xref: /openbmc/linux/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c (revision 0760aad038b5a032c31ea124feed63d88627d2f1)
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/net_tstamp.h>
15 #include <linux/fsl/mc.h>
16 #include <linux/bpf.h>
17 #include <linux/bpf_trace.h>
18 #include <net/pkt_cls.h>
19 #include <net/sock.h>
20 
21 #include "dpaa2-eth.h"
22 
23 /* CREATE_TRACE_POINTS only needs to be defined once. Other dpa files
24  * using trace events only need to #include <trace/events/sched.h>
25  */
26 #define CREATE_TRACE_POINTS
27 #include "dpaa2-eth-trace.h"
28 
29 MODULE_LICENSE("Dual BSD/GPL");
30 MODULE_AUTHOR("Freescale Semiconductor, Inc");
31 MODULE_DESCRIPTION("Freescale DPAA2 Ethernet Driver");
32 
33 static void *dpaa2_iova_to_virt(struct iommu_domain *domain,
34 				dma_addr_t iova_addr)
35 {
36 	phys_addr_t phys_addr;
37 
38 	phys_addr = domain ? iommu_iova_to_phys(domain, iova_addr) : iova_addr;
39 
40 	return phys_to_virt(phys_addr);
41 }
42 
43 static void dpaa2_eth_validate_rx_csum(struct dpaa2_eth_priv *priv,
44 				       u32 fd_status,
45 				       struct sk_buff *skb)
46 {
47 	skb_checksum_none_assert(skb);
48 
49 	/* HW checksum validation is disabled, nothing to do here */
50 	if (!(priv->net_dev->features & NETIF_F_RXCSUM))
51 		return;
52 
53 	/* Read checksum validation bits */
54 	if (!((fd_status & DPAA2_FAS_L3CV) &&
55 	      (fd_status & DPAA2_FAS_L4CV)))
56 		return;
57 
58 	/* Inform the stack there's no need to compute L3/L4 csum anymore */
59 	skb->ip_summed = CHECKSUM_UNNECESSARY;
60 }
61 
62 /* Free a received FD.
63  * Not to be used for Tx conf FDs or on any other paths.
64  */
65 static void dpaa2_eth_free_rx_fd(struct dpaa2_eth_priv *priv,
66 				 const struct dpaa2_fd *fd,
67 				 void *vaddr)
68 {
69 	struct device *dev = priv->net_dev->dev.parent;
70 	dma_addr_t addr = dpaa2_fd_get_addr(fd);
71 	u8 fd_format = dpaa2_fd_get_format(fd);
72 	struct dpaa2_sg_entry *sgt;
73 	void *sg_vaddr;
74 	int i;
75 
76 	/* If single buffer frame, just free the data buffer */
77 	if (fd_format == dpaa2_fd_single)
78 		goto free_buf;
79 	else if (fd_format != dpaa2_fd_sg)
80 		/* We don't support any other format */
81 		return;
82 
83 	/* For S/G frames, we first need to free all SG entries
84 	 * except the first one, which was taken care of already
85 	 */
86 	sgt = vaddr + dpaa2_fd_get_offset(fd);
87 	for (i = 1; i < DPAA2_ETH_MAX_SG_ENTRIES; i++) {
88 		addr = dpaa2_sg_get_addr(&sgt[i]);
89 		sg_vaddr = dpaa2_iova_to_virt(priv->iommu_domain, addr);
90 		dma_unmap_page(dev, addr, priv->rx_buf_size,
91 			       DMA_BIDIRECTIONAL);
92 
93 		free_pages((unsigned long)sg_vaddr, 0);
94 		if (dpaa2_sg_is_final(&sgt[i]))
95 			break;
96 	}
97 
98 free_buf:
99 	free_pages((unsigned long)vaddr, 0);
100 }
101 
102 /* Build a linear skb based on a single-buffer frame descriptor */
103 static struct sk_buff *dpaa2_eth_build_linear_skb(struct dpaa2_eth_channel *ch,
104 						  const struct dpaa2_fd *fd,
105 						  void *fd_vaddr)
106 {
107 	struct sk_buff *skb = NULL;
108 	u16 fd_offset = dpaa2_fd_get_offset(fd);
109 	u32 fd_length = dpaa2_fd_get_len(fd);
110 
111 	ch->buf_count--;
112 
113 	skb = build_skb(fd_vaddr, DPAA2_ETH_RX_BUF_RAW_SIZE);
114 	if (unlikely(!skb))
115 		return NULL;
116 
117 	skb_reserve(skb, fd_offset);
118 	skb_put(skb, fd_length);
119 
120 	return skb;
121 }
122 
123 /* Build a non linear (fragmented) skb based on a S/G table */
124 static struct sk_buff *dpaa2_eth_build_frag_skb(struct dpaa2_eth_priv *priv,
125 						struct dpaa2_eth_channel *ch,
126 						struct dpaa2_sg_entry *sgt)
127 {
128 	struct sk_buff *skb = NULL;
129 	struct device *dev = priv->net_dev->dev.parent;
130 	void *sg_vaddr;
131 	dma_addr_t sg_addr;
132 	u16 sg_offset;
133 	u32 sg_length;
134 	struct page *page, *head_page;
135 	int page_offset;
136 	int i;
137 
138 	for (i = 0; i < DPAA2_ETH_MAX_SG_ENTRIES; i++) {
139 		struct dpaa2_sg_entry *sge = &sgt[i];
140 
141 		/* NOTE: We only support SG entries in dpaa2_sg_single format,
142 		 * but this is the only format we may receive from HW anyway
143 		 */
144 
145 		/* Get the address and length from the S/G entry */
146 		sg_addr = dpaa2_sg_get_addr(sge);
147 		sg_vaddr = dpaa2_iova_to_virt(priv->iommu_domain, sg_addr);
148 		dma_unmap_page(dev, sg_addr, priv->rx_buf_size,
149 			       DMA_BIDIRECTIONAL);
150 
151 		sg_length = dpaa2_sg_get_len(sge);
152 
153 		if (i == 0) {
154 			/* We build the skb around the first data buffer */
155 			skb = build_skb(sg_vaddr, DPAA2_ETH_RX_BUF_RAW_SIZE);
156 			if (unlikely(!skb)) {
157 				/* Free the first SG entry now, since we already
158 				 * unmapped it and obtained the virtual address
159 				 */
160 				free_pages((unsigned long)sg_vaddr, 0);
161 
162 				/* We still need to subtract the buffers used
163 				 * by this FD from our software counter
164 				 */
165 				while (!dpaa2_sg_is_final(&sgt[i]) &&
166 				       i < DPAA2_ETH_MAX_SG_ENTRIES)
167 					i++;
168 				break;
169 			}
170 
171 			sg_offset = dpaa2_sg_get_offset(sge);
172 			skb_reserve(skb, sg_offset);
173 			skb_put(skb, sg_length);
174 		} else {
175 			/* Rest of the data buffers are stored as skb frags */
176 			page = virt_to_page(sg_vaddr);
177 			head_page = virt_to_head_page(sg_vaddr);
178 
179 			/* Offset in page (which may be compound).
180 			 * Data in subsequent SG entries is stored from the
181 			 * beginning of the buffer, so we don't need to add the
182 			 * sg_offset.
183 			 */
184 			page_offset = ((unsigned long)sg_vaddr &
185 				(PAGE_SIZE - 1)) +
186 				(page_address(page) - page_address(head_page));
187 
188 			skb_add_rx_frag(skb, i - 1, head_page, page_offset,
189 					sg_length, priv->rx_buf_size);
190 		}
191 
192 		if (dpaa2_sg_is_final(sge))
193 			break;
194 	}
195 
196 	WARN_ONCE(i == DPAA2_ETH_MAX_SG_ENTRIES, "Final bit not set in SGT");
197 
198 	/* Count all data buffers + SG table buffer */
199 	ch->buf_count -= i + 2;
200 
201 	return skb;
202 }
203 
204 /* Free buffers acquired from the buffer pool or which were meant to
205  * be released in the pool
206  */
207 static void dpaa2_eth_free_bufs(struct dpaa2_eth_priv *priv, u64 *buf_array,
208 				int count)
209 {
210 	struct device *dev = priv->net_dev->dev.parent;
211 	void *vaddr;
212 	int i;
213 
214 	for (i = 0; i < count; i++) {
215 		vaddr = dpaa2_iova_to_virt(priv->iommu_domain, buf_array[i]);
216 		dma_unmap_page(dev, buf_array[i], priv->rx_buf_size,
217 			       DMA_BIDIRECTIONAL);
218 		free_pages((unsigned long)vaddr, 0);
219 	}
220 }
221 
222 static void dpaa2_eth_xdp_release_buf(struct dpaa2_eth_priv *priv,
223 				      struct dpaa2_eth_channel *ch,
224 				      dma_addr_t addr)
225 {
226 	int retries = 0;
227 	int err;
228 
229 	ch->xdp.drop_bufs[ch->xdp.drop_cnt++] = addr;
230 	if (ch->xdp.drop_cnt < DPAA2_ETH_BUFS_PER_CMD)
231 		return;
232 
233 	while ((err = dpaa2_io_service_release(ch->dpio, priv->bpid,
234 					       ch->xdp.drop_bufs,
235 					       ch->xdp.drop_cnt)) == -EBUSY) {
236 		if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES)
237 			break;
238 		cpu_relax();
239 	}
240 
241 	if (err) {
242 		dpaa2_eth_free_bufs(priv, ch->xdp.drop_bufs, ch->xdp.drop_cnt);
243 		ch->buf_count -= ch->xdp.drop_cnt;
244 	}
245 
246 	ch->xdp.drop_cnt = 0;
247 }
248 
249 static int dpaa2_eth_xdp_flush(struct dpaa2_eth_priv *priv,
250 			       struct dpaa2_eth_fq *fq,
251 			       struct dpaa2_eth_xdp_fds *xdp_fds)
252 {
253 	int total_enqueued = 0, retries = 0, enqueued;
254 	struct dpaa2_eth_drv_stats *percpu_extras;
255 	int num_fds, err, max_retries;
256 	struct dpaa2_fd *fds;
257 
258 	percpu_extras = this_cpu_ptr(priv->percpu_extras);
259 
260 	/* try to enqueue all the FDs until the max number of retries is hit */
261 	fds = xdp_fds->fds;
262 	num_fds = xdp_fds->num;
263 	max_retries = num_fds * DPAA2_ETH_ENQUEUE_RETRIES;
264 	while (total_enqueued < num_fds && retries < max_retries) {
265 		err = priv->enqueue(priv, fq, &fds[total_enqueued],
266 				    0, num_fds - total_enqueued, &enqueued);
267 		if (err == -EBUSY) {
268 			percpu_extras->tx_portal_busy += ++retries;
269 			continue;
270 		}
271 		total_enqueued += enqueued;
272 	}
273 	xdp_fds->num = 0;
274 
275 	return total_enqueued;
276 }
277 
278 static void dpaa2_eth_xdp_tx_flush(struct dpaa2_eth_priv *priv,
279 				   struct dpaa2_eth_channel *ch,
280 				   struct dpaa2_eth_fq *fq)
281 {
282 	struct rtnl_link_stats64 *percpu_stats;
283 	struct dpaa2_fd *fds;
284 	int enqueued, i;
285 
286 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
287 
288 	// enqueue the array of XDP_TX frames
289 	enqueued = dpaa2_eth_xdp_flush(priv, fq, &fq->xdp_tx_fds);
290 
291 	/* update statistics */
292 	percpu_stats->tx_packets += enqueued;
293 	fds = fq->xdp_tx_fds.fds;
294 	for (i = 0; i < enqueued; i++) {
295 		percpu_stats->tx_bytes += dpaa2_fd_get_len(&fds[i]);
296 		ch->stats.xdp_tx++;
297 	}
298 	for (i = enqueued; i < fq->xdp_tx_fds.num; i++) {
299 		dpaa2_eth_xdp_release_buf(priv, ch, dpaa2_fd_get_addr(&fds[i]));
300 		percpu_stats->tx_errors++;
301 		ch->stats.xdp_tx_err++;
302 	}
303 	fq->xdp_tx_fds.num = 0;
304 }
305 
306 static void dpaa2_eth_xdp_enqueue(struct dpaa2_eth_priv *priv,
307 				  struct dpaa2_eth_channel *ch,
308 				  struct dpaa2_fd *fd,
309 				  void *buf_start, u16 queue_id)
310 {
311 	struct dpaa2_faead *faead;
312 	struct dpaa2_fd *dest_fd;
313 	struct dpaa2_eth_fq *fq;
314 	u32 ctrl, frc;
315 
316 	/* Mark the egress frame hardware annotation area as valid */
317 	frc = dpaa2_fd_get_frc(fd);
318 	dpaa2_fd_set_frc(fd, frc | DPAA2_FD_FRC_FAEADV);
319 	dpaa2_fd_set_ctrl(fd, DPAA2_FD_CTRL_ASAL);
320 
321 	/* Instruct hardware to release the FD buffer directly into
322 	 * the buffer pool once transmission is completed, instead of
323 	 * sending a Tx confirmation frame to us
324 	 */
325 	ctrl = DPAA2_FAEAD_A4V | DPAA2_FAEAD_A2V | DPAA2_FAEAD_EBDDV;
326 	faead = dpaa2_get_faead(buf_start, false);
327 	faead->ctrl = cpu_to_le32(ctrl);
328 	faead->conf_fqid = 0;
329 
330 	fq = &priv->fq[queue_id];
331 	dest_fd = &fq->xdp_tx_fds.fds[fq->xdp_tx_fds.num++];
332 	memcpy(dest_fd, fd, sizeof(*dest_fd));
333 
334 	if (fq->xdp_tx_fds.num < DEV_MAP_BULK_SIZE)
335 		return;
336 
337 	dpaa2_eth_xdp_tx_flush(priv, ch, fq);
338 }
339 
340 static u32 dpaa2_eth_run_xdp(struct dpaa2_eth_priv *priv,
341 			     struct dpaa2_eth_channel *ch,
342 			     struct dpaa2_eth_fq *rx_fq,
343 			     struct dpaa2_fd *fd, void *vaddr)
344 {
345 	dma_addr_t addr = dpaa2_fd_get_addr(fd);
346 	struct bpf_prog *xdp_prog;
347 	struct xdp_buff xdp;
348 	u32 xdp_act = XDP_PASS;
349 	int err;
350 
351 	rcu_read_lock();
352 
353 	xdp_prog = READ_ONCE(ch->xdp.prog);
354 	if (!xdp_prog)
355 		goto out;
356 
357 	xdp.data = vaddr + dpaa2_fd_get_offset(fd);
358 	xdp.data_end = xdp.data + dpaa2_fd_get_len(fd);
359 	xdp.data_hard_start = xdp.data - XDP_PACKET_HEADROOM;
360 	xdp_set_data_meta_invalid(&xdp);
361 	xdp.rxq = &ch->xdp_rxq;
362 
363 	xdp.frame_sz = DPAA2_ETH_RX_BUF_RAW_SIZE -
364 		(dpaa2_fd_get_offset(fd) - XDP_PACKET_HEADROOM);
365 
366 	xdp_act = bpf_prog_run_xdp(xdp_prog, &xdp);
367 
368 	/* xdp.data pointer may have changed */
369 	dpaa2_fd_set_offset(fd, xdp.data - vaddr);
370 	dpaa2_fd_set_len(fd, xdp.data_end - xdp.data);
371 
372 	switch (xdp_act) {
373 	case XDP_PASS:
374 		break;
375 	case XDP_TX:
376 		dpaa2_eth_xdp_enqueue(priv, ch, fd, vaddr, rx_fq->flowid);
377 		break;
378 	default:
379 		bpf_warn_invalid_xdp_action(xdp_act);
380 		fallthrough;
381 	case XDP_ABORTED:
382 		trace_xdp_exception(priv->net_dev, xdp_prog, xdp_act);
383 		fallthrough;
384 	case XDP_DROP:
385 		dpaa2_eth_xdp_release_buf(priv, ch, addr);
386 		ch->stats.xdp_drop++;
387 		break;
388 	case XDP_REDIRECT:
389 		dma_unmap_page(priv->net_dev->dev.parent, addr,
390 			       priv->rx_buf_size, DMA_BIDIRECTIONAL);
391 		ch->buf_count--;
392 
393 		/* Allow redirect use of full headroom */
394 		xdp.data_hard_start = vaddr;
395 		xdp.frame_sz = DPAA2_ETH_RX_BUF_RAW_SIZE;
396 
397 		err = xdp_do_redirect(priv->net_dev, &xdp, xdp_prog);
398 		if (unlikely(err))
399 			ch->stats.xdp_drop++;
400 		else
401 			ch->stats.xdp_redirect++;
402 		break;
403 	}
404 
405 	ch->xdp.res |= xdp_act;
406 out:
407 	rcu_read_unlock();
408 	return xdp_act;
409 }
410 
411 /* Main Rx frame processing routine */
412 static void dpaa2_eth_rx(struct dpaa2_eth_priv *priv,
413 			 struct dpaa2_eth_channel *ch,
414 			 const struct dpaa2_fd *fd,
415 			 struct dpaa2_eth_fq *fq)
416 {
417 	dma_addr_t addr = dpaa2_fd_get_addr(fd);
418 	u8 fd_format = dpaa2_fd_get_format(fd);
419 	void *vaddr;
420 	struct sk_buff *skb;
421 	struct rtnl_link_stats64 *percpu_stats;
422 	struct dpaa2_eth_drv_stats *percpu_extras;
423 	struct device *dev = priv->net_dev->dev.parent;
424 	struct dpaa2_fas *fas;
425 	void *buf_data;
426 	u32 status = 0;
427 	u32 xdp_act;
428 
429 	/* Tracing point */
430 	trace_dpaa2_rx_fd(priv->net_dev, fd);
431 
432 	vaddr = dpaa2_iova_to_virt(priv->iommu_domain, addr);
433 	dma_sync_single_for_cpu(dev, addr, priv->rx_buf_size,
434 				DMA_BIDIRECTIONAL);
435 
436 	fas = dpaa2_get_fas(vaddr, false);
437 	prefetch(fas);
438 	buf_data = vaddr + dpaa2_fd_get_offset(fd);
439 	prefetch(buf_data);
440 
441 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
442 	percpu_extras = this_cpu_ptr(priv->percpu_extras);
443 
444 	if (fd_format == dpaa2_fd_single) {
445 		xdp_act = dpaa2_eth_run_xdp(priv, ch, fq, (struct dpaa2_fd *)fd, vaddr);
446 		if (xdp_act != XDP_PASS) {
447 			percpu_stats->rx_packets++;
448 			percpu_stats->rx_bytes += dpaa2_fd_get_len(fd);
449 			return;
450 		}
451 
452 		dma_unmap_page(dev, addr, priv->rx_buf_size,
453 			       DMA_BIDIRECTIONAL);
454 		skb = dpaa2_eth_build_linear_skb(ch, fd, vaddr);
455 	} else if (fd_format == dpaa2_fd_sg) {
456 		WARN_ON(priv->xdp_prog);
457 
458 		dma_unmap_page(dev, addr, priv->rx_buf_size,
459 			       DMA_BIDIRECTIONAL);
460 		skb = dpaa2_eth_build_frag_skb(priv, ch, buf_data);
461 		free_pages((unsigned long)vaddr, 0);
462 		percpu_extras->rx_sg_frames++;
463 		percpu_extras->rx_sg_bytes += dpaa2_fd_get_len(fd);
464 	} else {
465 		/* We don't support any other format */
466 		goto err_frame_format;
467 	}
468 
469 	if (unlikely(!skb))
470 		goto err_build_skb;
471 
472 	prefetch(skb->data);
473 
474 	/* Get the timestamp value */
475 	if (priv->rx_tstamp) {
476 		struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
477 		__le64 *ts = dpaa2_get_ts(vaddr, false);
478 		u64 ns;
479 
480 		memset(shhwtstamps, 0, sizeof(*shhwtstamps));
481 
482 		ns = DPAA2_PTP_CLK_PERIOD_NS * le64_to_cpup(ts);
483 		shhwtstamps->hwtstamp = ns_to_ktime(ns);
484 	}
485 
486 	/* Check if we need to validate the L4 csum */
487 	if (likely(dpaa2_fd_get_frc(fd) & DPAA2_FD_FRC_FASV)) {
488 		status = le32_to_cpu(fas->status);
489 		dpaa2_eth_validate_rx_csum(priv, status, skb);
490 	}
491 
492 	skb->protocol = eth_type_trans(skb, priv->net_dev);
493 	skb_record_rx_queue(skb, fq->flowid);
494 
495 	percpu_stats->rx_packets++;
496 	percpu_stats->rx_bytes += dpaa2_fd_get_len(fd);
497 
498 	list_add_tail(&skb->list, ch->rx_list);
499 
500 	return;
501 
502 err_build_skb:
503 	dpaa2_eth_free_rx_fd(priv, fd, vaddr);
504 err_frame_format:
505 	percpu_stats->rx_dropped++;
506 }
507 
508 /* Consume all frames pull-dequeued into the store. This is the simplest way to
509  * make sure we don't accidentally issue another volatile dequeue which would
510  * overwrite (leak) frames already in the store.
511  *
512  * Observance of NAPI budget is not our concern, leaving that to the caller.
513  */
514 static int dpaa2_eth_consume_frames(struct dpaa2_eth_channel *ch,
515 				    struct dpaa2_eth_fq **src)
516 {
517 	struct dpaa2_eth_priv *priv = ch->priv;
518 	struct dpaa2_eth_fq *fq = NULL;
519 	struct dpaa2_dq *dq;
520 	const struct dpaa2_fd *fd;
521 	int cleaned = 0, retries = 0;
522 	int is_last;
523 
524 	do {
525 		dq = dpaa2_io_store_next(ch->store, &is_last);
526 		if (unlikely(!dq)) {
527 			/* If we're here, we *must* have placed a
528 			 * volatile dequeue comnmand, so keep reading through
529 			 * the store until we get some sort of valid response
530 			 * token (either a valid frame or an "empty dequeue")
531 			 */
532 			if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES) {
533 				netdev_err_once(priv->net_dev,
534 						"Unable to read a valid dequeue response\n");
535 				return -ETIMEDOUT;
536 			}
537 			continue;
538 		}
539 
540 		fd = dpaa2_dq_fd(dq);
541 		fq = (struct dpaa2_eth_fq *)(uintptr_t)dpaa2_dq_fqd_ctx(dq);
542 
543 		fq->consume(priv, ch, fd, fq);
544 		cleaned++;
545 		retries = 0;
546 	} while (!is_last);
547 
548 	if (!cleaned)
549 		return 0;
550 
551 	fq->stats.frames += cleaned;
552 	ch->stats.frames += cleaned;
553 
554 	/* A dequeue operation only pulls frames from a single queue
555 	 * into the store. Return the frame queue as an out param.
556 	 */
557 	if (src)
558 		*src = fq;
559 
560 	return cleaned;
561 }
562 
563 /* Configure the egress frame annotation for timestamp update */
564 static void dpaa2_eth_enable_tx_tstamp(struct dpaa2_fd *fd, void *buf_start)
565 {
566 	struct dpaa2_faead *faead;
567 	u32 ctrl, frc;
568 
569 	/* Mark the egress frame annotation area as valid */
570 	frc = dpaa2_fd_get_frc(fd);
571 	dpaa2_fd_set_frc(fd, frc | DPAA2_FD_FRC_FAEADV);
572 
573 	/* Set hardware annotation size */
574 	ctrl = dpaa2_fd_get_ctrl(fd);
575 	dpaa2_fd_set_ctrl(fd, ctrl | DPAA2_FD_CTRL_ASAL);
576 
577 	/* enable UPD (update prepanded data) bit in FAEAD field of
578 	 * hardware frame annotation area
579 	 */
580 	ctrl = DPAA2_FAEAD_A2V | DPAA2_FAEAD_UPDV | DPAA2_FAEAD_UPD;
581 	faead = dpaa2_get_faead(buf_start, true);
582 	faead->ctrl = cpu_to_le32(ctrl);
583 }
584 
585 /* Create a frame descriptor based on a fragmented skb */
586 static int dpaa2_eth_build_sg_fd(struct dpaa2_eth_priv *priv,
587 				 struct sk_buff *skb,
588 				 struct dpaa2_fd *fd)
589 {
590 	struct device *dev = priv->net_dev->dev.parent;
591 	void *sgt_buf = NULL;
592 	dma_addr_t addr;
593 	int nr_frags = skb_shinfo(skb)->nr_frags;
594 	struct dpaa2_sg_entry *sgt;
595 	int i, err;
596 	int sgt_buf_size;
597 	struct scatterlist *scl, *crt_scl;
598 	int num_sg;
599 	int num_dma_bufs;
600 	struct dpaa2_eth_swa *swa;
601 
602 	/* Create and map scatterlist.
603 	 * We don't advertise NETIF_F_FRAGLIST, so skb_to_sgvec() will not have
604 	 * to go beyond nr_frags+1.
605 	 * Note: We don't support chained scatterlists
606 	 */
607 	if (unlikely(PAGE_SIZE / sizeof(struct scatterlist) < nr_frags + 1))
608 		return -EINVAL;
609 
610 	scl = kcalloc(nr_frags + 1, sizeof(struct scatterlist), GFP_ATOMIC);
611 	if (unlikely(!scl))
612 		return -ENOMEM;
613 
614 	sg_init_table(scl, nr_frags + 1);
615 	num_sg = skb_to_sgvec(skb, scl, 0, skb->len);
616 	if (unlikely(num_sg < 0)) {
617 		err = -ENOMEM;
618 		goto dma_map_sg_failed;
619 	}
620 	num_dma_bufs = dma_map_sg(dev, scl, num_sg, DMA_BIDIRECTIONAL);
621 	if (unlikely(!num_dma_bufs)) {
622 		err = -ENOMEM;
623 		goto dma_map_sg_failed;
624 	}
625 
626 	/* Prepare the HW SGT structure */
627 	sgt_buf_size = priv->tx_data_offset +
628 		       sizeof(struct dpaa2_sg_entry) *  num_dma_bufs;
629 	sgt_buf = napi_alloc_frag(sgt_buf_size + DPAA2_ETH_TX_BUF_ALIGN);
630 	if (unlikely(!sgt_buf)) {
631 		err = -ENOMEM;
632 		goto sgt_buf_alloc_failed;
633 	}
634 	sgt_buf = PTR_ALIGN(sgt_buf, DPAA2_ETH_TX_BUF_ALIGN);
635 	memset(sgt_buf, 0, sgt_buf_size);
636 
637 	sgt = (struct dpaa2_sg_entry *)(sgt_buf + priv->tx_data_offset);
638 
639 	/* Fill in the HW SGT structure.
640 	 *
641 	 * sgt_buf is zeroed out, so the following fields are implicit
642 	 * in all sgt entries:
643 	 *   - offset is 0
644 	 *   - format is 'dpaa2_sg_single'
645 	 */
646 	for_each_sg(scl, crt_scl, num_dma_bufs, i) {
647 		dpaa2_sg_set_addr(&sgt[i], sg_dma_address(crt_scl));
648 		dpaa2_sg_set_len(&sgt[i], sg_dma_len(crt_scl));
649 	}
650 	dpaa2_sg_set_final(&sgt[i - 1], true);
651 
652 	/* Store the skb backpointer in the SGT buffer.
653 	 * Fit the scatterlist and the number of buffers alongside the
654 	 * skb backpointer in the software annotation area. We'll need
655 	 * all of them on Tx Conf.
656 	 */
657 	swa = (struct dpaa2_eth_swa *)sgt_buf;
658 	swa->type = DPAA2_ETH_SWA_SG;
659 	swa->sg.skb = skb;
660 	swa->sg.scl = scl;
661 	swa->sg.num_sg = num_sg;
662 	swa->sg.sgt_size = sgt_buf_size;
663 
664 	/* Separately map the SGT buffer */
665 	addr = dma_map_single(dev, sgt_buf, sgt_buf_size, DMA_BIDIRECTIONAL);
666 	if (unlikely(dma_mapping_error(dev, addr))) {
667 		err = -ENOMEM;
668 		goto dma_map_single_failed;
669 	}
670 	dpaa2_fd_set_offset(fd, priv->tx_data_offset);
671 	dpaa2_fd_set_format(fd, dpaa2_fd_sg);
672 	dpaa2_fd_set_addr(fd, addr);
673 	dpaa2_fd_set_len(fd, skb->len);
674 	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
675 
676 	if (priv->tx_tstamp && skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)
677 		dpaa2_eth_enable_tx_tstamp(fd, sgt_buf);
678 
679 	return 0;
680 
681 dma_map_single_failed:
682 	skb_free_frag(sgt_buf);
683 sgt_buf_alloc_failed:
684 	dma_unmap_sg(dev, scl, num_sg, DMA_BIDIRECTIONAL);
685 dma_map_sg_failed:
686 	kfree(scl);
687 	return err;
688 }
689 
690 /* Create a SG frame descriptor based on a linear skb.
691  *
692  * This function is used on the Tx path when the skb headroom is not large
693  * enough for the HW requirements, thus instead of realloc-ing the skb we
694  * create a SG frame descriptor with only one entry.
695  */
696 static int dpaa2_eth_build_sg_fd_single_buf(struct dpaa2_eth_priv *priv,
697 					    struct sk_buff *skb,
698 					    struct dpaa2_fd *fd)
699 {
700 	struct device *dev = priv->net_dev->dev.parent;
701 	struct dpaa2_eth_sgt_cache *sgt_cache;
702 	struct dpaa2_sg_entry *sgt;
703 	struct dpaa2_eth_swa *swa;
704 	dma_addr_t addr, sgt_addr;
705 	void *sgt_buf = NULL;
706 	int sgt_buf_size;
707 	int err;
708 
709 	/* Prepare the HW SGT structure */
710 	sgt_cache = this_cpu_ptr(priv->sgt_cache);
711 	sgt_buf_size = priv->tx_data_offset + sizeof(struct dpaa2_sg_entry);
712 
713 	if (sgt_cache->count == 0)
714 		sgt_buf = kzalloc(sgt_buf_size + DPAA2_ETH_TX_BUF_ALIGN,
715 				  GFP_ATOMIC);
716 	else
717 		sgt_buf = sgt_cache->buf[--sgt_cache->count];
718 	if (unlikely(!sgt_buf))
719 		return -ENOMEM;
720 
721 	sgt_buf = PTR_ALIGN(sgt_buf, DPAA2_ETH_TX_BUF_ALIGN);
722 	sgt = (struct dpaa2_sg_entry *)(sgt_buf + priv->tx_data_offset);
723 
724 	addr = dma_map_single(dev, skb->data, skb->len, DMA_BIDIRECTIONAL);
725 	if (unlikely(dma_mapping_error(dev, addr))) {
726 		err = -ENOMEM;
727 		goto data_map_failed;
728 	}
729 
730 	/* Fill in the HW SGT structure */
731 	dpaa2_sg_set_addr(sgt, addr);
732 	dpaa2_sg_set_len(sgt, skb->len);
733 	dpaa2_sg_set_final(sgt, true);
734 
735 	/* Store the skb backpointer in the SGT buffer */
736 	swa = (struct dpaa2_eth_swa *)sgt_buf;
737 	swa->type = DPAA2_ETH_SWA_SINGLE;
738 	swa->single.skb = skb;
739 	swa->sg.sgt_size = sgt_buf_size;
740 
741 	/* Separately map the SGT buffer */
742 	sgt_addr = dma_map_single(dev, sgt_buf, sgt_buf_size, DMA_BIDIRECTIONAL);
743 	if (unlikely(dma_mapping_error(dev, sgt_addr))) {
744 		err = -ENOMEM;
745 		goto sgt_map_failed;
746 	}
747 
748 	dpaa2_fd_set_offset(fd, priv->tx_data_offset);
749 	dpaa2_fd_set_format(fd, dpaa2_fd_sg);
750 	dpaa2_fd_set_addr(fd, sgt_addr);
751 	dpaa2_fd_set_len(fd, skb->len);
752 	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
753 
754 	if (priv->tx_tstamp && skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)
755 		dpaa2_eth_enable_tx_tstamp(fd, sgt_buf);
756 
757 	return 0;
758 
759 sgt_map_failed:
760 	dma_unmap_single(dev, addr, skb->len, DMA_BIDIRECTIONAL);
761 data_map_failed:
762 	if (sgt_cache->count >= DPAA2_ETH_SGT_CACHE_SIZE)
763 		kfree(sgt_buf);
764 	else
765 		sgt_cache->buf[sgt_cache->count++] = sgt_buf;
766 
767 	return err;
768 }
769 
770 /* Create a frame descriptor based on a linear skb */
771 static int dpaa2_eth_build_single_fd(struct dpaa2_eth_priv *priv,
772 				     struct sk_buff *skb,
773 				     struct dpaa2_fd *fd)
774 {
775 	struct device *dev = priv->net_dev->dev.parent;
776 	u8 *buffer_start, *aligned_start;
777 	struct dpaa2_eth_swa *swa;
778 	dma_addr_t addr;
779 
780 	buffer_start = skb->data - dpaa2_eth_needed_headroom(priv, skb);
781 
782 	/* If there's enough room to align the FD address, do it.
783 	 * It will help hardware optimize accesses.
784 	 */
785 	aligned_start = PTR_ALIGN(buffer_start - DPAA2_ETH_TX_BUF_ALIGN,
786 				  DPAA2_ETH_TX_BUF_ALIGN);
787 	if (aligned_start >= skb->head)
788 		buffer_start = aligned_start;
789 
790 	/* Store a backpointer to the skb at the beginning of the buffer
791 	 * (in the private data area) such that we can release it
792 	 * on Tx confirm
793 	 */
794 	swa = (struct dpaa2_eth_swa *)buffer_start;
795 	swa->type = DPAA2_ETH_SWA_SINGLE;
796 	swa->single.skb = skb;
797 
798 	addr = dma_map_single(dev, buffer_start,
799 			      skb_tail_pointer(skb) - buffer_start,
800 			      DMA_BIDIRECTIONAL);
801 	if (unlikely(dma_mapping_error(dev, addr)))
802 		return -ENOMEM;
803 
804 	dpaa2_fd_set_addr(fd, addr);
805 	dpaa2_fd_set_offset(fd, (u16)(skb->data - buffer_start));
806 	dpaa2_fd_set_len(fd, skb->len);
807 	dpaa2_fd_set_format(fd, dpaa2_fd_single);
808 	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
809 
810 	if (priv->tx_tstamp && skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)
811 		dpaa2_eth_enable_tx_tstamp(fd, buffer_start);
812 
813 	return 0;
814 }
815 
816 /* FD freeing routine on the Tx path
817  *
818  * DMA-unmap and free FD and possibly SGT buffer allocated on Tx. The skb
819  * back-pointed to is also freed.
820  * This can be called either from dpaa2_eth_tx_conf() or on the error path of
821  * dpaa2_eth_tx().
822  */
823 static void dpaa2_eth_free_tx_fd(const struct dpaa2_eth_priv *priv,
824 				 struct dpaa2_eth_fq *fq,
825 				 const struct dpaa2_fd *fd, bool in_napi)
826 {
827 	struct device *dev = priv->net_dev->dev.parent;
828 	dma_addr_t fd_addr, sg_addr;
829 	struct sk_buff *skb = NULL;
830 	unsigned char *buffer_start;
831 	struct dpaa2_eth_swa *swa;
832 	u8 fd_format = dpaa2_fd_get_format(fd);
833 	u32 fd_len = dpaa2_fd_get_len(fd);
834 
835 	struct dpaa2_eth_sgt_cache *sgt_cache;
836 	struct dpaa2_sg_entry *sgt;
837 
838 	fd_addr = dpaa2_fd_get_addr(fd);
839 	buffer_start = dpaa2_iova_to_virt(priv->iommu_domain, fd_addr);
840 	swa = (struct dpaa2_eth_swa *)buffer_start;
841 
842 	if (fd_format == dpaa2_fd_single) {
843 		if (swa->type == DPAA2_ETH_SWA_SINGLE) {
844 			skb = swa->single.skb;
845 			/* Accessing the skb buffer is safe before dma unmap,
846 			 * because we didn't map the actual skb shell.
847 			 */
848 			dma_unmap_single(dev, fd_addr,
849 					 skb_tail_pointer(skb) - buffer_start,
850 					 DMA_BIDIRECTIONAL);
851 		} else {
852 			WARN_ONCE(swa->type != DPAA2_ETH_SWA_XDP, "Wrong SWA type");
853 			dma_unmap_single(dev, fd_addr, swa->xdp.dma_size,
854 					 DMA_BIDIRECTIONAL);
855 		}
856 	} else if (fd_format == dpaa2_fd_sg) {
857 		if (swa->type == DPAA2_ETH_SWA_SG) {
858 			skb = swa->sg.skb;
859 
860 			/* Unmap the scatterlist */
861 			dma_unmap_sg(dev, swa->sg.scl, swa->sg.num_sg,
862 				     DMA_BIDIRECTIONAL);
863 			kfree(swa->sg.scl);
864 
865 			/* Unmap the SGT buffer */
866 			dma_unmap_single(dev, fd_addr, swa->sg.sgt_size,
867 					 DMA_BIDIRECTIONAL);
868 		} else {
869 			skb = swa->single.skb;
870 
871 			/* Unmap the SGT Buffer */
872 			dma_unmap_single(dev, fd_addr, swa->single.sgt_size,
873 					 DMA_BIDIRECTIONAL);
874 
875 			sgt = (struct dpaa2_sg_entry *)(buffer_start +
876 							priv->tx_data_offset);
877 			sg_addr = dpaa2_sg_get_addr(sgt);
878 			dma_unmap_single(dev, sg_addr, skb->len, DMA_BIDIRECTIONAL);
879 		}
880 	} else {
881 		netdev_dbg(priv->net_dev, "Invalid FD format\n");
882 		return;
883 	}
884 
885 	if (swa->type != DPAA2_ETH_SWA_XDP && in_napi) {
886 		fq->dq_frames++;
887 		fq->dq_bytes += fd_len;
888 	}
889 
890 	if (swa->type == DPAA2_ETH_SWA_XDP) {
891 		xdp_return_frame(swa->xdp.xdpf);
892 		return;
893 	}
894 
895 	/* Get the timestamp value */
896 	if (priv->tx_tstamp && skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) {
897 		struct skb_shared_hwtstamps shhwtstamps;
898 		__le64 *ts = dpaa2_get_ts(buffer_start, true);
899 		u64 ns;
900 
901 		memset(&shhwtstamps, 0, sizeof(shhwtstamps));
902 
903 		ns = DPAA2_PTP_CLK_PERIOD_NS * le64_to_cpup(ts);
904 		shhwtstamps.hwtstamp = ns_to_ktime(ns);
905 		skb_tstamp_tx(skb, &shhwtstamps);
906 	}
907 
908 	/* Free SGT buffer allocated on tx */
909 	if (fd_format != dpaa2_fd_single) {
910 		sgt_cache = this_cpu_ptr(priv->sgt_cache);
911 		if (swa->type == DPAA2_ETH_SWA_SG) {
912 			skb_free_frag(buffer_start);
913 		} else {
914 			if (sgt_cache->count >= DPAA2_ETH_SGT_CACHE_SIZE)
915 				kfree(buffer_start);
916 			else
917 				sgt_cache->buf[sgt_cache->count++] = buffer_start;
918 		}
919 	}
920 
921 	/* Move on with skb release */
922 	napi_consume_skb(skb, in_napi);
923 }
924 
925 static netdev_tx_t dpaa2_eth_tx(struct sk_buff *skb, struct net_device *net_dev)
926 {
927 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
928 	struct dpaa2_fd fd;
929 	struct rtnl_link_stats64 *percpu_stats;
930 	struct dpaa2_eth_drv_stats *percpu_extras;
931 	struct dpaa2_eth_fq *fq;
932 	struct netdev_queue *nq;
933 	u16 queue_mapping;
934 	unsigned int needed_headroom;
935 	u32 fd_len;
936 	u8 prio = 0;
937 	int err, i;
938 
939 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
940 	percpu_extras = this_cpu_ptr(priv->percpu_extras);
941 
942 	needed_headroom = dpaa2_eth_needed_headroom(priv, skb);
943 
944 	/* We'll be holding a back-reference to the skb until Tx Confirmation;
945 	 * we don't want that overwritten by a concurrent Tx with a cloned skb.
946 	 */
947 	skb = skb_unshare(skb, GFP_ATOMIC);
948 	if (unlikely(!skb)) {
949 		/* skb_unshare() has already freed the skb */
950 		percpu_stats->tx_dropped++;
951 		return NETDEV_TX_OK;
952 	}
953 
954 	/* Setup the FD fields */
955 	memset(&fd, 0, sizeof(fd));
956 
957 	if (skb_is_nonlinear(skb)) {
958 		err = dpaa2_eth_build_sg_fd(priv, skb, &fd);
959 		percpu_extras->tx_sg_frames++;
960 		percpu_extras->tx_sg_bytes += skb->len;
961 	} else if (skb_headroom(skb) < needed_headroom) {
962 		err = dpaa2_eth_build_sg_fd_single_buf(priv, skb, &fd);
963 		percpu_extras->tx_sg_frames++;
964 		percpu_extras->tx_sg_bytes += skb->len;
965 		percpu_extras->tx_converted_sg_frames++;
966 		percpu_extras->tx_converted_sg_bytes += skb->len;
967 	} else {
968 		err = dpaa2_eth_build_single_fd(priv, skb, &fd);
969 	}
970 
971 	if (unlikely(err)) {
972 		percpu_stats->tx_dropped++;
973 		goto err_build_fd;
974 	}
975 
976 	/* Tracing point */
977 	trace_dpaa2_tx_fd(net_dev, &fd);
978 
979 	/* TxConf FQ selection relies on queue id from the stack.
980 	 * In case of a forwarded frame from another DPNI interface, we choose
981 	 * a queue affined to the same core that processed the Rx frame
982 	 */
983 	queue_mapping = skb_get_queue_mapping(skb);
984 
985 	if (net_dev->num_tc) {
986 		prio = netdev_txq_to_tc(net_dev, queue_mapping);
987 		/* Hardware interprets priority level 0 as being the highest,
988 		 * so we need to do a reverse mapping to the netdev tc index
989 		 */
990 		prio = net_dev->num_tc - prio - 1;
991 		/* We have only one FQ array entry for all Tx hardware queues
992 		 * with the same flow id (but different priority levels)
993 		 */
994 		queue_mapping %= dpaa2_eth_queue_count(priv);
995 	}
996 	fq = &priv->fq[queue_mapping];
997 
998 	fd_len = dpaa2_fd_get_len(&fd);
999 	nq = netdev_get_tx_queue(net_dev, queue_mapping);
1000 	netdev_tx_sent_queue(nq, fd_len);
1001 
1002 	/* Everything that happens after this enqueues might race with
1003 	 * the Tx confirmation callback for this frame
1004 	 */
1005 	for (i = 0; i < DPAA2_ETH_ENQUEUE_RETRIES; i++) {
1006 		err = priv->enqueue(priv, fq, &fd, prio, 1, NULL);
1007 		if (err != -EBUSY)
1008 			break;
1009 	}
1010 	percpu_extras->tx_portal_busy += i;
1011 	if (unlikely(err < 0)) {
1012 		percpu_stats->tx_errors++;
1013 		/* Clean up everything, including freeing the skb */
1014 		dpaa2_eth_free_tx_fd(priv, fq, &fd, false);
1015 		netdev_tx_completed_queue(nq, 1, fd_len);
1016 	} else {
1017 		percpu_stats->tx_packets++;
1018 		percpu_stats->tx_bytes += fd_len;
1019 	}
1020 
1021 	return NETDEV_TX_OK;
1022 
1023 err_build_fd:
1024 	dev_kfree_skb(skb);
1025 
1026 	return NETDEV_TX_OK;
1027 }
1028 
1029 /* Tx confirmation frame processing routine */
1030 static void dpaa2_eth_tx_conf(struct dpaa2_eth_priv *priv,
1031 			      struct dpaa2_eth_channel *ch __always_unused,
1032 			      const struct dpaa2_fd *fd,
1033 			      struct dpaa2_eth_fq *fq)
1034 {
1035 	struct rtnl_link_stats64 *percpu_stats;
1036 	struct dpaa2_eth_drv_stats *percpu_extras;
1037 	u32 fd_len = dpaa2_fd_get_len(fd);
1038 	u32 fd_errors;
1039 
1040 	/* Tracing point */
1041 	trace_dpaa2_tx_conf_fd(priv->net_dev, fd);
1042 
1043 	percpu_extras = this_cpu_ptr(priv->percpu_extras);
1044 	percpu_extras->tx_conf_frames++;
1045 	percpu_extras->tx_conf_bytes += fd_len;
1046 
1047 	/* Check frame errors in the FD field */
1048 	fd_errors = dpaa2_fd_get_ctrl(fd) & DPAA2_FD_TX_ERR_MASK;
1049 	dpaa2_eth_free_tx_fd(priv, fq, fd, true);
1050 
1051 	if (likely(!fd_errors))
1052 		return;
1053 
1054 	if (net_ratelimit())
1055 		netdev_dbg(priv->net_dev, "TX frame FD error: 0x%08x\n",
1056 			   fd_errors);
1057 
1058 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
1059 	/* Tx-conf logically pertains to the egress path. */
1060 	percpu_stats->tx_errors++;
1061 }
1062 
1063 static int dpaa2_eth_set_rx_csum(struct dpaa2_eth_priv *priv, bool enable)
1064 {
1065 	int err;
1066 
1067 	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1068 			       DPNI_OFF_RX_L3_CSUM, enable);
1069 	if (err) {
1070 		netdev_err(priv->net_dev,
1071 			   "dpni_set_offload(RX_L3_CSUM) failed\n");
1072 		return err;
1073 	}
1074 
1075 	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1076 			       DPNI_OFF_RX_L4_CSUM, enable);
1077 	if (err) {
1078 		netdev_err(priv->net_dev,
1079 			   "dpni_set_offload(RX_L4_CSUM) failed\n");
1080 		return err;
1081 	}
1082 
1083 	return 0;
1084 }
1085 
1086 static int dpaa2_eth_set_tx_csum(struct dpaa2_eth_priv *priv, bool enable)
1087 {
1088 	int err;
1089 
1090 	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1091 			       DPNI_OFF_TX_L3_CSUM, enable);
1092 	if (err) {
1093 		netdev_err(priv->net_dev, "dpni_set_offload(TX_L3_CSUM) failed\n");
1094 		return err;
1095 	}
1096 
1097 	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1098 			       DPNI_OFF_TX_L4_CSUM, enable);
1099 	if (err) {
1100 		netdev_err(priv->net_dev, "dpni_set_offload(TX_L4_CSUM) failed\n");
1101 		return err;
1102 	}
1103 
1104 	return 0;
1105 }
1106 
1107 /* Perform a single release command to add buffers
1108  * to the specified buffer pool
1109  */
1110 static int dpaa2_eth_add_bufs(struct dpaa2_eth_priv *priv,
1111 			      struct dpaa2_eth_channel *ch, u16 bpid)
1112 {
1113 	struct device *dev = priv->net_dev->dev.parent;
1114 	u64 buf_array[DPAA2_ETH_BUFS_PER_CMD];
1115 	struct page *page;
1116 	dma_addr_t addr;
1117 	int retries = 0;
1118 	int i, err;
1119 
1120 	for (i = 0; i < DPAA2_ETH_BUFS_PER_CMD; i++) {
1121 		/* Allocate buffer visible to WRIOP + skb shared info +
1122 		 * alignment padding
1123 		 */
1124 		/* allocate one page for each Rx buffer. WRIOP sees
1125 		 * the entire page except for a tailroom reserved for
1126 		 * skb shared info
1127 		 */
1128 		page = dev_alloc_pages(0);
1129 		if (!page)
1130 			goto err_alloc;
1131 
1132 		addr = dma_map_page(dev, page, 0, priv->rx_buf_size,
1133 				    DMA_BIDIRECTIONAL);
1134 		if (unlikely(dma_mapping_error(dev, addr)))
1135 			goto err_map;
1136 
1137 		buf_array[i] = addr;
1138 
1139 		/* tracing point */
1140 		trace_dpaa2_eth_buf_seed(priv->net_dev,
1141 					 page, DPAA2_ETH_RX_BUF_RAW_SIZE,
1142 					 addr, priv->rx_buf_size,
1143 					 bpid);
1144 	}
1145 
1146 release_bufs:
1147 	/* In case the portal is busy, retry until successful */
1148 	while ((err = dpaa2_io_service_release(ch->dpio, bpid,
1149 					       buf_array, i)) == -EBUSY) {
1150 		if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES)
1151 			break;
1152 		cpu_relax();
1153 	}
1154 
1155 	/* If release command failed, clean up and bail out;
1156 	 * not much else we can do about it
1157 	 */
1158 	if (err) {
1159 		dpaa2_eth_free_bufs(priv, buf_array, i);
1160 		return 0;
1161 	}
1162 
1163 	return i;
1164 
1165 err_map:
1166 	__free_pages(page, 0);
1167 err_alloc:
1168 	/* If we managed to allocate at least some buffers,
1169 	 * release them to hardware
1170 	 */
1171 	if (i)
1172 		goto release_bufs;
1173 
1174 	return 0;
1175 }
1176 
1177 static int dpaa2_eth_seed_pool(struct dpaa2_eth_priv *priv, u16 bpid)
1178 {
1179 	int i, j;
1180 	int new_count;
1181 
1182 	for (j = 0; j < priv->num_channels; j++) {
1183 		for (i = 0; i < DPAA2_ETH_NUM_BUFS;
1184 		     i += DPAA2_ETH_BUFS_PER_CMD) {
1185 			new_count = dpaa2_eth_add_bufs(priv, priv->channel[j], bpid);
1186 			priv->channel[j]->buf_count += new_count;
1187 
1188 			if (new_count < DPAA2_ETH_BUFS_PER_CMD) {
1189 				return -ENOMEM;
1190 			}
1191 		}
1192 	}
1193 
1194 	return 0;
1195 }
1196 
1197 /**
1198  * Drain the specified number of buffers from the DPNI's private buffer pool.
1199  * @count must not exceeed DPAA2_ETH_BUFS_PER_CMD
1200  */
1201 static void dpaa2_eth_drain_bufs(struct dpaa2_eth_priv *priv, int count)
1202 {
1203 	u64 buf_array[DPAA2_ETH_BUFS_PER_CMD];
1204 	int retries = 0;
1205 	int ret;
1206 
1207 	do {
1208 		ret = dpaa2_io_service_acquire(NULL, priv->bpid,
1209 					       buf_array, count);
1210 		if (ret < 0) {
1211 			if (ret == -EBUSY &&
1212 			    retries++ < DPAA2_ETH_SWP_BUSY_RETRIES)
1213 				continue;
1214 			netdev_err(priv->net_dev, "dpaa2_io_service_acquire() failed\n");
1215 			return;
1216 		}
1217 		dpaa2_eth_free_bufs(priv, buf_array, ret);
1218 		retries = 0;
1219 	} while (ret);
1220 }
1221 
1222 static void dpaa2_eth_drain_pool(struct dpaa2_eth_priv *priv)
1223 {
1224 	int i;
1225 
1226 	dpaa2_eth_drain_bufs(priv, DPAA2_ETH_BUFS_PER_CMD);
1227 	dpaa2_eth_drain_bufs(priv, 1);
1228 
1229 	for (i = 0; i < priv->num_channels; i++)
1230 		priv->channel[i]->buf_count = 0;
1231 }
1232 
1233 /* Function is called from softirq context only, so we don't need to guard
1234  * the access to percpu count
1235  */
1236 static int dpaa2_eth_refill_pool(struct dpaa2_eth_priv *priv,
1237 				 struct dpaa2_eth_channel *ch,
1238 				 u16 bpid)
1239 {
1240 	int new_count;
1241 
1242 	if (likely(ch->buf_count >= DPAA2_ETH_REFILL_THRESH))
1243 		return 0;
1244 
1245 	do {
1246 		new_count = dpaa2_eth_add_bufs(priv, ch, bpid);
1247 		if (unlikely(!new_count)) {
1248 			/* Out of memory; abort for now, we'll try later on */
1249 			break;
1250 		}
1251 		ch->buf_count += new_count;
1252 	} while (ch->buf_count < DPAA2_ETH_NUM_BUFS);
1253 
1254 	if (unlikely(ch->buf_count < DPAA2_ETH_NUM_BUFS))
1255 		return -ENOMEM;
1256 
1257 	return 0;
1258 }
1259 
1260 static void dpaa2_eth_sgt_cache_drain(struct dpaa2_eth_priv *priv)
1261 {
1262 	struct dpaa2_eth_sgt_cache *sgt_cache;
1263 	u16 count;
1264 	int k, i;
1265 
1266 	for_each_possible_cpu(k) {
1267 		sgt_cache = per_cpu_ptr(priv->sgt_cache, k);
1268 		count = sgt_cache->count;
1269 
1270 		for (i = 0; i < count; i++)
1271 			kfree(sgt_cache->buf[i]);
1272 		sgt_cache->count = 0;
1273 	}
1274 }
1275 
1276 static int dpaa2_eth_pull_channel(struct dpaa2_eth_channel *ch)
1277 {
1278 	int err;
1279 	int dequeues = -1;
1280 
1281 	/* Retry while portal is busy */
1282 	do {
1283 		err = dpaa2_io_service_pull_channel(ch->dpio, ch->ch_id,
1284 						    ch->store);
1285 		dequeues++;
1286 		cpu_relax();
1287 	} while (err == -EBUSY && dequeues < DPAA2_ETH_SWP_BUSY_RETRIES);
1288 
1289 	ch->stats.dequeue_portal_busy += dequeues;
1290 	if (unlikely(err))
1291 		ch->stats.pull_err++;
1292 
1293 	return err;
1294 }
1295 
1296 /* NAPI poll routine
1297  *
1298  * Frames are dequeued from the QMan channel associated with this NAPI context.
1299  * Rx, Tx confirmation and (if configured) Rx error frames all count
1300  * towards the NAPI budget.
1301  */
1302 static int dpaa2_eth_poll(struct napi_struct *napi, int budget)
1303 {
1304 	struct dpaa2_eth_channel *ch;
1305 	struct dpaa2_eth_priv *priv;
1306 	int rx_cleaned = 0, txconf_cleaned = 0;
1307 	struct dpaa2_eth_fq *fq, *txc_fq = NULL;
1308 	struct netdev_queue *nq;
1309 	int store_cleaned, work_done;
1310 	struct list_head rx_list;
1311 	int retries = 0;
1312 	u16 flowid;
1313 	int err;
1314 
1315 	ch = container_of(napi, struct dpaa2_eth_channel, napi);
1316 	ch->xdp.res = 0;
1317 	priv = ch->priv;
1318 
1319 	INIT_LIST_HEAD(&rx_list);
1320 	ch->rx_list = &rx_list;
1321 
1322 	do {
1323 		err = dpaa2_eth_pull_channel(ch);
1324 		if (unlikely(err))
1325 			break;
1326 
1327 		/* Refill pool if appropriate */
1328 		dpaa2_eth_refill_pool(priv, ch, priv->bpid);
1329 
1330 		store_cleaned = dpaa2_eth_consume_frames(ch, &fq);
1331 		if (store_cleaned <= 0)
1332 			break;
1333 		if (fq->type == DPAA2_RX_FQ) {
1334 			rx_cleaned += store_cleaned;
1335 			flowid = fq->flowid;
1336 		} else {
1337 			txconf_cleaned += store_cleaned;
1338 			/* We have a single Tx conf FQ on this channel */
1339 			txc_fq = fq;
1340 		}
1341 
1342 		/* If we either consumed the whole NAPI budget with Rx frames
1343 		 * or we reached the Tx confirmations threshold, we're done.
1344 		 */
1345 		if (rx_cleaned >= budget ||
1346 		    txconf_cleaned >= DPAA2_ETH_TXCONF_PER_NAPI) {
1347 			work_done = budget;
1348 			goto out;
1349 		}
1350 	} while (store_cleaned);
1351 
1352 	/* We didn't consume the entire budget, so finish napi and
1353 	 * re-enable data availability notifications
1354 	 */
1355 	napi_complete_done(napi, rx_cleaned);
1356 	do {
1357 		err = dpaa2_io_service_rearm(ch->dpio, &ch->nctx);
1358 		cpu_relax();
1359 	} while (err == -EBUSY && retries++ < DPAA2_ETH_SWP_BUSY_RETRIES);
1360 	WARN_ONCE(err, "CDAN notifications rearm failed on core %d",
1361 		  ch->nctx.desired_cpu);
1362 
1363 	work_done = max(rx_cleaned, 1);
1364 
1365 out:
1366 	netif_receive_skb_list(ch->rx_list);
1367 
1368 	if (txc_fq && txc_fq->dq_frames) {
1369 		nq = netdev_get_tx_queue(priv->net_dev, txc_fq->flowid);
1370 		netdev_tx_completed_queue(nq, txc_fq->dq_frames,
1371 					  txc_fq->dq_bytes);
1372 		txc_fq->dq_frames = 0;
1373 		txc_fq->dq_bytes = 0;
1374 	}
1375 
1376 	if (ch->xdp.res & XDP_REDIRECT)
1377 		xdp_do_flush_map();
1378 	else if (rx_cleaned && ch->xdp.res & XDP_TX)
1379 		dpaa2_eth_xdp_tx_flush(priv, ch, &priv->fq[flowid]);
1380 
1381 	return work_done;
1382 }
1383 
1384 static void dpaa2_eth_enable_ch_napi(struct dpaa2_eth_priv *priv)
1385 {
1386 	struct dpaa2_eth_channel *ch;
1387 	int i;
1388 
1389 	for (i = 0; i < priv->num_channels; i++) {
1390 		ch = priv->channel[i];
1391 		napi_enable(&ch->napi);
1392 	}
1393 }
1394 
1395 static void dpaa2_eth_disable_ch_napi(struct dpaa2_eth_priv *priv)
1396 {
1397 	struct dpaa2_eth_channel *ch;
1398 	int i;
1399 
1400 	for (i = 0; i < priv->num_channels; i++) {
1401 		ch = priv->channel[i];
1402 		napi_disable(&ch->napi);
1403 	}
1404 }
1405 
1406 void dpaa2_eth_set_rx_taildrop(struct dpaa2_eth_priv *priv,
1407 			       bool tx_pause, bool pfc)
1408 {
1409 	struct dpni_taildrop td = {0};
1410 	struct dpaa2_eth_fq *fq;
1411 	int i, err;
1412 
1413 	/* FQ taildrop: threshold is in bytes, per frame queue. Enabled if
1414 	 * flow control is disabled (as it might interfere with either the
1415 	 * buffer pool depletion trigger for pause frames or with the group
1416 	 * congestion trigger for PFC frames)
1417 	 */
1418 	td.enable = !tx_pause;
1419 	if (priv->rx_fqtd_enabled == td.enable)
1420 		goto set_cgtd;
1421 
1422 	td.threshold = DPAA2_ETH_FQ_TAILDROP_THRESH;
1423 	td.units = DPNI_CONGESTION_UNIT_BYTES;
1424 
1425 	for (i = 0; i < priv->num_fqs; i++) {
1426 		fq = &priv->fq[i];
1427 		if (fq->type != DPAA2_RX_FQ)
1428 			continue;
1429 		err = dpni_set_taildrop(priv->mc_io, 0, priv->mc_token,
1430 					DPNI_CP_QUEUE, DPNI_QUEUE_RX,
1431 					fq->tc, fq->flowid, &td);
1432 		if (err) {
1433 			netdev_err(priv->net_dev,
1434 				   "dpni_set_taildrop(FQ) failed\n");
1435 			return;
1436 		}
1437 	}
1438 
1439 	priv->rx_fqtd_enabled = td.enable;
1440 
1441 set_cgtd:
1442 	/* Congestion group taildrop: threshold is in frames, per group
1443 	 * of FQs belonging to the same traffic class
1444 	 * Enabled if general Tx pause disabled or if PFCs are enabled
1445 	 * (congestion group threhsold for PFC generation is lower than the
1446 	 * CG taildrop threshold, so it won't interfere with it; we also
1447 	 * want frames in non-PFC enabled traffic classes to be kept in check)
1448 	 */
1449 	td.enable = !tx_pause || (tx_pause && pfc);
1450 	if (priv->rx_cgtd_enabled == td.enable)
1451 		return;
1452 
1453 	td.threshold = DPAA2_ETH_CG_TAILDROP_THRESH(priv);
1454 	td.units = DPNI_CONGESTION_UNIT_FRAMES;
1455 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
1456 		err = dpni_set_taildrop(priv->mc_io, 0, priv->mc_token,
1457 					DPNI_CP_GROUP, DPNI_QUEUE_RX,
1458 					i, 0, &td);
1459 		if (err) {
1460 			netdev_err(priv->net_dev,
1461 				   "dpni_set_taildrop(CG) failed\n");
1462 			return;
1463 		}
1464 	}
1465 
1466 	priv->rx_cgtd_enabled = td.enable;
1467 }
1468 
1469 static int dpaa2_eth_link_state_update(struct dpaa2_eth_priv *priv)
1470 {
1471 	struct dpni_link_state state = {0};
1472 	bool tx_pause;
1473 	int err;
1474 
1475 	err = dpni_get_link_state(priv->mc_io, 0, priv->mc_token, &state);
1476 	if (unlikely(err)) {
1477 		netdev_err(priv->net_dev,
1478 			   "dpni_get_link_state() failed\n");
1479 		return err;
1480 	}
1481 
1482 	/* If Tx pause frame settings have changed, we need to update
1483 	 * Rx FQ taildrop configuration as well. We configure taildrop
1484 	 * only when pause frame generation is disabled.
1485 	 */
1486 	tx_pause = dpaa2_eth_tx_pause_enabled(state.options);
1487 	dpaa2_eth_set_rx_taildrop(priv, tx_pause, priv->pfc_enabled);
1488 
1489 	/* When we manage the MAC/PHY using phylink there is no need
1490 	 * to manually update the netif_carrier.
1491 	 */
1492 	if (priv->mac)
1493 		goto out;
1494 
1495 	/* Chech link state; speed / duplex changes are not treated yet */
1496 	if (priv->link_state.up == state.up)
1497 		goto out;
1498 
1499 	if (state.up) {
1500 		netif_carrier_on(priv->net_dev);
1501 		netif_tx_start_all_queues(priv->net_dev);
1502 	} else {
1503 		netif_tx_stop_all_queues(priv->net_dev);
1504 		netif_carrier_off(priv->net_dev);
1505 	}
1506 
1507 	netdev_info(priv->net_dev, "Link Event: state %s\n",
1508 		    state.up ? "up" : "down");
1509 
1510 out:
1511 	priv->link_state = state;
1512 
1513 	return 0;
1514 }
1515 
1516 static int dpaa2_eth_open(struct net_device *net_dev)
1517 {
1518 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1519 	int err;
1520 
1521 	err = dpaa2_eth_seed_pool(priv, priv->bpid);
1522 	if (err) {
1523 		/* Not much to do; the buffer pool, though not filled up,
1524 		 * may still contain some buffers which would enable us
1525 		 * to limp on.
1526 		 */
1527 		netdev_err(net_dev, "Buffer seeding failed for DPBP %d (bpid=%d)\n",
1528 			   priv->dpbp_dev->obj_desc.id, priv->bpid);
1529 	}
1530 
1531 	if (!priv->mac) {
1532 		/* We'll only start the txqs when the link is actually ready;
1533 		 * make sure we don't race against the link up notification,
1534 		 * which may come immediately after dpni_enable();
1535 		 */
1536 		netif_tx_stop_all_queues(net_dev);
1537 
1538 		/* Also, explicitly set carrier off, otherwise
1539 		 * netif_carrier_ok() will return true and cause 'ip link show'
1540 		 * to report the LOWER_UP flag, even though the link
1541 		 * notification wasn't even received.
1542 		 */
1543 		netif_carrier_off(net_dev);
1544 	}
1545 	dpaa2_eth_enable_ch_napi(priv);
1546 
1547 	err = dpni_enable(priv->mc_io, 0, priv->mc_token);
1548 	if (err < 0) {
1549 		netdev_err(net_dev, "dpni_enable() failed\n");
1550 		goto enable_err;
1551 	}
1552 
1553 	if (!priv->mac) {
1554 		/* If the DPMAC object has already processed the link up
1555 		 * interrupt, we have to learn the link state ourselves.
1556 		 */
1557 		err = dpaa2_eth_link_state_update(priv);
1558 		if (err < 0) {
1559 			netdev_err(net_dev, "Can't update link state\n");
1560 			goto link_state_err;
1561 		}
1562 	} else {
1563 		phylink_start(priv->mac->phylink);
1564 	}
1565 
1566 	return 0;
1567 
1568 link_state_err:
1569 enable_err:
1570 	dpaa2_eth_disable_ch_napi(priv);
1571 	dpaa2_eth_drain_pool(priv);
1572 	return err;
1573 }
1574 
1575 /* Total number of in-flight frames on ingress queues */
1576 static u32 dpaa2_eth_ingress_fq_count(struct dpaa2_eth_priv *priv)
1577 {
1578 	struct dpaa2_eth_fq *fq;
1579 	u32 fcnt = 0, bcnt = 0, total = 0;
1580 	int i, err;
1581 
1582 	for (i = 0; i < priv->num_fqs; i++) {
1583 		fq = &priv->fq[i];
1584 		err = dpaa2_io_query_fq_count(NULL, fq->fqid, &fcnt, &bcnt);
1585 		if (err) {
1586 			netdev_warn(priv->net_dev, "query_fq_count failed");
1587 			break;
1588 		}
1589 		total += fcnt;
1590 	}
1591 
1592 	return total;
1593 }
1594 
1595 static void dpaa2_eth_wait_for_ingress_fq_empty(struct dpaa2_eth_priv *priv)
1596 {
1597 	int retries = 10;
1598 	u32 pending;
1599 
1600 	do {
1601 		pending = dpaa2_eth_ingress_fq_count(priv);
1602 		if (pending)
1603 			msleep(100);
1604 	} while (pending && --retries);
1605 }
1606 
1607 #define DPNI_TX_PENDING_VER_MAJOR	7
1608 #define DPNI_TX_PENDING_VER_MINOR	13
1609 static void dpaa2_eth_wait_for_egress_fq_empty(struct dpaa2_eth_priv *priv)
1610 {
1611 	union dpni_statistics stats;
1612 	int retries = 10;
1613 	int err;
1614 
1615 	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_TX_PENDING_VER_MAJOR,
1616 				   DPNI_TX_PENDING_VER_MINOR) < 0)
1617 		goto out;
1618 
1619 	do {
1620 		err = dpni_get_statistics(priv->mc_io, 0, priv->mc_token, 6,
1621 					  &stats);
1622 		if (err)
1623 			goto out;
1624 		if (stats.page_6.tx_pending_frames == 0)
1625 			return;
1626 	} while (--retries);
1627 
1628 out:
1629 	msleep(500);
1630 }
1631 
1632 static int dpaa2_eth_stop(struct net_device *net_dev)
1633 {
1634 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1635 	int dpni_enabled = 0;
1636 	int retries = 10;
1637 
1638 	if (!priv->mac) {
1639 		netif_tx_stop_all_queues(net_dev);
1640 		netif_carrier_off(net_dev);
1641 	} else {
1642 		phylink_stop(priv->mac->phylink);
1643 	}
1644 
1645 	/* On dpni_disable(), the MC firmware will:
1646 	 * - stop MAC Rx and wait for all Rx frames to be enqueued to software
1647 	 * - cut off WRIOP dequeues from egress FQs and wait until transmission
1648 	 * of all in flight Tx frames is finished (and corresponding Tx conf
1649 	 * frames are enqueued back to software)
1650 	 *
1651 	 * Before calling dpni_disable(), we wait for all Tx frames to arrive
1652 	 * on WRIOP. After it finishes, wait until all remaining frames on Rx
1653 	 * and Tx conf queues are consumed on NAPI poll.
1654 	 */
1655 	dpaa2_eth_wait_for_egress_fq_empty(priv);
1656 
1657 	do {
1658 		dpni_disable(priv->mc_io, 0, priv->mc_token);
1659 		dpni_is_enabled(priv->mc_io, 0, priv->mc_token, &dpni_enabled);
1660 		if (dpni_enabled)
1661 			/* Allow the hardware some slack */
1662 			msleep(100);
1663 	} while (dpni_enabled && --retries);
1664 	if (!retries) {
1665 		netdev_warn(net_dev, "Retry count exceeded disabling DPNI\n");
1666 		/* Must go on and disable NAPI nonetheless, so we don't crash at
1667 		 * the next "ifconfig up"
1668 		 */
1669 	}
1670 
1671 	dpaa2_eth_wait_for_ingress_fq_empty(priv);
1672 	dpaa2_eth_disable_ch_napi(priv);
1673 
1674 	/* Empty the buffer pool */
1675 	dpaa2_eth_drain_pool(priv);
1676 
1677 	/* Empty the Scatter-Gather Buffer cache */
1678 	dpaa2_eth_sgt_cache_drain(priv);
1679 
1680 	return 0;
1681 }
1682 
1683 static int dpaa2_eth_set_addr(struct net_device *net_dev, void *addr)
1684 {
1685 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1686 	struct device *dev = net_dev->dev.parent;
1687 	int err;
1688 
1689 	err = eth_mac_addr(net_dev, addr);
1690 	if (err < 0) {
1691 		dev_err(dev, "eth_mac_addr() failed (%d)\n", err);
1692 		return err;
1693 	}
1694 
1695 	err = dpni_set_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
1696 					net_dev->dev_addr);
1697 	if (err) {
1698 		dev_err(dev, "dpni_set_primary_mac_addr() failed (%d)\n", err);
1699 		return err;
1700 	}
1701 
1702 	return 0;
1703 }
1704 
1705 /** Fill in counters maintained by the GPP driver. These may be different from
1706  * the hardware counters obtained by ethtool.
1707  */
1708 static void dpaa2_eth_get_stats(struct net_device *net_dev,
1709 				struct rtnl_link_stats64 *stats)
1710 {
1711 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1712 	struct rtnl_link_stats64 *percpu_stats;
1713 	u64 *cpustats;
1714 	u64 *netstats = (u64 *)stats;
1715 	int i, j;
1716 	int num = sizeof(struct rtnl_link_stats64) / sizeof(u64);
1717 
1718 	for_each_possible_cpu(i) {
1719 		percpu_stats = per_cpu_ptr(priv->percpu_stats, i);
1720 		cpustats = (u64 *)percpu_stats;
1721 		for (j = 0; j < num; j++)
1722 			netstats[j] += cpustats[j];
1723 	}
1724 }
1725 
1726 /* Copy mac unicast addresses from @net_dev to @priv.
1727  * Its sole purpose is to make dpaa2_eth_set_rx_mode() more readable.
1728  */
1729 static void dpaa2_eth_add_uc_hw_addr(const struct net_device *net_dev,
1730 				     struct dpaa2_eth_priv *priv)
1731 {
1732 	struct netdev_hw_addr *ha;
1733 	int err;
1734 
1735 	netdev_for_each_uc_addr(ha, net_dev) {
1736 		err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token,
1737 					ha->addr);
1738 		if (err)
1739 			netdev_warn(priv->net_dev,
1740 				    "Could not add ucast MAC %pM to the filtering table (err %d)\n",
1741 				    ha->addr, err);
1742 	}
1743 }
1744 
1745 /* Copy mac multicast addresses from @net_dev to @priv
1746  * Its sole purpose is to make dpaa2_eth_set_rx_mode() more readable.
1747  */
1748 static void dpaa2_eth_add_mc_hw_addr(const struct net_device *net_dev,
1749 				     struct dpaa2_eth_priv *priv)
1750 {
1751 	struct netdev_hw_addr *ha;
1752 	int err;
1753 
1754 	netdev_for_each_mc_addr(ha, net_dev) {
1755 		err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token,
1756 					ha->addr);
1757 		if (err)
1758 			netdev_warn(priv->net_dev,
1759 				    "Could not add mcast MAC %pM to the filtering table (err %d)\n",
1760 				    ha->addr, err);
1761 	}
1762 }
1763 
1764 static void dpaa2_eth_set_rx_mode(struct net_device *net_dev)
1765 {
1766 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1767 	int uc_count = netdev_uc_count(net_dev);
1768 	int mc_count = netdev_mc_count(net_dev);
1769 	u8 max_mac = priv->dpni_attrs.mac_filter_entries;
1770 	u32 options = priv->dpni_attrs.options;
1771 	u16 mc_token = priv->mc_token;
1772 	struct fsl_mc_io *mc_io = priv->mc_io;
1773 	int err;
1774 
1775 	/* Basic sanity checks; these probably indicate a misconfiguration */
1776 	if (options & DPNI_OPT_NO_MAC_FILTER && max_mac != 0)
1777 		netdev_info(net_dev,
1778 			    "mac_filter_entries=%d, DPNI_OPT_NO_MAC_FILTER option must be disabled\n",
1779 			    max_mac);
1780 
1781 	/* Force promiscuous if the uc or mc counts exceed our capabilities. */
1782 	if (uc_count > max_mac) {
1783 		netdev_info(net_dev,
1784 			    "Unicast addr count reached %d, max allowed is %d; forcing promisc\n",
1785 			    uc_count, max_mac);
1786 		goto force_promisc;
1787 	}
1788 	if (mc_count + uc_count > max_mac) {
1789 		netdev_info(net_dev,
1790 			    "Unicast + multicast addr count reached %d, max allowed is %d; forcing promisc\n",
1791 			    uc_count + mc_count, max_mac);
1792 		goto force_mc_promisc;
1793 	}
1794 
1795 	/* Adjust promisc settings due to flag combinations */
1796 	if (net_dev->flags & IFF_PROMISC)
1797 		goto force_promisc;
1798 	if (net_dev->flags & IFF_ALLMULTI) {
1799 		/* First, rebuild unicast filtering table. This should be done
1800 		 * in promisc mode, in order to avoid frame loss while we
1801 		 * progressively add entries to the table.
1802 		 * We don't know whether we had been in promisc already, and
1803 		 * making an MC call to find out is expensive; so set uc promisc
1804 		 * nonetheless.
1805 		 */
1806 		err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 1);
1807 		if (err)
1808 			netdev_warn(net_dev, "Can't set uc promisc\n");
1809 
1810 		/* Actual uc table reconstruction. */
1811 		err = dpni_clear_mac_filters(mc_io, 0, mc_token, 1, 0);
1812 		if (err)
1813 			netdev_warn(net_dev, "Can't clear uc filters\n");
1814 		dpaa2_eth_add_uc_hw_addr(net_dev, priv);
1815 
1816 		/* Finally, clear uc promisc and set mc promisc as requested. */
1817 		err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 0);
1818 		if (err)
1819 			netdev_warn(net_dev, "Can't clear uc promisc\n");
1820 		goto force_mc_promisc;
1821 	}
1822 
1823 	/* Neither unicast, nor multicast promisc will be on... eventually.
1824 	 * For now, rebuild mac filtering tables while forcing both of them on.
1825 	 */
1826 	err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 1);
1827 	if (err)
1828 		netdev_warn(net_dev, "Can't set uc promisc (%d)\n", err);
1829 	err = dpni_set_multicast_promisc(mc_io, 0, mc_token, 1);
1830 	if (err)
1831 		netdev_warn(net_dev, "Can't set mc promisc (%d)\n", err);
1832 
1833 	/* Actual mac filtering tables reconstruction */
1834 	err = dpni_clear_mac_filters(mc_io, 0, mc_token, 1, 1);
1835 	if (err)
1836 		netdev_warn(net_dev, "Can't clear mac filters\n");
1837 	dpaa2_eth_add_mc_hw_addr(net_dev, priv);
1838 	dpaa2_eth_add_uc_hw_addr(net_dev, priv);
1839 
1840 	/* Now we can clear both ucast and mcast promisc, without risking
1841 	 * to drop legitimate frames anymore.
1842 	 */
1843 	err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 0);
1844 	if (err)
1845 		netdev_warn(net_dev, "Can't clear ucast promisc\n");
1846 	err = dpni_set_multicast_promisc(mc_io, 0, mc_token, 0);
1847 	if (err)
1848 		netdev_warn(net_dev, "Can't clear mcast promisc\n");
1849 
1850 	return;
1851 
1852 force_promisc:
1853 	err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 1);
1854 	if (err)
1855 		netdev_warn(net_dev, "Can't set ucast promisc\n");
1856 force_mc_promisc:
1857 	err = dpni_set_multicast_promisc(mc_io, 0, mc_token, 1);
1858 	if (err)
1859 		netdev_warn(net_dev, "Can't set mcast promisc\n");
1860 }
1861 
1862 static int dpaa2_eth_set_features(struct net_device *net_dev,
1863 				  netdev_features_t features)
1864 {
1865 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1866 	netdev_features_t changed = features ^ net_dev->features;
1867 	bool enable;
1868 	int err;
1869 
1870 	if (changed & NETIF_F_RXCSUM) {
1871 		enable = !!(features & NETIF_F_RXCSUM);
1872 		err = dpaa2_eth_set_rx_csum(priv, enable);
1873 		if (err)
1874 			return err;
1875 	}
1876 
1877 	if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
1878 		enable = !!(features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM));
1879 		err = dpaa2_eth_set_tx_csum(priv, enable);
1880 		if (err)
1881 			return err;
1882 	}
1883 
1884 	return 0;
1885 }
1886 
1887 static int dpaa2_eth_ts_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
1888 {
1889 	struct dpaa2_eth_priv *priv = netdev_priv(dev);
1890 	struct hwtstamp_config config;
1891 
1892 	if (copy_from_user(&config, rq->ifr_data, sizeof(config)))
1893 		return -EFAULT;
1894 
1895 	switch (config.tx_type) {
1896 	case HWTSTAMP_TX_OFF:
1897 		priv->tx_tstamp = false;
1898 		break;
1899 	case HWTSTAMP_TX_ON:
1900 		priv->tx_tstamp = true;
1901 		break;
1902 	default:
1903 		return -ERANGE;
1904 	}
1905 
1906 	if (config.rx_filter == HWTSTAMP_FILTER_NONE) {
1907 		priv->rx_tstamp = false;
1908 	} else {
1909 		priv->rx_tstamp = true;
1910 		/* TS is set for all frame types, not only those requested */
1911 		config.rx_filter = HWTSTAMP_FILTER_ALL;
1912 	}
1913 
1914 	return copy_to_user(rq->ifr_data, &config, sizeof(config)) ?
1915 			-EFAULT : 0;
1916 }
1917 
1918 static int dpaa2_eth_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
1919 {
1920 	struct dpaa2_eth_priv *priv = netdev_priv(dev);
1921 
1922 	if (cmd == SIOCSHWTSTAMP)
1923 		return dpaa2_eth_ts_ioctl(dev, rq, cmd);
1924 
1925 	if (priv->mac)
1926 		return phylink_mii_ioctl(priv->mac->phylink, rq, cmd);
1927 
1928 	return -EOPNOTSUPP;
1929 }
1930 
1931 static bool xdp_mtu_valid(struct dpaa2_eth_priv *priv, int mtu)
1932 {
1933 	int mfl, linear_mfl;
1934 
1935 	mfl = DPAA2_ETH_L2_MAX_FRM(mtu);
1936 	linear_mfl = priv->rx_buf_size - DPAA2_ETH_RX_HWA_SIZE -
1937 		     dpaa2_eth_rx_head_room(priv) - XDP_PACKET_HEADROOM;
1938 
1939 	if (mfl > linear_mfl) {
1940 		netdev_warn(priv->net_dev, "Maximum MTU for XDP is %d\n",
1941 			    linear_mfl - VLAN_ETH_HLEN);
1942 		return false;
1943 	}
1944 
1945 	return true;
1946 }
1947 
1948 static int dpaa2_eth_set_rx_mfl(struct dpaa2_eth_priv *priv, int mtu, bool has_xdp)
1949 {
1950 	int mfl, err;
1951 
1952 	/* We enforce a maximum Rx frame length based on MTU only if we have
1953 	 * an XDP program attached (in order to avoid Rx S/G frames).
1954 	 * Otherwise, we accept all incoming frames as long as they are not
1955 	 * larger than maximum size supported in hardware
1956 	 */
1957 	if (has_xdp)
1958 		mfl = DPAA2_ETH_L2_MAX_FRM(mtu);
1959 	else
1960 		mfl = DPAA2_ETH_MFL;
1961 
1962 	err = dpni_set_max_frame_length(priv->mc_io, 0, priv->mc_token, mfl);
1963 	if (err) {
1964 		netdev_err(priv->net_dev, "dpni_set_max_frame_length failed\n");
1965 		return err;
1966 	}
1967 
1968 	return 0;
1969 }
1970 
1971 static int dpaa2_eth_change_mtu(struct net_device *dev, int new_mtu)
1972 {
1973 	struct dpaa2_eth_priv *priv = netdev_priv(dev);
1974 	int err;
1975 
1976 	if (!priv->xdp_prog)
1977 		goto out;
1978 
1979 	if (!xdp_mtu_valid(priv, new_mtu))
1980 		return -EINVAL;
1981 
1982 	err = dpaa2_eth_set_rx_mfl(priv, new_mtu, true);
1983 	if (err)
1984 		return err;
1985 
1986 out:
1987 	dev->mtu = new_mtu;
1988 	return 0;
1989 }
1990 
1991 static int dpaa2_eth_update_rx_buffer_headroom(struct dpaa2_eth_priv *priv, bool has_xdp)
1992 {
1993 	struct dpni_buffer_layout buf_layout = {0};
1994 	int err;
1995 
1996 	err = dpni_get_buffer_layout(priv->mc_io, 0, priv->mc_token,
1997 				     DPNI_QUEUE_RX, &buf_layout);
1998 	if (err) {
1999 		netdev_err(priv->net_dev, "dpni_get_buffer_layout failed\n");
2000 		return err;
2001 	}
2002 
2003 	/* Reserve extra headroom for XDP header size changes */
2004 	buf_layout.data_head_room = dpaa2_eth_rx_head_room(priv) +
2005 				    (has_xdp ? XDP_PACKET_HEADROOM : 0);
2006 	buf_layout.options = DPNI_BUF_LAYOUT_OPT_DATA_HEAD_ROOM;
2007 	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
2008 				     DPNI_QUEUE_RX, &buf_layout);
2009 	if (err) {
2010 		netdev_err(priv->net_dev, "dpni_set_buffer_layout failed\n");
2011 		return err;
2012 	}
2013 
2014 	return 0;
2015 }
2016 
2017 static int dpaa2_eth_setup_xdp(struct net_device *dev, struct bpf_prog *prog)
2018 {
2019 	struct dpaa2_eth_priv *priv = netdev_priv(dev);
2020 	struct dpaa2_eth_channel *ch;
2021 	struct bpf_prog *old;
2022 	bool up, need_update;
2023 	int i, err;
2024 
2025 	if (prog && !xdp_mtu_valid(priv, dev->mtu))
2026 		return -EINVAL;
2027 
2028 	if (prog)
2029 		bpf_prog_add(prog, priv->num_channels);
2030 
2031 	up = netif_running(dev);
2032 	need_update = (!!priv->xdp_prog != !!prog);
2033 
2034 	if (up)
2035 		dpaa2_eth_stop(dev);
2036 
2037 	/* While in xdp mode, enforce a maximum Rx frame size based on MTU.
2038 	 * Also, when switching between xdp/non-xdp modes we need to reconfigure
2039 	 * our Rx buffer layout. Buffer pool was drained on dpaa2_eth_stop,
2040 	 * so we are sure no old format buffers will be used from now on.
2041 	 */
2042 	if (need_update) {
2043 		err = dpaa2_eth_set_rx_mfl(priv, dev->mtu, !!prog);
2044 		if (err)
2045 			goto out_err;
2046 		err = dpaa2_eth_update_rx_buffer_headroom(priv, !!prog);
2047 		if (err)
2048 			goto out_err;
2049 	}
2050 
2051 	old = xchg(&priv->xdp_prog, prog);
2052 	if (old)
2053 		bpf_prog_put(old);
2054 
2055 	for (i = 0; i < priv->num_channels; i++) {
2056 		ch = priv->channel[i];
2057 		old = xchg(&ch->xdp.prog, prog);
2058 		if (old)
2059 			bpf_prog_put(old);
2060 	}
2061 
2062 	if (up) {
2063 		err = dpaa2_eth_open(dev);
2064 		if (err)
2065 			return err;
2066 	}
2067 
2068 	return 0;
2069 
2070 out_err:
2071 	if (prog)
2072 		bpf_prog_sub(prog, priv->num_channels);
2073 	if (up)
2074 		dpaa2_eth_open(dev);
2075 
2076 	return err;
2077 }
2078 
2079 static int dpaa2_eth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2080 {
2081 	switch (xdp->command) {
2082 	case XDP_SETUP_PROG:
2083 		return dpaa2_eth_setup_xdp(dev, xdp->prog);
2084 	default:
2085 		return -EINVAL;
2086 	}
2087 
2088 	return 0;
2089 }
2090 
2091 static int dpaa2_eth_xdp_create_fd(struct net_device *net_dev,
2092 				   struct xdp_frame *xdpf,
2093 				   struct dpaa2_fd *fd)
2094 {
2095 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2096 	struct device *dev = net_dev->dev.parent;
2097 	unsigned int needed_headroom;
2098 	struct dpaa2_eth_swa *swa;
2099 	void *buffer_start, *aligned_start;
2100 	dma_addr_t addr;
2101 
2102 	/* We require a minimum headroom to be able to transmit the frame.
2103 	 * Otherwise return an error and let the original net_device handle it
2104 	 */
2105 	needed_headroom = dpaa2_eth_needed_headroom(priv, NULL);
2106 	if (xdpf->headroom < needed_headroom)
2107 		return -EINVAL;
2108 
2109 	/* Setup the FD fields */
2110 	memset(fd, 0, sizeof(*fd));
2111 
2112 	/* Align FD address, if possible */
2113 	buffer_start = xdpf->data - needed_headroom;
2114 	aligned_start = PTR_ALIGN(buffer_start - DPAA2_ETH_TX_BUF_ALIGN,
2115 				  DPAA2_ETH_TX_BUF_ALIGN);
2116 	if (aligned_start >= xdpf->data - xdpf->headroom)
2117 		buffer_start = aligned_start;
2118 
2119 	swa = (struct dpaa2_eth_swa *)buffer_start;
2120 	/* fill in necessary fields here */
2121 	swa->type = DPAA2_ETH_SWA_XDP;
2122 	swa->xdp.dma_size = xdpf->data + xdpf->len - buffer_start;
2123 	swa->xdp.xdpf = xdpf;
2124 
2125 	addr = dma_map_single(dev, buffer_start,
2126 			      swa->xdp.dma_size,
2127 			      DMA_BIDIRECTIONAL);
2128 	if (unlikely(dma_mapping_error(dev, addr)))
2129 		return -ENOMEM;
2130 
2131 	dpaa2_fd_set_addr(fd, addr);
2132 	dpaa2_fd_set_offset(fd, xdpf->data - buffer_start);
2133 	dpaa2_fd_set_len(fd, xdpf->len);
2134 	dpaa2_fd_set_format(fd, dpaa2_fd_single);
2135 	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
2136 
2137 	return 0;
2138 }
2139 
2140 static int dpaa2_eth_xdp_xmit(struct net_device *net_dev, int n,
2141 			      struct xdp_frame **frames, u32 flags)
2142 {
2143 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2144 	struct dpaa2_eth_xdp_fds *xdp_redirect_fds;
2145 	struct rtnl_link_stats64 *percpu_stats;
2146 	struct dpaa2_eth_fq *fq;
2147 	struct dpaa2_fd *fds;
2148 	int enqueued, i, err;
2149 
2150 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
2151 		return -EINVAL;
2152 
2153 	if (!netif_running(net_dev))
2154 		return -ENETDOWN;
2155 
2156 	fq = &priv->fq[smp_processor_id()];
2157 	xdp_redirect_fds = &fq->xdp_redirect_fds;
2158 	fds = xdp_redirect_fds->fds;
2159 
2160 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
2161 
2162 	/* create a FD for each xdp_frame in the list received */
2163 	for (i = 0; i < n; i++) {
2164 		err = dpaa2_eth_xdp_create_fd(net_dev, frames[i], &fds[i]);
2165 		if (err)
2166 			break;
2167 	}
2168 	xdp_redirect_fds->num = i;
2169 
2170 	/* enqueue all the frame descriptors */
2171 	enqueued = dpaa2_eth_xdp_flush(priv, fq, xdp_redirect_fds);
2172 
2173 	/* update statistics */
2174 	percpu_stats->tx_packets += enqueued;
2175 	for (i = 0; i < enqueued; i++)
2176 		percpu_stats->tx_bytes += dpaa2_fd_get_len(&fds[i]);
2177 	for (i = enqueued; i < n; i++)
2178 		xdp_return_frame_rx_napi(frames[i]);
2179 
2180 	return enqueued;
2181 }
2182 
2183 static int update_xps(struct dpaa2_eth_priv *priv)
2184 {
2185 	struct net_device *net_dev = priv->net_dev;
2186 	struct cpumask xps_mask;
2187 	struct dpaa2_eth_fq *fq;
2188 	int i, num_queues, netdev_queues;
2189 	int err = 0;
2190 
2191 	num_queues = dpaa2_eth_queue_count(priv);
2192 	netdev_queues = (net_dev->num_tc ? : 1) * num_queues;
2193 
2194 	/* The first <num_queues> entries in priv->fq array are Tx/Tx conf
2195 	 * queues, so only process those
2196 	 */
2197 	for (i = 0; i < netdev_queues; i++) {
2198 		fq = &priv->fq[i % num_queues];
2199 
2200 		cpumask_clear(&xps_mask);
2201 		cpumask_set_cpu(fq->target_cpu, &xps_mask);
2202 
2203 		err = netif_set_xps_queue(net_dev, &xps_mask, i);
2204 		if (err) {
2205 			netdev_warn_once(net_dev, "Error setting XPS queue\n");
2206 			break;
2207 		}
2208 	}
2209 
2210 	return err;
2211 }
2212 
2213 static int dpaa2_eth_setup_mqprio(struct net_device *net_dev,
2214 				  struct tc_mqprio_qopt *mqprio)
2215 {
2216 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2217 	u8 num_tc, num_queues;
2218 	int i;
2219 
2220 	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
2221 	num_queues = dpaa2_eth_queue_count(priv);
2222 	num_tc = mqprio->num_tc;
2223 
2224 	if (num_tc == net_dev->num_tc)
2225 		return 0;
2226 
2227 	if (num_tc  > dpaa2_eth_tc_count(priv)) {
2228 		netdev_err(net_dev, "Max %d traffic classes supported\n",
2229 			   dpaa2_eth_tc_count(priv));
2230 		return -EOPNOTSUPP;
2231 	}
2232 
2233 	if (!num_tc) {
2234 		netdev_reset_tc(net_dev);
2235 		netif_set_real_num_tx_queues(net_dev, num_queues);
2236 		goto out;
2237 	}
2238 
2239 	netdev_set_num_tc(net_dev, num_tc);
2240 	netif_set_real_num_tx_queues(net_dev, num_tc * num_queues);
2241 
2242 	for (i = 0; i < num_tc; i++)
2243 		netdev_set_tc_queue(net_dev, i, num_queues, i * num_queues);
2244 
2245 out:
2246 	update_xps(priv);
2247 
2248 	return 0;
2249 }
2250 
2251 #define bps_to_mbits(rate) (div_u64((rate), 1000000) * 8)
2252 
2253 static int dpaa2_eth_setup_tbf(struct net_device *net_dev, struct tc_tbf_qopt_offload *p)
2254 {
2255 	struct tc_tbf_qopt_offload_replace_params *cfg = &p->replace_params;
2256 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2257 	struct dpni_tx_shaping_cfg tx_cr_shaper = { 0 };
2258 	struct dpni_tx_shaping_cfg tx_er_shaper = { 0 };
2259 	int err;
2260 
2261 	if (p->command == TC_TBF_STATS)
2262 		return -EOPNOTSUPP;
2263 
2264 	/* Only per port Tx shaping */
2265 	if (p->parent != TC_H_ROOT)
2266 		return -EOPNOTSUPP;
2267 
2268 	if (p->command == TC_TBF_REPLACE) {
2269 		if (cfg->max_size > DPAA2_ETH_MAX_BURST_SIZE) {
2270 			netdev_err(net_dev, "burst size cannot be greater than %d\n",
2271 				   DPAA2_ETH_MAX_BURST_SIZE);
2272 			return -EINVAL;
2273 		}
2274 
2275 		tx_cr_shaper.max_burst_size = cfg->max_size;
2276 		/* The TBF interface is in bytes/s, whereas DPAA2 expects the
2277 		 * rate in Mbits/s
2278 		 */
2279 		tx_cr_shaper.rate_limit = bps_to_mbits(cfg->rate.rate_bytes_ps);
2280 	}
2281 
2282 	err = dpni_set_tx_shaping(priv->mc_io, 0, priv->mc_token, &tx_cr_shaper,
2283 				  &tx_er_shaper, 0);
2284 	if (err) {
2285 		netdev_err(net_dev, "dpni_set_tx_shaping() = %d\n", err);
2286 		return err;
2287 	}
2288 
2289 	return 0;
2290 }
2291 
2292 static int dpaa2_eth_setup_tc(struct net_device *net_dev,
2293 			      enum tc_setup_type type, void *type_data)
2294 {
2295 	switch (type) {
2296 	case TC_SETUP_QDISC_MQPRIO:
2297 		return dpaa2_eth_setup_mqprio(net_dev, type_data);
2298 	case TC_SETUP_QDISC_TBF:
2299 		return dpaa2_eth_setup_tbf(net_dev, type_data);
2300 	default:
2301 		return -EOPNOTSUPP;
2302 	}
2303 }
2304 
2305 static const struct net_device_ops dpaa2_eth_ops = {
2306 	.ndo_open = dpaa2_eth_open,
2307 	.ndo_start_xmit = dpaa2_eth_tx,
2308 	.ndo_stop = dpaa2_eth_stop,
2309 	.ndo_set_mac_address = dpaa2_eth_set_addr,
2310 	.ndo_get_stats64 = dpaa2_eth_get_stats,
2311 	.ndo_set_rx_mode = dpaa2_eth_set_rx_mode,
2312 	.ndo_set_features = dpaa2_eth_set_features,
2313 	.ndo_do_ioctl = dpaa2_eth_ioctl,
2314 	.ndo_change_mtu = dpaa2_eth_change_mtu,
2315 	.ndo_bpf = dpaa2_eth_xdp,
2316 	.ndo_xdp_xmit = dpaa2_eth_xdp_xmit,
2317 	.ndo_setup_tc = dpaa2_eth_setup_tc,
2318 };
2319 
2320 static void dpaa2_eth_cdan_cb(struct dpaa2_io_notification_ctx *ctx)
2321 {
2322 	struct dpaa2_eth_channel *ch;
2323 
2324 	ch = container_of(ctx, struct dpaa2_eth_channel, nctx);
2325 
2326 	/* Update NAPI statistics */
2327 	ch->stats.cdan++;
2328 
2329 	napi_schedule(&ch->napi);
2330 }
2331 
2332 /* Allocate and configure a DPCON object */
2333 static struct fsl_mc_device *dpaa2_eth_setup_dpcon(struct dpaa2_eth_priv *priv)
2334 {
2335 	struct fsl_mc_device *dpcon;
2336 	struct device *dev = priv->net_dev->dev.parent;
2337 	int err;
2338 
2339 	err = fsl_mc_object_allocate(to_fsl_mc_device(dev),
2340 				     FSL_MC_POOL_DPCON, &dpcon);
2341 	if (err) {
2342 		if (err == -ENXIO)
2343 			err = -EPROBE_DEFER;
2344 		else
2345 			dev_info(dev, "Not enough DPCONs, will go on as-is\n");
2346 		return ERR_PTR(err);
2347 	}
2348 
2349 	err = dpcon_open(priv->mc_io, 0, dpcon->obj_desc.id, &dpcon->mc_handle);
2350 	if (err) {
2351 		dev_err(dev, "dpcon_open() failed\n");
2352 		goto free;
2353 	}
2354 
2355 	err = dpcon_reset(priv->mc_io, 0, dpcon->mc_handle);
2356 	if (err) {
2357 		dev_err(dev, "dpcon_reset() failed\n");
2358 		goto close;
2359 	}
2360 
2361 	err = dpcon_enable(priv->mc_io, 0, dpcon->mc_handle);
2362 	if (err) {
2363 		dev_err(dev, "dpcon_enable() failed\n");
2364 		goto close;
2365 	}
2366 
2367 	return dpcon;
2368 
2369 close:
2370 	dpcon_close(priv->mc_io, 0, dpcon->mc_handle);
2371 free:
2372 	fsl_mc_object_free(dpcon);
2373 
2374 	return ERR_PTR(err);
2375 }
2376 
2377 static void dpaa2_eth_free_dpcon(struct dpaa2_eth_priv *priv,
2378 				 struct fsl_mc_device *dpcon)
2379 {
2380 	dpcon_disable(priv->mc_io, 0, dpcon->mc_handle);
2381 	dpcon_close(priv->mc_io, 0, dpcon->mc_handle);
2382 	fsl_mc_object_free(dpcon);
2383 }
2384 
2385 static struct dpaa2_eth_channel *dpaa2_eth_alloc_channel(struct dpaa2_eth_priv *priv)
2386 {
2387 	struct dpaa2_eth_channel *channel;
2388 	struct dpcon_attr attr;
2389 	struct device *dev = priv->net_dev->dev.parent;
2390 	int err;
2391 
2392 	channel = kzalloc(sizeof(*channel), GFP_KERNEL);
2393 	if (!channel)
2394 		return NULL;
2395 
2396 	channel->dpcon = dpaa2_eth_setup_dpcon(priv);
2397 	if (IS_ERR(channel->dpcon)) {
2398 		err = PTR_ERR(channel->dpcon);
2399 		goto err_setup;
2400 	}
2401 
2402 	err = dpcon_get_attributes(priv->mc_io, 0, channel->dpcon->mc_handle,
2403 				   &attr);
2404 	if (err) {
2405 		dev_err(dev, "dpcon_get_attributes() failed\n");
2406 		goto err_get_attr;
2407 	}
2408 
2409 	channel->dpcon_id = attr.id;
2410 	channel->ch_id = attr.qbman_ch_id;
2411 	channel->priv = priv;
2412 
2413 	return channel;
2414 
2415 err_get_attr:
2416 	dpaa2_eth_free_dpcon(priv, channel->dpcon);
2417 err_setup:
2418 	kfree(channel);
2419 	return ERR_PTR(err);
2420 }
2421 
2422 static void dpaa2_eth_free_channel(struct dpaa2_eth_priv *priv,
2423 				   struct dpaa2_eth_channel *channel)
2424 {
2425 	dpaa2_eth_free_dpcon(priv, channel->dpcon);
2426 	kfree(channel);
2427 }
2428 
2429 /* DPIO setup: allocate and configure QBMan channels, setup core affinity
2430  * and register data availability notifications
2431  */
2432 static int dpaa2_eth_setup_dpio(struct dpaa2_eth_priv *priv)
2433 {
2434 	struct dpaa2_io_notification_ctx *nctx;
2435 	struct dpaa2_eth_channel *channel;
2436 	struct dpcon_notification_cfg dpcon_notif_cfg;
2437 	struct device *dev = priv->net_dev->dev.parent;
2438 	int i, err;
2439 
2440 	/* We want the ability to spread ingress traffic (RX, TX conf) to as
2441 	 * many cores as possible, so we need one channel for each core
2442 	 * (unless there's fewer queues than cores, in which case the extra
2443 	 * channels would be wasted).
2444 	 * Allocate one channel per core and register it to the core's
2445 	 * affine DPIO. If not enough channels are available for all cores
2446 	 * or if some cores don't have an affine DPIO, there will be no
2447 	 * ingress frame processing on those cores.
2448 	 */
2449 	cpumask_clear(&priv->dpio_cpumask);
2450 	for_each_online_cpu(i) {
2451 		/* Try to allocate a channel */
2452 		channel = dpaa2_eth_alloc_channel(priv);
2453 		if (IS_ERR_OR_NULL(channel)) {
2454 			err = PTR_ERR_OR_ZERO(channel);
2455 			if (err != -EPROBE_DEFER)
2456 				dev_info(dev,
2457 					 "No affine channel for cpu %d and above\n", i);
2458 			goto err_alloc_ch;
2459 		}
2460 
2461 		priv->channel[priv->num_channels] = channel;
2462 
2463 		nctx = &channel->nctx;
2464 		nctx->is_cdan = 1;
2465 		nctx->cb = dpaa2_eth_cdan_cb;
2466 		nctx->id = channel->ch_id;
2467 		nctx->desired_cpu = i;
2468 
2469 		/* Register the new context */
2470 		channel->dpio = dpaa2_io_service_select(i);
2471 		err = dpaa2_io_service_register(channel->dpio, nctx, dev);
2472 		if (err) {
2473 			dev_dbg(dev, "No affine DPIO for cpu %d\n", i);
2474 			/* If no affine DPIO for this core, there's probably
2475 			 * none available for next cores either. Signal we want
2476 			 * to retry later, in case the DPIO devices weren't
2477 			 * probed yet.
2478 			 */
2479 			err = -EPROBE_DEFER;
2480 			goto err_service_reg;
2481 		}
2482 
2483 		/* Register DPCON notification with MC */
2484 		dpcon_notif_cfg.dpio_id = nctx->dpio_id;
2485 		dpcon_notif_cfg.priority = 0;
2486 		dpcon_notif_cfg.user_ctx = nctx->qman64;
2487 		err = dpcon_set_notification(priv->mc_io, 0,
2488 					     channel->dpcon->mc_handle,
2489 					     &dpcon_notif_cfg);
2490 		if (err) {
2491 			dev_err(dev, "dpcon_set_notification failed()\n");
2492 			goto err_set_cdan;
2493 		}
2494 
2495 		/* If we managed to allocate a channel and also found an affine
2496 		 * DPIO for this core, add it to the final mask
2497 		 */
2498 		cpumask_set_cpu(i, &priv->dpio_cpumask);
2499 		priv->num_channels++;
2500 
2501 		/* Stop if we already have enough channels to accommodate all
2502 		 * RX and TX conf queues
2503 		 */
2504 		if (priv->num_channels == priv->dpni_attrs.num_queues)
2505 			break;
2506 	}
2507 
2508 	return 0;
2509 
2510 err_set_cdan:
2511 	dpaa2_io_service_deregister(channel->dpio, nctx, dev);
2512 err_service_reg:
2513 	dpaa2_eth_free_channel(priv, channel);
2514 err_alloc_ch:
2515 	if (err == -EPROBE_DEFER) {
2516 		for (i = 0; i < priv->num_channels; i++) {
2517 			channel = priv->channel[i];
2518 			nctx = &channel->nctx;
2519 			dpaa2_io_service_deregister(channel->dpio, nctx, dev);
2520 			dpaa2_eth_free_channel(priv, channel);
2521 		}
2522 		priv->num_channels = 0;
2523 		return err;
2524 	}
2525 
2526 	if (cpumask_empty(&priv->dpio_cpumask)) {
2527 		dev_err(dev, "No cpu with an affine DPIO/DPCON\n");
2528 		return -ENODEV;
2529 	}
2530 
2531 	dev_info(dev, "Cores %*pbl available for processing ingress traffic\n",
2532 		 cpumask_pr_args(&priv->dpio_cpumask));
2533 
2534 	return 0;
2535 }
2536 
2537 static void dpaa2_eth_free_dpio(struct dpaa2_eth_priv *priv)
2538 {
2539 	struct device *dev = priv->net_dev->dev.parent;
2540 	struct dpaa2_eth_channel *ch;
2541 	int i;
2542 
2543 	/* deregister CDAN notifications and free channels */
2544 	for (i = 0; i < priv->num_channels; i++) {
2545 		ch = priv->channel[i];
2546 		dpaa2_io_service_deregister(ch->dpio, &ch->nctx, dev);
2547 		dpaa2_eth_free_channel(priv, ch);
2548 	}
2549 }
2550 
2551 static struct dpaa2_eth_channel *dpaa2_eth_get_affine_channel(struct dpaa2_eth_priv *priv,
2552 							      int cpu)
2553 {
2554 	struct device *dev = priv->net_dev->dev.parent;
2555 	int i;
2556 
2557 	for (i = 0; i < priv->num_channels; i++)
2558 		if (priv->channel[i]->nctx.desired_cpu == cpu)
2559 			return priv->channel[i];
2560 
2561 	/* We should never get here. Issue a warning and return
2562 	 * the first channel, because it's still better than nothing
2563 	 */
2564 	dev_warn(dev, "No affine channel found for cpu %d\n", cpu);
2565 
2566 	return priv->channel[0];
2567 }
2568 
2569 static void dpaa2_eth_set_fq_affinity(struct dpaa2_eth_priv *priv)
2570 {
2571 	struct device *dev = priv->net_dev->dev.parent;
2572 	struct dpaa2_eth_fq *fq;
2573 	int rx_cpu, txc_cpu;
2574 	int i;
2575 
2576 	/* For each FQ, pick one channel/CPU to deliver frames to.
2577 	 * This may well change at runtime, either through irqbalance or
2578 	 * through direct user intervention.
2579 	 */
2580 	rx_cpu = txc_cpu = cpumask_first(&priv->dpio_cpumask);
2581 
2582 	for (i = 0; i < priv->num_fqs; i++) {
2583 		fq = &priv->fq[i];
2584 		switch (fq->type) {
2585 		case DPAA2_RX_FQ:
2586 			fq->target_cpu = rx_cpu;
2587 			rx_cpu = cpumask_next(rx_cpu, &priv->dpio_cpumask);
2588 			if (rx_cpu >= nr_cpu_ids)
2589 				rx_cpu = cpumask_first(&priv->dpio_cpumask);
2590 			break;
2591 		case DPAA2_TX_CONF_FQ:
2592 			fq->target_cpu = txc_cpu;
2593 			txc_cpu = cpumask_next(txc_cpu, &priv->dpio_cpumask);
2594 			if (txc_cpu >= nr_cpu_ids)
2595 				txc_cpu = cpumask_first(&priv->dpio_cpumask);
2596 			break;
2597 		default:
2598 			dev_err(dev, "Unknown FQ type: %d\n", fq->type);
2599 		}
2600 		fq->channel = dpaa2_eth_get_affine_channel(priv, fq->target_cpu);
2601 	}
2602 
2603 	update_xps(priv);
2604 }
2605 
2606 static void dpaa2_eth_setup_fqs(struct dpaa2_eth_priv *priv)
2607 {
2608 	int i, j;
2609 
2610 	/* We have one TxConf FQ per Tx flow.
2611 	 * The number of Tx and Rx queues is the same.
2612 	 * Tx queues come first in the fq array.
2613 	 */
2614 	for (i = 0; i < dpaa2_eth_queue_count(priv); i++) {
2615 		priv->fq[priv->num_fqs].type = DPAA2_TX_CONF_FQ;
2616 		priv->fq[priv->num_fqs].consume = dpaa2_eth_tx_conf;
2617 		priv->fq[priv->num_fqs++].flowid = (u16)i;
2618 	}
2619 
2620 	for (j = 0; j < dpaa2_eth_tc_count(priv); j++) {
2621 		for (i = 0; i < dpaa2_eth_queue_count(priv); i++) {
2622 			priv->fq[priv->num_fqs].type = DPAA2_RX_FQ;
2623 			priv->fq[priv->num_fqs].consume = dpaa2_eth_rx;
2624 			priv->fq[priv->num_fqs].tc = (u8)j;
2625 			priv->fq[priv->num_fqs++].flowid = (u16)i;
2626 		}
2627 	}
2628 
2629 	/* For each FQ, decide on which core to process incoming frames */
2630 	dpaa2_eth_set_fq_affinity(priv);
2631 }
2632 
2633 /* Allocate and configure one buffer pool for each interface */
2634 static int dpaa2_eth_setup_dpbp(struct dpaa2_eth_priv *priv)
2635 {
2636 	int err;
2637 	struct fsl_mc_device *dpbp_dev;
2638 	struct device *dev = priv->net_dev->dev.parent;
2639 	struct dpbp_attr dpbp_attrs;
2640 
2641 	err = fsl_mc_object_allocate(to_fsl_mc_device(dev), FSL_MC_POOL_DPBP,
2642 				     &dpbp_dev);
2643 	if (err) {
2644 		if (err == -ENXIO)
2645 			err = -EPROBE_DEFER;
2646 		else
2647 			dev_err(dev, "DPBP device allocation failed\n");
2648 		return err;
2649 	}
2650 
2651 	priv->dpbp_dev = dpbp_dev;
2652 
2653 	err = dpbp_open(priv->mc_io, 0, priv->dpbp_dev->obj_desc.id,
2654 			&dpbp_dev->mc_handle);
2655 	if (err) {
2656 		dev_err(dev, "dpbp_open() failed\n");
2657 		goto err_open;
2658 	}
2659 
2660 	err = dpbp_reset(priv->mc_io, 0, dpbp_dev->mc_handle);
2661 	if (err) {
2662 		dev_err(dev, "dpbp_reset() failed\n");
2663 		goto err_reset;
2664 	}
2665 
2666 	err = dpbp_enable(priv->mc_io, 0, dpbp_dev->mc_handle);
2667 	if (err) {
2668 		dev_err(dev, "dpbp_enable() failed\n");
2669 		goto err_enable;
2670 	}
2671 
2672 	err = dpbp_get_attributes(priv->mc_io, 0, dpbp_dev->mc_handle,
2673 				  &dpbp_attrs);
2674 	if (err) {
2675 		dev_err(dev, "dpbp_get_attributes() failed\n");
2676 		goto err_get_attr;
2677 	}
2678 	priv->bpid = dpbp_attrs.bpid;
2679 
2680 	return 0;
2681 
2682 err_get_attr:
2683 	dpbp_disable(priv->mc_io, 0, dpbp_dev->mc_handle);
2684 err_enable:
2685 err_reset:
2686 	dpbp_close(priv->mc_io, 0, dpbp_dev->mc_handle);
2687 err_open:
2688 	fsl_mc_object_free(dpbp_dev);
2689 
2690 	return err;
2691 }
2692 
2693 static void dpaa2_eth_free_dpbp(struct dpaa2_eth_priv *priv)
2694 {
2695 	dpaa2_eth_drain_pool(priv);
2696 	dpbp_disable(priv->mc_io, 0, priv->dpbp_dev->mc_handle);
2697 	dpbp_close(priv->mc_io, 0, priv->dpbp_dev->mc_handle);
2698 	fsl_mc_object_free(priv->dpbp_dev);
2699 }
2700 
2701 static int dpaa2_eth_set_buffer_layout(struct dpaa2_eth_priv *priv)
2702 {
2703 	struct device *dev = priv->net_dev->dev.parent;
2704 	struct dpni_buffer_layout buf_layout = {0};
2705 	u16 rx_buf_align;
2706 	int err;
2707 
2708 	/* We need to check for WRIOP version 1.0.0, but depending on the MC
2709 	 * version, this number is not always provided correctly on rev1.
2710 	 * We need to check for both alternatives in this situation.
2711 	 */
2712 	if (priv->dpni_attrs.wriop_version == DPAA2_WRIOP_VERSION(0, 0, 0) ||
2713 	    priv->dpni_attrs.wriop_version == DPAA2_WRIOP_VERSION(1, 0, 0))
2714 		rx_buf_align = DPAA2_ETH_RX_BUF_ALIGN_REV1;
2715 	else
2716 		rx_buf_align = DPAA2_ETH_RX_BUF_ALIGN;
2717 
2718 	/* We need to ensure that the buffer size seen by WRIOP is a multiple
2719 	 * of 64 or 256 bytes depending on the WRIOP version.
2720 	 */
2721 	priv->rx_buf_size = ALIGN_DOWN(DPAA2_ETH_RX_BUF_SIZE, rx_buf_align);
2722 
2723 	/* tx buffer */
2724 	buf_layout.private_data_size = DPAA2_ETH_SWA_SIZE;
2725 	buf_layout.pass_timestamp = true;
2726 	buf_layout.options = DPNI_BUF_LAYOUT_OPT_PRIVATE_DATA_SIZE |
2727 			     DPNI_BUF_LAYOUT_OPT_TIMESTAMP;
2728 	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
2729 				     DPNI_QUEUE_TX, &buf_layout);
2730 	if (err) {
2731 		dev_err(dev, "dpni_set_buffer_layout(TX) failed\n");
2732 		return err;
2733 	}
2734 
2735 	/* tx-confirm buffer */
2736 	buf_layout.options = DPNI_BUF_LAYOUT_OPT_TIMESTAMP;
2737 	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
2738 				     DPNI_QUEUE_TX_CONFIRM, &buf_layout);
2739 	if (err) {
2740 		dev_err(dev, "dpni_set_buffer_layout(TX_CONF) failed\n");
2741 		return err;
2742 	}
2743 
2744 	/* Now that we've set our tx buffer layout, retrieve the minimum
2745 	 * required tx data offset.
2746 	 */
2747 	err = dpni_get_tx_data_offset(priv->mc_io, 0, priv->mc_token,
2748 				      &priv->tx_data_offset);
2749 	if (err) {
2750 		dev_err(dev, "dpni_get_tx_data_offset() failed\n");
2751 		return err;
2752 	}
2753 
2754 	if ((priv->tx_data_offset % 64) != 0)
2755 		dev_warn(dev, "Tx data offset (%d) not a multiple of 64B\n",
2756 			 priv->tx_data_offset);
2757 
2758 	/* rx buffer */
2759 	buf_layout.pass_frame_status = true;
2760 	buf_layout.pass_parser_result = true;
2761 	buf_layout.data_align = rx_buf_align;
2762 	buf_layout.data_head_room = dpaa2_eth_rx_head_room(priv);
2763 	buf_layout.private_data_size = 0;
2764 	buf_layout.options = DPNI_BUF_LAYOUT_OPT_PARSER_RESULT |
2765 			     DPNI_BUF_LAYOUT_OPT_FRAME_STATUS |
2766 			     DPNI_BUF_LAYOUT_OPT_DATA_ALIGN |
2767 			     DPNI_BUF_LAYOUT_OPT_DATA_HEAD_ROOM |
2768 			     DPNI_BUF_LAYOUT_OPT_TIMESTAMP;
2769 	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
2770 				     DPNI_QUEUE_RX, &buf_layout);
2771 	if (err) {
2772 		dev_err(dev, "dpni_set_buffer_layout(RX) failed\n");
2773 		return err;
2774 	}
2775 
2776 	return 0;
2777 }
2778 
2779 #define DPNI_ENQUEUE_FQID_VER_MAJOR	7
2780 #define DPNI_ENQUEUE_FQID_VER_MINOR	9
2781 
2782 static inline int dpaa2_eth_enqueue_qd(struct dpaa2_eth_priv *priv,
2783 				       struct dpaa2_eth_fq *fq,
2784 				       struct dpaa2_fd *fd, u8 prio,
2785 				       u32 num_frames __always_unused,
2786 				       int *frames_enqueued)
2787 {
2788 	int err;
2789 
2790 	err = dpaa2_io_service_enqueue_qd(fq->channel->dpio,
2791 					  priv->tx_qdid, prio,
2792 					  fq->tx_qdbin, fd);
2793 	if (!err && frames_enqueued)
2794 		*frames_enqueued = 1;
2795 	return err;
2796 }
2797 
2798 static inline int dpaa2_eth_enqueue_fq_multiple(struct dpaa2_eth_priv *priv,
2799 						struct dpaa2_eth_fq *fq,
2800 						struct dpaa2_fd *fd,
2801 						u8 prio, u32 num_frames,
2802 						int *frames_enqueued)
2803 {
2804 	int err;
2805 
2806 	err = dpaa2_io_service_enqueue_multiple_fq(fq->channel->dpio,
2807 						   fq->tx_fqid[prio],
2808 						   fd, num_frames);
2809 
2810 	if (err == 0)
2811 		return -EBUSY;
2812 
2813 	if (frames_enqueued)
2814 		*frames_enqueued = err;
2815 	return 0;
2816 }
2817 
2818 static void dpaa2_eth_set_enqueue_mode(struct dpaa2_eth_priv *priv)
2819 {
2820 	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_ENQUEUE_FQID_VER_MAJOR,
2821 				   DPNI_ENQUEUE_FQID_VER_MINOR) < 0)
2822 		priv->enqueue = dpaa2_eth_enqueue_qd;
2823 	else
2824 		priv->enqueue = dpaa2_eth_enqueue_fq_multiple;
2825 }
2826 
2827 static int dpaa2_eth_set_pause(struct dpaa2_eth_priv *priv)
2828 {
2829 	struct device *dev = priv->net_dev->dev.parent;
2830 	struct dpni_link_cfg link_cfg = {0};
2831 	int err;
2832 
2833 	/* Get the default link options so we don't override other flags */
2834 	err = dpni_get_link_cfg(priv->mc_io, 0, priv->mc_token, &link_cfg);
2835 	if (err) {
2836 		dev_err(dev, "dpni_get_link_cfg() failed\n");
2837 		return err;
2838 	}
2839 
2840 	/* By default, enable both Rx and Tx pause frames */
2841 	link_cfg.options |= DPNI_LINK_OPT_PAUSE;
2842 	link_cfg.options &= ~DPNI_LINK_OPT_ASYM_PAUSE;
2843 	err = dpni_set_link_cfg(priv->mc_io, 0, priv->mc_token, &link_cfg);
2844 	if (err) {
2845 		dev_err(dev, "dpni_set_link_cfg() failed\n");
2846 		return err;
2847 	}
2848 
2849 	priv->link_state.options = link_cfg.options;
2850 
2851 	return 0;
2852 }
2853 
2854 static void dpaa2_eth_update_tx_fqids(struct dpaa2_eth_priv *priv)
2855 {
2856 	struct dpni_queue_id qid = {0};
2857 	struct dpaa2_eth_fq *fq;
2858 	struct dpni_queue queue;
2859 	int i, j, err;
2860 
2861 	/* We only use Tx FQIDs for FQID-based enqueue, so check
2862 	 * if DPNI version supports it before updating FQIDs
2863 	 */
2864 	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_ENQUEUE_FQID_VER_MAJOR,
2865 				   DPNI_ENQUEUE_FQID_VER_MINOR) < 0)
2866 		return;
2867 
2868 	for (i = 0; i < priv->num_fqs; i++) {
2869 		fq = &priv->fq[i];
2870 		if (fq->type != DPAA2_TX_CONF_FQ)
2871 			continue;
2872 		for (j = 0; j < dpaa2_eth_tc_count(priv); j++) {
2873 			err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
2874 					     DPNI_QUEUE_TX, j, fq->flowid,
2875 					     &queue, &qid);
2876 			if (err)
2877 				goto out_err;
2878 
2879 			fq->tx_fqid[j] = qid.fqid;
2880 			if (fq->tx_fqid[j] == 0)
2881 				goto out_err;
2882 		}
2883 	}
2884 
2885 	priv->enqueue = dpaa2_eth_enqueue_fq_multiple;
2886 
2887 	return;
2888 
2889 out_err:
2890 	netdev_info(priv->net_dev,
2891 		    "Error reading Tx FQID, fallback to QDID-based enqueue\n");
2892 	priv->enqueue = dpaa2_eth_enqueue_qd;
2893 }
2894 
2895 /* Configure ingress classification based on VLAN PCP */
2896 static int dpaa2_eth_set_vlan_qos(struct dpaa2_eth_priv *priv)
2897 {
2898 	struct device *dev = priv->net_dev->dev.parent;
2899 	struct dpkg_profile_cfg kg_cfg = {0};
2900 	struct dpni_qos_tbl_cfg qos_cfg = {0};
2901 	struct dpni_rule_cfg key_params;
2902 	void *dma_mem, *key, *mask;
2903 	u8 key_size = 2;	/* VLAN TCI field */
2904 	int i, pcp, err;
2905 
2906 	/* VLAN-based classification only makes sense if we have multiple
2907 	 * traffic classes.
2908 	 * Also, we need to extract just the 3-bit PCP field from the VLAN
2909 	 * header and we can only do that by using a mask
2910 	 */
2911 	if (dpaa2_eth_tc_count(priv) == 1 || !dpaa2_eth_fs_mask_enabled(priv)) {
2912 		dev_dbg(dev, "VLAN-based QoS classification not supported\n");
2913 		return -EOPNOTSUPP;
2914 	}
2915 
2916 	dma_mem = kzalloc(DPAA2_CLASSIFIER_DMA_SIZE, GFP_KERNEL);
2917 	if (!dma_mem)
2918 		return -ENOMEM;
2919 
2920 	kg_cfg.num_extracts = 1;
2921 	kg_cfg.extracts[0].type = DPKG_EXTRACT_FROM_HDR;
2922 	kg_cfg.extracts[0].extract.from_hdr.prot = NET_PROT_VLAN;
2923 	kg_cfg.extracts[0].extract.from_hdr.type = DPKG_FULL_FIELD;
2924 	kg_cfg.extracts[0].extract.from_hdr.field = NH_FLD_VLAN_TCI;
2925 
2926 	err = dpni_prepare_key_cfg(&kg_cfg, dma_mem);
2927 	if (err) {
2928 		dev_err(dev, "dpni_prepare_key_cfg failed\n");
2929 		goto out_free_tbl;
2930 	}
2931 
2932 	/* set QoS table */
2933 	qos_cfg.default_tc = 0;
2934 	qos_cfg.discard_on_miss = 0;
2935 	qos_cfg.key_cfg_iova = dma_map_single(dev, dma_mem,
2936 					      DPAA2_CLASSIFIER_DMA_SIZE,
2937 					      DMA_TO_DEVICE);
2938 	if (dma_mapping_error(dev, qos_cfg.key_cfg_iova)) {
2939 		dev_err(dev, "QoS table DMA mapping failed\n");
2940 		err = -ENOMEM;
2941 		goto out_free_tbl;
2942 	}
2943 
2944 	err = dpni_set_qos_table(priv->mc_io, 0, priv->mc_token, &qos_cfg);
2945 	if (err) {
2946 		dev_err(dev, "dpni_set_qos_table failed\n");
2947 		goto out_unmap_tbl;
2948 	}
2949 
2950 	/* Add QoS table entries */
2951 	key = kzalloc(key_size * 2, GFP_KERNEL);
2952 	if (!key) {
2953 		err = -ENOMEM;
2954 		goto out_unmap_tbl;
2955 	}
2956 	mask = key + key_size;
2957 	*(__be16 *)mask = cpu_to_be16(VLAN_PRIO_MASK);
2958 
2959 	key_params.key_iova = dma_map_single(dev, key, key_size * 2,
2960 					     DMA_TO_DEVICE);
2961 	if (dma_mapping_error(dev, key_params.key_iova)) {
2962 		dev_err(dev, "Qos table entry DMA mapping failed\n");
2963 		err = -ENOMEM;
2964 		goto out_free_key;
2965 	}
2966 
2967 	key_params.mask_iova = key_params.key_iova + key_size;
2968 	key_params.key_size = key_size;
2969 
2970 	/* We add rules for PCP-based distribution starting with highest
2971 	 * priority (VLAN PCP = 7). If this DPNI doesn't have enough traffic
2972 	 * classes to accommodate all priority levels, the lowest ones end up
2973 	 * on TC 0 which was configured as default
2974 	 */
2975 	for (i = dpaa2_eth_tc_count(priv) - 1, pcp = 7; i >= 0; i--, pcp--) {
2976 		*(__be16 *)key = cpu_to_be16(pcp << VLAN_PRIO_SHIFT);
2977 		dma_sync_single_for_device(dev, key_params.key_iova,
2978 					   key_size * 2, DMA_TO_DEVICE);
2979 
2980 		err = dpni_add_qos_entry(priv->mc_io, 0, priv->mc_token,
2981 					 &key_params, i, i);
2982 		if (err) {
2983 			dev_err(dev, "dpni_add_qos_entry failed\n");
2984 			dpni_clear_qos_table(priv->mc_io, 0, priv->mc_token);
2985 			goto out_unmap_key;
2986 		}
2987 	}
2988 
2989 	priv->vlan_cls_enabled = true;
2990 
2991 	/* Table and key memory is not persistent, clean everything up after
2992 	 * configuration is finished
2993 	 */
2994 out_unmap_key:
2995 	dma_unmap_single(dev, key_params.key_iova, key_size * 2, DMA_TO_DEVICE);
2996 out_free_key:
2997 	kfree(key);
2998 out_unmap_tbl:
2999 	dma_unmap_single(dev, qos_cfg.key_cfg_iova, DPAA2_CLASSIFIER_DMA_SIZE,
3000 			 DMA_TO_DEVICE);
3001 out_free_tbl:
3002 	kfree(dma_mem);
3003 
3004 	return err;
3005 }
3006 
3007 /* Configure the DPNI object this interface is associated with */
3008 static int dpaa2_eth_setup_dpni(struct fsl_mc_device *ls_dev)
3009 {
3010 	struct device *dev = &ls_dev->dev;
3011 	struct dpaa2_eth_priv *priv;
3012 	struct net_device *net_dev;
3013 	int err;
3014 
3015 	net_dev = dev_get_drvdata(dev);
3016 	priv = netdev_priv(net_dev);
3017 
3018 	/* get a handle for the DPNI object */
3019 	err = dpni_open(priv->mc_io, 0, ls_dev->obj_desc.id, &priv->mc_token);
3020 	if (err) {
3021 		dev_err(dev, "dpni_open() failed\n");
3022 		return err;
3023 	}
3024 
3025 	/* Check if we can work with this DPNI object */
3026 	err = dpni_get_api_version(priv->mc_io, 0, &priv->dpni_ver_major,
3027 				   &priv->dpni_ver_minor);
3028 	if (err) {
3029 		dev_err(dev, "dpni_get_api_version() failed\n");
3030 		goto close;
3031 	}
3032 	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_VER_MAJOR, DPNI_VER_MINOR) < 0) {
3033 		dev_err(dev, "DPNI version %u.%u not supported, need >= %u.%u\n",
3034 			priv->dpni_ver_major, priv->dpni_ver_minor,
3035 			DPNI_VER_MAJOR, DPNI_VER_MINOR);
3036 		err = -ENOTSUPP;
3037 		goto close;
3038 	}
3039 
3040 	ls_dev->mc_io = priv->mc_io;
3041 	ls_dev->mc_handle = priv->mc_token;
3042 
3043 	err = dpni_reset(priv->mc_io, 0, priv->mc_token);
3044 	if (err) {
3045 		dev_err(dev, "dpni_reset() failed\n");
3046 		goto close;
3047 	}
3048 
3049 	err = dpni_get_attributes(priv->mc_io, 0, priv->mc_token,
3050 				  &priv->dpni_attrs);
3051 	if (err) {
3052 		dev_err(dev, "dpni_get_attributes() failed (err=%d)\n", err);
3053 		goto close;
3054 	}
3055 
3056 	err = dpaa2_eth_set_buffer_layout(priv);
3057 	if (err)
3058 		goto close;
3059 
3060 	dpaa2_eth_set_enqueue_mode(priv);
3061 
3062 	/* Enable pause frame support */
3063 	if (dpaa2_eth_has_pause_support(priv)) {
3064 		err = dpaa2_eth_set_pause(priv);
3065 		if (err)
3066 			goto close;
3067 	}
3068 
3069 	err = dpaa2_eth_set_vlan_qos(priv);
3070 	if (err && err != -EOPNOTSUPP)
3071 		goto close;
3072 
3073 	priv->cls_rules = devm_kcalloc(dev, dpaa2_eth_fs_count(priv),
3074 				       sizeof(struct dpaa2_eth_cls_rule),
3075 				       GFP_KERNEL);
3076 	if (!priv->cls_rules) {
3077 		err = -ENOMEM;
3078 		goto close;
3079 	}
3080 
3081 	return 0;
3082 
3083 close:
3084 	dpni_close(priv->mc_io, 0, priv->mc_token);
3085 
3086 	return err;
3087 }
3088 
3089 static void dpaa2_eth_free_dpni(struct dpaa2_eth_priv *priv)
3090 {
3091 	int err;
3092 
3093 	err = dpni_reset(priv->mc_io, 0, priv->mc_token);
3094 	if (err)
3095 		netdev_warn(priv->net_dev, "dpni_reset() failed (err %d)\n",
3096 			    err);
3097 
3098 	dpni_close(priv->mc_io, 0, priv->mc_token);
3099 }
3100 
3101 static int dpaa2_eth_setup_rx_flow(struct dpaa2_eth_priv *priv,
3102 				   struct dpaa2_eth_fq *fq)
3103 {
3104 	struct device *dev = priv->net_dev->dev.parent;
3105 	struct dpni_queue queue;
3106 	struct dpni_queue_id qid;
3107 	int err;
3108 
3109 	err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3110 			     DPNI_QUEUE_RX, fq->tc, fq->flowid, &queue, &qid);
3111 	if (err) {
3112 		dev_err(dev, "dpni_get_queue(RX) failed\n");
3113 		return err;
3114 	}
3115 
3116 	fq->fqid = qid.fqid;
3117 
3118 	queue.destination.id = fq->channel->dpcon_id;
3119 	queue.destination.type = DPNI_DEST_DPCON;
3120 	queue.destination.priority = 1;
3121 	queue.user_context = (u64)(uintptr_t)fq;
3122 	err = dpni_set_queue(priv->mc_io, 0, priv->mc_token,
3123 			     DPNI_QUEUE_RX, fq->tc, fq->flowid,
3124 			     DPNI_QUEUE_OPT_USER_CTX | DPNI_QUEUE_OPT_DEST,
3125 			     &queue);
3126 	if (err) {
3127 		dev_err(dev, "dpni_set_queue(RX) failed\n");
3128 		return err;
3129 	}
3130 
3131 	/* xdp_rxq setup */
3132 	/* only once for each channel */
3133 	if (fq->tc > 0)
3134 		return 0;
3135 
3136 	err = xdp_rxq_info_reg(&fq->channel->xdp_rxq, priv->net_dev,
3137 			       fq->flowid);
3138 	if (err) {
3139 		dev_err(dev, "xdp_rxq_info_reg failed\n");
3140 		return err;
3141 	}
3142 
3143 	err = xdp_rxq_info_reg_mem_model(&fq->channel->xdp_rxq,
3144 					 MEM_TYPE_PAGE_ORDER0, NULL);
3145 	if (err) {
3146 		dev_err(dev, "xdp_rxq_info_reg_mem_model failed\n");
3147 		return err;
3148 	}
3149 
3150 	return 0;
3151 }
3152 
3153 static int dpaa2_eth_setup_tx_flow(struct dpaa2_eth_priv *priv,
3154 				   struct dpaa2_eth_fq *fq)
3155 {
3156 	struct device *dev = priv->net_dev->dev.parent;
3157 	struct dpni_queue queue;
3158 	struct dpni_queue_id qid;
3159 	int i, err;
3160 
3161 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3162 		err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3163 				     DPNI_QUEUE_TX, i, fq->flowid,
3164 				     &queue, &qid);
3165 		if (err) {
3166 			dev_err(dev, "dpni_get_queue(TX) failed\n");
3167 			return err;
3168 		}
3169 		fq->tx_fqid[i] = qid.fqid;
3170 	}
3171 
3172 	/* All Tx queues belonging to the same flowid have the same qdbin */
3173 	fq->tx_qdbin = qid.qdbin;
3174 
3175 	err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3176 			     DPNI_QUEUE_TX_CONFIRM, 0, fq->flowid,
3177 			     &queue, &qid);
3178 	if (err) {
3179 		dev_err(dev, "dpni_get_queue(TX_CONF) failed\n");
3180 		return err;
3181 	}
3182 
3183 	fq->fqid = qid.fqid;
3184 
3185 	queue.destination.id = fq->channel->dpcon_id;
3186 	queue.destination.type = DPNI_DEST_DPCON;
3187 	queue.destination.priority = 0;
3188 	queue.user_context = (u64)(uintptr_t)fq;
3189 	err = dpni_set_queue(priv->mc_io, 0, priv->mc_token,
3190 			     DPNI_QUEUE_TX_CONFIRM, 0, fq->flowid,
3191 			     DPNI_QUEUE_OPT_USER_CTX | DPNI_QUEUE_OPT_DEST,
3192 			     &queue);
3193 	if (err) {
3194 		dev_err(dev, "dpni_set_queue(TX_CONF) failed\n");
3195 		return err;
3196 	}
3197 
3198 	return 0;
3199 }
3200 
3201 /* Supported header fields for Rx hash distribution key */
3202 static const struct dpaa2_eth_dist_fields dist_fields[] = {
3203 	{
3204 		/* L2 header */
3205 		.rxnfc_field = RXH_L2DA,
3206 		.cls_prot = NET_PROT_ETH,
3207 		.cls_field = NH_FLD_ETH_DA,
3208 		.id = DPAA2_ETH_DIST_ETHDST,
3209 		.size = 6,
3210 	}, {
3211 		.cls_prot = NET_PROT_ETH,
3212 		.cls_field = NH_FLD_ETH_SA,
3213 		.id = DPAA2_ETH_DIST_ETHSRC,
3214 		.size = 6,
3215 	}, {
3216 		/* This is the last ethertype field parsed:
3217 		 * depending on frame format, it can be the MAC ethertype
3218 		 * or the VLAN etype.
3219 		 */
3220 		.cls_prot = NET_PROT_ETH,
3221 		.cls_field = NH_FLD_ETH_TYPE,
3222 		.id = DPAA2_ETH_DIST_ETHTYPE,
3223 		.size = 2,
3224 	}, {
3225 		/* VLAN header */
3226 		.rxnfc_field = RXH_VLAN,
3227 		.cls_prot = NET_PROT_VLAN,
3228 		.cls_field = NH_FLD_VLAN_TCI,
3229 		.id = DPAA2_ETH_DIST_VLAN,
3230 		.size = 2,
3231 	}, {
3232 		/* IP header */
3233 		.rxnfc_field = RXH_IP_SRC,
3234 		.cls_prot = NET_PROT_IP,
3235 		.cls_field = NH_FLD_IP_SRC,
3236 		.id = DPAA2_ETH_DIST_IPSRC,
3237 		.size = 4,
3238 	}, {
3239 		.rxnfc_field = RXH_IP_DST,
3240 		.cls_prot = NET_PROT_IP,
3241 		.cls_field = NH_FLD_IP_DST,
3242 		.id = DPAA2_ETH_DIST_IPDST,
3243 		.size = 4,
3244 	}, {
3245 		.rxnfc_field = RXH_L3_PROTO,
3246 		.cls_prot = NET_PROT_IP,
3247 		.cls_field = NH_FLD_IP_PROTO,
3248 		.id = DPAA2_ETH_DIST_IPPROTO,
3249 		.size = 1,
3250 	}, {
3251 		/* Using UDP ports, this is functionally equivalent to raw
3252 		 * byte pairs from L4 header.
3253 		 */
3254 		.rxnfc_field = RXH_L4_B_0_1,
3255 		.cls_prot = NET_PROT_UDP,
3256 		.cls_field = NH_FLD_UDP_PORT_SRC,
3257 		.id = DPAA2_ETH_DIST_L4SRC,
3258 		.size = 2,
3259 	}, {
3260 		.rxnfc_field = RXH_L4_B_2_3,
3261 		.cls_prot = NET_PROT_UDP,
3262 		.cls_field = NH_FLD_UDP_PORT_DST,
3263 		.id = DPAA2_ETH_DIST_L4DST,
3264 		.size = 2,
3265 	},
3266 };
3267 
3268 /* Configure the Rx hash key using the legacy API */
3269 static int dpaa2_eth_config_legacy_hash_key(struct dpaa2_eth_priv *priv, dma_addr_t key)
3270 {
3271 	struct device *dev = priv->net_dev->dev.parent;
3272 	struct dpni_rx_tc_dist_cfg dist_cfg;
3273 	int i, err = 0;
3274 
3275 	memset(&dist_cfg, 0, sizeof(dist_cfg));
3276 
3277 	dist_cfg.key_cfg_iova = key;
3278 	dist_cfg.dist_size = dpaa2_eth_queue_count(priv);
3279 	dist_cfg.dist_mode = DPNI_DIST_MODE_HASH;
3280 
3281 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3282 		err = dpni_set_rx_tc_dist(priv->mc_io, 0, priv->mc_token,
3283 					  i, &dist_cfg);
3284 		if (err) {
3285 			dev_err(dev, "dpni_set_rx_tc_dist failed\n");
3286 			break;
3287 		}
3288 	}
3289 
3290 	return err;
3291 }
3292 
3293 /* Configure the Rx hash key using the new API */
3294 static int dpaa2_eth_config_hash_key(struct dpaa2_eth_priv *priv, dma_addr_t key)
3295 {
3296 	struct device *dev = priv->net_dev->dev.parent;
3297 	struct dpni_rx_dist_cfg dist_cfg;
3298 	int i, err = 0;
3299 
3300 	memset(&dist_cfg, 0, sizeof(dist_cfg));
3301 
3302 	dist_cfg.key_cfg_iova = key;
3303 	dist_cfg.dist_size = dpaa2_eth_queue_count(priv);
3304 	dist_cfg.enable = 1;
3305 
3306 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3307 		dist_cfg.tc = i;
3308 		err = dpni_set_rx_hash_dist(priv->mc_io, 0, priv->mc_token,
3309 					    &dist_cfg);
3310 		if (err) {
3311 			dev_err(dev, "dpni_set_rx_hash_dist failed\n");
3312 			break;
3313 		}
3314 	}
3315 
3316 	return err;
3317 }
3318 
3319 /* Configure the Rx flow classification key */
3320 static int dpaa2_eth_config_cls_key(struct dpaa2_eth_priv *priv, dma_addr_t key)
3321 {
3322 	struct device *dev = priv->net_dev->dev.parent;
3323 	struct dpni_rx_dist_cfg dist_cfg;
3324 	int i, err = 0;
3325 
3326 	memset(&dist_cfg, 0, sizeof(dist_cfg));
3327 
3328 	dist_cfg.key_cfg_iova = key;
3329 	dist_cfg.dist_size = dpaa2_eth_queue_count(priv);
3330 	dist_cfg.enable = 1;
3331 
3332 	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3333 		dist_cfg.tc = i;
3334 		err = dpni_set_rx_fs_dist(priv->mc_io, 0, priv->mc_token,
3335 					  &dist_cfg);
3336 		if (err) {
3337 			dev_err(dev, "dpni_set_rx_fs_dist failed\n");
3338 			break;
3339 		}
3340 	}
3341 
3342 	return err;
3343 }
3344 
3345 /* Size of the Rx flow classification key */
3346 int dpaa2_eth_cls_key_size(u64 fields)
3347 {
3348 	int i, size = 0;
3349 
3350 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3351 		if (!(fields & dist_fields[i].id))
3352 			continue;
3353 		size += dist_fields[i].size;
3354 	}
3355 
3356 	return size;
3357 }
3358 
3359 /* Offset of header field in Rx classification key */
3360 int dpaa2_eth_cls_fld_off(int prot, int field)
3361 {
3362 	int i, off = 0;
3363 
3364 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3365 		if (dist_fields[i].cls_prot == prot &&
3366 		    dist_fields[i].cls_field == field)
3367 			return off;
3368 		off += dist_fields[i].size;
3369 	}
3370 
3371 	WARN_ONCE(1, "Unsupported header field used for Rx flow cls\n");
3372 	return 0;
3373 }
3374 
3375 /* Prune unused fields from the classification rule.
3376  * Used when masking is not supported
3377  */
3378 void dpaa2_eth_cls_trim_rule(void *key_mem, u64 fields)
3379 {
3380 	int off = 0, new_off = 0;
3381 	int i, size;
3382 
3383 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3384 		size = dist_fields[i].size;
3385 		if (dist_fields[i].id & fields) {
3386 			memcpy(key_mem + new_off, key_mem + off, size);
3387 			new_off += size;
3388 		}
3389 		off += size;
3390 	}
3391 }
3392 
3393 /* Set Rx distribution (hash or flow classification) key
3394  * flags is a combination of RXH_ bits
3395  */
3396 static int dpaa2_eth_set_dist_key(struct net_device *net_dev,
3397 				  enum dpaa2_eth_rx_dist type, u64 flags)
3398 {
3399 	struct device *dev = net_dev->dev.parent;
3400 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
3401 	struct dpkg_profile_cfg cls_cfg;
3402 	u32 rx_hash_fields = 0;
3403 	dma_addr_t key_iova;
3404 	u8 *dma_mem;
3405 	int i;
3406 	int err = 0;
3407 
3408 	memset(&cls_cfg, 0, sizeof(cls_cfg));
3409 
3410 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3411 		struct dpkg_extract *key =
3412 			&cls_cfg.extracts[cls_cfg.num_extracts];
3413 
3414 		/* For both Rx hashing and classification keys
3415 		 * we set only the selected fields.
3416 		 */
3417 		if (!(flags & dist_fields[i].id))
3418 			continue;
3419 		if (type == DPAA2_ETH_RX_DIST_HASH)
3420 			rx_hash_fields |= dist_fields[i].rxnfc_field;
3421 
3422 		if (cls_cfg.num_extracts >= DPKG_MAX_NUM_OF_EXTRACTS) {
3423 			dev_err(dev, "error adding key extraction rule, too many rules?\n");
3424 			return -E2BIG;
3425 		}
3426 
3427 		key->type = DPKG_EXTRACT_FROM_HDR;
3428 		key->extract.from_hdr.prot = dist_fields[i].cls_prot;
3429 		key->extract.from_hdr.type = DPKG_FULL_FIELD;
3430 		key->extract.from_hdr.field = dist_fields[i].cls_field;
3431 		cls_cfg.num_extracts++;
3432 	}
3433 
3434 	dma_mem = kzalloc(DPAA2_CLASSIFIER_DMA_SIZE, GFP_KERNEL);
3435 	if (!dma_mem)
3436 		return -ENOMEM;
3437 
3438 	err = dpni_prepare_key_cfg(&cls_cfg, dma_mem);
3439 	if (err) {
3440 		dev_err(dev, "dpni_prepare_key_cfg error %d\n", err);
3441 		goto free_key;
3442 	}
3443 
3444 	/* Prepare for setting the rx dist */
3445 	key_iova = dma_map_single(dev, dma_mem, DPAA2_CLASSIFIER_DMA_SIZE,
3446 				  DMA_TO_DEVICE);
3447 	if (dma_mapping_error(dev, key_iova)) {
3448 		dev_err(dev, "DMA mapping failed\n");
3449 		err = -ENOMEM;
3450 		goto free_key;
3451 	}
3452 
3453 	if (type == DPAA2_ETH_RX_DIST_HASH) {
3454 		if (dpaa2_eth_has_legacy_dist(priv))
3455 			err = dpaa2_eth_config_legacy_hash_key(priv, key_iova);
3456 		else
3457 			err = dpaa2_eth_config_hash_key(priv, key_iova);
3458 	} else {
3459 		err = dpaa2_eth_config_cls_key(priv, key_iova);
3460 	}
3461 
3462 	dma_unmap_single(dev, key_iova, DPAA2_CLASSIFIER_DMA_SIZE,
3463 			 DMA_TO_DEVICE);
3464 	if (!err && type == DPAA2_ETH_RX_DIST_HASH)
3465 		priv->rx_hash_fields = rx_hash_fields;
3466 
3467 free_key:
3468 	kfree(dma_mem);
3469 	return err;
3470 }
3471 
3472 int dpaa2_eth_set_hash(struct net_device *net_dev, u64 flags)
3473 {
3474 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
3475 	u64 key = 0;
3476 	int i;
3477 
3478 	if (!dpaa2_eth_hash_enabled(priv))
3479 		return -EOPNOTSUPP;
3480 
3481 	for (i = 0; i < ARRAY_SIZE(dist_fields); i++)
3482 		if (dist_fields[i].rxnfc_field & flags)
3483 			key |= dist_fields[i].id;
3484 
3485 	return dpaa2_eth_set_dist_key(net_dev, DPAA2_ETH_RX_DIST_HASH, key);
3486 }
3487 
3488 int dpaa2_eth_set_cls(struct net_device *net_dev, u64 flags)
3489 {
3490 	return dpaa2_eth_set_dist_key(net_dev, DPAA2_ETH_RX_DIST_CLS, flags);
3491 }
3492 
3493 static int dpaa2_eth_set_default_cls(struct dpaa2_eth_priv *priv)
3494 {
3495 	struct device *dev = priv->net_dev->dev.parent;
3496 	int err;
3497 
3498 	/* Check if we actually support Rx flow classification */
3499 	if (dpaa2_eth_has_legacy_dist(priv)) {
3500 		dev_dbg(dev, "Rx cls not supported by current MC version\n");
3501 		return -EOPNOTSUPP;
3502 	}
3503 
3504 	if (!dpaa2_eth_fs_enabled(priv)) {
3505 		dev_dbg(dev, "Rx cls disabled in DPNI options\n");
3506 		return -EOPNOTSUPP;
3507 	}
3508 
3509 	if (!dpaa2_eth_hash_enabled(priv)) {
3510 		dev_dbg(dev, "Rx cls disabled for single queue DPNIs\n");
3511 		return -EOPNOTSUPP;
3512 	}
3513 
3514 	/* If there is no support for masking in the classification table,
3515 	 * we don't set a default key, as it will depend on the rules
3516 	 * added by the user at runtime.
3517 	 */
3518 	if (!dpaa2_eth_fs_mask_enabled(priv))
3519 		goto out;
3520 
3521 	err = dpaa2_eth_set_cls(priv->net_dev, DPAA2_ETH_DIST_ALL);
3522 	if (err)
3523 		return err;
3524 
3525 out:
3526 	priv->rx_cls_enabled = 1;
3527 
3528 	return 0;
3529 }
3530 
3531 /* Bind the DPNI to its needed objects and resources: buffer pool, DPIOs,
3532  * frame queues and channels
3533  */
3534 static int dpaa2_eth_bind_dpni(struct dpaa2_eth_priv *priv)
3535 {
3536 	struct net_device *net_dev = priv->net_dev;
3537 	struct device *dev = net_dev->dev.parent;
3538 	struct dpni_pools_cfg pools_params;
3539 	struct dpni_error_cfg err_cfg;
3540 	int err = 0;
3541 	int i;
3542 
3543 	pools_params.num_dpbp = 1;
3544 	pools_params.pools[0].dpbp_id = priv->dpbp_dev->obj_desc.id;
3545 	pools_params.pools[0].backup_pool = 0;
3546 	pools_params.pools[0].buffer_size = priv->rx_buf_size;
3547 	err = dpni_set_pools(priv->mc_io, 0, priv->mc_token, &pools_params);
3548 	if (err) {
3549 		dev_err(dev, "dpni_set_pools() failed\n");
3550 		return err;
3551 	}
3552 
3553 	/* have the interface implicitly distribute traffic based on
3554 	 * the default hash key
3555 	 */
3556 	err = dpaa2_eth_set_hash(net_dev, DPAA2_RXH_DEFAULT);
3557 	if (err && err != -EOPNOTSUPP)
3558 		dev_err(dev, "Failed to configure hashing\n");
3559 
3560 	/* Configure the flow classification key; it includes all
3561 	 * supported header fields and cannot be modified at runtime
3562 	 */
3563 	err = dpaa2_eth_set_default_cls(priv);
3564 	if (err && err != -EOPNOTSUPP)
3565 		dev_err(dev, "Failed to configure Rx classification key\n");
3566 
3567 	/* Configure handling of error frames */
3568 	err_cfg.errors = DPAA2_FAS_RX_ERR_MASK;
3569 	err_cfg.set_frame_annotation = 1;
3570 	err_cfg.error_action = DPNI_ERROR_ACTION_DISCARD;
3571 	err = dpni_set_errors_behavior(priv->mc_io, 0, priv->mc_token,
3572 				       &err_cfg);
3573 	if (err) {
3574 		dev_err(dev, "dpni_set_errors_behavior failed\n");
3575 		return err;
3576 	}
3577 
3578 	/* Configure Rx and Tx conf queues to generate CDANs */
3579 	for (i = 0; i < priv->num_fqs; i++) {
3580 		switch (priv->fq[i].type) {
3581 		case DPAA2_RX_FQ:
3582 			err = dpaa2_eth_setup_rx_flow(priv, &priv->fq[i]);
3583 			break;
3584 		case DPAA2_TX_CONF_FQ:
3585 			err = dpaa2_eth_setup_tx_flow(priv, &priv->fq[i]);
3586 			break;
3587 		default:
3588 			dev_err(dev, "Invalid FQ type %d\n", priv->fq[i].type);
3589 			return -EINVAL;
3590 		}
3591 		if (err)
3592 			return err;
3593 	}
3594 
3595 	err = dpni_get_qdid(priv->mc_io, 0, priv->mc_token,
3596 			    DPNI_QUEUE_TX, &priv->tx_qdid);
3597 	if (err) {
3598 		dev_err(dev, "dpni_get_qdid() failed\n");
3599 		return err;
3600 	}
3601 
3602 	return 0;
3603 }
3604 
3605 /* Allocate rings for storing incoming frame descriptors */
3606 static int dpaa2_eth_alloc_rings(struct dpaa2_eth_priv *priv)
3607 {
3608 	struct net_device *net_dev = priv->net_dev;
3609 	struct device *dev = net_dev->dev.parent;
3610 	int i;
3611 
3612 	for (i = 0; i < priv->num_channels; i++) {
3613 		priv->channel[i]->store =
3614 			dpaa2_io_store_create(DPAA2_ETH_STORE_SIZE, dev);
3615 		if (!priv->channel[i]->store) {
3616 			netdev_err(net_dev, "dpaa2_io_store_create() failed\n");
3617 			goto err_ring;
3618 		}
3619 	}
3620 
3621 	return 0;
3622 
3623 err_ring:
3624 	for (i = 0; i < priv->num_channels; i++) {
3625 		if (!priv->channel[i]->store)
3626 			break;
3627 		dpaa2_io_store_destroy(priv->channel[i]->store);
3628 	}
3629 
3630 	return -ENOMEM;
3631 }
3632 
3633 static void dpaa2_eth_free_rings(struct dpaa2_eth_priv *priv)
3634 {
3635 	int i;
3636 
3637 	for (i = 0; i < priv->num_channels; i++)
3638 		dpaa2_io_store_destroy(priv->channel[i]->store);
3639 }
3640 
3641 static int dpaa2_eth_set_mac_addr(struct dpaa2_eth_priv *priv)
3642 {
3643 	struct net_device *net_dev = priv->net_dev;
3644 	struct device *dev = net_dev->dev.parent;
3645 	u8 mac_addr[ETH_ALEN], dpni_mac_addr[ETH_ALEN];
3646 	int err;
3647 
3648 	/* Get firmware address, if any */
3649 	err = dpni_get_port_mac_addr(priv->mc_io, 0, priv->mc_token, mac_addr);
3650 	if (err) {
3651 		dev_err(dev, "dpni_get_port_mac_addr() failed\n");
3652 		return err;
3653 	}
3654 
3655 	/* Get DPNI attributes address, if any */
3656 	err = dpni_get_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
3657 					dpni_mac_addr);
3658 	if (err) {
3659 		dev_err(dev, "dpni_get_primary_mac_addr() failed\n");
3660 		return err;
3661 	}
3662 
3663 	/* First check if firmware has any address configured by bootloader */
3664 	if (!is_zero_ether_addr(mac_addr)) {
3665 		/* If the DPMAC addr != DPNI addr, update it */
3666 		if (!ether_addr_equal(mac_addr, dpni_mac_addr)) {
3667 			err = dpni_set_primary_mac_addr(priv->mc_io, 0,
3668 							priv->mc_token,
3669 							mac_addr);
3670 			if (err) {
3671 				dev_err(dev, "dpni_set_primary_mac_addr() failed\n");
3672 				return err;
3673 			}
3674 		}
3675 		memcpy(net_dev->dev_addr, mac_addr, net_dev->addr_len);
3676 	} else if (is_zero_ether_addr(dpni_mac_addr)) {
3677 		/* No MAC address configured, fill in net_dev->dev_addr
3678 		 * with a random one
3679 		 */
3680 		eth_hw_addr_random(net_dev);
3681 		dev_dbg_once(dev, "device(s) have all-zero hwaddr, replaced with random\n");
3682 
3683 		err = dpni_set_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
3684 						net_dev->dev_addr);
3685 		if (err) {
3686 			dev_err(dev, "dpni_set_primary_mac_addr() failed\n");
3687 			return err;
3688 		}
3689 
3690 		/* Override NET_ADDR_RANDOM set by eth_hw_addr_random(); for all
3691 		 * practical purposes, this will be our "permanent" mac address,
3692 		 * at least until the next reboot. This move will also permit
3693 		 * register_netdevice() to properly fill up net_dev->perm_addr.
3694 		 */
3695 		net_dev->addr_assign_type = NET_ADDR_PERM;
3696 	} else {
3697 		/* NET_ADDR_PERM is default, all we have to do is
3698 		 * fill in the device addr.
3699 		 */
3700 		memcpy(net_dev->dev_addr, dpni_mac_addr, net_dev->addr_len);
3701 	}
3702 
3703 	return 0;
3704 }
3705 
3706 static int dpaa2_eth_netdev_init(struct net_device *net_dev)
3707 {
3708 	struct device *dev = net_dev->dev.parent;
3709 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
3710 	u32 options = priv->dpni_attrs.options;
3711 	u64 supported = 0, not_supported = 0;
3712 	u8 bcast_addr[ETH_ALEN];
3713 	u8 num_queues;
3714 	int err;
3715 
3716 	net_dev->netdev_ops = &dpaa2_eth_ops;
3717 	net_dev->ethtool_ops = &dpaa2_ethtool_ops;
3718 
3719 	err = dpaa2_eth_set_mac_addr(priv);
3720 	if (err)
3721 		return err;
3722 
3723 	/* Explicitly add the broadcast address to the MAC filtering table */
3724 	eth_broadcast_addr(bcast_addr);
3725 	err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token, bcast_addr);
3726 	if (err) {
3727 		dev_err(dev, "dpni_add_mac_addr() failed\n");
3728 		return err;
3729 	}
3730 
3731 	/* Set MTU upper limit; lower limit is 68B (default value) */
3732 	net_dev->max_mtu = DPAA2_ETH_MAX_MTU;
3733 	err = dpni_set_max_frame_length(priv->mc_io, 0, priv->mc_token,
3734 					DPAA2_ETH_MFL);
3735 	if (err) {
3736 		dev_err(dev, "dpni_set_max_frame_length() failed\n");
3737 		return err;
3738 	}
3739 
3740 	/* Set actual number of queues in the net device */
3741 	num_queues = dpaa2_eth_queue_count(priv);
3742 	err = netif_set_real_num_tx_queues(net_dev, num_queues);
3743 	if (err) {
3744 		dev_err(dev, "netif_set_real_num_tx_queues() failed\n");
3745 		return err;
3746 	}
3747 	err = netif_set_real_num_rx_queues(net_dev, num_queues);
3748 	if (err) {
3749 		dev_err(dev, "netif_set_real_num_rx_queues() failed\n");
3750 		return err;
3751 	}
3752 
3753 	/* Capabilities listing */
3754 	supported |= IFF_LIVE_ADDR_CHANGE;
3755 
3756 	if (options & DPNI_OPT_NO_MAC_FILTER)
3757 		not_supported |= IFF_UNICAST_FLT;
3758 	else
3759 		supported |= IFF_UNICAST_FLT;
3760 
3761 	net_dev->priv_flags |= supported;
3762 	net_dev->priv_flags &= ~not_supported;
3763 
3764 	/* Features */
3765 	net_dev->features = NETIF_F_RXCSUM |
3766 			    NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
3767 			    NETIF_F_SG | NETIF_F_HIGHDMA |
3768 			    NETIF_F_LLTX | NETIF_F_HW_TC;
3769 	net_dev->hw_features = net_dev->features;
3770 
3771 	return 0;
3772 }
3773 
3774 static int dpaa2_eth_poll_link_state(void *arg)
3775 {
3776 	struct dpaa2_eth_priv *priv = (struct dpaa2_eth_priv *)arg;
3777 	int err;
3778 
3779 	while (!kthread_should_stop()) {
3780 		err = dpaa2_eth_link_state_update(priv);
3781 		if (unlikely(err))
3782 			return err;
3783 
3784 		msleep(DPAA2_ETH_LINK_STATE_REFRESH);
3785 	}
3786 
3787 	return 0;
3788 }
3789 
3790 static int dpaa2_eth_connect_mac(struct dpaa2_eth_priv *priv)
3791 {
3792 	struct fsl_mc_device *dpni_dev, *dpmac_dev;
3793 	struct dpaa2_mac *mac;
3794 	int err;
3795 
3796 	dpni_dev = to_fsl_mc_device(priv->net_dev->dev.parent);
3797 	dpmac_dev = fsl_mc_get_endpoint(dpni_dev);
3798 	if (IS_ERR_OR_NULL(dpmac_dev) || dpmac_dev->dev.type != &fsl_mc_bus_dpmac_type)
3799 		return 0;
3800 
3801 	if (dpaa2_mac_is_type_fixed(dpmac_dev, priv->mc_io))
3802 		return 0;
3803 
3804 	mac = kzalloc(sizeof(struct dpaa2_mac), GFP_KERNEL);
3805 	if (!mac)
3806 		return -ENOMEM;
3807 
3808 	mac->mc_dev = dpmac_dev;
3809 	mac->mc_io = priv->mc_io;
3810 	mac->net_dev = priv->net_dev;
3811 
3812 	err = dpaa2_mac_connect(mac);
3813 	if (err) {
3814 		netdev_err(priv->net_dev, "Error connecting to the MAC endpoint\n");
3815 		kfree(mac);
3816 		return err;
3817 	}
3818 	priv->mac = mac;
3819 
3820 	return 0;
3821 }
3822 
3823 static void dpaa2_eth_disconnect_mac(struct dpaa2_eth_priv *priv)
3824 {
3825 	if (!priv->mac)
3826 		return;
3827 
3828 	dpaa2_mac_disconnect(priv->mac);
3829 	kfree(priv->mac);
3830 	priv->mac = NULL;
3831 }
3832 
3833 static irqreturn_t dpni_irq0_handler_thread(int irq_num, void *arg)
3834 {
3835 	u32 status = ~0;
3836 	struct device *dev = (struct device *)arg;
3837 	struct fsl_mc_device *dpni_dev = to_fsl_mc_device(dev);
3838 	struct net_device *net_dev = dev_get_drvdata(dev);
3839 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
3840 	int err;
3841 
3842 	err = dpni_get_irq_status(dpni_dev->mc_io, 0, dpni_dev->mc_handle,
3843 				  DPNI_IRQ_INDEX, &status);
3844 	if (unlikely(err)) {
3845 		netdev_err(net_dev, "Can't get irq status (err %d)\n", err);
3846 		return IRQ_HANDLED;
3847 	}
3848 
3849 	if (status & DPNI_IRQ_EVENT_LINK_CHANGED)
3850 		dpaa2_eth_link_state_update(netdev_priv(net_dev));
3851 
3852 	if (status & DPNI_IRQ_EVENT_ENDPOINT_CHANGED) {
3853 		dpaa2_eth_set_mac_addr(netdev_priv(net_dev));
3854 		dpaa2_eth_update_tx_fqids(priv);
3855 
3856 		rtnl_lock();
3857 		if (priv->mac)
3858 			dpaa2_eth_disconnect_mac(priv);
3859 		else
3860 			dpaa2_eth_connect_mac(priv);
3861 		rtnl_unlock();
3862 	}
3863 
3864 	return IRQ_HANDLED;
3865 }
3866 
3867 static int dpaa2_eth_setup_irqs(struct fsl_mc_device *ls_dev)
3868 {
3869 	int err = 0;
3870 	struct fsl_mc_device_irq *irq;
3871 
3872 	err = fsl_mc_allocate_irqs(ls_dev);
3873 	if (err) {
3874 		dev_err(&ls_dev->dev, "MC irqs allocation failed\n");
3875 		return err;
3876 	}
3877 
3878 	irq = ls_dev->irqs[0];
3879 	err = devm_request_threaded_irq(&ls_dev->dev, irq->msi_desc->irq,
3880 					NULL, dpni_irq0_handler_thread,
3881 					IRQF_NO_SUSPEND | IRQF_ONESHOT,
3882 					dev_name(&ls_dev->dev), &ls_dev->dev);
3883 	if (err < 0) {
3884 		dev_err(&ls_dev->dev, "devm_request_threaded_irq(): %d\n", err);
3885 		goto free_mc_irq;
3886 	}
3887 
3888 	err = dpni_set_irq_mask(ls_dev->mc_io, 0, ls_dev->mc_handle,
3889 				DPNI_IRQ_INDEX, DPNI_IRQ_EVENT_LINK_CHANGED |
3890 				DPNI_IRQ_EVENT_ENDPOINT_CHANGED);
3891 	if (err < 0) {
3892 		dev_err(&ls_dev->dev, "dpni_set_irq_mask(): %d\n", err);
3893 		goto free_irq;
3894 	}
3895 
3896 	err = dpni_set_irq_enable(ls_dev->mc_io, 0, ls_dev->mc_handle,
3897 				  DPNI_IRQ_INDEX, 1);
3898 	if (err < 0) {
3899 		dev_err(&ls_dev->dev, "dpni_set_irq_enable(): %d\n", err);
3900 		goto free_irq;
3901 	}
3902 
3903 	return 0;
3904 
3905 free_irq:
3906 	devm_free_irq(&ls_dev->dev, irq->msi_desc->irq, &ls_dev->dev);
3907 free_mc_irq:
3908 	fsl_mc_free_irqs(ls_dev);
3909 
3910 	return err;
3911 }
3912 
3913 static void dpaa2_eth_add_ch_napi(struct dpaa2_eth_priv *priv)
3914 {
3915 	int i;
3916 	struct dpaa2_eth_channel *ch;
3917 
3918 	for (i = 0; i < priv->num_channels; i++) {
3919 		ch = priv->channel[i];
3920 		/* NAPI weight *MUST* be a multiple of DPAA2_ETH_STORE_SIZE */
3921 		netif_napi_add(priv->net_dev, &ch->napi, dpaa2_eth_poll,
3922 			       NAPI_POLL_WEIGHT);
3923 	}
3924 }
3925 
3926 static void dpaa2_eth_del_ch_napi(struct dpaa2_eth_priv *priv)
3927 {
3928 	int i;
3929 	struct dpaa2_eth_channel *ch;
3930 
3931 	for (i = 0; i < priv->num_channels; i++) {
3932 		ch = priv->channel[i];
3933 		netif_napi_del(&ch->napi);
3934 	}
3935 }
3936 
3937 static int dpaa2_eth_probe(struct fsl_mc_device *dpni_dev)
3938 {
3939 	struct device *dev;
3940 	struct net_device *net_dev = NULL;
3941 	struct dpaa2_eth_priv *priv = NULL;
3942 	int err = 0;
3943 
3944 	dev = &dpni_dev->dev;
3945 
3946 	/* Net device */
3947 	net_dev = alloc_etherdev_mq(sizeof(*priv), DPAA2_ETH_MAX_NETDEV_QUEUES);
3948 	if (!net_dev) {
3949 		dev_err(dev, "alloc_etherdev_mq() failed\n");
3950 		return -ENOMEM;
3951 	}
3952 
3953 	SET_NETDEV_DEV(net_dev, dev);
3954 	dev_set_drvdata(dev, net_dev);
3955 
3956 	priv = netdev_priv(net_dev);
3957 	priv->net_dev = net_dev;
3958 
3959 	priv->iommu_domain = iommu_get_domain_for_dev(dev);
3960 
3961 	/* Obtain a MC portal */
3962 	err = fsl_mc_portal_allocate(dpni_dev, FSL_MC_IO_ATOMIC_CONTEXT_PORTAL,
3963 				     &priv->mc_io);
3964 	if (err) {
3965 		if (err == -ENXIO)
3966 			err = -EPROBE_DEFER;
3967 		else
3968 			dev_err(dev, "MC portal allocation failed\n");
3969 		goto err_portal_alloc;
3970 	}
3971 
3972 	/* MC objects initialization and configuration */
3973 	err = dpaa2_eth_setup_dpni(dpni_dev);
3974 	if (err)
3975 		goto err_dpni_setup;
3976 
3977 	err = dpaa2_eth_setup_dpio(priv);
3978 	if (err)
3979 		goto err_dpio_setup;
3980 
3981 	dpaa2_eth_setup_fqs(priv);
3982 
3983 	err = dpaa2_eth_setup_dpbp(priv);
3984 	if (err)
3985 		goto err_dpbp_setup;
3986 
3987 	err = dpaa2_eth_bind_dpni(priv);
3988 	if (err)
3989 		goto err_bind;
3990 
3991 	/* Add a NAPI context for each channel */
3992 	dpaa2_eth_add_ch_napi(priv);
3993 
3994 	/* Percpu statistics */
3995 	priv->percpu_stats = alloc_percpu(*priv->percpu_stats);
3996 	if (!priv->percpu_stats) {
3997 		dev_err(dev, "alloc_percpu(percpu_stats) failed\n");
3998 		err = -ENOMEM;
3999 		goto err_alloc_percpu_stats;
4000 	}
4001 	priv->percpu_extras = alloc_percpu(*priv->percpu_extras);
4002 	if (!priv->percpu_extras) {
4003 		dev_err(dev, "alloc_percpu(percpu_extras) failed\n");
4004 		err = -ENOMEM;
4005 		goto err_alloc_percpu_extras;
4006 	}
4007 
4008 	priv->sgt_cache = alloc_percpu(*priv->sgt_cache);
4009 	if (!priv->sgt_cache) {
4010 		dev_err(dev, "alloc_percpu(sgt_cache) failed\n");
4011 		err = -ENOMEM;
4012 		goto err_alloc_sgt_cache;
4013 	}
4014 
4015 	err = dpaa2_eth_netdev_init(net_dev);
4016 	if (err)
4017 		goto err_netdev_init;
4018 
4019 	/* Configure checksum offload based on current interface flags */
4020 	err = dpaa2_eth_set_rx_csum(priv, !!(net_dev->features & NETIF_F_RXCSUM));
4021 	if (err)
4022 		goto err_csum;
4023 
4024 	err = dpaa2_eth_set_tx_csum(priv,
4025 				    !!(net_dev->features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)));
4026 	if (err)
4027 		goto err_csum;
4028 
4029 	err = dpaa2_eth_alloc_rings(priv);
4030 	if (err)
4031 		goto err_alloc_rings;
4032 
4033 #ifdef CONFIG_FSL_DPAA2_ETH_DCB
4034 	if (dpaa2_eth_has_pause_support(priv) && priv->vlan_cls_enabled) {
4035 		priv->dcbx_mode = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_IEEE;
4036 		net_dev->dcbnl_ops = &dpaa2_eth_dcbnl_ops;
4037 	} else {
4038 		dev_dbg(dev, "PFC not supported\n");
4039 	}
4040 #endif
4041 
4042 	err = dpaa2_eth_setup_irqs(dpni_dev);
4043 	if (err) {
4044 		netdev_warn(net_dev, "Failed to set link interrupt, fall back to polling\n");
4045 		priv->poll_thread = kthread_run(dpaa2_eth_poll_link_state, priv,
4046 						"%s_poll_link", net_dev->name);
4047 		if (IS_ERR(priv->poll_thread)) {
4048 			dev_err(dev, "Error starting polling thread\n");
4049 			goto err_poll_thread;
4050 		}
4051 		priv->do_link_poll = true;
4052 	}
4053 
4054 	err = dpaa2_eth_connect_mac(priv);
4055 	if (err)
4056 		goto err_connect_mac;
4057 
4058 	err = register_netdev(net_dev);
4059 	if (err < 0) {
4060 		dev_err(dev, "register_netdev() failed\n");
4061 		goto err_netdev_reg;
4062 	}
4063 
4064 #ifdef CONFIG_DEBUG_FS
4065 	dpaa2_dbg_add(priv);
4066 #endif
4067 
4068 	dev_info(dev, "Probed interface %s\n", net_dev->name);
4069 	return 0;
4070 
4071 err_netdev_reg:
4072 	dpaa2_eth_disconnect_mac(priv);
4073 err_connect_mac:
4074 	if (priv->do_link_poll)
4075 		kthread_stop(priv->poll_thread);
4076 	else
4077 		fsl_mc_free_irqs(dpni_dev);
4078 err_poll_thread:
4079 	dpaa2_eth_free_rings(priv);
4080 err_alloc_rings:
4081 err_csum:
4082 err_netdev_init:
4083 	free_percpu(priv->sgt_cache);
4084 err_alloc_sgt_cache:
4085 	free_percpu(priv->percpu_extras);
4086 err_alloc_percpu_extras:
4087 	free_percpu(priv->percpu_stats);
4088 err_alloc_percpu_stats:
4089 	dpaa2_eth_del_ch_napi(priv);
4090 err_bind:
4091 	dpaa2_eth_free_dpbp(priv);
4092 err_dpbp_setup:
4093 	dpaa2_eth_free_dpio(priv);
4094 err_dpio_setup:
4095 	dpaa2_eth_free_dpni(priv);
4096 err_dpni_setup:
4097 	fsl_mc_portal_free(priv->mc_io);
4098 err_portal_alloc:
4099 	dev_set_drvdata(dev, NULL);
4100 	free_netdev(net_dev);
4101 
4102 	return err;
4103 }
4104 
4105 static int dpaa2_eth_remove(struct fsl_mc_device *ls_dev)
4106 {
4107 	struct device *dev;
4108 	struct net_device *net_dev;
4109 	struct dpaa2_eth_priv *priv;
4110 
4111 	dev = &ls_dev->dev;
4112 	net_dev = dev_get_drvdata(dev);
4113 	priv = netdev_priv(net_dev);
4114 
4115 #ifdef CONFIG_DEBUG_FS
4116 	dpaa2_dbg_remove(priv);
4117 #endif
4118 	rtnl_lock();
4119 	dpaa2_eth_disconnect_mac(priv);
4120 	rtnl_unlock();
4121 
4122 	unregister_netdev(net_dev);
4123 
4124 	if (priv->do_link_poll)
4125 		kthread_stop(priv->poll_thread);
4126 	else
4127 		fsl_mc_free_irqs(ls_dev);
4128 
4129 	dpaa2_eth_free_rings(priv);
4130 	free_percpu(priv->sgt_cache);
4131 	free_percpu(priv->percpu_stats);
4132 	free_percpu(priv->percpu_extras);
4133 
4134 	dpaa2_eth_del_ch_napi(priv);
4135 	dpaa2_eth_free_dpbp(priv);
4136 	dpaa2_eth_free_dpio(priv);
4137 	dpaa2_eth_free_dpni(priv);
4138 
4139 	fsl_mc_portal_free(priv->mc_io);
4140 
4141 	free_netdev(net_dev);
4142 
4143 	dev_dbg(net_dev->dev.parent, "Removed interface %s\n", net_dev->name);
4144 
4145 	return 0;
4146 }
4147 
4148 static const struct fsl_mc_device_id dpaa2_eth_match_id_table[] = {
4149 	{
4150 		.vendor = FSL_MC_VENDOR_FREESCALE,
4151 		.obj_type = "dpni",
4152 	},
4153 	{ .vendor = 0x0 }
4154 };
4155 MODULE_DEVICE_TABLE(fslmc, dpaa2_eth_match_id_table);
4156 
4157 static struct fsl_mc_driver dpaa2_eth_driver = {
4158 	.driver = {
4159 		.name = KBUILD_MODNAME,
4160 		.owner = THIS_MODULE,
4161 	},
4162 	.probe = dpaa2_eth_probe,
4163 	.remove = dpaa2_eth_remove,
4164 	.match_id_table = dpaa2_eth_match_id_table
4165 };
4166 
4167 static int __init dpaa2_eth_driver_init(void)
4168 {
4169 	int err;
4170 
4171 	dpaa2_eth_dbg_init();
4172 	err = fsl_mc_driver_register(&dpaa2_eth_driver);
4173 	if (err) {
4174 		dpaa2_eth_dbg_exit();
4175 		return err;
4176 	}
4177 
4178 	return 0;
4179 }
4180 
4181 static void __exit dpaa2_eth_driver_exit(void)
4182 {
4183 	dpaa2_eth_dbg_exit();
4184 	fsl_mc_driver_unregister(&dpaa2_eth_driver);
4185 }
4186 
4187 module_init(dpaa2_eth_driver_init);
4188 module_exit(dpaa2_eth_driver_exit);
4189