1 /* Copyright 2008 - 2016 Freescale Semiconductor Inc.
2  *
3  * Redistribution and use in source and binary forms, with or without
4  * modification, are permitted provided that the following conditions are met:
5  *     * Redistributions of source code must retain the above copyright
6  *	 notice, this list of conditions and the following disclaimer.
7  *     * Redistributions in binary form must reproduce the above copyright
8  *	 notice, this list of conditions and the following disclaimer in the
9  *	 documentation and/or other materials provided with the distribution.
10  *     * Neither the name of Freescale Semiconductor nor the
11  *	 names of its contributors may be used to endorse or promote products
12  *	 derived from this software without specific prior written permission.
13  *
14  * ALTERNATIVELY, this software may be distributed under the terms of the
15  * GNU General Public License ("GPL") as published by the Free Software
16  * Foundation, either version 2 of that License or (at your option) any
17  * later version.
18  *
19  * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
20  * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
21  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
22  * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
23  * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
24  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
25  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
26  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
28  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
32 
33 #include <linux/init.h>
34 #include <linux/module.h>
35 #include <linux/of_platform.h>
36 #include <linux/of_mdio.h>
37 #include <linux/of_net.h>
38 #include <linux/io.h>
39 #include <linux/if_arp.h>
40 #include <linux/if_vlan.h>
41 #include <linux/icmp.h>
42 #include <linux/ip.h>
43 #include <linux/ipv6.h>
44 #include <linux/udp.h>
45 #include <linux/tcp.h>
46 #include <linux/net.h>
47 #include <linux/skbuff.h>
48 #include <linux/etherdevice.h>
49 #include <linux/if_ether.h>
50 #include <linux/highmem.h>
51 #include <linux/percpu.h>
52 #include <linux/dma-mapping.h>
53 #include <linux/sort.h>
54 #include <soc/fsl/bman.h>
55 #include <soc/fsl/qman.h>
56 
57 #include "fman.h"
58 #include "fman_port.h"
59 #include "mac.h"
60 #include "dpaa_eth.h"
61 
62 /* CREATE_TRACE_POINTS only needs to be defined once. Other dpaa files
63  * using trace events only need to #include <trace/events/sched.h>
64  */
65 #define CREATE_TRACE_POINTS
66 #include "dpaa_eth_trace.h"
67 
68 static int debug = -1;
69 module_param(debug, int, 0444);
70 MODULE_PARM_DESC(debug, "Module/Driver verbosity level (0=none,...,16=all)");
71 
72 static u16 tx_timeout = 1000;
73 module_param(tx_timeout, ushort, 0444);
74 MODULE_PARM_DESC(tx_timeout, "The Tx timeout in ms");
75 
76 #define FM_FD_STAT_RX_ERRORS						\
77 	(FM_FD_ERR_DMA | FM_FD_ERR_PHYSICAL	| \
78 	 FM_FD_ERR_SIZE | FM_FD_ERR_CLS_DISCARD | \
79 	 FM_FD_ERR_EXTRACTION | FM_FD_ERR_NO_SCHEME	| \
80 	 FM_FD_ERR_PRS_TIMEOUT | FM_FD_ERR_PRS_ILL_INSTRUCT | \
81 	 FM_FD_ERR_PRS_HDR_ERR)
82 
83 #define FM_FD_STAT_TX_ERRORS \
84 	(FM_FD_ERR_UNSUPPORTED_FORMAT | \
85 	 FM_FD_ERR_LENGTH | FM_FD_ERR_DMA)
86 
87 #define DPAA_MSG_DEFAULT (NETIF_MSG_DRV | NETIF_MSG_PROBE | \
88 			  NETIF_MSG_LINK | NETIF_MSG_IFUP | \
89 			  NETIF_MSG_IFDOWN)
90 
91 #define DPAA_INGRESS_CS_THRESHOLD 0x10000000
92 /* Ingress congestion threshold on FMan ports
93  * The size in bytes of the ingress tail-drop threshold on FMan ports.
94  * Traffic piling up above this value will be rejected by QMan and discarded
95  * by FMan.
96  */
97 
98 /* Size in bytes of the FQ taildrop threshold */
99 #define DPAA_FQ_TD 0x200000
100 
101 #define DPAA_CS_THRESHOLD_1G 0x06000000
102 /* Egress congestion threshold on 1G ports, range 0x1000 .. 0x10000000
103  * The size in bytes of the egress Congestion State notification threshold on
104  * 1G ports. The 1G dTSECs can quite easily be flooded by cores doing Tx in a
105  * tight loop (e.g. by sending UDP datagrams at "while(1) speed"),
106  * and the larger the frame size, the more acute the problem.
107  * So we have to find a balance between these factors:
108  * - avoiding the device staying congested for a prolonged time (risking
109  *   the netdev watchdog to fire - see also the tx_timeout module param);
110  * - affecting performance of protocols such as TCP, which otherwise
111  *   behave well under the congestion notification mechanism;
112  * - preventing the Tx cores from tightly-looping (as if the congestion
113  *   threshold was too low to be effective);
114  * - running out of memory if the CS threshold is set too high.
115  */
116 
117 #define DPAA_CS_THRESHOLD_10G 0x10000000
118 /* The size in bytes of the egress Congestion State notification threshold on
119  * 10G ports, range 0x1000 .. 0x10000000
120  */
121 
122 /* Largest value that the FQD's OAL field can hold */
123 #define FSL_QMAN_MAX_OAL	127
124 
125 /* Default alignment for start of data in an Rx FD */
126 #define DPAA_FD_DATA_ALIGNMENT  16
127 
128 /* The DPAA requires 256 bytes reserved and mapped for the SGT */
129 #define DPAA_SGT_SIZE 256
130 
131 /* Values for the L3R field of the FM Parse Results
132  */
133 /* L3 Type field: First IP Present IPv4 */
134 #define FM_L3_PARSE_RESULT_IPV4	0x8000
135 /* L3 Type field: First IP Present IPv6 */
136 #define FM_L3_PARSE_RESULT_IPV6	0x4000
137 /* Values for the L4R field of the FM Parse Results */
138 /* L4 Type field: UDP */
139 #define FM_L4_PARSE_RESULT_UDP	0x40
140 /* L4 Type field: TCP */
141 #define FM_L4_PARSE_RESULT_TCP	0x20
142 
143 /* FD status field indicating whether the FM Parser has attempted to validate
144  * the L4 csum of the frame.
145  * Note that having this bit set doesn't necessarily imply that the checksum
146  * is valid. One would have to check the parse results to find that out.
147  */
148 #define FM_FD_STAT_L4CV         0x00000004
149 
150 #define DPAA_SGT_MAX_ENTRIES 16 /* maximum number of entries in SG Table */
151 #define DPAA_BUFF_RELEASE_MAX 8 /* maximum number of buffers released at once */
152 
153 #define FSL_DPAA_BPID_INV		0xff
154 #define FSL_DPAA_ETH_MAX_BUF_COUNT	128
155 #define FSL_DPAA_ETH_REFILL_THRESHOLD	80
156 
157 #define DPAA_TX_PRIV_DATA_SIZE	16
158 #define DPAA_PARSE_RESULTS_SIZE sizeof(struct fman_prs_result)
159 #define DPAA_TIME_STAMP_SIZE 8
160 #define DPAA_HASH_RESULTS_SIZE 8
161 #define DPAA_RX_PRIV_DATA_SIZE	(u16)(DPAA_TX_PRIV_DATA_SIZE + \
162 					dpaa_rx_extra_headroom)
163 
164 #define DPAA_ETH_PCD_RXQ_NUM	128
165 
166 #define DPAA_ENQUEUE_RETRIES	100000
167 
168 enum port_type {RX, TX};
169 
170 struct fm_port_fqs {
171 	struct dpaa_fq *tx_defq;
172 	struct dpaa_fq *tx_errq;
173 	struct dpaa_fq *rx_defq;
174 	struct dpaa_fq *rx_errq;
175 	struct dpaa_fq *rx_pcdq;
176 };
177 
178 /* All the dpa bps in use at any moment */
179 static struct dpaa_bp *dpaa_bp_array[BM_MAX_NUM_OF_POOLS];
180 
181 /* The raw buffer size must be cacheline aligned */
182 #define DPAA_BP_RAW_SIZE 4096
183 /* When using more than one buffer pool, the raw sizes are as follows:
184  * 1 bp: 4KB
185  * 2 bp: 2KB, 4KB
186  * 3 bp: 1KB, 2KB, 4KB
187  * 4 bp: 1KB, 2KB, 4KB, 8KB
188  */
189 static inline size_t bpool_buffer_raw_size(u8 index, u8 cnt)
190 {
191 	size_t res = DPAA_BP_RAW_SIZE / 4;
192 	u8 i;
193 
194 	for (i = (cnt < 3) ? cnt : 3; i < 3 + index; i++)
195 		res *= 2;
196 	return res;
197 }
198 
199 /* FMan-DMA requires 16-byte alignment for Rx buffers, but SKB_DATA_ALIGN is
200  * even stronger (SMP_CACHE_BYTES-aligned), so we just get away with that,
201  * via SKB_WITH_OVERHEAD(). We can't rely on netdev_alloc_frag() giving us
202  * half-page-aligned buffers, so we reserve some more space for start-of-buffer
203  * alignment.
204  */
205 #define dpaa_bp_size(raw_size) SKB_WITH_OVERHEAD((raw_size) - SMP_CACHE_BYTES)
206 
207 static int dpaa_max_frm;
208 
209 static int dpaa_rx_extra_headroom;
210 
211 #define dpaa_get_max_mtu()	\
212 	(dpaa_max_frm - (VLAN_ETH_HLEN + ETH_FCS_LEN))
213 
214 static int dpaa_netdev_init(struct net_device *net_dev,
215 			    const struct net_device_ops *dpaa_ops,
216 			    u16 tx_timeout)
217 {
218 	struct dpaa_priv *priv = netdev_priv(net_dev);
219 	struct device *dev = net_dev->dev.parent;
220 	struct dpaa_percpu_priv *percpu_priv;
221 	const u8 *mac_addr;
222 	int i, err;
223 
224 	/* Although we access another CPU's private data here
225 	 * we do it at initialization so it is safe
226 	 */
227 	for_each_possible_cpu(i) {
228 		percpu_priv = per_cpu_ptr(priv->percpu_priv, i);
229 		percpu_priv->net_dev = net_dev;
230 	}
231 
232 	net_dev->netdev_ops = dpaa_ops;
233 	mac_addr = priv->mac_dev->addr;
234 
235 	net_dev->mem_start = priv->mac_dev->res->start;
236 	net_dev->mem_end = priv->mac_dev->res->end;
237 
238 	net_dev->min_mtu = ETH_MIN_MTU;
239 	net_dev->max_mtu = dpaa_get_max_mtu();
240 
241 	net_dev->hw_features |= (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
242 				 NETIF_F_LLTX | NETIF_F_RXHASH);
243 
244 	net_dev->hw_features |= NETIF_F_SG | NETIF_F_HIGHDMA;
245 	/* The kernels enables GSO automatically, if we declare NETIF_F_SG.
246 	 * For conformity, we'll still declare GSO explicitly.
247 	 */
248 	net_dev->features |= NETIF_F_GSO;
249 	net_dev->features |= NETIF_F_RXCSUM;
250 
251 	net_dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
252 	/* we do not want shared skbs on TX */
253 	net_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
254 
255 	net_dev->features |= net_dev->hw_features;
256 	net_dev->vlan_features = net_dev->features;
257 
258 	memcpy(net_dev->perm_addr, mac_addr, net_dev->addr_len);
259 	memcpy(net_dev->dev_addr, mac_addr, net_dev->addr_len);
260 
261 	net_dev->ethtool_ops = &dpaa_ethtool_ops;
262 
263 	net_dev->needed_headroom = priv->tx_headroom;
264 	net_dev->watchdog_timeo = msecs_to_jiffies(tx_timeout);
265 
266 	/* start without the RUNNING flag, phylib controls it later */
267 	netif_carrier_off(net_dev);
268 
269 	err = register_netdev(net_dev);
270 	if (err < 0) {
271 		dev_err(dev, "register_netdev() = %d\n", err);
272 		return err;
273 	}
274 
275 	return 0;
276 }
277 
278 static int dpaa_stop(struct net_device *net_dev)
279 {
280 	struct mac_device *mac_dev;
281 	struct dpaa_priv *priv;
282 	int i, err, error;
283 
284 	priv = netdev_priv(net_dev);
285 	mac_dev = priv->mac_dev;
286 
287 	netif_tx_stop_all_queues(net_dev);
288 	/* Allow the Fman (Tx) port to process in-flight frames before we
289 	 * try switching it off.
290 	 */
291 	usleep_range(5000, 10000);
292 
293 	err = mac_dev->stop(mac_dev);
294 	if (err < 0)
295 		netif_err(priv, ifdown, net_dev, "mac_dev->stop() = %d\n",
296 			  err);
297 
298 	for (i = 0; i < ARRAY_SIZE(mac_dev->port); i++) {
299 		error = fman_port_disable(mac_dev->port[i]);
300 		if (error)
301 			err = error;
302 	}
303 
304 	if (net_dev->phydev)
305 		phy_disconnect(net_dev->phydev);
306 	net_dev->phydev = NULL;
307 
308 	return err;
309 }
310 
311 static void dpaa_tx_timeout(struct net_device *net_dev)
312 {
313 	struct dpaa_percpu_priv *percpu_priv;
314 	const struct dpaa_priv	*priv;
315 
316 	priv = netdev_priv(net_dev);
317 	percpu_priv = this_cpu_ptr(priv->percpu_priv);
318 
319 	netif_crit(priv, timer, net_dev, "Transmit timeout latency: %u ms\n",
320 		   jiffies_to_msecs(jiffies - dev_trans_start(net_dev)));
321 
322 	percpu_priv->stats.tx_errors++;
323 }
324 
325 /* Calculates the statistics for the given device by adding the statistics
326  * collected by each CPU.
327  */
328 static void dpaa_get_stats64(struct net_device *net_dev,
329 			     struct rtnl_link_stats64 *s)
330 {
331 	int numstats = sizeof(struct rtnl_link_stats64) / sizeof(u64);
332 	struct dpaa_priv *priv = netdev_priv(net_dev);
333 	struct dpaa_percpu_priv *percpu_priv;
334 	u64 *netstats = (u64 *)s;
335 	u64 *cpustats;
336 	int i, j;
337 
338 	for_each_possible_cpu(i) {
339 		percpu_priv = per_cpu_ptr(priv->percpu_priv, i);
340 
341 		cpustats = (u64 *)&percpu_priv->stats;
342 
343 		/* add stats from all CPUs */
344 		for (j = 0; j < numstats; j++)
345 			netstats[j] += cpustats[j];
346 	}
347 }
348 
349 static int dpaa_setup_tc(struct net_device *net_dev, enum tc_setup_type type,
350 			 void *type_data)
351 {
352 	struct dpaa_priv *priv = netdev_priv(net_dev);
353 	struct tc_mqprio_qopt *mqprio = type_data;
354 	u8 num_tc;
355 	int i;
356 
357 	if (type != TC_SETUP_QDISC_MQPRIO)
358 		return -EOPNOTSUPP;
359 
360 	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
361 	num_tc = mqprio->num_tc;
362 
363 	if (num_tc == priv->num_tc)
364 		return 0;
365 
366 	if (!num_tc) {
367 		netdev_reset_tc(net_dev);
368 		goto out;
369 	}
370 
371 	if (num_tc > DPAA_TC_NUM) {
372 		netdev_err(net_dev, "Too many traffic classes: max %d supported.\n",
373 			   DPAA_TC_NUM);
374 		return -EINVAL;
375 	}
376 
377 	netdev_set_num_tc(net_dev, num_tc);
378 
379 	for (i = 0; i < num_tc; i++)
380 		netdev_set_tc_queue(net_dev, i, DPAA_TC_TXQ_NUM,
381 				    i * DPAA_TC_TXQ_NUM);
382 
383 out:
384 	priv->num_tc = num_tc ? : 1;
385 	netif_set_real_num_tx_queues(net_dev, priv->num_tc * DPAA_TC_TXQ_NUM);
386 	return 0;
387 }
388 
389 static struct mac_device *dpaa_mac_dev_get(struct platform_device *pdev)
390 {
391 	struct dpaa_eth_data *eth_data;
392 	struct device *dpaa_dev;
393 	struct mac_device *mac_dev;
394 
395 	dpaa_dev = &pdev->dev;
396 	eth_data = dpaa_dev->platform_data;
397 	if (!eth_data) {
398 		dev_err(dpaa_dev, "eth_data missing\n");
399 		return ERR_PTR(-ENODEV);
400 	}
401 	mac_dev = eth_data->mac_dev;
402 	if (!mac_dev) {
403 		dev_err(dpaa_dev, "mac_dev missing\n");
404 		return ERR_PTR(-EINVAL);
405 	}
406 
407 	return mac_dev;
408 }
409 
410 static int dpaa_set_mac_address(struct net_device *net_dev, void *addr)
411 {
412 	const struct dpaa_priv *priv;
413 	struct mac_device *mac_dev;
414 	struct sockaddr old_addr;
415 	int err;
416 
417 	priv = netdev_priv(net_dev);
418 
419 	memcpy(old_addr.sa_data, net_dev->dev_addr,  ETH_ALEN);
420 
421 	err = eth_mac_addr(net_dev, addr);
422 	if (err < 0) {
423 		netif_err(priv, drv, net_dev, "eth_mac_addr() = %d\n", err);
424 		return err;
425 	}
426 
427 	mac_dev = priv->mac_dev;
428 
429 	err = mac_dev->change_addr(mac_dev->fman_mac,
430 				   (enet_addr_t *)net_dev->dev_addr);
431 	if (err < 0) {
432 		netif_err(priv, drv, net_dev, "mac_dev->change_addr() = %d\n",
433 			  err);
434 		/* reverting to previous address */
435 		eth_mac_addr(net_dev, &old_addr);
436 
437 		return err;
438 	}
439 
440 	return 0;
441 }
442 
443 static void dpaa_set_rx_mode(struct net_device *net_dev)
444 {
445 	const struct dpaa_priv	*priv;
446 	int err;
447 
448 	priv = netdev_priv(net_dev);
449 
450 	if (!!(net_dev->flags & IFF_PROMISC) != priv->mac_dev->promisc) {
451 		priv->mac_dev->promisc = !priv->mac_dev->promisc;
452 		err = priv->mac_dev->set_promisc(priv->mac_dev->fman_mac,
453 						 priv->mac_dev->promisc);
454 		if (err < 0)
455 			netif_err(priv, drv, net_dev,
456 				  "mac_dev->set_promisc() = %d\n",
457 				  err);
458 	}
459 
460 	if (!!(net_dev->flags & IFF_ALLMULTI) != priv->mac_dev->allmulti) {
461 		priv->mac_dev->allmulti = !priv->mac_dev->allmulti;
462 		err = priv->mac_dev->set_allmulti(priv->mac_dev->fman_mac,
463 						  priv->mac_dev->allmulti);
464 		if (err < 0)
465 			netif_err(priv, drv, net_dev,
466 				  "mac_dev->set_allmulti() = %d\n",
467 				  err);
468 	}
469 
470 	err = priv->mac_dev->set_multi(net_dev, priv->mac_dev);
471 	if (err < 0)
472 		netif_err(priv, drv, net_dev, "mac_dev->set_multi() = %d\n",
473 			  err);
474 }
475 
476 static struct dpaa_bp *dpaa_bpid2pool(int bpid)
477 {
478 	if (WARN_ON(bpid < 0 || bpid >= BM_MAX_NUM_OF_POOLS))
479 		return NULL;
480 
481 	return dpaa_bp_array[bpid];
482 }
483 
484 /* checks if this bpool is already allocated */
485 static bool dpaa_bpid2pool_use(int bpid)
486 {
487 	if (dpaa_bpid2pool(bpid)) {
488 		atomic_inc(&dpaa_bp_array[bpid]->refs);
489 		return true;
490 	}
491 
492 	return false;
493 }
494 
495 /* called only once per bpid by dpaa_bp_alloc_pool() */
496 static void dpaa_bpid2pool_map(int bpid, struct dpaa_bp *dpaa_bp)
497 {
498 	dpaa_bp_array[bpid] = dpaa_bp;
499 	atomic_set(&dpaa_bp->refs, 1);
500 }
501 
502 static int dpaa_bp_alloc_pool(struct dpaa_bp *dpaa_bp)
503 {
504 	int err;
505 
506 	if (dpaa_bp->size == 0 || dpaa_bp->config_count == 0) {
507 		pr_err("%s: Buffer pool is not properly initialized! Missing size or initial number of buffers\n",
508 		       __func__);
509 		return -EINVAL;
510 	}
511 
512 	/* If the pool is already specified, we only create one per bpid */
513 	if (dpaa_bp->bpid != FSL_DPAA_BPID_INV &&
514 	    dpaa_bpid2pool_use(dpaa_bp->bpid))
515 		return 0;
516 
517 	if (dpaa_bp->bpid == FSL_DPAA_BPID_INV) {
518 		dpaa_bp->pool = bman_new_pool();
519 		if (!dpaa_bp->pool) {
520 			pr_err("%s: bman_new_pool() failed\n",
521 			       __func__);
522 			return -ENODEV;
523 		}
524 
525 		dpaa_bp->bpid = (u8)bman_get_bpid(dpaa_bp->pool);
526 	}
527 
528 	if (dpaa_bp->seed_cb) {
529 		err = dpaa_bp->seed_cb(dpaa_bp);
530 		if (err)
531 			goto pool_seed_failed;
532 	}
533 
534 	dpaa_bpid2pool_map(dpaa_bp->bpid, dpaa_bp);
535 
536 	return 0;
537 
538 pool_seed_failed:
539 	pr_err("%s: pool seeding failed\n", __func__);
540 	bman_free_pool(dpaa_bp->pool);
541 
542 	return err;
543 }
544 
545 /* remove and free all the buffers from the given buffer pool */
546 static void dpaa_bp_drain(struct dpaa_bp *bp)
547 {
548 	u8 num = 8;
549 	int ret;
550 
551 	do {
552 		struct bm_buffer bmb[8];
553 		int i;
554 
555 		ret = bman_acquire(bp->pool, bmb, num);
556 		if (ret < 0) {
557 			if (num == 8) {
558 				/* we have less than 8 buffers left;
559 				 * drain them one by one
560 				 */
561 				num = 1;
562 				ret = 1;
563 				continue;
564 			} else {
565 				/* Pool is fully drained */
566 				break;
567 			}
568 		}
569 
570 		if (bp->free_buf_cb)
571 			for (i = 0; i < num; i++)
572 				bp->free_buf_cb(bp, &bmb[i]);
573 	} while (ret > 0);
574 }
575 
576 static void dpaa_bp_free(struct dpaa_bp *dpaa_bp)
577 {
578 	struct dpaa_bp *bp = dpaa_bpid2pool(dpaa_bp->bpid);
579 
580 	/* the mapping between bpid and dpaa_bp is done very late in the
581 	 * allocation procedure; if something failed before the mapping, the bp
582 	 * was not configured, therefore we don't need the below instructions
583 	 */
584 	if (!bp)
585 		return;
586 
587 	if (!atomic_dec_and_test(&bp->refs))
588 		return;
589 
590 	if (bp->free_buf_cb)
591 		dpaa_bp_drain(bp);
592 
593 	dpaa_bp_array[bp->bpid] = NULL;
594 	bman_free_pool(bp->pool);
595 }
596 
597 static void dpaa_bps_free(struct dpaa_priv *priv)
598 {
599 	int i;
600 
601 	for (i = 0; i < DPAA_BPS_NUM; i++)
602 		dpaa_bp_free(priv->dpaa_bps[i]);
603 }
604 
605 /* Use multiple WQs for FQ assignment:
606  *	- Tx Confirmation queues go to WQ1.
607  *	- Rx Error and Tx Error queues go to WQ5 (giving them a better chance
608  *	  to be scheduled, in case there are many more FQs in WQ6).
609  *	- Rx Default goes to WQ6.
610  *	- Tx queues go to different WQs depending on their priority. Equal
611  *	  chunks of NR_CPUS queues go to WQ6 (lowest priority), WQ2, WQ1 and
612  *	  WQ0 (highest priority).
613  * This ensures that Tx-confirmed buffers are timely released. In particular,
614  * it avoids congestion on the Tx Confirm FQs, which can pile up PFDRs if they
615  * are greatly outnumbered by other FQs in the system, while
616  * dequeue scheduling is round-robin.
617  */
618 static inline void dpaa_assign_wq(struct dpaa_fq *fq, int idx)
619 {
620 	switch (fq->fq_type) {
621 	case FQ_TYPE_TX_CONFIRM:
622 	case FQ_TYPE_TX_CONF_MQ:
623 		fq->wq = 1;
624 		break;
625 	case FQ_TYPE_RX_ERROR:
626 	case FQ_TYPE_TX_ERROR:
627 		fq->wq = 5;
628 		break;
629 	case FQ_TYPE_RX_DEFAULT:
630 	case FQ_TYPE_RX_PCD:
631 		fq->wq = 6;
632 		break;
633 	case FQ_TYPE_TX:
634 		switch (idx / DPAA_TC_TXQ_NUM) {
635 		case 0:
636 			/* Low priority (best effort) */
637 			fq->wq = 6;
638 			break;
639 		case 1:
640 			/* Medium priority */
641 			fq->wq = 2;
642 			break;
643 		case 2:
644 			/* High priority */
645 			fq->wq = 1;
646 			break;
647 		case 3:
648 			/* Very high priority */
649 			fq->wq = 0;
650 			break;
651 		default:
652 			WARN(1, "Too many TX FQs: more than %d!\n",
653 			     DPAA_ETH_TXQ_NUM);
654 		}
655 		break;
656 	default:
657 		WARN(1, "Invalid FQ type %d for FQID %d!\n",
658 		     fq->fq_type, fq->fqid);
659 	}
660 }
661 
662 static struct dpaa_fq *dpaa_fq_alloc(struct device *dev,
663 				     u32 start, u32 count,
664 				     struct list_head *list,
665 				     enum dpaa_fq_type fq_type)
666 {
667 	struct dpaa_fq *dpaa_fq;
668 	int i;
669 
670 	dpaa_fq = devm_kcalloc(dev, count, sizeof(*dpaa_fq),
671 			       GFP_KERNEL);
672 	if (!dpaa_fq)
673 		return NULL;
674 
675 	for (i = 0; i < count; i++) {
676 		dpaa_fq[i].fq_type = fq_type;
677 		dpaa_fq[i].fqid = start ? start + i : 0;
678 		list_add_tail(&dpaa_fq[i].list, list);
679 	}
680 
681 	for (i = 0; i < count; i++)
682 		dpaa_assign_wq(dpaa_fq + i, i);
683 
684 	return dpaa_fq;
685 }
686 
687 static int dpaa_alloc_all_fqs(struct device *dev, struct list_head *list,
688 			      struct fm_port_fqs *port_fqs)
689 {
690 	struct dpaa_fq *dpaa_fq;
691 	u32 fq_base, fq_base_aligned, i;
692 
693 	dpaa_fq = dpaa_fq_alloc(dev, 0, 1, list, FQ_TYPE_RX_ERROR);
694 	if (!dpaa_fq)
695 		goto fq_alloc_failed;
696 
697 	port_fqs->rx_errq = &dpaa_fq[0];
698 
699 	dpaa_fq = dpaa_fq_alloc(dev, 0, 1, list, FQ_TYPE_RX_DEFAULT);
700 	if (!dpaa_fq)
701 		goto fq_alloc_failed;
702 
703 	port_fqs->rx_defq = &dpaa_fq[0];
704 
705 	/* the PCD FQIDs range needs to be aligned for correct operation */
706 	if (qman_alloc_fqid_range(&fq_base, 2 * DPAA_ETH_PCD_RXQ_NUM))
707 		goto fq_alloc_failed;
708 
709 	fq_base_aligned = ALIGN(fq_base, DPAA_ETH_PCD_RXQ_NUM);
710 
711 	for (i = fq_base; i < fq_base_aligned; i++)
712 		qman_release_fqid(i);
713 
714 	for (i = fq_base_aligned + DPAA_ETH_PCD_RXQ_NUM;
715 	     i < (fq_base + 2 * DPAA_ETH_PCD_RXQ_NUM); i++)
716 		qman_release_fqid(i);
717 
718 	dpaa_fq = dpaa_fq_alloc(dev, fq_base_aligned, DPAA_ETH_PCD_RXQ_NUM,
719 				list, FQ_TYPE_RX_PCD);
720 	if (!dpaa_fq)
721 		goto fq_alloc_failed;
722 
723 	port_fqs->rx_pcdq = &dpaa_fq[0];
724 
725 	if (!dpaa_fq_alloc(dev, 0, DPAA_ETH_TXQ_NUM, list, FQ_TYPE_TX_CONF_MQ))
726 		goto fq_alloc_failed;
727 
728 	dpaa_fq = dpaa_fq_alloc(dev, 0, 1, list, FQ_TYPE_TX_ERROR);
729 	if (!dpaa_fq)
730 		goto fq_alloc_failed;
731 
732 	port_fqs->tx_errq = &dpaa_fq[0];
733 
734 	dpaa_fq = dpaa_fq_alloc(dev, 0, 1, list, FQ_TYPE_TX_CONFIRM);
735 	if (!dpaa_fq)
736 		goto fq_alloc_failed;
737 
738 	port_fqs->tx_defq = &dpaa_fq[0];
739 
740 	if (!dpaa_fq_alloc(dev, 0, DPAA_ETH_TXQ_NUM, list, FQ_TYPE_TX))
741 		goto fq_alloc_failed;
742 
743 	return 0;
744 
745 fq_alloc_failed:
746 	dev_err(dev, "dpaa_fq_alloc() failed\n");
747 	return -ENOMEM;
748 }
749 
750 static u32 rx_pool_channel;
751 static DEFINE_SPINLOCK(rx_pool_channel_init);
752 
753 static int dpaa_get_channel(void)
754 {
755 	spin_lock(&rx_pool_channel_init);
756 	if (!rx_pool_channel) {
757 		u32 pool;
758 		int ret;
759 
760 		ret = qman_alloc_pool(&pool);
761 
762 		if (!ret)
763 			rx_pool_channel = pool;
764 	}
765 	spin_unlock(&rx_pool_channel_init);
766 	if (!rx_pool_channel)
767 		return -ENOMEM;
768 	return rx_pool_channel;
769 }
770 
771 static void dpaa_release_channel(void)
772 {
773 	qman_release_pool(rx_pool_channel);
774 }
775 
776 static void dpaa_eth_add_channel(u16 channel)
777 {
778 	u32 pool = QM_SDQCR_CHANNELS_POOL_CONV(channel);
779 	const cpumask_t *cpus = qman_affine_cpus();
780 	struct qman_portal *portal;
781 	int cpu;
782 
783 	for_each_cpu(cpu, cpus) {
784 		portal = qman_get_affine_portal(cpu);
785 		qman_p_static_dequeue_add(portal, pool);
786 	}
787 }
788 
789 /* Congestion group state change notification callback.
790  * Stops the device's egress queues while they are congested and
791  * wakes them upon exiting congested state.
792  * Also updates some CGR-related stats.
793  */
794 static void dpaa_eth_cgscn(struct qman_portal *qm, struct qman_cgr *cgr,
795 			   int congested)
796 {
797 	struct dpaa_priv *priv = (struct dpaa_priv *)container_of(cgr,
798 		struct dpaa_priv, cgr_data.cgr);
799 
800 	if (congested) {
801 		priv->cgr_data.congestion_start_jiffies = jiffies;
802 		netif_tx_stop_all_queues(priv->net_dev);
803 		priv->cgr_data.cgr_congested_count++;
804 	} else {
805 		priv->cgr_data.congested_jiffies +=
806 			(jiffies - priv->cgr_data.congestion_start_jiffies);
807 		netif_tx_wake_all_queues(priv->net_dev);
808 	}
809 }
810 
811 static int dpaa_eth_cgr_init(struct dpaa_priv *priv)
812 {
813 	struct qm_mcc_initcgr initcgr;
814 	u32 cs_th;
815 	int err;
816 
817 	err = qman_alloc_cgrid(&priv->cgr_data.cgr.cgrid);
818 	if (err < 0) {
819 		if (netif_msg_drv(priv))
820 			pr_err("%s: Error %d allocating CGR ID\n",
821 			       __func__, err);
822 		goto out_error;
823 	}
824 	priv->cgr_data.cgr.cb = dpaa_eth_cgscn;
825 
826 	/* Enable Congestion State Change Notifications and CS taildrop */
827 	memset(&initcgr, 0, sizeof(initcgr));
828 	initcgr.we_mask = cpu_to_be16(QM_CGR_WE_CSCN_EN | QM_CGR_WE_CS_THRES);
829 	initcgr.cgr.cscn_en = QM_CGR_EN;
830 
831 	/* Set different thresholds based on the MAC speed.
832 	 * This may turn suboptimal if the MAC is reconfigured at a speed
833 	 * lower than its max, e.g. if a dTSEC later negotiates a 100Mbps link.
834 	 * In such cases, we ought to reconfigure the threshold, too.
835 	 */
836 	if (priv->mac_dev->if_support & SUPPORTED_10000baseT_Full)
837 		cs_th = DPAA_CS_THRESHOLD_10G;
838 	else
839 		cs_th = DPAA_CS_THRESHOLD_1G;
840 	qm_cgr_cs_thres_set64(&initcgr.cgr.cs_thres, cs_th, 1);
841 
842 	initcgr.we_mask |= cpu_to_be16(QM_CGR_WE_CSTD_EN);
843 	initcgr.cgr.cstd_en = QM_CGR_EN;
844 
845 	err = qman_create_cgr(&priv->cgr_data.cgr, QMAN_CGR_FLAG_USE_INIT,
846 			      &initcgr);
847 	if (err < 0) {
848 		if (netif_msg_drv(priv))
849 			pr_err("%s: Error %d creating CGR with ID %d\n",
850 			       __func__, err, priv->cgr_data.cgr.cgrid);
851 		qman_release_cgrid(priv->cgr_data.cgr.cgrid);
852 		goto out_error;
853 	}
854 	if (netif_msg_drv(priv))
855 		pr_debug("Created CGR %d for netdev with hwaddr %pM on QMan channel %d\n",
856 			 priv->cgr_data.cgr.cgrid, priv->mac_dev->addr,
857 			 priv->cgr_data.cgr.chan);
858 
859 out_error:
860 	return err;
861 }
862 
863 static inline void dpaa_setup_ingress(const struct dpaa_priv *priv,
864 				      struct dpaa_fq *fq,
865 				      const struct qman_fq *template)
866 {
867 	fq->fq_base = *template;
868 	fq->net_dev = priv->net_dev;
869 
870 	fq->flags = QMAN_FQ_FLAG_NO_ENQUEUE;
871 	fq->channel = priv->channel;
872 }
873 
874 static inline void dpaa_setup_egress(const struct dpaa_priv *priv,
875 				     struct dpaa_fq *fq,
876 				     struct fman_port *port,
877 				     const struct qman_fq *template)
878 {
879 	fq->fq_base = *template;
880 	fq->net_dev = priv->net_dev;
881 
882 	if (port) {
883 		fq->flags = QMAN_FQ_FLAG_TO_DCPORTAL;
884 		fq->channel = (u16)fman_port_get_qman_channel_id(port);
885 	} else {
886 		fq->flags = QMAN_FQ_FLAG_NO_MODIFY;
887 	}
888 }
889 
890 static void dpaa_fq_setup(struct dpaa_priv *priv,
891 			  const struct dpaa_fq_cbs *fq_cbs,
892 			  struct fman_port *tx_port)
893 {
894 	int egress_cnt = 0, conf_cnt = 0, num_portals = 0, portal_cnt = 0, cpu;
895 	const cpumask_t *affine_cpus = qman_affine_cpus();
896 	u16 channels[NR_CPUS];
897 	struct dpaa_fq *fq;
898 
899 	for_each_cpu(cpu, affine_cpus)
900 		channels[num_portals++] = qman_affine_channel(cpu);
901 
902 	if (num_portals == 0)
903 		dev_err(priv->net_dev->dev.parent,
904 			"No Qman software (affine) channels found");
905 
906 	/* Initialize each FQ in the list */
907 	list_for_each_entry(fq, &priv->dpaa_fq_list, list) {
908 		switch (fq->fq_type) {
909 		case FQ_TYPE_RX_DEFAULT:
910 			dpaa_setup_ingress(priv, fq, &fq_cbs->rx_defq);
911 			break;
912 		case FQ_TYPE_RX_ERROR:
913 			dpaa_setup_ingress(priv, fq, &fq_cbs->rx_errq);
914 			break;
915 		case FQ_TYPE_RX_PCD:
916 			if (!num_portals)
917 				continue;
918 			dpaa_setup_ingress(priv, fq, &fq_cbs->rx_defq);
919 			fq->channel = channels[portal_cnt++ % num_portals];
920 			break;
921 		case FQ_TYPE_TX:
922 			dpaa_setup_egress(priv, fq, tx_port,
923 					  &fq_cbs->egress_ern);
924 			/* If we have more Tx queues than the number of cores,
925 			 * just ignore the extra ones.
926 			 */
927 			if (egress_cnt < DPAA_ETH_TXQ_NUM)
928 				priv->egress_fqs[egress_cnt++] = &fq->fq_base;
929 			break;
930 		case FQ_TYPE_TX_CONF_MQ:
931 			priv->conf_fqs[conf_cnt++] = &fq->fq_base;
932 			/* fall through */
933 		case FQ_TYPE_TX_CONFIRM:
934 			dpaa_setup_ingress(priv, fq, &fq_cbs->tx_defq);
935 			break;
936 		case FQ_TYPE_TX_ERROR:
937 			dpaa_setup_ingress(priv, fq, &fq_cbs->tx_errq);
938 			break;
939 		default:
940 			dev_warn(priv->net_dev->dev.parent,
941 				 "Unknown FQ type detected!\n");
942 			break;
943 		}
944 	}
945 
946 	 /* Make sure all CPUs receive a corresponding Tx queue. */
947 	while (egress_cnt < DPAA_ETH_TXQ_NUM) {
948 		list_for_each_entry(fq, &priv->dpaa_fq_list, list) {
949 			if (fq->fq_type != FQ_TYPE_TX)
950 				continue;
951 			priv->egress_fqs[egress_cnt++] = &fq->fq_base;
952 			if (egress_cnt == DPAA_ETH_TXQ_NUM)
953 				break;
954 		}
955 	}
956 }
957 
958 static inline int dpaa_tx_fq_to_id(const struct dpaa_priv *priv,
959 				   struct qman_fq *tx_fq)
960 {
961 	int i;
962 
963 	for (i = 0; i < DPAA_ETH_TXQ_NUM; i++)
964 		if (priv->egress_fqs[i] == tx_fq)
965 			return i;
966 
967 	return -EINVAL;
968 }
969 
970 static int dpaa_fq_init(struct dpaa_fq *dpaa_fq, bool td_enable)
971 {
972 	const struct dpaa_priv	*priv;
973 	struct qman_fq *confq = NULL;
974 	struct qm_mcc_initfq initfq;
975 	struct device *dev;
976 	struct qman_fq *fq;
977 	int queue_id;
978 	int err;
979 
980 	priv = netdev_priv(dpaa_fq->net_dev);
981 	dev = dpaa_fq->net_dev->dev.parent;
982 
983 	if (dpaa_fq->fqid == 0)
984 		dpaa_fq->flags |= QMAN_FQ_FLAG_DYNAMIC_FQID;
985 
986 	dpaa_fq->init = !(dpaa_fq->flags & QMAN_FQ_FLAG_NO_MODIFY);
987 
988 	err = qman_create_fq(dpaa_fq->fqid, dpaa_fq->flags, &dpaa_fq->fq_base);
989 	if (err) {
990 		dev_err(dev, "qman_create_fq() failed\n");
991 		return err;
992 	}
993 	fq = &dpaa_fq->fq_base;
994 
995 	if (dpaa_fq->init) {
996 		memset(&initfq, 0, sizeof(initfq));
997 
998 		initfq.we_mask = cpu_to_be16(QM_INITFQ_WE_FQCTRL);
999 		/* Note: we may get to keep an empty FQ in cache */
1000 		initfq.fqd.fq_ctrl = cpu_to_be16(QM_FQCTRL_PREFERINCACHE);
1001 
1002 		/* Try to reduce the number of portal interrupts for
1003 		 * Tx Confirmation FQs.
1004 		 */
1005 		if (dpaa_fq->fq_type == FQ_TYPE_TX_CONFIRM)
1006 			initfq.fqd.fq_ctrl |= cpu_to_be16(QM_FQCTRL_AVOIDBLOCK);
1007 
1008 		/* FQ placement */
1009 		initfq.we_mask |= cpu_to_be16(QM_INITFQ_WE_DESTWQ);
1010 
1011 		qm_fqd_set_destwq(&initfq.fqd, dpaa_fq->channel, dpaa_fq->wq);
1012 
1013 		/* Put all egress queues in a congestion group of their own.
1014 		 * Sensu stricto, the Tx confirmation queues are Rx FQs,
1015 		 * rather than Tx - but they nonetheless account for the
1016 		 * memory footprint on behalf of egress traffic. We therefore
1017 		 * place them in the netdev's CGR, along with the Tx FQs.
1018 		 */
1019 		if (dpaa_fq->fq_type == FQ_TYPE_TX ||
1020 		    dpaa_fq->fq_type == FQ_TYPE_TX_CONFIRM ||
1021 		    dpaa_fq->fq_type == FQ_TYPE_TX_CONF_MQ) {
1022 			initfq.we_mask |= cpu_to_be16(QM_INITFQ_WE_CGID);
1023 			initfq.fqd.fq_ctrl |= cpu_to_be16(QM_FQCTRL_CGE);
1024 			initfq.fqd.cgid = (u8)priv->cgr_data.cgr.cgrid;
1025 			/* Set a fixed overhead accounting, in an attempt to
1026 			 * reduce the impact of fixed-size skb shells and the
1027 			 * driver's needed headroom on system memory. This is
1028 			 * especially the case when the egress traffic is
1029 			 * composed of small datagrams.
1030 			 * Unfortunately, QMan's OAL value is capped to an
1031 			 * insufficient value, but even that is better than
1032 			 * no overhead accounting at all.
1033 			 */
1034 			initfq.we_mask |= cpu_to_be16(QM_INITFQ_WE_OAC);
1035 			qm_fqd_set_oac(&initfq.fqd, QM_OAC_CG);
1036 			qm_fqd_set_oal(&initfq.fqd,
1037 				       min(sizeof(struct sk_buff) +
1038 				       priv->tx_headroom,
1039 				       (size_t)FSL_QMAN_MAX_OAL));
1040 		}
1041 
1042 		if (td_enable) {
1043 			initfq.we_mask |= cpu_to_be16(QM_INITFQ_WE_TDTHRESH);
1044 			qm_fqd_set_taildrop(&initfq.fqd, DPAA_FQ_TD, 1);
1045 			initfq.fqd.fq_ctrl = cpu_to_be16(QM_FQCTRL_TDE);
1046 		}
1047 
1048 		if (dpaa_fq->fq_type == FQ_TYPE_TX) {
1049 			queue_id = dpaa_tx_fq_to_id(priv, &dpaa_fq->fq_base);
1050 			if (queue_id >= 0)
1051 				confq = priv->conf_fqs[queue_id];
1052 			if (confq) {
1053 				initfq.we_mask |=
1054 					cpu_to_be16(QM_INITFQ_WE_CONTEXTA);
1055 			/* ContextA: OVOM=1(use contextA2 bits instead of ICAD)
1056 			 *	     A2V=1 (contextA A2 field is valid)
1057 			 *	     A0V=1 (contextA A0 field is valid)
1058 			 *	     B0V=1 (contextB field is valid)
1059 			 * ContextA A2: EBD=1 (deallocate buffers inside FMan)
1060 			 * ContextB B0(ASPID): 0 (absolute Virtual Storage ID)
1061 			 */
1062 				qm_fqd_context_a_set64(&initfq.fqd,
1063 						       0x1e00000080000000ULL);
1064 			}
1065 		}
1066 
1067 		/* Put all the ingress queues in our "ingress CGR". */
1068 		if (priv->use_ingress_cgr &&
1069 		    (dpaa_fq->fq_type == FQ_TYPE_RX_DEFAULT ||
1070 		     dpaa_fq->fq_type == FQ_TYPE_RX_ERROR ||
1071 		     dpaa_fq->fq_type == FQ_TYPE_RX_PCD)) {
1072 			initfq.we_mask |= cpu_to_be16(QM_INITFQ_WE_CGID);
1073 			initfq.fqd.fq_ctrl |= cpu_to_be16(QM_FQCTRL_CGE);
1074 			initfq.fqd.cgid = (u8)priv->ingress_cgr.cgrid;
1075 			/* Set a fixed overhead accounting, just like for the
1076 			 * egress CGR.
1077 			 */
1078 			initfq.we_mask |= cpu_to_be16(QM_INITFQ_WE_OAC);
1079 			qm_fqd_set_oac(&initfq.fqd, QM_OAC_CG);
1080 			qm_fqd_set_oal(&initfq.fqd,
1081 				       min(sizeof(struct sk_buff) +
1082 				       priv->tx_headroom,
1083 				       (size_t)FSL_QMAN_MAX_OAL));
1084 		}
1085 
1086 		/* Initialization common to all ingress queues */
1087 		if (dpaa_fq->flags & QMAN_FQ_FLAG_NO_ENQUEUE) {
1088 			initfq.we_mask |= cpu_to_be16(QM_INITFQ_WE_CONTEXTA);
1089 			initfq.fqd.fq_ctrl |= cpu_to_be16(QM_FQCTRL_HOLDACTIVE |
1090 						QM_FQCTRL_CTXASTASHING);
1091 			initfq.fqd.context_a.stashing.exclusive =
1092 				QM_STASHING_EXCL_DATA | QM_STASHING_EXCL_CTX |
1093 				QM_STASHING_EXCL_ANNOTATION;
1094 			qm_fqd_set_stashing(&initfq.fqd, 1, 2,
1095 					    DIV_ROUND_UP(sizeof(struct qman_fq),
1096 							 64));
1097 		}
1098 
1099 		err = qman_init_fq(fq, QMAN_INITFQ_FLAG_SCHED, &initfq);
1100 		if (err < 0) {
1101 			dev_err(dev, "qman_init_fq(%u) = %d\n",
1102 				qman_fq_fqid(fq), err);
1103 			qman_destroy_fq(fq);
1104 			return err;
1105 		}
1106 	}
1107 
1108 	dpaa_fq->fqid = qman_fq_fqid(fq);
1109 
1110 	return 0;
1111 }
1112 
1113 static int dpaa_fq_free_entry(struct device *dev, struct qman_fq *fq)
1114 {
1115 	const struct dpaa_priv  *priv;
1116 	struct dpaa_fq *dpaa_fq;
1117 	int err, error;
1118 
1119 	err = 0;
1120 
1121 	dpaa_fq = container_of(fq, struct dpaa_fq, fq_base);
1122 	priv = netdev_priv(dpaa_fq->net_dev);
1123 
1124 	if (dpaa_fq->init) {
1125 		err = qman_retire_fq(fq, NULL);
1126 		if (err < 0 && netif_msg_drv(priv))
1127 			dev_err(dev, "qman_retire_fq(%u) = %d\n",
1128 				qman_fq_fqid(fq), err);
1129 
1130 		error = qman_oos_fq(fq);
1131 		if (error < 0 && netif_msg_drv(priv)) {
1132 			dev_err(dev, "qman_oos_fq(%u) = %d\n",
1133 				qman_fq_fqid(fq), error);
1134 			if (err >= 0)
1135 				err = error;
1136 		}
1137 	}
1138 
1139 	qman_destroy_fq(fq);
1140 	list_del(&dpaa_fq->list);
1141 
1142 	return err;
1143 }
1144 
1145 static int dpaa_fq_free(struct device *dev, struct list_head *list)
1146 {
1147 	struct dpaa_fq *dpaa_fq, *tmp;
1148 	int err, error;
1149 
1150 	err = 0;
1151 	list_for_each_entry_safe(dpaa_fq, tmp, list, list) {
1152 		error = dpaa_fq_free_entry(dev, (struct qman_fq *)dpaa_fq);
1153 		if (error < 0 && err >= 0)
1154 			err = error;
1155 	}
1156 
1157 	return err;
1158 }
1159 
1160 static int dpaa_eth_init_tx_port(struct fman_port *port, struct dpaa_fq *errq,
1161 				 struct dpaa_fq *defq,
1162 				 struct dpaa_buffer_layout *buf_layout)
1163 {
1164 	struct fman_buffer_prefix_content buf_prefix_content;
1165 	struct fman_port_params params;
1166 	int err;
1167 
1168 	memset(&params, 0, sizeof(params));
1169 	memset(&buf_prefix_content, 0, sizeof(buf_prefix_content));
1170 
1171 	buf_prefix_content.priv_data_size = buf_layout->priv_data_size;
1172 	buf_prefix_content.pass_prs_result = true;
1173 	buf_prefix_content.pass_hash_result = true;
1174 	buf_prefix_content.pass_time_stamp = true;
1175 	buf_prefix_content.data_align = DPAA_FD_DATA_ALIGNMENT;
1176 
1177 	params.specific_params.non_rx_params.err_fqid = errq->fqid;
1178 	params.specific_params.non_rx_params.dflt_fqid = defq->fqid;
1179 
1180 	err = fman_port_config(port, &params);
1181 	if (err) {
1182 		pr_err("%s: fman_port_config failed\n", __func__);
1183 		return err;
1184 	}
1185 
1186 	err = fman_port_cfg_buf_prefix_content(port, &buf_prefix_content);
1187 	if (err) {
1188 		pr_err("%s: fman_port_cfg_buf_prefix_content failed\n",
1189 		       __func__);
1190 		return err;
1191 	}
1192 
1193 	err = fman_port_init(port);
1194 	if (err)
1195 		pr_err("%s: fm_port_init failed\n", __func__);
1196 
1197 	return err;
1198 }
1199 
1200 static int dpaa_eth_init_rx_port(struct fman_port *port, struct dpaa_bp **bps,
1201 				 size_t count, struct dpaa_fq *errq,
1202 				 struct dpaa_fq *defq, struct dpaa_fq *pcdq,
1203 				 struct dpaa_buffer_layout *buf_layout)
1204 {
1205 	struct fman_buffer_prefix_content buf_prefix_content;
1206 	struct fman_port_rx_params *rx_p;
1207 	struct fman_port_params params;
1208 	int i, err;
1209 
1210 	memset(&params, 0, sizeof(params));
1211 	memset(&buf_prefix_content, 0, sizeof(buf_prefix_content));
1212 
1213 	buf_prefix_content.priv_data_size = buf_layout->priv_data_size;
1214 	buf_prefix_content.pass_prs_result = true;
1215 	buf_prefix_content.pass_hash_result = true;
1216 	buf_prefix_content.pass_time_stamp = true;
1217 	buf_prefix_content.data_align = DPAA_FD_DATA_ALIGNMENT;
1218 
1219 	rx_p = &params.specific_params.rx_params;
1220 	rx_p->err_fqid = errq->fqid;
1221 	rx_p->dflt_fqid = defq->fqid;
1222 	if (pcdq) {
1223 		rx_p->pcd_base_fqid = pcdq->fqid;
1224 		rx_p->pcd_fqs_count = DPAA_ETH_PCD_RXQ_NUM;
1225 	}
1226 
1227 	count = min(ARRAY_SIZE(rx_p->ext_buf_pools.ext_buf_pool), count);
1228 	rx_p->ext_buf_pools.num_of_pools_used = (u8)count;
1229 	for (i = 0; i < count; i++) {
1230 		rx_p->ext_buf_pools.ext_buf_pool[i].id =  bps[i]->bpid;
1231 		rx_p->ext_buf_pools.ext_buf_pool[i].size = (u16)bps[i]->size;
1232 	}
1233 
1234 	err = fman_port_config(port, &params);
1235 	if (err) {
1236 		pr_err("%s: fman_port_config failed\n", __func__);
1237 		return err;
1238 	}
1239 
1240 	err = fman_port_cfg_buf_prefix_content(port, &buf_prefix_content);
1241 	if (err) {
1242 		pr_err("%s: fman_port_cfg_buf_prefix_content failed\n",
1243 		       __func__);
1244 		return err;
1245 	}
1246 
1247 	err = fman_port_init(port);
1248 	if (err)
1249 		pr_err("%s: fm_port_init failed\n", __func__);
1250 
1251 	return err;
1252 }
1253 
1254 static int dpaa_eth_init_ports(struct mac_device *mac_dev,
1255 			       struct dpaa_bp **bps, size_t count,
1256 			       struct fm_port_fqs *port_fqs,
1257 			       struct dpaa_buffer_layout *buf_layout,
1258 			       struct device *dev)
1259 {
1260 	struct fman_port *rxport = mac_dev->port[RX];
1261 	struct fman_port *txport = mac_dev->port[TX];
1262 	int err;
1263 
1264 	err = dpaa_eth_init_tx_port(txport, port_fqs->tx_errq,
1265 				    port_fqs->tx_defq, &buf_layout[TX]);
1266 	if (err)
1267 		return err;
1268 
1269 	err = dpaa_eth_init_rx_port(rxport, bps, count, port_fqs->rx_errq,
1270 				    port_fqs->rx_defq, port_fqs->rx_pcdq,
1271 				    &buf_layout[RX]);
1272 
1273 	return err;
1274 }
1275 
1276 static int dpaa_bman_release(const struct dpaa_bp *dpaa_bp,
1277 			     struct bm_buffer *bmb, int cnt)
1278 {
1279 	int err;
1280 
1281 	err = bman_release(dpaa_bp->pool, bmb, cnt);
1282 	/* Should never occur, address anyway to avoid leaking the buffers */
1283 	if (WARN_ON(err) && dpaa_bp->free_buf_cb)
1284 		while (cnt-- > 0)
1285 			dpaa_bp->free_buf_cb(dpaa_bp, &bmb[cnt]);
1286 
1287 	return cnt;
1288 }
1289 
1290 static void dpaa_release_sgt_members(struct qm_sg_entry *sgt)
1291 {
1292 	struct bm_buffer bmb[DPAA_BUFF_RELEASE_MAX];
1293 	struct dpaa_bp *dpaa_bp;
1294 	int i = 0, j;
1295 
1296 	memset(bmb, 0, sizeof(bmb));
1297 
1298 	do {
1299 		dpaa_bp = dpaa_bpid2pool(sgt[i].bpid);
1300 		if (!dpaa_bp)
1301 			return;
1302 
1303 		j = 0;
1304 		do {
1305 			WARN_ON(qm_sg_entry_is_ext(&sgt[i]));
1306 
1307 			bm_buffer_set64(&bmb[j], qm_sg_entry_get64(&sgt[i]));
1308 
1309 			j++; i++;
1310 		} while (j < ARRAY_SIZE(bmb) &&
1311 				!qm_sg_entry_is_final(&sgt[i - 1]) &&
1312 				sgt[i - 1].bpid == sgt[i].bpid);
1313 
1314 		dpaa_bman_release(dpaa_bp, bmb, j);
1315 	} while (!qm_sg_entry_is_final(&sgt[i - 1]));
1316 }
1317 
1318 static void dpaa_fd_release(const struct net_device *net_dev,
1319 			    const struct qm_fd *fd)
1320 {
1321 	struct qm_sg_entry *sgt;
1322 	struct dpaa_bp *dpaa_bp;
1323 	struct bm_buffer bmb;
1324 	dma_addr_t addr;
1325 	void *vaddr;
1326 
1327 	bmb.data = 0;
1328 	bm_buffer_set64(&bmb, qm_fd_addr(fd));
1329 
1330 	dpaa_bp = dpaa_bpid2pool(fd->bpid);
1331 	if (!dpaa_bp)
1332 		return;
1333 
1334 	if (qm_fd_get_format(fd) == qm_fd_sg) {
1335 		vaddr = phys_to_virt(qm_fd_addr(fd));
1336 		sgt = vaddr + qm_fd_get_offset(fd);
1337 
1338 		dma_unmap_single(dpaa_bp->dev, qm_fd_addr(fd), dpaa_bp->size,
1339 				 DMA_FROM_DEVICE);
1340 
1341 		dpaa_release_sgt_members(sgt);
1342 
1343 		addr = dma_map_single(dpaa_bp->dev, vaddr, dpaa_bp->size,
1344 				      DMA_FROM_DEVICE);
1345 		if (dma_mapping_error(dpaa_bp->dev, addr)) {
1346 			dev_err(dpaa_bp->dev, "DMA mapping failed");
1347 			return;
1348 		}
1349 		bm_buffer_set64(&bmb, addr);
1350 	}
1351 
1352 	dpaa_bman_release(dpaa_bp, &bmb, 1);
1353 }
1354 
1355 static void count_ern(struct dpaa_percpu_priv *percpu_priv,
1356 		      const union qm_mr_entry *msg)
1357 {
1358 	switch (msg->ern.rc & QM_MR_RC_MASK) {
1359 	case QM_MR_RC_CGR_TAILDROP:
1360 		percpu_priv->ern_cnt.cg_tdrop++;
1361 		break;
1362 	case QM_MR_RC_WRED:
1363 		percpu_priv->ern_cnt.wred++;
1364 		break;
1365 	case QM_MR_RC_ERROR:
1366 		percpu_priv->ern_cnt.err_cond++;
1367 		break;
1368 	case QM_MR_RC_ORPWINDOW_EARLY:
1369 		percpu_priv->ern_cnt.early_window++;
1370 		break;
1371 	case QM_MR_RC_ORPWINDOW_LATE:
1372 		percpu_priv->ern_cnt.late_window++;
1373 		break;
1374 	case QM_MR_RC_FQ_TAILDROP:
1375 		percpu_priv->ern_cnt.fq_tdrop++;
1376 		break;
1377 	case QM_MR_RC_ORPWINDOW_RETIRED:
1378 		percpu_priv->ern_cnt.fq_retired++;
1379 		break;
1380 	case QM_MR_RC_ORP_ZERO:
1381 		percpu_priv->ern_cnt.orp_zero++;
1382 		break;
1383 	}
1384 }
1385 
1386 /* Turn on HW checksum computation for this outgoing frame.
1387  * If the current protocol is not something we support in this regard
1388  * (or if the stack has already computed the SW checksum), we do nothing.
1389  *
1390  * Returns 0 if all goes well (or HW csum doesn't apply), and a negative value
1391  * otherwise.
1392  *
1393  * Note that this function may modify the fd->cmd field and the skb data buffer
1394  * (the Parse Results area).
1395  */
1396 static int dpaa_enable_tx_csum(struct dpaa_priv *priv,
1397 			       struct sk_buff *skb,
1398 			       struct qm_fd *fd,
1399 			       char *parse_results)
1400 {
1401 	struct fman_prs_result *parse_result;
1402 	u16 ethertype = ntohs(skb->protocol);
1403 	struct ipv6hdr *ipv6h = NULL;
1404 	struct iphdr *iph;
1405 	int retval = 0;
1406 	u8 l4_proto;
1407 
1408 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1409 		return 0;
1410 
1411 	/* Note: L3 csum seems to be already computed in sw, but we can't choose
1412 	 * L4 alone from the FM configuration anyway.
1413 	 */
1414 
1415 	/* Fill in some fields of the Parse Results array, so the FMan
1416 	 * can find them as if they came from the FMan Parser.
1417 	 */
1418 	parse_result = (struct fman_prs_result *)parse_results;
1419 
1420 	/* If we're dealing with VLAN, get the real Ethernet type */
1421 	if (ethertype == ETH_P_8021Q) {
1422 		/* We can't always assume the MAC header is set correctly
1423 		 * by the stack, so reset to beginning of skb->data
1424 		 */
1425 		skb_reset_mac_header(skb);
1426 		ethertype = ntohs(vlan_eth_hdr(skb)->h_vlan_encapsulated_proto);
1427 	}
1428 
1429 	/* Fill in the relevant L3 parse result fields
1430 	 * and read the L4 protocol type
1431 	 */
1432 	switch (ethertype) {
1433 	case ETH_P_IP:
1434 		parse_result->l3r = cpu_to_be16(FM_L3_PARSE_RESULT_IPV4);
1435 		iph = ip_hdr(skb);
1436 		WARN_ON(!iph);
1437 		l4_proto = iph->protocol;
1438 		break;
1439 	case ETH_P_IPV6:
1440 		parse_result->l3r = cpu_to_be16(FM_L3_PARSE_RESULT_IPV6);
1441 		ipv6h = ipv6_hdr(skb);
1442 		WARN_ON(!ipv6h);
1443 		l4_proto = ipv6h->nexthdr;
1444 		break;
1445 	default:
1446 		/* We shouldn't even be here */
1447 		if (net_ratelimit())
1448 			netif_alert(priv, tx_err, priv->net_dev,
1449 				    "Can't compute HW csum for L3 proto 0x%x\n",
1450 				    ntohs(skb->protocol));
1451 		retval = -EIO;
1452 		goto return_error;
1453 	}
1454 
1455 	/* Fill in the relevant L4 parse result fields */
1456 	switch (l4_proto) {
1457 	case IPPROTO_UDP:
1458 		parse_result->l4r = FM_L4_PARSE_RESULT_UDP;
1459 		break;
1460 	case IPPROTO_TCP:
1461 		parse_result->l4r = FM_L4_PARSE_RESULT_TCP;
1462 		break;
1463 	default:
1464 		if (net_ratelimit())
1465 			netif_alert(priv, tx_err, priv->net_dev,
1466 				    "Can't compute HW csum for L4 proto 0x%x\n",
1467 				    l4_proto);
1468 		retval = -EIO;
1469 		goto return_error;
1470 	}
1471 
1472 	/* At index 0 is IPOffset_1 as defined in the Parse Results */
1473 	parse_result->ip_off[0] = (u8)skb_network_offset(skb);
1474 	parse_result->l4_off = (u8)skb_transport_offset(skb);
1475 
1476 	/* Enable L3 (and L4, if TCP or UDP) HW checksum. */
1477 	fd->cmd |= cpu_to_be32(FM_FD_CMD_RPD | FM_FD_CMD_DTC);
1478 
1479 	/* On P1023 and similar platforms fd->cmd interpretation could
1480 	 * be disabled by setting CONTEXT_A bit ICMD; currently this bit
1481 	 * is not set so we do not need to check; in the future, if/when
1482 	 * using context_a we need to check this bit
1483 	 */
1484 
1485 return_error:
1486 	return retval;
1487 }
1488 
1489 static int dpaa_bp_add_8_bufs(const struct dpaa_bp *dpaa_bp)
1490 {
1491 	struct device *dev = dpaa_bp->dev;
1492 	struct bm_buffer bmb[8];
1493 	dma_addr_t addr;
1494 	void *new_buf;
1495 	u8 i;
1496 
1497 	for (i = 0; i < 8; i++) {
1498 		new_buf = netdev_alloc_frag(dpaa_bp->raw_size);
1499 		if (unlikely(!new_buf)) {
1500 			dev_err(dev, "netdev_alloc_frag() failed, size %zu\n",
1501 				dpaa_bp->raw_size);
1502 			goto release_previous_buffs;
1503 		}
1504 		new_buf = PTR_ALIGN(new_buf, SMP_CACHE_BYTES);
1505 
1506 		addr = dma_map_single(dev, new_buf,
1507 				      dpaa_bp->size, DMA_FROM_DEVICE);
1508 		if (unlikely(dma_mapping_error(dev, addr))) {
1509 			dev_err(dpaa_bp->dev, "DMA map failed");
1510 			goto release_previous_buffs;
1511 		}
1512 
1513 		bmb[i].data = 0;
1514 		bm_buffer_set64(&bmb[i], addr);
1515 	}
1516 
1517 release_bufs:
1518 	return dpaa_bman_release(dpaa_bp, bmb, i);
1519 
1520 release_previous_buffs:
1521 	WARN_ONCE(1, "dpaa_eth: failed to add buffers on Rx\n");
1522 
1523 	bm_buffer_set64(&bmb[i], 0);
1524 	/* Avoid releasing a completely null buffer; bman_release() requires
1525 	 * at least one buffer.
1526 	 */
1527 	if (likely(i))
1528 		goto release_bufs;
1529 
1530 	return 0;
1531 }
1532 
1533 static int dpaa_bp_seed(struct dpaa_bp *dpaa_bp)
1534 {
1535 	int i;
1536 
1537 	/* Give each CPU an allotment of "config_count" buffers */
1538 	for_each_possible_cpu(i) {
1539 		int *count_ptr = per_cpu_ptr(dpaa_bp->percpu_count, i);
1540 		int j;
1541 
1542 		/* Although we access another CPU's counters here
1543 		 * we do it at boot time so it is safe
1544 		 */
1545 		for (j = 0; j < dpaa_bp->config_count; j += 8)
1546 			*count_ptr += dpaa_bp_add_8_bufs(dpaa_bp);
1547 	}
1548 	return 0;
1549 }
1550 
1551 /* Add buffers/(pages) for Rx processing whenever bpool count falls below
1552  * REFILL_THRESHOLD.
1553  */
1554 static int dpaa_eth_refill_bpool(struct dpaa_bp *dpaa_bp, int *countptr)
1555 {
1556 	int count = *countptr;
1557 	int new_bufs;
1558 
1559 	if (unlikely(count < FSL_DPAA_ETH_REFILL_THRESHOLD)) {
1560 		do {
1561 			new_bufs = dpaa_bp_add_8_bufs(dpaa_bp);
1562 			if (unlikely(!new_bufs)) {
1563 				/* Avoid looping forever if we've temporarily
1564 				 * run out of memory. We'll try again at the
1565 				 * next NAPI cycle.
1566 				 */
1567 				break;
1568 			}
1569 			count += new_bufs;
1570 		} while (count < FSL_DPAA_ETH_MAX_BUF_COUNT);
1571 
1572 		*countptr = count;
1573 		if (unlikely(count < FSL_DPAA_ETH_MAX_BUF_COUNT))
1574 			return -ENOMEM;
1575 	}
1576 
1577 	return 0;
1578 }
1579 
1580 static int dpaa_eth_refill_bpools(struct dpaa_priv *priv)
1581 {
1582 	struct dpaa_bp *dpaa_bp;
1583 	int *countptr;
1584 	int res, i;
1585 
1586 	for (i = 0; i < DPAA_BPS_NUM; i++) {
1587 		dpaa_bp = priv->dpaa_bps[i];
1588 		if (!dpaa_bp)
1589 			return -EINVAL;
1590 		countptr = this_cpu_ptr(dpaa_bp->percpu_count);
1591 		res  = dpaa_eth_refill_bpool(dpaa_bp, countptr);
1592 		if (res)
1593 			return res;
1594 	}
1595 	return 0;
1596 }
1597 
1598 /* Cleanup function for outgoing frame descriptors that were built on Tx path,
1599  * either contiguous frames or scatter/gather ones.
1600  * Skb freeing is not handled here.
1601  *
1602  * This function may be called on error paths in the Tx function, so guard
1603  * against cases when not all fd relevant fields were filled in.
1604  *
1605  * Return the skb backpointer, since for S/G frames the buffer containing it
1606  * gets freed here.
1607  */
1608 static struct sk_buff *dpaa_cleanup_tx_fd(const struct dpaa_priv *priv,
1609 					  const struct qm_fd *fd)
1610 {
1611 	const enum dma_data_direction dma_dir = DMA_TO_DEVICE;
1612 	struct device *dev = priv->net_dev->dev.parent;
1613 	struct skb_shared_hwtstamps shhwtstamps;
1614 	dma_addr_t addr = qm_fd_addr(fd);
1615 	const struct qm_sg_entry *sgt;
1616 	struct sk_buff **skbh, *skb;
1617 	int nr_frags, i;
1618 	u64 ns;
1619 
1620 	skbh = (struct sk_buff **)phys_to_virt(addr);
1621 	skb = *skbh;
1622 
1623 	if (priv->tx_tstamp && skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) {
1624 		memset(&shhwtstamps, 0, sizeof(shhwtstamps));
1625 
1626 		if (!fman_port_get_tstamp(priv->mac_dev->port[TX], (void *)skbh,
1627 					  &ns)) {
1628 			shhwtstamps.hwtstamp = ns_to_ktime(ns);
1629 			skb_tstamp_tx(skb, &shhwtstamps);
1630 		} else {
1631 			dev_warn(dev, "fman_port_get_tstamp failed!\n");
1632 		}
1633 	}
1634 
1635 	if (unlikely(qm_fd_get_format(fd) == qm_fd_sg)) {
1636 		nr_frags = skb_shinfo(skb)->nr_frags;
1637 		dma_unmap_single(dev, addr,
1638 				 qm_fd_get_offset(fd) + DPAA_SGT_SIZE,
1639 				 dma_dir);
1640 
1641 		/* The sgt buffer has been allocated with netdev_alloc_frag(),
1642 		 * it's from lowmem.
1643 		 */
1644 		sgt = phys_to_virt(addr + qm_fd_get_offset(fd));
1645 
1646 		/* sgt[0] is from lowmem, was dma_map_single()-ed */
1647 		dma_unmap_single(dev, qm_sg_addr(&sgt[0]),
1648 				 qm_sg_entry_get_len(&sgt[0]), dma_dir);
1649 
1650 		/* remaining pages were mapped with skb_frag_dma_map() */
1651 		for (i = 1; i < nr_frags; i++) {
1652 			WARN_ON(qm_sg_entry_is_ext(&sgt[i]));
1653 
1654 			dma_unmap_page(dev, qm_sg_addr(&sgt[i]),
1655 				       qm_sg_entry_get_len(&sgt[i]), dma_dir);
1656 		}
1657 
1658 		/* Free the page frag that we allocated on Tx */
1659 		skb_free_frag(phys_to_virt(addr));
1660 	} else {
1661 		dma_unmap_single(dev, addr,
1662 				 skb_tail_pointer(skb) - (u8 *)skbh, dma_dir);
1663 	}
1664 
1665 	return skb;
1666 }
1667 
1668 static u8 rx_csum_offload(const struct dpaa_priv *priv, const struct qm_fd *fd)
1669 {
1670 	/* The parser has run and performed L4 checksum validation.
1671 	 * We know there were no parser errors (and implicitly no
1672 	 * L4 csum error), otherwise we wouldn't be here.
1673 	 */
1674 	if ((priv->net_dev->features & NETIF_F_RXCSUM) &&
1675 	    (be32_to_cpu(fd->status) & FM_FD_STAT_L4CV))
1676 		return CHECKSUM_UNNECESSARY;
1677 
1678 	/* We're here because either the parser didn't run or the L4 checksum
1679 	 * was not verified. This may include the case of a UDP frame with
1680 	 * checksum zero or an L4 proto other than TCP/UDP
1681 	 */
1682 	return CHECKSUM_NONE;
1683 }
1684 
1685 /* Build a linear skb around the received buffer.
1686  * We are guaranteed there is enough room at the end of the data buffer to
1687  * accommodate the shared info area of the skb.
1688  */
1689 static struct sk_buff *contig_fd_to_skb(const struct dpaa_priv *priv,
1690 					const struct qm_fd *fd)
1691 {
1692 	ssize_t fd_off = qm_fd_get_offset(fd);
1693 	dma_addr_t addr = qm_fd_addr(fd);
1694 	struct dpaa_bp *dpaa_bp;
1695 	struct sk_buff *skb;
1696 	void *vaddr;
1697 
1698 	vaddr = phys_to_virt(addr);
1699 	WARN_ON(!IS_ALIGNED((unsigned long)vaddr, SMP_CACHE_BYTES));
1700 
1701 	dpaa_bp = dpaa_bpid2pool(fd->bpid);
1702 	if (!dpaa_bp)
1703 		goto free_buffer;
1704 
1705 	skb = build_skb(vaddr, dpaa_bp->size +
1706 			SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
1707 	if (WARN_ONCE(!skb, "Build skb failure on Rx\n"))
1708 		goto free_buffer;
1709 	WARN_ON(fd_off != priv->rx_headroom);
1710 	skb_reserve(skb, fd_off);
1711 	skb_put(skb, qm_fd_get_length(fd));
1712 
1713 	skb->ip_summed = rx_csum_offload(priv, fd);
1714 
1715 	return skb;
1716 
1717 free_buffer:
1718 	skb_free_frag(vaddr);
1719 	return NULL;
1720 }
1721 
1722 /* Build an skb with the data of the first S/G entry in the linear portion and
1723  * the rest of the frame as skb fragments.
1724  *
1725  * The page fragment holding the S/G Table is recycled here.
1726  */
1727 static struct sk_buff *sg_fd_to_skb(const struct dpaa_priv *priv,
1728 				    const struct qm_fd *fd)
1729 {
1730 	ssize_t fd_off = qm_fd_get_offset(fd);
1731 	dma_addr_t addr = qm_fd_addr(fd);
1732 	const struct qm_sg_entry *sgt;
1733 	struct page *page, *head_page;
1734 	struct dpaa_bp *dpaa_bp;
1735 	void *vaddr, *sg_vaddr;
1736 	int frag_off, frag_len;
1737 	struct sk_buff *skb;
1738 	dma_addr_t sg_addr;
1739 	int page_offset;
1740 	unsigned int sz;
1741 	int *count_ptr;
1742 	int i;
1743 
1744 	vaddr = phys_to_virt(addr);
1745 	WARN_ON(!IS_ALIGNED((unsigned long)vaddr, SMP_CACHE_BYTES));
1746 
1747 	/* Iterate through the SGT entries and add data buffers to the skb */
1748 	sgt = vaddr + fd_off;
1749 	skb = NULL;
1750 	for (i = 0; i < DPAA_SGT_MAX_ENTRIES; i++) {
1751 		/* Extension bit is not supported */
1752 		WARN_ON(qm_sg_entry_is_ext(&sgt[i]));
1753 
1754 		sg_addr = qm_sg_addr(&sgt[i]);
1755 		sg_vaddr = phys_to_virt(sg_addr);
1756 		WARN_ON(!IS_ALIGNED((unsigned long)sg_vaddr,
1757 				    SMP_CACHE_BYTES));
1758 
1759 		/* We may use multiple Rx pools */
1760 		dpaa_bp = dpaa_bpid2pool(sgt[i].bpid);
1761 		if (!dpaa_bp)
1762 			goto free_buffers;
1763 
1764 		count_ptr = this_cpu_ptr(dpaa_bp->percpu_count);
1765 		dma_unmap_single(dpaa_bp->dev, sg_addr, dpaa_bp->size,
1766 				 DMA_FROM_DEVICE);
1767 		if (!skb) {
1768 			sz = dpaa_bp->size +
1769 				SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1770 			skb = build_skb(sg_vaddr, sz);
1771 			if (WARN_ON(!skb))
1772 				goto free_buffers;
1773 
1774 			skb->ip_summed = rx_csum_offload(priv, fd);
1775 
1776 			/* Make sure forwarded skbs will have enough space
1777 			 * on Tx, if extra headers are added.
1778 			 */
1779 			WARN_ON(fd_off != priv->rx_headroom);
1780 			skb_reserve(skb, fd_off);
1781 			skb_put(skb, qm_sg_entry_get_len(&sgt[i]));
1782 		} else {
1783 			/* Not the first S/G entry; all data from buffer will
1784 			 * be added in an skb fragment; fragment index is offset
1785 			 * by one since first S/G entry was incorporated in the
1786 			 * linear part of the skb.
1787 			 *
1788 			 * Caution: 'page' may be a tail page.
1789 			 */
1790 			page = virt_to_page(sg_vaddr);
1791 			head_page = virt_to_head_page(sg_vaddr);
1792 
1793 			/* Compute offset in (possibly tail) page */
1794 			page_offset = ((unsigned long)sg_vaddr &
1795 					(PAGE_SIZE - 1)) +
1796 				(page_address(page) - page_address(head_page));
1797 			/* page_offset only refers to the beginning of sgt[i];
1798 			 * but the buffer itself may have an internal offset.
1799 			 */
1800 			frag_off = qm_sg_entry_get_off(&sgt[i]) + page_offset;
1801 			frag_len = qm_sg_entry_get_len(&sgt[i]);
1802 			/* skb_add_rx_frag() does no checking on the page; if
1803 			 * we pass it a tail page, we'll end up with
1804 			 * bad page accounting and eventually with segafults.
1805 			 */
1806 			skb_add_rx_frag(skb, i - 1, head_page, frag_off,
1807 					frag_len, dpaa_bp->size);
1808 		}
1809 		/* Update the pool count for the current {cpu x bpool} */
1810 		(*count_ptr)--;
1811 
1812 		if (qm_sg_entry_is_final(&sgt[i]))
1813 			break;
1814 	}
1815 	WARN_ONCE(i == DPAA_SGT_MAX_ENTRIES, "No final bit on SGT\n");
1816 
1817 	/* free the SG table buffer */
1818 	skb_free_frag(vaddr);
1819 
1820 	return skb;
1821 
1822 free_buffers:
1823 	/* compensate sw bpool counter changes */
1824 	for (i--; i >= 0; i--) {
1825 		dpaa_bp = dpaa_bpid2pool(sgt[i].bpid);
1826 		if (dpaa_bp) {
1827 			count_ptr = this_cpu_ptr(dpaa_bp->percpu_count);
1828 			(*count_ptr)++;
1829 		}
1830 	}
1831 	/* free all the SG entries */
1832 	for (i = 0; i < DPAA_SGT_MAX_ENTRIES ; i++) {
1833 		sg_addr = qm_sg_addr(&sgt[i]);
1834 		sg_vaddr = phys_to_virt(sg_addr);
1835 		skb_free_frag(sg_vaddr);
1836 		dpaa_bp = dpaa_bpid2pool(sgt[i].bpid);
1837 		if (dpaa_bp) {
1838 			count_ptr = this_cpu_ptr(dpaa_bp->percpu_count);
1839 			(*count_ptr)--;
1840 		}
1841 
1842 		if (qm_sg_entry_is_final(&sgt[i]))
1843 			break;
1844 	}
1845 	/* free the SGT fragment */
1846 	skb_free_frag(vaddr);
1847 
1848 	return NULL;
1849 }
1850 
1851 static int skb_to_contig_fd(struct dpaa_priv *priv,
1852 			    struct sk_buff *skb, struct qm_fd *fd,
1853 			    int *offset)
1854 {
1855 	struct net_device *net_dev = priv->net_dev;
1856 	struct device *dev = net_dev->dev.parent;
1857 	enum dma_data_direction dma_dir;
1858 	unsigned char *buffer_start;
1859 	struct sk_buff **skbh;
1860 	dma_addr_t addr;
1861 	int err;
1862 
1863 	/* We are guaranteed to have at least tx_headroom bytes
1864 	 * available, so just use that for offset.
1865 	 */
1866 	fd->bpid = FSL_DPAA_BPID_INV;
1867 	buffer_start = skb->data - priv->tx_headroom;
1868 	dma_dir = DMA_TO_DEVICE;
1869 
1870 	skbh = (struct sk_buff **)buffer_start;
1871 	*skbh = skb;
1872 
1873 	/* Enable L3/L4 hardware checksum computation.
1874 	 *
1875 	 * We must do this before dma_map_single(DMA_TO_DEVICE), because we may
1876 	 * need to write into the skb.
1877 	 */
1878 	err = dpaa_enable_tx_csum(priv, skb, fd,
1879 				  ((char *)skbh) + DPAA_TX_PRIV_DATA_SIZE);
1880 	if (unlikely(err < 0)) {
1881 		if (net_ratelimit())
1882 			netif_err(priv, tx_err, net_dev, "HW csum error: %d\n",
1883 				  err);
1884 		return err;
1885 	}
1886 
1887 	/* Fill in the rest of the FD fields */
1888 	qm_fd_set_contig(fd, priv->tx_headroom, skb->len);
1889 	fd->cmd |= cpu_to_be32(FM_FD_CMD_FCO);
1890 
1891 	/* Map the entire buffer size that may be seen by FMan, but no more */
1892 	addr = dma_map_single(dev, skbh,
1893 			      skb_tail_pointer(skb) - buffer_start, dma_dir);
1894 	if (unlikely(dma_mapping_error(dev, addr))) {
1895 		if (net_ratelimit())
1896 			netif_err(priv, tx_err, net_dev, "dma_map_single() failed\n");
1897 		return -EINVAL;
1898 	}
1899 	qm_fd_addr_set64(fd, addr);
1900 
1901 	return 0;
1902 }
1903 
1904 static int skb_to_sg_fd(struct dpaa_priv *priv,
1905 			struct sk_buff *skb, struct qm_fd *fd)
1906 {
1907 	const enum dma_data_direction dma_dir = DMA_TO_DEVICE;
1908 	const int nr_frags = skb_shinfo(skb)->nr_frags;
1909 	struct net_device *net_dev = priv->net_dev;
1910 	struct device *dev = net_dev->dev.parent;
1911 	struct qm_sg_entry *sgt;
1912 	struct sk_buff **skbh;
1913 	int i, j, err, sz;
1914 	void *buffer_start;
1915 	skb_frag_t *frag;
1916 	dma_addr_t addr;
1917 	size_t frag_len;
1918 	void *sgt_buf;
1919 
1920 	/* get a page frag to store the SGTable */
1921 	sz = SKB_DATA_ALIGN(priv->tx_headroom + DPAA_SGT_SIZE);
1922 	sgt_buf = netdev_alloc_frag(sz);
1923 	if (unlikely(!sgt_buf)) {
1924 		netdev_err(net_dev, "netdev_alloc_frag() failed for size %d\n",
1925 			   sz);
1926 		return -ENOMEM;
1927 	}
1928 
1929 	/* Enable L3/L4 hardware checksum computation.
1930 	 *
1931 	 * We must do this before dma_map_single(DMA_TO_DEVICE), because we may
1932 	 * need to write into the skb.
1933 	 */
1934 	err = dpaa_enable_tx_csum(priv, skb, fd,
1935 				  sgt_buf + DPAA_TX_PRIV_DATA_SIZE);
1936 	if (unlikely(err < 0)) {
1937 		if (net_ratelimit())
1938 			netif_err(priv, tx_err, net_dev, "HW csum error: %d\n",
1939 				  err);
1940 		goto csum_failed;
1941 	}
1942 
1943 	/* SGT[0] is used by the linear part */
1944 	sgt = (struct qm_sg_entry *)(sgt_buf + priv->tx_headroom);
1945 	frag_len = skb_headlen(skb);
1946 	qm_sg_entry_set_len(&sgt[0], frag_len);
1947 	sgt[0].bpid = FSL_DPAA_BPID_INV;
1948 	sgt[0].offset = 0;
1949 	addr = dma_map_single(dev, skb->data,
1950 			      skb_headlen(skb), dma_dir);
1951 	if (unlikely(dma_mapping_error(dev, addr))) {
1952 		dev_err(dev, "DMA mapping failed");
1953 		err = -EINVAL;
1954 		goto sg0_map_failed;
1955 	}
1956 	qm_sg_entry_set64(&sgt[0], addr);
1957 
1958 	/* populate the rest of SGT entries */
1959 	for (i = 0; i < nr_frags; i++) {
1960 		frag = &skb_shinfo(skb)->frags[i];
1961 		frag_len = frag->size;
1962 		WARN_ON(!skb_frag_page(frag));
1963 		addr = skb_frag_dma_map(dev, frag, 0,
1964 					frag_len, dma_dir);
1965 		if (unlikely(dma_mapping_error(dev, addr))) {
1966 			dev_err(dev, "DMA mapping failed");
1967 			err = -EINVAL;
1968 			goto sg_map_failed;
1969 		}
1970 
1971 		qm_sg_entry_set_len(&sgt[i + 1], frag_len);
1972 		sgt[i + 1].bpid = FSL_DPAA_BPID_INV;
1973 		sgt[i + 1].offset = 0;
1974 
1975 		/* keep the offset in the address */
1976 		qm_sg_entry_set64(&sgt[i + 1], addr);
1977 	}
1978 
1979 	/* Set the final bit in the last used entry of the SGT */
1980 	qm_sg_entry_set_f(&sgt[nr_frags], frag_len);
1981 
1982 	qm_fd_set_sg(fd, priv->tx_headroom, skb->len);
1983 
1984 	/* DMA map the SGT page */
1985 	buffer_start = (void *)sgt - priv->tx_headroom;
1986 	skbh = (struct sk_buff **)buffer_start;
1987 	*skbh = skb;
1988 
1989 	addr = dma_map_single(dev, buffer_start,
1990 			      priv->tx_headroom + DPAA_SGT_SIZE, dma_dir);
1991 	if (unlikely(dma_mapping_error(dev, addr))) {
1992 		dev_err(dev, "DMA mapping failed");
1993 		err = -EINVAL;
1994 		goto sgt_map_failed;
1995 	}
1996 
1997 	fd->bpid = FSL_DPAA_BPID_INV;
1998 	fd->cmd |= cpu_to_be32(FM_FD_CMD_FCO);
1999 	qm_fd_addr_set64(fd, addr);
2000 
2001 	return 0;
2002 
2003 sgt_map_failed:
2004 sg_map_failed:
2005 	for (j = 0; j < i; j++)
2006 		dma_unmap_page(dev, qm_sg_addr(&sgt[j]),
2007 			       qm_sg_entry_get_len(&sgt[j]), dma_dir);
2008 sg0_map_failed:
2009 csum_failed:
2010 	skb_free_frag(sgt_buf);
2011 
2012 	return err;
2013 }
2014 
2015 static inline int dpaa_xmit(struct dpaa_priv *priv,
2016 			    struct rtnl_link_stats64 *percpu_stats,
2017 			    int queue,
2018 			    struct qm_fd *fd)
2019 {
2020 	struct qman_fq *egress_fq;
2021 	int err, i;
2022 
2023 	egress_fq = priv->egress_fqs[queue];
2024 	if (fd->bpid == FSL_DPAA_BPID_INV)
2025 		fd->cmd |= cpu_to_be32(qman_fq_fqid(priv->conf_fqs[queue]));
2026 
2027 	/* Trace this Tx fd */
2028 	trace_dpaa_tx_fd(priv->net_dev, egress_fq, fd);
2029 
2030 	for (i = 0; i < DPAA_ENQUEUE_RETRIES; i++) {
2031 		err = qman_enqueue(egress_fq, fd);
2032 		if (err != -EBUSY)
2033 			break;
2034 	}
2035 
2036 	if (unlikely(err < 0)) {
2037 		percpu_stats->tx_fifo_errors++;
2038 		return err;
2039 	}
2040 
2041 	percpu_stats->tx_packets++;
2042 	percpu_stats->tx_bytes += qm_fd_get_length(fd);
2043 
2044 	return 0;
2045 }
2046 
2047 static netdev_tx_t
2048 dpaa_start_xmit(struct sk_buff *skb, struct net_device *net_dev)
2049 {
2050 	const int queue_mapping = skb_get_queue_mapping(skb);
2051 	bool nonlinear = skb_is_nonlinear(skb);
2052 	struct rtnl_link_stats64 *percpu_stats;
2053 	struct dpaa_percpu_priv *percpu_priv;
2054 	struct dpaa_priv *priv;
2055 	struct qm_fd fd;
2056 	int offset = 0;
2057 	int err = 0;
2058 
2059 	priv = netdev_priv(net_dev);
2060 	percpu_priv = this_cpu_ptr(priv->percpu_priv);
2061 	percpu_stats = &percpu_priv->stats;
2062 
2063 	qm_fd_clear_fd(&fd);
2064 
2065 	if (!nonlinear) {
2066 		/* We're going to store the skb backpointer at the beginning
2067 		 * of the data buffer, so we need a privately owned skb
2068 		 *
2069 		 * We've made sure skb is not shared in dev->priv_flags,
2070 		 * we need to verify the skb head is not cloned
2071 		 */
2072 		if (skb_cow_head(skb, priv->tx_headroom))
2073 			goto enomem;
2074 
2075 		WARN_ON(skb_is_nonlinear(skb));
2076 	}
2077 
2078 	/* MAX_SKB_FRAGS is equal or larger than our dpaa_SGT_MAX_ENTRIES;
2079 	 * make sure we don't feed FMan with more fragments than it supports.
2080 	 */
2081 	if (unlikely(nonlinear &&
2082 		     (skb_shinfo(skb)->nr_frags >= DPAA_SGT_MAX_ENTRIES))) {
2083 		/* If the egress skb contains more fragments than we support
2084 		 * we have no choice but to linearize it ourselves.
2085 		 */
2086 		if (__skb_linearize(skb))
2087 			goto enomem;
2088 
2089 		nonlinear = skb_is_nonlinear(skb);
2090 	}
2091 
2092 	if (nonlinear) {
2093 		/* Just create a S/G fd based on the skb */
2094 		err = skb_to_sg_fd(priv, skb, &fd);
2095 		percpu_priv->tx_frag_skbuffs++;
2096 	} else {
2097 		/* Create a contig FD from this skb */
2098 		err = skb_to_contig_fd(priv, skb, &fd, &offset);
2099 	}
2100 	if (unlikely(err < 0))
2101 		goto skb_to_fd_failed;
2102 
2103 	if (priv->tx_tstamp && skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) {
2104 		fd.cmd |= cpu_to_be32(FM_FD_CMD_UPD);
2105 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
2106 	}
2107 
2108 	if (likely(dpaa_xmit(priv, percpu_stats, queue_mapping, &fd) == 0))
2109 		return NETDEV_TX_OK;
2110 
2111 	dpaa_cleanup_tx_fd(priv, &fd);
2112 skb_to_fd_failed:
2113 enomem:
2114 	percpu_stats->tx_errors++;
2115 	dev_kfree_skb(skb);
2116 	return NETDEV_TX_OK;
2117 }
2118 
2119 static void dpaa_rx_error(struct net_device *net_dev,
2120 			  const struct dpaa_priv *priv,
2121 			  struct dpaa_percpu_priv *percpu_priv,
2122 			  const struct qm_fd *fd,
2123 			  u32 fqid)
2124 {
2125 	if (net_ratelimit())
2126 		netif_err(priv, hw, net_dev, "Err FD status = 0x%08x\n",
2127 			  be32_to_cpu(fd->status) & FM_FD_STAT_RX_ERRORS);
2128 
2129 	percpu_priv->stats.rx_errors++;
2130 
2131 	if (be32_to_cpu(fd->status) & FM_FD_ERR_DMA)
2132 		percpu_priv->rx_errors.dme++;
2133 	if (be32_to_cpu(fd->status) & FM_FD_ERR_PHYSICAL)
2134 		percpu_priv->rx_errors.fpe++;
2135 	if (be32_to_cpu(fd->status) & FM_FD_ERR_SIZE)
2136 		percpu_priv->rx_errors.fse++;
2137 	if (be32_to_cpu(fd->status) & FM_FD_ERR_PRS_HDR_ERR)
2138 		percpu_priv->rx_errors.phe++;
2139 
2140 	dpaa_fd_release(net_dev, fd);
2141 }
2142 
2143 static void dpaa_tx_error(struct net_device *net_dev,
2144 			  const struct dpaa_priv *priv,
2145 			  struct dpaa_percpu_priv *percpu_priv,
2146 			  const struct qm_fd *fd,
2147 			  u32 fqid)
2148 {
2149 	struct sk_buff *skb;
2150 
2151 	if (net_ratelimit())
2152 		netif_warn(priv, hw, net_dev, "FD status = 0x%08x\n",
2153 			   be32_to_cpu(fd->status) & FM_FD_STAT_TX_ERRORS);
2154 
2155 	percpu_priv->stats.tx_errors++;
2156 
2157 	skb = dpaa_cleanup_tx_fd(priv, fd);
2158 	dev_kfree_skb(skb);
2159 }
2160 
2161 static int dpaa_eth_poll(struct napi_struct *napi, int budget)
2162 {
2163 	struct dpaa_napi_portal *np =
2164 			container_of(napi, struct dpaa_napi_portal, napi);
2165 
2166 	int cleaned = qman_p_poll_dqrr(np->p, budget);
2167 
2168 	if (cleaned < budget) {
2169 		napi_complete_done(napi, cleaned);
2170 		qman_p_irqsource_add(np->p, QM_PIRQ_DQRI);
2171 
2172 	} else if (np->down) {
2173 		qman_p_irqsource_add(np->p, QM_PIRQ_DQRI);
2174 	}
2175 
2176 	return cleaned;
2177 }
2178 
2179 static void dpaa_tx_conf(struct net_device *net_dev,
2180 			 const struct dpaa_priv *priv,
2181 			 struct dpaa_percpu_priv *percpu_priv,
2182 			 const struct qm_fd *fd,
2183 			 u32 fqid)
2184 {
2185 	struct sk_buff	*skb;
2186 
2187 	if (unlikely(be32_to_cpu(fd->status) & FM_FD_STAT_TX_ERRORS)) {
2188 		if (net_ratelimit())
2189 			netif_warn(priv, hw, net_dev, "FD status = 0x%08x\n",
2190 				   be32_to_cpu(fd->status) &
2191 				   FM_FD_STAT_TX_ERRORS);
2192 
2193 		percpu_priv->stats.tx_errors++;
2194 	}
2195 
2196 	percpu_priv->tx_confirm++;
2197 
2198 	skb = dpaa_cleanup_tx_fd(priv, fd);
2199 
2200 	consume_skb(skb);
2201 }
2202 
2203 static inline int dpaa_eth_napi_schedule(struct dpaa_percpu_priv *percpu_priv,
2204 					 struct qman_portal *portal)
2205 {
2206 	if (unlikely(in_irq() || !in_serving_softirq())) {
2207 		/* Disable QMan IRQ and invoke NAPI */
2208 		qman_p_irqsource_remove(portal, QM_PIRQ_DQRI);
2209 
2210 		percpu_priv->np.p = portal;
2211 		napi_schedule(&percpu_priv->np.napi);
2212 		percpu_priv->in_interrupt++;
2213 		return 1;
2214 	}
2215 	return 0;
2216 }
2217 
2218 static enum qman_cb_dqrr_result rx_error_dqrr(struct qman_portal *portal,
2219 					      struct qman_fq *fq,
2220 					      const struct qm_dqrr_entry *dq)
2221 {
2222 	struct dpaa_fq *dpaa_fq = container_of(fq, struct dpaa_fq, fq_base);
2223 	struct dpaa_percpu_priv *percpu_priv;
2224 	struct net_device *net_dev;
2225 	struct dpaa_bp *dpaa_bp;
2226 	struct dpaa_priv *priv;
2227 
2228 	net_dev = dpaa_fq->net_dev;
2229 	priv = netdev_priv(net_dev);
2230 	dpaa_bp = dpaa_bpid2pool(dq->fd.bpid);
2231 	if (!dpaa_bp)
2232 		return qman_cb_dqrr_consume;
2233 
2234 	percpu_priv = this_cpu_ptr(priv->percpu_priv);
2235 
2236 	if (dpaa_eth_napi_schedule(percpu_priv, portal))
2237 		return qman_cb_dqrr_stop;
2238 
2239 	dpaa_eth_refill_bpools(priv);
2240 	dpaa_rx_error(net_dev, priv, percpu_priv, &dq->fd, fq->fqid);
2241 
2242 	return qman_cb_dqrr_consume;
2243 }
2244 
2245 static enum qman_cb_dqrr_result rx_default_dqrr(struct qman_portal *portal,
2246 						struct qman_fq *fq,
2247 						const struct qm_dqrr_entry *dq)
2248 {
2249 	struct skb_shared_hwtstamps *shhwtstamps;
2250 	struct rtnl_link_stats64 *percpu_stats;
2251 	struct dpaa_percpu_priv *percpu_priv;
2252 	const struct qm_fd *fd = &dq->fd;
2253 	dma_addr_t addr = qm_fd_addr(fd);
2254 	enum qm_fd_format fd_format;
2255 	struct net_device *net_dev;
2256 	u32 fd_status, hash_offset;
2257 	struct dpaa_bp *dpaa_bp;
2258 	struct dpaa_priv *priv;
2259 	unsigned int skb_len;
2260 	struct sk_buff *skb;
2261 	int *count_ptr;
2262 	void *vaddr;
2263 	u64 ns;
2264 
2265 	fd_status = be32_to_cpu(fd->status);
2266 	fd_format = qm_fd_get_format(fd);
2267 	net_dev = ((struct dpaa_fq *)fq)->net_dev;
2268 	priv = netdev_priv(net_dev);
2269 	dpaa_bp = dpaa_bpid2pool(dq->fd.bpid);
2270 	if (!dpaa_bp)
2271 		return qman_cb_dqrr_consume;
2272 
2273 	/* Trace the Rx fd */
2274 	trace_dpaa_rx_fd(net_dev, fq, &dq->fd);
2275 
2276 	percpu_priv = this_cpu_ptr(priv->percpu_priv);
2277 	percpu_stats = &percpu_priv->stats;
2278 
2279 	if (unlikely(dpaa_eth_napi_schedule(percpu_priv, portal)))
2280 		return qman_cb_dqrr_stop;
2281 
2282 	/* Make sure we didn't run out of buffers */
2283 	if (unlikely(dpaa_eth_refill_bpools(priv))) {
2284 		/* Unable to refill the buffer pool due to insufficient
2285 		 * system memory. Just release the frame back into the pool,
2286 		 * otherwise we'll soon end up with an empty buffer pool.
2287 		 */
2288 		dpaa_fd_release(net_dev, &dq->fd);
2289 		return qman_cb_dqrr_consume;
2290 	}
2291 
2292 	if (unlikely(fd_status & FM_FD_STAT_RX_ERRORS) != 0) {
2293 		if (net_ratelimit())
2294 			netif_warn(priv, hw, net_dev, "FD status = 0x%08x\n",
2295 				   fd_status & FM_FD_STAT_RX_ERRORS);
2296 
2297 		percpu_stats->rx_errors++;
2298 		dpaa_fd_release(net_dev, fd);
2299 		return qman_cb_dqrr_consume;
2300 	}
2301 
2302 	dpaa_bp = dpaa_bpid2pool(fd->bpid);
2303 	if (!dpaa_bp)
2304 		return qman_cb_dqrr_consume;
2305 
2306 	dma_unmap_single(dpaa_bp->dev, addr, dpaa_bp->size, DMA_FROM_DEVICE);
2307 
2308 	/* prefetch the first 64 bytes of the frame or the SGT start */
2309 	vaddr = phys_to_virt(addr);
2310 	prefetch(vaddr + qm_fd_get_offset(fd));
2311 
2312 	/* The only FD types that we may receive are contig and S/G */
2313 	WARN_ON((fd_format != qm_fd_contig) && (fd_format != qm_fd_sg));
2314 
2315 	/* Account for either the contig buffer or the SGT buffer (depending on
2316 	 * which case we were in) having been removed from the pool.
2317 	 */
2318 	count_ptr = this_cpu_ptr(dpaa_bp->percpu_count);
2319 	(*count_ptr)--;
2320 
2321 	if (likely(fd_format == qm_fd_contig))
2322 		skb = contig_fd_to_skb(priv, fd);
2323 	else
2324 		skb = sg_fd_to_skb(priv, fd);
2325 	if (!skb)
2326 		return qman_cb_dqrr_consume;
2327 
2328 	if (priv->rx_tstamp) {
2329 		shhwtstamps = skb_hwtstamps(skb);
2330 		memset(shhwtstamps, 0, sizeof(*shhwtstamps));
2331 
2332 		if (!fman_port_get_tstamp(priv->mac_dev->port[RX], vaddr, &ns))
2333 			shhwtstamps->hwtstamp = ns_to_ktime(ns);
2334 		else
2335 			dev_warn(net_dev->dev.parent, "fman_port_get_tstamp failed!\n");
2336 	}
2337 
2338 	skb->protocol = eth_type_trans(skb, net_dev);
2339 
2340 	if (net_dev->features & NETIF_F_RXHASH && priv->keygen_in_use &&
2341 	    !fman_port_get_hash_result_offset(priv->mac_dev->port[RX],
2342 					      &hash_offset)) {
2343 		enum pkt_hash_types type;
2344 
2345 		/* if L4 exists, it was used in the hash generation */
2346 		type = be32_to_cpu(fd->status) & FM_FD_STAT_L4CV ?
2347 			PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3;
2348 		skb_set_hash(skb, be32_to_cpu(*(u32 *)(vaddr + hash_offset)),
2349 			     type);
2350 	}
2351 
2352 	skb_len = skb->len;
2353 
2354 	if (unlikely(netif_receive_skb(skb) == NET_RX_DROP)) {
2355 		percpu_stats->rx_dropped++;
2356 		return qman_cb_dqrr_consume;
2357 	}
2358 
2359 	percpu_stats->rx_packets++;
2360 	percpu_stats->rx_bytes += skb_len;
2361 
2362 	return qman_cb_dqrr_consume;
2363 }
2364 
2365 static enum qman_cb_dqrr_result conf_error_dqrr(struct qman_portal *portal,
2366 						struct qman_fq *fq,
2367 						const struct qm_dqrr_entry *dq)
2368 {
2369 	struct dpaa_percpu_priv *percpu_priv;
2370 	struct net_device *net_dev;
2371 	struct dpaa_priv *priv;
2372 
2373 	net_dev = ((struct dpaa_fq *)fq)->net_dev;
2374 	priv = netdev_priv(net_dev);
2375 
2376 	percpu_priv = this_cpu_ptr(priv->percpu_priv);
2377 
2378 	if (dpaa_eth_napi_schedule(percpu_priv, portal))
2379 		return qman_cb_dqrr_stop;
2380 
2381 	dpaa_tx_error(net_dev, priv, percpu_priv, &dq->fd, fq->fqid);
2382 
2383 	return qman_cb_dqrr_consume;
2384 }
2385 
2386 static enum qman_cb_dqrr_result conf_dflt_dqrr(struct qman_portal *portal,
2387 					       struct qman_fq *fq,
2388 					       const struct qm_dqrr_entry *dq)
2389 {
2390 	struct dpaa_percpu_priv *percpu_priv;
2391 	struct net_device *net_dev;
2392 	struct dpaa_priv *priv;
2393 
2394 	net_dev = ((struct dpaa_fq *)fq)->net_dev;
2395 	priv = netdev_priv(net_dev);
2396 
2397 	/* Trace the fd */
2398 	trace_dpaa_tx_conf_fd(net_dev, fq, &dq->fd);
2399 
2400 	percpu_priv = this_cpu_ptr(priv->percpu_priv);
2401 
2402 	if (dpaa_eth_napi_schedule(percpu_priv, portal))
2403 		return qman_cb_dqrr_stop;
2404 
2405 	dpaa_tx_conf(net_dev, priv, percpu_priv, &dq->fd, fq->fqid);
2406 
2407 	return qman_cb_dqrr_consume;
2408 }
2409 
2410 static void egress_ern(struct qman_portal *portal,
2411 		       struct qman_fq *fq,
2412 		       const union qm_mr_entry *msg)
2413 {
2414 	const struct qm_fd *fd = &msg->ern.fd;
2415 	struct dpaa_percpu_priv *percpu_priv;
2416 	const struct dpaa_priv *priv;
2417 	struct net_device *net_dev;
2418 	struct sk_buff *skb;
2419 
2420 	net_dev = ((struct dpaa_fq *)fq)->net_dev;
2421 	priv = netdev_priv(net_dev);
2422 	percpu_priv = this_cpu_ptr(priv->percpu_priv);
2423 
2424 	percpu_priv->stats.tx_dropped++;
2425 	percpu_priv->stats.tx_fifo_errors++;
2426 	count_ern(percpu_priv, msg);
2427 
2428 	skb = dpaa_cleanup_tx_fd(priv, fd);
2429 	dev_kfree_skb_any(skb);
2430 }
2431 
2432 static const struct dpaa_fq_cbs dpaa_fq_cbs = {
2433 	.rx_defq = { .cb = { .dqrr = rx_default_dqrr } },
2434 	.tx_defq = { .cb = { .dqrr = conf_dflt_dqrr } },
2435 	.rx_errq = { .cb = { .dqrr = rx_error_dqrr } },
2436 	.tx_errq = { .cb = { .dqrr = conf_error_dqrr } },
2437 	.egress_ern = { .cb = { .ern = egress_ern } }
2438 };
2439 
2440 static void dpaa_eth_napi_enable(struct dpaa_priv *priv)
2441 {
2442 	struct dpaa_percpu_priv *percpu_priv;
2443 	int i;
2444 
2445 	for_each_possible_cpu(i) {
2446 		percpu_priv = per_cpu_ptr(priv->percpu_priv, i);
2447 
2448 		percpu_priv->np.down = 0;
2449 		napi_enable(&percpu_priv->np.napi);
2450 	}
2451 }
2452 
2453 static void dpaa_eth_napi_disable(struct dpaa_priv *priv)
2454 {
2455 	struct dpaa_percpu_priv *percpu_priv;
2456 	int i;
2457 
2458 	for_each_possible_cpu(i) {
2459 		percpu_priv = per_cpu_ptr(priv->percpu_priv, i);
2460 
2461 		percpu_priv->np.down = 1;
2462 		napi_disable(&percpu_priv->np.napi);
2463 	}
2464 }
2465 
2466 static void dpaa_adjust_link(struct net_device *net_dev)
2467 {
2468 	struct mac_device *mac_dev;
2469 	struct dpaa_priv *priv;
2470 
2471 	priv = netdev_priv(net_dev);
2472 	mac_dev = priv->mac_dev;
2473 	mac_dev->adjust_link(mac_dev);
2474 }
2475 
2476 static int dpaa_phy_init(struct net_device *net_dev)
2477 {
2478 	struct mac_device *mac_dev;
2479 	struct phy_device *phy_dev;
2480 	struct dpaa_priv *priv;
2481 
2482 	priv = netdev_priv(net_dev);
2483 	mac_dev = priv->mac_dev;
2484 
2485 	phy_dev = of_phy_connect(net_dev, mac_dev->phy_node,
2486 				 &dpaa_adjust_link, 0,
2487 				 mac_dev->phy_if);
2488 	if (!phy_dev) {
2489 		netif_err(priv, ifup, net_dev, "init_phy() failed\n");
2490 		return -ENODEV;
2491 	}
2492 
2493 	/* Remove any features not supported by the controller */
2494 	phy_dev->supported &= mac_dev->if_support;
2495 	phy_support_asym_pause(phy_dev);
2496 
2497 	mac_dev->phy_dev = phy_dev;
2498 	net_dev->phydev = phy_dev;
2499 
2500 	return 0;
2501 }
2502 
2503 static int dpaa_open(struct net_device *net_dev)
2504 {
2505 	struct mac_device *mac_dev;
2506 	struct dpaa_priv *priv;
2507 	int err, i;
2508 
2509 	priv = netdev_priv(net_dev);
2510 	mac_dev = priv->mac_dev;
2511 	dpaa_eth_napi_enable(priv);
2512 
2513 	err = dpaa_phy_init(net_dev);
2514 	if (err)
2515 		goto phy_init_failed;
2516 
2517 	for (i = 0; i < ARRAY_SIZE(mac_dev->port); i++) {
2518 		err = fman_port_enable(mac_dev->port[i]);
2519 		if (err)
2520 			goto mac_start_failed;
2521 	}
2522 
2523 	err = priv->mac_dev->start(mac_dev);
2524 	if (err < 0) {
2525 		netif_err(priv, ifup, net_dev, "mac_dev->start() = %d\n", err);
2526 		goto mac_start_failed;
2527 	}
2528 
2529 	netif_tx_start_all_queues(net_dev);
2530 
2531 	return 0;
2532 
2533 mac_start_failed:
2534 	for (i = 0; i < ARRAY_SIZE(mac_dev->port); i++)
2535 		fman_port_disable(mac_dev->port[i]);
2536 
2537 phy_init_failed:
2538 	dpaa_eth_napi_disable(priv);
2539 
2540 	return err;
2541 }
2542 
2543 static int dpaa_eth_stop(struct net_device *net_dev)
2544 {
2545 	struct dpaa_priv *priv;
2546 	int err;
2547 
2548 	err = dpaa_stop(net_dev);
2549 
2550 	priv = netdev_priv(net_dev);
2551 	dpaa_eth_napi_disable(priv);
2552 
2553 	return err;
2554 }
2555 
2556 static int dpaa_ts_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
2557 {
2558 	struct dpaa_priv *priv = netdev_priv(dev);
2559 	struct hwtstamp_config config;
2560 
2561 	if (copy_from_user(&config, rq->ifr_data, sizeof(config)))
2562 		return -EFAULT;
2563 
2564 	switch (config.tx_type) {
2565 	case HWTSTAMP_TX_OFF:
2566 		/* Couldn't disable rx/tx timestamping separately.
2567 		 * Do nothing here.
2568 		 */
2569 		priv->tx_tstamp = false;
2570 		break;
2571 	case HWTSTAMP_TX_ON:
2572 		priv->mac_dev->set_tstamp(priv->mac_dev->fman_mac, true);
2573 		priv->tx_tstamp = true;
2574 		break;
2575 	default:
2576 		return -ERANGE;
2577 	}
2578 
2579 	if (config.rx_filter == HWTSTAMP_FILTER_NONE) {
2580 		/* Couldn't disable rx/tx timestamping separately.
2581 		 * Do nothing here.
2582 		 */
2583 		priv->rx_tstamp = false;
2584 	} else {
2585 		priv->mac_dev->set_tstamp(priv->mac_dev->fman_mac, true);
2586 		priv->rx_tstamp = true;
2587 		/* TS is set for all frame types, not only those requested */
2588 		config.rx_filter = HWTSTAMP_FILTER_ALL;
2589 	}
2590 
2591 	return copy_to_user(rq->ifr_data, &config, sizeof(config)) ?
2592 			-EFAULT : 0;
2593 }
2594 
2595 static int dpaa_ioctl(struct net_device *net_dev, struct ifreq *rq, int cmd)
2596 {
2597 	int ret = -EINVAL;
2598 
2599 	if (cmd == SIOCGMIIREG) {
2600 		if (net_dev->phydev)
2601 			return phy_mii_ioctl(net_dev->phydev, rq, cmd);
2602 	}
2603 
2604 	if (cmd == SIOCSHWTSTAMP)
2605 		return dpaa_ts_ioctl(net_dev, rq, cmd);
2606 
2607 	return ret;
2608 }
2609 
2610 static const struct net_device_ops dpaa_ops = {
2611 	.ndo_open = dpaa_open,
2612 	.ndo_start_xmit = dpaa_start_xmit,
2613 	.ndo_stop = dpaa_eth_stop,
2614 	.ndo_tx_timeout = dpaa_tx_timeout,
2615 	.ndo_get_stats64 = dpaa_get_stats64,
2616 	.ndo_set_mac_address = dpaa_set_mac_address,
2617 	.ndo_validate_addr = eth_validate_addr,
2618 	.ndo_set_rx_mode = dpaa_set_rx_mode,
2619 	.ndo_do_ioctl = dpaa_ioctl,
2620 	.ndo_setup_tc = dpaa_setup_tc,
2621 };
2622 
2623 static int dpaa_napi_add(struct net_device *net_dev)
2624 {
2625 	struct dpaa_priv *priv = netdev_priv(net_dev);
2626 	struct dpaa_percpu_priv *percpu_priv;
2627 	int cpu;
2628 
2629 	for_each_possible_cpu(cpu) {
2630 		percpu_priv = per_cpu_ptr(priv->percpu_priv, cpu);
2631 
2632 		netif_napi_add(net_dev, &percpu_priv->np.napi,
2633 			       dpaa_eth_poll, NAPI_POLL_WEIGHT);
2634 	}
2635 
2636 	return 0;
2637 }
2638 
2639 static void dpaa_napi_del(struct net_device *net_dev)
2640 {
2641 	struct dpaa_priv *priv = netdev_priv(net_dev);
2642 	struct dpaa_percpu_priv *percpu_priv;
2643 	int cpu;
2644 
2645 	for_each_possible_cpu(cpu) {
2646 		percpu_priv = per_cpu_ptr(priv->percpu_priv, cpu);
2647 
2648 		netif_napi_del(&percpu_priv->np.napi);
2649 	}
2650 }
2651 
2652 static inline void dpaa_bp_free_pf(const struct dpaa_bp *bp,
2653 				   struct bm_buffer *bmb)
2654 {
2655 	dma_addr_t addr = bm_buf_addr(bmb);
2656 
2657 	dma_unmap_single(bp->dev, addr, bp->size, DMA_FROM_DEVICE);
2658 
2659 	skb_free_frag(phys_to_virt(addr));
2660 }
2661 
2662 /* Alloc the dpaa_bp struct and configure default values */
2663 static struct dpaa_bp *dpaa_bp_alloc(struct device *dev)
2664 {
2665 	struct dpaa_bp *dpaa_bp;
2666 
2667 	dpaa_bp = devm_kzalloc(dev, sizeof(*dpaa_bp), GFP_KERNEL);
2668 	if (!dpaa_bp)
2669 		return ERR_PTR(-ENOMEM);
2670 
2671 	dpaa_bp->bpid = FSL_DPAA_BPID_INV;
2672 	dpaa_bp->percpu_count = devm_alloc_percpu(dev, *dpaa_bp->percpu_count);
2673 	if (!dpaa_bp->percpu_count)
2674 		return ERR_PTR(-ENOMEM);
2675 
2676 	dpaa_bp->config_count = FSL_DPAA_ETH_MAX_BUF_COUNT;
2677 
2678 	dpaa_bp->seed_cb = dpaa_bp_seed;
2679 	dpaa_bp->free_buf_cb = dpaa_bp_free_pf;
2680 
2681 	return dpaa_bp;
2682 }
2683 
2684 /* Place all ingress FQs (Rx Default, Rx Error) in a dedicated CGR.
2685  * We won't be sending congestion notifications to FMan; for now, we just use
2686  * this CGR to generate enqueue rejections to FMan in order to drop the frames
2687  * before they reach our ingress queues and eat up memory.
2688  */
2689 static int dpaa_ingress_cgr_init(struct dpaa_priv *priv)
2690 {
2691 	struct qm_mcc_initcgr initcgr;
2692 	u32 cs_th;
2693 	int err;
2694 
2695 	err = qman_alloc_cgrid(&priv->ingress_cgr.cgrid);
2696 	if (err < 0) {
2697 		if (netif_msg_drv(priv))
2698 			pr_err("Error %d allocating CGR ID\n", err);
2699 		goto out_error;
2700 	}
2701 
2702 	/* Enable CS TD, but disable Congestion State Change Notifications. */
2703 	memset(&initcgr, 0, sizeof(initcgr));
2704 	initcgr.we_mask = cpu_to_be16(QM_CGR_WE_CS_THRES);
2705 	initcgr.cgr.cscn_en = QM_CGR_EN;
2706 	cs_th = DPAA_INGRESS_CS_THRESHOLD;
2707 	qm_cgr_cs_thres_set64(&initcgr.cgr.cs_thres, cs_th, 1);
2708 
2709 	initcgr.we_mask |= cpu_to_be16(QM_CGR_WE_CSTD_EN);
2710 	initcgr.cgr.cstd_en = QM_CGR_EN;
2711 
2712 	/* This CGR will be associated with the SWP affined to the current CPU.
2713 	 * However, we'll place all our ingress FQs in it.
2714 	 */
2715 	err = qman_create_cgr(&priv->ingress_cgr, QMAN_CGR_FLAG_USE_INIT,
2716 			      &initcgr);
2717 	if (err < 0) {
2718 		if (netif_msg_drv(priv))
2719 			pr_err("Error %d creating ingress CGR with ID %d\n",
2720 			       err, priv->ingress_cgr.cgrid);
2721 		qman_release_cgrid(priv->ingress_cgr.cgrid);
2722 		goto out_error;
2723 	}
2724 	if (netif_msg_drv(priv))
2725 		pr_debug("Created ingress CGR %d for netdev with hwaddr %pM\n",
2726 			 priv->ingress_cgr.cgrid, priv->mac_dev->addr);
2727 
2728 	priv->use_ingress_cgr = true;
2729 
2730 out_error:
2731 	return err;
2732 }
2733 
2734 static inline u16 dpaa_get_headroom(struct dpaa_buffer_layout *bl)
2735 {
2736 	u16 headroom;
2737 
2738 	/* The frame headroom must accommodate:
2739 	 * - the driver private data area
2740 	 * - parse results, hash results, timestamp if selected
2741 	 * If either hash results or time stamp are selected, both will
2742 	 * be copied to/from the frame headroom, as TS is located between PR and
2743 	 * HR in the IC and IC copy size has a granularity of 16bytes
2744 	 * (see description of FMBM_RICP and FMBM_TICP registers in DPAARM)
2745 	 *
2746 	 * Also make sure the headroom is a multiple of data_align bytes
2747 	 */
2748 	headroom = (u16)(bl->priv_data_size + DPAA_PARSE_RESULTS_SIZE +
2749 		DPAA_TIME_STAMP_SIZE + DPAA_HASH_RESULTS_SIZE);
2750 
2751 	return DPAA_FD_DATA_ALIGNMENT ? ALIGN(headroom,
2752 					      DPAA_FD_DATA_ALIGNMENT) :
2753 					headroom;
2754 }
2755 
2756 static int dpaa_eth_probe(struct platform_device *pdev)
2757 {
2758 	struct dpaa_bp *dpaa_bps[DPAA_BPS_NUM] = {NULL};
2759 	struct net_device *net_dev = NULL;
2760 	struct dpaa_fq *dpaa_fq, *tmp;
2761 	struct dpaa_priv *priv = NULL;
2762 	struct fm_port_fqs port_fqs;
2763 	struct mac_device *mac_dev;
2764 	int err = 0, i, channel;
2765 	struct device *dev;
2766 
2767 	/* device used for DMA mapping */
2768 	dev = pdev->dev.parent;
2769 	err = dma_coerce_mask_and_coherent(dev, DMA_BIT_MASK(40));
2770 	if (err) {
2771 		dev_err(dev, "dma_coerce_mask_and_coherent() failed\n");
2772 		return err;
2773 	}
2774 
2775 	/* Allocate this early, so we can store relevant information in
2776 	 * the private area
2777 	 */
2778 	net_dev = alloc_etherdev_mq(sizeof(*priv), DPAA_ETH_TXQ_NUM);
2779 	if (!net_dev) {
2780 		dev_err(dev, "alloc_etherdev_mq() failed\n");
2781 		return -ENOMEM;
2782 	}
2783 
2784 	/* Do this here, so we can be verbose early */
2785 	SET_NETDEV_DEV(net_dev, dev);
2786 	dev_set_drvdata(dev, net_dev);
2787 
2788 	priv = netdev_priv(net_dev);
2789 	priv->net_dev = net_dev;
2790 
2791 	priv->msg_enable = netif_msg_init(debug, DPAA_MSG_DEFAULT);
2792 
2793 	mac_dev = dpaa_mac_dev_get(pdev);
2794 	if (IS_ERR(mac_dev)) {
2795 		dev_err(dev, "dpaa_mac_dev_get() failed\n");
2796 		err = PTR_ERR(mac_dev);
2797 		goto free_netdev;
2798 	}
2799 
2800 	/* If fsl_fm_max_frm is set to a higher value than the all-common 1500,
2801 	 * we choose conservatively and let the user explicitly set a higher
2802 	 * MTU via ifconfig. Otherwise, the user may end up with different MTUs
2803 	 * in the same LAN.
2804 	 * If on the other hand fsl_fm_max_frm has been chosen below 1500,
2805 	 * start with the maximum allowed.
2806 	 */
2807 	net_dev->mtu = min(dpaa_get_max_mtu(), ETH_DATA_LEN);
2808 
2809 	netdev_dbg(net_dev, "Setting initial MTU on net device: %d\n",
2810 		   net_dev->mtu);
2811 
2812 	priv->buf_layout[RX].priv_data_size = DPAA_RX_PRIV_DATA_SIZE; /* Rx */
2813 	priv->buf_layout[TX].priv_data_size = DPAA_TX_PRIV_DATA_SIZE; /* Tx */
2814 
2815 	/* bp init */
2816 	for (i = 0; i < DPAA_BPS_NUM; i++) {
2817 		dpaa_bps[i] = dpaa_bp_alloc(dev);
2818 		if (IS_ERR(dpaa_bps[i])) {
2819 			err = PTR_ERR(dpaa_bps[i]);
2820 			goto free_dpaa_bps;
2821 		}
2822 		/* the raw size of the buffers used for reception */
2823 		dpaa_bps[i]->raw_size = bpool_buffer_raw_size(i, DPAA_BPS_NUM);
2824 		/* avoid runtime computations by keeping the usable size here */
2825 		dpaa_bps[i]->size = dpaa_bp_size(dpaa_bps[i]->raw_size);
2826 		dpaa_bps[i]->dev = dev;
2827 
2828 		err = dpaa_bp_alloc_pool(dpaa_bps[i]);
2829 		if (err < 0)
2830 			goto free_dpaa_bps;
2831 		priv->dpaa_bps[i] = dpaa_bps[i];
2832 	}
2833 
2834 	INIT_LIST_HEAD(&priv->dpaa_fq_list);
2835 
2836 	memset(&port_fqs, 0, sizeof(port_fqs));
2837 
2838 	err = dpaa_alloc_all_fqs(dev, &priv->dpaa_fq_list, &port_fqs);
2839 	if (err < 0) {
2840 		dev_err(dev, "dpaa_alloc_all_fqs() failed\n");
2841 		goto free_dpaa_bps;
2842 	}
2843 
2844 	priv->mac_dev = mac_dev;
2845 
2846 	channel = dpaa_get_channel();
2847 	if (channel < 0) {
2848 		dev_err(dev, "dpaa_get_channel() failed\n");
2849 		err = channel;
2850 		goto free_dpaa_bps;
2851 	}
2852 
2853 	priv->channel = (u16)channel;
2854 
2855 	/* Walk the CPUs with affine portals
2856 	 * and add this pool channel to each's dequeue mask.
2857 	 */
2858 	dpaa_eth_add_channel(priv->channel);
2859 
2860 	dpaa_fq_setup(priv, &dpaa_fq_cbs, priv->mac_dev->port[TX]);
2861 
2862 	/* Create a congestion group for this netdev, with
2863 	 * dynamically-allocated CGR ID.
2864 	 * Must be executed after probing the MAC, but before
2865 	 * assigning the egress FQs to the CGRs.
2866 	 */
2867 	err = dpaa_eth_cgr_init(priv);
2868 	if (err < 0) {
2869 		dev_err(dev, "Error initializing CGR\n");
2870 		goto free_dpaa_bps;
2871 	}
2872 
2873 	err = dpaa_ingress_cgr_init(priv);
2874 	if (err < 0) {
2875 		dev_err(dev, "Error initializing ingress CGR\n");
2876 		goto delete_egress_cgr;
2877 	}
2878 
2879 	/* Add the FQs to the interface, and make them active */
2880 	list_for_each_entry_safe(dpaa_fq, tmp, &priv->dpaa_fq_list, list) {
2881 		err = dpaa_fq_init(dpaa_fq, false);
2882 		if (err < 0)
2883 			goto free_dpaa_fqs;
2884 	}
2885 
2886 	priv->tx_headroom = dpaa_get_headroom(&priv->buf_layout[TX]);
2887 	priv->rx_headroom = dpaa_get_headroom(&priv->buf_layout[RX]);
2888 
2889 	/* All real interfaces need their ports initialized */
2890 	err = dpaa_eth_init_ports(mac_dev, dpaa_bps, DPAA_BPS_NUM, &port_fqs,
2891 				  &priv->buf_layout[0], dev);
2892 	if (err)
2893 		goto free_dpaa_fqs;
2894 
2895 	/* Rx traffic distribution based on keygen hashing defaults to on */
2896 	priv->keygen_in_use = true;
2897 
2898 	priv->percpu_priv = devm_alloc_percpu(dev, *priv->percpu_priv);
2899 	if (!priv->percpu_priv) {
2900 		dev_err(dev, "devm_alloc_percpu() failed\n");
2901 		err = -ENOMEM;
2902 		goto free_dpaa_fqs;
2903 	}
2904 
2905 	priv->num_tc = 1;
2906 	netif_set_real_num_tx_queues(net_dev, priv->num_tc * DPAA_TC_TXQ_NUM);
2907 
2908 	/* Initialize NAPI */
2909 	err = dpaa_napi_add(net_dev);
2910 	if (err < 0)
2911 		goto delete_dpaa_napi;
2912 
2913 	err = dpaa_netdev_init(net_dev, &dpaa_ops, tx_timeout);
2914 	if (err < 0)
2915 		goto delete_dpaa_napi;
2916 
2917 	dpaa_eth_sysfs_init(&net_dev->dev);
2918 
2919 	netif_info(priv, probe, net_dev, "Probed interface %s\n",
2920 		   net_dev->name);
2921 
2922 	return 0;
2923 
2924 delete_dpaa_napi:
2925 	dpaa_napi_del(net_dev);
2926 free_dpaa_fqs:
2927 	dpaa_fq_free(dev, &priv->dpaa_fq_list);
2928 	qman_delete_cgr_safe(&priv->ingress_cgr);
2929 	qman_release_cgrid(priv->ingress_cgr.cgrid);
2930 delete_egress_cgr:
2931 	qman_delete_cgr_safe(&priv->cgr_data.cgr);
2932 	qman_release_cgrid(priv->cgr_data.cgr.cgrid);
2933 free_dpaa_bps:
2934 	dpaa_bps_free(priv);
2935 free_netdev:
2936 	dev_set_drvdata(dev, NULL);
2937 	free_netdev(net_dev);
2938 
2939 	return err;
2940 }
2941 
2942 static int dpaa_remove(struct platform_device *pdev)
2943 {
2944 	struct net_device *net_dev;
2945 	struct dpaa_priv *priv;
2946 	struct device *dev;
2947 	int err;
2948 
2949 	dev = pdev->dev.parent;
2950 	net_dev = dev_get_drvdata(dev);
2951 
2952 	priv = netdev_priv(net_dev);
2953 
2954 	dpaa_eth_sysfs_remove(dev);
2955 
2956 	dev_set_drvdata(dev, NULL);
2957 	unregister_netdev(net_dev);
2958 
2959 	err = dpaa_fq_free(dev, &priv->dpaa_fq_list);
2960 
2961 	qman_delete_cgr_safe(&priv->ingress_cgr);
2962 	qman_release_cgrid(priv->ingress_cgr.cgrid);
2963 	qman_delete_cgr_safe(&priv->cgr_data.cgr);
2964 	qman_release_cgrid(priv->cgr_data.cgr.cgrid);
2965 
2966 	dpaa_napi_del(net_dev);
2967 
2968 	dpaa_bps_free(priv);
2969 
2970 	free_netdev(net_dev);
2971 
2972 	return err;
2973 }
2974 
2975 static const struct platform_device_id dpaa_devtype[] = {
2976 	{
2977 		.name = "dpaa-ethernet",
2978 		.driver_data = 0,
2979 	}, {
2980 	}
2981 };
2982 MODULE_DEVICE_TABLE(platform, dpaa_devtype);
2983 
2984 static struct platform_driver dpaa_driver = {
2985 	.driver = {
2986 		.name = KBUILD_MODNAME,
2987 	},
2988 	.id_table = dpaa_devtype,
2989 	.probe = dpaa_eth_probe,
2990 	.remove = dpaa_remove
2991 };
2992 
2993 static int __init dpaa_load(void)
2994 {
2995 	int err;
2996 
2997 	pr_debug("FSL DPAA Ethernet driver\n");
2998 
2999 	/* initialize dpaa_eth mirror values */
3000 	dpaa_rx_extra_headroom = fman_get_rx_extra_headroom();
3001 	dpaa_max_frm = fman_get_max_frm();
3002 
3003 	err = platform_driver_register(&dpaa_driver);
3004 	if (err < 0)
3005 		pr_err("Error, platform_driver_register() = %d\n", err);
3006 
3007 	return err;
3008 }
3009 module_init(dpaa_load);
3010 
3011 static void __exit dpaa_unload(void)
3012 {
3013 	platform_driver_unregister(&dpaa_driver);
3014 
3015 	/* Only one channel is used and needs to be released after all
3016 	 * interfaces are removed
3017 	 */
3018 	dpaa_release_channel();
3019 }
3020 module_exit(dpaa_unload);
3021 
3022 MODULE_LICENSE("Dual BSD/GPL");
3023 MODULE_DESCRIPTION("FSL DPAA Ethernet driver");
3024