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(priv->net_dev, xdp_prog, 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 	ch->stats.bytes_per_cdan += dpaa2_fd_get_len(fd);
537 
538 	list_add_tail(&skb->list, ch->rx_list);
539 
540 	return;
541 
542 err_build_skb:
543 	dpaa2_eth_free_rx_fd(priv, fd, vaddr);
544 err_frame_format:
545 	percpu_stats->rx_dropped++;
546 }
547 
548 /* Processing of Rx frames received on the error FQ
549  * We check and print the error bits and then free the frame
550  */
551 static void dpaa2_eth_rx_err(struct dpaa2_eth_priv *priv,
552 			     struct dpaa2_eth_channel *ch,
553 			     const struct dpaa2_fd *fd,
554 			     struct dpaa2_eth_fq *fq __always_unused)
555 {
556 	struct device *dev = priv->net_dev->dev.parent;
557 	dma_addr_t addr = dpaa2_fd_get_addr(fd);
558 	u8 fd_format = dpaa2_fd_get_format(fd);
559 	struct rtnl_link_stats64 *percpu_stats;
560 	struct dpaa2_eth_trap_item *trap_item;
561 	struct dpaa2_fapr *fapr;
562 	struct sk_buff *skb;
563 	void *buf_data;
564 	void *vaddr;
565 
566 	vaddr = dpaa2_iova_to_virt(priv->iommu_domain, addr);
567 	dma_sync_single_for_cpu(dev, addr, priv->rx_buf_size,
568 				DMA_BIDIRECTIONAL);
569 
570 	buf_data = vaddr + dpaa2_fd_get_offset(fd);
571 
572 	if (fd_format == dpaa2_fd_single) {
573 		dma_unmap_page(dev, addr, priv->rx_buf_size,
574 			       DMA_BIDIRECTIONAL);
575 		skb = dpaa2_eth_build_linear_skb(ch, fd, vaddr);
576 	} else if (fd_format == dpaa2_fd_sg) {
577 		dma_unmap_page(dev, addr, priv->rx_buf_size,
578 			       DMA_BIDIRECTIONAL);
579 		skb = dpaa2_eth_build_frag_skb(priv, ch, buf_data);
580 		free_pages((unsigned long)vaddr, 0);
581 	} else {
582 		/* We don't support any other format */
583 		dpaa2_eth_free_rx_fd(priv, fd, vaddr);
584 		goto err_frame_format;
585 	}
586 
587 	fapr = dpaa2_get_fapr(vaddr, false);
588 	trap_item = dpaa2_eth_dl_get_trap(priv, fapr);
589 	if (trap_item)
590 		devlink_trap_report(priv->devlink, skb, trap_item->trap_ctx,
591 				    &priv->devlink_port, NULL);
592 	consume_skb(skb);
593 
594 err_frame_format:
595 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
596 	percpu_stats->rx_errors++;
597 	ch->buf_count--;
598 }
599 
600 /* Consume all frames pull-dequeued into the store. This is the simplest way to
601  * make sure we don't accidentally issue another volatile dequeue which would
602  * overwrite (leak) frames already in the store.
603  *
604  * Observance of NAPI budget is not our concern, leaving that to the caller.
605  */
606 static int dpaa2_eth_consume_frames(struct dpaa2_eth_channel *ch,
607 				    struct dpaa2_eth_fq **src)
608 {
609 	struct dpaa2_eth_priv *priv = ch->priv;
610 	struct dpaa2_eth_fq *fq = NULL;
611 	struct dpaa2_dq *dq;
612 	const struct dpaa2_fd *fd;
613 	int cleaned = 0, retries = 0;
614 	int is_last;
615 
616 	do {
617 		dq = dpaa2_io_store_next(ch->store, &is_last);
618 		if (unlikely(!dq)) {
619 			/* If we're here, we *must* have placed a
620 			 * volatile dequeue comnmand, so keep reading through
621 			 * the store until we get some sort of valid response
622 			 * token (either a valid frame or an "empty dequeue")
623 			 */
624 			if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES) {
625 				netdev_err_once(priv->net_dev,
626 						"Unable to read a valid dequeue response\n");
627 				return -ETIMEDOUT;
628 			}
629 			continue;
630 		}
631 
632 		fd = dpaa2_dq_fd(dq);
633 		fq = (struct dpaa2_eth_fq *)(uintptr_t)dpaa2_dq_fqd_ctx(dq);
634 
635 		fq->consume(priv, ch, fd, fq);
636 		cleaned++;
637 		retries = 0;
638 	} while (!is_last);
639 
640 	if (!cleaned)
641 		return 0;
642 
643 	fq->stats.frames += cleaned;
644 	ch->stats.frames += cleaned;
645 	ch->stats.frames_per_cdan += cleaned;
646 
647 	/* A dequeue operation only pulls frames from a single queue
648 	 * into the store. Return the frame queue as an out param.
649 	 */
650 	if (src)
651 		*src = fq;
652 
653 	return cleaned;
654 }
655 
656 static int dpaa2_eth_ptp_parse(struct sk_buff *skb,
657 			       u8 *msgtype, u8 *twostep, u8 *udp,
658 			       u16 *correction_offset,
659 			       u16 *origintimestamp_offset)
660 {
661 	unsigned int ptp_class;
662 	struct ptp_header *hdr;
663 	unsigned int type;
664 	u8 *base;
665 
666 	ptp_class = ptp_classify_raw(skb);
667 	if (ptp_class == PTP_CLASS_NONE)
668 		return -EINVAL;
669 
670 	hdr = ptp_parse_header(skb, ptp_class);
671 	if (!hdr)
672 		return -EINVAL;
673 
674 	*msgtype = ptp_get_msgtype(hdr, ptp_class);
675 	*twostep = hdr->flag_field[0] & 0x2;
676 
677 	type = ptp_class & PTP_CLASS_PMASK;
678 	if (type == PTP_CLASS_IPV4 ||
679 	    type == PTP_CLASS_IPV6)
680 		*udp = 1;
681 	else
682 		*udp = 0;
683 
684 	base = skb_mac_header(skb);
685 	*correction_offset = (u8 *)&hdr->correction - base;
686 	*origintimestamp_offset = (u8 *)hdr + sizeof(struct ptp_header) - base;
687 
688 	return 0;
689 }
690 
691 /* Configure the egress frame annotation for timestamp update */
692 static void dpaa2_eth_enable_tx_tstamp(struct dpaa2_eth_priv *priv,
693 				       struct dpaa2_fd *fd,
694 				       void *buf_start,
695 				       struct sk_buff *skb)
696 {
697 	struct ptp_tstamp origin_timestamp;
698 	struct dpni_single_step_cfg cfg;
699 	u8 msgtype, twostep, udp;
700 	struct dpaa2_faead *faead;
701 	struct dpaa2_fas *fas;
702 	struct timespec64 ts;
703 	u16 offset1, offset2;
704 	u32 ctrl, frc;
705 	__le64 *ns;
706 	u8 *data;
707 
708 	/* Mark the egress frame annotation area as valid */
709 	frc = dpaa2_fd_get_frc(fd);
710 	dpaa2_fd_set_frc(fd, frc | DPAA2_FD_FRC_FAEADV);
711 
712 	/* Set hardware annotation size */
713 	ctrl = dpaa2_fd_get_ctrl(fd);
714 	dpaa2_fd_set_ctrl(fd, ctrl | DPAA2_FD_CTRL_ASAL);
715 
716 	/* enable UPD (update prepanded data) bit in FAEAD field of
717 	 * hardware frame annotation area
718 	 */
719 	ctrl = DPAA2_FAEAD_A2V | DPAA2_FAEAD_UPDV | DPAA2_FAEAD_UPD;
720 	faead = dpaa2_get_faead(buf_start, true);
721 	faead->ctrl = cpu_to_le32(ctrl);
722 
723 	if (skb->cb[0] == TX_TSTAMP_ONESTEP_SYNC) {
724 		if (dpaa2_eth_ptp_parse(skb, &msgtype, &twostep, &udp,
725 					&offset1, &offset2) ||
726 		    msgtype != PTP_MSGTYPE_SYNC || twostep) {
727 			WARN_ONCE(1, "Bad packet for one-step timestamping\n");
728 			return;
729 		}
730 
731 		/* Mark the frame annotation status as valid */
732 		frc = dpaa2_fd_get_frc(fd);
733 		dpaa2_fd_set_frc(fd, frc | DPAA2_FD_FRC_FASV);
734 
735 		/* Mark the PTP flag for one step timestamping */
736 		fas = dpaa2_get_fas(buf_start, true);
737 		fas->status = cpu_to_le32(DPAA2_FAS_PTP);
738 
739 		dpaa2_ptp->caps.gettime64(&dpaa2_ptp->caps, &ts);
740 		ns = dpaa2_get_ts(buf_start, true);
741 		*ns = cpu_to_le64(timespec64_to_ns(&ts) /
742 				  DPAA2_PTP_CLK_PERIOD_NS);
743 
744 		/* Update current time to PTP message originTimestamp field */
745 		ns_to_ptp_tstamp(&origin_timestamp, le64_to_cpup(ns));
746 		data = skb_mac_header(skb);
747 		*(__be16 *)(data + offset2) = htons(origin_timestamp.sec_msb);
748 		*(__be32 *)(data + offset2 + 2) =
749 			htonl(origin_timestamp.sec_lsb);
750 		*(__be32 *)(data + offset2 + 6) = htonl(origin_timestamp.nsec);
751 
752 		cfg.en = 1;
753 		cfg.ch_update = udp;
754 		cfg.offset = offset1;
755 		cfg.peer_delay = 0;
756 
757 		if (dpni_set_single_step_cfg(priv->mc_io, 0, priv->mc_token,
758 					     &cfg))
759 			WARN_ONCE(1, "Failed to set single step register");
760 	}
761 }
762 
763 /* Create a frame descriptor based on a fragmented skb */
764 static int dpaa2_eth_build_sg_fd(struct dpaa2_eth_priv *priv,
765 				 struct sk_buff *skb,
766 				 struct dpaa2_fd *fd,
767 				 void **swa_addr)
768 {
769 	struct device *dev = priv->net_dev->dev.parent;
770 	void *sgt_buf = NULL;
771 	dma_addr_t addr;
772 	int nr_frags = skb_shinfo(skb)->nr_frags;
773 	struct dpaa2_sg_entry *sgt;
774 	int i, err;
775 	int sgt_buf_size;
776 	struct scatterlist *scl, *crt_scl;
777 	int num_sg;
778 	int num_dma_bufs;
779 	struct dpaa2_eth_swa *swa;
780 
781 	/* Create and map scatterlist.
782 	 * We don't advertise NETIF_F_FRAGLIST, so skb_to_sgvec() will not have
783 	 * to go beyond nr_frags+1.
784 	 * Note: We don't support chained scatterlists
785 	 */
786 	if (unlikely(PAGE_SIZE / sizeof(struct scatterlist) < nr_frags + 1))
787 		return -EINVAL;
788 
789 	scl = kmalloc_array(nr_frags + 1, sizeof(struct scatterlist), GFP_ATOMIC);
790 	if (unlikely(!scl))
791 		return -ENOMEM;
792 
793 	sg_init_table(scl, nr_frags + 1);
794 	num_sg = skb_to_sgvec(skb, scl, 0, skb->len);
795 	if (unlikely(num_sg < 0)) {
796 		err = -ENOMEM;
797 		goto dma_map_sg_failed;
798 	}
799 	num_dma_bufs = dma_map_sg(dev, scl, num_sg, DMA_BIDIRECTIONAL);
800 	if (unlikely(!num_dma_bufs)) {
801 		err = -ENOMEM;
802 		goto dma_map_sg_failed;
803 	}
804 
805 	/* Prepare the HW SGT structure */
806 	sgt_buf_size = priv->tx_data_offset +
807 		       sizeof(struct dpaa2_sg_entry) *  num_dma_bufs;
808 	sgt_buf = napi_alloc_frag_align(sgt_buf_size, DPAA2_ETH_TX_BUF_ALIGN);
809 	if (unlikely(!sgt_buf)) {
810 		err = -ENOMEM;
811 		goto sgt_buf_alloc_failed;
812 	}
813 	memset(sgt_buf, 0, sgt_buf_size);
814 
815 	sgt = (struct dpaa2_sg_entry *)(sgt_buf + priv->tx_data_offset);
816 
817 	/* Fill in the HW SGT structure.
818 	 *
819 	 * sgt_buf is zeroed out, so the following fields are implicit
820 	 * in all sgt entries:
821 	 *   - offset is 0
822 	 *   - format is 'dpaa2_sg_single'
823 	 */
824 	for_each_sg(scl, crt_scl, num_dma_bufs, i) {
825 		dpaa2_sg_set_addr(&sgt[i], sg_dma_address(crt_scl));
826 		dpaa2_sg_set_len(&sgt[i], sg_dma_len(crt_scl));
827 	}
828 	dpaa2_sg_set_final(&sgt[i - 1], true);
829 
830 	/* Store the skb backpointer in the SGT buffer.
831 	 * Fit the scatterlist and the number of buffers alongside the
832 	 * skb backpointer in the software annotation area. We'll need
833 	 * all of them on Tx Conf.
834 	 */
835 	*swa_addr = (void *)sgt_buf;
836 	swa = (struct dpaa2_eth_swa *)sgt_buf;
837 	swa->type = DPAA2_ETH_SWA_SG;
838 	swa->sg.skb = skb;
839 	swa->sg.scl = scl;
840 	swa->sg.num_sg = num_sg;
841 	swa->sg.sgt_size = sgt_buf_size;
842 
843 	/* Separately map the SGT buffer */
844 	addr = dma_map_single(dev, sgt_buf, sgt_buf_size, DMA_BIDIRECTIONAL);
845 	if (unlikely(dma_mapping_error(dev, addr))) {
846 		err = -ENOMEM;
847 		goto dma_map_single_failed;
848 	}
849 	dpaa2_fd_set_offset(fd, priv->tx_data_offset);
850 	dpaa2_fd_set_format(fd, dpaa2_fd_sg);
851 	dpaa2_fd_set_addr(fd, addr);
852 	dpaa2_fd_set_len(fd, skb->len);
853 	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
854 
855 	return 0;
856 
857 dma_map_single_failed:
858 	skb_free_frag(sgt_buf);
859 sgt_buf_alloc_failed:
860 	dma_unmap_sg(dev, scl, num_sg, DMA_BIDIRECTIONAL);
861 dma_map_sg_failed:
862 	kfree(scl);
863 	return err;
864 }
865 
866 /* Create a SG frame descriptor based on a linear skb.
867  *
868  * This function is used on the Tx path when the skb headroom is not large
869  * enough for the HW requirements, thus instead of realloc-ing the skb we
870  * create a SG frame descriptor with only one entry.
871  */
872 static int dpaa2_eth_build_sg_fd_single_buf(struct dpaa2_eth_priv *priv,
873 					    struct sk_buff *skb,
874 					    struct dpaa2_fd *fd,
875 					    void **swa_addr)
876 {
877 	struct device *dev = priv->net_dev->dev.parent;
878 	struct dpaa2_eth_sgt_cache *sgt_cache;
879 	struct dpaa2_sg_entry *sgt;
880 	struct dpaa2_eth_swa *swa;
881 	dma_addr_t addr, sgt_addr;
882 	void *sgt_buf = NULL;
883 	int sgt_buf_size;
884 	int err;
885 
886 	/* Prepare the HW SGT structure */
887 	sgt_cache = this_cpu_ptr(priv->sgt_cache);
888 	sgt_buf_size = priv->tx_data_offset + sizeof(struct dpaa2_sg_entry);
889 
890 	if (sgt_cache->count == 0)
891 		sgt_buf = kzalloc(sgt_buf_size + DPAA2_ETH_TX_BUF_ALIGN,
892 				  GFP_ATOMIC);
893 	else
894 		sgt_buf = sgt_cache->buf[--sgt_cache->count];
895 	if (unlikely(!sgt_buf))
896 		return -ENOMEM;
897 
898 	sgt_buf = PTR_ALIGN(sgt_buf, DPAA2_ETH_TX_BUF_ALIGN);
899 	sgt = (struct dpaa2_sg_entry *)(sgt_buf + priv->tx_data_offset);
900 
901 	addr = dma_map_single(dev, skb->data, skb->len, DMA_BIDIRECTIONAL);
902 	if (unlikely(dma_mapping_error(dev, addr))) {
903 		err = -ENOMEM;
904 		goto data_map_failed;
905 	}
906 
907 	/* Fill in the HW SGT structure */
908 	dpaa2_sg_set_addr(sgt, addr);
909 	dpaa2_sg_set_len(sgt, skb->len);
910 	dpaa2_sg_set_final(sgt, true);
911 
912 	/* Store the skb backpointer in the SGT buffer */
913 	*swa_addr = (void *)sgt_buf;
914 	swa = (struct dpaa2_eth_swa *)sgt_buf;
915 	swa->type = DPAA2_ETH_SWA_SINGLE;
916 	swa->single.skb = skb;
917 	swa->single.sgt_size = sgt_buf_size;
918 
919 	/* Separately map the SGT buffer */
920 	sgt_addr = dma_map_single(dev, sgt_buf, sgt_buf_size, DMA_BIDIRECTIONAL);
921 	if (unlikely(dma_mapping_error(dev, sgt_addr))) {
922 		err = -ENOMEM;
923 		goto sgt_map_failed;
924 	}
925 
926 	dpaa2_fd_set_offset(fd, priv->tx_data_offset);
927 	dpaa2_fd_set_format(fd, dpaa2_fd_sg);
928 	dpaa2_fd_set_addr(fd, sgt_addr);
929 	dpaa2_fd_set_len(fd, skb->len);
930 	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
931 
932 	return 0;
933 
934 sgt_map_failed:
935 	dma_unmap_single(dev, addr, skb->len, DMA_BIDIRECTIONAL);
936 data_map_failed:
937 	if (sgt_cache->count >= DPAA2_ETH_SGT_CACHE_SIZE)
938 		kfree(sgt_buf);
939 	else
940 		sgt_cache->buf[sgt_cache->count++] = sgt_buf;
941 
942 	return err;
943 }
944 
945 /* Create a frame descriptor based on a linear skb */
946 static int dpaa2_eth_build_single_fd(struct dpaa2_eth_priv *priv,
947 				     struct sk_buff *skb,
948 				     struct dpaa2_fd *fd,
949 				     void **swa_addr)
950 {
951 	struct device *dev = priv->net_dev->dev.parent;
952 	u8 *buffer_start, *aligned_start;
953 	struct dpaa2_eth_swa *swa;
954 	dma_addr_t addr;
955 
956 	buffer_start = skb->data - dpaa2_eth_needed_headroom(skb);
957 
958 	/* If there's enough room to align the FD address, do it.
959 	 * It will help hardware optimize accesses.
960 	 */
961 	aligned_start = PTR_ALIGN(buffer_start - DPAA2_ETH_TX_BUF_ALIGN,
962 				  DPAA2_ETH_TX_BUF_ALIGN);
963 	if (aligned_start >= skb->head)
964 		buffer_start = aligned_start;
965 
966 	/* Store a backpointer to the skb at the beginning of the buffer
967 	 * (in the private data area) such that we can release it
968 	 * on Tx confirm
969 	 */
970 	*swa_addr = (void *)buffer_start;
971 	swa = (struct dpaa2_eth_swa *)buffer_start;
972 	swa->type = DPAA2_ETH_SWA_SINGLE;
973 	swa->single.skb = skb;
974 
975 	addr = dma_map_single(dev, buffer_start,
976 			      skb_tail_pointer(skb) - buffer_start,
977 			      DMA_BIDIRECTIONAL);
978 	if (unlikely(dma_mapping_error(dev, addr)))
979 		return -ENOMEM;
980 
981 	dpaa2_fd_set_addr(fd, addr);
982 	dpaa2_fd_set_offset(fd, (u16)(skb->data - buffer_start));
983 	dpaa2_fd_set_len(fd, skb->len);
984 	dpaa2_fd_set_format(fd, dpaa2_fd_single);
985 	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
986 
987 	return 0;
988 }
989 
990 /* FD freeing routine on the Tx path
991  *
992  * DMA-unmap and free FD and possibly SGT buffer allocated on Tx. The skb
993  * back-pointed to is also freed.
994  * This can be called either from dpaa2_eth_tx_conf() or on the error path of
995  * dpaa2_eth_tx().
996  */
997 static void dpaa2_eth_free_tx_fd(struct dpaa2_eth_priv *priv,
998 				 struct dpaa2_eth_fq *fq,
999 				 const struct dpaa2_fd *fd, bool in_napi)
1000 {
1001 	struct device *dev = priv->net_dev->dev.parent;
1002 	dma_addr_t fd_addr, sg_addr;
1003 	struct sk_buff *skb = NULL;
1004 	unsigned char *buffer_start;
1005 	struct dpaa2_eth_swa *swa;
1006 	u8 fd_format = dpaa2_fd_get_format(fd);
1007 	u32 fd_len = dpaa2_fd_get_len(fd);
1008 
1009 	struct dpaa2_eth_sgt_cache *sgt_cache;
1010 	struct dpaa2_sg_entry *sgt;
1011 
1012 	fd_addr = dpaa2_fd_get_addr(fd);
1013 	buffer_start = dpaa2_iova_to_virt(priv->iommu_domain, fd_addr);
1014 	swa = (struct dpaa2_eth_swa *)buffer_start;
1015 
1016 	if (fd_format == dpaa2_fd_single) {
1017 		if (swa->type == DPAA2_ETH_SWA_SINGLE) {
1018 			skb = swa->single.skb;
1019 			/* Accessing the skb buffer is safe before dma unmap,
1020 			 * because we didn't map the actual skb shell.
1021 			 */
1022 			dma_unmap_single(dev, fd_addr,
1023 					 skb_tail_pointer(skb) - buffer_start,
1024 					 DMA_BIDIRECTIONAL);
1025 		} else {
1026 			WARN_ONCE(swa->type != DPAA2_ETH_SWA_XDP, "Wrong SWA type");
1027 			dma_unmap_single(dev, fd_addr, swa->xdp.dma_size,
1028 					 DMA_BIDIRECTIONAL);
1029 		}
1030 	} else if (fd_format == dpaa2_fd_sg) {
1031 		if (swa->type == DPAA2_ETH_SWA_SG) {
1032 			skb = swa->sg.skb;
1033 
1034 			/* Unmap the scatterlist */
1035 			dma_unmap_sg(dev, swa->sg.scl, swa->sg.num_sg,
1036 				     DMA_BIDIRECTIONAL);
1037 			kfree(swa->sg.scl);
1038 
1039 			/* Unmap the SGT buffer */
1040 			dma_unmap_single(dev, fd_addr, swa->sg.sgt_size,
1041 					 DMA_BIDIRECTIONAL);
1042 		} else {
1043 			skb = swa->single.skb;
1044 
1045 			/* Unmap the SGT Buffer */
1046 			dma_unmap_single(dev, fd_addr, swa->single.sgt_size,
1047 					 DMA_BIDIRECTIONAL);
1048 
1049 			sgt = (struct dpaa2_sg_entry *)(buffer_start +
1050 							priv->tx_data_offset);
1051 			sg_addr = dpaa2_sg_get_addr(sgt);
1052 			dma_unmap_single(dev, sg_addr, skb->len, DMA_BIDIRECTIONAL);
1053 		}
1054 	} else {
1055 		netdev_dbg(priv->net_dev, "Invalid FD format\n");
1056 		return;
1057 	}
1058 
1059 	if (swa->type != DPAA2_ETH_SWA_XDP && in_napi) {
1060 		fq->dq_frames++;
1061 		fq->dq_bytes += fd_len;
1062 	}
1063 
1064 	if (swa->type == DPAA2_ETH_SWA_XDP) {
1065 		xdp_return_frame(swa->xdp.xdpf);
1066 		return;
1067 	}
1068 
1069 	/* Get the timestamp value */
1070 	if (skb->cb[0] == TX_TSTAMP) {
1071 		struct skb_shared_hwtstamps shhwtstamps;
1072 		__le64 *ts = dpaa2_get_ts(buffer_start, true);
1073 		u64 ns;
1074 
1075 		memset(&shhwtstamps, 0, sizeof(shhwtstamps));
1076 
1077 		ns = DPAA2_PTP_CLK_PERIOD_NS * le64_to_cpup(ts);
1078 		shhwtstamps.hwtstamp = ns_to_ktime(ns);
1079 		skb_tstamp_tx(skb, &shhwtstamps);
1080 	} else if (skb->cb[0] == TX_TSTAMP_ONESTEP_SYNC) {
1081 		mutex_unlock(&priv->onestep_tstamp_lock);
1082 	}
1083 
1084 	/* Free SGT buffer allocated on tx */
1085 	if (fd_format != dpaa2_fd_single) {
1086 		sgt_cache = this_cpu_ptr(priv->sgt_cache);
1087 		if (swa->type == DPAA2_ETH_SWA_SG) {
1088 			skb_free_frag(buffer_start);
1089 		} else {
1090 			if (sgt_cache->count >= DPAA2_ETH_SGT_CACHE_SIZE)
1091 				kfree(buffer_start);
1092 			else
1093 				sgt_cache->buf[sgt_cache->count++] = buffer_start;
1094 		}
1095 	}
1096 
1097 	/* Move on with skb release */
1098 	napi_consume_skb(skb, in_napi);
1099 }
1100 
1101 static netdev_tx_t __dpaa2_eth_tx(struct sk_buff *skb,
1102 				  struct net_device *net_dev)
1103 {
1104 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1105 	struct dpaa2_fd fd;
1106 	struct rtnl_link_stats64 *percpu_stats;
1107 	struct dpaa2_eth_drv_stats *percpu_extras;
1108 	struct dpaa2_eth_fq *fq;
1109 	struct netdev_queue *nq;
1110 	u16 queue_mapping;
1111 	unsigned int needed_headroom;
1112 	u32 fd_len;
1113 	u8 prio = 0;
1114 	int err, i;
1115 	void *swa;
1116 
1117 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
1118 	percpu_extras = this_cpu_ptr(priv->percpu_extras);
1119 
1120 	needed_headroom = dpaa2_eth_needed_headroom(skb);
1121 
1122 	/* We'll be holding a back-reference to the skb until Tx Confirmation;
1123 	 * we don't want that overwritten by a concurrent Tx with a cloned skb.
1124 	 */
1125 	skb = skb_unshare(skb, GFP_ATOMIC);
1126 	if (unlikely(!skb)) {
1127 		/* skb_unshare() has already freed the skb */
1128 		percpu_stats->tx_dropped++;
1129 		return NETDEV_TX_OK;
1130 	}
1131 
1132 	/* Setup the FD fields */
1133 	memset(&fd, 0, sizeof(fd));
1134 
1135 	if (skb_is_nonlinear(skb)) {
1136 		err = dpaa2_eth_build_sg_fd(priv, skb, &fd, &swa);
1137 		percpu_extras->tx_sg_frames++;
1138 		percpu_extras->tx_sg_bytes += skb->len;
1139 	} else if (skb_headroom(skb) < needed_headroom) {
1140 		err = dpaa2_eth_build_sg_fd_single_buf(priv, skb, &fd, &swa);
1141 		percpu_extras->tx_sg_frames++;
1142 		percpu_extras->tx_sg_bytes += skb->len;
1143 		percpu_extras->tx_converted_sg_frames++;
1144 		percpu_extras->tx_converted_sg_bytes += skb->len;
1145 	} else {
1146 		err = dpaa2_eth_build_single_fd(priv, skb, &fd, &swa);
1147 	}
1148 
1149 	if (unlikely(err)) {
1150 		percpu_stats->tx_dropped++;
1151 		goto err_build_fd;
1152 	}
1153 
1154 	if (skb->cb[0])
1155 		dpaa2_eth_enable_tx_tstamp(priv, &fd, swa, skb);
1156 
1157 	/* Tracing point */
1158 	trace_dpaa2_tx_fd(net_dev, &fd);
1159 
1160 	/* TxConf FQ selection relies on queue id from the stack.
1161 	 * In case of a forwarded frame from another DPNI interface, we choose
1162 	 * a queue affined to the same core that processed the Rx frame
1163 	 */
1164 	queue_mapping = skb_get_queue_mapping(skb);
1165 
1166 	if (net_dev->num_tc) {
1167 		prio = netdev_txq_to_tc(net_dev, queue_mapping);
1168 		/* Hardware interprets priority level 0 as being the highest,
1169 		 * so we need to do a reverse mapping to the netdev tc index
1170 		 */
1171 		prio = net_dev->num_tc - prio - 1;
1172 		/* We have only one FQ array entry for all Tx hardware queues
1173 		 * with the same flow id (but different priority levels)
1174 		 */
1175 		queue_mapping %= dpaa2_eth_queue_count(priv);
1176 	}
1177 	fq = &priv->fq[queue_mapping];
1178 
1179 	fd_len = dpaa2_fd_get_len(&fd);
1180 	nq = netdev_get_tx_queue(net_dev, queue_mapping);
1181 	netdev_tx_sent_queue(nq, fd_len);
1182 
1183 	/* Everything that happens after this enqueues might race with
1184 	 * the Tx confirmation callback for this frame
1185 	 */
1186 	for (i = 0; i < DPAA2_ETH_ENQUEUE_RETRIES; i++) {
1187 		err = priv->enqueue(priv, fq, &fd, prio, 1, NULL);
1188 		if (err != -EBUSY)
1189 			break;
1190 	}
1191 	percpu_extras->tx_portal_busy += i;
1192 	if (unlikely(err < 0)) {
1193 		percpu_stats->tx_errors++;
1194 		/* Clean up everything, including freeing the skb */
1195 		dpaa2_eth_free_tx_fd(priv, fq, &fd, false);
1196 		netdev_tx_completed_queue(nq, 1, fd_len);
1197 	} else {
1198 		percpu_stats->tx_packets++;
1199 		percpu_stats->tx_bytes += fd_len;
1200 	}
1201 
1202 	return NETDEV_TX_OK;
1203 
1204 err_build_fd:
1205 	dev_kfree_skb(skb);
1206 
1207 	return NETDEV_TX_OK;
1208 }
1209 
1210 static void dpaa2_eth_tx_onestep_tstamp(struct work_struct *work)
1211 {
1212 	struct dpaa2_eth_priv *priv = container_of(work, struct dpaa2_eth_priv,
1213 						   tx_onestep_tstamp);
1214 	struct sk_buff *skb;
1215 
1216 	while (true) {
1217 		skb = skb_dequeue(&priv->tx_skbs);
1218 		if (!skb)
1219 			return;
1220 
1221 		/* Lock just before TX one-step timestamping packet,
1222 		 * and release the lock in dpaa2_eth_free_tx_fd when
1223 		 * confirm the packet has been sent on hardware, or
1224 		 * when clean up during transmit failure.
1225 		 */
1226 		mutex_lock(&priv->onestep_tstamp_lock);
1227 		__dpaa2_eth_tx(skb, priv->net_dev);
1228 	}
1229 }
1230 
1231 static netdev_tx_t dpaa2_eth_tx(struct sk_buff *skb, struct net_device *net_dev)
1232 {
1233 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1234 	u8 msgtype, twostep, udp;
1235 	u16 offset1, offset2;
1236 
1237 	/* Utilize skb->cb[0] for timestamping request per skb */
1238 	skb->cb[0] = 0;
1239 
1240 	if ((skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) && dpaa2_ptp) {
1241 		if (priv->tx_tstamp_type == HWTSTAMP_TX_ON)
1242 			skb->cb[0] = TX_TSTAMP;
1243 		else if (priv->tx_tstamp_type == HWTSTAMP_TX_ONESTEP_SYNC)
1244 			skb->cb[0] = TX_TSTAMP_ONESTEP_SYNC;
1245 	}
1246 
1247 	/* TX for one-step timestamping PTP Sync packet */
1248 	if (skb->cb[0] == TX_TSTAMP_ONESTEP_SYNC) {
1249 		if (!dpaa2_eth_ptp_parse(skb, &msgtype, &twostep, &udp,
1250 					 &offset1, &offset2))
1251 			if (msgtype == PTP_MSGTYPE_SYNC && twostep == 0) {
1252 				skb_queue_tail(&priv->tx_skbs, skb);
1253 				queue_work(priv->dpaa2_ptp_wq,
1254 					   &priv->tx_onestep_tstamp);
1255 				return NETDEV_TX_OK;
1256 			}
1257 		/* Use two-step timestamping if not one-step timestamping
1258 		 * PTP Sync packet
1259 		 */
1260 		skb->cb[0] = TX_TSTAMP;
1261 	}
1262 
1263 	/* TX for other packets */
1264 	return __dpaa2_eth_tx(skb, net_dev);
1265 }
1266 
1267 /* Tx confirmation frame processing routine */
1268 static void dpaa2_eth_tx_conf(struct dpaa2_eth_priv *priv,
1269 			      struct dpaa2_eth_channel *ch,
1270 			      const struct dpaa2_fd *fd,
1271 			      struct dpaa2_eth_fq *fq)
1272 {
1273 	struct rtnl_link_stats64 *percpu_stats;
1274 	struct dpaa2_eth_drv_stats *percpu_extras;
1275 	u32 fd_len = dpaa2_fd_get_len(fd);
1276 	u32 fd_errors;
1277 
1278 	/* Tracing point */
1279 	trace_dpaa2_tx_conf_fd(priv->net_dev, fd);
1280 
1281 	percpu_extras = this_cpu_ptr(priv->percpu_extras);
1282 	percpu_extras->tx_conf_frames++;
1283 	percpu_extras->tx_conf_bytes += fd_len;
1284 	ch->stats.bytes_per_cdan += fd_len;
1285 
1286 	/* Check frame errors in the FD field */
1287 	fd_errors = dpaa2_fd_get_ctrl(fd) & DPAA2_FD_TX_ERR_MASK;
1288 	dpaa2_eth_free_tx_fd(priv, fq, fd, true);
1289 
1290 	if (likely(!fd_errors))
1291 		return;
1292 
1293 	if (net_ratelimit())
1294 		netdev_dbg(priv->net_dev, "TX frame FD error: 0x%08x\n",
1295 			   fd_errors);
1296 
1297 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
1298 	/* Tx-conf logically pertains to the egress path. */
1299 	percpu_stats->tx_errors++;
1300 }
1301 
1302 static int dpaa2_eth_set_rx_vlan_filtering(struct dpaa2_eth_priv *priv,
1303 					   bool enable)
1304 {
1305 	int err;
1306 
1307 	err = dpni_enable_vlan_filter(priv->mc_io, 0, priv->mc_token, enable);
1308 
1309 	if (err) {
1310 		netdev_err(priv->net_dev,
1311 			   "dpni_enable_vlan_filter failed\n");
1312 		return err;
1313 	}
1314 
1315 	return 0;
1316 }
1317 
1318 static int dpaa2_eth_set_rx_csum(struct dpaa2_eth_priv *priv, bool enable)
1319 {
1320 	int err;
1321 
1322 	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1323 			       DPNI_OFF_RX_L3_CSUM, enable);
1324 	if (err) {
1325 		netdev_err(priv->net_dev,
1326 			   "dpni_set_offload(RX_L3_CSUM) failed\n");
1327 		return err;
1328 	}
1329 
1330 	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1331 			       DPNI_OFF_RX_L4_CSUM, enable);
1332 	if (err) {
1333 		netdev_err(priv->net_dev,
1334 			   "dpni_set_offload(RX_L4_CSUM) failed\n");
1335 		return err;
1336 	}
1337 
1338 	return 0;
1339 }
1340 
1341 static int dpaa2_eth_set_tx_csum(struct dpaa2_eth_priv *priv, bool enable)
1342 {
1343 	int err;
1344 
1345 	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1346 			       DPNI_OFF_TX_L3_CSUM, enable);
1347 	if (err) {
1348 		netdev_err(priv->net_dev, "dpni_set_offload(TX_L3_CSUM) failed\n");
1349 		return err;
1350 	}
1351 
1352 	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1353 			       DPNI_OFF_TX_L4_CSUM, enable);
1354 	if (err) {
1355 		netdev_err(priv->net_dev, "dpni_set_offload(TX_L4_CSUM) failed\n");
1356 		return err;
1357 	}
1358 
1359 	return 0;
1360 }
1361 
1362 /* Perform a single release command to add buffers
1363  * to the specified buffer pool
1364  */
1365 static int dpaa2_eth_add_bufs(struct dpaa2_eth_priv *priv,
1366 			      struct dpaa2_eth_channel *ch, u16 bpid)
1367 {
1368 	struct device *dev = priv->net_dev->dev.parent;
1369 	u64 buf_array[DPAA2_ETH_BUFS_PER_CMD];
1370 	struct page *page;
1371 	dma_addr_t addr;
1372 	int retries = 0;
1373 	int i, err;
1374 
1375 	for (i = 0; i < DPAA2_ETH_BUFS_PER_CMD; i++) {
1376 		/* Allocate buffer visible to WRIOP + skb shared info +
1377 		 * alignment padding
1378 		 */
1379 		/* allocate one page for each Rx buffer. WRIOP sees
1380 		 * the entire page except for a tailroom reserved for
1381 		 * skb shared info
1382 		 */
1383 		page = dev_alloc_pages(0);
1384 		if (!page)
1385 			goto err_alloc;
1386 
1387 		addr = dma_map_page(dev, page, 0, priv->rx_buf_size,
1388 				    DMA_BIDIRECTIONAL);
1389 		if (unlikely(dma_mapping_error(dev, addr)))
1390 			goto err_map;
1391 
1392 		buf_array[i] = addr;
1393 
1394 		/* tracing point */
1395 		trace_dpaa2_eth_buf_seed(priv->net_dev,
1396 					 page, DPAA2_ETH_RX_BUF_RAW_SIZE,
1397 					 addr, priv->rx_buf_size,
1398 					 bpid);
1399 	}
1400 
1401 release_bufs:
1402 	/* In case the portal is busy, retry until successful */
1403 	while ((err = dpaa2_io_service_release(ch->dpio, bpid,
1404 					       buf_array, i)) == -EBUSY) {
1405 		if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES)
1406 			break;
1407 		cpu_relax();
1408 	}
1409 
1410 	/* If release command failed, clean up and bail out;
1411 	 * not much else we can do about it
1412 	 */
1413 	if (err) {
1414 		dpaa2_eth_free_bufs(priv, buf_array, i);
1415 		return 0;
1416 	}
1417 
1418 	return i;
1419 
1420 err_map:
1421 	__free_pages(page, 0);
1422 err_alloc:
1423 	/* If we managed to allocate at least some buffers,
1424 	 * release them to hardware
1425 	 */
1426 	if (i)
1427 		goto release_bufs;
1428 
1429 	return 0;
1430 }
1431 
1432 static int dpaa2_eth_seed_pool(struct dpaa2_eth_priv *priv, u16 bpid)
1433 {
1434 	int i, j;
1435 	int new_count;
1436 
1437 	for (j = 0; j < priv->num_channels; j++) {
1438 		for (i = 0; i < DPAA2_ETH_NUM_BUFS;
1439 		     i += DPAA2_ETH_BUFS_PER_CMD) {
1440 			new_count = dpaa2_eth_add_bufs(priv, priv->channel[j], bpid);
1441 			priv->channel[j]->buf_count += new_count;
1442 
1443 			if (new_count < DPAA2_ETH_BUFS_PER_CMD) {
1444 				return -ENOMEM;
1445 			}
1446 		}
1447 	}
1448 
1449 	return 0;
1450 }
1451 
1452 /*
1453  * Drain the specified number of buffers from the DPNI's private buffer pool.
1454  * @count must not exceeed DPAA2_ETH_BUFS_PER_CMD
1455  */
1456 static void dpaa2_eth_drain_bufs(struct dpaa2_eth_priv *priv, int count)
1457 {
1458 	u64 buf_array[DPAA2_ETH_BUFS_PER_CMD];
1459 	int retries = 0;
1460 	int ret;
1461 
1462 	do {
1463 		ret = dpaa2_io_service_acquire(NULL, priv->bpid,
1464 					       buf_array, count);
1465 		if (ret < 0) {
1466 			if (ret == -EBUSY &&
1467 			    retries++ < DPAA2_ETH_SWP_BUSY_RETRIES)
1468 				continue;
1469 			netdev_err(priv->net_dev, "dpaa2_io_service_acquire() failed\n");
1470 			return;
1471 		}
1472 		dpaa2_eth_free_bufs(priv, buf_array, ret);
1473 		retries = 0;
1474 	} while (ret);
1475 }
1476 
1477 static void dpaa2_eth_drain_pool(struct dpaa2_eth_priv *priv)
1478 {
1479 	int i;
1480 
1481 	dpaa2_eth_drain_bufs(priv, DPAA2_ETH_BUFS_PER_CMD);
1482 	dpaa2_eth_drain_bufs(priv, 1);
1483 
1484 	for (i = 0; i < priv->num_channels; i++)
1485 		priv->channel[i]->buf_count = 0;
1486 }
1487 
1488 /* Function is called from softirq context only, so we don't need to guard
1489  * the access to percpu count
1490  */
1491 static int dpaa2_eth_refill_pool(struct dpaa2_eth_priv *priv,
1492 				 struct dpaa2_eth_channel *ch,
1493 				 u16 bpid)
1494 {
1495 	int new_count;
1496 
1497 	if (likely(ch->buf_count >= DPAA2_ETH_REFILL_THRESH))
1498 		return 0;
1499 
1500 	do {
1501 		new_count = dpaa2_eth_add_bufs(priv, ch, bpid);
1502 		if (unlikely(!new_count)) {
1503 			/* Out of memory; abort for now, we'll try later on */
1504 			break;
1505 		}
1506 		ch->buf_count += new_count;
1507 	} while (ch->buf_count < DPAA2_ETH_NUM_BUFS);
1508 
1509 	if (unlikely(ch->buf_count < DPAA2_ETH_NUM_BUFS))
1510 		return -ENOMEM;
1511 
1512 	return 0;
1513 }
1514 
1515 static void dpaa2_eth_sgt_cache_drain(struct dpaa2_eth_priv *priv)
1516 {
1517 	struct dpaa2_eth_sgt_cache *sgt_cache;
1518 	u16 count;
1519 	int k, i;
1520 
1521 	for_each_possible_cpu(k) {
1522 		sgt_cache = per_cpu_ptr(priv->sgt_cache, k);
1523 		count = sgt_cache->count;
1524 
1525 		for (i = 0; i < count; i++)
1526 			kfree(sgt_cache->buf[i]);
1527 		sgt_cache->count = 0;
1528 	}
1529 }
1530 
1531 static int dpaa2_eth_pull_channel(struct dpaa2_eth_channel *ch)
1532 {
1533 	int err;
1534 	int dequeues = -1;
1535 
1536 	/* Retry while portal is busy */
1537 	do {
1538 		err = dpaa2_io_service_pull_channel(ch->dpio, ch->ch_id,
1539 						    ch->store);
1540 		dequeues++;
1541 		cpu_relax();
1542 	} while (err == -EBUSY && dequeues < DPAA2_ETH_SWP_BUSY_RETRIES);
1543 
1544 	ch->stats.dequeue_portal_busy += dequeues;
1545 	if (unlikely(err))
1546 		ch->stats.pull_err++;
1547 
1548 	return err;
1549 }
1550 
1551 /* NAPI poll routine
1552  *
1553  * Frames are dequeued from the QMan channel associated with this NAPI context.
1554  * Rx, Tx confirmation and (if configured) Rx error frames all count
1555  * towards the NAPI budget.
1556  */
1557 static int dpaa2_eth_poll(struct napi_struct *napi, int budget)
1558 {
1559 	struct dpaa2_eth_channel *ch;
1560 	struct dpaa2_eth_priv *priv;
1561 	int rx_cleaned = 0, txconf_cleaned = 0;
1562 	struct dpaa2_eth_fq *fq, *txc_fq = NULL;
1563 	struct netdev_queue *nq;
1564 	int store_cleaned, work_done;
1565 	struct list_head rx_list;
1566 	int retries = 0;
1567 	u16 flowid;
1568 	int err;
1569 
1570 	ch = container_of(napi, struct dpaa2_eth_channel, napi);
1571 	ch->xdp.res = 0;
1572 	priv = ch->priv;
1573 
1574 	INIT_LIST_HEAD(&rx_list);
1575 	ch->rx_list = &rx_list;
1576 
1577 	do {
1578 		err = dpaa2_eth_pull_channel(ch);
1579 		if (unlikely(err))
1580 			break;
1581 
1582 		/* Refill pool if appropriate */
1583 		dpaa2_eth_refill_pool(priv, ch, priv->bpid);
1584 
1585 		store_cleaned = dpaa2_eth_consume_frames(ch, &fq);
1586 		if (store_cleaned <= 0)
1587 			break;
1588 		if (fq->type == DPAA2_RX_FQ) {
1589 			rx_cleaned += store_cleaned;
1590 			flowid = fq->flowid;
1591 		} else {
1592 			txconf_cleaned += store_cleaned;
1593 			/* We have a single Tx conf FQ on this channel */
1594 			txc_fq = fq;
1595 		}
1596 
1597 		/* If we either consumed the whole NAPI budget with Rx frames
1598 		 * or we reached the Tx confirmations threshold, we're done.
1599 		 */
1600 		if (rx_cleaned >= budget ||
1601 		    txconf_cleaned >= DPAA2_ETH_TXCONF_PER_NAPI) {
1602 			work_done = budget;
1603 			goto out;
1604 		}
1605 	} while (store_cleaned);
1606 
1607 	/* Update NET DIM with the values for this CDAN */
1608 	dpaa2_io_update_net_dim(ch->dpio, ch->stats.frames_per_cdan,
1609 				ch->stats.bytes_per_cdan);
1610 	ch->stats.frames_per_cdan = 0;
1611 	ch->stats.bytes_per_cdan = 0;
1612 
1613 	/* We didn't consume the entire budget, so finish napi and
1614 	 * re-enable data availability notifications
1615 	 */
1616 	napi_complete_done(napi, rx_cleaned);
1617 	do {
1618 		err = dpaa2_io_service_rearm(ch->dpio, &ch->nctx);
1619 		cpu_relax();
1620 	} while (err == -EBUSY && retries++ < DPAA2_ETH_SWP_BUSY_RETRIES);
1621 	WARN_ONCE(err, "CDAN notifications rearm failed on core %d",
1622 		  ch->nctx.desired_cpu);
1623 
1624 	work_done = max(rx_cleaned, 1);
1625 
1626 out:
1627 	netif_receive_skb_list(ch->rx_list);
1628 
1629 	if (txc_fq && txc_fq->dq_frames) {
1630 		nq = netdev_get_tx_queue(priv->net_dev, txc_fq->flowid);
1631 		netdev_tx_completed_queue(nq, txc_fq->dq_frames,
1632 					  txc_fq->dq_bytes);
1633 		txc_fq->dq_frames = 0;
1634 		txc_fq->dq_bytes = 0;
1635 	}
1636 
1637 	if (ch->xdp.res & XDP_REDIRECT)
1638 		xdp_do_flush_map();
1639 	else if (rx_cleaned && ch->xdp.res & XDP_TX)
1640 		dpaa2_eth_xdp_tx_flush(priv, ch, &priv->fq[flowid]);
1641 
1642 	return work_done;
1643 }
1644 
1645 static void dpaa2_eth_enable_ch_napi(struct dpaa2_eth_priv *priv)
1646 {
1647 	struct dpaa2_eth_channel *ch;
1648 	int i;
1649 
1650 	for (i = 0; i < priv->num_channels; i++) {
1651 		ch = priv->channel[i];
1652 		napi_enable(&ch->napi);
1653 	}
1654 }
1655 
1656 static void dpaa2_eth_disable_ch_napi(struct dpaa2_eth_priv *priv)
1657 {
1658 	struct dpaa2_eth_channel *ch;
1659 	int i;
1660 
1661 	for (i = 0; i < priv->num_channels; i++) {
1662 		ch = priv->channel[i];
1663 		napi_disable(&ch->napi);
1664 	}
1665 }
1666 
1667 void dpaa2_eth_set_rx_taildrop(struct dpaa2_eth_priv *priv,
1668 			       bool tx_pause, bool pfc)
1669 {
1670 	struct dpni_taildrop td = {0};
1671 	struct dpaa2_eth_fq *fq;
1672 	int i, err;
1673 
1674 	/* FQ taildrop: threshold is in bytes, per frame queue. Enabled if
1675 	 * flow control is disabled (as it might interfere with either the
1676 	 * buffer pool depletion trigger for pause frames or with the group
1677 	 * congestion trigger for PFC frames)
1678 	 */
1679 	td.enable = !tx_pause;
1680 	if (priv->rx_fqtd_enabled == td.enable)
1681 		goto set_cgtd;
1682 
1683 	td.threshold = DPAA2_ETH_FQ_TAILDROP_THRESH;
1684 	td.units = DPNI_CONGESTION_UNIT_BYTES;
1685 
1686 	for (i = 0; i < priv->num_fqs; i++) {
1687 		fq = &priv->fq[i];
1688 		if (fq->type != DPAA2_RX_FQ)
1689 			continue;
1690 		err = dpni_set_taildrop(priv->mc_io, 0, priv->mc_token,
1691 					DPNI_CP_QUEUE, DPNI_QUEUE_RX,
1692 					fq->tc, fq->flowid, &td);
1693 		if (err) {
1694 			netdev_err(priv->net_dev,
1695 				   "dpni_set_taildrop(FQ) failed\n");
1696 			return;
1697 		}
1698 	}
1699 
1700 	priv->rx_fqtd_enabled = td.enable;
1701 
1702 set_cgtd:
1703 	/* Congestion group taildrop: threshold is in frames, per group
1704 	 * of FQs belonging to the same traffic class
1705 	 * Enabled if general Tx pause disabled or if PFCs are enabled
1706 	 * (congestion group threhsold for PFC generation is lower than the
1707 	 * CG taildrop threshold, so it won't interfere with it; we also
1708 	 * want frames in non-PFC enabled traffic classes to be kept in check)
1709 	 */
1710 	td.enable = !tx_pause || pfc;
1711 	if (priv->rx_cgtd_enabled == td.enable)
1712 		return;
1713 
1714 	td.threshold = DPAA2_ETH_CG_TAILDROP_THRESH(priv);
1715 	td.units = DPNI_CONGESTION_UNIT_FRAMES;
1716 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
1717 		err = dpni_set_taildrop(priv->mc_io, 0, priv->mc_token,
1718 					DPNI_CP_GROUP, DPNI_QUEUE_RX,
1719 					i, 0, &td);
1720 		if (err) {
1721 			netdev_err(priv->net_dev,
1722 				   "dpni_set_taildrop(CG) failed\n");
1723 			return;
1724 		}
1725 	}
1726 
1727 	priv->rx_cgtd_enabled = td.enable;
1728 }
1729 
1730 static int dpaa2_eth_link_state_update(struct dpaa2_eth_priv *priv)
1731 {
1732 	struct dpni_link_state state = {0};
1733 	bool tx_pause;
1734 	int err;
1735 
1736 	err = dpni_get_link_state(priv->mc_io, 0, priv->mc_token, &state);
1737 	if (unlikely(err)) {
1738 		netdev_err(priv->net_dev,
1739 			   "dpni_get_link_state() failed\n");
1740 		return err;
1741 	}
1742 
1743 	/* If Tx pause frame settings have changed, we need to update
1744 	 * Rx FQ taildrop configuration as well. We configure taildrop
1745 	 * only when pause frame generation is disabled.
1746 	 */
1747 	tx_pause = dpaa2_eth_tx_pause_enabled(state.options);
1748 	dpaa2_eth_set_rx_taildrop(priv, tx_pause, priv->pfc_enabled);
1749 
1750 	/* When we manage the MAC/PHY using phylink there is no need
1751 	 * to manually update the netif_carrier.
1752 	 */
1753 	if (dpaa2_eth_is_type_phy(priv))
1754 		goto out;
1755 
1756 	/* Chech link state; speed / duplex changes are not treated yet */
1757 	if (priv->link_state.up == state.up)
1758 		goto out;
1759 
1760 	if (state.up) {
1761 		netif_carrier_on(priv->net_dev);
1762 		netif_tx_start_all_queues(priv->net_dev);
1763 	} else {
1764 		netif_tx_stop_all_queues(priv->net_dev);
1765 		netif_carrier_off(priv->net_dev);
1766 	}
1767 
1768 	netdev_info(priv->net_dev, "Link Event: state %s\n",
1769 		    state.up ? "up" : "down");
1770 
1771 out:
1772 	priv->link_state = state;
1773 
1774 	return 0;
1775 }
1776 
1777 static int dpaa2_eth_open(struct net_device *net_dev)
1778 {
1779 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1780 	int err;
1781 
1782 	err = dpaa2_eth_seed_pool(priv, priv->bpid);
1783 	if (err) {
1784 		/* Not much to do; the buffer pool, though not filled up,
1785 		 * may still contain some buffers which would enable us
1786 		 * to limp on.
1787 		 */
1788 		netdev_err(net_dev, "Buffer seeding failed for DPBP %d (bpid=%d)\n",
1789 			   priv->dpbp_dev->obj_desc.id, priv->bpid);
1790 	}
1791 
1792 	if (!dpaa2_eth_is_type_phy(priv)) {
1793 		/* We'll only start the txqs when the link is actually ready;
1794 		 * make sure we don't race against the link up notification,
1795 		 * which may come immediately after dpni_enable();
1796 		 */
1797 		netif_tx_stop_all_queues(net_dev);
1798 
1799 		/* Also, explicitly set carrier off, otherwise
1800 		 * netif_carrier_ok() will return true and cause 'ip link show'
1801 		 * to report the LOWER_UP flag, even though the link
1802 		 * notification wasn't even received.
1803 		 */
1804 		netif_carrier_off(net_dev);
1805 	}
1806 	dpaa2_eth_enable_ch_napi(priv);
1807 
1808 	err = dpni_enable(priv->mc_io, 0, priv->mc_token);
1809 	if (err < 0) {
1810 		netdev_err(net_dev, "dpni_enable() failed\n");
1811 		goto enable_err;
1812 	}
1813 
1814 	if (dpaa2_eth_is_type_phy(priv))
1815 		phylink_start(priv->mac->phylink);
1816 
1817 	return 0;
1818 
1819 enable_err:
1820 	dpaa2_eth_disable_ch_napi(priv);
1821 	dpaa2_eth_drain_pool(priv);
1822 	return err;
1823 }
1824 
1825 /* Total number of in-flight frames on ingress queues */
1826 static u32 dpaa2_eth_ingress_fq_count(struct dpaa2_eth_priv *priv)
1827 {
1828 	struct dpaa2_eth_fq *fq;
1829 	u32 fcnt = 0, bcnt = 0, total = 0;
1830 	int i, err;
1831 
1832 	for (i = 0; i < priv->num_fqs; i++) {
1833 		fq = &priv->fq[i];
1834 		err = dpaa2_io_query_fq_count(NULL, fq->fqid, &fcnt, &bcnt);
1835 		if (err) {
1836 			netdev_warn(priv->net_dev, "query_fq_count failed");
1837 			break;
1838 		}
1839 		total += fcnt;
1840 	}
1841 
1842 	return total;
1843 }
1844 
1845 static void dpaa2_eth_wait_for_ingress_fq_empty(struct dpaa2_eth_priv *priv)
1846 {
1847 	int retries = 10;
1848 	u32 pending;
1849 
1850 	do {
1851 		pending = dpaa2_eth_ingress_fq_count(priv);
1852 		if (pending)
1853 			msleep(100);
1854 	} while (pending && --retries);
1855 }
1856 
1857 #define DPNI_TX_PENDING_VER_MAJOR	7
1858 #define DPNI_TX_PENDING_VER_MINOR	13
1859 static void dpaa2_eth_wait_for_egress_fq_empty(struct dpaa2_eth_priv *priv)
1860 {
1861 	union dpni_statistics stats;
1862 	int retries = 10;
1863 	int err;
1864 
1865 	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_TX_PENDING_VER_MAJOR,
1866 				   DPNI_TX_PENDING_VER_MINOR) < 0)
1867 		goto out;
1868 
1869 	do {
1870 		err = dpni_get_statistics(priv->mc_io, 0, priv->mc_token, 6,
1871 					  &stats);
1872 		if (err)
1873 			goto out;
1874 		if (stats.page_6.tx_pending_frames == 0)
1875 			return;
1876 	} while (--retries);
1877 
1878 out:
1879 	msleep(500);
1880 }
1881 
1882 static int dpaa2_eth_stop(struct net_device *net_dev)
1883 {
1884 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1885 	int dpni_enabled = 0;
1886 	int retries = 10;
1887 
1888 	if (dpaa2_eth_is_type_phy(priv)) {
1889 		phylink_stop(priv->mac->phylink);
1890 	} else {
1891 		netif_tx_stop_all_queues(net_dev);
1892 		netif_carrier_off(net_dev);
1893 	}
1894 
1895 	/* On dpni_disable(), the MC firmware will:
1896 	 * - stop MAC Rx and wait for all Rx frames to be enqueued to software
1897 	 * - cut off WRIOP dequeues from egress FQs and wait until transmission
1898 	 * of all in flight Tx frames is finished (and corresponding Tx conf
1899 	 * frames are enqueued back to software)
1900 	 *
1901 	 * Before calling dpni_disable(), we wait for all Tx frames to arrive
1902 	 * on WRIOP. After it finishes, wait until all remaining frames on Rx
1903 	 * and Tx conf queues are consumed on NAPI poll.
1904 	 */
1905 	dpaa2_eth_wait_for_egress_fq_empty(priv);
1906 
1907 	do {
1908 		dpni_disable(priv->mc_io, 0, priv->mc_token);
1909 		dpni_is_enabled(priv->mc_io, 0, priv->mc_token, &dpni_enabled);
1910 		if (dpni_enabled)
1911 			/* Allow the hardware some slack */
1912 			msleep(100);
1913 	} while (dpni_enabled && --retries);
1914 	if (!retries) {
1915 		netdev_warn(net_dev, "Retry count exceeded disabling DPNI\n");
1916 		/* Must go on and disable NAPI nonetheless, so we don't crash at
1917 		 * the next "ifconfig up"
1918 		 */
1919 	}
1920 
1921 	dpaa2_eth_wait_for_ingress_fq_empty(priv);
1922 	dpaa2_eth_disable_ch_napi(priv);
1923 
1924 	/* Empty the buffer pool */
1925 	dpaa2_eth_drain_pool(priv);
1926 
1927 	/* Empty the Scatter-Gather Buffer cache */
1928 	dpaa2_eth_sgt_cache_drain(priv);
1929 
1930 	return 0;
1931 }
1932 
1933 static int dpaa2_eth_set_addr(struct net_device *net_dev, void *addr)
1934 {
1935 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1936 	struct device *dev = net_dev->dev.parent;
1937 	int err;
1938 
1939 	err = eth_mac_addr(net_dev, addr);
1940 	if (err < 0) {
1941 		dev_err(dev, "eth_mac_addr() failed (%d)\n", err);
1942 		return err;
1943 	}
1944 
1945 	err = dpni_set_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
1946 					net_dev->dev_addr);
1947 	if (err) {
1948 		dev_err(dev, "dpni_set_primary_mac_addr() failed (%d)\n", err);
1949 		return err;
1950 	}
1951 
1952 	return 0;
1953 }
1954 
1955 /** Fill in counters maintained by the GPP driver. These may be different from
1956  * the hardware counters obtained by ethtool.
1957  */
1958 static void dpaa2_eth_get_stats(struct net_device *net_dev,
1959 				struct rtnl_link_stats64 *stats)
1960 {
1961 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1962 	struct rtnl_link_stats64 *percpu_stats;
1963 	u64 *cpustats;
1964 	u64 *netstats = (u64 *)stats;
1965 	int i, j;
1966 	int num = sizeof(struct rtnl_link_stats64) / sizeof(u64);
1967 
1968 	for_each_possible_cpu(i) {
1969 		percpu_stats = per_cpu_ptr(priv->percpu_stats, i);
1970 		cpustats = (u64 *)percpu_stats;
1971 		for (j = 0; j < num; j++)
1972 			netstats[j] += cpustats[j];
1973 	}
1974 }
1975 
1976 /* Copy mac unicast addresses from @net_dev to @priv.
1977  * Its sole purpose is to make dpaa2_eth_set_rx_mode() more readable.
1978  */
1979 static void dpaa2_eth_add_uc_hw_addr(const struct net_device *net_dev,
1980 				     struct dpaa2_eth_priv *priv)
1981 {
1982 	struct netdev_hw_addr *ha;
1983 	int err;
1984 
1985 	netdev_for_each_uc_addr(ha, net_dev) {
1986 		err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token,
1987 					ha->addr);
1988 		if (err)
1989 			netdev_warn(priv->net_dev,
1990 				    "Could not add ucast MAC %pM to the filtering table (err %d)\n",
1991 				    ha->addr, err);
1992 	}
1993 }
1994 
1995 /* Copy mac multicast addresses from @net_dev to @priv
1996  * Its sole purpose is to make dpaa2_eth_set_rx_mode() more readable.
1997  */
1998 static void dpaa2_eth_add_mc_hw_addr(const struct net_device *net_dev,
1999 				     struct dpaa2_eth_priv *priv)
2000 {
2001 	struct netdev_hw_addr *ha;
2002 	int err;
2003 
2004 	netdev_for_each_mc_addr(ha, net_dev) {
2005 		err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token,
2006 					ha->addr);
2007 		if (err)
2008 			netdev_warn(priv->net_dev,
2009 				    "Could not add mcast MAC %pM to the filtering table (err %d)\n",
2010 				    ha->addr, err);
2011 	}
2012 }
2013 
2014 static int dpaa2_eth_rx_add_vid(struct net_device *net_dev,
2015 				__be16 vlan_proto, u16 vid)
2016 {
2017 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2018 	int err;
2019 
2020 	err = dpni_add_vlan_id(priv->mc_io, 0, priv->mc_token,
2021 			       vid, 0, 0, 0);
2022 
2023 	if (err) {
2024 		netdev_warn(priv->net_dev,
2025 			    "Could not add the vlan id %u\n",
2026 			    vid);
2027 		return err;
2028 	}
2029 
2030 	return 0;
2031 }
2032 
2033 static int dpaa2_eth_rx_kill_vid(struct net_device *net_dev,
2034 				 __be16 vlan_proto, u16 vid)
2035 {
2036 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2037 	int err;
2038 
2039 	err = dpni_remove_vlan_id(priv->mc_io, 0, priv->mc_token, vid);
2040 
2041 	if (err) {
2042 		netdev_warn(priv->net_dev,
2043 			    "Could not remove the vlan id %u\n",
2044 			    vid);
2045 		return err;
2046 	}
2047 
2048 	return 0;
2049 }
2050 
2051 static void dpaa2_eth_set_rx_mode(struct net_device *net_dev)
2052 {
2053 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2054 	int uc_count = netdev_uc_count(net_dev);
2055 	int mc_count = netdev_mc_count(net_dev);
2056 	u8 max_mac = priv->dpni_attrs.mac_filter_entries;
2057 	u32 options = priv->dpni_attrs.options;
2058 	u16 mc_token = priv->mc_token;
2059 	struct fsl_mc_io *mc_io = priv->mc_io;
2060 	int err;
2061 
2062 	/* Basic sanity checks; these probably indicate a misconfiguration */
2063 	if (options & DPNI_OPT_NO_MAC_FILTER && max_mac != 0)
2064 		netdev_info(net_dev,
2065 			    "mac_filter_entries=%d, DPNI_OPT_NO_MAC_FILTER option must be disabled\n",
2066 			    max_mac);
2067 
2068 	/* Force promiscuous if the uc or mc counts exceed our capabilities. */
2069 	if (uc_count > max_mac) {
2070 		netdev_info(net_dev,
2071 			    "Unicast addr count reached %d, max allowed is %d; forcing promisc\n",
2072 			    uc_count, max_mac);
2073 		goto force_promisc;
2074 	}
2075 	if (mc_count + uc_count > max_mac) {
2076 		netdev_info(net_dev,
2077 			    "Unicast + multicast addr count reached %d, max allowed is %d; forcing promisc\n",
2078 			    uc_count + mc_count, max_mac);
2079 		goto force_mc_promisc;
2080 	}
2081 
2082 	/* Adjust promisc settings due to flag combinations */
2083 	if (net_dev->flags & IFF_PROMISC)
2084 		goto force_promisc;
2085 	if (net_dev->flags & IFF_ALLMULTI) {
2086 		/* First, rebuild unicast filtering table. This should be done
2087 		 * in promisc mode, in order to avoid frame loss while we
2088 		 * progressively add entries to the table.
2089 		 * We don't know whether we had been in promisc already, and
2090 		 * making an MC call to find out is expensive; so set uc promisc
2091 		 * nonetheless.
2092 		 */
2093 		err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 1);
2094 		if (err)
2095 			netdev_warn(net_dev, "Can't set uc promisc\n");
2096 
2097 		/* Actual uc table reconstruction. */
2098 		err = dpni_clear_mac_filters(mc_io, 0, mc_token, 1, 0);
2099 		if (err)
2100 			netdev_warn(net_dev, "Can't clear uc filters\n");
2101 		dpaa2_eth_add_uc_hw_addr(net_dev, priv);
2102 
2103 		/* Finally, clear uc promisc and set mc promisc as requested. */
2104 		err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 0);
2105 		if (err)
2106 			netdev_warn(net_dev, "Can't clear uc promisc\n");
2107 		goto force_mc_promisc;
2108 	}
2109 
2110 	/* Neither unicast, nor multicast promisc will be on... eventually.
2111 	 * For now, rebuild mac filtering tables while forcing both of them on.
2112 	 */
2113 	err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 1);
2114 	if (err)
2115 		netdev_warn(net_dev, "Can't set uc promisc (%d)\n", err);
2116 	err = dpni_set_multicast_promisc(mc_io, 0, mc_token, 1);
2117 	if (err)
2118 		netdev_warn(net_dev, "Can't set mc promisc (%d)\n", err);
2119 
2120 	/* Actual mac filtering tables reconstruction */
2121 	err = dpni_clear_mac_filters(mc_io, 0, mc_token, 1, 1);
2122 	if (err)
2123 		netdev_warn(net_dev, "Can't clear mac filters\n");
2124 	dpaa2_eth_add_mc_hw_addr(net_dev, priv);
2125 	dpaa2_eth_add_uc_hw_addr(net_dev, priv);
2126 
2127 	/* Now we can clear both ucast and mcast promisc, without risking
2128 	 * to drop legitimate frames anymore.
2129 	 */
2130 	err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 0);
2131 	if (err)
2132 		netdev_warn(net_dev, "Can't clear ucast promisc\n");
2133 	err = dpni_set_multicast_promisc(mc_io, 0, mc_token, 0);
2134 	if (err)
2135 		netdev_warn(net_dev, "Can't clear mcast promisc\n");
2136 
2137 	return;
2138 
2139 force_promisc:
2140 	err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 1);
2141 	if (err)
2142 		netdev_warn(net_dev, "Can't set ucast promisc\n");
2143 force_mc_promisc:
2144 	err = dpni_set_multicast_promisc(mc_io, 0, mc_token, 1);
2145 	if (err)
2146 		netdev_warn(net_dev, "Can't set mcast promisc\n");
2147 }
2148 
2149 static int dpaa2_eth_set_features(struct net_device *net_dev,
2150 				  netdev_features_t features)
2151 {
2152 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2153 	netdev_features_t changed = features ^ net_dev->features;
2154 	bool enable;
2155 	int err;
2156 
2157 	if (changed & NETIF_F_HW_VLAN_CTAG_FILTER) {
2158 		enable = !!(features & NETIF_F_HW_VLAN_CTAG_FILTER);
2159 		err = dpaa2_eth_set_rx_vlan_filtering(priv, enable);
2160 		if (err)
2161 			return err;
2162 	}
2163 
2164 	if (changed & NETIF_F_RXCSUM) {
2165 		enable = !!(features & NETIF_F_RXCSUM);
2166 		err = dpaa2_eth_set_rx_csum(priv, enable);
2167 		if (err)
2168 			return err;
2169 	}
2170 
2171 	if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
2172 		enable = !!(features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM));
2173 		err = dpaa2_eth_set_tx_csum(priv, enable);
2174 		if (err)
2175 			return err;
2176 	}
2177 
2178 	return 0;
2179 }
2180 
2181 static int dpaa2_eth_ts_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
2182 {
2183 	struct dpaa2_eth_priv *priv = netdev_priv(dev);
2184 	struct hwtstamp_config config;
2185 
2186 	if (!dpaa2_ptp)
2187 		return -EINVAL;
2188 
2189 	if (copy_from_user(&config, rq->ifr_data, sizeof(config)))
2190 		return -EFAULT;
2191 
2192 	switch (config.tx_type) {
2193 	case HWTSTAMP_TX_OFF:
2194 	case HWTSTAMP_TX_ON:
2195 	case HWTSTAMP_TX_ONESTEP_SYNC:
2196 		priv->tx_tstamp_type = config.tx_type;
2197 		break;
2198 	default:
2199 		return -ERANGE;
2200 	}
2201 
2202 	if (config.rx_filter == HWTSTAMP_FILTER_NONE) {
2203 		priv->rx_tstamp = false;
2204 	} else {
2205 		priv->rx_tstamp = true;
2206 		/* TS is set for all frame types, not only those requested */
2207 		config.rx_filter = HWTSTAMP_FILTER_ALL;
2208 	}
2209 
2210 	return copy_to_user(rq->ifr_data, &config, sizeof(config)) ?
2211 			-EFAULT : 0;
2212 }
2213 
2214 static int dpaa2_eth_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
2215 {
2216 	struct dpaa2_eth_priv *priv = netdev_priv(dev);
2217 
2218 	if (cmd == SIOCSHWTSTAMP)
2219 		return dpaa2_eth_ts_ioctl(dev, rq, cmd);
2220 
2221 	if (dpaa2_eth_is_type_phy(priv))
2222 		return phylink_mii_ioctl(priv->mac->phylink, rq, cmd);
2223 
2224 	return -EOPNOTSUPP;
2225 }
2226 
2227 static bool xdp_mtu_valid(struct dpaa2_eth_priv *priv, int mtu)
2228 {
2229 	int mfl, linear_mfl;
2230 
2231 	mfl = DPAA2_ETH_L2_MAX_FRM(mtu);
2232 	linear_mfl = priv->rx_buf_size - DPAA2_ETH_RX_HWA_SIZE -
2233 		     dpaa2_eth_rx_head_room(priv) - XDP_PACKET_HEADROOM;
2234 
2235 	if (mfl > linear_mfl) {
2236 		netdev_warn(priv->net_dev, "Maximum MTU for XDP is %d\n",
2237 			    linear_mfl - VLAN_ETH_HLEN);
2238 		return false;
2239 	}
2240 
2241 	return true;
2242 }
2243 
2244 static int dpaa2_eth_set_rx_mfl(struct dpaa2_eth_priv *priv, int mtu, bool has_xdp)
2245 {
2246 	int mfl, err;
2247 
2248 	/* We enforce a maximum Rx frame length based on MTU only if we have
2249 	 * an XDP program attached (in order to avoid Rx S/G frames).
2250 	 * Otherwise, we accept all incoming frames as long as they are not
2251 	 * larger than maximum size supported in hardware
2252 	 */
2253 	if (has_xdp)
2254 		mfl = DPAA2_ETH_L2_MAX_FRM(mtu);
2255 	else
2256 		mfl = DPAA2_ETH_MFL;
2257 
2258 	err = dpni_set_max_frame_length(priv->mc_io, 0, priv->mc_token, mfl);
2259 	if (err) {
2260 		netdev_err(priv->net_dev, "dpni_set_max_frame_length failed\n");
2261 		return err;
2262 	}
2263 
2264 	return 0;
2265 }
2266 
2267 static int dpaa2_eth_change_mtu(struct net_device *dev, int new_mtu)
2268 {
2269 	struct dpaa2_eth_priv *priv = netdev_priv(dev);
2270 	int err;
2271 
2272 	if (!priv->xdp_prog)
2273 		goto out;
2274 
2275 	if (!xdp_mtu_valid(priv, new_mtu))
2276 		return -EINVAL;
2277 
2278 	err = dpaa2_eth_set_rx_mfl(priv, new_mtu, true);
2279 	if (err)
2280 		return err;
2281 
2282 out:
2283 	dev->mtu = new_mtu;
2284 	return 0;
2285 }
2286 
2287 static int dpaa2_eth_update_rx_buffer_headroom(struct dpaa2_eth_priv *priv, bool has_xdp)
2288 {
2289 	struct dpni_buffer_layout buf_layout = {0};
2290 	int err;
2291 
2292 	err = dpni_get_buffer_layout(priv->mc_io, 0, priv->mc_token,
2293 				     DPNI_QUEUE_RX, &buf_layout);
2294 	if (err) {
2295 		netdev_err(priv->net_dev, "dpni_get_buffer_layout failed\n");
2296 		return err;
2297 	}
2298 
2299 	/* Reserve extra headroom for XDP header size changes */
2300 	buf_layout.data_head_room = dpaa2_eth_rx_head_room(priv) +
2301 				    (has_xdp ? XDP_PACKET_HEADROOM : 0);
2302 	buf_layout.options = DPNI_BUF_LAYOUT_OPT_DATA_HEAD_ROOM;
2303 	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
2304 				     DPNI_QUEUE_RX, &buf_layout);
2305 	if (err) {
2306 		netdev_err(priv->net_dev, "dpni_set_buffer_layout failed\n");
2307 		return err;
2308 	}
2309 
2310 	return 0;
2311 }
2312 
2313 static int dpaa2_eth_setup_xdp(struct net_device *dev, struct bpf_prog *prog)
2314 {
2315 	struct dpaa2_eth_priv *priv = netdev_priv(dev);
2316 	struct dpaa2_eth_channel *ch;
2317 	struct bpf_prog *old;
2318 	bool up, need_update;
2319 	int i, err;
2320 
2321 	if (prog && !xdp_mtu_valid(priv, dev->mtu))
2322 		return -EINVAL;
2323 
2324 	if (prog)
2325 		bpf_prog_add(prog, priv->num_channels);
2326 
2327 	up = netif_running(dev);
2328 	need_update = (!!priv->xdp_prog != !!prog);
2329 
2330 	if (up)
2331 		dpaa2_eth_stop(dev);
2332 
2333 	/* While in xdp mode, enforce a maximum Rx frame size based on MTU.
2334 	 * Also, when switching between xdp/non-xdp modes we need to reconfigure
2335 	 * our Rx buffer layout. Buffer pool was drained on dpaa2_eth_stop,
2336 	 * so we are sure no old format buffers will be used from now on.
2337 	 */
2338 	if (need_update) {
2339 		err = dpaa2_eth_set_rx_mfl(priv, dev->mtu, !!prog);
2340 		if (err)
2341 			goto out_err;
2342 		err = dpaa2_eth_update_rx_buffer_headroom(priv, !!prog);
2343 		if (err)
2344 			goto out_err;
2345 	}
2346 
2347 	old = xchg(&priv->xdp_prog, prog);
2348 	if (old)
2349 		bpf_prog_put(old);
2350 
2351 	for (i = 0; i < priv->num_channels; i++) {
2352 		ch = priv->channel[i];
2353 		old = xchg(&ch->xdp.prog, prog);
2354 		if (old)
2355 			bpf_prog_put(old);
2356 	}
2357 
2358 	if (up) {
2359 		err = dpaa2_eth_open(dev);
2360 		if (err)
2361 			return err;
2362 	}
2363 
2364 	return 0;
2365 
2366 out_err:
2367 	if (prog)
2368 		bpf_prog_sub(prog, priv->num_channels);
2369 	if (up)
2370 		dpaa2_eth_open(dev);
2371 
2372 	return err;
2373 }
2374 
2375 static int dpaa2_eth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2376 {
2377 	switch (xdp->command) {
2378 	case XDP_SETUP_PROG:
2379 		return dpaa2_eth_setup_xdp(dev, xdp->prog);
2380 	default:
2381 		return -EINVAL;
2382 	}
2383 
2384 	return 0;
2385 }
2386 
2387 static int dpaa2_eth_xdp_create_fd(struct net_device *net_dev,
2388 				   struct xdp_frame *xdpf,
2389 				   struct dpaa2_fd *fd)
2390 {
2391 	struct device *dev = net_dev->dev.parent;
2392 	unsigned int needed_headroom;
2393 	struct dpaa2_eth_swa *swa;
2394 	void *buffer_start, *aligned_start;
2395 	dma_addr_t addr;
2396 
2397 	/* We require a minimum headroom to be able to transmit the frame.
2398 	 * Otherwise return an error and let the original net_device handle it
2399 	 */
2400 	needed_headroom = dpaa2_eth_needed_headroom(NULL);
2401 	if (xdpf->headroom < needed_headroom)
2402 		return -EINVAL;
2403 
2404 	/* Setup the FD fields */
2405 	memset(fd, 0, sizeof(*fd));
2406 
2407 	/* Align FD address, if possible */
2408 	buffer_start = xdpf->data - needed_headroom;
2409 	aligned_start = PTR_ALIGN(buffer_start - DPAA2_ETH_TX_BUF_ALIGN,
2410 				  DPAA2_ETH_TX_BUF_ALIGN);
2411 	if (aligned_start >= xdpf->data - xdpf->headroom)
2412 		buffer_start = aligned_start;
2413 
2414 	swa = (struct dpaa2_eth_swa *)buffer_start;
2415 	/* fill in necessary fields here */
2416 	swa->type = DPAA2_ETH_SWA_XDP;
2417 	swa->xdp.dma_size = xdpf->data + xdpf->len - buffer_start;
2418 	swa->xdp.xdpf = xdpf;
2419 
2420 	addr = dma_map_single(dev, buffer_start,
2421 			      swa->xdp.dma_size,
2422 			      DMA_BIDIRECTIONAL);
2423 	if (unlikely(dma_mapping_error(dev, addr)))
2424 		return -ENOMEM;
2425 
2426 	dpaa2_fd_set_addr(fd, addr);
2427 	dpaa2_fd_set_offset(fd, xdpf->data - buffer_start);
2428 	dpaa2_fd_set_len(fd, xdpf->len);
2429 	dpaa2_fd_set_format(fd, dpaa2_fd_single);
2430 	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
2431 
2432 	return 0;
2433 }
2434 
2435 static int dpaa2_eth_xdp_xmit(struct net_device *net_dev, int n,
2436 			      struct xdp_frame **frames, u32 flags)
2437 {
2438 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2439 	struct dpaa2_eth_xdp_fds *xdp_redirect_fds;
2440 	struct rtnl_link_stats64 *percpu_stats;
2441 	struct dpaa2_eth_fq *fq;
2442 	struct dpaa2_fd *fds;
2443 	int enqueued, i, err;
2444 
2445 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
2446 		return -EINVAL;
2447 
2448 	if (!netif_running(net_dev))
2449 		return -ENETDOWN;
2450 
2451 	fq = &priv->fq[smp_processor_id()];
2452 	xdp_redirect_fds = &fq->xdp_redirect_fds;
2453 	fds = xdp_redirect_fds->fds;
2454 
2455 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
2456 
2457 	/* create a FD for each xdp_frame in the list received */
2458 	for (i = 0; i < n; i++) {
2459 		err = dpaa2_eth_xdp_create_fd(net_dev, frames[i], &fds[i]);
2460 		if (err)
2461 			break;
2462 	}
2463 	xdp_redirect_fds->num = i;
2464 
2465 	/* enqueue all the frame descriptors */
2466 	enqueued = dpaa2_eth_xdp_flush(priv, fq, xdp_redirect_fds);
2467 
2468 	/* update statistics */
2469 	percpu_stats->tx_packets += enqueued;
2470 	for (i = 0; i < enqueued; i++)
2471 		percpu_stats->tx_bytes += dpaa2_fd_get_len(&fds[i]);
2472 
2473 	return enqueued;
2474 }
2475 
2476 static int update_xps(struct dpaa2_eth_priv *priv)
2477 {
2478 	struct net_device *net_dev = priv->net_dev;
2479 	struct cpumask xps_mask;
2480 	struct dpaa2_eth_fq *fq;
2481 	int i, num_queues, netdev_queues;
2482 	int err = 0;
2483 
2484 	num_queues = dpaa2_eth_queue_count(priv);
2485 	netdev_queues = (net_dev->num_tc ? : 1) * num_queues;
2486 
2487 	/* The first <num_queues> entries in priv->fq array are Tx/Tx conf
2488 	 * queues, so only process those
2489 	 */
2490 	for (i = 0; i < netdev_queues; i++) {
2491 		fq = &priv->fq[i % num_queues];
2492 
2493 		cpumask_clear(&xps_mask);
2494 		cpumask_set_cpu(fq->target_cpu, &xps_mask);
2495 
2496 		err = netif_set_xps_queue(net_dev, &xps_mask, i);
2497 		if (err) {
2498 			netdev_warn_once(net_dev, "Error setting XPS queue\n");
2499 			break;
2500 		}
2501 	}
2502 
2503 	return err;
2504 }
2505 
2506 static int dpaa2_eth_setup_mqprio(struct net_device *net_dev,
2507 				  struct tc_mqprio_qopt *mqprio)
2508 {
2509 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2510 	u8 num_tc, num_queues;
2511 	int i;
2512 
2513 	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
2514 	num_queues = dpaa2_eth_queue_count(priv);
2515 	num_tc = mqprio->num_tc;
2516 
2517 	if (num_tc == net_dev->num_tc)
2518 		return 0;
2519 
2520 	if (num_tc  > dpaa2_eth_tc_count(priv)) {
2521 		netdev_err(net_dev, "Max %d traffic classes supported\n",
2522 			   dpaa2_eth_tc_count(priv));
2523 		return -EOPNOTSUPP;
2524 	}
2525 
2526 	if (!num_tc) {
2527 		netdev_reset_tc(net_dev);
2528 		netif_set_real_num_tx_queues(net_dev, num_queues);
2529 		goto out;
2530 	}
2531 
2532 	netdev_set_num_tc(net_dev, num_tc);
2533 	netif_set_real_num_tx_queues(net_dev, num_tc * num_queues);
2534 
2535 	for (i = 0; i < num_tc; i++)
2536 		netdev_set_tc_queue(net_dev, i, num_queues, i * num_queues);
2537 
2538 out:
2539 	update_xps(priv);
2540 
2541 	return 0;
2542 }
2543 
2544 #define bps_to_mbits(rate) (div_u64((rate), 1000000) * 8)
2545 
2546 static int dpaa2_eth_setup_tbf(struct net_device *net_dev, struct tc_tbf_qopt_offload *p)
2547 {
2548 	struct tc_tbf_qopt_offload_replace_params *cfg = &p->replace_params;
2549 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2550 	struct dpni_tx_shaping_cfg tx_cr_shaper = { 0 };
2551 	struct dpni_tx_shaping_cfg tx_er_shaper = { 0 };
2552 	int err;
2553 
2554 	if (p->command == TC_TBF_STATS)
2555 		return -EOPNOTSUPP;
2556 
2557 	/* Only per port Tx shaping */
2558 	if (p->parent != TC_H_ROOT)
2559 		return -EOPNOTSUPP;
2560 
2561 	if (p->command == TC_TBF_REPLACE) {
2562 		if (cfg->max_size > DPAA2_ETH_MAX_BURST_SIZE) {
2563 			netdev_err(net_dev, "burst size cannot be greater than %d\n",
2564 				   DPAA2_ETH_MAX_BURST_SIZE);
2565 			return -EINVAL;
2566 		}
2567 
2568 		tx_cr_shaper.max_burst_size = cfg->max_size;
2569 		/* The TBF interface is in bytes/s, whereas DPAA2 expects the
2570 		 * rate in Mbits/s
2571 		 */
2572 		tx_cr_shaper.rate_limit = bps_to_mbits(cfg->rate.rate_bytes_ps);
2573 	}
2574 
2575 	err = dpni_set_tx_shaping(priv->mc_io, 0, priv->mc_token, &tx_cr_shaper,
2576 				  &tx_er_shaper, 0);
2577 	if (err) {
2578 		netdev_err(net_dev, "dpni_set_tx_shaping() = %d\n", err);
2579 		return err;
2580 	}
2581 
2582 	return 0;
2583 }
2584 
2585 static int dpaa2_eth_setup_tc(struct net_device *net_dev,
2586 			      enum tc_setup_type type, void *type_data)
2587 {
2588 	switch (type) {
2589 	case TC_SETUP_QDISC_MQPRIO:
2590 		return dpaa2_eth_setup_mqprio(net_dev, type_data);
2591 	case TC_SETUP_QDISC_TBF:
2592 		return dpaa2_eth_setup_tbf(net_dev, type_data);
2593 	default:
2594 		return -EOPNOTSUPP;
2595 	}
2596 }
2597 
2598 static const struct net_device_ops dpaa2_eth_ops = {
2599 	.ndo_open = dpaa2_eth_open,
2600 	.ndo_start_xmit = dpaa2_eth_tx,
2601 	.ndo_stop = dpaa2_eth_stop,
2602 	.ndo_set_mac_address = dpaa2_eth_set_addr,
2603 	.ndo_get_stats64 = dpaa2_eth_get_stats,
2604 	.ndo_set_rx_mode = dpaa2_eth_set_rx_mode,
2605 	.ndo_set_features = dpaa2_eth_set_features,
2606 	.ndo_eth_ioctl = dpaa2_eth_ioctl,
2607 	.ndo_change_mtu = dpaa2_eth_change_mtu,
2608 	.ndo_bpf = dpaa2_eth_xdp,
2609 	.ndo_xdp_xmit = dpaa2_eth_xdp_xmit,
2610 	.ndo_setup_tc = dpaa2_eth_setup_tc,
2611 	.ndo_vlan_rx_add_vid = dpaa2_eth_rx_add_vid,
2612 	.ndo_vlan_rx_kill_vid = dpaa2_eth_rx_kill_vid
2613 };
2614 
2615 static void dpaa2_eth_cdan_cb(struct dpaa2_io_notification_ctx *ctx)
2616 {
2617 	struct dpaa2_eth_channel *ch;
2618 
2619 	ch = container_of(ctx, struct dpaa2_eth_channel, nctx);
2620 
2621 	/* Update NAPI statistics */
2622 	ch->stats.cdan++;
2623 
2624 	napi_schedule(&ch->napi);
2625 }
2626 
2627 /* Allocate and configure a DPCON object */
2628 static struct fsl_mc_device *dpaa2_eth_setup_dpcon(struct dpaa2_eth_priv *priv)
2629 {
2630 	struct fsl_mc_device *dpcon;
2631 	struct device *dev = priv->net_dev->dev.parent;
2632 	int err;
2633 
2634 	err = fsl_mc_object_allocate(to_fsl_mc_device(dev),
2635 				     FSL_MC_POOL_DPCON, &dpcon);
2636 	if (err) {
2637 		if (err == -ENXIO)
2638 			err = -EPROBE_DEFER;
2639 		else
2640 			dev_info(dev, "Not enough DPCONs, will go on as-is\n");
2641 		return ERR_PTR(err);
2642 	}
2643 
2644 	err = dpcon_open(priv->mc_io, 0, dpcon->obj_desc.id, &dpcon->mc_handle);
2645 	if (err) {
2646 		dev_err(dev, "dpcon_open() failed\n");
2647 		goto free;
2648 	}
2649 
2650 	err = dpcon_reset(priv->mc_io, 0, dpcon->mc_handle);
2651 	if (err) {
2652 		dev_err(dev, "dpcon_reset() failed\n");
2653 		goto close;
2654 	}
2655 
2656 	err = dpcon_enable(priv->mc_io, 0, dpcon->mc_handle);
2657 	if (err) {
2658 		dev_err(dev, "dpcon_enable() failed\n");
2659 		goto close;
2660 	}
2661 
2662 	return dpcon;
2663 
2664 close:
2665 	dpcon_close(priv->mc_io, 0, dpcon->mc_handle);
2666 free:
2667 	fsl_mc_object_free(dpcon);
2668 
2669 	return ERR_PTR(err);
2670 }
2671 
2672 static void dpaa2_eth_free_dpcon(struct dpaa2_eth_priv *priv,
2673 				 struct fsl_mc_device *dpcon)
2674 {
2675 	dpcon_disable(priv->mc_io, 0, dpcon->mc_handle);
2676 	dpcon_close(priv->mc_io, 0, dpcon->mc_handle);
2677 	fsl_mc_object_free(dpcon);
2678 }
2679 
2680 static struct dpaa2_eth_channel *dpaa2_eth_alloc_channel(struct dpaa2_eth_priv *priv)
2681 {
2682 	struct dpaa2_eth_channel *channel;
2683 	struct dpcon_attr attr;
2684 	struct device *dev = priv->net_dev->dev.parent;
2685 	int err;
2686 
2687 	channel = kzalloc(sizeof(*channel), GFP_KERNEL);
2688 	if (!channel)
2689 		return NULL;
2690 
2691 	channel->dpcon = dpaa2_eth_setup_dpcon(priv);
2692 	if (IS_ERR(channel->dpcon)) {
2693 		err = PTR_ERR(channel->dpcon);
2694 		goto err_setup;
2695 	}
2696 
2697 	err = dpcon_get_attributes(priv->mc_io, 0, channel->dpcon->mc_handle,
2698 				   &attr);
2699 	if (err) {
2700 		dev_err(dev, "dpcon_get_attributes() failed\n");
2701 		goto err_get_attr;
2702 	}
2703 
2704 	channel->dpcon_id = attr.id;
2705 	channel->ch_id = attr.qbman_ch_id;
2706 	channel->priv = priv;
2707 
2708 	return channel;
2709 
2710 err_get_attr:
2711 	dpaa2_eth_free_dpcon(priv, channel->dpcon);
2712 err_setup:
2713 	kfree(channel);
2714 	return ERR_PTR(err);
2715 }
2716 
2717 static void dpaa2_eth_free_channel(struct dpaa2_eth_priv *priv,
2718 				   struct dpaa2_eth_channel *channel)
2719 {
2720 	dpaa2_eth_free_dpcon(priv, channel->dpcon);
2721 	kfree(channel);
2722 }
2723 
2724 /* DPIO setup: allocate and configure QBMan channels, setup core affinity
2725  * and register data availability notifications
2726  */
2727 static int dpaa2_eth_setup_dpio(struct dpaa2_eth_priv *priv)
2728 {
2729 	struct dpaa2_io_notification_ctx *nctx;
2730 	struct dpaa2_eth_channel *channel;
2731 	struct dpcon_notification_cfg dpcon_notif_cfg;
2732 	struct device *dev = priv->net_dev->dev.parent;
2733 	int i, err;
2734 
2735 	/* We want the ability to spread ingress traffic (RX, TX conf) to as
2736 	 * many cores as possible, so we need one channel for each core
2737 	 * (unless there's fewer queues than cores, in which case the extra
2738 	 * channels would be wasted).
2739 	 * Allocate one channel per core and register it to the core's
2740 	 * affine DPIO. If not enough channels are available for all cores
2741 	 * or if some cores don't have an affine DPIO, there will be no
2742 	 * ingress frame processing on those cores.
2743 	 */
2744 	cpumask_clear(&priv->dpio_cpumask);
2745 	for_each_online_cpu(i) {
2746 		/* Try to allocate a channel */
2747 		channel = dpaa2_eth_alloc_channel(priv);
2748 		if (IS_ERR_OR_NULL(channel)) {
2749 			err = PTR_ERR_OR_ZERO(channel);
2750 			if (err != -EPROBE_DEFER)
2751 				dev_info(dev,
2752 					 "No affine channel for cpu %d and above\n", i);
2753 			goto err_alloc_ch;
2754 		}
2755 
2756 		priv->channel[priv->num_channels] = channel;
2757 
2758 		nctx = &channel->nctx;
2759 		nctx->is_cdan = 1;
2760 		nctx->cb = dpaa2_eth_cdan_cb;
2761 		nctx->id = channel->ch_id;
2762 		nctx->desired_cpu = i;
2763 
2764 		/* Register the new context */
2765 		channel->dpio = dpaa2_io_service_select(i);
2766 		err = dpaa2_io_service_register(channel->dpio, nctx, dev);
2767 		if (err) {
2768 			dev_dbg(dev, "No affine DPIO for cpu %d\n", i);
2769 			/* If no affine DPIO for this core, there's probably
2770 			 * none available for next cores either. Signal we want
2771 			 * to retry later, in case the DPIO devices weren't
2772 			 * probed yet.
2773 			 */
2774 			err = -EPROBE_DEFER;
2775 			goto err_service_reg;
2776 		}
2777 
2778 		/* Register DPCON notification with MC */
2779 		dpcon_notif_cfg.dpio_id = nctx->dpio_id;
2780 		dpcon_notif_cfg.priority = 0;
2781 		dpcon_notif_cfg.user_ctx = nctx->qman64;
2782 		err = dpcon_set_notification(priv->mc_io, 0,
2783 					     channel->dpcon->mc_handle,
2784 					     &dpcon_notif_cfg);
2785 		if (err) {
2786 			dev_err(dev, "dpcon_set_notification failed()\n");
2787 			goto err_set_cdan;
2788 		}
2789 
2790 		/* If we managed to allocate a channel and also found an affine
2791 		 * DPIO for this core, add it to the final mask
2792 		 */
2793 		cpumask_set_cpu(i, &priv->dpio_cpumask);
2794 		priv->num_channels++;
2795 
2796 		/* Stop if we already have enough channels to accommodate all
2797 		 * RX and TX conf queues
2798 		 */
2799 		if (priv->num_channels == priv->dpni_attrs.num_queues)
2800 			break;
2801 	}
2802 
2803 	return 0;
2804 
2805 err_set_cdan:
2806 	dpaa2_io_service_deregister(channel->dpio, nctx, dev);
2807 err_service_reg:
2808 	dpaa2_eth_free_channel(priv, channel);
2809 err_alloc_ch:
2810 	if (err == -EPROBE_DEFER) {
2811 		for (i = 0; i < priv->num_channels; i++) {
2812 			channel = priv->channel[i];
2813 			nctx = &channel->nctx;
2814 			dpaa2_io_service_deregister(channel->dpio, nctx, dev);
2815 			dpaa2_eth_free_channel(priv, channel);
2816 		}
2817 		priv->num_channels = 0;
2818 		return err;
2819 	}
2820 
2821 	if (cpumask_empty(&priv->dpio_cpumask)) {
2822 		dev_err(dev, "No cpu with an affine DPIO/DPCON\n");
2823 		return -ENODEV;
2824 	}
2825 
2826 	dev_info(dev, "Cores %*pbl available for processing ingress traffic\n",
2827 		 cpumask_pr_args(&priv->dpio_cpumask));
2828 
2829 	return 0;
2830 }
2831 
2832 static void dpaa2_eth_free_dpio(struct dpaa2_eth_priv *priv)
2833 {
2834 	struct device *dev = priv->net_dev->dev.parent;
2835 	struct dpaa2_eth_channel *ch;
2836 	int i;
2837 
2838 	/* deregister CDAN notifications and free channels */
2839 	for (i = 0; i < priv->num_channels; i++) {
2840 		ch = priv->channel[i];
2841 		dpaa2_io_service_deregister(ch->dpio, &ch->nctx, dev);
2842 		dpaa2_eth_free_channel(priv, ch);
2843 	}
2844 }
2845 
2846 static struct dpaa2_eth_channel *dpaa2_eth_get_affine_channel(struct dpaa2_eth_priv *priv,
2847 							      int cpu)
2848 {
2849 	struct device *dev = priv->net_dev->dev.parent;
2850 	int i;
2851 
2852 	for (i = 0; i < priv->num_channels; i++)
2853 		if (priv->channel[i]->nctx.desired_cpu == cpu)
2854 			return priv->channel[i];
2855 
2856 	/* We should never get here. Issue a warning and return
2857 	 * the first channel, because it's still better than nothing
2858 	 */
2859 	dev_warn(dev, "No affine channel found for cpu %d\n", cpu);
2860 
2861 	return priv->channel[0];
2862 }
2863 
2864 static void dpaa2_eth_set_fq_affinity(struct dpaa2_eth_priv *priv)
2865 {
2866 	struct device *dev = priv->net_dev->dev.parent;
2867 	struct dpaa2_eth_fq *fq;
2868 	int rx_cpu, txc_cpu;
2869 	int i;
2870 
2871 	/* For each FQ, pick one channel/CPU to deliver frames to.
2872 	 * This may well change at runtime, either through irqbalance or
2873 	 * through direct user intervention.
2874 	 */
2875 	rx_cpu = txc_cpu = cpumask_first(&priv->dpio_cpumask);
2876 
2877 	for (i = 0; i < priv->num_fqs; i++) {
2878 		fq = &priv->fq[i];
2879 		switch (fq->type) {
2880 		case DPAA2_RX_FQ:
2881 		case DPAA2_RX_ERR_FQ:
2882 			fq->target_cpu = rx_cpu;
2883 			rx_cpu = cpumask_next(rx_cpu, &priv->dpio_cpumask);
2884 			if (rx_cpu >= nr_cpu_ids)
2885 				rx_cpu = cpumask_first(&priv->dpio_cpumask);
2886 			break;
2887 		case DPAA2_TX_CONF_FQ:
2888 			fq->target_cpu = txc_cpu;
2889 			txc_cpu = cpumask_next(txc_cpu, &priv->dpio_cpumask);
2890 			if (txc_cpu >= nr_cpu_ids)
2891 				txc_cpu = cpumask_first(&priv->dpio_cpumask);
2892 			break;
2893 		default:
2894 			dev_err(dev, "Unknown FQ type: %d\n", fq->type);
2895 		}
2896 		fq->channel = dpaa2_eth_get_affine_channel(priv, fq->target_cpu);
2897 	}
2898 
2899 	update_xps(priv);
2900 }
2901 
2902 static void dpaa2_eth_setup_fqs(struct dpaa2_eth_priv *priv)
2903 {
2904 	int i, j;
2905 
2906 	/* We have one TxConf FQ per Tx flow.
2907 	 * The number of Tx and Rx queues is the same.
2908 	 * Tx queues come first in the fq array.
2909 	 */
2910 	for (i = 0; i < dpaa2_eth_queue_count(priv); i++) {
2911 		priv->fq[priv->num_fqs].type = DPAA2_TX_CONF_FQ;
2912 		priv->fq[priv->num_fqs].consume = dpaa2_eth_tx_conf;
2913 		priv->fq[priv->num_fqs++].flowid = (u16)i;
2914 	}
2915 
2916 	for (j = 0; j < dpaa2_eth_tc_count(priv); j++) {
2917 		for (i = 0; i < dpaa2_eth_queue_count(priv); i++) {
2918 			priv->fq[priv->num_fqs].type = DPAA2_RX_FQ;
2919 			priv->fq[priv->num_fqs].consume = dpaa2_eth_rx;
2920 			priv->fq[priv->num_fqs].tc = (u8)j;
2921 			priv->fq[priv->num_fqs++].flowid = (u16)i;
2922 		}
2923 	}
2924 
2925 	/* We have exactly one Rx error queue per DPNI */
2926 	priv->fq[priv->num_fqs].type = DPAA2_RX_ERR_FQ;
2927 	priv->fq[priv->num_fqs++].consume = dpaa2_eth_rx_err;
2928 
2929 	/* For each FQ, decide on which core to process incoming frames */
2930 	dpaa2_eth_set_fq_affinity(priv);
2931 }
2932 
2933 /* Allocate and configure one buffer pool for each interface */
2934 static int dpaa2_eth_setup_dpbp(struct dpaa2_eth_priv *priv)
2935 {
2936 	int err;
2937 	struct fsl_mc_device *dpbp_dev;
2938 	struct device *dev = priv->net_dev->dev.parent;
2939 	struct dpbp_attr dpbp_attrs;
2940 
2941 	err = fsl_mc_object_allocate(to_fsl_mc_device(dev), FSL_MC_POOL_DPBP,
2942 				     &dpbp_dev);
2943 	if (err) {
2944 		if (err == -ENXIO)
2945 			err = -EPROBE_DEFER;
2946 		else
2947 			dev_err(dev, "DPBP device allocation failed\n");
2948 		return err;
2949 	}
2950 
2951 	priv->dpbp_dev = dpbp_dev;
2952 
2953 	err = dpbp_open(priv->mc_io, 0, priv->dpbp_dev->obj_desc.id,
2954 			&dpbp_dev->mc_handle);
2955 	if (err) {
2956 		dev_err(dev, "dpbp_open() failed\n");
2957 		goto err_open;
2958 	}
2959 
2960 	err = dpbp_reset(priv->mc_io, 0, dpbp_dev->mc_handle);
2961 	if (err) {
2962 		dev_err(dev, "dpbp_reset() failed\n");
2963 		goto err_reset;
2964 	}
2965 
2966 	err = dpbp_enable(priv->mc_io, 0, dpbp_dev->mc_handle);
2967 	if (err) {
2968 		dev_err(dev, "dpbp_enable() failed\n");
2969 		goto err_enable;
2970 	}
2971 
2972 	err = dpbp_get_attributes(priv->mc_io, 0, dpbp_dev->mc_handle,
2973 				  &dpbp_attrs);
2974 	if (err) {
2975 		dev_err(dev, "dpbp_get_attributes() failed\n");
2976 		goto err_get_attr;
2977 	}
2978 	priv->bpid = dpbp_attrs.bpid;
2979 
2980 	return 0;
2981 
2982 err_get_attr:
2983 	dpbp_disable(priv->mc_io, 0, dpbp_dev->mc_handle);
2984 err_enable:
2985 err_reset:
2986 	dpbp_close(priv->mc_io, 0, dpbp_dev->mc_handle);
2987 err_open:
2988 	fsl_mc_object_free(dpbp_dev);
2989 
2990 	return err;
2991 }
2992 
2993 static void dpaa2_eth_free_dpbp(struct dpaa2_eth_priv *priv)
2994 {
2995 	dpaa2_eth_drain_pool(priv);
2996 	dpbp_disable(priv->mc_io, 0, priv->dpbp_dev->mc_handle);
2997 	dpbp_close(priv->mc_io, 0, priv->dpbp_dev->mc_handle);
2998 	fsl_mc_object_free(priv->dpbp_dev);
2999 }
3000 
3001 static int dpaa2_eth_set_buffer_layout(struct dpaa2_eth_priv *priv)
3002 {
3003 	struct device *dev = priv->net_dev->dev.parent;
3004 	struct dpni_buffer_layout buf_layout = {0};
3005 	u16 rx_buf_align;
3006 	int err;
3007 
3008 	/* We need to check for WRIOP version 1.0.0, but depending on the MC
3009 	 * version, this number is not always provided correctly on rev1.
3010 	 * We need to check for both alternatives in this situation.
3011 	 */
3012 	if (priv->dpni_attrs.wriop_version == DPAA2_WRIOP_VERSION(0, 0, 0) ||
3013 	    priv->dpni_attrs.wriop_version == DPAA2_WRIOP_VERSION(1, 0, 0))
3014 		rx_buf_align = DPAA2_ETH_RX_BUF_ALIGN_REV1;
3015 	else
3016 		rx_buf_align = DPAA2_ETH_RX_BUF_ALIGN;
3017 
3018 	/* We need to ensure that the buffer size seen by WRIOP is a multiple
3019 	 * of 64 or 256 bytes depending on the WRIOP version.
3020 	 */
3021 	priv->rx_buf_size = ALIGN_DOWN(DPAA2_ETH_RX_BUF_SIZE, rx_buf_align);
3022 
3023 	/* tx buffer */
3024 	buf_layout.private_data_size = DPAA2_ETH_SWA_SIZE;
3025 	buf_layout.pass_timestamp = true;
3026 	buf_layout.pass_frame_status = true;
3027 	buf_layout.options = DPNI_BUF_LAYOUT_OPT_PRIVATE_DATA_SIZE |
3028 			     DPNI_BUF_LAYOUT_OPT_TIMESTAMP |
3029 			     DPNI_BUF_LAYOUT_OPT_FRAME_STATUS;
3030 	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
3031 				     DPNI_QUEUE_TX, &buf_layout);
3032 	if (err) {
3033 		dev_err(dev, "dpni_set_buffer_layout(TX) failed\n");
3034 		return err;
3035 	}
3036 
3037 	/* tx-confirm buffer */
3038 	buf_layout.options = DPNI_BUF_LAYOUT_OPT_TIMESTAMP |
3039 			     DPNI_BUF_LAYOUT_OPT_FRAME_STATUS;
3040 	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
3041 				     DPNI_QUEUE_TX_CONFIRM, &buf_layout);
3042 	if (err) {
3043 		dev_err(dev, "dpni_set_buffer_layout(TX_CONF) failed\n");
3044 		return err;
3045 	}
3046 
3047 	/* Now that we've set our tx buffer layout, retrieve the minimum
3048 	 * required tx data offset.
3049 	 */
3050 	err = dpni_get_tx_data_offset(priv->mc_io, 0, priv->mc_token,
3051 				      &priv->tx_data_offset);
3052 	if (err) {
3053 		dev_err(dev, "dpni_get_tx_data_offset() failed\n");
3054 		return err;
3055 	}
3056 
3057 	if ((priv->tx_data_offset % 64) != 0)
3058 		dev_warn(dev, "Tx data offset (%d) not a multiple of 64B\n",
3059 			 priv->tx_data_offset);
3060 
3061 	/* rx buffer */
3062 	buf_layout.pass_frame_status = true;
3063 	buf_layout.pass_parser_result = true;
3064 	buf_layout.data_align = rx_buf_align;
3065 	buf_layout.data_head_room = dpaa2_eth_rx_head_room(priv);
3066 	buf_layout.private_data_size = 0;
3067 	buf_layout.options = DPNI_BUF_LAYOUT_OPT_PARSER_RESULT |
3068 			     DPNI_BUF_LAYOUT_OPT_FRAME_STATUS |
3069 			     DPNI_BUF_LAYOUT_OPT_DATA_ALIGN |
3070 			     DPNI_BUF_LAYOUT_OPT_DATA_HEAD_ROOM |
3071 			     DPNI_BUF_LAYOUT_OPT_TIMESTAMP;
3072 	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
3073 				     DPNI_QUEUE_RX, &buf_layout);
3074 	if (err) {
3075 		dev_err(dev, "dpni_set_buffer_layout(RX) failed\n");
3076 		return err;
3077 	}
3078 
3079 	return 0;
3080 }
3081 
3082 #define DPNI_ENQUEUE_FQID_VER_MAJOR	7
3083 #define DPNI_ENQUEUE_FQID_VER_MINOR	9
3084 
3085 static inline int dpaa2_eth_enqueue_qd(struct dpaa2_eth_priv *priv,
3086 				       struct dpaa2_eth_fq *fq,
3087 				       struct dpaa2_fd *fd, u8 prio,
3088 				       u32 num_frames __always_unused,
3089 				       int *frames_enqueued)
3090 {
3091 	int err;
3092 
3093 	err = dpaa2_io_service_enqueue_qd(fq->channel->dpio,
3094 					  priv->tx_qdid, prio,
3095 					  fq->tx_qdbin, fd);
3096 	if (!err && frames_enqueued)
3097 		*frames_enqueued = 1;
3098 	return err;
3099 }
3100 
3101 static inline int dpaa2_eth_enqueue_fq_multiple(struct dpaa2_eth_priv *priv,
3102 						struct dpaa2_eth_fq *fq,
3103 						struct dpaa2_fd *fd,
3104 						u8 prio, u32 num_frames,
3105 						int *frames_enqueued)
3106 {
3107 	int err;
3108 
3109 	err = dpaa2_io_service_enqueue_multiple_fq(fq->channel->dpio,
3110 						   fq->tx_fqid[prio],
3111 						   fd, num_frames);
3112 
3113 	if (err == 0)
3114 		return -EBUSY;
3115 
3116 	if (frames_enqueued)
3117 		*frames_enqueued = err;
3118 	return 0;
3119 }
3120 
3121 static void dpaa2_eth_set_enqueue_mode(struct dpaa2_eth_priv *priv)
3122 {
3123 	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_ENQUEUE_FQID_VER_MAJOR,
3124 				   DPNI_ENQUEUE_FQID_VER_MINOR) < 0)
3125 		priv->enqueue = dpaa2_eth_enqueue_qd;
3126 	else
3127 		priv->enqueue = dpaa2_eth_enqueue_fq_multiple;
3128 }
3129 
3130 static int dpaa2_eth_set_pause(struct dpaa2_eth_priv *priv)
3131 {
3132 	struct device *dev = priv->net_dev->dev.parent;
3133 	struct dpni_link_cfg link_cfg = {0};
3134 	int err;
3135 
3136 	/* Get the default link options so we don't override other flags */
3137 	err = dpni_get_link_cfg(priv->mc_io, 0, priv->mc_token, &link_cfg);
3138 	if (err) {
3139 		dev_err(dev, "dpni_get_link_cfg() failed\n");
3140 		return err;
3141 	}
3142 
3143 	/* By default, enable both Rx and Tx pause frames */
3144 	link_cfg.options |= DPNI_LINK_OPT_PAUSE;
3145 	link_cfg.options &= ~DPNI_LINK_OPT_ASYM_PAUSE;
3146 	err = dpni_set_link_cfg(priv->mc_io, 0, priv->mc_token, &link_cfg);
3147 	if (err) {
3148 		dev_err(dev, "dpni_set_link_cfg() failed\n");
3149 		return err;
3150 	}
3151 
3152 	priv->link_state.options = link_cfg.options;
3153 
3154 	return 0;
3155 }
3156 
3157 static void dpaa2_eth_update_tx_fqids(struct dpaa2_eth_priv *priv)
3158 {
3159 	struct dpni_queue_id qid = {0};
3160 	struct dpaa2_eth_fq *fq;
3161 	struct dpni_queue queue;
3162 	int i, j, err;
3163 
3164 	/* We only use Tx FQIDs for FQID-based enqueue, so check
3165 	 * if DPNI version supports it before updating FQIDs
3166 	 */
3167 	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_ENQUEUE_FQID_VER_MAJOR,
3168 				   DPNI_ENQUEUE_FQID_VER_MINOR) < 0)
3169 		return;
3170 
3171 	for (i = 0; i < priv->num_fqs; i++) {
3172 		fq = &priv->fq[i];
3173 		if (fq->type != DPAA2_TX_CONF_FQ)
3174 			continue;
3175 		for (j = 0; j < dpaa2_eth_tc_count(priv); j++) {
3176 			err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3177 					     DPNI_QUEUE_TX, j, fq->flowid,
3178 					     &queue, &qid);
3179 			if (err)
3180 				goto out_err;
3181 
3182 			fq->tx_fqid[j] = qid.fqid;
3183 			if (fq->tx_fqid[j] == 0)
3184 				goto out_err;
3185 		}
3186 	}
3187 
3188 	priv->enqueue = dpaa2_eth_enqueue_fq_multiple;
3189 
3190 	return;
3191 
3192 out_err:
3193 	netdev_info(priv->net_dev,
3194 		    "Error reading Tx FQID, fallback to QDID-based enqueue\n");
3195 	priv->enqueue = dpaa2_eth_enqueue_qd;
3196 }
3197 
3198 /* Configure ingress classification based on VLAN PCP */
3199 static int dpaa2_eth_set_vlan_qos(struct dpaa2_eth_priv *priv)
3200 {
3201 	struct device *dev = priv->net_dev->dev.parent;
3202 	struct dpkg_profile_cfg kg_cfg = {0};
3203 	struct dpni_qos_tbl_cfg qos_cfg = {0};
3204 	struct dpni_rule_cfg key_params;
3205 	void *dma_mem, *key, *mask;
3206 	u8 key_size = 2;	/* VLAN TCI field */
3207 	int i, pcp, err;
3208 
3209 	/* VLAN-based classification only makes sense if we have multiple
3210 	 * traffic classes.
3211 	 * Also, we need to extract just the 3-bit PCP field from the VLAN
3212 	 * header and we can only do that by using a mask
3213 	 */
3214 	if (dpaa2_eth_tc_count(priv) == 1 || !dpaa2_eth_fs_mask_enabled(priv)) {
3215 		dev_dbg(dev, "VLAN-based QoS classification not supported\n");
3216 		return -EOPNOTSUPP;
3217 	}
3218 
3219 	dma_mem = kzalloc(DPAA2_CLASSIFIER_DMA_SIZE, GFP_KERNEL);
3220 	if (!dma_mem)
3221 		return -ENOMEM;
3222 
3223 	kg_cfg.num_extracts = 1;
3224 	kg_cfg.extracts[0].type = DPKG_EXTRACT_FROM_HDR;
3225 	kg_cfg.extracts[0].extract.from_hdr.prot = NET_PROT_VLAN;
3226 	kg_cfg.extracts[0].extract.from_hdr.type = DPKG_FULL_FIELD;
3227 	kg_cfg.extracts[0].extract.from_hdr.field = NH_FLD_VLAN_TCI;
3228 
3229 	err = dpni_prepare_key_cfg(&kg_cfg, dma_mem);
3230 	if (err) {
3231 		dev_err(dev, "dpni_prepare_key_cfg failed\n");
3232 		goto out_free_tbl;
3233 	}
3234 
3235 	/* set QoS table */
3236 	qos_cfg.default_tc = 0;
3237 	qos_cfg.discard_on_miss = 0;
3238 	qos_cfg.key_cfg_iova = dma_map_single(dev, dma_mem,
3239 					      DPAA2_CLASSIFIER_DMA_SIZE,
3240 					      DMA_TO_DEVICE);
3241 	if (dma_mapping_error(dev, qos_cfg.key_cfg_iova)) {
3242 		dev_err(dev, "QoS table DMA mapping failed\n");
3243 		err = -ENOMEM;
3244 		goto out_free_tbl;
3245 	}
3246 
3247 	err = dpni_set_qos_table(priv->mc_io, 0, priv->mc_token, &qos_cfg);
3248 	if (err) {
3249 		dev_err(dev, "dpni_set_qos_table failed\n");
3250 		goto out_unmap_tbl;
3251 	}
3252 
3253 	/* Add QoS table entries */
3254 	key = kzalloc(key_size * 2, GFP_KERNEL);
3255 	if (!key) {
3256 		err = -ENOMEM;
3257 		goto out_unmap_tbl;
3258 	}
3259 	mask = key + key_size;
3260 	*(__be16 *)mask = cpu_to_be16(VLAN_PRIO_MASK);
3261 
3262 	key_params.key_iova = dma_map_single(dev, key, key_size * 2,
3263 					     DMA_TO_DEVICE);
3264 	if (dma_mapping_error(dev, key_params.key_iova)) {
3265 		dev_err(dev, "Qos table entry DMA mapping failed\n");
3266 		err = -ENOMEM;
3267 		goto out_free_key;
3268 	}
3269 
3270 	key_params.mask_iova = key_params.key_iova + key_size;
3271 	key_params.key_size = key_size;
3272 
3273 	/* We add rules for PCP-based distribution starting with highest
3274 	 * priority (VLAN PCP = 7). If this DPNI doesn't have enough traffic
3275 	 * classes to accommodate all priority levels, the lowest ones end up
3276 	 * on TC 0 which was configured as default
3277 	 */
3278 	for (i = dpaa2_eth_tc_count(priv) - 1, pcp = 7; i >= 0; i--, pcp--) {
3279 		*(__be16 *)key = cpu_to_be16(pcp << VLAN_PRIO_SHIFT);
3280 		dma_sync_single_for_device(dev, key_params.key_iova,
3281 					   key_size * 2, DMA_TO_DEVICE);
3282 
3283 		err = dpni_add_qos_entry(priv->mc_io, 0, priv->mc_token,
3284 					 &key_params, i, i);
3285 		if (err) {
3286 			dev_err(dev, "dpni_add_qos_entry failed\n");
3287 			dpni_clear_qos_table(priv->mc_io, 0, priv->mc_token);
3288 			goto out_unmap_key;
3289 		}
3290 	}
3291 
3292 	priv->vlan_cls_enabled = true;
3293 
3294 	/* Table and key memory is not persistent, clean everything up after
3295 	 * configuration is finished
3296 	 */
3297 out_unmap_key:
3298 	dma_unmap_single(dev, key_params.key_iova, key_size * 2, DMA_TO_DEVICE);
3299 out_free_key:
3300 	kfree(key);
3301 out_unmap_tbl:
3302 	dma_unmap_single(dev, qos_cfg.key_cfg_iova, DPAA2_CLASSIFIER_DMA_SIZE,
3303 			 DMA_TO_DEVICE);
3304 out_free_tbl:
3305 	kfree(dma_mem);
3306 
3307 	return err;
3308 }
3309 
3310 /* Configure the DPNI object this interface is associated with */
3311 static int dpaa2_eth_setup_dpni(struct fsl_mc_device *ls_dev)
3312 {
3313 	struct device *dev = &ls_dev->dev;
3314 	struct dpaa2_eth_priv *priv;
3315 	struct net_device *net_dev;
3316 	int err;
3317 
3318 	net_dev = dev_get_drvdata(dev);
3319 	priv = netdev_priv(net_dev);
3320 
3321 	/* get a handle for the DPNI object */
3322 	err = dpni_open(priv->mc_io, 0, ls_dev->obj_desc.id, &priv->mc_token);
3323 	if (err) {
3324 		dev_err(dev, "dpni_open() failed\n");
3325 		return err;
3326 	}
3327 
3328 	/* Check if we can work with this DPNI object */
3329 	err = dpni_get_api_version(priv->mc_io, 0, &priv->dpni_ver_major,
3330 				   &priv->dpni_ver_minor);
3331 	if (err) {
3332 		dev_err(dev, "dpni_get_api_version() failed\n");
3333 		goto close;
3334 	}
3335 	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_VER_MAJOR, DPNI_VER_MINOR) < 0) {
3336 		dev_err(dev, "DPNI version %u.%u not supported, need >= %u.%u\n",
3337 			priv->dpni_ver_major, priv->dpni_ver_minor,
3338 			DPNI_VER_MAJOR, DPNI_VER_MINOR);
3339 		err = -ENOTSUPP;
3340 		goto close;
3341 	}
3342 
3343 	ls_dev->mc_io = priv->mc_io;
3344 	ls_dev->mc_handle = priv->mc_token;
3345 
3346 	err = dpni_reset(priv->mc_io, 0, priv->mc_token);
3347 	if (err) {
3348 		dev_err(dev, "dpni_reset() failed\n");
3349 		goto close;
3350 	}
3351 
3352 	err = dpni_get_attributes(priv->mc_io, 0, priv->mc_token,
3353 				  &priv->dpni_attrs);
3354 	if (err) {
3355 		dev_err(dev, "dpni_get_attributes() failed (err=%d)\n", err);
3356 		goto close;
3357 	}
3358 
3359 	err = dpaa2_eth_set_buffer_layout(priv);
3360 	if (err)
3361 		goto close;
3362 
3363 	dpaa2_eth_set_enqueue_mode(priv);
3364 
3365 	/* Enable pause frame support */
3366 	if (dpaa2_eth_has_pause_support(priv)) {
3367 		err = dpaa2_eth_set_pause(priv);
3368 		if (err)
3369 			goto close;
3370 	}
3371 
3372 	err = dpaa2_eth_set_vlan_qos(priv);
3373 	if (err && err != -EOPNOTSUPP)
3374 		goto close;
3375 
3376 	priv->cls_rules = devm_kcalloc(dev, dpaa2_eth_fs_count(priv),
3377 				       sizeof(struct dpaa2_eth_cls_rule),
3378 				       GFP_KERNEL);
3379 	if (!priv->cls_rules) {
3380 		err = -ENOMEM;
3381 		goto close;
3382 	}
3383 
3384 	return 0;
3385 
3386 close:
3387 	dpni_close(priv->mc_io, 0, priv->mc_token);
3388 
3389 	return err;
3390 }
3391 
3392 static void dpaa2_eth_free_dpni(struct dpaa2_eth_priv *priv)
3393 {
3394 	int err;
3395 
3396 	err = dpni_reset(priv->mc_io, 0, priv->mc_token);
3397 	if (err)
3398 		netdev_warn(priv->net_dev, "dpni_reset() failed (err %d)\n",
3399 			    err);
3400 
3401 	dpni_close(priv->mc_io, 0, priv->mc_token);
3402 }
3403 
3404 static int dpaa2_eth_setup_rx_flow(struct dpaa2_eth_priv *priv,
3405 				   struct dpaa2_eth_fq *fq)
3406 {
3407 	struct device *dev = priv->net_dev->dev.parent;
3408 	struct dpni_queue queue;
3409 	struct dpni_queue_id qid;
3410 	int err;
3411 
3412 	err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3413 			     DPNI_QUEUE_RX, fq->tc, fq->flowid, &queue, &qid);
3414 	if (err) {
3415 		dev_err(dev, "dpni_get_queue(RX) failed\n");
3416 		return err;
3417 	}
3418 
3419 	fq->fqid = qid.fqid;
3420 
3421 	queue.destination.id = fq->channel->dpcon_id;
3422 	queue.destination.type = DPNI_DEST_DPCON;
3423 	queue.destination.priority = 1;
3424 	queue.user_context = (u64)(uintptr_t)fq;
3425 	err = dpni_set_queue(priv->mc_io, 0, priv->mc_token,
3426 			     DPNI_QUEUE_RX, fq->tc, fq->flowid,
3427 			     DPNI_QUEUE_OPT_USER_CTX | DPNI_QUEUE_OPT_DEST,
3428 			     &queue);
3429 	if (err) {
3430 		dev_err(dev, "dpni_set_queue(RX) failed\n");
3431 		return err;
3432 	}
3433 
3434 	/* xdp_rxq setup */
3435 	/* only once for each channel */
3436 	if (fq->tc > 0)
3437 		return 0;
3438 
3439 	err = xdp_rxq_info_reg(&fq->channel->xdp_rxq, priv->net_dev,
3440 			       fq->flowid, 0);
3441 	if (err) {
3442 		dev_err(dev, "xdp_rxq_info_reg failed\n");
3443 		return err;
3444 	}
3445 
3446 	err = xdp_rxq_info_reg_mem_model(&fq->channel->xdp_rxq,
3447 					 MEM_TYPE_PAGE_ORDER0, NULL);
3448 	if (err) {
3449 		dev_err(dev, "xdp_rxq_info_reg_mem_model failed\n");
3450 		return err;
3451 	}
3452 
3453 	return 0;
3454 }
3455 
3456 static int dpaa2_eth_setup_tx_flow(struct dpaa2_eth_priv *priv,
3457 				   struct dpaa2_eth_fq *fq)
3458 {
3459 	struct device *dev = priv->net_dev->dev.parent;
3460 	struct dpni_queue queue;
3461 	struct dpni_queue_id qid;
3462 	int i, err;
3463 
3464 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3465 		err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3466 				     DPNI_QUEUE_TX, i, fq->flowid,
3467 				     &queue, &qid);
3468 		if (err) {
3469 			dev_err(dev, "dpni_get_queue(TX) failed\n");
3470 			return err;
3471 		}
3472 		fq->tx_fqid[i] = qid.fqid;
3473 	}
3474 
3475 	/* All Tx queues belonging to the same flowid have the same qdbin */
3476 	fq->tx_qdbin = qid.qdbin;
3477 
3478 	err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3479 			     DPNI_QUEUE_TX_CONFIRM, 0, fq->flowid,
3480 			     &queue, &qid);
3481 	if (err) {
3482 		dev_err(dev, "dpni_get_queue(TX_CONF) failed\n");
3483 		return err;
3484 	}
3485 
3486 	fq->fqid = qid.fqid;
3487 
3488 	queue.destination.id = fq->channel->dpcon_id;
3489 	queue.destination.type = DPNI_DEST_DPCON;
3490 	queue.destination.priority = 0;
3491 	queue.user_context = (u64)(uintptr_t)fq;
3492 	err = dpni_set_queue(priv->mc_io, 0, priv->mc_token,
3493 			     DPNI_QUEUE_TX_CONFIRM, 0, fq->flowid,
3494 			     DPNI_QUEUE_OPT_USER_CTX | DPNI_QUEUE_OPT_DEST,
3495 			     &queue);
3496 	if (err) {
3497 		dev_err(dev, "dpni_set_queue(TX_CONF) failed\n");
3498 		return err;
3499 	}
3500 
3501 	return 0;
3502 }
3503 
3504 static int setup_rx_err_flow(struct dpaa2_eth_priv *priv,
3505 			     struct dpaa2_eth_fq *fq)
3506 {
3507 	struct device *dev = priv->net_dev->dev.parent;
3508 	struct dpni_queue q = { { 0 } };
3509 	struct dpni_queue_id qid;
3510 	u8 q_opt = DPNI_QUEUE_OPT_USER_CTX | DPNI_QUEUE_OPT_DEST;
3511 	int err;
3512 
3513 	err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3514 			     DPNI_QUEUE_RX_ERR, 0, 0, &q, &qid);
3515 	if (err) {
3516 		dev_err(dev, "dpni_get_queue() failed (%d)\n", err);
3517 		return err;
3518 	}
3519 
3520 	fq->fqid = qid.fqid;
3521 
3522 	q.destination.id = fq->channel->dpcon_id;
3523 	q.destination.type = DPNI_DEST_DPCON;
3524 	q.destination.priority = 1;
3525 	q.user_context = (u64)(uintptr_t)fq;
3526 	err = dpni_set_queue(priv->mc_io, 0, priv->mc_token,
3527 			     DPNI_QUEUE_RX_ERR, 0, 0, q_opt, &q);
3528 	if (err) {
3529 		dev_err(dev, "dpni_set_queue() failed (%d)\n", err);
3530 		return err;
3531 	}
3532 
3533 	return 0;
3534 }
3535 
3536 /* Supported header fields for Rx hash distribution key */
3537 static const struct dpaa2_eth_dist_fields dist_fields[] = {
3538 	{
3539 		/* L2 header */
3540 		.rxnfc_field = RXH_L2DA,
3541 		.cls_prot = NET_PROT_ETH,
3542 		.cls_field = NH_FLD_ETH_DA,
3543 		.id = DPAA2_ETH_DIST_ETHDST,
3544 		.size = 6,
3545 	}, {
3546 		.cls_prot = NET_PROT_ETH,
3547 		.cls_field = NH_FLD_ETH_SA,
3548 		.id = DPAA2_ETH_DIST_ETHSRC,
3549 		.size = 6,
3550 	}, {
3551 		/* This is the last ethertype field parsed:
3552 		 * depending on frame format, it can be the MAC ethertype
3553 		 * or the VLAN etype.
3554 		 */
3555 		.cls_prot = NET_PROT_ETH,
3556 		.cls_field = NH_FLD_ETH_TYPE,
3557 		.id = DPAA2_ETH_DIST_ETHTYPE,
3558 		.size = 2,
3559 	}, {
3560 		/* VLAN header */
3561 		.rxnfc_field = RXH_VLAN,
3562 		.cls_prot = NET_PROT_VLAN,
3563 		.cls_field = NH_FLD_VLAN_TCI,
3564 		.id = DPAA2_ETH_DIST_VLAN,
3565 		.size = 2,
3566 	}, {
3567 		/* IP header */
3568 		.rxnfc_field = RXH_IP_SRC,
3569 		.cls_prot = NET_PROT_IP,
3570 		.cls_field = NH_FLD_IP_SRC,
3571 		.id = DPAA2_ETH_DIST_IPSRC,
3572 		.size = 4,
3573 	}, {
3574 		.rxnfc_field = RXH_IP_DST,
3575 		.cls_prot = NET_PROT_IP,
3576 		.cls_field = NH_FLD_IP_DST,
3577 		.id = DPAA2_ETH_DIST_IPDST,
3578 		.size = 4,
3579 	}, {
3580 		.rxnfc_field = RXH_L3_PROTO,
3581 		.cls_prot = NET_PROT_IP,
3582 		.cls_field = NH_FLD_IP_PROTO,
3583 		.id = DPAA2_ETH_DIST_IPPROTO,
3584 		.size = 1,
3585 	}, {
3586 		/* Using UDP ports, this is functionally equivalent to raw
3587 		 * byte pairs from L4 header.
3588 		 */
3589 		.rxnfc_field = RXH_L4_B_0_1,
3590 		.cls_prot = NET_PROT_UDP,
3591 		.cls_field = NH_FLD_UDP_PORT_SRC,
3592 		.id = DPAA2_ETH_DIST_L4SRC,
3593 		.size = 2,
3594 	}, {
3595 		.rxnfc_field = RXH_L4_B_2_3,
3596 		.cls_prot = NET_PROT_UDP,
3597 		.cls_field = NH_FLD_UDP_PORT_DST,
3598 		.id = DPAA2_ETH_DIST_L4DST,
3599 		.size = 2,
3600 	},
3601 };
3602 
3603 /* Configure the Rx hash key using the legacy API */
3604 static int dpaa2_eth_config_legacy_hash_key(struct dpaa2_eth_priv *priv, dma_addr_t key)
3605 {
3606 	struct device *dev = priv->net_dev->dev.parent;
3607 	struct dpni_rx_tc_dist_cfg dist_cfg;
3608 	int i, err = 0;
3609 
3610 	memset(&dist_cfg, 0, sizeof(dist_cfg));
3611 
3612 	dist_cfg.key_cfg_iova = key;
3613 	dist_cfg.dist_size = dpaa2_eth_queue_count(priv);
3614 	dist_cfg.dist_mode = DPNI_DIST_MODE_HASH;
3615 
3616 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3617 		err = dpni_set_rx_tc_dist(priv->mc_io, 0, priv->mc_token,
3618 					  i, &dist_cfg);
3619 		if (err) {
3620 			dev_err(dev, "dpni_set_rx_tc_dist failed\n");
3621 			break;
3622 		}
3623 	}
3624 
3625 	return err;
3626 }
3627 
3628 /* Configure the Rx hash key using the new API */
3629 static int dpaa2_eth_config_hash_key(struct dpaa2_eth_priv *priv, dma_addr_t key)
3630 {
3631 	struct device *dev = priv->net_dev->dev.parent;
3632 	struct dpni_rx_dist_cfg dist_cfg;
3633 	int i, err = 0;
3634 
3635 	memset(&dist_cfg, 0, sizeof(dist_cfg));
3636 
3637 	dist_cfg.key_cfg_iova = key;
3638 	dist_cfg.dist_size = dpaa2_eth_queue_count(priv);
3639 	dist_cfg.enable = 1;
3640 
3641 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3642 		dist_cfg.tc = i;
3643 		err = dpni_set_rx_hash_dist(priv->mc_io, 0, priv->mc_token,
3644 					    &dist_cfg);
3645 		if (err) {
3646 			dev_err(dev, "dpni_set_rx_hash_dist failed\n");
3647 			break;
3648 		}
3649 
3650 		/* If the flow steering / hashing key is shared between all
3651 		 * traffic classes, install it just once
3652 		 */
3653 		if (priv->dpni_attrs.options & DPNI_OPT_SHARED_FS)
3654 			break;
3655 	}
3656 
3657 	return err;
3658 }
3659 
3660 /* Configure the Rx flow classification key */
3661 static int dpaa2_eth_config_cls_key(struct dpaa2_eth_priv *priv, dma_addr_t key)
3662 {
3663 	struct device *dev = priv->net_dev->dev.parent;
3664 	struct dpni_rx_dist_cfg dist_cfg;
3665 	int i, err = 0;
3666 
3667 	memset(&dist_cfg, 0, sizeof(dist_cfg));
3668 
3669 	dist_cfg.key_cfg_iova = key;
3670 	dist_cfg.dist_size = dpaa2_eth_queue_count(priv);
3671 	dist_cfg.enable = 1;
3672 
3673 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3674 		dist_cfg.tc = i;
3675 		err = dpni_set_rx_fs_dist(priv->mc_io, 0, priv->mc_token,
3676 					  &dist_cfg);
3677 		if (err) {
3678 			dev_err(dev, "dpni_set_rx_fs_dist failed\n");
3679 			break;
3680 		}
3681 
3682 		/* If the flow steering / hashing key is shared between all
3683 		 * traffic classes, install it just once
3684 		 */
3685 		if (priv->dpni_attrs.options & DPNI_OPT_SHARED_FS)
3686 			break;
3687 	}
3688 
3689 	return err;
3690 }
3691 
3692 /* Size of the Rx flow classification key */
3693 int dpaa2_eth_cls_key_size(u64 fields)
3694 {
3695 	int i, size = 0;
3696 
3697 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3698 		if (!(fields & dist_fields[i].id))
3699 			continue;
3700 		size += dist_fields[i].size;
3701 	}
3702 
3703 	return size;
3704 }
3705 
3706 /* Offset of header field in Rx classification key */
3707 int dpaa2_eth_cls_fld_off(int prot, int field)
3708 {
3709 	int i, off = 0;
3710 
3711 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3712 		if (dist_fields[i].cls_prot == prot &&
3713 		    dist_fields[i].cls_field == field)
3714 			return off;
3715 		off += dist_fields[i].size;
3716 	}
3717 
3718 	WARN_ONCE(1, "Unsupported header field used for Rx flow cls\n");
3719 	return 0;
3720 }
3721 
3722 /* Prune unused fields from the classification rule.
3723  * Used when masking is not supported
3724  */
3725 void dpaa2_eth_cls_trim_rule(void *key_mem, u64 fields)
3726 {
3727 	int off = 0, new_off = 0;
3728 	int i, size;
3729 
3730 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3731 		size = dist_fields[i].size;
3732 		if (dist_fields[i].id & fields) {
3733 			memcpy(key_mem + new_off, key_mem + off, size);
3734 			new_off += size;
3735 		}
3736 		off += size;
3737 	}
3738 }
3739 
3740 /* Set Rx distribution (hash or flow classification) key
3741  * flags is a combination of RXH_ bits
3742  */
3743 static int dpaa2_eth_set_dist_key(struct net_device *net_dev,
3744 				  enum dpaa2_eth_rx_dist type, u64 flags)
3745 {
3746 	struct device *dev = net_dev->dev.parent;
3747 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
3748 	struct dpkg_profile_cfg cls_cfg;
3749 	u32 rx_hash_fields = 0;
3750 	dma_addr_t key_iova;
3751 	u8 *dma_mem;
3752 	int i;
3753 	int err = 0;
3754 
3755 	memset(&cls_cfg, 0, sizeof(cls_cfg));
3756 
3757 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3758 		struct dpkg_extract *key =
3759 			&cls_cfg.extracts[cls_cfg.num_extracts];
3760 
3761 		/* For both Rx hashing and classification keys
3762 		 * we set only the selected fields.
3763 		 */
3764 		if (!(flags & dist_fields[i].id))
3765 			continue;
3766 		if (type == DPAA2_ETH_RX_DIST_HASH)
3767 			rx_hash_fields |= dist_fields[i].rxnfc_field;
3768 
3769 		if (cls_cfg.num_extracts >= DPKG_MAX_NUM_OF_EXTRACTS) {
3770 			dev_err(dev, "error adding key extraction rule, too many rules?\n");
3771 			return -E2BIG;
3772 		}
3773 
3774 		key->type = DPKG_EXTRACT_FROM_HDR;
3775 		key->extract.from_hdr.prot = dist_fields[i].cls_prot;
3776 		key->extract.from_hdr.type = DPKG_FULL_FIELD;
3777 		key->extract.from_hdr.field = dist_fields[i].cls_field;
3778 		cls_cfg.num_extracts++;
3779 	}
3780 
3781 	dma_mem = kzalloc(DPAA2_CLASSIFIER_DMA_SIZE, GFP_KERNEL);
3782 	if (!dma_mem)
3783 		return -ENOMEM;
3784 
3785 	err = dpni_prepare_key_cfg(&cls_cfg, dma_mem);
3786 	if (err) {
3787 		dev_err(dev, "dpni_prepare_key_cfg error %d\n", err);
3788 		goto free_key;
3789 	}
3790 
3791 	/* Prepare for setting the rx dist */
3792 	key_iova = dma_map_single(dev, dma_mem, DPAA2_CLASSIFIER_DMA_SIZE,
3793 				  DMA_TO_DEVICE);
3794 	if (dma_mapping_error(dev, key_iova)) {
3795 		dev_err(dev, "DMA mapping failed\n");
3796 		err = -ENOMEM;
3797 		goto free_key;
3798 	}
3799 
3800 	if (type == DPAA2_ETH_RX_DIST_HASH) {
3801 		if (dpaa2_eth_has_legacy_dist(priv))
3802 			err = dpaa2_eth_config_legacy_hash_key(priv, key_iova);
3803 		else
3804 			err = dpaa2_eth_config_hash_key(priv, key_iova);
3805 	} else {
3806 		err = dpaa2_eth_config_cls_key(priv, key_iova);
3807 	}
3808 
3809 	dma_unmap_single(dev, key_iova, DPAA2_CLASSIFIER_DMA_SIZE,
3810 			 DMA_TO_DEVICE);
3811 	if (!err && type == DPAA2_ETH_RX_DIST_HASH)
3812 		priv->rx_hash_fields = rx_hash_fields;
3813 
3814 free_key:
3815 	kfree(dma_mem);
3816 	return err;
3817 }
3818 
3819 int dpaa2_eth_set_hash(struct net_device *net_dev, u64 flags)
3820 {
3821 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
3822 	u64 key = 0;
3823 	int i;
3824 
3825 	if (!dpaa2_eth_hash_enabled(priv))
3826 		return -EOPNOTSUPP;
3827 
3828 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++)
3829 		if (dist_fields[i].rxnfc_field & flags)
3830 			key |= dist_fields[i].id;
3831 
3832 	return dpaa2_eth_set_dist_key(net_dev, DPAA2_ETH_RX_DIST_HASH, key);
3833 }
3834 
3835 int dpaa2_eth_set_cls(struct net_device *net_dev, u64 flags)
3836 {
3837 	return dpaa2_eth_set_dist_key(net_dev, DPAA2_ETH_RX_DIST_CLS, flags);
3838 }
3839 
3840 static int dpaa2_eth_set_default_cls(struct dpaa2_eth_priv *priv)
3841 {
3842 	struct device *dev = priv->net_dev->dev.parent;
3843 	int err;
3844 
3845 	/* Check if we actually support Rx flow classification */
3846 	if (dpaa2_eth_has_legacy_dist(priv)) {
3847 		dev_dbg(dev, "Rx cls not supported by current MC version\n");
3848 		return -EOPNOTSUPP;
3849 	}
3850 
3851 	if (!dpaa2_eth_fs_enabled(priv)) {
3852 		dev_dbg(dev, "Rx cls disabled in DPNI options\n");
3853 		return -EOPNOTSUPP;
3854 	}
3855 
3856 	if (!dpaa2_eth_hash_enabled(priv)) {
3857 		dev_dbg(dev, "Rx cls disabled for single queue DPNIs\n");
3858 		return -EOPNOTSUPP;
3859 	}
3860 
3861 	/* If there is no support for masking in the classification table,
3862 	 * we don't set a default key, as it will depend on the rules
3863 	 * added by the user at runtime.
3864 	 */
3865 	if (!dpaa2_eth_fs_mask_enabled(priv))
3866 		goto out;
3867 
3868 	err = dpaa2_eth_set_cls(priv->net_dev, DPAA2_ETH_DIST_ALL);
3869 	if (err)
3870 		return err;
3871 
3872 out:
3873 	priv->rx_cls_enabled = 1;
3874 
3875 	return 0;
3876 }
3877 
3878 /* Bind the DPNI to its needed objects and resources: buffer pool, DPIOs,
3879  * frame queues and channels
3880  */
3881 static int dpaa2_eth_bind_dpni(struct dpaa2_eth_priv *priv)
3882 {
3883 	struct net_device *net_dev = priv->net_dev;
3884 	struct device *dev = net_dev->dev.parent;
3885 	struct dpni_pools_cfg pools_params;
3886 	struct dpni_error_cfg err_cfg;
3887 	int err = 0;
3888 	int i;
3889 
3890 	pools_params.num_dpbp = 1;
3891 	pools_params.pools[0].dpbp_id = priv->dpbp_dev->obj_desc.id;
3892 	pools_params.pools[0].backup_pool = 0;
3893 	pools_params.pools[0].buffer_size = priv->rx_buf_size;
3894 	err = dpni_set_pools(priv->mc_io, 0, priv->mc_token, &pools_params);
3895 	if (err) {
3896 		dev_err(dev, "dpni_set_pools() failed\n");
3897 		return err;
3898 	}
3899 
3900 	/* have the interface implicitly distribute traffic based on
3901 	 * the default hash key
3902 	 */
3903 	err = dpaa2_eth_set_hash(net_dev, DPAA2_RXH_DEFAULT);
3904 	if (err && err != -EOPNOTSUPP)
3905 		dev_err(dev, "Failed to configure hashing\n");
3906 
3907 	/* Configure the flow classification key; it includes all
3908 	 * supported header fields and cannot be modified at runtime
3909 	 */
3910 	err = dpaa2_eth_set_default_cls(priv);
3911 	if (err && err != -EOPNOTSUPP)
3912 		dev_err(dev, "Failed to configure Rx classification key\n");
3913 
3914 	/* Configure handling of error frames */
3915 	err_cfg.errors = DPAA2_FAS_RX_ERR_MASK;
3916 	err_cfg.set_frame_annotation = 1;
3917 	err_cfg.error_action = DPNI_ERROR_ACTION_DISCARD;
3918 	err = dpni_set_errors_behavior(priv->mc_io, 0, priv->mc_token,
3919 				       &err_cfg);
3920 	if (err) {
3921 		dev_err(dev, "dpni_set_errors_behavior failed\n");
3922 		return err;
3923 	}
3924 
3925 	/* Configure Rx and Tx conf queues to generate CDANs */
3926 	for (i = 0; i < priv->num_fqs; i++) {
3927 		switch (priv->fq[i].type) {
3928 		case DPAA2_RX_FQ:
3929 			err = dpaa2_eth_setup_rx_flow(priv, &priv->fq[i]);
3930 			break;
3931 		case DPAA2_TX_CONF_FQ:
3932 			err = dpaa2_eth_setup_tx_flow(priv, &priv->fq[i]);
3933 			break;
3934 		case DPAA2_RX_ERR_FQ:
3935 			err = setup_rx_err_flow(priv, &priv->fq[i]);
3936 			break;
3937 		default:
3938 			dev_err(dev, "Invalid FQ type %d\n", priv->fq[i].type);
3939 			return -EINVAL;
3940 		}
3941 		if (err)
3942 			return err;
3943 	}
3944 
3945 	err = dpni_get_qdid(priv->mc_io, 0, priv->mc_token,
3946 			    DPNI_QUEUE_TX, &priv->tx_qdid);
3947 	if (err) {
3948 		dev_err(dev, "dpni_get_qdid() failed\n");
3949 		return err;
3950 	}
3951 
3952 	return 0;
3953 }
3954 
3955 /* Allocate rings for storing incoming frame descriptors */
3956 static int dpaa2_eth_alloc_rings(struct dpaa2_eth_priv *priv)
3957 {
3958 	struct net_device *net_dev = priv->net_dev;
3959 	struct device *dev = net_dev->dev.parent;
3960 	int i;
3961 
3962 	for (i = 0; i < priv->num_channels; i++) {
3963 		priv->channel[i]->store =
3964 			dpaa2_io_store_create(DPAA2_ETH_STORE_SIZE, dev);
3965 		if (!priv->channel[i]->store) {
3966 			netdev_err(net_dev, "dpaa2_io_store_create() failed\n");
3967 			goto err_ring;
3968 		}
3969 	}
3970 
3971 	return 0;
3972 
3973 err_ring:
3974 	for (i = 0; i < priv->num_channels; i++) {
3975 		if (!priv->channel[i]->store)
3976 			break;
3977 		dpaa2_io_store_destroy(priv->channel[i]->store);
3978 	}
3979 
3980 	return -ENOMEM;
3981 }
3982 
3983 static void dpaa2_eth_free_rings(struct dpaa2_eth_priv *priv)
3984 {
3985 	int i;
3986 
3987 	for (i = 0; i < priv->num_channels; i++)
3988 		dpaa2_io_store_destroy(priv->channel[i]->store);
3989 }
3990 
3991 static int dpaa2_eth_set_mac_addr(struct dpaa2_eth_priv *priv)
3992 {
3993 	struct net_device *net_dev = priv->net_dev;
3994 	struct device *dev = net_dev->dev.parent;
3995 	u8 mac_addr[ETH_ALEN], dpni_mac_addr[ETH_ALEN];
3996 	int err;
3997 
3998 	/* Get firmware address, if any */
3999 	err = dpni_get_port_mac_addr(priv->mc_io, 0, priv->mc_token, mac_addr);
4000 	if (err) {
4001 		dev_err(dev, "dpni_get_port_mac_addr() failed\n");
4002 		return err;
4003 	}
4004 
4005 	/* Get DPNI attributes address, if any */
4006 	err = dpni_get_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
4007 					dpni_mac_addr);
4008 	if (err) {
4009 		dev_err(dev, "dpni_get_primary_mac_addr() failed\n");
4010 		return err;
4011 	}
4012 
4013 	/* First check if firmware has any address configured by bootloader */
4014 	if (!is_zero_ether_addr(mac_addr)) {
4015 		/* If the DPMAC addr != DPNI addr, update it */
4016 		if (!ether_addr_equal(mac_addr, dpni_mac_addr)) {
4017 			err = dpni_set_primary_mac_addr(priv->mc_io, 0,
4018 							priv->mc_token,
4019 							mac_addr);
4020 			if (err) {
4021 				dev_err(dev, "dpni_set_primary_mac_addr() failed\n");
4022 				return err;
4023 			}
4024 		}
4025 		eth_hw_addr_set(net_dev, mac_addr);
4026 	} else if (is_zero_ether_addr(dpni_mac_addr)) {
4027 		/* No MAC address configured, fill in net_dev->dev_addr
4028 		 * with a random one
4029 		 */
4030 		eth_hw_addr_random(net_dev);
4031 		dev_dbg_once(dev, "device(s) have all-zero hwaddr, replaced with random\n");
4032 
4033 		err = dpni_set_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
4034 						net_dev->dev_addr);
4035 		if (err) {
4036 			dev_err(dev, "dpni_set_primary_mac_addr() failed\n");
4037 			return err;
4038 		}
4039 
4040 		/* Override NET_ADDR_RANDOM set by eth_hw_addr_random(); for all
4041 		 * practical purposes, this will be our "permanent" mac address,
4042 		 * at least until the next reboot. This move will also permit
4043 		 * register_netdevice() to properly fill up net_dev->perm_addr.
4044 		 */
4045 		net_dev->addr_assign_type = NET_ADDR_PERM;
4046 	} else {
4047 		/* NET_ADDR_PERM is default, all we have to do is
4048 		 * fill in the device addr.
4049 		 */
4050 		eth_hw_addr_set(net_dev, dpni_mac_addr);
4051 	}
4052 
4053 	return 0;
4054 }
4055 
4056 static int dpaa2_eth_netdev_init(struct net_device *net_dev)
4057 {
4058 	struct device *dev = net_dev->dev.parent;
4059 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
4060 	u32 options = priv->dpni_attrs.options;
4061 	u64 supported = 0, not_supported = 0;
4062 	u8 bcast_addr[ETH_ALEN];
4063 	u8 num_queues;
4064 	int err;
4065 
4066 	net_dev->netdev_ops = &dpaa2_eth_ops;
4067 	net_dev->ethtool_ops = &dpaa2_ethtool_ops;
4068 
4069 	err = dpaa2_eth_set_mac_addr(priv);
4070 	if (err)
4071 		return err;
4072 
4073 	/* Explicitly add the broadcast address to the MAC filtering table */
4074 	eth_broadcast_addr(bcast_addr);
4075 	err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token, bcast_addr);
4076 	if (err) {
4077 		dev_err(dev, "dpni_add_mac_addr() failed\n");
4078 		return err;
4079 	}
4080 
4081 	/* Set MTU upper limit; lower limit is 68B (default value) */
4082 	net_dev->max_mtu = DPAA2_ETH_MAX_MTU;
4083 	err = dpni_set_max_frame_length(priv->mc_io, 0, priv->mc_token,
4084 					DPAA2_ETH_MFL);
4085 	if (err) {
4086 		dev_err(dev, "dpni_set_max_frame_length() failed\n");
4087 		return err;
4088 	}
4089 
4090 	/* Set actual number of queues in the net device */
4091 	num_queues = dpaa2_eth_queue_count(priv);
4092 	err = netif_set_real_num_tx_queues(net_dev, num_queues);
4093 	if (err) {
4094 		dev_err(dev, "netif_set_real_num_tx_queues() failed\n");
4095 		return err;
4096 	}
4097 	err = netif_set_real_num_rx_queues(net_dev, num_queues);
4098 	if (err) {
4099 		dev_err(dev, "netif_set_real_num_rx_queues() failed\n");
4100 		return err;
4101 	}
4102 
4103 	/* Capabilities listing */
4104 	supported |= IFF_LIVE_ADDR_CHANGE;
4105 
4106 	if (options & DPNI_OPT_NO_MAC_FILTER)
4107 		not_supported |= IFF_UNICAST_FLT;
4108 	else
4109 		supported |= IFF_UNICAST_FLT;
4110 
4111 	net_dev->priv_flags |= supported;
4112 	net_dev->priv_flags &= ~not_supported;
4113 
4114 	/* Features */
4115 	net_dev->features = NETIF_F_RXCSUM |
4116 			    NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4117 			    NETIF_F_SG | NETIF_F_HIGHDMA |
4118 			    NETIF_F_LLTX | NETIF_F_HW_TC;
4119 	net_dev->hw_features = net_dev->features;
4120 
4121 	if (priv->dpni_attrs.vlan_filter_entries)
4122 		net_dev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
4123 
4124 	return 0;
4125 }
4126 
4127 static int dpaa2_eth_poll_link_state(void *arg)
4128 {
4129 	struct dpaa2_eth_priv *priv = (struct dpaa2_eth_priv *)arg;
4130 	int err;
4131 
4132 	while (!kthread_should_stop()) {
4133 		err = dpaa2_eth_link_state_update(priv);
4134 		if (unlikely(err))
4135 			return err;
4136 
4137 		msleep(DPAA2_ETH_LINK_STATE_REFRESH);
4138 	}
4139 
4140 	return 0;
4141 }
4142 
4143 static int dpaa2_eth_connect_mac(struct dpaa2_eth_priv *priv)
4144 {
4145 	struct fsl_mc_device *dpni_dev, *dpmac_dev;
4146 	struct dpaa2_mac *mac;
4147 	int err;
4148 
4149 	dpni_dev = to_fsl_mc_device(priv->net_dev->dev.parent);
4150 	dpmac_dev = fsl_mc_get_endpoint(dpni_dev, 0);
4151 
4152 	if (PTR_ERR(dpmac_dev) == -EPROBE_DEFER)
4153 		return PTR_ERR(dpmac_dev);
4154 
4155 	if (IS_ERR(dpmac_dev) || dpmac_dev->dev.type != &fsl_mc_bus_dpmac_type)
4156 		return 0;
4157 
4158 	mac = kzalloc(sizeof(struct dpaa2_mac), GFP_KERNEL);
4159 	if (!mac)
4160 		return -ENOMEM;
4161 
4162 	mac->mc_dev = dpmac_dev;
4163 	mac->mc_io = priv->mc_io;
4164 	mac->net_dev = priv->net_dev;
4165 
4166 	err = dpaa2_mac_open(mac);
4167 	if (err)
4168 		goto err_free_mac;
4169 	priv->mac = mac;
4170 
4171 	if (dpaa2_eth_is_type_phy(priv)) {
4172 		err = dpaa2_mac_connect(mac);
4173 		if (err && err != -EPROBE_DEFER)
4174 			netdev_err(priv->net_dev, "Error connecting to the MAC endpoint: %pe",
4175 				   ERR_PTR(err));
4176 		if (err)
4177 			goto err_close_mac;
4178 	}
4179 
4180 	return 0;
4181 
4182 err_close_mac:
4183 	dpaa2_mac_close(mac);
4184 	priv->mac = NULL;
4185 err_free_mac:
4186 	kfree(mac);
4187 	return err;
4188 }
4189 
4190 static void dpaa2_eth_disconnect_mac(struct dpaa2_eth_priv *priv)
4191 {
4192 	if (dpaa2_eth_is_type_phy(priv))
4193 		dpaa2_mac_disconnect(priv->mac);
4194 
4195 	if (!dpaa2_eth_has_mac(priv))
4196 		return;
4197 
4198 	dpaa2_mac_close(priv->mac);
4199 	kfree(priv->mac);
4200 	priv->mac = NULL;
4201 }
4202 
4203 static irqreturn_t dpni_irq0_handler_thread(int irq_num, void *arg)
4204 {
4205 	u32 status = ~0;
4206 	struct device *dev = (struct device *)arg;
4207 	struct fsl_mc_device *dpni_dev = to_fsl_mc_device(dev);
4208 	struct net_device *net_dev = dev_get_drvdata(dev);
4209 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
4210 	int err;
4211 
4212 	err = dpni_get_irq_status(dpni_dev->mc_io, 0, dpni_dev->mc_handle,
4213 				  DPNI_IRQ_INDEX, &status);
4214 	if (unlikely(err)) {
4215 		netdev_err(net_dev, "Can't get irq status (err %d)\n", err);
4216 		return IRQ_HANDLED;
4217 	}
4218 
4219 	if (status & DPNI_IRQ_EVENT_LINK_CHANGED)
4220 		dpaa2_eth_link_state_update(netdev_priv(net_dev));
4221 
4222 	if (status & DPNI_IRQ_EVENT_ENDPOINT_CHANGED) {
4223 		dpaa2_eth_set_mac_addr(netdev_priv(net_dev));
4224 		dpaa2_eth_update_tx_fqids(priv);
4225 
4226 		rtnl_lock();
4227 		if (dpaa2_eth_has_mac(priv))
4228 			dpaa2_eth_disconnect_mac(priv);
4229 		else
4230 			dpaa2_eth_connect_mac(priv);
4231 		rtnl_unlock();
4232 	}
4233 
4234 	return IRQ_HANDLED;
4235 }
4236 
4237 static int dpaa2_eth_setup_irqs(struct fsl_mc_device *ls_dev)
4238 {
4239 	int err = 0;
4240 	struct fsl_mc_device_irq *irq;
4241 
4242 	err = fsl_mc_allocate_irqs(ls_dev);
4243 	if (err) {
4244 		dev_err(&ls_dev->dev, "MC irqs allocation failed\n");
4245 		return err;
4246 	}
4247 
4248 	irq = ls_dev->irqs[0];
4249 	err = devm_request_threaded_irq(&ls_dev->dev, irq->virq,
4250 					NULL, dpni_irq0_handler_thread,
4251 					IRQF_NO_SUSPEND | IRQF_ONESHOT,
4252 					dev_name(&ls_dev->dev), &ls_dev->dev);
4253 	if (err < 0) {
4254 		dev_err(&ls_dev->dev, "devm_request_threaded_irq(): %d\n", err);
4255 		goto free_mc_irq;
4256 	}
4257 
4258 	err = dpni_set_irq_mask(ls_dev->mc_io, 0, ls_dev->mc_handle,
4259 				DPNI_IRQ_INDEX, DPNI_IRQ_EVENT_LINK_CHANGED |
4260 				DPNI_IRQ_EVENT_ENDPOINT_CHANGED);
4261 	if (err < 0) {
4262 		dev_err(&ls_dev->dev, "dpni_set_irq_mask(): %d\n", err);
4263 		goto free_irq;
4264 	}
4265 
4266 	err = dpni_set_irq_enable(ls_dev->mc_io, 0, ls_dev->mc_handle,
4267 				  DPNI_IRQ_INDEX, 1);
4268 	if (err < 0) {
4269 		dev_err(&ls_dev->dev, "dpni_set_irq_enable(): %d\n", err);
4270 		goto free_irq;
4271 	}
4272 
4273 	return 0;
4274 
4275 free_irq:
4276 	devm_free_irq(&ls_dev->dev, irq->virq, &ls_dev->dev);
4277 free_mc_irq:
4278 	fsl_mc_free_irqs(ls_dev);
4279 
4280 	return err;
4281 }
4282 
4283 static void dpaa2_eth_add_ch_napi(struct dpaa2_eth_priv *priv)
4284 {
4285 	int i;
4286 	struct dpaa2_eth_channel *ch;
4287 
4288 	for (i = 0; i < priv->num_channels; i++) {
4289 		ch = priv->channel[i];
4290 		/* NAPI weight *MUST* be a multiple of DPAA2_ETH_STORE_SIZE */
4291 		netif_napi_add(priv->net_dev, &ch->napi, dpaa2_eth_poll,
4292 			       NAPI_POLL_WEIGHT);
4293 	}
4294 }
4295 
4296 static void dpaa2_eth_del_ch_napi(struct dpaa2_eth_priv *priv)
4297 {
4298 	int i;
4299 	struct dpaa2_eth_channel *ch;
4300 
4301 	for (i = 0; i < priv->num_channels; i++) {
4302 		ch = priv->channel[i];
4303 		netif_napi_del(&ch->napi);
4304 	}
4305 }
4306 
4307 static int dpaa2_eth_probe(struct fsl_mc_device *dpni_dev)
4308 {
4309 	struct device *dev;
4310 	struct net_device *net_dev = NULL;
4311 	struct dpaa2_eth_priv *priv = NULL;
4312 	int err = 0;
4313 
4314 	dev = &dpni_dev->dev;
4315 
4316 	/* Net device */
4317 	net_dev = alloc_etherdev_mq(sizeof(*priv), DPAA2_ETH_MAX_NETDEV_QUEUES);
4318 	if (!net_dev) {
4319 		dev_err(dev, "alloc_etherdev_mq() failed\n");
4320 		return -ENOMEM;
4321 	}
4322 
4323 	SET_NETDEV_DEV(net_dev, dev);
4324 	dev_set_drvdata(dev, net_dev);
4325 
4326 	priv = netdev_priv(net_dev);
4327 	priv->net_dev = net_dev;
4328 
4329 	priv->iommu_domain = iommu_get_domain_for_dev(dev);
4330 
4331 	priv->tx_tstamp_type = HWTSTAMP_TX_OFF;
4332 	priv->rx_tstamp = false;
4333 
4334 	priv->dpaa2_ptp_wq = alloc_workqueue("dpaa2_ptp_wq", 0, 0);
4335 	if (!priv->dpaa2_ptp_wq) {
4336 		err = -ENOMEM;
4337 		goto err_wq_alloc;
4338 	}
4339 
4340 	INIT_WORK(&priv->tx_onestep_tstamp, dpaa2_eth_tx_onestep_tstamp);
4341 	mutex_init(&priv->onestep_tstamp_lock);
4342 	skb_queue_head_init(&priv->tx_skbs);
4343 
4344 	priv->rx_copybreak = DPAA2_ETH_DEFAULT_COPYBREAK;
4345 
4346 	/* Obtain a MC portal */
4347 	err = fsl_mc_portal_allocate(dpni_dev, FSL_MC_IO_ATOMIC_CONTEXT_PORTAL,
4348 				     &priv->mc_io);
4349 	if (err) {
4350 		if (err == -ENXIO)
4351 			err = -EPROBE_DEFER;
4352 		else
4353 			dev_err(dev, "MC portal allocation failed\n");
4354 		goto err_portal_alloc;
4355 	}
4356 
4357 	/* MC objects initialization and configuration */
4358 	err = dpaa2_eth_setup_dpni(dpni_dev);
4359 	if (err)
4360 		goto err_dpni_setup;
4361 
4362 	err = dpaa2_eth_setup_dpio(priv);
4363 	if (err)
4364 		goto err_dpio_setup;
4365 
4366 	dpaa2_eth_setup_fqs(priv);
4367 
4368 	err = dpaa2_eth_setup_dpbp(priv);
4369 	if (err)
4370 		goto err_dpbp_setup;
4371 
4372 	err = dpaa2_eth_bind_dpni(priv);
4373 	if (err)
4374 		goto err_bind;
4375 
4376 	/* Add a NAPI context for each channel */
4377 	dpaa2_eth_add_ch_napi(priv);
4378 
4379 	/* Percpu statistics */
4380 	priv->percpu_stats = alloc_percpu(*priv->percpu_stats);
4381 	if (!priv->percpu_stats) {
4382 		dev_err(dev, "alloc_percpu(percpu_stats) failed\n");
4383 		err = -ENOMEM;
4384 		goto err_alloc_percpu_stats;
4385 	}
4386 	priv->percpu_extras = alloc_percpu(*priv->percpu_extras);
4387 	if (!priv->percpu_extras) {
4388 		dev_err(dev, "alloc_percpu(percpu_extras) failed\n");
4389 		err = -ENOMEM;
4390 		goto err_alloc_percpu_extras;
4391 	}
4392 
4393 	priv->sgt_cache = alloc_percpu(*priv->sgt_cache);
4394 	if (!priv->sgt_cache) {
4395 		dev_err(dev, "alloc_percpu(sgt_cache) failed\n");
4396 		err = -ENOMEM;
4397 		goto err_alloc_sgt_cache;
4398 	}
4399 
4400 	err = dpaa2_eth_netdev_init(net_dev);
4401 	if (err)
4402 		goto err_netdev_init;
4403 
4404 	/* Configure checksum offload based on current interface flags */
4405 	err = dpaa2_eth_set_rx_csum(priv, !!(net_dev->features & NETIF_F_RXCSUM));
4406 	if (err)
4407 		goto err_csum;
4408 
4409 	err = dpaa2_eth_set_tx_csum(priv,
4410 				    !!(net_dev->features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)));
4411 	if (err)
4412 		goto err_csum;
4413 
4414 	err = dpaa2_eth_alloc_rings(priv);
4415 	if (err)
4416 		goto err_alloc_rings;
4417 
4418 #ifdef CONFIG_FSL_DPAA2_ETH_DCB
4419 	if (dpaa2_eth_has_pause_support(priv) && priv->vlan_cls_enabled) {
4420 		priv->dcbx_mode = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_IEEE;
4421 		net_dev->dcbnl_ops = &dpaa2_eth_dcbnl_ops;
4422 	} else {
4423 		dev_dbg(dev, "PFC not supported\n");
4424 	}
4425 #endif
4426 
4427 	err = dpaa2_eth_setup_irqs(dpni_dev);
4428 	if (err) {
4429 		netdev_warn(net_dev, "Failed to set link interrupt, fall back to polling\n");
4430 		priv->poll_thread = kthread_run(dpaa2_eth_poll_link_state, priv,
4431 						"%s_poll_link", net_dev->name);
4432 		if (IS_ERR(priv->poll_thread)) {
4433 			dev_err(dev, "Error starting polling thread\n");
4434 			goto err_poll_thread;
4435 		}
4436 		priv->do_link_poll = true;
4437 	}
4438 
4439 	err = dpaa2_eth_connect_mac(priv);
4440 	if (err)
4441 		goto err_connect_mac;
4442 
4443 	err = dpaa2_eth_dl_alloc(priv);
4444 	if (err)
4445 		goto err_dl_register;
4446 
4447 	err = dpaa2_eth_dl_traps_register(priv);
4448 	if (err)
4449 		goto err_dl_trap_register;
4450 
4451 	err = dpaa2_eth_dl_port_add(priv);
4452 	if (err)
4453 		goto err_dl_port_add;
4454 
4455 	err = register_netdev(net_dev);
4456 	if (err < 0) {
4457 		dev_err(dev, "register_netdev() failed\n");
4458 		goto err_netdev_reg;
4459 	}
4460 
4461 #ifdef CONFIG_DEBUG_FS
4462 	dpaa2_dbg_add(priv);
4463 #endif
4464 
4465 	dpaa2_eth_dl_register(priv);
4466 	dev_info(dev, "Probed interface %s\n", net_dev->name);
4467 	return 0;
4468 
4469 err_netdev_reg:
4470 	dpaa2_eth_dl_port_del(priv);
4471 err_dl_port_add:
4472 	dpaa2_eth_dl_traps_unregister(priv);
4473 err_dl_trap_register:
4474 	dpaa2_eth_dl_free(priv);
4475 err_dl_register:
4476 	dpaa2_eth_disconnect_mac(priv);
4477 err_connect_mac:
4478 	if (priv->do_link_poll)
4479 		kthread_stop(priv->poll_thread);
4480 	else
4481 		fsl_mc_free_irqs(dpni_dev);
4482 err_poll_thread:
4483 	dpaa2_eth_free_rings(priv);
4484 err_alloc_rings:
4485 err_csum:
4486 err_netdev_init:
4487 	free_percpu(priv->sgt_cache);
4488 err_alloc_sgt_cache:
4489 	free_percpu(priv->percpu_extras);
4490 err_alloc_percpu_extras:
4491 	free_percpu(priv->percpu_stats);
4492 err_alloc_percpu_stats:
4493 	dpaa2_eth_del_ch_napi(priv);
4494 err_bind:
4495 	dpaa2_eth_free_dpbp(priv);
4496 err_dpbp_setup:
4497 	dpaa2_eth_free_dpio(priv);
4498 err_dpio_setup:
4499 	dpaa2_eth_free_dpni(priv);
4500 err_dpni_setup:
4501 	fsl_mc_portal_free(priv->mc_io);
4502 err_portal_alloc:
4503 	destroy_workqueue(priv->dpaa2_ptp_wq);
4504 err_wq_alloc:
4505 	dev_set_drvdata(dev, NULL);
4506 	free_netdev(net_dev);
4507 
4508 	return err;
4509 }
4510 
4511 static int dpaa2_eth_remove(struct fsl_mc_device *ls_dev)
4512 {
4513 	struct device *dev;
4514 	struct net_device *net_dev;
4515 	struct dpaa2_eth_priv *priv;
4516 
4517 	dev = &ls_dev->dev;
4518 	net_dev = dev_get_drvdata(dev);
4519 	priv = netdev_priv(net_dev);
4520 
4521 	dpaa2_eth_dl_unregister(priv);
4522 
4523 #ifdef CONFIG_DEBUG_FS
4524 	dpaa2_dbg_remove(priv);
4525 #endif
4526 
4527 	unregister_netdev(net_dev);
4528 	rtnl_lock();
4529 	dpaa2_eth_disconnect_mac(priv);
4530 	rtnl_unlock();
4531 
4532 	dpaa2_eth_dl_port_del(priv);
4533 	dpaa2_eth_dl_traps_unregister(priv);
4534 	dpaa2_eth_dl_free(priv);
4535 
4536 	if (priv->do_link_poll)
4537 		kthread_stop(priv->poll_thread);
4538 	else
4539 		fsl_mc_free_irqs(ls_dev);
4540 
4541 	dpaa2_eth_free_rings(priv);
4542 	free_percpu(priv->sgt_cache);
4543 	free_percpu(priv->percpu_stats);
4544 	free_percpu(priv->percpu_extras);
4545 
4546 	dpaa2_eth_del_ch_napi(priv);
4547 	dpaa2_eth_free_dpbp(priv);
4548 	dpaa2_eth_free_dpio(priv);
4549 	dpaa2_eth_free_dpni(priv);
4550 
4551 	fsl_mc_portal_free(priv->mc_io);
4552 
4553 	destroy_workqueue(priv->dpaa2_ptp_wq);
4554 
4555 	dev_dbg(net_dev->dev.parent, "Removed interface %s\n", net_dev->name);
4556 
4557 	free_netdev(net_dev);
4558 
4559 	return 0;
4560 }
4561 
4562 static const struct fsl_mc_device_id dpaa2_eth_match_id_table[] = {
4563 	{
4564 		.vendor = FSL_MC_VENDOR_FREESCALE,
4565 		.obj_type = "dpni",
4566 	},
4567 	{ .vendor = 0x0 }
4568 };
4569 MODULE_DEVICE_TABLE(fslmc, dpaa2_eth_match_id_table);
4570 
4571 static struct fsl_mc_driver dpaa2_eth_driver = {
4572 	.driver = {
4573 		.name = KBUILD_MODNAME,
4574 		.owner = THIS_MODULE,
4575 	},
4576 	.probe = dpaa2_eth_probe,
4577 	.remove = dpaa2_eth_remove,
4578 	.match_id_table = dpaa2_eth_match_id_table
4579 };
4580 
4581 static int __init dpaa2_eth_driver_init(void)
4582 {
4583 	int err;
4584 
4585 	dpaa2_eth_dbg_init();
4586 	err = fsl_mc_driver_register(&dpaa2_eth_driver);
4587 	if (err) {
4588 		dpaa2_eth_dbg_exit();
4589 		return err;
4590 	}
4591 
4592 	return 0;
4593 }
4594 
4595 static void __exit dpaa2_eth_driver_exit(void)
4596 {
4597 	dpaa2_eth_dbg_exit();
4598 	fsl_mc_driver_unregister(&dpaa2_eth_driver);
4599 }
4600 
4601 module_init(dpaa2_eth_driver_init);
4602 module_exit(dpaa2_eth_driver_exit);
4603