1// SPDX-License-Identifier: (GPL-2.0+ OR BSD-3-Clause)
2/* Copyright 2014-2016 Freescale Semiconductor Inc.
3 * Copyright 2016-2020 NXP
4 */
5#include <linux/init.h>
6#include <linux/module.h>
7#include <linux/platform_device.h>
8#include <linux/etherdevice.h>
9#include <linux/of_net.h>
10#include <linux/interrupt.h>
11#include <linux/msi.h>
12#include <linux/kthread.h>
13#include <linux/iommu.h>
14#include <linux/fsl/mc.h>
15#include <linux/bpf.h>
16#include <linux/bpf_trace.h>
17#include <linux/fsl/ptp_qoriq.h>
18#include <linux/ptp_classify.h>
19#include <net/pkt_cls.h>
20#include <net/sock.h>
21
22#include "dpaa2-eth.h"
23
24/* CREATE_TRACE_POINTS only needs to be defined once. Other dpa files
25 * using trace events only need to #include <trace/events/sched.h>
26 */
27#define CREATE_TRACE_POINTS
28#include "dpaa2-eth-trace.h"
29
30MODULE_LICENSE("Dual BSD/GPL");
31MODULE_AUTHOR("Freescale Semiconductor, Inc");
32MODULE_DESCRIPTION("Freescale DPAA2 Ethernet Driver");
33
34struct ptp_qoriq *dpaa2_ptp;
35EXPORT_SYMBOL(dpaa2_ptp);
36
37static void *dpaa2_iova_to_virt(struct iommu_domain *domain,
38				dma_addr_t iova_addr)
39{
40	phys_addr_t phys_addr;
41
42	phys_addr = domain ? iommu_iova_to_phys(domain, iova_addr) : iova_addr;
43
44	return phys_to_virt(phys_addr);
45}
46
47static void dpaa2_eth_validate_rx_csum(struct dpaa2_eth_priv *priv,
48				       u32 fd_status,
49				       struct sk_buff *skb)
50{
51	skb_checksum_none_assert(skb);
52
53	/* HW checksum validation is disabled, nothing to do here */
54	if (!(priv->net_dev->features & NETIF_F_RXCSUM))
55		return;
56
57	/* Read checksum validation bits */
58	if (!((fd_status & DPAA2_FAS_L3CV) &&
59	      (fd_status & DPAA2_FAS_L4CV)))
60		return;
61
62	/* Inform the stack there's no need to compute L3/L4 csum anymore */
63	skb->ip_summed = CHECKSUM_UNNECESSARY;
64}
65
66/* Free a received FD.
67 * Not to be used for Tx conf FDs or on any other paths.
68 */
69static void dpaa2_eth_free_rx_fd(struct dpaa2_eth_priv *priv,
70				 const struct dpaa2_fd *fd,
71				 void *vaddr)
72{
73	struct device *dev = priv->net_dev->dev.parent;
74	dma_addr_t addr = dpaa2_fd_get_addr(fd);
75	u8 fd_format = dpaa2_fd_get_format(fd);
76	struct dpaa2_sg_entry *sgt;
77	void *sg_vaddr;
78	int i;
79
80	/* If single buffer frame, just free the data buffer */
81	if (fd_format == dpaa2_fd_single)
82		goto free_buf;
83	else if (fd_format != dpaa2_fd_sg)
84		/* We don't support any other format */
85		return;
86
87	/* For S/G frames, we first need to free all SG entries
88	 * except the first one, which was taken care of already
89	 */
90	sgt = vaddr + dpaa2_fd_get_offset(fd);
91	for (i = 1; i < DPAA2_ETH_MAX_SG_ENTRIES; i++) {
92		addr = dpaa2_sg_get_addr(&sgt[i]);
93		sg_vaddr = dpaa2_iova_to_virt(priv->iommu_domain, addr);
94		dma_unmap_page(dev, addr, priv->rx_buf_size,
95			       DMA_BIDIRECTIONAL);
96
97		free_pages((unsigned long)sg_vaddr, 0);
98		if (dpaa2_sg_is_final(&sgt[i]))
99			break;
100	}
101
102free_buf:
103	free_pages((unsigned long)vaddr, 0);
104}
105
106/* Build a linear skb based on a single-buffer frame descriptor */
107static struct sk_buff *dpaa2_eth_build_linear_skb(struct dpaa2_eth_channel *ch,
108						  const struct dpaa2_fd *fd,
109						  void *fd_vaddr)
110{
111	struct sk_buff *skb = NULL;
112	u16 fd_offset = dpaa2_fd_get_offset(fd);
113	u32 fd_length = dpaa2_fd_get_len(fd);
114
115	ch->buf_count--;
116
117	skb = build_skb(fd_vaddr, DPAA2_ETH_RX_BUF_RAW_SIZE);
118	if (unlikely(!skb))
119		return NULL;
120
121	skb_reserve(skb, fd_offset);
122	skb_put(skb, fd_length);
123
124	return skb;
125}
126
127/* Build a non linear (fragmented) skb based on a S/G table */
128static struct sk_buff *dpaa2_eth_build_frag_skb(struct dpaa2_eth_priv *priv,
129						struct dpaa2_eth_channel *ch,
130						struct dpaa2_sg_entry *sgt)
131{
132	struct sk_buff *skb = NULL;
133	struct device *dev = priv->net_dev->dev.parent;
134	void *sg_vaddr;
135	dma_addr_t sg_addr;
136	u16 sg_offset;
137	u32 sg_length;
138	struct page *page, *head_page;
139	int page_offset;
140	int i;
141
142	for (i = 0; i < DPAA2_ETH_MAX_SG_ENTRIES; i++) {
143		struct dpaa2_sg_entry *sge = &sgt[i];
144
145		/* NOTE: We only support SG entries in dpaa2_sg_single format,
146		 * but this is the only format we may receive from HW anyway
147		 */
148
149		/* Get the address and length from the S/G entry */
150		sg_addr = dpaa2_sg_get_addr(sge);
151		sg_vaddr = dpaa2_iova_to_virt(priv->iommu_domain, sg_addr);
152		dma_unmap_page(dev, sg_addr, priv->rx_buf_size,
153			       DMA_BIDIRECTIONAL);
154
155		sg_length = dpaa2_sg_get_len(sge);
156
157		if (i == 0) {
158			/* We build the skb around the first data buffer */
159			skb = build_skb(sg_vaddr, DPAA2_ETH_RX_BUF_RAW_SIZE);
160			if (unlikely(!skb)) {
161				/* Free the first SG entry now, since we already
162				 * unmapped it and obtained the virtual address
163				 */
164				free_pages((unsigned long)sg_vaddr, 0);
165
166				/* We still need to subtract the buffers used
167				 * by this FD from our software counter
168				 */
169				while (!dpaa2_sg_is_final(&sgt[i]) &&
170				       i < DPAA2_ETH_MAX_SG_ENTRIES)
171					i++;
172				break;
173			}
174
175			sg_offset = dpaa2_sg_get_offset(sge);
176			skb_reserve(skb, sg_offset);
177			skb_put(skb, sg_length);
178		} else {
179			/* Rest of the data buffers are stored as skb frags */
180			page = virt_to_page(sg_vaddr);
181			head_page = virt_to_head_page(sg_vaddr);
182
183			/* Offset in page (which may be compound).
184			 * Data in subsequent SG entries is stored from the
185			 * beginning of the buffer, so we don't need to add the
186			 * sg_offset.
187			 */
188			page_offset = ((unsigned long)sg_vaddr &
189				(PAGE_SIZE - 1)) +
190				(page_address(page) - page_address(head_page));
191
192			skb_add_rx_frag(skb, i - 1, head_page, page_offset,
193					sg_length, priv->rx_buf_size);
194		}
195
196		if (dpaa2_sg_is_final(sge))
197			break;
198	}
199
200	WARN_ONCE(i == DPAA2_ETH_MAX_SG_ENTRIES, "Final bit not set in SGT");
201
202	/* Count all data buffers + SG table buffer */
203	ch->buf_count -= i + 2;
204
205	return skb;
206}
207
208/* Free buffers acquired from the buffer pool or which were meant to
209 * be released in the pool
210 */
211static void dpaa2_eth_free_bufs(struct dpaa2_eth_priv *priv, u64 *buf_array,
212				int count)
213{
214	struct device *dev = priv->net_dev->dev.parent;
215	void *vaddr;
216	int i;
217
218	for (i = 0; i < count; i++) {
219		vaddr = dpaa2_iova_to_virt(priv->iommu_domain, buf_array[i]);
220		dma_unmap_page(dev, buf_array[i], priv->rx_buf_size,
221			       DMA_BIDIRECTIONAL);
222		free_pages((unsigned long)vaddr, 0);
223	}
224}
225
226static void dpaa2_eth_xdp_release_buf(struct dpaa2_eth_priv *priv,
227				      struct dpaa2_eth_channel *ch,
228				      dma_addr_t addr)
229{
230	int retries = 0;
231	int err;
232
233	ch->xdp.drop_bufs[ch->xdp.drop_cnt++] = addr;
234	if (ch->xdp.drop_cnt < DPAA2_ETH_BUFS_PER_CMD)
235		return;
236
237	while ((err = dpaa2_io_service_release(ch->dpio, priv->bpid,
238					       ch->xdp.drop_bufs,
239					       ch->xdp.drop_cnt)) == -EBUSY) {
240		if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES)
241			break;
242		cpu_relax();
243	}
244
245	if (err) {
246		dpaa2_eth_free_bufs(priv, ch->xdp.drop_bufs, ch->xdp.drop_cnt);
247		ch->buf_count -= ch->xdp.drop_cnt;
248	}
249
250	ch->xdp.drop_cnt = 0;
251}
252
253static int dpaa2_eth_xdp_flush(struct dpaa2_eth_priv *priv,
254			       struct dpaa2_eth_fq *fq,
255			       struct dpaa2_eth_xdp_fds *xdp_fds)
256{
257	int total_enqueued = 0, retries = 0, enqueued;
258	struct dpaa2_eth_drv_stats *percpu_extras;
259	int num_fds, err, max_retries;
260	struct dpaa2_fd *fds;
261
262	percpu_extras = this_cpu_ptr(priv->percpu_extras);
263
264	/* try to enqueue all the FDs until the max number of retries is hit */
265	fds = xdp_fds->fds;
266	num_fds = xdp_fds->num;
267	max_retries = num_fds * DPAA2_ETH_ENQUEUE_RETRIES;
268	while (total_enqueued < num_fds && retries < max_retries) {
269		err = priv->enqueue(priv, fq, &fds[total_enqueued],
270				    0, num_fds - total_enqueued, &enqueued);
271		if (err == -EBUSY) {
272			percpu_extras->tx_portal_busy += ++retries;
273			continue;
274		}
275		total_enqueued += enqueued;
276	}
277	xdp_fds->num = 0;
278
279	return total_enqueued;
280}
281
282static void dpaa2_eth_xdp_tx_flush(struct dpaa2_eth_priv *priv,
283				   struct dpaa2_eth_channel *ch,
284				   struct dpaa2_eth_fq *fq)
285{
286	struct rtnl_link_stats64 *percpu_stats;
287	struct dpaa2_fd *fds;
288	int enqueued, i;
289
290	percpu_stats = this_cpu_ptr(priv->percpu_stats);
291
292	// enqueue the array of XDP_TX frames
293	enqueued = dpaa2_eth_xdp_flush(priv, fq, &fq->xdp_tx_fds);
294
295	/* update statistics */
296	percpu_stats->tx_packets += enqueued;
297	fds = fq->xdp_tx_fds.fds;
298	for (i = 0; i < enqueued; i++) {
299		percpu_stats->tx_bytes += dpaa2_fd_get_len(&fds[i]);
300		ch->stats.xdp_tx++;
301	}
302	for (i = enqueued; i < fq->xdp_tx_fds.num; i++) {
303		dpaa2_eth_xdp_release_buf(priv, ch, dpaa2_fd_get_addr(&fds[i]));
304		percpu_stats->tx_errors++;
305		ch->stats.xdp_tx_err++;
306	}
307	fq->xdp_tx_fds.num = 0;
308}
309
310static void dpaa2_eth_xdp_enqueue(struct dpaa2_eth_priv *priv,
311				  struct dpaa2_eth_channel *ch,
312				  struct dpaa2_fd *fd,
313				  void *buf_start, u16 queue_id)
314{
315	struct dpaa2_faead *faead;
316	struct dpaa2_fd *dest_fd;
317	struct dpaa2_eth_fq *fq;
318	u32 ctrl, frc;
319
320	/* Mark the egress frame hardware annotation area as valid */
321	frc = dpaa2_fd_get_frc(fd);
322	dpaa2_fd_set_frc(fd, frc | DPAA2_FD_FRC_FAEADV);
323	dpaa2_fd_set_ctrl(fd, DPAA2_FD_CTRL_ASAL);
324
325	/* Instruct hardware to release the FD buffer directly into
326	 * the buffer pool once transmission is completed, instead of
327	 * sending a Tx confirmation frame to us
328	 */
329	ctrl = DPAA2_FAEAD_A4V | DPAA2_FAEAD_A2V | DPAA2_FAEAD_EBDDV;
330	faead = dpaa2_get_faead(buf_start, false);
331	faead->ctrl = cpu_to_le32(ctrl);
332	faead->conf_fqid = 0;
333
334	fq = &priv->fq[queue_id];
335	dest_fd = &fq->xdp_tx_fds.fds[fq->xdp_tx_fds.num++];
336	memcpy(dest_fd, fd, sizeof(*dest_fd));
337
338	if (fq->xdp_tx_fds.num < DEV_MAP_BULK_SIZE)
339		return;
340
341	dpaa2_eth_xdp_tx_flush(priv, ch, fq);
342}
343
344static u32 dpaa2_eth_run_xdp(struct dpaa2_eth_priv *priv,
345			     struct dpaa2_eth_channel *ch,
346			     struct dpaa2_eth_fq *rx_fq,
347			     struct dpaa2_fd *fd, void *vaddr)
348{
349	dma_addr_t addr = dpaa2_fd_get_addr(fd);
350	struct bpf_prog *xdp_prog;
351	struct xdp_buff xdp;
352	u32 xdp_act = XDP_PASS;
353	int err;
354
355	rcu_read_lock();
356
357	xdp_prog = READ_ONCE(ch->xdp.prog);
358	if (!xdp_prog)
359		goto out;
360
361	xdp.data = vaddr + dpaa2_fd_get_offset(fd);
362	xdp.data_end = xdp.data + dpaa2_fd_get_len(fd);
363	xdp.data_hard_start = xdp.data - XDP_PACKET_HEADROOM;
364	xdp_set_data_meta_invalid(&xdp);
365	xdp.rxq = &ch->xdp_rxq;
366
367	xdp.frame_sz = DPAA2_ETH_RX_BUF_RAW_SIZE -
368		(dpaa2_fd_get_offset(fd) - XDP_PACKET_HEADROOM);
369
370	xdp_act = bpf_prog_run_xdp(xdp_prog, &xdp);
371
372	/* xdp.data pointer may have changed */
373	dpaa2_fd_set_offset(fd, xdp.data - vaddr);
374	dpaa2_fd_set_len(fd, xdp.data_end - xdp.data);
375
376	switch (xdp_act) {
377	case XDP_PASS:
378		break;
379	case XDP_TX:
380		dpaa2_eth_xdp_enqueue(priv, ch, fd, vaddr, rx_fq->flowid);
381		break;
382	default:
383		bpf_warn_invalid_xdp_action(xdp_act);
384		fallthrough;
385	case XDP_ABORTED:
386		trace_xdp_exception(priv->net_dev, xdp_prog, xdp_act);
387		fallthrough;
388	case XDP_DROP:
389		dpaa2_eth_xdp_release_buf(priv, ch, addr);
390		ch->stats.xdp_drop++;
391		break;
392	case XDP_REDIRECT:
393		dma_unmap_page(priv->net_dev->dev.parent, addr,
394			       priv->rx_buf_size, DMA_BIDIRECTIONAL);
395		ch->buf_count--;
396
397		/* Allow redirect use of full headroom */
398		xdp.data_hard_start = vaddr;
399		xdp.frame_sz = DPAA2_ETH_RX_BUF_RAW_SIZE;
400
401		err = xdp_do_redirect(priv->net_dev, &xdp, xdp_prog);
402		if (unlikely(err)) {
403			addr = dma_map_page(priv->net_dev->dev.parent,
404					    virt_to_page(vaddr), 0,
405					    priv->rx_buf_size, DMA_BIDIRECTIONAL);
406			if (unlikely(dma_mapping_error(priv->net_dev->dev.parent, addr))) {
407				free_pages((unsigned long)vaddr, 0);
408			} else {
409				ch->buf_count++;
410				dpaa2_eth_xdp_release_buf(priv, ch, addr);
411			}
412			ch->stats.xdp_drop++;
413		} else {
414			ch->stats.xdp_redirect++;
415		}
416		break;
417	}
418
419	ch->xdp.res |= xdp_act;
420out:
421	rcu_read_unlock();
422	return xdp_act;
423}
424
425/* Main Rx frame processing routine */
426static void dpaa2_eth_rx(struct dpaa2_eth_priv *priv,
427			 struct dpaa2_eth_channel *ch,
428			 const struct dpaa2_fd *fd,
429			 struct dpaa2_eth_fq *fq)
430{
431	dma_addr_t addr = dpaa2_fd_get_addr(fd);
432	u8 fd_format = dpaa2_fd_get_format(fd);
433	void *vaddr;
434	struct sk_buff *skb;
435	struct rtnl_link_stats64 *percpu_stats;
436	struct dpaa2_eth_drv_stats *percpu_extras;
437	struct device *dev = priv->net_dev->dev.parent;
438	struct dpaa2_fas *fas;
439	void *buf_data;
440	u32 status = 0;
441	u32 xdp_act;
442
443	/* Tracing point */
444	trace_dpaa2_rx_fd(priv->net_dev, fd);
445
446	vaddr = dpaa2_iova_to_virt(priv->iommu_domain, addr);
447	dma_sync_single_for_cpu(dev, addr, priv->rx_buf_size,
448				DMA_BIDIRECTIONAL);
449
450	fas = dpaa2_get_fas(vaddr, false);
451	prefetch(fas);
452	buf_data = vaddr + dpaa2_fd_get_offset(fd);
453	prefetch(buf_data);
454
455	percpu_stats = this_cpu_ptr(priv->percpu_stats);
456	percpu_extras = this_cpu_ptr(priv->percpu_extras);
457
458	if (fd_format == dpaa2_fd_single) {
459		xdp_act = dpaa2_eth_run_xdp(priv, ch, fq, (struct dpaa2_fd *)fd, vaddr);
460		if (xdp_act != XDP_PASS) {
461			percpu_stats->rx_packets++;
462			percpu_stats->rx_bytes += dpaa2_fd_get_len(fd);
463			return;
464		}
465
466		dma_unmap_page(dev, addr, priv->rx_buf_size,
467			       DMA_BIDIRECTIONAL);
468		skb = dpaa2_eth_build_linear_skb(ch, fd, vaddr);
469	} else if (fd_format == dpaa2_fd_sg) {
470		WARN_ON(priv->xdp_prog);
471
472		dma_unmap_page(dev, addr, priv->rx_buf_size,
473			       DMA_BIDIRECTIONAL);
474		skb = dpaa2_eth_build_frag_skb(priv, ch, buf_data);
475		free_pages((unsigned long)vaddr, 0);
476		percpu_extras->rx_sg_frames++;
477		percpu_extras->rx_sg_bytes += dpaa2_fd_get_len(fd);
478	} else {
479		/* We don't support any other format */
480		goto err_frame_format;
481	}
482
483	if (unlikely(!skb))
484		goto err_build_skb;
485
486	prefetch(skb->data);
487
488	/* Get the timestamp value */
489	if (priv->rx_tstamp) {
490		struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
491		__le64 *ts = dpaa2_get_ts(vaddr, false);
492		u64 ns;
493
494		memset(shhwtstamps, 0, sizeof(*shhwtstamps));
495
496		ns = DPAA2_PTP_CLK_PERIOD_NS * le64_to_cpup(ts);
497		shhwtstamps->hwtstamp = ns_to_ktime(ns);
498	}
499
500	/* Check if we need to validate the L4 csum */
501	if (likely(dpaa2_fd_get_frc(fd) & DPAA2_FD_FRC_FASV)) {
502		status = le32_to_cpu(fas->status);
503		dpaa2_eth_validate_rx_csum(priv, status, skb);
504	}
505
506	skb->protocol = eth_type_trans(skb, priv->net_dev);
507	skb_record_rx_queue(skb, fq->flowid);
508
509	percpu_stats->rx_packets++;
510	percpu_stats->rx_bytes += dpaa2_fd_get_len(fd);
511
512	list_add_tail(&skb->list, ch->rx_list);
513
514	return;
515
516err_build_skb:
517	dpaa2_eth_free_rx_fd(priv, fd, vaddr);
518err_frame_format:
519	percpu_stats->rx_dropped++;
520}
521
522/* Processing of Rx frames received on the error FQ
523 * We check and print the error bits and then free the frame
524 */
525static void dpaa2_eth_rx_err(struct dpaa2_eth_priv *priv,
526			     struct dpaa2_eth_channel *ch,
527			     const struct dpaa2_fd *fd,
528			     struct dpaa2_eth_fq *fq __always_unused)
529{
530	struct device *dev = priv->net_dev->dev.parent;
531	dma_addr_t addr = dpaa2_fd_get_addr(fd);
532	u8 fd_format = dpaa2_fd_get_format(fd);
533	struct rtnl_link_stats64 *percpu_stats;
534	struct dpaa2_eth_trap_item *trap_item;
535	struct dpaa2_fapr *fapr;
536	struct sk_buff *skb;
537	void *buf_data;
538	void *vaddr;
539
540	vaddr = dpaa2_iova_to_virt(priv->iommu_domain, addr);
541	dma_sync_single_for_cpu(dev, addr, priv->rx_buf_size,
542				DMA_BIDIRECTIONAL);
543
544	buf_data = vaddr + dpaa2_fd_get_offset(fd);
545
546	if (fd_format == dpaa2_fd_single) {
547		dma_unmap_page(dev, addr, priv->rx_buf_size,
548			       DMA_BIDIRECTIONAL);
549		skb = dpaa2_eth_build_linear_skb(ch, fd, vaddr);
550	} else if (fd_format == dpaa2_fd_sg) {
551		dma_unmap_page(dev, addr, priv->rx_buf_size,
552			       DMA_BIDIRECTIONAL);
553		skb = dpaa2_eth_build_frag_skb(priv, ch, buf_data);
554		free_pages((unsigned long)vaddr, 0);
555	} else {
556		/* We don't support any other format */
557		dpaa2_eth_free_rx_fd(priv, fd, vaddr);
558		goto err_frame_format;
559	}
560
561	fapr = dpaa2_get_fapr(vaddr, false);
562	trap_item = dpaa2_eth_dl_get_trap(priv, fapr);
563	if (trap_item)
564		devlink_trap_report(priv->devlink, skb, trap_item->trap_ctx,
565				    &priv->devlink_port, NULL);
566	consume_skb(skb);
567
568err_frame_format:
569	percpu_stats = this_cpu_ptr(priv->percpu_stats);
570	percpu_stats->rx_errors++;
571	ch->buf_count--;
572}
573
574/* Consume all frames pull-dequeued into the store. This is the simplest way to
575 * make sure we don't accidentally issue another volatile dequeue which would
576 * overwrite (leak) frames already in the store.
577 *
578 * Observance of NAPI budget is not our concern, leaving that to the caller.
579 */
580static int dpaa2_eth_consume_frames(struct dpaa2_eth_channel *ch,
581				    struct dpaa2_eth_fq **src)
582{
583	struct dpaa2_eth_priv *priv = ch->priv;
584	struct dpaa2_eth_fq *fq = NULL;
585	struct dpaa2_dq *dq;
586	const struct dpaa2_fd *fd;
587	int cleaned = 0, retries = 0;
588	int is_last;
589
590	do {
591		dq = dpaa2_io_store_next(ch->store, &is_last);
592		if (unlikely(!dq)) {
593			/* If we're here, we *must* have placed a
594			 * volatile dequeue comnmand, so keep reading through
595			 * the store until we get some sort of valid response
596			 * token (either a valid frame or an "empty dequeue")
597			 */
598			if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES) {
599				netdev_err_once(priv->net_dev,
600						"Unable to read a valid dequeue response\n");
601				return -ETIMEDOUT;
602			}
603			continue;
604		}
605
606		fd = dpaa2_dq_fd(dq);
607		fq = (struct dpaa2_eth_fq *)(uintptr_t)dpaa2_dq_fqd_ctx(dq);
608
609		fq->consume(priv, ch, fd, fq);
610		cleaned++;
611		retries = 0;
612	} while (!is_last);
613
614	if (!cleaned)
615		return 0;
616
617	fq->stats.frames += cleaned;
618	ch->stats.frames += cleaned;
619
620	/* A dequeue operation only pulls frames from a single queue
621	 * into the store. Return the frame queue as an out param.
622	 */
623	if (src)
624		*src = fq;
625
626	return cleaned;
627}
628
629static int dpaa2_eth_ptp_parse(struct sk_buff *skb,
630			       u8 *msgtype, u8 *twostep, u8 *udp,
631			       u16 *correction_offset,
632			       u16 *origintimestamp_offset)
633{
634	unsigned int ptp_class;
635	struct ptp_header *hdr;
636	unsigned int type;
637	u8 *base;
638
639	ptp_class = ptp_classify_raw(skb);
640	if (ptp_class == PTP_CLASS_NONE)
641		return -EINVAL;
642
643	hdr = ptp_parse_header(skb, ptp_class);
644	if (!hdr)
645		return -EINVAL;
646
647	*msgtype = ptp_get_msgtype(hdr, ptp_class);
648	*twostep = hdr->flag_field[0] & 0x2;
649
650	type = ptp_class & PTP_CLASS_PMASK;
651	if (type == PTP_CLASS_IPV4 ||
652	    type == PTP_CLASS_IPV6)
653		*udp = 1;
654	else
655		*udp = 0;
656
657	base = skb_mac_header(skb);
658	*correction_offset = (u8 *)&hdr->correction - base;
659	*origintimestamp_offset = (u8 *)hdr + sizeof(struct ptp_header) - base;
660
661	return 0;
662}
663
664/* Configure the egress frame annotation for timestamp update */
665static void dpaa2_eth_enable_tx_tstamp(struct dpaa2_eth_priv *priv,
666				       struct dpaa2_fd *fd,
667				       void *buf_start,
668				       struct sk_buff *skb)
669{
670	struct ptp_tstamp origin_timestamp;
671	struct dpni_single_step_cfg cfg;
672	u8 msgtype, twostep, udp;
673	struct dpaa2_faead *faead;
674	struct dpaa2_fas *fas;
675	struct timespec64 ts;
676	u16 offset1, offset2;
677	u32 ctrl, frc;
678	__le64 *ns;
679	u8 *data;
680
681	/* Mark the egress frame annotation area as valid */
682	frc = dpaa2_fd_get_frc(fd);
683	dpaa2_fd_set_frc(fd, frc | DPAA2_FD_FRC_FAEADV);
684
685	/* Set hardware annotation size */
686	ctrl = dpaa2_fd_get_ctrl(fd);
687	dpaa2_fd_set_ctrl(fd, ctrl | DPAA2_FD_CTRL_ASAL);
688
689	/* enable UPD (update prepanded data) bit in FAEAD field of
690	 * hardware frame annotation area
691	 */
692	ctrl = DPAA2_FAEAD_A2V | DPAA2_FAEAD_UPDV | DPAA2_FAEAD_UPD;
693	faead = dpaa2_get_faead(buf_start, true);
694	faead->ctrl = cpu_to_le32(ctrl);
695
696	if (skb->cb[0] == TX_TSTAMP_ONESTEP_SYNC) {
697		if (dpaa2_eth_ptp_parse(skb, &msgtype, &twostep, &udp,
698					&offset1, &offset2) ||
699		    msgtype != 0 || twostep) {
700			WARN_ONCE(1, "Bad packet for one-step timestamping\n");
701			return;
702		}
703
704		/* Mark the frame annotation status as valid */
705		frc = dpaa2_fd_get_frc(fd);
706		dpaa2_fd_set_frc(fd, frc | DPAA2_FD_FRC_FASV);
707
708		/* Mark the PTP flag for one step timestamping */
709		fas = dpaa2_get_fas(buf_start, true);
710		fas->status = cpu_to_le32(DPAA2_FAS_PTP);
711
712		dpaa2_ptp->caps.gettime64(&dpaa2_ptp->caps, &ts);
713		ns = dpaa2_get_ts(buf_start, true);
714		*ns = cpu_to_le64(timespec64_to_ns(&ts) /
715				  DPAA2_PTP_CLK_PERIOD_NS);
716
717		/* Update current time to PTP message originTimestamp field */
718		ns_to_ptp_tstamp(&origin_timestamp, le64_to_cpup(ns));
719		data = skb_mac_header(skb);
720		*(__be16 *)(data + offset2) = htons(origin_timestamp.sec_msb);
721		*(__be32 *)(data + offset2 + 2) =
722			htonl(origin_timestamp.sec_lsb);
723		*(__be32 *)(data + offset2 + 6) = htonl(origin_timestamp.nsec);
724
725		cfg.en = 1;
726		cfg.ch_update = udp;
727		cfg.offset = offset1;
728		cfg.peer_delay = 0;
729
730		if (dpni_set_single_step_cfg(priv->mc_io, 0, priv->mc_token,
731					     &cfg))
732			WARN_ONCE(1, "Failed to set single step register");
733	}
734}
735
736/* Create a frame descriptor based on a fragmented skb */
737static int dpaa2_eth_build_sg_fd(struct dpaa2_eth_priv *priv,
738				 struct sk_buff *skb,
739				 struct dpaa2_fd *fd,
740				 void **swa_addr)
741{
742	struct device *dev = priv->net_dev->dev.parent;
743	void *sgt_buf = NULL;
744	dma_addr_t addr;
745	int nr_frags = skb_shinfo(skb)->nr_frags;
746	struct dpaa2_sg_entry *sgt;
747	int i, err;
748	int sgt_buf_size;
749	struct scatterlist *scl, *crt_scl;
750	int num_sg;
751	int num_dma_bufs;
752	struct dpaa2_eth_swa *swa;
753
754	/* Create and map scatterlist.
755	 * We don't advertise NETIF_F_FRAGLIST, so skb_to_sgvec() will not have
756	 * to go beyond nr_frags+1.
757	 * Note: We don't support chained scatterlists
758	 */
759	if (unlikely(PAGE_SIZE / sizeof(struct scatterlist) < nr_frags + 1))
760		return -EINVAL;
761
762	scl = kmalloc_array(nr_frags + 1, sizeof(struct scatterlist), GFP_ATOMIC);
763	if (unlikely(!scl))
764		return -ENOMEM;
765
766	sg_init_table(scl, nr_frags + 1);
767	num_sg = skb_to_sgvec(skb, scl, 0, skb->len);
768	if (unlikely(num_sg < 0)) {
769		err = -ENOMEM;
770		goto dma_map_sg_failed;
771	}
772	num_dma_bufs = dma_map_sg(dev, scl, num_sg, DMA_BIDIRECTIONAL);
773	if (unlikely(!num_dma_bufs)) {
774		err = -ENOMEM;
775		goto dma_map_sg_failed;
776	}
777
778	/* Prepare the HW SGT structure */
779	sgt_buf_size = priv->tx_data_offset +
780		       sizeof(struct dpaa2_sg_entry) *  num_dma_bufs;
781	sgt_buf = napi_alloc_frag(sgt_buf_size + DPAA2_ETH_TX_BUF_ALIGN);
782	if (unlikely(!sgt_buf)) {
783		err = -ENOMEM;
784		goto sgt_buf_alloc_failed;
785	}
786	sgt_buf = PTR_ALIGN(sgt_buf, DPAA2_ETH_TX_BUF_ALIGN);
787	memset(sgt_buf, 0, sgt_buf_size);
788
789	sgt = (struct dpaa2_sg_entry *)(sgt_buf + priv->tx_data_offset);
790
791	/* Fill in the HW SGT structure.
792	 *
793	 * sgt_buf is zeroed out, so the following fields are implicit
794	 * in all sgt entries:
795	 *   - offset is 0
796	 *   - format is 'dpaa2_sg_single'
797	 */
798	for_each_sg(scl, crt_scl, num_dma_bufs, i) {
799		dpaa2_sg_set_addr(&sgt[i], sg_dma_address(crt_scl));
800		dpaa2_sg_set_len(&sgt[i], sg_dma_len(crt_scl));
801	}
802	dpaa2_sg_set_final(&sgt[i - 1], true);
803
804	/* Store the skb backpointer in the SGT buffer.
805	 * Fit the scatterlist and the number of buffers alongside the
806	 * skb backpointer in the software annotation area. We'll need
807	 * all of them on Tx Conf.
808	 */
809	*swa_addr = (void *)sgt_buf;
810	swa = (struct dpaa2_eth_swa *)sgt_buf;
811	swa->type = DPAA2_ETH_SWA_SG;
812	swa->sg.skb = skb;
813	swa->sg.scl = scl;
814	swa->sg.num_sg = num_sg;
815	swa->sg.sgt_size = sgt_buf_size;
816
817	/* Separately map the SGT buffer */
818	addr = dma_map_single(dev, sgt_buf, sgt_buf_size, DMA_BIDIRECTIONAL);
819	if (unlikely(dma_mapping_error(dev, addr))) {
820		err = -ENOMEM;
821		goto dma_map_single_failed;
822	}
823	dpaa2_fd_set_offset(fd, priv->tx_data_offset);
824	dpaa2_fd_set_format(fd, dpaa2_fd_sg);
825	dpaa2_fd_set_addr(fd, addr);
826	dpaa2_fd_set_len(fd, skb->len);
827	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
828
829	return 0;
830
831dma_map_single_failed:
832	skb_free_frag(sgt_buf);
833sgt_buf_alloc_failed:
834	dma_unmap_sg(dev, scl, num_sg, DMA_BIDIRECTIONAL);
835dma_map_sg_failed:
836	kfree(scl);
837	return err;
838}
839
840/* Create a SG frame descriptor based on a linear skb.
841 *
842 * This function is used on the Tx path when the skb headroom is not large
843 * enough for the HW requirements, thus instead of realloc-ing the skb we
844 * create a SG frame descriptor with only one entry.
845 */
846static int dpaa2_eth_build_sg_fd_single_buf(struct dpaa2_eth_priv *priv,
847					    struct sk_buff *skb,
848					    struct dpaa2_fd *fd,
849					    void **swa_addr)
850{
851	struct device *dev = priv->net_dev->dev.parent;
852	struct dpaa2_eth_sgt_cache *sgt_cache;
853	struct dpaa2_sg_entry *sgt;
854	struct dpaa2_eth_swa *swa;
855	dma_addr_t addr, sgt_addr;
856	void *sgt_buf = NULL;
857	int sgt_buf_size;
858	int err;
859
860	/* Prepare the HW SGT structure */
861	sgt_cache = this_cpu_ptr(priv->sgt_cache);
862	sgt_buf_size = priv->tx_data_offset + sizeof(struct dpaa2_sg_entry);
863
864	if (sgt_cache->count == 0)
865		sgt_buf = kzalloc(sgt_buf_size + DPAA2_ETH_TX_BUF_ALIGN,
866				  GFP_ATOMIC);
867	else
868		sgt_buf = sgt_cache->buf[--sgt_cache->count];
869	if (unlikely(!sgt_buf))
870		return -ENOMEM;
871
872	sgt_buf = PTR_ALIGN(sgt_buf, DPAA2_ETH_TX_BUF_ALIGN);
873	sgt = (struct dpaa2_sg_entry *)(sgt_buf + priv->tx_data_offset);
874
875	addr = dma_map_single(dev, skb->data, skb->len, DMA_BIDIRECTIONAL);
876	if (unlikely(dma_mapping_error(dev, addr))) {
877		err = -ENOMEM;
878		goto data_map_failed;
879	}
880
881	/* Fill in the HW SGT structure */
882	dpaa2_sg_set_addr(sgt, addr);
883	dpaa2_sg_set_len(sgt, skb->len);
884	dpaa2_sg_set_final(sgt, true);
885
886	/* Store the skb backpointer in the SGT buffer */
887	*swa_addr = (void *)sgt_buf;
888	swa = (struct dpaa2_eth_swa *)sgt_buf;
889	swa->type = DPAA2_ETH_SWA_SINGLE;
890	swa->single.skb = skb;
891	swa->single.sgt_size = sgt_buf_size;
892
893	/* Separately map the SGT buffer */
894	sgt_addr = dma_map_single(dev, sgt_buf, sgt_buf_size, DMA_BIDIRECTIONAL);
895	if (unlikely(dma_mapping_error(dev, sgt_addr))) {
896		err = -ENOMEM;
897		goto sgt_map_failed;
898	}
899
900	dpaa2_fd_set_offset(fd, priv->tx_data_offset);
901	dpaa2_fd_set_format(fd, dpaa2_fd_sg);
902	dpaa2_fd_set_addr(fd, sgt_addr);
903	dpaa2_fd_set_len(fd, skb->len);
904	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
905
906	return 0;
907
908sgt_map_failed:
909	dma_unmap_single(dev, addr, skb->len, DMA_BIDIRECTIONAL);
910data_map_failed:
911	if (sgt_cache->count >= DPAA2_ETH_SGT_CACHE_SIZE)
912		kfree(sgt_buf);
913	else
914		sgt_cache->buf[sgt_cache->count++] = sgt_buf;
915
916	return err;
917}
918
919/* Create a frame descriptor based on a linear skb */
920static int dpaa2_eth_build_single_fd(struct dpaa2_eth_priv *priv,
921				     struct sk_buff *skb,
922				     struct dpaa2_fd *fd,
923				     void **swa_addr)
924{
925	struct device *dev = priv->net_dev->dev.parent;
926	u8 *buffer_start, *aligned_start;
927	struct dpaa2_eth_swa *swa;
928	dma_addr_t addr;
929
930	buffer_start = skb->data - dpaa2_eth_needed_headroom(skb);
931	aligned_start = PTR_ALIGN(buffer_start - DPAA2_ETH_TX_BUF_ALIGN,
932				  DPAA2_ETH_TX_BUF_ALIGN);
933	if (aligned_start >= skb->head)
934		buffer_start = aligned_start;
935	else
936		return -ENOMEM;
937
938	/* Store a backpointer to the skb at the beginning of the buffer
939	 * (in the private data area) such that we can release it
940	 * on Tx confirm
941	 */
942	*swa_addr = (void *)buffer_start;
943	swa = (struct dpaa2_eth_swa *)buffer_start;
944	swa->type = DPAA2_ETH_SWA_SINGLE;
945	swa->single.skb = skb;
946
947	addr = dma_map_single(dev, buffer_start,
948			      skb_tail_pointer(skb) - buffer_start,
949			      DMA_BIDIRECTIONAL);
950	if (unlikely(dma_mapping_error(dev, addr)))
951		return -ENOMEM;
952
953	dpaa2_fd_set_addr(fd, addr);
954	dpaa2_fd_set_offset(fd, (u16)(skb->data - buffer_start));
955	dpaa2_fd_set_len(fd, skb->len);
956	dpaa2_fd_set_format(fd, dpaa2_fd_single);
957	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
958
959	return 0;
960}
961
962/* FD freeing routine on the Tx path
963 *
964 * DMA-unmap and free FD and possibly SGT buffer allocated on Tx. The skb
965 * back-pointed to is also freed.
966 * This can be called either from dpaa2_eth_tx_conf() or on the error path of
967 * dpaa2_eth_tx().
968 */
969static void dpaa2_eth_free_tx_fd(struct dpaa2_eth_priv *priv,
970				 struct dpaa2_eth_fq *fq,
971				 const struct dpaa2_fd *fd, bool in_napi)
972{
973	struct device *dev = priv->net_dev->dev.parent;
974	dma_addr_t fd_addr, sg_addr;
975	struct sk_buff *skb = NULL;
976	unsigned char *buffer_start;
977	struct dpaa2_eth_swa *swa;
978	u8 fd_format = dpaa2_fd_get_format(fd);
979	u32 fd_len = dpaa2_fd_get_len(fd);
980
981	struct dpaa2_eth_sgt_cache *sgt_cache;
982	struct dpaa2_sg_entry *sgt;
983
984	fd_addr = dpaa2_fd_get_addr(fd);
985	buffer_start = dpaa2_iova_to_virt(priv->iommu_domain, fd_addr);
986	swa = (struct dpaa2_eth_swa *)buffer_start;
987
988	if (fd_format == dpaa2_fd_single) {
989		if (swa->type == DPAA2_ETH_SWA_SINGLE) {
990			skb = swa->single.skb;
991			/* Accessing the skb buffer is safe before dma unmap,
992			 * because we didn't map the actual skb shell.
993			 */
994			dma_unmap_single(dev, fd_addr,
995					 skb_tail_pointer(skb) - buffer_start,
996					 DMA_BIDIRECTIONAL);
997		} else {
998			WARN_ONCE(swa->type != DPAA2_ETH_SWA_XDP, "Wrong SWA type");
999			dma_unmap_single(dev, fd_addr, swa->xdp.dma_size,
1000					 DMA_BIDIRECTIONAL);
1001		}
1002	} else if (fd_format == dpaa2_fd_sg) {
1003		if (swa->type == DPAA2_ETH_SWA_SG) {
1004			skb = swa->sg.skb;
1005
1006			/* Unmap the scatterlist */
1007			dma_unmap_sg(dev, swa->sg.scl, swa->sg.num_sg,
1008				     DMA_BIDIRECTIONAL);
1009			kfree(swa->sg.scl);
1010
1011			/* Unmap the SGT buffer */
1012			dma_unmap_single(dev, fd_addr, swa->sg.sgt_size,
1013					 DMA_BIDIRECTIONAL);
1014		} else {
1015			skb = swa->single.skb;
1016
1017			/* Unmap the SGT Buffer */
1018			dma_unmap_single(dev, fd_addr, swa->single.sgt_size,
1019					 DMA_BIDIRECTIONAL);
1020
1021			sgt = (struct dpaa2_sg_entry *)(buffer_start +
1022							priv->tx_data_offset);
1023			sg_addr = dpaa2_sg_get_addr(sgt);
1024			dma_unmap_single(dev, sg_addr, skb->len, DMA_BIDIRECTIONAL);
1025		}
1026	} else {
1027		netdev_dbg(priv->net_dev, "Invalid FD format\n");
1028		return;
1029	}
1030
1031	if (swa->type != DPAA2_ETH_SWA_XDP && in_napi) {
1032		fq->dq_frames++;
1033		fq->dq_bytes += fd_len;
1034	}
1035
1036	if (swa->type == DPAA2_ETH_SWA_XDP) {
1037		xdp_return_frame(swa->xdp.xdpf);
1038		return;
1039	}
1040
1041	/* Get the timestamp value */
1042	if (skb->cb[0] == TX_TSTAMP) {
1043		struct skb_shared_hwtstamps shhwtstamps;
1044		__le64 *ts = dpaa2_get_ts(buffer_start, true);
1045		u64 ns;
1046
1047		memset(&shhwtstamps, 0, sizeof(shhwtstamps));
1048
1049		ns = DPAA2_PTP_CLK_PERIOD_NS * le64_to_cpup(ts);
1050		shhwtstamps.hwtstamp = ns_to_ktime(ns);
1051		skb_tstamp_tx(skb, &shhwtstamps);
1052	} else if (skb->cb[0] == TX_TSTAMP_ONESTEP_SYNC) {
1053		mutex_unlock(&priv->onestep_tstamp_lock);
1054	}
1055
1056	/* Free SGT buffer allocated on tx */
1057	if (fd_format != dpaa2_fd_single) {
1058		sgt_cache = this_cpu_ptr(priv->sgt_cache);
1059		if (swa->type == DPAA2_ETH_SWA_SG) {
1060			skb_free_frag(buffer_start);
1061		} else {
1062			if (sgt_cache->count >= DPAA2_ETH_SGT_CACHE_SIZE)
1063				kfree(buffer_start);
1064			else
1065				sgt_cache->buf[sgt_cache->count++] = buffer_start;
1066		}
1067	}
1068
1069	/* Move on with skb release */
1070	napi_consume_skb(skb, in_napi);
1071}
1072
1073static netdev_tx_t __dpaa2_eth_tx(struct sk_buff *skb,
1074				  struct net_device *net_dev)
1075{
1076	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1077	struct dpaa2_fd fd;
1078	struct rtnl_link_stats64 *percpu_stats;
1079	struct dpaa2_eth_drv_stats *percpu_extras;
1080	struct dpaa2_eth_fq *fq;
1081	struct netdev_queue *nq;
1082	u16 queue_mapping;
1083	unsigned int needed_headroom;
1084	u32 fd_len;
1085	u8 prio = 0;
1086	int err, i;
1087	void *swa;
1088
1089	percpu_stats = this_cpu_ptr(priv->percpu_stats);
1090	percpu_extras = this_cpu_ptr(priv->percpu_extras);
1091
1092	needed_headroom = dpaa2_eth_needed_headroom(skb);
1093
1094	/* We'll be holding a back-reference to the skb until Tx Confirmation;
1095	 * we don't want that overwritten by a concurrent Tx with a cloned skb.
1096	 */
1097	skb = skb_unshare(skb, GFP_ATOMIC);
1098	if (unlikely(!skb)) {
1099		/* skb_unshare() has already freed the skb */
1100		percpu_stats->tx_dropped++;
1101		return NETDEV_TX_OK;
1102	}
1103
1104	/* Setup the FD fields */
1105	memset(&fd, 0, sizeof(fd));
1106
1107	if (skb_is_nonlinear(skb)) {
1108		err = dpaa2_eth_build_sg_fd(priv, skb, &fd, &swa);
1109		percpu_extras->tx_sg_frames++;
1110		percpu_extras->tx_sg_bytes += skb->len;
1111	} else if (skb_headroom(skb) < needed_headroom) {
1112		err = dpaa2_eth_build_sg_fd_single_buf(priv, skb, &fd, &swa);
1113		percpu_extras->tx_sg_frames++;
1114		percpu_extras->tx_sg_bytes += skb->len;
1115		percpu_extras->tx_converted_sg_frames++;
1116		percpu_extras->tx_converted_sg_bytes += skb->len;
1117	} else {
1118		err = dpaa2_eth_build_single_fd(priv, skb, &fd, &swa);
1119	}
1120
1121	if (unlikely(err)) {
1122		percpu_stats->tx_dropped++;
1123		goto err_build_fd;
1124	}
1125
1126	if (skb->cb[0])
1127		dpaa2_eth_enable_tx_tstamp(priv, &fd, swa, skb);
1128
1129	/* Tracing point */
1130	trace_dpaa2_tx_fd(net_dev, &fd);
1131
1132	/* TxConf FQ selection relies on queue id from the stack.
1133	 * In case of a forwarded frame from another DPNI interface, we choose
1134	 * a queue affined to the same core that processed the Rx frame
1135	 */
1136	queue_mapping = skb_get_queue_mapping(skb);
1137
1138	if (net_dev->num_tc) {
1139		prio = netdev_txq_to_tc(net_dev, queue_mapping);
1140		/* Hardware interprets priority level 0 as being the highest,
1141		 * so we need to do a reverse mapping to the netdev tc index
1142		 */
1143		prio = net_dev->num_tc - prio - 1;
1144		/* We have only one FQ array entry for all Tx hardware queues
1145		 * with the same flow id (but different priority levels)
1146		 */
1147		queue_mapping %= dpaa2_eth_queue_count(priv);
1148	}
1149	fq = &priv->fq[queue_mapping];
1150
1151	fd_len = dpaa2_fd_get_len(&fd);
1152	nq = netdev_get_tx_queue(net_dev, queue_mapping);
1153	netdev_tx_sent_queue(nq, fd_len);
1154
1155	/* Everything that happens after this enqueues might race with
1156	 * the Tx confirmation callback for this frame
1157	 */
1158	for (i = 0; i < DPAA2_ETH_ENQUEUE_RETRIES; i++) {
1159		err = priv->enqueue(priv, fq, &fd, prio, 1, NULL);
1160		if (err != -EBUSY)
1161			break;
1162	}
1163	percpu_extras->tx_portal_busy += i;
1164	if (unlikely(err < 0)) {
1165		percpu_stats->tx_errors++;
1166		/* Clean up everything, including freeing the skb */
1167		dpaa2_eth_free_tx_fd(priv, fq, &fd, false);
1168		netdev_tx_completed_queue(nq, 1, fd_len);
1169	} else {
1170		percpu_stats->tx_packets++;
1171		percpu_stats->tx_bytes += fd_len;
1172	}
1173
1174	return NETDEV_TX_OK;
1175
1176err_build_fd:
1177	dev_kfree_skb(skb);
1178
1179	return NETDEV_TX_OK;
1180}
1181
1182static void dpaa2_eth_tx_onestep_tstamp(struct work_struct *work)
1183{
1184	struct dpaa2_eth_priv *priv = container_of(work, struct dpaa2_eth_priv,
1185						   tx_onestep_tstamp);
1186	struct sk_buff *skb;
1187
1188	while (true) {
1189		skb = skb_dequeue(&priv->tx_skbs);
1190		if (!skb)
1191			return;
1192
1193		/* Lock just before TX one-step timestamping packet,
1194		 * and release the lock in dpaa2_eth_free_tx_fd when
1195		 * confirm the packet has been sent on hardware, or
1196		 * when clean up during transmit failure.
1197		 */
1198		mutex_lock(&priv->onestep_tstamp_lock);
1199		__dpaa2_eth_tx(skb, priv->net_dev);
1200	}
1201}
1202
1203static netdev_tx_t dpaa2_eth_tx(struct sk_buff *skb, struct net_device *net_dev)
1204{
1205	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1206	u8 msgtype, twostep, udp;
1207	u16 offset1, offset2;
1208
1209	/* Utilize skb->cb[0] for timestamping request per skb */
1210	skb->cb[0] = 0;
1211
1212	if ((skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) && dpaa2_ptp) {
1213		if (priv->tx_tstamp_type == HWTSTAMP_TX_ON)
1214			skb->cb[0] = TX_TSTAMP;
1215		else if (priv->tx_tstamp_type == HWTSTAMP_TX_ONESTEP_SYNC)
1216			skb->cb[0] = TX_TSTAMP_ONESTEP_SYNC;
1217	}
1218
1219	/* TX for one-step timestamping PTP Sync packet */
1220	if (skb->cb[0] == TX_TSTAMP_ONESTEP_SYNC) {
1221		if (!dpaa2_eth_ptp_parse(skb, &msgtype, &twostep, &udp,
1222					 &offset1, &offset2))
1223			if (msgtype == 0 && twostep == 0) {
1224				skb_queue_tail(&priv->tx_skbs, skb);
1225				queue_work(priv->dpaa2_ptp_wq,
1226					   &priv->tx_onestep_tstamp);
1227				return NETDEV_TX_OK;
1228			}
1229		/* Use two-step timestamping if not one-step timestamping
1230		 * PTP Sync packet
1231		 */
1232		skb->cb[0] = TX_TSTAMP;
1233	}
1234
1235	/* TX for other packets */
1236	return __dpaa2_eth_tx(skb, net_dev);
1237}
1238
1239/* Tx confirmation frame processing routine */
1240static void dpaa2_eth_tx_conf(struct dpaa2_eth_priv *priv,
1241			      struct dpaa2_eth_channel *ch __always_unused,
1242			      const struct dpaa2_fd *fd,
1243			      struct dpaa2_eth_fq *fq)
1244{
1245	struct rtnl_link_stats64 *percpu_stats;
1246	struct dpaa2_eth_drv_stats *percpu_extras;
1247	u32 fd_len = dpaa2_fd_get_len(fd);
1248	u32 fd_errors;
1249
1250	/* Tracing point */
1251	trace_dpaa2_tx_conf_fd(priv->net_dev, fd);
1252
1253	percpu_extras = this_cpu_ptr(priv->percpu_extras);
1254	percpu_extras->tx_conf_frames++;
1255	percpu_extras->tx_conf_bytes += fd_len;
1256
1257	/* Check frame errors in the FD field */
1258	fd_errors = dpaa2_fd_get_ctrl(fd) & DPAA2_FD_TX_ERR_MASK;
1259	dpaa2_eth_free_tx_fd(priv, fq, fd, true);
1260
1261	if (likely(!fd_errors))
1262		return;
1263
1264	if (net_ratelimit())
1265		netdev_dbg(priv->net_dev, "TX frame FD error: 0x%08x\n",
1266			   fd_errors);
1267
1268	percpu_stats = this_cpu_ptr(priv->percpu_stats);
1269	/* Tx-conf logically pertains to the egress path. */
1270	percpu_stats->tx_errors++;
1271}
1272
1273static int dpaa2_eth_set_rx_csum(struct dpaa2_eth_priv *priv, bool enable)
1274{
1275	int err;
1276
1277	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1278			       DPNI_OFF_RX_L3_CSUM, enable);
1279	if (err) {
1280		netdev_err(priv->net_dev,
1281			   "dpni_set_offload(RX_L3_CSUM) failed\n");
1282		return err;
1283	}
1284
1285	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1286			       DPNI_OFF_RX_L4_CSUM, enable);
1287	if (err) {
1288		netdev_err(priv->net_dev,
1289			   "dpni_set_offload(RX_L4_CSUM) failed\n");
1290		return err;
1291	}
1292
1293	return 0;
1294}
1295
1296static int dpaa2_eth_set_tx_csum(struct dpaa2_eth_priv *priv, bool enable)
1297{
1298	int err;
1299
1300	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1301			       DPNI_OFF_TX_L3_CSUM, enable);
1302	if (err) {
1303		netdev_err(priv->net_dev, "dpni_set_offload(TX_L3_CSUM) failed\n");
1304		return err;
1305	}
1306
1307	err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
1308			       DPNI_OFF_TX_L4_CSUM, enable);
1309	if (err) {
1310		netdev_err(priv->net_dev, "dpni_set_offload(TX_L4_CSUM) failed\n");
1311		return err;
1312	}
1313
1314	return 0;
1315}
1316
1317/* Perform a single release command to add buffers
1318 * to the specified buffer pool
1319 */
1320static int dpaa2_eth_add_bufs(struct dpaa2_eth_priv *priv,
1321			      struct dpaa2_eth_channel *ch, u16 bpid)
1322{
1323	struct device *dev = priv->net_dev->dev.parent;
1324	u64 buf_array[DPAA2_ETH_BUFS_PER_CMD];
1325	struct page *page;
1326	dma_addr_t addr;
1327	int retries = 0;
1328	int i, err;
1329
1330	for (i = 0; i < DPAA2_ETH_BUFS_PER_CMD; i++) {
1331		/* Allocate buffer visible to WRIOP + skb shared info +
1332		 * alignment padding
1333		 */
1334		/* allocate one page for each Rx buffer. WRIOP sees
1335		 * the entire page except for a tailroom reserved for
1336		 * skb shared info
1337		 */
1338		page = dev_alloc_pages(0);
1339		if (!page)
1340			goto err_alloc;
1341
1342		addr = dma_map_page(dev, page, 0, priv->rx_buf_size,
1343				    DMA_BIDIRECTIONAL);
1344		if (unlikely(dma_mapping_error(dev, addr)))
1345			goto err_map;
1346
1347		buf_array[i] = addr;
1348
1349		/* tracing point */
1350		trace_dpaa2_eth_buf_seed(priv->net_dev, page_address(page),
1351					 DPAA2_ETH_RX_BUF_RAW_SIZE,
1352					 addr, priv->rx_buf_size,
1353					 bpid);
1354	}
1355
1356release_bufs:
1357	/* In case the portal is busy, retry until successful */
1358	while ((err = dpaa2_io_service_release(ch->dpio, bpid,
1359					       buf_array, i)) == -EBUSY) {
1360		if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES)
1361			break;
1362		cpu_relax();
1363	}
1364
1365	/* If release command failed, clean up and bail out;
1366	 * not much else we can do about it
1367	 */
1368	if (err) {
1369		dpaa2_eth_free_bufs(priv, buf_array, i);
1370		return 0;
1371	}
1372
1373	return i;
1374
1375err_map:
1376	__free_pages(page, 0);
1377err_alloc:
1378	/* If we managed to allocate at least some buffers,
1379	 * release them to hardware
1380	 */
1381	if (i)
1382		goto release_bufs;
1383
1384	return 0;
1385}
1386
1387static int dpaa2_eth_seed_pool(struct dpaa2_eth_priv *priv, u16 bpid)
1388{
1389	int i, j;
1390	int new_count;
1391
1392	for (j = 0; j < priv->num_channels; j++) {
1393		for (i = 0; i < DPAA2_ETH_NUM_BUFS;
1394		     i += DPAA2_ETH_BUFS_PER_CMD) {
1395			new_count = dpaa2_eth_add_bufs(priv, priv->channel[j], bpid);
1396			priv->channel[j]->buf_count += new_count;
1397
1398			if (new_count < DPAA2_ETH_BUFS_PER_CMD) {
1399				return -ENOMEM;
1400			}
1401		}
1402	}
1403
1404	return 0;
1405}
1406
1407/*
1408 * Drain the specified number of buffers from the DPNI's private buffer pool.
1409 * @count must not exceeed DPAA2_ETH_BUFS_PER_CMD
1410 */
1411static void dpaa2_eth_drain_bufs(struct dpaa2_eth_priv *priv, int count)
1412{
1413	u64 buf_array[DPAA2_ETH_BUFS_PER_CMD];
1414	int retries = 0;
1415	int ret;
1416
1417	do {
1418		ret = dpaa2_io_service_acquire(NULL, priv->bpid,
1419					       buf_array, count);
1420		if (ret < 0) {
1421			if (ret == -EBUSY &&
1422			    retries++ < DPAA2_ETH_SWP_BUSY_RETRIES)
1423				continue;
1424			netdev_err(priv->net_dev, "dpaa2_io_service_acquire() failed\n");
1425			return;
1426		}
1427		dpaa2_eth_free_bufs(priv, buf_array, ret);
1428		retries = 0;
1429	} while (ret);
1430}
1431
1432static void dpaa2_eth_drain_pool(struct dpaa2_eth_priv *priv)
1433{
1434	int i;
1435
1436	dpaa2_eth_drain_bufs(priv, DPAA2_ETH_BUFS_PER_CMD);
1437	dpaa2_eth_drain_bufs(priv, 1);
1438
1439	for (i = 0; i < priv->num_channels; i++)
1440		priv->channel[i]->buf_count = 0;
1441}
1442
1443/* Function is called from softirq context only, so we don't need to guard
1444 * the access to percpu count
1445 */
1446static int dpaa2_eth_refill_pool(struct dpaa2_eth_priv *priv,
1447				 struct dpaa2_eth_channel *ch,
1448				 u16 bpid)
1449{
1450	int new_count;
1451
1452	if (likely(ch->buf_count >= DPAA2_ETH_REFILL_THRESH))
1453		return 0;
1454
1455	do {
1456		new_count = dpaa2_eth_add_bufs(priv, ch, bpid);
1457		if (unlikely(!new_count)) {
1458			/* Out of memory; abort for now, we'll try later on */
1459			break;
1460		}
1461		ch->buf_count += new_count;
1462	} while (ch->buf_count < DPAA2_ETH_NUM_BUFS);
1463
1464	if (unlikely(ch->buf_count < DPAA2_ETH_NUM_BUFS))
1465		return -ENOMEM;
1466
1467	return 0;
1468}
1469
1470static void dpaa2_eth_sgt_cache_drain(struct dpaa2_eth_priv *priv)
1471{
1472	struct dpaa2_eth_sgt_cache *sgt_cache;
1473	u16 count;
1474	int k, i;
1475
1476	for_each_possible_cpu(k) {
1477		sgt_cache = per_cpu_ptr(priv->sgt_cache, k);
1478		count = sgt_cache->count;
1479
1480		for (i = 0; i < count; i++)
1481			kfree(sgt_cache->buf[i]);
1482		sgt_cache->count = 0;
1483	}
1484}
1485
1486static int dpaa2_eth_pull_channel(struct dpaa2_eth_channel *ch)
1487{
1488	int err;
1489	int dequeues = -1;
1490
1491	/* Retry while portal is busy */
1492	do {
1493		err = dpaa2_io_service_pull_channel(ch->dpio, ch->ch_id,
1494						    ch->store);
1495		dequeues++;
1496		cpu_relax();
1497	} while (err == -EBUSY && dequeues < DPAA2_ETH_SWP_BUSY_RETRIES);
1498
1499	ch->stats.dequeue_portal_busy += dequeues;
1500	if (unlikely(err))
1501		ch->stats.pull_err++;
1502
1503	return err;
1504}
1505
1506/* NAPI poll routine
1507 *
1508 * Frames are dequeued from the QMan channel associated with this NAPI context.
1509 * Rx, Tx confirmation and (if configured) Rx error frames all count
1510 * towards the NAPI budget.
1511 */
1512static int dpaa2_eth_poll(struct napi_struct *napi, int budget)
1513{
1514	struct dpaa2_eth_channel *ch;
1515	struct dpaa2_eth_priv *priv;
1516	int rx_cleaned = 0, txconf_cleaned = 0;
1517	struct dpaa2_eth_fq *fq, *txc_fq = NULL;
1518	struct netdev_queue *nq;
1519	int store_cleaned, work_done;
1520	struct list_head rx_list;
1521	int retries = 0;
1522	u16 flowid;
1523	int err;
1524
1525	ch = container_of(napi, struct dpaa2_eth_channel, napi);
1526	ch->xdp.res = 0;
1527	priv = ch->priv;
1528
1529	INIT_LIST_HEAD(&rx_list);
1530	ch->rx_list = &rx_list;
1531
1532	do {
1533		err = dpaa2_eth_pull_channel(ch);
1534		if (unlikely(err))
1535			break;
1536
1537		/* Refill pool if appropriate */
1538		dpaa2_eth_refill_pool(priv, ch, priv->bpid);
1539
1540		store_cleaned = dpaa2_eth_consume_frames(ch, &fq);
1541		if (store_cleaned <= 0)
1542			break;
1543		if (fq->type == DPAA2_RX_FQ) {
1544			rx_cleaned += store_cleaned;
1545			flowid = fq->flowid;
1546		} else {
1547			txconf_cleaned += store_cleaned;
1548			/* We have a single Tx conf FQ on this channel */
1549			txc_fq = fq;
1550		}
1551
1552		/* If we either consumed the whole NAPI budget with Rx frames
1553		 * or we reached the Tx confirmations threshold, we're done.
1554		 */
1555		if (rx_cleaned >= budget ||
1556		    txconf_cleaned >= DPAA2_ETH_TXCONF_PER_NAPI) {
1557			work_done = budget;
1558			goto out;
1559		}
1560	} while (store_cleaned);
1561
1562	/* We didn't consume the entire budget, so finish napi and
1563	 * re-enable data availability notifications
1564	 */
1565	napi_complete_done(napi, rx_cleaned);
1566	do {
1567		err = dpaa2_io_service_rearm(ch->dpio, &ch->nctx);
1568		cpu_relax();
1569	} while (err == -EBUSY && retries++ < DPAA2_ETH_SWP_BUSY_RETRIES);
1570	WARN_ONCE(err, "CDAN notifications rearm failed on core %d",
1571		  ch->nctx.desired_cpu);
1572
1573	work_done = max(rx_cleaned, 1);
1574
1575out:
1576	netif_receive_skb_list(ch->rx_list);
1577
1578	if (txc_fq && txc_fq->dq_frames) {
1579		nq = netdev_get_tx_queue(priv->net_dev, txc_fq->flowid);
1580		netdev_tx_completed_queue(nq, txc_fq->dq_frames,
1581					  txc_fq->dq_bytes);
1582		txc_fq->dq_frames = 0;
1583		txc_fq->dq_bytes = 0;
1584	}
1585
1586	if (ch->xdp.res & XDP_REDIRECT)
1587		xdp_do_flush_map();
1588	else if (rx_cleaned && ch->xdp.res & XDP_TX)
1589		dpaa2_eth_xdp_tx_flush(priv, ch, &priv->fq[flowid]);
1590
1591	return work_done;
1592}
1593
1594static void dpaa2_eth_enable_ch_napi(struct dpaa2_eth_priv *priv)
1595{
1596	struct dpaa2_eth_channel *ch;
1597	int i;
1598
1599	for (i = 0; i < priv->num_channels; i++) {
1600		ch = priv->channel[i];
1601		napi_enable(&ch->napi);
1602	}
1603}
1604
1605static void dpaa2_eth_disable_ch_napi(struct dpaa2_eth_priv *priv)
1606{
1607	struct dpaa2_eth_channel *ch;
1608	int i;
1609
1610	for (i = 0; i < priv->num_channels; i++) {
1611		ch = priv->channel[i];
1612		napi_disable(&ch->napi);
1613	}
1614}
1615
1616void dpaa2_eth_set_rx_taildrop(struct dpaa2_eth_priv *priv,
1617			       bool tx_pause, bool pfc)
1618{
1619	struct dpni_taildrop td = {0};
1620	struct dpaa2_eth_fq *fq;
1621	int i, err;
1622
1623	/* FQ taildrop: threshold is in bytes, per frame queue. Enabled if
1624	 * flow control is disabled (as it might interfere with either the
1625	 * buffer pool depletion trigger for pause frames or with the group
1626	 * congestion trigger for PFC frames)
1627	 */
1628	td.enable = !tx_pause;
1629	if (priv->rx_fqtd_enabled == td.enable)
1630		goto set_cgtd;
1631
1632	td.threshold = DPAA2_ETH_FQ_TAILDROP_THRESH;
1633	td.units = DPNI_CONGESTION_UNIT_BYTES;
1634
1635	for (i = 0; i < priv->num_fqs; i++) {
1636		fq = &priv->fq[i];
1637		if (fq->type != DPAA2_RX_FQ)
1638			continue;
1639		err = dpni_set_taildrop(priv->mc_io, 0, priv->mc_token,
1640					DPNI_CP_QUEUE, DPNI_QUEUE_RX,
1641					fq->tc, fq->flowid, &td);
1642		if (err) {
1643			netdev_err(priv->net_dev,
1644				   "dpni_set_taildrop(FQ) failed\n");
1645			return;
1646		}
1647	}
1648
1649	priv->rx_fqtd_enabled = td.enable;
1650
1651set_cgtd:
1652	/* Congestion group taildrop: threshold is in frames, per group
1653	 * of FQs belonging to the same traffic class
1654	 * Enabled if general Tx pause disabled or if PFCs are enabled
1655	 * (congestion group threhsold for PFC generation is lower than the
1656	 * CG taildrop threshold, so it won't interfere with it; we also
1657	 * want frames in non-PFC enabled traffic classes to be kept in check)
1658	 */
1659	td.enable = !tx_pause || (tx_pause && pfc);
1660	if (priv->rx_cgtd_enabled == td.enable)
1661		return;
1662
1663	td.threshold = DPAA2_ETH_CG_TAILDROP_THRESH(priv);
1664	td.units = DPNI_CONGESTION_UNIT_FRAMES;
1665	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
1666		err = dpni_set_taildrop(priv->mc_io, 0, priv->mc_token,
1667					DPNI_CP_GROUP, DPNI_QUEUE_RX,
1668					i, 0, &td);
1669		if (err) {
1670			netdev_err(priv->net_dev,
1671				   "dpni_set_taildrop(CG) failed\n");
1672			return;
1673		}
1674	}
1675
1676	priv->rx_cgtd_enabled = td.enable;
1677}
1678
1679static int dpaa2_eth_link_state_update(struct dpaa2_eth_priv *priv)
1680{
1681	struct dpni_link_state state = {0};
1682	bool tx_pause;
1683	int err;
1684
1685	err = dpni_get_link_state(priv->mc_io, 0, priv->mc_token, &state);
1686	if (unlikely(err)) {
1687		netdev_err(priv->net_dev,
1688			   "dpni_get_link_state() failed\n");
1689		return err;
1690	}
1691
1692	/* If Tx pause frame settings have changed, we need to update
1693	 * Rx FQ taildrop configuration as well. We configure taildrop
1694	 * only when pause frame generation is disabled.
1695	 */
1696	tx_pause = dpaa2_eth_tx_pause_enabled(state.options);
1697	dpaa2_eth_set_rx_taildrop(priv, tx_pause, priv->pfc_enabled);
1698
1699	/* When we manage the MAC/PHY using phylink there is no need
1700	 * to manually update the netif_carrier.
1701	 */
1702	if (priv->mac)
1703		goto out;
1704
1705	/* Chech link state; speed / duplex changes are not treated yet */
1706	if (priv->link_state.up == state.up)
1707		goto out;
1708
1709	if (state.up) {
1710		netif_carrier_on(priv->net_dev);
1711		netif_tx_start_all_queues(priv->net_dev);
1712	} else {
1713		netif_tx_stop_all_queues(priv->net_dev);
1714		netif_carrier_off(priv->net_dev);
1715	}
1716
1717	netdev_info(priv->net_dev, "Link Event: state %s\n",
1718		    state.up ? "up" : "down");
1719
1720out:
1721	priv->link_state = state;
1722
1723	return 0;
1724}
1725
1726static int dpaa2_eth_open(struct net_device *net_dev)
1727{
1728	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1729	int err;
1730
1731	err = dpaa2_eth_seed_pool(priv, priv->bpid);
1732	if (err) {
1733		/* Not much to do; the buffer pool, though not filled up,
1734		 * may still contain some buffers which would enable us
1735		 * to limp on.
1736		 */
1737		netdev_err(net_dev, "Buffer seeding failed for DPBP %d (bpid=%d)\n",
1738			   priv->dpbp_dev->obj_desc.id, priv->bpid);
1739	}
1740
1741	if (!priv->mac) {
1742		/* We'll only start the txqs when the link is actually ready;
1743		 * make sure we don't race against the link up notification,
1744		 * which may come immediately after dpni_enable();
1745		 */
1746		netif_tx_stop_all_queues(net_dev);
1747
1748		/* Also, explicitly set carrier off, otherwise
1749		 * netif_carrier_ok() will return true and cause 'ip link show'
1750		 * to report the LOWER_UP flag, even though the link
1751		 * notification wasn't even received.
1752		 */
1753		netif_carrier_off(net_dev);
1754	}
1755	dpaa2_eth_enable_ch_napi(priv);
1756
1757	err = dpni_enable(priv->mc_io, 0, priv->mc_token);
1758	if (err < 0) {
1759		netdev_err(net_dev, "dpni_enable() failed\n");
1760		goto enable_err;
1761	}
1762
1763	if (priv->mac)
1764		phylink_start(priv->mac->phylink);
1765
1766	return 0;
1767
1768enable_err:
1769	dpaa2_eth_disable_ch_napi(priv);
1770	dpaa2_eth_drain_pool(priv);
1771	return err;
1772}
1773
1774/* Total number of in-flight frames on ingress queues */
1775static u32 dpaa2_eth_ingress_fq_count(struct dpaa2_eth_priv *priv)
1776{
1777	struct dpaa2_eth_fq *fq;
1778	u32 fcnt = 0, bcnt = 0, total = 0;
1779	int i, err;
1780
1781	for (i = 0; i < priv->num_fqs; i++) {
1782		fq = &priv->fq[i];
1783		err = dpaa2_io_query_fq_count(NULL, fq->fqid, &fcnt, &bcnt);
1784		if (err) {
1785			netdev_warn(priv->net_dev, "query_fq_count failed");
1786			break;
1787		}
1788		total += fcnt;
1789	}
1790
1791	return total;
1792}
1793
1794static void dpaa2_eth_wait_for_ingress_fq_empty(struct dpaa2_eth_priv *priv)
1795{
1796	int retries = 10;
1797	u32 pending;
1798
1799	do {
1800		pending = dpaa2_eth_ingress_fq_count(priv);
1801		if (pending)
1802			msleep(100);
1803	} while (pending && --retries);
1804}
1805
1806#define DPNI_TX_PENDING_VER_MAJOR	7
1807#define DPNI_TX_PENDING_VER_MINOR	13
1808static void dpaa2_eth_wait_for_egress_fq_empty(struct dpaa2_eth_priv *priv)
1809{
1810	union dpni_statistics stats;
1811	int retries = 10;
1812	int err;
1813
1814	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_TX_PENDING_VER_MAJOR,
1815				   DPNI_TX_PENDING_VER_MINOR) < 0)
1816		goto out;
1817
1818	do {
1819		err = dpni_get_statistics(priv->mc_io, 0, priv->mc_token, 6,
1820					  &stats);
1821		if (err)
1822			goto out;
1823		if (stats.page_6.tx_pending_frames == 0)
1824			return;
1825	} while (--retries);
1826
1827out:
1828	msleep(500);
1829}
1830
1831static int dpaa2_eth_stop(struct net_device *net_dev)
1832{
1833	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1834	int dpni_enabled = 0;
1835	int retries = 10;
1836
1837	if (!priv->mac) {
1838		netif_tx_stop_all_queues(net_dev);
1839		netif_carrier_off(net_dev);
1840	} else {
1841		phylink_stop(priv->mac->phylink);
1842	}
1843
1844	/* On dpni_disable(), the MC firmware will:
1845	 * - stop MAC Rx and wait for all Rx frames to be enqueued to software
1846	 * - cut off WRIOP dequeues from egress FQs and wait until transmission
1847	 * of all in flight Tx frames is finished (and corresponding Tx conf
1848	 * frames are enqueued back to software)
1849	 *
1850	 * Before calling dpni_disable(), we wait for all Tx frames to arrive
1851	 * on WRIOP. After it finishes, wait until all remaining frames on Rx
1852	 * and Tx conf queues are consumed on NAPI poll.
1853	 */
1854	dpaa2_eth_wait_for_egress_fq_empty(priv);
1855
1856	do {
1857		dpni_disable(priv->mc_io, 0, priv->mc_token);
1858		dpni_is_enabled(priv->mc_io, 0, priv->mc_token, &dpni_enabled);
1859		if (dpni_enabled)
1860			/* Allow the hardware some slack */
1861			msleep(100);
1862	} while (dpni_enabled && --retries);
1863	if (!retries) {
1864		netdev_warn(net_dev, "Retry count exceeded disabling DPNI\n");
1865		/* Must go on and disable NAPI nonetheless, so we don't crash at
1866		 * the next "ifconfig up"
1867		 */
1868	}
1869
1870	dpaa2_eth_wait_for_ingress_fq_empty(priv);
1871	dpaa2_eth_disable_ch_napi(priv);
1872
1873	/* Empty the buffer pool */
1874	dpaa2_eth_drain_pool(priv);
1875
1876	/* Empty the Scatter-Gather Buffer cache */
1877	dpaa2_eth_sgt_cache_drain(priv);
1878
1879	return 0;
1880}
1881
1882static int dpaa2_eth_set_addr(struct net_device *net_dev, void *addr)
1883{
1884	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1885	struct device *dev = net_dev->dev.parent;
1886	int err;
1887
1888	err = eth_mac_addr(net_dev, addr);
1889	if (err < 0) {
1890		dev_err(dev, "eth_mac_addr() failed (%d)\n", err);
1891		return err;
1892	}
1893
1894	err = dpni_set_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
1895					net_dev->dev_addr);
1896	if (err) {
1897		dev_err(dev, "dpni_set_primary_mac_addr() failed (%d)\n", err);
1898		return err;
1899	}
1900
1901	return 0;
1902}
1903
1904/** Fill in counters maintained by the GPP driver. These may be different from
1905 * the hardware counters obtained by ethtool.
1906 */
1907static void dpaa2_eth_get_stats(struct net_device *net_dev,
1908				struct rtnl_link_stats64 *stats)
1909{
1910	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1911	struct rtnl_link_stats64 *percpu_stats;
1912	u64 *cpustats;
1913	u64 *netstats = (u64 *)stats;
1914	int i, j;
1915	int num = sizeof(struct rtnl_link_stats64) / sizeof(u64);
1916
1917	for_each_possible_cpu(i) {
1918		percpu_stats = per_cpu_ptr(priv->percpu_stats, i);
1919		cpustats = (u64 *)percpu_stats;
1920		for (j = 0; j < num; j++)
1921			netstats[j] += cpustats[j];
1922	}
1923}
1924
1925/* Copy mac unicast addresses from @net_dev to @priv.
1926 * Its sole purpose is to make dpaa2_eth_set_rx_mode() more readable.
1927 */
1928static void dpaa2_eth_add_uc_hw_addr(const struct net_device *net_dev,
1929				     struct dpaa2_eth_priv *priv)
1930{
1931	struct netdev_hw_addr *ha;
1932	int err;
1933
1934	netdev_for_each_uc_addr(ha, net_dev) {
1935		err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token,
1936					ha->addr);
1937		if (err)
1938			netdev_warn(priv->net_dev,
1939				    "Could not add ucast MAC %pM to the filtering table (err %d)\n",
1940				    ha->addr, err);
1941	}
1942}
1943
1944/* Copy mac multicast addresses from @net_dev to @priv
1945 * Its sole purpose is to make dpaa2_eth_set_rx_mode() more readable.
1946 */
1947static void dpaa2_eth_add_mc_hw_addr(const struct net_device *net_dev,
1948				     struct dpaa2_eth_priv *priv)
1949{
1950	struct netdev_hw_addr *ha;
1951	int err;
1952
1953	netdev_for_each_mc_addr(ha, net_dev) {
1954		err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token,
1955					ha->addr);
1956		if (err)
1957			netdev_warn(priv->net_dev,
1958				    "Could not add mcast MAC %pM to the filtering table (err %d)\n",
1959				    ha->addr, err);
1960	}
1961}
1962
1963static void dpaa2_eth_set_rx_mode(struct net_device *net_dev)
1964{
1965	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1966	int uc_count = netdev_uc_count(net_dev);
1967	int mc_count = netdev_mc_count(net_dev);
1968	u8 max_mac = priv->dpni_attrs.mac_filter_entries;
1969	u32 options = priv->dpni_attrs.options;
1970	u16 mc_token = priv->mc_token;
1971	struct fsl_mc_io *mc_io = priv->mc_io;
1972	int err;
1973
1974	/* Basic sanity checks; these probably indicate a misconfiguration */
1975	if (options & DPNI_OPT_NO_MAC_FILTER && max_mac != 0)
1976		netdev_info(net_dev,
1977			    "mac_filter_entries=%d, DPNI_OPT_NO_MAC_FILTER option must be disabled\n",
1978			    max_mac);
1979
1980	/* Force promiscuous if the uc or mc counts exceed our capabilities. */
1981	if (uc_count > max_mac) {
1982		netdev_info(net_dev,
1983			    "Unicast addr count reached %d, max allowed is %d; forcing promisc\n",
1984			    uc_count, max_mac);
1985		goto force_promisc;
1986	}
1987	if (mc_count + uc_count > max_mac) {
1988		netdev_info(net_dev,
1989			    "Unicast + multicast addr count reached %d, max allowed is %d; forcing promisc\n",
1990			    uc_count + mc_count, max_mac);
1991		goto force_mc_promisc;
1992	}
1993
1994	/* Adjust promisc settings due to flag combinations */
1995	if (net_dev->flags & IFF_PROMISC)
1996		goto force_promisc;
1997	if (net_dev->flags & IFF_ALLMULTI) {
1998		/* First, rebuild unicast filtering table. This should be done
1999		 * in promisc mode, in order to avoid frame loss while we
2000		 * progressively add entries to the table.
2001		 * We don't know whether we had been in promisc already, and
2002		 * making an MC call to find out is expensive; so set uc promisc
2003		 * nonetheless.
2004		 */
2005		err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 1);
2006		if (err)
2007			netdev_warn(net_dev, "Can't set uc promisc\n");
2008
2009		/* Actual uc table reconstruction. */
2010		err = dpni_clear_mac_filters(mc_io, 0, mc_token, 1, 0);
2011		if (err)
2012			netdev_warn(net_dev, "Can't clear uc filters\n");
2013		dpaa2_eth_add_uc_hw_addr(net_dev, priv);
2014
2015		/* Finally, clear uc promisc and set mc promisc as requested. */
2016		err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 0);
2017		if (err)
2018			netdev_warn(net_dev, "Can't clear uc promisc\n");
2019		goto force_mc_promisc;
2020	}
2021
2022	/* Neither unicast, nor multicast promisc will be on... eventually.
2023	 * For now, rebuild mac filtering tables while forcing both of them on.
2024	 */
2025	err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 1);
2026	if (err)
2027		netdev_warn(net_dev, "Can't set uc promisc (%d)\n", err);
2028	err = dpni_set_multicast_promisc(mc_io, 0, mc_token, 1);
2029	if (err)
2030		netdev_warn(net_dev, "Can't set mc promisc (%d)\n", err);
2031
2032	/* Actual mac filtering tables reconstruction */
2033	err = dpni_clear_mac_filters(mc_io, 0, mc_token, 1, 1);
2034	if (err)
2035		netdev_warn(net_dev, "Can't clear mac filters\n");
2036	dpaa2_eth_add_mc_hw_addr(net_dev, priv);
2037	dpaa2_eth_add_uc_hw_addr(net_dev, priv);
2038
2039	/* Now we can clear both ucast and mcast promisc, without risking
2040	 * to drop legitimate frames anymore.
2041	 */
2042	err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 0);
2043	if (err)
2044		netdev_warn(net_dev, "Can't clear ucast promisc\n");
2045	err = dpni_set_multicast_promisc(mc_io, 0, mc_token, 0);
2046	if (err)
2047		netdev_warn(net_dev, "Can't clear mcast promisc\n");
2048
2049	return;
2050
2051force_promisc:
2052	err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 1);
2053	if (err)
2054		netdev_warn(net_dev, "Can't set ucast promisc\n");
2055force_mc_promisc:
2056	err = dpni_set_multicast_promisc(mc_io, 0, mc_token, 1);
2057	if (err)
2058		netdev_warn(net_dev, "Can't set mcast promisc\n");
2059}
2060
2061static int dpaa2_eth_set_features(struct net_device *net_dev,
2062				  netdev_features_t features)
2063{
2064	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2065	netdev_features_t changed = features ^ net_dev->features;
2066	bool enable;
2067	int err;
2068
2069	if (changed & NETIF_F_RXCSUM) {
2070		enable = !!(features & NETIF_F_RXCSUM);
2071		err = dpaa2_eth_set_rx_csum(priv, enable);
2072		if (err)
2073			return err;
2074	}
2075
2076	if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
2077		enable = !!(features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM));
2078		err = dpaa2_eth_set_tx_csum(priv, enable);
2079		if (err)
2080			return err;
2081	}
2082
2083	return 0;
2084}
2085
2086static int dpaa2_eth_ts_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
2087{
2088	struct dpaa2_eth_priv *priv = netdev_priv(dev);
2089	struct hwtstamp_config config;
2090
2091	if (!dpaa2_ptp)
2092		return -EINVAL;
2093
2094	if (copy_from_user(&config, rq->ifr_data, sizeof(config)))
2095		return -EFAULT;
2096
2097	switch (config.tx_type) {
2098	case HWTSTAMP_TX_OFF:
2099	case HWTSTAMP_TX_ON:
2100	case HWTSTAMP_TX_ONESTEP_SYNC:
2101		priv->tx_tstamp_type = config.tx_type;
2102		break;
2103	default:
2104		return -ERANGE;
2105	}
2106
2107	if (config.rx_filter == HWTSTAMP_FILTER_NONE) {
2108		priv->rx_tstamp = false;
2109	} else {
2110		priv->rx_tstamp = true;
2111		/* TS is set for all frame types, not only those requested */
2112		config.rx_filter = HWTSTAMP_FILTER_ALL;
2113	}
2114
2115	return copy_to_user(rq->ifr_data, &config, sizeof(config)) ?
2116			-EFAULT : 0;
2117}
2118
2119static int dpaa2_eth_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
2120{
2121	struct dpaa2_eth_priv *priv = netdev_priv(dev);
2122
2123	if (cmd == SIOCSHWTSTAMP)
2124		return dpaa2_eth_ts_ioctl(dev, rq, cmd);
2125
2126	if (priv->mac)
2127		return phylink_mii_ioctl(priv->mac->phylink, rq, cmd);
2128
2129	return -EOPNOTSUPP;
2130}
2131
2132static bool xdp_mtu_valid(struct dpaa2_eth_priv *priv, int mtu)
2133{
2134	int mfl, linear_mfl;
2135
2136	mfl = DPAA2_ETH_L2_MAX_FRM(mtu);
2137	linear_mfl = priv->rx_buf_size - DPAA2_ETH_RX_HWA_SIZE -
2138		     dpaa2_eth_rx_head_room(priv) - XDP_PACKET_HEADROOM;
2139
2140	if (mfl > linear_mfl) {
2141		netdev_warn(priv->net_dev, "Maximum MTU for XDP is %d\n",
2142			    linear_mfl - VLAN_ETH_HLEN);
2143		return false;
2144	}
2145
2146	return true;
2147}
2148
2149static int dpaa2_eth_set_rx_mfl(struct dpaa2_eth_priv *priv, int mtu, bool has_xdp)
2150{
2151	int mfl, err;
2152
2153	/* We enforce a maximum Rx frame length based on MTU only if we have
2154	 * an XDP program attached (in order to avoid Rx S/G frames).
2155	 * Otherwise, we accept all incoming frames as long as they are not
2156	 * larger than maximum size supported in hardware
2157	 */
2158	if (has_xdp)
2159		mfl = DPAA2_ETH_L2_MAX_FRM(mtu);
2160	else
2161		mfl = DPAA2_ETH_MFL;
2162
2163	err = dpni_set_max_frame_length(priv->mc_io, 0, priv->mc_token, mfl);
2164	if (err) {
2165		netdev_err(priv->net_dev, "dpni_set_max_frame_length failed\n");
2166		return err;
2167	}
2168
2169	return 0;
2170}
2171
2172static int dpaa2_eth_change_mtu(struct net_device *dev, int new_mtu)
2173{
2174	struct dpaa2_eth_priv *priv = netdev_priv(dev);
2175	int err;
2176
2177	if (!priv->xdp_prog)
2178		goto out;
2179
2180	if (!xdp_mtu_valid(priv, new_mtu))
2181		return -EINVAL;
2182
2183	err = dpaa2_eth_set_rx_mfl(priv, new_mtu, true);
2184	if (err)
2185		return err;
2186
2187out:
2188	dev->mtu = new_mtu;
2189	return 0;
2190}
2191
2192static int dpaa2_eth_update_rx_buffer_headroom(struct dpaa2_eth_priv *priv, bool has_xdp)
2193{
2194	struct dpni_buffer_layout buf_layout = {0};
2195	int err;
2196
2197	err = dpni_get_buffer_layout(priv->mc_io, 0, priv->mc_token,
2198				     DPNI_QUEUE_RX, &buf_layout);
2199	if (err) {
2200		netdev_err(priv->net_dev, "dpni_get_buffer_layout failed\n");
2201		return err;
2202	}
2203
2204	/* Reserve extra headroom for XDP header size changes */
2205	buf_layout.data_head_room = dpaa2_eth_rx_head_room(priv) +
2206				    (has_xdp ? XDP_PACKET_HEADROOM : 0);
2207	buf_layout.options = DPNI_BUF_LAYOUT_OPT_DATA_HEAD_ROOM;
2208	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
2209				     DPNI_QUEUE_RX, &buf_layout);
2210	if (err) {
2211		netdev_err(priv->net_dev, "dpni_set_buffer_layout failed\n");
2212		return err;
2213	}
2214
2215	return 0;
2216}
2217
2218static int dpaa2_eth_setup_xdp(struct net_device *dev, struct bpf_prog *prog)
2219{
2220	struct dpaa2_eth_priv *priv = netdev_priv(dev);
2221	struct dpaa2_eth_channel *ch;
2222	struct bpf_prog *old;
2223	bool up, need_update;
2224	int i, err;
2225
2226	if (prog && !xdp_mtu_valid(priv, dev->mtu))
2227		return -EINVAL;
2228
2229	if (prog)
2230		bpf_prog_add(prog, priv->num_channels);
2231
2232	up = netif_running(dev);
2233	need_update = (!!priv->xdp_prog != !!prog);
2234
2235	if (up)
2236		dpaa2_eth_stop(dev);
2237
2238	/* While in xdp mode, enforce a maximum Rx frame size based on MTU.
2239	 * Also, when switching between xdp/non-xdp modes we need to reconfigure
2240	 * our Rx buffer layout. Buffer pool was drained on dpaa2_eth_stop,
2241	 * so we are sure no old format buffers will be used from now on.
2242	 */
2243	if (need_update) {
2244		err = dpaa2_eth_set_rx_mfl(priv, dev->mtu, !!prog);
2245		if (err)
2246			goto out_err;
2247		err = dpaa2_eth_update_rx_buffer_headroom(priv, !!prog);
2248		if (err)
2249			goto out_err;
2250	}
2251
2252	old = xchg(&priv->xdp_prog, prog);
2253	if (old)
2254		bpf_prog_put(old);
2255
2256	for (i = 0; i < priv->num_channels; i++) {
2257		ch = priv->channel[i];
2258		old = xchg(&ch->xdp.prog, prog);
2259		if (old)
2260			bpf_prog_put(old);
2261	}
2262
2263	if (up) {
2264		err = dpaa2_eth_open(dev);
2265		if (err)
2266			return err;
2267	}
2268
2269	return 0;
2270
2271out_err:
2272	if (prog)
2273		bpf_prog_sub(prog, priv->num_channels);
2274	if (up)
2275		dpaa2_eth_open(dev);
2276
2277	return err;
2278}
2279
2280static int dpaa2_eth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2281{
2282	switch (xdp->command) {
2283	case XDP_SETUP_PROG:
2284		return dpaa2_eth_setup_xdp(dev, xdp->prog);
2285	default:
2286		return -EINVAL;
2287	}
2288
2289	return 0;
2290}
2291
2292static int dpaa2_eth_xdp_create_fd(struct net_device *net_dev,
2293				   struct xdp_frame *xdpf,
2294				   struct dpaa2_fd *fd)
2295{
2296	struct device *dev = net_dev->dev.parent;
2297	unsigned int needed_headroom;
2298	struct dpaa2_eth_swa *swa;
2299	void *buffer_start, *aligned_start;
2300	dma_addr_t addr;
2301
2302	/* We require a minimum headroom to be able to transmit the frame.
2303	 * Otherwise return an error and let the original net_device handle it
2304	 */
2305	needed_headroom = dpaa2_eth_needed_headroom(NULL);
2306	if (xdpf->headroom < needed_headroom)
2307		return -EINVAL;
2308
2309	/* Setup the FD fields */
2310	memset(fd, 0, sizeof(*fd));
2311
2312	/* Align FD address, if possible */
2313	buffer_start = xdpf->data - needed_headroom;
2314	aligned_start = PTR_ALIGN(buffer_start - DPAA2_ETH_TX_BUF_ALIGN,
2315				  DPAA2_ETH_TX_BUF_ALIGN);
2316	if (aligned_start >= xdpf->data - xdpf->headroom)
2317		buffer_start = aligned_start;
2318
2319	swa = (struct dpaa2_eth_swa *)buffer_start;
2320	/* fill in necessary fields here */
2321	swa->type = DPAA2_ETH_SWA_XDP;
2322	swa->xdp.dma_size = xdpf->data + xdpf->len - buffer_start;
2323	swa->xdp.xdpf = xdpf;
2324
2325	addr = dma_map_single(dev, buffer_start,
2326			      swa->xdp.dma_size,
2327			      DMA_BIDIRECTIONAL);
2328	if (unlikely(dma_mapping_error(dev, addr)))
2329		return -ENOMEM;
2330
2331	dpaa2_fd_set_addr(fd, addr);
2332	dpaa2_fd_set_offset(fd, xdpf->data - buffer_start);
2333	dpaa2_fd_set_len(fd, xdpf->len);
2334	dpaa2_fd_set_format(fd, dpaa2_fd_single);
2335	dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA);
2336
2337	return 0;
2338}
2339
2340static int dpaa2_eth_xdp_xmit(struct net_device *net_dev, int n,
2341			      struct xdp_frame **frames, u32 flags)
2342{
2343	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2344	struct dpaa2_eth_xdp_fds *xdp_redirect_fds;
2345	struct rtnl_link_stats64 *percpu_stats;
2346	struct dpaa2_eth_fq *fq;
2347	struct dpaa2_fd *fds;
2348	int enqueued, i, err;
2349
2350	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
2351		return -EINVAL;
2352
2353	if (!netif_running(net_dev))
2354		return -ENETDOWN;
2355
2356	fq = &priv->fq[smp_processor_id()];
2357	xdp_redirect_fds = &fq->xdp_redirect_fds;
2358	fds = xdp_redirect_fds->fds;
2359
2360	percpu_stats = this_cpu_ptr(priv->percpu_stats);
2361
2362	/* create a FD for each xdp_frame in the list received */
2363	for (i = 0; i < n; i++) {
2364		err = dpaa2_eth_xdp_create_fd(net_dev, frames[i], &fds[i]);
2365		if (err)
2366			break;
2367	}
2368	xdp_redirect_fds->num = i;
2369
2370	/* enqueue all the frame descriptors */
2371	enqueued = dpaa2_eth_xdp_flush(priv, fq, xdp_redirect_fds);
2372
2373	/* update statistics */
2374	percpu_stats->tx_packets += enqueued;
2375	for (i = 0; i < enqueued; i++)
2376		percpu_stats->tx_bytes += dpaa2_fd_get_len(&fds[i]);
2377	for (i = enqueued; i < n; i++)
2378		xdp_return_frame_rx_napi(frames[i]);
2379
2380	return enqueued;
2381}
2382
2383static int update_xps(struct dpaa2_eth_priv *priv)
2384{
2385	struct net_device *net_dev = priv->net_dev;
2386	struct cpumask xps_mask;
2387	struct dpaa2_eth_fq *fq;
2388	int i, num_queues, netdev_queues;
2389	int err = 0;
2390
2391	num_queues = dpaa2_eth_queue_count(priv);
2392	netdev_queues = (net_dev->num_tc ? : 1) * num_queues;
2393
2394	/* The first <num_queues> entries in priv->fq array are Tx/Tx conf
2395	 * queues, so only process those
2396	 */
2397	for (i = 0; i < netdev_queues; i++) {
2398		fq = &priv->fq[i % num_queues];
2399
2400		cpumask_clear(&xps_mask);
2401		cpumask_set_cpu(fq->target_cpu, &xps_mask);
2402
2403		err = netif_set_xps_queue(net_dev, &xps_mask, i);
2404		if (err) {
2405			netdev_warn_once(net_dev, "Error setting XPS queue\n");
2406			break;
2407		}
2408	}
2409
2410	return err;
2411}
2412
2413static int dpaa2_eth_setup_mqprio(struct net_device *net_dev,
2414				  struct tc_mqprio_qopt *mqprio)
2415{
2416	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2417	u8 num_tc, num_queues;
2418	int i;
2419
2420	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
2421	num_queues = dpaa2_eth_queue_count(priv);
2422	num_tc = mqprio->num_tc;
2423
2424	if (num_tc == net_dev->num_tc)
2425		return 0;
2426
2427	if (num_tc  > dpaa2_eth_tc_count(priv)) {
2428		netdev_err(net_dev, "Max %d traffic classes supported\n",
2429			   dpaa2_eth_tc_count(priv));
2430		return -EOPNOTSUPP;
2431	}
2432
2433	if (!num_tc) {
2434		netdev_reset_tc(net_dev);
2435		netif_set_real_num_tx_queues(net_dev, num_queues);
2436		goto out;
2437	}
2438
2439	netdev_set_num_tc(net_dev, num_tc);
2440	netif_set_real_num_tx_queues(net_dev, num_tc * num_queues);
2441
2442	for (i = 0; i < num_tc; i++)
2443		netdev_set_tc_queue(net_dev, i, num_queues, i * num_queues);
2444
2445out:
2446	update_xps(priv);
2447
2448	return 0;
2449}
2450
2451#define bps_to_mbits(rate) (div_u64((rate), 1000000) * 8)
2452
2453static int dpaa2_eth_setup_tbf(struct net_device *net_dev, struct tc_tbf_qopt_offload *p)
2454{
2455	struct tc_tbf_qopt_offload_replace_params *cfg = &p->replace_params;
2456	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2457	struct dpni_tx_shaping_cfg tx_cr_shaper = { 0 };
2458	struct dpni_tx_shaping_cfg tx_er_shaper = { 0 };
2459	int err;
2460
2461	if (p->command == TC_TBF_STATS)
2462		return -EOPNOTSUPP;
2463
2464	/* Only per port Tx shaping */
2465	if (p->parent != TC_H_ROOT)
2466		return -EOPNOTSUPP;
2467
2468	if (p->command == TC_TBF_REPLACE) {
2469		if (cfg->max_size > DPAA2_ETH_MAX_BURST_SIZE) {
2470			netdev_err(net_dev, "burst size cannot be greater than %d\n",
2471				   DPAA2_ETH_MAX_BURST_SIZE);
2472			return -EINVAL;
2473		}
2474
2475		tx_cr_shaper.max_burst_size = cfg->max_size;
2476		/* The TBF interface is in bytes/s, whereas DPAA2 expects the
2477		 * rate in Mbits/s
2478		 */
2479		tx_cr_shaper.rate_limit = bps_to_mbits(cfg->rate.rate_bytes_ps);
2480	}
2481
2482	err = dpni_set_tx_shaping(priv->mc_io, 0, priv->mc_token, &tx_cr_shaper,
2483				  &tx_er_shaper, 0);
2484	if (err) {
2485		netdev_err(net_dev, "dpni_set_tx_shaping() = %d\n", err);
2486		return err;
2487	}
2488
2489	return 0;
2490}
2491
2492static int dpaa2_eth_setup_tc(struct net_device *net_dev,
2493			      enum tc_setup_type type, void *type_data)
2494{
2495	switch (type) {
2496	case TC_SETUP_QDISC_MQPRIO:
2497		return dpaa2_eth_setup_mqprio(net_dev, type_data);
2498	case TC_SETUP_QDISC_TBF:
2499		return dpaa2_eth_setup_tbf(net_dev, type_data);
2500	default:
2501		return -EOPNOTSUPP;
2502	}
2503}
2504
2505static const struct net_device_ops dpaa2_eth_ops = {
2506	.ndo_open = dpaa2_eth_open,
2507	.ndo_start_xmit = dpaa2_eth_tx,
2508	.ndo_stop = dpaa2_eth_stop,
2509	.ndo_set_mac_address = dpaa2_eth_set_addr,
2510	.ndo_get_stats64 = dpaa2_eth_get_stats,
2511	.ndo_set_rx_mode = dpaa2_eth_set_rx_mode,
2512	.ndo_set_features = dpaa2_eth_set_features,
2513	.ndo_do_ioctl = dpaa2_eth_ioctl,
2514	.ndo_change_mtu = dpaa2_eth_change_mtu,
2515	.ndo_bpf = dpaa2_eth_xdp,
2516	.ndo_xdp_xmit = dpaa2_eth_xdp_xmit,
2517	.ndo_setup_tc = dpaa2_eth_setup_tc,
2518};
2519
2520static void dpaa2_eth_cdan_cb(struct dpaa2_io_notification_ctx *ctx)
2521{
2522	struct dpaa2_eth_channel *ch;
2523
2524	ch = container_of(ctx, struct dpaa2_eth_channel, nctx);
2525
2526	/* Update NAPI statistics */
2527	ch->stats.cdan++;
2528
2529	napi_schedule(&ch->napi);
2530}
2531
2532/* Allocate and configure a DPCON object */
2533static struct fsl_mc_device *dpaa2_eth_setup_dpcon(struct dpaa2_eth_priv *priv)
2534{
2535	struct fsl_mc_device *dpcon;
2536	struct device *dev = priv->net_dev->dev.parent;
2537	int err;
2538
2539	err = fsl_mc_object_allocate(to_fsl_mc_device(dev),
2540				     FSL_MC_POOL_DPCON, &dpcon);
2541	if (err) {
2542		if (err == -ENXIO)
2543			err = -EPROBE_DEFER;
2544		else
2545			dev_info(dev, "Not enough DPCONs, will go on as-is\n");
2546		return ERR_PTR(err);
2547	}
2548
2549	err = dpcon_open(priv->mc_io, 0, dpcon->obj_desc.id, &dpcon->mc_handle);
2550	if (err) {
2551		dev_err(dev, "dpcon_open() failed\n");
2552		goto free;
2553	}
2554
2555	err = dpcon_reset(priv->mc_io, 0, dpcon->mc_handle);
2556	if (err) {
2557		dev_err(dev, "dpcon_reset() failed\n");
2558		goto close;
2559	}
2560
2561	err = dpcon_enable(priv->mc_io, 0, dpcon->mc_handle);
2562	if (err) {
2563		dev_err(dev, "dpcon_enable() failed\n");
2564		goto close;
2565	}
2566
2567	return dpcon;
2568
2569close:
2570	dpcon_close(priv->mc_io, 0, dpcon->mc_handle);
2571free:
2572	fsl_mc_object_free(dpcon);
2573
2574	return ERR_PTR(err);
2575}
2576
2577static void dpaa2_eth_free_dpcon(struct dpaa2_eth_priv *priv,
2578				 struct fsl_mc_device *dpcon)
2579{
2580	dpcon_disable(priv->mc_io, 0, dpcon->mc_handle);
2581	dpcon_close(priv->mc_io, 0, dpcon->mc_handle);
2582	fsl_mc_object_free(dpcon);
2583}
2584
2585static struct dpaa2_eth_channel *dpaa2_eth_alloc_channel(struct dpaa2_eth_priv *priv)
2586{
2587	struct dpaa2_eth_channel *channel;
2588	struct dpcon_attr attr;
2589	struct device *dev = priv->net_dev->dev.parent;
2590	int err;
2591
2592	channel = kzalloc(sizeof(*channel), GFP_KERNEL);
2593	if (!channel)
2594		return NULL;
2595
2596	channel->dpcon = dpaa2_eth_setup_dpcon(priv);
2597	if (IS_ERR(channel->dpcon)) {
2598		err = PTR_ERR(channel->dpcon);
2599		goto err_setup;
2600	}
2601
2602	err = dpcon_get_attributes(priv->mc_io, 0, channel->dpcon->mc_handle,
2603				   &attr);
2604	if (err) {
2605		dev_err(dev, "dpcon_get_attributes() failed\n");
2606		goto err_get_attr;
2607	}
2608
2609	channel->dpcon_id = attr.id;
2610	channel->ch_id = attr.qbman_ch_id;
2611	channel->priv = priv;
2612
2613	return channel;
2614
2615err_get_attr:
2616	dpaa2_eth_free_dpcon(priv, channel->dpcon);
2617err_setup:
2618	kfree(channel);
2619	return ERR_PTR(err);
2620}
2621
2622static void dpaa2_eth_free_channel(struct dpaa2_eth_priv *priv,
2623				   struct dpaa2_eth_channel *channel)
2624{
2625	dpaa2_eth_free_dpcon(priv, channel->dpcon);
2626	kfree(channel);
2627}
2628
2629/* DPIO setup: allocate and configure QBMan channels, setup core affinity
2630 * and register data availability notifications
2631 */
2632static int dpaa2_eth_setup_dpio(struct dpaa2_eth_priv *priv)
2633{
2634	struct dpaa2_io_notification_ctx *nctx;
2635	struct dpaa2_eth_channel *channel;
2636	struct dpcon_notification_cfg dpcon_notif_cfg;
2637	struct device *dev = priv->net_dev->dev.parent;
2638	int i, err;
2639
2640	/* We want the ability to spread ingress traffic (RX, TX conf) to as
2641	 * many cores as possible, so we need one channel for each core
2642	 * (unless there's fewer queues than cores, in which case the extra
2643	 * channels would be wasted).
2644	 * Allocate one channel per core and register it to the core's
2645	 * affine DPIO. If not enough channels are available for all cores
2646	 * or if some cores don't have an affine DPIO, there will be no
2647	 * ingress frame processing on those cores.
2648	 */
2649	cpumask_clear(&priv->dpio_cpumask);
2650	for_each_online_cpu(i) {
2651		/* Try to allocate a channel */
2652		channel = dpaa2_eth_alloc_channel(priv);
2653		if (IS_ERR_OR_NULL(channel)) {
2654			err = PTR_ERR_OR_ZERO(channel);
2655			if (err != -EPROBE_DEFER)
2656				dev_info(dev,
2657					 "No affine channel for cpu %d and above\n", i);
2658			goto err_alloc_ch;
2659		}
2660
2661		priv->channel[priv->num_channels] = channel;
2662
2663		nctx = &channel->nctx;
2664		nctx->is_cdan = 1;
2665		nctx->cb = dpaa2_eth_cdan_cb;
2666		nctx->id = channel->ch_id;
2667		nctx->desired_cpu = i;
2668
2669		/* Register the new context */
2670		channel->dpio = dpaa2_io_service_select(i);
2671		err = dpaa2_io_service_register(channel->dpio, nctx, dev);
2672		if (err) {
2673			dev_dbg(dev, "No affine DPIO for cpu %d\n", i);
2674			/* If no affine DPIO for this core, there's probably
2675			 * none available for next cores either. Signal we want
2676			 * to retry later, in case the DPIO devices weren't
2677			 * probed yet.
2678			 */
2679			err = -EPROBE_DEFER;
2680			goto err_service_reg;
2681		}
2682
2683		/* Register DPCON notification with MC */
2684		dpcon_notif_cfg.dpio_id = nctx->dpio_id;
2685		dpcon_notif_cfg.priority = 0;
2686		dpcon_notif_cfg.user_ctx = nctx->qman64;
2687		err = dpcon_set_notification(priv->mc_io, 0,
2688					     channel->dpcon->mc_handle,
2689					     &dpcon_notif_cfg);
2690		if (err) {
2691			dev_err(dev, "dpcon_set_notification failed()\n");
2692			goto err_set_cdan;
2693		}
2694
2695		/* If we managed to allocate a channel and also found an affine
2696		 * DPIO for this core, add it to the final mask
2697		 */
2698		cpumask_set_cpu(i, &priv->dpio_cpumask);
2699		priv->num_channels++;
2700
2701		/* Stop if we already have enough channels to accommodate all
2702		 * RX and TX conf queues
2703		 */
2704		if (priv->num_channels == priv->dpni_attrs.num_queues)
2705			break;
2706	}
2707
2708	return 0;
2709
2710err_set_cdan:
2711	dpaa2_io_service_deregister(channel->dpio, nctx, dev);
2712err_service_reg:
2713	dpaa2_eth_free_channel(priv, channel);
2714err_alloc_ch:
2715	if (err == -EPROBE_DEFER) {
2716		for (i = 0; i < priv->num_channels; i++) {
2717			channel = priv->channel[i];
2718			nctx = &channel->nctx;
2719			dpaa2_io_service_deregister(channel->dpio, nctx, dev);
2720			dpaa2_eth_free_channel(priv, channel);
2721		}
2722		priv->num_channels = 0;
2723		return err;
2724	}
2725
2726	if (cpumask_empty(&priv->dpio_cpumask)) {
2727		dev_err(dev, "No cpu with an affine DPIO/DPCON\n");
2728		return -ENODEV;
2729	}
2730
2731	dev_info(dev, "Cores %*pbl available for processing ingress traffic\n",
2732		 cpumask_pr_args(&priv->dpio_cpumask));
2733
2734	return 0;
2735}
2736
2737static void dpaa2_eth_free_dpio(struct dpaa2_eth_priv *priv)
2738{
2739	struct device *dev = priv->net_dev->dev.parent;
2740	struct dpaa2_eth_channel *ch;
2741	int i;
2742
2743	/* deregister CDAN notifications and free channels */
2744	for (i = 0; i < priv->num_channels; i++) {
2745		ch = priv->channel[i];
2746		dpaa2_io_service_deregister(ch->dpio, &ch->nctx, dev);
2747		dpaa2_eth_free_channel(priv, ch);
2748	}
2749}
2750
2751static struct dpaa2_eth_channel *dpaa2_eth_get_affine_channel(struct dpaa2_eth_priv *priv,
2752							      int cpu)
2753{
2754	struct device *dev = priv->net_dev->dev.parent;
2755	int i;
2756
2757	for (i = 0; i < priv->num_channels; i++)
2758		if (priv->channel[i]->nctx.desired_cpu == cpu)
2759			return priv->channel[i];
2760
2761	/* We should never get here. Issue a warning and return
2762	 * the first channel, because it's still better than nothing
2763	 */
2764	dev_warn(dev, "No affine channel found for cpu %d\n", cpu);
2765
2766	return priv->channel[0];
2767}
2768
2769static void dpaa2_eth_set_fq_affinity(struct dpaa2_eth_priv *priv)
2770{
2771	struct device *dev = priv->net_dev->dev.parent;
2772	struct dpaa2_eth_fq *fq;
2773	int rx_cpu, txc_cpu;
2774	int i;
2775
2776	/* For each FQ, pick one channel/CPU to deliver frames to.
2777	 * This may well change at runtime, either through irqbalance or
2778	 * through direct user intervention.
2779	 */
2780	rx_cpu = txc_cpu = cpumask_first(&priv->dpio_cpumask);
2781
2782	for (i = 0; i < priv->num_fqs; i++) {
2783		fq = &priv->fq[i];
2784		switch (fq->type) {
2785		case DPAA2_RX_FQ:
2786		case DPAA2_RX_ERR_FQ:
2787			fq->target_cpu = rx_cpu;
2788			rx_cpu = cpumask_next(rx_cpu, &priv->dpio_cpumask);
2789			if (rx_cpu >= nr_cpu_ids)
2790				rx_cpu = cpumask_first(&priv->dpio_cpumask);
2791			break;
2792		case DPAA2_TX_CONF_FQ:
2793			fq->target_cpu = txc_cpu;
2794			txc_cpu = cpumask_next(txc_cpu, &priv->dpio_cpumask);
2795			if (txc_cpu >= nr_cpu_ids)
2796				txc_cpu = cpumask_first(&priv->dpio_cpumask);
2797			break;
2798		default:
2799			dev_err(dev, "Unknown FQ type: %d\n", fq->type);
2800		}
2801		fq->channel = dpaa2_eth_get_affine_channel(priv, fq->target_cpu);
2802	}
2803
2804	update_xps(priv);
2805}
2806
2807static void dpaa2_eth_setup_fqs(struct dpaa2_eth_priv *priv)
2808{
2809	int i, j;
2810
2811	/* We have one TxConf FQ per Tx flow.
2812	 * The number of Tx and Rx queues is the same.
2813	 * Tx queues come first in the fq array.
2814	 */
2815	for (i = 0; i < dpaa2_eth_queue_count(priv); i++) {
2816		priv->fq[priv->num_fqs].type = DPAA2_TX_CONF_FQ;
2817		priv->fq[priv->num_fqs].consume = dpaa2_eth_tx_conf;
2818		priv->fq[priv->num_fqs++].flowid = (u16)i;
2819	}
2820
2821	for (j = 0; j < dpaa2_eth_tc_count(priv); j++) {
2822		for (i = 0; i < dpaa2_eth_queue_count(priv); i++) {
2823			priv->fq[priv->num_fqs].type = DPAA2_RX_FQ;
2824			priv->fq[priv->num_fqs].consume = dpaa2_eth_rx;
2825			priv->fq[priv->num_fqs].tc = (u8)j;
2826			priv->fq[priv->num_fqs++].flowid = (u16)i;
2827		}
2828	}
2829
2830	/* We have exactly one Rx error queue per DPNI */
2831	priv->fq[priv->num_fqs].type = DPAA2_RX_ERR_FQ;
2832	priv->fq[priv->num_fqs++].consume = dpaa2_eth_rx_err;
2833
2834	/* For each FQ, decide on which core to process incoming frames */
2835	dpaa2_eth_set_fq_affinity(priv);
2836}
2837
2838/* Allocate and configure one buffer pool for each interface */
2839static int dpaa2_eth_setup_dpbp(struct dpaa2_eth_priv *priv)
2840{
2841	int err;
2842	struct fsl_mc_device *dpbp_dev;
2843	struct device *dev = priv->net_dev->dev.parent;
2844	struct dpbp_attr dpbp_attrs;
2845
2846	err = fsl_mc_object_allocate(to_fsl_mc_device(dev), FSL_MC_POOL_DPBP,
2847				     &dpbp_dev);
2848	if (err) {
2849		if (err == -ENXIO)
2850			err = -EPROBE_DEFER;
2851		else
2852			dev_err(dev, "DPBP device allocation failed\n");
2853		return err;
2854	}
2855
2856	priv->dpbp_dev = dpbp_dev;
2857
2858	err = dpbp_open(priv->mc_io, 0, priv->dpbp_dev->obj_desc.id,
2859			&dpbp_dev->mc_handle);
2860	if (err) {
2861		dev_err(dev, "dpbp_open() failed\n");
2862		goto err_open;
2863	}
2864
2865	err = dpbp_reset(priv->mc_io, 0, dpbp_dev->mc_handle);
2866	if (err) {
2867		dev_err(dev, "dpbp_reset() failed\n");
2868		goto err_reset;
2869	}
2870
2871	err = dpbp_enable(priv->mc_io, 0, dpbp_dev->mc_handle);
2872	if (err) {
2873		dev_err(dev, "dpbp_enable() failed\n");
2874		goto err_enable;
2875	}
2876
2877	err = dpbp_get_attributes(priv->mc_io, 0, dpbp_dev->mc_handle,
2878				  &dpbp_attrs);
2879	if (err) {
2880		dev_err(dev, "dpbp_get_attributes() failed\n");
2881		goto err_get_attr;
2882	}
2883	priv->bpid = dpbp_attrs.bpid;
2884
2885	return 0;
2886
2887err_get_attr:
2888	dpbp_disable(priv->mc_io, 0, dpbp_dev->mc_handle);
2889err_enable:
2890err_reset:
2891	dpbp_close(priv->mc_io, 0, dpbp_dev->mc_handle);
2892err_open:
2893	fsl_mc_object_free(dpbp_dev);
2894
2895	return err;
2896}
2897
2898static void dpaa2_eth_free_dpbp(struct dpaa2_eth_priv *priv)
2899{
2900	dpaa2_eth_drain_pool(priv);
2901	dpbp_disable(priv->mc_io, 0, priv->dpbp_dev->mc_handle);
2902	dpbp_close(priv->mc_io, 0, priv->dpbp_dev->mc_handle);
2903	fsl_mc_object_free(priv->dpbp_dev);
2904}
2905
2906static int dpaa2_eth_set_buffer_layout(struct dpaa2_eth_priv *priv)
2907{
2908	struct device *dev = priv->net_dev->dev.parent;
2909	struct dpni_buffer_layout buf_layout = {0};
2910	u16 rx_buf_align;
2911	int err;
2912
2913	/* We need to check for WRIOP version 1.0.0, but depending on the MC
2914	 * version, this number is not always provided correctly on rev1.
2915	 * We need to check for both alternatives in this situation.
2916	 */
2917	if (priv->dpni_attrs.wriop_version == DPAA2_WRIOP_VERSION(0, 0, 0) ||
2918	    priv->dpni_attrs.wriop_version == DPAA2_WRIOP_VERSION(1, 0, 0))
2919		rx_buf_align = DPAA2_ETH_RX_BUF_ALIGN_REV1;
2920	else
2921		rx_buf_align = DPAA2_ETH_RX_BUF_ALIGN;
2922
2923	/* We need to ensure that the buffer size seen by WRIOP is a multiple
2924	 * of 64 or 256 bytes depending on the WRIOP version.
2925	 */
2926	priv->rx_buf_size = ALIGN_DOWN(DPAA2_ETH_RX_BUF_SIZE, rx_buf_align);
2927
2928	/* tx buffer */
2929	buf_layout.private_data_size = DPAA2_ETH_SWA_SIZE;
2930	buf_layout.pass_timestamp = true;
2931	buf_layout.pass_frame_status = true;
2932	buf_layout.options = DPNI_BUF_LAYOUT_OPT_PRIVATE_DATA_SIZE |
2933			     DPNI_BUF_LAYOUT_OPT_TIMESTAMP |
2934			     DPNI_BUF_LAYOUT_OPT_FRAME_STATUS;
2935	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
2936				     DPNI_QUEUE_TX, &buf_layout);
2937	if (err) {
2938		dev_err(dev, "dpni_set_buffer_layout(TX) failed\n");
2939		return err;
2940	}
2941
2942	/* tx-confirm buffer */
2943	buf_layout.options = DPNI_BUF_LAYOUT_OPT_TIMESTAMP |
2944			     DPNI_BUF_LAYOUT_OPT_FRAME_STATUS;
2945	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
2946				     DPNI_QUEUE_TX_CONFIRM, &buf_layout);
2947	if (err) {
2948		dev_err(dev, "dpni_set_buffer_layout(TX_CONF) failed\n");
2949		return err;
2950	}
2951
2952	/* Now that we've set our tx buffer layout, retrieve the minimum
2953	 * required tx data offset.
2954	 */
2955	err = dpni_get_tx_data_offset(priv->mc_io, 0, priv->mc_token,
2956				      &priv->tx_data_offset);
2957	if (err) {
2958		dev_err(dev, "dpni_get_tx_data_offset() failed\n");
2959		return err;
2960	}
2961
2962	if ((priv->tx_data_offset % 64) != 0)
2963		dev_warn(dev, "Tx data offset (%d) not a multiple of 64B\n",
2964			 priv->tx_data_offset);
2965
2966	/* rx buffer */
2967	buf_layout.pass_frame_status = true;
2968	buf_layout.pass_parser_result = true;
2969	buf_layout.data_align = rx_buf_align;
2970	buf_layout.data_head_room = dpaa2_eth_rx_head_room(priv);
2971	buf_layout.private_data_size = 0;
2972	buf_layout.options = DPNI_BUF_LAYOUT_OPT_PARSER_RESULT |
2973			     DPNI_BUF_LAYOUT_OPT_FRAME_STATUS |
2974			     DPNI_BUF_LAYOUT_OPT_DATA_ALIGN |
2975			     DPNI_BUF_LAYOUT_OPT_DATA_HEAD_ROOM |
2976			     DPNI_BUF_LAYOUT_OPT_TIMESTAMP;
2977	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
2978				     DPNI_QUEUE_RX, &buf_layout);
2979	if (err) {
2980		dev_err(dev, "dpni_set_buffer_layout(RX) failed\n");
2981		return err;
2982	}
2983
2984	return 0;
2985}
2986
2987#define DPNI_ENQUEUE_FQID_VER_MAJOR	7
2988#define DPNI_ENQUEUE_FQID_VER_MINOR	9
2989
2990static inline int dpaa2_eth_enqueue_qd(struct dpaa2_eth_priv *priv,
2991				       struct dpaa2_eth_fq *fq,
2992				       struct dpaa2_fd *fd, u8 prio,
2993				       u32 num_frames __always_unused,
2994				       int *frames_enqueued)
2995{
2996	int err;
2997
2998	err = dpaa2_io_service_enqueue_qd(fq->channel->dpio,
2999					  priv->tx_qdid, prio,
3000					  fq->tx_qdbin, fd);
3001	if (!err && frames_enqueued)
3002		*frames_enqueued = 1;
3003	return err;
3004}
3005
3006static inline int dpaa2_eth_enqueue_fq_multiple(struct dpaa2_eth_priv *priv,
3007						struct dpaa2_eth_fq *fq,
3008						struct dpaa2_fd *fd,
3009						u8 prio, u32 num_frames,
3010						int *frames_enqueued)
3011{
3012	int err;
3013
3014	err = dpaa2_io_service_enqueue_multiple_fq(fq->channel->dpio,
3015						   fq->tx_fqid[prio],
3016						   fd, num_frames);
3017
3018	if (err == 0)
3019		return -EBUSY;
3020
3021	if (frames_enqueued)
3022		*frames_enqueued = err;
3023	return 0;
3024}
3025
3026static void dpaa2_eth_set_enqueue_mode(struct dpaa2_eth_priv *priv)
3027{
3028	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_ENQUEUE_FQID_VER_MAJOR,
3029				   DPNI_ENQUEUE_FQID_VER_MINOR) < 0)
3030		priv->enqueue = dpaa2_eth_enqueue_qd;
3031	else
3032		priv->enqueue = dpaa2_eth_enqueue_fq_multiple;
3033}
3034
3035static int dpaa2_eth_set_pause(struct dpaa2_eth_priv *priv)
3036{
3037	struct device *dev = priv->net_dev->dev.parent;
3038	struct dpni_link_cfg link_cfg = {0};
3039	int err;
3040
3041	/* Get the default link options so we don't override other flags */
3042	err = dpni_get_link_cfg(priv->mc_io, 0, priv->mc_token, &link_cfg);
3043	if (err) {
3044		dev_err(dev, "dpni_get_link_cfg() failed\n");
3045		return err;
3046	}
3047
3048	/* By default, enable both Rx and Tx pause frames */
3049	link_cfg.options |= DPNI_LINK_OPT_PAUSE;
3050	link_cfg.options &= ~DPNI_LINK_OPT_ASYM_PAUSE;
3051	err = dpni_set_link_cfg(priv->mc_io, 0, priv->mc_token, &link_cfg);
3052	if (err) {
3053		dev_err(dev, "dpni_set_link_cfg() failed\n");
3054		return err;
3055	}
3056
3057	priv->link_state.options = link_cfg.options;
3058
3059	return 0;
3060}
3061
3062static void dpaa2_eth_update_tx_fqids(struct dpaa2_eth_priv *priv)
3063{
3064	struct dpni_queue_id qid = {0};
3065	struct dpaa2_eth_fq *fq;
3066	struct dpni_queue queue;
3067	int i, j, err;
3068
3069	/* We only use Tx FQIDs for FQID-based enqueue, so check
3070	 * if DPNI version supports it before updating FQIDs
3071	 */
3072	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_ENQUEUE_FQID_VER_MAJOR,
3073				   DPNI_ENQUEUE_FQID_VER_MINOR) < 0)
3074		return;
3075
3076	for (i = 0; i < priv->num_fqs; i++) {
3077		fq = &priv->fq[i];
3078		if (fq->type != DPAA2_TX_CONF_FQ)
3079			continue;
3080		for (j = 0; j < dpaa2_eth_tc_count(priv); j++) {
3081			err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3082					     DPNI_QUEUE_TX, j, fq->flowid,
3083					     &queue, &qid);
3084			if (err)
3085				goto out_err;
3086
3087			fq->tx_fqid[j] = qid.fqid;
3088			if (fq->tx_fqid[j] == 0)
3089				goto out_err;
3090		}
3091	}
3092
3093	priv->enqueue = dpaa2_eth_enqueue_fq_multiple;
3094
3095	return;
3096
3097out_err:
3098	netdev_info(priv->net_dev,
3099		    "Error reading Tx FQID, fallback to QDID-based enqueue\n");
3100	priv->enqueue = dpaa2_eth_enqueue_qd;
3101}
3102
3103/* Configure ingress classification based on VLAN PCP */
3104static int dpaa2_eth_set_vlan_qos(struct dpaa2_eth_priv *priv)
3105{
3106	struct device *dev = priv->net_dev->dev.parent;
3107	struct dpkg_profile_cfg kg_cfg = {0};
3108	struct dpni_qos_tbl_cfg qos_cfg = {0};
3109	struct dpni_rule_cfg key_params;
3110	void *dma_mem, *key, *mask;
3111	u8 key_size = 2;	/* VLAN TCI field */
3112	int i, pcp, err;
3113
3114	/* VLAN-based classification only makes sense if we have multiple
3115	 * traffic classes.
3116	 * Also, we need to extract just the 3-bit PCP field from the VLAN
3117	 * header and we can only do that by using a mask
3118	 */
3119	if (dpaa2_eth_tc_count(priv) == 1 || !dpaa2_eth_fs_mask_enabled(priv)) {
3120		dev_dbg(dev, "VLAN-based QoS classification not supported\n");
3121		return -EOPNOTSUPP;
3122	}
3123
3124	dma_mem = kzalloc(DPAA2_CLASSIFIER_DMA_SIZE, GFP_KERNEL);
3125	if (!dma_mem)
3126		return -ENOMEM;
3127
3128	kg_cfg.num_extracts = 1;
3129	kg_cfg.extracts[0].type = DPKG_EXTRACT_FROM_HDR;
3130	kg_cfg.extracts[0].extract.from_hdr.prot = NET_PROT_VLAN;
3131	kg_cfg.extracts[0].extract.from_hdr.type = DPKG_FULL_FIELD;
3132	kg_cfg.extracts[0].extract.from_hdr.field = NH_FLD_VLAN_TCI;
3133
3134	err = dpni_prepare_key_cfg(&kg_cfg, dma_mem);
3135	if (err) {
3136		dev_err(dev, "dpni_prepare_key_cfg failed\n");
3137		goto out_free_tbl;
3138	}
3139
3140	/* set QoS table */
3141	qos_cfg.default_tc = 0;
3142	qos_cfg.discard_on_miss = 0;
3143	qos_cfg.key_cfg_iova = dma_map_single(dev, dma_mem,
3144					      DPAA2_CLASSIFIER_DMA_SIZE,
3145					      DMA_TO_DEVICE);
3146	if (dma_mapping_error(dev, qos_cfg.key_cfg_iova)) {
3147		dev_err(dev, "QoS table DMA mapping failed\n");
3148		err = -ENOMEM;
3149		goto out_free_tbl;
3150	}
3151
3152	err = dpni_set_qos_table(priv->mc_io, 0, priv->mc_token, &qos_cfg);
3153	if (err) {
3154		dev_err(dev, "dpni_set_qos_table failed\n");
3155		goto out_unmap_tbl;
3156	}
3157
3158	/* Add QoS table entries */
3159	key = kzalloc(key_size * 2, GFP_KERNEL);
3160	if (!key) {
3161		err = -ENOMEM;
3162		goto out_unmap_tbl;
3163	}
3164	mask = key + key_size;
3165	*(__be16 *)mask = cpu_to_be16(VLAN_PRIO_MASK);
3166
3167	key_params.key_iova = dma_map_single(dev, key, key_size * 2,
3168					     DMA_TO_DEVICE);
3169	if (dma_mapping_error(dev, key_params.key_iova)) {
3170		dev_err(dev, "Qos table entry DMA mapping failed\n");
3171		err = -ENOMEM;
3172		goto out_free_key;
3173	}
3174
3175	key_params.mask_iova = key_params.key_iova + key_size;
3176	key_params.key_size = key_size;
3177
3178	/* We add rules for PCP-based distribution starting with highest
3179	 * priority (VLAN PCP = 7). If this DPNI doesn't have enough traffic
3180	 * classes to accommodate all priority levels, the lowest ones end up
3181	 * on TC 0 which was configured as default
3182	 */
3183	for (i = dpaa2_eth_tc_count(priv) - 1, pcp = 7; i >= 0; i--, pcp--) {
3184		*(__be16 *)key = cpu_to_be16(pcp << VLAN_PRIO_SHIFT);
3185		dma_sync_single_for_device(dev, key_params.key_iova,
3186					   key_size * 2, DMA_TO_DEVICE);
3187
3188		err = dpni_add_qos_entry(priv->mc_io, 0, priv->mc_token,
3189					 &key_params, i, i);
3190		if (err) {
3191			dev_err(dev, "dpni_add_qos_entry failed\n");
3192			dpni_clear_qos_table(priv->mc_io, 0, priv->mc_token);
3193			goto out_unmap_key;
3194		}
3195	}
3196
3197	priv->vlan_cls_enabled = true;
3198
3199	/* Table and key memory is not persistent, clean everything up after
3200	 * configuration is finished
3201	 */
3202out_unmap_key:
3203	dma_unmap_single(dev, key_params.key_iova, key_size * 2, DMA_TO_DEVICE);
3204out_free_key:
3205	kfree(key);
3206out_unmap_tbl:
3207	dma_unmap_single(dev, qos_cfg.key_cfg_iova, DPAA2_CLASSIFIER_DMA_SIZE,
3208			 DMA_TO_DEVICE);
3209out_free_tbl:
3210	kfree(dma_mem);
3211
3212	return err;
3213}
3214
3215/* Configure the DPNI object this interface is associated with */
3216static int dpaa2_eth_setup_dpni(struct fsl_mc_device *ls_dev)
3217{
3218	struct device *dev = &ls_dev->dev;
3219	struct dpaa2_eth_priv *priv;
3220	struct net_device *net_dev;
3221	int err;
3222
3223	net_dev = dev_get_drvdata(dev);
3224	priv = netdev_priv(net_dev);
3225
3226	/* get a handle for the DPNI object */
3227	err = dpni_open(priv->mc_io, 0, ls_dev->obj_desc.id, &priv->mc_token);
3228	if (err) {
3229		dev_err(dev, "dpni_open() failed\n");
3230		return err;
3231	}
3232
3233	/* Check if we can work with this DPNI object */
3234	err = dpni_get_api_version(priv->mc_io, 0, &priv->dpni_ver_major,
3235				   &priv->dpni_ver_minor);
3236	if (err) {
3237		dev_err(dev, "dpni_get_api_version() failed\n");
3238		goto close;
3239	}
3240	if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_VER_MAJOR, DPNI_VER_MINOR) < 0) {
3241		dev_err(dev, "DPNI version %u.%u not supported, need >= %u.%u\n",
3242			priv->dpni_ver_major, priv->dpni_ver_minor,
3243			DPNI_VER_MAJOR, DPNI_VER_MINOR);
3244		err = -ENOTSUPP;
3245		goto close;
3246	}
3247
3248	ls_dev->mc_io = priv->mc_io;
3249	ls_dev->mc_handle = priv->mc_token;
3250
3251	err = dpni_reset(priv->mc_io, 0, priv->mc_token);
3252	if (err) {
3253		dev_err(dev, "dpni_reset() failed\n");
3254		goto close;
3255	}
3256
3257	err = dpni_get_attributes(priv->mc_io, 0, priv->mc_token,
3258				  &priv->dpni_attrs);
3259	if (err) {
3260		dev_err(dev, "dpni_get_attributes() failed (err=%d)\n", err);
3261		goto close;
3262	}
3263
3264	err = dpaa2_eth_set_buffer_layout(priv);
3265	if (err)
3266		goto close;
3267
3268	dpaa2_eth_set_enqueue_mode(priv);
3269
3270	/* Enable pause frame support */
3271	if (dpaa2_eth_has_pause_support(priv)) {
3272		err = dpaa2_eth_set_pause(priv);
3273		if (err)
3274			goto close;
3275	}
3276
3277	err = dpaa2_eth_set_vlan_qos(priv);
3278	if (err && err != -EOPNOTSUPP)
3279		goto close;
3280
3281	priv->cls_rules = devm_kcalloc(dev, dpaa2_eth_fs_count(priv),
3282				       sizeof(struct dpaa2_eth_cls_rule),
3283				       GFP_KERNEL);
3284	if (!priv->cls_rules) {
3285		err = -ENOMEM;
3286		goto close;
3287	}
3288
3289	return 0;
3290
3291close:
3292	dpni_close(priv->mc_io, 0, priv->mc_token);
3293
3294	return err;
3295}
3296
3297static void dpaa2_eth_free_dpni(struct dpaa2_eth_priv *priv)
3298{
3299	int err;
3300
3301	err = dpni_reset(priv->mc_io, 0, priv->mc_token);
3302	if (err)
3303		netdev_warn(priv->net_dev, "dpni_reset() failed (err %d)\n",
3304			    err);
3305
3306	dpni_close(priv->mc_io, 0, priv->mc_token);
3307}
3308
3309static int dpaa2_eth_setup_rx_flow(struct dpaa2_eth_priv *priv,
3310				   struct dpaa2_eth_fq *fq)
3311{
3312	struct device *dev = priv->net_dev->dev.parent;
3313	struct dpni_queue queue;
3314	struct dpni_queue_id qid;
3315	int err;
3316
3317	err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3318			     DPNI_QUEUE_RX, fq->tc, fq->flowid, &queue, &qid);
3319	if (err) {
3320		dev_err(dev, "dpni_get_queue(RX) failed\n");
3321		return err;
3322	}
3323
3324	fq->fqid = qid.fqid;
3325
3326	queue.destination.id = fq->channel->dpcon_id;
3327	queue.destination.type = DPNI_DEST_DPCON;
3328	queue.destination.priority = 1;
3329	queue.user_context = (u64)(uintptr_t)fq;
3330	err = dpni_set_queue(priv->mc_io, 0, priv->mc_token,
3331			     DPNI_QUEUE_RX, fq->tc, fq->flowid,
3332			     DPNI_QUEUE_OPT_USER_CTX | DPNI_QUEUE_OPT_DEST,
3333			     &queue);
3334	if (err) {
3335		dev_err(dev, "dpni_set_queue(RX) failed\n");
3336		return err;
3337	}
3338
3339	/* xdp_rxq setup */
3340	/* only once for each channel */
3341	if (fq->tc > 0)
3342		return 0;
3343
3344	err = xdp_rxq_info_reg(&fq->channel->xdp_rxq, priv->net_dev,
3345			       fq->flowid);
3346	if (err) {
3347		dev_err(dev, "xdp_rxq_info_reg failed\n");
3348		return err;
3349	}
3350
3351	err = xdp_rxq_info_reg_mem_model(&fq->channel->xdp_rxq,
3352					 MEM_TYPE_PAGE_ORDER0, NULL);
3353	if (err) {
3354		dev_err(dev, "xdp_rxq_info_reg_mem_model failed\n");
3355		return err;
3356	}
3357
3358	return 0;
3359}
3360
3361static int dpaa2_eth_setup_tx_flow(struct dpaa2_eth_priv *priv,
3362				   struct dpaa2_eth_fq *fq)
3363{
3364	struct device *dev = priv->net_dev->dev.parent;
3365	struct dpni_queue queue;
3366	struct dpni_queue_id qid;
3367	int i, err;
3368
3369	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3370		err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3371				     DPNI_QUEUE_TX, i, fq->flowid,
3372				     &queue, &qid);
3373		if (err) {
3374			dev_err(dev, "dpni_get_queue(TX) failed\n");
3375			return err;
3376		}
3377		fq->tx_fqid[i] = qid.fqid;
3378	}
3379
3380	/* All Tx queues belonging to the same flowid have the same qdbin */
3381	fq->tx_qdbin = qid.qdbin;
3382
3383	err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3384			     DPNI_QUEUE_TX_CONFIRM, 0, fq->flowid,
3385			     &queue, &qid);
3386	if (err) {
3387		dev_err(dev, "dpni_get_queue(TX_CONF) failed\n");
3388		return err;
3389	}
3390
3391	fq->fqid = qid.fqid;
3392
3393	queue.destination.id = fq->channel->dpcon_id;
3394	queue.destination.type = DPNI_DEST_DPCON;
3395	queue.destination.priority = 0;
3396	queue.user_context = (u64)(uintptr_t)fq;
3397	err = dpni_set_queue(priv->mc_io, 0, priv->mc_token,
3398			     DPNI_QUEUE_TX_CONFIRM, 0, fq->flowid,
3399			     DPNI_QUEUE_OPT_USER_CTX | DPNI_QUEUE_OPT_DEST,
3400			     &queue);
3401	if (err) {
3402		dev_err(dev, "dpni_set_queue(TX_CONF) failed\n");
3403		return err;
3404	}
3405
3406	return 0;
3407}
3408
3409static int setup_rx_err_flow(struct dpaa2_eth_priv *priv,
3410			     struct dpaa2_eth_fq *fq)
3411{
3412	struct device *dev = priv->net_dev->dev.parent;
3413	struct dpni_queue q = { { 0 } };
3414	struct dpni_queue_id qid;
3415	u8 q_opt = DPNI_QUEUE_OPT_USER_CTX | DPNI_QUEUE_OPT_DEST;
3416	int err;
3417
3418	err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
3419			     DPNI_QUEUE_RX_ERR, 0, 0, &q, &qid);
3420	if (err) {
3421		dev_err(dev, "dpni_get_queue() failed (%d)\n", err);
3422		return err;
3423	}
3424
3425	fq->fqid = qid.fqid;
3426
3427	q.destination.id = fq->channel->dpcon_id;
3428	q.destination.type = DPNI_DEST_DPCON;
3429	q.destination.priority = 1;
3430	q.user_context = (u64)(uintptr_t)fq;
3431	err = dpni_set_queue(priv->mc_io, 0, priv->mc_token,
3432			     DPNI_QUEUE_RX_ERR, 0, 0, q_opt, &q);
3433	if (err) {
3434		dev_err(dev, "dpni_set_queue() failed (%d)\n", err);
3435		return err;
3436	}
3437
3438	return 0;
3439}
3440
3441/* Supported header fields for Rx hash distribution key */
3442static const struct dpaa2_eth_dist_fields dist_fields[] = {
3443	{
3444		/* L2 header */
3445		.rxnfc_field = RXH_L2DA,
3446		.cls_prot = NET_PROT_ETH,
3447		.cls_field = NH_FLD_ETH_DA,
3448		.id = DPAA2_ETH_DIST_ETHDST,
3449		.size = 6,
3450	}, {
3451		.cls_prot = NET_PROT_ETH,
3452		.cls_field = NH_FLD_ETH_SA,
3453		.id = DPAA2_ETH_DIST_ETHSRC,
3454		.size = 6,
3455	}, {
3456		/* This is the last ethertype field parsed:
3457		 * depending on frame format, it can be the MAC ethertype
3458		 * or the VLAN etype.
3459		 */
3460		.cls_prot = NET_PROT_ETH,
3461		.cls_field = NH_FLD_ETH_TYPE,
3462		.id = DPAA2_ETH_DIST_ETHTYPE,
3463		.size = 2,
3464	}, {
3465		/* VLAN header */
3466		.rxnfc_field = RXH_VLAN,
3467		.cls_prot = NET_PROT_VLAN,
3468		.cls_field = NH_FLD_VLAN_TCI,
3469		.id = DPAA2_ETH_DIST_VLAN,
3470		.size = 2,
3471	}, {
3472		/* IP header */
3473		.rxnfc_field = RXH_IP_SRC,
3474		.cls_prot = NET_PROT_IP,
3475		.cls_field = NH_FLD_IP_SRC,
3476		.id = DPAA2_ETH_DIST_IPSRC,
3477		.size = 4,
3478	}, {
3479		.rxnfc_field = RXH_IP_DST,
3480		.cls_prot = NET_PROT_IP,
3481		.cls_field = NH_FLD_IP_DST,
3482		.id = DPAA2_ETH_DIST_IPDST,
3483		.size = 4,
3484	}, {
3485		.rxnfc_field = RXH_L3_PROTO,
3486		.cls_prot = NET_PROT_IP,
3487		.cls_field = NH_FLD_IP_PROTO,
3488		.id = DPAA2_ETH_DIST_IPPROTO,
3489		.size = 1,
3490	}, {
3491		/* Using UDP ports, this is functionally equivalent to raw
3492		 * byte pairs from L4 header.
3493		 */
3494		.rxnfc_field = RXH_L4_B_0_1,
3495		.cls_prot = NET_PROT_UDP,
3496		.cls_field = NH_FLD_UDP_PORT_SRC,
3497		.id = DPAA2_ETH_DIST_L4SRC,
3498		.size = 2,
3499	}, {
3500		.rxnfc_field = RXH_L4_B_2_3,
3501		.cls_prot = NET_PROT_UDP,
3502		.cls_field = NH_FLD_UDP_PORT_DST,
3503		.id = DPAA2_ETH_DIST_L4DST,
3504		.size = 2,
3505	},
3506};
3507
3508/* Configure the Rx hash key using the legacy API */
3509static int dpaa2_eth_config_legacy_hash_key(struct dpaa2_eth_priv *priv, dma_addr_t key)
3510{
3511	struct device *dev = priv->net_dev->dev.parent;
3512	struct dpni_rx_tc_dist_cfg dist_cfg;
3513	int i, err = 0;
3514
3515	memset(&dist_cfg, 0, sizeof(dist_cfg));
3516
3517	dist_cfg.key_cfg_iova = key;
3518	dist_cfg.dist_size = dpaa2_eth_queue_count(priv);
3519	dist_cfg.dist_mode = DPNI_DIST_MODE_HASH;
3520
3521	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3522		err = dpni_set_rx_tc_dist(priv->mc_io, 0, priv->mc_token,
3523					  i, &dist_cfg);
3524		if (err) {
3525			dev_err(dev, "dpni_set_rx_tc_dist failed\n");
3526			break;
3527		}
3528	}
3529
3530	return err;
3531}
3532
3533/* Configure the Rx hash key using the new API */
3534static int dpaa2_eth_config_hash_key(struct dpaa2_eth_priv *priv, dma_addr_t key)
3535{
3536	struct device *dev = priv->net_dev->dev.parent;
3537	struct dpni_rx_dist_cfg dist_cfg;
3538	int i, err = 0;
3539
3540	memset(&dist_cfg, 0, sizeof(dist_cfg));
3541
3542	dist_cfg.key_cfg_iova = key;
3543	dist_cfg.dist_size = dpaa2_eth_queue_count(priv);
3544	dist_cfg.enable = 1;
3545
3546	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3547		dist_cfg.tc = i;
3548		err = dpni_set_rx_hash_dist(priv->mc_io, 0, priv->mc_token,
3549					    &dist_cfg);
3550		if (err) {
3551			dev_err(dev, "dpni_set_rx_hash_dist failed\n");
3552			break;
3553		}
3554
3555		/* If the flow steering / hashing key is shared between all
3556		 * traffic classes, install it just once
3557		 */
3558		if (priv->dpni_attrs.options & DPNI_OPT_SHARED_FS)
3559			break;
3560	}
3561
3562	return err;
3563}
3564
3565/* Configure the Rx flow classification key */
3566static int dpaa2_eth_config_cls_key(struct dpaa2_eth_priv *priv, dma_addr_t key)
3567{
3568	struct device *dev = priv->net_dev->dev.parent;
3569	struct dpni_rx_dist_cfg dist_cfg;
3570	int i, err = 0;
3571
3572	memset(&dist_cfg, 0, sizeof(dist_cfg));
3573
3574	dist_cfg.key_cfg_iova = key;
3575	dist_cfg.dist_size = dpaa2_eth_queue_count(priv);
3576	dist_cfg.enable = 1;
3577
3578	for (i = 0; i < dpaa2_eth_tc_count(priv); i++) {
3579		dist_cfg.tc = i;
3580		err = dpni_set_rx_fs_dist(priv->mc_io, 0, priv->mc_token,
3581					  &dist_cfg);
3582		if (err) {
3583			dev_err(dev, "dpni_set_rx_fs_dist failed\n");
3584			break;
3585		}
3586
3587		/* If the flow steering / hashing key is shared between all
3588		 * traffic classes, install it just once
3589		 */
3590		if (priv->dpni_attrs.options & DPNI_OPT_SHARED_FS)
3591			break;
3592	}
3593
3594	return err;
3595}
3596
3597/* Size of the Rx flow classification key */
3598int dpaa2_eth_cls_key_size(u64 fields)
3599{
3600	int i, size = 0;
3601
3602	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3603		if (!(fields & dist_fields[i].id))
3604			continue;
3605		size += dist_fields[i].size;
3606	}
3607
3608	return size;
3609}
3610
3611/* Offset of header field in Rx classification key */
3612int dpaa2_eth_cls_fld_off(int prot, int field)
3613{
3614	int i, off = 0;
3615
3616	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3617		if (dist_fields[i].cls_prot == prot &&
3618		    dist_fields[i].cls_field == field)
3619			return off;
3620		off += dist_fields[i].size;
3621	}
3622
3623	WARN_ONCE(1, "Unsupported header field used for Rx flow cls\n");
3624	return 0;
3625}
3626
3627/* Prune unused fields from the classification rule.
3628 * Used when masking is not supported
3629 */
3630void dpaa2_eth_cls_trim_rule(void *key_mem, u64 fields)
3631{
3632	int off = 0, new_off = 0;
3633	int i, size;
3634
3635	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3636		size = dist_fields[i].size;
3637		if (dist_fields[i].id & fields) {
3638			memcpy(key_mem + new_off, key_mem + off, size);
3639			new_off += size;
3640		}
3641		off += size;
3642	}
3643}
3644
3645/* Set Rx distribution (hash or flow classification) key
3646 * flags is a combination of RXH_ bits
3647 */
3648static int dpaa2_eth_set_dist_key(struct net_device *net_dev,
3649				  enum dpaa2_eth_rx_dist type, u64 flags)
3650{
3651	struct device *dev = net_dev->dev.parent;
3652	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
3653	struct dpkg_profile_cfg cls_cfg;
3654	u32 rx_hash_fields = 0;
3655	dma_addr_t key_iova;
3656	u8 *dma_mem;
3657	int i;
3658	int err = 0;
3659
3660	memset(&cls_cfg, 0, sizeof(cls_cfg));
3661
3662	for (i = 0; i < ARRAY_SIZE(dist_fields); i++) {
3663		struct dpkg_extract *key =
3664			&cls_cfg.extracts[cls_cfg.num_extracts];
3665
3666		/* For both Rx hashing and classification keys
3667		 * we set only the selected fields.
3668		 */
3669		if (!(flags & dist_fields[i].id))
3670			continue;
3671		if (type == DPAA2_ETH_RX_DIST_HASH)
3672			rx_hash_fields |= dist_fields[i].rxnfc_field;
3673
3674		if (cls_cfg.num_extracts >= DPKG_MAX_NUM_OF_EXTRACTS) {
3675			dev_err(dev, "error adding key extraction rule, too many rules?\n");
3676			return -E2BIG;
3677		}
3678
3679		key->type = DPKG_EXTRACT_FROM_HDR;
3680		key->extract.from_hdr.prot = dist_fields[i].cls_prot;
3681		key->extract.from_hdr.type = DPKG_FULL_FIELD;
3682		key->extract.from_hdr.field = dist_fields[i].cls_field;
3683		cls_cfg.num_extracts++;
3684	}
3685
3686	dma_mem = kzalloc(DPAA2_CLASSIFIER_DMA_SIZE, GFP_KERNEL);
3687	if (!dma_mem)
3688		return -ENOMEM;
3689
3690	err = dpni_prepare_key_cfg(&cls_cfg, dma_mem);
3691	if (err) {
3692		dev_err(dev, "dpni_prepare_key_cfg error %d\n", err);
3693		goto free_key;
3694	}
3695
3696	/* Prepare for setting the rx dist */
3697	key_iova = dma_map_single(dev, dma_mem, DPAA2_CLASSIFIER_DMA_SIZE,
3698				  DMA_TO_DEVICE);
3699	if (dma_mapping_error(dev, key_iova)) {
3700		dev_err(dev, "DMA mapping failed\n");
3701		err = -ENOMEM;
3702		goto free_key;
3703	}
3704
3705	if (type == DPAA2_ETH_RX_DIST_HASH) {
3706		if (dpaa2_eth_has_legacy_dist(priv))
3707			err = dpaa2_eth_config_legacy_hash_key(priv, key_iova);
3708		else
3709			err = dpaa2_eth_config_hash_key(priv, key_iova);
3710	} else {
3711		err = dpaa2_eth_config_cls_key(priv, key_iova);
3712	}
3713
3714	dma_unmap_single(dev, key_iova, DPAA2_CLASSIFIER_DMA_SIZE,
3715			 DMA_TO_DEVICE);
3716	if (!err && type == DPAA2_ETH_RX_DIST_HASH)
3717		priv->rx_hash_fields = rx_hash_fields;
3718
3719free_key:
3720	kfree(dma_mem);
3721	return err;
3722}
3723
3724int dpaa2_eth_set_hash(struct net_device *net_dev, u64 flags)
3725{
3726	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
3727	u64 key = 0;
3728	int i;
3729
3730	if (!dpaa2_eth_hash_enabled(priv))
3731		return -EOPNOTSUPP;
3732
3733	for (i = 0; i < ARRAY_SIZE(dist_fields); i++)
3734		if (dist_fields[i].rxnfc_field & flags)
3735			key |= dist_fields[i].id;
3736
3737	return dpaa2_eth_set_dist_key(net_dev, DPAA2_ETH_RX_DIST_HASH, key);
3738}
3739
3740int dpaa2_eth_set_cls(struct net_device *net_dev, u64 flags)
3741{
3742	return dpaa2_eth_set_dist_key(net_dev, DPAA2_ETH_RX_DIST_CLS, flags);
3743}
3744
3745static int dpaa2_eth_set_default_cls(struct dpaa2_eth_priv *priv)
3746{
3747	struct device *dev = priv->net_dev->dev.parent;
3748	int err;
3749
3750	/* Check if we actually support Rx flow classification */
3751	if (dpaa2_eth_has_legacy_dist(priv)) {
3752		dev_dbg(dev, "Rx cls not supported by current MC version\n");
3753		return -EOPNOTSUPP;
3754	}
3755
3756	if (!dpaa2_eth_fs_enabled(priv)) {
3757		dev_dbg(dev, "Rx cls disabled in DPNI options\n");
3758		return -EOPNOTSUPP;
3759	}
3760
3761	if (!dpaa2_eth_hash_enabled(priv)) {
3762		dev_dbg(dev, "Rx cls disabled for single queue DPNIs\n");
3763		return -EOPNOTSUPP;
3764	}
3765
3766	/* If there is no support for masking in the classification table,
3767	 * we don't set a default key, as it will depend on the rules
3768	 * added by the user at runtime.
3769	 */
3770	if (!dpaa2_eth_fs_mask_enabled(priv))
3771		goto out;
3772
3773	err = dpaa2_eth_set_cls(priv->net_dev, DPAA2_ETH_DIST_ALL);
3774	if (err)
3775		return err;
3776
3777out:
3778	priv->rx_cls_enabled = 1;
3779
3780	return 0;
3781}
3782
3783/* Bind the DPNI to its needed objects and resources: buffer pool, DPIOs,
3784 * frame queues and channels
3785 */
3786static int dpaa2_eth_bind_dpni(struct dpaa2_eth_priv *priv)
3787{
3788	struct net_device *net_dev = priv->net_dev;
3789	struct device *dev = net_dev->dev.parent;
3790	struct dpni_pools_cfg pools_params;
3791	struct dpni_error_cfg err_cfg;
3792	int err = 0;
3793	int i;
3794
3795	pools_params.num_dpbp = 1;
3796	pools_params.pools[0].dpbp_id = priv->dpbp_dev->obj_desc.id;
3797	pools_params.pools[0].backup_pool = 0;
3798	pools_params.pools[0].buffer_size = priv->rx_buf_size;
3799	err = dpni_set_pools(priv->mc_io, 0, priv->mc_token, &pools_params);
3800	if (err) {
3801		dev_err(dev, "dpni_set_pools() failed\n");
3802		return err;
3803	}
3804
3805	/* have the interface implicitly distribute traffic based on
3806	 * the default hash key
3807	 */
3808	err = dpaa2_eth_set_hash(net_dev, DPAA2_RXH_DEFAULT);
3809	if (err && err != -EOPNOTSUPP)
3810		dev_err(dev, "Failed to configure hashing\n");
3811
3812	/* Configure the flow classification key; it includes all
3813	 * supported header fields and cannot be modified at runtime
3814	 */
3815	err = dpaa2_eth_set_default_cls(priv);
3816	if (err && err != -EOPNOTSUPP)
3817		dev_err(dev, "Failed to configure Rx classification key\n");
3818
3819	/* Configure handling of error frames */
3820	err_cfg.errors = DPAA2_FAS_RX_ERR_MASK;
3821	err_cfg.set_frame_annotation = 1;
3822	err_cfg.error_action = DPNI_ERROR_ACTION_DISCARD;
3823	err = dpni_set_errors_behavior(priv->mc_io, 0, priv->mc_token,
3824				       &err_cfg);
3825	if (err) {
3826		dev_err(dev, "dpni_set_errors_behavior failed\n");
3827		return err;
3828	}
3829
3830	/* Configure Rx and Tx conf queues to generate CDANs */
3831	for (i = 0; i < priv->num_fqs; i++) {
3832		switch (priv->fq[i].type) {
3833		case DPAA2_RX_FQ:
3834			err = dpaa2_eth_setup_rx_flow(priv, &priv->fq[i]);
3835			break;
3836		case DPAA2_TX_CONF_FQ:
3837			err = dpaa2_eth_setup_tx_flow(priv, &priv->fq[i]);
3838			break;
3839		case DPAA2_RX_ERR_FQ:
3840			err = setup_rx_err_flow(priv, &priv->fq[i]);
3841			break;
3842		default:
3843			dev_err(dev, "Invalid FQ type %d\n", priv->fq[i].type);
3844			return -EINVAL;
3845		}
3846		if (err)
3847			return err;
3848	}
3849
3850	err = dpni_get_qdid(priv->mc_io, 0, priv->mc_token,
3851			    DPNI_QUEUE_TX, &priv->tx_qdid);
3852	if (err) {
3853		dev_err(dev, "dpni_get_qdid() failed\n");
3854		return err;
3855	}
3856
3857	return 0;
3858}
3859
3860/* Allocate rings for storing incoming frame descriptors */
3861static int dpaa2_eth_alloc_rings(struct dpaa2_eth_priv *priv)
3862{
3863	struct net_device *net_dev = priv->net_dev;
3864	struct device *dev = net_dev->dev.parent;
3865	int i;
3866
3867	for (i = 0; i < priv->num_channels; i++) {
3868		priv->channel[i]->store =
3869			dpaa2_io_store_create(DPAA2_ETH_STORE_SIZE, dev);
3870		if (!priv->channel[i]->store) {
3871			netdev_err(net_dev, "dpaa2_io_store_create() failed\n");
3872			goto err_ring;
3873		}
3874	}
3875
3876	return 0;
3877
3878err_ring:
3879	for (i = 0; i < priv->num_channels; i++) {
3880		if (!priv->channel[i]->store)
3881			break;
3882		dpaa2_io_store_destroy(priv->channel[i]->store);
3883	}
3884
3885	return -ENOMEM;
3886}
3887
3888static void dpaa2_eth_free_rings(struct dpaa2_eth_priv *priv)
3889{
3890	int i;
3891
3892	for (i = 0; i < priv->num_channels; i++)
3893		dpaa2_io_store_destroy(priv->channel[i]->store);
3894}
3895
3896static int dpaa2_eth_set_mac_addr(struct dpaa2_eth_priv *priv)
3897{
3898	struct net_device *net_dev = priv->net_dev;
3899	struct device *dev = net_dev->dev.parent;
3900	u8 mac_addr[ETH_ALEN], dpni_mac_addr[ETH_ALEN];
3901	int err;
3902
3903	/* Get firmware address, if any */
3904	err = dpni_get_port_mac_addr(priv->mc_io, 0, priv->mc_token, mac_addr);
3905	if (err) {
3906		dev_err(dev, "dpni_get_port_mac_addr() failed\n");
3907		return err;
3908	}
3909
3910	/* Get DPNI attributes address, if any */
3911	err = dpni_get_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
3912					dpni_mac_addr);
3913	if (err) {
3914		dev_err(dev, "dpni_get_primary_mac_addr() failed\n");
3915		return err;
3916	}
3917
3918	/* First check if firmware has any address configured by bootloader */
3919	if (!is_zero_ether_addr(mac_addr)) {
3920		/* If the DPMAC addr != DPNI addr, update it */
3921		if (!ether_addr_equal(mac_addr, dpni_mac_addr)) {
3922			err = dpni_set_primary_mac_addr(priv->mc_io, 0,
3923							priv->mc_token,
3924							mac_addr);
3925			if (err) {
3926				dev_err(dev, "dpni_set_primary_mac_addr() failed\n");
3927				return err;
3928			}
3929		}
3930		memcpy(net_dev->dev_addr, mac_addr, net_dev->addr_len);
3931	} else if (is_zero_ether_addr(dpni_mac_addr)) {
3932		/* No MAC address configured, fill in net_dev->dev_addr
3933		 * with a random one
3934		 */
3935		eth_hw_addr_random(net_dev);
3936		dev_dbg_once(dev, "device(s) have all-zero hwaddr, replaced with random\n");
3937
3938		err = dpni_set_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
3939						net_dev->dev_addr);
3940		if (err) {
3941			dev_err(dev, "dpni_set_primary_mac_addr() failed\n");
3942			return err;
3943		}
3944
3945		/* Override NET_ADDR_RANDOM set by eth_hw_addr_random(); for all
3946		 * practical purposes, this will be our "permanent" mac address,
3947		 * at least until the next reboot. This move will also permit
3948		 * register_netdevice() to properly fill up net_dev->perm_addr.
3949		 */
3950		net_dev->addr_assign_type = NET_ADDR_PERM;
3951	} else {
3952		/* NET_ADDR_PERM is default, all we have to do is
3953		 * fill in the device addr.
3954		 */
3955		memcpy(net_dev->dev_addr, dpni_mac_addr, net_dev->addr_len);
3956	}
3957
3958	return 0;
3959}
3960
3961static int dpaa2_eth_netdev_init(struct net_device *net_dev)
3962{
3963	struct device *dev = net_dev->dev.parent;
3964	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
3965	u32 options = priv->dpni_attrs.options;
3966	u64 supported = 0, not_supported = 0;
3967	u8 bcast_addr[ETH_ALEN];
3968	u8 num_queues;
3969	int err;
3970
3971	net_dev->netdev_ops = &dpaa2_eth_ops;
3972	net_dev->ethtool_ops = &dpaa2_ethtool_ops;
3973
3974	err = dpaa2_eth_set_mac_addr(priv);
3975	if (err)
3976		return err;
3977
3978	/* Explicitly add the broadcast address to the MAC filtering table */
3979	eth_broadcast_addr(bcast_addr);
3980	err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token, bcast_addr);
3981	if (err) {
3982		dev_err(dev, "dpni_add_mac_addr() failed\n");
3983		return err;
3984	}
3985
3986	/* Set MTU upper limit; lower limit is 68B (default value) */
3987	net_dev->max_mtu = DPAA2_ETH_MAX_MTU;
3988	err = dpni_set_max_frame_length(priv->mc_io, 0, priv->mc_token,
3989					DPAA2_ETH_MFL);
3990	if (err) {
3991		dev_err(dev, "dpni_set_max_frame_length() failed\n");
3992		return err;
3993	}
3994
3995	/* Set actual number of queues in the net device */
3996	num_queues = dpaa2_eth_queue_count(priv);
3997	err = netif_set_real_num_tx_queues(net_dev, num_queues);
3998	if (err) {
3999		dev_err(dev, "netif_set_real_num_tx_queues() failed\n");
4000		return err;
4001	}
4002	err = netif_set_real_num_rx_queues(net_dev, num_queues);
4003	if (err) {
4004		dev_err(dev, "netif_set_real_num_rx_queues() failed\n");
4005		return err;
4006	}
4007
4008	/* Capabilities listing */
4009	supported |= IFF_LIVE_ADDR_CHANGE;
4010
4011	if (options & DPNI_OPT_NO_MAC_FILTER)
4012		not_supported |= IFF_UNICAST_FLT;
4013	else
4014		supported |= IFF_UNICAST_FLT;
4015
4016	net_dev->priv_flags |= supported;
4017	net_dev->priv_flags &= ~not_supported;
4018
4019	/* Features */
4020	net_dev->features = NETIF_F_RXCSUM |
4021			    NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4022			    NETIF_F_SG | NETIF_F_HIGHDMA |
4023			    NETIF_F_LLTX | NETIF_F_HW_TC;
4024	net_dev->hw_features = net_dev->features;
4025
4026	return 0;
4027}
4028
4029static int dpaa2_eth_poll_link_state(void *arg)
4030{
4031	struct dpaa2_eth_priv *priv = (struct dpaa2_eth_priv *)arg;
4032	int err;
4033
4034	while (!kthread_should_stop()) {
4035		err = dpaa2_eth_link_state_update(priv);
4036		if (unlikely(err))
4037			return err;
4038
4039		msleep(DPAA2_ETH_LINK_STATE_REFRESH);
4040	}
4041
4042	return 0;
4043}
4044
4045static int dpaa2_eth_connect_mac(struct dpaa2_eth_priv *priv)
4046{
4047	struct fsl_mc_device *dpni_dev, *dpmac_dev;
4048	struct dpaa2_mac *mac;
4049	int err;
4050
4051	dpni_dev = to_fsl_mc_device(priv->net_dev->dev.parent);
4052	dpmac_dev = fsl_mc_get_endpoint(dpni_dev);
4053	if (IS_ERR_OR_NULL(dpmac_dev) || dpmac_dev->dev.type != &fsl_mc_bus_dpmac_type)
4054		return 0;
4055
4056	if (dpaa2_mac_is_type_fixed(dpmac_dev, priv->mc_io))
4057		return 0;
4058
4059	mac = kzalloc(sizeof(struct dpaa2_mac), GFP_KERNEL);
4060	if (!mac)
4061		return -ENOMEM;
4062
4063	mac->mc_dev = dpmac_dev;
4064	mac->mc_io = priv->mc_io;
4065	mac->net_dev = priv->net_dev;
4066
4067	err = dpaa2_mac_connect(mac);
4068	if (err) {
4069		netdev_err(priv->net_dev, "Error connecting to the MAC endpoint\n");
4070		kfree(mac);
4071		return err;
4072	}
4073	priv->mac = mac;
4074
4075	return 0;
4076}
4077
4078static void dpaa2_eth_disconnect_mac(struct dpaa2_eth_priv *priv)
4079{
4080	if (!priv->mac)
4081		return;
4082
4083	dpaa2_mac_disconnect(priv->mac);
4084	kfree(priv->mac);
4085	priv->mac = NULL;
4086}
4087
4088static irqreturn_t dpni_irq0_handler_thread(int irq_num, void *arg)
4089{
4090	u32 status = ~0;
4091	struct device *dev = (struct device *)arg;
4092	struct fsl_mc_device *dpni_dev = to_fsl_mc_device(dev);
4093	struct net_device *net_dev = dev_get_drvdata(dev);
4094	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
4095	int err;
4096
4097	err = dpni_get_irq_status(dpni_dev->mc_io, 0, dpni_dev->mc_handle,
4098				  DPNI_IRQ_INDEX, &status);
4099	if (unlikely(err)) {
4100		netdev_err(net_dev, "Can't get irq status (err %d)\n", err);
4101		return IRQ_HANDLED;
4102	}
4103
4104	if (status & DPNI_IRQ_EVENT_LINK_CHANGED)
4105		dpaa2_eth_link_state_update(netdev_priv(net_dev));
4106
4107	if (status & DPNI_IRQ_EVENT_ENDPOINT_CHANGED) {
4108		dpaa2_eth_set_mac_addr(netdev_priv(net_dev));
4109		dpaa2_eth_update_tx_fqids(priv);
4110
4111		rtnl_lock();
4112		if (priv->mac)
4113			dpaa2_eth_disconnect_mac(priv);
4114		else
4115			dpaa2_eth_connect_mac(priv);
4116		rtnl_unlock();
4117	}
4118
4119	return IRQ_HANDLED;
4120}
4121
4122static int dpaa2_eth_setup_irqs(struct fsl_mc_device *ls_dev)
4123{
4124	int err = 0;
4125	struct fsl_mc_device_irq *irq;
4126
4127	err = fsl_mc_allocate_irqs(ls_dev);
4128	if (err) {
4129		dev_err(&ls_dev->dev, "MC irqs allocation failed\n");
4130		return err;
4131	}
4132
4133	irq = ls_dev->irqs[0];
4134	err = devm_request_threaded_irq(&ls_dev->dev, irq->msi_desc->irq,
4135					NULL, dpni_irq0_handler_thread,
4136					IRQF_NO_SUSPEND | IRQF_ONESHOT,
4137					dev_name(&ls_dev->dev), &ls_dev->dev);
4138	if (err < 0) {
4139		dev_err(&ls_dev->dev, "devm_request_threaded_irq(): %d\n", err);
4140		goto free_mc_irq;
4141	}
4142
4143	err = dpni_set_irq_mask(ls_dev->mc_io, 0, ls_dev->mc_handle,
4144				DPNI_IRQ_INDEX, DPNI_IRQ_EVENT_LINK_CHANGED |
4145				DPNI_IRQ_EVENT_ENDPOINT_CHANGED);
4146	if (err < 0) {
4147		dev_err(&ls_dev->dev, "dpni_set_irq_mask(): %d\n", err);
4148		goto free_irq;
4149	}
4150
4151	err = dpni_set_irq_enable(ls_dev->mc_io, 0, ls_dev->mc_handle,
4152				  DPNI_IRQ_INDEX, 1);
4153	if (err < 0) {
4154		dev_err(&ls_dev->dev, "dpni_set_irq_enable(): %d\n", err);
4155		goto free_irq;
4156	}
4157
4158	return 0;
4159
4160free_irq:
4161	devm_free_irq(&ls_dev->dev, irq->msi_desc->irq, &ls_dev->dev);
4162free_mc_irq:
4163	fsl_mc_free_irqs(ls_dev);
4164
4165	return err;
4166}
4167
4168static void dpaa2_eth_add_ch_napi(struct dpaa2_eth_priv *priv)
4169{
4170	int i;
4171	struct dpaa2_eth_channel *ch;
4172
4173	for (i = 0; i < priv->num_channels; i++) {
4174		ch = priv->channel[i];
4175		/* NAPI weight *MUST* be a multiple of DPAA2_ETH_STORE_SIZE */
4176		netif_napi_add(priv->net_dev, &ch->napi, dpaa2_eth_poll,
4177			       NAPI_POLL_WEIGHT);
4178	}
4179}
4180
4181static void dpaa2_eth_del_ch_napi(struct dpaa2_eth_priv *priv)
4182{
4183	int i;
4184	struct dpaa2_eth_channel *ch;
4185
4186	for (i = 0; i < priv->num_channels; i++) {
4187		ch = priv->channel[i];
4188		netif_napi_del(&ch->napi);
4189	}
4190}
4191
4192static int dpaa2_eth_probe(struct fsl_mc_device *dpni_dev)
4193{
4194	struct device *dev;
4195	struct net_device *net_dev = NULL;
4196	struct dpaa2_eth_priv *priv = NULL;
4197	int err = 0;
4198
4199	dev = &dpni_dev->dev;
4200
4201	/* Net device */
4202	net_dev = alloc_etherdev_mq(sizeof(*priv), DPAA2_ETH_MAX_NETDEV_QUEUES);
4203	if (!net_dev) {
4204		dev_err(dev, "alloc_etherdev_mq() failed\n");
4205		return -ENOMEM;
4206	}
4207
4208	SET_NETDEV_DEV(net_dev, dev);
4209	dev_set_drvdata(dev, net_dev);
4210
4211	priv = netdev_priv(net_dev);
4212	priv->net_dev = net_dev;
4213
4214	priv->iommu_domain = iommu_get_domain_for_dev(dev);
4215
4216	priv->tx_tstamp_type = HWTSTAMP_TX_OFF;
4217	priv->rx_tstamp = false;
4218
4219	priv->dpaa2_ptp_wq = alloc_workqueue("dpaa2_ptp_wq", 0, 0);
4220	if (!priv->dpaa2_ptp_wq) {
4221		err = -ENOMEM;
4222		goto err_wq_alloc;
4223	}
4224
4225	INIT_WORK(&priv->tx_onestep_tstamp, dpaa2_eth_tx_onestep_tstamp);
4226	mutex_init(&priv->onestep_tstamp_lock);
4227	skb_queue_head_init(&priv->tx_skbs);
4228
4229	/* Obtain a MC portal */
4230	err = fsl_mc_portal_allocate(dpni_dev, FSL_MC_IO_ATOMIC_CONTEXT_PORTAL,
4231				     &priv->mc_io);
4232	if (err) {
4233		if (err == -ENXIO)
4234			err = -EPROBE_DEFER;
4235		else
4236			dev_err(dev, "MC portal allocation failed\n");
4237		goto err_portal_alloc;
4238	}
4239
4240	/* MC objects initialization and configuration */
4241	err = dpaa2_eth_setup_dpni(dpni_dev);
4242	if (err)
4243		goto err_dpni_setup;
4244
4245	err = dpaa2_eth_setup_dpio(priv);
4246	if (err)
4247		goto err_dpio_setup;
4248
4249	dpaa2_eth_setup_fqs(priv);
4250
4251	err = dpaa2_eth_setup_dpbp(priv);
4252	if (err)
4253		goto err_dpbp_setup;
4254
4255	err = dpaa2_eth_bind_dpni(priv);
4256	if (err)
4257		goto err_bind;
4258
4259	/* Add a NAPI context for each channel */
4260	dpaa2_eth_add_ch_napi(priv);
4261
4262	/* Percpu statistics */
4263	priv->percpu_stats = alloc_percpu(*priv->percpu_stats);
4264	if (!priv->percpu_stats) {
4265		dev_err(dev, "alloc_percpu(percpu_stats) failed\n");
4266		err = -ENOMEM;
4267		goto err_alloc_percpu_stats;
4268	}
4269	priv->percpu_extras = alloc_percpu(*priv->percpu_extras);
4270	if (!priv->percpu_extras) {
4271		dev_err(dev, "alloc_percpu(percpu_extras) failed\n");
4272		err = -ENOMEM;
4273		goto err_alloc_percpu_extras;
4274	}
4275
4276	priv->sgt_cache = alloc_percpu(*priv->sgt_cache);
4277	if (!priv->sgt_cache) {
4278		dev_err(dev, "alloc_percpu(sgt_cache) failed\n");
4279		err = -ENOMEM;
4280		goto err_alloc_sgt_cache;
4281	}
4282
4283	err = dpaa2_eth_netdev_init(net_dev);
4284	if (err)
4285		goto err_netdev_init;
4286
4287	/* Configure checksum offload based on current interface flags */
4288	err = dpaa2_eth_set_rx_csum(priv, !!(net_dev->features & NETIF_F_RXCSUM));
4289	if (err)
4290		goto err_csum;
4291
4292	err = dpaa2_eth_set_tx_csum(priv,
4293				    !!(net_dev->features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)));
4294	if (err)
4295		goto err_csum;
4296
4297	err = dpaa2_eth_alloc_rings(priv);
4298	if (err)
4299		goto err_alloc_rings;
4300
4301#ifdef CONFIG_FSL_DPAA2_ETH_DCB
4302	if (dpaa2_eth_has_pause_support(priv) && priv->vlan_cls_enabled) {
4303		priv->dcbx_mode = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_IEEE;
4304		net_dev->dcbnl_ops = &dpaa2_eth_dcbnl_ops;
4305	} else {
4306		dev_dbg(dev, "PFC not supported\n");
4307	}
4308#endif
4309
4310	err = dpaa2_eth_setup_irqs(dpni_dev);
4311	if (err) {
4312		netdev_warn(net_dev, "Failed to set link interrupt, fall back to polling\n");
4313		priv->poll_thread = kthread_run(dpaa2_eth_poll_link_state, priv,
4314						"%s_poll_link", net_dev->name);
4315		if (IS_ERR(priv->poll_thread)) {
4316			dev_err(dev, "Error starting polling thread\n");
4317			goto err_poll_thread;
4318		}
4319		priv->do_link_poll = true;
4320	}
4321
4322	err = dpaa2_eth_connect_mac(priv);
4323	if (err)
4324		goto err_connect_mac;
4325
4326	err = dpaa2_eth_dl_register(priv);
4327	if (err)
4328		goto err_dl_register;
4329
4330	err = dpaa2_eth_dl_traps_register(priv);
4331	if (err)
4332		goto err_dl_trap_register;
4333
4334	err = dpaa2_eth_dl_port_add(priv);
4335	if (err)
4336		goto err_dl_port_add;
4337
4338	net_dev->needed_headroom = DPAA2_ETH_SWA_SIZE + DPAA2_ETH_TX_BUF_ALIGN;
4339
4340	err = register_netdev(net_dev);
4341	if (err < 0) {
4342		dev_err(dev, "register_netdev() failed\n");
4343		goto err_netdev_reg;
4344	}
4345
4346#ifdef CONFIG_DEBUG_FS
4347	dpaa2_dbg_add(priv);
4348#endif
4349
4350	dev_info(dev, "Probed interface %s\n", net_dev->name);
4351	return 0;
4352
4353err_netdev_reg:
4354	dpaa2_eth_dl_port_del(priv);
4355err_dl_port_add:
4356	dpaa2_eth_dl_traps_unregister(priv);
4357err_dl_trap_register:
4358	dpaa2_eth_dl_unregister(priv);
4359err_dl_register:
4360	dpaa2_eth_disconnect_mac(priv);
4361err_connect_mac:
4362	if (priv->do_link_poll)
4363		kthread_stop(priv->poll_thread);
4364	else
4365		fsl_mc_free_irqs(dpni_dev);
4366err_poll_thread:
4367	dpaa2_eth_free_rings(priv);
4368err_alloc_rings:
4369err_csum:
4370err_netdev_init:
4371	free_percpu(priv->sgt_cache);
4372err_alloc_sgt_cache:
4373	free_percpu(priv->percpu_extras);
4374err_alloc_percpu_extras:
4375	free_percpu(priv->percpu_stats);
4376err_alloc_percpu_stats:
4377	dpaa2_eth_del_ch_napi(priv);
4378err_bind:
4379	dpaa2_eth_free_dpbp(priv);
4380err_dpbp_setup:
4381	dpaa2_eth_free_dpio(priv);
4382err_dpio_setup:
4383	dpaa2_eth_free_dpni(priv);
4384err_dpni_setup:
4385	fsl_mc_portal_free(priv->mc_io);
4386err_portal_alloc:
4387	destroy_workqueue(priv->dpaa2_ptp_wq);
4388err_wq_alloc:
4389	dev_set_drvdata(dev, NULL);
4390	free_netdev(net_dev);
4391
4392	return err;
4393}
4394
4395static int dpaa2_eth_remove(struct fsl_mc_device *ls_dev)
4396{
4397	struct device *dev;
4398	struct net_device *net_dev;
4399	struct dpaa2_eth_priv *priv;
4400
4401	dev = &ls_dev->dev;
4402	net_dev = dev_get_drvdata(dev);
4403	priv = netdev_priv(net_dev);
4404
4405#ifdef CONFIG_DEBUG_FS
4406	dpaa2_dbg_remove(priv);
4407#endif
4408	rtnl_lock();
4409	dpaa2_eth_disconnect_mac(priv);
4410	rtnl_unlock();
4411
4412	unregister_netdev(net_dev);
4413
4414	dpaa2_eth_dl_port_del(priv);
4415	dpaa2_eth_dl_traps_unregister(priv);
4416	dpaa2_eth_dl_unregister(priv);
4417
4418	if (priv->do_link_poll)
4419		kthread_stop(priv->poll_thread);
4420	else
4421		fsl_mc_free_irqs(ls_dev);
4422
4423	dpaa2_eth_free_rings(priv);
4424	free_percpu(priv->sgt_cache);
4425	free_percpu(priv->percpu_stats);
4426	free_percpu(priv->percpu_extras);
4427
4428	dpaa2_eth_del_ch_napi(priv);
4429	dpaa2_eth_free_dpbp(priv);
4430	dpaa2_eth_free_dpio(priv);
4431	dpaa2_eth_free_dpni(priv);
4432
4433	fsl_mc_portal_free(priv->mc_io);
4434
4435	destroy_workqueue(priv->dpaa2_ptp_wq);
4436
4437	dev_dbg(net_dev->dev.parent, "Removed interface %s\n", net_dev->name);
4438
4439	free_netdev(net_dev);
4440
4441	return 0;
4442}
4443
4444static const struct fsl_mc_device_id dpaa2_eth_match_id_table[] = {
4445	{
4446		.vendor = FSL_MC_VENDOR_FREESCALE,
4447		.obj_type = "dpni",
4448	},
4449	{ .vendor = 0x0 }
4450};
4451MODULE_DEVICE_TABLE(fslmc, dpaa2_eth_match_id_table);
4452
4453static struct fsl_mc_driver dpaa2_eth_driver = {
4454	.driver = {
4455		.name = KBUILD_MODNAME,
4456		.owner = THIS_MODULE,
4457	},
4458	.probe = dpaa2_eth_probe,
4459	.remove = dpaa2_eth_remove,
4460	.match_id_table = dpaa2_eth_match_id_table
4461};
4462
4463static int __init dpaa2_eth_driver_init(void)
4464{
4465	int err;
4466
4467	dpaa2_eth_dbg_init();
4468	err = fsl_mc_driver_register(&dpaa2_eth_driver);
4469	if (err) {
4470		dpaa2_eth_dbg_exit();
4471		return err;
4472	}
4473
4474	return 0;
4475}
4476
4477static void __exit dpaa2_eth_driver_exit(void)
4478{
4479	dpaa2_eth_dbg_exit();
4480	fsl_mc_driver_unregister(&dpaa2_eth_driver);
4481}
4482
4483module_init(dpaa2_eth_driver_init);
4484module_exit(dpaa2_eth_driver_exit);
4485