1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Atlantic Network Driver
3  *
4  * Copyright (C) 2014-2019 aQuantia Corporation
5  * Copyright (C) 2019-2020 Marvell International Ltd.
6  */
7 
8 /* File aq_ring.c: Definition of functions for Rx/Tx rings. */
9 
10 #include "aq_nic.h"
11 #include "aq_hw.h"
12 #include "aq_hw_utils.h"
13 #include "aq_ptp.h"
14 #include "aq_vec.h"
15 #include "aq_main.h"
16 
17 #include <net/xdp.h>
18 #include <linux/filter.h>
19 #include <linux/bpf_trace.h>
20 #include <linux/netdevice.h>
21 #include <linux/etherdevice.h>
22 
23 static void aq_get_rxpages_xdp(struct aq_ring_buff_s *buff,
24 			       struct xdp_buff *xdp)
25 {
26 	struct skb_shared_info *sinfo;
27 	int i;
28 
29 	if (xdp_buff_has_frags(xdp)) {
30 		sinfo = xdp_get_shared_info_from_buff(xdp);
31 
32 		for (i = 0; i < sinfo->nr_frags; i++) {
33 			skb_frag_t *frag = &sinfo->frags[i];
34 
35 			page_ref_inc(skb_frag_page(frag));
36 		}
37 	}
38 	page_ref_inc(buff->rxdata.page);
39 }
40 
41 static inline void aq_free_rxpage(struct aq_rxpage *rxpage, struct device *dev)
42 {
43 	unsigned int len = PAGE_SIZE << rxpage->order;
44 
45 	dma_unmap_page(dev, rxpage->daddr, len, DMA_FROM_DEVICE);
46 
47 	/* Drop the ref for being in the ring. */
48 	__free_pages(rxpage->page, rxpage->order);
49 	rxpage->page = NULL;
50 }
51 
52 static int aq_alloc_rxpages(struct aq_rxpage *rxpage, struct aq_ring_s *rx_ring)
53 {
54 	struct device *dev = aq_nic_get_dev(rx_ring->aq_nic);
55 	unsigned int order = rx_ring->page_order;
56 	struct page *page;
57 	int ret = -ENOMEM;
58 	dma_addr_t daddr;
59 
60 	page = dev_alloc_pages(order);
61 	if (unlikely(!page))
62 		goto err_exit;
63 
64 	daddr = dma_map_page(dev, page, 0, PAGE_SIZE << order,
65 			     DMA_FROM_DEVICE);
66 
67 	if (unlikely(dma_mapping_error(dev, daddr)))
68 		goto free_page;
69 
70 	rxpage->page = page;
71 	rxpage->daddr = daddr;
72 	rxpage->order = order;
73 	rxpage->pg_off = rx_ring->page_offset;
74 
75 	return 0;
76 
77 free_page:
78 	__free_pages(page, order);
79 
80 err_exit:
81 	return ret;
82 }
83 
84 static int aq_get_rxpages(struct aq_ring_s *self, struct aq_ring_buff_s *rxbuf)
85 {
86 	unsigned int order = self->page_order;
87 	u16 page_offset = self->page_offset;
88 	u16 frame_max = self->frame_max;
89 	u16 tail_size = self->tail_size;
90 	int ret;
91 
92 	if (rxbuf->rxdata.page) {
93 		/* One means ring is the only user and can reuse */
94 		if (page_ref_count(rxbuf->rxdata.page) > 1) {
95 			/* Try reuse buffer */
96 			rxbuf->rxdata.pg_off += frame_max + page_offset +
97 						tail_size;
98 			if (rxbuf->rxdata.pg_off + frame_max + tail_size <=
99 			    (PAGE_SIZE << order)) {
100 				u64_stats_update_begin(&self->stats.rx.syncp);
101 				self->stats.rx.pg_flips++;
102 				u64_stats_update_end(&self->stats.rx.syncp);
103 
104 			} else {
105 				/* Buffer exhausted. We have other users and
106 				 * should release this page and realloc
107 				 */
108 				aq_free_rxpage(&rxbuf->rxdata,
109 					       aq_nic_get_dev(self->aq_nic));
110 				u64_stats_update_begin(&self->stats.rx.syncp);
111 				self->stats.rx.pg_losts++;
112 				u64_stats_update_end(&self->stats.rx.syncp);
113 			}
114 		} else {
115 			rxbuf->rxdata.pg_off = page_offset;
116 			u64_stats_update_begin(&self->stats.rx.syncp);
117 			self->stats.rx.pg_reuses++;
118 			u64_stats_update_end(&self->stats.rx.syncp);
119 		}
120 	}
121 
122 	if (!rxbuf->rxdata.page) {
123 		ret = aq_alloc_rxpages(&rxbuf->rxdata, self);
124 		if (ret) {
125 			u64_stats_update_begin(&self->stats.rx.syncp);
126 			self->stats.rx.alloc_fails++;
127 			u64_stats_update_end(&self->stats.rx.syncp);
128 		}
129 		return ret;
130 	}
131 
132 	return 0;
133 }
134 
135 static struct aq_ring_s *aq_ring_alloc(struct aq_ring_s *self,
136 				       struct aq_nic_s *aq_nic)
137 {
138 	int err = 0;
139 
140 	self->buff_ring =
141 		kcalloc(self->size, sizeof(struct aq_ring_buff_s), GFP_KERNEL);
142 
143 	if (!self->buff_ring) {
144 		err = -ENOMEM;
145 		goto err_exit;
146 	}
147 
148 	self->dx_ring = dma_alloc_coherent(aq_nic_get_dev(aq_nic),
149 					   self->size * self->dx_size,
150 					   &self->dx_ring_pa, GFP_KERNEL);
151 	if (!self->dx_ring) {
152 		err = -ENOMEM;
153 		goto err_exit;
154 	}
155 
156 err_exit:
157 	if (err < 0) {
158 		aq_ring_free(self);
159 		self = NULL;
160 	}
161 
162 	return self;
163 }
164 
165 struct aq_ring_s *aq_ring_tx_alloc(struct aq_ring_s *self,
166 				   struct aq_nic_s *aq_nic,
167 				   unsigned int idx,
168 				   struct aq_nic_cfg_s *aq_nic_cfg)
169 {
170 	int err = 0;
171 
172 	self->aq_nic = aq_nic;
173 	self->idx = idx;
174 	self->size = aq_nic_cfg->txds;
175 	self->dx_size = aq_nic_cfg->aq_hw_caps->txd_size;
176 
177 	self = aq_ring_alloc(self, aq_nic);
178 	if (!self) {
179 		err = -ENOMEM;
180 		goto err_exit;
181 	}
182 
183 err_exit:
184 	if (err < 0) {
185 		aq_ring_free(self);
186 		self = NULL;
187 	}
188 
189 	return self;
190 }
191 
192 struct aq_ring_s *aq_ring_rx_alloc(struct aq_ring_s *self,
193 				   struct aq_nic_s *aq_nic,
194 				   unsigned int idx,
195 				   struct aq_nic_cfg_s *aq_nic_cfg)
196 {
197 	int err = 0;
198 
199 	self->aq_nic = aq_nic;
200 	self->idx = idx;
201 	self->size = aq_nic_cfg->rxds;
202 	self->dx_size = aq_nic_cfg->aq_hw_caps->rxd_size;
203 	self->xdp_prog = aq_nic->xdp_prog;
204 	self->frame_max = AQ_CFG_RX_FRAME_MAX;
205 
206 	/* Only order-2 is allowed if XDP is enabled */
207 	if (READ_ONCE(self->xdp_prog)) {
208 		self->page_offset = AQ_XDP_HEADROOM;
209 		self->page_order = AQ_CFG_XDP_PAGEORDER;
210 		self->tail_size = AQ_XDP_TAILROOM;
211 	} else {
212 		self->page_offset = 0;
213 		self->page_order = fls(self->frame_max / PAGE_SIZE +
214 				       (self->frame_max % PAGE_SIZE ? 1 : 0)) - 1;
215 		if (aq_nic_cfg->rxpageorder > self->page_order)
216 			self->page_order = aq_nic_cfg->rxpageorder;
217 		self->tail_size = 0;
218 	}
219 
220 	self = aq_ring_alloc(self, aq_nic);
221 	if (!self) {
222 		err = -ENOMEM;
223 		goto err_exit;
224 	}
225 
226 err_exit:
227 	if (err < 0) {
228 		aq_ring_free(self);
229 		self = NULL;
230 	}
231 
232 	return self;
233 }
234 
235 struct aq_ring_s *
236 aq_ring_hwts_rx_alloc(struct aq_ring_s *self, struct aq_nic_s *aq_nic,
237 		      unsigned int idx, unsigned int size, unsigned int dx_size)
238 {
239 	struct device *dev = aq_nic_get_dev(aq_nic);
240 	size_t sz = size * dx_size + AQ_CFG_RXDS_DEF;
241 
242 	memset(self, 0, sizeof(*self));
243 
244 	self->aq_nic = aq_nic;
245 	self->idx = idx;
246 	self->size = size;
247 	self->dx_size = dx_size;
248 
249 	self->dx_ring = dma_alloc_coherent(dev, sz, &self->dx_ring_pa,
250 					   GFP_KERNEL);
251 	if (!self->dx_ring) {
252 		aq_ring_free(self);
253 		return NULL;
254 	}
255 
256 	return self;
257 }
258 
259 int aq_ring_init(struct aq_ring_s *self, const enum atl_ring_type ring_type)
260 {
261 	self->hw_head = 0;
262 	self->sw_head = 0;
263 	self->sw_tail = 0;
264 	self->ring_type = ring_type;
265 
266 	if (self->ring_type == ATL_RING_RX)
267 		u64_stats_init(&self->stats.rx.syncp);
268 	else
269 		u64_stats_init(&self->stats.tx.syncp);
270 
271 	return 0;
272 }
273 
274 static inline bool aq_ring_dx_in_range(unsigned int h, unsigned int i,
275 				       unsigned int t)
276 {
277 	return (h < t) ? ((h < i) && (i < t)) : ((h < i) || (i < t));
278 }
279 
280 void aq_ring_update_queue_state(struct aq_ring_s *ring)
281 {
282 	if (aq_ring_avail_dx(ring) <= AQ_CFG_SKB_FRAGS_MAX)
283 		aq_ring_queue_stop(ring);
284 	else if (aq_ring_avail_dx(ring) > AQ_CFG_RESTART_DESC_THRES)
285 		aq_ring_queue_wake(ring);
286 }
287 
288 void aq_ring_queue_wake(struct aq_ring_s *ring)
289 {
290 	struct net_device *ndev = aq_nic_get_ndev(ring->aq_nic);
291 
292 	if (__netif_subqueue_stopped(ndev,
293 				     AQ_NIC_RING2QMAP(ring->aq_nic,
294 						      ring->idx))) {
295 		netif_wake_subqueue(ndev,
296 				    AQ_NIC_RING2QMAP(ring->aq_nic, ring->idx));
297 		u64_stats_update_begin(&ring->stats.tx.syncp);
298 		ring->stats.tx.queue_restarts++;
299 		u64_stats_update_end(&ring->stats.tx.syncp);
300 	}
301 }
302 
303 void aq_ring_queue_stop(struct aq_ring_s *ring)
304 {
305 	struct net_device *ndev = aq_nic_get_ndev(ring->aq_nic);
306 
307 	if (!__netif_subqueue_stopped(ndev,
308 				      AQ_NIC_RING2QMAP(ring->aq_nic,
309 						       ring->idx)))
310 		netif_stop_subqueue(ndev,
311 				    AQ_NIC_RING2QMAP(ring->aq_nic, ring->idx));
312 }
313 
314 bool aq_ring_tx_clean(struct aq_ring_s *self)
315 {
316 	struct device *dev = aq_nic_get_dev(self->aq_nic);
317 	unsigned int budget;
318 
319 	for (budget = AQ_CFG_TX_CLEAN_BUDGET;
320 	     budget && self->sw_head != self->hw_head; budget--) {
321 		struct aq_ring_buff_s *buff = &self->buff_ring[self->sw_head];
322 
323 		if (likely(buff->is_mapped)) {
324 			if (unlikely(buff->is_sop)) {
325 				if (!buff->is_eop &&
326 				    buff->eop_index != 0xffffU &&
327 				    (!aq_ring_dx_in_range(self->sw_head,
328 						buff->eop_index,
329 						self->hw_head)))
330 					break;
331 
332 				dma_unmap_single(dev, buff->pa, buff->len,
333 						 DMA_TO_DEVICE);
334 			} else {
335 				dma_unmap_page(dev, buff->pa, buff->len,
336 					       DMA_TO_DEVICE);
337 			}
338 		}
339 
340 		if (likely(!buff->is_eop))
341 			goto out;
342 
343 		if (buff->skb) {
344 			u64_stats_update_begin(&self->stats.tx.syncp);
345 			++self->stats.tx.packets;
346 			self->stats.tx.bytes += buff->skb->len;
347 			u64_stats_update_end(&self->stats.tx.syncp);
348 			dev_kfree_skb_any(buff->skb);
349 		} else if (buff->xdpf) {
350 			u64_stats_update_begin(&self->stats.tx.syncp);
351 			++self->stats.tx.packets;
352 			self->stats.tx.bytes += xdp_get_frame_len(buff->xdpf);
353 			u64_stats_update_end(&self->stats.tx.syncp);
354 			xdp_return_frame_rx_napi(buff->xdpf);
355 		}
356 
357 out:
358 		buff->skb = NULL;
359 		buff->xdpf = NULL;
360 		buff->pa = 0U;
361 		buff->eop_index = 0xffffU;
362 		self->sw_head = aq_ring_next_dx(self, self->sw_head);
363 	}
364 
365 	return !!budget;
366 }
367 
368 static void aq_rx_checksum(struct aq_ring_s *self,
369 			   struct aq_ring_buff_s *buff,
370 			   struct sk_buff *skb)
371 {
372 	if (!(self->aq_nic->ndev->features & NETIF_F_RXCSUM))
373 		return;
374 
375 	if (unlikely(buff->is_cso_err)) {
376 		u64_stats_update_begin(&self->stats.rx.syncp);
377 		++self->stats.rx.errors;
378 		u64_stats_update_end(&self->stats.rx.syncp);
379 		skb->ip_summed = CHECKSUM_NONE;
380 		return;
381 	}
382 	if (buff->is_ip_cso) {
383 		__skb_incr_checksum_unnecessary(skb);
384 	} else {
385 		skb->ip_summed = CHECKSUM_NONE;
386 	}
387 
388 	if (buff->is_udp_cso || buff->is_tcp_cso)
389 		__skb_incr_checksum_unnecessary(skb);
390 }
391 
392 int aq_xdp_xmit(struct net_device *dev, int num_frames,
393 		struct xdp_frame **frames, u32 flags)
394 {
395 	struct aq_nic_s *aq_nic = netdev_priv(dev);
396 	unsigned int vec, i, drop = 0;
397 	int cpu = smp_processor_id();
398 	struct aq_nic_cfg_s *aq_cfg;
399 	struct aq_ring_s *ring;
400 
401 	aq_cfg = aq_nic_get_cfg(aq_nic);
402 	vec = cpu % aq_cfg->vecs;
403 	ring = aq_nic->aq_ring_tx[AQ_NIC_CFG_TCVEC2RING(aq_cfg, 0, vec)];
404 
405 	for (i = 0; i < num_frames; i++) {
406 		struct xdp_frame *xdpf = frames[i];
407 
408 		if (aq_nic_xmit_xdpf(aq_nic, ring, xdpf) == NETDEV_TX_BUSY)
409 			drop++;
410 	}
411 
412 	return num_frames - drop;
413 }
414 
415 static struct sk_buff *aq_xdp_build_skb(struct xdp_buff *xdp,
416 					struct net_device *dev,
417 					struct aq_ring_buff_s *buff)
418 {
419 	struct xdp_frame *xdpf;
420 	struct sk_buff *skb;
421 
422 	xdpf = xdp_convert_buff_to_frame(xdp);
423 	if (unlikely(!xdpf))
424 		return NULL;
425 
426 	skb = xdp_build_skb_from_frame(xdpf, dev);
427 	if (!skb)
428 		return NULL;
429 
430 	aq_get_rxpages_xdp(buff, xdp);
431 	return skb;
432 }
433 
434 static struct sk_buff *aq_xdp_run_prog(struct aq_nic_s *aq_nic,
435 				       struct xdp_buff *xdp,
436 				       struct aq_ring_s *rx_ring,
437 				       struct aq_ring_buff_s *buff)
438 {
439 	int result = NETDEV_TX_BUSY;
440 	struct aq_ring_s *tx_ring;
441 	struct xdp_frame *xdpf;
442 	struct bpf_prog *prog;
443 	u32 act = XDP_ABORTED;
444 	struct sk_buff *skb;
445 
446 	u64_stats_update_begin(&rx_ring->stats.rx.syncp);
447 	++rx_ring->stats.rx.packets;
448 	rx_ring->stats.rx.bytes += xdp_get_buff_len(xdp);
449 	u64_stats_update_end(&rx_ring->stats.rx.syncp);
450 
451 	prog = READ_ONCE(rx_ring->xdp_prog);
452 	if (!prog)
453 		return aq_xdp_build_skb(xdp, aq_nic->ndev, buff);
454 
455 	prefetchw(xdp->data_hard_start); /* xdp_frame write */
456 
457 	/* single buffer XDP program, but packet is multi buffer, aborted */
458 	if (xdp_buff_has_frags(xdp) && !prog->aux->xdp_has_frags)
459 		goto out_aborted;
460 
461 	act = bpf_prog_run_xdp(prog, xdp);
462 	switch (act) {
463 	case XDP_PASS:
464 		skb = aq_xdp_build_skb(xdp, aq_nic->ndev, buff);
465 		if (!skb)
466 			goto out_aborted;
467 		u64_stats_update_begin(&rx_ring->stats.rx.syncp);
468 		++rx_ring->stats.rx.xdp_pass;
469 		u64_stats_update_end(&rx_ring->stats.rx.syncp);
470 		return skb;
471 	case XDP_TX:
472 		xdpf = xdp_convert_buff_to_frame(xdp);
473 		if (unlikely(!xdpf))
474 			goto out_aborted;
475 		tx_ring = aq_nic->aq_ring_tx[rx_ring->idx];
476 		result = aq_nic_xmit_xdpf(aq_nic, tx_ring, xdpf);
477 		if (result == NETDEV_TX_BUSY)
478 			goto out_aborted;
479 		u64_stats_update_begin(&rx_ring->stats.rx.syncp);
480 		++rx_ring->stats.rx.xdp_tx;
481 		u64_stats_update_end(&rx_ring->stats.rx.syncp);
482 		aq_get_rxpages_xdp(buff, xdp);
483 		break;
484 	case XDP_REDIRECT:
485 		if (xdp_do_redirect(aq_nic->ndev, xdp, prog) < 0)
486 			goto out_aborted;
487 		xdp_do_flush();
488 		u64_stats_update_begin(&rx_ring->stats.rx.syncp);
489 		++rx_ring->stats.rx.xdp_redirect;
490 		u64_stats_update_end(&rx_ring->stats.rx.syncp);
491 		aq_get_rxpages_xdp(buff, xdp);
492 		break;
493 	default:
494 		fallthrough;
495 	case XDP_ABORTED:
496 out_aborted:
497 		u64_stats_update_begin(&rx_ring->stats.rx.syncp);
498 		++rx_ring->stats.rx.xdp_aborted;
499 		u64_stats_update_end(&rx_ring->stats.rx.syncp);
500 		trace_xdp_exception(aq_nic->ndev, prog, act);
501 		bpf_warn_invalid_xdp_action(aq_nic->ndev, prog, act);
502 		break;
503 	case XDP_DROP:
504 		u64_stats_update_begin(&rx_ring->stats.rx.syncp);
505 		++rx_ring->stats.rx.xdp_drop;
506 		u64_stats_update_end(&rx_ring->stats.rx.syncp);
507 		break;
508 	}
509 
510 	return ERR_PTR(-result);
511 }
512 
513 static bool aq_add_rx_fragment(struct device *dev,
514 			       struct aq_ring_s *ring,
515 			       struct aq_ring_buff_s *buff,
516 			       struct xdp_buff *xdp)
517 {
518 	struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
519 	struct aq_ring_buff_s *buff_ = buff;
520 
521 	memset(sinfo, 0, sizeof(*sinfo));
522 	do {
523 		skb_frag_t *frag;
524 
525 		if (unlikely(sinfo->nr_frags >= MAX_SKB_FRAGS))
526 			return true;
527 
528 		frag = &sinfo->frags[sinfo->nr_frags++];
529 		buff_ = &ring->buff_ring[buff_->next];
530 		dma_sync_single_range_for_cpu(dev,
531 					      buff_->rxdata.daddr,
532 					      buff_->rxdata.pg_off,
533 					      buff_->len,
534 					      DMA_FROM_DEVICE);
535 		sinfo->xdp_frags_size += buff_->len;
536 		skb_frag_fill_page_desc(frag, buff_->rxdata.page,
537 					buff_->rxdata.pg_off,
538 					buff_->len);
539 
540 		buff_->is_cleaned = 1;
541 
542 		buff->is_ip_cso &= buff_->is_ip_cso;
543 		buff->is_udp_cso &= buff_->is_udp_cso;
544 		buff->is_tcp_cso &= buff_->is_tcp_cso;
545 		buff->is_cso_err |= buff_->is_cso_err;
546 
547 		if (page_is_pfmemalloc(buff_->rxdata.page))
548 			xdp_buff_set_frag_pfmemalloc(xdp);
549 
550 	} while (!buff_->is_eop);
551 
552 	xdp_buff_set_frags_flag(xdp);
553 
554 	return false;
555 }
556 
557 static int __aq_ring_rx_clean(struct aq_ring_s *self, struct napi_struct *napi,
558 			      int *work_done, int budget)
559 {
560 	struct net_device *ndev = aq_nic_get_ndev(self->aq_nic);
561 	int err = 0;
562 
563 	for (; (self->sw_head != self->hw_head) && budget;
564 		self->sw_head = aq_ring_next_dx(self, self->sw_head),
565 		--budget, ++(*work_done)) {
566 		struct aq_ring_buff_s *buff = &self->buff_ring[self->sw_head];
567 		bool is_ptp_ring = aq_ptp_ring(self->aq_nic, self);
568 		struct aq_ring_buff_s *buff_ = NULL;
569 		struct sk_buff *skb = NULL;
570 		unsigned int next_ = 0U;
571 		unsigned int i = 0U;
572 		u16 hdr_len;
573 
574 		if (buff->is_cleaned)
575 			continue;
576 
577 		if (!buff->is_eop) {
578 			unsigned int frag_cnt = 0U;
579 			buff_ = buff;
580 			do {
581 				bool is_rsc_completed = true;
582 
583 				if (buff_->next >= self->size) {
584 					err = -EIO;
585 					goto err_exit;
586 				}
587 
588 				frag_cnt++;
589 				next_ = buff_->next,
590 				buff_ = &self->buff_ring[next_];
591 				is_rsc_completed =
592 					aq_ring_dx_in_range(self->sw_head,
593 							    next_,
594 							    self->hw_head);
595 
596 				if (unlikely(!is_rsc_completed) ||
597 						frag_cnt > MAX_SKB_FRAGS) {
598 					err = 0;
599 					goto err_exit;
600 				}
601 
602 				buff->is_error |= buff_->is_error;
603 				buff->is_cso_err |= buff_->is_cso_err;
604 
605 			} while (!buff_->is_eop);
606 
607 			if (buff->is_error ||
608 			    (buff->is_lro && buff->is_cso_err)) {
609 				buff_ = buff;
610 				do {
611 					if (buff_->next >= self->size) {
612 						err = -EIO;
613 						goto err_exit;
614 					}
615 					next_ = buff_->next,
616 					buff_ = &self->buff_ring[next_];
617 
618 					buff_->is_cleaned = true;
619 				} while (!buff_->is_eop);
620 
621 				u64_stats_update_begin(&self->stats.rx.syncp);
622 				++self->stats.rx.errors;
623 				u64_stats_update_end(&self->stats.rx.syncp);
624 				continue;
625 			}
626 		}
627 
628 		if (buff->is_error) {
629 			u64_stats_update_begin(&self->stats.rx.syncp);
630 			++self->stats.rx.errors;
631 			u64_stats_update_end(&self->stats.rx.syncp);
632 			continue;
633 		}
634 
635 		dma_sync_single_range_for_cpu(aq_nic_get_dev(self->aq_nic),
636 					      buff->rxdata.daddr,
637 					      buff->rxdata.pg_off,
638 					      buff->len, DMA_FROM_DEVICE);
639 
640 		skb = napi_alloc_skb(napi, AQ_CFG_RX_HDR_SIZE);
641 		if (unlikely(!skb)) {
642 			u64_stats_update_begin(&self->stats.rx.syncp);
643 			self->stats.rx.skb_alloc_fails++;
644 			u64_stats_update_end(&self->stats.rx.syncp);
645 			err = -ENOMEM;
646 			goto err_exit;
647 		}
648 		if (is_ptp_ring)
649 			buff->len -=
650 				aq_ptp_extract_ts(self->aq_nic, skb_hwtstamps(skb),
651 						  aq_buf_vaddr(&buff->rxdata),
652 						  buff->len);
653 
654 		hdr_len = buff->len;
655 		if (hdr_len > AQ_CFG_RX_HDR_SIZE)
656 			hdr_len = eth_get_headlen(skb->dev,
657 						  aq_buf_vaddr(&buff->rxdata),
658 						  AQ_CFG_RX_HDR_SIZE);
659 
660 		memcpy(__skb_put(skb, hdr_len), aq_buf_vaddr(&buff->rxdata),
661 		       ALIGN(hdr_len, sizeof(long)));
662 
663 		if (buff->len - hdr_len > 0) {
664 			skb_add_rx_frag(skb, i++, buff->rxdata.page,
665 					buff->rxdata.pg_off + hdr_len,
666 					buff->len - hdr_len,
667 					self->frame_max);
668 			page_ref_inc(buff->rxdata.page);
669 		}
670 
671 		if (!buff->is_eop) {
672 			buff_ = buff;
673 			do {
674 				next_ = buff_->next;
675 				buff_ = &self->buff_ring[next_];
676 
677 				dma_sync_single_range_for_cpu(aq_nic_get_dev(self->aq_nic),
678 							      buff_->rxdata.daddr,
679 							      buff_->rxdata.pg_off,
680 							      buff_->len,
681 							      DMA_FROM_DEVICE);
682 				skb_add_rx_frag(skb, i++,
683 						buff_->rxdata.page,
684 						buff_->rxdata.pg_off,
685 						buff_->len,
686 						self->frame_max);
687 				page_ref_inc(buff_->rxdata.page);
688 				buff_->is_cleaned = 1;
689 
690 				buff->is_ip_cso &= buff_->is_ip_cso;
691 				buff->is_udp_cso &= buff_->is_udp_cso;
692 				buff->is_tcp_cso &= buff_->is_tcp_cso;
693 				buff->is_cso_err |= buff_->is_cso_err;
694 
695 			} while (!buff_->is_eop);
696 		}
697 
698 		if (buff->is_vlan)
699 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
700 					       buff->vlan_rx_tag);
701 
702 		skb->protocol = eth_type_trans(skb, ndev);
703 
704 		aq_rx_checksum(self, buff, skb);
705 
706 		skb_set_hash(skb, buff->rss_hash,
707 			     buff->is_hash_l4 ? PKT_HASH_TYPE_L4 :
708 			     PKT_HASH_TYPE_NONE);
709 		/* Send all PTP traffic to 0 queue */
710 		skb_record_rx_queue(skb,
711 				    is_ptp_ring ? 0
712 						: AQ_NIC_RING2QMAP(self->aq_nic,
713 								   self->idx));
714 
715 		u64_stats_update_begin(&self->stats.rx.syncp);
716 		++self->stats.rx.packets;
717 		self->stats.rx.bytes += skb->len;
718 		u64_stats_update_end(&self->stats.rx.syncp);
719 
720 		napi_gro_receive(napi, skb);
721 	}
722 
723 err_exit:
724 	return err;
725 }
726 
727 static int __aq_ring_xdp_clean(struct aq_ring_s *rx_ring,
728 			       struct napi_struct *napi, int *work_done,
729 			       int budget)
730 {
731 	int frame_sz = rx_ring->page_offset + rx_ring->frame_max +
732 		       rx_ring->tail_size;
733 	struct aq_nic_s *aq_nic = rx_ring->aq_nic;
734 	bool is_rsc_completed = true;
735 	struct device *dev;
736 	int err = 0;
737 
738 	dev = aq_nic_get_dev(aq_nic);
739 	for (; (rx_ring->sw_head != rx_ring->hw_head) && budget;
740 		rx_ring->sw_head = aq_ring_next_dx(rx_ring, rx_ring->sw_head),
741 		--budget, ++(*work_done)) {
742 		struct aq_ring_buff_s *buff = &rx_ring->buff_ring[rx_ring->sw_head];
743 		bool is_ptp_ring = aq_ptp_ring(rx_ring->aq_nic, rx_ring);
744 		struct aq_ring_buff_s *buff_ = NULL;
745 		u16 ptp_hwtstamp_len = 0;
746 		struct skb_shared_hwtstamps shhwtstamps;
747 		struct sk_buff *skb = NULL;
748 		unsigned int next_ = 0U;
749 		struct xdp_buff xdp;
750 		void *hard_start;
751 
752 		if (buff->is_cleaned)
753 			continue;
754 
755 		if (!buff->is_eop) {
756 			buff_ = buff;
757 			do {
758 				if (buff_->next >= rx_ring->size) {
759 					err = -EIO;
760 					goto err_exit;
761 				}
762 				next_ = buff_->next;
763 				buff_ = &rx_ring->buff_ring[next_];
764 				is_rsc_completed =
765 					aq_ring_dx_in_range(rx_ring->sw_head,
766 							    next_,
767 							    rx_ring->hw_head);
768 
769 				if (unlikely(!is_rsc_completed))
770 					break;
771 
772 				buff->is_error |= buff_->is_error;
773 				buff->is_cso_err |= buff_->is_cso_err;
774 			} while (!buff_->is_eop);
775 
776 			if (!is_rsc_completed) {
777 				err = 0;
778 				goto err_exit;
779 			}
780 			if (buff->is_error ||
781 			    (buff->is_lro && buff->is_cso_err)) {
782 				buff_ = buff;
783 				do {
784 					if (buff_->next >= rx_ring->size) {
785 						err = -EIO;
786 						goto err_exit;
787 					}
788 					next_ = buff_->next;
789 					buff_ = &rx_ring->buff_ring[next_];
790 
791 					buff_->is_cleaned = true;
792 				} while (!buff_->is_eop);
793 
794 				u64_stats_update_begin(&rx_ring->stats.rx.syncp);
795 				++rx_ring->stats.rx.errors;
796 				u64_stats_update_end(&rx_ring->stats.rx.syncp);
797 				continue;
798 			}
799 		}
800 
801 		if (buff->is_error) {
802 			u64_stats_update_begin(&rx_ring->stats.rx.syncp);
803 			++rx_ring->stats.rx.errors;
804 			u64_stats_update_end(&rx_ring->stats.rx.syncp);
805 			continue;
806 		}
807 
808 		dma_sync_single_range_for_cpu(dev,
809 					      buff->rxdata.daddr,
810 					      buff->rxdata.pg_off,
811 					      buff->len, DMA_FROM_DEVICE);
812 		hard_start = page_address(buff->rxdata.page) +
813 			     buff->rxdata.pg_off - rx_ring->page_offset;
814 
815 		if (is_ptp_ring) {
816 			ptp_hwtstamp_len = aq_ptp_extract_ts(rx_ring->aq_nic, &shhwtstamps,
817 							     aq_buf_vaddr(&buff->rxdata),
818 							     buff->len);
819 			buff->len -= ptp_hwtstamp_len;
820 		}
821 
822 		xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
823 		xdp_prepare_buff(&xdp, hard_start, rx_ring->page_offset,
824 				 buff->len, false);
825 		if (!buff->is_eop) {
826 			if (aq_add_rx_fragment(dev, rx_ring, buff, &xdp)) {
827 				u64_stats_update_begin(&rx_ring->stats.rx.syncp);
828 				++rx_ring->stats.rx.packets;
829 				rx_ring->stats.rx.bytes += xdp_get_buff_len(&xdp);
830 				++rx_ring->stats.rx.xdp_aborted;
831 				u64_stats_update_end(&rx_ring->stats.rx.syncp);
832 				continue;
833 			}
834 		}
835 
836 		skb = aq_xdp_run_prog(aq_nic, &xdp, rx_ring, buff);
837 		if (IS_ERR(skb) || !skb)
838 			continue;
839 
840 		if (ptp_hwtstamp_len > 0)
841 			*skb_hwtstamps(skb) = shhwtstamps;
842 
843 		if (buff->is_vlan)
844 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
845 					       buff->vlan_rx_tag);
846 
847 		aq_rx_checksum(rx_ring, buff, skb);
848 
849 		skb_set_hash(skb, buff->rss_hash,
850 			     buff->is_hash_l4 ? PKT_HASH_TYPE_L4 :
851 			     PKT_HASH_TYPE_NONE);
852 		/* Send all PTP traffic to 0 queue */
853 		skb_record_rx_queue(skb,
854 				    is_ptp_ring ? 0
855 						: AQ_NIC_RING2QMAP(rx_ring->aq_nic,
856 								   rx_ring->idx));
857 
858 		napi_gro_receive(napi, skb);
859 	}
860 
861 err_exit:
862 	return err;
863 }
864 
865 int aq_ring_rx_clean(struct aq_ring_s *self,
866 		     struct napi_struct *napi,
867 		     int *work_done,
868 		     int budget)
869 {
870 	if (static_branch_unlikely(&aq_xdp_locking_key))
871 		return __aq_ring_xdp_clean(self, napi, work_done, budget);
872 	else
873 		return __aq_ring_rx_clean(self, napi, work_done, budget);
874 }
875 
876 void aq_ring_hwts_rx_clean(struct aq_ring_s *self, struct aq_nic_s *aq_nic)
877 {
878 #if IS_REACHABLE(CONFIG_PTP_1588_CLOCK)
879 	while (self->sw_head != self->hw_head) {
880 		u64 ns;
881 
882 		aq_nic->aq_hw_ops->extract_hwts(aq_nic->aq_hw,
883 						self->dx_ring +
884 						(self->sw_head * self->dx_size),
885 						self->dx_size, &ns);
886 		aq_ptp_tx_hwtstamp(aq_nic, ns);
887 
888 		self->sw_head = aq_ring_next_dx(self, self->sw_head);
889 	}
890 #endif
891 }
892 
893 int aq_ring_rx_fill(struct aq_ring_s *self)
894 {
895 	struct aq_ring_buff_s *buff = NULL;
896 	int err = 0;
897 	int i = 0;
898 
899 	if (aq_ring_avail_dx(self) < min_t(unsigned int, AQ_CFG_RX_REFILL_THRES,
900 					   self->size / 2))
901 		return err;
902 
903 	for (i = aq_ring_avail_dx(self); i--;
904 		self->sw_tail = aq_ring_next_dx(self, self->sw_tail)) {
905 		buff = &self->buff_ring[self->sw_tail];
906 
907 		buff->flags = 0U;
908 		buff->len = self->frame_max;
909 
910 		err = aq_get_rxpages(self, buff);
911 		if (err)
912 			goto err_exit;
913 
914 		buff->pa = aq_buf_daddr(&buff->rxdata);
915 		buff = NULL;
916 	}
917 
918 err_exit:
919 	return err;
920 }
921 
922 void aq_ring_rx_deinit(struct aq_ring_s *self)
923 {
924 	if (!self)
925 		return;
926 
927 	for (; self->sw_head != self->sw_tail;
928 		self->sw_head = aq_ring_next_dx(self, self->sw_head)) {
929 		struct aq_ring_buff_s *buff = &self->buff_ring[self->sw_head];
930 
931 		aq_free_rxpage(&buff->rxdata, aq_nic_get_dev(self->aq_nic));
932 	}
933 }
934 
935 void aq_ring_free(struct aq_ring_s *self)
936 {
937 	if (!self)
938 		return;
939 
940 	kfree(self->buff_ring);
941 	self->buff_ring = NULL;
942 
943 	if (self->dx_ring) {
944 		dma_free_coherent(aq_nic_get_dev(self->aq_nic),
945 				  self->size * self->dx_size, self->dx_ring,
946 				  self->dx_ring_pa);
947 		self->dx_ring = NULL;
948 	}
949 }
950 
951 unsigned int aq_ring_fill_stats_data(struct aq_ring_s *self, u64 *data)
952 {
953 	unsigned int count;
954 	unsigned int start;
955 
956 	if (self->ring_type == ATL_RING_RX) {
957 		/* This data should mimic aq_ethtool_queue_rx_stat_names structure */
958 		do {
959 			count = 0;
960 			start = u64_stats_fetch_begin(&self->stats.rx.syncp);
961 			data[count] = self->stats.rx.packets;
962 			data[++count] = self->stats.rx.jumbo_packets;
963 			data[++count] = self->stats.rx.lro_packets;
964 			data[++count] = self->stats.rx.errors;
965 			data[++count] = self->stats.rx.alloc_fails;
966 			data[++count] = self->stats.rx.skb_alloc_fails;
967 			data[++count] = self->stats.rx.polls;
968 			data[++count] = self->stats.rx.pg_flips;
969 			data[++count] = self->stats.rx.pg_reuses;
970 			data[++count] = self->stats.rx.pg_losts;
971 			data[++count] = self->stats.rx.xdp_aborted;
972 			data[++count] = self->stats.rx.xdp_drop;
973 			data[++count] = self->stats.rx.xdp_pass;
974 			data[++count] = self->stats.rx.xdp_tx;
975 			data[++count] = self->stats.rx.xdp_invalid;
976 			data[++count] = self->stats.rx.xdp_redirect;
977 		} while (u64_stats_fetch_retry(&self->stats.rx.syncp, start));
978 	} else {
979 		/* This data should mimic aq_ethtool_queue_tx_stat_names structure */
980 		do {
981 			count = 0;
982 			start = u64_stats_fetch_begin(&self->stats.tx.syncp);
983 			data[count] = self->stats.tx.packets;
984 			data[++count] = self->stats.tx.queue_restarts;
985 		} while (u64_stats_fetch_retry(&self->stats.tx.syncp, start));
986 	}
987 
988 	return ++count;
989 }
990