xref: /linux/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c (revision 0ad53fe3ae82443c74ff8cfd7bd13377cc1134a3)
1 // SPDX-License-Identifier: (GPL-2.0+ OR BSD-3-Clause)
2 /* Copyright 2014-2016 Freescale Semiconductor Inc.
3  * Copyright 2016-2020 NXP
4  */
5 #include <linux/init.h>
6 #include <linux/module.h>
7 #include <linux/platform_device.h>
8 #include <linux/etherdevice.h>
9 #include <linux/of_net.h>
10 #include <linux/interrupt.h>
11 #include <linux/msi.h>
12 #include <linux/kthread.h>
13 #include <linux/iommu.h>
14 #include <linux/fsl/mc.h>
15 #include <linux/bpf.h>
16 #include <linux/bpf_trace.h>
17 #include <linux/fsl/ptp_qoriq.h>
18 #include <linux/ptp_classify.h>
19 #include <net/pkt_cls.h>
20 #include <net/sock.h>
21 
22 #include "dpaa2-eth.h"
23 
24 /* CREATE_TRACE_POINTS only needs to be defined once. Other dpa files
25  * using trace events only need to #include <trace/events/sched.h>
26  */
27 #define CREATE_TRACE_POINTS
28 #include "dpaa2-eth-trace.h"
29 
30 MODULE_LICENSE("Dual BSD/GPL");
31 MODULE_AUTHOR("Freescale Semiconductor, Inc");
32 MODULE_DESCRIPTION("Freescale DPAA2 Ethernet Driver");
33 
34 struct ptp_qoriq *dpaa2_ptp;
35 EXPORT_SYMBOL(dpaa2_ptp);
36 
37 static void *dpaa2_iova_to_virt(struct iommu_domain *domain,
38 				dma_addr_t iova_addr)
39 {
40 	phys_addr_t phys_addr;
41 
42 	phys_addr = domain ? iommu_iova_to_phys(domain, iova_addr) : iova_addr;
43 
44 	return phys_to_virt(phys_addr);
45 }
46 
47 static void dpaa2_eth_validate_rx_csum(struct dpaa2_eth_priv *priv,
48 				       u32 fd_status,
49 				       struct sk_buff *skb)
50 {
51 	skb_checksum_none_assert(skb);
52 
53 	/* HW checksum validation is disabled, nothing to do here */
54 	if (!(priv->net_dev->features & NETIF_F_RXCSUM))
55 		return;
56 
57 	/* Read checksum validation bits */
58 	if (!((fd_status & DPAA2_FAS_L3CV) &&
59 	      (fd_status & DPAA2_FAS_L4CV)))
60 		return;
61 
62 	/* Inform the stack there's no need to compute L3/L4 csum anymore */
63 	skb->ip_summed = CHECKSUM_UNNECESSARY;
64 }
65 
66 /* Free a received FD.
67  * Not to be used for Tx conf FDs or on any other paths.
68  */
69 static void dpaa2_eth_free_rx_fd(struct dpaa2_eth_priv *priv,
70 				 const struct dpaa2_fd *fd,
71 				 void *vaddr)
72 {
73 	struct device *dev = priv->net_dev->dev.parent;
74 	dma_addr_t addr = dpaa2_fd_get_addr(fd);
75 	u8 fd_format = dpaa2_fd_get_format(fd);
76 	struct dpaa2_sg_entry *sgt;
77 	void *sg_vaddr;
78 	int i;
79 
80 	/* If single buffer frame, just free the data buffer */
81 	if (fd_format == dpaa2_fd_single)
82 		goto free_buf;
83 	else if (fd_format != dpaa2_fd_sg)
84 		/* We don't support any other format */
85 		return;
86 
87 	/* For S/G frames, we first need to free all SG entries
88 	 * except the first one, which was taken care of already
89 	 */
90 	sgt = vaddr + dpaa2_fd_get_offset(fd);
91 	for (i = 1; i < DPAA2_ETH_MAX_SG_ENTRIES; i++) {
92 		addr = dpaa2_sg_get_addr(&sgt[i]);
93 		sg_vaddr = dpaa2_iova_to_virt(priv->iommu_domain, addr);
94 		dma_unmap_page(dev, addr, priv->rx_buf_size,
95 			       DMA_BIDIRECTIONAL);
96 
97 		free_pages((unsigned long)sg_vaddr, 0);
98 		if (dpaa2_sg_is_final(&sgt[i]))
99 			break;
100 	}
101 
102 free_buf:
103 	free_pages((unsigned long)vaddr, 0);
104 }
105 
106 /* Build a linear skb based on a single-buffer frame descriptor */
107 static struct sk_buff *dpaa2_eth_build_linear_skb(struct dpaa2_eth_channel *ch,
108 						  const struct dpaa2_fd *fd,
109 						  void *fd_vaddr)
110 {
111 	struct sk_buff *skb = NULL;
112 	u16 fd_offset = dpaa2_fd_get_offset(fd);
113 	u32 fd_length = dpaa2_fd_get_len(fd);
114 
115 	ch->buf_count--;
116 
117 	skb = build_skb(fd_vaddr, DPAA2_ETH_RX_BUF_RAW_SIZE);
118 	if (unlikely(!skb))
119 		return NULL;
120 
121 	skb_reserve(skb, fd_offset);
122 	skb_put(skb, fd_length);
123 
124 	return skb;
125 }
126 
127 /* Build a non linear (fragmented) skb based on a S/G table */
128 static struct sk_buff *dpaa2_eth_build_frag_skb(struct dpaa2_eth_priv *priv,
129 						struct dpaa2_eth_channel *ch,
130 						struct dpaa2_sg_entry *sgt)
131 {
132 	struct sk_buff *skb = NULL;
133 	struct device *dev = priv->net_dev->dev.parent;
134 	void *sg_vaddr;
135 	dma_addr_t sg_addr;
136 	u16 sg_offset;
137 	u32 sg_length;
138 	struct page *page, *head_page;
139 	int page_offset;
140 	int i;
141 
142 	for (i = 0; i < DPAA2_ETH_MAX_SG_ENTRIES; i++) {
143 		struct dpaa2_sg_entry *sge = &sgt[i];
144 
145 		/* NOTE: We only support SG entries in dpaa2_sg_single format,
146 		 * but this is the only format we may receive from HW anyway
147 		 */
148 
149 		/* Get the address and length from the S/G entry */
150 		sg_addr = dpaa2_sg_get_addr(sge);
151 		sg_vaddr = dpaa2_iova_to_virt(priv->iommu_domain, sg_addr);
152 		dma_unmap_page(dev, sg_addr, priv->rx_buf_size,
153 			       DMA_BIDIRECTIONAL);
154 
155 		sg_length = dpaa2_sg_get_len(sge);
156 
157 		if (i == 0) {
158 			/* We build the skb around the first data buffer */
159 			skb = build_skb(sg_vaddr, DPAA2_ETH_RX_BUF_RAW_SIZE);
160 			if (unlikely(!skb)) {
161 				/* Free the first SG entry now, since we already
162 				 * unmapped it and obtained the virtual address
163 				 */
164 				free_pages((unsigned long)sg_vaddr, 0);
165 
166 				/* We still need to subtract the buffers used
167 				 * by this FD from our software counter
168 				 */
169 				while (!dpaa2_sg_is_final(&sgt[i]) &&
170 				       i < DPAA2_ETH_MAX_SG_ENTRIES)
171 					i++;
172 				break;
173 			}
174 
175 			sg_offset = dpaa2_sg_get_offset(sge);
176 			skb_reserve(skb, sg_offset);
177 			skb_put(skb, sg_length);
178 		} else {
179 			/* Rest of the data buffers are stored as skb frags */
180 			page = virt_to_page(sg_vaddr);
181 			head_page = virt_to_head_page(sg_vaddr);
182 
183 			/* Offset in page (which may be compound).
184 			 * Data in subsequent SG entries is stored from the
185 			 * beginning of the buffer, so we don't need to add the
186 			 * sg_offset.
187 			 */
188 			page_offset = ((unsigned long)sg_vaddr &
189 				(PAGE_SIZE - 1)) +
190 				(page_address(page) - page_address(head_page));
191 
192 			skb_add_rx_frag(skb, i - 1, head_page, page_offset,
193 					sg_length, priv->rx_buf_size);
194 		}
195 
196 		if (dpaa2_sg_is_final(sge))
197 			break;
198 	}
199 
200 	WARN_ONCE(i == DPAA2_ETH_MAX_SG_ENTRIES, "Final bit not set in SGT");
201 
202 	/* Count all data buffers + SG table buffer */
203 	ch->buf_count -= i + 2;
204 
205 	return skb;
206 }
207 
208 /* Free buffers acquired from the buffer pool or which were meant to
209  * be released in the pool
210  */
211 static void dpaa2_eth_free_bufs(struct dpaa2_eth_priv *priv, u64 *buf_array,
212 				int count)
213 {
214 	struct device *dev = priv->net_dev->dev.parent;
215 	void *vaddr;
216 	int i;
217 
218 	for (i = 0; i < count; i++) {
219 		vaddr = dpaa2_iova_to_virt(priv->iommu_domain, buf_array[i]);
220 		dma_unmap_page(dev, buf_array[i], priv->rx_buf_size,
221 			       DMA_BIDIRECTIONAL);
222 		free_pages((unsigned long)vaddr, 0);
223 	}
224 }
225 
226 static void dpaa2_eth_recycle_buf(struct dpaa2_eth_priv *priv,
227 				  struct dpaa2_eth_channel *ch,
228 				  dma_addr_t addr)
229 {
230 	int retries = 0;
231 	int err;
232 
233 	ch->recycled_bufs[ch->recycled_bufs_cnt++] = addr;
234 	if (ch->recycled_bufs_cnt < DPAA2_ETH_BUFS_PER_CMD)
235 		return;
236 
237 	while ((err = dpaa2_io_service_release(ch->dpio, priv->bpid,
238 					       ch->recycled_bufs,
239 					       ch->recycled_bufs_cnt)) == -EBUSY) {
240 		if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES)
241 			break;
242 		cpu_relax();
243 	}
244 
245 	if (err) {
246 		dpaa2_eth_free_bufs(priv, ch->recycled_bufs, ch->recycled_bufs_cnt);
247 		ch->buf_count -= ch->recycled_bufs_cnt;
248 	}
249 
250 	ch->recycled_bufs_cnt = 0;
251 }
252 
253 static int dpaa2_eth_xdp_flush(struct dpaa2_eth_priv *priv,
254 			       struct dpaa2_eth_fq *fq,
255 			       struct dpaa2_eth_xdp_fds *xdp_fds)
256 {
257 	int total_enqueued = 0, retries = 0, enqueued;
258 	struct dpaa2_eth_drv_stats *percpu_extras;
259 	int num_fds, err, max_retries;
260 	struct dpaa2_fd *fds;
261 
262 	percpu_extras = this_cpu_ptr(priv->percpu_extras);
263 
264 	/* try to enqueue all the FDs until the max number of retries is hit */
265 	fds = xdp_fds->fds;
266 	num_fds = xdp_fds->num;
267 	max_retries = num_fds * DPAA2_ETH_ENQUEUE_RETRIES;
268 	while (total_enqueued < num_fds && retries < max_retries) {
269 		err = priv->enqueue(priv, fq, &fds[total_enqueued],
270 				    0, num_fds - total_enqueued, &enqueued);
271 		if (err == -EBUSY) {
272 			percpu_extras->tx_portal_busy += ++retries;
273 			continue;
274 		}
275 		total_enqueued += enqueued;
276 	}
277 	xdp_fds->num = 0;
278 
279 	return total_enqueued;
280 }
281 
282 static void dpaa2_eth_xdp_tx_flush(struct dpaa2_eth_priv *priv,
283 				   struct dpaa2_eth_channel *ch,
284 				   struct dpaa2_eth_fq *fq)
285 {
286 	struct rtnl_link_stats64 *percpu_stats;
287 	struct dpaa2_fd *fds;
288 	int enqueued, i;
289 
290 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
291 
292 	// enqueue the array of XDP_TX frames
293 	enqueued = dpaa2_eth_xdp_flush(priv, fq, &fq->xdp_tx_fds);
294 
295 	/* update statistics */
296 	percpu_stats->tx_packets += enqueued;
297 	fds = fq->xdp_tx_fds.fds;
298 	for (i = 0; i < enqueued; i++) {
299 		percpu_stats->tx_bytes += dpaa2_fd_get_len(&fds[i]);
300 		ch->stats.xdp_tx++;
301 	}
302 	for (i = enqueued; i < fq->xdp_tx_fds.num; i++) {
303 		dpaa2_eth_recycle_buf(priv, ch, dpaa2_fd_get_addr(&fds[i]));
304 		percpu_stats->tx_errors++;
305 		ch->stats.xdp_tx_err++;
306 	}
307 	fq->xdp_tx_fds.num = 0;
308 }
309 
310 static void dpaa2_eth_xdp_enqueue(struct dpaa2_eth_priv *priv,
311 				  struct dpaa2_eth_channel *ch,
312 				  struct dpaa2_fd *fd,
313 				  void *buf_start, u16 queue_id)
314 {
315 	struct dpaa2_faead *faead;
316 	struct dpaa2_fd *dest_fd;
317 	struct dpaa2_eth_fq *fq;
318 	u32 ctrl, frc;
319 
320 	/* Mark the egress frame hardware annotation area as valid */
321 	frc = dpaa2_fd_get_frc(fd);
322 	dpaa2_fd_set_frc(fd, frc | DPAA2_FD_FRC_FAEADV);
323 	dpaa2_fd_set_ctrl(fd, DPAA2_FD_CTRL_ASAL);
324 
325 	/* Instruct hardware to release the FD buffer directly into
326 	 * the buffer pool once transmission is completed, instead of
327 	 * sending a Tx confirmation frame to us
328 	 */
329 	ctrl = DPAA2_FAEAD_A4V | DPAA2_FAEAD_A2V | DPAA2_FAEAD_EBDDV;
330 	faead = dpaa2_get_faead(buf_start, false);
331 	faead->ctrl = cpu_to_le32(ctrl);
332 	faead->conf_fqid = 0;
333 
334 	fq = &priv->fq[queue_id];
335 	dest_fd = &fq->xdp_tx_fds.fds[fq->xdp_tx_fds.num++];
336 	memcpy(dest_fd, fd, sizeof(*dest_fd));
337 
338 	if (fq->xdp_tx_fds.num < DEV_MAP_BULK_SIZE)
339 		return;
340 
341 	dpaa2_eth_xdp_tx_flush(priv, ch, fq);
342 }
343 
344 static u32 dpaa2_eth_run_xdp(struct dpaa2_eth_priv *priv,
345 			     struct dpaa2_eth_channel *ch,
346 			     struct dpaa2_eth_fq *rx_fq,
347 			     struct dpaa2_fd *fd, void *vaddr)
348 {
349 	dma_addr_t addr = dpaa2_fd_get_addr(fd);
350 	struct bpf_prog *xdp_prog;
351 	struct xdp_buff xdp;
352 	u32 xdp_act = XDP_PASS;
353 	int err, offset;
354 
355 	xdp_prog = READ_ONCE(ch->xdp.prog);
356 	if (!xdp_prog)
357 		goto out;
358 
359 	offset = dpaa2_fd_get_offset(fd) - XDP_PACKET_HEADROOM;
360 	xdp_init_buff(&xdp, DPAA2_ETH_RX_BUF_RAW_SIZE - offset, &ch->xdp_rxq);
361 	xdp_prepare_buff(&xdp, vaddr + offset, XDP_PACKET_HEADROOM,
362 			 dpaa2_fd_get_len(fd), false);
363 
364 	xdp_act = bpf_prog_run_xdp(xdp_prog, &xdp);
365 
366 	/* xdp.data pointer may have changed */
367 	dpaa2_fd_set_offset(fd, xdp.data - vaddr);
368 	dpaa2_fd_set_len(fd, xdp.data_end - xdp.data);
369 
370 	switch (xdp_act) {
371 	case XDP_PASS:
372 		break;
373 	case XDP_TX:
374 		dpaa2_eth_xdp_enqueue(priv, ch, fd, vaddr, rx_fq->flowid);
375 		break;
376 	default:
377 		bpf_warn_invalid_xdp_action(xdp_act);
378 		fallthrough;
379 	case XDP_ABORTED:
380 		trace_xdp_exception(priv->net_dev, xdp_prog, xdp_act);
381 		fallthrough;
382 	case XDP_DROP:
383 		dpaa2_eth_recycle_buf(priv, ch, addr);
384 		ch->stats.xdp_drop++;
385 		break;
386 	case XDP_REDIRECT:
387 		dma_unmap_page(priv->net_dev->dev.parent, addr,
388 			       priv->rx_buf_size, DMA_BIDIRECTIONAL);
389 		ch->buf_count--;
390 
391 		/* Allow redirect use of full headroom */
392 		xdp.data_hard_start = vaddr;
393 		xdp.frame_sz = DPAA2_ETH_RX_BUF_RAW_SIZE;
394 
395 		err = xdp_do_redirect(priv->net_dev, &xdp, xdp_prog);
396 		if (unlikely(err)) {
397 			addr = dma_map_page(priv->net_dev->dev.parent,
398 					    virt_to_page(vaddr), 0,
399 					    priv->rx_buf_size, DMA_BIDIRECTIONAL);
400 			if (unlikely(dma_mapping_error(priv->net_dev->dev.parent, addr))) {
401 				free_pages((unsigned long)vaddr, 0);
402 			} else {
403 				ch->buf_count++;
404 				dpaa2_eth_recycle_buf(priv, ch, addr);
405 			}
406 			ch->stats.xdp_drop++;
407 		} else {
408 			ch->stats.xdp_redirect++;
409 		}
410 		break;
411 	}
412 
413 	ch->xdp.res |= xdp_act;
414 out:
415 	return xdp_act;
416 }
417 
418 static struct sk_buff *dpaa2_eth_copybreak(struct dpaa2_eth_channel *ch,
419 					   const struct dpaa2_fd *fd,
420 					   void *fd_vaddr)
421 {
422 	u16 fd_offset = dpaa2_fd_get_offset(fd);
423 	struct dpaa2_eth_priv *priv = ch->priv;
424 	u32 fd_length = dpaa2_fd_get_len(fd);
425 	struct sk_buff *skb = NULL;
426 	unsigned int skb_len;
427 
428 	if (fd_length > priv->rx_copybreak)
429 		return NULL;
430 
431 	skb_len = fd_length + dpaa2_eth_needed_headroom(NULL);
432 
433 	skb = napi_alloc_skb(&ch->napi, skb_len);
434 	if (!skb)
435 		return NULL;
436 
437 	skb_reserve(skb, dpaa2_eth_needed_headroom(NULL));
438 	skb_put(skb, fd_length);
439 
440 	memcpy(skb->data, fd_vaddr + fd_offset, fd_length);
441 
442 	dpaa2_eth_recycle_buf(priv, ch, dpaa2_fd_get_addr(fd));
443 
444 	return skb;
445 }
446 
447 /* Main Rx frame processing routine */
448 static void dpaa2_eth_rx(struct dpaa2_eth_priv *priv,
449 			 struct dpaa2_eth_channel *ch,
450 			 const struct dpaa2_fd *fd,
451 			 struct dpaa2_eth_fq *fq)
452 {
453 	dma_addr_t addr = dpaa2_fd_get_addr(fd);
454 	u8 fd_format = dpaa2_fd_get_format(fd);
455 	void *vaddr;
456 	struct sk_buff *skb;
457 	struct rtnl_link_stats64 *percpu_stats;
458 	struct dpaa2_eth_drv_stats *percpu_extras;
459 	struct device *dev = priv->net_dev->dev.parent;
460 	struct dpaa2_fas *fas;
461 	void *buf_data;
462 	u32 status = 0;
463 	u32 xdp_act;
464 
465 	/* Tracing point */
466 	trace_dpaa2_rx_fd(priv->net_dev, fd);
467 
468 	vaddr = dpaa2_iova_to_virt(priv->iommu_domain, addr);
469 	dma_sync_single_for_cpu(dev, addr, priv->rx_buf_size,
470 				DMA_BIDIRECTIONAL);
471 
472 	fas = dpaa2_get_fas(vaddr, false);
473 	prefetch(fas);
474 	buf_data = vaddr + dpaa2_fd_get_offset(fd);
475 	prefetch(buf_data);
476 
477 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
478 	percpu_extras = this_cpu_ptr(priv->percpu_extras);
479 
480 	if (fd_format == dpaa2_fd_single) {
481 		xdp_act = dpaa2_eth_run_xdp(priv, ch, fq, (struct dpaa2_fd *)fd, vaddr);
482 		if (xdp_act != XDP_PASS) {
483 			percpu_stats->rx_packets++;
484 			percpu_stats->rx_bytes += dpaa2_fd_get_len(fd);
485 			return;
486 		}
487 
488 		skb = dpaa2_eth_copybreak(ch, fd, vaddr);
489 		if (!skb) {
490 			dma_unmap_page(dev, addr, priv->rx_buf_size,
491 				       DMA_BIDIRECTIONAL);
492 			skb = dpaa2_eth_build_linear_skb(ch, fd, vaddr);
493 		}
494 	} else if (fd_format == dpaa2_fd_sg) {
495 		WARN_ON(priv->xdp_prog);
496 
497 		dma_unmap_page(dev, addr, priv->rx_buf_size,
498 			       DMA_BIDIRECTIONAL);
499 		skb = dpaa2_eth_build_frag_skb(priv, ch, buf_data);
500 		free_pages((unsigned long)vaddr, 0);
501 		percpu_extras->rx_sg_frames++;
502 		percpu_extras->rx_sg_bytes += dpaa2_fd_get_len(fd);
503 	} else {
504 		/* We don't support any other format */
505 		goto err_frame_format;
506 	}
507 
508 	if (unlikely(!skb))
509 		goto err_build_skb;
510 
511 	prefetch(skb->data);
512 
513 	/* Get the timestamp value */
514 	if (priv->rx_tstamp) {
515 		struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
516 		__le64 *ts = dpaa2_get_ts(vaddr, false);
517 		u64 ns;
518 
519 		memset(shhwtstamps, 0, sizeof(*shhwtstamps));
520 
521 		ns = DPAA2_PTP_CLK_PERIOD_NS * le64_to_cpup(ts);
522 		shhwtstamps->hwtstamp = ns_to_ktime(ns);
523 	}
524 
525 	/* Check if we need to validate the L4 csum */
526 	if (likely(dpaa2_fd_get_frc(fd) & DPAA2_FD_FRC_FASV)) {
527 		status = le32_to_cpu(fas->status);
528 		dpaa2_eth_validate_rx_csum(priv, status, skb);
529 	}
530 
531 	skb->protocol = eth_type_trans(skb, priv->net_dev);
532 	skb_record_rx_queue(skb, fq->flowid);
533 
534 	percpu_stats->rx_packets++;
535 	percpu_stats->rx_bytes += dpaa2_fd_get_len(fd);
536 
537 	list_add_tail(&skb->list, ch->rx_list);
538 
539 	return;
540 
541 err_build_skb:
542 	dpaa2_eth_free_rx_fd(priv, fd, vaddr);
543 err_frame_format:
544 	percpu_stats->rx_dropped++;
545 }
546 
547 /* Processing of Rx frames received on the error FQ
548  * We check and print the error bits and then free the frame
549  */
550 static void dpaa2_eth_rx_err(struct dpaa2_eth_priv *priv,
551 			     struct dpaa2_eth_channel *ch,
552 			     const struct dpaa2_fd *fd,
553 			     struct dpaa2_eth_fq *fq __always_unused)
554 {
555 	struct device *dev = priv->net_dev->dev.parent;
556 	dma_addr_t addr = dpaa2_fd_get_addr(fd);
557 	u8 fd_format = dpaa2_fd_get_format(fd);
558 	struct rtnl_link_stats64 *percpu_stats;
559 	struct dpaa2_eth_trap_item *trap_item;
560 	struct dpaa2_fapr *fapr;
561 	struct sk_buff *skb;
562 	void *buf_data;
563 	void *vaddr;
564 
565 	vaddr = dpaa2_iova_to_virt(priv->iommu_domain, addr);
566 	dma_sync_single_for_cpu(dev, addr, priv->rx_buf_size,
567 				DMA_BIDIRECTIONAL);
568 
569 	buf_data = vaddr + dpaa2_fd_get_offset(fd);
570 
571 	if (fd_format == dpaa2_fd_single) {
572 		dma_unmap_page(dev, addr, priv->rx_buf_size,
573 			       DMA_BIDIRECTIONAL);
574 		skb = dpaa2_eth_build_linear_skb(ch, fd, vaddr);
575 	} else if (fd_format == dpaa2_fd_sg) {
576 		dma_unmap_page(dev, addr, priv->rx_buf_size,
577 			       DMA_BIDIRECTIONAL);
578 		skb = dpaa2_eth_build_frag_skb(priv, ch, buf_data);
579 		free_pages((unsigned long)vaddr, 0);
580 	} else {
581 		/* We don't support any other format */
582 		dpaa2_eth_free_rx_fd(priv, fd, vaddr);
583 		goto err_frame_format;
584 	}
585 
586 	fapr = dpaa2_get_fapr(vaddr, false);
587 	trap_item = dpaa2_eth_dl_get_trap(priv, fapr);
588 	if (trap_item)
589 		devlink_trap_report(priv->devlink, skb, trap_item->trap_ctx,
590 				    &priv->devlink_port, NULL);
591 	consume_skb(skb);
592 
593 err_frame_format:
594 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
595 	percpu_stats->rx_errors++;
596 	ch->buf_count--;
597 }
598 
599 /* Consume all frames pull-dequeued into the store. This is the simplest way to
600  * make sure we don't accidentally issue another volatile dequeue which would
601  * overwrite (leak) frames already in the store.
602  *
603  * Observance of NAPI budget is not our concern, leaving that to the caller.
604  */
605 static int dpaa2_eth_consume_frames(struct dpaa2_eth_channel *ch,
606 				    struct dpaa2_eth_fq **src)
607 {
608 	struct dpaa2_eth_priv *priv = ch->priv;
609 	struct dpaa2_eth_fq *fq = NULL;
610 	struct dpaa2_dq *dq;
611 	const struct dpaa2_fd *fd;
612 	int cleaned = 0, retries = 0;
613 	int is_last;
614 
615 	do {
616 		dq = dpaa2_io_store_next(ch->store, &is_last);
617 		if (unlikely(!dq)) {
618 			/* If we're here, we *must* have placed a
619 			 * volatile dequeue comnmand, so keep reading through
620 			 * the store until we get some sort of valid response
621 			 * token (either a valid frame or an "empty dequeue")
622 			 */
623 			if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES) {
624 				netdev_err_once(priv->net_dev,
625 						"Unable to read a valid dequeue response\n");
626 				return -ETIMEDOUT;
627 			}
628 			continue;
629 		}
630 
631 		fd = dpaa2_dq_fd(dq);
632 		fq = (struct dpaa2_eth_fq *)(uintptr_t)dpaa2_dq_fqd_ctx(dq);
633 
634 		fq->consume(priv, ch, fd, fq);
635 		cleaned++;
636 		retries = 0;
637 	} while (!is_last);
638 
639 	if (!cleaned)
640 		return 0;
641 
642 	fq->stats.frames += cleaned;
643 	ch->stats.frames += cleaned;
644 
645 	/* A dequeue operation only pulls frames from a single queue
646 	 * into the store. Return the frame queue as an out param.
647 	 */
648 	if (src)
649 		*src = fq;
650 
651 	return cleaned;
652 }
653 
654 static int dpaa2_eth_ptp_parse(struct sk_buff *skb,
655 			       u8 *msgtype, u8 *twostep, u8 *udp,
656 			       u16 *correction_offset,
657 			       u16 *origintimestamp_offset)
658 {
659 	unsigned int ptp_class;
660 	struct ptp_header *hdr;
661 	unsigned int type;
662 	u8 *base;
663 
664 	ptp_class = ptp_classify_raw(skb);
665 	if (ptp_class == PTP_CLASS_NONE)
666 		return -EINVAL;
667 
668 	hdr = ptp_parse_header(skb, ptp_class);
669 	if (!hdr)
670 		return -EINVAL;
671 
672 	*msgtype = ptp_get_msgtype(hdr, ptp_class);
673 	*twostep = hdr->flag_field[0] & 0x2;
674 
675 	type = ptp_class & PTP_CLASS_PMASK;
676 	if (type == PTP_CLASS_IPV4 ||
677 	    type == PTP_CLASS_IPV6)
678 		*udp = 1;
679 	else
680 		*udp = 0;
681 
682 	base = skb_mac_header(skb);
683 	*correction_offset = (u8 *)&hdr->correction - base;
684 	*origintimestamp_offset = (u8 *)hdr + sizeof(struct ptp_header) - base;
685 
686 	return 0;
687 }
688 
689 /* Configure the egress frame annotation for timestamp update */
690 static void dpaa2_eth_enable_tx_tstamp(struct dpaa2_eth_priv *priv,
691 				       struct dpaa2_fd *fd,
692 				       void *buf_start,
693 				       struct sk_buff *skb)
694 {
695 	struct ptp_tstamp origin_timestamp;
696 	struct dpni_single_step_cfg cfg;
697 	u8 msgtype, twostep, udp;
698 	struct dpaa2_faead *faead;
699 	struct dpaa2_fas *fas;
700 	struct timespec64 ts;
701 	u16 offset1, offset2;
702 	u32 ctrl, frc;
703 	__le64 *ns;
704 	u8 *data;
705 
706 	/* Mark the egress frame annotation area as valid */
707 	frc = dpaa2_fd_get_frc(fd);
708 	dpaa2_fd_set_frc(fd, frc | DPAA2_FD_FRC_FAEADV);
709 
710 	/* Set hardware annotation size */
711 	ctrl = dpaa2_fd_get_ctrl(fd);
712 	dpaa2_fd_set_ctrl(fd, ctrl | DPAA2_FD_CTRL_ASAL);
713 
714 	/* enable UPD (update prepanded data) bit in FAEAD field of
715 	 * hardware frame annotation area
716 	 */
717 	ctrl = DPAA2_FAEAD_A2V | DPAA2_FAEAD_UPDV | DPAA2_FAEAD_UPD;
718 	faead = dpaa2_get_faead(buf_start, true);
719 	faead->ctrl = cpu_to_le32(ctrl);
720 
721 	if (skb->cb[0] == TX_TSTAMP_ONESTEP_SYNC) {
722 		if (dpaa2_eth_ptp_parse(skb, &msgtype, &twostep, &udp,
723 					&offset1, &offset2) ||
724 		    msgtype != PTP_MSGTYPE_SYNC || twostep) {
725 			WARN_ONCE(1, "Bad packet for one-step timestamping\n");
726 			return;
727 		}
728 
729 		/* Mark the frame annotation status as valid */
730 		frc = dpaa2_fd_get_frc(fd);
731 		dpaa2_fd_set_frc(fd, frc | DPAA2_FD_FRC_FASV);
732 
733 		/* Mark the PTP flag for one step timestamping */
734 		fas = dpaa2_get_fas(buf_start, true);
735 		fas->status = cpu_to_le32(DPAA2_FAS_PTP);
736 
737 		dpaa2_ptp->caps.gettime64(&dpaa2_ptp->caps, &ts);
738 		ns = dpaa2_get_ts(buf_start, true);
739 		*ns = cpu_to_le64(timespec64_to_ns(&ts) /
740 				  DPAA2_PTP_CLK_PERIOD_NS);
741 
742 		/* Update current time to PTP message originTimestamp field */
743 		ns_to_ptp_tstamp(&origin_timestamp, le64_to_cpup(ns));
744 		data = skb_mac_header(skb);
745 		*(__be16 *)(data + offset2) = htons(origin_timestamp.sec_msb);
746 		*(__be32 *)(data + offset2 + 2) =
747 			htonl(origin_timestamp.sec_lsb);
748 		*(__be32 *)(data + offset2 + 6) = htonl(origin_timestamp.nsec);
749 
750 		cfg.en = 1;
751 		cfg.ch_update = udp;
752 		cfg.offset = offset1;
753 		cfg.peer_delay = 0;
754 
755 		if (dpni_set_single_step_cfg(priv->mc_io, 0, priv->mc_token,
756 					     &cfg))
757 			WARN_ONCE(1, "Failed to set single step register");
758 	}
759 }
760 
761 /* Create a frame descriptor based on a fragmented skb */
762 static int dpaa2_eth_build_sg_fd(struct dpaa2_eth_priv *priv,
763 				 struct sk_buff *skb,
764 				 struct dpaa2_fd *fd,
765 				 void **swa_addr)
766 {
767 	struct device *dev = priv->net_dev->dev.parent;
768 	void *sgt_buf = NULL;
769 	dma_addr_t addr;
770 	int nr_frags = skb_shinfo(skb)->nr_frags;
771 	struct dpaa2_sg_entry *sgt;
772 	int i, err;
773 	int sgt_buf_size;
774 	struct scatterlist *scl, *crt_scl;
775 	int num_sg;
776 	int num_dma_bufs;
777 	struct dpaa2_eth_swa *swa;
778 
779 	/* Create and map scatterlist.
780 	 * We don't advertise NETIF_F_FRAGLIST, so skb_to_sgvec() will not have
781 	 * to go beyond nr_frags+1.
782 	 * Note: We don't support chained scatterlists
783 	 */
784 	if (unlikely(PAGE_SIZE / sizeof(struct scatterlist) < nr_frags + 1))
785 		return -EINVAL;
786 
787 	scl = kmalloc_array(nr_frags + 1, sizeof(struct scatterlist), GFP_ATOMIC);
788 	if (unlikely(!scl))
789 		return -ENOMEM;
790 
791 	sg_init_table(scl, nr_frags + 1);
792 	num_sg = skb_to_sgvec(skb, scl, 0, skb->len);
793 	if (unlikely(num_sg < 0)) {
794 		err = -ENOMEM;
795 		goto dma_map_sg_failed;
796 	}
797 	num_dma_bufs = dma_map_sg(dev, scl, num_sg, DMA_BIDIRECTIONAL);
798 	if (unlikely(!num_dma_bufs)) {
799 		err = -ENOMEM;
800 		goto dma_map_sg_failed;
801 	}
802 
803 	/* Prepare the HW SGT structure */
804 	sgt_buf_size = priv->tx_data_offset +
805 		       sizeof(struct dpaa2_sg_entry) *  num_dma_bufs;
806 	sgt_buf = napi_alloc_frag_align(sgt_buf_size, DPAA2_ETH_TX_BUF_ALIGN);
807 	if (unlikely(!sgt_buf)) {
808 		err = -ENOMEM;
809 		goto sgt_buf_alloc_failed;
810 	}
811 	memset(sgt_buf, 0, sgt_buf_size);
812 
813 	sgt = (struct dpaa2_sg_entry *)(sgt_buf + priv->tx_data_offset);
814 
815 	/* Fill in the HW SGT structure.
816 	 *
817 	 * sgt_buf is zeroed out, so the following fields are implicit
818 	 * in all sgt entries:
819 	 *   - offset is 0
820 	 *   - format is 'dpaa2_sg_single'
821 	 */
822 	for_each_sg(scl, crt_scl, num_dma_bufs, i) {
823 		dpaa2_sg_set_addr(&sgt[i], sg_dma_address(crt_scl));
824 		dpaa2_sg_set_len(&sgt[i], sg_dma_len(crt_scl));
825 	}
826 	dpaa2_sg_set_final(&sgt[i - 1], true);
827 
828 	/* Store the skb backpointer in the SGT buffer.
829 	 * Fit the scatterlist and the number of buffers alongside the
830 	 * skb backpointer in the software annotation area. We'll need
831 	 * all of them on Tx Conf.
832 	 */
833 	*swa_addr = (void *)sgt_buf;
834 	swa = (struct dpaa2_eth_swa *)sgt_buf;
835 	swa->type = DPAA2_ETH_SWA_SG;
836 	swa->sg.skb = skb;
837 	swa->sg.scl = scl;
838 	swa->sg.num_sg = num_sg;
839 	swa->sg.sgt_size = sgt_buf_size;
840 
841 	/* Separately map the SGT buffer */
842 	addr = dma_map_single(dev, sgt_buf, sgt_buf_size, DMA_BIDIRECTIONAL);
843 	if (unlikely(dma_mapping_error(dev, addr))) {
844 		err = -ENOMEM;
845 		goto dma_map_single_failed;
846 	}
847 	dpaa2_fd_set_offset(fd, priv->tx_data_offset);
848 	dpaa2_fd_set_format(fd, dpaa2_fd_sg);
849 	dpaa2_fd_set_addr(fd, addr);
850 	dpaa2_fd_set_len(fd, skb->len);
851 	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
852 
853 	return 0;
854 
855 dma_map_single_failed:
856 	skb_free_frag(sgt_buf);
857 sgt_buf_alloc_failed:
858 	dma_unmap_sg(dev, scl, num_sg, DMA_BIDIRECTIONAL);
859 dma_map_sg_failed:
860 	kfree(scl);
861 	return err;
862 }
863 
864 /* Create a SG frame descriptor based on a linear skb.
865  *
866  * This function is used on the Tx path when the skb headroom is not large
867  * enough for the HW requirements, thus instead of realloc-ing the skb we
868  * create a SG frame descriptor with only one entry.
869  */
870 static int dpaa2_eth_build_sg_fd_single_buf(struct dpaa2_eth_priv *priv,
871 					    struct sk_buff *skb,
872 					    struct dpaa2_fd *fd,
873 					    void **swa_addr)
874 {
875 	struct device *dev = priv->net_dev->dev.parent;
876 	struct dpaa2_eth_sgt_cache *sgt_cache;
877 	struct dpaa2_sg_entry *sgt;
878 	struct dpaa2_eth_swa *swa;
879 	dma_addr_t addr, sgt_addr;
880 	void *sgt_buf = NULL;
881 	int sgt_buf_size;
882 	int err;
883 
884 	/* Prepare the HW SGT structure */
885 	sgt_cache = this_cpu_ptr(priv->sgt_cache);
886 	sgt_buf_size = priv->tx_data_offset + sizeof(struct dpaa2_sg_entry);
887 
888 	if (sgt_cache->count == 0)
889 		sgt_buf = kzalloc(sgt_buf_size + DPAA2_ETH_TX_BUF_ALIGN,
890 				  GFP_ATOMIC);
891 	else
892 		sgt_buf = sgt_cache->buf[--sgt_cache->count];
893 	if (unlikely(!sgt_buf))
894 		return -ENOMEM;
895 
896 	sgt_buf = PTR_ALIGN(sgt_buf, DPAA2_ETH_TX_BUF_ALIGN);
897 	sgt = (struct dpaa2_sg_entry *)(sgt_buf + priv->tx_data_offset);
898 
899 	addr = dma_map_single(dev, skb->data, skb->len, DMA_BIDIRECTIONAL);
900 	if (unlikely(dma_mapping_error(dev, addr))) {
901 		err = -ENOMEM;
902 		goto data_map_failed;
903 	}
904 
905 	/* Fill in the HW SGT structure */
906 	dpaa2_sg_set_addr(sgt, addr);
907 	dpaa2_sg_set_len(sgt, skb->len);
908 	dpaa2_sg_set_final(sgt, true);
909 
910 	/* Store the skb backpointer in the SGT buffer */
911 	*swa_addr = (void *)sgt_buf;
912 	swa = (struct dpaa2_eth_swa *)sgt_buf;
913 	swa->type = DPAA2_ETH_SWA_SINGLE;
914 	swa->single.skb = skb;
915 	swa->single.sgt_size = sgt_buf_size;
916 
917 	/* Separately map the SGT buffer */
918 	sgt_addr = dma_map_single(dev, sgt_buf, sgt_buf_size, DMA_BIDIRECTIONAL);
919 	if (unlikely(dma_mapping_error(dev, sgt_addr))) {
920 		err = -ENOMEM;
921 		goto sgt_map_failed;
922 	}
923 
924 	dpaa2_fd_set_offset(fd, priv->tx_data_offset);
925 	dpaa2_fd_set_format(fd, dpaa2_fd_sg);
926 	dpaa2_fd_set_addr(fd, sgt_addr);
927 	dpaa2_fd_set_len(fd, skb->len);
928 	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
929 
930 	return 0;
931 
932 sgt_map_failed:
933 	dma_unmap_single(dev, addr, skb->len, DMA_BIDIRECTIONAL);
934 data_map_failed:
935 	if (sgt_cache->count >= DPAA2_ETH_SGT_CACHE_SIZE)
936 		kfree(sgt_buf);
937 	else
938 		sgt_cache->buf[sgt_cache->count++] = sgt_buf;
939 
940 	return err;
941 }
942 
943 /* Create a frame descriptor based on a linear skb */
944 static int dpaa2_eth_build_single_fd(struct dpaa2_eth_priv *priv,
945 				     struct sk_buff *skb,
946 				     struct dpaa2_fd *fd,
947 				     void **swa_addr)
948 {
949 	struct device *dev = priv->net_dev->dev.parent;
950 	u8 *buffer_start, *aligned_start;
951 	struct dpaa2_eth_swa *swa;
952 	dma_addr_t addr;
953 
954 	buffer_start = skb->data - dpaa2_eth_needed_headroom(skb);
955 
956 	/* If there's enough room to align the FD address, do it.
957 	 * It will help hardware optimize accesses.
958 	 */
959 	aligned_start = PTR_ALIGN(buffer_start - DPAA2_ETH_TX_BUF_ALIGN,
960 				  DPAA2_ETH_TX_BUF_ALIGN);
961 	if (aligned_start >= skb->head)
962 		buffer_start = aligned_start;
963 
964 	/* Store a backpointer to the skb at the beginning of the buffer
965 	 * (in the private data area) such that we can release it
966 	 * on Tx confirm
967 	 */
968 	*swa_addr = (void *)buffer_start;
969 	swa = (struct dpaa2_eth_swa *)buffer_start;
970 	swa->type = DPAA2_ETH_SWA_SINGLE;
971 	swa->single.skb = skb;
972 
973 	addr = dma_map_single(dev, buffer_start,
974 			      skb_tail_pointer(skb) - buffer_start,
975 			      DMA_BIDIRECTIONAL);
976 	if (unlikely(dma_mapping_error(dev, addr)))
977 		return -ENOMEM;
978 
979 	dpaa2_fd_set_addr(fd, addr);
980 	dpaa2_fd_set_offset(fd, (u16)(skb->data - buffer_start));
981 	dpaa2_fd_set_len(fd, skb->len);
982 	dpaa2_fd_set_format(fd, dpaa2_fd_single);
983 	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
984 
985 	return 0;
986 }
987 
988 /* FD freeing routine on the Tx path
989  *
990  * DMA-unmap and free FD and possibly SGT buffer allocated on Tx. The skb
991  * back-pointed to is also freed.
992  * This can be called either from dpaa2_eth_tx_conf() or on the error path of
993  * dpaa2_eth_tx().
994  */
995 static void dpaa2_eth_free_tx_fd(struct dpaa2_eth_priv *priv,
996 				 struct dpaa2_eth_fq *fq,
997 				 const struct dpaa2_fd *fd, bool in_napi)
998 {
999 	struct device *dev = priv->net_dev->dev.parent;
1000 	dma_addr_t fd_addr, sg_addr;
1001 	struct sk_buff *skb = NULL;
1002 	unsigned char *buffer_start;
1003 	struct dpaa2_eth_swa *swa;
1004 	u8 fd_format = dpaa2_fd_get_format(fd);
1005 	u32 fd_len = dpaa2_fd_get_len(fd);
1006 
1007 	struct dpaa2_eth_sgt_cache *sgt_cache;
1008 	struct dpaa2_sg_entry *sgt;
1009 
1010 	fd_addr = dpaa2_fd_get_addr(fd);
1011 	buffer_start = dpaa2_iova_to_virt(priv->iommu_domain, fd_addr);
1012 	swa = (struct dpaa2_eth_swa *)buffer_start;
1013 
1014 	if (fd_format == dpaa2_fd_single) {
1015 		if (swa->type == DPAA2_ETH_SWA_SINGLE) {
1016 			skb = swa->single.skb;
1017 			/* Accessing the skb buffer is safe before dma unmap,
1018 			 * because we didn't map the actual skb shell.
1019 			 */
1020 			dma_unmap_single(dev, fd_addr,
1021 					 skb_tail_pointer(skb) - buffer_start,
1022 					 DMA_BIDIRECTIONAL);
1023 		} else {
1024 			WARN_ONCE(swa->type != DPAA2_ETH_SWA_XDP, "Wrong SWA type");
1025 			dma_unmap_single(dev, fd_addr, swa->xdp.dma_size,
1026 					 DMA_BIDIRECTIONAL);
1027 		}
1028 	} else if (fd_format == dpaa2_fd_sg) {
1029 		if (swa->type == DPAA2_ETH_SWA_SG) {
1030 			skb = swa->sg.skb;
1031 
1032 			/* Unmap the scatterlist */
1033 			dma_unmap_sg(dev, swa->sg.scl, swa->sg.num_sg,
1034 				     DMA_BIDIRECTIONAL);
1035 			kfree(swa->sg.scl);
1036 
1037 			/* Unmap the SGT buffer */
1038 			dma_unmap_single(dev, fd_addr, swa->sg.sgt_size,
1039 					 DMA_BIDIRECTIONAL);
1040 		} else {
1041 			skb = swa->single.skb;
1042 
1043 			/* Unmap the SGT Buffer */
1044 			dma_unmap_single(dev, fd_addr, swa->single.sgt_size,
1045 					 DMA_BIDIRECTIONAL);
1046 
1047 			sgt = (struct dpaa2_sg_entry *)(buffer_start +
1048 							priv->tx_data_offset);
1049 			sg_addr = dpaa2_sg_get_addr(sgt);
1050 			dma_unmap_single(dev, sg_addr, skb->len, DMA_BIDIRECTIONAL);
1051 		}
1052 	} else {
1053 		netdev_dbg(priv->net_dev, "Invalid FD format\n");
1054 		return;
1055 	}
1056 
1057 	if (swa->type != DPAA2_ETH_SWA_XDP && in_napi) {
1058 		fq->dq_frames++;
1059 		fq->dq_bytes += fd_len;
1060 	}
1061 
1062 	if (swa->type == DPAA2_ETH_SWA_XDP) {
1063 		xdp_return_frame(swa->xdp.xdpf);
1064 		return;
1065 	}
1066 
1067 	/* Get the timestamp value */
1068 	if (skb->cb[0] == TX_TSTAMP) {
1069 		struct skb_shared_hwtstamps shhwtstamps;
1070 		__le64 *ts = dpaa2_get_ts(buffer_start, true);
1071 		u64 ns;
1072 
1073 		memset(&shhwtstamps, 0, sizeof(shhwtstamps));
1074 
1075 		ns = DPAA2_PTP_CLK_PERIOD_NS * le64_to_cpup(ts);
1076 		shhwtstamps.hwtstamp = ns_to_ktime(ns);
1077 		skb_tstamp_tx(skb, &shhwtstamps);
1078 	} else if (skb->cb[0] == TX_TSTAMP_ONESTEP_SYNC) {
1079 		mutex_unlock(&priv->onestep_tstamp_lock);
1080 	}
1081 
1082 	/* Free SGT buffer allocated on tx */
1083 	if (fd_format != dpaa2_fd_single) {
1084 		sgt_cache = this_cpu_ptr(priv->sgt_cache);
1085 		if (swa->type == DPAA2_ETH_SWA_SG) {
1086 			skb_free_frag(buffer_start);
1087 		} else {
1088 			if (sgt_cache->count >= DPAA2_ETH_SGT_CACHE_SIZE)
1089 				kfree(buffer_start);
1090 			else
1091 				sgt_cache->buf[sgt_cache->count++] = buffer_start;
1092 		}
1093 	}
1094 
1095 	/* Move on with skb release */
1096 	napi_consume_skb(skb, in_napi);
1097 }
1098 
1099 static netdev_tx_t __dpaa2_eth_tx(struct sk_buff *skb,
1100 				  struct net_device *net_dev)
1101 {
1102 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1103 	struct dpaa2_fd fd;
1104 	struct rtnl_link_stats64 *percpu_stats;
1105 	struct dpaa2_eth_drv_stats *percpu_extras;
1106 	struct dpaa2_eth_fq *fq;
1107 	struct netdev_queue *nq;
1108 	u16 queue_mapping;
1109 	unsigned int needed_headroom;
1110 	u32 fd_len;
1111 	u8 prio = 0;
1112 	int err, i;
1113 	void *swa;
1114 
1115 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
1116 	percpu_extras = this_cpu_ptr(priv->percpu_extras);
1117 
1118 	needed_headroom = dpaa2_eth_needed_headroom(skb);
1119 
1120 	/* We'll be holding a back-reference to the skb until Tx Confirmation;
1121 	 * we don't want that overwritten by a concurrent Tx with a cloned skb.
1122 	 */
1123 	skb = skb_unshare(skb, GFP_ATOMIC);
1124 	if (unlikely(!skb)) {
1125 		/* skb_unshare() has already freed the skb */
1126 		percpu_stats->tx_dropped++;
1127 		return NETDEV_TX_OK;
1128 	}
1129 
1130 	/* Setup the FD fields */
1131 	memset(&fd, 0, sizeof(fd));
1132 
1133 	if (skb_is_nonlinear(skb)) {
1134 		err = dpaa2_eth_build_sg_fd(priv, skb, &fd, &swa);
1135 		percpu_extras->tx_sg_frames++;
1136 		percpu_extras->tx_sg_bytes += skb->len;
1137 	} else if (skb_headroom(skb) < needed_headroom) {
1138 		err = dpaa2_eth_build_sg_fd_single_buf(priv, skb, &fd, &swa);
1139 		percpu_extras->tx_sg_frames++;
1140 		percpu_extras->tx_sg_bytes += skb->len;
1141 		percpu_extras->tx_converted_sg_frames++;
1142 		percpu_extras->tx_converted_sg_bytes += skb->len;
1143 	} else {
1144 		err = dpaa2_eth_build_single_fd(priv, skb, &fd, &swa);
1145 	}
1146 
1147 	if (unlikely(err)) {
1148 		percpu_stats->tx_dropped++;
1149 		goto err_build_fd;
1150 	}
1151 
1152 	if (skb->cb[0])
1153 		dpaa2_eth_enable_tx_tstamp(priv, &fd, swa, skb);
1154 
1155 	/* Tracing point */
1156 	trace_dpaa2_tx_fd(net_dev, &fd);
1157 
1158 	/* TxConf FQ selection relies on queue id from the stack.
1159 	 * In case of a forwarded frame from another DPNI interface, we choose
1160 	 * a queue affined to the same core that processed the Rx frame
1161 	 */
1162 	queue_mapping = skb_get_queue_mapping(skb);
1163 
1164 	if (net_dev->num_tc) {
1165 		prio = netdev_txq_to_tc(net_dev, queue_mapping);
1166 		/* Hardware interprets priority level 0 as being the highest,
1167 		 * so we need to do a reverse mapping to the netdev tc index
1168 		 */
1169 		prio = net_dev->num_tc - prio - 1;
1170 		/* We have only one FQ array entry for all Tx hardware queues
1171 		 * with the same flow id (but different priority levels)
1172 		 */
1173 		queue_mapping %= dpaa2_eth_queue_count(priv);
1174 	}
1175 	fq = &priv->fq[queue_mapping];
1176 
1177 	fd_len = dpaa2_fd_get_len(&fd);
1178 	nq = netdev_get_tx_queue(net_dev, queue_mapping);
1179 	netdev_tx_sent_queue(nq, fd_len);
1180 
1181 	/* Everything that happens after this enqueues might race with
1182 	 * the Tx confirmation callback for this frame
1183 	 */
1184 	for (i = 0; i < DPAA2_ETH_ENQUEUE_RETRIES; i++) {
1185 		err = priv->enqueue(priv, fq, &fd, prio, 1, NULL);
1186 		if (err != -EBUSY)
1187 			break;
1188 	}
1189 	percpu_extras->tx_portal_busy += i;
1190 	if (unlikely(err < 0)) {
1191 		percpu_stats->tx_errors++;
1192 		/* Clean up everything, including freeing the skb */
1193 		dpaa2_eth_free_tx_fd(priv, fq, &fd, false);
1194 		netdev_tx_completed_queue(nq, 1, fd_len);
1195 	} else {
1196 		percpu_stats->tx_packets++;
1197 		percpu_stats->tx_bytes += fd_len;
1198 	}
1199 
1200 	return NETDEV_TX_OK;
1201 
1202 err_build_fd:
1203 	dev_kfree_skb(skb);
1204 
1205 	return NETDEV_TX_OK;
1206 }
1207 
1208 static void dpaa2_eth_tx_onestep_tstamp(struct work_struct *work)
1209 {
1210 	struct dpaa2_eth_priv *priv = container_of(work, struct dpaa2_eth_priv,
1211 						   tx_onestep_tstamp);
1212 	struct sk_buff *skb;
1213 
1214 	while (true) {
1215 		skb = skb_dequeue(&priv->tx_skbs);
1216 		if (!skb)
1217 			return;
1218 
1219 		/* Lock just before TX one-step timestamping packet,
1220 		 * and release the lock in dpaa2_eth_free_tx_fd when
1221 		 * confirm the packet has been sent on hardware, or
1222 		 * when clean up during transmit failure.
1223 		 */
1224 		mutex_lock(&priv->onestep_tstamp_lock);
1225 		__dpaa2_eth_tx(skb, priv->net_dev);
1226 	}
1227 }
1228 
1229 static netdev_tx_t dpaa2_eth_tx(struct sk_buff *skb, struct net_device *net_dev)
1230 {
1231 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1232 	u8 msgtype, twostep, udp;
1233 	u16 offset1, offset2;
1234 
1235 	/* Utilize skb->cb[0] for timestamping request per skb */
1236 	skb->cb[0] = 0;
1237 
1238 	if ((skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) && dpaa2_ptp) {
1239 		if (priv->tx_tstamp_type == HWTSTAMP_TX_ON)
1240 			skb->cb[0] = TX_TSTAMP;
1241 		else if (priv->tx_tstamp_type == HWTSTAMP_TX_ONESTEP_SYNC)
1242 			skb->cb[0] = TX_TSTAMP_ONESTEP_SYNC;
1243 	}
1244 
1245 	/* TX for one-step timestamping PTP Sync packet */
1246 	if (skb->cb[0] == TX_TSTAMP_ONESTEP_SYNC) {
1247 		if (!dpaa2_eth_ptp_parse(skb, &msgtype, &twostep, &udp,
1248 					 &offset1, &offset2))
1249 			if (msgtype == PTP_MSGTYPE_SYNC && twostep == 0) {
1250 				skb_queue_tail(&priv->tx_skbs, skb);
1251 				queue_work(priv->dpaa2_ptp_wq,
1252 					   &priv->tx_onestep_tstamp);
1253 				return NETDEV_TX_OK;
1254 			}
1255 		/* Use two-step timestamping if not one-step timestamping
1256 		 * PTP Sync packet
1257 		 */
1258 		skb->cb[0] = TX_TSTAMP;
1259 	}
1260 
1261 	/* TX for other packets */
1262 	return __dpaa2_eth_tx(skb, net_dev);
1263 }
1264 
1265 /* Tx confirmation frame processing routine */
1266 static void dpaa2_eth_tx_conf(struct dpaa2_eth_priv *priv,
1267 			      struct dpaa2_eth_channel *ch __always_unused,
1268 			      const struct dpaa2_fd *fd,
1269 			      struct dpaa2_eth_fq *fq)
1270 {
1271 	struct rtnl_link_stats64 *percpu_stats;
1272 	struct dpaa2_eth_drv_stats *percpu_extras;
1273 	u32 fd_len = dpaa2_fd_get_len(fd);
1274 	u32 fd_errors;
1275 
1276 	/* Tracing point */
1277 	trace_dpaa2_tx_conf_fd(priv->net_dev, fd);
1278 
1279 	percpu_extras = this_cpu_ptr(priv->percpu_extras);
1280 	percpu_extras->tx_conf_frames++;
1281 	percpu_extras->tx_conf_bytes += fd_len;
1282 
1283 	/* Check frame errors in the FD field */
1284 	fd_errors = dpaa2_fd_get_ctrl(fd) & DPAA2_FD_TX_ERR_MASK;
1285 	dpaa2_eth_free_tx_fd(priv, fq, fd, true);
1286 
1287 	if (likely(!fd_errors))
1288 		return;
1289 
1290 	if (net_ratelimit())
1291 		netdev_dbg(priv->net_dev, "TX frame FD error: 0x%08x\n",
1292 			   fd_errors);
1293 
1294 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
1295 	/* Tx-conf logically pertains to the egress path. */
1296 	percpu_stats->tx_errors++;
1297 }
1298 
1299 static int dpaa2_eth_set_rx_vlan_filtering(struct dpaa2_eth_priv *priv,
1300 					   bool enable)
1301 {
1302 	int err;
1303 
1304 	err = dpni_enable_vlan_filter(priv->mc_io, 0, priv->mc_token, enable);
1305 
1306 	if (err) {
1307 		netdev_err(priv->net_dev,
1308 			   "dpni_enable_vlan_filter failed\n");
1309 		return err;
1310 	}
1311 
1312 	return 0;
1313 }
1314 
1315 static int dpaa2_eth_set_rx_csum(struct dpaa2_eth_priv *priv, bool enable)
1316 {
1317 	int err;
1318 
1319 	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1320 			       DPNI_OFF_RX_L3_CSUM, enable);
1321 	if (err) {
1322 		netdev_err(priv->net_dev,
1323 			   "dpni_set_offload(RX_L3_CSUM) failed\n");
1324 		return err;
1325 	}
1326 
1327 	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1328 			       DPNI_OFF_RX_L4_CSUM, enable);
1329 	if (err) {
1330 		netdev_err(priv->net_dev,
1331 			   "dpni_set_offload(RX_L4_CSUM) failed\n");
1332 		return err;
1333 	}
1334 
1335 	return 0;
1336 }
1337 
1338 static int dpaa2_eth_set_tx_csum(struct dpaa2_eth_priv *priv, bool enable)
1339 {
1340 	int err;
1341 
1342 	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1343 			       DPNI_OFF_TX_L3_CSUM, enable);
1344 	if (err) {
1345 		netdev_err(priv->net_dev, "dpni_set_offload(TX_L3_CSUM) failed\n");
1346 		return err;
1347 	}
1348 
1349 	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1350 			       DPNI_OFF_TX_L4_CSUM, enable);
1351 	if (err) {
1352 		netdev_err(priv->net_dev, "dpni_set_offload(TX_L4_CSUM) failed\n");
1353 		return err;
1354 	}
1355 
1356 	return 0;
1357 }
1358 
1359 /* Perform a single release command to add buffers
1360  * to the specified buffer pool
1361  */
1362 static int dpaa2_eth_add_bufs(struct dpaa2_eth_priv *priv,
1363 			      struct dpaa2_eth_channel *ch, u16 bpid)
1364 {
1365 	struct device *dev = priv->net_dev->dev.parent;
1366 	u64 buf_array[DPAA2_ETH_BUFS_PER_CMD];
1367 	struct page *page;
1368 	dma_addr_t addr;
1369 	int retries = 0;
1370 	int i, err;
1371 
1372 	for (i = 0; i < DPAA2_ETH_BUFS_PER_CMD; i++) {
1373 		/* Allocate buffer visible to WRIOP + skb shared info +
1374 		 * alignment padding
1375 		 */
1376 		/* allocate one page for each Rx buffer. WRIOP sees
1377 		 * the entire page except for a tailroom reserved for
1378 		 * skb shared info
1379 		 */
1380 		page = dev_alloc_pages(0);
1381 		if (!page)
1382 			goto err_alloc;
1383 
1384 		addr = dma_map_page(dev, page, 0, priv->rx_buf_size,
1385 				    DMA_BIDIRECTIONAL);
1386 		if (unlikely(dma_mapping_error(dev, addr)))
1387 			goto err_map;
1388 
1389 		buf_array[i] = addr;
1390 
1391 		/* tracing point */
1392 		trace_dpaa2_eth_buf_seed(priv->net_dev,
1393 					 page, DPAA2_ETH_RX_BUF_RAW_SIZE,
1394 					 addr, priv->rx_buf_size,
1395 					 bpid);
1396 	}
1397 
1398 release_bufs:
1399 	/* In case the portal is busy, retry until successful */
1400 	while ((err = dpaa2_io_service_release(ch->dpio, bpid,
1401 					       buf_array, i)) == -EBUSY) {
1402 		if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES)
1403 			break;
1404 		cpu_relax();
1405 	}
1406 
1407 	/* If release command failed, clean up and bail out;
1408 	 * not much else we can do about it
1409 	 */
1410 	if (err) {
1411 		dpaa2_eth_free_bufs(priv, buf_array, i);
1412 		return 0;
1413 	}
1414 
1415 	return i;
1416 
1417 err_map:
1418 	__free_pages(page, 0);
1419 err_alloc:
1420 	/* If we managed to allocate at least some buffers,
1421 	 * release them to hardware
1422 	 */
1423 	if (i)
1424 		goto release_bufs;
1425 
1426 	return 0;
1427 }
1428 
1429 static int dpaa2_eth_seed_pool(struct dpaa2_eth_priv *priv, u16 bpid)
1430 {
1431 	int i, j;
1432 	int new_count;
1433 
1434 	for (j = 0; j < priv->num_channels; j++) {
1435 		for (i = 0; i < DPAA2_ETH_NUM_BUFS;
1436 		     i += DPAA2_ETH_BUFS_PER_CMD) {
1437 			new_count = dpaa2_eth_add_bufs(priv, priv->channel[j], bpid);
1438 			priv->channel[j]->buf_count += new_count;
1439 
1440 			if (new_count < DPAA2_ETH_BUFS_PER_CMD) {
1441 				return -ENOMEM;
1442 			}
1443 		}
1444 	}
1445 
1446 	return 0;
1447 }
1448 
1449 /*
1450  * Drain the specified number of buffers from the DPNI's private buffer pool.
1451  * @count must not exceeed DPAA2_ETH_BUFS_PER_CMD
1452  */
1453 static void dpaa2_eth_drain_bufs(struct dpaa2_eth_priv *priv, int count)
1454 {
1455 	u64 buf_array[DPAA2_ETH_BUFS_PER_CMD];
1456 	int retries = 0;
1457 	int ret;
1458 
1459 	do {
1460 		ret = dpaa2_io_service_acquire(NULL, priv->bpid,
1461 					       buf_array, count);
1462 		if (ret < 0) {
1463 			if (ret == -EBUSY &&
1464 			    retries++ < DPAA2_ETH_SWP_BUSY_RETRIES)
1465 				continue;
1466 			netdev_err(priv->net_dev, "dpaa2_io_service_acquire() failed\n");
1467 			return;
1468 		}
1469 		dpaa2_eth_free_bufs(priv, buf_array, ret);
1470 		retries = 0;
1471 	} while (ret);
1472 }
1473 
1474 static void dpaa2_eth_drain_pool(struct dpaa2_eth_priv *priv)
1475 {
1476 	int i;
1477 
1478 	dpaa2_eth_drain_bufs(priv, DPAA2_ETH_BUFS_PER_CMD);
1479 	dpaa2_eth_drain_bufs(priv, 1);
1480 
1481 	for (i = 0; i < priv->num_channels; i++)
1482 		priv->channel[i]->buf_count = 0;
1483 }
1484 
1485 /* Function is called from softirq context only, so we don't need to guard
1486  * the access to percpu count
1487  */
1488 static int dpaa2_eth_refill_pool(struct dpaa2_eth_priv *priv,
1489 				 struct dpaa2_eth_channel *ch,
1490 				 u16 bpid)
1491 {
1492 	int new_count;
1493 
1494 	if (likely(ch->buf_count >= DPAA2_ETH_REFILL_THRESH))
1495 		return 0;
1496 
1497 	do {
1498 		new_count = dpaa2_eth_add_bufs(priv, ch, bpid);
1499 		if (unlikely(!new_count)) {
1500 			/* Out of memory; abort for now, we'll try later on */
1501 			break;
1502 		}
1503 		ch->buf_count += new_count;
1504 	} while (ch->buf_count < DPAA2_ETH_NUM_BUFS);
1505 
1506 	if (unlikely(ch->buf_count < DPAA2_ETH_NUM_BUFS))
1507 		return -ENOMEM;
1508 
1509 	return 0;
1510 }
1511 
1512 static void dpaa2_eth_sgt_cache_drain(struct dpaa2_eth_priv *priv)
1513 {
1514 	struct dpaa2_eth_sgt_cache *sgt_cache;
1515 	u16 count;
1516 	int k, i;
1517 
1518 	for_each_possible_cpu(k) {
1519 		sgt_cache = per_cpu_ptr(priv->sgt_cache, k);
1520 		count = sgt_cache->count;
1521 
1522 		for (i = 0; i < count; i++)
1523 			kfree(sgt_cache->buf[i]);
1524 		sgt_cache->count = 0;
1525 	}
1526 }
1527 
1528 static int dpaa2_eth_pull_channel(struct dpaa2_eth_channel *ch)
1529 {
1530 	int err;
1531 	int dequeues = -1;
1532 
1533 	/* Retry while portal is busy */
1534 	do {
1535 		err = dpaa2_io_service_pull_channel(ch->dpio, ch->ch_id,
1536 						    ch->store);
1537 		dequeues++;
1538 		cpu_relax();
1539 	} while (err == -EBUSY && dequeues < DPAA2_ETH_SWP_BUSY_RETRIES);
1540 
1541 	ch->stats.dequeue_portal_busy += dequeues;
1542 	if (unlikely(err))
1543 		ch->stats.pull_err++;
1544 
1545 	return err;
1546 }
1547 
1548 /* NAPI poll routine
1549  *
1550  * Frames are dequeued from the QMan channel associated with this NAPI context.
1551  * Rx, Tx confirmation and (if configured) Rx error frames all count
1552  * towards the NAPI budget.
1553  */
1554 static int dpaa2_eth_poll(struct napi_struct *napi, int budget)
1555 {
1556 	struct dpaa2_eth_channel *ch;
1557 	struct dpaa2_eth_priv *priv;
1558 	int rx_cleaned = 0, txconf_cleaned = 0;
1559 	struct dpaa2_eth_fq *fq, *txc_fq = NULL;
1560 	struct netdev_queue *nq;
1561 	int store_cleaned, work_done;
1562 	struct list_head rx_list;
1563 	int retries = 0;
1564 	u16 flowid;
1565 	int err;
1566 
1567 	ch = container_of(napi, struct dpaa2_eth_channel, napi);
1568 	ch->xdp.res = 0;
1569 	priv = ch->priv;
1570 
1571 	INIT_LIST_HEAD(&rx_list);
1572 	ch->rx_list = &rx_list;
1573 
1574 	do {
1575 		err = dpaa2_eth_pull_channel(ch);
1576 		if (unlikely(err))
1577 			break;
1578 
1579 		/* Refill pool if appropriate */
1580 		dpaa2_eth_refill_pool(priv, ch, priv->bpid);
1581 
1582 		store_cleaned = dpaa2_eth_consume_frames(ch, &fq);
1583 		if (store_cleaned <= 0)
1584 			break;
1585 		if (fq->type == DPAA2_RX_FQ) {
1586 			rx_cleaned += store_cleaned;
1587 			flowid = fq->flowid;
1588 		} else {
1589 			txconf_cleaned += store_cleaned;
1590 			/* We have a single Tx conf FQ on this channel */
1591 			txc_fq = fq;
1592 		}
1593 
1594 		/* If we either consumed the whole NAPI budget with Rx frames
1595 		 * or we reached the Tx confirmations threshold, we're done.
1596 		 */
1597 		if (rx_cleaned >= budget ||
1598 		    txconf_cleaned >= DPAA2_ETH_TXCONF_PER_NAPI) {
1599 			work_done = budget;
1600 			goto out;
1601 		}
1602 	} while (store_cleaned);
1603 
1604 	/* We didn't consume the entire budget, so finish napi and
1605 	 * re-enable data availability notifications
1606 	 */
1607 	napi_complete_done(napi, rx_cleaned);
1608 	do {
1609 		err = dpaa2_io_service_rearm(ch->dpio, &ch->nctx);
1610 		cpu_relax();
1611 	} while (err == -EBUSY && retries++ < DPAA2_ETH_SWP_BUSY_RETRIES);
1612 	WARN_ONCE(err, "CDAN notifications rearm failed on core %d",
1613 		  ch->nctx.desired_cpu);
1614 
1615 	work_done = max(rx_cleaned, 1);
1616 
1617 out:
1618 	netif_receive_skb_list(ch->rx_list);
1619 
1620 	if (txc_fq && txc_fq->dq_frames) {
1621 		nq = netdev_get_tx_queue(priv->net_dev, txc_fq->flowid);
1622 		netdev_tx_completed_queue(nq, txc_fq->dq_frames,
1623 					  txc_fq->dq_bytes);
1624 		txc_fq->dq_frames = 0;
1625 		txc_fq->dq_bytes = 0;
1626 	}
1627 
1628 	if (ch->xdp.res & XDP_REDIRECT)
1629 		xdp_do_flush_map();
1630 	else if (rx_cleaned && ch->xdp.res & XDP_TX)
1631 		dpaa2_eth_xdp_tx_flush(priv, ch, &priv->fq[flowid]);
1632 
1633 	return work_done;
1634 }
1635 
1636 static void dpaa2_eth_enable_ch_napi(struct dpaa2_eth_priv *priv)
1637 {
1638 	struct dpaa2_eth_channel *ch;
1639 	int i;
1640 
1641 	for (i = 0; i < priv->num_channels; i++) {
1642 		ch = priv->channel[i];
1643 		napi_enable(&ch->napi);
1644 	}
1645 }
1646 
1647 static void dpaa2_eth_disable_ch_napi(struct dpaa2_eth_priv *priv)
1648 {
1649 	struct dpaa2_eth_channel *ch;
1650 	int i;
1651 
1652 	for (i = 0; i < priv->num_channels; i++) {
1653 		ch = priv->channel[i];
1654 		napi_disable(&ch->napi);
1655 	}
1656 }
1657 
1658 void dpaa2_eth_set_rx_taildrop(struct dpaa2_eth_priv *priv,
1659 			       bool tx_pause, bool pfc)
1660 {
1661 	struct dpni_taildrop td = {0};
1662 	struct dpaa2_eth_fq *fq;
1663 	int i, err;
1664 
1665 	/* FQ taildrop: threshold is in bytes, per frame queue. Enabled if
1666 	 * flow control is disabled (as it might interfere with either the
1667 	 * buffer pool depletion trigger for pause frames or with the group
1668 	 * congestion trigger for PFC frames)
1669 	 */
1670 	td.enable = !tx_pause;
1671 	if (priv->rx_fqtd_enabled == td.enable)
1672 		goto set_cgtd;
1673 
1674 	td.threshold = DPAA2_ETH_FQ_TAILDROP_THRESH;
1675 	td.units = DPNI_CONGESTION_UNIT_BYTES;
1676 
1677 	for (i = 0; i < priv->num_fqs; i++) {
1678 		fq = &priv->fq[i];
1679 		if (fq->type != DPAA2_RX_FQ)
1680 			continue;
1681 		err = dpni_set_taildrop(priv->mc_io, 0, priv->mc_token,
1682 					DPNI_CP_QUEUE, DPNI_QUEUE_RX,
1683 					fq->tc, fq->flowid, &td);
1684 		if (err) {
1685 			netdev_err(priv->net_dev,
1686 				   "dpni_set_taildrop(FQ) failed\n");
1687 			return;
1688 		}
1689 	}
1690 
1691 	priv->rx_fqtd_enabled = td.enable;
1692 
1693 set_cgtd:
1694 	/* Congestion group taildrop: threshold is in frames, per group
1695 	 * of FQs belonging to the same traffic class
1696 	 * Enabled if general Tx pause disabled or if PFCs are enabled
1697 	 * (congestion group threhsold for PFC generation is lower than the
1698 	 * CG taildrop threshold, so it won't interfere with it; we also
1699 	 * want frames in non-PFC enabled traffic classes to be kept in check)
1700 	 */
1701 	td.enable = !tx_pause || pfc;
1702 	if (priv->rx_cgtd_enabled == td.enable)
1703 		return;
1704 
1705 	td.threshold = DPAA2_ETH_CG_TAILDROP_THRESH(priv);
1706 	td.units = DPNI_CONGESTION_UNIT_FRAMES;
1707 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
1708 		err = dpni_set_taildrop(priv->mc_io, 0, priv->mc_token,
1709 					DPNI_CP_GROUP, DPNI_QUEUE_RX,
1710 					i, 0, &td);
1711 		if (err) {
1712 			netdev_err(priv->net_dev,
1713 				   "dpni_set_taildrop(CG) failed\n");
1714 			return;
1715 		}
1716 	}
1717 
1718 	priv->rx_cgtd_enabled = td.enable;
1719 }
1720 
1721 static int dpaa2_eth_link_state_update(struct dpaa2_eth_priv *priv)
1722 {
1723 	struct dpni_link_state state = {0};
1724 	bool tx_pause;
1725 	int err;
1726 
1727 	err = dpni_get_link_state(priv->mc_io, 0, priv->mc_token, &state);
1728 	if (unlikely(err)) {
1729 		netdev_err(priv->net_dev,
1730 			   "dpni_get_link_state() failed\n");
1731 		return err;
1732 	}
1733 
1734 	/* If Tx pause frame settings have changed, we need to update
1735 	 * Rx FQ taildrop configuration as well. We configure taildrop
1736 	 * only when pause frame generation is disabled.
1737 	 */
1738 	tx_pause = dpaa2_eth_tx_pause_enabled(state.options);
1739 	dpaa2_eth_set_rx_taildrop(priv, tx_pause, priv->pfc_enabled);
1740 
1741 	/* When we manage the MAC/PHY using phylink there is no need
1742 	 * to manually update the netif_carrier.
1743 	 */
1744 	if (dpaa2_eth_is_type_phy(priv))
1745 		goto out;
1746 
1747 	/* Chech link state; speed / duplex changes are not treated yet */
1748 	if (priv->link_state.up == state.up)
1749 		goto out;
1750 
1751 	if (state.up) {
1752 		netif_carrier_on(priv->net_dev);
1753 		netif_tx_start_all_queues(priv->net_dev);
1754 	} else {
1755 		netif_tx_stop_all_queues(priv->net_dev);
1756 		netif_carrier_off(priv->net_dev);
1757 	}
1758 
1759 	netdev_info(priv->net_dev, "Link Event: state %s\n",
1760 		    state.up ? "up" : "down");
1761 
1762 out:
1763 	priv->link_state = state;
1764 
1765 	return 0;
1766 }
1767 
1768 static int dpaa2_eth_open(struct net_device *net_dev)
1769 {
1770 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1771 	int err;
1772 
1773 	err = dpaa2_eth_seed_pool(priv, priv->bpid);
1774 	if (err) {
1775 		/* Not much to do; the buffer pool, though not filled up,
1776 		 * may still contain some buffers which would enable us
1777 		 * to limp on.
1778 		 */
1779 		netdev_err(net_dev, "Buffer seeding failed for DPBP %d (bpid=%d)\n",
1780 			   priv->dpbp_dev->obj_desc.id, priv->bpid);
1781 	}
1782 
1783 	if (!dpaa2_eth_is_type_phy(priv)) {
1784 		/* We'll only start the txqs when the link is actually ready;
1785 		 * make sure we don't race against the link up notification,
1786 		 * which may come immediately after dpni_enable();
1787 		 */
1788 		netif_tx_stop_all_queues(net_dev);
1789 
1790 		/* Also, explicitly set carrier off, otherwise
1791 		 * netif_carrier_ok() will return true and cause 'ip link show'
1792 		 * to report the LOWER_UP flag, even though the link
1793 		 * notification wasn't even received.
1794 		 */
1795 		netif_carrier_off(net_dev);
1796 	}
1797 	dpaa2_eth_enable_ch_napi(priv);
1798 
1799 	err = dpni_enable(priv->mc_io, 0, priv->mc_token);
1800 	if (err < 0) {
1801 		netdev_err(net_dev, "dpni_enable() failed\n");
1802 		goto enable_err;
1803 	}
1804 
1805 	if (dpaa2_eth_is_type_phy(priv))
1806 		phylink_start(priv->mac->phylink);
1807 
1808 	return 0;
1809 
1810 enable_err:
1811 	dpaa2_eth_disable_ch_napi(priv);
1812 	dpaa2_eth_drain_pool(priv);
1813 	return err;
1814 }
1815 
1816 /* Total number of in-flight frames on ingress queues */
1817 static u32 dpaa2_eth_ingress_fq_count(struct dpaa2_eth_priv *priv)
1818 {
1819 	struct dpaa2_eth_fq *fq;
1820 	u32 fcnt = 0, bcnt = 0, total = 0;
1821 	int i, err;
1822 
1823 	for (i = 0; i < priv->num_fqs; i++) {
1824 		fq = &priv->fq[i];
1825 		err = dpaa2_io_query_fq_count(NULL, fq->fqid, &fcnt, &bcnt);
1826 		if (err) {
1827 			netdev_warn(priv->net_dev, "query_fq_count failed");
1828 			break;
1829 		}
1830 		total += fcnt;
1831 	}
1832 
1833 	return total;
1834 }
1835 
1836 static void dpaa2_eth_wait_for_ingress_fq_empty(struct dpaa2_eth_priv *priv)
1837 {
1838 	int retries = 10;
1839 	u32 pending;
1840 
1841 	do {
1842 		pending = dpaa2_eth_ingress_fq_count(priv);
1843 		if (pending)
1844 			msleep(100);
1845 	} while (pending && --retries);
1846 }
1847 
1848 #define DPNI_TX_PENDING_VER_MAJOR	7
1849 #define DPNI_TX_PENDING_VER_MINOR	13
1850 static void dpaa2_eth_wait_for_egress_fq_empty(struct dpaa2_eth_priv *priv)
1851 {
1852 	union dpni_statistics stats;
1853 	int retries = 10;
1854 	int err;
1855 
1856 	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_TX_PENDING_VER_MAJOR,
1857 				   DPNI_TX_PENDING_VER_MINOR) < 0)
1858 		goto out;
1859 
1860 	do {
1861 		err = dpni_get_statistics(priv->mc_io, 0, priv->mc_token, 6,
1862 					  &stats);
1863 		if (err)
1864 			goto out;
1865 		if (stats.page_6.tx_pending_frames == 0)
1866 			return;
1867 	} while (--retries);
1868 
1869 out:
1870 	msleep(500);
1871 }
1872 
1873 static int dpaa2_eth_stop(struct net_device *net_dev)
1874 {
1875 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1876 	int dpni_enabled = 0;
1877 	int retries = 10;
1878 
1879 	if (dpaa2_eth_is_type_phy(priv)) {
1880 		phylink_stop(priv->mac->phylink);
1881 	} else {
1882 		netif_tx_stop_all_queues(net_dev);
1883 		netif_carrier_off(net_dev);
1884 	}
1885 
1886 	/* On dpni_disable(), the MC firmware will:
1887 	 * - stop MAC Rx and wait for all Rx frames to be enqueued to software
1888 	 * - cut off WRIOP dequeues from egress FQs and wait until transmission
1889 	 * of all in flight Tx frames is finished (and corresponding Tx conf
1890 	 * frames are enqueued back to software)
1891 	 *
1892 	 * Before calling dpni_disable(), we wait for all Tx frames to arrive
1893 	 * on WRIOP. After it finishes, wait until all remaining frames on Rx
1894 	 * and Tx conf queues are consumed on NAPI poll.
1895 	 */
1896 	dpaa2_eth_wait_for_egress_fq_empty(priv);
1897 
1898 	do {
1899 		dpni_disable(priv->mc_io, 0, priv->mc_token);
1900 		dpni_is_enabled(priv->mc_io, 0, priv->mc_token, &dpni_enabled);
1901 		if (dpni_enabled)
1902 			/* Allow the hardware some slack */
1903 			msleep(100);
1904 	} while (dpni_enabled && --retries);
1905 	if (!retries) {
1906 		netdev_warn(net_dev, "Retry count exceeded disabling DPNI\n");
1907 		/* Must go on and disable NAPI nonetheless, so we don't crash at
1908 		 * the next "ifconfig up"
1909 		 */
1910 	}
1911 
1912 	dpaa2_eth_wait_for_ingress_fq_empty(priv);
1913 	dpaa2_eth_disable_ch_napi(priv);
1914 
1915 	/* Empty the buffer pool */
1916 	dpaa2_eth_drain_pool(priv);
1917 
1918 	/* Empty the Scatter-Gather Buffer cache */
1919 	dpaa2_eth_sgt_cache_drain(priv);
1920 
1921 	return 0;
1922 }
1923 
1924 static int dpaa2_eth_set_addr(struct net_device *net_dev, void *addr)
1925 {
1926 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1927 	struct device *dev = net_dev->dev.parent;
1928 	int err;
1929 
1930 	err = eth_mac_addr(net_dev, addr);
1931 	if (err < 0) {
1932 		dev_err(dev, "eth_mac_addr() failed (%d)\n", err);
1933 		return err;
1934 	}
1935 
1936 	err = dpni_set_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
1937 					net_dev->dev_addr);
1938 	if (err) {
1939 		dev_err(dev, "dpni_set_primary_mac_addr() failed (%d)\n", err);
1940 		return err;
1941 	}
1942 
1943 	return 0;
1944 }
1945 
1946 /** Fill in counters maintained by the GPP driver. These may be different from
1947  * the hardware counters obtained by ethtool.
1948  */
1949 static void dpaa2_eth_get_stats(struct net_device *net_dev,
1950 				struct rtnl_link_stats64 *stats)
1951 {
1952 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1953 	struct rtnl_link_stats64 *percpu_stats;
1954 	u64 *cpustats;
1955 	u64 *netstats = (u64 *)stats;
1956 	int i, j;
1957 	int num = sizeof(struct rtnl_link_stats64) / sizeof(u64);
1958 
1959 	for_each_possible_cpu(i) {
1960 		percpu_stats = per_cpu_ptr(priv->percpu_stats, i);
1961 		cpustats = (u64 *)percpu_stats;
1962 		for (j = 0; j < num; j++)
1963 			netstats[j] += cpustats[j];
1964 	}
1965 }
1966 
1967 /* Copy mac unicast addresses from @net_dev to @priv.
1968  * Its sole purpose is to make dpaa2_eth_set_rx_mode() more readable.
1969  */
1970 static void dpaa2_eth_add_uc_hw_addr(const struct net_device *net_dev,
1971 				     struct dpaa2_eth_priv *priv)
1972 {
1973 	struct netdev_hw_addr *ha;
1974 	int err;
1975 
1976 	netdev_for_each_uc_addr(ha, net_dev) {
1977 		err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token,
1978 					ha->addr);
1979 		if (err)
1980 			netdev_warn(priv->net_dev,
1981 				    "Could not add ucast MAC %pM to the filtering table (err %d)\n",
1982 				    ha->addr, err);
1983 	}
1984 }
1985 
1986 /* Copy mac multicast addresses from @net_dev to @priv
1987  * Its sole purpose is to make dpaa2_eth_set_rx_mode() more readable.
1988  */
1989 static void dpaa2_eth_add_mc_hw_addr(const struct net_device *net_dev,
1990 				     struct dpaa2_eth_priv *priv)
1991 {
1992 	struct netdev_hw_addr *ha;
1993 	int err;
1994 
1995 	netdev_for_each_mc_addr(ha, net_dev) {
1996 		err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token,
1997 					ha->addr);
1998 		if (err)
1999 			netdev_warn(priv->net_dev,
2000 				    "Could not add mcast MAC %pM to the filtering table (err %d)\n",
2001 				    ha->addr, err);
2002 	}
2003 }
2004 
2005 static int dpaa2_eth_rx_add_vid(struct net_device *net_dev,
2006 				__be16 vlan_proto, u16 vid)
2007 {
2008 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2009 	int err;
2010 
2011 	err = dpni_add_vlan_id(priv->mc_io, 0, priv->mc_token,
2012 			       vid, 0, 0, 0);
2013 
2014 	if (err) {
2015 		netdev_warn(priv->net_dev,
2016 			    "Could not add the vlan id %u\n",
2017 			    vid);
2018 		return err;
2019 	}
2020 
2021 	return 0;
2022 }
2023 
2024 static int dpaa2_eth_rx_kill_vid(struct net_device *net_dev,
2025 				 __be16 vlan_proto, u16 vid)
2026 {
2027 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2028 	int err;
2029 
2030 	err = dpni_remove_vlan_id(priv->mc_io, 0, priv->mc_token, vid);
2031 
2032 	if (err) {
2033 		netdev_warn(priv->net_dev,
2034 			    "Could not remove the vlan id %u\n",
2035 			    vid);
2036 		return err;
2037 	}
2038 
2039 	return 0;
2040 }
2041 
2042 static void dpaa2_eth_set_rx_mode(struct net_device *net_dev)
2043 {
2044 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2045 	int uc_count = netdev_uc_count(net_dev);
2046 	int mc_count = netdev_mc_count(net_dev);
2047 	u8 max_mac = priv->dpni_attrs.mac_filter_entries;
2048 	u32 options = priv->dpni_attrs.options;
2049 	u16 mc_token = priv->mc_token;
2050 	struct fsl_mc_io *mc_io = priv->mc_io;
2051 	int err;
2052 
2053 	/* Basic sanity checks; these probably indicate a misconfiguration */
2054 	if (options & DPNI_OPT_NO_MAC_FILTER && max_mac != 0)
2055 		netdev_info(net_dev,
2056 			    "mac_filter_entries=%d, DPNI_OPT_NO_MAC_FILTER option must be disabled\n",
2057 			    max_mac);
2058 
2059 	/* Force promiscuous if the uc or mc counts exceed our capabilities. */
2060 	if (uc_count > max_mac) {
2061 		netdev_info(net_dev,
2062 			    "Unicast addr count reached %d, max allowed is %d; forcing promisc\n",
2063 			    uc_count, max_mac);
2064 		goto force_promisc;
2065 	}
2066 	if (mc_count + uc_count > max_mac) {
2067 		netdev_info(net_dev,
2068 			    "Unicast + multicast addr count reached %d, max allowed is %d; forcing promisc\n",
2069 			    uc_count + mc_count, max_mac);
2070 		goto force_mc_promisc;
2071 	}
2072 
2073 	/* Adjust promisc settings due to flag combinations */
2074 	if (net_dev->flags & IFF_PROMISC)
2075 		goto force_promisc;
2076 	if (net_dev->flags & IFF_ALLMULTI) {
2077 		/* First, rebuild unicast filtering table. This should be done
2078 		 * in promisc mode, in order to avoid frame loss while we
2079 		 * progressively add entries to the table.
2080 		 * We don't know whether we had been in promisc already, and
2081 		 * making an MC call to find out is expensive; so set uc promisc
2082 		 * nonetheless.
2083 		 */
2084 		err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 1);
2085 		if (err)
2086 			netdev_warn(net_dev, "Can't set uc promisc\n");
2087 
2088 		/* Actual uc table reconstruction. */
2089 		err = dpni_clear_mac_filters(mc_io, 0, mc_token, 1, 0);
2090 		if (err)
2091 			netdev_warn(net_dev, "Can't clear uc filters\n");
2092 		dpaa2_eth_add_uc_hw_addr(net_dev, priv);
2093 
2094 		/* Finally, clear uc promisc and set mc promisc as requested. */
2095 		err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 0);
2096 		if (err)
2097 			netdev_warn(net_dev, "Can't clear uc promisc\n");
2098 		goto force_mc_promisc;
2099 	}
2100 
2101 	/* Neither unicast, nor multicast promisc will be on... eventually.
2102 	 * For now, rebuild mac filtering tables while forcing both of them on.
2103 	 */
2104 	err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 1);
2105 	if (err)
2106 		netdev_warn(net_dev, "Can't set uc promisc (%d)\n", err);
2107 	err = dpni_set_multicast_promisc(mc_io, 0, mc_token, 1);
2108 	if (err)
2109 		netdev_warn(net_dev, "Can't set mc promisc (%d)\n", err);
2110 
2111 	/* Actual mac filtering tables reconstruction */
2112 	err = dpni_clear_mac_filters(mc_io, 0, mc_token, 1, 1);
2113 	if (err)
2114 		netdev_warn(net_dev, "Can't clear mac filters\n");
2115 	dpaa2_eth_add_mc_hw_addr(net_dev, priv);
2116 	dpaa2_eth_add_uc_hw_addr(net_dev, priv);
2117 
2118 	/* Now we can clear both ucast and mcast promisc, without risking
2119 	 * to drop legitimate frames anymore.
2120 	 */
2121 	err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 0);
2122 	if (err)
2123 		netdev_warn(net_dev, "Can't clear ucast promisc\n");
2124 	err = dpni_set_multicast_promisc(mc_io, 0, mc_token, 0);
2125 	if (err)
2126 		netdev_warn(net_dev, "Can't clear mcast promisc\n");
2127 
2128 	return;
2129 
2130 force_promisc:
2131 	err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 1);
2132 	if (err)
2133 		netdev_warn(net_dev, "Can't set ucast promisc\n");
2134 force_mc_promisc:
2135 	err = dpni_set_multicast_promisc(mc_io, 0, mc_token, 1);
2136 	if (err)
2137 		netdev_warn(net_dev, "Can't set mcast promisc\n");
2138 }
2139 
2140 static int dpaa2_eth_set_features(struct net_device *net_dev,
2141 				  netdev_features_t features)
2142 {
2143 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2144 	netdev_features_t changed = features ^ net_dev->features;
2145 	bool enable;
2146 	int err;
2147 
2148 	if (changed & NETIF_F_HW_VLAN_CTAG_FILTER) {
2149 		enable = !!(features & NETIF_F_HW_VLAN_CTAG_FILTER);
2150 		err = dpaa2_eth_set_rx_vlan_filtering(priv, enable);
2151 		if (err)
2152 			return err;
2153 	}
2154 
2155 	if (changed & NETIF_F_RXCSUM) {
2156 		enable = !!(features & NETIF_F_RXCSUM);
2157 		err = dpaa2_eth_set_rx_csum(priv, enable);
2158 		if (err)
2159 			return err;
2160 	}
2161 
2162 	if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
2163 		enable = !!(features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM));
2164 		err = dpaa2_eth_set_tx_csum(priv, enable);
2165 		if (err)
2166 			return err;
2167 	}
2168 
2169 	return 0;
2170 }
2171 
2172 static int dpaa2_eth_ts_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
2173 {
2174 	struct dpaa2_eth_priv *priv = netdev_priv(dev);
2175 	struct hwtstamp_config config;
2176 
2177 	if (!dpaa2_ptp)
2178 		return -EINVAL;
2179 
2180 	if (copy_from_user(&config, rq->ifr_data, sizeof(config)))
2181 		return -EFAULT;
2182 
2183 	switch (config.tx_type) {
2184 	case HWTSTAMP_TX_OFF:
2185 	case HWTSTAMP_TX_ON:
2186 	case HWTSTAMP_TX_ONESTEP_SYNC:
2187 		priv->tx_tstamp_type = config.tx_type;
2188 		break;
2189 	default:
2190 		return -ERANGE;
2191 	}
2192 
2193 	if (config.rx_filter == HWTSTAMP_FILTER_NONE) {
2194 		priv->rx_tstamp = false;
2195 	} else {
2196 		priv->rx_tstamp = true;
2197 		/* TS is set for all frame types, not only those requested */
2198 		config.rx_filter = HWTSTAMP_FILTER_ALL;
2199 	}
2200 
2201 	return copy_to_user(rq->ifr_data, &config, sizeof(config)) ?
2202 			-EFAULT : 0;
2203 }
2204 
2205 static int dpaa2_eth_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
2206 {
2207 	struct dpaa2_eth_priv *priv = netdev_priv(dev);
2208 
2209 	if (cmd == SIOCSHWTSTAMP)
2210 		return dpaa2_eth_ts_ioctl(dev, rq, cmd);
2211 
2212 	if (dpaa2_eth_is_type_phy(priv))
2213 		return phylink_mii_ioctl(priv->mac->phylink, rq, cmd);
2214 
2215 	return -EOPNOTSUPP;
2216 }
2217 
2218 static bool xdp_mtu_valid(struct dpaa2_eth_priv *priv, int mtu)
2219 {
2220 	int mfl, linear_mfl;
2221 
2222 	mfl = DPAA2_ETH_L2_MAX_FRM(mtu);
2223 	linear_mfl = priv->rx_buf_size - DPAA2_ETH_RX_HWA_SIZE -
2224 		     dpaa2_eth_rx_head_room(priv) - XDP_PACKET_HEADROOM;
2225 
2226 	if (mfl > linear_mfl) {
2227 		netdev_warn(priv->net_dev, "Maximum MTU for XDP is %d\n",
2228 			    linear_mfl - VLAN_ETH_HLEN);
2229 		return false;
2230 	}
2231 
2232 	return true;
2233 }
2234 
2235 static int dpaa2_eth_set_rx_mfl(struct dpaa2_eth_priv *priv, int mtu, bool has_xdp)
2236 {
2237 	int mfl, err;
2238 
2239 	/* We enforce a maximum Rx frame length based on MTU only if we have
2240 	 * an XDP program attached (in order to avoid Rx S/G frames).
2241 	 * Otherwise, we accept all incoming frames as long as they are not
2242 	 * larger than maximum size supported in hardware
2243 	 */
2244 	if (has_xdp)
2245 		mfl = DPAA2_ETH_L2_MAX_FRM(mtu);
2246 	else
2247 		mfl = DPAA2_ETH_MFL;
2248 
2249 	err = dpni_set_max_frame_length(priv->mc_io, 0, priv->mc_token, mfl);
2250 	if (err) {
2251 		netdev_err(priv->net_dev, "dpni_set_max_frame_length failed\n");
2252 		return err;
2253 	}
2254 
2255 	return 0;
2256 }
2257 
2258 static int dpaa2_eth_change_mtu(struct net_device *dev, int new_mtu)
2259 {
2260 	struct dpaa2_eth_priv *priv = netdev_priv(dev);
2261 	int err;
2262 
2263 	if (!priv->xdp_prog)
2264 		goto out;
2265 
2266 	if (!xdp_mtu_valid(priv, new_mtu))
2267 		return -EINVAL;
2268 
2269 	err = dpaa2_eth_set_rx_mfl(priv, new_mtu, true);
2270 	if (err)
2271 		return err;
2272 
2273 out:
2274 	dev->mtu = new_mtu;
2275 	return 0;
2276 }
2277 
2278 static int dpaa2_eth_update_rx_buffer_headroom(struct dpaa2_eth_priv *priv, bool has_xdp)
2279 {
2280 	struct dpni_buffer_layout buf_layout = {0};
2281 	int err;
2282 
2283 	err = dpni_get_buffer_layout(priv->mc_io, 0, priv->mc_token,
2284 				     DPNI_QUEUE_RX, &buf_layout);
2285 	if (err) {
2286 		netdev_err(priv->net_dev, "dpni_get_buffer_layout failed\n");
2287 		return err;
2288 	}
2289 
2290 	/* Reserve extra headroom for XDP header size changes */
2291 	buf_layout.data_head_room = dpaa2_eth_rx_head_room(priv) +
2292 				    (has_xdp ? XDP_PACKET_HEADROOM : 0);
2293 	buf_layout.options = DPNI_BUF_LAYOUT_OPT_DATA_HEAD_ROOM;
2294 	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
2295 				     DPNI_QUEUE_RX, &buf_layout);
2296 	if (err) {
2297 		netdev_err(priv->net_dev, "dpni_set_buffer_layout failed\n");
2298 		return err;
2299 	}
2300 
2301 	return 0;
2302 }
2303 
2304 static int dpaa2_eth_setup_xdp(struct net_device *dev, struct bpf_prog *prog)
2305 {
2306 	struct dpaa2_eth_priv *priv = netdev_priv(dev);
2307 	struct dpaa2_eth_channel *ch;
2308 	struct bpf_prog *old;
2309 	bool up, need_update;
2310 	int i, err;
2311 
2312 	if (prog && !xdp_mtu_valid(priv, dev->mtu))
2313 		return -EINVAL;
2314 
2315 	if (prog)
2316 		bpf_prog_add(prog, priv->num_channels);
2317 
2318 	up = netif_running(dev);
2319 	need_update = (!!priv->xdp_prog != !!prog);
2320 
2321 	if (up)
2322 		dpaa2_eth_stop(dev);
2323 
2324 	/* While in xdp mode, enforce a maximum Rx frame size based on MTU.
2325 	 * Also, when switching between xdp/non-xdp modes we need to reconfigure
2326 	 * our Rx buffer layout. Buffer pool was drained on dpaa2_eth_stop,
2327 	 * so we are sure no old format buffers will be used from now on.
2328 	 */
2329 	if (need_update) {
2330 		err = dpaa2_eth_set_rx_mfl(priv, dev->mtu, !!prog);
2331 		if (err)
2332 			goto out_err;
2333 		err = dpaa2_eth_update_rx_buffer_headroom(priv, !!prog);
2334 		if (err)
2335 			goto out_err;
2336 	}
2337 
2338 	old = xchg(&priv->xdp_prog, prog);
2339 	if (old)
2340 		bpf_prog_put(old);
2341 
2342 	for (i = 0; i < priv->num_channels; i++) {
2343 		ch = priv->channel[i];
2344 		old = xchg(&ch->xdp.prog, prog);
2345 		if (old)
2346 			bpf_prog_put(old);
2347 	}
2348 
2349 	if (up) {
2350 		err = dpaa2_eth_open(dev);
2351 		if (err)
2352 			return err;
2353 	}
2354 
2355 	return 0;
2356 
2357 out_err:
2358 	if (prog)
2359 		bpf_prog_sub(prog, priv->num_channels);
2360 	if (up)
2361 		dpaa2_eth_open(dev);
2362 
2363 	return err;
2364 }
2365 
2366 static int dpaa2_eth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2367 {
2368 	switch (xdp->command) {
2369 	case XDP_SETUP_PROG:
2370 		return dpaa2_eth_setup_xdp(dev, xdp->prog);
2371 	default:
2372 		return -EINVAL;
2373 	}
2374 
2375 	return 0;
2376 }
2377 
2378 static int dpaa2_eth_xdp_create_fd(struct net_device *net_dev,
2379 				   struct xdp_frame *xdpf,
2380 				   struct dpaa2_fd *fd)
2381 {
2382 	struct device *dev = net_dev->dev.parent;
2383 	unsigned int needed_headroom;
2384 	struct dpaa2_eth_swa *swa;
2385 	void *buffer_start, *aligned_start;
2386 	dma_addr_t addr;
2387 
2388 	/* We require a minimum headroom to be able to transmit the frame.
2389 	 * Otherwise return an error and let the original net_device handle it
2390 	 */
2391 	needed_headroom = dpaa2_eth_needed_headroom(NULL);
2392 	if (xdpf->headroom < needed_headroom)
2393 		return -EINVAL;
2394 
2395 	/* Setup the FD fields */
2396 	memset(fd, 0, sizeof(*fd));
2397 
2398 	/* Align FD address, if possible */
2399 	buffer_start = xdpf->data - needed_headroom;
2400 	aligned_start = PTR_ALIGN(buffer_start - DPAA2_ETH_TX_BUF_ALIGN,
2401 				  DPAA2_ETH_TX_BUF_ALIGN);
2402 	if (aligned_start >= xdpf->data - xdpf->headroom)
2403 		buffer_start = aligned_start;
2404 
2405 	swa = (struct dpaa2_eth_swa *)buffer_start;
2406 	/* fill in necessary fields here */
2407 	swa->type = DPAA2_ETH_SWA_XDP;
2408 	swa->xdp.dma_size = xdpf->data + xdpf->len - buffer_start;
2409 	swa->xdp.xdpf = xdpf;
2410 
2411 	addr = dma_map_single(dev, buffer_start,
2412 			      swa->xdp.dma_size,
2413 			      DMA_BIDIRECTIONAL);
2414 	if (unlikely(dma_mapping_error(dev, addr)))
2415 		return -ENOMEM;
2416 
2417 	dpaa2_fd_set_addr(fd, addr);
2418 	dpaa2_fd_set_offset(fd, xdpf->data - buffer_start);
2419 	dpaa2_fd_set_len(fd, xdpf->len);
2420 	dpaa2_fd_set_format(fd, dpaa2_fd_single);
2421 	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
2422 
2423 	return 0;
2424 }
2425 
2426 static int dpaa2_eth_xdp_xmit(struct net_device *net_dev, int n,
2427 			      struct xdp_frame **frames, u32 flags)
2428 {
2429 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2430 	struct dpaa2_eth_xdp_fds *xdp_redirect_fds;
2431 	struct rtnl_link_stats64 *percpu_stats;
2432 	struct dpaa2_eth_fq *fq;
2433 	struct dpaa2_fd *fds;
2434 	int enqueued, i, err;
2435 
2436 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
2437 		return -EINVAL;
2438 
2439 	if (!netif_running(net_dev))
2440 		return -ENETDOWN;
2441 
2442 	fq = &priv->fq[smp_processor_id()];
2443 	xdp_redirect_fds = &fq->xdp_redirect_fds;
2444 	fds = xdp_redirect_fds->fds;
2445 
2446 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
2447 
2448 	/* create a FD for each xdp_frame in the list received */
2449 	for (i = 0; i < n; i++) {
2450 		err = dpaa2_eth_xdp_create_fd(net_dev, frames[i], &fds[i]);
2451 		if (err)
2452 			break;
2453 	}
2454 	xdp_redirect_fds->num = i;
2455 
2456 	/* enqueue all the frame descriptors */
2457 	enqueued = dpaa2_eth_xdp_flush(priv, fq, xdp_redirect_fds);
2458 
2459 	/* update statistics */
2460 	percpu_stats->tx_packets += enqueued;
2461 	for (i = 0; i < enqueued; i++)
2462 		percpu_stats->tx_bytes += dpaa2_fd_get_len(&fds[i]);
2463 
2464 	return enqueued;
2465 }
2466 
2467 static int update_xps(struct dpaa2_eth_priv *priv)
2468 {
2469 	struct net_device *net_dev = priv->net_dev;
2470 	struct cpumask xps_mask;
2471 	struct dpaa2_eth_fq *fq;
2472 	int i, num_queues, netdev_queues;
2473 	int err = 0;
2474 
2475 	num_queues = dpaa2_eth_queue_count(priv);
2476 	netdev_queues = (net_dev->num_tc ? : 1) * num_queues;
2477 
2478 	/* The first <num_queues> entries in priv->fq array are Tx/Tx conf
2479 	 * queues, so only process those
2480 	 */
2481 	for (i = 0; i < netdev_queues; i++) {
2482 		fq = &priv->fq[i % num_queues];
2483 
2484 		cpumask_clear(&xps_mask);
2485 		cpumask_set_cpu(fq->target_cpu, &xps_mask);
2486 
2487 		err = netif_set_xps_queue(net_dev, &xps_mask, i);
2488 		if (err) {
2489 			netdev_warn_once(net_dev, "Error setting XPS queue\n");
2490 			break;
2491 		}
2492 	}
2493 
2494 	return err;
2495 }
2496 
2497 static int dpaa2_eth_setup_mqprio(struct net_device *net_dev,
2498 				  struct tc_mqprio_qopt *mqprio)
2499 {
2500 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2501 	u8 num_tc, num_queues;
2502 	int i;
2503 
2504 	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
2505 	num_queues = dpaa2_eth_queue_count(priv);
2506 	num_tc = mqprio->num_tc;
2507 
2508 	if (num_tc == net_dev->num_tc)
2509 		return 0;
2510 
2511 	if (num_tc  > dpaa2_eth_tc_count(priv)) {
2512 		netdev_err(net_dev, "Max %d traffic classes supported\n",
2513 			   dpaa2_eth_tc_count(priv));
2514 		return -EOPNOTSUPP;
2515 	}
2516 
2517 	if (!num_tc) {
2518 		netdev_reset_tc(net_dev);
2519 		netif_set_real_num_tx_queues(net_dev, num_queues);
2520 		goto out;
2521 	}
2522 
2523 	netdev_set_num_tc(net_dev, num_tc);
2524 	netif_set_real_num_tx_queues(net_dev, num_tc * num_queues);
2525 
2526 	for (i = 0; i < num_tc; i++)
2527 		netdev_set_tc_queue(net_dev, i, num_queues, i * num_queues);
2528 
2529 out:
2530 	update_xps(priv);
2531 
2532 	return 0;
2533 }
2534 
2535 #define bps_to_mbits(rate) (div_u64((rate), 1000000) * 8)
2536 
2537 static int dpaa2_eth_setup_tbf(struct net_device *net_dev, struct tc_tbf_qopt_offload *p)
2538 {
2539 	struct tc_tbf_qopt_offload_replace_params *cfg = &p->replace_params;
2540 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2541 	struct dpni_tx_shaping_cfg tx_cr_shaper = { 0 };
2542 	struct dpni_tx_shaping_cfg tx_er_shaper = { 0 };
2543 	int err;
2544 
2545 	if (p->command == TC_TBF_STATS)
2546 		return -EOPNOTSUPP;
2547 
2548 	/* Only per port Tx shaping */
2549 	if (p->parent != TC_H_ROOT)
2550 		return -EOPNOTSUPP;
2551 
2552 	if (p->command == TC_TBF_REPLACE) {
2553 		if (cfg->max_size > DPAA2_ETH_MAX_BURST_SIZE) {
2554 			netdev_err(net_dev, "burst size cannot be greater than %d\n",
2555 				   DPAA2_ETH_MAX_BURST_SIZE);
2556 			return -EINVAL;
2557 		}
2558 
2559 		tx_cr_shaper.max_burst_size = cfg->max_size;
2560 		/* The TBF interface is in bytes/s, whereas DPAA2 expects the
2561 		 * rate in Mbits/s
2562 		 */
2563 		tx_cr_shaper.rate_limit = bps_to_mbits(cfg->rate.rate_bytes_ps);
2564 	}
2565 
2566 	err = dpni_set_tx_shaping(priv->mc_io, 0, priv->mc_token, &tx_cr_shaper,
2567 				  &tx_er_shaper, 0);
2568 	if (err) {
2569 		netdev_err(net_dev, "dpni_set_tx_shaping() = %d\n", err);
2570 		return err;
2571 	}
2572 
2573 	return 0;
2574 }
2575 
2576 static int dpaa2_eth_setup_tc(struct net_device *net_dev,
2577 			      enum tc_setup_type type, void *type_data)
2578 {
2579 	switch (type) {
2580 	case TC_SETUP_QDISC_MQPRIO:
2581 		return dpaa2_eth_setup_mqprio(net_dev, type_data);
2582 	case TC_SETUP_QDISC_TBF:
2583 		return dpaa2_eth_setup_tbf(net_dev, type_data);
2584 	default:
2585 		return -EOPNOTSUPP;
2586 	}
2587 }
2588 
2589 static const struct net_device_ops dpaa2_eth_ops = {
2590 	.ndo_open = dpaa2_eth_open,
2591 	.ndo_start_xmit = dpaa2_eth_tx,
2592 	.ndo_stop = dpaa2_eth_stop,
2593 	.ndo_set_mac_address = dpaa2_eth_set_addr,
2594 	.ndo_get_stats64 = dpaa2_eth_get_stats,
2595 	.ndo_set_rx_mode = dpaa2_eth_set_rx_mode,
2596 	.ndo_set_features = dpaa2_eth_set_features,
2597 	.ndo_eth_ioctl = dpaa2_eth_ioctl,
2598 	.ndo_change_mtu = dpaa2_eth_change_mtu,
2599 	.ndo_bpf = dpaa2_eth_xdp,
2600 	.ndo_xdp_xmit = dpaa2_eth_xdp_xmit,
2601 	.ndo_setup_tc = dpaa2_eth_setup_tc,
2602 	.ndo_vlan_rx_add_vid = dpaa2_eth_rx_add_vid,
2603 	.ndo_vlan_rx_kill_vid = dpaa2_eth_rx_kill_vid
2604 };
2605 
2606 static void dpaa2_eth_cdan_cb(struct dpaa2_io_notification_ctx *ctx)
2607 {
2608 	struct dpaa2_eth_channel *ch;
2609 
2610 	ch = container_of(ctx, struct dpaa2_eth_channel, nctx);
2611 
2612 	/* Update NAPI statistics */
2613 	ch->stats.cdan++;
2614 
2615 	napi_schedule(&ch->napi);
2616 }
2617 
2618 /* Allocate and configure a DPCON object */
2619 static struct fsl_mc_device *dpaa2_eth_setup_dpcon(struct dpaa2_eth_priv *priv)
2620 {
2621 	struct fsl_mc_device *dpcon;
2622 	struct device *dev = priv->net_dev->dev.parent;
2623 	int err;
2624 
2625 	err = fsl_mc_object_allocate(to_fsl_mc_device(dev),
2626 				     FSL_MC_POOL_DPCON, &dpcon);
2627 	if (err) {
2628 		if (err == -ENXIO)
2629 			err = -EPROBE_DEFER;
2630 		else
2631 			dev_info(dev, "Not enough DPCONs, will go on as-is\n");
2632 		return ERR_PTR(err);
2633 	}
2634 
2635 	err = dpcon_open(priv->mc_io, 0, dpcon->obj_desc.id, &dpcon->mc_handle);
2636 	if (err) {
2637 		dev_err(dev, "dpcon_open() failed\n");
2638 		goto free;
2639 	}
2640 
2641 	err = dpcon_reset(priv->mc_io, 0, dpcon->mc_handle);
2642 	if (err) {
2643 		dev_err(dev, "dpcon_reset() failed\n");
2644 		goto close;
2645 	}
2646 
2647 	err = dpcon_enable(priv->mc_io, 0, dpcon->mc_handle);
2648 	if (err) {
2649 		dev_err(dev, "dpcon_enable() failed\n");
2650 		goto close;
2651 	}
2652 
2653 	return dpcon;
2654 
2655 close:
2656 	dpcon_close(priv->mc_io, 0, dpcon->mc_handle);
2657 free:
2658 	fsl_mc_object_free(dpcon);
2659 
2660 	return ERR_PTR(err);
2661 }
2662 
2663 static void dpaa2_eth_free_dpcon(struct dpaa2_eth_priv *priv,
2664 				 struct fsl_mc_device *dpcon)
2665 {
2666 	dpcon_disable(priv->mc_io, 0, dpcon->mc_handle);
2667 	dpcon_close(priv->mc_io, 0, dpcon->mc_handle);
2668 	fsl_mc_object_free(dpcon);
2669 }
2670 
2671 static struct dpaa2_eth_channel *dpaa2_eth_alloc_channel(struct dpaa2_eth_priv *priv)
2672 {
2673 	struct dpaa2_eth_channel *channel;
2674 	struct dpcon_attr attr;
2675 	struct device *dev = priv->net_dev->dev.parent;
2676 	int err;
2677 
2678 	channel = kzalloc(sizeof(*channel), GFP_KERNEL);
2679 	if (!channel)
2680 		return NULL;
2681 
2682 	channel->dpcon = dpaa2_eth_setup_dpcon(priv);
2683 	if (IS_ERR(channel->dpcon)) {
2684 		err = PTR_ERR(channel->dpcon);
2685 		goto err_setup;
2686 	}
2687 
2688 	err = dpcon_get_attributes(priv->mc_io, 0, channel->dpcon->mc_handle,
2689 				   &attr);
2690 	if (err) {
2691 		dev_err(dev, "dpcon_get_attributes() failed\n");
2692 		goto err_get_attr;
2693 	}
2694 
2695 	channel->dpcon_id = attr.id;
2696 	channel->ch_id = attr.qbman_ch_id;
2697 	channel->priv = priv;
2698 
2699 	return channel;
2700 
2701 err_get_attr:
2702 	dpaa2_eth_free_dpcon(priv, channel->dpcon);
2703 err_setup:
2704 	kfree(channel);
2705 	return ERR_PTR(err);
2706 }
2707 
2708 static void dpaa2_eth_free_channel(struct dpaa2_eth_priv *priv,
2709 				   struct dpaa2_eth_channel *channel)
2710 {
2711 	dpaa2_eth_free_dpcon(priv, channel->dpcon);
2712 	kfree(channel);
2713 }
2714 
2715 /* DPIO setup: allocate and configure QBMan channels, setup core affinity
2716  * and register data availability notifications
2717  */
2718 static int dpaa2_eth_setup_dpio(struct dpaa2_eth_priv *priv)
2719 {
2720 	struct dpaa2_io_notification_ctx *nctx;
2721 	struct dpaa2_eth_channel *channel;
2722 	struct dpcon_notification_cfg dpcon_notif_cfg;
2723 	struct device *dev = priv->net_dev->dev.parent;
2724 	int i, err;
2725 
2726 	/* We want the ability to spread ingress traffic (RX, TX conf) to as
2727 	 * many cores as possible, so we need one channel for each core
2728 	 * (unless there's fewer queues than cores, in which case the extra
2729 	 * channels would be wasted).
2730 	 * Allocate one channel per core and register it to the core's
2731 	 * affine DPIO. If not enough channels are available for all cores
2732 	 * or if some cores don't have an affine DPIO, there will be no
2733 	 * ingress frame processing on those cores.
2734 	 */
2735 	cpumask_clear(&priv->dpio_cpumask);
2736 	for_each_online_cpu(i) {
2737 		/* Try to allocate a channel */
2738 		channel = dpaa2_eth_alloc_channel(priv);
2739 		if (IS_ERR_OR_NULL(channel)) {
2740 			err = PTR_ERR_OR_ZERO(channel);
2741 			if (err != -EPROBE_DEFER)
2742 				dev_info(dev,
2743 					 "No affine channel for cpu %d and above\n", i);
2744 			goto err_alloc_ch;
2745 		}
2746 
2747 		priv->channel[priv->num_channels] = channel;
2748 
2749 		nctx = &channel->nctx;
2750 		nctx->is_cdan = 1;
2751 		nctx->cb = dpaa2_eth_cdan_cb;
2752 		nctx->id = channel->ch_id;
2753 		nctx->desired_cpu = i;
2754 
2755 		/* Register the new context */
2756 		channel->dpio = dpaa2_io_service_select(i);
2757 		err = dpaa2_io_service_register(channel->dpio, nctx, dev);
2758 		if (err) {
2759 			dev_dbg(dev, "No affine DPIO for cpu %d\n", i);
2760 			/* If no affine DPIO for this core, there's probably
2761 			 * none available for next cores either. Signal we want
2762 			 * to retry later, in case the DPIO devices weren't
2763 			 * probed yet.
2764 			 */
2765 			err = -EPROBE_DEFER;
2766 			goto err_service_reg;
2767 		}
2768 
2769 		/* Register DPCON notification with MC */
2770 		dpcon_notif_cfg.dpio_id = nctx->dpio_id;
2771 		dpcon_notif_cfg.priority = 0;
2772 		dpcon_notif_cfg.user_ctx = nctx->qman64;
2773 		err = dpcon_set_notification(priv->mc_io, 0,
2774 					     channel->dpcon->mc_handle,
2775 					     &dpcon_notif_cfg);
2776 		if (err) {
2777 			dev_err(dev, "dpcon_set_notification failed()\n");
2778 			goto err_set_cdan;
2779 		}
2780 
2781 		/* If we managed to allocate a channel and also found an affine
2782 		 * DPIO for this core, add it to the final mask
2783 		 */
2784 		cpumask_set_cpu(i, &priv->dpio_cpumask);
2785 		priv->num_channels++;
2786 
2787 		/* Stop if we already have enough channels to accommodate all
2788 		 * RX and TX conf queues
2789 		 */
2790 		if (priv->num_channels == priv->dpni_attrs.num_queues)
2791 			break;
2792 	}
2793 
2794 	return 0;
2795 
2796 err_set_cdan:
2797 	dpaa2_io_service_deregister(channel->dpio, nctx, dev);
2798 err_service_reg:
2799 	dpaa2_eth_free_channel(priv, channel);
2800 err_alloc_ch:
2801 	if (err == -EPROBE_DEFER) {
2802 		for (i = 0; i < priv->num_channels; i++) {
2803 			channel = priv->channel[i];
2804 			nctx = &channel->nctx;
2805 			dpaa2_io_service_deregister(channel->dpio, nctx, dev);
2806 			dpaa2_eth_free_channel(priv, channel);
2807 		}
2808 		priv->num_channels = 0;
2809 		return err;
2810 	}
2811 
2812 	if (cpumask_empty(&priv->dpio_cpumask)) {
2813 		dev_err(dev, "No cpu with an affine DPIO/DPCON\n");
2814 		return -ENODEV;
2815 	}
2816 
2817 	dev_info(dev, "Cores %*pbl available for processing ingress traffic\n",
2818 		 cpumask_pr_args(&priv->dpio_cpumask));
2819 
2820 	return 0;
2821 }
2822 
2823 static void dpaa2_eth_free_dpio(struct dpaa2_eth_priv *priv)
2824 {
2825 	struct device *dev = priv->net_dev->dev.parent;
2826 	struct dpaa2_eth_channel *ch;
2827 	int i;
2828 
2829 	/* deregister CDAN notifications and free channels */
2830 	for (i = 0; i < priv->num_channels; i++) {
2831 		ch = priv->channel[i];
2832 		dpaa2_io_service_deregister(ch->dpio, &ch->nctx, dev);
2833 		dpaa2_eth_free_channel(priv, ch);
2834 	}
2835 }
2836 
2837 static struct dpaa2_eth_channel *dpaa2_eth_get_affine_channel(struct dpaa2_eth_priv *priv,
2838 							      int cpu)
2839 {
2840 	struct device *dev = priv->net_dev->dev.parent;
2841 	int i;
2842 
2843 	for (i = 0; i < priv->num_channels; i++)
2844 		if (priv->channel[i]->nctx.desired_cpu == cpu)
2845 			return priv->channel[i];
2846 
2847 	/* We should never get here. Issue a warning and return
2848 	 * the first channel, because it's still better than nothing
2849 	 */
2850 	dev_warn(dev, "No affine channel found for cpu %d\n", cpu);
2851 
2852 	return priv->channel[0];
2853 }
2854 
2855 static void dpaa2_eth_set_fq_affinity(struct dpaa2_eth_priv *priv)
2856 {
2857 	struct device *dev = priv->net_dev->dev.parent;
2858 	struct dpaa2_eth_fq *fq;
2859 	int rx_cpu, txc_cpu;
2860 	int i;
2861 
2862 	/* For each FQ, pick one channel/CPU to deliver frames to.
2863 	 * This may well change at runtime, either through irqbalance or
2864 	 * through direct user intervention.
2865 	 */
2866 	rx_cpu = txc_cpu = cpumask_first(&priv->dpio_cpumask);
2867 
2868 	for (i = 0; i < priv->num_fqs; i++) {
2869 		fq = &priv->fq[i];
2870 		switch (fq->type) {
2871 		case DPAA2_RX_FQ:
2872 		case DPAA2_RX_ERR_FQ:
2873 			fq->target_cpu = rx_cpu;
2874 			rx_cpu = cpumask_next(rx_cpu, &priv->dpio_cpumask);
2875 			if (rx_cpu >= nr_cpu_ids)
2876 				rx_cpu = cpumask_first(&priv->dpio_cpumask);
2877 			break;
2878 		case DPAA2_TX_CONF_FQ:
2879 			fq->target_cpu = txc_cpu;
2880 			txc_cpu = cpumask_next(txc_cpu, &priv->dpio_cpumask);
2881 			if (txc_cpu >= nr_cpu_ids)
2882 				txc_cpu = cpumask_first(&priv->dpio_cpumask);
2883 			break;
2884 		default:
2885 			dev_err(dev, "Unknown FQ type: %d\n", fq->type);
2886 		}
2887 		fq->channel = dpaa2_eth_get_affine_channel(priv, fq->target_cpu);
2888 	}
2889 
2890 	update_xps(priv);
2891 }
2892 
2893 static void dpaa2_eth_setup_fqs(struct dpaa2_eth_priv *priv)
2894 {
2895 	int i, j;
2896 
2897 	/* We have one TxConf FQ per Tx flow.
2898 	 * The number of Tx and Rx queues is the same.
2899 	 * Tx queues come first in the fq array.
2900 	 */
2901 	for (i = 0; i < dpaa2_eth_queue_count(priv); i++) {
2902 		priv->fq[priv->num_fqs].type = DPAA2_TX_CONF_FQ;
2903 		priv->fq[priv->num_fqs].consume = dpaa2_eth_tx_conf;
2904 		priv->fq[priv->num_fqs++].flowid = (u16)i;
2905 	}
2906 
2907 	for (j = 0; j < dpaa2_eth_tc_count(priv); j++) {
2908 		for (i = 0; i < dpaa2_eth_queue_count(priv); i++) {
2909 			priv->fq[priv->num_fqs].type = DPAA2_RX_FQ;
2910 			priv->fq[priv->num_fqs].consume = dpaa2_eth_rx;
2911 			priv->fq[priv->num_fqs].tc = (u8)j;
2912 			priv->fq[priv->num_fqs++].flowid = (u16)i;
2913 		}
2914 	}
2915 
2916 	/* We have exactly one Rx error queue per DPNI */
2917 	priv->fq[priv->num_fqs].type = DPAA2_RX_ERR_FQ;
2918 	priv->fq[priv->num_fqs++].consume = dpaa2_eth_rx_err;
2919 
2920 	/* For each FQ, decide on which core to process incoming frames */
2921 	dpaa2_eth_set_fq_affinity(priv);
2922 }
2923 
2924 /* Allocate and configure one buffer pool for each interface */
2925 static int dpaa2_eth_setup_dpbp(struct dpaa2_eth_priv *priv)
2926 {
2927 	int err;
2928 	struct fsl_mc_device *dpbp_dev;
2929 	struct device *dev = priv->net_dev->dev.parent;
2930 	struct dpbp_attr dpbp_attrs;
2931 
2932 	err = fsl_mc_object_allocate(to_fsl_mc_device(dev), FSL_MC_POOL_DPBP,
2933 				     &dpbp_dev);
2934 	if (err) {
2935 		if (err == -ENXIO)
2936 			err = -EPROBE_DEFER;
2937 		else
2938 			dev_err(dev, "DPBP device allocation failed\n");
2939 		return err;
2940 	}
2941 
2942 	priv->dpbp_dev = dpbp_dev;
2943 
2944 	err = dpbp_open(priv->mc_io, 0, priv->dpbp_dev->obj_desc.id,
2945 			&dpbp_dev->mc_handle);
2946 	if (err) {
2947 		dev_err(dev, "dpbp_open() failed\n");
2948 		goto err_open;
2949 	}
2950 
2951 	err = dpbp_reset(priv->mc_io, 0, dpbp_dev->mc_handle);
2952 	if (err) {
2953 		dev_err(dev, "dpbp_reset() failed\n");
2954 		goto err_reset;
2955 	}
2956 
2957 	err = dpbp_enable(priv->mc_io, 0, dpbp_dev->mc_handle);
2958 	if (err) {
2959 		dev_err(dev, "dpbp_enable() failed\n");
2960 		goto err_enable;
2961 	}
2962 
2963 	err = dpbp_get_attributes(priv->mc_io, 0, dpbp_dev->mc_handle,
2964 				  &dpbp_attrs);
2965 	if (err) {
2966 		dev_err(dev, "dpbp_get_attributes() failed\n");
2967 		goto err_get_attr;
2968 	}
2969 	priv->bpid = dpbp_attrs.bpid;
2970 
2971 	return 0;
2972 
2973 err_get_attr:
2974 	dpbp_disable(priv->mc_io, 0, dpbp_dev->mc_handle);
2975 err_enable:
2976 err_reset:
2977 	dpbp_close(priv->mc_io, 0, dpbp_dev->mc_handle);
2978 err_open:
2979 	fsl_mc_object_free(dpbp_dev);
2980 
2981 	return err;
2982 }
2983 
2984 static void dpaa2_eth_free_dpbp(struct dpaa2_eth_priv *priv)
2985 {
2986 	dpaa2_eth_drain_pool(priv);
2987 	dpbp_disable(priv->mc_io, 0, priv->dpbp_dev->mc_handle);
2988 	dpbp_close(priv->mc_io, 0, priv->dpbp_dev->mc_handle);
2989 	fsl_mc_object_free(priv->dpbp_dev);
2990 }
2991 
2992 static int dpaa2_eth_set_buffer_layout(struct dpaa2_eth_priv *priv)
2993 {
2994 	struct device *dev = priv->net_dev->dev.parent;
2995 	struct dpni_buffer_layout buf_layout = {0};
2996 	u16 rx_buf_align;
2997 	int err;
2998 
2999 	/* We need to check for WRIOP version 1.0.0, but depending on the MC
3000 	 * version, this number is not always provided correctly on rev1.
3001 	 * We need to check for both alternatives in this situation.
3002 	 */
3003 	if (priv->dpni_attrs.wriop_version == DPAA2_WRIOP_VERSION(0, 0, 0) ||
3004 	    priv->dpni_attrs.wriop_version == DPAA2_WRIOP_VERSION(1, 0, 0))
3005 		rx_buf_align = DPAA2_ETH_RX_BUF_ALIGN_REV1;
3006 	else
3007 		rx_buf_align = DPAA2_ETH_RX_BUF_ALIGN;
3008 
3009 	/* We need to ensure that the buffer size seen by WRIOP is a multiple
3010 	 * of 64 or 256 bytes depending on the WRIOP version.
3011 	 */
3012 	priv->rx_buf_size = ALIGN_DOWN(DPAA2_ETH_RX_BUF_SIZE, rx_buf_align);
3013 
3014 	/* tx buffer */
3015 	buf_layout.private_data_size = DPAA2_ETH_SWA_SIZE;
3016 	buf_layout.pass_timestamp = true;
3017 	buf_layout.pass_frame_status = true;
3018 	buf_layout.options = DPNI_BUF_LAYOUT_OPT_PRIVATE_DATA_SIZE |
3019 			     DPNI_BUF_LAYOUT_OPT_TIMESTAMP |
3020 			     DPNI_BUF_LAYOUT_OPT_FRAME_STATUS;
3021 	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
3022 				     DPNI_QUEUE_TX, &buf_layout);
3023 	if (err) {
3024 		dev_err(dev, "dpni_set_buffer_layout(TX) failed\n");
3025 		return err;
3026 	}
3027 
3028 	/* tx-confirm buffer */
3029 	buf_layout.options = DPNI_BUF_LAYOUT_OPT_TIMESTAMP |
3030 			     DPNI_BUF_LAYOUT_OPT_FRAME_STATUS;
3031 	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
3032 				     DPNI_QUEUE_TX_CONFIRM, &buf_layout);
3033 	if (err) {
3034 		dev_err(dev, "dpni_set_buffer_layout(TX_CONF) failed\n");
3035 		return err;
3036 	}
3037 
3038 	/* Now that we've set our tx buffer layout, retrieve the minimum
3039 	 * required tx data offset.
3040 	 */
3041 	err = dpni_get_tx_data_offset(priv->mc_io, 0, priv->mc_token,
3042 				      &priv->tx_data_offset);
3043 	if (err) {
3044 		dev_err(dev, "dpni_get_tx_data_offset() failed\n");
3045 		return err;
3046 	}
3047 
3048 	if ((priv->tx_data_offset % 64) != 0)
3049 		dev_warn(dev, "Tx data offset (%d) not a multiple of 64B\n",
3050 			 priv->tx_data_offset);
3051 
3052 	/* rx buffer */
3053 	buf_layout.pass_frame_status = true;
3054 	buf_layout.pass_parser_result = true;
3055 	buf_layout.data_align = rx_buf_align;
3056 	buf_layout.data_head_room = dpaa2_eth_rx_head_room(priv);
3057 	buf_layout.private_data_size = 0;
3058 	buf_layout.options = DPNI_BUF_LAYOUT_OPT_PARSER_RESULT |
3059 			     DPNI_BUF_LAYOUT_OPT_FRAME_STATUS |
3060 			     DPNI_BUF_LAYOUT_OPT_DATA_ALIGN |
3061 			     DPNI_BUF_LAYOUT_OPT_DATA_HEAD_ROOM |
3062 			     DPNI_BUF_LAYOUT_OPT_TIMESTAMP;
3063 	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
3064 				     DPNI_QUEUE_RX, &buf_layout);
3065 	if (err) {
3066 		dev_err(dev, "dpni_set_buffer_layout(RX) failed\n");
3067 		return err;
3068 	}
3069 
3070 	return 0;
3071 }
3072 
3073 #define DPNI_ENQUEUE_FQID_VER_MAJOR	7
3074 #define DPNI_ENQUEUE_FQID_VER_MINOR	9
3075 
3076 static inline int dpaa2_eth_enqueue_qd(struct dpaa2_eth_priv *priv,
3077 				       struct dpaa2_eth_fq *fq,
3078 				       struct dpaa2_fd *fd, u8 prio,
3079 				       u32 num_frames __always_unused,
3080 				       int *frames_enqueued)
3081 {
3082 	int err;
3083 
3084 	err = dpaa2_io_service_enqueue_qd(fq->channel->dpio,
3085 					  priv->tx_qdid, prio,
3086 					  fq->tx_qdbin, fd);
3087 	if (!err && frames_enqueued)
3088 		*frames_enqueued = 1;
3089 	return err;
3090 }
3091 
3092 static inline int dpaa2_eth_enqueue_fq_multiple(struct dpaa2_eth_priv *priv,
3093 						struct dpaa2_eth_fq *fq,
3094 						struct dpaa2_fd *fd,
3095 						u8 prio, u32 num_frames,
3096 						int *frames_enqueued)
3097 {
3098 	int err;
3099 
3100 	err = dpaa2_io_service_enqueue_multiple_fq(fq->channel->dpio,
3101 						   fq->tx_fqid[prio],
3102 						   fd, num_frames);
3103 
3104 	if (err == 0)
3105 		return -EBUSY;
3106 
3107 	if (frames_enqueued)
3108 		*frames_enqueued = err;
3109 	return 0;
3110 }
3111 
3112 static void dpaa2_eth_set_enqueue_mode(struct dpaa2_eth_priv *priv)
3113 {
3114 	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_ENQUEUE_FQID_VER_MAJOR,
3115 				   DPNI_ENQUEUE_FQID_VER_MINOR) < 0)
3116 		priv->enqueue = dpaa2_eth_enqueue_qd;
3117 	else
3118 		priv->enqueue = dpaa2_eth_enqueue_fq_multiple;
3119 }
3120 
3121 static int dpaa2_eth_set_pause(struct dpaa2_eth_priv *priv)
3122 {
3123 	struct device *dev = priv->net_dev->dev.parent;
3124 	struct dpni_link_cfg link_cfg = {0};
3125 	int err;
3126 
3127 	/* Get the default link options so we don't override other flags */
3128 	err = dpni_get_link_cfg(priv->mc_io, 0, priv->mc_token, &link_cfg);
3129 	if (err) {
3130 		dev_err(dev, "dpni_get_link_cfg() failed\n");
3131 		return err;
3132 	}
3133 
3134 	/* By default, enable both Rx and Tx pause frames */
3135 	link_cfg.options |= DPNI_LINK_OPT_PAUSE;
3136 	link_cfg.options &= ~DPNI_LINK_OPT_ASYM_PAUSE;
3137 	err = dpni_set_link_cfg(priv->mc_io, 0, priv->mc_token, &link_cfg);
3138 	if (err) {
3139 		dev_err(dev, "dpni_set_link_cfg() failed\n");
3140 		return err;
3141 	}
3142 
3143 	priv->link_state.options = link_cfg.options;
3144 
3145 	return 0;
3146 }
3147 
3148 static void dpaa2_eth_update_tx_fqids(struct dpaa2_eth_priv *priv)
3149 {
3150 	struct dpni_queue_id qid = {0};
3151 	struct dpaa2_eth_fq *fq;
3152 	struct dpni_queue queue;
3153 	int i, j, err;
3154 
3155 	/* We only use Tx FQIDs for FQID-based enqueue, so check
3156 	 * if DPNI version supports it before updating FQIDs
3157 	 */
3158 	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_ENQUEUE_FQID_VER_MAJOR,
3159 				   DPNI_ENQUEUE_FQID_VER_MINOR) < 0)
3160 		return;
3161 
3162 	for (i = 0; i < priv->num_fqs; i++) {
3163 		fq = &priv->fq[i];
3164 		if (fq->type != DPAA2_TX_CONF_FQ)
3165 			continue;
3166 		for (j = 0; j < dpaa2_eth_tc_count(priv); j++) {
3167 			err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3168 					     DPNI_QUEUE_TX, j, fq->flowid,
3169 					     &queue, &qid);
3170 			if (err)
3171 				goto out_err;
3172 
3173 			fq->tx_fqid[j] = qid.fqid;
3174 			if (fq->tx_fqid[j] == 0)
3175 				goto out_err;
3176 		}
3177 	}
3178 
3179 	priv->enqueue = dpaa2_eth_enqueue_fq_multiple;
3180 
3181 	return;
3182 
3183 out_err:
3184 	netdev_info(priv->net_dev,
3185 		    "Error reading Tx FQID, fallback to QDID-based enqueue\n");
3186 	priv->enqueue = dpaa2_eth_enqueue_qd;
3187 }
3188 
3189 /* Configure ingress classification based on VLAN PCP */
3190 static int dpaa2_eth_set_vlan_qos(struct dpaa2_eth_priv *priv)
3191 {
3192 	struct device *dev = priv->net_dev->dev.parent;
3193 	struct dpkg_profile_cfg kg_cfg = {0};
3194 	struct dpni_qos_tbl_cfg qos_cfg = {0};
3195 	struct dpni_rule_cfg key_params;
3196 	void *dma_mem, *key, *mask;
3197 	u8 key_size = 2;	/* VLAN TCI field */
3198 	int i, pcp, err;
3199 
3200 	/* VLAN-based classification only makes sense if we have multiple
3201 	 * traffic classes.
3202 	 * Also, we need to extract just the 3-bit PCP field from the VLAN
3203 	 * header and we can only do that by using a mask
3204 	 */
3205 	if (dpaa2_eth_tc_count(priv) == 1 || !dpaa2_eth_fs_mask_enabled(priv)) {
3206 		dev_dbg(dev, "VLAN-based QoS classification not supported\n");
3207 		return -EOPNOTSUPP;
3208 	}
3209 
3210 	dma_mem = kzalloc(DPAA2_CLASSIFIER_DMA_SIZE, GFP_KERNEL);
3211 	if (!dma_mem)
3212 		return -ENOMEM;
3213 
3214 	kg_cfg.num_extracts = 1;
3215 	kg_cfg.extracts[0].type = DPKG_EXTRACT_FROM_HDR;
3216 	kg_cfg.extracts[0].extract.from_hdr.prot = NET_PROT_VLAN;
3217 	kg_cfg.extracts[0].extract.from_hdr.type = DPKG_FULL_FIELD;
3218 	kg_cfg.extracts[0].extract.from_hdr.field = NH_FLD_VLAN_TCI;
3219 
3220 	err = dpni_prepare_key_cfg(&kg_cfg, dma_mem);
3221 	if (err) {
3222 		dev_err(dev, "dpni_prepare_key_cfg failed\n");
3223 		goto out_free_tbl;
3224 	}
3225 
3226 	/* set QoS table */
3227 	qos_cfg.default_tc = 0;
3228 	qos_cfg.discard_on_miss = 0;
3229 	qos_cfg.key_cfg_iova = dma_map_single(dev, dma_mem,
3230 					      DPAA2_CLASSIFIER_DMA_SIZE,
3231 					      DMA_TO_DEVICE);
3232 	if (dma_mapping_error(dev, qos_cfg.key_cfg_iova)) {
3233 		dev_err(dev, "QoS table DMA mapping failed\n");
3234 		err = -ENOMEM;
3235 		goto out_free_tbl;
3236 	}
3237 
3238 	err = dpni_set_qos_table(priv->mc_io, 0, priv->mc_token, &qos_cfg);
3239 	if (err) {
3240 		dev_err(dev, "dpni_set_qos_table failed\n");
3241 		goto out_unmap_tbl;
3242 	}
3243 
3244 	/* Add QoS table entries */
3245 	key = kzalloc(key_size * 2, GFP_KERNEL);
3246 	if (!key) {
3247 		err = -ENOMEM;
3248 		goto out_unmap_tbl;
3249 	}
3250 	mask = key + key_size;
3251 	*(__be16 *)mask = cpu_to_be16(VLAN_PRIO_MASK);
3252 
3253 	key_params.key_iova = dma_map_single(dev, key, key_size * 2,
3254 					     DMA_TO_DEVICE);
3255 	if (dma_mapping_error(dev, key_params.key_iova)) {
3256 		dev_err(dev, "Qos table entry DMA mapping failed\n");
3257 		err = -ENOMEM;
3258 		goto out_free_key;
3259 	}
3260 
3261 	key_params.mask_iova = key_params.key_iova + key_size;
3262 	key_params.key_size = key_size;
3263 
3264 	/* We add rules for PCP-based distribution starting with highest
3265 	 * priority (VLAN PCP = 7). If this DPNI doesn't have enough traffic
3266 	 * classes to accommodate all priority levels, the lowest ones end up
3267 	 * on TC 0 which was configured as default
3268 	 */
3269 	for (i = dpaa2_eth_tc_count(priv) - 1, pcp = 7; i >= 0; i--, pcp--) {
3270 		*(__be16 *)key = cpu_to_be16(pcp << VLAN_PRIO_SHIFT);
3271 		dma_sync_single_for_device(dev, key_params.key_iova,
3272 					   key_size * 2, DMA_TO_DEVICE);
3273 
3274 		err = dpni_add_qos_entry(priv->mc_io, 0, priv->mc_token,
3275 					 &key_params, i, i);
3276 		if (err) {
3277 			dev_err(dev, "dpni_add_qos_entry failed\n");
3278 			dpni_clear_qos_table(priv->mc_io, 0, priv->mc_token);
3279 			goto out_unmap_key;
3280 		}
3281 	}
3282 
3283 	priv->vlan_cls_enabled = true;
3284 
3285 	/* Table and key memory is not persistent, clean everything up after
3286 	 * configuration is finished
3287 	 */
3288 out_unmap_key:
3289 	dma_unmap_single(dev, key_params.key_iova, key_size * 2, DMA_TO_DEVICE);
3290 out_free_key:
3291 	kfree(key);
3292 out_unmap_tbl:
3293 	dma_unmap_single(dev, qos_cfg.key_cfg_iova, DPAA2_CLASSIFIER_DMA_SIZE,
3294 			 DMA_TO_DEVICE);
3295 out_free_tbl:
3296 	kfree(dma_mem);
3297 
3298 	return err;
3299 }
3300 
3301 /* Configure the DPNI object this interface is associated with */
3302 static int dpaa2_eth_setup_dpni(struct fsl_mc_device *ls_dev)
3303 {
3304 	struct device *dev = &ls_dev->dev;
3305 	struct dpaa2_eth_priv *priv;
3306 	struct net_device *net_dev;
3307 	int err;
3308 
3309 	net_dev = dev_get_drvdata(dev);
3310 	priv = netdev_priv(net_dev);
3311 
3312 	/* get a handle for the DPNI object */
3313 	err = dpni_open(priv->mc_io, 0, ls_dev->obj_desc.id, &priv->mc_token);
3314 	if (err) {
3315 		dev_err(dev, "dpni_open() failed\n");
3316 		return err;
3317 	}
3318 
3319 	/* Check if we can work with this DPNI object */
3320 	err = dpni_get_api_version(priv->mc_io, 0, &priv->dpni_ver_major,
3321 				   &priv->dpni_ver_minor);
3322 	if (err) {
3323 		dev_err(dev, "dpni_get_api_version() failed\n");
3324 		goto close;
3325 	}
3326 	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_VER_MAJOR, DPNI_VER_MINOR) < 0) {
3327 		dev_err(dev, "DPNI version %u.%u not supported, need >= %u.%u\n",
3328 			priv->dpni_ver_major, priv->dpni_ver_minor,
3329 			DPNI_VER_MAJOR, DPNI_VER_MINOR);
3330 		err = -ENOTSUPP;
3331 		goto close;
3332 	}
3333 
3334 	ls_dev->mc_io = priv->mc_io;
3335 	ls_dev->mc_handle = priv->mc_token;
3336 
3337 	err = dpni_reset(priv->mc_io, 0, priv->mc_token);
3338 	if (err) {
3339 		dev_err(dev, "dpni_reset() failed\n");
3340 		goto close;
3341 	}
3342 
3343 	err = dpni_get_attributes(priv->mc_io, 0, priv->mc_token,
3344 				  &priv->dpni_attrs);
3345 	if (err) {
3346 		dev_err(dev, "dpni_get_attributes() failed (err=%d)\n", err);
3347 		goto close;
3348 	}
3349 
3350 	err = dpaa2_eth_set_buffer_layout(priv);
3351 	if (err)
3352 		goto close;
3353 
3354 	dpaa2_eth_set_enqueue_mode(priv);
3355 
3356 	/* Enable pause frame support */
3357 	if (dpaa2_eth_has_pause_support(priv)) {
3358 		err = dpaa2_eth_set_pause(priv);
3359 		if (err)
3360 			goto close;
3361 	}
3362 
3363 	err = dpaa2_eth_set_vlan_qos(priv);
3364 	if (err && err != -EOPNOTSUPP)
3365 		goto close;
3366 
3367 	priv->cls_rules = devm_kcalloc(dev, dpaa2_eth_fs_count(priv),
3368 				       sizeof(struct dpaa2_eth_cls_rule),
3369 				       GFP_KERNEL);
3370 	if (!priv->cls_rules) {
3371 		err = -ENOMEM;
3372 		goto close;
3373 	}
3374 
3375 	return 0;
3376 
3377 close:
3378 	dpni_close(priv->mc_io, 0, priv->mc_token);
3379 
3380 	return err;
3381 }
3382 
3383 static void dpaa2_eth_free_dpni(struct dpaa2_eth_priv *priv)
3384 {
3385 	int err;
3386 
3387 	err = dpni_reset(priv->mc_io, 0, priv->mc_token);
3388 	if (err)
3389 		netdev_warn(priv->net_dev, "dpni_reset() failed (err %d)\n",
3390 			    err);
3391 
3392 	dpni_close(priv->mc_io, 0, priv->mc_token);
3393 }
3394 
3395 static int dpaa2_eth_setup_rx_flow(struct dpaa2_eth_priv *priv,
3396 				   struct dpaa2_eth_fq *fq)
3397 {
3398 	struct device *dev = priv->net_dev->dev.parent;
3399 	struct dpni_queue queue;
3400 	struct dpni_queue_id qid;
3401 	int err;
3402 
3403 	err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3404 			     DPNI_QUEUE_RX, fq->tc, fq->flowid, &queue, &qid);
3405 	if (err) {
3406 		dev_err(dev, "dpni_get_queue(RX) failed\n");
3407 		return err;
3408 	}
3409 
3410 	fq->fqid = qid.fqid;
3411 
3412 	queue.destination.id = fq->channel->dpcon_id;
3413 	queue.destination.type = DPNI_DEST_DPCON;
3414 	queue.destination.priority = 1;
3415 	queue.user_context = (u64)(uintptr_t)fq;
3416 	err = dpni_set_queue(priv->mc_io, 0, priv->mc_token,
3417 			     DPNI_QUEUE_RX, fq->tc, fq->flowid,
3418 			     DPNI_QUEUE_OPT_USER_CTX | DPNI_QUEUE_OPT_DEST,
3419 			     &queue);
3420 	if (err) {
3421 		dev_err(dev, "dpni_set_queue(RX) failed\n");
3422 		return err;
3423 	}
3424 
3425 	/* xdp_rxq setup */
3426 	/* only once for each channel */
3427 	if (fq->tc > 0)
3428 		return 0;
3429 
3430 	err = xdp_rxq_info_reg(&fq->channel->xdp_rxq, priv->net_dev,
3431 			       fq->flowid, 0);
3432 	if (err) {
3433 		dev_err(dev, "xdp_rxq_info_reg failed\n");
3434 		return err;
3435 	}
3436 
3437 	err = xdp_rxq_info_reg_mem_model(&fq->channel->xdp_rxq,
3438 					 MEM_TYPE_PAGE_ORDER0, NULL);
3439 	if (err) {
3440 		dev_err(dev, "xdp_rxq_info_reg_mem_model failed\n");
3441 		return err;
3442 	}
3443 
3444 	return 0;
3445 }
3446 
3447 static int dpaa2_eth_setup_tx_flow(struct dpaa2_eth_priv *priv,
3448 				   struct dpaa2_eth_fq *fq)
3449 {
3450 	struct device *dev = priv->net_dev->dev.parent;
3451 	struct dpni_queue queue;
3452 	struct dpni_queue_id qid;
3453 	int i, err;
3454 
3455 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3456 		err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3457 				     DPNI_QUEUE_TX, i, fq->flowid,
3458 				     &queue, &qid);
3459 		if (err) {
3460 			dev_err(dev, "dpni_get_queue(TX) failed\n");
3461 			return err;
3462 		}
3463 		fq->tx_fqid[i] = qid.fqid;
3464 	}
3465 
3466 	/* All Tx queues belonging to the same flowid have the same qdbin */
3467 	fq->tx_qdbin = qid.qdbin;
3468 
3469 	err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3470 			     DPNI_QUEUE_TX_CONFIRM, 0, fq->flowid,
3471 			     &queue, &qid);
3472 	if (err) {
3473 		dev_err(dev, "dpni_get_queue(TX_CONF) failed\n");
3474 		return err;
3475 	}
3476 
3477 	fq->fqid = qid.fqid;
3478 
3479 	queue.destination.id = fq->channel->dpcon_id;
3480 	queue.destination.type = DPNI_DEST_DPCON;
3481 	queue.destination.priority = 0;
3482 	queue.user_context = (u64)(uintptr_t)fq;
3483 	err = dpni_set_queue(priv->mc_io, 0, priv->mc_token,
3484 			     DPNI_QUEUE_TX_CONFIRM, 0, fq->flowid,
3485 			     DPNI_QUEUE_OPT_USER_CTX | DPNI_QUEUE_OPT_DEST,
3486 			     &queue);
3487 	if (err) {
3488 		dev_err(dev, "dpni_set_queue(TX_CONF) failed\n");
3489 		return err;
3490 	}
3491 
3492 	return 0;
3493 }
3494 
3495 static int setup_rx_err_flow(struct dpaa2_eth_priv *priv,
3496 			     struct dpaa2_eth_fq *fq)
3497 {
3498 	struct device *dev = priv->net_dev->dev.parent;
3499 	struct dpni_queue q = { { 0 } };
3500 	struct dpni_queue_id qid;
3501 	u8 q_opt = DPNI_QUEUE_OPT_USER_CTX | DPNI_QUEUE_OPT_DEST;
3502 	int err;
3503 
3504 	err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3505 			     DPNI_QUEUE_RX_ERR, 0, 0, &q, &qid);
3506 	if (err) {
3507 		dev_err(dev, "dpni_get_queue() failed (%d)\n", err);
3508 		return err;
3509 	}
3510 
3511 	fq->fqid = qid.fqid;
3512 
3513 	q.destination.id = fq->channel->dpcon_id;
3514 	q.destination.type = DPNI_DEST_DPCON;
3515 	q.destination.priority = 1;
3516 	q.user_context = (u64)(uintptr_t)fq;
3517 	err = dpni_set_queue(priv->mc_io, 0, priv->mc_token,
3518 			     DPNI_QUEUE_RX_ERR, 0, 0, q_opt, &q);
3519 	if (err) {
3520 		dev_err(dev, "dpni_set_queue() failed (%d)\n", err);
3521 		return err;
3522 	}
3523 
3524 	return 0;
3525 }
3526 
3527 /* Supported header fields for Rx hash distribution key */
3528 static const struct dpaa2_eth_dist_fields dist_fields[] = {
3529 	{
3530 		/* L2 header */
3531 		.rxnfc_field = RXH_L2DA,
3532 		.cls_prot = NET_PROT_ETH,
3533 		.cls_field = NH_FLD_ETH_DA,
3534 		.id = DPAA2_ETH_DIST_ETHDST,
3535 		.size = 6,
3536 	}, {
3537 		.cls_prot = NET_PROT_ETH,
3538 		.cls_field = NH_FLD_ETH_SA,
3539 		.id = DPAA2_ETH_DIST_ETHSRC,
3540 		.size = 6,
3541 	}, {
3542 		/* This is the last ethertype field parsed:
3543 		 * depending on frame format, it can be the MAC ethertype
3544 		 * or the VLAN etype.
3545 		 */
3546 		.cls_prot = NET_PROT_ETH,
3547 		.cls_field = NH_FLD_ETH_TYPE,
3548 		.id = DPAA2_ETH_DIST_ETHTYPE,
3549 		.size = 2,
3550 	}, {
3551 		/* VLAN header */
3552 		.rxnfc_field = RXH_VLAN,
3553 		.cls_prot = NET_PROT_VLAN,
3554 		.cls_field = NH_FLD_VLAN_TCI,
3555 		.id = DPAA2_ETH_DIST_VLAN,
3556 		.size = 2,
3557 	}, {
3558 		/* IP header */
3559 		.rxnfc_field = RXH_IP_SRC,
3560 		.cls_prot = NET_PROT_IP,
3561 		.cls_field = NH_FLD_IP_SRC,
3562 		.id = DPAA2_ETH_DIST_IPSRC,
3563 		.size = 4,
3564 	}, {
3565 		.rxnfc_field = RXH_IP_DST,
3566 		.cls_prot = NET_PROT_IP,
3567 		.cls_field = NH_FLD_IP_DST,
3568 		.id = DPAA2_ETH_DIST_IPDST,
3569 		.size = 4,
3570 	}, {
3571 		.rxnfc_field = RXH_L3_PROTO,
3572 		.cls_prot = NET_PROT_IP,
3573 		.cls_field = NH_FLD_IP_PROTO,
3574 		.id = DPAA2_ETH_DIST_IPPROTO,
3575 		.size = 1,
3576 	}, {
3577 		/* Using UDP ports, this is functionally equivalent to raw
3578 		 * byte pairs from L4 header.
3579 		 */
3580 		.rxnfc_field = RXH_L4_B_0_1,
3581 		.cls_prot = NET_PROT_UDP,
3582 		.cls_field = NH_FLD_UDP_PORT_SRC,
3583 		.id = DPAA2_ETH_DIST_L4SRC,
3584 		.size = 2,
3585 	}, {
3586 		.rxnfc_field = RXH_L4_B_2_3,
3587 		.cls_prot = NET_PROT_UDP,
3588 		.cls_field = NH_FLD_UDP_PORT_DST,
3589 		.id = DPAA2_ETH_DIST_L4DST,
3590 		.size = 2,
3591 	},
3592 };
3593 
3594 /* Configure the Rx hash key using the legacy API */
3595 static int dpaa2_eth_config_legacy_hash_key(struct dpaa2_eth_priv *priv, dma_addr_t key)
3596 {
3597 	struct device *dev = priv->net_dev->dev.parent;
3598 	struct dpni_rx_tc_dist_cfg dist_cfg;
3599 	int i, err = 0;
3600 
3601 	memset(&dist_cfg, 0, sizeof(dist_cfg));
3602 
3603 	dist_cfg.key_cfg_iova = key;
3604 	dist_cfg.dist_size = dpaa2_eth_queue_count(priv);
3605 	dist_cfg.dist_mode = DPNI_DIST_MODE_HASH;
3606 
3607 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3608 		err = dpni_set_rx_tc_dist(priv->mc_io, 0, priv->mc_token,
3609 					  i, &dist_cfg);
3610 		if (err) {
3611 			dev_err(dev, "dpni_set_rx_tc_dist failed\n");
3612 			break;
3613 		}
3614 	}
3615 
3616 	return err;
3617 }
3618 
3619 /* Configure the Rx hash key using the new API */
3620 static int dpaa2_eth_config_hash_key(struct dpaa2_eth_priv *priv, dma_addr_t key)
3621 {
3622 	struct device *dev = priv->net_dev->dev.parent;
3623 	struct dpni_rx_dist_cfg dist_cfg;
3624 	int i, err = 0;
3625 
3626 	memset(&dist_cfg, 0, sizeof(dist_cfg));
3627 
3628 	dist_cfg.key_cfg_iova = key;
3629 	dist_cfg.dist_size = dpaa2_eth_queue_count(priv);
3630 	dist_cfg.enable = 1;
3631 
3632 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3633 		dist_cfg.tc = i;
3634 		err = dpni_set_rx_hash_dist(priv->mc_io, 0, priv->mc_token,
3635 					    &dist_cfg);
3636 		if (err) {
3637 			dev_err(dev, "dpni_set_rx_hash_dist failed\n");
3638 			break;
3639 		}
3640 
3641 		/* If the flow steering / hashing key is shared between all
3642 		 * traffic classes, install it just once
3643 		 */
3644 		if (priv->dpni_attrs.options & DPNI_OPT_SHARED_FS)
3645 			break;
3646 	}
3647 
3648 	return err;
3649 }
3650 
3651 /* Configure the Rx flow classification key */
3652 static int dpaa2_eth_config_cls_key(struct dpaa2_eth_priv *priv, dma_addr_t key)
3653 {
3654 	struct device *dev = priv->net_dev->dev.parent;
3655 	struct dpni_rx_dist_cfg dist_cfg;
3656 	int i, err = 0;
3657 
3658 	memset(&dist_cfg, 0, sizeof(dist_cfg));
3659 
3660 	dist_cfg.key_cfg_iova = key;
3661 	dist_cfg.dist_size = dpaa2_eth_queue_count(priv);
3662 	dist_cfg.enable = 1;
3663 
3664 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3665 		dist_cfg.tc = i;
3666 		err = dpni_set_rx_fs_dist(priv->mc_io, 0, priv->mc_token,
3667 					  &dist_cfg);
3668 		if (err) {
3669 			dev_err(dev, "dpni_set_rx_fs_dist failed\n");
3670 			break;
3671 		}
3672 
3673 		/* If the flow steering / hashing key is shared between all
3674 		 * traffic classes, install it just once
3675 		 */
3676 		if (priv->dpni_attrs.options & DPNI_OPT_SHARED_FS)
3677 			break;
3678 	}
3679 
3680 	return err;
3681 }
3682 
3683 /* Size of the Rx flow classification key */
3684 int dpaa2_eth_cls_key_size(u64 fields)
3685 {
3686 	int i, size = 0;
3687 
3688 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3689 		if (!(fields & dist_fields[i].id))
3690 			continue;
3691 		size += dist_fields[i].size;
3692 	}
3693 
3694 	return size;
3695 }
3696 
3697 /* Offset of header field in Rx classification key */
3698 int dpaa2_eth_cls_fld_off(int prot, int field)
3699 {
3700 	int i, off = 0;
3701 
3702 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3703 		if (dist_fields[i].cls_prot == prot &&
3704 		    dist_fields[i].cls_field == field)
3705 			return off;
3706 		off += dist_fields[i].size;
3707 	}
3708 
3709 	WARN_ONCE(1, "Unsupported header field used for Rx flow cls\n");
3710 	return 0;
3711 }
3712 
3713 /* Prune unused fields from the classification rule.
3714  * Used when masking is not supported
3715  */
3716 void dpaa2_eth_cls_trim_rule(void *key_mem, u64 fields)
3717 {
3718 	int off = 0, new_off = 0;
3719 	int i, size;
3720 
3721 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3722 		size = dist_fields[i].size;
3723 		if (dist_fields[i].id & fields) {
3724 			memcpy(key_mem + new_off, key_mem + off, size);
3725 			new_off += size;
3726 		}
3727 		off += size;
3728 	}
3729 }
3730 
3731 /* Set Rx distribution (hash or flow classification) key
3732  * flags is a combination of RXH_ bits
3733  */
3734 static int dpaa2_eth_set_dist_key(struct net_device *net_dev,
3735 				  enum dpaa2_eth_rx_dist type, u64 flags)
3736 {
3737 	struct device *dev = net_dev->dev.parent;
3738 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
3739 	struct dpkg_profile_cfg cls_cfg;
3740 	u32 rx_hash_fields = 0;
3741 	dma_addr_t key_iova;
3742 	u8 *dma_mem;
3743 	int i;
3744 	int err = 0;
3745 
3746 	memset(&cls_cfg, 0, sizeof(cls_cfg));
3747 
3748 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3749 		struct dpkg_extract *key =
3750 			&cls_cfg.extracts[cls_cfg.num_extracts];
3751 
3752 		/* For both Rx hashing and classification keys
3753 		 * we set only the selected fields.
3754 		 */
3755 		if (!(flags & dist_fields[i].id))
3756 			continue;
3757 		if (type == DPAA2_ETH_RX_DIST_HASH)
3758 			rx_hash_fields |= dist_fields[i].rxnfc_field;
3759 
3760 		if (cls_cfg.num_extracts >= DPKG_MAX_NUM_OF_EXTRACTS) {
3761 			dev_err(dev, "error adding key extraction rule, too many rules?\n");
3762 			return -E2BIG;
3763 		}
3764 
3765 		key->type = DPKG_EXTRACT_FROM_HDR;
3766 		key->extract.from_hdr.prot = dist_fields[i].cls_prot;
3767 		key->extract.from_hdr.type = DPKG_FULL_FIELD;
3768 		key->extract.from_hdr.field = dist_fields[i].cls_field;
3769 		cls_cfg.num_extracts++;
3770 	}
3771 
3772 	dma_mem = kzalloc(DPAA2_CLASSIFIER_DMA_SIZE, GFP_KERNEL);
3773 	if (!dma_mem)
3774 		return -ENOMEM;
3775 
3776 	err = dpni_prepare_key_cfg(&cls_cfg, dma_mem);
3777 	if (err) {
3778 		dev_err(dev, "dpni_prepare_key_cfg error %d\n", err);
3779 		goto free_key;
3780 	}
3781 
3782 	/* Prepare for setting the rx dist */
3783 	key_iova = dma_map_single(dev, dma_mem, DPAA2_CLASSIFIER_DMA_SIZE,
3784 				  DMA_TO_DEVICE);
3785 	if (dma_mapping_error(dev, key_iova)) {
3786 		dev_err(dev, "DMA mapping failed\n");
3787 		err = -ENOMEM;
3788 		goto free_key;
3789 	}
3790 
3791 	if (type == DPAA2_ETH_RX_DIST_HASH) {
3792 		if (dpaa2_eth_has_legacy_dist(priv))
3793 			err = dpaa2_eth_config_legacy_hash_key(priv, key_iova);
3794 		else
3795 			err = dpaa2_eth_config_hash_key(priv, key_iova);
3796 	} else {
3797 		err = dpaa2_eth_config_cls_key(priv, key_iova);
3798 	}
3799 
3800 	dma_unmap_single(dev, key_iova, DPAA2_CLASSIFIER_DMA_SIZE,
3801 			 DMA_TO_DEVICE);
3802 	if (!err && type == DPAA2_ETH_RX_DIST_HASH)
3803 		priv->rx_hash_fields = rx_hash_fields;
3804 
3805 free_key:
3806 	kfree(dma_mem);
3807 	return err;
3808 }
3809 
3810 int dpaa2_eth_set_hash(struct net_device *net_dev, u64 flags)
3811 {
3812 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
3813 	u64 key = 0;
3814 	int i;
3815 
3816 	if (!dpaa2_eth_hash_enabled(priv))
3817 		return -EOPNOTSUPP;
3818 
3819 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++)
3820 		if (dist_fields[i].rxnfc_field & flags)
3821 			key |= dist_fields[i].id;
3822 
3823 	return dpaa2_eth_set_dist_key(net_dev, DPAA2_ETH_RX_DIST_HASH, key);
3824 }
3825 
3826 int dpaa2_eth_set_cls(struct net_device *net_dev, u64 flags)
3827 {
3828 	return dpaa2_eth_set_dist_key(net_dev, DPAA2_ETH_RX_DIST_CLS, flags);
3829 }
3830 
3831 static int dpaa2_eth_set_default_cls(struct dpaa2_eth_priv *priv)
3832 {
3833 	struct device *dev = priv->net_dev->dev.parent;
3834 	int err;
3835 
3836 	/* Check if we actually support Rx flow classification */
3837 	if (dpaa2_eth_has_legacy_dist(priv)) {
3838 		dev_dbg(dev, "Rx cls not supported by current MC version\n");
3839 		return -EOPNOTSUPP;
3840 	}
3841 
3842 	if (!dpaa2_eth_fs_enabled(priv)) {
3843 		dev_dbg(dev, "Rx cls disabled in DPNI options\n");
3844 		return -EOPNOTSUPP;
3845 	}
3846 
3847 	if (!dpaa2_eth_hash_enabled(priv)) {
3848 		dev_dbg(dev, "Rx cls disabled for single queue DPNIs\n");
3849 		return -EOPNOTSUPP;
3850 	}
3851 
3852 	/* If there is no support for masking in the classification table,
3853 	 * we don't set a default key, as it will depend on the rules
3854 	 * added by the user at runtime.
3855 	 */
3856 	if (!dpaa2_eth_fs_mask_enabled(priv))
3857 		goto out;
3858 
3859 	err = dpaa2_eth_set_cls(priv->net_dev, DPAA2_ETH_DIST_ALL);
3860 	if (err)
3861 		return err;
3862 
3863 out:
3864 	priv->rx_cls_enabled = 1;
3865 
3866 	return 0;
3867 }
3868 
3869 /* Bind the DPNI to its needed objects and resources: buffer pool, DPIOs,
3870  * frame queues and channels
3871  */
3872 static int dpaa2_eth_bind_dpni(struct dpaa2_eth_priv *priv)
3873 {
3874 	struct net_device *net_dev = priv->net_dev;
3875 	struct device *dev = net_dev->dev.parent;
3876 	struct dpni_pools_cfg pools_params;
3877 	struct dpni_error_cfg err_cfg;
3878 	int err = 0;
3879 	int i;
3880 
3881 	pools_params.num_dpbp = 1;
3882 	pools_params.pools[0].dpbp_id = priv->dpbp_dev->obj_desc.id;
3883 	pools_params.pools[0].backup_pool = 0;
3884 	pools_params.pools[0].buffer_size = priv->rx_buf_size;
3885 	err = dpni_set_pools(priv->mc_io, 0, priv->mc_token, &pools_params);
3886 	if (err) {
3887 		dev_err(dev, "dpni_set_pools() failed\n");
3888 		return err;
3889 	}
3890 
3891 	/* have the interface implicitly distribute traffic based on
3892 	 * the default hash key
3893 	 */
3894 	err = dpaa2_eth_set_hash(net_dev, DPAA2_RXH_DEFAULT);
3895 	if (err && err != -EOPNOTSUPP)
3896 		dev_err(dev, "Failed to configure hashing\n");
3897 
3898 	/* Configure the flow classification key; it includes all
3899 	 * supported header fields and cannot be modified at runtime
3900 	 */
3901 	err = dpaa2_eth_set_default_cls(priv);
3902 	if (err && err != -EOPNOTSUPP)
3903 		dev_err(dev, "Failed to configure Rx classification key\n");
3904 
3905 	/* Configure handling of error frames */
3906 	err_cfg.errors = DPAA2_FAS_RX_ERR_MASK;
3907 	err_cfg.set_frame_annotation = 1;
3908 	err_cfg.error_action = DPNI_ERROR_ACTION_DISCARD;
3909 	err = dpni_set_errors_behavior(priv->mc_io, 0, priv->mc_token,
3910 				       &err_cfg);
3911 	if (err) {
3912 		dev_err(dev, "dpni_set_errors_behavior failed\n");
3913 		return err;
3914 	}
3915 
3916 	/* Configure Rx and Tx conf queues to generate CDANs */
3917 	for (i = 0; i < priv->num_fqs; i++) {
3918 		switch (priv->fq[i].type) {
3919 		case DPAA2_RX_FQ:
3920 			err = dpaa2_eth_setup_rx_flow(priv, &priv->fq[i]);
3921 			break;
3922 		case DPAA2_TX_CONF_FQ:
3923 			err = dpaa2_eth_setup_tx_flow(priv, &priv->fq[i]);
3924 			break;
3925 		case DPAA2_RX_ERR_FQ:
3926 			err = setup_rx_err_flow(priv, &priv->fq[i]);
3927 			break;
3928 		default:
3929 			dev_err(dev, "Invalid FQ type %d\n", priv->fq[i].type);
3930 			return -EINVAL;
3931 		}
3932 		if (err)
3933 			return err;
3934 	}
3935 
3936 	err = dpni_get_qdid(priv->mc_io, 0, priv->mc_token,
3937 			    DPNI_QUEUE_TX, &priv->tx_qdid);
3938 	if (err) {
3939 		dev_err(dev, "dpni_get_qdid() failed\n");
3940 		return err;
3941 	}
3942 
3943 	return 0;
3944 }
3945 
3946 /* Allocate rings for storing incoming frame descriptors */
3947 static int dpaa2_eth_alloc_rings(struct dpaa2_eth_priv *priv)
3948 {
3949 	struct net_device *net_dev = priv->net_dev;
3950 	struct device *dev = net_dev->dev.parent;
3951 	int i;
3952 
3953 	for (i = 0; i < priv->num_channels; i++) {
3954 		priv->channel[i]->store =
3955 			dpaa2_io_store_create(DPAA2_ETH_STORE_SIZE, dev);
3956 		if (!priv->channel[i]->store) {
3957 			netdev_err(net_dev, "dpaa2_io_store_create() failed\n");
3958 			goto err_ring;
3959 		}
3960 	}
3961 
3962 	return 0;
3963 
3964 err_ring:
3965 	for (i = 0; i < priv->num_channels; i++) {
3966 		if (!priv->channel[i]->store)
3967 			break;
3968 		dpaa2_io_store_destroy(priv->channel[i]->store);
3969 	}
3970 
3971 	return -ENOMEM;
3972 }
3973 
3974 static void dpaa2_eth_free_rings(struct dpaa2_eth_priv *priv)
3975 {
3976 	int i;
3977 
3978 	for (i = 0; i < priv->num_channels; i++)
3979 		dpaa2_io_store_destroy(priv->channel[i]->store);
3980 }
3981 
3982 static int dpaa2_eth_set_mac_addr(struct dpaa2_eth_priv *priv)
3983 {
3984 	struct net_device *net_dev = priv->net_dev;
3985 	struct device *dev = net_dev->dev.parent;
3986 	u8 mac_addr[ETH_ALEN], dpni_mac_addr[ETH_ALEN];
3987 	int err;
3988 
3989 	/* Get firmware address, if any */
3990 	err = dpni_get_port_mac_addr(priv->mc_io, 0, priv->mc_token, mac_addr);
3991 	if (err) {
3992 		dev_err(dev, "dpni_get_port_mac_addr() failed\n");
3993 		return err;
3994 	}
3995 
3996 	/* Get DPNI attributes address, if any */
3997 	err = dpni_get_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
3998 					dpni_mac_addr);
3999 	if (err) {
4000 		dev_err(dev, "dpni_get_primary_mac_addr() failed\n");
4001 		return err;
4002 	}
4003 
4004 	/* First check if firmware has any address configured by bootloader */
4005 	if (!is_zero_ether_addr(mac_addr)) {
4006 		/* If the DPMAC addr != DPNI addr, update it */
4007 		if (!ether_addr_equal(mac_addr, dpni_mac_addr)) {
4008 			err = dpni_set_primary_mac_addr(priv->mc_io, 0,
4009 							priv->mc_token,
4010 							mac_addr);
4011 			if (err) {
4012 				dev_err(dev, "dpni_set_primary_mac_addr() failed\n");
4013 				return err;
4014 			}
4015 		}
4016 		memcpy(net_dev->dev_addr, mac_addr, net_dev->addr_len);
4017 	} else if (is_zero_ether_addr(dpni_mac_addr)) {
4018 		/* No MAC address configured, fill in net_dev->dev_addr
4019 		 * with a random one
4020 		 */
4021 		eth_hw_addr_random(net_dev);
4022 		dev_dbg_once(dev, "device(s) have all-zero hwaddr, replaced with random\n");
4023 
4024 		err = dpni_set_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
4025 						net_dev->dev_addr);
4026 		if (err) {
4027 			dev_err(dev, "dpni_set_primary_mac_addr() failed\n");
4028 			return err;
4029 		}
4030 
4031 		/* Override NET_ADDR_RANDOM set by eth_hw_addr_random(); for all
4032 		 * practical purposes, this will be our "permanent" mac address,
4033 		 * at least until the next reboot. This move will also permit
4034 		 * register_netdevice() to properly fill up net_dev->perm_addr.
4035 		 */
4036 		net_dev->addr_assign_type = NET_ADDR_PERM;
4037 	} else {
4038 		/* NET_ADDR_PERM is default, all we have to do is
4039 		 * fill in the device addr.
4040 		 */
4041 		memcpy(net_dev->dev_addr, dpni_mac_addr, net_dev->addr_len);
4042 	}
4043 
4044 	return 0;
4045 }
4046 
4047 static int dpaa2_eth_netdev_init(struct net_device *net_dev)
4048 {
4049 	struct device *dev = net_dev->dev.parent;
4050 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
4051 	u32 options = priv->dpni_attrs.options;
4052 	u64 supported = 0, not_supported = 0;
4053 	u8 bcast_addr[ETH_ALEN];
4054 	u8 num_queues;
4055 	int err;
4056 
4057 	net_dev->netdev_ops = &dpaa2_eth_ops;
4058 	net_dev->ethtool_ops = &dpaa2_ethtool_ops;
4059 
4060 	err = dpaa2_eth_set_mac_addr(priv);
4061 	if (err)
4062 		return err;
4063 
4064 	/* Explicitly add the broadcast address to the MAC filtering table */
4065 	eth_broadcast_addr(bcast_addr);
4066 	err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token, bcast_addr);
4067 	if (err) {
4068 		dev_err(dev, "dpni_add_mac_addr() failed\n");
4069 		return err;
4070 	}
4071 
4072 	/* Set MTU upper limit; lower limit is 68B (default value) */
4073 	net_dev->max_mtu = DPAA2_ETH_MAX_MTU;
4074 	err = dpni_set_max_frame_length(priv->mc_io, 0, priv->mc_token,
4075 					DPAA2_ETH_MFL);
4076 	if (err) {
4077 		dev_err(dev, "dpni_set_max_frame_length() failed\n");
4078 		return err;
4079 	}
4080 
4081 	/* Set actual number of queues in the net device */
4082 	num_queues = dpaa2_eth_queue_count(priv);
4083 	err = netif_set_real_num_tx_queues(net_dev, num_queues);
4084 	if (err) {
4085 		dev_err(dev, "netif_set_real_num_tx_queues() failed\n");
4086 		return err;
4087 	}
4088 	err = netif_set_real_num_rx_queues(net_dev, num_queues);
4089 	if (err) {
4090 		dev_err(dev, "netif_set_real_num_rx_queues() failed\n");
4091 		return err;
4092 	}
4093 
4094 	/* Capabilities listing */
4095 	supported |= IFF_LIVE_ADDR_CHANGE;
4096 
4097 	if (options & DPNI_OPT_NO_MAC_FILTER)
4098 		not_supported |= IFF_UNICAST_FLT;
4099 	else
4100 		supported |= IFF_UNICAST_FLT;
4101 
4102 	net_dev->priv_flags |= supported;
4103 	net_dev->priv_flags &= ~not_supported;
4104 
4105 	/* Features */
4106 	net_dev->features = NETIF_F_RXCSUM |
4107 			    NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4108 			    NETIF_F_SG | NETIF_F_HIGHDMA |
4109 			    NETIF_F_LLTX | NETIF_F_HW_TC;
4110 	net_dev->hw_features = net_dev->features;
4111 
4112 	if (priv->dpni_attrs.vlan_filter_entries)
4113 		net_dev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
4114 
4115 	return 0;
4116 }
4117 
4118 static int dpaa2_eth_poll_link_state(void *arg)
4119 {
4120 	struct dpaa2_eth_priv *priv = (struct dpaa2_eth_priv *)arg;
4121 	int err;
4122 
4123 	while (!kthread_should_stop()) {
4124 		err = dpaa2_eth_link_state_update(priv);
4125 		if (unlikely(err))
4126 			return err;
4127 
4128 		msleep(DPAA2_ETH_LINK_STATE_REFRESH);
4129 	}
4130 
4131 	return 0;
4132 }
4133 
4134 static int dpaa2_eth_connect_mac(struct dpaa2_eth_priv *priv)
4135 {
4136 	struct fsl_mc_device *dpni_dev, *dpmac_dev;
4137 	struct dpaa2_mac *mac;
4138 	int err;
4139 
4140 	dpni_dev = to_fsl_mc_device(priv->net_dev->dev.parent);
4141 	dpmac_dev = fsl_mc_get_endpoint(dpni_dev, 0);
4142 
4143 	if (PTR_ERR(dpmac_dev) == -EPROBE_DEFER)
4144 		return PTR_ERR(dpmac_dev);
4145 
4146 	if (IS_ERR(dpmac_dev) || dpmac_dev->dev.type != &fsl_mc_bus_dpmac_type)
4147 		return 0;
4148 
4149 	mac = kzalloc(sizeof(struct dpaa2_mac), GFP_KERNEL);
4150 	if (!mac)
4151 		return -ENOMEM;
4152 
4153 	mac->mc_dev = dpmac_dev;
4154 	mac->mc_io = priv->mc_io;
4155 	mac->net_dev = priv->net_dev;
4156 
4157 	err = dpaa2_mac_open(mac);
4158 	if (err)
4159 		goto err_free_mac;
4160 	priv->mac = mac;
4161 
4162 	if (dpaa2_eth_is_type_phy(priv)) {
4163 		err = dpaa2_mac_connect(mac);
4164 		if (err && err != -EPROBE_DEFER)
4165 			netdev_err(priv->net_dev, "Error connecting to the MAC endpoint: %pe",
4166 				   ERR_PTR(err));
4167 		if (err)
4168 			goto err_close_mac;
4169 	}
4170 
4171 	return 0;
4172 
4173 err_close_mac:
4174 	dpaa2_mac_close(mac);
4175 	priv->mac = NULL;
4176 err_free_mac:
4177 	kfree(mac);
4178 	return err;
4179 }
4180 
4181 static void dpaa2_eth_disconnect_mac(struct dpaa2_eth_priv *priv)
4182 {
4183 	if (dpaa2_eth_is_type_phy(priv))
4184 		dpaa2_mac_disconnect(priv->mac);
4185 
4186 	if (!dpaa2_eth_has_mac(priv))
4187 		return;
4188 
4189 	dpaa2_mac_close(priv->mac);
4190 	kfree(priv->mac);
4191 	priv->mac = NULL;
4192 }
4193 
4194 static irqreturn_t dpni_irq0_handler_thread(int irq_num, void *arg)
4195 {
4196 	u32 status = ~0;
4197 	struct device *dev = (struct device *)arg;
4198 	struct fsl_mc_device *dpni_dev = to_fsl_mc_device(dev);
4199 	struct net_device *net_dev = dev_get_drvdata(dev);
4200 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
4201 	int err;
4202 
4203 	err = dpni_get_irq_status(dpni_dev->mc_io, 0, dpni_dev->mc_handle,
4204 				  DPNI_IRQ_INDEX, &status);
4205 	if (unlikely(err)) {
4206 		netdev_err(net_dev, "Can't get irq status (err %d)\n", err);
4207 		return IRQ_HANDLED;
4208 	}
4209 
4210 	if (status & DPNI_IRQ_EVENT_LINK_CHANGED)
4211 		dpaa2_eth_link_state_update(netdev_priv(net_dev));
4212 
4213 	if (status & DPNI_IRQ_EVENT_ENDPOINT_CHANGED) {
4214 		dpaa2_eth_set_mac_addr(netdev_priv(net_dev));
4215 		dpaa2_eth_update_tx_fqids(priv);
4216 
4217 		rtnl_lock();
4218 		if (dpaa2_eth_has_mac(priv))
4219 			dpaa2_eth_disconnect_mac(priv);
4220 		else
4221 			dpaa2_eth_connect_mac(priv);
4222 		rtnl_unlock();
4223 	}
4224 
4225 	return IRQ_HANDLED;
4226 }
4227 
4228 static int dpaa2_eth_setup_irqs(struct fsl_mc_device *ls_dev)
4229 {
4230 	int err = 0;
4231 	struct fsl_mc_device_irq *irq;
4232 
4233 	err = fsl_mc_allocate_irqs(ls_dev);
4234 	if (err) {
4235 		dev_err(&ls_dev->dev, "MC irqs allocation failed\n");
4236 		return err;
4237 	}
4238 
4239 	irq = ls_dev->irqs[0];
4240 	err = devm_request_threaded_irq(&ls_dev->dev, irq->msi_desc->irq,
4241 					NULL, dpni_irq0_handler_thread,
4242 					IRQF_NO_SUSPEND | IRQF_ONESHOT,
4243 					dev_name(&ls_dev->dev), &ls_dev->dev);
4244 	if (err < 0) {
4245 		dev_err(&ls_dev->dev, "devm_request_threaded_irq(): %d\n", err);
4246 		goto free_mc_irq;
4247 	}
4248 
4249 	err = dpni_set_irq_mask(ls_dev->mc_io, 0, ls_dev->mc_handle,
4250 				DPNI_IRQ_INDEX, DPNI_IRQ_EVENT_LINK_CHANGED |
4251 				DPNI_IRQ_EVENT_ENDPOINT_CHANGED);
4252 	if (err < 0) {
4253 		dev_err(&ls_dev->dev, "dpni_set_irq_mask(): %d\n", err);
4254 		goto free_irq;
4255 	}
4256 
4257 	err = dpni_set_irq_enable(ls_dev->mc_io, 0, ls_dev->mc_handle,
4258 				  DPNI_IRQ_INDEX, 1);
4259 	if (err < 0) {
4260 		dev_err(&ls_dev->dev, "dpni_set_irq_enable(): %d\n", err);
4261 		goto free_irq;
4262 	}
4263 
4264 	return 0;
4265 
4266 free_irq:
4267 	devm_free_irq(&ls_dev->dev, irq->msi_desc->irq, &ls_dev->dev);
4268 free_mc_irq:
4269 	fsl_mc_free_irqs(ls_dev);
4270 
4271 	return err;
4272 }
4273 
4274 static void dpaa2_eth_add_ch_napi(struct dpaa2_eth_priv *priv)
4275 {
4276 	int i;
4277 	struct dpaa2_eth_channel *ch;
4278 
4279 	for (i = 0; i < priv->num_channels; i++) {
4280 		ch = priv->channel[i];
4281 		/* NAPI weight *MUST* be a multiple of DPAA2_ETH_STORE_SIZE */
4282 		netif_napi_add(priv->net_dev, &ch->napi, dpaa2_eth_poll,
4283 			       NAPI_POLL_WEIGHT);
4284 	}
4285 }
4286 
4287 static void dpaa2_eth_del_ch_napi(struct dpaa2_eth_priv *priv)
4288 {
4289 	int i;
4290 	struct dpaa2_eth_channel *ch;
4291 
4292 	for (i = 0; i < priv->num_channels; i++) {
4293 		ch = priv->channel[i];
4294 		netif_napi_del(&ch->napi);
4295 	}
4296 }
4297 
4298 static int dpaa2_eth_probe(struct fsl_mc_device *dpni_dev)
4299 {
4300 	struct device *dev;
4301 	struct net_device *net_dev = NULL;
4302 	struct dpaa2_eth_priv *priv = NULL;
4303 	int err = 0;
4304 
4305 	dev = &dpni_dev->dev;
4306 
4307 	/* Net device */
4308 	net_dev = alloc_etherdev_mq(sizeof(*priv), DPAA2_ETH_MAX_NETDEV_QUEUES);
4309 	if (!net_dev) {
4310 		dev_err(dev, "alloc_etherdev_mq() failed\n");
4311 		return -ENOMEM;
4312 	}
4313 
4314 	SET_NETDEV_DEV(net_dev, dev);
4315 	dev_set_drvdata(dev, net_dev);
4316 
4317 	priv = netdev_priv(net_dev);
4318 	priv->net_dev = net_dev;
4319 
4320 	priv->iommu_domain = iommu_get_domain_for_dev(dev);
4321 
4322 	priv->tx_tstamp_type = HWTSTAMP_TX_OFF;
4323 	priv->rx_tstamp = false;
4324 
4325 	priv->dpaa2_ptp_wq = alloc_workqueue("dpaa2_ptp_wq", 0, 0);
4326 	if (!priv->dpaa2_ptp_wq) {
4327 		err = -ENOMEM;
4328 		goto err_wq_alloc;
4329 	}
4330 
4331 	INIT_WORK(&priv->tx_onestep_tstamp, dpaa2_eth_tx_onestep_tstamp);
4332 
4333 	skb_queue_head_init(&priv->tx_skbs);
4334 
4335 	priv->rx_copybreak = DPAA2_ETH_DEFAULT_COPYBREAK;
4336 
4337 	/* Obtain a MC portal */
4338 	err = fsl_mc_portal_allocate(dpni_dev, FSL_MC_IO_ATOMIC_CONTEXT_PORTAL,
4339 				     &priv->mc_io);
4340 	if (err) {
4341 		if (err == -ENXIO)
4342 			err = -EPROBE_DEFER;
4343 		else
4344 			dev_err(dev, "MC portal allocation failed\n");
4345 		goto err_portal_alloc;
4346 	}
4347 
4348 	/* MC objects initialization and configuration */
4349 	err = dpaa2_eth_setup_dpni(dpni_dev);
4350 	if (err)
4351 		goto err_dpni_setup;
4352 
4353 	err = dpaa2_eth_setup_dpio(priv);
4354 	if (err)
4355 		goto err_dpio_setup;
4356 
4357 	dpaa2_eth_setup_fqs(priv);
4358 
4359 	err = dpaa2_eth_setup_dpbp(priv);
4360 	if (err)
4361 		goto err_dpbp_setup;
4362 
4363 	err = dpaa2_eth_bind_dpni(priv);
4364 	if (err)
4365 		goto err_bind;
4366 
4367 	/* Add a NAPI context for each channel */
4368 	dpaa2_eth_add_ch_napi(priv);
4369 
4370 	/* Percpu statistics */
4371 	priv->percpu_stats = alloc_percpu(*priv->percpu_stats);
4372 	if (!priv->percpu_stats) {
4373 		dev_err(dev, "alloc_percpu(percpu_stats) failed\n");
4374 		err = -ENOMEM;
4375 		goto err_alloc_percpu_stats;
4376 	}
4377 	priv->percpu_extras = alloc_percpu(*priv->percpu_extras);
4378 	if (!priv->percpu_extras) {
4379 		dev_err(dev, "alloc_percpu(percpu_extras) failed\n");
4380 		err = -ENOMEM;
4381 		goto err_alloc_percpu_extras;
4382 	}
4383 
4384 	priv->sgt_cache = alloc_percpu(*priv->sgt_cache);
4385 	if (!priv->sgt_cache) {
4386 		dev_err(dev, "alloc_percpu(sgt_cache) failed\n");
4387 		err = -ENOMEM;
4388 		goto err_alloc_sgt_cache;
4389 	}
4390 
4391 	err = dpaa2_eth_netdev_init(net_dev);
4392 	if (err)
4393 		goto err_netdev_init;
4394 
4395 	/* Configure checksum offload based on current interface flags */
4396 	err = dpaa2_eth_set_rx_csum(priv, !!(net_dev->features & NETIF_F_RXCSUM));
4397 	if (err)
4398 		goto err_csum;
4399 
4400 	err = dpaa2_eth_set_tx_csum(priv,
4401 				    !!(net_dev->features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)));
4402 	if (err)
4403 		goto err_csum;
4404 
4405 	err = dpaa2_eth_alloc_rings(priv);
4406 	if (err)
4407 		goto err_alloc_rings;
4408 
4409 #ifdef CONFIG_FSL_DPAA2_ETH_DCB
4410 	if (dpaa2_eth_has_pause_support(priv) && priv->vlan_cls_enabled) {
4411 		priv->dcbx_mode = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_IEEE;
4412 		net_dev->dcbnl_ops = &dpaa2_eth_dcbnl_ops;
4413 	} else {
4414 		dev_dbg(dev, "PFC not supported\n");
4415 	}
4416 #endif
4417 
4418 	err = dpaa2_eth_setup_irqs(dpni_dev);
4419 	if (err) {
4420 		netdev_warn(net_dev, "Failed to set link interrupt, fall back to polling\n");
4421 		priv->poll_thread = kthread_run(dpaa2_eth_poll_link_state, priv,
4422 						"%s_poll_link", net_dev->name);
4423 		if (IS_ERR(priv->poll_thread)) {
4424 			dev_err(dev, "Error starting polling thread\n");
4425 			goto err_poll_thread;
4426 		}
4427 		priv->do_link_poll = true;
4428 	}
4429 
4430 	err = dpaa2_eth_connect_mac(priv);
4431 	if (err)
4432 		goto err_connect_mac;
4433 
4434 	err = dpaa2_eth_dl_register(priv);
4435 	if (err)
4436 		goto err_dl_register;
4437 
4438 	err = dpaa2_eth_dl_traps_register(priv);
4439 	if (err)
4440 		goto err_dl_trap_register;
4441 
4442 	err = dpaa2_eth_dl_port_add(priv);
4443 	if (err)
4444 		goto err_dl_port_add;
4445 
4446 	err = register_netdev(net_dev);
4447 	if (err < 0) {
4448 		dev_err(dev, "register_netdev() failed\n");
4449 		goto err_netdev_reg;
4450 	}
4451 
4452 #ifdef CONFIG_DEBUG_FS
4453 	dpaa2_dbg_add(priv);
4454 #endif
4455 
4456 	dev_info(dev, "Probed interface %s\n", net_dev->name);
4457 	return 0;
4458 
4459 err_netdev_reg:
4460 	dpaa2_eth_dl_port_del(priv);
4461 err_dl_port_add:
4462 	dpaa2_eth_dl_traps_unregister(priv);
4463 err_dl_trap_register:
4464 	dpaa2_eth_dl_unregister(priv);
4465 err_dl_register:
4466 	dpaa2_eth_disconnect_mac(priv);
4467 err_connect_mac:
4468 	if (priv->do_link_poll)
4469 		kthread_stop(priv->poll_thread);
4470 	else
4471 		fsl_mc_free_irqs(dpni_dev);
4472 err_poll_thread:
4473 	dpaa2_eth_free_rings(priv);
4474 err_alloc_rings:
4475 err_csum:
4476 err_netdev_init:
4477 	free_percpu(priv->sgt_cache);
4478 err_alloc_sgt_cache:
4479 	free_percpu(priv->percpu_extras);
4480 err_alloc_percpu_extras:
4481 	free_percpu(priv->percpu_stats);
4482 err_alloc_percpu_stats:
4483 	dpaa2_eth_del_ch_napi(priv);
4484 err_bind:
4485 	dpaa2_eth_free_dpbp(priv);
4486 err_dpbp_setup:
4487 	dpaa2_eth_free_dpio(priv);
4488 err_dpio_setup:
4489 	dpaa2_eth_free_dpni(priv);
4490 err_dpni_setup:
4491 	fsl_mc_portal_free(priv->mc_io);
4492 err_portal_alloc:
4493 	destroy_workqueue(priv->dpaa2_ptp_wq);
4494 err_wq_alloc:
4495 	dev_set_drvdata(dev, NULL);
4496 	free_netdev(net_dev);
4497 
4498 	return err;
4499 }
4500 
4501 static int dpaa2_eth_remove(struct fsl_mc_device *ls_dev)
4502 {
4503 	struct device *dev;
4504 	struct net_device *net_dev;
4505 	struct dpaa2_eth_priv *priv;
4506 
4507 	dev = &ls_dev->dev;
4508 	net_dev = dev_get_drvdata(dev);
4509 	priv = netdev_priv(net_dev);
4510 
4511 #ifdef CONFIG_DEBUG_FS
4512 	dpaa2_dbg_remove(priv);
4513 #endif
4514 	rtnl_lock();
4515 	dpaa2_eth_disconnect_mac(priv);
4516 	rtnl_unlock();
4517 
4518 	unregister_netdev(net_dev);
4519 
4520 	dpaa2_eth_dl_port_del(priv);
4521 	dpaa2_eth_dl_traps_unregister(priv);
4522 	dpaa2_eth_dl_unregister(priv);
4523 
4524 	if (priv->do_link_poll)
4525 		kthread_stop(priv->poll_thread);
4526 	else
4527 		fsl_mc_free_irqs(ls_dev);
4528 
4529 	dpaa2_eth_free_rings(priv);
4530 	free_percpu(priv->sgt_cache);
4531 	free_percpu(priv->percpu_stats);
4532 	free_percpu(priv->percpu_extras);
4533 
4534 	dpaa2_eth_del_ch_napi(priv);
4535 	dpaa2_eth_free_dpbp(priv);
4536 	dpaa2_eth_free_dpio(priv);
4537 	dpaa2_eth_free_dpni(priv);
4538 
4539 	fsl_mc_portal_free(priv->mc_io);
4540 
4541 	free_netdev(net_dev);
4542 
4543 	dev_dbg(net_dev->dev.parent, "Removed interface %s\n", net_dev->name);
4544 
4545 	return 0;
4546 }
4547 
4548 static const struct fsl_mc_device_id dpaa2_eth_match_id_table[] = {
4549 	{
4550 		.vendor = FSL_MC_VENDOR_FREESCALE,
4551 		.obj_type = "dpni",
4552 	},
4553 	{ .vendor = 0x0 }
4554 };
4555 MODULE_DEVICE_TABLE(fslmc, dpaa2_eth_match_id_table);
4556 
4557 static struct fsl_mc_driver dpaa2_eth_driver = {
4558 	.driver = {
4559 		.name = KBUILD_MODNAME,
4560 		.owner = THIS_MODULE,
4561 	},
4562 	.probe = dpaa2_eth_probe,
4563 	.remove = dpaa2_eth_remove,
4564 	.match_id_table = dpaa2_eth_match_id_table
4565 };
4566 
4567 static int __init dpaa2_eth_driver_init(void)
4568 {
4569 	int err;
4570 
4571 	dpaa2_eth_dbg_init();
4572 	err = fsl_mc_driver_register(&dpaa2_eth_driver);
4573 	if (err) {
4574 		dpaa2_eth_dbg_exit();
4575 		return err;
4576 	}
4577 
4578 	return 0;
4579 }
4580 
4581 static void __exit dpaa2_eth_driver_exit(void)
4582 {
4583 	dpaa2_eth_dbg_exit();
4584 	fsl_mc_driver_unregister(&dpaa2_eth_driver);
4585 }
4586 
4587 module_init(dpaa2_eth_driver_init);
4588 module_exit(dpaa2_eth_driver_exit);
4589