1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 2018 Intel Corporation. */
3
4 #include <linux/bpf_trace.h>
5 #include <net/xdp_sock_drv.h>
6 #include "i40e_txrx_common.h"
7 #include "i40e_xsk.h"
8
i40e_clear_rx_bi_zc(struct i40e_ring * rx_ring)9 void i40e_clear_rx_bi_zc(struct i40e_ring *rx_ring)
10 {
11 memset(rx_ring->rx_bi_zc, 0,
12 sizeof(*rx_ring->rx_bi_zc) * rx_ring->count);
13 }
14
i40e_rx_bi(struct i40e_ring * rx_ring,u32 idx)15 static struct xdp_buff **i40e_rx_bi(struct i40e_ring *rx_ring, u32 idx)
16 {
17 return &rx_ring->rx_bi_zc[idx];
18 }
19
20 /**
21 * i40e_realloc_rx_xdp_bi - reallocate SW ring for either XSK or normal buffer
22 * @rx_ring: Current rx ring
23 * @pool_present: is pool for XSK present
24 *
25 * Try allocating memory and return ENOMEM, if failed to allocate.
26 * If allocation was successful, substitute buffer with allocated one.
27 * Returns 0 on success, negative on failure
28 */
i40e_realloc_rx_xdp_bi(struct i40e_ring * rx_ring,bool pool_present)29 static int i40e_realloc_rx_xdp_bi(struct i40e_ring *rx_ring, bool pool_present)
30 {
31 size_t elem_size = pool_present ? sizeof(*rx_ring->rx_bi_zc) :
32 sizeof(*rx_ring->rx_bi);
33 void *sw_ring = kcalloc(rx_ring->count, elem_size, GFP_KERNEL);
34
35 if (!sw_ring)
36 return -ENOMEM;
37
38 if (pool_present) {
39 kfree(rx_ring->rx_bi);
40 rx_ring->rx_bi = NULL;
41 rx_ring->rx_bi_zc = sw_ring;
42 } else {
43 kfree(rx_ring->rx_bi_zc);
44 rx_ring->rx_bi_zc = NULL;
45 rx_ring->rx_bi = sw_ring;
46 }
47 return 0;
48 }
49
50 /**
51 * i40e_realloc_rx_bi_zc - reallocate rx SW rings
52 * @vsi: Current VSI
53 * @zc: is zero copy set
54 *
55 * Reallocate buffer for rx_rings that might be used by XSK.
56 * XDP requires more memory, than rx_buf provides.
57 * Returns 0 on success, negative on failure
58 */
i40e_realloc_rx_bi_zc(struct i40e_vsi * vsi,bool zc)59 int i40e_realloc_rx_bi_zc(struct i40e_vsi *vsi, bool zc)
60 {
61 struct i40e_ring *rx_ring;
62 unsigned long q;
63
64 for_each_set_bit(q, vsi->af_xdp_zc_qps, vsi->alloc_queue_pairs) {
65 rx_ring = vsi->rx_rings[q];
66 if (i40e_realloc_rx_xdp_bi(rx_ring, zc))
67 return -ENOMEM;
68 }
69 return 0;
70 }
71
72 /**
73 * i40e_xsk_pool_enable - Enable/associate an AF_XDP buffer pool to a
74 * certain ring/qid
75 * @vsi: Current VSI
76 * @pool: buffer pool
77 * @qid: Rx ring to associate buffer pool with
78 *
79 * Returns 0 on success, <0 on failure
80 **/
i40e_xsk_pool_enable(struct i40e_vsi * vsi,struct xsk_buff_pool * pool,u16 qid)81 static int i40e_xsk_pool_enable(struct i40e_vsi *vsi,
82 struct xsk_buff_pool *pool,
83 u16 qid)
84 {
85 struct net_device *netdev = vsi->netdev;
86 bool if_running;
87 int err;
88
89 if (vsi->type != I40E_VSI_MAIN)
90 return -EINVAL;
91
92 if (qid >= vsi->num_queue_pairs)
93 return -EINVAL;
94
95 if (qid >= netdev->real_num_rx_queues ||
96 qid >= netdev->real_num_tx_queues)
97 return -EINVAL;
98
99 err = xsk_pool_dma_map(pool, &vsi->back->pdev->dev, I40E_RX_DMA_ATTR);
100 if (err)
101 return err;
102
103 set_bit(qid, vsi->af_xdp_zc_qps);
104
105 if_running = netif_running(vsi->netdev) && i40e_enabled_xdp_vsi(vsi);
106
107 if (if_running) {
108 err = i40e_queue_pair_disable(vsi, qid);
109 if (err)
110 return err;
111
112 err = i40e_realloc_rx_xdp_bi(vsi->rx_rings[qid], true);
113 if (err)
114 return err;
115
116 err = i40e_queue_pair_enable(vsi, qid);
117 if (err)
118 return err;
119
120 /* Kick start the NAPI context so that receiving will start */
121 err = i40e_xsk_wakeup(vsi->netdev, qid, XDP_WAKEUP_RX);
122 if (err)
123 return err;
124 }
125
126 return 0;
127 }
128
129 /**
130 * i40e_xsk_pool_disable - Disassociate an AF_XDP buffer pool from a
131 * certain ring/qid
132 * @vsi: Current VSI
133 * @qid: Rx ring to associate buffer pool with
134 *
135 * Returns 0 on success, <0 on failure
136 **/
i40e_xsk_pool_disable(struct i40e_vsi * vsi,u16 qid)137 static int i40e_xsk_pool_disable(struct i40e_vsi *vsi, u16 qid)
138 {
139 struct net_device *netdev = vsi->netdev;
140 struct xsk_buff_pool *pool;
141 bool if_running;
142 int err;
143
144 pool = xsk_get_pool_from_qid(netdev, qid);
145 if (!pool)
146 return -EINVAL;
147
148 if_running = netif_running(vsi->netdev) && i40e_enabled_xdp_vsi(vsi);
149
150 if (if_running) {
151 err = i40e_queue_pair_disable(vsi, qid);
152 if (err)
153 return err;
154 }
155
156 clear_bit(qid, vsi->af_xdp_zc_qps);
157 xsk_pool_dma_unmap(pool, I40E_RX_DMA_ATTR);
158
159 if (if_running) {
160 err = i40e_realloc_rx_xdp_bi(vsi->rx_rings[qid], false);
161 if (err)
162 return err;
163 err = i40e_queue_pair_enable(vsi, qid);
164 if (err)
165 return err;
166 }
167
168 return 0;
169 }
170
171 /**
172 * i40e_xsk_pool_setup - Enable/disassociate an AF_XDP buffer pool to/from
173 * a ring/qid
174 * @vsi: Current VSI
175 * @pool: Buffer pool to enable/associate to a ring, or NULL to disable
176 * @qid: Rx ring to (dis)associate buffer pool (from)to
177 *
178 * This function enables or disables a buffer pool to a certain ring.
179 *
180 * Returns 0 on success, <0 on failure
181 **/
i40e_xsk_pool_setup(struct i40e_vsi * vsi,struct xsk_buff_pool * pool,u16 qid)182 int i40e_xsk_pool_setup(struct i40e_vsi *vsi, struct xsk_buff_pool *pool,
183 u16 qid)
184 {
185 return pool ? i40e_xsk_pool_enable(vsi, pool, qid) :
186 i40e_xsk_pool_disable(vsi, qid);
187 }
188
189 /**
190 * i40e_run_xdp_zc - Executes an XDP program on an xdp_buff
191 * @rx_ring: Rx ring
192 * @xdp: xdp_buff used as input to the XDP program
193 * @xdp_prog: XDP program to run
194 *
195 * Returns any of I40E_XDP_{PASS, CONSUMED, TX, REDIR}
196 **/
i40e_run_xdp_zc(struct i40e_ring * rx_ring,struct xdp_buff * xdp,struct bpf_prog * xdp_prog)197 static int i40e_run_xdp_zc(struct i40e_ring *rx_ring, struct xdp_buff *xdp,
198 struct bpf_prog *xdp_prog)
199 {
200 int err, result = I40E_XDP_PASS;
201 struct i40e_ring *xdp_ring;
202 u32 act;
203
204 act = bpf_prog_run_xdp(xdp_prog, xdp);
205
206 if (likely(act == XDP_REDIRECT)) {
207 err = xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog);
208 if (!err)
209 return I40E_XDP_REDIR;
210 if (xsk_uses_need_wakeup(rx_ring->xsk_pool) && err == -ENOBUFS)
211 result = I40E_XDP_EXIT;
212 else
213 result = I40E_XDP_CONSUMED;
214 goto out_failure;
215 }
216
217 switch (act) {
218 case XDP_PASS:
219 break;
220 case XDP_TX:
221 xdp_ring = rx_ring->vsi->xdp_rings[rx_ring->queue_index];
222 result = i40e_xmit_xdp_tx_ring(xdp, xdp_ring);
223 if (result == I40E_XDP_CONSUMED)
224 goto out_failure;
225 break;
226 case XDP_DROP:
227 result = I40E_XDP_CONSUMED;
228 break;
229 default:
230 bpf_warn_invalid_xdp_action(rx_ring->netdev, xdp_prog, act);
231 fallthrough;
232 case XDP_ABORTED:
233 result = I40E_XDP_CONSUMED;
234 out_failure:
235 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
236 }
237 return result;
238 }
239
i40e_alloc_rx_buffers_zc(struct i40e_ring * rx_ring,u16 count)240 bool i40e_alloc_rx_buffers_zc(struct i40e_ring *rx_ring, u16 count)
241 {
242 u16 ntu = rx_ring->next_to_use;
243 union i40e_rx_desc *rx_desc;
244 struct xdp_buff **xdp;
245 u32 nb_buffs, i;
246 dma_addr_t dma;
247
248 rx_desc = I40E_RX_DESC(rx_ring, ntu);
249 xdp = i40e_rx_bi(rx_ring, ntu);
250
251 nb_buffs = min_t(u16, count, rx_ring->count - ntu);
252 nb_buffs = xsk_buff_alloc_batch(rx_ring->xsk_pool, xdp, nb_buffs);
253 if (!nb_buffs)
254 return false;
255
256 i = nb_buffs;
257 while (i--) {
258 dma = xsk_buff_xdp_get_dma(*xdp);
259 rx_desc->read.pkt_addr = cpu_to_le64(dma);
260 rx_desc->read.hdr_addr = 0;
261
262 rx_desc++;
263 xdp++;
264 }
265
266 ntu += nb_buffs;
267 if (ntu == rx_ring->count) {
268 rx_desc = I40E_RX_DESC(rx_ring, 0);
269 ntu = 0;
270 }
271
272 /* clear the status bits for the next_to_use descriptor */
273 rx_desc->wb.qword1.status_error_len = 0;
274 i40e_release_rx_desc(rx_ring, ntu);
275
276 return count == nb_buffs;
277 }
278
279 /**
280 * i40e_construct_skb_zc - Create skbuff from zero-copy Rx buffer
281 * @rx_ring: Rx ring
282 * @xdp: xdp_buff
283 *
284 * This functions allocates a new skb from a zero-copy Rx buffer.
285 *
286 * Returns the skb, or NULL on failure.
287 **/
i40e_construct_skb_zc(struct i40e_ring * rx_ring,struct xdp_buff * xdp)288 static struct sk_buff *i40e_construct_skb_zc(struct i40e_ring *rx_ring,
289 struct xdp_buff *xdp)
290 {
291 unsigned int totalsize = xdp->data_end - xdp->data_meta;
292 unsigned int metasize = xdp->data - xdp->data_meta;
293 struct skb_shared_info *sinfo = NULL;
294 struct sk_buff *skb;
295 u32 nr_frags = 0;
296
297 if (unlikely(xdp_buff_has_frags(xdp))) {
298 sinfo = xdp_get_shared_info_from_buff(xdp);
299 nr_frags = sinfo->nr_frags;
300 }
301 net_prefetch(xdp->data_meta);
302
303 /* allocate a skb to store the frags */
304 skb = napi_alloc_skb(&rx_ring->q_vector->napi, totalsize);
305 if (unlikely(!skb))
306 goto out;
307
308 memcpy(__skb_put(skb, totalsize), xdp->data_meta,
309 ALIGN(totalsize, sizeof(long)));
310
311 if (metasize) {
312 skb_metadata_set(skb, metasize);
313 __skb_pull(skb, metasize);
314 }
315
316 if (likely(!xdp_buff_has_frags(xdp)))
317 goto out;
318
319 for (int i = 0; i < nr_frags; i++) {
320 struct skb_shared_info *skinfo = skb_shinfo(skb);
321 skb_frag_t *frag = &sinfo->frags[i];
322 struct page *page;
323 void *addr;
324
325 page = dev_alloc_page();
326 if (!page) {
327 dev_kfree_skb(skb);
328 return NULL;
329 }
330 addr = page_to_virt(page);
331
332 memcpy(addr, skb_frag_page(frag), skb_frag_size(frag));
333
334 __skb_fill_page_desc_noacc(skinfo, skinfo->nr_frags++,
335 addr, 0, skb_frag_size(frag));
336 }
337
338 out:
339 xsk_buff_free(xdp);
340 return skb;
341 }
342
i40e_handle_xdp_result_zc(struct i40e_ring * rx_ring,struct xdp_buff * xdp_buff,union i40e_rx_desc * rx_desc,unsigned int * rx_packets,unsigned int * rx_bytes,unsigned int xdp_res,bool * failure)343 static void i40e_handle_xdp_result_zc(struct i40e_ring *rx_ring,
344 struct xdp_buff *xdp_buff,
345 union i40e_rx_desc *rx_desc,
346 unsigned int *rx_packets,
347 unsigned int *rx_bytes,
348 unsigned int xdp_res,
349 bool *failure)
350 {
351 struct sk_buff *skb;
352
353 *rx_packets = 1;
354 *rx_bytes = xdp_get_buff_len(xdp_buff);
355
356 if (likely(xdp_res == I40E_XDP_REDIR) || xdp_res == I40E_XDP_TX)
357 return;
358
359 if (xdp_res == I40E_XDP_EXIT) {
360 *failure = true;
361 return;
362 }
363
364 if (xdp_res == I40E_XDP_CONSUMED) {
365 xsk_buff_free(xdp_buff);
366 return;
367 }
368 if (xdp_res == I40E_XDP_PASS) {
369 /* NB! We are not checking for errors using
370 * i40e_test_staterr with
371 * BIT(I40E_RXD_QW1_ERROR_SHIFT). This is due to that
372 * SBP is *not* set in PRT_SBPVSI (default not set).
373 */
374 skb = i40e_construct_skb_zc(rx_ring, xdp_buff);
375 if (!skb) {
376 rx_ring->rx_stats.alloc_buff_failed++;
377 *rx_packets = 0;
378 *rx_bytes = 0;
379 return;
380 }
381
382 if (eth_skb_pad(skb)) {
383 *rx_packets = 0;
384 *rx_bytes = 0;
385 return;
386 }
387
388 i40e_process_skb_fields(rx_ring, rx_desc, skb);
389 napi_gro_receive(&rx_ring->q_vector->napi, skb);
390 return;
391 }
392
393 /* Should never get here, as all valid cases have been handled already.
394 */
395 WARN_ON_ONCE(1);
396 }
397
398 static int
i40e_add_xsk_frag(struct i40e_ring * rx_ring,struct xdp_buff * first,struct xdp_buff * xdp,const unsigned int size)399 i40e_add_xsk_frag(struct i40e_ring *rx_ring, struct xdp_buff *first,
400 struct xdp_buff *xdp, const unsigned int size)
401 {
402 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(first);
403
404 if (!xdp_buff_has_frags(first)) {
405 sinfo->nr_frags = 0;
406 sinfo->xdp_frags_size = 0;
407 xdp_buff_set_frags_flag(first);
408 }
409
410 if (unlikely(sinfo->nr_frags == MAX_SKB_FRAGS)) {
411 xsk_buff_free(first);
412 return -ENOMEM;
413 }
414
415 __skb_fill_page_desc_noacc(sinfo, sinfo->nr_frags++,
416 virt_to_page(xdp->data_hard_start),
417 XDP_PACKET_HEADROOM, size);
418 sinfo->xdp_frags_size += size;
419 xsk_buff_add_frag(xdp);
420
421 return 0;
422 }
423
424 /**
425 * i40e_clean_rx_irq_zc - Consumes Rx packets from the hardware ring
426 * @rx_ring: Rx ring
427 * @budget: NAPI budget
428 *
429 * Returns amount of work completed
430 **/
i40e_clean_rx_irq_zc(struct i40e_ring * rx_ring,int budget)431 int i40e_clean_rx_irq_zc(struct i40e_ring *rx_ring, int budget)
432 {
433 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
434 u16 next_to_process = rx_ring->next_to_process;
435 u16 next_to_clean = rx_ring->next_to_clean;
436 unsigned int xdp_res, xdp_xmit = 0;
437 struct xdp_buff *first = NULL;
438 u32 count = rx_ring->count;
439 struct bpf_prog *xdp_prog;
440 u32 entries_to_alloc;
441 bool failure = false;
442
443 if (next_to_process != next_to_clean)
444 first = *i40e_rx_bi(rx_ring, next_to_clean);
445
446 /* NB! xdp_prog will always be !NULL, due to the fact that
447 * this path is enabled by setting an XDP program.
448 */
449 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
450
451 while (likely(total_rx_packets < (unsigned int)budget)) {
452 union i40e_rx_desc *rx_desc;
453 unsigned int rx_packets;
454 unsigned int rx_bytes;
455 struct xdp_buff *bi;
456 unsigned int size;
457 u64 qword;
458
459 rx_desc = I40E_RX_DESC(rx_ring, next_to_process);
460 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
461
462 /* This memory barrier is needed to keep us from reading
463 * any other fields out of the rx_desc until we have
464 * verified the descriptor has been written back.
465 */
466 dma_rmb();
467
468 if (i40e_rx_is_programming_status(qword)) {
469 i40e_clean_programming_status(rx_ring,
470 rx_desc->raw.qword[0],
471 qword);
472 bi = *i40e_rx_bi(rx_ring, next_to_process);
473 xsk_buff_free(bi);
474 if (++next_to_process == count)
475 next_to_process = 0;
476 continue;
477 }
478
479 size = FIELD_GET(I40E_RXD_QW1_LENGTH_PBUF_MASK, qword);
480 if (!size)
481 break;
482
483 bi = *i40e_rx_bi(rx_ring, next_to_process);
484 xsk_buff_set_size(bi, size);
485 xsk_buff_dma_sync_for_cpu(bi);
486
487 if (!first)
488 first = bi;
489 else if (i40e_add_xsk_frag(rx_ring, first, bi, size))
490 break;
491
492 if (++next_to_process == count)
493 next_to_process = 0;
494
495 if (i40e_is_non_eop(rx_ring, rx_desc))
496 continue;
497
498 xdp_res = i40e_run_xdp_zc(rx_ring, first, xdp_prog);
499 i40e_handle_xdp_result_zc(rx_ring, first, rx_desc, &rx_packets,
500 &rx_bytes, xdp_res, &failure);
501 next_to_clean = next_to_process;
502 if (failure)
503 break;
504 total_rx_packets += rx_packets;
505 total_rx_bytes += rx_bytes;
506 xdp_xmit |= xdp_res & (I40E_XDP_TX | I40E_XDP_REDIR);
507 first = NULL;
508 }
509
510 rx_ring->next_to_clean = next_to_clean;
511 rx_ring->next_to_process = next_to_process;
512
513 entries_to_alloc = I40E_DESC_UNUSED(rx_ring);
514 if (entries_to_alloc >= I40E_RX_BUFFER_WRITE)
515 failure |= !i40e_alloc_rx_buffers_zc(rx_ring, entries_to_alloc);
516
517 i40e_finalize_xdp_rx(rx_ring, xdp_xmit);
518 i40e_update_rx_stats(rx_ring, total_rx_bytes, total_rx_packets);
519
520 if (xsk_uses_need_wakeup(rx_ring->xsk_pool)) {
521 if (failure || next_to_clean == rx_ring->next_to_use)
522 xsk_set_rx_need_wakeup(rx_ring->xsk_pool);
523 else
524 xsk_clear_rx_need_wakeup(rx_ring->xsk_pool);
525
526 return (int)total_rx_packets;
527 }
528 return failure ? budget : (int)total_rx_packets;
529 }
530
i40e_xmit_pkt(struct i40e_ring * xdp_ring,struct xdp_desc * desc,unsigned int * total_bytes)531 static void i40e_xmit_pkt(struct i40e_ring *xdp_ring, struct xdp_desc *desc,
532 unsigned int *total_bytes)
533 {
534 u32 cmd = I40E_TX_DESC_CMD_ICRC | xsk_is_eop_desc(desc);
535 struct i40e_tx_desc *tx_desc;
536 dma_addr_t dma;
537
538 dma = xsk_buff_raw_get_dma(xdp_ring->xsk_pool, desc->addr);
539 xsk_buff_raw_dma_sync_for_device(xdp_ring->xsk_pool, dma, desc->len);
540
541 tx_desc = I40E_TX_DESC(xdp_ring, xdp_ring->next_to_use++);
542 tx_desc->buffer_addr = cpu_to_le64(dma);
543 tx_desc->cmd_type_offset_bsz = build_ctob(cmd, 0, desc->len, 0);
544
545 *total_bytes += desc->len;
546 }
547
i40e_xmit_pkt_batch(struct i40e_ring * xdp_ring,struct xdp_desc * desc,unsigned int * total_bytes)548 static void i40e_xmit_pkt_batch(struct i40e_ring *xdp_ring, struct xdp_desc *desc,
549 unsigned int *total_bytes)
550 {
551 u16 ntu = xdp_ring->next_to_use;
552 struct i40e_tx_desc *tx_desc;
553 dma_addr_t dma;
554 u32 i;
555
556 loop_unrolled_for(i = 0; i < PKTS_PER_BATCH; i++) {
557 u32 cmd = I40E_TX_DESC_CMD_ICRC | xsk_is_eop_desc(&desc[i]);
558
559 dma = xsk_buff_raw_get_dma(xdp_ring->xsk_pool, desc[i].addr);
560 xsk_buff_raw_dma_sync_for_device(xdp_ring->xsk_pool, dma, desc[i].len);
561
562 tx_desc = I40E_TX_DESC(xdp_ring, ntu++);
563 tx_desc->buffer_addr = cpu_to_le64(dma);
564 tx_desc->cmd_type_offset_bsz = build_ctob(cmd, 0, desc[i].len, 0);
565
566 *total_bytes += desc[i].len;
567 }
568
569 xdp_ring->next_to_use = ntu;
570 }
571
i40e_fill_tx_hw_ring(struct i40e_ring * xdp_ring,struct xdp_desc * descs,u32 nb_pkts,unsigned int * total_bytes)572 static void i40e_fill_tx_hw_ring(struct i40e_ring *xdp_ring, struct xdp_desc *descs, u32 nb_pkts,
573 unsigned int *total_bytes)
574 {
575 u32 batched, leftover, i;
576
577 batched = nb_pkts & ~(PKTS_PER_BATCH - 1);
578 leftover = nb_pkts & (PKTS_PER_BATCH - 1);
579 for (i = 0; i < batched; i += PKTS_PER_BATCH)
580 i40e_xmit_pkt_batch(xdp_ring, &descs[i], total_bytes);
581 for (i = batched; i < batched + leftover; i++)
582 i40e_xmit_pkt(xdp_ring, &descs[i], total_bytes);
583 }
584
i40e_set_rs_bit(struct i40e_ring * xdp_ring)585 static void i40e_set_rs_bit(struct i40e_ring *xdp_ring)
586 {
587 u16 ntu = xdp_ring->next_to_use ? xdp_ring->next_to_use - 1 : xdp_ring->count - 1;
588 struct i40e_tx_desc *tx_desc;
589
590 tx_desc = I40E_TX_DESC(xdp_ring, ntu);
591 tx_desc->cmd_type_offset_bsz |= cpu_to_le64(I40E_TX_DESC_CMD_RS << I40E_TXD_QW1_CMD_SHIFT);
592 }
593
594 /**
595 * i40e_xmit_zc - Performs zero-copy Tx AF_XDP
596 * @xdp_ring: XDP Tx ring
597 * @budget: NAPI budget
598 *
599 * Returns true if the work is finished.
600 **/
i40e_xmit_zc(struct i40e_ring * xdp_ring,unsigned int budget)601 static bool i40e_xmit_zc(struct i40e_ring *xdp_ring, unsigned int budget)
602 {
603 struct xdp_desc *descs = xdp_ring->xsk_pool->tx_descs;
604 u32 nb_pkts, nb_processed = 0;
605 unsigned int total_bytes = 0;
606
607 nb_pkts = xsk_tx_peek_release_desc_batch(xdp_ring->xsk_pool, budget);
608 if (!nb_pkts)
609 return true;
610
611 if (xdp_ring->next_to_use + nb_pkts >= xdp_ring->count) {
612 nb_processed = xdp_ring->count - xdp_ring->next_to_use;
613 i40e_fill_tx_hw_ring(xdp_ring, descs, nb_processed, &total_bytes);
614 xdp_ring->next_to_use = 0;
615 }
616
617 i40e_fill_tx_hw_ring(xdp_ring, &descs[nb_processed], nb_pkts - nb_processed,
618 &total_bytes);
619
620 /* Request an interrupt for the last frame and bump tail ptr. */
621 i40e_set_rs_bit(xdp_ring);
622 i40e_xdp_ring_update_tail(xdp_ring);
623
624 i40e_update_tx_stats(xdp_ring, nb_pkts, total_bytes);
625
626 return nb_pkts < budget;
627 }
628
629 /**
630 * i40e_clean_xdp_tx_buffer - Frees and unmaps an XDP Tx entry
631 * @tx_ring: XDP Tx ring
632 * @tx_bi: Tx buffer info to clean
633 **/
i40e_clean_xdp_tx_buffer(struct i40e_ring * tx_ring,struct i40e_tx_buffer * tx_bi)634 static void i40e_clean_xdp_tx_buffer(struct i40e_ring *tx_ring,
635 struct i40e_tx_buffer *tx_bi)
636 {
637 xdp_return_frame(tx_bi->xdpf);
638 tx_ring->xdp_tx_active--;
639 dma_unmap_single(tx_ring->dev,
640 dma_unmap_addr(tx_bi, dma),
641 dma_unmap_len(tx_bi, len), DMA_TO_DEVICE);
642 dma_unmap_len_set(tx_bi, len, 0);
643 }
644
645 /**
646 * i40e_clean_xdp_tx_irq - Completes AF_XDP entries, and cleans XDP entries
647 * @vsi: Current VSI
648 * @tx_ring: XDP Tx ring
649 *
650 * Returns true if cleanup/transmission is done.
651 **/
i40e_clean_xdp_tx_irq(struct i40e_vsi * vsi,struct i40e_ring * tx_ring)652 bool i40e_clean_xdp_tx_irq(struct i40e_vsi *vsi, struct i40e_ring *tx_ring)
653 {
654 struct xsk_buff_pool *bp = tx_ring->xsk_pool;
655 u32 i, completed_frames, xsk_frames = 0;
656 u32 head_idx = i40e_get_head(tx_ring);
657 struct i40e_tx_buffer *tx_bi;
658 unsigned int ntc;
659
660 if (head_idx < tx_ring->next_to_clean)
661 head_idx += tx_ring->count;
662 completed_frames = head_idx - tx_ring->next_to_clean;
663
664 if (completed_frames == 0)
665 goto out_xmit;
666
667 if (likely(!tx_ring->xdp_tx_active)) {
668 xsk_frames = completed_frames;
669 goto skip;
670 }
671
672 ntc = tx_ring->next_to_clean;
673
674 for (i = 0; i < completed_frames; i++) {
675 tx_bi = &tx_ring->tx_bi[ntc];
676
677 if (tx_bi->xdpf) {
678 i40e_clean_xdp_tx_buffer(tx_ring, tx_bi);
679 tx_bi->xdpf = NULL;
680 } else {
681 xsk_frames++;
682 }
683
684 if (++ntc >= tx_ring->count)
685 ntc = 0;
686 }
687
688 skip:
689 tx_ring->next_to_clean += completed_frames;
690 if (unlikely(tx_ring->next_to_clean >= tx_ring->count))
691 tx_ring->next_to_clean -= tx_ring->count;
692
693 if (xsk_frames)
694 xsk_tx_completed(bp, xsk_frames);
695
696 i40e_arm_wb(tx_ring, vsi, completed_frames);
697
698 out_xmit:
699 if (xsk_uses_need_wakeup(tx_ring->xsk_pool))
700 xsk_set_tx_need_wakeup(tx_ring->xsk_pool);
701
702 return i40e_xmit_zc(tx_ring, I40E_DESC_UNUSED(tx_ring));
703 }
704
705 /**
706 * i40e_xsk_wakeup - Implements the ndo_xsk_wakeup
707 * @dev: the netdevice
708 * @queue_id: queue id to wake up
709 * @flags: ignored in our case since we have Rx and Tx in the same NAPI.
710 *
711 * Returns <0 for errors, 0 otherwise.
712 **/
i40e_xsk_wakeup(struct net_device * dev,u32 queue_id,u32 flags)713 int i40e_xsk_wakeup(struct net_device *dev, u32 queue_id, u32 flags)
714 {
715 struct i40e_netdev_priv *np = netdev_priv(dev);
716 struct i40e_vsi *vsi = np->vsi;
717 struct i40e_pf *pf = vsi->back;
718 struct i40e_ring *ring;
719
720 if (test_bit(__I40E_CONFIG_BUSY, pf->state))
721 return -EAGAIN;
722
723 if (test_bit(__I40E_VSI_DOWN, vsi->state))
724 return -ENETDOWN;
725
726 if (!i40e_enabled_xdp_vsi(vsi))
727 return -EINVAL;
728
729 if (queue_id >= vsi->num_queue_pairs)
730 return -EINVAL;
731
732 if (!vsi->xdp_rings[queue_id]->xsk_pool)
733 return -EINVAL;
734
735 ring = vsi->xdp_rings[queue_id];
736
737 /* The idea here is that if NAPI is running, mark a miss, so
738 * it will run again. If not, trigger an interrupt and
739 * schedule the NAPI from interrupt context. If NAPI would be
740 * scheduled here, the interrupt affinity would not be
741 * honored.
742 */
743 if (!napi_if_scheduled_mark_missed(&ring->q_vector->napi))
744 i40e_force_wb(vsi, ring->q_vector);
745
746 return 0;
747 }
748
i40e_xsk_clean_rx_ring(struct i40e_ring * rx_ring)749 void i40e_xsk_clean_rx_ring(struct i40e_ring *rx_ring)
750 {
751 u16 ntc = rx_ring->next_to_clean;
752 u16 ntu = rx_ring->next_to_use;
753
754 while (ntc != ntu) {
755 struct xdp_buff *rx_bi = *i40e_rx_bi(rx_ring, ntc);
756
757 xsk_buff_free(rx_bi);
758 ntc++;
759 if (ntc >= rx_ring->count)
760 ntc = 0;
761 }
762 }
763
764 /**
765 * i40e_xsk_clean_tx_ring - Clean the XDP Tx ring on shutdown
766 * @tx_ring: XDP Tx ring
767 **/
i40e_xsk_clean_tx_ring(struct i40e_ring * tx_ring)768 void i40e_xsk_clean_tx_ring(struct i40e_ring *tx_ring)
769 {
770 u16 ntc = tx_ring->next_to_clean, ntu = tx_ring->next_to_use;
771 struct xsk_buff_pool *bp = tx_ring->xsk_pool;
772 struct i40e_tx_buffer *tx_bi;
773 u32 xsk_frames = 0;
774
775 while (ntc != ntu) {
776 tx_bi = &tx_ring->tx_bi[ntc];
777
778 if (tx_bi->xdpf)
779 i40e_clean_xdp_tx_buffer(tx_ring, tx_bi);
780 else
781 xsk_frames++;
782
783 tx_bi->xdpf = NULL;
784
785 ntc++;
786 if (ntc >= tx_ring->count)
787 ntc = 0;
788 }
789
790 if (xsk_frames)
791 xsk_tx_completed(bp, xsk_frames);
792 }
793
794 /**
795 * i40e_xsk_any_rx_ring_enabled - Checks if Rx rings have an AF_XDP
796 * buffer pool attached
797 * @vsi: vsi
798 *
799 * Returns true if any of the Rx rings has an AF_XDP buffer pool attached
800 **/
i40e_xsk_any_rx_ring_enabled(struct i40e_vsi * vsi)801 bool i40e_xsk_any_rx_ring_enabled(struct i40e_vsi *vsi)
802 {
803 struct net_device *netdev = vsi->netdev;
804 int i;
805
806 for (i = 0; i < vsi->num_queue_pairs; i++) {
807 if (xsk_get_pool_from_qid(netdev, i))
808 return true;
809 }
810
811 return false;
812 }
813