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 <linux/phy_fixed.h>
55 #include <soc/fsl/bman.h>
56 #include <soc/fsl/qman.h>
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 #define DPAA_BP_RAW_SIZE 4096
182 
183 #define dpaa_bp_size(raw_size) SKB_WITH_OVERHEAD(raw_size)
184 
185 static int dpaa_max_frm;
186 
187 static int dpaa_rx_extra_headroom;
188 
189 #define dpaa_get_max_mtu()	\
190 	(dpaa_max_frm - (VLAN_ETH_HLEN + ETH_FCS_LEN))
191 
192 static int dpaa_netdev_init(struct net_device *net_dev,
193 			    const struct net_device_ops *dpaa_ops,
194 			    u16 tx_timeout)
195 {
196 	struct dpaa_priv *priv = netdev_priv(net_dev);
197 	struct device *dev = net_dev->dev.parent;
198 	struct dpaa_percpu_priv *percpu_priv;
199 	const u8 *mac_addr;
200 	int i, err;
201 
202 	/* Although we access another CPU's private data here
203 	 * we do it at initialization so it is safe
204 	 */
205 	for_each_possible_cpu(i) {
206 		percpu_priv = per_cpu_ptr(priv->percpu_priv, i);
207 		percpu_priv->net_dev = net_dev;
208 	}
209 
210 	net_dev->netdev_ops = dpaa_ops;
211 	mac_addr = priv->mac_dev->addr;
212 
213 	net_dev->mem_start = priv->mac_dev->res->start;
214 	net_dev->mem_end = priv->mac_dev->res->end;
215 
216 	net_dev->min_mtu = ETH_MIN_MTU;
217 	net_dev->max_mtu = dpaa_get_max_mtu();
218 
219 	net_dev->hw_features |= (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
220 				 NETIF_F_LLTX | NETIF_F_RXHASH);
221 
222 	net_dev->hw_features |= NETIF_F_SG | NETIF_F_HIGHDMA;
223 	/* The kernels enables GSO automatically, if we declare NETIF_F_SG.
224 	 * For conformity, we'll still declare GSO explicitly.
225 	 */
226 	net_dev->features |= NETIF_F_GSO;
227 	net_dev->features |= NETIF_F_RXCSUM;
228 
229 	net_dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
230 	/* we do not want shared skbs on TX */
231 	net_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
232 
233 	net_dev->features |= net_dev->hw_features;
234 	net_dev->vlan_features = net_dev->features;
235 
236 	memcpy(net_dev->perm_addr, mac_addr, net_dev->addr_len);
237 	memcpy(net_dev->dev_addr, mac_addr, net_dev->addr_len);
238 
239 	net_dev->ethtool_ops = &dpaa_ethtool_ops;
240 
241 	net_dev->needed_headroom = priv->tx_headroom;
242 	net_dev->watchdog_timeo = msecs_to_jiffies(tx_timeout);
243 
244 	/* start without the RUNNING flag, phylib controls it later */
245 	netif_carrier_off(net_dev);
246 
247 	err = register_netdev(net_dev);
248 	if (err < 0) {
249 		dev_err(dev, "register_netdev() = %d\n", err);
250 		return err;
251 	}
252 
253 	return 0;
254 }
255 
256 static int dpaa_stop(struct net_device *net_dev)
257 {
258 	struct mac_device *mac_dev;
259 	struct dpaa_priv *priv;
260 	int i, err, error;
261 
262 	priv = netdev_priv(net_dev);
263 	mac_dev = priv->mac_dev;
264 
265 	netif_tx_stop_all_queues(net_dev);
266 	/* Allow the Fman (Tx) port to process in-flight frames before we
267 	 * try switching it off.
268 	 */
269 	msleep(200);
270 
271 	err = mac_dev->stop(mac_dev);
272 	if (err < 0)
273 		netif_err(priv, ifdown, net_dev, "mac_dev->stop() = %d\n",
274 			  err);
275 
276 	for (i = 0; i < ARRAY_SIZE(mac_dev->port); i++) {
277 		error = fman_port_disable(mac_dev->port[i]);
278 		if (error)
279 			err = error;
280 	}
281 
282 	if (net_dev->phydev)
283 		phy_disconnect(net_dev->phydev);
284 	net_dev->phydev = NULL;
285 
286 	msleep(200);
287 
288 	return err;
289 }
290 
291 static void dpaa_tx_timeout(struct net_device *net_dev, unsigned int txqueue)
292 {
293 	struct dpaa_percpu_priv *percpu_priv;
294 	const struct dpaa_priv	*priv;
295 
296 	priv = netdev_priv(net_dev);
297 	percpu_priv = this_cpu_ptr(priv->percpu_priv);
298 
299 	netif_crit(priv, timer, net_dev, "Transmit timeout latency: %u ms\n",
300 		   jiffies_to_msecs(jiffies - dev_trans_start(net_dev)));
301 
302 	percpu_priv->stats.tx_errors++;
303 }
304 
305 /* Calculates the statistics for the given device by adding the statistics
306  * collected by each CPU.
307  */
308 static void dpaa_get_stats64(struct net_device *net_dev,
309 			     struct rtnl_link_stats64 *s)
310 {
311 	int numstats = sizeof(struct rtnl_link_stats64) / sizeof(u64);
312 	struct dpaa_priv *priv = netdev_priv(net_dev);
313 	struct dpaa_percpu_priv *percpu_priv;
314 	u64 *netstats = (u64 *)s;
315 	u64 *cpustats;
316 	int i, j;
317 
318 	for_each_possible_cpu(i) {
319 		percpu_priv = per_cpu_ptr(priv->percpu_priv, i);
320 
321 		cpustats = (u64 *)&percpu_priv->stats;
322 
323 		/* add stats from all CPUs */
324 		for (j = 0; j < numstats; j++)
325 			netstats[j] += cpustats[j];
326 	}
327 }
328 
329 static int dpaa_setup_tc(struct net_device *net_dev, enum tc_setup_type type,
330 			 void *type_data)
331 {
332 	struct dpaa_priv *priv = netdev_priv(net_dev);
333 	struct tc_mqprio_qopt *mqprio = type_data;
334 	u8 num_tc;
335 	int i;
336 
337 	if (type != TC_SETUP_QDISC_MQPRIO)
338 		return -EOPNOTSUPP;
339 
340 	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
341 	num_tc = mqprio->num_tc;
342 
343 	if (num_tc == priv->num_tc)
344 		return 0;
345 
346 	if (!num_tc) {
347 		netdev_reset_tc(net_dev);
348 		goto out;
349 	}
350 
351 	if (num_tc > DPAA_TC_NUM) {
352 		netdev_err(net_dev, "Too many traffic classes: max %d supported.\n",
353 			   DPAA_TC_NUM);
354 		return -EINVAL;
355 	}
356 
357 	netdev_set_num_tc(net_dev, num_tc);
358 
359 	for (i = 0; i < num_tc; i++)
360 		netdev_set_tc_queue(net_dev, i, DPAA_TC_TXQ_NUM,
361 				    i * DPAA_TC_TXQ_NUM);
362 
363 out:
364 	priv->num_tc = num_tc ? : 1;
365 	netif_set_real_num_tx_queues(net_dev, priv->num_tc * DPAA_TC_TXQ_NUM);
366 	return 0;
367 }
368 
369 static struct mac_device *dpaa_mac_dev_get(struct platform_device *pdev)
370 {
371 	struct dpaa_eth_data *eth_data;
372 	struct device *dpaa_dev;
373 	struct mac_device *mac_dev;
374 
375 	dpaa_dev = &pdev->dev;
376 	eth_data = dpaa_dev->platform_data;
377 	if (!eth_data) {
378 		dev_err(dpaa_dev, "eth_data missing\n");
379 		return ERR_PTR(-ENODEV);
380 	}
381 	mac_dev = eth_data->mac_dev;
382 	if (!mac_dev) {
383 		dev_err(dpaa_dev, "mac_dev missing\n");
384 		return ERR_PTR(-EINVAL);
385 	}
386 
387 	return mac_dev;
388 }
389 
390 static int dpaa_set_mac_address(struct net_device *net_dev, void *addr)
391 {
392 	const struct dpaa_priv *priv;
393 	struct mac_device *mac_dev;
394 	struct sockaddr old_addr;
395 	int err;
396 
397 	priv = netdev_priv(net_dev);
398 
399 	memcpy(old_addr.sa_data, net_dev->dev_addr,  ETH_ALEN);
400 
401 	err = eth_mac_addr(net_dev, addr);
402 	if (err < 0) {
403 		netif_err(priv, drv, net_dev, "eth_mac_addr() = %d\n", err);
404 		return err;
405 	}
406 
407 	mac_dev = priv->mac_dev;
408 
409 	err = mac_dev->change_addr(mac_dev->fman_mac,
410 				   (enet_addr_t *)net_dev->dev_addr);
411 	if (err < 0) {
412 		netif_err(priv, drv, net_dev, "mac_dev->change_addr() = %d\n",
413 			  err);
414 		/* reverting to previous address */
415 		eth_mac_addr(net_dev, &old_addr);
416 
417 		return err;
418 	}
419 
420 	return 0;
421 }
422 
423 static void dpaa_set_rx_mode(struct net_device *net_dev)
424 {
425 	const struct dpaa_priv	*priv;
426 	int err;
427 
428 	priv = netdev_priv(net_dev);
429 
430 	if (!!(net_dev->flags & IFF_PROMISC) != priv->mac_dev->promisc) {
431 		priv->mac_dev->promisc = !priv->mac_dev->promisc;
432 		err = priv->mac_dev->set_promisc(priv->mac_dev->fman_mac,
433 						 priv->mac_dev->promisc);
434 		if (err < 0)
435 			netif_err(priv, drv, net_dev,
436 				  "mac_dev->set_promisc() = %d\n",
437 				  err);
438 	}
439 
440 	if (!!(net_dev->flags & IFF_ALLMULTI) != priv->mac_dev->allmulti) {
441 		priv->mac_dev->allmulti = !priv->mac_dev->allmulti;
442 		err = priv->mac_dev->set_allmulti(priv->mac_dev->fman_mac,
443 						  priv->mac_dev->allmulti);
444 		if (err < 0)
445 			netif_err(priv, drv, net_dev,
446 				  "mac_dev->set_allmulti() = %d\n",
447 				  err);
448 	}
449 
450 	err = priv->mac_dev->set_multi(net_dev, priv->mac_dev);
451 	if (err < 0)
452 		netif_err(priv, drv, net_dev, "mac_dev->set_multi() = %d\n",
453 			  err);
454 }
455 
456 static struct dpaa_bp *dpaa_bpid2pool(int bpid)
457 {
458 	if (WARN_ON(bpid < 0 || bpid >= BM_MAX_NUM_OF_POOLS))
459 		return NULL;
460 
461 	return dpaa_bp_array[bpid];
462 }
463 
464 /* checks if this bpool is already allocated */
465 static bool dpaa_bpid2pool_use(int bpid)
466 {
467 	if (dpaa_bpid2pool(bpid)) {
468 		refcount_inc(&dpaa_bp_array[bpid]->refs);
469 		return true;
470 	}
471 
472 	return false;
473 }
474 
475 /* called only once per bpid by dpaa_bp_alloc_pool() */
476 static void dpaa_bpid2pool_map(int bpid, struct dpaa_bp *dpaa_bp)
477 {
478 	dpaa_bp_array[bpid] = dpaa_bp;
479 	refcount_set(&dpaa_bp->refs, 1);
480 }
481 
482 static int dpaa_bp_alloc_pool(struct dpaa_bp *dpaa_bp)
483 {
484 	int err;
485 
486 	if (dpaa_bp->size == 0 || dpaa_bp->config_count == 0) {
487 		pr_err("%s: Buffer pool is not properly initialized! Missing size or initial number of buffers\n",
488 		       __func__);
489 		return -EINVAL;
490 	}
491 
492 	/* If the pool is already specified, we only create one per bpid */
493 	if (dpaa_bp->bpid != FSL_DPAA_BPID_INV &&
494 	    dpaa_bpid2pool_use(dpaa_bp->bpid))
495 		return 0;
496 
497 	if (dpaa_bp->bpid == FSL_DPAA_BPID_INV) {
498 		dpaa_bp->pool = bman_new_pool();
499 		if (!dpaa_bp->pool) {
500 			pr_err("%s: bman_new_pool() failed\n",
501 			       __func__);
502 			return -ENODEV;
503 		}
504 
505 		dpaa_bp->bpid = (u8)bman_get_bpid(dpaa_bp->pool);
506 	}
507 
508 	if (dpaa_bp->seed_cb) {
509 		err = dpaa_bp->seed_cb(dpaa_bp);
510 		if (err)
511 			goto pool_seed_failed;
512 	}
513 
514 	dpaa_bpid2pool_map(dpaa_bp->bpid, dpaa_bp);
515 
516 	return 0;
517 
518 pool_seed_failed:
519 	pr_err("%s: pool seeding failed\n", __func__);
520 	bman_free_pool(dpaa_bp->pool);
521 
522 	return err;
523 }
524 
525 /* remove and free all the buffers from the given buffer pool */
526 static void dpaa_bp_drain(struct dpaa_bp *bp)
527 {
528 	u8 num = 8;
529 	int ret;
530 
531 	do {
532 		struct bm_buffer bmb[8];
533 		int i;
534 
535 		ret = bman_acquire(bp->pool, bmb, num);
536 		if (ret < 0) {
537 			if (num == 8) {
538 				/* we have less than 8 buffers left;
539 				 * drain them one by one
540 				 */
541 				num = 1;
542 				ret = 1;
543 				continue;
544 			} else {
545 				/* Pool is fully drained */
546 				break;
547 			}
548 		}
549 
550 		if (bp->free_buf_cb)
551 			for (i = 0; i < num; i++)
552 				bp->free_buf_cb(bp, &bmb[i]);
553 	} while (ret > 0);
554 }
555 
556 static void dpaa_bp_free(struct dpaa_bp *dpaa_bp)
557 {
558 	struct dpaa_bp *bp = dpaa_bpid2pool(dpaa_bp->bpid);
559 
560 	/* the mapping between bpid and dpaa_bp is done very late in the
561 	 * allocation procedure; if something failed before the mapping, the bp
562 	 * was not configured, therefore we don't need the below instructions
563 	 */
564 	if (!bp)
565 		return;
566 
567 	if (!refcount_dec_and_test(&bp->refs))
568 		return;
569 
570 	if (bp->free_buf_cb)
571 		dpaa_bp_drain(bp);
572 
573 	dpaa_bp_array[bp->bpid] = NULL;
574 	bman_free_pool(bp->pool);
575 }
576 
577 static void dpaa_bps_free(struct dpaa_priv *priv)
578 {
579 	dpaa_bp_free(priv->dpaa_bp);
580 }
581 
582 /* Use multiple WQs for FQ assignment:
583  *	- Tx Confirmation queues go to WQ1.
584  *	- Rx Error and Tx Error queues go to WQ5 (giving them a better chance
585  *	  to be scheduled, in case there are many more FQs in WQ6).
586  *	- Rx Default goes to WQ6.
587  *	- Tx queues go to different WQs depending on their priority. Equal
588  *	  chunks of NR_CPUS queues go to WQ6 (lowest priority), WQ2, WQ1 and
589  *	  WQ0 (highest priority).
590  * This ensures that Tx-confirmed buffers are timely released. In particular,
591  * it avoids congestion on the Tx Confirm FQs, which can pile up PFDRs if they
592  * are greatly outnumbered by other FQs in the system, while
593  * dequeue scheduling is round-robin.
594  */
595 static inline void dpaa_assign_wq(struct dpaa_fq *fq, int idx)
596 {
597 	switch (fq->fq_type) {
598 	case FQ_TYPE_TX_CONFIRM:
599 	case FQ_TYPE_TX_CONF_MQ:
600 		fq->wq = 1;
601 		break;
602 	case FQ_TYPE_RX_ERROR:
603 	case FQ_TYPE_TX_ERROR:
604 		fq->wq = 5;
605 		break;
606 	case FQ_TYPE_RX_DEFAULT:
607 	case FQ_TYPE_RX_PCD:
608 		fq->wq = 6;
609 		break;
610 	case FQ_TYPE_TX:
611 		switch (idx / DPAA_TC_TXQ_NUM) {
612 		case 0:
613 			/* Low priority (best effort) */
614 			fq->wq = 6;
615 			break;
616 		case 1:
617 			/* Medium priority */
618 			fq->wq = 2;
619 			break;
620 		case 2:
621 			/* High priority */
622 			fq->wq = 1;
623 			break;
624 		case 3:
625 			/* Very high priority */
626 			fq->wq = 0;
627 			break;
628 		default:
629 			WARN(1, "Too many TX FQs: more than %d!\n",
630 			     DPAA_ETH_TXQ_NUM);
631 		}
632 		break;
633 	default:
634 		WARN(1, "Invalid FQ type %d for FQID %d!\n",
635 		     fq->fq_type, fq->fqid);
636 	}
637 }
638 
639 static struct dpaa_fq *dpaa_fq_alloc(struct device *dev,
640 				     u32 start, u32 count,
641 				     struct list_head *list,
642 				     enum dpaa_fq_type fq_type)
643 {
644 	struct dpaa_fq *dpaa_fq;
645 	int i;
646 
647 	dpaa_fq = devm_kcalloc(dev, count, sizeof(*dpaa_fq),
648 			       GFP_KERNEL);
649 	if (!dpaa_fq)
650 		return NULL;
651 
652 	for (i = 0; i < count; i++) {
653 		dpaa_fq[i].fq_type = fq_type;
654 		dpaa_fq[i].fqid = start ? start + i : 0;
655 		list_add_tail(&dpaa_fq[i].list, list);
656 	}
657 
658 	for (i = 0; i < count; i++)
659 		dpaa_assign_wq(dpaa_fq + i, i);
660 
661 	return dpaa_fq;
662 }
663 
664 static int dpaa_alloc_all_fqs(struct device *dev, struct list_head *list,
665 			      struct fm_port_fqs *port_fqs)
666 {
667 	struct dpaa_fq *dpaa_fq;
668 	u32 fq_base, fq_base_aligned, i;
669 
670 	dpaa_fq = dpaa_fq_alloc(dev, 0, 1, list, FQ_TYPE_RX_ERROR);
671 	if (!dpaa_fq)
672 		goto fq_alloc_failed;
673 
674 	port_fqs->rx_errq = &dpaa_fq[0];
675 
676 	dpaa_fq = dpaa_fq_alloc(dev, 0, 1, list, FQ_TYPE_RX_DEFAULT);
677 	if (!dpaa_fq)
678 		goto fq_alloc_failed;
679 
680 	port_fqs->rx_defq = &dpaa_fq[0];
681 
682 	/* the PCD FQIDs range needs to be aligned for correct operation */
683 	if (qman_alloc_fqid_range(&fq_base, 2 * DPAA_ETH_PCD_RXQ_NUM))
684 		goto fq_alloc_failed;
685 
686 	fq_base_aligned = ALIGN(fq_base, DPAA_ETH_PCD_RXQ_NUM);
687 
688 	for (i = fq_base; i < fq_base_aligned; i++)
689 		qman_release_fqid(i);
690 
691 	for (i = fq_base_aligned + DPAA_ETH_PCD_RXQ_NUM;
692 	     i < (fq_base + 2 * DPAA_ETH_PCD_RXQ_NUM); i++)
693 		qman_release_fqid(i);
694 
695 	dpaa_fq = dpaa_fq_alloc(dev, fq_base_aligned, DPAA_ETH_PCD_RXQ_NUM,
696 				list, FQ_TYPE_RX_PCD);
697 	if (!dpaa_fq)
698 		goto fq_alloc_failed;
699 
700 	port_fqs->rx_pcdq = &dpaa_fq[0];
701 
702 	if (!dpaa_fq_alloc(dev, 0, DPAA_ETH_TXQ_NUM, list, FQ_TYPE_TX_CONF_MQ))
703 		goto fq_alloc_failed;
704 
705 	dpaa_fq = dpaa_fq_alloc(dev, 0, 1, list, FQ_TYPE_TX_ERROR);
706 	if (!dpaa_fq)
707 		goto fq_alloc_failed;
708 
709 	port_fqs->tx_errq = &dpaa_fq[0];
710 
711 	dpaa_fq = dpaa_fq_alloc(dev, 0, 1, list, FQ_TYPE_TX_CONFIRM);
712 	if (!dpaa_fq)
713 		goto fq_alloc_failed;
714 
715 	port_fqs->tx_defq = &dpaa_fq[0];
716 
717 	if (!dpaa_fq_alloc(dev, 0, DPAA_ETH_TXQ_NUM, list, FQ_TYPE_TX))
718 		goto fq_alloc_failed;
719 
720 	return 0;
721 
722 fq_alloc_failed:
723 	dev_err(dev, "dpaa_fq_alloc() failed\n");
724 	return -ENOMEM;
725 }
726 
727 static u32 rx_pool_channel;
728 static DEFINE_SPINLOCK(rx_pool_channel_init);
729 
730 static int dpaa_get_channel(void)
731 {
732 	spin_lock(&rx_pool_channel_init);
733 	if (!rx_pool_channel) {
734 		u32 pool;
735 		int ret;
736 
737 		ret = qman_alloc_pool(&pool);
738 
739 		if (!ret)
740 			rx_pool_channel = pool;
741 	}
742 	spin_unlock(&rx_pool_channel_init);
743 	if (!rx_pool_channel)
744 		return -ENOMEM;
745 	return rx_pool_channel;
746 }
747 
748 static void dpaa_release_channel(void)
749 {
750 	qman_release_pool(rx_pool_channel);
751 }
752 
753 static void dpaa_eth_add_channel(u16 channel, struct device *dev)
754 {
755 	u32 pool = QM_SDQCR_CHANNELS_POOL_CONV(channel);
756 	const cpumask_t *cpus = qman_affine_cpus();
757 	struct qman_portal *portal;
758 	int cpu;
759 
760 	for_each_cpu_and(cpu, cpus, cpu_online_mask) {
761 		portal = qman_get_affine_portal(cpu);
762 		qman_p_static_dequeue_add(portal, pool);
763 		qman_start_using_portal(portal, dev);
764 	}
765 }
766 
767 /* Congestion group state change notification callback.
768  * Stops the device's egress queues while they are congested and
769  * wakes them upon exiting congested state.
770  * Also updates some CGR-related stats.
771  */
772 static void dpaa_eth_cgscn(struct qman_portal *qm, struct qman_cgr *cgr,
773 			   int congested)
774 {
775 	struct dpaa_priv *priv = (struct dpaa_priv *)container_of(cgr,
776 		struct dpaa_priv, cgr_data.cgr);
777 
778 	if (congested) {
779 		priv->cgr_data.congestion_start_jiffies = jiffies;
780 		netif_tx_stop_all_queues(priv->net_dev);
781 		priv->cgr_data.cgr_congested_count++;
782 	} else {
783 		priv->cgr_data.congested_jiffies +=
784 			(jiffies - priv->cgr_data.congestion_start_jiffies);
785 		netif_tx_wake_all_queues(priv->net_dev);
786 	}
787 }
788 
789 static int dpaa_eth_cgr_init(struct dpaa_priv *priv)
790 {
791 	struct qm_mcc_initcgr initcgr;
792 	u32 cs_th;
793 	int err;
794 
795 	err = qman_alloc_cgrid(&priv->cgr_data.cgr.cgrid);
796 	if (err < 0) {
797 		if (netif_msg_drv(priv))
798 			pr_err("%s: Error %d allocating CGR ID\n",
799 			       __func__, err);
800 		goto out_error;
801 	}
802 	priv->cgr_data.cgr.cb = dpaa_eth_cgscn;
803 
804 	/* Enable Congestion State Change Notifications and CS taildrop */
805 	memset(&initcgr, 0, sizeof(initcgr));
806 	initcgr.we_mask = cpu_to_be16(QM_CGR_WE_CSCN_EN | QM_CGR_WE_CS_THRES);
807 	initcgr.cgr.cscn_en = QM_CGR_EN;
808 
809 	/* Set different thresholds based on the MAC speed.
810 	 * This may turn suboptimal if the MAC is reconfigured at a speed
811 	 * lower than its max, e.g. if a dTSEC later negotiates a 100Mbps link.
812 	 * In such cases, we ought to reconfigure the threshold, too.
813 	 */
814 	if (priv->mac_dev->if_support & SUPPORTED_10000baseT_Full)
815 		cs_th = DPAA_CS_THRESHOLD_10G;
816 	else
817 		cs_th = DPAA_CS_THRESHOLD_1G;
818 	qm_cgr_cs_thres_set64(&initcgr.cgr.cs_thres, cs_th, 1);
819 
820 	initcgr.we_mask |= cpu_to_be16(QM_CGR_WE_CSTD_EN);
821 	initcgr.cgr.cstd_en = QM_CGR_EN;
822 
823 	err = qman_create_cgr(&priv->cgr_data.cgr, QMAN_CGR_FLAG_USE_INIT,
824 			      &initcgr);
825 	if (err < 0) {
826 		if (netif_msg_drv(priv))
827 			pr_err("%s: Error %d creating CGR with ID %d\n",
828 			       __func__, err, priv->cgr_data.cgr.cgrid);
829 		qman_release_cgrid(priv->cgr_data.cgr.cgrid);
830 		goto out_error;
831 	}
832 	if (netif_msg_drv(priv))
833 		pr_debug("Created CGR %d for netdev with hwaddr %pM on QMan channel %d\n",
834 			 priv->cgr_data.cgr.cgrid, priv->mac_dev->addr,
835 			 priv->cgr_data.cgr.chan);
836 
837 out_error:
838 	return err;
839 }
840 
841 static inline void dpaa_setup_ingress(const struct dpaa_priv *priv,
842 				      struct dpaa_fq *fq,
843 				      const struct qman_fq *template)
844 {
845 	fq->fq_base = *template;
846 	fq->net_dev = priv->net_dev;
847 
848 	fq->flags = QMAN_FQ_FLAG_NO_ENQUEUE;
849 	fq->channel = priv->channel;
850 }
851 
852 static inline void dpaa_setup_egress(const struct dpaa_priv *priv,
853 				     struct dpaa_fq *fq,
854 				     struct fman_port *port,
855 				     const struct qman_fq *template)
856 {
857 	fq->fq_base = *template;
858 	fq->net_dev = priv->net_dev;
859 
860 	if (port) {
861 		fq->flags = QMAN_FQ_FLAG_TO_DCPORTAL;
862 		fq->channel = (u16)fman_port_get_qman_channel_id(port);
863 	} else {
864 		fq->flags = QMAN_FQ_FLAG_NO_MODIFY;
865 	}
866 }
867 
868 static void dpaa_fq_setup(struct dpaa_priv *priv,
869 			  const struct dpaa_fq_cbs *fq_cbs,
870 			  struct fman_port *tx_port)
871 {
872 	int egress_cnt = 0, conf_cnt = 0, num_portals = 0, portal_cnt = 0, cpu;
873 	const cpumask_t *affine_cpus = qman_affine_cpus();
874 	u16 channels[NR_CPUS];
875 	struct dpaa_fq *fq;
876 
877 	for_each_cpu_and(cpu, affine_cpus, cpu_online_mask)
878 		channels[num_portals++] = qman_affine_channel(cpu);
879 
880 	if (num_portals == 0)
881 		dev_err(priv->net_dev->dev.parent,
882 			"No Qman software (affine) channels found\n");
883 
884 	/* Initialize each FQ in the list */
885 	list_for_each_entry(fq, &priv->dpaa_fq_list, list) {
886 		switch (fq->fq_type) {
887 		case FQ_TYPE_RX_DEFAULT:
888 			dpaa_setup_ingress(priv, fq, &fq_cbs->rx_defq);
889 			break;
890 		case FQ_TYPE_RX_ERROR:
891 			dpaa_setup_ingress(priv, fq, &fq_cbs->rx_errq);
892 			break;
893 		case FQ_TYPE_RX_PCD:
894 			if (!num_portals)
895 				continue;
896 			dpaa_setup_ingress(priv, fq, &fq_cbs->rx_defq);
897 			fq->channel = channels[portal_cnt++ % num_portals];
898 			break;
899 		case FQ_TYPE_TX:
900 			dpaa_setup_egress(priv, fq, tx_port,
901 					  &fq_cbs->egress_ern);
902 			/* If we have more Tx queues than the number of cores,
903 			 * just ignore the extra ones.
904 			 */
905 			if (egress_cnt < DPAA_ETH_TXQ_NUM)
906 				priv->egress_fqs[egress_cnt++] = &fq->fq_base;
907 			break;
908 		case FQ_TYPE_TX_CONF_MQ:
909 			priv->conf_fqs[conf_cnt++] = &fq->fq_base;
910 			/* fall through */
911 		case FQ_TYPE_TX_CONFIRM:
912 			dpaa_setup_ingress(priv, fq, &fq_cbs->tx_defq);
913 			break;
914 		case FQ_TYPE_TX_ERROR:
915 			dpaa_setup_ingress(priv, fq, &fq_cbs->tx_errq);
916 			break;
917 		default:
918 			dev_warn(priv->net_dev->dev.parent,
919 				 "Unknown FQ type detected!\n");
920 			break;
921 		}
922 	}
923 
924 	 /* Make sure all CPUs receive a corresponding Tx queue. */
925 	while (egress_cnt < DPAA_ETH_TXQ_NUM) {
926 		list_for_each_entry(fq, &priv->dpaa_fq_list, list) {
927 			if (fq->fq_type != FQ_TYPE_TX)
928 				continue;
929 			priv->egress_fqs[egress_cnt++] = &fq->fq_base;
930 			if (egress_cnt == DPAA_ETH_TXQ_NUM)
931 				break;
932 		}
933 	}
934 }
935 
936 static inline int dpaa_tx_fq_to_id(const struct dpaa_priv *priv,
937 				   struct qman_fq *tx_fq)
938 {
939 	int i;
940 
941 	for (i = 0; i < DPAA_ETH_TXQ_NUM; i++)
942 		if (priv->egress_fqs[i] == tx_fq)
943 			return i;
944 
945 	return -EINVAL;
946 }
947 
948 static int dpaa_fq_init(struct dpaa_fq *dpaa_fq, bool td_enable)
949 {
950 	const struct dpaa_priv	*priv;
951 	struct qman_fq *confq = NULL;
952 	struct qm_mcc_initfq initfq;
953 	struct device *dev;
954 	struct qman_fq *fq;
955 	int queue_id;
956 	int err;
957 
958 	priv = netdev_priv(dpaa_fq->net_dev);
959 	dev = dpaa_fq->net_dev->dev.parent;
960 
961 	if (dpaa_fq->fqid == 0)
962 		dpaa_fq->flags |= QMAN_FQ_FLAG_DYNAMIC_FQID;
963 
964 	dpaa_fq->init = !(dpaa_fq->flags & QMAN_FQ_FLAG_NO_MODIFY);
965 
966 	err = qman_create_fq(dpaa_fq->fqid, dpaa_fq->flags, &dpaa_fq->fq_base);
967 	if (err) {
968 		dev_err(dev, "qman_create_fq() failed\n");
969 		return err;
970 	}
971 	fq = &dpaa_fq->fq_base;
972 
973 	if (dpaa_fq->init) {
974 		memset(&initfq, 0, sizeof(initfq));
975 
976 		initfq.we_mask = cpu_to_be16(QM_INITFQ_WE_FQCTRL);
977 		/* Note: we may get to keep an empty FQ in cache */
978 		initfq.fqd.fq_ctrl = cpu_to_be16(QM_FQCTRL_PREFERINCACHE);
979 
980 		/* Try to reduce the number of portal interrupts for
981 		 * Tx Confirmation FQs.
982 		 */
983 		if (dpaa_fq->fq_type == FQ_TYPE_TX_CONFIRM)
984 			initfq.fqd.fq_ctrl |= cpu_to_be16(QM_FQCTRL_AVOIDBLOCK);
985 
986 		/* FQ placement */
987 		initfq.we_mask |= cpu_to_be16(QM_INITFQ_WE_DESTWQ);
988 
989 		qm_fqd_set_destwq(&initfq.fqd, dpaa_fq->channel, dpaa_fq->wq);
990 
991 		/* Put all egress queues in a congestion group of their own.
992 		 * Sensu stricto, the Tx confirmation queues are Rx FQs,
993 		 * rather than Tx - but they nonetheless account for the
994 		 * memory footprint on behalf of egress traffic. We therefore
995 		 * place them in the netdev's CGR, along with the Tx FQs.
996 		 */
997 		if (dpaa_fq->fq_type == FQ_TYPE_TX ||
998 		    dpaa_fq->fq_type == FQ_TYPE_TX_CONFIRM ||
999 		    dpaa_fq->fq_type == FQ_TYPE_TX_CONF_MQ) {
1000 			initfq.we_mask |= cpu_to_be16(QM_INITFQ_WE_CGID);
1001 			initfq.fqd.fq_ctrl |= cpu_to_be16(QM_FQCTRL_CGE);
1002 			initfq.fqd.cgid = (u8)priv->cgr_data.cgr.cgrid;
1003 			/* Set a fixed overhead accounting, in an attempt to
1004 			 * reduce the impact of fixed-size skb shells and the
1005 			 * driver's needed headroom on system memory. This is
1006 			 * especially the case when the egress traffic is
1007 			 * composed of small datagrams.
1008 			 * Unfortunately, QMan's OAL value is capped to an
1009 			 * insufficient value, but even that is better than
1010 			 * no overhead accounting at all.
1011 			 */
1012 			initfq.we_mask |= cpu_to_be16(QM_INITFQ_WE_OAC);
1013 			qm_fqd_set_oac(&initfq.fqd, QM_OAC_CG);
1014 			qm_fqd_set_oal(&initfq.fqd,
1015 				       min(sizeof(struct sk_buff) +
1016 				       priv->tx_headroom,
1017 				       (size_t)FSL_QMAN_MAX_OAL));
1018 		}
1019 
1020 		if (td_enable) {
1021 			initfq.we_mask |= cpu_to_be16(QM_INITFQ_WE_TDTHRESH);
1022 			qm_fqd_set_taildrop(&initfq.fqd, DPAA_FQ_TD, 1);
1023 			initfq.fqd.fq_ctrl = cpu_to_be16(QM_FQCTRL_TDE);
1024 		}
1025 
1026 		if (dpaa_fq->fq_type == FQ_TYPE_TX) {
1027 			queue_id = dpaa_tx_fq_to_id(priv, &dpaa_fq->fq_base);
1028 			if (queue_id >= 0)
1029 				confq = priv->conf_fqs[queue_id];
1030 			if (confq) {
1031 				initfq.we_mask |=
1032 					cpu_to_be16(QM_INITFQ_WE_CONTEXTA);
1033 			/* ContextA: OVOM=1(use contextA2 bits instead of ICAD)
1034 			 *	     A2V=1 (contextA A2 field is valid)
1035 			 *	     A0V=1 (contextA A0 field is valid)
1036 			 *	     B0V=1 (contextB field is valid)
1037 			 * ContextA A2: EBD=1 (deallocate buffers inside FMan)
1038 			 * ContextB B0(ASPID): 0 (absolute Virtual Storage ID)
1039 			 */
1040 				qm_fqd_context_a_set64(&initfq.fqd,
1041 						       0x1e00000080000000ULL);
1042 			}
1043 		}
1044 
1045 		/* Put all the ingress queues in our "ingress CGR". */
1046 		if (priv->use_ingress_cgr &&
1047 		    (dpaa_fq->fq_type == FQ_TYPE_RX_DEFAULT ||
1048 		     dpaa_fq->fq_type == FQ_TYPE_RX_ERROR ||
1049 		     dpaa_fq->fq_type == FQ_TYPE_RX_PCD)) {
1050 			initfq.we_mask |= cpu_to_be16(QM_INITFQ_WE_CGID);
1051 			initfq.fqd.fq_ctrl |= cpu_to_be16(QM_FQCTRL_CGE);
1052 			initfq.fqd.cgid = (u8)priv->ingress_cgr.cgrid;
1053 			/* Set a fixed overhead accounting, just like for the
1054 			 * egress CGR.
1055 			 */
1056 			initfq.we_mask |= cpu_to_be16(QM_INITFQ_WE_OAC);
1057 			qm_fqd_set_oac(&initfq.fqd, QM_OAC_CG);
1058 			qm_fqd_set_oal(&initfq.fqd,
1059 				       min(sizeof(struct sk_buff) +
1060 				       priv->tx_headroom,
1061 				       (size_t)FSL_QMAN_MAX_OAL));
1062 		}
1063 
1064 		/* Initialization common to all ingress queues */
1065 		if (dpaa_fq->flags & QMAN_FQ_FLAG_NO_ENQUEUE) {
1066 			initfq.we_mask |= cpu_to_be16(QM_INITFQ_WE_CONTEXTA);
1067 			initfq.fqd.fq_ctrl |= cpu_to_be16(QM_FQCTRL_HOLDACTIVE |
1068 						QM_FQCTRL_CTXASTASHING);
1069 			initfq.fqd.context_a.stashing.exclusive =
1070 				QM_STASHING_EXCL_DATA | QM_STASHING_EXCL_CTX |
1071 				QM_STASHING_EXCL_ANNOTATION;
1072 			qm_fqd_set_stashing(&initfq.fqd, 1, 2,
1073 					    DIV_ROUND_UP(sizeof(struct qman_fq),
1074 							 64));
1075 		}
1076 
1077 		err = qman_init_fq(fq, QMAN_INITFQ_FLAG_SCHED, &initfq);
1078 		if (err < 0) {
1079 			dev_err(dev, "qman_init_fq(%u) = %d\n",
1080 				qman_fq_fqid(fq), err);
1081 			qman_destroy_fq(fq);
1082 			return err;
1083 		}
1084 	}
1085 
1086 	dpaa_fq->fqid = qman_fq_fqid(fq);
1087 
1088 	return 0;
1089 }
1090 
1091 static int dpaa_fq_free_entry(struct device *dev, struct qman_fq *fq)
1092 {
1093 	const struct dpaa_priv  *priv;
1094 	struct dpaa_fq *dpaa_fq;
1095 	int err, error;
1096 
1097 	err = 0;
1098 
1099 	dpaa_fq = container_of(fq, struct dpaa_fq, fq_base);
1100 	priv = netdev_priv(dpaa_fq->net_dev);
1101 
1102 	if (dpaa_fq->init) {
1103 		err = qman_retire_fq(fq, NULL);
1104 		if (err < 0 && netif_msg_drv(priv))
1105 			dev_err(dev, "qman_retire_fq(%u) = %d\n",
1106 				qman_fq_fqid(fq), err);
1107 
1108 		error = qman_oos_fq(fq);
1109 		if (error < 0 && netif_msg_drv(priv)) {
1110 			dev_err(dev, "qman_oos_fq(%u) = %d\n",
1111 				qman_fq_fqid(fq), error);
1112 			if (err >= 0)
1113 				err = error;
1114 		}
1115 	}
1116 
1117 	qman_destroy_fq(fq);
1118 	list_del(&dpaa_fq->list);
1119 
1120 	return err;
1121 }
1122 
1123 static int dpaa_fq_free(struct device *dev, struct list_head *list)
1124 {
1125 	struct dpaa_fq *dpaa_fq, *tmp;
1126 	int err, error;
1127 
1128 	err = 0;
1129 	list_for_each_entry_safe(dpaa_fq, tmp, list, list) {
1130 		error = dpaa_fq_free_entry(dev, (struct qman_fq *)dpaa_fq);
1131 		if (error < 0 && err >= 0)
1132 			err = error;
1133 	}
1134 
1135 	return err;
1136 }
1137 
1138 static int dpaa_eth_init_tx_port(struct fman_port *port, struct dpaa_fq *errq,
1139 				 struct dpaa_fq *defq,
1140 				 struct dpaa_buffer_layout *buf_layout)
1141 {
1142 	struct fman_buffer_prefix_content buf_prefix_content;
1143 	struct fman_port_params params;
1144 	int err;
1145 
1146 	memset(&params, 0, sizeof(params));
1147 	memset(&buf_prefix_content, 0, sizeof(buf_prefix_content));
1148 
1149 	buf_prefix_content.priv_data_size = buf_layout->priv_data_size;
1150 	buf_prefix_content.pass_prs_result = true;
1151 	buf_prefix_content.pass_hash_result = true;
1152 	buf_prefix_content.pass_time_stamp = true;
1153 	buf_prefix_content.data_align = DPAA_FD_DATA_ALIGNMENT;
1154 
1155 	params.specific_params.non_rx_params.err_fqid = errq->fqid;
1156 	params.specific_params.non_rx_params.dflt_fqid = defq->fqid;
1157 
1158 	err = fman_port_config(port, &params);
1159 	if (err) {
1160 		pr_err("%s: fman_port_config failed\n", __func__);
1161 		return err;
1162 	}
1163 
1164 	err = fman_port_cfg_buf_prefix_content(port, &buf_prefix_content);
1165 	if (err) {
1166 		pr_err("%s: fman_port_cfg_buf_prefix_content failed\n",
1167 		       __func__);
1168 		return err;
1169 	}
1170 
1171 	err = fman_port_init(port);
1172 	if (err)
1173 		pr_err("%s: fm_port_init failed\n", __func__);
1174 
1175 	return err;
1176 }
1177 
1178 static int dpaa_eth_init_rx_port(struct fman_port *port, struct dpaa_bp *bp,
1179 				 struct dpaa_fq *errq,
1180 				 struct dpaa_fq *defq, struct dpaa_fq *pcdq,
1181 				 struct dpaa_buffer_layout *buf_layout)
1182 {
1183 	struct fman_buffer_prefix_content buf_prefix_content;
1184 	struct fman_port_rx_params *rx_p;
1185 	struct fman_port_params params;
1186 	int err;
1187 
1188 	memset(&params, 0, sizeof(params));
1189 	memset(&buf_prefix_content, 0, sizeof(buf_prefix_content));
1190 
1191 	buf_prefix_content.priv_data_size = buf_layout->priv_data_size;
1192 	buf_prefix_content.pass_prs_result = true;
1193 	buf_prefix_content.pass_hash_result = true;
1194 	buf_prefix_content.pass_time_stamp = true;
1195 	buf_prefix_content.data_align = DPAA_FD_DATA_ALIGNMENT;
1196 
1197 	rx_p = &params.specific_params.rx_params;
1198 	rx_p->err_fqid = errq->fqid;
1199 	rx_p->dflt_fqid = defq->fqid;
1200 	if (pcdq) {
1201 		rx_p->pcd_base_fqid = pcdq->fqid;
1202 		rx_p->pcd_fqs_count = DPAA_ETH_PCD_RXQ_NUM;
1203 	}
1204 
1205 	rx_p->ext_buf_pools.num_of_pools_used = 1;
1206 	rx_p->ext_buf_pools.ext_buf_pool[0].id =  bp->bpid;
1207 	rx_p->ext_buf_pools.ext_buf_pool[0].size = (u16)bp->size;
1208 
1209 	err = fman_port_config(port, &params);
1210 	if (err) {
1211 		pr_err("%s: fman_port_config failed\n", __func__);
1212 		return err;
1213 	}
1214 
1215 	err = fman_port_cfg_buf_prefix_content(port, &buf_prefix_content);
1216 	if (err) {
1217 		pr_err("%s: fman_port_cfg_buf_prefix_content failed\n",
1218 		       __func__);
1219 		return err;
1220 	}
1221 
1222 	err = fman_port_init(port);
1223 	if (err)
1224 		pr_err("%s: fm_port_init failed\n", __func__);
1225 
1226 	return err;
1227 }
1228 
1229 static int dpaa_eth_init_ports(struct mac_device *mac_dev,
1230 			       struct dpaa_bp *bp,
1231 			       struct fm_port_fqs *port_fqs,
1232 			       struct dpaa_buffer_layout *buf_layout,
1233 			       struct device *dev)
1234 {
1235 	struct fman_port *rxport = mac_dev->port[RX];
1236 	struct fman_port *txport = mac_dev->port[TX];
1237 	int err;
1238 
1239 	err = dpaa_eth_init_tx_port(txport, port_fqs->tx_errq,
1240 				    port_fqs->tx_defq, &buf_layout[TX]);
1241 	if (err)
1242 		return err;
1243 
1244 	err = dpaa_eth_init_rx_port(rxport, bp, port_fqs->rx_errq,
1245 				    port_fqs->rx_defq, port_fqs->rx_pcdq,
1246 				    &buf_layout[RX]);
1247 
1248 	return err;
1249 }
1250 
1251 static int dpaa_bman_release(const struct dpaa_bp *dpaa_bp,
1252 			     struct bm_buffer *bmb, int cnt)
1253 {
1254 	int err;
1255 
1256 	err = bman_release(dpaa_bp->pool, bmb, cnt);
1257 	/* Should never occur, address anyway to avoid leaking the buffers */
1258 	if (WARN_ON(err) && dpaa_bp->free_buf_cb)
1259 		while (cnt-- > 0)
1260 			dpaa_bp->free_buf_cb(dpaa_bp, &bmb[cnt]);
1261 
1262 	return cnt;
1263 }
1264 
1265 static void dpaa_release_sgt_members(struct qm_sg_entry *sgt)
1266 {
1267 	struct bm_buffer bmb[DPAA_BUFF_RELEASE_MAX];
1268 	struct dpaa_bp *dpaa_bp;
1269 	int i = 0, j;
1270 
1271 	memset(bmb, 0, sizeof(bmb));
1272 
1273 	do {
1274 		dpaa_bp = dpaa_bpid2pool(sgt[i].bpid);
1275 		if (!dpaa_bp)
1276 			return;
1277 
1278 		j = 0;
1279 		do {
1280 			WARN_ON(qm_sg_entry_is_ext(&sgt[i]));
1281 
1282 			bm_buffer_set64(&bmb[j], qm_sg_entry_get64(&sgt[i]));
1283 
1284 			j++; i++;
1285 		} while (j < ARRAY_SIZE(bmb) &&
1286 				!qm_sg_entry_is_final(&sgt[i - 1]) &&
1287 				sgt[i - 1].bpid == sgt[i].bpid);
1288 
1289 		dpaa_bman_release(dpaa_bp, bmb, j);
1290 	} while (!qm_sg_entry_is_final(&sgt[i - 1]));
1291 }
1292 
1293 static void dpaa_fd_release(const struct net_device *net_dev,
1294 			    const struct qm_fd *fd)
1295 {
1296 	struct qm_sg_entry *sgt;
1297 	struct dpaa_bp *dpaa_bp;
1298 	struct bm_buffer bmb;
1299 	dma_addr_t addr;
1300 	void *vaddr;
1301 
1302 	bmb.data = 0;
1303 	bm_buffer_set64(&bmb, qm_fd_addr(fd));
1304 
1305 	dpaa_bp = dpaa_bpid2pool(fd->bpid);
1306 	if (!dpaa_bp)
1307 		return;
1308 
1309 	if (qm_fd_get_format(fd) == qm_fd_sg) {
1310 		vaddr = phys_to_virt(qm_fd_addr(fd));
1311 		sgt = vaddr + qm_fd_get_offset(fd);
1312 
1313 		dma_unmap_page(dpaa_bp->priv->rx_dma_dev, qm_fd_addr(fd),
1314 			       DPAA_BP_RAW_SIZE, DMA_FROM_DEVICE);
1315 
1316 		dpaa_release_sgt_members(sgt);
1317 
1318 		addr = dma_map_page(dpaa_bp->priv->rx_dma_dev,
1319 				    virt_to_page(vaddr), 0, DPAA_BP_RAW_SIZE,
1320 				    DMA_FROM_DEVICE);
1321 		if (dma_mapping_error(dpaa_bp->priv->rx_dma_dev, addr)) {
1322 			netdev_err(net_dev, "DMA mapping failed\n");
1323 			return;
1324 		}
1325 		bm_buffer_set64(&bmb, addr);
1326 	}
1327 
1328 	dpaa_bman_release(dpaa_bp, &bmb, 1);
1329 }
1330 
1331 static void count_ern(struct dpaa_percpu_priv *percpu_priv,
1332 		      const union qm_mr_entry *msg)
1333 {
1334 	switch (msg->ern.rc & QM_MR_RC_MASK) {
1335 	case QM_MR_RC_CGR_TAILDROP:
1336 		percpu_priv->ern_cnt.cg_tdrop++;
1337 		break;
1338 	case QM_MR_RC_WRED:
1339 		percpu_priv->ern_cnt.wred++;
1340 		break;
1341 	case QM_MR_RC_ERROR:
1342 		percpu_priv->ern_cnt.err_cond++;
1343 		break;
1344 	case QM_MR_RC_ORPWINDOW_EARLY:
1345 		percpu_priv->ern_cnt.early_window++;
1346 		break;
1347 	case QM_MR_RC_ORPWINDOW_LATE:
1348 		percpu_priv->ern_cnt.late_window++;
1349 		break;
1350 	case QM_MR_RC_FQ_TAILDROP:
1351 		percpu_priv->ern_cnt.fq_tdrop++;
1352 		break;
1353 	case QM_MR_RC_ORPWINDOW_RETIRED:
1354 		percpu_priv->ern_cnt.fq_retired++;
1355 		break;
1356 	case QM_MR_RC_ORP_ZERO:
1357 		percpu_priv->ern_cnt.orp_zero++;
1358 		break;
1359 	}
1360 }
1361 
1362 /* Turn on HW checksum computation for this outgoing frame.
1363  * If the current protocol is not something we support in this regard
1364  * (or if the stack has already computed the SW checksum), we do nothing.
1365  *
1366  * Returns 0 if all goes well (or HW csum doesn't apply), and a negative value
1367  * otherwise.
1368  *
1369  * Note that this function may modify the fd->cmd field and the skb data buffer
1370  * (the Parse Results area).
1371  */
1372 static int dpaa_enable_tx_csum(struct dpaa_priv *priv,
1373 			       struct sk_buff *skb,
1374 			       struct qm_fd *fd,
1375 			       void *parse_results)
1376 {
1377 	struct fman_prs_result *parse_result;
1378 	u16 ethertype = ntohs(skb->protocol);
1379 	struct ipv6hdr *ipv6h = NULL;
1380 	struct iphdr *iph;
1381 	int retval = 0;
1382 	u8 l4_proto;
1383 
1384 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1385 		return 0;
1386 
1387 	/* Note: L3 csum seems to be already computed in sw, but we can't choose
1388 	 * L4 alone from the FM configuration anyway.
1389 	 */
1390 
1391 	/* Fill in some fields of the Parse Results array, so the FMan
1392 	 * can find them as if they came from the FMan Parser.
1393 	 */
1394 	parse_result = (struct fman_prs_result *)parse_results;
1395 
1396 	/* If we're dealing with VLAN, get the real Ethernet type */
1397 	if (ethertype == ETH_P_8021Q) {
1398 		/* We can't always assume the MAC header is set correctly
1399 		 * by the stack, so reset to beginning of skb->data
1400 		 */
1401 		skb_reset_mac_header(skb);
1402 		ethertype = ntohs(vlan_eth_hdr(skb)->h_vlan_encapsulated_proto);
1403 	}
1404 
1405 	/* Fill in the relevant L3 parse result fields
1406 	 * and read the L4 protocol type
1407 	 */
1408 	switch (ethertype) {
1409 	case ETH_P_IP:
1410 		parse_result->l3r = cpu_to_be16(FM_L3_PARSE_RESULT_IPV4);
1411 		iph = ip_hdr(skb);
1412 		WARN_ON(!iph);
1413 		l4_proto = iph->protocol;
1414 		break;
1415 	case ETH_P_IPV6:
1416 		parse_result->l3r = cpu_to_be16(FM_L3_PARSE_RESULT_IPV6);
1417 		ipv6h = ipv6_hdr(skb);
1418 		WARN_ON(!ipv6h);
1419 		l4_proto = ipv6h->nexthdr;
1420 		break;
1421 	default:
1422 		/* We shouldn't even be here */
1423 		if (net_ratelimit())
1424 			netif_alert(priv, tx_err, priv->net_dev,
1425 				    "Can't compute HW csum for L3 proto 0x%x\n",
1426 				    ntohs(skb->protocol));
1427 		retval = -EIO;
1428 		goto return_error;
1429 	}
1430 
1431 	/* Fill in the relevant L4 parse result fields */
1432 	switch (l4_proto) {
1433 	case IPPROTO_UDP:
1434 		parse_result->l4r = FM_L4_PARSE_RESULT_UDP;
1435 		break;
1436 	case IPPROTO_TCP:
1437 		parse_result->l4r = FM_L4_PARSE_RESULT_TCP;
1438 		break;
1439 	default:
1440 		if (net_ratelimit())
1441 			netif_alert(priv, tx_err, priv->net_dev,
1442 				    "Can't compute HW csum for L4 proto 0x%x\n",
1443 				    l4_proto);
1444 		retval = -EIO;
1445 		goto return_error;
1446 	}
1447 
1448 	/* At index 0 is IPOffset_1 as defined in the Parse Results */
1449 	parse_result->ip_off[0] = (u8)skb_network_offset(skb);
1450 	parse_result->l4_off = (u8)skb_transport_offset(skb);
1451 
1452 	/* Enable L3 (and L4, if TCP or UDP) HW checksum. */
1453 	fd->cmd |= cpu_to_be32(FM_FD_CMD_RPD | FM_FD_CMD_DTC);
1454 
1455 	/* On P1023 and similar platforms fd->cmd interpretation could
1456 	 * be disabled by setting CONTEXT_A bit ICMD; currently this bit
1457 	 * is not set so we do not need to check; in the future, if/when
1458 	 * using context_a we need to check this bit
1459 	 */
1460 
1461 return_error:
1462 	return retval;
1463 }
1464 
1465 static int dpaa_bp_add_8_bufs(const struct dpaa_bp *dpaa_bp)
1466 {
1467 	struct net_device *net_dev = dpaa_bp->priv->net_dev;
1468 	struct bm_buffer bmb[8];
1469 	dma_addr_t addr;
1470 	struct page *p;
1471 	u8 i;
1472 
1473 	for (i = 0; i < 8; i++) {
1474 		p = dev_alloc_pages(0);
1475 		if (unlikely(!p)) {
1476 			netdev_err(net_dev, "dev_alloc_pages() failed\n");
1477 			goto release_previous_buffs;
1478 		}
1479 
1480 		addr = dma_map_page(dpaa_bp->priv->rx_dma_dev, p, 0,
1481 				    DPAA_BP_RAW_SIZE, DMA_FROM_DEVICE);
1482 		if (unlikely(dma_mapping_error(dpaa_bp->priv->rx_dma_dev,
1483 					       addr))) {
1484 			netdev_err(net_dev, "DMA map failed\n");
1485 			goto release_previous_buffs;
1486 		}
1487 
1488 		bmb[i].data = 0;
1489 		bm_buffer_set64(&bmb[i], addr);
1490 	}
1491 
1492 release_bufs:
1493 	return dpaa_bman_release(dpaa_bp, bmb, i);
1494 
1495 release_previous_buffs:
1496 	WARN_ONCE(1, "dpaa_eth: failed to add buffers on Rx\n");
1497 
1498 	bm_buffer_set64(&bmb[i], 0);
1499 	/* Avoid releasing a completely null buffer; bman_release() requires
1500 	 * at least one buffer.
1501 	 */
1502 	if (likely(i))
1503 		goto release_bufs;
1504 
1505 	return 0;
1506 }
1507 
1508 static int dpaa_bp_seed(struct dpaa_bp *dpaa_bp)
1509 {
1510 	int i;
1511 
1512 	/* Give each CPU an allotment of "config_count" buffers */
1513 	for_each_possible_cpu(i) {
1514 		int *count_ptr = per_cpu_ptr(dpaa_bp->percpu_count, i);
1515 		int j;
1516 
1517 		/* Although we access another CPU's counters here
1518 		 * we do it at boot time so it is safe
1519 		 */
1520 		for (j = 0; j < dpaa_bp->config_count; j += 8)
1521 			*count_ptr += dpaa_bp_add_8_bufs(dpaa_bp);
1522 	}
1523 	return 0;
1524 }
1525 
1526 /* Add buffers/(pages) for Rx processing whenever bpool count falls below
1527  * REFILL_THRESHOLD.
1528  */
1529 static int dpaa_eth_refill_bpool(struct dpaa_bp *dpaa_bp, int *countptr)
1530 {
1531 	int count = *countptr;
1532 	int new_bufs;
1533 
1534 	if (unlikely(count < FSL_DPAA_ETH_REFILL_THRESHOLD)) {
1535 		do {
1536 			new_bufs = dpaa_bp_add_8_bufs(dpaa_bp);
1537 			if (unlikely(!new_bufs)) {
1538 				/* Avoid looping forever if we've temporarily
1539 				 * run out of memory. We'll try again at the
1540 				 * next NAPI cycle.
1541 				 */
1542 				break;
1543 			}
1544 			count += new_bufs;
1545 		} while (count < FSL_DPAA_ETH_MAX_BUF_COUNT);
1546 
1547 		*countptr = count;
1548 		if (unlikely(count < FSL_DPAA_ETH_MAX_BUF_COUNT))
1549 			return -ENOMEM;
1550 	}
1551 
1552 	return 0;
1553 }
1554 
1555 static int dpaa_eth_refill_bpools(struct dpaa_priv *priv)
1556 {
1557 	struct dpaa_bp *dpaa_bp;
1558 	int *countptr;
1559 	int res;
1560 
1561 	dpaa_bp = priv->dpaa_bp;
1562 	if (!dpaa_bp)
1563 		return -EINVAL;
1564 	countptr = this_cpu_ptr(dpaa_bp->percpu_count);
1565 	res  = dpaa_eth_refill_bpool(dpaa_bp, countptr);
1566 	if (res)
1567 		return res;
1568 
1569 	return 0;
1570 }
1571 
1572 /* Cleanup function for outgoing frame descriptors that were built on Tx path,
1573  * either contiguous frames or scatter/gather ones.
1574  * Skb freeing is not handled here.
1575  *
1576  * This function may be called on error paths in the Tx function, so guard
1577  * against cases when not all fd relevant fields were filled in. To avoid
1578  * reading the invalid transmission timestamp for the error paths set ts to
1579  * false.
1580  *
1581  * Return the skb backpointer, since for S/G frames the buffer containing it
1582  * gets freed here.
1583  */
1584 static struct sk_buff *dpaa_cleanup_tx_fd(const struct dpaa_priv *priv,
1585 					  const struct qm_fd *fd, bool ts)
1586 {
1587 	const enum dma_data_direction dma_dir = DMA_TO_DEVICE;
1588 	struct device *dev = priv->net_dev->dev.parent;
1589 	struct skb_shared_hwtstamps shhwtstamps;
1590 	dma_addr_t addr = qm_fd_addr(fd);
1591 	void *vaddr = phys_to_virt(addr);
1592 	const struct qm_sg_entry *sgt;
1593 	struct sk_buff *skb;
1594 	u64 ns;
1595 	int i;
1596 
1597 	if (unlikely(qm_fd_get_format(fd) == qm_fd_sg)) {
1598 		dma_unmap_page(priv->tx_dma_dev, addr,
1599 			       qm_fd_get_offset(fd) + DPAA_SGT_SIZE,
1600 			       dma_dir);
1601 
1602 		/* The sgt buffer has been allocated with netdev_alloc_frag(),
1603 		 * it's from lowmem.
1604 		 */
1605 		sgt = vaddr + qm_fd_get_offset(fd);
1606 
1607 		/* sgt[0] is from lowmem, was dma_map_single()-ed */
1608 		dma_unmap_single(priv->tx_dma_dev, qm_sg_addr(&sgt[0]),
1609 				 qm_sg_entry_get_len(&sgt[0]), dma_dir);
1610 
1611 		/* remaining pages were mapped with skb_frag_dma_map() */
1612 		for (i = 1; (i < DPAA_SGT_MAX_ENTRIES) &&
1613 		     !qm_sg_entry_is_final(&sgt[i - 1]); i++) {
1614 			WARN_ON(qm_sg_entry_is_ext(&sgt[i]));
1615 
1616 			dma_unmap_page(priv->tx_dma_dev, qm_sg_addr(&sgt[i]),
1617 				       qm_sg_entry_get_len(&sgt[i]), dma_dir);
1618 		}
1619 	} else {
1620 		dma_unmap_single(priv->tx_dma_dev, addr,
1621 				 priv->tx_headroom + qm_fd_get_length(fd),
1622 				 dma_dir);
1623 	}
1624 
1625 	skb = *(struct sk_buff **)vaddr;
1626 
1627 	/* DMA unmapping is required before accessing the HW provided info */
1628 	if (ts && priv->tx_tstamp &&
1629 	    skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) {
1630 		memset(&shhwtstamps, 0, sizeof(shhwtstamps));
1631 
1632 		if (!fman_port_get_tstamp(priv->mac_dev->port[TX], vaddr,
1633 					  &ns)) {
1634 			shhwtstamps.hwtstamp = ns_to_ktime(ns);
1635 			skb_tstamp_tx(skb, &shhwtstamps);
1636 		} else {
1637 			dev_warn(dev, "fman_port_get_tstamp failed!\n");
1638 		}
1639 	}
1640 
1641 	if (qm_fd_get_format(fd) == qm_fd_sg)
1642 		/* Free the page that we allocated on Tx for the SGT */
1643 		free_pages((unsigned long)vaddr, 0);
1644 
1645 	return skb;
1646 }
1647 
1648 static u8 rx_csum_offload(const struct dpaa_priv *priv, const struct qm_fd *fd)
1649 {
1650 	/* The parser has run and performed L4 checksum validation.
1651 	 * We know there were no parser errors (and implicitly no
1652 	 * L4 csum error), otherwise we wouldn't be here.
1653 	 */
1654 	if ((priv->net_dev->features & NETIF_F_RXCSUM) &&
1655 	    (be32_to_cpu(fd->status) & FM_FD_STAT_L4CV))
1656 		return CHECKSUM_UNNECESSARY;
1657 
1658 	/* We're here because either the parser didn't run or the L4 checksum
1659 	 * was not verified. This may include the case of a UDP frame with
1660 	 * checksum zero or an L4 proto other than TCP/UDP
1661 	 */
1662 	return CHECKSUM_NONE;
1663 }
1664 
1665 /* Build a linear skb around the received buffer.
1666  * We are guaranteed there is enough room at the end of the data buffer to
1667  * accommodate the shared info area of the skb.
1668  */
1669 static struct sk_buff *contig_fd_to_skb(const struct dpaa_priv *priv,
1670 					const struct qm_fd *fd)
1671 {
1672 	ssize_t fd_off = qm_fd_get_offset(fd);
1673 	dma_addr_t addr = qm_fd_addr(fd);
1674 	struct dpaa_bp *dpaa_bp;
1675 	struct sk_buff *skb;
1676 	void *vaddr;
1677 
1678 	vaddr = phys_to_virt(addr);
1679 	WARN_ON(!IS_ALIGNED((unsigned long)vaddr, SMP_CACHE_BYTES));
1680 
1681 	dpaa_bp = dpaa_bpid2pool(fd->bpid);
1682 	if (!dpaa_bp)
1683 		goto free_buffer;
1684 
1685 	skb = build_skb(vaddr, dpaa_bp->size +
1686 			SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
1687 	if (WARN_ONCE(!skb, "Build skb failure on Rx\n"))
1688 		goto free_buffer;
1689 	WARN_ON(fd_off != priv->rx_headroom);
1690 	skb_reserve(skb, fd_off);
1691 	skb_put(skb, qm_fd_get_length(fd));
1692 
1693 	skb->ip_summed = rx_csum_offload(priv, fd);
1694 
1695 	return skb;
1696 
1697 free_buffer:
1698 	free_pages((unsigned long)vaddr, 0);
1699 	return NULL;
1700 }
1701 
1702 /* Build an skb with the data of the first S/G entry in the linear portion and
1703  * the rest of the frame as skb fragments.
1704  *
1705  * The page fragment holding the S/G Table is recycled here.
1706  */
1707 static struct sk_buff *sg_fd_to_skb(const struct dpaa_priv *priv,
1708 				    const struct qm_fd *fd)
1709 {
1710 	ssize_t fd_off = qm_fd_get_offset(fd);
1711 	dma_addr_t addr = qm_fd_addr(fd);
1712 	const struct qm_sg_entry *sgt;
1713 	struct page *page, *head_page;
1714 	struct dpaa_bp *dpaa_bp;
1715 	void *vaddr, *sg_vaddr;
1716 	int frag_off, frag_len;
1717 	struct sk_buff *skb;
1718 	dma_addr_t sg_addr;
1719 	int page_offset;
1720 	unsigned int sz;
1721 	int *count_ptr;
1722 	int i, j;
1723 
1724 	vaddr = phys_to_virt(addr);
1725 	WARN_ON(!IS_ALIGNED((unsigned long)vaddr, SMP_CACHE_BYTES));
1726 
1727 	/* Iterate through the SGT entries and add data buffers to the skb */
1728 	sgt = vaddr + fd_off;
1729 	skb = NULL;
1730 	for (i = 0; i < DPAA_SGT_MAX_ENTRIES; i++) {
1731 		/* Extension bit is not supported */
1732 		WARN_ON(qm_sg_entry_is_ext(&sgt[i]));
1733 
1734 		sg_addr = qm_sg_addr(&sgt[i]);
1735 		sg_vaddr = phys_to_virt(sg_addr);
1736 		WARN_ON(!IS_ALIGNED((unsigned long)sg_vaddr,
1737 				    SMP_CACHE_BYTES));
1738 
1739 		dma_unmap_page(priv->rx_dma_dev, sg_addr,
1740 			       DPAA_BP_RAW_SIZE, DMA_FROM_DEVICE);
1741 
1742 		/* We may use multiple Rx pools */
1743 		dpaa_bp = dpaa_bpid2pool(sgt[i].bpid);
1744 		if (!dpaa_bp)
1745 			goto free_buffers;
1746 
1747 		if (!skb) {
1748 			sz = dpaa_bp->size +
1749 				SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1750 			skb = build_skb(sg_vaddr, sz);
1751 			if (WARN_ON(!skb))
1752 				goto free_buffers;
1753 
1754 			skb->ip_summed = rx_csum_offload(priv, fd);
1755 
1756 			/* Make sure forwarded skbs will have enough space
1757 			 * on Tx, if extra headers are added.
1758 			 */
1759 			WARN_ON(fd_off != priv->rx_headroom);
1760 			skb_reserve(skb, fd_off);
1761 			skb_put(skb, qm_sg_entry_get_len(&sgt[i]));
1762 		} else {
1763 			/* Not the first S/G entry; all data from buffer will
1764 			 * be added in an skb fragment; fragment index is offset
1765 			 * by one since first S/G entry was incorporated in the
1766 			 * linear part of the skb.
1767 			 *
1768 			 * Caution: 'page' may be a tail page.
1769 			 */
1770 			page = virt_to_page(sg_vaddr);
1771 			head_page = virt_to_head_page(sg_vaddr);
1772 
1773 			/* Compute offset in (possibly tail) page */
1774 			page_offset = ((unsigned long)sg_vaddr &
1775 					(PAGE_SIZE - 1)) +
1776 				(page_address(page) - page_address(head_page));
1777 			/* page_offset only refers to the beginning of sgt[i];
1778 			 * but the buffer itself may have an internal offset.
1779 			 */
1780 			frag_off = qm_sg_entry_get_off(&sgt[i]) + page_offset;
1781 			frag_len = qm_sg_entry_get_len(&sgt[i]);
1782 			/* skb_add_rx_frag() does no checking on the page; if
1783 			 * we pass it a tail page, we'll end up with
1784 			 * bad page accounting and eventually with segafults.
1785 			 */
1786 			skb_add_rx_frag(skb, i - 1, head_page, frag_off,
1787 					frag_len, dpaa_bp->size);
1788 		}
1789 
1790 		/* Update the pool count for the current {cpu x bpool} */
1791 		count_ptr = this_cpu_ptr(dpaa_bp->percpu_count);
1792 		(*count_ptr)--;
1793 
1794 		if (qm_sg_entry_is_final(&sgt[i]))
1795 			break;
1796 	}
1797 	WARN_ONCE(i == DPAA_SGT_MAX_ENTRIES, "No final bit on SGT\n");
1798 
1799 	/* free the SG table buffer */
1800 	free_pages((unsigned long)vaddr, 0);
1801 
1802 	return skb;
1803 
1804 free_buffers:
1805 	/* free all the SG entries */
1806 	for (j = 0; j < DPAA_SGT_MAX_ENTRIES ; j++) {
1807 		sg_addr = qm_sg_addr(&sgt[j]);
1808 		sg_vaddr = phys_to_virt(sg_addr);
1809 		/* all pages 0..i were unmaped */
1810 		if (j > i)
1811 			dma_unmap_page(priv->rx_dma_dev, qm_sg_addr(&sgt[j]),
1812 				       DPAA_BP_RAW_SIZE, DMA_FROM_DEVICE);
1813 		free_pages((unsigned long)sg_vaddr, 0);
1814 		/* counters 0..i-1 were decremented */
1815 		if (j >= i) {
1816 			dpaa_bp = dpaa_bpid2pool(sgt[j].bpid);
1817 			if (dpaa_bp) {
1818 				count_ptr = this_cpu_ptr(dpaa_bp->percpu_count);
1819 				(*count_ptr)--;
1820 			}
1821 		}
1822 
1823 		if (qm_sg_entry_is_final(&sgt[j]))
1824 			break;
1825 	}
1826 	/* free the SGT fragment */
1827 	free_pages((unsigned long)vaddr, 0);
1828 
1829 	return NULL;
1830 }
1831 
1832 static int skb_to_contig_fd(struct dpaa_priv *priv,
1833 			    struct sk_buff *skb, struct qm_fd *fd,
1834 			    int *offset)
1835 {
1836 	struct net_device *net_dev = priv->net_dev;
1837 	enum dma_data_direction dma_dir;
1838 	unsigned char *buff_start;
1839 	struct sk_buff **skbh;
1840 	dma_addr_t addr;
1841 	int err;
1842 
1843 	/* We are guaranteed to have at least tx_headroom bytes
1844 	 * available, so just use that for offset.
1845 	 */
1846 	fd->bpid = FSL_DPAA_BPID_INV;
1847 	buff_start = skb->data - priv->tx_headroom;
1848 	dma_dir = DMA_TO_DEVICE;
1849 
1850 	skbh = (struct sk_buff **)buff_start;
1851 	*skbh = skb;
1852 
1853 	/* Enable L3/L4 hardware checksum computation.
1854 	 *
1855 	 * We must do this before dma_map_single(DMA_TO_DEVICE), because we may
1856 	 * need to write into the skb.
1857 	 */
1858 	err = dpaa_enable_tx_csum(priv, skb, fd,
1859 				  buff_start + DPAA_TX_PRIV_DATA_SIZE);
1860 	if (unlikely(err < 0)) {
1861 		if (net_ratelimit())
1862 			netif_err(priv, tx_err, net_dev, "HW csum error: %d\n",
1863 				  err);
1864 		return err;
1865 	}
1866 
1867 	/* Fill in the rest of the FD fields */
1868 	qm_fd_set_contig(fd, priv->tx_headroom, skb->len);
1869 	fd->cmd |= cpu_to_be32(FM_FD_CMD_FCO);
1870 
1871 	/* Map the entire buffer size that may be seen by FMan, but no more */
1872 	addr = dma_map_single(priv->tx_dma_dev, buff_start,
1873 			      priv->tx_headroom + skb->len, dma_dir);
1874 	if (unlikely(dma_mapping_error(priv->tx_dma_dev, addr))) {
1875 		if (net_ratelimit())
1876 			netif_err(priv, tx_err, net_dev, "dma_map_single() failed\n");
1877 		return -EINVAL;
1878 	}
1879 	qm_fd_addr_set64(fd, addr);
1880 
1881 	return 0;
1882 }
1883 
1884 static int skb_to_sg_fd(struct dpaa_priv *priv,
1885 			struct sk_buff *skb, struct qm_fd *fd)
1886 {
1887 	const enum dma_data_direction dma_dir = DMA_TO_DEVICE;
1888 	const int nr_frags = skb_shinfo(skb)->nr_frags;
1889 	struct net_device *net_dev = priv->net_dev;
1890 	struct qm_sg_entry *sgt;
1891 	struct sk_buff **skbh;
1892 	void *buff_start;
1893 	skb_frag_t *frag;
1894 	dma_addr_t addr;
1895 	size_t frag_len;
1896 	struct page *p;
1897 	int i, j, err;
1898 
1899 	/* get a page to store the SGTable */
1900 	p = dev_alloc_pages(0);
1901 	if (unlikely(!p)) {
1902 		netdev_err(net_dev, "dev_alloc_pages() failed\n");
1903 		return -ENOMEM;
1904 	}
1905 	buff_start = page_address(p);
1906 
1907 	/* Enable L3/L4 hardware checksum computation.
1908 	 *
1909 	 * We must do this before dma_map_single(DMA_TO_DEVICE), because we may
1910 	 * need to write into the skb.
1911 	 */
1912 	err = dpaa_enable_tx_csum(priv, skb, fd,
1913 				  buff_start + DPAA_TX_PRIV_DATA_SIZE);
1914 	if (unlikely(err < 0)) {
1915 		if (net_ratelimit())
1916 			netif_err(priv, tx_err, net_dev, "HW csum error: %d\n",
1917 				  err);
1918 		goto csum_failed;
1919 	}
1920 
1921 	/* SGT[0] is used by the linear part */
1922 	sgt = (struct qm_sg_entry *)(buff_start + priv->tx_headroom);
1923 	frag_len = skb_headlen(skb);
1924 	qm_sg_entry_set_len(&sgt[0], frag_len);
1925 	sgt[0].bpid = FSL_DPAA_BPID_INV;
1926 	sgt[0].offset = 0;
1927 	addr = dma_map_single(priv->tx_dma_dev, skb->data,
1928 			      skb_headlen(skb), dma_dir);
1929 	if (unlikely(dma_mapping_error(priv->tx_dma_dev, addr))) {
1930 		netdev_err(priv->net_dev, "DMA mapping failed\n");
1931 		err = -EINVAL;
1932 		goto sg0_map_failed;
1933 	}
1934 	qm_sg_entry_set64(&sgt[0], addr);
1935 
1936 	/* populate the rest of SGT entries */
1937 	for (i = 0; i < nr_frags; i++) {
1938 		frag = &skb_shinfo(skb)->frags[i];
1939 		frag_len = skb_frag_size(frag);
1940 		WARN_ON(!skb_frag_page(frag));
1941 		addr = skb_frag_dma_map(priv->tx_dma_dev, frag, 0,
1942 					frag_len, dma_dir);
1943 		if (unlikely(dma_mapping_error(priv->tx_dma_dev, addr))) {
1944 			netdev_err(priv->net_dev, "DMA mapping failed\n");
1945 			err = -EINVAL;
1946 			goto sg_map_failed;
1947 		}
1948 
1949 		qm_sg_entry_set_len(&sgt[i + 1], frag_len);
1950 		sgt[i + 1].bpid = FSL_DPAA_BPID_INV;
1951 		sgt[i + 1].offset = 0;
1952 
1953 		/* keep the offset in the address */
1954 		qm_sg_entry_set64(&sgt[i + 1], addr);
1955 	}
1956 
1957 	/* Set the final bit in the last used entry of the SGT */
1958 	qm_sg_entry_set_f(&sgt[nr_frags], frag_len);
1959 
1960 	/* set fd offset to priv->tx_headroom */
1961 	qm_fd_set_sg(fd, priv->tx_headroom, skb->len);
1962 
1963 	/* DMA map the SGT page */
1964 	skbh = (struct sk_buff **)buff_start;
1965 	*skbh = skb;
1966 
1967 	addr = dma_map_page(priv->tx_dma_dev, p, 0,
1968 			    priv->tx_headroom + DPAA_SGT_SIZE, dma_dir);
1969 	if (unlikely(dma_mapping_error(priv->tx_dma_dev, addr))) {
1970 		netdev_err(priv->net_dev, "DMA mapping failed\n");
1971 		err = -EINVAL;
1972 		goto sgt_map_failed;
1973 	}
1974 
1975 	fd->bpid = FSL_DPAA_BPID_INV;
1976 	fd->cmd |= cpu_to_be32(FM_FD_CMD_FCO);
1977 	qm_fd_addr_set64(fd, addr);
1978 
1979 	return 0;
1980 
1981 sgt_map_failed:
1982 sg_map_failed:
1983 	for (j = 0; j < i; j++)
1984 		dma_unmap_page(priv->tx_dma_dev, qm_sg_addr(&sgt[j]),
1985 			       qm_sg_entry_get_len(&sgt[j]), dma_dir);
1986 sg0_map_failed:
1987 csum_failed:
1988 	free_pages((unsigned long)buff_start, 0);
1989 
1990 	return err;
1991 }
1992 
1993 static inline int dpaa_xmit(struct dpaa_priv *priv,
1994 			    struct rtnl_link_stats64 *percpu_stats,
1995 			    int queue,
1996 			    struct qm_fd *fd)
1997 {
1998 	struct qman_fq *egress_fq;
1999 	int err, i;
2000 
2001 	egress_fq = priv->egress_fqs[queue];
2002 	if (fd->bpid == FSL_DPAA_BPID_INV)
2003 		fd->cmd |= cpu_to_be32(qman_fq_fqid(priv->conf_fqs[queue]));
2004 
2005 	/* Trace this Tx fd */
2006 	trace_dpaa_tx_fd(priv->net_dev, egress_fq, fd);
2007 
2008 	for (i = 0; i < DPAA_ENQUEUE_RETRIES; i++) {
2009 		err = qman_enqueue(egress_fq, fd);
2010 		if (err != -EBUSY)
2011 			break;
2012 	}
2013 
2014 	if (unlikely(err < 0)) {
2015 		percpu_stats->tx_fifo_errors++;
2016 		return err;
2017 	}
2018 
2019 	percpu_stats->tx_packets++;
2020 	percpu_stats->tx_bytes += qm_fd_get_length(fd);
2021 
2022 	return 0;
2023 }
2024 
2025 static netdev_tx_t
2026 dpaa_start_xmit(struct sk_buff *skb, struct net_device *net_dev)
2027 {
2028 	const int queue_mapping = skb_get_queue_mapping(skb);
2029 	bool nonlinear = skb_is_nonlinear(skb);
2030 	struct rtnl_link_stats64 *percpu_stats;
2031 	struct dpaa_percpu_priv *percpu_priv;
2032 	struct netdev_queue *txq;
2033 	struct dpaa_priv *priv;
2034 	struct qm_fd fd;
2035 	int offset = 0;
2036 	int err = 0;
2037 
2038 	priv = netdev_priv(net_dev);
2039 	percpu_priv = this_cpu_ptr(priv->percpu_priv);
2040 	percpu_stats = &percpu_priv->stats;
2041 
2042 	qm_fd_clear_fd(&fd);
2043 
2044 	if (!nonlinear) {
2045 		/* We're going to store the skb backpointer at the beginning
2046 		 * of the data buffer, so we need a privately owned skb
2047 		 *
2048 		 * We've made sure skb is not shared in dev->priv_flags,
2049 		 * we need to verify the skb head is not cloned
2050 		 */
2051 		if (skb_cow_head(skb, priv->tx_headroom))
2052 			goto enomem;
2053 
2054 		WARN_ON(skb_is_nonlinear(skb));
2055 	}
2056 
2057 	/* MAX_SKB_FRAGS is equal or larger than our dpaa_SGT_MAX_ENTRIES;
2058 	 * make sure we don't feed FMan with more fragments than it supports.
2059 	 */
2060 	if (unlikely(nonlinear &&
2061 		     (skb_shinfo(skb)->nr_frags >= DPAA_SGT_MAX_ENTRIES))) {
2062 		/* If the egress skb contains more fragments than we support
2063 		 * we have no choice but to linearize it ourselves.
2064 		 */
2065 		if (__skb_linearize(skb))
2066 			goto enomem;
2067 
2068 		nonlinear = skb_is_nonlinear(skb);
2069 	}
2070 
2071 	if (nonlinear) {
2072 		/* Just create a S/G fd based on the skb */
2073 		err = skb_to_sg_fd(priv, skb, &fd);
2074 		percpu_priv->tx_frag_skbuffs++;
2075 	} else {
2076 		/* Create a contig FD from this skb */
2077 		err = skb_to_contig_fd(priv, skb, &fd, &offset);
2078 	}
2079 	if (unlikely(err < 0))
2080 		goto skb_to_fd_failed;
2081 
2082 	txq = netdev_get_tx_queue(net_dev, queue_mapping);
2083 
2084 	/* LLTX requires to do our own update of trans_start */
2085 	txq->trans_start = jiffies;
2086 
2087 	if (priv->tx_tstamp && skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) {
2088 		fd.cmd |= cpu_to_be32(FM_FD_CMD_UPD);
2089 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
2090 	}
2091 
2092 	if (likely(dpaa_xmit(priv, percpu_stats, queue_mapping, &fd) == 0))
2093 		return NETDEV_TX_OK;
2094 
2095 	dpaa_cleanup_tx_fd(priv, &fd, false);
2096 skb_to_fd_failed:
2097 enomem:
2098 	percpu_stats->tx_errors++;
2099 	dev_kfree_skb(skb);
2100 	return NETDEV_TX_OK;
2101 }
2102 
2103 static void dpaa_rx_error(struct net_device *net_dev,
2104 			  const struct dpaa_priv *priv,
2105 			  struct dpaa_percpu_priv *percpu_priv,
2106 			  const struct qm_fd *fd,
2107 			  u32 fqid)
2108 {
2109 	if (net_ratelimit())
2110 		netif_err(priv, hw, net_dev, "Err FD status = 0x%08x\n",
2111 			  be32_to_cpu(fd->status) & FM_FD_STAT_RX_ERRORS);
2112 
2113 	percpu_priv->stats.rx_errors++;
2114 
2115 	if (be32_to_cpu(fd->status) & FM_FD_ERR_DMA)
2116 		percpu_priv->rx_errors.dme++;
2117 	if (be32_to_cpu(fd->status) & FM_FD_ERR_PHYSICAL)
2118 		percpu_priv->rx_errors.fpe++;
2119 	if (be32_to_cpu(fd->status) & FM_FD_ERR_SIZE)
2120 		percpu_priv->rx_errors.fse++;
2121 	if (be32_to_cpu(fd->status) & FM_FD_ERR_PRS_HDR_ERR)
2122 		percpu_priv->rx_errors.phe++;
2123 
2124 	dpaa_fd_release(net_dev, fd);
2125 }
2126 
2127 static void dpaa_tx_error(struct net_device *net_dev,
2128 			  const struct dpaa_priv *priv,
2129 			  struct dpaa_percpu_priv *percpu_priv,
2130 			  const struct qm_fd *fd,
2131 			  u32 fqid)
2132 {
2133 	struct sk_buff *skb;
2134 
2135 	if (net_ratelimit())
2136 		netif_warn(priv, hw, net_dev, "FD status = 0x%08x\n",
2137 			   be32_to_cpu(fd->status) & FM_FD_STAT_TX_ERRORS);
2138 
2139 	percpu_priv->stats.tx_errors++;
2140 
2141 	skb = dpaa_cleanup_tx_fd(priv, fd, false);
2142 	dev_kfree_skb(skb);
2143 }
2144 
2145 static int dpaa_eth_poll(struct napi_struct *napi, int budget)
2146 {
2147 	struct dpaa_napi_portal *np =
2148 			container_of(napi, struct dpaa_napi_portal, napi);
2149 
2150 	int cleaned = qman_p_poll_dqrr(np->p, budget);
2151 
2152 	if (cleaned < budget) {
2153 		napi_complete_done(napi, cleaned);
2154 		qman_p_irqsource_add(np->p, QM_PIRQ_DQRI);
2155 	} else if (np->down) {
2156 		qman_p_irqsource_add(np->p, QM_PIRQ_DQRI);
2157 	}
2158 
2159 	return cleaned;
2160 }
2161 
2162 static void dpaa_tx_conf(struct net_device *net_dev,
2163 			 const struct dpaa_priv *priv,
2164 			 struct dpaa_percpu_priv *percpu_priv,
2165 			 const struct qm_fd *fd,
2166 			 u32 fqid)
2167 {
2168 	struct sk_buff	*skb;
2169 
2170 	if (unlikely(be32_to_cpu(fd->status) & FM_FD_STAT_TX_ERRORS)) {
2171 		if (net_ratelimit())
2172 			netif_warn(priv, hw, net_dev, "FD status = 0x%08x\n",
2173 				   be32_to_cpu(fd->status) &
2174 				   FM_FD_STAT_TX_ERRORS);
2175 
2176 		percpu_priv->stats.tx_errors++;
2177 	}
2178 
2179 	percpu_priv->tx_confirm++;
2180 
2181 	skb = dpaa_cleanup_tx_fd(priv, fd, true);
2182 
2183 	consume_skb(skb);
2184 }
2185 
2186 static inline int dpaa_eth_napi_schedule(struct dpaa_percpu_priv *percpu_priv,
2187 					 struct qman_portal *portal)
2188 {
2189 	if (unlikely(in_irq() || !in_serving_softirq())) {
2190 		/* Disable QMan IRQ and invoke NAPI */
2191 		qman_p_irqsource_remove(portal, QM_PIRQ_DQRI);
2192 
2193 		percpu_priv->np.p = portal;
2194 		napi_schedule(&percpu_priv->np.napi);
2195 		percpu_priv->in_interrupt++;
2196 		return 1;
2197 	}
2198 	return 0;
2199 }
2200 
2201 static enum qman_cb_dqrr_result rx_error_dqrr(struct qman_portal *portal,
2202 					      struct qman_fq *fq,
2203 					      const struct qm_dqrr_entry *dq)
2204 {
2205 	struct dpaa_fq *dpaa_fq = container_of(fq, struct dpaa_fq, fq_base);
2206 	struct dpaa_percpu_priv *percpu_priv;
2207 	struct net_device *net_dev;
2208 	struct dpaa_bp *dpaa_bp;
2209 	struct dpaa_priv *priv;
2210 
2211 	net_dev = dpaa_fq->net_dev;
2212 	priv = netdev_priv(net_dev);
2213 	dpaa_bp = dpaa_bpid2pool(dq->fd.bpid);
2214 	if (!dpaa_bp)
2215 		return qman_cb_dqrr_consume;
2216 
2217 	percpu_priv = this_cpu_ptr(priv->percpu_priv);
2218 
2219 	if (dpaa_eth_napi_schedule(percpu_priv, portal))
2220 		return qman_cb_dqrr_stop;
2221 
2222 	dpaa_eth_refill_bpools(priv);
2223 	dpaa_rx_error(net_dev, priv, percpu_priv, &dq->fd, fq->fqid);
2224 
2225 	return qman_cb_dqrr_consume;
2226 }
2227 
2228 static enum qman_cb_dqrr_result rx_default_dqrr(struct qman_portal *portal,
2229 						struct qman_fq *fq,
2230 						const struct qm_dqrr_entry *dq)
2231 {
2232 	struct skb_shared_hwtstamps *shhwtstamps;
2233 	struct rtnl_link_stats64 *percpu_stats;
2234 	struct dpaa_percpu_priv *percpu_priv;
2235 	const struct qm_fd *fd = &dq->fd;
2236 	dma_addr_t addr = qm_fd_addr(fd);
2237 	enum qm_fd_format fd_format;
2238 	struct net_device *net_dev;
2239 	u32 fd_status, hash_offset;
2240 	struct dpaa_bp *dpaa_bp;
2241 	struct dpaa_priv *priv;
2242 	unsigned int skb_len;
2243 	struct sk_buff *skb;
2244 	int *count_ptr;
2245 	void *vaddr;
2246 	u64 ns;
2247 
2248 	fd_status = be32_to_cpu(fd->status);
2249 	fd_format = qm_fd_get_format(fd);
2250 	net_dev = ((struct dpaa_fq *)fq)->net_dev;
2251 	priv = netdev_priv(net_dev);
2252 	dpaa_bp = dpaa_bpid2pool(dq->fd.bpid);
2253 	if (!dpaa_bp)
2254 		return qman_cb_dqrr_consume;
2255 
2256 	/* Trace the Rx fd */
2257 	trace_dpaa_rx_fd(net_dev, fq, &dq->fd);
2258 
2259 	percpu_priv = this_cpu_ptr(priv->percpu_priv);
2260 	percpu_stats = &percpu_priv->stats;
2261 
2262 	if (unlikely(dpaa_eth_napi_schedule(percpu_priv, portal)))
2263 		return qman_cb_dqrr_stop;
2264 
2265 	/* Make sure we didn't run out of buffers */
2266 	if (unlikely(dpaa_eth_refill_bpools(priv))) {
2267 		/* Unable to refill the buffer pool due to insufficient
2268 		 * system memory. Just release the frame back into the pool,
2269 		 * otherwise we'll soon end up with an empty buffer pool.
2270 		 */
2271 		dpaa_fd_release(net_dev, &dq->fd);
2272 		return qman_cb_dqrr_consume;
2273 	}
2274 
2275 	if (unlikely(fd_status & FM_FD_STAT_RX_ERRORS) != 0) {
2276 		if (net_ratelimit())
2277 			netif_warn(priv, hw, net_dev, "FD status = 0x%08x\n",
2278 				   fd_status & FM_FD_STAT_RX_ERRORS);
2279 
2280 		percpu_stats->rx_errors++;
2281 		dpaa_fd_release(net_dev, fd);
2282 		return qman_cb_dqrr_consume;
2283 	}
2284 
2285 	dma_unmap_page(dpaa_bp->priv->rx_dma_dev, addr, DPAA_BP_RAW_SIZE,
2286 		       DMA_FROM_DEVICE);
2287 
2288 	/* prefetch the first 64 bytes of the frame or the SGT start */
2289 	vaddr = phys_to_virt(addr);
2290 	prefetch(vaddr + qm_fd_get_offset(fd));
2291 
2292 	/* The only FD types that we may receive are contig and S/G */
2293 	WARN_ON((fd_format != qm_fd_contig) && (fd_format != qm_fd_sg));
2294 
2295 	/* Account for either the contig buffer or the SGT buffer (depending on
2296 	 * which case we were in) having been removed from the pool.
2297 	 */
2298 	count_ptr = this_cpu_ptr(dpaa_bp->percpu_count);
2299 	(*count_ptr)--;
2300 
2301 	if (likely(fd_format == qm_fd_contig))
2302 		skb = contig_fd_to_skb(priv, fd);
2303 	else
2304 		skb = sg_fd_to_skb(priv, fd);
2305 	if (!skb)
2306 		return qman_cb_dqrr_consume;
2307 
2308 	if (priv->rx_tstamp) {
2309 		shhwtstamps = skb_hwtstamps(skb);
2310 		memset(shhwtstamps, 0, sizeof(*shhwtstamps));
2311 
2312 		if (!fman_port_get_tstamp(priv->mac_dev->port[RX], vaddr, &ns))
2313 			shhwtstamps->hwtstamp = ns_to_ktime(ns);
2314 		else
2315 			dev_warn(net_dev->dev.parent, "fman_port_get_tstamp failed!\n");
2316 	}
2317 
2318 	skb->protocol = eth_type_trans(skb, net_dev);
2319 
2320 	if (net_dev->features & NETIF_F_RXHASH && priv->keygen_in_use &&
2321 	    !fman_port_get_hash_result_offset(priv->mac_dev->port[RX],
2322 					      &hash_offset)) {
2323 		enum pkt_hash_types type;
2324 
2325 		/* if L4 exists, it was used in the hash generation */
2326 		type = be32_to_cpu(fd->status) & FM_FD_STAT_L4CV ?
2327 			PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3;
2328 		skb_set_hash(skb, be32_to_cpu(*(u32 *)(vaddr + hash_offset)),
2329 			     type);
2330 	}
2331 
2332 	skb_len = skb->len;
2333 
2334 	if (unlikely(netif_receive_skb(skb) == NET_RX_DROP)) {
2335 		percpu_stats->rx_dropped++;
2336 		return qman_cb_dqrr_consume;
2337 	}
2338 
2339 	percpu_stats->rx_packets++;
2340 	percpu_stats->rx_bytes += skb_len;
2341 
2342 	return qman_cb_dqrr_consume;
2343 }
2344 
2345 static enum qman_cb_dqrr_result conf_error_dqrr(struct qman_portal *portal,
2346 						struct qman_fq *fq,
2347 						const struct qm_dqrr_entry *dq)
2348 {
2349 	struct dpaa_percpu_priv *percpu_priv;
2350 	struct net_device *net_dev;
2351 	struct dpaa_priv *priv;
2352 
2353 	net_dev = ((struct dpaa_fq *)fq)->net_dev;
2354 	priv = netdev_priv(net_dev);
2355 
2356 	percpu_priv = this_cpu_ptr(priv->percpu_priv);
2357 
2358 	if (dpaa_eth_napi_schedule(percpu_priv, portal))
2359 		return qman_cb_dqrr_stop;
2360 
2361 	dpaa_tx_error(net_dev, priv, percpu_priv, &dq->fd, fq->fqid);
2362 
2363 	return qman_cb_dqrr_consume;
2364 }
2365 
2366 static enum qman_cb_dqrr_result conf_dflt_dqrr(struct qman_portal *portal,
2367 					       struct qman_fq *fq,
2368 					       const struct qm_dqrr_entry *dq)
2369 {
2370 	struct dpaa_percpu_priv *percpu_priv;
2371 	struct net_device *net_dev;
2372 	struct dpaa_priv *priv;
2373 
2374 	net_dev = ((struct dpaa_fq *)fq)->net_dev;
2375 	priv = netdev_priv(net_dev);
2376 
2377 	/* Trace the fd */
2378 	trace_dpaa_tx_conf_fd(net_dev, fq, &dq->fd);
2379 
2380 	percpu_priv = this_cpu_ptr(priv->percpu_priv);
2381 
2382 	if (dpaa_eth_napi_schedule(percpu_priv, portal))
2383 		return qman_cb_dqrr_stop;
2384 
2385 	dpaa_tx_conf(net_dev, priv, percpu_priv, &dq->fd, fq->fqid);
2386 
2387 	return qman_cb_dqrr_consume;
2388 }
2389 
2390 static void egress_ern(struct qman_portal *portal,
2391 		       struct qman_fq *fq,
2392 		       const union qm_mr_entry *msg)
2393 {
2394 	const struct qm_fd *fd = &msg->ern.fd;
2395 	struct dpaa_percpu_priv *percpu_priv;
2396 	const struct dpaa_priv *priv;
2397 	struct net_device *net_dev;
2398 	struct sk_buff *skb;
2399 
2400 	net_dev = ((struct dpaa_fq *)fq)->net_dev;
2401 	priv = netdev_priv(net_dev);
2402 	percpu_priv = this_cpu_ptr(priv->percpu_priv);
2403 
2404 	percpu_priv->stats.tx_dropped++;
2405 	percpu_priv->stats.tx_fifo_errors++;
2406 	count_ern(percpu_priv, msg);
2407 
2408 	skb = dpaa_cleanup_tx_fd(priv, fd, false);
2409 	dev_kfree_skb_any(skb);
2410 }
2411 
2412 static const struct dpaa_fq_cbs dpaa_fq_cbs = {
2413 	.rx_defq = { .cb = { .dqrr = rx_default_dqrr } },
2414 	.tx_defq = { .cb = { .dqrr = conf_dflt_dqrr } },
2415 	.rx_errq = { .cb = { .dqrr = rx_error_dqrr } },
2416 	.tx_errq = { .cb = { .dqrr = conf_error_dqrr } },
2417 	.egress_ern = { .cb = { .ern = egress_ern } }
2418 };
2419 
2420 static void dpaa_eth_napi_enable(struct dpaa_priv *priv)
2421 {
2422 	struct dpaa_percpu_priv *percpu_priv;
2423 	int i;
2424 
2425 	for_each_online_cpu(i) {
2426 		percpu_priv = per_cpu_ptr(priv->percpu_priv, i);
2427 
2428 		percpu_priv->np.down = 0;
2429 		napi_enable(&percpu_priv->np.napi);
2430 	}
2431 }
2432 
2433 static void dpaa_eth_napi_disable(struct dpaa_priv *priv)
2434 {
2435 	struct dpaa_percpu_priv *percpu_priv;
2436 	int i;
2437 
2438 	for_each_online_cpu(i) {
2439 		percpu_priv = per_cpu_ptr(priv->percpu_priv, i);
2440 
2441 		percpu_priv->np.down = 1;
2442 		napi_disable(&percpu_priv->np.napi);
2443 	}
2444 }
2445 
2446 static void dpaa_adjust_link(struct net_device *net_dev)
2447 {
2448 	struct mac_device *mac_dev;
2449 	struct dpaa_priv *priv;
2450 
2451 	priv = netdev_priv(net_dev);
2452 	mac_dev = priv->mac_dev;
2453 	mac_dev->adjust_link(mac_dev);
2454 }
2455 
2456 /* The Aquantia PHYs are capable of performing rate adaptation */
2457 #define PHY_VEND_AQUANTIA	0x03a1b400
2458 
2459 static int dpaa_phy_init(struct net_device *net_dev)
2460 {
2461 	__ETHTOOL_DECLARE_LINK_MODE_MASK(mask) = { 0, };
2462 	struct mac_device *mac_dev;
2463 	struct phy_device *phy_dev;
2464 	struct dpaa_priv *priv;
2465 
2466 	priv = netdev_priv(net_dev);
2467 	mac_dev = priv->mac_dev;
2468 
2469 	phy_dev = of_phy_connect(net_dev, mac_dev->phy_node,
2470 				 &dpaa_adjust_link, 0,
2471 				 mac_dev->phy_if);
2472 	if (!phy_dev) {
2473 		netif_err(priv, ifup, net_dev, "init_phy() failed\n");
2474 		return -ENODEV;
2475 	}
2476 
2477 	/* Unless the PHY is capable of rate adaptation */
2478 	if (mac_dev->phy_if != PHY_INTERFACE_MODE_XGMII ||
2479 	    ((phy_dev->drv->phy_id & GENMASK(31, 10)) != PHY_VEND_AQUANTIA)) {
2480 		/* remove any features not supported by the controller */
2481 		ethtool_convert_legacy_u32_to_link_mode(mask,
2482 							mac_dev->if_support);
2483 		linkmode_and(phy_dev->supported, phy_dev->supported, mask);
2484 	}
2485 
2486 	phy_support_asym_pause(phy_dev);
2487 
2488 	mac_dev->phy_dev = phy_dev;
2489 	net_dev->phydev = phy_dev;
2490 
2491 	return 0;
2492 }
2493 
2494 static int dpaa_open(struct net_device *net_dev)
2495 {
2496 	struct mac_device *mac_dev;
2497 	struct dpaa_priv *priv;
2498 	int err, i;
2499 
2500 	priv = netdev_priv(net_dev);
2501 	mac_dev = priv->mac_dev;
2502 	dpaa_eth_napi_enable(priv);
2503 
2504 	err = dpaa_phy_init(net_dev);
2505 	if (err)
2506 		goto phy_init_failed;
2507 
2508 	for (i = 0; i < ARRAY_SIZE(mac_dev->port); i++) {
2509 		err = fman_port_enable(mac_dev->port[i]);
2510 		if (err)
2511 			goto mac_start_failed;
2512 	}
2513 
2514 	err = priv->mac_dev->start(mac_dev);
2515 	if (err < 0) {
2516 		netif_err(priv, ifup, net_dev, "mac_dev->start() = %d\n", err);
2517 		goto mac_start_failed;
2518 	}
2519 
2520 	netif_tx_start_all_queues(net_dev);
2521 
2522 	return 0;
2523 
2524 mac_start_failed:
2525 	for (i = 0; i < ARRAY_SIZE(mac_dev->port); i++)
2526 		fman_port_disable(mac_dev->port[i]);
2527 
2528 phy_init_failed:
2529 	dpaa_eth_napi_disable(priv);
2530 
2531 	return err;
2532 }
2533 
2534 static int dpaa_eth_stop(struct net_device *net_dev)
2535 {
2536 	struct dpaa_priv *priv;
2537 	int err;
2538 
2539 	err = dpaa_stop(net_dev);
2540 
2541 	priv = netdev_priv(net_dev);
2542 	dpaa_eth_napi_disable(priv);
2543 
2544 	return err;
2545 }
2546 
2547 static int dpaa_ts_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
2548 {
2549 	struct dpaa_priv *priv = netdev_priv(dev);
2550 	struct hwtstamp_config config;
2551 
2552 	if (copy_from_user(&config, rq->ifr_data, sizeof(config)))
2553 		return -EFAULT;
2554 
2555 	switch (config.tx_type) {
2556 	case HWTSTAMP_TX_OFF:
2557 		/* Couldn't disable rx/tx timestamping separately.
2558 		 * Do nothing here.
2559 		 */
2560 		priv->tx_tstamp = false;
2561 		break;
2562 	case HWTSTAMP_TX_ON:
2563 		priv->mac_dev->set_tstamp(priv->mac_dev->fman_mac, true);
2564 		priv->tx_tstamp = true;
2565 		break;
2566 	default:
2567 		return -ERANGE;
2568 	}
2569 
2570 	if (config.rx_filter == HWTSTAMP_FILTER_NONE) {
2571 		/* Couldn't disable rx/tx timestamping separately.
2572 		 * Do nothing here.
2573 		 */
2574 		priv->rx_tstamp = false;
2575 	} else {
2576 		priv->mac_dev->set_tstamp(priv->mac_dev->fman_mac, true);
2577 		priv->rx_tstamp = true;
2578 		/* TS is set for all frame types, not only those requested */
2579 		config.rx_filter = HWTSTAMP_FILTER_ALL;
2580 	}
2581 
2582 	return copy_to_user(rq->ifr_data, &config, sizeof(config)) ?
2583 			-EFAULT : 0;
2584 }
2585 
2586 static int dpaa_ioctl(struct net_device *net_dev, struct ifreq *rq, int cmd)
2587 {
2588 	int ret = -EINVAL;
2589 
2590 	if (cmd == SIOCGMIIREG) {
2591 		if (net_dev->phydev)
2592 			return phy_mii_ioctl(net_dev->phydev, rq, cmd);
2593 	}
2594 
2595 	if (cmd == SIOCSHWTSTAMP)
2596 		return dpaa_ts_ioctl(net_dev, rq, cmd);
2597 
2598 	return ret;
2599 }
2600 
2601 static const struct net_device_ops dpaa_ops = {
2602 	.ndo_open = dpaa_open,
2603 	.ndo_start_xmit = dpaa_start_xmit,
2604 	.ndo_stop = dpaa_eth_stop,
2605 	.ndo_tx_timeout = dpaa_tx_timeout,
2606 	.ndo_get_stats64 = dpaa_get_stats64,
2607 	.ndo_change_carrier = fixed_phy_change_carrier,
2608 	.ndo_set_mac_address = dpaa_set_mac_address,
2609 	.ndo_validate_addr = eth_validate_addr,
2610 	.ndo_set_rx_mode = dpaa_set_rx_mode,
2611 	.ndo_do_ioctl = dpaa_ioctl,
2612 	.ndo_setup_tc = dpaa_setup_tc,
2613 };
2614 
2615 static int dpaa_napi_add(struct net_device *net_dev)
2616 {
2617 	struct dpaa_priv *priv = netdev_priv(net_dev);
2618 	struct dpaa_percpu_priv *percpu_priv;
2619 	int cpu;
2620 
2621 	for_each_possible_cpu(cpu) {
2622 		percpu_priv = per_cpu_ptr(priv->percpu_priv, cpu);
2623 
2624 		netif_napi_add(net_dev, &percpu_priv->np.napi,
2625 			       dpaa_eth_poll, NAPI_POLL_WEIGHT);
2626 	}
2627 
2628 	return 0;
2629 }
2630 
2631 static void dpaa_napi_del(struct net_device *net_dev)
2632 {
2633 	struct dpaa_priv *priv = netdev_priv(net_dev);
2634 	struct dpaa_percpu_priv *percpu_priv;
2635 	int cpu;
2636 
2637 	for_each_possible_cpu(cpu) {
2638 		percpu_priv = per_cpu_ptr(priv->percpu_priv, cpu);
2639 
2640 		netif_napi_del(&percpu_priv->np.napi);
2641 	}
2642 }
2643 
2644 static inline void dpaa_bp_free_pf(const struct dpaa_bp *bp,
2645 				   struct bm_buffer *bmb)
2646 {
2647 	dma_addr_t addr = bm_buf_addr(bmb);
2648 
2649 	dma_unmap_page(bp->priv->rx_dma_dev, addr, DPAA_BP_RAW_SIZE,
2650 		       DMA_FROM_DEVICE);
2651 
2652 	skb_free_frag(phys_to_virt(addr));
2653 }
2654 
2655 /* Alloc the dpaa_bp struct and configure default values */
2656 static struct dpaa_bp *dpaa_bp_alloc(struct device *dev)
2657 {
2658 	struct dpaa_bp *dpaa_bp;
2659 
2660 	dpaa_bp = devm_kzalloc(dev, sizeof(*dpaa_bp), GFP_KERNEL);
2661 	if (!dpaa_bp)
2662 		return ERR_PTR(-ENOMEM);
2663 
2664 	dpaa_bp->bpid = FSL_DPAA_BPID_INV;
2665 	dpaa_bp->percpu_count = devm_alloc_percpu(dev, *dpaa_bp->percpu_count);
2666 	if (!dpaa_bp->percpu_count)
2667 		return ERR_PTR(-ENOMEM);
2668 
2669 	dpaa_bp->config_count = FSL_DPAA_ETH_MAX_BUF_COUNT;
2670 
2671 	dpaa_bp->seed_cb = dpaa_bp_seed;
2672 	dpaa_bp->free_buf_cb = dpaa_bp_free_pf;
2673 
2674 	return dpaa_bp;
2675 }
2676 
2677 /* Place all ingress FQs (Rx Default, Rx Error) in a dedicated CGR.
2678  * We won't be sending congestion notifications to FMan; for now, we just use
2679  * this CGR to generate enqueue rejections to FMan in order to drop the frames
2680  * before they reach our ingress queues and eat up memory.
2681  */
2682 static int dpaa_ingress_cgr_init(struct dpaa_priv *priv)
2683 {
2684 	struct qm_mcc_initcgr initcgr;
2685 	u32 cs_th;
2686 	int err;
2687 
2688 	err = qman_alloc_cgrid(&priv->ingress_cgr.cgrid);
2689 	if (err < 0) {
2690 		if (netif_msg_drv(priv))
2691 			pr_err("Error %d allocating CGR ID\n", err);
2692 		goto out_error;
2693 	}
2694 
2695 	/* Enable CS TD, but disable Congestion State Change Notifications. */
2696 	memset(&initcgr, 0, sizeof(initcgr));
2697 	initcgr.we_mask = cpu_to_be16(QM_CGR_WE_CS_THRES);
2698 	initcgr.cgr.cscn_en = QM_CGR_EN;
2699 	cs_th = DPAA_INGRESS_CS_THRESHOLD;
2700 	qm_cgr_cs_thres_set64(&initcgr.cgr.cs_thres, cs_th, 1);
2701 
2702 	initcgr.we_mask |= cpu_to_be16(QM_CGR_WE_CSTD_EN);
2703 	initcgr.cgr.cstd_en = QM_CGR_EN;
2704 
2705 	/* This CGR will be associated with the SWP affined to the current CPU.
2706 	 * However, we'll place all our ingress FQs in it.
2707 	 */
2708 	err = qman_create_cgr(&priv->ingress_cgr, QMAN_CGR_FLAG_USE_INIT,
2709 			      &initcgr);
2710 	if (err < 0) {
2711 		if (netif_msg_drv(priv))
2712 			pr_err("Error %d creating ingress CGR with ID %d\n",
2713 			       err, priv->ingress_cgr.cgrid);
2714 		qman_release_cgrid(priv->ingress_cgr.cgrid);
2715 		goto out_error;
2716 	}
2717 	if (netif_msg_drv(priv))
2718 		pr_debug("Created ingress CGR %d for netdev with hwaddr %pM\n",
2719 			 priv->ingress_cgr.cgrid, priv->mac_dev->addr);
2720 
2721 	priv->use_ingress_cgr = true;
2722 
2723 out_error:
2724 	return err;
2725 }
2726 
2727 static inline u16 dpaa_get_headroom(struct dpaa_buffer_layout *bl)
2728 {
2729 	u16 headroom;
2730 
2731 	/* The frame headroom must accommodate:
2732 	 * - the driver private data area
2733 	 * - parse results, hash results, timestamp if selected
2734 	 * If either hash results or time stamp are selected, both will
2735 	 * be copied to/from the frame headroom, as TS is located between PR and
2736 	 * HR in the IC and IC copy size has a granularity of 16bytes
2737 	 * (see description of FMBM_RICP and FMBM_TICP registers in DPAARM)
2738 	 *
2739 	 * Also make sure the headroom is a multiple of data_align bytes
2740 	 */
2741 	headroom = (u16)(bl->priv_data_size + DPAA_PARSE_RESULTS_SIZE +
2742 		DPAA_TIME_STAMP_SIZE + DPAA_HASH_RESULTS_SIZE);
2743 
2744 	return DPAA_FD_DATA_ALIGNMENT ? ALIGN(headroom,
2745 					      DPAA_FD_DATA_ALIGNMENT) :
2746 					headroom;
2747 }
2748 
2749 static int dpaa_eth_probe(struct platform_device *pdev)
2750 {
2751 	struct net_device *net_dev = NULL;
2752 	struct dpaa_bp *dpaa_bp = NULL;
2753 	struct dpaa_fq *dpaa_fq, *tmp;
2754 	struct dpaa_priv *priv = NULL;
2755 	struct fm_port_fqs port_fqs;
2756 	struct mac_device *mac_dev;
2757 	int err = 0, channel;
2758 	struct device *dev;
2759 
2760 	dev = &pdev->dev;
2761 
2762 	err = bman_is_probed();
2763 	if (!err)
2764 		return -EPROBE_DEFER;
2765 	if (err < 0) {
2766 		dev_err(dev, "failing probe due to bman probe error\n");
2767 		return -ENODEV;
2768 	}
2769 	err = qman_is_probed();
2770 	if (!err)
2771 		return -EPROBE_DEFER;
2772 	if (err < 0) {
2773 		dev_err(dev, "failing probe due to qman probe error\n");
2774 		return -ENODEV;
2775 	}
2776 	err = bman_portals_probed();
2777 	if (!err)
2778 		return -EPROBE_DEFER;
2779 	if (err < 0) {
2780 		dev_err(dev,
2781 			"failing probe due to bman portals probe error\n");
2782 		return -ENODEV;
2783 	}
2784 	err = qman_portals_probed();
2785 	if (!err)
2786 		return -EPROBE_DEFER;
2787 	if (err < 0) {
2788 		dev_err(dev,
2789 			"failing probe due to qman portals probe error\n");
2790 		return -ENODEV;
2791 	}
2792 
2793 	/* Allocate this early, so we can store relevant information in
2794 	 * the private area
2795 	 */
2796 	net_dev = alloc_etherdev_mq(sizeof(*priv), DPAA_ETH_TXQ_NUM);
2797 	if (!net_dev) {
2798 		dev_err(dev, "alloc_etherdev_mq() failed\n");
2799 		return -ENOMEM;
2800 	}
2801 
2802 	/* Do this here, so we can be verbose early */
2803 	SET_NETDEV_DEV(net_dev, dev);
2804 	dev_set_drvdata(dev, net_dev);
2805 
2806 	priv = netdev_priv(net_dev);
2807 	priv->net_dev = net_dev;
2808 
2809 	priv->msg_enable = netif_msg_init(debug, DPAA_MSG_DEFAULT);
2810 
2811 	mac_dev = dpaa_mac_dev_get(pdev);
2812 	if (IS_ERR(mac_dev)) {
2813 		netdev_err(net_dev, "dpaa_mac_dev_get() failed\n");
2814 		err = PTR_ERR(mac_dev);
2815 		goto free_netdev;
2816 	}
2817 
2818 	/* Devices used for DMA mapping */
2819 	priv->rx_dma_dev = fman_port_get_device(mac_dev->port[RX]);
2820 	priv->tx_dma_dev = fman_port_get_device(mac_dev->port[TX]);
2821 	err = dma_coerce_mask_and_coherent(priv->rx_dma_dev, DMA_BIT_MASK(40));
2822 	if (!err)
2823 		err = dma_coerce_mask_and_coherent(priv->tx_dma_dev,
2824 						   DMA_BIT_MASK(40));
2825 	if (err) {
2826 		netdev_err(net_dev, "dma_coerce_mask_and_coherent() failed\n");
2827 		return err;
2828 	}
2829 
2830 	/* If fsl_fm_max_frm is set to a higher value than the all-common 1500,
2831 	 * we choose conservatively and let the user explicitly set a higher
2832 	 * MTU via ifconfig. Otherwise, the user may end up with different MTUs
2833 	 * in the same LAN.
2834 	 * If on the other hand fsl_fm_max_frm has been chosen below 1500,
2835 	 * start with the maximum allowed.
2836 	 */
2837 	net_dev->mtu = min(dpaa_get_max_mtu(), ETH_DATA_LEN);
2838 
2839 	netdev_dbg(net_dev, "Setting initial MTU on net device: %d\n",
2840 		   net_dev->mtu);
2841 
2842 	priv->buf_layout[RX].priv_data_size = DPAA_RX_PRIV_DATA_SIZE; /* Rx */
2843 	priv->buf_layout[TX].priv_data_size = DPAA_TX_PRIV_DATA_SIZE; /* Tx */
2844 
2845 	/* bp init */
2846 	dpaa_bp = dpaa_bp_alloc(dev);
2847 	if (IS_ERR(dpaa_bp)) {
2848 		err = PTR_ERR(dpaa_bp);
2849 		goto free_dpaa_bps;
2850 	}
2851 	/* the raw size of the buffers used for reception */
2852 	dpaa_bp->raw_size = DPAA_BP_RAW_SIZE;
2853 	/* avoid runtime computations by keeping the usable size here */
2854 	dpaa_bp->size = dpaa_bp_size(dpaa_bp->raw_size);
2855 	dpaa_bp->priv = priv;
2856 
2857 	err = dpaa_bp_alloc_pool(dpaa_bp);
2858 	if (err < 0)
2859 		goto free_dpaa_bps;
2860 	priv->dpaa_bp = dpaa_bp;
2861 
2862 	INIT_LIST_HEAD(&priv->dpaa_fq_list);
2863 
2864 	memset(&port_fqs, 0, sizeof(port_fqs));
2865 
2866 	err = dpaa_alloc_all_fqs(dev, &priv->dpaa_fq_list, &port_fqs);
2867 	if (err < 0) {
2868 		dev_err(dev, "dpaa_alloc_all_fqs() failed\n");
2869 		goto free_dpaa_bps;
2870 	}
2871 
2872 	priv->mac_dev = mac_dev;
2873 
2874 	channel = dpaa_get_channel();
2875 	if (channel < 0) {
2876 		dev_err(dev, "dpaa_get_channel() failed\n");
2877 		err = channel;
2878 		goto free_dpaa_bps;
2879 	}
2880 
2881 	priv->channel = (u16)channel;
2882 
2883 	/* Walk the CPUs with affine portals
2884 	 * and add this pool channel to each's dequeue mask.
2885 	 */
2886 	dpaa_eth_add_channel(priv->channel, &pdev->dev);
2887 
2888 	dpaa_fq_setup(priv, &dpaa_fq_cbs, priv->mac_dev->port[TX]);
2889 
2890 	/* Create a congestion group for this netdev, with
2891 	 * dynamically-allocated CGR ID.
2892 	 * Must be executed after probing the MAC, but before
2893 	 * assigning the egress FQs to the CGRs.
2894 	 */
2895 	err = dpaa_eth_cgr_init(priv);
2896 	if (err < 0) {
2897 		dev_err(dev, "Error initializing CGR\n");
2898 		goto free_dpaa_bps;
2899 	}
2900 
2901 	err = dpaa_ingress_cgr_init(priv);
2902 	if (err < 0) {
2903 		dev_err(dev, "Error initializing ingress CGR\n");
2904 		goto delete_egress_cgr;
2905 	}
2906 
2907 	/* Add the FQs to the interface, and make them active */
2908 	list_for_each_entry_safe(dpaa_fq, tmp, &priv->dpaa_fq_list, list) {
2909 		err = dpaa_fq_init(dpaa_fq, false);
2910 		if (err < 0)
2911 			goto free_dpaa_fqs;
2912 	}
2913 
2914 	priv->tx_headroom = dpaa_get_headroom(&priv->buf_layout[TX]);
2915 	priv->rx_headroom = dpaa_get_headroom(&priv->buf_layout[RX]);
2916 
2917 	/* All real interfaces need their ports initialized */
2918 	err = dpaa_eth_init_ports(mac_dev, dpaa_bp, &port_fqs,
2919 				  &priv->buf_layout[0], dev);
2920 	if (err)
2921 		goto free_dpaa_fqs;
2922 
2923 	/* Rx traffic distribution based on keygen hashing defaults to on */
2924 	priv->keygen_in_use = true;
2925 
2926 	priv->percpu_priv = devm_alloc_percpu(dev, *priv->percpu_priv);
2927 	if (!priv->percpu_priv) {
2928 		dev_err(dev, "devm_alloc_percpu() failed\n");
2929 		err = -ENOMEM;
2930 		goto free_dpaa_fqs;
2931 	}
2932 
2933 	priv->num_tc = 1;
2934 	netif_set_real_num_tx_queues(net_dev, priv->num_tc * DPAA_TC_TXQ_NUM);
2935 
2936 	/* Initialize NAPI */
2937 	err = dpaa_napi_add(net_dev);
2938 	if (err < 0)
2939 		goto delete_dpaa_napi;
2940 
2941 	err = dpaa_netdev_init(net_dev, &dpaa_ops, tx_timeout);
2942 	if (err < 0)
2943 		goto delete_dpaa_napi;
2944 
2945 	dpaa_eth_sysfs_init(&net_dev->dev);
2946 
2947 	netif_info(priv, probe, net_dev, "Probed interface %s\n",
2948 		   net_dev->name);
2949 
2950 	return 0;
2951 
2952 delete_dpaa_napi:
2953 	dpaa_napi_del(net_dev);
2954 free_dpaa_fqs:
2955 	dpaa_fq_free(dev, &priv->dpaa_fq_list);
2956 	qman_delete_cgr_safe(&priv->ingress_cgr);
2957 	qman_release_cgrid(priv->ingress_cgr.cgrid);
2958 delete_egress_cgr:
2959 	qman_delete_cgr_safe(&priv->cgr_data.cgr);
2960 	qman_release_cgrid(priv->cgr_data.cgr.cgrid);
2961 free_dpaa_bps:
2962 	dpaa_bps_free(priv);
2963 free_netdev:
2964 	dev_set_drvdata(dev, NULL);
2965 	free_netdev(net_dev);
2966 
2967 	return err;
2968 }
2969 
2970 static int dpaa_remove(struct platform_device *pdev)
2971 {
2972 	struct net_device *net_dev;
2973 	struct dpaa_priv *priv;
2974 	struct device *dev;
2975 	int err;
2976 
2977 	dev = &pdev->dev;
2978 	net_dev = dev_get_drvdata(dev);
2979 
2980 	priv = netdev_priv(net_dev);
2981 
2982 	dpaa_eth_sysfs_remove(dev);
2983 
2984 	dev_set_drvdata(dev, NULL);
2985 	unregister_netdev(net_dev);
2986 
2987 	err = dpaa_fq_free(dev, &priv->dpaa_fq_list);
2988 
2989 	qman_delete_cgr_safe(&priv->ingress_cgr);
2990 	qman_release_cgrid(priv->ingress_cgr.cgrid);
2991 	qman_delete_cgr_safe(&priv->cgr_data.cgr);
2992 	qman_release_cgrid(priv->cgr_data.cgr.cgrid);
2993 
2994 	dpaa_napi_del(net_dev);
2995 
2996 	dpaa_bps_free(priv);
2997 
2998 	free_netdev(net_dev);
2999 
3000 	return err;
3001 }
3002 
3003 static const struct platform_device_id dpaa_devtype[] = {
3004 	{
3005 		.name = "dpaa-ethernet",
3006 		.driver_data = 0,
3007 	}, {
3008 	}
3009 };
3010 MODULE_DEVICE_TABLE(platform, dpaa_devtype);
3011 
3012 static struct platform_driver dpaa_driver = {
3013 	.driver = {
3014 		.name = KBUILD_MODNAME,
3015 	},
3016 	.id_table = dpaa_devtype,
3017 	.probe = dpaa_eth_probe,
3018 	.remove = dpaa_remove
3019 };
3020 
3021 static int __init dpaa_load(void)
3022 {
3023 	int err;
3024 
3025 	pr_debug("FSL DPAA Ethernet driver\n");
3026 
3027 	/* initialize dpaa_eth mirror values */
3028 	dpaa_rx_extra_headroom = fman_get_rx_extra_headroom();
3029 	dpaa_max_frm = fman_get_max_frm();
3030 
3031 	err = platform_driver_register(&dpaa_driver);
3032 	if (err < 0)
3033 		pr_err("Error, platform_driver_register() = %d\n", err);
3034 
3035 	return err;
3036 }
3037 module_init(dpaa_load);
3038 
3039 static void __exit dpaa_unload(void)
3040 {
3041 	platform_driver_unregister(&dpaa_driver);
3042 
3043 	/* Only one channel is used and needs to be released after all
3044 	 * interfaces are removed
3045 	 */
3046 	dpaa_release_channel();
3047 }
3048 module_exit(dpaa_unload);
3049 
3050 MODULE_LICENSE("Dual BSD/GPL");
3051 MODULE_DESCRIPTION("FSL DPAA Ethernet driver");
3052