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_xdp_release_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->xdp.drop_bufs[ch->xdp.drop_cnt++] = addr;
234 	if (ch->xdp.drop_cnt < DPAA2_ETH_BUFS_PER_CMD)
235 		return;
236 
237 	while ((err = dpaa2_io_service_release(ch->dpio, priv->bpid,
238 					       ch->xdp.drop_bufs,
239 					       ch->xdp.drop_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->xdp.drop_bufs, ch->xdp.drop_cnt);
247 		ch->buf_count -= ch->xdp.drop_cnt;
248 	}
249 
250 	ch->xdp.drop_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_xdp_release_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;
354 
355 	rcu_read_lock();
356 
357 	xdp_prog = READ_ONCE(ch->xdp.prog);
358 	if (!xdp_prog)
359 		goto out;
360 
361 	xdp.data = vaddr + dpaa2_fd_get_offset(fd);
362 	xdp.data_end = xdp.data + dpaa2_fd_get_len(fd);
363 	xdp.data_hard_start = xdp.data - XDP_PACKET_HEADROOM;
364 	xdp_set_data_meta_invalid(&xdp);
365 	xdp.rxq = &ch->xdp_rxq;
366 
367 	xdp.frame_sz = DPAA2_ETH_RX_BUF_RAW_SIZE -
368 		(dpaa2_fd_get_offset(fd) - XDP_PACKET_HEADROOM);
369 
370 	xdp_act = bpf_prog_run_xdp(xdp_prog, &xdp);
371 
372 	/* xdp.data pointer may have changed */
373 	dpaa2_fd_set_offset(fd, xdp.data - vaddr);
374 	dpaa2_fd_set_len(fd, xdp.data_end - xdp.data);
375 
376 	switch (xdp_act) {
377 	case XDP_PASS:
378 		break;
379 	case XDP_TX:
380 		dpaa2_eth_xdp_enqueue(priv, ch, fd, vaddr, rx_fq->flowid);
381 		break;
382 	default:
383 		bpf_warn_invalid_xdp_action(xdp_act);
384 		fallthrough;
385 	case XDP_ABORTED:
386 		trace_xdp_exception(priv->net_dev, xdp_prog, xdp_act);
387 		fallthrough;
388 	case XDP_DROP:
389 		dpaa2_eth_xdp_release_buf(priv, ch, addr);
390 		ch->stats.xdp_drop++;
391 		break;
392 	case XDP_REDIRECT:
393 		dma_unmap_page(priv->net_dev->dev.parent, addr,
394 			       priv->rx_buf_size, DMA_BIDIRECTIONAL);
395 		ch->buf_count--;
396 
397 		/* Allow redirect use of full headroom */
398 		xdp.data_hard_start = vaddr;
399 		xdp.frame_sz = DPAA2_ETH_RX_BUF_RAW_SIZE;
400 
401 		err = xdp_do_redirect(priv->net_dev, &xdp, xdp_prog);
402 		if (unlikely(err))
403 			ch->stats.xdp_drop++;
404 		else
405 			ch->stats.xdp_redirect++;
406 		break;
407 	}
408 
409 	ch->xdp.res |= xdp_act;
410 out:
411 	rcu_read_unlock();
412 	return xdp_act;
413 }
414 
415 /* Main Rx frame processing routine */
416 static void dpaa2_eth_rx(struct dpaa2_eth_priv *priv,
417 			 struct dpaa2_eth_channel *ch,
418 			 const struct dpaa2_fd *fd,
419 			 struct dpaa2_eth_fq *fq)
420 {
421 	dma_addr_t addr = dpaa2_fd_get_addr(fd);
422 	u8 fd_format = dpaa2_fd_get_format(fd);
423 	void *vaddr;
424 	struct sk_buff *skb;
425 	struct rtnl_link_stats64 *percpu_stats;
426 	struct dpaa2_eth_drv_stats *percpu_extras;
427 	struct device *dev = priv->net_dev->dev.parent;
428 	struct dpaa2_fas *fas;
429 	void *buf_data;
430 	u32 status = 0;
431 	u32 xdp_act;
432 
433 	/* Tracing point */
434 	trace_dpaa2_rx_fd(priv->net_dev, fd);
435 
436 	vaddr = dpaa2_iova_to_virt(priv->iommu_domain, addr);
437 	dma_sync_single_for_cpu(dev, addr, priv->rx_buf_size,
438 				DMA_BIDIRECTIONAL);
439 
440 	fas = dpaa2_get_fas(vaddr, false);
441 	prefetch(fas);
442 	buf_data = vaddr + dpaa2_fd_get_offset(fd);
443 	prefetch(buf_data);
444 
445 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
446 	percpu_extras = this_cpu_ptr(priv->percpu_extras);
447 
448 	if (fd_format == dpaa2_fd_single) {
449 		xdp_act = dpaa2_eth_run_xdp(priv, ch, fq, (struct dpaa2_fd *)fd, vaddr);
450 		if (xdp_act != XDP_PASS) {
451 			percpu_stats->rx_packets++;
452 			percpu_stats->rx_bytes += dpaa2_fd_get_len(fd);
453 			return;
454 		}
455 
456 		dma_unmap_page(dev, addr, priv->rx_buf_size,
457 			       DMA_BIDIRECTIONAL);
458 		skb = dpaa2_eth_build_linear_skb(ch, fd, vaddr);
459 	} else if (fd_format == dpaa2_fd_sg) {
460 		WARN_ON(priv->xdp_prog);
461 
462 		dma_unmap_page(dev, addr, priv->rx_buf_size,
463 			       DMA_BIDIRECTIONAL);
464 		skb = dpaa2_eth_build_frag_skb(priv, ch, buf_data);
465 		free_pages((unsigned long)vaddr, 0);
466 		percpu_extras->rx_sg_frames++;
467 		percpu_extras->rx_sg_bytes += dpaa2_fd_get_len(fd);
468 	} else {
469 		/* We don't support any other format */
470 		goto err_frame_format;
471 	}
472 
473 	if (unlikely(!skb))
474 		goto err_build_skb;
475 
476 	prefetch(skb->data);
477 
478 	/* Get the timestamp value */
479 	if (priv->rx_tstamp) {
480 		struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
481 		__le64 *ts = dpaa2_get_ts(vaddr, false);
482 		u64 ns;
483 
484 		memset(shhwtstamps, 0, sizeof(*shhwtstamps));
485 
486 		ns = DPAA2_PTP_CLK_PERIOD_NS * le64_to_cpup(ts);
487 		shhwtstamps->hwtstamp = ns_to_ktime(ns);
488 	}
489 
490 	/* Check if we need to validate the L4 csum */
491 	if (likely(dpaa2_fd_get_frc(fd) & DPAA2_FD_FRC_FASV)) {
492 		status = le32_to_cpu(fas->status);
493 		dpaa2_eth_validate_rx_csum(priv, status, skb);
494 	}
495 
496 	skb->protocol = eth_type_trans(skb, priv->net_dev);
497 	skb_record_rx_queue(skb, fq->flowid);
498 
499 	percpu_stats->rx_packets++;
500 	percpu_stats->rx_bytes += dpaa2_fd_get_len(fd);
501 
502 	list_add_tail(&skb->list, ch->rx_list);
503 
504 	return;
505 
506 err_build_skb:
507 	dpaa2_eth_free_rx_fd(priv, fd, vaddr);
508 err_frame_format:
509 	percpu_stats->rx_dropped++;
510 }
511 
512 /* Processing of Rx frames received on the error FQ
513  * We check and print the error bits and then free the frame
514  */
515 static void dpaa2_eth_rx_err(struct dpaa2_eth_priv *priv,
516 			     struct dpaa2_eth_channel *ch,
517 			     const struct dpaa2_fd *fd,
518 			     struct dpaa2_eth_fq *fq __always_unused)
519 {
520 	struct device *dev = priv->net_dev->dev.parent;
521 	dma_addr_t addr = dpaa2_fd_get_addr(fd);
522 	u8 fd_format = dpaa2_fd_get_format(fd);
523 	struct rtnl_link_stats64 *percpu_stats;
524 	struct dpaa2_eth_trap_item *trap_item;
525 	struct dpaa2_fapr *fapr;
526 	struct sk_buff *skb;
527 	void *buf_data;
528 	void *vaddr;
529 
530 	vaddr = dpaa2_iova_to_virt(priv->iommu_domain, addr);
531 	dma_sync_single_for_cpu(dev, addr, priv->rx_buf_size,
532 				DMA_BIDIRECTIONAL);
533 
534 	buf_data = vaddr + dpaa2_fd_get_offset(fd);
535 
536 	if (fd_format == dpaa2_fd_single) {
537 		dma_unmap_page(dev, addr, priv->rx_buf_size,
538 			       DMA_BIDIRECTIONAL);
539 		skb = dpaa2_eth_build_linear_skb(ch, fd, vaddr);
540 	} else if (fd_format == dpaa2_fd_sg) {
541 		dma_unmap_page(dev, addr, priv->rx_buf_size,
542 			       DMA_BIDIRECTIONAL);
543 		skb = dpaa2_eth_build_frag_skb(priv, ch, buf_data);
544 		free_pages((unsigned long)vaddr, 0);
545 	} else {
546 		/* We don't support any other format */
547 		dpaa2_eth_free_rx_fd(priv, fd, vaddr);
548 		goto err_frame_format;
549 	}
550 
551 	fapr = dpaa2_get_fapr(vaddr, false);
552 	trap_item = dpaa2_eth_dl_get_trap(priv, fapr);
553 	if (trap_item)
554 		devlink_trap_report(priv->devlink, skb, trap_item->trap_ctx,
555 				    &priv->devlink_port, NULL);
556 	consume_skb(skb);
557 
558 err_frame_format:
559 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
560 	percpu_stats->rx_errors++;
561 	ch->buf_count--;
562 }
563 
564 /* Consume all frames pull-dequeued into the store. This is the simplest way to
565  * make sure we don't accidentally issue another volatile dequeue which would
566  * overwrite (leak) frames already in the store.
567  *
568  * Observance of NAPI budget is not our concern, leaving that to the caller.
569  */
570 static int dpaa2_eth_consume_frames(struct dpaa2_eth_channel *ch,
571 				    struct dpaa2_eth_fq **src)
572 {
573 	struct dpaa2_eth_priv *priv = ch->priv;
574 	struct dpaa2_eth_fq *fq = NULL;
575 	struct dpaa2_dq *dq;
576 	const struct dpaa2_fd *fd;
577 	int cleaned = 0, retries = 0;
578 	int is_last;
579 
580 	do {
581 		dq = dpaa2_io_store_next(ch->store, &is_last);
582 		if (unlikely(!dq)) {
583 			/* If we're here, we *must* have placed a
584 			 * volatile dequeue comnmand, so keep reading through
585 			 * the store until we get some sort of valid response
586 			 * token (either a valid frame or an "empty dequeue")
587 			 */
588 			if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES) {
589 				netdev_err_once(priv->net_dev,
590 						"Unable to read a valid dequeue response\n");
591 				return -ETIMEDOUT;
592 			}
593 			continue;
594 		}
595 
596 		fd = dpaa2_dq_fd(dq);
597 		fq = (struct dpaa2_eth_fq *)(uintptr_t)dpaa2_dq_fqd_ctx(dq);
598 
599 		fq->consume(priv, ch, fd, fq);
600 		cleaned++;
601 		retries = 0;
602 	} while (!is_last);
603 
604 	if (!cleaned)
605 		return 0;
606 
607 	fq->stats.frames += cleaned;
608 	ch->stats.frames += cleaned;
609 
610 	/* A dequeue operation only pulls frames from a single queue
611 	 * into the store. Return the frame queue as an out param.
612 	 */
613 	if (src)
614 		*src = fq;
615 
616 	return cleaned;
617 }
618 
619 static int dpaa2_eth_ptp_parse(struct sk_buff *skb,
620 			       u8 *msgtype, u8 *twostep, u8 *udp,
621 			       u16 *correction_offset,
622 			       u16 *origintimestamp_offset)
623 {
624 	unsigned int ptp_class;
625 	struct ptp_header *hdr;
626 	unsigned int type;
627 	u8 *base;
628 
629 	ptp_class = ptp_classify_raw(skb);
630 	if (ptp_class == PTP_CLASS_NONE)
631 		return -EINVAL;
632 
633 	hdr = ptp_parse_header(skb, ptp_class);
634 	if (!hdr)
635 		return -EINVAL;
636 
637 	*msgtype = ptp_get_msgtype(hdr, ptp_class);
638 	*twostep = hdr->flag_field[0] & 0x2;
639 
640 	type = ptp_class & PTP_CLASS_PMASK;
641 	if (type == PTP_CLASS_IPV4 ||
642 	    type == PTP_CLASS_IPV6)
643 		*udp = 1;
644 	else
645 		*udp = 0;
646 
647 	base = skb_mac_header(skb);
648 	*correction_offset = (u8 *)&hdr->correction - base;
649 	*origintimestamp_offset = (u8 *)hdr + sizeof(struct ptp_header) - base;
650 
651 	return 0;
652 }
653 
654 /* Configure the egress frame annotation for timestamp update */
655 static void dpaa2_eth_enable_tx_tstamp(struct dpaa2_eth_priv *priv,
656 				       struct dpaa2_fd *fd,
657 				       void *buf_start,
658 				       struct sk_buff *skb)
659 {
660 	struct ptp_tstamp origin_timestamp;
661 	struct dpni_single_step_cfg cfg;
662 	u8 msgtype, twostep, udp;
663 	struct dpaa2_faead *faead;
664 	struct dpaa2_fas *fas;
665 	struct timespec64 ts;
666 	u16 offset1, offset2;
667 	u32 ctrl, frc;
668 	__le64 *ns;
669 	u8 *data;
670 
671 	/* Mark the egress frame annotation area as valid */
672 	frc = dpaa2_fd_get_frc(fd);
673 	dpaa2_fd_set_frc(fd, frc | DPAA2_FD_FRC_FAEADV);
674 
675 	/* Set hardware annotation size */
676 	ctrl = dpaa2_fd_get_ctrl(fd);
677 	dpaa2_fd_set_ctrl(fd, ctrl | DPAA2_FD_CTRL_ASAL);
678 
679 	/* enable UPD (update prepanded data) bit in FAEAD field of
680 	 * hardware frame annotation area
681 	 */
682 	ctrl = DPAA2_FAEAD_A2V | DPAA2_FAEAD_UPDV | DPAA2_FAEAD_UPD;
683 	faead = dpaa2_get_faead(buf_start, true);
684 	faead->ctrl = cpu_to_le32(ctrl);
685 
686 	if (skb->cb[0] == TX_TSTAMP_ONESTEP_SYNC) {
687 		if (dpaa2_eth_ptp_parse(skb, &msgtype, &twostep, &udp,
688 					&offset1, &offset2) ||
689 		    msgtype != PTP_MSGTYPE_SYNC || twostep) {
690 			WARN_ONCE(1, "Bad packet for one-step timestamping\n");
691 			return;
692 		}
693 
694 		/* Mark the frame annotation status as valid */
695 		frc = dpaa2_fd_get_frc(fd);
696 		dpaa2_fd_set_frc(fd, frc | DPAA2_FD_FRC_FASV);
697 
698 		/* Mark the PTP flag for one step timestamping */
699 		fas = dpaa2_get_fas(buf_start, true);
700 		fas->status = cpu_to_le32(DPAA2_FAS_PTP);
701 
702 		dpaa2_ptp->caps.gettime64(&dpaa2_ptp->caps, &ts);
703 		ns = dpaa2_get_ts(buf_start, true);
704 		*ns = cpu_to_le64(timespec64_to_ns(&ts) /
705 				  DPAA2_PTP_CLK_PERIOD_NS);
706 
707 		/* Update current time to PTP message originTimestamp field */
708 		ns_to_ptp_tstamp(&origin_timestamp, le64_to_cpup(ns));
709 		data = skb_mac_header(skb);
710 		*(__be16 *)(data + offset2) = htons(origin_timestamp.sec_msb);
711 		*(__be32 *)(data + offset2 + 2) =
712 			htonl(origin_timestamp.sec_lsb);
713 		*(__be32 *)(data + offset2 + 6) = htonl(origin_timestamp.nsec);
714 
715 		cfg.en = 1;
716 		cfg.ch_update = udp;
717 		cfg.offset = offset1;
718 		cfg.peer_delay = 0;
719 
720 		if (dpni_set_single_step_cfg(priv->mc_io, 0, priv->mc_token,
721 					     &cfg))
722 			WARN_ONCE(1, "Failed to set single step register");
723 	}
724 }
725 
726 /* Create a frame descriptor based on a fragmented skb */
727 static int dpaa2_eth_build_sg_fd(struct dpaa2_eth_priv *priv,
728 				 struct sk_buff *skb,
729 				 struct dpaa2_fd *fd,
730 				 void **swa_addr)
731 {
732 	struct device *dev = priv->net_dev->dev.parent;
733 	void *sgt_buf = NULL;
734 	dma_addr_t addr;
735 	int nr_frags = skb_shinfo(skb)->nr_frags;
736 	struct dpaa2_sg_entry *sgt;
737 	int i, err;
738 	int sgt_buf_size;
739 	struct scatterlist *scl, *crt_scl;
740 	int num_sg;
741 	int num_dma_bufs;
742 	struct dpaa2_eth_swa *swa;
743 
744 	/* Create and map scatterlist.
745 	 * We don't advertise NETIF_F_FRAGLIST, so skb_to_sgvec() will not have
746 	 * to go beyond nr_frags+1.
747 	 * Note: We don't support chained scatterlists
748 	 */
749 	if (unlikely(PAGE_SIZE / sizeof(struct scatterlist) < nr_frags + 1))
750 		return -EINVAL;
751 
752 	scl = kmalloc_array(nr_frags + 1, sizeof(struct scatterlist), GFP_ATOMIC);
753 	if (unlikely(!scl))
754 		return -ENOMEM;
755 
756 	sg_init_table(scl, nr_frags + 1);
757 	num_sg = skb_to_sgvec(skb, scl, 0, skb->len);
758 	if (unlikely(num_sg < 0)) {
759 		err = -ENOMEM;
760 		goto dma_map_sg_failed;
761 	}
762 	num_dma_bufs = dma_map_sg(dev, scl, num_sg, DMA_BIDIRECTIONAL);
763 	if (unlikely(!num_dma_bufs)) {
764 		err = -ENOMEM;
765 		goto dma_map_sg_failed;
766 	}
767 
768 	/* Prepare the HW SGT structure */
769 	sgt_buf_size = priv->tx_data_offset +
770 		       sizeof(struct dpaa2_sg_entry) *  num_dma_bufs;
771 	sgt_buf = napi_alloc_frag(sgt_buf_size + DPAA2_ETH_TX_BUF_ALIGN);
772 	if (unlikely(!sgt_buf)) {
773 		err = -ENOMEM;
774 		goto sgt_buf_alloc_failed;
775 	}
776 	sgt_buf = PTR_ALIGN(sgt_buf, DPAA2_ETH_TX_BUF_ALIGN);
777 	memset(sgt_buf, 0, sgt_buf_size);
778 
779 	sgt = (struct dpaa2_sg_entry *)(sgt_buf + priv->tx_data_offset);
780 
781 	/* Fill in the HW SGT structure.
782 	 *
783 	 * sgt_buf is zeroed out, so the following fields are implicit
784 	 * in all sgt entries:
785 	 *   - offset is 0
786 	 *   - format is 'dpaa2_sg_single'
787 	 */
788 	for_each_sg(scl, crt_scl, num_dma_bufs, i) {
789 		dpaa2_sg_set_addr(&sgt[i], sg_dma_address(crt_scl));
790 		dpaa2_sg_set_len(&sgt[i], sg_dma_len(crt_scl));
791 	}
792 	dpaa2_sg_set_final(&sgt[i - 1], true);
793 
794 	/* Store the skb backpointer in the SGT buffer.
795 	 * Fit the scatterlist and the number of buffers alongside the
796 	 * skb backpointer in the software annotation area. We'll need
797 	 * all of them on Tx Conf.
798 	 */
799 	*swa_addr = (void *)sgt_buf;
800 	swa = (struct dpaa2_eth_swa *)sgt_buf;
801 	swa->type = DPAA2_ETH_SWA_SG;
802 	swa->sg.skb = skb;
803 	swa->sg.scl = scl;
804 	swa->sg.num_sg = num_sg;
805 	swa->sg.sgt_size = sgt_buf_size;
806 
807 	/* Separately map the SGT buffer */
808 	addr = dma_map_single(dev, sgt_buf, sgt_buf_size, DMA_BIDIRECTIONAL);
809 	if (unlikely(dma_mapping_error(dev, addr))) {
810 		err = -ENOMEM;
811 		goto dma_map_single_failed;
812 	}
813 	dpaa2_fd_set_offset(fd, priv->tx_data_offset);
814 	dpaa2_fd_set_format(fd, dpaa2_fd_sg);
815 	dpaa2_fd_set_addr(fd, addr);
816 	dpaa2_fd_set_len(fd, skb->len);
817 	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
818 
819 	return 0;
820 
821 dma_map_single_failed:
822 	skb_free_frag(sgt_buf);
823 sgt_buf_alloc_failed:
824 	dma_unmap_sg(dev, scl, num_sg, DMA_BIDIRECTIONAL);
825 dma_map_sg_failed:
826 	kfree(scl);
827 	return err;
828 }
829 
830 /* Create a SG frame descriptor based on a linear skb.
831  *
832  * This function is used on the Tx path when the skb headroom is not large
833  * enough for the HW requirements, thus instead of realloc-ing the skb we
834  * create a SG frame descriptor with only one entry.
835  */
836 static int dpaa2_eth_build_sg_fd_single_buf(struct dpaa2_eth_priv *priv,
837 					    struct sk_buff *skb,
838 					    struct dpaa2_fd *fd,
839 					    void **swa_addr)
840 {
841 	struct device *dev = priv->net_dev->dev.parent;
842 	struct dpaa2_eth_sgt_cache *sgt_cache;
843 	struct dpaa2_sg_entry *sgt;
844 	struct dpaa2_eth_swa *swa;
845 	dma_addr_t addr, sgt_addr;
846 	void *sgt_buf = NULL;
847 	int sgt_buf_size;
848 	int err;
849 
850 	/* Prepare the HW SGT structure */
851 	sgt_cache = this_cpu_ptr(priv->sgt_cache);
852 	sgt_buf_size = priv->tx_data_offset + sizeof(struct dpaa2_sg_entry);
853 
854 	if (sgt_cache->count == 0)
855 		sgt_buf = kzalloc(sgt_buf_size + DPAA2_ETH_TX_BUF_ALIGN,
856 				  GFP_ATOMIC);
857 	else
858 		sgt_buf = sgt_cache->buf[--sgt_cache->count];
859 	if (unlikely(!sgt_buf))
860 		return -ENOMEM;
861 
862 	sgt_buf = PTR_ALIGN(sgt_buf, DPAA2_ETH_TX_BUF_ALIGN);
863 	sgt = (struct dpaa2_sg_entry *)(sgt_buf + priv->tx_data_offset);
864 
865 	addr = dma_map_single(dev, skb->data, skb->len, DMA_BIDIRECTIONAL);
866 	if (unlikely(dma_mapping_error(dev, addr))) {
867 		err = -ENOMEM;
868 		goto data_map_failed;
869 	}
870 
871 	/* Fill in the HW SGT structure */
872 	dpaa2_sg_set_addr(sgt, addr);
873 	dpaa2_sg_set_len(sgt, skb->len);
874 	dpaa2_sg_set_final(sgt, true);
875 
876 	/* Store the skb backpointer in the SGT buffer */
877 	*swa_addr = (void *)sgt_buf;
878 	swa = (struct dpaa2_eth_swa *)sgt_buf;
879 	swa->type = DPAA2_ETH_SWA_SINGLE;
880 	swa->single.skb = skb;
881 	swa->single.sgt_size = sgt_buf_size;
882 
883 	/* Separately map the SGT buffer */
884 	sgt_addr = dma_map_single(dev, sgt_buf, sgt_buf_size, DMA_BIDIRECTIONAL);
885 	if (unlikely(dma_mapping_error(dev, sgt_addr))) {
886 		err = -ENOMEM;
887 		goto sgt_map_failed;
888 	}
889 
890 	dpaa2_fd_set_offset(fd, priv->tx_data_offset);
891 	dpaa2_fd_set_format(fd, dpaa2_fd_sg);
892 	dpaa2_fd_set_addr(fd, sgt_addr);
893 	dpaa2_fd_set_len(fd, skb->len);
894 	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
895 
896 	return 0;
897 
898 sgt_map_failed:
899 	dma_unmap_single(dev, addr, skb->len, DMA_BIDIRECTIONAL);
900 data_map_failed:
901 	if (sgt_cache->count >= DPAA2_ETH_SGT_CACHE_SIZE)
902 		kfree(sgt_buf);
903 	else
904 		sgt_cache->buf[sgt_cache->count++] = sgt_buf;
905 
906 	return err;
907 }
908 
909 /* Create a frame descriptor based on a linear skb */
910 static int dpaa2_eth_build_single_fd(struct dpaa2_eth_priv *priv,
911 				     struct sk_buff *skb,
912 				     struct dpaa2_fd *fd,
913 				     void **swa_addr)
914 {
915 	struct device *dev = priv->net_dev->dev.parent;
916 	u8 *buffer_start, *aligned_start;
917 	struct dpaa2_eth_swa *swa;
918 	dma_addr_t addr;
919 
920 	buffer_start = skb->data - dpaa2_eth_needed_headroom(skb);
921 
922 	/* If there's enough room to align the FD address, do it.
923 	 * It will help hardware optimize accesses.
924 	 */
925 	aligned_start = PTR_ALIGN(buffer_start - DPAA2_ETH_TX_BUF_ALIGN,
926 				  DPAA2_ETH_TX_BUF_ALIGN);
927 	if (aligned_start >= skb->head)
928 		buffer_start = aligned_start;
929 
930 	/* Store a backpointer to the skb at the beginning of the buffer
931 	 * (in the private data area) such that we can release it
932 	 * on Tx confirm
933 	 */
934 	*swa_addr = (void *)buffer_start;
935 	swa = (struct dpaa2_eth_swa *)buffer_start;
936 	swa->type = DPAA2_ETH_SWA_SINGLE;
937 	swa->single.skb = skb;
938 
939 	addr = dma_map_single(dev, buffer_start,
940 			      skb_tail_pointer(skb) - buffer_start,
941 			      DMA_BIDIRECTIONAL);
942 	if (unlikely(dma_mapping_error(dev, addr)))
943 		return -ENOMEM;
944 
945 	dpaa2_fd_set_addr(fd, addr);
946 	dpaa2_fd_set_offset(fd, (u16)(skb->data - buffer_start));
947 	dpaa2_fd_set_len(fd, skb->len);
948 	dpaa2_fd_set_format(fd, dpaa2_fd_single);
949 	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
950 
951 	return 0;
952 }
953 
954 /* FD freeing routine on the Tx path
955  *
956  * DMA-unmap and free FD and possibly SGT buffer allocated on Tx. The skb
957  * back-pointed to is also freed.
958  * This can be called either from dpaa2_eth_tx_conf() or on the error path of
959  * dpaa2_eth_tx().
960  */
961 static void dpaa2_eth_free_tx_fd(struct dpaa2_eth_priv *priv,
962 				 struct dpaa2_eth_fq *fq,
963 				 const struct dpaa2_fd *fd, bool in_napi)
964 {
965 	struct device *dev = priv->net_dev->dev.parent;
966 	dma_addr_t fd_addr, sg_addr;
967 	struct sk_buff *skb = NULL;
968 	unsigned char *buffer_start;
969 	struct dpaa2_eth_swa *swa;
970 	u8 fd_format = dpaa2_fd_get_format(fd);
971 	u32 fd_len = dpaa2_fd_get_len(fd);
972 
973 	struct dpaa2_eth_sgt_cache *sgt_cache;
974 	struct dpaa2_sg_entry *sgt;
975 
976 	fd_addr = dpaa2_fd_get_addr(fd);
977 	buffer_start = dpaa2_iova_to_virt(priv->iommu_domain, fd_addr);
978 	swa = (struct dpaa2_eth_swa *)buffer_start;
979 
980 	if (fd_format == dpaa2_fd_single) {
981 		if (swa->type == DPAA2_ETH_SWA_SINGLE) {
982 			skb = swa->single.skb;
983 			/* Accessing the skb buffer is safe before dma unmap,
984 			 * because we didn't map the actual skb shell.
985 			 */
986 			dma_unmap_single(dev, fd_addr,
987 					 skb_tail_pointer(skb) - buffer_start,
988 					 DMA_BIDIRECTIONAL);
989 		} else {
990 			WARN_ONCE(swa->type != DPAA2_ETH_SWA_XDP, "Wrong SWA type");
991 			dma_unmap_single(dev, fd_addr, swa->xdp.dma_size,
992 					 DMA_BIDIRECTIONAL);
993 		}
994 	} else if (fd_format == dpaa2_fd_sg) {
995 		if (swa->type == DPAA2_ETH_SWA_SG) {
996 			skb = swa->sg.skb;
997 
998 			/* Unmap the scatterlist */
999 			dma_unmap_sg(dev, swa->sg.scl, swa->sg.num_sg,
1000 				     DMA_BIDIRECTIONAL);
1001 			kfree(swa->sg.scl);
1002 
1003 			/* Unmap the SGT buffer */
1004 			dma_unmap_single(dev, fd_addr, swa->sg.sgt_size,
1005 					 DMA_BIDIRECTIONAL);
1006 		} else {
1007 			skb = swa->single.skb;
1008 
1009 			/* Unmap the SGT Buffer */
1010 			dma_unmap_single(dev, fd_addr, swa->single.sgt_size,
1011 					 DMA_BIDIRECTIONAL);
1012 
1013 			sgt = (struct dpaa2_sg_entry *)(buffer_start +
1014 							priv->tx_data_offset);
1015 			sg_addr = dpaa2_sg_get_addr(sgt);
1016 			dma_unmap_single(dev, sg_addr, skb->len, DMA_BIDIRECTIONAL);
1017 		}
1018 	} else {
1019 		netdev_dbg(priv->net_dev, "Invalid FD format\n");
1020 		return;
1021 	}
1022 
1023 	if (swa->type != DPAA2_ETH_SWA_XDP && in_napi) {
1024 		fq->dq_frames++;
1025 		fq->dq_bytes += fd_len;
1026 	}
1027 
1028 	if (swa->type == DPAA2_ETH_SWA_XDP) {
1029 		xdp_return_frame(swa->xdp.xdpf);
1030 		return;
1031 	}
1032 
1033 	/* Get the timestamp value */
1034 	if (skb->cb[0] == TX_TSTAMP) {
1035 		struct skb_shared_hwtstamps shhwtstamps;
1036 		__le64 *ts = dpaa2_get_ts(buffer_start, true);
1037 		u64 ns;
1038 
1039 		memset(&shhwtstamps, 0, sizeof(shhwtstamps));
1040 
1041 		ns = DPAA2_PTP_CLK_PERIOD_NS * le64_to_cpup(ts);
1042 		shhwtstamps.hwtstamp = ns_to_ktime(ns);
1043 		skb_tstamp_tx(skb, &shhwtstamps);
1044 	} else if (skb->cb[0] == TX_TSTAMP_ONESTEP_SYNC) {
1045 		mutex_unlock(&priv->onestep_tstamp_lock);
1046 	}
1047 
1048 	/* Free SGT buffer allocated on tx */
1049 	if (fd_format != dpaa2_fd_single) {
1050 		sgt_cache = this_cpu_ptr(priv->sgt_cache);
1051 		if (swa->type == DPAA2_ETH_SWA_SG) {
1052 			skb_free_frag(buffer_start);
1053 		} else {
1054 			if (sgt_cache->count >= DPAA2_ETH_SGT_CACHE_SIZE)
1055 				kfree(buffer_start);
1056 			else
1057 				sgt_cache->buf[sgt_cache->count++] = buffer_start;
1058 		}
1059 	}
1060 
1061 	/* Move on with skb release */
1062 	napi_consume_skb(skb, in_napi);
1063 }
1064 
1065 static netdev_tx_t __dpaa2_eth_tx(struct sk_buff *skb,
1066 				  struct net_device *net_dev)
1067 {
1068 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1069 	struct dpaa2_fd fd;
1070 	struct rtnl_link_stats64 *percpu_stats;
1071 	struct dpaa2_eth_drv_stats *percpu_extras;
1072 	struct dpaa2_eth_fq *fq;
1073 	struct netdev_queue *nq;
1074 	u16 queue_mapping;
1075 	unsigned int needed_headroom;
1076 	u32 fd_len;
1077 	u8 prio = 0;
1078 	int err, i;
1079 	void *swa;
1080 
1081 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
1082 	percpu_extras = this_cpu_ptr(priv->percpu_extras);
1083 
1084 	needed_headroom = dpaa2_eth_needed_headroom(skb);
1085 
1086 	/* We'll be holding a back-reference to the skb until Tx Confirmation;
1087 	 * we don't want that overwritten by a concurrent Tx with a cloned skb.
1088 	 */
1089 	skb = skb_unshare(skb, GFP_ATOMIC);
1090 	if (unlikely(!skb)) {
1091 		/* skb_unshare() has already freed the skb */
1092 		percpu_stats->tx_dropped++;
1093 		return NETDEV_TX_OK;
1094 	}
1095 
1096 	/* Setup the FD fields */
1097 	memset(&fd, 0, sizeof(fd));
1098 
1099 	if (skb_is_nonlinear(skb)) {
1100 		err = dpaa2_eth_build_sg_fd(priv, skb, &fd, &swa);
1101 		percpu_extras->tx_sg_frames++;
1102 		percpu_extras->tx_sg_bytes += skb->len;
1103 	} else if (skb_headroom(skb) < needed_headroom) {
1104 		err = dpaa2_eth_build_sg_fd_single_buf(priv, skb, &fd, &swa);
1105 		percpu_extras->tx_sg_frames++;
1106 		percpu_extras->tx_sg_bytes += skb->len;
1107 		percpu_extras->tx_converted_sg_frames++;
1108 		percpu_extras->tx_converted_sg_bytes += skb->len;
1109 	} else {
1110 		err = dpaa2_eth_build_single_fd(priv, skb, &fd, &swa);
1111 	}
1112 
1113 	if (unlikely(err)) {
1114 		percpu_stats->tx_dropped++;
1115 		goto err_build_fd;
1116 	}
1117 
1118 	if (skb->cb[0])
1119 		dpaa2_eth_enable_tx_tstamp(priv, &fd, swa, skb);
1120 
1121 	/* Tracing point */
1122 	trace_dpaa2_tx_fd(net_dev, &fd);
1123 
1124 	/* TxConf FQ selection relies on queue id from the stack.
1125 	 * In case of a forwarded frame from another DPNI interface, we choose
1126 	 * a queue affined to the same core that processed the Rx frame
1127 	 */
1128 	queue_mapping = skb_get_queue_mapping(skb);
1129 
1130 	if (net_dev->num_tc) {
1131 		prio = netdev_txq_to_tc(net_dev, queue_mapping);
1132 		/* Hardware interprets priority level 0 as being the highest,
1133 		 * so we need to do a reverse mapping to the netdev tc index
1134 		 */
1135 		prio = net_dev->num_tc - prio - 1;
1136 		/* We have only one FQ array entry for all Tx hardware queues
1137 		 * with the same flow id (but different priority levels)
1138 		 */
1139 		queue_mapping %= dpaa2_eth_queue_count(priv);
1140 	}
1141 	fq = &priv->fq[queue_mapping];
1142 
1143 	fd_len = dpaa2_fd_get_len(&fd);
1144 	nq = netdev_get_tx_queue(net_dev, queue_mapping);
1145 	netdev_tx_sent_queue(nq, fd_len);
1146 
1147 	/* Everything that happens after this enqueues might race with
1148 	 * the Tx confirmation callback for this frame
1149 	 */
1150 	for (i = 0; i < DPAA2_ETH_ENQUEUE_RETRIES; i++) {
1151 		err = priv->enqueue(priv, fq, &fd, prio, 1, NULL);
1152 		if (err != -EBUSY)
1153 			break;
1154 	}
1155 	percpu_extras->tx_portal_busy += i;
1156 	if (unlikely(err < 0)) {
1157 		percpu_stats->tx_errors++;
1158 		/* Clean up everything, including freeing the skb */
1159 		dpaa2_eth_free_tx_fd(priv, fq, &fd, false);
1160 		netdev_tx_completed_queue(nq, 1, fd_len);
1161 	} else {
1162 		percpu_stats->tx_packets++;
1163 		percpu_stats->tx_bytes += fd_len;
1164 	}
1165 
1166 	return NETDEV_TX_OK;
1167 
1168 err_build_fd:
1169 	dev_kfree_skb(skb);
1170 
1171 	return NETDEV_TX_OK;
1172 }
1173 
1174 static void dpaa2_eth_tx_onestep_tstamp(struct work_struct *work)
1175 {
1176 	struct dpaa2_eth_priv *priv = container_of(work, struct dpaa2_eth_priv,
1177 						   tx_onestep_tstamp);
1178 	struct sk_buff *skb;
1179 
1180 	while (true) {
1181 		skb = skb_dequeue(&priv->tx_skbs);
1182 		if (!skb)
1183 			return;
1184 
1185 		/* Lock just before TX one-step timestamping packet,
1186 		 * and release the lock in dpaa2_eth_free_tx_fd when
1187 		 * confirm the packet has been sent on hardware, or
1188 		 * when clean up during transmit failure.
1189 		 */
1190 		mutex_lock(&priv->onestep_tstamp_lock);
1191 		__dpaa2_eth_tx(skb, priv->net_dev);
1192 	}
1193 }
1194 
1195 static netdev_tx_t dpaa2_eth_tx(struct sk_buff *skb, struct net_device *net_dev)
1196 {
1197 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1198 	u8 msgtype, twostep, udp;
1199 	u16 offset1, offset2;
1200 
1201 	/* Utilize skb->cb[0] for timestamping request per skb */
1202 	skb->cb[0] = 0;
1203 
1204 	if ((skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) && dpaa2_ptp) {
1205 		if (priv->tx_tstamp_type == HWTSTAMP_TX_ON)
1206 			skb->cb[0] = TX_TSTAMP;
1207 		else if (priv->tx_tstamp_type == HWTSTAMP_TX_ONESTEP_SYNC)
1208 			skb->cb[0] = TX_TSTAMP_ONESTEP_SYNC;
1209 	}
1210 
1211 	/* TX for one-step timestamping PTP Sync packet */
1212 	if (skb->cb[0] == TX_TSTAMP_ONESTEP_SYNC) {
1213 		if (!dpaa2_eth_ptp_parse(skb, &msgtype, &twostep, &udp,
1214 					 &offset1, &offset2))
1215 			if (msgtype == PTP_MSGTYPE_SYNC && twostep == 0) {
1216 				skb_queue_tail(&priv->tx_skbs, skb);
1217 				queue_work(priv->dpaa2_ptp_wq,
1218 					   &priv->tx_onestep_tstamp);
1219 				return NETDEV_TX_OK;
1220 			}
1221 		/* Use two-step timestamping if not one-step timestamping
1222 		 * PTP Sync packet
1223 		 */
1224 		skb->cb[0] = TX_TSTAMP;
1225 	}
1226 
1227 	/* TX for other packets */
1228 	return __dpaa2_eth_tx(skb, net_dev);
1229 }
1230 
1231 /* Tx confirmation frame processing routine */
1232 static void dpaa2_eth_tx_conf(struct dpaa2_eth_priv *priv,
1233 			      struct dpaa2_eth_channel *ch __always_unused,
1234 			      const struct dpaa2_fd *fd,
1235 			      struct dpaa2_eth_fq *fq)
1236 {
1237 	struct rtnl_link_stats64 *percpu_stats;
1238 	struct dpaa2_eth_drv_stats *percpu_extras;
1239 	u32 fd_len = dpaa2_fd_get_len(fd);
1240 	u32 fd_errors;
1241 
1242 	/* Tracing point */
1243 	trace_dpaa2_tx_conf_fd(priv->net_dev, fd);
1244 
1245 	percpu_extras = this_cpu_ptr(priv->percpu_extras);
1246 	percpu_extras->tx_conf_frames++;
1247 	percpu_extras->tx_conf_bytes += fd_len;
1248 
1249 	/* Check frame errors in the FD field */
1250 	fd_errors = dpaa2_fd_get_ctrl(fd) & DPAA2_FD_TX_ERR_MASK;
1251 	dpaa2_eth_free_tx_fd(priv, fq, fd, true);
1252 
1253 	if (likely(!fd_errors))
1254 		return;
1255 
1256 	if (net_ratelimit())
1257 		netdev_dbg(priv->net_dev, "TX frame FD error: 0x%08x\n",
1258 			   fd_errors);
1259 
1260 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
1261 	/* Tx-conf logically pertains to the egress path. */
1262 	percpu_stats->tx_errors++;
1263 }
1264 
1265 static int dpaa2_eth_set_rx_csum(struct dpaa2_eth_priv *priv, bool enable)
1266 {
1267 	int err;
1268 
1269 	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1270 			       DPNI_OFF_RX_L3_CSUM, enable);
1271 	if (err) {
1272 		netdev_err(priv->net_dev,
1273 			   "dpni_set_offload(RX_L3_CSUM) failed\n");
1274 		return err;
1275 	}
1276 
1277 	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1278 			       DPNI_OFF_RX_L4_CSUM, enable);
1279 	if (err) {
1280 		netdev_err(priv->net_dev,
1281 			   "dpni_set_offload(RX_L4_CSUM) failed\n");
1282 		return err;
1283 	}
1284 
1285 	return 0;
1286 }
1287 
1288 static int dpaa2_eth_set_tx_csum(struct dpaa2_eth_priv *priv, bool enable)
1289 {
1290 	int err;
1291 
1292 	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1293 			       DPNI_OFF_TX_L3_CSUM, enable);
1294 	if (err) {
1295 		netdev_err(priv->net_dev, "dpni_set_offload(TX_L3_CSUM) failed\n");
1296 		return err;
1297 	}
1298 
1299 	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1300 			       DPNI_OFF_TX_L4_CSUM, enable);
1301 	if (err) {
1302 		netdev_err(priv->net_dev, "dpni_set_offload(TX_L4_CSUM) failed\n");
1303 		return err;
1304 	}
1305 
1306 	return 0;
1307 }
1308 
1309 /* Perform a single release command to add buffers
1310  * to the specified buffer pool
1311  */
1312 static int dpaa2_eth_add_bufs(struct dpaa2_eth_priv *priv,
1313 			      struct dpaa2_eth_channel *ch, u16 bpid)
1314 {
1315 	struct device *dev = priv->net_dev->dev.parent;
1316 	u64 buf_array[DPAA2_ETH_BUFS_PER_CMD];
1317 	struct page *page;
1318 	dma_addr_t addr;
1319 	int retries = 0;
1320 	int i, err;
1321 
1322 	for (i = 0; i < DPAA2_ETH_BUFS_PER_CMD; i++) {
1323 		/* Allocate buffer visible to WRIOP + skb shared info +
1324 		 * alignment padding
1325 		 */
1326 		/* allocate one page for each Rx buffer. WRIOP sees
1327 		 * the entire page except for a tailroom reserved for
1328 		 * skb shared info
1329 		 */
1330 		page = dev_alloc_pages(0);
1331 		if (!page)
1332 			goto err_alloc;
1333 
1334 		addr = dma_map_page(dev, page, 0, priv->rx_buf_size,
1335 				    DMA_BIDIRECTIONAL);
1336 		if (unlikely(dma_mapping_error(dev, addr)))
1337 			goto err_map;
1338 
1339 		buf_array[i] = addr;
1340 
1341 		/* tracing point */
1342 		trace_dpaa2_eth_buf_seed(priv->net_dev,
1343 					 page, DPAA2_ETH_RX_BUF_RAW_SIZE,
1344 					 addr, priv->rx_buf_size,
1345 					 bpid);
1346 	}
1347 
1348 release_bufs:
1349 	/* In case the portal is busy, retry until successful */
1350 	while ((err = dpaa2_io_service_release(ch->dpio, bpid,
1351 					       buf_array, i)) == -EBUSY) {
1352 		if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES)
1353 			break;
1354 		cpu_relax();
1355 	}
1356 
1357 	/* If release command failed, clean up and bail out;
1358 	 * not much else we can do about it
1359 	 */
1360 	if (err) {
1361 		dpaa2_eth_free_bufs(priv, buf_array, i);
1362 		return 0;
1363 	}
1364 
1365 	return i;
1366 
1367 err_map:
1368 	__free_pages(page, 0);
1369 err_alloc:
1370 	/* If we managed to allocate at least some buffers,
1371 	 * release them to hardware
1372 	 */
1373 	if (i)
1374 		goto release_bufs;
1375 
1376 	return 0;
1377 }
1378 
1379 static int dpaa2_eth_seed_pool(struct dpaa2_eth_priv *priv, u16 bpid)
1380 {
1381 	int i, j;
1382 	int new_count;
1383 
1384 	for (j = 0; j < priv->num_channels; j++) {
1385 		for (i = 0; i < DPAA2_ETH_NUM_BUFS;
1386 		     i += DPAA2_ETH_BUFS_PER_CMD) {
1387 			new_count = dpaa2_eth_add_bufs(priv, priv->channel[j], bpid);
1388 			priv->channel[j]->buf_count += new_count;
1389 
1390 			if (new_count < DPAA2_ETH_BUFS_PER_CMD) {
1391 				return -ENOMEM;
1392 			}
1393 		}
1394 	}
1395 
1396 	return 0;
1397 }
1398 
1399 /*
1400  * Drain the specified number of buffers from the DPNI's private buffer pool.
1401  * @count must not exceeed DPAA2_ETH_BUFS_PER_CMD
1402  */
1403 static void dpaa2_eth_drain_bufs(struct dpaa2_eth_priv *priv, int count)
1404 {
1405 	u64 buf_array[DPAA2_ETH_BUFS_PER_CMD];
1406 	int retries = 0;
1407 	int ret;
1408 
1409 	do {
1410 		ret = dpaa2_io_service_acquire(NULL, priv->bpid,
1411 					       buf_array, count);
1412 		if (ret < 0) {
1413 			if (ret == -EBUSY &&
1414 			    retries++ < DPAA2_ETH_SWP_BUSY_RETRIES)
1415 				continue;
1416 			netdev_err(priv->net_dev, "dpaa2_io_service_acquire() failed\n");
1417 			return;
1418 		}
1419 		dpaa2_eth_free_bufs(priv, buf_array, ret);
1420 		retries = 0;
1421 	} while (ret);
1422 }
1423 
1424 static void dpaa2_eth_drain_pool(struct dpaa2_eth_priv *priv)
1425 {
1426 	int i;
1427 
1428 	dpaa2_eth_drain_bufs(priv, DPAA2_ETH_BUFS_PER_CMD);
1429 	dpaa2_eth_drain_bufs(priv, 1);
1430 
1431 	for (i = 0; i < priv->num_channels; i++)
1432 		priv->channel[i]->buf_count = 0;
1433 }
1434 
1435 /* Function is called from softirq context only, so we don't need to guard
1436  * the access to percpu count
1437  */
1438 static int dpaa2_eth_refill_pool(struct dpaa2_eth_priv *priv,
1439 				 struct dpaa2_eth_channel *ch,
1440 				 u16 bpid)
1441 {
1442 	int new_count;
1443 
1444 	if (likely(ch->buf_count >= DPAA2_ETH_REFILL_THRESH))
1445 		return 0;
1446 
1447 	do {
1448 		new_count = dpaa2_eth_add_bufs(priv, ch, bpid);
1449 		if (unlikely(!new_count)) {
1450 			/* Out of memory; abort for now, we'll try later on */
1451 			break;
1452 		}
1453 		ch->buf_count += new_count;
1454 	} while (ch->buf_count < DPAA2_ETH_NUM_BUFS);
1455 
1456 	if (unlikely(ch->buf_count < DPAA2_ETH_NUM_BUFS))
1457 		return -ENOMEM;
1458 
1459 	return 0;
1460 }
1461 
1462 static void dpaa2_eth_sgt_cache_drain(struct dpaa2_eth_priv *priv)
1463 {
1464 	struct dpaa2_eth_sgt_cache *sgt_cache;
1465 	u16 count;
1466 	int k, i;
1467 
1468 	for_each_possible_cpu(k) {
1469 		sgt_cache = per_cpu_ptr(priv->sgt_cache, k);
1470 		count = sgt_cache->count;
1471 
1472 		for (i = 0; i < count; i++)
1473 			kfree(sgt_cache->buf[i]);
1474 		sgt_cache->count = 0;
1475 	}
1476 }
1477 
1478 static int dpaa2_eth_pull_channel(struct dpaa2_eth_channel *ch)
1479 {
1480 	int err;
1481 	int dequeues = -1;
1482 
1483 	/* Retry while portal is busy */
1484 	do {
1485 		err = dpaa2_io_service_pull_channel(ch->dpio, ch->ch_id,
1486 						    ch->store);
1487 		dequeues++;
1488 		cpu_relax();
1489 	} while (err == -EBUSY && dequeues < DPAA2_ETH_SWP_BUSY_RETRIES);
1490 
1491 	ch->stats.dequeue_portal_busy += dequeues;
1492 	if (unlikely(err))
1493 		ch->stats.pull_err++;
1494 
1495 	return err;
1496 }
1497 
1498 /* NAPI poll routine
1499  *
1500  * Frames are dequeued from the QMan channel associated with this NAPI context.
1501  * Rx, Tx confirmation and (if configured) Rx error frames all count
1502  * towards the NAPI budget.
1503  */
1504 static int dpaa2_eth_poll(struct napi_struct *napi, int budget)
1505 {
1506 	struct dpaa2_eth_channel *ch;
1507 	struct dpaa2_eth_priv *priv;
1508 	int rx_cleaned = 0, txconf_cleaned = 0;
1509 	struct dpaa2_eth_fq *fq, *txc_fq = NULL;
1510 	struct netdev_queue *nq;
1511 	int store_cleaned, work_done;
1512 	struct list_head rx_list;
1513 	int retries = 0;
1514 	u16 flowid;
1515 	int err;
1516 
1517 	ch = container_of(napi, struct dpaa2_eth_channel, napi);
1518 	ch->xdp.res = 0;
1519 	priv = ch->priv;
1520 
1521 	INIT_LIST_HEAD(&rx_list);
1522 	ch->rx_list = &rx_list;
1523 
1524 	do {
1525 		err = dpaa2_eth_pull_channel(ch);
1526 		if (unlikely(err))
1527 			break;
1528 
1529 		/* Refill pool if appropriate */
1530 		dpaa2_eth_refill_pool(priv, ch, priv->bpid);
1531 
1532 		store_cleaned = dpaa2_eth_consume_frames(ch, &fq);
1533 		if (store_cleaned <= 0)
1534 			break;
1535 		if (fq->type == DPAA2_RX_FQ) {
1536 			rx_cleaned += store_cleaned;
1537 			flowid = fq->flowid;
1538 		} else {
1539 			txconf_cleaned += store_cleaned;
1540 			/* We have a single Tx conf FQ on this channel */
1541 			txc_fq = fq;
1542 		}
1543 
1544 		/* If we either consumed the whole NAPI budget with Rx frames
1545 		 * or we reached the Tx confirmations threshold, we're done.
1546 		 */
1547 		if (rx_cleaned >= budget ||
1548 		    txconf_cleaned >= DPAA2_ETH_TXCONF_PER_NAPI) {
1549 			work_done = budget;
1550 			goto out;
1551 		}
1552 	} while (store_cleaned);
1553 
1554 	/* We didn't consume the entire budget, so finish napi and
1555 	 * re-enable data availability notifications
1556 	 */
1557 	napi_complete_done(napi, rx_cleaned);
1558 	do {
1559 		err = dpaa2_io_service_rearm(ch->dpio, &ch->nctx);
1560 		cpu_relax();
1561 	} while (err == -EBUSY && retries++ < DPAA2_ETH_SWP_BUSY_RETRIES);
1562 	WARN_ONCE(err, "CDAN notifications rearm failed on core %d",
1563 		  ch->nctx.desired_cpu);
1564 
1565 	work_done = max(rx_cleaned, 1);
1566 
1567 out:
1568 	netif_receive_skb_list(ch->rx_list);
1569 
1570 	if (txc_fq && txc_fq->dq_frames) {
1571 		nq = netdev_get_tx_queue(priv->net_dev, txc_fq->flowid);
1572 		netdev_tx_completed_queue(nq, txc_fq->dq_frames,
1573 					  txc_fq->dq_bytes);
1574 		txc_fq->dq_frames = 0;
1575 		txc_fq->dq_bytes = 0;
1576 	}
1577 
1578 	if (ch->xdp.res & XDP_REDIRECT)
1579 		xdp_do_flush_map();
1580 	else if (rx_cleaned && ch->xdp.res & XDP_TX)
1581 		dpaa2_eth_xdp_tx_flush(priv, ch, &priv->fq[flowid]);
1582 
1583 	return work_done;
1584 }
1585 
1586 static void dpaa2_eth_enable_ch_napi(struct dpaa2_eth_priv *priv)
1587 {
1588 	struct dpaa2_eth_channel *ch;
1589 	int i;
1590 
1591 	for (i = 0; i < priv->num_channels; i++) {
1592 		ch = priv->channel[i];
1593 		napi_enable(&ch->napi);
1594 	}
1595 }
1596 
1597 static void dpaa2_eth_disable_ch_napi(struct dpaa2_eth_priv *priv)
1598 {
1599 	struct dpaa2_eth_channel *ch;
1600 	int i;
1601 
1602 	for (i = 0; i < priv->num_channels; i++) {
1603 		ch = priv->channel[i];
1604 		napi_disable(&ch->napi);
1605 	}
1606 }
1607 
1608 void dpaa2_eth_set_rx_taildrop(struct dpaa2_eth_priv *priv,
1609 			       bool tx_pause, bool pfc)
1610 {
1611 	struct dpni_taildrop td = {0};
1612 	struct dpaa2_eth_fq *fq;
1613 	int i, err;
1614 
1615 	/* FQ taildrop: threshold is in bytes, per frame queue. Enabled if
1616 	 * flow control is disabled (as it might interfere with either the
1617 	 * buffer pool depletion trigger for pause frames or with the group
1618 	 * congestion trigger for PFC frames)
1619 	 */
1620 	td.enable = !tx_pause;
1621 	if (priv->rx_fqtd_enabled == td.enable)
1622 		goto set_cgtd;
1623 
1624 	td.threshold = DPAA2_ETH_FQ_TAILDROP_THRESH;
1625 	td.units = DPNI_CONGESTION_UNIT_BYTES;
1626 
1627 	for (i = 0; i < priv->num_fqs; i++) {
1628 		fq = &priv->fq[i];
1629 		if (fq->type != DPAA2_RX_FQ)
1630 			continue;
1631 		err = dpni_set_taildrop(priv->mc_io, 0, priv->mc_token,
1632 					DPNI_CP_QUEUE, DPNI_QUEUE_RX,
1633 					fq->tc, fq->flowid, &td);
1634 		if (err) {
1635 			netdev_err(priv->net_dev,
1636 				   "dpni_set_taildrop(FQ) failed\n");
1637 			return;
1638 		}
1639 	}
1640 
1641 	priv->rx_fqtd_enabled = td.enable;
1642 
1643 set_cgtd:
1644 	/* Congestion group taildrop: threshold is in frames, per group
1645 	 * of FQs belonging to the same traffic class
1646 	 * Enabled if general Tx pause disabled or if PFCs are enabled
1647 	 * (congestion group threhsold for PFC generation is lower than the
1648 	 * CG taildrop threshold, so it won't interfere with it; we also
1649 	 * want frames in non-PFC enabled traffic classes to be kept in check)
1650 	 */
1651 	td.enable = !tx_pause || (tx_pause && pfc);
1652 	if (priv->rx_cgtd_enabled == td.enable)
1653 		return;
1654 
1655 	td.threshold = DPAA2_ETH_CG_TAILDROP_THRESH(priv);
1656 	td.units = DPNI_CONGESTION_UNIT_FRAMES;
1657 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
1658 		err = dpni_set_taildrop(priv->mc_io, 0, priv->mc_token,
1659 					DPNI_CP_GROUP, DPNI_QUEUE_RX,
1660 					i, 0, &td);
1661 		if (err) {
1662 			netdev_err(priv->net_dev,
1663 				   "dpni_set_taildrop(CG) failed\n");
1664 			return;
1665 		}
1666 	}
1667 
1668 	priv->rx_cgtd_enabled = td.enable;
1669 }
1670 
1671 static int dpaa2_eth_link_state_update(struct dpaa2_eth_priv *priv)
1672 {
1673 	struct dpni_link_state state = {0};
1674 	bool tx_pause;
1675 	int err;
1676 
1677 	err = dpni_get_link_state(priv->mc_io, 0, priv->mc_token, &state);
1678 	if (unlikely(err)) {
1679 		netdev_err(priv->net_dev,
1680 			   "dpni_get_link_state() failed\n");
1681 		return err;
1682 	}
1683 
1684 	/* If Tx pause frame settings have changed, we need to update
1685 	 * Rx FQ taildrop configuration as well. We configure taildrop
1686 	 * only when pause frame generation is disabled.
1687 	 */
1688 	tx_pause = dpaa2_eth_tx_pause_enabled(state.options);
1689 	dpaa2_eth_set_rx_taildrop(priv, tx_pause, priv->pfc_enabled);
1690 
1691 	/* When we manage the MAC/PHY using phylink there is no need
1692 	 * to manually update the netif_carrier.
1693 	 */
1694 	if (dpaa2_eth_is_type_phy(priv))
1695 		goto out;
1696 
1697 	/* Chech link state; speed / duplex changes are not treated yet */
1698 	if (priv->link_state.up == state.up)
1699 		goto out;
1700 
1701 	if (state.up) {
1702 		netif_carrier_on(priv->net_dev);
1703 		netif_tx_start_all_queues(priv->net_dev);
1704 	} else {
1705 		netif_tx_stop_all_queues(priv->net_dev);
1706 		netif_carrier_off(priv->net_dev);
1707 	}
1708 
1709 	netdev_info(priv->net_dev, "Link Event: state %s\n",
1710 		    state.up ? "up" : "down");
1711 
1712 out:
1713 	priv->link_state = state;
1714 
1715 	return 0;
1716 }
1717 
1718 static int dpaa2_eth_open(struct net_device *net_dev)
1719 {
1720 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1721 	int err;
1722 
1723 	err = dpaa2_eth_seed_pool(priv, priv->bpid);
1724 	if (err) {
1725 		/* Not much to do; the buffer pool, though not filled up,
1726 		 * may still contain some buffers which would enable us
1727 		 * to limp on.
1728 		 */
1729 		netdev_err(net_dev, "Buffer seeding failed for DPBP %d (bpid=%d)\n",
1730 			   priv->dpbp_dev->obj_desc.id, priv->bpid);
1731 	}
1732 
1733 	if (!dpaa2_eth_is_type_phy(priv)) {
1734 		/* We'll only start the txqs when the link is actually ready;
1735 		 * make sure we don't race against the link up notification,
1736 		 * which may come immediately after dpni_enable();
1737 		 */
1738 		netif_tx_stop_all_queues(net_dev);
1739 
1740 		/* Also, explicitly set carrier off, otherwise
1741 		 * netif_carrier_ok() will return true and cause 'ip link show'
1742 		 * to report the LOWER_UP flag, even though the link
1743 		 * notification wasn't even received.
1744 		 */
1745 		netif_carrier_off(net_dev);
1746 	}
1747 	dpaa2_eth_enable_ch_napi(priv);
1748 
1749 	err = dpni_enable(priv->mc_io, 0, priv->mc_token);
1750 	if (err < 0) {
1751 		netdev_err(net_dev, "dpni_enable() failed\n");
1752 		goto enable_err;
1753 	}
1754 
1755 	if (dpaa2_eth_is_type_phy(priv))
1756 		phylink_start(priv->mac->phylink);
1757 
1758 	return 0;
1759 
1760 enable_err:
1761 	dpaa2_eth_disable_ch_napi(priv);
1762 	dpaa2_eth_drain_pool(priv);
1763 	return err;
1764 }
1765 
1766 /* Total number of in-flight frames on ingress queues */
1767 static u32 dpaa2_eth_ingress_fq_count(struct dpaa2_eth_priv *priv)
1768 {
1769 	struct dpaa2_eth_fq *fq;
1770 	u32 fcnt = 0, bcnt = 0, total = 0;
1771 	int i, err;
1772 
1773 	for (i = 0; i < priv->num_fqs; i++) {
1774 		fq = &priv->fq[i];
1775 		err = dpaa2_io_query_fq_count(NULL, fq->fqid, &fcnt, &bcnt);
1776 		if (err) {
1777 			netdev_warn(priv->net_dev, "query_fq_count failed");
1778 			break;
1779 		}
1780 		total += fcnt;
1781 	}
1782 
1783 	return total;
1784 }
1785 
1786 static void dpaa2_eth_wait_for_ingress_fq_empty(struct dpaa2_eth_priv *priv)
1787 {
1788 	int retries = 10;
1789 	u32 pending;
1790 
1791 	do {
1792 		pending = dpaa2_eth_ingress_fq_count(priv);
1793 		if (pending)
1794 			msleep(100);
1795 	} while (pending && --retries);
1796 }
1797 
1798 #define DPNI_TX_PENDING_VER_MAJOR	7
1799 #define DPNI_TX_PENDING_VER_MINOR	13
1800 static void dpaa2_eth_wait_for_egress_fq_empty(struct dpaa2_eth_priv *priv)
1801 {
1802 	union dpni_statistics stats;
1803 	int retries = 10;
1804 	int err;
1805 
1806 	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_TX_PENDING_VER_MAJOR,
1807 				   DPNI_TX_PENDING_VER_MINOR) < 0)
1808 		goto out;
1809 
1810 	do {
1811 		err = dpni_get_statistics(priv->mc_io, 0, priv->mc_token, 6,
1812 					  &stats);
1813 		if (err)
1814 			goto out;
1815 		if (stats.page_6.tx_pending_frames == 0)
1816 			return;
1817 	} while (--retries);
1818 
1819 out:
1820 	msleep(500);
1821 }
1822 
1823 static int dpaa2_eth_stop(struct net_device *net_dev)
1824 {
1825 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1826 	int dpni_enabled = 0;
1827 	int retries = 10;
1828 
1829 	if (dpaa2_eth_is_type_phy(priv)) {
1830 		phylink_stop(priv->mac->phylink);
1831 	} else {
1832 		netif_tx_stop_all_queues(net_dev);
1833 		netif_carrier_off(net_dev);
1834 	}
1835 
1836 	/* On dpni_disable(), the MC firmware will:
1837 	 * - stop MAC Rx and wait for all Rx frames to be enqueued to software
1838 	 * - cut off WRIOP dequeues from egress FQs and wait until transmission
1839 	 * of all in flight Tx frames is finished (and corresponding Tx conf
1840 	 * frames are enqueued back to software)
1841 	 *
1842 	 * Before calling dpni_disable(), we wait for all Tx frames to arrive
1843 	 * on WRIOP. After it finishes, wait until all remaining frames on Rx
1844 	 * and Tx conf queues are consumed on NAPI poll.
1845 	 */
1846 	dpaa2_eth_wait_for_egress_fq_empty(priv);
1847 
1848 	do {
1849 		dpni_disable(priv->mc_io, 0, priv->mc_token);
1850 		dpni_is_enabled(priv->mc_io, 0, priv->mc_token, &dpni_enabled);
1851 		if (dpni_enabled)
1852 			/* Allow the hardware some slack */
1853 			msleep(100);
1854 	} while (dpni_enabled && --retries);
1855 	if (!retries) {
1856 		netdev_warn(net_dev, "Retry count exceeded disabling DPNI\n");
1857 		/* Must go on and disable NAPI nonetheless, so we don't crash at
1858 		 * the next "ifconfig up"
1859 		 */
1860 	}
1861 
1862 	dpaa2_eth_wait_for_ingress_fq_empty(priv);
1863 	dpaa2_eth_disable_ch_napi(priv);
1864 
1865 	/* Empty the buffer pool */
1866 	dpaa2_eth_drain_pool(priv);
1867 
1868 	/* Empty the Scatter-Gather Buffer cache */
1869 	dpaa2_eth_sgt_cache_drain(priv);
1870 
1871 	return 0;
1872 }
1873 
1874 static int dpaa2_eth_set_addr(struct net_device *net_dev, void *addr)
1875 {
1876 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1877 	struct device *dev = net_dev->dev.parent;
1878 	int err;
1879 
1880 	err = eth_mac_addr(net_dev, addr);
1881 	if (err < 0) {
1882 		dev_err(dev, "eth_mac_addr() failed (%d)\n", err);
1883 		return err;
1884 	}
1885 
1886 	err = dpni_set_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
1887 					net_dev->dev_addr);
1888 	if (err) {
1889 		dev_err(dev, "dpni_set_primary_mac_addr() failed (%d)\n", err);
1890 		return err;
1891 	}
1892 
1893 	return 0;
1894 }
1895 
1896 /** Fill in counters maintained by the GPP driver. These may be different from
1897  * the hardware counters obtained by ethtool.
1898  */
1899 static void dpaa2_eth_get_stats(struct net_device *net_dev,
1900 				struct rtnl_link_stats64 *stats)
1901 {
1902 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1903 	struct rtnl_link_stats64 *percpu_stats;
1904 	u64 *cpustats;
1905 	u64 *netstats = (u64 *)stats;
1906 	int i, j;
1907 	int num = sizeof(struct rtnl_link_stats64) / sizeof(u64);
1908 
1909 	for_each_possible_cpu(i) {
1910 		percpu_stats = per_cpu_ptr(priv->percpu_stats, i);
1911 		cpustats = (u64 *)percpu_stats;
1912 		for (j = 0; j < num; j++)
1913 			netstats[j] += cpustats[j];
1914 	}
1915 }
1916 
1917 /* Copy mac unicast addresses from @net_dev to @priv.
1918  * Its sole purpose is to make dpaa2_eth_set_rx_mode() more readable.
1919  */
1920 static void dpaa2_eth_add_uc_hw_addr(const struct net_device *net_dev,
1921 				     struct dpaa2_eth_priv *priv)
1922 {
1923 	struct netdev_hw_addr *ha;
1924 	int err;
1925 
1926 	netdev_for_each_uc_addr(ha, net_dev) {
1927 		err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token,
1928 					ha->addr);
1929 		if (err)
1930 			netdev_warn(priv->net_dev,
1931 				    "Could not add ucast MAC %pM to the filtering table (err %d)\n",
1932 				    ha->addr, err);
1933 	}
1934 }
1935 
1936 /* Copy mac multicast addresses from @net_dev to @priv
1937  * Its sole purpose is to make dpaa2_eth_set_rx_mode() more readable.
1938  */
1939 static void dpaa2_eth_add_mc_hw_addr(const struct net_device *net_dev,
1940 				     struct dpaa2_eth_priv *priv)
1941 {
1942 	struct netdev_hw_addr *ha;
1943 	int err;
1944 
1945 	netdev_for_each_mc_addr(ha, net_dev) {
1946 		err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token,
1947 					ha->addr);
1948 		if (err)
1949 			netdev_warn(priv->net_dev,
1950 				    "Could not add mcast MAC %pM to the filtering table (err %d)\n",
1951 				    ha->addr, err);
1952 	}
1953 }
1954 
1955 static void dpaa2_eth_set_rx_mode(struct net_device *net_dev)
1956 {
1957 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1958 	int uc_count = netdev_uc_count(net_dev);
1959 	int mc_count = netdev_mc_count(net_dev);
1960 	u8 max_mac = priv->dpni_attrs.mac_filter_entries;
1961 	u32 options = priv->dpni_attrs.options;
1962 	u16 mc_token = priv->mc_token;
1963 	struct fsl_mc_io *mc_io = priv->mc_io;
1964 	int err;
1965 
1966 	/* Basic sanity checks; these probably indicate a misconfiguration */
1967 	if (options & DPNI_OPT_NO_MAC_FILTER && max_mac != 0)
1968 		netdev_info(net_dev,
1969 			    "mac_filter_entries=%d, DPNI_OPT_NO_MAC_FILTER option must be disabled\n",
1970 			    max_mac);
1971 
1972 	/* Force promiscuous if the uc or mc counts exceed our capabilities. */
1973 	if (uc_count > max_mac) {
1974 		netdev_info(net_dev,
1975 			    "Unicast addr count reached %d, max allowed is %d; forcing promisc\n",
1976 			    uc_count, max_mac);
1977 		goto force_promisc;
1978 	}
1979 	if (mc_count + uc_count > max_mac) {
1980 		netdev_info(net_dev,
1981 			    "Unicast + multicast addr count reached %d, max allowed is %d; forcing promisc\n",
1982 			    uc_count + mc_count, max_mac);
1983 		goto force_mc_promisc;
1984 	}
1985 
1986 	/* Adjust promisc settings due to flag combinations */
1987 	if (net_dev->flags & IFF_PROMISC)
1988 		goto force_promisc;
1989 	if (net_dev->flags & IFF_ALLMULTI) {
1990 		/* First, rebuild unicast filtering table. This should be done
1991 		 * in promisc mode, in order to avoid frame loss while we
1992 		 * progressively add entries to the table.
1993 		 * We don't know whether we had been in promisc already, and
1994 		 * making an MC call to find out is expensive; so set uc promisc
1995 		 * nonetheless.
1996 		 */
1997 		err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 1);
1998 		if (err)
1999 			netdev_warn(net_dev, "Can't set uc promisc\n");
2000 
2001 		/* Actual uc table reconstruction. */
2002 		err = dpni_clear_mac_filters(mc_io, 0, mc_token, 1, 0);
2003 		if (err)
2004 			netdev_warn(net_dev, "Can't clear uc filters\n");
2005 		dpaa2_eth_add_uc_hw_addr(net_dev, priv);
2006 
2007 		/* Finally, clear uc promisc and set mc promisc as requested. */
2008 		err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 0);
2009 		if (err)
2010 			netdev_warn(net_dev, "Can't clear uc promisc\n");
2011 		goto force_mc_promisc;
2012 	}
2013 
2014 	/* Neither unicast, nor multicast promisc will be on... eventually.
2015 	 * For now, rebuild mac filtering tables while forcing both of them on.
2016 	 */
2017 	err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 1);
2018 	if (err)
2019 		netdev_warn(net_dev, "Can't set uc promisc (%d)\n", err);
2020 	err = dpni_set_multicast_promisc(mc_io, 0, mc_token, 1);
2021 	if (err)
2022 		netdev_warn(net_dev, "Can't set mc promisc (%d)\n", err);
2023 
2024 	/* Actual mac filtering tables reconstruction */
2025 	err = dpni_clear_mac_filters(mc_io, 0, mc_token, 1, 1);
2026 	if (err)
2027 		netdev_warn(net_dev, "Can't clear mac filters\n");
2028 	dpaa2_eth_add_mc_hw_addr(net_dev, priv);
2029 	dpaa2_eth_add_uc_hw_addr(net_dev, priv);
2030 
2031 	/* Now we can clear both ucast and mcast promisc, without risking
2032 	 * to drop legitimate frames anymore.
2033 	 */
2034 	err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 0);
2035 	if (err)
2036 		netdev_warn(net_dev, "Can't clear ucast promisc\n");
2037 	err = dpni_set_multicast_promisc(mc_io, 0, mc_token, 0);
2038 	if (err)
2039 		netdev_warn(net_dev, "Can't clear mcast promisc\n");
2040 
2041 	return;
2042 
2043 force_promisc:
2044 	err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 1);
2045 	if (err)
2046 		netdev_warn(net_dev, "Can't set ucast promisc\n");
2047 force_mc_promisc:
2048 	err = dpni_set_multicast_promisc(mc_io, 0, mc_token, 1);
2049 	if (err)
2050 		netdev_warn(net_dev, "Can't set mcast promisc\n");
2051 }
2052 
2053 static int dpaa2_eth_set_features(struct net_device *net_dev,
2054 				  netdev_features_t features)
2055 {
2056 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2057 	netdev_features_t changed = features ^ net_dev->features;
2058 	bool enable;
2059 	int err;
2060 
2061 	if (changed & NETIF_F_RXCSUM) {
2062 		enable = !!(features & NETIF_F_RXCSUM);
2063 		err = dpaa2_eth_set_rx_csum(priv, enable);
2064 		if (err)
2065 			return err;
2066 	}
2067 
2068 	if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
2069 		enable = !!(features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM));
2070 		err = dpaa2_eth_set_tx_csum(priv, enable);
2071 		if (err)
2072 			return err;
2073 	}
2074 
2075 	return 0;
2076 }
2077 
2078 static int dpaa2_eth_ts_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
2079 {
2080 	struct dpaa2_eth_priv *priv = netdev_priv(dev);
2081 	struct hwtstamp_config config;
2082 
2083 	if (!dpaa2_ptp)
2084 		return -EINVAL;
2085 
2086 	if (copy_from_user(&config, rq->ifr_data, sizeof(config)))
2087 		return -EFAULT;
2088 
2089 	switch (config.tx_type) {
2090 	case HWTSTAMP_TX_OFF:
2091 	case HWTSTAMP_TX_ON:
2092 	case HWTSTAMP_TX_ONESTEP_SYNC:
2093 		priv->tx_tstamp_type = config.tx_type;
2094 		break;
2095 	default:
2096 		return -ERANGE;
2097 	}
2098 
2099 	if (config.rx_filter == HWTSTAMP_FILTER_NONE) {
2100 		priv->rx_tstamp = false;
2101 	} else {
2102 		priv->rx_tstamp = true;
2103 		/* TS is set for all frame types, not only those requested */
2104 		config.rx_filter = HWTSTAMP_FILTER_ALL;
2105 	}
2106 
2107 	return copy_to_user(rq->ifr_data, &config, sizeof(config)) ?
2108 			-EFAULT : 0;
2109 }
2110 
2111 static int dpaa2_eth_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
2112 {
2113 	struct dpaa2_eth_priv *priv = netdev_priv(dev);
2114 
2115 	if (cmd == SIOCSHWTSTAMP)
2116 		return dpaa2_eth_ts_ioctl(dev, rq, cmd);
2117 
2118 	if (dpaa2_eth_is_type_phy(priv))
2119 		return phylink_mii_ioctl(priv->mac->phylink, rq, cmd);
2120 
2121 	return -EOPNOTSUPP;
2122 }
2123 
2124 static bool xdp_mtu_valid(struct dpaa2_eth_priv *priv, int mtu)
2125 {
2126 	int mfl, linear_mfl;
2127 
2128 	mfl = DPAA2_ETH_L2_MAX_FRM(mtu);
2129 	linear_mfl = priv->rx_buf_size - DPAA2_ETH_RX_HWA_SIZE -
2130 		     dpaa2_eth_rx_head_room(priv) - XDP_PACKET_HEADROOM;
2131 
2132 	if (mfl > linear_mfl) {
2133 		netdev_warn(priv->net_dev, "Maximum MTU for XDP is %d\n",
2134 			    linear_mfl - VLAN_ETH_HLEN);
2135 		return false;
2136 	}
2137 
2138 	return true;
2139 }
2140 
2141 static int dpaa2_eth_set_rx_mfl(struct dpaa2_eth_priv *priv, int mtu, bool has_xdp)
2142 {
2143 	int mfl, err;
2144 
2145 	/* We enforce a maximum Rx frame length based on MTU only if we have
2146 	 * an XDP program attached (in order to avoid Rx S/G frames).
2147 	 * Otherwise, we accept all incoming frames as long as they are not
2148 	 * larger than maximum size supported in hardware
2149 	 */
2150 	if (has_xdp)
2151 		mfl = DPAA2_ETH_L2_MAX_FRM(mtu);
2152 	else
2153 		mfl = DPAA2_ETH_MFL;
2154 
2155 	err = dpni_set_max_frame_length(priv->mc_io, 0, priv->mc_token, mfl);
2156 	if (err) {
2157 		netdev_err(priv->net_dev, "dpni_set_max_frame_length failed\n");
2158 		return err;
2159 	}
2160 
2161 	return 0;
2162 }
2163 
2164 static int dpaa2_eth_change_mtu(struct net_device *dev, int new_mtu)
2165 {
2166 	struct dpaa2_eth_priv *priv = netdev_priv(dev);
2167 	int err;
2168 
2169 	if (!priv->xdp_prog)
2170 		goto out;
2171 
2172 	if (!xdp_mtu_valid(priv, new_mtu))
2173 		return -EINVAL;
2174 
2175 	err = dpaa2_eth_set_rx_mfl(priv, new_mtu, true);
2176 	if (err)
2177 		return err;
2178 
2179 out:
2180 	dev->mtu = new_mtu;
2181 	return 0;
2182 }
2183 
2184 static int dpaa2_eth_update_rx_buffer_headroom(struct dpaa2_eth_priv *priv, bool has_xdp)
2185 {
2186 	struct dpni_buffer_layout buf_layout = {0};
2187 	int err;
2188 
2189 	err = dpni_get_buffer_layout(priv->mc_io, 0, priv->mc_token,
2190 				     DPNI_QUEUE_RX, &buf_layout);
2191 	if (err) {
2192 		netdev_err(priv->net_dev, "dpni_get_buffer_layout failed\n");
2193 		return err;
2194 	}
2195 
2196 	/* Reserve extra headroom for XDP header size changes */
2197 	buf_layout.data_head_room = dpaa2_eth_rx_head_room(priv) +
2198 				    (has_xdp ? XDP_PACKET_HEADROOM : 0);
2199 	buf_layout.options = DPNI_BUF_LAYOUT_OPT_DATA_HEAD_ROOM;
2200 	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
2201 				     DPNI_QUEUE_RX, &buf_layout);
2202 	if (err) {
2203 		netdev_err(priv->net_dev, "dpni_set_buffer_layout failed\n");
2204 		return err;
2205 	}
2206 
2207 	return 0;
2208 }
2209 
2210 static int dpaa2_eth_setup_xdp(struct net_device *dev, struct bpf_prog *prog)
2211 {
2212 	struct dpaa2_eth_priv *priv = netdev_priv(dev);
2213 	struct dpaa2_eth_channel *ch;
2214 	struct bpf_prog *old;
2215 	bool up, need_update;
2216 	int i, err;
2217 
2218 	if (prog && !xdp_mtu_valid(priv, dev->mtu))
2219 		return -EINVAL;
2220 
2221 	if (prog)
2222 		bpf_prog_add(prog, priv->num_channels);
2223 
2224 	up = netif_running(dev);
2225 	need_update = (!!priv->xdp_prog != !!prog);
2226 
2227 	if (up)
2228 		dpaa2_eth_stop(dev);
2229 
2230 	/* While in xdp mode, enforce a maximum Rx frame size based on MTU.
2231 	 * Also, when switching between xdp/non-xdp modes we need to reconfigure
2232 	 * our Rx buffer layout. Buffer pool was drained on dpaa2_eth_stop,
2233 	 * so we are sure no old format buffers will be used from now on.
2234 	 */
2235 	if (need_update) {
2236 		err = dpaa2_eth_set_rx_mfl(priv, dev->mtu, !!prog);
2237 		if (err)
2238 			goto out_err;
2239 		err = dpaa2_eth_update_rx_buffer_headroom(priv, !!prog);
2240 		if (err)
2241 			goto out_err;
2242 	}
2243 
2244 	old = xchg(&priv->xdp_prog, prog);
2245 	if (old)
2246 		bpf_prog_put(old);
2247 
2248 	for (i = 0; i < priv->num_channels; i++) {
2249 		ch = priv->channel[i];
2250 		old = xchg(&ch->xdp.prog, prog);
2251 		if (old)
2252 			bpf_prog_put(old);
2253 	}
2254 
2255 	if (up) {
2256 		err = dpaa2_eth_open(dev);
2257 		if (err)
2258 			return err;
2259 	}
2260 
2261 	return 0;
2262 
2263 out_err:
2264 	if (prog)
2265 		bpf_prog_sub(prog, priv->num_channels);
2266 	if (up)
2267 		dpaa2_eth_open(dev);
2268 
2269 	return err;
2270 }
2271 
2272 static int dpaa2_eth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2273 {
2274 	switch (xdp->command) {
2275 	case XDP_SETUP_PROG:
2276 		return dpaa2_eth_setup_xdp(dev, xdp->prog);
2277 	default:
2278 		return -EINVAL;
2279 	}
2280 
2281 	return 0;
2282 }
2283 
2284 static int dpaa2_eth_xdp_create_fd(struct net_device *net_dev,
2285 				   struct xdp_frame *xdpf,
2286 				   struct dpaa2_fd *fd)
2287 {
2288 	struct device *dev = net_dev->dev.parent;
2289 	unsigned int needed_headroom;
2290 	struct dpaa2_eth_swa *swa;
2291 	void *buffer_start, *aligned_start;
2292 	dma_addr_t addr;
2293 
2294 	/* We require a minimum headroom to be able to transmit the frame.
2295 	 * Otherwise return an error and let the original net_device handle it
2296 	 */
2297 	needed_headroom = dpaa2_eth_needed_headroom(NULL);
2298 	if (xdpf->headroom < needed_headroom)
2299 		return -EINVAL;
2300 
2301 	/* Setup the FD fields */
2302 	memset(fd, 0, sizeof(*fd));
2303 
2304 	/* Align FD address, if possible */
2305 	buffer_start = xdpf->data - needed_headroom;
2306 	aligned_start = PTR_ALIGN(buffer_start - DPAA2_ETH_TX_BUF_ALIGN,
2307 				  DPAA2_ETH_TX_BUF_ALIGN);
2308 	if (aligned_start >= xdpf->data - xdpf->headroom)
2309 		buffer_start = aligned_start;
2310 
2311 	swa = (struct dpaa2_eth_swa *)buffer_start;
2312 	/* fill in necessary fields here */
2313 	swa->type = DPAA2_ETH_SWA_XDP;
2314 	swa->xdp.dma_size = xdpf->data + xdpf->len - buffer_start;
2315 	swa->xdp.xdpf = xdpf;
2316 
2317 	addr = dma_map_single(dev, buffer_start,
2318 			      swa->xdp.dma_size,
2319 			      DMA_BIDIRECTIONAL);
2320 	if (unlikely(dma_mapping_error(dev, addr)))
2321 		return -ENOMEM;
2322 
2323 	dpaa2_fd_set_addr(fd, addr);
2324 	dpaa2_fd_set_offset(fd, xdpf->data - buffer_start);
2325 	dpaa2_fd_set_len(fd, xdpf->len);
2326 	dpaa2_fd_set_format(fd, dpaa2_fd_single);
2327 	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
2328 
2329 	return 0;
2330 }
2331 
2332 static int dpaa2_eth_xdp_xmit(struct net_device *net_dev, int n,
2333 			      struct xdp_frame **frames, u32 flags)
2334 {
2335 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2336 	struct dpaa2_eth_xdp_fds *xdp_redirect_fds;
2337 	struct rtnl_link_stats64 *percpu_stats;
2338 	struct dpaa2_eth_fq *fq;
2339 	struct dpaa2_fd *fds;
2340 	int enqueued, i, err;
2341 
2342 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
2343 		return -EINVAL;
2344 
2345 	if (!netif_running(net_dev))
2346 		return -ENETDOWN;
2347 
2348 	fq = &priv->fq[smp_processor_id()];
2349 	xdp_redirect_fds = &fq->xdp_redirect_fds;
2350 	fds = xdp_redirect_fds->fds;
2351 
2352 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
2353 
2354 	/* create a FD for each xdp_frame in the list received */
2355 	for (i = 0; i < n; i++) {
2356 		err = dpaa2_eth_xdp_create_fd(net_dev, frames[i], &fds[i]);
2357 		if (err)
2358 			break;
2359 	}
2360 	xdp_redirect_fds->num = i;
2361 
2362 	/* enqueue all the frame descriptors */
2363 	enqueued = dpaa2_eth_xdp_flush(priv, fq, xdp_redirect_fds);
2364 
2365 	/* update statistics */
2366 	percpu_stats->tx_packets += enqueued;
2367 	for (i = 0; i < enqueued; i++)
2368 		percpu_stats->tx_bytes += dpaa2_fd_get_len(&fds[i]);
2369 	for (i = enqueued; i < n; i++)
2370 		xdp_return_frame_rx_napi(frames[i]);
2371 
2372 	return enqueued;
2373 }
2374 
2375 static int update_xps(struct dpaa2_eth_priv *priv)
2376 {
2377 	struct net_device *net_dev = priv->net_dev;
2378 	struct cpumask xps_mask;
2379 	struct dpaa2_eth_fq *fq;
2380 	int i, num_queues, netdev_queues;
2381 	int err = 0;
2382 
2383 	num_queues = dpaa2_eth_queue_count(priv);
2384 	netdev_queues = (net_dev->num_tc ? : 1) * num_queues;
2385 
2386 	/* The first <num_queues> entries in priv->fq array are Tx/Tx conf
2387 	 * queues, so only process those
2388 	 */
2389 	for (i = 0; i < netdev_queues; i++) {
2390 		fq = &priv->fq[i % num_queues];
2391 
2392 		cpumask_clear(&xps_mask);
2393 		cpumask_set_cpu(fq->target_cpu, &xps_mask);
2394 
2395 		err = netif_set_xps_queue(net_dev, &xps_mask, i);
2396 		if (err) {
2397 			netdev_warn_once(net_dev, "Error setting XPS queue\n");
2398 			break;
2399 		}
2400 	}
2401 
2402 	return err;
2403 }
2404 
2405 static int dpaa2_eth_setup_mqprio(struct net_device *net_dev,
2406 				  struct tc_mqprio_qopt *mqprio)
2407 {
2408 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2409 	u8 num_tc, num_queues;
2410 	int i;
2411 
2412 	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
2413 	num_queues = dpaa2_eth_queue_count(priv);
2414 	num_tc = mqprio->num_tc;
2415 
2416 	if (num_tc == net_dev->num_tc)
2417 		return 0;
2418 
2419 	if (num_tc  > dpaa2_eth_tc_count(priv)) {
2420 		netdev_err(net_dev, "Max %d traffic classes supported\n",
2421 			   dpaa2_eth_tc_count(priv));
2422 		return -EOPNOTSUPP;
2423 	}
2424 
2425 	if (!num_tc) {
2426 		netdev_reset_tc(net_dev);
2427 		netif_set_real_num_tx_queues(net_dev, num_queues);
2428 		goto out;
2429 	}
2430 
2431 	netdev_set_num_tc(net_dev, num_tc);
2432 	netif_set_real_num_tx_queues(net_dev, num_tc * num_queues);
2433 
2434 	for (i = 0; i < num_tc; i++)
2435 		netdev_set_tc_queue(net_dev, i, num_queues, i * num_queues);
2436 
2437 out:
2438 	update_xps(priv);
2439 
2440 	return 0;
2441 }
2442 
2443 #define bps_to_mbits(rate) (div_u64((rate), 1000000) * 8)
2444 
2445 static int dpaa2_eth_setup_tbf(struct net_device *net_dev, struct tc_tbf_qopt_offload *p)
2446 {
2447 	struct tc_tbf_qopt_offload_replace_params *cfg = &p->replace_params;
2448 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2449 	struct dpni_tx_shaping_cfg tx_cr_shaper = { 0 };
2450 	struct dpni_tx_shaping_cfg tx_er_shaper = { 0 };
2451 	int err;
2452 
2453 	if (p->command == TC_TBF_STATS)
2454 		return -EOPNOTSUPP;
2455 
2456 	/* Only per port Tx shaping */
2457 	if (p->parent != TC_H_ROOT)
2458 		return -EOPNOTSUPP;
2459 
2460 	if (p->command == TC_TBF_REPLACE) {
2461 		if (cfg->max_size > DPAA2_ETH_MAX_BURST_SIZE) {
2462 			netdev_err(net_dev, "burst size cannot be greater than %d\n",
2463 				   DPAA2_ETH_MAX_BURST_SIZE);
2464 			return -EINVAL;
2465 		}
2466 
2467 		tx_cr_shaper.max_burst_size = cfg->max_size;
2468 		/* The TBF interface is in bytes/s, whereas DPAA2 expects the
2469 		 * rate in Mbits/s
2470 		 */
2471 		tx_cr_shaper.rate_limit = bps_to_mbits(cfg->rate.rate_bytes_ps);
2472 	}
2473 
2474 	err = dpni_set_tx_shaping(priv->mc_io, 0, priv->mc_token, &tx_cr_shaper,
2475 				  &tx_er_shaper, 0);
2476 	if (err) {
2477 		netdev_err(net_dev, "dpni_set_tx_shaping() = %d\n", err);
2478 		return err;
2479 	}
2480 
2481 	return 0;
2482 }
2483 
2484 static int dpaa2_eth_setup_tc(struct net_device *net_dev,
2485 			      enum tc_setup_type type, void *type_data)
2486 {
2487 	switch (type) {
2488 	case TC_SETUP_QDISC_MQPRIO:
2489 		return dpaa2_eth_setup_mqprio(net_dev, type_data);
2490 	case TC_SETUP_QDISC_TBF:
2491 		return dpaa2_eth_setup_tbf(net_dev, type_data);
2492 	default:
2493 		return -EOPNOTSUPP;
2494 	}
2495 }
2496 
2497 static const struct net_device_ops dpaa2_eth_ops = {
2498 	.ndo_open = dpaa2_eth_open,
2499 	.ndo_start_xmit = dpaa2_eth_tx,
2500 	.ndo_stop = dpaa2_eth_stop,
2501 	.ndo_set_mac_address = dpaa2_eth_set_addr,
2502 	.ndo_get_stats64 = dpaa2_eth_get_stats,
2503 	.ndo_set_rx_mode = dpaa2_eth_set_rx_mode,
2504 	.ndo_set_features = dpaa2_eth_set_features,
2505 	.ndo_do_ioctl = dpaa2_eth_ioctl,
2506 	.ndo_change_mtu = dpaa2_eth_change_mtu,
2507 	.ndo_bpf = dpaa2_eth_xdp,
2508 	.ndo_xdp_xmit = dpaa2_eth_xdp_xmit,
2509 	.ndo_setup_tc = dpaa2_eth_setup_tc,
2510 };
2511 
2512 static void dpaa2_eth_cdan_cb(struct dpaa2_io_notification_ctx *ctx)
2513 {
2514 	struct dpaa2_eth_channel *ch;
2515 
2516 	ch = container_of(ctx, struct dpaa2_eth_channel, nctx);
2517 
2518 	/* Update NAPI statistics */
2519 	ch->stats.cdan++;
2520 
2521 	napi_schedule(&ch->napi);
2522 }
2523 
2524 /* Allocate and configure a DPCON object */
2525 static struct fsl_mc_device *dpaa2_eth_setup_dpcon(struct dpaa2_eth_priv *priv)
2526 {
2527 	struct fsl_mc_device *dpcon;
2528 	struct device *dev = priv->net_dev->dev.parent;
2529 	int err;
2530 
2531 	err = fsl_mc_object_allocate(to_fsl_mc_device(dev),
2532 				     FSL_MC_POOL_DPCON, &dpcon);
2533 	if (err) {
2534 		if (err == -ENXIO)
2535 			err = -EPROBE_DEFER;
2536 		else
2537 			dev_info(dev, "Not enough DPCONs, will go on as-is\n");
2538 		return ERR_PTR(err);
2539 	}
2540 
2541 	err = dpcon_open(priv->mc_io, 0, dpcon->obj_desc.id, &dpcon->mc_handle);
2542 	if (err) {
2543 		dev_err(dev, "dpcon_open() failed\n");
2544 		goto free;
2545 	}
2546 
2547 	err = dpcon_reset(priv->mc_io, 0, dpcon->mc_handle);
2548 	if (err) {
2549 		dev_err(dev, "dpcon_reset() failed\n");
2550 		goto close;
2551 	}
2552 
2553 	err = dpcon_enable(priv->mc_io, 0, dpcon->mc_handle);
2554 	if (err) {
2555 		dev_err(dev, "dpcon_enable() failed\n");
2556 		goto close;
2557 	}
2558 
2559 	return dpcon;
2560 
2561 close:
2562 	dpcon_close(priv->mc_io, 0, dpcon->mc_handle);
2563 free:
2564 	fsl_mc_object_free(dpcon);
2565 
2566 	return ERR_PTR(err);
2567 }
2568 
2569 static void dpaa2_eth_free_dpcon(struct dpaa2_eth_priv *priv,
2570 				 struct fsl_mc_device *dpcon)
2571 {
2572 	dpcon_disable(priv->mc_io, 0, dpcon->mc_handle);
2573 	dpcon_close(priv->mc_io, 0, dpcon->mc_handle);
2574 	fsl_mc_object_free(dpcon);
2575 }
2576 
2577 static struct dpaa2_eth_channel *dpaa2_eth_alloc_channel(struct dpaa2_eth_priv *priv)
2578 {
2579 	struct dpaa2_eth_channel *channel;
2580 	struct dpcon_attr attr;
2581 	struct device *dev = priv->net_dev->dev.parent;
2582 	int err;
2583 
2584 	channel = kzalloc(sizeof(*channel), GFP_KERNEL);
2585 	if (!channel)
2586 		return NULL;
2587 
2588 	channel->dpcon = dpaa2_eth_setup_dpcon(priv);
2589 	if (IS_ERR(channel->dpcon)) {
2590 		err = PTR_ERR(channel->dpcon);
2591 		goto err_setup;
2592 	}
2593 
2594 	err = dpcon_get_attributes(priv->mc_io, 0, channel->dpcon->mc_handle,
2595 				   &attr);
2596 	if (err) {
2597 		dev_err(dev, "dpcon_get_attributes() failed\n");
2598 		goto err_get_attr;
2599 	}
2600 
2601 	channel->dpcon_id = attr.id;
2602 	channel->ch_id = attr.qbman_ch_id;
2603 	channel->priv = priv;
2604 
2605 	return channel;
2606 
2607 err_get_attr:
2608 	dpaa2_eth_free_dpcon(priv, channel->dpcon);
2609 err_setup:
2610 	kfree(channel);
2611 	return ERR_PTR(err);
2612 }
2613 
2614 static void dpaa2_eth_free_channel(struct dpaa2_eth_priv *priv,
2615 				   struct dpaa2_eth_channel *channel)
2616 {
2617 	dpaa2_eth_free_dpcon(priv, channel->dpcon);
2618 	kfree(channel);
2619 }
2620 
2621 /* DPIO setup: allocate and configure QBMan channels, setup core affinity
2622  * and register data availability notifications
2623  */
2624 static int dpaa2_eth_setup_dpio(struct dpaa2_eth_priv *priv)
2625 {
2626 	struct dpaa2_io_notification_ctx *nctx;
2627 	struct dpaa2_eth_channel *channel;
2628 	struct dpcon_notification_cfg dpcon_notif_cfg;
2629 	struct device *dev = priv->net_dev->dev.parent;
2630 	int i, err;
2631 
2632 	/* We want the ability to spread ingress traffic (RX, TX conf) to as
2633 	 * many cores as possible, so we need one channel for each core
2634 	 * (unless there's fewer queues than cores, in which case the extra
2635 	 * channels would be wasted).
2636 	 * Allocate one channel per core and register it to the core's
2637 	 * affine DPIO. If not enough channels are available for all cores
2638 	 * or if some cores don't have an affine DPIO, there will be no
2639 	 * ingress frame processing on those cores.
2640 	 */
2641 	cpumask_clear(&priv->dpio_cpumask);
2642 	for_each_online_cpu(i) {
2643 		/* Try to allocate a channel */
2644 		channel = dpaa2_eth_alloc_channel(priv);
2645 		if (IS_ERR_OR_NULL(channel)) {
2646 			err = PTR_ERR_OR_ZERO(channel);
2647 			if (err != -EPROBE_DEFER)
2648 				dev_info(dev,
2649 					 "No affine channel for cpu %d and above\n", i);
2650 			goto err_alloc_ch;
2651 		}
2652 
2653 		priv->channel[priv->num_channels] = channel;
2654 
2655 		nctx = &channel->nctx;
2656 		nctx->is_cdan = 1;
2657 		nctx->cb = dpaa2_eth_cdan_cb;
2658 		nctx->id = channel->ch_id;
2659 		nctx->desired_cpu = i;
2660 
2661 		/* Register the new context */
2662 		channel->dpio = dpaa2_io_service_select(i);
2663 		err = dpaa2_io_service_register(channel->dpio, nctx, dev);
2664 		if (err) {
2665 			dev_dbg(dev, "No affine DPIO for cpu %d\n", i);
2666 			/* If no affine DPIO for this core, there's probably
2667 			 * none available for next cores either. Signal we want
2668 			 * to retry later, in case the DPIO devices weren't
2669 			 * probed yet.
2670 			 */
2671 			err = -EPROBE_DEFER;
2672 			goto err_service_reg;
2673 		}
2674 
2675 		/* Register DPCON notification with MC */
2676 		dpcon_notif_cfg.dpio_id = nctx->dpio_id;
2677 		dpcon_notif_cfg.priority = 0;
2678 		dpcon_notif_cfg.user_ctx = nctx->qman64;
2679 		err = dpcon_set_notification(priv->mc_io, 0,
2680 					     channel->dpcon->mc_handle,
2681 					     &dpcon_notif_cfg);
2682 		if (err) {
2683 			dev_err(dev, "dpcon_set_notification failed()\n");
2684 			goto err_set_cdan;
2685 		}
2686 
2687 		/* If we managed to allocate a channel and also found an affine
2688 		 * DPIO for this core, add it to the final mask
2689 		 */
2690 		cpumask_set_cpu(i, &priv->dpio_cpumask);
2691 		priv->num_channels++;
2692 
2693 		/* Stop if we already have enough channels to accommodate all
2694 		 * RX and TX conf queues
2695 		 */
2696 		if (priv->num_channels == priv->dpni_attrs.num_queues)
2697 			break;
2698 	}
2699 
2700 	return 0;
2701 
2702 err_set_cdan:
2703 	dpaa2_io_service_deregister(channel->dpio, nctx, dev);
2704 err_service_reg:
2705 	dpaa2_eth_free_channel(priv, channel);
2706 err_alloc_ch:
2707 	if (err == -EPROBE_DEFER) {
2708 		for (i = 0; i < priv->num_channels; i++) {
2709 			channel = priv->channel[i];
2710 			nctx = &channel->nctx;
2711 			dpaa2_io_service_deregister(channel->dpio, nctx, dev);
2712 			dpaa2_eth_free_channel(priv, channel);
2713 		}
2714 		priv->num_channels = 0;
2715 		return err;
2716 	}
2717 
2718 	if (cpumask_empty(&priv->dpio_cpumask)) {
2719 		dev_err(dev, "No cpu with an affine DPIO/DPCON\n");
2720 		return -ENODEV;
2721 	}
2722 
2723 	dev_info(dev, "Cores %*pbl available for processing ingress traffic\n",
2724 		 cpumask_pr_args(&priv->dpio_cpumask));
2725 
2726 	return 0;
2727 }
2728 
2729 static void dpaa2_eth_free_dpio(struct dpaa2_eth_priv *priv)
2730 {
2731 	struct device *dev = priv->net_dev->dev.parent;
2732 	struct dpaa2_eth_channel *ch;
2733 	int i;
2734 
2735 	/* deregister CDAN notifications and free channels */
2736 	for (i = 0; i < priv->num_channels; i++) {
2737 		ch = priv->channel[i];
2738 		dpaa2_io_service_deregister(ch->dpio, &ch->nctx, dev);
2739 		dpaa2_eth_free_channel(priv, ch);
2740 	}
2741 }
2742 
2743 static struct dpaa2_eth_channel *dpaa2_eth_get_affine_channel(struct dpaa2_eth_priv *priv,
2744 							      int cpu)
2745 {
2746 	struct device *dev = priv->net_dev->dev.parent;
2747 	int i;
2748 
2749 	for (i = 0; i < priv->num_channels; i++)
2750 		if (priv->channel[i]->nctx.desired_cpu == cpu)
2751 			return priv->channel[i];
2752 
2753 	/* We should never get here. Issue a warning and return
2754 	 * the first channel, because it's still better than nothing
2755 	 */
2756 	dev_warn(dev, "No affine channel found for cpu %d\n", cpu);
2757 
2758 	return priv->channel[0];
2759 }
2760 
2761 static void dpaa2_eth_set_fq_affinity(struct dpaa2_eth_priv *priv)
2762 {
2763 	struct device *dev = priv->net_dev->dev.parent;
2764 	struct dpaa2_eth_fq *fq;
2765 	int rx_cpu, txc_cpu;
2766 	int i;
2767 
2768 	/* For each FQ, pick one channel/CPU to deliver frames to.
2769 	 * This may well change at runtime, either through irqbalance or
2770 	 * through direct user intervention.
2771 	 */
2772 	rx_cpu = txc_cpu = cpumask_first(&priv->dpio_cpumask);
2773 
2774 	for (i = 0; i < priv->num_fqs; i++) {
2775 		fq = &priv->fq[i];
2776 		switch (fq->type) {
2777 		case DPAA2_RX_FQ:
2778 		case DPAA2_RX_ERR_FQ:
2779 			fq->target_cpu = rx_cpu;
2780 			rx_cpu = cpumask_next(rx_cpu, &priv->dpio_cpumask);
2781 			if (rx_cpu >= nr_cpu_ids)
2782 				rx_cpu = cpumask_first(&priv->dpio_cpumask);
2783 			break;
2784 		case DPAA2_TX_CONF_FQ:
2785 			fq->target_cpu = txc_cpu;
2786 			txc_cpu = cpumask_next(txc_cpu, &priv->dpio_cpumask);
2787 			if (txc_cpu >= nr_cpu_ids)
2788 				txc_cpu = cpumask_first(&priv->dpio_cpumask);
2789 			break;
2790 		default:
2791 			dev_err(dev, "Unknown FQ type: %d\n", fq->type);
2792 		}
2793 		fq->channel = dpaa2_eth_get_affine_channel(priv, fq->target_cpu);
2794 	}
2795 
2796 	update_xps(priv);
2797 }
2798 
2799 static void dpaa2_eth_setup_fqs(struct dpaa2_eth_priv *priv)
2800 {
2801 	int i, j;
2802 
2803 	/* We have one TxConf FQ per Tx flow.
2804 	 * The number of Tx and Rx queues is the same.
2805 	 * Tx queues come first in the fq array.
2806 	 */
2807 	for (i = 0; i < dpaa2_eth_queue_count(priv); i++) {
2808 		priv->fq[priv->num_fqs].type = DPAA2_TX_CONF_FQ;
2809 		priv->fq[priv->num_fqs].consume = dpaa2_eth_tx_conf;
2810 		priv->fq[priv->num_fqs++].flowid = (u16)i;
2811 	}
2812 
2813 	for (j = 0; j < dpaa2_eth_tc_count(priv); j++) {
2814 		for (i = 0; i < dpaa2_eth_queue_count(priv); i++) {
2815 			priv->fq[priv->num_fqs].type = DPAA2_RX_FQ;
2816 			priv->fq[priv->num_fqs].consume = dpaa2_eth_rx;
2817 			priv->fq[priv->num_fqs].tc = (u8)j;
2818 			priv->fq[priv->num_fqs++].flowid = (u16)i;
2819 		}
2820 	}
2821 
2822 	/* We have exactly one Rx error queue per DPNI */
2823 	priv->fq[priv->num_fqs].type = DPAA2_RX_ERR_FQ;
2824 	priv->fq[priv->num_fqs++].consume = dpaa2_eth_rx_err;
2825 
2826 	/* For each FQ, decide on which core to process incoming frames */
2827 	dpaa2_eth_set_fq_affinity(priv);
2828 }
2829 
2830 /* Allocate and configure one buffer pool for each interface */
2831 static int dpaa2_eth_setup_dpbp(struct dpaa2_eth_priv *priv)
2832 {
2833 	int err;
2834 	struct fsl_mc_device *dpbp_dev;
2835 	struct device *dev = priv->net_dev->dev.parent;
2836 	struct dpbp_attr dpbp_attrs;
2837 
2838 	err = fsl_mc_object_allocate(to_fsl_mc_device(dev), FSL_MC_POOL_DPBP,
2839 				     &dpbp_dev);
2840 	if (err) {
2841 		if (err == -ENXIO)
2842 			err = -EPROBE_DEFER;
2843 		else
2844 			dev_err(dev, "DPBP device allocation failed\n");
2845 		return err;
2846 	}
2847 
2848 	priv->dpbp_dev = dpbp_dev;
2849 
2850 	err = dpbp_open(priv->mc_io, 0, priv->dpbp_dev->obj_desc.id,
2851 			&dpbp_dev->mc_handle);
2852 	if (err) {
2853 		dev_err(dev, "dpbp_open() failed\n");
2854 		goto err_open;
2855 	}
2856 
2857 	err = dpbp_reset(priv->mc_io, 0, dpbp_dev->mc_handle);
2858 	if (err) {
2859 		dev_err(dev, "dpbp_reset() failed\n");
2860 		goto err_reset;
2861 	}
2862 
2863 	err = dpbp_enable(priv->mc_io, 0, dpbp_dev->mc_handle);
2864 	if (err) {
2865 		dev_err(dev, "dpbp_enable() failed\n");
2866 		goto err_enable;
2867 	}
2868 
2869 	err = dpbp_get_attributes(priv->mc_io, 0, dpbp_dev->mc_handle,
2870 				  &dpbp_attrs);
2871 	if (err) {
2872 		dev_err(dev, "dpbp_get_attributes() failed\n");
2873 		goto err_get_attr;
2874 	}
2875 	priv->bpid = dpbp_attrs.bpid;
2876 
2877 	return 0;
2878 
2879 err_get_attr:
2880 	dpbp_disable(priv->mc_io, 0, dpbp_dev->mc_handle);
2881 err_enable:
2882 err_reset:
2883 	dpbp_close(priv->mc_io, 0, dpbp_dev->mc_handle);
2884 err_open:
2885 	fsl_mc_object_free(dpbp_dev);
2886 
2887 	return err;
2888 }
2889 
2890 static void dpaa2_eth_free_dpbp(struct dpaa2_eth_priv *priv)
2891 {
2892 	dpaa2_eth_drain_pool(priv);
2893 	dpbp_disable(priv->mc_io, 0, priv->dpbp_dev->mc_handle);
2894 	dpbp_close(priv->mc_io, 0, priv->dpbp_dev->mc_handle);
2895 	fsl_mc_object_free(priv->dpbp_dev);
2896 }
2897 
2898 static int dpaa2_eth_set_buffer_layout(struct dpaa2_eth_priv *priv)
2899 {
2900 	struct device *dev = priv->net_dev->dev.parent;
2901 	struct dpni_buffer_layout buf_layout = {0};
2902 	u16 rx_buf_align;
2903 	int err;
2904 
2905 	/* We need to check for WRIOP version 1.0.0, but depending on the MC
2906 	 * version, this number is not always provided correctly on rev1.
2907 	 * We need to check for both alternatives in this situation.
2908 	 */
2909 	if (priv->dpni_attrs.wriop_version == DPAA2_WRIOP_VERSION(0, 0, 0) ||
2910 	    priv->dpni_attrs.wriop_version == DPAA2_WRIOP_VERSION(1, 0, 0))
2911 		rx_buf_align = DPAA2_ETH_RX_BUF_ALIGN_REV1;
2912 	else
2913 		rx_buf_align = DPAA2_ETH_RX_BUF_ALIGN;
2914 
2915 	/* We need to ensure that the buffer size seen by WRIOP is a multiple
2916 	 * of 64 or 256 bytes depending on the WRIOP version.
2917 	 */
2918 	priv->rx_buf_size = ALIGN_DOWN(DPAA2_ETH_RX_BUF_SIZE, rx_buf_align);
2919 
2920 	/* tx buffer */
2921 	buf_layout.private_data_size = DPAA2_ETH_SWA_SIZE;
2922 	buf_layout.pass_timestamp = true;
2923 	buf_layout.pass_frame_status = true;
2924 	buf_layout.options = DPNI_BUF_LAYOUT_OPT_PRIVATE_DATA_SIZE |
2925 			     DPNI_BUF_LAYOUT_OPT_TIMESTAMP |
2926 			     DPNI_BUF_LAYOUT_OPT_FRAME_STATUS;
2927 	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
2928 				     DPNI_QUEUE_TX, &buf_layout);
2929 	if (err) {
2930 		dev_err(dev, "dpni_set_buffer_layout(TX) failed\n");
2931 		return err;
2932 	}
2933 
2934 	/* tx-confirm buffer */
2935 	buf_layout.options = DPNI_BUF_LAYOUT_OPT_TIMESTAMP |
2936 			     DPNI_BUF_LAYOUT_OPT_FRAME_STATUS;
2937 	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
2938 				     DPNI_QUEUE_TX_CONFIRM, &buf_layout);
2939 	if (err) {
2940 		dev_err(dev, "dpni_set_buffer_layout(TX_CONF) failed\n");
2941 		return err;
2942 	}
2943 
2944 	/* Now that we've set our tx buffer layout, retrieve the minimum
2945 	 * required tx data offset.
2946 	 */
2947 	err = dpni_get_tx_data_offset(priv->mc_io, 0, priv->mc_token,
2948 				      &priv->tx_data_offset);
2949 	if (err) {
2950 		dev_err(dev, "dpni_get_tx_data_offset() failed\n");
2951 		return err;
2952 	}
2953 
2954 	if ((priv->tx_data_offset % 64) != 0)
2955 		dev_warn(dev, "Tx data offset (%d) not a multiple of 64B\n",
2956 			 priv->tx_data_offset);
2957 
2958 	/* rx buffer */
2959 	buf_layout.pass_frame_status = true;
2960 	buf_layout.pass_parser_result = true;
2961 	buf_layout.data_align = rx_buf_align;
2962 	buf_layout.data_head_room = dpaa2_eth_rx_head_room(priv);
2963 	buf_layout.private_data_size = 0;
2964 	buf_layout.options = DPNI_BUF_LAYOUT_OPT_PARSER_RESULT |
2965 			     DPNI_BUF_LAYOUT_OPT_FRAME_STATUS |
2966 			     DPNI_BUF_LAYOUT_OPT_DATA_ALIGN |
2967 			     DPNI_BUF_LAYOUT_OPT_DATA_HEAD_ROOM |
2968 			     DPNI_BUF_LAYOUT_OPT_TIMESTAMP;
2969 	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
2970 				     DPNI_QUEUE_RX, &buf_layout);
2971 	if (err) {
2972 		dev_err(dev, "dpni_set_buffer_layout(RX) failed\n");
2973 		return err;
2974 	}
2975 
2976 	return 0;
2977 }
2978 
2979 #define DPNI_ENQUEUE_FQID_VER_MAJOR	7
2980 #define DPNI_ENQUEUE_FQID_VER_MINOR	9
2981 
2982 static inline int dpaa2_eth_enqueue_qd(struct dpaa2_eth_priv *priv,
2983 				       struct dpaa2_eth_fq *fq,
2984 				       struct dpaa2_fd *fd, u8 prio,
2985 				       u32 num_frames __always_unused,
2986 				       int *frames_enqueued)
2987 {
2988 	int err;
2989 
2990 	err = dpaa2_io_service_enqueue_qd(fq->channel->dpio,
2991 					  priv->tx_qdid, prio,
2992 					  fq->tx_qdbin, fd);
2993 	if (!err && frames_enqueued)
2994 		*frames_enqueued = 1;
2995 	return err;
2996 }
2997 
2998 static inline int dpaa2_eth_enqueue_fq_multiple(struct dpaa2_eth_priv *priv,
2999 						struct dpaa2_eth_fq *fq,
3000 						struct dpaa2_fd *fd,
3001 						u8 prio, u32 num_frames,
3002 						int *frames_enqueued)
3003 {
3004 	int err;
3005 
3006 	err = dpaa2_io_service_enqueue_multiple_fq(fq->channel->dpio,
3007 						   fq->tx_fqid[prio],
3008 						   fd, num_frames);
3009 
3010 	if (err == 0)
3011 		return -EBUSY;
3012 
3013 	if (frames_enqueued)
3014 		*frames_enqueued = err;
3015 	return 0;
3016 }
3017 
3018 static void dpaa2_eth_set_enqueue_mode(struct dpaa2_eth_priv *priv)
3019 {
3020 	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_ENQUEUE_FQID_VER_MAJOR,
3021 				   DPNI_ENQUEUE_FQID_VER_MINOR) < 0)
3022 		priv->enqueue = dpaa2_eth_enqueue_qd;
3023 	else
3024 		priv->enqueue = dpaa2_eth_enqueue_fq_multiple;
3025 }
3026 
3027 static int dpaa2_eth_set_pause(struct dpaa2_eth_priv *priv)
3028 {
3029 	struct device *dev = priv->net_dev->dev.parent;
3030 	struct dpni_link_cfg link_cfg = {0};
3031 	int err;
3032 
3033 	/* Get the default link options so we don't override other flags */
3034 	err = dpni_get_link_cfg(priv->mc_io, 0, priv->mc_token, &link_cfg);
3035 	if (err) {
3036 		dev_err(dev, "dpni_get_link_cfg() failed\n");
3037 		return err;
3038 	}
3039 
3040 	/* By default, enable both Rx and Tx pause frames */
3041 	link_cfg.options |= DPNI_LINK_OPT_PAUSE;
3042 	link_cfg.options &= ~DPNI_LINK_OPT_ASYM_PAUSE;
3043 	err = dpni_set_link_cfg(priv->mc_io, 0, priv->mc_token, &link_cfg);
3044 	if (err) {
3045 		dev_err(dev, "dpni_set_link_cfg() failed\n");
3046 		return err;
3047 	}
3048 
3049 	priv->link_state.options = link_cfg.options;
3050 
3051 	return 0;
3052 }
3053 
3054 static void dpaa2_eth_update_tx_fqids(struct dpaa2_eth_priv *priv)
3055 {
3056 	struct dpni_queue_id qid = {0};
3057 	struct dpaa2_eth_fq *fq;
3058 	struct dpni_queue queue;
3059 	int i, j, err;
3060 
3061 	/* We only use Tx FQIDs for FQID-based enqueue, so check
3062 	 * if DPNI version supports it before updating FQIDs
3063 	 */
3064 	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_ENQUEUE_FQID_VER_MAJOR,
3065 				   DPNI_ENQUEUE_FQID_VER_MINOR) < 0)
3066 		return;
3067 
3068 	for (i = 0; i < priv->num_fqs; i++) {
3069 		fq = &priv->fq[i];
3070 		if (fq->type != DPAA2_TX_CONF_FQ)
3071 			continue;
3072 		for (j = 0; j < dpaa2_eth_tc_count(priv); j++) {
3073 			err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3074 					     DPNI_QUEUE_TX, j, fq->flowid,
3075 					     &queue, &qid);
3076 			if (err)
3077 				goto out_err;
3078 
3079 			fq->tx_fqid[j] = qid.fqid;
3080 			if (fq->tx_fqid[j] == 0)
3081 				goto out_err;
3082 		}
3083 	}
3084 
3085 	priv->enqueue = dpaa2_eth_enqueue_fq_multiple;
3086 
3087 	return;
3088 
3089 out_err:
3090 	netdev_info(priv->net_dev,
3091 		    "Error reading Tx FQID, fallback to QDID-based enqueue\n");
3092 	priv->enqueue = dpaa2_eth_enqueue_qd;
3093 }
3094 
3095 /* Configure ingress classification based on VLAN PCP */
3096 static int dpaa2_eth_set_vlan_qos(struct dpaa2_eth_priv *priv)
3097 {
3098 	struct device *dev = priv->net_dev->dev.parent;
3099 	struct dpkg_profile_cfg kg_cfg = {0};
3100 	struct dpni_qos_tbl_cfg qos_cfg = {0};
3101 	struct dpni_rule_cfg key_params;
3102 	void *dma_mem, *key, *mask;
3103 	u8 key_size = 2;	/* VLAN TCI field */
3104 	int i, pcp, err;
3105 
3106 	/* VLAN-based classification only makes sense if we have multiple
3107 	 * traffic classes.
3108 	 * Also, we need to extract just the 3-bit PCP field from the VLAN
3109 	 * header and we can only do that by using a mask
3110 	 */
3111 	if (dpaa2_eth_tc_count(priv) == 1 || !dpaa2_eth_fs_mask_enabled(priv)) {
3112 		dev_dbg(dev, "VLAN-based QoS classification not supported\n");
3113 		return -EOPNOTSUPP;
3114 	}
3115 
3116 	dma_mem = kzalloc(DPAA2_CLASSIFIER_DMA_SIZE, GFP_KERNEL);
3117 	if (!dma_mem)
3118 		return -ENOMEM;
3119 
3120 	kg_cfg.num_extracts = 1;
3121 	kg_cfg.extracts[0].type = DPKG_EXTRACT_FROM_HDR;
3122 	kg_cfg.extracts[0].extract.from_hdr.prot = NET_PROT_VLAN;
3123 	kg_cfg.extracts[0].extract.from_hdr.type = DPKG_FULL_FIELD;
3124 	kg_cfg.extracts[0].extract.from_hdr.field = NH_FLD_VLAN_TCI;
3125 
3126 	err = dpni_prepare_key_cfg(&kg_cfg, dma_mem);
3127 	if (err) {
3128 		dev_err(dev, "dpni_prepare_key_cfg failed\n");
3129 		goto out_free_tbl;
3130 	}
3131 
3132 	/* set QoS table */
3133 	qos_cfg.default_tc = 0;
3134 	qos_cfg.discard_on_miss = 0;
3135 	qos_cfg.key_cfg_iova = dma_map_single(dev, dma_mem,
3136 					      DPAA2_CLASSIFIER_DMA_SIZE,
3137 					      DMA_TO_DEVICE);
3138 	if (dma_mapping_error(dev, qos_cfg.key_cfg_iova)) {
3139 		dev_err(dev, "QoS table DMA mapping failed\n");
3140 		err = -ENOMEM;
3141 		goto out_free_tbl;
3142 	}
3143 
3144 	err = dpni_set_qos_table(priv->mc_io, 0, priv->mc_token, &qos_cfg);
3145 	if (err) {
3146 		dev_err(dev, "dpni_set_qos_table failed\n");
3147 		goto out_unmap_tbl;
3148 	}
3149 
3150 	/* Add QoS table entries */
3151 	key = kzalloc(key_size * 2, GFP_KERNEL);
3152 	if (!key) {
3153 		err = -ENOMEM;
3154 		goto out_unmap_tbl;
3155 	}
3156 	mask = key + key_size;
3157 	*(__be16 *)mask = cpu_to_be16(VLAN_PRIO_MASK);
3158 
3159 	key_params.key_iova = dma_map_single(dev, key, key_size * 2,
3160 					     DMA_TO_DEVICE);
3161 	if (dma_mapping_error(dev, key_params.key_iova)) {
3162 		dev_err(dev, "Qos table entry DMA mapping failed\n");
3163 		err = -ENOMEM;
3164 		goto out_free_key;
3165 	}
3166 
3167 	key_params.mask_iova = key_params.key_iova + key_size;
3168 	key_params.key_size = key_size;
3169 
3170 	/* We add rules for PCP-based distribution starting with highest
3171 	 * priority (VLAN PCP = 7). If this DPNI doesn't have enough traffic
3172 	 * classes to accommodate all priority levels, the lowest ones end up
3173 	 * on TC 0 which was configured as default
3174 	 */
3175 	for (i = dpaa2_eth_tc_count(priv) - 1, pcp = 7; i >= 0; i--, pcp--) {
3176 		*(__be16 *)key = cpu_to_be16(pcp << VLAN_PRIO_SHIFT);
3177 		dma_sync_single_for_device(dev, key_params.key_iova,
3178 					   key_size * 2, DMA_TO_DEVICE);
3179 
3180 		err = dpni_add_qos_entry(priv->mc_io, 0, priv->mc_token,
3181 					 &key_params, i, i);
3182 		if (err) {
3183 			dev_err(dev, "dpni_add_qos_entry failed\n");
3184 			dpni_clear_qos_table(priv->mc_io, 0, priv->mc_token);
3185 			goto out_unmap_key;
3186 		}
3187 	}
3188 
3189 	priv->vlan_cls_enabled = true;
3190 
3191 	/* Table and key memory is not persistent, clean everything up after
3192 	 * configuration is finished
3193 	 */
3194 out_unmap_key:
3195 	dma_unmap_single(dev, key_params.key_iova, key_size * 2, DMA_TO_DEVICE);
3196 out_free_key:
3197 	kfree(key);
3198 out_unmap_tbl:
3199 	dma_unmap_single(dev, qos_cfg.key_cfg_iova, DPAA2_CLASSIFIER_DMA_SIZE,
3200 			 DMA_TO_DEVICE);
3201 out_free_tbl:
3202 	kfree(dma_mem);
3203 
3204 	return err;
3205 }
3206 
3207 /* Configure the DPNI object this interface is associated with */
3208 static int dpaa2_eth_setup_dpni(struct fsl_mc_device *ls_dev)
3209 {
3210 	struct device *dev = &ls_dev->dev;
3211 	struct dpaa2_eth_priv *priv;
3212 	struct net_device *net_dev;
3213 	int err;
3214 
3215 	net_dev = dev_get_drvdata(dev);
3216 	priv = netdev_priv(net_dev);
3217 
3218 	/* get a handle for the DPNI object */
3219 	err = dpni_open(priv->mc_io, 0, ls_dev->obj_desc.id, &priv->mc_token);
3220 	if (err) {
3221 		dev_err(dev, "dpni_open() failed\n");
3222 		return err;
3223 	}
3224 
3225 	/* Check if we can work with this DPNI object */
3226 	err = dpni_get_api_version(priv->mc_io, 0, &priv->dpni_ver_major,
3227 				   &priv->dpni_ver_minor);
3228 	if (err) {
3229 		dev_err(dev, "dpni_get_api_version() failed\n");
3230 		goto close;
3231 	}
3232 	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_VER_MAJOR, DPNI_VER_MINOR) < 0) {
3233 		dev_err(dev, "DPNI version %u.%u not supported, need >= %u.%u\n",
3234 			priv->dpni_ver_major, priv->dpni_ver_minor,
3235 			DPNI_VER_MAJOR, DPNI_VER_MINOR);
3236 		err = -ENOTSUPP;
3237 		goto close;
3238 	}
3239 
3240 	ls_dev->mc_io = priv->mc_io;
3241 	ls_dev->mc_handle = priv->mc_token;
3242 
3243 	err = dpni_reset(priv->mc_io, 0, priv->mc_token);
3244 	if (err) {
3245 		dev_err(dev, "dpni_reset() failed\n");
3246 		goto close;
3247 	}
3248 
3249 	err = dpni_get_attributes(priv->mc_io, 0, priv->mc_token,
3250 				  &priv->dpni_attrs);
3251 	if (err) {
3252 		dev_err(dev, "dpni_get_attributes() failed (err=%d)\n", err);
3253 		goto close;
3254 	}
3255 
3256 	err = dpaa2_eth_set_buffer_layout(priv);
3257 	if (err)
3258 		goto close;
3259 
3260 	dpaa2_eth_set_enqueue_mode(priv);
3261 
3262 	/* Enable pause frame support */
3263 	if (dpaa2_eth_has_pause_support(priv)) {
3264 		err = dpaa2_eth_set_pause(priv);
3265 		if (err)
3266 			goto close;
3267 	}
3268 
3269 	err = dpaa2_eth_set_vlan_qos(priv);
3270 	if (err && err != -EOPNOTSUPP)
3271 		goto close;
3272 
3273 	priv->cls_rules = devm_kcalloc(dev, dpaa2_eth_fs_count(priv),
3274 				       sizeof(struct dpaa2_eth_cls_rule),
3275 				       GFP_KERNEL);
3276 	if (!priv->cls_rules) {
3277 		err = -ENOMEM;
3278 		goto close;
3279 	}
3280 
3281 	return 0;
3282 
3283 close:
3284 	dpni_close(priv->mc_io, 0, priv->mc_token);
3285 
3286 	return err;
3287 }
3288 
3289 static void dpaa2_eth_free_dpni(struct dpaa2_eth_priv *priv)
3290 {
3291 	int err;
3292 
3293 	err = dpni_reset(priv->mc_io, 0, priv->mc_token);
3294 	if (err)
3295 		netdev_warn(priv->net_dev, "dpni_reset() failed (err %d)\n",
3296 			    err);
3297 
3298 	dpni_close(priv->mc_io, 0, priv->mc_token);
3299 }
3300 
3301 static int dpaa2_eth_setup_rx_flow(struct dpaa2_eth_priv *priv,
3302 				   struct dpaa2_eth_fq *fq)
3303 {
3304 	struct device *dev = priv->net_dev->dev.parent;
3305 	struct dpni_queue queue;
3306 	struct dpni_queue_id qid;
3307 	int err;
3308 
3309 	err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3310 			     DPNI_QUEUE_RX, fq->tc, fq->flowid, &queue, &qid);
3311 	if (err) {
3312 		dev_err(dev, "dpni_get_queue(RX) failed\n");
3313 		return err;
3314 	}
3315 
3316 	fq->fqid = qid.fqid;
3317 
3318 	queue.destination.id = fq->channel->dpcon_id;
3319 	queue.destination.type = DPNI_DEST_DPCON;
3320 	queue.destination.priority = 1;
3321 	queue.user_context = (u64)(uintptr_t)fq;
3322 	err = dpni_set_queue(priv->mc_io, 0, priv->mc_token,
3323 			     DPNI_QUEUE_RX, fq->tc, fq->flowid,
3324 			     DPNI_QUEUE_OPT_USER_CTX | DPNI_QUEUE_OPT_DEST,
3325 			     &queue);
3326 	if (err) {
3327 		dev_err(dev, "dpni_set_queue(RX) failed\n");
3328 		return err;
3329 	}
3330 
3331 	/* xdp_rxq setup */
3332 	/* only once for each channel */
3333 	if (fq->tc > 0)
3334 		return 0;
3335 
3336 	err = xdp_rxq_info_reg(&fq->channel->xdp_rxq, priv->net_dev,
3337 			       fq->flowid, 0);
3338 	if (err) {
3339 		dev_err(dev, "xdp_rxq_info_reg failed\n");
3340 		return err;
3341 	}
3342 
3343 	err = xdp_rxq_info_reg_mem_model(&fq->channel->xdp_rxq,
3344 					 MEM_TYPE_PAGE_ORDER0, NULL);
3345 	if (err) {
3346 		dev_err(dev, "xdp_rxq_info_reg_mem_model failed\n");
3347 		return err;
3348 	}
3349 
3350 	return 0;
3351 }
3352 
3353 static int dpaa2_eth_setup_tx_flow(struct dpaa2_eth_priv *priv,
3354 				   struct dpaa2_eth_fq *fq)
3355 {
3356 	struct device *dev = priv->net_dev->dev.parent;
3357 	struct dpni_queue queue;
3358 	struct dpni_queue_id qid;
3359 	int i, err;
3360 
3361 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3362 		err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3363 				     DPNI_QUEUE_TX, i, fq->flowid,
3364 				     &queue, &qid);
3365 		if (err) {
3366 			dev_err(dev, "dpni_get_queue(TX) failed\n");
3367 			return err;
3368 		}
3369 		fq->tx_fqid[i] = qid.fqid;
3370 	}
3371 
3372 	/* All Tx queues belonging to the same flowid have the same qdbin */
3373 	fq->tx_qdbin = qid.qdbin;
3374 
3375 	err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3376 			     DPNI_QUEUE_TX_CONFIRM, 0, fq->flowid,
3377 			     &queue, &qid);
3378 	if (err) {
3379 		dev_err(dev, "dpni_get_queue(TX_CONF) failed\n");
3380 		return err;
3381 	}
3382 
3383 	fq->fqid = qid.fqid;
3384 
3385 	queue.destination.id = fq->channel->dpcon_id;
3386 	queue.destination.type = DPNI_DEST_DPCON;
3387 	queue.destination.priority = 0;
3388 	queue.user_context = (u64)(uintptr_t)fq;
3389 	err = dpni_set_queue(priv->mc_io, 0, priv->mc_token,
3390 			     DPNI_QUEUE_TX_CONFIRM, 0, fq->flowid,
3391 			     DPNI_QUEUE_OPT_USER_CTX | DPNI_QUEUE_OPT_DEST,
3392 			     &queue);
3393 	if (err) {
3394 		dev_err(dev, "dpni_set_queue(TX_CONF) failed\n");
3395 		return err;
3396 	}
3397 
3398 	return 0;
3399 }
3400 
3401 static int setup_rx_err_flow(struct dpaa2_eth_priv *priv,
3402 			     struct dpaa2_eth_fq *fq)
3403 {
3404 	struct device *dev = priv->net_dev->dev.parent;
3405 	struct dpni_queue q = { { 0 } };
3406 	struct dpni_queue_id qid;
3407 	u8 q_opt = DPNI_QUEUE_OPT_USER_CTX | DPNI_QUEUE_OPT_DEST;
3408 	int err;
3409 
3410 	err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3411 			     DPNI_QUEUE_RX_ERR, 0, 0, &q, &qid);
3412 	if (err) {
3413 		dev_err(dev, "dpni_get_queue() failed (%d)\n", err);
3414 		return err;
3415 	}
3416 
3417 	fq->fqid = qid.fqid;
3418 
3419 	q.destination.id = fq->channel->dpcon_id;
3420 	q.destination.type = DPNI_DEST_DPCON;
3421 	q.destination.priority = 1;
3422 	q.user_context = (u64)(uintptr_t)fq;
3423 	err = dpni_set_queue(priv->mc_io, 0, priv->mc_token,
3424 			     DPNI_QUEUE_RX_ERR, 0, 0, q_opt, &q);
3425 	if (err) {
3426 		dev_err(dev, "dpni_set_queue() failed (%d)\n", err);
3427 		return err;
3428 	}
3429 
3430 	return 0;
3431 }
3432 
3433 /* Supported header fields for Rx hash distribution key */
3434 static const struct dpaa2_eth_dist_fields dist_fields[] = {
3435 	{
3436 		/* L2 header */
3437 		.rxnfc_field = RXH_L2DA,
3438 		.cls_prot = NET_PROT_ETH,
3439 		.cls_field = NH_FLD_ETH_DA,
3440 		.id = DPAA2_ETH_DIST_ETHDST,
3441 		.size = 6,
3442 	}, {
3443 		.cls_prot = NET_PROT_ETH,
3444 		.cls_field = NH_FLD_ETH_SA,
3445 		.id = DPAA2_ETH_DIST_ETHSRC,
3446 		.size = 6,
3447 	}, {
3448 		/* This is the last ethertype field parsed:
3449 		 * depending on frame format, it can be the MAC ethertype
3450 		 * or the VLAN etype.
3451 		 */
3452 		.cls_prot = NET_PROT_ETH,
3453 		.cls_field = NH_FLD_ETH_TYPE,
3454 		.id = DPAA2_ETH_DIST_ETHTYPE,
3455 		.size = 2,
3456 	}, {
3457 		/* VLAN header */
3458 		.rxnfc_field = RXH_VLAN,
3459 		.cls_prot = NET_PROT_VLAN,
3460 		.cls_field = NH_FLD_VLAN_TCI,
3461 		.id = DPAA2_ETH_DIST_VLAN,
3462 		.size = 2,
3463 	}, {
3464 		/* IP header */
3465 		.rxnfc_field = RXH_IP_SRC,
3466 		.cls_prot = NET_PROT_IP,
3467 		.cls_field = NH_FLD_IP_SRC,
3468 		.id = DPAA2_ETH_DIST_IPSRC,
3469 		.size = 4,
3470 	}, {
3471 		.rxnfc_field = RXH_IP_DST,
3472 		.cls_prot = NET_PROT_IP,
3473 		.cls_field = NH_FLD_IP_DST,
3474 		.id = DPAA2_ETH_DIST_IPDST,
3475 		.size = 4,
3476 	}, {
3477 		.rxnfc_field = RXH_L3_PROTO,
3478 		.cls_prot = NET_PROT_IP,
3479 		.cls_field = NH_FLD_IP_PROTO,
3480 		.id = DPAA2_ETH_DIST_IPPROTO,
3481 		.size = 1,
3482 	}, {
3483 		/* Using UDP ports, this is functionally equivalent to raw
3484 		 * byte pairs from L4 header.
3485 		 */
3486 		.rxnfc_field = RXH_L4_B_0_1,
3487 		.cls_prot = NET_PROT_UDP,
3488 		.cls_field = NH_FLD_UDP_PORT_SRC,
3489 		.id = DPAA2_ETH_DIST_L4SRC,
3490 		.size = 2,
3491 	}, {
3492 		.rxnfc_field = RXH_L4_B_2_3,
3493 		.cls_prot = NET_PROT_UDP,
3494 		.cls_field = NH_FLD_UDP_PORT_DST,
3495 		.id = DPAA2_ETH_DIST_L4DST,
3496 		.size = 2,
3497 	},
3498 };
3499 
3500 /* Configure the Rx hash key using the legacy API */
3501 static int dpaa2_eth_config_legacy_hash_key(struct dpaa2_eth_priv *priv, dma_addr_t key)
3502 {
3503 	struct device *dev = priv->net_dev->dev.parent;
3504 	struct dpni_rx_tc_dist_cfg dist_cfg;
3505 	int i, err = 0;
3506 
3507 	memset(&dist_cfg, 0, sizeof(dist_cfg));
3508 
3509 	dist_cfg.key_cfg_iova = key;
3510 	dist_cfg.dist_size = dpaa2_eth_queue_count(priv);
3511 	dist_cfg.dist_mode = DPNI_DIST_MODE_HASH;
3512 
3513 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3514 		err = dpni_set_rx_tc_dist(priv->mc_io, 0, priv->mc_token,
3515 					  i, &dist_cfg);
3516 		if (err) {
3517 			dev_err(dev, "dpni_set_rx_tc_dist failed\n");
3518 			break;
3519 		}
3520 	}
3521 
3522 	return err;
3523 }
3524 
3525 /* Configure the Rx hash key using the new API */
3526 static int dpaa2_eth_config_hash_key(struct dpaa2_eth_priv *priv, dma_addr_t key)
3527 {
3528 	struct device *dev = priv->net_dev->dev.parent;
3529 	struct dpni_rx_dist_cfg dist_cfg;
3530 	int i, err = 0;
3531 
3532 	memset(&dist_cfg, 0, sizeof(dist_cfg));
3533 
3534 	dist_cfg.key_cfg_iova = key;
3535 	dist_cfg.dist_size = dpaa2_eth_queue_count(priv);
3536 	dist_cfg.enable = 1;
3537 
3538 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3539 		dist_cfg.tc = i;
3540 		err = dpni_set_rx_hash_dist(priv->mc_io, 0, priv->mc_token,
3541 					    &dist_cfg);
3542 		if (err) {
3543 			dev_err(dev, "dpni_set_rx_hash_dist failed\n");
3544 			break;
3545 		}
3546 
3547 		/* If the flow steering / hashing key is shared between all
3548 		 * traffic classes, install it just once
3549 		 */
3550 		if (priv->dpni_attrs.options & DPNI_OPT_SHARED_FS)
3551 			break;
3552 	}
3553 
3554 	return err;
3555 }
3556 
3557 /* Configure the Rx flow classification key */
3558 static int dpaa2_eth_config_cls_key(struct dpaa2_eth_priv *priv, dma_addr_t key)
3559 {
3560 	struct device *dev = priv->net_dev->dev.parent;
3561 	struct dpni_rx_dist_cfg dist_cfg;
3562 	int i, err = 0;
3563 
3564 	memset(&dist_cfg, 0, sizeof(dist_cfg));
3565 
3566 	dist_cfg.key_cfg_iova = key;
3567 	dist_cfg.dist_size = dpaa2_eth_queue_count(priv);
3568 	dist_cfg.enable = 1;
3569 
3570 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3571 		dist_cfg.tc = i;
3572 		err = dpni_set_rx_fs_dist(priv->mc_io, 0, priv->mc_token,
3573 					  &dist_cfg);
3574 		if (err) {
3575 			dev_err(dev, "dpni_set_rx_fs_dist failed\n");
3576 			break;
3577 		}
3578 
3579 		/* If the flow steering / hashing key is shared between all
3580 		 * traffic classes, install it just once
3581 		 */
3582 		if (priv->dpni_attrs.options & DPNI_OPT_SHARED_FS)
3583 			break;
3584 	}
3585 
3586 	return err;
3587 }
3588 
3589 /* Size of the Rx flow classification key */
3590 int dpaa2_eth_cls_key_size(u64 fields)
3591 {
3592 	int i, size = 0;
3593 
3594 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3595 		if (!(fields & dist_fields[i].id))
3596 			continue;
3597 		size += dist_fields[i].size;
3598 	}
3599 
3600 	return size;
3601 }
3602 
3603 /* Offset of header field in Rx classification key */
3604 int dpaa2_eth_cls_fld_off(int prot, int field)
3605 {
3606 	int i, off = 0;
3607 
3608 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3609 		if (dist_fields[i].cls_prot == prot &&
3610 		    dist_fields[i].cls_field == field)
3611 			return off;
3612 		off += dist_fields[i].size;
3613 	}
3614 
3615 	WARN_ONCE(1, "Unsupported header field used for Rx flow cls\n");
3616 	return 0;
3617 }
3618 
3619 /* Prune unused fields from the classification rule.
3620  * Used when masking is not supported
3621  */
3622 void dpaa2_eth_cls_trim_rule(void *key_mem, u64 fields)
3623 {
3624 	int off = 0, new_off = 0;
3625 	int i, size;
3626 
3627 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3628 		size = dist_fields[i].size;
3629 		if (dist_fields[i].id & fields) {
3630 			memcpy(key_mem + new_off, key_mem + off, size);
3631 			new_off += size;
3632 		}
3633 		off += size;
3634 	}
3635 }
3636 
3637 /* Set Rx distribution (hash or flow classification) key
3638  * flags is a combination of RXH_ bits
3639  */
3640 static int dpaa2_eth_set_dist_key(struct net_device *net_dev,
3641 				  enum dpaa2_eth_rx_dist type, u64 flags)
3642 {
3643 	struct device *dev = net_dev->dev.parent;
3644 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
3645 	struct dpkg_profile_cfg cls_cfg;
3646 	u32 rx_hash_fields = 0;
3647 	dma_addr_t key_iova;
3648 	u8 *dma_mem;
3649 	int i;
3650 	int err = 0;
3651 
3652 	memset(&cls_cfg, 0, sizeof(cls_cfg));
3653 
3654 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3655 		struct dpkg_extract *key =
3656 			&cls_cfg.extracts[cls_cfg.num_extracts];
3657 
3658 		/* For both Rx hashing and classification keys
3659 		 * we set only the selected fields.
3660 		 */
3661 		if (!(flags & dist_fields[i].id))
3662 			continue;
3663 		if (type == DPAA2_ETH_RX_DIST_HASH)
3664 			rx_hash_fields |= dist_fields[i].rxnfc_field;
3665 
3666 		if (cls_cfg.num_extracts >= DPKG_MAX_NUM_OF_EXTRACTS) {
3667 			dev_err(dev, "error adding key extraction rule, too many rules?\n");
3668 			return -E2BIG;
3669 		}
3670 
3671 		key->type = DPKG_EXTRACT_FROM_HDR;
3672 		key->extract.from_hdr.prot = dist_fields[i].cls_prot;
3673 		key->extract.from_hdr.type = DPKG_FULL_FIELD;
3674 		key->extract.from_hdr.field = dist_fields[i].cls_field;
3675 		cls_cfg.num_extracts++;
3676 	}
3677 
3678 	dma_mem = kzalloc(DPAA2_CLASSIFIER_DMA_SIZE, GFP_KERNEL);
3679 	if (!dma_mem)
3680 		return -ENOMEM;
3681 
3682 	err = dpni_prepare_key_cfg(&cls_cfg, dma_mem);
3683 	if (err) {
3684 		dev_err(dev, "dpni_prepare_key_cfg error %d\n", err);
3685 		goto free_key;
3686 	}
3687 
3688 	/* Prepare for setting the rx dist */
3689 	key_iova = dma_map_single(dev, dma_mem, DPAA2_CLASSIFIER_DMA_SIZE,
3690 				  DMA_TO_DEVICE);
3691 	if (dma_mapping_error(dev, key_iova)) {
3692 		dev_err(dev, "DMA mapping failed\n");
3693 		err = -ENOMEM;
3694 		goto free_key;
3695 	}
3696 
3697 	if (type == DPAA2_ETH_RX_DIST_HASH) {
3698 		if (dpaa2_eth_has_legacy_dist(priv))
3699 			err = dpaa2_eth_config_legacy_hash_key(priv, key_iova);
3700 		else
3701 			err = dpaa2_eth_config_hash_key(priv, key_iova);
3702 	} else {
3703 		err = dpaa2_eth_config_cls_key(priv, key_iova);
3704 	}
3705 
3706 	dma_unmap_single(dev, key_iova, DPAA2_CLASSIFIER_DMA_SIZE,
3707 			 DMA_TO_DEVICE);
3708 	if (!err && type == DPAA2_ETH_RX_DIST_HASH)
3709 		priv->rx_hash_fields = rx_hash_fields;
3710 
3711 free_key:
3712 	kfree(dma_mem);
3713 	return err;
3714 }
3715 
3716 int dpaa2_eth_set_hash(struct net_device *net_dev, u64 flags)
3717 {
3718 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
3719 	u64 key = 0;
3720 	int i;
3721 
3722 	if (!dpaa2_eth_hash_enabled(priv))
3723 		return -EOPNOTSUPP;
3724 
3725 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++)
3726 		if (dist_fields[i].rxnfc_field & flags)
3727 			key |= dist_fields[i].id;
3728 
3729 	return dpaa2_eth_set_dist_key(net_dev, DPAA2_ETH_RX_DIST_HASH, key);
3730 }
3731 
3732 int dpaa2_eth_set_cls(struct net_device *net_dev, u64 flags)
3733 {
3734 	return dpaa2_eth_set_dist_key(net_dev, DPAA2_ETH_RX_DIST_CLS, flags);
3735 }
3736 
3737 static int dpaa2_eth_set_default_cls(struct dpaa2_eth_priv *priv)
3738 {
3739 	struct device *dev = priv->net_dev->dev.parent;
3740 	int err;
3741 
3742 	/* Check if we actually support Rx flow classification */
3743 	if (dpaa2_eth_has_legacy_dist(priv)) {
3744 		dev_dbg(dev, "Rx cls not supported by current MC version\n");
3745 		return -EOPNOTSUPP;
3746 	}
3747 
3748 	if (!dpaa2_eth_fs_enabled(priv)) {
3749 		dev_dbg(dev, "Rx cls disabled in DPNI options\n");
3750 		return -EOPNOTSUPP;
3751 	}
3752 
3753 	if (!dpaa2_eth_hash_enabled(priv)) {
3754 		dev_dbg(dev, "Rx cls disabled for single queue DPNIs\n");
3755 		return -EOPNOTSUPP;
3756 	}
3757 
3758 	/* If there is no support for masking in the classification table,
3759 	 * we don't set a default key, as it will depend on the rules
3760 	 * added by the user at runtime.
3761 	 */
3762 	if (!dpaa2_eth_fs_mask_enabled(priv))
3763 		goto out;
3764 
3765 	err = dpaa2_eth_set_cls(priv->net_dev, DPAA2_ETH_DIST_ALL);
3766 	if (err)
3767 		return err;
3768 
3769 out:
3770 	priv->rx_cls_enabled = 1;
3771 
3772 	return 0;
3773 }
3774 
3775 /* Bind the DPNI to its needed objects and resources: buffer pool, DPIOs,
3776  * frame queues and channels
3777  */
3778 static int dpaa2_eth_bind_dpni(struct dpaa2_eth_priv *priv)
3779 {
3780 	struct net_device *net_dev = priv->net_dev;
3781 	struct device *dev = net_dev->dev.parent;
3782 	struct dpni_pools_cfg pools_params;
3783 	struct dpni_error_cfg err_cfg;
3784 	int err = 0;
3785 	int i;
3786 
3787 	pools_params.num_dpbp = 1;
3788 	pools_params.pools[0].dpbp_id = priv->dpbp_dev->obj_desc.id;
3789 	pools_params.pools[0].backup_pool = 0;
3790 	pools_params.pools[0].buffer_size = priv->rx_buf_size;
3791 	err = dpni_set_pools(priv->mc_io, 0, priv->mc_token, &pools_params);
3792 	if (err) {
3793 		dev_err(dev, "dpni_set_pools() failed\n");
3794 		return err;
3795 	}
3796 
3797 	/* have the interface implicitly distribute traffic based on
3798 	 * the default hash key
3799 	 */
3800 	err = dpaa2_eth_set_hash(net_dev, DPAA2_RXH_DEFAULT);
3801 	if (err && err != -EOPNOTSUPP)
3802 		dev_err(dev, "Failed to configure hashing\n");
3803 
3804 	/* Configure the flow classification key; it includes all
3805 	 * supported header fields and cannot be modified at runtime
3806 	 */
3807 	err = dpaa2_eth_set_default_cls(priv);
3808 	if (err && err != -EOPNOTSUPP)
3809 		dev_err(dev, "Failed to configure Rx classification key\n");
3810 
3811 	/* Configure handling of error frames */
3812 	err_cfg.errors = DPAA2_FAS_RX_ERR_MASK;
3813 	err_cfg.set_frame_annotation = 1;
3814 	err_cfg.error_action = DPNI_ERROR_ACTION_DISCARD;
3815 	err = dpni_set_errors_behavior(priv->mc_io, 0, priv->mc_token,
3816 				       &err_cfg);
3817 	if (err) {
3818 		dev_err(dev, "dpni_set_errors_behavior failed\n");
3819 		return err;
3820 	}
3821 
3822 	/* Configure Rx and Tx conf queues to generate CDANs */
3823 	for (i = 0; i < priv->num_fqs; i++) {
3824 		switch (priv->fq[i].type) {
3825 		case DPAA2_RX_FQ:
3826 			err = dpaa2_eth_setup_rx_flow(priv, &priv->fq[i]);
3827 			break;
3828 		case DPAA2_TX_CONF_FQ:
3829 			err = dpaa2_eth_setup_tx_flow(priv, &priv->fq[i]);
3830 			break;
3831 		case DPAA2_RX_ERR_FQ:
3832 			err = setup_rx_err_flow(priv, &priv->fq[i]);
3833 			break;
3834 		default:
3835 			dev_err(dev, "Invalid FQ type %d\n", priv->fq[i].type);
3836 			return -EINVAL;
3837 		}
3838 		if (err)
3839 			return err;
3840 	}
3841 
3842 	err = dpni_get_qdid(priv->mc_io, 0, priv->mc_token,
3843 			    DPNI_QUEUE_TX, &priv->tx_qdid);
3844 	if (err) {
3845 		dev_err(dev, "dpni_get_qdid() failed\n");
3846 		return err;
3847 	}
3848 
3849 	return 0;
3850 }
3851 
3852 /* Allocate rings for storing incoming frame descriptors */
3853 static int dpaa2_eth_alloc_rings(struct dpaa2_eth_priv *priv)
3854 {
3855 	struct net_device *net_dev = priv->net_dev;
3856 	struct device *dev = net_dev->dev.parent;
3857 	int i;
3858 
3859 	for (i = 0; i < priv->num_channels; i++) {
3860 		priv->channel[i]->store =
3861 			dpaa2_io_store_create(DPAA2_ETH_STORE_SIZE, dev);
3862 		if (!priv->channel[i]->store) {
3863 			netdev_err(net_dev, "dpaa2_io_store_create() failed\n");
3864 			goto err_ring;
3865 		}
3866 	}
3867 
3868 	return 0;
3869 
3870 err_ring:
3871 	for (i = 0; i < priv->num_channels; i++) {
3872 		if (!priv->channel[i]->store)
3873 			break;
3874 		dpaa2_io_store_destroy(priv->channel[i]->store);
3875 	}
3876 
3877 	return -ENOMEM;
3878 }
3879 
3880 static void dpaa2_eth_free_rings(struct dpaa2_eth_priv *priv)
3881 {
3882 	int i;
3883 
3884 	for (i = 0; i < priv->num_channels; i++)
3885 		dpaa2_io_store_destroy(priv->channel[i]->store);
3886 }
3887 
3888 static int dpaa2_eth_set_mac_addr(struct dpaa2_eth_priv *priv)
3889 {
3890 	struct net_device *net_dev = priv->net_dev;
3891 	struct device *dev = net_dev->dev.parent;
3892 	u8 mac_addr[ETH_ALEN], dpni_mac_addr[ETH_ALEN];
3893 	int err;
3894 
3895 	/* Get firmware address, if any */
3896 	err = dpni_get_port_mac_addr(priv->mc_io, 0, priv->mc_token, mac_addr);
3897 	if (err) {
3898 		dev_err(dev, "dpni_get_port_mac_addr() failed\n");
3899 		return err;
3900 	}
3901 
3902 	/* Get DPNI attributes address, if any */
3903 	err = dpni_get_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
3904 					dpni_mac_addr);
3905 	if (err) {
3906 		dev_err(dev, "dpni_get_primary_mac_addr() failed\n");
3907 		return err;
3908 	}
3909 
3910 	/* First check if firmware has any address configured by bootloader */
3911 	if (!is_zero_ether_addr(mac_addr)) {
3912 		/* If the DPMAC addr != DPNI addr, update it */
3913 		if (!ether_addr_equal(mac_addr, dpni_mac_addr)) {
3914 			err = dpni_set_primary_mac_addr(priv->mc_io, 0,
3915 							priv->mc_token,
3916 							mac_addr);
3917 			if (err) {
3918 				dev_err(dev, "dpni_set_primary_mac_addr() failed\n");
3919 				return err;
3920 			}
3921 		}
3922 		memcpy(net_dev->dev_addr, mac_addr, net_dev->addr_len);
3923 	} else if (is_zero_ether_addr(dpni_mac_addr)) {
3924 		/* No MAC address configured, fill in net_dev->dev_addr
3925 		 * with a random one
3926 		 */
3927 		eth_hw_addr_random(net_dev);
3928 		dev_dbg_once(dev, "device(s) have all-zero hwaddr, replaced with random\n");
3929 
3930 		err = dpni_set_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
3931 						net_dev->dev_addr);
3932 		if (err) {
3933 			dev_err(dev, "dpni_set_primary_mac_addr() failed\n");
3934 			return err;
3935 		}
3936 
3937 		/* Override NET_ADDR_RANDOM set by eth_hw_addr_random(); for all
3938 		 * practical purposes, this will be our "permanent" mac address,
3939 		 * at least until the next reboot. This move will also permit
3940 		 * register_netdevice() to properly fill up net_dev->perm_addr.
3941 		 */
3942 		net_dev->addr_assign_type = NET_ADDR_PERM;
3943 	} else {
3944 		/* NET_ADDR_PERM is default, all we have to do is
3945 		 * fill in the device addr.
3946 		 */
3947 		memcpy(net_dev->dev_addr, dpni_mac_addr, net_dev->addr_len);
3948 	}
3949 
3950 	return 0;
3951 }
3952 
3953 static int dpaa2_eth_netdev_init(struct net_device *net_dev)
3954 {
3955 	struct device *dev = net_dev->dev.parent;
3956 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
3957 	u32 options = priv->dpni_attrs.options;
3958 	u64 supported = 0, not_supported = 0;
3959 	u8 bcast_addr[ETH_ALEN];
3960 	u8 num_queues;
3961 	int err;
3962 
3963 	net_dev->netdev_ops = &dpaa2_eth_ops;
3964 	net_dev->ethtool_ops = &dpaa2_ethtool_ops;
3965 
3966 	err = dpaa2_eth_set_mac_addr(priv);
3967 	if (err)
3968 		return err;
3969 
3970 	/* Explicitly add the broadcast address to the MAC filtering table */
3971 	eth_broadcast_addr(bcast_addr);
3972 	err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token, bcast_addr);
3973 	if (err) {
3974 		dev_err(dev, "dpni_add_mac_addr() failed\n");
3975 		return err;
3976 	}
3977 
3978 	/* Set MTU upper limit; lower limit is 68B (default value) */
3979 	net_dev->max_mtu = DPAA2_ETH_MAX_MTU;
3980 	err = dpni_set_max_frame_length(priv->mc_io, 0, priv->mc_token,
3981 					DPAA2_ETH_MFL);
3982 	if (err) {
3983 		dev_err(dev, "dpni_set_max_frame_length() failed\n");
3984 		return err;
3985 	}
3986 
3987 	/* Set actual number of queues in the net device */
3988 	num_queues = dpaa2_eth_queue_count(priv);
3989 	err = netif_set_real_num_tx_queues(net_dev, num_queues);
3990 	if (err) {
3991 		dev_err(dev, "netif_set_real_num_tx_queues() failed\n");
3992 		return err;
3993 	}
3994 	err = netif_set_real_num_rx_queues(net_dev, num_queues);
3995 	if (err) {
3996 		dev_err(dev, "netif_set_real_num_rx_queues() failed\n");
3997 		return err;
3998 	}
3999 
4000 	/* Capabilities listing */
4001 	supported |= IFF_LIVE_ADDR_CHANGE;
4002 
4003 	if (options & DPNI_OPT_NO_MAC_FILTER)
4004 		not_supported |= IFF_UNICAST_FLT;
4005 	else
4006 		supported |= IFF_UNICAST_FLT;
4007 
4008 	net_dev->priv_flags |= supported;
4009 	net_dev->priv_flags &= ~not_supported;
4010 
4011 	/* Features */
4012 	net_dev->features = NETIF_F_RXCSUM |
4013 			    NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4014 			    NETIF_F_SG | NETIF_F_HIGHDMA |
4015 			    NETIF_F_LLTX | NETIF_F_HW_TC;
4016 	net_dev->hw_features = net_dev->features;
4017 
4018 	return 0;
4019 }
4020 
4021 static int dpaa2_eth_poll_link_state(void *arg)
4022 {
4023 	struct dpaa2_eth_priv *priv = (struct dpaa2_eth_priv *)arg;
4024 	int err;
4025 
4026 	while (!kthread_should_stop()) {
4027 		err = dpaa2_eth_link_state_update(priv);
4028 		if (unlikely(err))
4029 			return err;
4030 
4031 		msleep(DPAA2_ETH_LINK_STATE_REFRESH);
4032 	}
4033 
4034 	return 0;
4035 }
4036 
4037 static int dpaa2_eth_connect_mac(struct dpaa2_eth_priv *priv)
4038 {
4039 	struct fsl_mc_device *dpni_dev, *dpmac_dev;
4040 	struct dpaa2_mac *mac;
4041 	int err;
4042 
4043 	dpni_dev = to_fsl_mc_device(priv->net_dev->dev.parent);
4044 	dpmac_dev = fsl_mc_get_endpoint(dpni_dev);
4045 	if (IS_ERR_OR_NULL(dpmac_dev) || dpmac_dev->dev.type != &fsl_mc_bus_dpmac_type)
4046 		return 0;
4047 
4048 	mac = kzalloc(sizeof(struct dpaa2_mac), GFP_KERNEL);
4049 	if (!mac)
4050 		return -ENOMEM;
4051 
4052 	mac->mc_dev = dpmac_dev;
4053 	mac->mc_io = priv->mc_io;
4054 	mac->net_dev = priv->net_dev;
4055 
4056 	err = dpaa2_mac_open(mac);
4057 	if (err)
4058 		goto err_free_mac;
4059 	priv->mac = mac;
4060 
4061 	if (dpaa2_eth_is_type_phy(priv)) {
4062 		err = dpaa2_mac_connect(mac);
4063 		if (err) {
4064 			netdev_err(priv->net_dev, "Error connecting to the MAC endpoint\n");
4065 			goto err_close_mac;
4066 		}
4067 	}
4068 
4069 	return 0;
4070 
4071 err_close_mac:
4072 	dpaa2_mac_close(mac);
4073 	priv->mac = NULL;
4074 err_free_mac:
4075 	kfree(mac);
4076 	return err;
4077 }
4078 
4079 static void dpaa2_eth_disconnect_mac(struct dpaa2_eth_priv *priv)
4080 {
4081 	if (dpaa2_eth_is_type_phy(priv))
4082 		dpaa2_mac_disconnect(priv->mac);
4083 
4084 	dpaa2_mac_close(priv->mac);
4085 	kfree(priv->mac);
4086 	priv->mac = NULL;
4087 }
4088 
4089 static irqreturn_t dpni_irq0_handler_thread(int irq_num, void *arg)
4090 {
4091 	u32 status = ~0;
4092 	struct device *dev = (struct device *)arg;
4093 	struct fsl_mc_device *dpni_dev = to_fsl_mc_device(dev);
4094 	struct net_device *net_dev = dev_get_drvdata(dev);
4095 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
4096 	int err;
4097 
4098 	err = dpni_get_irq_status(dpni_dev->mc_io, 0, dpni_dev->mc_handle,
4099 				  DPNI_IRQ_INDEX, &status);
4100 	if (unlikely(err)) {
4101 		netdev_err(net_dev, "Can't get irq status (err %d)\n", err);
4102 		return IRQ_HANDLED;
4103 	}
4104 
4105 	if (status & DPNI_IRQ_EVENT_LINK_CHANGED)
4106 		dpaa2_eth_link_state_update(netdev_priv(net_dev));
4107 
4108 	if (status & DPNI_IRQ_EVENT_ENDPOINT_CHANGED) {
4109 		dpaa2_eth_set_mac_addr(netdev_priv(net_dev));
4110 		dpaa2_eth_update_tx_fqids(priv);
4111 
4112 		rtnl_lock();
4113 		if (dpaa2_eth_has_mac(priv))
4114 			dpaa2_eth_disconnect_mac(priv);
4115 		else
4116 			dpaa2_eth_connect_mac(priv);
4117 		rtnl_unlock();
4118 	}
4119 
4120 	return IRQ_HANDLED;
4121 }
4122 
4123 static int dpaa2_eth_setup_irqs(struct fsl_mc_device *ls_dev)
4124 {
4125 	int err = 0;
4126 	struct fsl_mc_device_irq *irq;
4127 
4128 	err = fsl_mc_allocate_irqs(ls_dev);
4129 	if (err) {
4130 		dev_err(&ls_dev->dev, "MC irqs allocation failed\n");
4131 		return err;
4132 	}
4133 
4134 	irq = ls_dev->irqs[0];
4135 	err = devm_request_threaded_irq(&ls_dev->dev, irq->msi_desc->irq,
4136 					NULL, dpni_irq0_handler_thread,
4137 					IRQF_NO_SUSPEND | IRQF_ONESHOT,
4138 					dev_name(&ls_dev->dev), &ls_dev->dev);
4139 	if (err < 0) {
4140 		dev_err(&ls_dev->dev, "devm_request_threaded_irq(): %d\n", err);
4141 		goto free_mc_irq;
4142 	}
4143 
4144 	err = dpni_set_irq_mask(ls_dev->mc_io, 0, ls_dev->mc_handle,
4145 				DPNI_IRQ_INDEX, DPNI_IRQ_EVENT_LINK_CHANGED |
4146 				DPNI_IRQ_EVENT_ENDPOINT_CHANGED);
4147 	if (err < 0) {
4148 		dev_err(&ls_dev->dev, "dpni_set_irq_mask(): %d\n", err);
4149 		goto free_irq;
4150 	}
4151 
4152 	err = dpni_set_irq_enable(ls_dev->mc_io, 0, ls_dev->mc_handle,
4153 				  DPNI_IRQ_INDEX, 1);
4154 	if (err < 0) {
4155 		dev_err(&ls_dev->dev, "dpni_set_irq_enable(): %d\n", err);
4156 		goto free_irq;
4157 	}
4158 
4159 	return 0;
4160 
4161 free_irq:
4162 	devm_free_irq(&ls_dev->dev, irq->msi_desc->irq, &ls_dev->dev);
4163 free_mc_irq:
4164 	fsl_mc_free_irqs(ls_dev);
4165 
4166 	return err;
4167 }
4168 
4169 static void dpaa2_eth_add_ch_napi(struct dpaa2_eth_priv *priv)
4170 {
4171 	int i;
4172 	struct dpaa2_eth_channel *ch;
4173 
4174 	for (i = 0; i < priv->num_channels; i++) {
4175 		ch = priv->channel[i];
4176 		/* NAPI weight *MUST* be a multiple of DPAA2_ETH_STORE_SIZE */
4177 		netif_napi_add(priv->net_dev, &ch->napi, dpaa2_eth_poll,
4178 			       NAPI_POLL_WEIGHT);
4179 	}
4180 }
4181 
4182 static void dpaa2_eth_del_ch_napi(struct dpaa2_eth_priv *priv)
4183 {
4184 	int i;
4185 	struct dpaa2_eth_channel *ch;
4186 
4187 	for (i = 0; i < priv->num_channels; i++) {
4188 		ch = priv->channel[i];
4189 		netif_napi_del(&ch->napi);
4190 	}
4191 }
4192 
4193 static int dpaa2_eth_probe(struct fsl_mc_device *dpni_dev)
4194 {
4195 	struct device *dev;
4196 	struct net_device *net_dev = NULL;
4197 	struct dpaa2_eth_priv *priv = NULL;
4198 	int err = 0;
4199 
4200 	dev = &dpni_dev->dev;
4201 
4202 	/* Net device */
4203 	net_dev = alloc_etherdev_mq(sizeof(*priv), DPAA2_ETH_MAX_NETDEV_QUEUES);
4204 	if (!net_dev) {
4205 		dev_err(dev, "alloc_etherdev_mq() failed\n");
4206 		return -ENOMEM;
4207 	}
4208 
4209 	SET_NETDEV_DEV(net_dev, dev);
4210 	dev_set_drvdata(dev, net_dev);
4211 
4212 	priv = netdev_priv(net_dev);
4213 	priv->net_dev = net_dev;
4214 
4215 	priv->iommu_domain = iommu_get_domain_for_dev(dev);
4216 
4217 	priv->tx_tstamp_type = HWTSTAMP_TX_OFF;
4218 	priv->rx_tstamp = false;
4219 
4220 	priv->dpaa2_ptp_wq = alloc_workqueue("dpaa2_ptp_wq", 0, 0);
4221 	if (!priv->dpaa2_ptp_wq) {
4222 		err = -ENOMEM;
4223 		goto err_wq_alloc;
4224 	}
4225 
4226 	INIT_WORK(&priv->tx_onestep_tstamp, dpaa2_eth_tx_onestep_tstamp);
4227 
4228 	skb_queue_head_init(&priv->tx_skbs);
4229 
4230 	/* Obtain a MC portal */
4231 	err = fsl_mc_portal_allocate(dpni_dev, FSL_MC_IO_ATOMIC_CONTEXT_PORTAL,
4232 				     &priv->mc_io);
4233 	if (err) {
4234 		if (err == -ENXIO)
4235 			err = -EPROBE_DEFER;
4236 		else
4237 			dev_err(dev, "MC portal allocation failed\n");
4238 		goto err_portal_alloc;
4239 	}
4240 
4241 	/* MC objects initialization and configuration */
4242 	err = dpaa2_eth_setup_dpni(dpni_dev);
4243 	if (err)
4244 		goto err_dpni_setup;
4245 
4246 	err = dpaa2_eth_setup_dpio(priv);
4247 	if (err)
4248 		goto err_dpio_setup;
4249 
4250 	dpaa2_eth_setup_fqs(priv);
4251 
4252 	err = dpaa2_eth_setup_dpbp(priv);
4253 	if (err)
4254 		goto err_dpbp_setup;
4255 
4256 	err = dpaa2_eth_bind_dpni(priv);
4257 	if (err)
4258 		goto err_bind;
4259 
4260 	/* Add a NAPI context for each channel */
4261 	dpaa2_eth_add_ch_napi(priv);
4262 
4263 	/* Percpu statistics */
4264 	priv->percpu_stats = alloc_percpu(*priv->percpu_stats);
4265 	if (!priv->percpu_stats) {
4266 		dev_err(dev, "alloc_percpu(percpu_stats) failed\n");
4267 		err = -ENOMEM;
4268 		goto err_alloc_percpu_stats;
4269 	}
4270 	priv->percpu_extras = alloc_percpu(*priv->percpu_extras);
4271 	if (!priv->percpu_extras) {
4272 		dev_err(dev, "alloc_percpu(percpu_extras) failed\n");
4273 		err = -ENOMEM;
4274 		goto err_alloc_percpu_extras;
4275 	}
4276 
4277 	priv->sgt_cache = alloc_percpu(*priv->sgt_cache);
4278 	if (!priv->sgt_cache) {
4279 		dev_err(dev, "alloc_percpu(sgt_cache) failed\n");
4280 		err = -ENOMEM;
4281 		goto err_alloc_sgt_cache;
4282 	}
4283 
4284 	err = dpaa2_eth_netdev_init(net_dev);
4285 	if (err)
4286 		goto err_netdev_init;
4287 
4288 	/* Configure checksum offload based on current interface flags */
4289 	err = dpaa2_eth_set_rx_csum(priv, !!(net_dev->features & NETIF_F_RXCSUM));
4290 	if (err)
4291 		goto err_csum;
4292 
4293 	err = dpaa2_eth_set_tx_csum(priv,
4294 				    !!(net_dev->features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)));
4295 	if (err)
4296 		goto err_csum;
4297 
4298 	err = dpaa2_eth_alloc_rings(priv);
4299 	if (err)
4300 		goto err_alloc_rings;
4301 
4302 #ifdef CONFIG_FSL_DPAA2_ETH_DCB
4303 	if (dpaa2_eth_has_pause_support(priv) && priv->vlan_cls_enabled) {
4304 		priv->dcbx_mode = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_IEEE;
4305 		net_dev->dcbnl_ops = &dpaa2_eth_dcbnl_ops;
4306 	} else {
4307 		dev_dbg(dev, "PFC not supported\n");
4308 	}
4309 #endif
4310 
4311 	err = dpaa2_eth_setup_irqs(dpni_dev);
4312 	if (err) {
4313 		netdev_warn(net_dev, "Failed to set link interrupt, fall back to polling\n");
4314 		priv->poll_thread = kthread_run(dpaa2_eth_poll_link_state, priv,
4315 						"%s_poll_link", net_dev->name);
4316 		if (IS_ERR(priv->poll_thread)) {
4317 			dev_err(dev, "Error starting polling thread\n");
4318 			goto err_poll_thread;
4319 		}
4320 		priv->do_link_poll = true;
4321 	}
4322 
4323 	err = dpaa2_eth_connect_mac(priv);
4324 	if (err)
4325 		goto err_connect_mac;
4326 
4327 	err = dpaa2_eth_dl_register(priv);
4328 	if (err)
4329 		goto err_dl_register;
4330 
4331 	err = dpaa2_eth_dl_traps_register(priv);
4332 	if (err)
4333 		goto err_dl_trap_register;
4334 
4335 	err = dpaa2_eth_dl_port_add(priv);
4336 	if (err)
4337 		goto err_dl_port_add;
4338 
4339 	err = register_netdev(net_dev);
4340 	if (err < 0) {
4341 		dev_err(dev, "register_netdev() failed\n");
4342 		goto err_netdev_reg;
4343 	}
4344 
4345 #ifdef CONFIG_DEBUG_FS
4346 	dpaa2_dbg_add(priv);
4347 #endif
4348 
4349 	dev_info(dev, "Probed interface %s\n", net_dev->name);
4350 	return 0;
4351 
4352 err_netdev_reg:
4353 	dpaa2_eth_dl_port_del(priv);
4354 err_dl_port_add:
4355 	dpaa2_eth_dl_traps_unregister(priv);
4356 err_dl_trap_register:
4357 	dpaa2_eth_dl_unregister(priv);
4358 err_dl_register:
4359 	dpaa2_eth_disconnect_mac(priv);
4360 err_connect_mac:
4361 	if (priv->do_link_poll)
4362 		kthread_stop(priv->poll_thread);
4363 	else
4364 		fsl_mc_free_irqs(dpni_dev);
4365 err_poll_thread:
4366 	dpaa2_eth_free_rings(priv);
4367 err_alloc_rings:
4368 err_csum:
4369 err_netdev_init:
4370 	free_percpu(priv->sgt_cache);
4371 err_alloc_sgt_cache:
4372 	free_percpu(priv->percpu_extras);
4373 err_alloc_percpu_extras:
4374 	free_percpu(priv->percpu_stats);
4375 err_alloc_percpu_stats:
4376 	dpaa2_eth_del_ch_napi(priv);
4377 err_bind:
4378 	dpaa2_eth_free_dpbp(priv);
4379 err_dpbp_setup:
4380 	dpaa2_eth_free_dpio(priv);
4381 err_dpio_setup:
4382 	dpaa2_eth_free_dpni(priv);
4383 err_dpni_setup:
4384 	fsl_mc_portal_free(priv->mc_io);
4385 err_portal_alloc:
4386 	destroy_workqueue(priv->dpaa2_ptp_wq);
4387 err_wq_alloc:
4388 	dev_set_drvdata(dev, NULL);
4389 	free_netdev(net_dev);
4390 
4391 	return err;
4392 }
4393 
4394 static int dpaa2_eth_remove(struct fsl_mc_device *ls_dev)
4395 {
4396 	struct device *dev;
4397 	struct net_device *net_dev;
4398 	struct dpaa2_eth_priv *priv;
4399 
4400 	dev = &ls_dev->dev;
4401 	net_dev = dev_get_drvdata(dev);
4402 	priv = netdev_priv(net_dev);
4403 
4404 #ifdef CONFIG_DEBUG_FS
4405 	dpaa2_dbg_remove(priv);
4406 #endif
4407 	rtnl_lock();
4408 	dpaa2_eth_disconnect_mac(priv);
4409 	rtnl_unlock();
4410 
4411 	unregister_netdev(net_dev);
4412 
4413 	dpaa2_eth_dl_port_del(priv);
4414 	dpaa2_eth_dl_traps_unregister(priv);
4415 	dpaa2_eth_dl_unregister(priv);
4416 
4417 	if (priv->do_link_poll)
4418 		kthread_stop(priv->poll_thread);
4419 	else
4420 		fsl_mc_free_irqs(ls_dev);
4421 
4422 	dpaa2_eth_free_rings(priv);
4423 	free_percpu(priv->sgt_cache);
4424 	free_percpu(priv->percpu_stats);
4425 	free_percpu(priv->percpu_extras);
4426 
4427 	dpaa2_eth_del_ch_napi(priv);
4428 	dpaa2_eth_free_dpbp(priv);
4429 	dpaa2_eth_free_dpio(priv);
4430 	dpaa2_eth_free_dpni(priv);
4431 
4432 	fsl_mc_portal_free(priv->mc_io);
4433 
4434 	free_netdev(net_dev);
4435 
4436 	dev_dbg(net_dev->dev.parent, "Removed interface %s\n", net_dev->name);
4437 
4438 	return 0;
4439 }
4440 
4441 static const struct fsl_mc_device_id dpaa2_eth_match_id_table[] = {
4442 	{
4443 		.vendor = FSL_MC_VENDOR_FREESCALE,
4444 		.obj_type = "dpni",
4445 	},
4446 	{ .vendor = 0x0 }
4447 };
4448 MODULE_DEVICE_TABLE(fslmc, dpaa2_eth_match_id_table);
4449 
4450 static struct fsl_mc_driver dpaa2_eth_driver = {
4451 	.driver = {
4452 		.name = KBUILD_MODNAME,
4453 		.owner = THIS_MODULE,
4454 	},
4455 	.probe = dpaa2_eth_probe,
4456 	.remove = dpaa2_eth_remove,
4457 	.match_id_table = dpaa2_eth_match_id_table
4458 };
4459 
4460 static int __init dpaa2_eth_driver_init(void)
4461 {
4462 	int err;
4463 
4464 	dpaa2_eth_dbg_init();
4465 	err = fsl_mc_driver_register(&dpaa2_eth_driver);
4466 	if (err) {
4467 		dpaa2_eth_dbg_exit();
4468 		return err;
4469 	}
4470 
4471 	return 0;
4472 }
4473 
4474 static void __exit dpaa2_eth_driver_exit(void)
4475 {
4476 	dpaa2_eth_dbg_exit();
4477 	fsl_mc_driver_unregister(&dpaa2_eth_driver);
4478 }
4479 
4480 module_init(dpaa2_eth_driver_init);
4481 module_exit(dpaa2_eth_driver_exit);
4482