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 
aq_get_rxpages_xdp(struct aq_ring_buff_s * buff,struct xdp_buff * xdp)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 
aq_free_rxpage(struct aq_rxpage * rxpage,struct device * dev)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 
aq_alloc_rxpages(struct aq_rxpage * rxpage,struct aq_ring_s * rx_ring)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 
aq_get_rxpages(struct aq_ring_s * self,struct aq_ring_buff_s * rxbuf)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 
aq_ring_alloc(struct aq_ring_s * self,struct aq_nic_s * aq_nic)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 
aq_ring_tx_alloc(struct aq_ring_s * self,struct aq_nic_s * aq_nic,unsigned int idx,struct aq_nic_cfg_s * aq_nic_cfg)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 
aq_ring_rx_alloc(struct aq_ring_s * self,struct aq_nic_s * aq_nic,unsigned int idx,struct aq_nic_cfg_s * aq_nic_cfg)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 *
aq_ring_hwts_rx_alloc(struct aq_ring_s * self,struct aq_nic_s * aq_nic,unsigned int idx,unsigned int size,unsigned int dx_size)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 
aq_ring_init(struct aq_ring_s * self,const enum atl_ring_type ring_type)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 
aq_ring_dx_in_range(unsigned int h,unsigned int i,unsigned int t)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 
aq_ring_update_queue_state(struct aq_ring_s * ring)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 
aq_ring_queue_wake(struct aq_ring_s * ring)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 
aq_ring_queue_stop(struct aq_ring_s * ring)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 
aq_ring_tx_clean(struct aq_ring_s * self)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 
aq_rx_checksum(struct aq_ring_s * self,struct aq_ring_buff_s * buff,struct sk_buff * skb)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 
aq_xdp_xmit(struct net_device * dev,int num_frames,struct xdp_frame ** frames,u32 flags)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 
aq_xdp_run_prog(struct aq_nic_s * aq_nic,struct xdp_buff * xdp,struct aq_ring_s * rx_ring,struct aq_ring_buff_s * buff)415 static struct sk_buff *aq_xdp_run_prog(struct aq_nic_s *aq_nic,
416 				       struct xdp_buff *xdp,
417 				       struct aq_ring_s *rx_ring,
418 				       struct aq_ring_buff_s *buff)
419 {
420 	int result = NETDEV_TX_BUSY;
421 	struct aq_ring_s *tx_ring;
422 	struct xdp_frame *xdpf;
423 	struct bpf_prog *prog;
424 	u32 act = XDP_ABORTED;
425 	struct sk_buff *skb;
426 
427 	u64_stats_update_begin(&rx_ring->stats.rx.syncp);
428 	++rx_ring->stats.rx.packets;
429 	rx_ring->stats.rx.bytes += xdp_get_buff_len(xdp);
430 	u64_stats_update_end(&rx_ring->stats.rx.syncp);
431 
432 	prog = READ_ONCE(rx_ring->xdp_prog);
433 	if (!prog)
434 		goto pass;
435 
436 	prefetchw(xdp->data_hard_start); /* xdp_frame write */
437 
438 	/* single buffer XDP program, but packet is multi buffer, aborted */
439 	if (xdp_buff_has_frags(xdp) && !prog->aux->xdp_has_frags)
440 		goto out_aborted;
441 
442 	act = bpf_prog_run_xdp(prog, xdp);
443 	switch (act) {
444 	case XDP_PASS:
445 pass:
446 		xdpf = xdp_convert_buff_to_frame(xdp);
447 		if (unlikely(!xdpf))
448 			goto out_aborted;
449 		skb = xdp_build_skb_from_frame(xdpf, aq_nic->ndev);
450 		if (!skb)
451 			goto out_aborted;
452 		u64_stats_update_begin(&rx_ring->stats.rx.syncp);
453 		++rx_ring->stats.rx.xdp_pass;
454 		u64_stats_update_end(&rx_ring->stats.rx.syncp);
455 		aq_get_rxpages_xdp(buff, xdp);
456 		return skb;
457 	case XDP_TX:
458 		xdpf = xdp_convert_buff_to_frame(xdp);
459 		if (unlikely(!xdpf))
460 			goto out_aborted;
461 		tx_ring = aq_nic->aq_ring_tx[rx_ring->idx];
462 		result = aq_nic_xmit_xdpf(aq_nic, tx_ring, xdpf);
463 		if (result == NETDEV_TX_BUSY)
464 			goto out_aborted;
465 		u64_stats_update_begin(&rx_ring->stats.rx.syncp);
466 		++rx_ring->stats.rx.xdp_tx;
467 		u64_stats_update_end(&rx_ring->stats.rx.syncp);
468 		aq_get_rxpages_xdp(buff, xdp);
469 		break;
470 	case XDP_REDIRECT:
471 		if (xdp_do_redirect(aq_nic->ndev, xdp, prog) < 0)
472 			goto out_aborted;
473 		xdp_do_flush();
474 		u64_stats_update_begin(&rx_ring->stats.rx.syncp);
475 		++rx_ring->stats.rx.xdp_redirect;
476 		u64_stats_update_end(&rx_ring->stats.rx.syncp);
477 		aq_get_rxpages_xdp(buff, xdp);
478 		break;
479 	default:
480 		fallthrough;
481 	case XDP_ABORTED:
482 out_aborted:
483 		u64_stats_update_begin(&rx_ring->stats.rx.syncp);
484 		++rx_ring->stats.rx.xdp_aborted;
485 		u64_stats_update_end(&rx_ring->stats.rx.syncp);
486 		trace_xdp_exception(aq_nic->ndev, prog, act);
487 		bpf_warn_invalid_xdp_action(aq_nic->ndev, prog, act);
488 		break;
489 	case XDP_DROP:
490 		u64_stats_update_begin(&rx_ring->stats.rx.syncp);
491 		++rx_ring->stats.rx.xdp_drop;
492 		u64_stats_update_end(&rx_ring->stats.rx.syncp);
493 		break;
494 	}
495 
496 	return ERR_PTR(-result);
497 }
498 
aq_add_rx_fragment(struct device * dev,struct aq_ring_s * ring,struct aq_ring_buff_s * buff,struct xdp_buff * xdp)499 static bool aq_add_rx_fragment(struct device *dev,
500 			       struct aq_ring_s *ring,
501 			       struct aq_ring_buff_s *buff,
502 			       struct xdp_buff *xdp)
503 {
504 	struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
505 	struct aq_ring_buff_s *buff_ = buff;
506 
507 	memset(sinfo, 0, sizeof(*sinfo));
508 	do {
509 		skb_frag_t *frag;
510 
511 		if (unlikely(sinfo->nr_frags >= MAX_SKB_FRAGS))
512 			return true;
513 
514 		frag = &sinfo->frags[sinfo->nr_frags++];
515 		buff_ = &ring->buff_ring[buff_->next];
516 		dma_sync_single_range_for_cpu(dev,
517 					      buff_->rxdata.daddr,
518 					      buff_->rxdata.pg_off,
519 					      buff_->len,
520 					      DMA_FROM_DEVICE);
521 		skb_frag_off_set(frag, buff_->rxdata.pg_off);
522 		skb_frag_size_set(frag, buff_->len);
523 		sinfo->xdp_frags_size += buff_->len;
524 		__skb_frag_set_page(frag, buff_->rxdata.page);
525 
526 		buff_->is_cleaned = 1;
527 
528 		buff->is_ip_cso &= buff_->is_ip_cso;
529 		buff->is_udp_cso &= buff_->is_udp_cso;
530 		buff->is_tcp_cso &= buff_->is_tcp_cso;
531 		buff->is_cso_err |= buff_->is_cso_err;
532 
533 		if (page_is_pfmemalloc(buff_->rxdata.page))
534 			xdp_buff_set_frag_pfmemalloc(xdp);
535 
536 	} while (!buff_->is_eop);
537 
538 	xdp_buff_set_frags_flag(xdp);
539 
540 	return false;
541 }
542 
__aq_ring_rx_clean(struct aq_ring_s * self,struct napi_struct * napi,int * work_done,int budget)543 static int __aq_ring_rx_clean(struct aq_ring_s *self, struct napi_struct *napi,
544 			      int *work_done, int budget)
545 {
546 	struct net_device *ndev = aq_nic_get_ndev(self->aq_nic);
547 	int err = 0;
548 
549 	for (; (self->sw_head != self->hw_head) && budget;
550 		self->sw_head = aq_ring_next_dx(self, self->sw_head),
551 		--budget, ++(*work_done)) {
552 		struct aq_ring_buff_s *buff = &self->buff_ring[self->sw_head];
553 		bool is_ptp_ring = aq_ptp_ring(self->aq_nic, self);
554 		struct aq_ring_buff_s *buff_ = NULL;
555 		struct sk_buff *skb = NULL;
556 		unsigned int next_ = 0U;
557 		unsigned int i = 0U;
558 		u16 hdr_len;
559 
560 		if (buff->is_cleaned)
561 			continue;
562 
563 		if (!buff->is_eop) {
564 			unsigned int frag_cnt = 0U;
565 			buff_ = buff;
566 			do {
567 				bool is_rsc_completed = true;
568 
569 				if (buff_->next >= self->size) {
570 					err = -EIO;
571 					goto err_exit;
572 				}
573 
574 				frag_cnt++;
575 				next_ = buff_->next,
576 				buff_ = &self->buff_ring[next_];
577 				is_rsc_completed =
578 					aq_ring_dx_in_range(self->sw_head,
579 							    next_,
580 							    self->hw_head);
581 
582 				if (unlikely(!is_rsc_completed) ||
583 						frag_cnt > MAX_SKB_FRAGS) {
584 					err = 0;
585 					goto err_exit;
586 				}
587 
588 				buff->is_error |= buff_->is_error;
589 				buff->is_cso_err |= buff_->is_cso_err;
590 
591 			} while (!buff_->is_eop);
592 
593 			if (buff->is_error ||
594 			    (buff->is_lro && buff->is_cso_err)) {
595 				buff_ = buff;
596 				do {
597 					if (buff_->next >= self->size) {
598 						err = -EIO;
599 						goto err_exit;
600 					}
601 					next_ = buff_->next,
602 					buff_ = &self->buff_ring[next_];
603 
604 					buff_->is_cleaned = true;
605 				} while (!buff_->is_eop);
606 
607 				u64_stats_update_begin(&self->stats.rx.syncp);
608 				++self->stats.rx.errors;
609 				u64_stats_update_end(&self->stats.rx.syncp);
610 				continue;
611 			}
612 		}
613 
614 		if (buff->is_error) {
615 			u64_stats_update_begin(&self->stats.rx.syncp);
616 			++self->stats.rx.errors;
617 			u64_stats_update_end(&self->stats.rx.syncp);
618 			continue;
619 		}
620 
621 		dma_sync_single_range_for_cpu(aq_nic_get_dev(self->aq_nic),
622 					      buff->rxdata.daddr,
623 					      buff->rxdata.pg_off,
624 					      buff->len, DMA_FROM_DEVICE);
625 
626 		skb = napi_alloc_skb(napi, AQ_CFG_RX_HDR_SIZE);
627 		if (unlikely(!skb)) {
628 			u64_stats_update_begin(&self->stats.rx.syncp);
629 			self->stats.rx.skb_alloc_fails++;
630 			u64_stats_update_end(&self->stats.rx.syncp);
631 			err = -ENOMEM;
632 			goto err_exit;
633 		}
634 		if (is_ptp_ring)
635 			buff->len -=
636 				aq_ptp_extract_ts(self->aq_nic, skb,
637 						  aq_buf_vaddr(&buff->rxdata),
638 						  buff->len);
639 
640 		hdr_len = buff->len;
641 		if (hdr_len > AQ_CFG_RX_HDR_SIZE)
642 			hdr_len = eth_get_headlen(skb->dev,
643 						  aq_buf_vaddr(&buff->rxdata),
644 						  AQ_CFG_RX_HDR_SIZE);
645 
646 		memcpy(__skb_put(skb, hdr_len), aq_buf_vaddr(&buff->rxdata),
647 		       ALIGN(hdr_len, sizeof(long)));
648 
649 		if (buff->len - hdr_len > 0) {
650 			skb_add_rx_frag(skb, i++, buff->rxdata.page,
651 					buff->rxdata.pg_off + hdr_len,
652 					buff->len - hdr_len,
653 					self->frame_max);
654 			page_ref_inc(buff->rxdata.page);
655 		}
656 
657 		if (!buff->is_eop) {
658 			buff_ = buff;
659 			do {
660 				next_ = buff_->next;
661 				buff_ = &self->buff_ring[next_];
662 
663 				dma_sync_single_range_for_cpu(aq_nic_get_dev(self->aq_nic),
664 							      buff_->rxdata.daddr,
665 							      buff_->rxdata.pg_off,
666 							      buff_->len,
667 							      DMA_FROM_DEVICE);
668 				skb_add_rx_frag(skb, i++,
669 						buff_->rxdata.page,
670 						buff_->rxdata.pg_off,
671 						buff_->len,
672 						self->frame_max);
673 				page_ref_inc(buff_->rxdata.page);
674 				buff_->is_cleaned = 1;
675 
676 				buff->is_ip_cso &= buff_->is_ip_cso;
677 				buff->is_udp_cso &= buff_->is_udp_cso;
678 				buff->is_tcp_cso &= buff_->is_tcp_cso;
679 				buff->is_cso_err |= buff_->is_cso_err;
680 
681 			} while (!buff_->is_eop);
682 		}
683 
684 		if (buff->is_vlan)
685 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
686 					       buff->vlan_rx_tag);
687 
688 		skb->protocol = eth_type_trans(skb, ndev);
689 
690 		aq_rx_checksum(self, buff, skb);
691 
692 		skb_set_hash(skb, buff->rss_hash,
693 			     buff->is_hash_l4 ? PKT_HASH_TYPE_L4 :
694 			     PKT_HASH_TYPE_NONE);
695 		/* Send all PTP traffic to 0 queue */
696 		skb_record_rx_queue(skb,
697 				    is_ptp_ring ? 0
698 						: AQ_NIC_RING2QMAP(self->aq_nic,
699 								   self->idx));
700 
701 		u64_stats_update_begin(&self->stats.rx.syncp);
702 		++self->stats.rx.packets;
703 		self->stats.rx.bytes += skb->len;
704 		u64_stats_update_end(&self->stats.rx.syncp);
705 
706 		napi_gro_receive(napi, skb);
707 	}
708 
709 err_exit:
710 	return err;
711 }
712 
__aq_ring_xdp_clean(struct aq_ring_s * rx_ring,struct napi_struct * napi,int * work_done,int budget)713 static int __aq_ring_xdp_clean(struct aq_ring_s *rx_ring,
714 			       struct napi_struct *napi, int *work_done,
715 			       int budget)
716 {
717 	int frame_sz = rx_ring->page_offset + rx_ring->frame_max +
718 		       rx_ring->tail_size;
719 	struct aq_nic_s *aq_nic = rx_ring->aq_nic;
720 	bool is_rsc_completed = true;
721 	struct device *dev;
722 	int err = 0;
723 
724 	dev = aq_nic_get_dev(aq_nic);
725 	for (; (rx_ring->sw_head != rx_ring->hw_head) && budget;
726 		rx_ring->sw_head = aq_ring_next_dx(rx_ring, rx_ring->sw_head),
727 		--budget, ++(*work_done)) {
728 		struct aq_ring_buff_s *buff = &rx_ring->buff_ring[rx_ring->sw_head];
729 		bool is_ptp_ring = aq_ptp_ring(rx_ring->aq_nic, rx_ring);
730 		struct aq_ring_buff_s *buff_ = NULL;
731 		struct sk_buff *skb = NULL;
732 		unsigned int next_ = 0U;
733 		struct xdp_buff xdp;
734 		void *hard_start;
735 
736 		if (buff->is_cleaned)
737 			continue;
738 
739 		if (!buff->is_eop) {
740 			buff_ = buff;
741 			do {
742 				if (buff_->next >= rx_ring->size) {
743 					err = -EIO;
744 					goto err_exit;
745 				}
746 				next_ = buff_->next;
747 				buff_ = &rx_ring->buff_ring[next_];
748 				is_rsc_completed =
749 					aq_ring_dx_in_range(rx_ring->sw_head,
750 							    next_,
751 							    rx_ring->hw_head);
752 
753 				if (unlikely(!is_rsc_completed))
754 					break;
755 
756 				buff->is_error |= buff_->is_error;
757 				buff->is_cso_err |= buff_->is_cso_err;
758 			} while (!buff_->is_eop);
759 
760 			if (!is_rsc_completed) {
761 				err = 0;
762 				goto err_exit;
763 			}
764 			if (buff->is_error ||
765 			    (buff->is_lro && buff->is_cso_err)) {
766 				buff_ = buff;
767 				do {
768 					if (buff_->next >= rx_ring->size) {
769 						err = -EIO;
770 						goto err_exit;
771 					}
772 					next_ = buff_->next;
773 					buff_ = &rx_ring->buff_ring[next_];
774 
775 					buff_->is_cleaned = true;
776 				} while (!buff_->is_eop);
777 
778 				u64_stats_update_begin(&rx_ring->stats.rx.syncp);
779 				++rx_ring->stats.rx.errors;
780 				u64_stats_update_end(&rx_ring->stats.rx.syncp);
781 				continue;
782 			}
783 		}
784 
785 		if (buff->is_error) {
786 			u64_stats_update_begin(&rx_ring->stats.rx.syncp);
787 			++rx_ring->stats.rx.errors;
788 			u64_stats_update_end(&rx_ring->stats.rx.syncp);
789 			continue;
790 		}
791 
792 		dma_sync_single_range_for_cpu(dev,
793 					      buff->rxdata.daddr,
794 					      buff->rxdata.pg_off,
795 					      buff->len, DMA_FROM_DEVICE);
796 		hard_start = page_address(buff->rxdata.page) +
797 			     buff->rxdata.pg_off - rx_ring->page_offset;
798 
799 		if (is_ptp_ring)
800 			buff->len -=
801 				aq_ptp_extract_ts(rx_ring->aq_nic, skb,
802 						  aq_buf_vaddr(&buff->rxdata),
803 						  buff->len);
804 
805 		xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
806 		xdp_prepare_buff(&xdp, hard_start, rx_ring->page_offset,
807 				 buff->len, false);
808 		if (!buff->is_eop) {
809 			if (aq_add_rx_fragment(dev, rx_ring, buff, &xdp)) {
810 				u64_stats_update_begin(&rx_ring->stats.rx.syncp);
811 				++rx_ring->stats.rx.packets;
812 				rx_ring->stats.rx.bytes += xdp_get_buff_len(&xdp);
813 				++rx_ring->stats.rx.xdp_aborted;
814 				u64_stats_update_end(&rx_ring->stats.rx.syncp);
815 				continue;
816 			}
817 		}
818 
819 		skb = aq_xdp_run_prog(aq_nic, &xdp, rx_ring, buff);
820 		if (IS_ERR(skb) || !skb)
821 			continue;
822 
823 		if (buff->is_vlan)
824 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
825 					       buff->vlan_rx_tag);
826 
827 		aq_rx_checksum(rx_ring, buff, skb);
828 
829 		skb_set_hash(skb, buff->rss_hash,
830 			     buff->is_hash_l4 ? PKT_HASH_TYPE_L4 :
831 			     PKT_HASH_TYPE_NONE);
832 		/* Send all PTP traffic to 0 queue */
833 		skb_record_rx_queue(skb,
834 				    is_ptp_ring ? 0
835 						: AQ_NIC_RING2QMAP(rx_ring->aq_nic,
836 								   rx_ring->idx));
837 
838 		napi_gro_receive(napi, skb);
839 	}
840 
841 err_exit:
842 	return err;
843 }
844 
aq_ring_rx_clean(struct aq_ring_s * self,struct napi_struct * napi,int * work_done,int budget)845 int aq_ring_rx_clean(struct aq_ring_s *self,
846 		     struct napi_struct *napi,
847 		     int *work_done,
848 		     int budget)
849 {
850 	if (static_branch_unlikely(&aq_xdp_locking_key))
851 		return __aq_ring_xdp_clean(self, napi, work_done, budget);
852 	else
853 		return __aq_ring_rx_clean(self, napi, work_done, budget);
854 }
855 
aq_ring_hwts_rx_clean(struct aq_ring_s * self,struct aq_nic_s * aq_nic)856 void aq_ring_hwts_rx_clean(struct aq_ring_s *self, struct aq_nic_s *aq_nic)
857 {
858 #if IS_REACHABLE(CONFIG_PTP_1588_CLOCK)
859 	while (self->sw_head != self->hw_head) {
860 		u64 ns;
861 
862 		aq_nic->aq_hw_ops->extract_hwts(aq_nic->aq_hw,
863 						self->dx_ring +
864 						(self->sw_head * self->dx_size),
865 						self->dx_size, &ns);
866 		aq_ptp_tx_hwtstamp(aq_nic, ns);
867 
868 		self->sw_head = aq_ring_next_dx(self, self->sw_head);
869 	}
870 #endif
871 }
872 
aq_ring_rx_fill(struct aq_ring_s * self)873 int aq_ring_rx_fill(struct aq_ring_s *self)
874 {
875 	struct aq_ring_buff_s *buff = NULL;
876 	int err = 0;
877 	int i = 0;
878 
879 	if (aq_ring_avail_dx(self) < min_t(unsigned int, AQ_CFG_RX_REFILL_THRES,
880 					   self->size / 2))
881 		return err;
882 
883 	for (i = aq_ring_avail_dx(self); i--;
884 		self->sw_tail = aq_ring_next_dx(self, self->sw_tail)) {
885 		buff = &self->buff_ring[self->sw_tail];
886 
887 		buff->flags = 0U;
888 		buff->len = self->frame_max;
889 
890 		err = aq_get_rxpages(self, buff);
891 		if (err)
892 			goto err_exit;
893 
894 		buff->pa = aq_buf_daddr(&buff->rxdata);
895 		buff = NULL;
896 	}
897 
898 err_exit:
899 	return err;
900 }
901 
aq_ring_rx_deinit(struct aq_ring_s * self)902 void aq_ring_rx_deinit(struct aq_ring_s *self)
903 {
904 	if (!self)
905 		return;
906 
907 	for (; self->sw_head != self->sw_tail;
908 		self->sw_head = aq_ring_next_dx(self, self->sw_head)) {
909 		struct aq_ring_buff_s *buff = &self->buff_ring[self->sw_head];
910 
911 		aq_free_rxpage(&buff->rxdata, aq_nic_get_dev(self->aq_nic));
912 	}
913 }
914 
aq_ring_free(struct aq_ring_s * self)915 void aq_ring_free(struct aq_ring_s *self)
916 {
917 	if (!self)
918 		return;
919 
920 	kfree(self->buff_ring);
921 
922 	if (self->dx_ring)
923 		dma_free_coherent(aq_nic_get_dev(self->aq_nic),
924 				  self->size * self->dx_size, self->dx_ring,
925 				  self->dx_ring_pa);
926 }
927 
aq_ring_fill_stats_data(struct aq_ring_s * self,u64 * data)928 unsigned int aq_ring_fill_stats_data(struct aq_ring_s *self, u64 *data)
929 {
930 	unsigned int count;
931 	unsigned int start;
932 
933 	if (self->ring_type == ATL_RING_RX) {
934 		/* This data should mimic aq_ethtool_queue_rx_stat_names structure */
935 		do {
936 			count = 0;
937 			start = u64_stats_fetch_begin_irq(&self->stats.rx.syncp);
938 			data[count] = self->stats.rx.packets;
939 			data[++count] = self->stats.rx.jumbo_packets;
940 			data[++count] = self->stats.rx.lro_packets;
941 			data[++count] = self->stats.rx.errors;
942 			data[++count] = self->stats.rx.alloc_fails;
943 			data[++count] = self->stats.rx.skb_alloc_fails;
944 			data[++count] = self->stats.rx.polls;
945 			data[++count] = self->stats.rx.pg_flips;
946 			data[++count] = self->stats.rx.pg_reuses;
947 			data[++count] = self->stats.rx.pg_losts;
948 			data[++count] = self->stats.rx.xdp_aborted;
949 			data[++count] = self->stats.rx.xdp_drop;
950 			data[++count] = self->stats.rx.xdp_pass;
951 			data[++count] = self->stats.rx.xdp_tx;
952 			data[++count] = self->stats.rx.xdp_invalid;
953 			data[++count] = self->stats.rx.xdp_redirect;
954 		} while (u64_stats_fetch_retry_irq(&self->stats.rx.syncp, start));
955 	} else {
956 		/* This data should mimic aq_ethtool_queue_tx_stat_names structure */
957 		do {
958 			count = 0;
959 			start = u64_stats_fetch_begin_irq(&self->stats.tx.syncp);
960 			data[count] = self->stats.tx.packets;
961 			data[++count] = self->stats.tx.queue_restarts;
962 		} while (u64_stats_fetch_retry_irq(&self->stats.tx.syncp, start));
963 	}
964 
965 	return ++count;
966 }
967