1// SPDX-License-Identifier: GPL-2.0-only
2/* Copyright (C) 2023 Intel Corporation */
3
4#include "idpf.h"
5
6/**
7 * idpf_tx_singleq_csum - Enable tx checksum offloads
8 * @skb: pointer to skb
9 * @off: pointer to struct that holds offload parameters
10 *
11 * Returns 0 or error (negative) if checksum offload cannot be executed, 1
12 * otherwise.
13 */
14static int idpf_tx_singleq_csum(struct sk_buff *skb,
15 struct idpf_tx_offload_params *off)
16{
17 u32 l4_len, l3_len, l2_len;
18 union {
19 struct iphdr *v4;
20 struct ipv6hdr *v6;
21 unsigned char *hdr;
22 } ip;
23 union {
24 struct tcphdr *tcp;
25 unsigned char *hdr;
26 } l4;
27 u32 offset, cmd = 0;
28 u8 l4_proto = 0;
29 __be16 frag_off;
30 bool is_tso;
31
32 if (skb->ip_summed != CHECKSUM_PARTIAL)
33 return 0;
34
35 ip.hdr = skb_network_header(skb);
36 l4.hdr = skb_transport_header(skb);
37
38 /* compute outer L2 header size */
39 l2_len = ip.hdr - skb->data;
40 offset = FIELD_PREP(0x3F << IDPF_TX_DESC_LEN_MACLEN_S, l2_len / 2);
41 is_tso = !!(off->tx_flags & IDPF_TX_FLAGS_TSO);
42 if (skb->encapsulation) {
43 u32 tunnel = 0;
44
45 /* define outer network header type */
46 if (off->tx_flags & IDPF_TX_FLAGS_IPV4) {
47 /* The stack computes the IP header already, the only
48 * time we need the hardware to recompute it is in the
49 * case of TSO.
50 */
51 tunnel |= is_tso ?
52 IDPF_TX_CTX_EXT_IP_IPV4 :
53 IDPF_TX_CTX_EXT_IP_IPV4_NO_CSUM;
54
55 l4_proto = ip.v4->protocol;
56 } else if (off->tx_flags & IDPF_TX_FLAGS_IPV6) {
57 tunnel |= IDPF_TX_CTX_EXT_IP_IPV6;
58
59 l4_proto = ip.v6->nexthdr;
60 if (ipv6_ext_hdr(nexthdr: l4_proto))
61 ipv6_skip_exthdr(skb, start: skb_network_offset(skb) +
62 sizeof(*ip.v6),
63 nexthdrp: &l4_proto, frag_offp: &frag_off);
64 }
65
66 /* define outer transport */
67 switch (l4_proto) {
68 case IPPROTO_UDP:
69 tunnel |= IDPF_TXD_CTX_UDP_TUNNELING;
70 break;
71 case IPPROTO_GRE:
72 tunnel |= IDPF_TXD_CTX_GRE_TUNNELING;
73 break;
74 case IPPROTO_IPIP:
75 case IPPROTO_IPV6:
76 l4.hdr = skb_inner_network_header(skb);
77 break;
78 default:
79 if (is_tso)
80 return -1;
81
82 skb_checksum_help(skb);
83
84 return 0;
85 }
86 off->tx_flags |= IDPF_TX_FLAGS_TUNNEL;
87
88 /* compute outer L3 header size */
89 tunnel |= FIELD_PREP(IDPF_TXD_CTX_QW0_TUNN_EXT_IPLEN_M,
90 (l4.hdr - ip.hdr) / 4);
91
92 /* switch IP header pointer from outer to inner header */
93 ip.hdr = skb_inner_network_header(skb);
94
95 /* compute tunnel header size */
96 tunnel |= FIELD_PREP(IDPF_TXD_CTX_QW0_TUNN_NATLEN_M,
97 (ip.hdr - l4.hdr) / 2);
98
99 /* indicate if we need to offload outer UDP header */
100 if (is_tso &&
101 !(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
102 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM))
103 tunnel |= IDPF_TXD_CTX_QW0_TUNN_L4T_CS_M;
104
105 /* record tunnel offload values */
106 off->cd_tunneling |= tunnel;
107
108 /* switch L4 header pointer from outer to inner */
109 l4.hdr = skb_inner_transport_header(skb);
110 l4_proto = 0;
111
112 /* reset type as we transition from outer to inner headers */
113 off->tx_flags &= ~(IDPF_TX_FLAGS_IPV4 | IDPF_TX_FLAGS_IPV6);
114 if (ip.v4->version == 4)
115 off->tx_flags |= IDPF_TX_FLAGS_IPV4;
116 if (ip.v6->version == 6)
117 off->tx_flags |= IDPF_TX_FLAGS_IPV6;
118 }
119
120 /* Enable IP checksum offloads */
121 if (off->tx_flags & IDPF_TX_FLAGS_IPV4) {
122 l4_proto = ip.v4->protocol;
123 /* See comment above regarding need for HW to recompute IP
124 * header checksum in the case of TSO.
125 */
126 if (is_tso)
127 cmd |= IDPF_TX_DESC_CMD_IIPT_IPV4_CSUM;
128 else
129 cmd |= IDPF_TX_DESC_CMD_IIPT_IPV4;
130
131 } else if (off->tx_flags & IDPF_TX_FLAGS_IPV6) {
132 cmd |= IDPF_TX_DESC_CMD_IIPT_IPV6;
133 l4_proto = ip.v6->nexthdr;
134 if (ipv6_ext_hdr(nexthdr: l4_proto))
135 ipv6_skip_exthdr(skb, start: skb_network_offset(skb) +
136 sizeof(*ip.v6), nexthdrp: &l4_proto,
137 frag_offp: &frag_off);
138 } else {
139 return -1;
140 }
141
142 /* compute inner L3 header size */
143 l3_len = l4.hdr - ip.hdr;
144 offset |= (l3_len / 4) << IDPF_TX_DESC_LEN_IPLEN_S;
145
146 /* Enable L4 checksum offloads */
147 switch (l4_proto) {
148 case IPPROTO_TCP:
149 /* enable checksum offloads */
150 cmd |= IDPF_TX_DESC_CMD_L4T_EOFT_TCP;
151 l4_len = l4.tcp->doff;
152 break;
153 case IPPROTO_UDP:
154 /* enable UDP checksum offload */
155 cmd |= IDPF_TX_DESC_CMD_L4T_EOFT_UDP;
156 l4_len = sizeof(struct udphdr) >> 2;
157 break;
158 case IPPROTO_SCTP:
159 /* enable SCTP checksum offload */
160 cmd |= IDPF_TX_DESC_CMD_L4T_EOFT_SCTP;
161 l4_len = sizeof(struct sctphdr) >> 2;
162 break;
163 default:
164 if (is_tso)
165 return -1;
166
167 skb_checksum_help(skb);
168
169 return 0;
170 }
171
172 offset |= l4_len << IDPF_TX_DESC_LEN_L4_LEN_S;
173 off->td_cmd |= cmd;
174 off->hdr_offsets |= offset;
175
176 return 1;
177}
178
179/**
180 * idpf_tx_singleq_map - Build the Tx base descriptor
181 * @tx_q: queue to send buffer on
182 * @first: first buffer info buffer to use
183 * @offloads: pointer to struct that holds offload parameters
184 *
185 * This function loops over the skb data pointed to by *first
186 * and gets a physical address for each memory location and programs
187 * it and the length into the transmit base mode descriptor.
188 */
189static void idpf_tx_singleq_map(struct idpf_queue *tx_q,
190 struct idpf_tx_buf *first,
191 struct idpf_tx_offload_params *offloads)
192{
193 u32 offsets = offloads->hdr_offsets;
194 struct idpf_tx_buf *tx_buf = first;
195 struct idpf_base_tx_desc *tx_desc;
196 struct sk_buff *skb = first->skb;
197 u64 td_cmd = offloads->td_cmd;
198 unsigned int data_len, size;
199 u16 i = tx_q->next_to_use;
200 struct netdev_queue *nq;
201 skb_frag_t *frag;
202 dma_addr_t dma;
203 u64 td_tag = 0;
204
205 data_len = skb->data_len;
206 size = skb_headlen(skb);
207
208 tx_desc = IDPF_BASE_TX_DESC(tx_q, i);
209
210 dma = dma_map_single(tx_q->dev, skb->data, size, DMA_TO_DEVICE);
211
212 /* write each descriptor with CRC bit */
213 if (tx_q->vport->crc_enable)
214 td_cmd |= IDPF_TX_DESC_CMD_ICRC;
215
216 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
217 unsigned int max_data = IDPF_TX_MAX_DESC_DATA_ALIGNED;
218
219 if (dma_mapping_error(dev: tx_q->dev, dma_addr: dma))
220 return idpf_tx_dma_map_error(txq: tx_q, skb, first, ring_idx: i);
221
222 /* record length, and DMA address */
223 dma_unmap_len_set(tx_buf, len, size);
224 dma_unmap_addr_set(tx_buf, dma, dma);
225
226 /* align size to end of page */
227 max_data += -dma & (IDPF_TX_MAX_READ_REQ_SIZE - 1);
228 tx_desc->buf_addr = cpu_to_le64(dma);
229
230 /* account for data chunks larger than the hardware
231 * can handle
232 */
233 while (unlikely(size > IDPF_TX_MAX_DESC_DATA)) {
234 tx_desc->qw1 = idpf_tx_singleq_build_ctob(td_cmd,
235 td_offset: offsets,
236 size: max_data,
237 td_tag);
238 tx_desc++;
239 i++;
240
241 if (i == tx_q->desc_count) {
242 tx_desc = IDPF_BASE_TX_DESC(tx_q, 0);
243 i = 0;
244 }
245
246 dma += max_data;
247 size -= max_data;
248
249 max_data = IDPF_TX_MAX_DESC_DATA_ALIGNED;
250 tx_desc->buf_addr = cpu_to_le64(dma);
251 }
252
253 if (!data_len)
254 break;
255
256 tx_desc->qw1 = idpf_tx_singleq_build_ctob(td_cmd, td_offset: offsets,
257 size, td_tag);
258 tx_desc++;
259 i++;
260
261 if (i == tx_q->desc_count) {
262 tx_desc = IDPF_BASE_TX_DESC(tx_q, 0);
263 i = 0;
264 }
265
266 size = skb_frag_size(frag);
267 data_len -= size;
268
269 dma = skb_frag_dma_map(dev: tx_q->dev, frag, offset: 0, size,
270 dir: DMA_TO_DEVICE);
271
272 tx_buf = &tx_q->tx_buf[i];
273 }
274
275 skb_tx_timestamp(skb: first->skb);
276
277 /* write last descriptor with RS and EOP bits */
278 td_cmd |= (u64)(IDPF_TX_DESC_CMD_EOP | IDPF_TX_DESC_CMD_RS);
279
280 tx_desc->qw1 = idpf_tx_singleq_build_ctob(td_cmd, td_offset: offsets,
281 size, td_tag);
282
283 IDPF_SINGLEQ_BUMP_RING_IDX(tx_q, i);
284
285 /* set next_to_watch value indicating a packet is present */
286 first->next_to_watch = tx_desc;
287
288 nq = netdev_get_tx_queue(dev: tx_q->vport->netdev, index: tx_q->idx);
289 netdev_tx_sent_queue(dev_queue: nq, bytes: first->bytecount);
290
291 idpf_tx_buf_hw_update(tx_q, val: i, xmit_more: netdev_xmit_more());
292}
293
294/**
295 * idpf_tx_singleq_get_ctx_desc - grab next desc and update buffer ring
296 * @txq: queue to put context descriptor on
297 *
298 * Since the TX buffer rings mimics the descriptor ring, update the tx buffer
299 * ring entry to reflect that this index is a context descriptor
300 */
301static struct idpf_base_tx_ctx_desc *
302idpf_tx_singleq_get_ctx_desc(struct idpf_queue *txq)
303{
304 struct idpf_base_tx_ctx_desc *ctx_desc;
305 int ntu = txq->next_to_use;
306
307 memset(&txq->tx_buf[ntu], 0, sizeof(struct idpf_tx_buf));
308 txq->tx_buf[ntu].ctx_entry = true;
309
310 ctx_desc = IDPF_BASE_TX_CTX_DESC(txq, ntu);
311
312 IDPF_SINGLEQ_BUMP_RING_IDX(txq, ntu);
313 txq->next_to_use = ntu;
314
315 return ctx_desc;
316}
317
318/**
319 * idpf_tx_singleq_build_ctx_desc - populate context descriptor
320 * @txq: queue to send buffer on
321 * @offload: offload parameter structure
322 **/
323static void idpf_tx_singleq_build_ctx_desc(struct idpf_queue *txq,
324 struct idpf_tx_offload_params *offload)
325{
326 struct idpf_base_tx_ctx_desc *desc = idpf_tx_singleq_get_ctx_desc(txq);
327 u64 qw1 = (u64)IDPF_TX_DESC_DTYPE_CTX;
328
329 if (offload->tso_segs) {
330 qw1 |= IDPF_TX_CTX_DESC_TSO << IDPF_TXD_CTX_QW1_CMD_S;
331 qw1 |= FIELD_PREP(IDPF_TXD_CTX_QW1_TSO_LEN_M,
332 offload->tso_len);
333 qw1 |= FIELD_PREP(IDPF_TXD_CTX_QW1_MSS_M, offload->mss);
334
335 u64_stats_update_begin(syncp: &txq->stats_sync);
336 u64_stats_inc(p: &txq->q_stats.tx.lso_pkts);
337 u64_stats_update_end(syncp: &txq->stats_sync);
338 }
339
340 desc->qw0.tunneling_params = cpu_to_le32(offload->cd_tunneling);
341
342 desc->qw0.l2tag2 = 0;
343 desc->qw0.rsvd1 = 0;
344 desc->qw1 = cpu_to_le64(qw1);
345}
346
347/**
348 * idpf_tx_singleq_frame - Sends buffer on Tx ring using base descriptors
349 * @skb: send buffer
350 * @tx_q: queue to send buffer on
351 *
352 * Returns NETDEV_TX_OK if sent, else an error code
353 */
354static netdev_tx_t idpf_tx_singleq_frame(struct sk_buff *skb,
355 struct idpf_queue *tx_q)
356{
357 struct idpf_tx_offload_params offload = { };
358 struct idpf_tx_buf *first;
359 unsigned int count;
360 __be16 protocol;
361 int csum, tso;
362
363 count = idpf_tx_desc_count_required(txq: tx_q, skb);
364 if (unlikely(!count))
365 return idpf_tx_drop_skb(tx_q, skb);
366
367 if (idpf_tx_maybe_stop_common(tx_q,
368 size: count + IDPF_TX_DESCS_PER_CACHE_LINE +
369 IDPF_TX_DESCS_FOR_CTX)) {
370 idpf_tx_buf_hw_update(tx_q, val: tx_q->next_to_use, xmit_more: false);
371
372 return NETDEV_TX_BUSY;
373 }
374
375 protocol = vlan_get_protocol(skb);
376 if (protocol == htons(ETH_P_IP))
377 offload.tx_flags |= IDPF_TX_FLAGS_IPV4;
378 else if (protocol == htons(ETH_P_IPV6))
379 offload.tx_flags |= IDPF_TX_FLAGS_IPV6;
380
381 tso = idpf_tso(skb, off: &offload);
382 if (tso < 0)
383 goto out_drop;
384
385 csum = idpf_tx_singleq_csum(skb, off: &offload);
386 if (csum < 0)
387 goto out_drop;
388
389 if (tso || offload.cd_tunneling)
390 idpf_tx_singleq_build_ctx_desc(txq: tx_q, offload: &offload);
391
392 /* record the location of the first descriptor for this packet */
393 first = &tx_q->tx_buf[tx_q->next_to_use];
394 first->skb = skb;
395
396 if (tso) {
397 first->gso_segs = offload.tso_segs;
398 first->bytecount = skb->len + ((first->gso_segs - 1) * offload.tso_hdr_len);
399 } else {
400 first->bytecount = max_t(unsigned int, skb->len, ETH_ZLEN);
401 first->gso_segs = 1;
402 }
403 idpf_tx_singleq_map(tx_q, first, offloads: &offload);
404
405 return NETDEV_TX_OK;
406
407out_drop:
408 return idpf_tx_drop_skb(tx_q, skb);
409}
410
411/**
412 * idpf_tx_singleq_start - Selects the right Tx queue to send buffer
413 * @skb: send buffer
414 * @netdev: network interface device structure
415 *
416 * Returns NETDEV_TX_OK if sent, else an error code
417 */
418netdev_tx_t idpf_tx_singleq_start(struct sk_buff *skb,
419 struct net_device *netdev)
420{
421 struct idpf_vport *vport = idpf_netdev_to_vport(netdev);
422 struct idpf_queue *tx_q;
423
424 tx_q = vport->txqs[skb_get_queue_mapping(skb)];
425
426 /* hardware can't handle really short frames, hardware padding works
427 * beyond this point
428 */
429 if (skb_put_padto(skb, IDPF_TX_MIN_PKT_LEN)) {
430 idpf_tx_buf_hw_update(tx_q, val: tx_q->next_to_use, xmit_more: false);
431
432 return NETDEV_TX_OK;
433 }
434
435 return idpf_tx_singleq_frame(skb, tx_q);
436}
437
438/**
439 * idpf_tx_singleq_clean - Reclaim resources from queue
440 * @tx_q: Tx queue to clean
441 * @napi_budget: Used to determine if we are in netpoll
442 * @cleaned: returns number of packets cleaned
443 *
444 */
445static bool idpf_tx_singleq_clean(struct idpf_queue *tx_q, int napi_budget,
446 int *cleaned)
447{
448 unsigned int budget = tx_q->vport->compln_clean_budget;
449 unsigned int total_bytes = 0, total_pkts = 0;
450 struct idpf_base_tx_desc *tx_desc;
451 s16 ntc = tx_q->next_to_clean;
452 struct idpf_netdev_priv *np;
453 struct idpf_tx_buf *tx_buf;
454 struct idpf_vport *vport;
455 struct netdev_queue *nq;
456 bool dont_wake;
457
458 tx_desc = IDPF_BASE_TX_DESC(tx_q, ntc);
459 tx_buf = &tx_q->tx_buf[ntc];
460 ntc -= tx_q->desc_count;
461
462 do {
463 struct idpf_base_tx_desc *eop_desc;
464
465 /* If this entry in the ring was used as a context descriptor,
466 * it's corresponding entry in the buffer ring will indicate as
467 * such. We can skip this descriptor since there is no buffer
468 * to clean.
469 */
470 if (tx_buf->ctx_entry) {
471 /* Clear this flag here to avoid stale flag values when
472 * this buffer is used for actual data in the future.
473 * There are cases where the tx_buf struct / the flags
474 * field will not be cleared before being reused.
475 */
476 tx_buf->ctx_entry = false;
477 goto fetch_next_txq_desc;
478 }
479
480 /* if next_to_watch is not set then no work pending */
481 eop_desc = (struct idpf_base_tx_desc *)tx_buf->next_to_watch;
482 if (!eop_desc)
483 break;
484
485 /* prevent any other reads prior to eop_desc */
486 smp_rmb();
487
488 /* if the descriptor isn't done, no work yet to do */
489 if (!(eop_desc->qw1 &
490 cpu_to_le64(IDPF_TX_DESC_DTYPE_DESC_DONE)))
491 break;
492
493 /* clear next_to_watch to prevent false hangs */
494 tx_buf->next_to_watch = NULL;
495
496 /* update the statistics for this packet */
497 total_bytes += tx_buf->bytecount;
498 total_pkts += tx_buf->gso_segs;
499
500 napi_consume_skb(skb: tx_buf->skb, budget: napi_budget);
501
502 /* unmap skb header data */
503 dma_unmap_single(tx_q->dev,
504 dma_unmap_addr(tx_buf, dma),
505 dma_unmap_len(tx_buf, len),
506 DMA_TO_DEVICE);
507
508 /* clear tx_buf data */
509 tx_buf->skb = NULL;
510 dma_unmap_len_set(tx_buf, len, 0);
511
512 /* unmap remaining buffers */
513 while (tx_desc != eop_desc) {
514 tx_buf++;
515 tx_desc++;
516 ntc++;
517 if (unlikely(!ntc)) {
518 ntc -= tx_q->desc_count;
519 tx_buf = tx_q->tx_buf;
520 tx_desc = IDPF_BASE_TX_DESC(tx_q, 0);
521 }
522
523 /* unmap any remaining paged data */
524 if (dma_unmap_len(tx_buf, len)) {
525 dma_unmap_page(tx_q->dev,
526 dma_unmap_addr(tx_buf, dma),
527 dma_unmap_len(tx_buf, len),
528 DMA_TO_DEVICE);
529 dma_unmap_len_set(tx_buf, len, 0);
530 }
531 }
532
533 /* update budget only if we did something */
534 budget--;
535
536fetch_next_txq_desc:
537 tx_buf++;
538 tx_desc++;
539 ntc++;
540 if (unlikely(!ntc)) {
541 ntc -= tx_q->desc_count;
542 tx_buf = tx_q->tx_buf;
543 tx_desc = IDPF_BASE_TX_DESC(tx_q, 0);
544 }
545 } while (likely(budget));
546
547 ntc += tx_q->desc_count;
548 tx_q->next_to_clean = ntc;
549
550 *cleaned += total_pkts;
551
552 u64_stats_update_begin(syncp: &tx_q->stats_sync);
553 u64_stats_add(p: &tx_q->q_stats.tx.packets, val: total_pkts);
554 u64_stats_add(p: &tx_q->q_stats.tx.bytes, val: total_bytes);
555 u64_stats_update_end(syncp: &tx_q->stats_sync);
556
557 vport = tx_q->vport;
558 np = netdev_priv(dev: vport->netdev);
559 nq = netdev_get_tx_queue(dev: vport->netdev, index: tx_q->idx);
560
561 dont_wake = np->state != __IDPF_VPORT_UP ||
562 !netif_carrier_ok(dev: vport->netdev);
563 __netif_txq_completed_wake(nq, total_pkts, total_bytes,
564 IDPF_DESC_UNUSED(tx_q), IDPF_TX_WAKE_THRESH,
565 dont_wake);
566
567 return !!budget;
568}
569
570/**
571 * idpf_tx_singleq_clean_all - Clean all Tx queues
572 * @q_vec: queue vector
573 * @budget: Used to determine if we are in netpoll
574 * @cleaned: returns number of packets cleaned
575 *
576 * Returns false if clean is not complete else returns true
577 */
578static bool idpf_tx_singleq_clean_all(struct idpf_q_vector *q_vec, int budget,
579 int *cleaned)
580{
581 u16 num_txq = q_vec->num_txq;
582 bool clean_complete = true;
583 int i, budget_per_q;
584
585 budget_per_q = num_txq ? max(budget / num_txq, 1) : 0;
586 for (i = 0; i < num_txq; i++) {
587 struct idpf_queue *q;
588
589 q = q_vec->tx[i];
590 clean_complete &= idpf_tx_singleq_clean(tx_q: q, napi_budget: budget_per_q,
591 cleaned);
592 }
593
594 return clean_complete;
595}
596
597/**
598 * idpf_rx_singleq_test_staterr - tests bits in Rx descriptor
599 * status and error fields
600 * @rx_desc: pointer to receive descriptor (in le64 format)
601 * @stat_err_bits: value to mask
602 *
603 * This function does some fast chicanery in order to return the
604 * value of the mask which is really only used for boolean tests.
605 * The status_error_ptype_len doesn't need to be shifted because it begins
606 * at offset zero.
607 */
608static bool idpf_rx_singleq_test_staterr(const union virtchnl2_rx_desc *rx_desc,
609 const u64 stat_err_bits)
610{
611 return !!(rx_desc->base_wb.qword1.status_error_ptype_len &
612 cpu_to_le64(stat_err_bits));
613}
614
615/**
616 * idpf_rx_singleq_is_non_eop - process handling of non-EOP buffers
617 * @rxq: Rx ring being processed
618 * @rx_desc: Rx descriptor for current buffer
619 * @skb: Current socket buffer containing buffer in progress
620 * @ntc: next to clean
621 */
622static bool idpf_rx_singleq_is_non_eop(struct idpf_queue *rxq,
623 union virtchnl2_rx_desc *rx_desc,
624 struct sk_buff *skb, u16 ntc)
625{
626 /* if we are the last buffer then there is nothing else to do */
627 if (likely(idpf_rx_singleq_test_staterr(rx_desc, IDPF_RXD_EOF_SINGLEQ)))
628 return false;
629
630 return true;
631}
632
633/**
634 * idpf_rx_singleq_csum - Indicate in skb if checksum is good
635 * @rxq: Rx ring being processed
636 * @skb: skb currently being received and modified
637 * @csum_bits: checksum bits from descriptor
638 * @ptype: the packet type decoded by hardware
639 *
640 * skb->protocol must be set before this function is called
641 */
642static void idpf_rx_singleq_csum(struct idpf_queue *rxq, struct sk_buff *skb,
643 struct idpf_rx_csum_decoded *csum_bits,
644 u16 ptype)
645{
646 struct idpf_rx_ptype_decoded decoded;
647 bool ipv4, ipv6;
648
649 /* check if Rx checksum is enabled */
650 if (unlikely(!(rxq->vport->netdev->features & NETIF_F_RXCSUM)))
651 return;
652
653 /* check if HW has decoded the packet and checksum */
654 if (unlikely(!(csum_bits->l3l4p)))
655 return;
656
657 decoded = rxq->vport->rx_ptype_lkup[ptype];
658 if (unlikely(!(decoded.known && decoded.outer_ip)))
659 return;
660
661 ipv4 = IDPF_RX_PTYPE_TO_IPV(&decoded, IDPF_RX_PTYPE_OUTER_IPV4);
662 ipv6 = IDPF_RX_PTYPE_TO_IPV(&decoded, IDPF_RX_PTYPE_OUTER_IPV6);
663
664 /* Check if there were any checksum errors */
665 if (unlikely(ipv4 && (csum_bits->ipe || csum_bits->eipe)))
666 goto checksum_fail;
667
668 /* Device could not do any checksum offload for certain extension
669 * headers as indicated by setting IPV6EXADD bit
670 */
671 if (unlikely(ipv6 && csum_bits->ipv6exadd))
672 return;
673
674 /* check for L4 errors and handle packets that were not able to be
675 * checksummed due to arrival speed
676 */
677 if (unlikely(csum_bits->l4e))
678 goto checksum_fail;
679
680 if (unlikely(csum_bits->nat && csum_bits->eudpe))
681 goto checksum_fail;
682
683 /* Handle packets that were not able to be checksummed due to arrival
684 * speed, in this case the stack can compute the csum.
685 */
686 if (unlikely(csum_bits->pprs))
687 return;
688
689 /* If there is an outer header present that might contain a checksum
690 * we need to bump the checksum level by 1 to reflect the fact that
691 * we are indicating we validated the inner checksum.
692 */
693 if (decoded.tunnel_type >= IDPF_RX_PTYPE_TUNNEL_IP_GRENAT)
694 skb->csum_level = 1;
695
696 /* Only report checksum unnecessary for ICMP, TCP, UDP, or SCTP */
697 switch (decoded.inner_prot) {
698 case IDPF_RX_PTYPE_INNER_PROT_ICMP:
699 case IDPF_RX_PTYPE_INNER_PROT_TCP:
700 case IDPF_RX_PTYPE_INNER_PROT_UDP:
701 case IDPF_RX_PTYPE_INNER_PROT_SCTP:
702 skb->ip_summed = CHECKSUM_UNNECESSARY;
703 return;
704 default:
705 return;
706 }
707
708checksum_fail:
709 u64_stats_update_begin(syncp: &rxq->stats_sync);
710 u64_stats_inc(p: &rxq->q_stats.rx.hw_csum_err);
711 u64_stats_update_end(syncp: &rxq->stats_sync);
712}
713
714/**
715 * idpf_rx_singleq_base_csum - Indicate in skb if hw indicated a good cksum
716 * @rx_q: Rx completion queue
717 * @skb: skb currently being received and modified
718 * @rx_desc: the receive descriptor
719 * @ptype: Rx packet type
720 *
721 * This function only operates on the VIRTCHNL2_RXDID_1_32B_BASE_M base 32byte
722 * descriptor writeback format.
723 **/
724static void idpf_rx_singleq_base_csum(struct idpf_queue *rx_q,
725 struct sk_buff *skb,
726 union virtchnl2_rx_desc *rx_desc,
727 u16 ptype)
728{
729 struct idpf_rx_csum_decoded csum_bits;
730 u32 rx_error, rx_status;
731 u64 qword;
732
733 qword = le64_to_cpu(rx_desc->base_wb.qword1.status_error_ptype_len);
734
735 rx_status = FIELD_GET(VIRTCHNL2_RX_BASE_DESC_QW1_STATUS_M, qword);
736 rx_error = FIELD_GET(VIRTCHNL2_RX_BASE_DESC_QW1_ERROR_M, qword);
737
738 csum_bits.ipe = FIELD_GET(VIRTCHNL2_RX_BASE_DESC_ERROR_IPE_M, rx_error);
739 csum_bits.eipe = FIELD_GET(VIRTCHNL2_RX_BASE_DESC_ERROR_EIPE_M,
740 rx_error);
741 csum_bits.l4e = FIELD_GET(VIRTCHNL2_RX_BASE_DESC_ERROR_L4E_M, rx_error);
742 csum_bits.pprs = FIELD_GET(VIRTCHNL2_RX_BASE_DESC_ERROR_PPRS_M,
743 rx_error);
744 csum_bits.l3l4p = FIELD_GET(VIRTCHNL2_RX_BASE_DESC_STATUS_L3L4P_M,
745 rx_status);
746 csum_bits.ipv6exadd = FIELD_GET(VIRTCHNL2_RX_BASE_DESC_STATUS_IPV6EXADD_M,
747 rx_status);
748 csum_bits.nat = 0;
749 csum_bits.eudpe = 0;
750
751 idpf_rx_singleq_csum(rxq: rx_q, skb, csum_bits: &csum_bits, ptype);
752}
753
754/**
755 * idpf_rx_singleq_flex_csum - Indicate in skb if hw indicated a good cksum
756 * @rx_q: Rx completion queue
757 * @skb: skb currently being received and modified
758 * @rx_desc: the receive descriptor
759 * @ptype: Rx packet type
760 *
761 * This function only operates on the VIRTCHNL2_RXDID_2_FLEX_SQ_NIC flexible
762 * descriptor writeback format.
763 **/
764static void idpf_rx_singleq_flex_csum(struct idpf_queue *rx_q,
765 struct sk_buff *skb,
766 union virtchnl2_rx_desc *rx_desc,
767 u16 ptype)
768{
769 struct idpf_rx_csum_decoded csum_bits;
770 u16 rx_status0, rx_status1;
771
772 rx_status0 = le16_to_cpu(rx_desc->flex_nic_wb.status_error0);
773 rx_status1 = le16_to_cpu(rx_desc->flex_nic_wb.status_error1);
774
775 csum_bits.ipe = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_STATUS0_XSUM_IPE_M,
776 rx_status0);
777 csum_bits.eipe = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_STATUS0_XSUM_EIPE_M,
778 rx_status0);
779 csum_bits.l4e = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_STATUS0_XSUM_L4E_M,
780 rx_status0);
781 csum_bits.eudpe = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_STATUS0_XSUM_EUDPE_M,
782 rx_status0);
783 csum_bits.l3l4p = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_STATUS0_L3L4P_M,
784 rx_status0);
785 csum_bits.ipv6exadd = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_STATUS0_IPV6EXADD_M,
786 rx_status0);
787 csum_bits.nat = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_STATUS1_NAT_M,
788 rx_status1);
789 csum_bits.pprs = 0;
790
791 idpf_rx_singleq_csum(rxq: rx_q, skb, csum_bits: &csum_bits, ptype);
792}
793
794/**
795 * idpf_rx_singleq_base_hash - set the hash value in the skb
796 * @rx_q: Rx completion queue
797 * @skb: skb currently being received and modified
798 * @rx_desc: specific descriptor
799 * @decoded: Decoded Rx packet type related fields
800 *
801 * This function only operates on the VIRTCHNL2_RXDID_1_32B_BASE_M base 32byte
802 * descriptor writeback format.
803 **/
804static void idpf_rx_singleq_base_hash(struct idpf_queue *rx_q,
805 struct sk_buff *skb,
806 union virtchnl2_rx_desc *rx_desc,
807 struct idpf_rx_ptype_decoded *decoded)
808{
809 u64 mask, qw1;
810
811 if (unlikely(!(rx_q->vport->netdev->features & NETIF_F_RXHASH)))
812 return;
813
814 mask = VIRTCHNL2_RX_BASE_DESC_FLTSTAT_RSS_HASH_M;
815 qw1 = le64_to_cpu(rx_desc->base_wb.qword1.status_error_ptype_len);
816
817 if (FIELD_GET(mask, qw1) == mask) {
818 u32 hash = le32_to_cpu(rx_desc->base_wb.qword0.hi_dword.rss);
819
820 skb_set_hash(skb, hash, type: idpf_ptype_to_htype(decoded));
821 }
822}
823
824/**
825 * idpf_rx_singleq_flex_hash - set the hash value in the skb
826 * @rx_q: Rx completion queue
827 * @skb: skb currently being received and modified
828 * @rx_desc: specific descriptor
829 * @decoded: Decoded Rx packet type related fields
830 *
831 * This function only operates on the VIRTCHNL2_RXDID_2_FLEX_SQ_NIC flexible
832 * descriptor writeback format.
833 **/
834static void idpf_rx_singleq_flex_hash(struct idpf_queue *rx_q,
835 struct sk_buff *skb,
836 union virtchnl2_rx_desc *rx_desc,
837 struct idpf_rx_ptype_decoded *decoded)
838{
839 if (unlikely(!(rx_q->vport->netdev->features & NETIF_F_RXHASH)))
840 return;
841
842 if (FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_STATUS0_RSS_VALID_M,
843 le16_to_cpu(rx_desc->flex_nic_wb.status_error0)))
844 skb_set_hash(skb, le32_to_cpu(rx_desc->flex_nic_wb.rss_hash),
845 type: idpf_ptype_to_htype(decoded));
846}
847
848/**
849 * idpf_rx_singleq_process_skb_fields - Populate skb header fields from Rx
850 * descriptor
851 * @rx_q: Rx ring being processed
852 * @skb: pointer to current skb being populated
853 * @rx_desc: descriptor for skb
854 * @ptype: packet type
855 *
856 * This function checks the ring, descriptor, and packet information in
857 * order to populate the hash, checksum, VLAN, protocol, and
858 * other fields within the skb.
859 */
860static void idpf_rx_singleq_process_skb_fields(struct idpf_queue *rx_q,
861 struct sk_buff *skb,
862 union virtchnl2_rx_desc *rx_desc,
863 u16 ptype)
864{
865 struct idpf_rx_ptype_decoded decoded =
866 rx_q->vport->rx_ptype_lkup[ptype];
867
868 /* modifies the skb - consumes the enet header */
869 skb->protocol = eth_type_trans(skb, dev: rx_q->vport->netdev);
870
871 /* Check if we're using base mode descriptor IDs */
872 if (rx_q->rxdids == VIRTCHNL2_RXDID_1_32B_BASE_M) {
873 idpf_rx_singleq_base_hash(rx_q, skb, rx_desc, decoded: &decoded);
874 idpf_rx_singleq_base_csum(rx_q, skb, rx_desc, ptype);
875 } else {
876 idpf_rx_singleq_flex_hash(rx_q, skb, rx_desc, decoded: &decoded);
877 idpf_rx_singleq_flex_csum(rx_q, skb, rx_desc, ptype);
878 }
879}
880
881/**
882 * idpf_rx_singleq_buf_hw_alloc_all - Replace used receive buffers
883 * @rx_q: queue for which the hw buffers are allocated
884 * @cleaned_count: number of buffers to replace
885 *
886 * Returns false if all allocations were successful, true if any fail
887 */
888bool idpf_rx_singleq_buf_hw_alloc_all(struct idpf_queue *rx_q,
889 u16 cleaned_count)
890{
891 struct virtchnl2_singleq_rx_buf_desc *desc;
892 u16 nta = rx_q->next_to_alloc;
893 struct idpf_rx_buf *buf;
894
895 if (!cleaned_count)
896 return false;
897
898 desc = IDPF_SINGLEQ_RX_BUF_DESC(rx_q, nta);
899 buf = &rx_q->rx_buf.buf[nta];
900
901 do {
902 dma_addr_t addr;
903
904 addr = idpf_alloc_page(pool: rx_q->pp, buf, buf_size: rx_q->rx_buf_size);
905 if (unlikely(addr == DMA_MAPPING_ERROR))
906 break;
907
908 /* Refresh the desc even if buffer_addrs didn't change
909 * because each write-back erases this info.
910 */
911 desc->pkt_addr = cpu_to_le64(addr);
912 desc->hdr_addr = 0;
913 desc++;
914
915 buf++;
916 nta++;
917 if (unlikely(nta == rx_q->desc_count)) {
918 desc = IDPF_SINGLEQ_RX_BUF_DESC(rx_q, 0);
919 buf = rx_q->rx_buf.buf;
920 nta = 0;
921 }
922
923 cleaned_count--;
924 } while (cleaned_count);
925
926 if (rx_q->next_to_alloc != nta) {
927 idpf_rx_buf_hw_update(rxq: rx_q, val: nta);
928 rx_q->next_to_alloc = nta;
929 }
930
931 return !!cleaned_count;
932}
933
934/**
935 * idpf_rx_singleq_extract_base_fields - Extract fields from the Rx descriptor
936 * @rx_q: Rx descriptor queue
937 * @rx_desc: the descriptor to process
938 * @fields: storage for extracted values
939 *
940 * Decode the Rx descriptor and extract relevant information including the
941 * size and Rx packet type.
942 *
943 * This function only operates on the VIRTCHNL2_RXDID_1_32B_BASE_M base 32byte
944 * descriptor writeback format.
945 */
946static void idpf_rx_singleq_extract_base_fields(struct idpf_queue *rx_q,
947 union virtchnl2_rx_desc *rx_desc,
948 struct idpf_rx_extracted *fields)
949{
950 u64 qword;
951
952 qword = le64_to_cpu(rx_desc->base_wb.qword1.status_error_ptype_len);
953
954 fields->size = FIELD_GET(VIRTCHNL2_RX_BASE_DESC_QW1_LEN_PBUF_M, qword);
955 fields->rx_ptype = FIELD_GET(VIRTCHNL2_RX_BASE_DESC_QW1_PTYPE_M, qword);
956}
957
958/**
959 * idpf_rx_singleq_extract_flex_fields - Extract fields from the Rx descriptor
960 * @rx_q: Rx descriptor queue
961 * @rx_desc: the descriptor to process
962 * @fields: storage for extracted values
963 *
964 * Decode the Rx descriptor and extract relevant information including the
965 * size and Rx packet type.
966 *
967 * This function only operates on the VIRTCHNL2_RXDID_2_FLEX_SQ_NIC flexible
968 * descriptor writeback format.
969 */
970static void idpf_rx_singleq_extract_flex_fields(struct idpf_queue *rx_q,
971 union virtchnl2_rx_desc *rx_desc,
972 struct idpf_rx_extracted *fields)
973{
974 fields->size = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_PKT_LEN_M,
975 le16_to_cpu(rx_desc->flex_nic_wb.pkt_len));
976 fields->rx_ptype = FIELD_GET(VIRTCHNL2_RX_FLEX_DESC_PTYPE_M,
977 le16_to_cpu(rx_desc->flex_nic_wb.ptype_flex_flags0));
978}
979
980/**
981 * idpf_rx_singleq_extract_fields - Extract fields from the Rx descriptor
982 * @rx_q: Rx descriptor queue
983 * @rx_desc: the descriptor to process
984 * @fields: storage for extracted values
985 *
986 */
987static void idpf_rx_singleq_extract_fields(struct idpf_queue *rx_q,
988 union virtchnl2_rx_desc *rx_desc,
989 struct idpf_rx_extracted *fields)
990{
991 if (rx_q->rxdids == VIRTCHNL2_RXDID_1_32B_BASE_M)
992 idpf_rx_singleq_extract_base_fields(rx_q, rx_desc, fields);
993 else
994 idpf_rx_singleq_extract_flex_fields(rx_q, rx_desc, fields);
995}
996
997/**
998 * idpf_rx_singleq_clean - Reclaim resources after receive completes
999 * @rx_q: rx queue to clean
1000 * @budget: Total limit on number of packets to process
1001 *
1002 * Returns true if there's any budget left (e.g. the clean is finished)
1003 */
1004static int idpf_rx_singleq_clean(struct idpf_queue *rx_q, int budget)
1005{
1006 unsigned int total_rx_bytes = 0, total_rx_pkts = 0;
1007 struct sk_buff *skb = rx_q->skb;
1008 u16 ntc = rx_q->next_to_clean;
1009 u16 cleaned_count = 0;
1010 bool failure = false;
1011
1012 /* Process Rx packets bounded by budget */
1013 while (likely(total_rx_pkts < (unsigned int)budget)) {
1014 struct idpf_rx_extracted fields = { };
1015 union virtchnl2_rx_desc *rx_desc;
1016 struct idpf_rx_buf *rx_buf;
1017
1018 /* get the Rx desc from Rx queue based on 'next_to_clean' */
1019 rx_desc = IDPF_RX_DESC(rx_q, ntc);
1020
1021 /* status_error_ptype_len will always be zero for unused
1022 * descriptors because it's cleared in cleanup, and overlaps
1023 * with hdr_addr which is always zero because packet split
1024 * isn't used, if the hardware wrote DD then the length will be
1025 * non-zero
1026 */
1027#define IDPF_RXD_DD VIRTCHNL2_RX_BASE_DESC_STATUS_DD_M
1028 if (!idpf_rx_singleq_test_staterr(rx_desc,
1029 IDPF_RXD_DD))
1030 break;
1031
1032 /* This memory barrier is needed to keep us from reading
1033 * any other fields out of the rx_desc
1034 */
1035 dma_rmb();
1036
1037 idpf_rx_singleq_extract_fields(rx_q, rx_desc, fields: &fields);
1038
1039 rx_buf = &rx_q->rx_buf.buf[ntc];
1040 if (!fields.size) {
1041 idpf_rx_put_page(rx_buf);
1042 goto skip_data;
1043 }
1044
1045 idpf_rx_sync_for_cpu(rx_buf, len: fields.size);
1046 if (skb)
1047 idpf_rx_add_frag(rx_buf, skb, size: fields.size);
1048 else
1049 skb = idpf_rx_construct_skb(rxq: rx_q, rx_buf, size: fields.size);
1050
1051 /* exit if we failed to retrieve a buffer */
1052 if (!skb)
1053 break;
1054
1055skip_data:
1056 IDPF_SINGLEQ_BUMP_RING_IDX(rx_q, ntc);
1057
1058 cleaned_count++;
1059
1060 /* skip if it is non EOP desc */
1061 if (idpf_rx_singleq_is_non_eop(rxq: rx_q, rx_desc, skb, ntc))
1062 continue;
1063
1064#define IDPF_RXD_ERR_S FIELD_PREP(VIRTCHNL2_RX_BASE_DESC_QW1_ERROR_M, \
1065 VIRTCHNL2_RX_BASE_DESC_ERROR_RXE_M)
1066 if (unlikely(idpf_rx_singleq_test_staterr(rx_desc,
1067 IDPF_RXD_ERR_S))) {
1068 dev_kfree_skb_any(skb);
1069 skb = NULL;
1070 continue;
1071 }
1072
1073 /* pad skb if needed (to make valid ethernet frame) */
1074 if (eth_skb_pad(skb)) {
1075 skb = NULL;
1076 continue;
1077 }
1078
1079 /* probably a little skewed due to removing CRC */
1080 total_rx_bytes += skb->len;
1081
1082 /* protocol */
1083 idpf_rx_singleq_process_skb_fields(rx_q, skb,
1084 rx_desc, ptype: fields.rx_ptype);
1085
1086 /* send completed skb up the stack */
1087 napi_gro_receive(napi: &rx_q->q_vector->napi, skb);
1088 skb = NULL;
1089
1090 /* update budget accounting */
1091 total_rx_pkts++;
1092 }
1093
1094 rx_q->skb = skb;
1095
1096 rx_q->next_to_clean = ntc;
1097
1098 if (cleaned_count)
1099 failure = idpf_rx_singleq_buf_hw_alloc_all(rx_q, cleaned_count);
1100
1101 u64_stats_update_begin(syncp: &rx_q->stats_sync);
1102 u64_stats_add(p: &rx_q->q_stats.rx.packets, val: total_rx_pkts);
1103 u64_stats_add(p: &rx_q->q_stats.rx.bytes, val: total_rx_bytes);
1104 u64_stats_update_end(syncp: &rx_q->stats_sync);
1105
1106 /* guarantee a trip back through this routine if there was a failure */
1107 return failure ? budget : (int)total_rx_pkts;
1108}
1109
1110/**
1111 * idpf_rx_singleq_clean_all - Clean all Rx queues
1112 * @q_vec: queue vector
1113 * @budget: Used to determine if we are in netpoll
1114 * @cleaned: returns number of packets cleaned
1115 *
1116 * Returns false if clean is not complete else returns true
1117 */
1118static bool idpf_rx_singleq_clean_all(struct idpf_q_vector *q_vec, int budget,
1119 int *cleaned)
1120{
1121 u16 num_rxq = q_vec->num_rxq;
1122 bool clean_complete = true;
1123 int budget_per_q, i;
1124
1125 /* We attempt to distribute budget to each Rx queue fairly, but don't
1126 * allow the budget to go below 1 because that would exit polling early.
1127 */
1128 budget_per_q = num_rxq ? max(budget / num_rxq, 1) : 0;
1129 for (i = 0; i < num_rxq; i++) {
1130 struct idpf_queue *rxq = q_vec->rx[i];
1131 int pkts_cleaned_per_q;
1132
1133 pkts_cleaned_per_q = idpf_rx_singleq_clean(rx_q: rxq, budget: budget_per_q);
1134
1135 /* if we clean as many as budgeted, we must not be done */
1136 if (pkts_cleaned_per_q >= budget_per_q)
1137 clean_complete = false;
1138 *cleaned += pkts_cleaned_per_q;
1139 }
1140
1141 return clean_complete;
1142}
1143
1144/**
1145 * idpf_vport_singleq_napi_poll - NAPI handler
1146 * @napi: struct from which you get q_vector
1147 * @budget: budget provided by stack
1148 */
1149int idpf_vport_singleq_napi_poll(struct napi_struct *napi, int budget)
1150{
1151 struct idpf_q_vector *q_vector =
1152 container_of(napi, struct idpf_q_vector, napi);
1153 bool clean_complete;
1154 int work_done = 0;
1155
1156 /* Handle case where we are called by netpoll with a budget of 0 */
1157 if (budget <= 0) {
1158 idpf_tx_singleq_clean_all(q_vec: q_vector, budget, cleaned: &work_done);
1159
1160 return budget;
1161 }
1162
1163 clean_complete = idpf_rx_singleq_clean_all(q_vec: q_vector, budget,
1164 cleaned: &work_done);
1165 clean_complete &= idpf_tx_singleq_clean_all(q_vec: q_vector, budget,
1166 cleaned: &work_done);
1167
1168 /* If work not completed, return budget and polling will return */
1169 if (!clean_complete)
1170 return budget;
1171
1172 work_done = min_t(int, work_done, budget - 1);
1173
1174 /* Exit the polling mode, but don't re-enable interrupts if stack might
1175 * poll us due to busy-polling
1176 */
1177 if (likely(napi_complete_done(napi, work_done)))
1178 idpf_vport_intr_update_itr_ena_irq(q_vector);
1179
1180 return work_done;
1181}
1182

source code of linux/drivers/net/ethernet/intel/idpf/idpf_singleq_txrx.c