(__skb_put(skb, len), va, len); | |
1115 | copysize = len; | |
a132af24 | 1116 | rx_packet_len -= len; |
7f12ad74 | 1117 | } |
7f12ad74 | 1118 | /* Get the rest of the data if this was a header split */ |
a132af24 | 1119 | if (rx_packet_len) { |
f16704e5 MW |
1120 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, |
1121 | rx_bi->page, | |
1122 | rx_bi->page_offset + copysize, | |
1123 | rx_packet_len, I40E_RXBUFFER_2048); | |
1124 | ||
1125 | get_page(rx_bi->page); | |
1126 | /* switch to the other half-page here; the allocation | |
1127 | * code programs the right addr into HW. If we haven't | |
1128 | * used this half-page, the address won't be changed, | |
1129 | * and HW can just use it next time through. | |
1130 | */ | |
1131 | rx_bi->page_offset ^= PAGE_SIZE / 2; | |
1132 | /* If the page count is more than 2, then both halves | |
1133 | * of the page are used and we need to free it. Do it | |
1134 | * here instead of in the alloc code. Otherwise one | |
1135 | * of the half-pages might be released between now and | |
1136 | * then, and we wouldn't know which one to use. | |
1137 | */ | |
1138 | if (page_count(rx_bi->page) > 2) { | |
1139 | dma_unmap_page(rx_ring->dev, | |
1140 | rx_bi->page_dma, | |
1141 | PAGE_SIZE, | |
1142 | DMA_FROM_DEVICE); | |
1143 | __free_page(rx_bi->page); | |
7f12ad74 | 1144 | rx_bi->page = NULL; |
f16704e5 MW |
1145 | rx_bi->page_dma = 0; |
1146 | rx_ring->rx_stats.realloc_count++; | |
1147 | } | |
7f12ad74 | 1148 | |
7f12ad74 | 1149 | } |
a132af24 | 1150 | I40E_RX_INCREMENT(rx_ring, i); |
7f12ad74 GR |
1151 | |
1152 | if (unlikely( | |
41a1d04b | 1153 | !(rx_status & BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)))) { |
7f12ad74 GR |
1154 | struct i40e_rx_buffer *next_buffer; |
1155 | ||
1156 | next_buffer = &rx_ring->rx_bi[i]; | |
a132af24 | 1157 | next_buffer->skb = skb; |
7f12ad74 | 1158 | rx_ring->rx_stats.non_eop_descs++; |
a132af24 | 1159 | continue; |
7f12ad74 GR |
1160 | } |
1161 | ||
1162 | /* ERR_MASK will only have valid bits if EOP set */ | |
41a1d04b | 1163 | if (unlikely(rx_error & BIT(I40E_RX_DESC_ERROR_RXE_SHIFT))) { |
7f12ad74 | 1164 | dev_kfree_skb_any(skb); |
a132af24 | 1165 | continue; |
7f12ad74 GR |
1166 | } |
1167 | ||
857942fd ASJ |
1168 | i40e_rx_hash(rx_ring, rx_desc, skb, rx_ptype); |
1169 | ||
7f12ad74 GR |
1170 | /* probably a little skewed due to removing CRC */ |
1171 | total_rx_bytes += skb->len; | |
1172 | total_rx_packets++; | |
1173 | ||
1174 | skb->protocol = eth_type_trans(skb, rx_ring->netdev); | |
1175 | ||
1176 | i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype); | |
1177 | ||
41a1d04b | 1178 | vlan_tag = rx_status & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT) |
7f12ad74 GR |
1179 | ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1) |
1180 | : 0; | |
a132af24 MW |
1181 | #ifdef I40E_FCOE |
1182 | if (!i40e_fcoe_handle_offload(rx_ring, rx_desc, skb)) { | |
1183 | dev_kfree_skb_any(skb); | |
1184 | continue; | |
1185 | } | |
1186 | #endif | |
7f12ad74 GR |
1187 | i40e_receive_skb(rx_ring, skb, vlan_tag); |
1188 | ||
7f12ad74 | 1189 | rx_desc->wb.qword1.status_error_len = 0; |
7f12ad74 | 1190 | |
a132af24 MW |
1191 | } while (likely(total_rx_packets < budget)); |
1192 | ||
1193 | u64_stats_update_begin(&rx_ring->syncp); | |
1194 | rx_ring->stats.packets += total_rx_packets; | |
1195 | rx_ring->stats.bytes += total_rx_bytes; | |
1196 | u64_stats_update_end(&rx_ring->syncp); | |
1197 | rx_ring->q_vector->rx.total_packets += total_rx_packets; | |
1198 | rx_ring->q_vector->rx.total_bytes += total_rx_bytes; | |
1199 | ||
c2e245ab | 1200 | return failure ? budget : total_rx_packets; |
a132af24 MW |
1201 | } |
1202 | ||
1203 | /** | |
1204 | * i40e_clean_rx_irq_1buf - Reclaim resources after receive; single buffer | |
1205 | * @rx_ring: rx ring to clean | |
1206 | * @budget: how many cleans we're allowed | |
1207 | * | |
1208 | * Returns number of packets cleaned | |
1209 | **/ | |
1210 | static int i40e_clean_rx_irq_1buf(struct i40e_ring *rx_ring, int budget) | |
1211 | { | |
1212 | unsigned int total_rx_bytes = 0, total_rx_packets = 0; | |
1213 | u16 cleaned_count = I40E_DESC_UNUSED(rx_ring); | |
1214 | struct i40e_vsi *vsi = rx_ring->vsi; | |
1215 | union i40e_rx_desc *rx_desc; | |
1216 | u32 rx_error, rx_status; | |
1217 | u16 rx_packet_len; | |
c2e245ab | 1218 | bool failure = false; |
a132af24 MW |
1219 | u8 rx_ptype; |
1220 | u64 qword; | |
1221 | u16 i; | |
1222 | ||
1223 | do { | |
1224 | struct i40e_rx_buffer *rx_bi; | |
1225 | struct sk_buff *skb; | |
1226 | u16 vlan_tag; | |
7f12ad74 GR |
1227 | /* return some buffers to hardware, one at a time is too slow */ |
1228 | if (cleaned_count >= I40E_RX_BUFFER_WRITE) { | |
c2e245ab JB |
1229 | failure = failure || |
1230 | i40evf_alloc_rx_buffers_1buf(rx_ring, | |
1231 | cleaned_count); | |
7f12ad74 GR |
1232 | cleaned_count = 0; |
1233 | } | |
1234 | ||
a132af24 MW |
1235 | i = rx_ring->next_to_clean; |
1236 | rx_desc = I40E_RX_DESC(rx_ring, i); | |
7f12ad74 GR |
1237 | qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len); |
1238 | rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >> | |
a132af24 MW |
1239 | I40E_RXD_QW1_STATUS_SHIFT; |
1240 | ||
41a1d04b | 1241 | if (!(rx_status & BIT(I40E_RX_DESC_STATUS_DD_SHIFT))) |
a132af24 MW |
1242 | break; |
1243 | ||
1244 | /* This memory barrier is needed to keep us from reading | |
1245 | * any other fields out of the rx_desc until we know the | |
1246 | * DD bit is set. | |
1247 | */ | |
67317166 | 1248 | dma_rmb(); |
a132af24 MW |
1249 | |
1250 | rx_bi = &rx_ring->rx_bi[i]; | |
1251 | skb = rx_bi->skb; | |
1252 | prefetch(skb->data); | |
1253 | ||
1254 | rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >> | |
1255 | I40E_RXD_QW1_LENGTH_PBUF_SHIFT; | |
1256 | ||
1257 | rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >> | |
1258 | I40E_RXD_QW1_ERROR_SHIFT; | |
41a1d04b | 1259 | rx_error &= ~BIT(I40E_RX_DESC_ERROR_HBO_SHIFT); |
a132af24 MW |
1260 | |
1261 | rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >> | |
1262 | I40E_RXD_QW1_PTYPE_SHIFT; | |
1263 | rx_bi->skb = NULL; | |
1264 | cleaned_count++; | |
1265 | ||
1266 | /* Get the header and possibly the whole packet | |
1267 | * If this is an skb from previous receive dma will be 0 | |
1268 | */ | |
1269 | skb_put(skb, rx_packet_len); | |
1270 | dma_unmap_single(rx_ring->dev, rx_bi->dma, rx_ring->rx_buf_len, | |
1271 | DMA_FROM_DEVICE); | |
1272 | rx_bi->dma = 0; | |
1273 | ||
1274 | I40E_RX_INCREMENT(rx_ring, i); | |
1275 | ||
1276 | if (unlikely( | |
41a1d04b | 1277 | !(rx_status & BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)))) { |
a132af24 MW |
1278 | rx_ring->rx_stats.non_eop_descs++; |
1279 | continue; | |
1280 | } | |
1281 | ||
1282 | /* ERR_MASK will only have valid bits if EOP set */ | |
41a1d04b | 1283 | if (unlikely(rx_error & BIT(I40E_RX_DESC_ERROR_RXE_SHIFT))) { |
a132af24 | 1284 | dev_kfree_skb_any(skb); |
a132af24 MW |
1285 | continue; |
1286 | } | |
1287 | ||
857942fd | 1288 | i40e_rx_hash(rx_ring, rx_desc, skb, rx_ptype); |
a132af24 MW |
1289 | /* probably a little skewed due to removing CRC */ |
1290 | total_rx_bytes += skb->len; | |
1291 | total_rx_packets++; | |
1292 | ||
1293 | skb->protocol = eth_type_trans(skb, rx_ring->netdev); | |
1294 | ||
1295 | i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype); | |
1296 | ||
41a1d04b | 1297 | vlan_tag = rx_status & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT) |
a132af24 MW |
1298 | ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1) |
1299 | : 0; | |
1300 | i40e_receive_skb(rx_ring, skb, vlan_tag); | |
1301 | ||
a132af24 MW |
1302 | rx_desc->wb.qword1.status_error_len = 0; |
1303 | } while (likely(total_rx_packets < budget)); | |
7f12ad74 | 1304 | |
7f12ad74 GR |
1305 | u64_stats_update_begin(&rx_ring->syncp); |
1306 | rx_ring->stats.packets += total_rx_packets; | |
1307 | rx_ring->stats.bytes += total_rx_bytes; | |
1308 | u64_stats_update_end(&rx_ring->syncp); | |
1309 | rx_ring->q_vector->rx.total_packets += total_rx_packets; | |
1310 | rx_ring->q_vector->rx.total_bytes += total_rx_bytes; | |
1311 | ||
c2e245ab | 1312 | return failure ? budget : total_rx_packets; |
7f12ad74 GR |
1313 | } |
1314 | ||
8f5e39ce JB |
1315 | static u32 i40e_buildreg_itr(const int type, const u16 itr) |
1316 | { | |
1317 | u32 val; | |
1318 | ||
1319 | val = I40E_VFINT_DYN_CTLN1_INTENA_MASK | | |
40d72a50 JB |
1320 | /* Don't clear PBA because that can cause lost interrupts that |
1321 | * came in while we were cleaning/polling | |
1322 | */ | |
8f5e39ce JB |
1323 | (type << I40E_VFINT_DYN_CTLN1_ITR_INDX_SHIFT) | |
1324 | (itr << I40E_VFINT_DYN_CTLN1_INTERVAL_SHIFT); | |
1325 | ||
1326 | return val; | |
1327 | } | |
1328 | ||
1329 | /* a small macro to shorten up some long lines */ | |
1330 | #define INTREG I40E_VFINT_DYN_CTLN1 | |
1331 | ||
de32e3ef CW |
1332 | /** |
1333 | * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt | |
1334 | * @vsi: the VSI we care about | |
1335 | * @q_vector: q_vector for which itr is being updated and interrupt enabled | |
1336 | * | |
1337 | **/ | |
1338 | static inline void i40e_update_enable_itr(struct i40e_vsi *vsi, | |
1339 | struct i40e_q_vector *q_vector) | |
1340 | { | |
1341 | struct i40e_hw *hw = &vsi->back->hw; | |
8f5e39ce JB |
1342 | bool rx = false, tx = false; |
1343 | u32 rxval, txval; | |
de32e3ef | 1344 | int vector; |
de32e3ef CW |
1345 | |
1346 | vector = (q_vector->v_idx + vsi->base_vector); | |
ee2319cf JB |
1347 | |
1348 | /* avoid dynamic calculation if in countdown mode OR if | |
1349 | * all dynamic is disabled | |
1350 | */ | |
8f5e39ce JB |
1351 | rxval = txval = i40e_buildreg_itr(I40E_ITR_NONE, 0); |
1352 | ||
ee2319cf JB |
1353 | if (q_vector->itr_countdown > 0 || |
1354 | (!ITR_IS_DYNAMIC(vsi->rx_itr_setting) && | |
1355 | !ITR_IS_DYNAMIC(vsi->tx_itr_setting))) { | |
1356 | goto enable_int; | |
1357 | } | |
1358 | ||
de32e3ef | 1359 | if (ITR_IS_DYNAMIC(vsi->rx_itr_setting)) { |
8f5e39ce JB |
1360 | rx = i40e_set_new_dynamic_itr(&q_vector->rx); |
1361 | rxval = i40e_buildreg_itr(I40E_RX_ITR, q_vector->rx.itr); | |
de32e3ef | 1362 | } |
4eeb1fff | 1363 | |
de32e3ef | 1364 | if (ITR_IS_DYNAMIC(vsi->tx_itr_setting)) { |
8f5e39ce JB |
1365 | tx = i40e_set_new_dynamic_itr(&q_vector->tx); |
1366 | txval = i40e_buildreg_itr(I40E_TX_ITR, q_vector->tx.itr); | |
1367 | } | |
4eeb1fff | 1368 | |
8f5e39ce JB |
1369 | if (rx || tx) { |
1370 | /* get the higher of the two ITR adjustments and | |
1371 | * use the same value for both ITR registers | |
1372 | * when in adaptive mode (Rx and/or Tx) | |
1373 | */ | |
1374 | u16 itr = max(q_vector->tx.itr, q_vector->rx.itr); | |
1375 | ||
1376 | q_vector->tx.itr = q_vector->rx.itr = itr; | |
1377 | txval = i40e_buildreg_itr(I40E_TX_ITR, itr); | |
1378 | tx = true; | |
1379 | rxval = i40e_buildreg_itr(I40E_RX_ITR, itr); | |
1380 | rx = true; | |
de32e3ef | 1381 | } |
8f5e39ce JB |
1382 | |
1383 | /* only need to enable the interrupt once, but need | |
1384 | * to possibly update both ITR values | |
1385 | */ | |
1386 | if (rx) { | |
1387 | /* set the INTENA_MSK_MASK so that this first write | |
1388 | * won't actually enable the interrupt, instead just | |
1389 | * updating the ITR (it's bit 31 PF and VF) | |
1390 | */ | |
1391 | rxval |= BIT(31); | |
1392 | /* don't check _DOWN because interrupt isn't being enabled */ | |
1393 | wr32(hw, INTREG(vector - 1), rxval); | |
1394 | } | |
1395 | ||
ee2319cf | 1396 | enable_int: |
8f5e39ce JB |
1397 | if (!test_bit(__I40E_DOWN, &vsi->state)) |
1398 | wr32(hw, INTREG(vector - 1), txval); | |
ee2319cf JB |
1399 | |
1400 | if (q_vector->itr_countdown) | |
1401 | q_vector->itr_countdown--; | |
1402 | else | |
1403 | q_vector->itr_countdown = ITR_COUNTDOWN_START; | |
de32e3ef CW |
1404 | } |
1405 | ||
7f12ad74 GR |
1406 | /** |
1407 | * i40evf_napi_poll - NAPI polling Rx/Tx cleanup routine | |
1408 | * @napi: napi struct with our devices info in it | |
1409 | * @budget: amount of work driver is allowed to do this pass, in packets | |
1410 | * | |
1411 | * This function will clean all queues associated with a q_vector. | |
1412 | * | |
1413 | * Returns the amount of work done | |
1414 | **/ | |
1415 | int i40evf_napi_poll(struct napi_struct *napi, int budget) | |
1416 | { | |
1417 | struct i40e_q_vector *q_vector = | |
1418 | container_of(napi, struct i40e_q_vector, napi); | |
1419 | struct i40e_vsi *vsi = q_vector->vsi; | |
1420 | struct i40e_ring *ring; | |
1421 | bool clean_complete = true; | |
c29af37f | 1422 | bool arm_wb = false; |
7f12ad74 | 1423 | int budget_per_ring; |
32b3e08f | 1424 | int work_done = 0; |
7f12ad74 GR |
1425 | |
1426 | if (test_bit(__I40E_DOWN, &vsi->state)) { | |
1427 | napi_complete(napi); | |
1428 | return 0; | |
1429 | } | |
1430 | ||
1431 | /* Since the actual Tx work is minimal, we can give the Tx a larger | |
1432 | * budget and be more aggressive about cleaning up the Tx descriptors. | |
1433 | */ | |
c29af37f | 1434 | i40e_for_each_ring(ring, q_vector->tx) { |
1a36d7fa MW |
1435 | clean_complete = clean_complete && |
1436 | i40e_clean_tx_irq(ring, vsi->work_limit); | |
44cdb791 | 1437 | arm_wb = arm_wb || ring->arm_wb; |
0deda868 | 1438 | ring->arm_wb = false; |
c29af37f | 1439 | } |
7f12ad74 | 1440 | |
c67caceb AD |
1441 | /* Handle case where we are called by netpoll with a budget of 0 */ |
1442 | if (budget <= 0) | |
1443 | goto tx_only; | |
1444 | ||
7f12ad74 GR |
1445 | /* We attempt to distribute budget to each Rx queue fairly, but don't |
1446 | * allow the budget to go below 1 because that would exit polling early. | |
1447 | */ | |
1448 | budget_per_ring = max(budget/q_vector->num_ringpairs, 1); | |
1449 | ||
a132af24 | 1450 | i40e_for_each_ring(ring, q_vector->rx) { |
32b3e08f JB |
1451 | int cleaned; |
1452 | ||
a132af24 MW |
1453 | if (ring_is_ps_enabled(ring)) |
1454 | cleaned = i40e_clean_rx_irq_ps(ring, budget_per_ring); | |
1455 | else | |
1456 | cleaned = i40e_clean_rx_irq_1buf(ring, budget_per_ring); | |
32b3e08f JB |
1457 | |
1458 | work_done += cleaned; | |
a132af24 | 1459 | /* if we didn't clean as many as budgeted, we must be done */ |
1a36d7fa | 1460 | clean_complete = clean_complete && (budget_per_ring > cleaned); |
a132af24 | 1461 | } |
7f12ad74 GR |
1462 | |
1463 | /* If work not completed, return budget and polling will return */ | |
c29af37f | 1464 | if (!clean_complete) { |
c67caceb | 1465 | tx_only: |
164c9f54 ASJ |
1466 | if (arm_wb) { |
1467 | q_vector->tx.ring[0].tx_stats.tx_force_wb++; | |
ecc6a239 | 1468 | i40e_enable_wb_on_itr(vsi, q_vector); |
164c9f54 | 1469 | } |
7f12ad74 | 1470 | return budget; |
c29af37f | 1471 | } |
7f12ad74 | 1472 | |
8e0764b4 ASJ |
1473 | if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR) |
1474 | q_vector->arm_wb_state = false; | |
1475 | ||
7f12ad74 | 1476 | /* Work is done so exit the polling mode and re-enable the interrupt */ |
32b3e08f | 1477 | napi_complete_done(napi, work_done); |
de32e3ef | 1478 | i40e_update_enable_itr(vsi, q_vector); |
7f12ad74 GR |
1479 | return 0; |
1480 | } | |
1481 | ||
1482 | /** | |
3e587cf3 | 1483 | * i40evf_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW |
7f12ad74 GR |
1484 | * @skb: send buffer |
1485 | * @tx_ring: ring to send buffer on | |
1486 | * @flags: the tx flags to be set | |
1487 | * | |
1488 | * Checks the skb and set up correspondingly several generic transmit flags | |
1489 | * related to VLAN tagging for the HW, such as VLAN, DCB, etc. | |
1490 | * | |
1491 | * Returns error code indicate the frame should be dropped upon error and the | |
1492 | * otherwise returns 0 to indicate the flags has been set properly. | |
1493 | **/ | |
3e587cf3 JB |
1494 | static inline int i40evf_tx_prepare_vlan_flags(struct sk_buff *skb, |
1495 | struct i40e_ring *tx_ring, | |
1496 | u32 *flags) | |
7f12ad74 GR |
1497 | { |
1498 | __be16 protocol = skb->protocol; | |
1499 | u32 tx_flags = 0; | |
1500 | ||
31eaaccf GR |
1501 | if (protocol == htons(ETH_P_8021Q) && |
1502 | !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) { | |
1503 | /* When HW VLAN acceleration is turned off by the user the | |
1504 | * stack sets the protocol to 8021q so that the driver | |
1505 | * can take any steps required to support the SW only | |
1506 | * VLAN handling. In our case the driver doesn't need | |
1507 | * to take any further steps so just set the protocol | |
1508 | * to the encapsulated ethertype. | |
1509 | */ | |
1510 | skb->protocol = vlan_get_protocol(skb); | |
1511 | goto out; | |
1512 | } | |
1513 | ||
7f12ad74 | 1514 | /* if we have a HW VLAN tag being added, default to the HW one */ |
df8a39de JP |
1515 | if (skb_vlan_tag_present(skb)) { |
1516 | tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT; | |
7f12ad74 GR |
1517 | tx_flags |= I40E_TX_FLAGS_HW_VLAN; |
1518 | /* else if it is a SW VLAN, check the next protocol and store the tag */ | |
1519 | } else if (protocol == htons(ETH_P_8021Q)) { | |
1520 | struct vlan_hdr *vhdr, _vhdr; | |
6995b36c | 1521 | |
7f12ad74 GR |
1522 | vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr); |
1523 | if (!vhdr) | |
1524 | return -EINVAL; | |
1525 | ||
1526 | protocol = vhdr->h_vlan_encapsulated_proto; | |
1527 | tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT; | |
1528 | tx_flags |= I40E_TX_FLAGS_SW_VLAN; | |
1529 | } | |
1530 | ||
31eaaccf | 1531 | out: |
7f12ad74 GR |
1532 | *flags = tx_flags; |
1533 | return 0; | |
1534 | } | |
1535 | ||
1536 | /** | |
1537 | * i40e_tso - set up the tso context descriptor | |
1538 | * @tx_ring: ptr to the ring to send | |
1539 | * @skb: ptr to the skb we're sending | |
7f12ad74 | 1540 | * @hdr_len: ptr to the size of the packet header |
9c883bd3 | 1541 | * @cd_type_cmd_tso_mss: Quad Word 1 |
7f12ad74 GR |
1542 | * |
1543 | * Returns 0 if no TSO can happen, 1 if tso is going, or error | |
1544 | **/ | |
1545 | static int i40e_tso(struct i40e_ring *tx_ring, struct sk_buff *skb, | |
9c883bd3 | 1546 | u8 *hdr_len, u64 *cd_type_cmd_tso_mss) |
7f12ad74 GR |
1547 | { |
1548 | u32 cd_cmd, cd_tso_len, cd_mss; | |
fe6d4aa4 | 1549 | struct ipv6hdr *ipv6h; |
7f12ad74 GR |
1550 | struct tcphdr *tcph; |
1551 | struct iphdr *iph; | |
1552 | u32 l4len; | |
1553 | int err; | |
7f12ad74 | 1554 | |
e9f6563d SN |
1555 | if (skb->ip_summed != CHECKSUM_PARTIAL) |
1556 | return 0; | |
1557 | ||
7f12ad74 GR |
1558 | if (!skb_is_gso(skb)) |
1559 | return 0; | |
1560 | ||
fe6d4aa4 FR |
1561 | err = skb_cow_head(skb, 0); |
1562 | if (err < 0) | |
1563 | return err; | |
7f12ad74 | 1564 | |
85e76d03 AS |
1565 | iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb); |
1566 | ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) : ipv6_hdr(skb); | |
1567 | ||
1568 | if (iph->version == 4) { | |
7f12ad74 GR |
1569 | tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb); |
1570 | iph->tot_len = 0; | |
1571 | iph->check = 0; | |
1572 | tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, | |
1573 | 0, IPPROTO_TCP, 0); | |
85e76d03 | 1574 | } else if (ipv6h->version == 6) { |
7f12ad74 GR |
1575 | tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb); |
1576 | ipv6h->payload_len = 0; | |
1577 | tcph->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, | |
1578 | 0, IPPROTO_TCP, 0); | |
1579 | } | |
1580 | ||
1581 | l4len = skb->encapsulation ? inner_tcp_hdrlen(skb) : tcp_hdrlen(skb); | |
1582 | *hdr_len = (skb->encapsulation | |
1583 | ? (skb_inner_transport_header(skb) - skb->data) | |
1584 | : skb_transport_offset(skb)) + l4len; | |
1585 | ||
1586 | /* find the field values */ | |
1587 | cd_cmd = I40E_TX_CTX_DESC_TSO; | |
1588 | cd_tso_len = skb->len - *hdr_len; | |
1589 | cd_mss = skb_shinfo(skb)->gso_size; | |
1590 | *cd_type_cmd_tso_mss |= ((u64)cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) | | |
1591 | ((u64)cd_tso_len << | |
1592 | I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) | | |
1593 | ((u64)cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT); | |
1594 | return 1; | |
1595 | } | |
1596 | ||
1597 | /** | |
1598 | * i40e_tx_enable_csum - Enable Tx checksum offloads | |
1599 | * @skb: send buffer | |
89232c3b | 1600 | * @tx_flags: pointer to Tx flags currently set |
7f12ad74 GR |
1601 | * @td_cmd: Tx descriptor command bits to set |
1602 | * @td_offset: Tx descriptor header offsets to set | |
1603 | * @cd_tunneling: ptr to context desc bits | |
1604 | **/ | |
89232c3b | 1605 | static void i40e_tx_enable_csum(struct sk_buff *skb, u32 *tx_flags, |
7f12ad74 GR |
1606 | u32 *td_cmd, u32 *td_offset, |
1607 | struct i40e_ring *tx_ring, | |
1608 | u32 *cd_tunneling) | |
1609 | { | |
1610 | struct ipv6hdr *this_ipv6_hdr; | |
1611 | unsigned int this_tcp_hdrlen; | |
1612 | struct iphdr *this_ip_hdr; | |
1613 | u32 network_hdr_len; | |
1614 | u8 l4_hdr = 0; | |
527274c7 ASJ |
1615 | struct udphdr *oudph; |
1616 | struct iphdr *oiph; | |
45991204 | 1617 | u32 l4_tunnel = 0; |
7f12ad74 GR |
1618 | |
1619 | if (skb->encapsulation) { | |
45991204 ASJ |
1620 | switch (ip_hdr(skb)->protocol) { |
1621 | case IPPROTO_UDP: | |
527274c7 ASJ |
1622 | oudph = udp_hdr(skb); |
1623 | oiph = ip_hdr(skb); | |
45991204 | 1624 | l4_tunnel = I40E_TXD_CTX_UDP_TUNNELING; |
89232c3b | 1625 | *tx_flags |= I40E_TX_FLAGS_VXLAN_TUNNEL; |
45991204 ASJ |
1626 | break; |
1627 | default: | |
1628 | return; | |
1629 | } | |
7f12ad74 GR |
1630 | network_hdr_len = skb_inner_network_header_len(skb); |
1631 | this_ip_hdr = inner_ip_hdr(skb); | |
1632 | this_ipv6_hdr = inner_ipv6_hdr(skb); | |
1633 | this_tcp_hdrlen = inner_tcp_hdrlen(skb); | |
1634 | ||
89232c3b ASJ |
1635 | if (*tx_flags & I40E_TX_FLAGS_IPV4) { |
1636 | if (*tx_flags & I40E_TX_FLAGS_TSO) { | |
7f12ad74 GR |
1637 | *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4; |
1638 | ip_hdr(skb)->check = 0; | |
1639 | } else { | |
1640 | *cd_tunneling |= | |
1641 | I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM; | |
1642 | } | |
89232c3b | 1643 | } else if (*tx_flags & I40E_TX_FLAGS_IPV6) { |
85e76d03 | 1644 | *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV6; |
89232c3b | 1645 | if (*tx_flags & I40E_TX_FLAGS_TSO) |
7f12ad74 | 1646 | ip_hdr(skb)->check = 0; |
7f12ad74 GR |
1647 | } |
1648 | ||
1649 | /* Now set the ctx descriptor fields */ | |
1650 | *cd_tunneling |= (skb_network_header_len(skb) >> 2) << | |
45991204 ASJ |
1651 | I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT | |
1652 | l4_tunnel | | |
7f12ad74 GR |
1653 | ((skb_inner_network_offset(skb) - |
1654 | skb_transport_offset(skb)) >> 1) << | |
1655 | I40E_TXD_CTX_QW0_NATLEN_SHIFT; | |
85e76d03 | 1656 | if (this_ip_hdr->version == 6) { |
89232c3b ASJ |
1657 | *tx_flags &= ~I40E_TX_FLAGS_IPV4; |
1658 | *tx_flags |= I40E_TX_FLAGS_IPV6; | |
85e76d03 AS |
1659 | } |
1660 | ||
527274c7 ASJ |
1661 | if ((tx_ring->flags & I40E_TXR_FLAGS_OUTER_UDP_CSUM) && |
1662 | (l4_tunnel == I40E_TXD_CTX_UDP_TUNNELING) && | |
1663 | (*cd_tunneling & I40E_TXD_CTX_QW0_EXT_IP_MASK)) { | |
1664 | oudph->check = ~csum_tcpudp_magic(oiph->saddr, | |
1665 | oiph->daddr, | |
1666 | (skb->len - skb_transport_offset(skb)), | |
1667 | IPPROTO_UDP, 0); | |
1668 | *cd_tunneling |= I40E_TXD_CTX_QW0_L4T_CS_MASK; | |
1669 | } | |
7f12ad74 GR |
1670 | } else { |
1671 | network_hdr_len = skb_network_header_len(skb); | |
1672 | this_ip_hdr = ip_hdr(skb); | |
1673 | this_ipv6_hdr = ipv6_hdr(skb); | |
1674 | this_tcp_hdrlen = tcp_hdrlen(skb); | |
1675 | } | |
1676 | ||
1677 | /* Enable IP checksum offloads */ | |
89232c3b | 1678 | if (*tx_flags & I40E_TX_FLAGS_IPV4) { |
7f12ad74 GR |
1679 | l4_hdr = this_ip_hdr->protocol; |
1680 | /* the stack computes the IP header already, the only time we | |
1681 | * need the hardware to recompute it is in the case of TSO. | |
1682 | */ | |
89232c3b | 1683 | if (*tx_flags & I40E_TX_FLAGS_TSO) { |
7f12ad74 GR |
1684 | *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM; |
1685 | this_ip_hdr->check = 0; | |
1686 | } else { | |
1687 | *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4; | |
1688 | } | |
1689 | /* Now set the td_offset for IP header length */ | |
1690 | *td_offset = (network_hdr_len >> 2) << | |
1691 | I40E_TX_DESC_LENGTH_IPLEN_SHIFT; | |
89232c3b | 1692 | } else if (*tx_flags & I40E_TX_FLAGS_IPV6) { |
7f12ad74 GR |
1693 | l4_hdr = this_ipv6_hdr->nexthdr; |
1694 | *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6; | |
1695 | /* Now set the td_offset for IP header length */ | |
1696 | *td_offset = (network_hdr_len >> 2) << | |
1697 | I40E_TX_DESC_LENGTH_IPLEN_SHIFT; | |
1698 | } | |
1699 | /* words in MACLEN + dwords in IPLEN + dwords in L4Len */ | |
1700 | *td_offset |= (skb_network_offset(skb) >> 1) << | |
1701 | I40E_TX_DESC_LENGTH_MACLEN_SHIFT; | |
1702 | ||
1703 | /* Enable L4 checksum offloads */ | |
1704 | switch (l4_hdr) { | |
1705 | case IPPROTO_TCP: | |
1706 | /* enable checksum offloads */ | |
1707 | *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP; | |
1708 | *td_offset |= (this_tcp_hdrlen >> 2) << | |
1709 | I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT; | |
1710 | break; | |
1711 | case IPPROTO_SCTP: | |
1712 | /* enable SCTP checksum offload */ | |
1713 | *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP; | |
1714 | *td_offset |= (sizeof(struct sctphdr) >> 2) << | |
1715 | I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT; | |
1716 | break; | |
1717 | case IPPROTO_UDP: | |
1718 | /* enable UDP checksum offload */ | |
1719 | *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP; | |
1720 | *td_offset |= (sizeof(struct udphdr) >> 2) << | |
1721 | I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT; | |
1722 | break; | |
1723 | default: | |
1724 | break; | |
1725 | } | |
1726 | } | |
1727 | ||
1728 | /** | |
1729 | * i40e_create_tx_ctx Build the Tx context descriptor | |
1730 | * @tx_ring: ring to create the descriptor on | |
1731 | * @cd_type_cmd_tso_mss: Quad Word 1 | |
1732 | * @cd_tunneling: Quad Word 0 - bits 0-31 | |
1733 | * @cd_l2tag2: Quad Word 0 - bits 32-63 | |
1734 | **/ | |
1735 | static void i40e_create_tx_ctx(struct i40e_ring *tx_ring, | |
1736 | const u64 cd_type_cmd_tso_mss, | |
1737 | const u32 cd_tunneling, const u32 cd_l2tag2) | |
1738 | { | |
1739 | struct i40e_tx_context_desc *context_desc; | |
1740 | int i = tx_ring->next_to_use; | |
1741 | ||
ff40dd5d JB |
1742 | if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) && |
1743 | !cd_tunneling && !cd_l2tag2) | |
7f12ad74 GR |
1744 | return; |
1745 | ||
1746 | /* grab the next descriptor */ | |
1747 | context_desc = I40E_TX_CTXTDESC(tx_ring, i); | |
1748 | ||
1749 | i++; | |
1750 | tx_ring->next_to_use = (i < tx_ring->count) ? i : 0; | |
1751 | ||
1752 | /* cpu_to_le32 and assign to struct fields */ | |
1753 | context_desc->tunneling_params = cpu_to_le32(cd_tunneling); | |
1754 | context_desc->l2tag2 = cpu_to_le16(cd_l2tag2); | |
3efbbb20 | 1755 | context_desc->rsvd = cpu_to_le16(0); |
7f12ad74 GR |
1756 | context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss); |
1757 | } | |
1758 | ||
4eeb1fff | 1759 | /** |
71da6197 AS |
1760 | * i40e_chk_linearize - Check if there are more than 8 fragments per packet |
1761 | * @skb: send buffer | |
1762 | * @tx_flags: collected send information | |
71da6197 AS |
1763 | * |
1764 | * Note: Our HW can't scatter-gather more than 8 fragments to build | |
1765 | * a packet on the wire and so we need to figure out the cases where we | |
1766 | * need to linearize the skb. | |
1767 | **/ | |
30520831 | 1768 | static bool i40e_chk_linearize(struct sk_buff *skb, u32 tx_flags) |
71da6197 AS |
1769 | { |
1770 | struct skb_frag_struct *frag; | |
1771 | bool linearize = false; | |
1772 | unsigned int size = 0; | |
1773 | u16 num_frags; | |
1774 | u16 gso_segs; | |
1775 | ||
1776 | num_frags = skb_shinfo(skb)->nr_frags; | |
1777 | gso_segs = skb_shinfo(skb)->gso_segs; | |
1778 | ||
1779 | if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO)) { | |
30520831 | 1780 | u16 j = 0; |
71da6197 AS |
1781 | |
1782 | if (num_frags < (I40E_MAX_BUFFER_TXD)) | |
1783 | goto linearize_chk_done; | |
1784 | /* try the simple math, if we have too many frags per segment */ | |
1785 | if (DIV_ROUND_UP((num_frags + gso_segs), gso_segs) > | |
1786 | I40E_MAX_BUFFER_TXD) { | |
1787 | linearize = true; | |
1788 | goto linearize_chk_done; | |
1789 | } | |
1790 | frag = &skb_shinfo(skb)->frags[0]; | |
71da6197 AS |
1791 | /* we might still have more fragments per segment */ |
1792 | do { | |
1793 | size += skb_frag_size(frag); | |
1794 | frag++; j++; | |
30520831 ASJ |
1795 | if ((size >= skb_shinfo(skb)->gso_size) && |
1796 | (j < I40E_MAX_BUFFER_TXD)) { | |
1797 | size = (size % skb_shinfo(skb)->gso_size); | |
1798 | j = (size) ? 1 : 0; | |
1799 | } | |
71da6197 | 1800 | if (j == I40E_MAX_BUFFER_TXD) { |
30520831 ASJ |
1801 | linearize = true; |
1802 | break; | |
71da6197 AS |
1803 | } |
1804 | num_frags--; | |
1805 | } while (num_frags); | |
1806 | } else { | |
1807 | if (num_frags >= I40E_MAX_BUFFER_TXD) | |
1808 | linearize = true; | |
1809 | } | |
1810 | ||
1811 | linearize_chk_done: | |
1812 | return linearize; | |
1813 | } | |
1814 | ||
8f6a2b05 JB |
1815 | /** |
1816 | * __i40evf_maybe_stop_tx - 2nd level check for tx stop conditions | |
1817 | * @tx_ring: the ring to be checked | |
1818 | * @size: the size buffer we want to assure is available | |
1819 | * | |
1820 | * Returns -EBUSY if a stop is needed, else 0 | |
1821 | **/ | |
1822 | static inline int __i40evf_maybe_stop_tx(struct i40e_ring *tx_ring, int size) | |
1823 | { | |
1824 | netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index); | |
1825 | /* Memory barrier before checking head and tail */ | |
1826 | smp_mb(); | |
1827 | ||
1828 | /* Check again in a case another CPU has just made room available. */ | |
1829 | if (likely(I40E_DESC_UNUSED(tx_ring) < size)) | |
1830 | return -EBUSY; | |
1831 | ||
1832 | /* A reprieve! - use start_queue because it doesn't call schedule */ | |
1833 | netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index); | |
1834 | ++tx_ring->tx_stats.restart_queue; | |
1835 | return 0; | |
1836 | } | |
1837 | ||
1838 | /** | |
1839 | * i40evf_maybe_stop_tx - 1st level check for tx stop conditions | |
1840 | * @tx_ring: the ring to be checked | |
1841 | * @size: the size buffer we want to assure is available | |
1842 | * | |
1843 | * Returns 0 if stop is not needed | |
1844 | **/ | |
3e587cf3 | 1845 | static inline int i40evf_maybe_stop_tx(struct i40e_ring *tx_ring, int size) |
8f6a2b05 JB |
1846 | { |
1847 | if (likely(I40E_DESC_UNUSED(tx_ring) >= size)) | |
1848 | return 0; | |
1849 | return __i40evf_maybe_stop_tx(tx_ring, size); | |
1850 | } | |
1851 | ||
7f12ad74 | 1852 | /** |
3e587cf3 | 1853 | * i40evf_tx_map - Build the Tx descriptor |
7f12ad74 GR |
1854 | * @tx_ring: ring to send buffer on |
1855 | * @skb: send buffer | |
1856 | * @first: first buffer info buffer to use | |
1857 | * @tx_flags: collected send information | |
1858 | * @hdr_len: size of the packet header | |
1859 | * @td_cmd: the command field in the descriptor | |
1860 | * @td_offset: offset for checksum or crc | |
1861 | **/ | |
3e587cf3 JB |
1862 | static inline void i40evf_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb, |
1863 | struct i40e_tx_buffer *first, u32 tx_flags, | |
1864 | const u8 hdr_len, u32 td_cmd, u32 td_offset) | |
7f12ad74 GR |
1865 | { |
1866 | unsigned int data_len = skb->data_len; | |
1867 | unsigned int size = skb_headlen(skb); | |
1868 | struct skb_frag_struct *frag; | |
1869 | struct i40e_tx_buffer *tx_bi; | |
1870 | struct i40e_tx_desc *tx_desc; | |
1871 | u16 i = tx_ring->next_to_use; | |
1872 | u32 td_tag = 0; | |
1873 | dma_addr_t dma; | |
1874 | u16 gso_segs; | |
6a7fded7 ASJ |
1875 | u16 desc_count = 0; |
1876 | bool tail_bump = true; | |
1877 | bool do_rs = false; | |
7f12ad74 GR |
1878 | |
1879 | if (tx_flags & I40E_TX_FLAGS_HW_VLAN) { | |
1880 | td_cmd |= I40E_TX_DESC_CMD_IL2TAG1; | |
1881 | td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >> | |
1882 | I40E_TX_FLAGS_VLAN_SHIFT; | |
1883 | } | |
1884 | ||
1885 | if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO)) | |
1886 | gso_segs = skb_shinfo(skb)->gso_segs; | |
1887 | else | |
1888 | gso_segs = 1; | |
1889 | ||
1890 | /* multiply data chunks by size of headers */ | |
1891 | first->bytecount = skb->len - hdr_len + (gso_segs * hdr_len); | |
1892 | first->gso_segs = gso_segs; | |
1893 | first->skb = skb; | |
1894 | first->tx_flags = tx_flags; | |
1895 | ||
1896 | dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE); | |
1897 | ||
1898 | tx_desc = I40E_TX_DESC(tx_ring, i); | |
1899 | tx_bi = first; | |
1900 | ||
1901 | for (frag = &skb_shinfo(skb)->frags[0];; frag++) { | |
1902 | if (dma_mapping_error(tx_ring->dev, dma)) | |
1903 | goto dma_error; | |
1904 | ||
1905 | /* record length, and DMA address */ | |
1906 | dma_unmap_len_set(tx_bi, len, size); | |
1907 | dma_unmap_addr_set(tx_bi, dma, dma); | |
1908 | ||
1909 | tx_desc->buffer_addr = cpu_to_le64(dma); | |
1910 | ||
1911 | while (unlikely(size > I40E_MAX_DATA_PER_TXD)) { | |
1912 | tx_desc->cmd_type_offset_bsz = | |
1913 | build_ctob(td_cmd, td_offset, | |
1914 | I40E_MAX_DATA_PER_TXD, td_tag); | |
1915 | ||
1916 | tx_desc++; | |
1917 | i++; | |
6a7fded7 ASJ |
1918 | desc_count++; |
1919 | ||
7f12ad74 GR |
1920 | if (i == tx_ring->count) { |
1921 | tx_desc = I40E_TX_DESC(tx_ring, 0); | |
1922 | i = 0; | |
1923 | } | |
1924 | ||
1925 | dma += I40E_MAX_DATA_PER_TXD; | |
1926 | size -= I40E_MAX_DATA_PER_TXD; | |
1927 | ||
1928 | tx_desc->buffer_addr = cpu_to_le64(dma); | |
1929 | } | |
1930 | ||
1931 | if (likely(!data_len)) | |
1932 | break; | |
1933 | ||
1934 | tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset, | |
1935 | size, td_tag); | |
1936 | ||
1937 | tx_desc++; | |
1938 | i++; | |
6a7fded7 ASJ |
1939 | desc_count++; |
1940 | ||
7f12ad74 GR |
1941 | if (i == tx_ring->count) { |
1942 | tx_desc = I40E_TX_DESC(tx_ring, 0); | |
1943 | i = 0; | |
1944 | } | |
1945 | ||
1946 | size = skb_frag_size(frag); | |
1947 | data_len -= size; | |
1948 | ||
1949 | dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size, | |
1950 | DMA_TO_DEVICE); | |
1951 | ||
1952 | tx_bi = &tx_ring->tx_bi[i]; | |
1953 | } | |
1954 | ||
7f12ad74 GR |
1955 | /* set next_to_watch value indicating a packet is present */ |
1956 | first->next_to_watch = tx_desc; | |
1957 | ||
1958 | i++; | |
1959 | if (i == tx_ring->count) | |
1960 | i = 0; | |
1961 | ||
1962 | tx_ring->next_to_use = i; | |
1963 | ||
6a7fded7 ASJ |
1964 | netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev, |
1965 | tx_ring->queue_index), | |
1966 | first->bytecount); | |
8f6a2b05 | 1967 | i40evf_maybe_stop_tx(tx_ring, DESC_NEEDED); |
6a7fded7 ASJ |
1968 | |
1969 | /* Algorithm to optimize tail and RS bit setting: | |
1970 | * if xmit_more is supported | |
1971 | * if xmit_more is true | |
1972 | * do not update tail and do not mark RS bit. | |
1973 | * if xmit_more is false and last xmit_more was false | |
1974 | * if every packet spanned less than 4 desc | |
1975 | * then set RS bit on 4th packet and update tail | |
1976 | * on every packet | |
1977 | * else | |
1978 | * update tail and set RS bit on every packet. | |
1979 | * if xmit_more is false and last_xmit_more was true | |
1980 | * update tail and set RS bit. | |
6a7fded7 ASJ |
1981 | * |
1982 | * Optimization: wmb to be issued only in case of tail update. | |
1983 | * Also optimize the Descriptor WB path for RS bit with the same | |
1984 | * algorithm. | |
1985 | * | |
1986 | * Note: If there are less than 4 packets | |
1987 | * pending and interrupts were disabled the service task will | |
1988 | * trigger a force WB. | |
1989 | */ | |
1990 | if (skb->xmit_more && | |
1991 | !netif_xmit_stopped(netdev_get_tx_queue(tx_ring->netdev, | |
1992 | tx_ring->queue_index))) { | |
1993 | tx_ring->flags |= I40E_TXR_FLAGS_LAST_XMIT_MORE_SET; | |
1994 | tail_bump = false; | |
1995 | } else if (!skb->xmit_more && | |
1996 | !netif_xmit_stopped(netdev_get_tx_queue(tx_ring->netdev, | |
1997 | tx_ring->queue_index)) && | |
1998 | (!(tx_ring->flags & I40E_TXR_FLAGS_LAST_XMIT_MORE_SET)) && | |
1999 | (tx_ring->packet_stride < WB_STRIDE) && | |
2000 | (desc_count < WB_STRIDE)) { | |
2001 | tx_ring->packet_stride++; | |
2002 | } else { | |
2003 | tx_ring->packet_stride = 0; | |
2004 | tx_ring->flags &= ~I40E_TXR_FLAGS_LAST_XMIT_MORE_SET; | |
2005 | do_rs = true; | |
2006 | } | |
2007 | if (do_rs) | |
2008 | tx_ring->packet_stride = 0; | |
2009 | ||
2010 | tx_desc->cmd_type_offset_bsz = | |
2011 | build_ctob(td_cmd, td_offset, size, td_tag) | | |
2012 | cpu_to_le64((u64)(do_rs ? I40E_TXD_CMD : | |
2013 | I40E_TX_DESC_CMD_EOP) << | |
2014 | I40E_TXD_QW1_CMD_SHIFT); | |
2015 | ||
7f12ad74 | 2016 | /* notify HW of packet */ |
6a7fded7 | 2017 | if (!tail_bump) |
489ce7a4 | 2018 | prefetchw(tx_desc + 1); |
7f12ad74 | 2019 | |
6a7fded7 ASJ |
2020 | if (tail_bump) { |
2021 | /* Force memory writes to complete before letting h/w | |
2022 | * know there are new descriptors to fetch. (Only | |
2023 | * applicable for weak-ordered memory model archs, | |
2024 | * such as IA-64). | |
2025 | */ | |
2026 | wmb(); | |
2027 | writel(i, tx_ring->tail); | |
2028 | } | |
2029 | ||
7f12ad74 GR |
2030 | return; |
2031 | ||
2032 | dma_error: | |
2033 | dev_info(tx_ring->dev, "TX DMA map failed\n"); | |
2034 | ||
2035 | /* clear dma mappings for failed tx_bi map */ | |
2036 | for (;;) { | |
2037 | tx_bi = &tx_ring->tx_bi[i]; | |
2038 | i40e_unmap_and_free_tx_resource(tx_ring, tx_bi); | |
2039 | if (tx_bi == first) | |
2040 | break; | |
2041 | if (i == 0) | |
2042 | i = tx_ring->count; | |
2043 | i--; | |
2044 | } | |
2045 | ||
2046 | tx_ring->next_to_use = i; | |
2047 | } | |
2048 | ||
7f12ad74 | 2049 | /** |
3e587cf3 | 2050 | * i40evf_xmit_descriptor_count - calculate number of tx descriptors needed |
7f12ad74 GR |
2051 | * @skb: send buffer |
2052 | * @tx_ring: ring to send buffer on | |
2053 | * | |
2054 | * Returns number of data descriptors needed for this skb. Returns 0 to indicate | |
2055 | * there is not enough descriptors available in this ring since we need at least | |
2056 | * one descriptor. | |
2057 | **/ | |
3e587cf3 JB |
2058 | static inline int i40evf_xmit_descriptor_count(struct sk_buff *skb, |
2059 | struct i40e_ring *tx_ring) | |
7f12ad74 | 2060 | { |
7f12ad74 | 2061 | unsigned int f; |
7f12ad74 GR |
2062 | int count = 0; |
2063 | ||
2064 | /* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD, | |
2065 | * + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD, | |
be560521 | 2066 | * + 4 desc gap to avoid the cache line where head is, |
7f12ad74 GR |
2067 | * + 1 desc for context descriptor, |
2068 | * otherwise try next time | |
2069 | */ | |
7f12ad74 GR |
2070 | for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) |
2071 | count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size); | |
980093eb | 2072 | |
7f12ad74 | 2073 | count += TXD_USE_COUNT(skb_headlen(skb)); |
8f6a2b05 | 2074 | if (i40evf_maybe_stop_tx(tx_ring, count + 4 + 1)) { |
7f12ad74 GR |
2075 | tx_ring->tx_stats.tx_busy++; |
2076 | return 0; | |
2077 | } | |
2078 | return count; | |
2079 | } | |
2080 | ||
2081 | /** | |
2082 | * i40e_xmit_frame_ring - Sends buffer on Tx ring | |
2083 | * @skb: send buffer | |
2084 | * @tx_ring: ring to send buffer on | |
2085 | * | |
2086 | * Returns NETDEV_TX_OK if sent, else an error code | |
2087 | **/ | |
2088 | static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb, | |
2089 | struct i40e_ring *tx_ring) | |
2090 | { | |
2091 | u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT; | |
2092 | u32 cd_tunneling = 0, cd_l2tag2 = 0; | |
2093 | struct i40e_tx_buffer *first; | |
2094 | u32 td_offset = 0; | |
2095 | u32 tx_flags = 0; | |
2096 | __be16 protocol; | |
2097 | u32 td_cmd = 0; | |
2098 | u8 hdr_len = 0; | |
2099 | int tso; | |
6995b36c | 2100 | |
b74118f0 JB |
2101 | /* prefetch the data, we'll need it later */ |
2102 | prefetch(skb->data); | |
2103 | ||
3e587cf3 | 2104 | if (0 == i40evf_xmit_descriptor_count(skb, tx_ring)) |
7f12ad74 GR |
2105 | return NETDEV_TX_BUSY; |
2106 | ||
2107 | /* prepare the xmit flags */ | |
3e587cf3 | 2108 | if (i40evf_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags)) |
7f12ad74 GR |
2109 | goto out_drop; |
2110 | ||
2111 | /* obtain protocol of skb */ | |
a12c4158 | 2112 | protocol = vlan_get_protocol(skb); |
7f12ad74 GR |
2113 | |
2114 | /* record the location of the first descriptor for this packet */ | |
2115 | first = &tx_ring->tx_bi[tx_ring->next_to_use]; | |
2116 | ||
2117 | /* setup IPv4/IPv6 offloads */ | |
2118 | if (protocol == htons(ETH_P_IP)) | |
2119 | tx_flags |= I40E_TX_FLAGS_IPV4; | |
2120 | else if (protocol == htons(ETH_P_IPV6)) | |
2121 | tx_flags |= I40E_TX_FLAGS_IPV6; | |
2122 | ||
9c883bd3 | 2123 | tso = i40e_tso(tx_ring, skb, &hdr_len, &cd_type_cmd_tso_mss); |
7f12ad74 GR |
2124 | |
2125 | if (tso < 0) | |
2126 | goto out_drop; | |
2127 | else if (tso) | |
2128 | tx_flags |= I40E_TX_FLAGS_TSO; | |
2129 | ||
2fc3d715 | 2130 | if (i40e_chk_linearize(skb, tx_flags)) { |
71da6197 AS |
2131 | if (skb_linearize(skb)) |
2132 | goto out_drop; | |
2fc3d715 ASJ |
2133 | tx_ring->tx_stats.tx_linearize++; |
2134 | } | |
7f12ad74 GR |
2135 | skb_tx_timestamp(skb); |
2136 | ||
2137 | /* always enable CRC insertion offload */ | |
2138 | td_cmd |= I40E_TX_DESC_CMD_ICRC; | |
2139 | ||
2140 | /* Always offload the checksum, since it's in the data descriptor */ | |
2141 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
2142 | tx_flags |= I40E_TX_FLAGS_CSUM; | |
2143 | ||
89232c3b | 2144 | i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset, |
7f12ad74 GR |
2145 | tx_ring, &cd_tunneling); |
2146 | } | |
2147 | ||
2148 | i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss, | |
2149 | cd_tunneling, cd_l2tag2); | |
2150 | ||
3e587cf3 JB |
2151 | i40evf_tx_map(tx_ring, skb, first, tx_flags, hdr_len, |
2152 | td_cmd, td_offset); | |
7f12ad74 | 2153 | |
7f12ad74 GR |
2154 | return NETDEV_TX_OK; |
2155 | ||
2156 | out_drop: | |
2157 | dev_kfree_skb_any(skb); | |
2158 | return NETDEV_TX_OK; | |
2159 | } | |
2160 | ||
2161 | /** | |
2162 | * i40evf_xmit_frame - Selects the correct VSI and Tx queue to send buffer | |
2163 | * @skb: send buffer | |
2164 | * @netdev: network interface device structure | |
2165 | * | |
2166 | * Returns NETDEV_TX_OK if sent, else an error code | |
2167 | **/ | |
2168 | netdev_tx_t i40evf_xmit_frame(struct sk_buff *skb, struct net_device *netdev) | |
2169 | { | |
2170 | struct i40evf_adapter *adapter = netdev_priv(netdev); | |
0dd438d8 | 2171 | struct i40e_ring *tx_ring = &adapter->tx_rings[skb->queue_mapping]; |
7f12ad74 GR |
2172 | |
2173 | /* hardware can't handle really short frames, hardware padding works | |
2174 | * beyond this point | |
2175 | */ | |
2176 | if (unlikely(skb->len < I40E_MIN_TX_LEN)) { | |
2177 | if (skb_pad(skb, I40E_MIN_TX_LEN - skb->len)) | |
2178 | return NETDEV_TX_OK; | |
2179 | skb->len = I40E_MIN_TX_LEN; | |
2180 | skb_set_tail_pointer(skb, I40E_MIN_TX_LEN); | |
2181 | } | |
2182 | ||
2183 | return i40e_xmit_frame_ring(skb, tx_ring); | |
2184 | } |