1 /* QLogic qede NIC Driver
2 * Copyright (c) 2015 QLogic Corporation
3 *
4 * This software is available under the terms of the GNU General Public License
5 * (GPL) Version 2, available from the file COPYING in the main directory of
6 * this source tree.
7 */
8 
9 #include <linux/module.h>
10 #include <linux/pci.h>
11 #include <linux/version.h>
12 #include <linux/device.h>
13 #include <linux/netdevice.h>
14 #include <linux/etherdevice.h>
15 #include <linux/skbuff.h>
16 #include <linux/errno.h>
17 #include <linux/list.h>
18 #include <linux/string.h>
19 #include <linux/dma-mapping.h>
20 #include <linux/interrupt.h>
21 #include <asm/byteorder.h>
22 #include <asm/param.h>
23 #include <linux/io.h>
24 #include <linux/netdev_features.h>
25 #include <linux/udp.h>
26 #include <linux/tcp.h>
27 #include <net/vxlan.h>
28 #include <linux/ip.h>
29 #include <net/ipv6.h>
30 #include <net/tcp.h>
31 #include <linux/if_ether.h>
32 #include <linux/if_vlan.h>
33 #include <linux/pkt_sched.h>
34 #include <linux/ethtool.h>
35 #include <linux/in.h>
36 #include <linux/random.h>
37 #include <net/ip6_checksum.h>
38 #include <linux/bitops.h>
39 
40 #include "qede.h"
41 
42 static char version[] =
43 	"QLogic FastLinQ 4xxxx Ethernet Driver qede " DRV_MODULE_VERSION "\n";
44 
45 MODULE_DESCRIPTION("QLogic FastLinQ 4xxxx Ethernet Driver");
46 MODULE_LICENSE("GPL");
47 MODULE_VERSION(DRV_MODULE_VERSION);
48 
49 static uint debug;
50 module_param(debug, uint, 0);
51 MODULE_PARM_DESC(debug, " Default debug msglevel");
52 
53 static const struct qed_eth_ops *qed_ops;
54 
55 #define CHIP_NUM_57980S_40		0x1634
56 #define CHIP_NUM_57980S_10		0x1666
57 #define CHIP_NUM_57980S_MF		0x1636
58 #define CHIP_NUM_57980S_100		0x1644
59 #define CHIP_NUM_57980S_50		0x1654
60 #define CHIP_NUM_57980S_25		0x1656
61 
62 #ifndef PCI_DEVICE_ID_NX2_57980E
63 #define PCI_DEVICE_ID_57980S_40		CHIP_NUM_57980S_40
64 #define PCI_DEVICE_ID_57980S_10		CHIP_NUM_57980S_10
65 #define PCI_DEVICE_ID_57980S_MF		CHIP_NUM_57980S_MF
66 #define PCI_DEVICE_ID_57980S_100	CHIP_NUM_57980S_100
67 #define PCI_DEVICE_ID_57980S_50		CHIP_NUM_57980S_50
68 #define PCI_DEVICE_ID_57980S_25		CHIP_NUM_57980S_25
69 #endif
70 
71 static const struct pci_device_id qede_pci_tbl[] = {
72 	{ PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_40), 0 },
73 	{ PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_10), 0 },
74 	{ PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_MF), 0 },
75 	{ PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_100), 0 },
76 	{ PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_50), 0 },
77 	{ PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_25), 0 },
78 	{ 0 }
79 };
80 
81 MODULE_DEVICE_TABLE(pci, qede_pci_tbl);
82 
83 static int qede_probe(struct pci_dev *pdev, const struct pci_device_id *id);
84 
85 #define TX_TIMEOUT		(5 * HZ)
86 
87 static void qede_remove(struct pci_dev *pdev);
88 static int qede_alloc_rx_buffer(struct qede_dev *edev,
89 				struct qede_rx_queue *rxq);
90 static void qede_link_update(void *dev, struct qed_link_output *link);
91 
92 static struct pci_driver qede_pci_driver = {
93 	.name = "qede",
94 	.id_table = qede_pci_tbl,
95 	.probe = qede_probe,
96 	.remove = qede_remove,
97 };
98 
99 static struct qed_eth_cb_ops qede_ll_ops = {
100 	{
101 		.link_update = qede_link_update,
102 	},
103 };
104 
105 static int qede_netdev_event(struct notifier_block *this, unsigned long event,
106 			     void *ptr)
107 {
108 	struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
109 	struct ethtool_drvinfo drvinfo;
110 	struct qede_dev *edev;
111 
112 	/* Currently only support name change */
113 	if (event != NETDEV_CHANGENAME)
114 		goto done;
115 
116 	/* Check whether this is a qede device */
117 	if (!ndev || !ndev->ethtool_ops || !ndev->ethtool_ops->get_drvinfo)
118 		goto done;
119 
120 	memset(&drvinfo, 0, sizeof(drvinfo));
121 	ndev->ethtool_ops->get_drvinfo(ndev, &drvinfo);
122 	if (strcmp(drvinfo.driver, "qede"))
123 		goto done;
124 	edev = netdev_priv(ndev);
125 
126 	/* Notify qed of the name change */
127 	if (!edev->ops || !edev->ops->common)
128 		goto done;
129 	edev->ops->common->set_id(edev->cdev, edev->ndev->name,
130 				  "qede");
131 
132 done:
133 	return NOTIFY_DONE;
134 }
135 
136 static struct notifier_block qede_netdev_notifier = {
137 	.notifier_call = qede_netdev_event,
138 };
139 
140 static
141 int __init qede_init(void)
142 {
143 	int ret;
144 	u32 qed_ver;
145 
146 	pr_notice("qede_init: %s\n", version);
147 
148 	qed_ver = qed_get_protocol_version(QED_PROTOCOL_ETH);
149 	if (qed_ver !=  QEDE_ETH_INTERFACE_VERSION) {
150 		pr_notice("Version mismatch [%08x != %08x]\n",
151 			  qed_ver,
152 			  QEDE_ETH_INTERFACE_VERSION);
153 		return -EINVAL;
154 	}
155 
156 	qed_ops = qed_get_eth_ops(QEDE_ETH_INTERFACE_VERSION);
157 	if (!qed_ops) {
158 		pr_notice("Failed to get qed ethtool operations\n");
159 		return -EINVAL;
160 	}
161 
162 	/* Must register notifier before pci ops, since we might miss
163 	 * interface rename after pci probe and netdev registeration.
164 	 */
165 	ret = register_netdevice_notifier(&qede_netdev_notifier);
166 	if (ret) {
167 		pr_notice("Failed to register netdevice_notifier\n");
168 		qed_put_eth_ops();
169 		return -EINVAL;
170 	}
171 
172 	ret = pci_register_driver(&qede_pci_driver);
173 	if (ret) {
174 		pr_notice("Failed to register driver\n");
175 		unregister_netdevice_notifier(&qede_netdev_notifier);
176 		qed_put_eth_ops();
177 		return -EINVAL;
178 	}
179 
180 	return 0;
181 }
182 
183 static void __exit qede_cleanup(void)
184 {
185 	pr_notice("qede_cleanup called\n");
186 
187 	unregister_netdevice_notifier(&qede_netdev_notifier);
188 	pci_unregister_driver(&qede_pci_driver);
189 	qed_put_eth_ops();
190 }
191 
192 module_init(qede_init);
193 module_exit(qede_cleanup);
194 
195 /* -------------------------------------------------------------------------
196  * START OF FAST-PATH
197  * -------------------------------------------------------------------------
198  */
199 
200 /* Unmap the data and free skb */
201 static int qede_free_tx_pkt(struct qede_dev *edev,
202 			    struct qede_tx_queue *txq,
203 			    int *len)
204 {
205 	u16 idx = txq->sw_tx_cons & NUM_TX_BDS_MAX;
206 	struct sk_buff *skb = txq->sw_tx_ring[idx].skb;
207 	struct eth_tx_1st_bd *first_bd;
208 	struct eth_tx_bd *tx_data_bd;
209 	int bds_consumed = 0;
210 	int nbds;
211 	bool data_split = txq->sw_tx_ring[idx].flags & QEDE_TSO_SPLIT_BD;
212 	int i, split_bd_len = 0;
213 
214 	if (unlikely(!skb)) {
215 		DP_ERR(edev,
216 		       "skb is null for txq idx=%d txq->sw_tx_cons=%d txq->sw_tx_prod=%d\n",
217 		       idx, txq->sw_tx_cons, txq->sw_tx_prod);
218 		return -1;
219 	}
220 
221 	*len = skb->len;
222 
223 	first_bd = (struct eth_tx_1st_bd *)qed_chain_consume(&txq->tx_pbl);
224 
225 	bds_consumed++;
226 
227 	nbds = first_bd->data.nbds;
228 
229 	if (data_split) {
230 		struct eth_tx_bd *split = (struct eth_tx_bd *)
231 			qed_chain_consume(&txq->tx_pbl);
232 		split_bd_len = BD_UNMAP_LEN(split);
233 		bds_consumed++;
234 	}
235 	dma_unmap_page(&edev->pdev->dev, BD_UNMAP_ADDR(first_bd),
236 		       BD_UNMAP_LEN(first_bd) + split_bd_len, DMA_TO_DEVICE);
237 
238 	/* Unmap the data of the skb frags */
239 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++, bds_consumed++) {
240 		tx_data_bd = (struct eth_tx_bd *)
241 			qed_chain_consume(&txq->tx_pbl);
242 		dma_unmap_page(&edev->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
243 			       BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
244 	}
245 
246 	while (bds_consumed++ < nbds)
247 		qed_chain_consume(&txq->tx_pbl);
248 
249 	/* Free skb */
250 	dev_kfree_skb_any(skb);
251 	txq->sw_tx_ring[idx].skb = NULL;
252 	txq->sw_tx_ring[idx].flags = 0;
253 
254 	return 0;
255 }
256 
257 /* Unmap the data and free skb when mapping failed during start_xmit */
258 static void qede_free_failed_tx_pkt(struct qede_dev *edev,
259 				    struct qede_tx_queue *txq,
260 				    struct eth_tx_1st_bd *first_bd,
261 				    int nbd,
262 				    bool data_split)
263 {
264 	u16 idx = txq->sw_tx_prod & NUM_TX_BDS_MAX;
265 	struct sk_buff *skb = txq->sw_tx_ring[idx].skb;
266 	struct eth_tx_bd *tx_data_bd;
267 	int i, split_bd_len = 0;
268 
269 	/* Return prod to its position before this skb was handled */
270 	qed_chain_set_prod(&txq->tx_pbl,
271 			   le16_to_cpu(txq->tx_db.data.bd_prod),
272 			   first_bd);
273 
274 	first_bd = (struct eth_tx_1st_bd *)qed_chain_produce(&txq->tx_pbl);
275 
276 	if (data_split) {
277 		struct eth_tx_bd *split = (struct eth_tx_bd *)
278 					  qed_chain_produce(&txq->tx_pbl);
279 		split_bd_len = BD_UNMAP_LEN(split);
280 		nbd--;
281 	}
282 
283 	dma_unmap_page(&edev->pdev->dev, BD_UNMAP_ADDR(first_bd),
284 		       BD_UNMAP_LEN(first_bd) + split_bd_len, DMA_TO_DEVICE);
285 
286 	/* Unmap the data of the skb frags */
287 	for (i = 0; i < nbd; i++) {
288 		tx_data_bd = (struct eth_tx_bd *)
289 			qed_chain_produce(&txq->tx_pbl);
290 		if (tx_data_bd->nbytes)
291 			dma_unmap_page(&edev->pdev->dev,
292 				       BD_UNMAP_ADDR(tx_data_bd),
293 				       BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
294 	}
295 
296 	/* Return again prod to its position before this skb was handled */
297 	qed_chain_set_prod(&txq->tx_pbl,
298 			   le16_to_cpu(txq->tx_db.data.bd_prod),
299 			   first_bd);
300 
301 	/* Free skb */
302 	dev_kfree_skb_any(skb);
303 	txq->sw_tx_ring[idx].skb = NULL;
304 	txq->sw_tx_ring[idx].flags = 0;
305 }
306 
307 static u32 qede_xmit_type(struct qede_dev *edev,
308 			  struct sk_buff *skb,
309 			  int *ipv6_ext)
310 {
311 	u32 rc = XMIT_L4_CSUM;
312 	__be16 l3_proto;
313 
314 	if (skb->ip_summed != CHECKSUM_PARTIAL)
315 		return XMIT_PLAIN;
316 
317 	l3_proto = vlan_get_protocol(skb);
318 	if (l3_proto == htons(ETH_P_IPV6) &&
319 	    (ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
320 		*ipv6_ext = 1;
321 
322 	if (skb_is_gso(skb))
323 		rc |= XMIT_LSO;
324 
325 	return rc;
326 }
327 
328 static void qede_set_params_for_ipv6_ext(struct sk_buff *skb,
329 					 struct eth_tx_2nd_bd *second_bd,
330 					 struct eth_tx_3rd_bd *third_bd)
331 {
332 	u8 l4_proto;
333 	u16 bd2_bits1 = 0, bd2_bits2 = 0;
334 
335 	bd2_bits1 |= (1 << ETH_TX_DATA_2ND_BD_IPV6_EXT_SHIFT);
336 
337 	bd2_bits2 |= ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) &
338 		     ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_MASK)
339 		    << ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_SHIFT;
340 
341 	bd2_bits1 |= (ETH_L4_PSEUDO_CSUM_CORRECT_LENGTH <<
342 		      ETH_TX_DATA_2ND_BD_L4_PSEUDO_CSUM_MODE_SHIFT);
343 
344 	if (vlan_get_protocol(skb) == htons(ETH_P_IPV6))
345 		l4_proto = ipv6_hdr(skb)->nexthdr;
346 	else
347 		l4_proto = ip_hdr(skb)->protocol;
348 
349 	if (l4_proto == IPPROTO_UDP)
350 		bd2_bits1 |= 1 << ETH_TX_DATA_2ND_BD_L4_UDP_SHIFT;
351 
352 	if (third_bd)
353 		third_bd->data.bitfields |=
354 			cpu_to_le16(((tcp_hdrlen(skb) / 4) &
355 				ETH_TX_DATA_3RD_BD_TCP_HDR_LEN_DW_MASK) <<
356 				ETH_TX_DATA_3RD_BD_TCP_HDR_LEN_DW_SHIFT);
357 
358 	second_bd->data.bitfields1 = cpu_to_le16(bd2_bits1);
359 	second_bd->data.bitfields2 = cpu_to_le16(bd2_bits2);
360 }
361 
362 static int map_frag_to_bd(struct qede_dev *edev,
363 			  skb_frag_t *frag,
364 			  struct eth_tx_bd *bd)
365 {
366 	dma_addr_t mapping;
367 
368 	/* Map skb non-linear frag data for DMA */
369 	mapping = skb_frag_dma_map(&edev->pdev->dev, frag, 0,
370 				   skb_frag_size(frag),
371 				   DMA_TO_DEVICE);
372 	if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
373 		DP_NOTICE(edev, "Unable to map frag - dropping packet\n");
374 		return -ENOMEM;
375 	}
376 
377 	/* Setup the data pointer of the frag data */
378 	BD_SET_UNMAP_ADDR_LEN(bd, mapping, skb_frag_size(frag));
379 
380 	return 0;
381 }
382 
383 /* +2 for 1st BD for headers and 2nd BD for headlen (if required) */
384 #if ((MAX_SKB_FRAGS + 2) > ETH_TX_MAX_BDS_PER_NON_LSO_PACKET)
385 static bool qede_pkt_req_lin(struct qede_dev *edev, struct sk_buff *skb,
386 			     u8 xmit_type)
387 {
388 	int allowed_frags = ETH_TX_MAX_BDS_PER_NON_LSO_PACKET - 1;
389 
390 	if (xmit_type & XMIT_LSO) {
391 		int hlen;
392 
393 		hlen = skb_transport_header(skb) +
394 		       tcp_hdrlen(skb) - skb->data;
395 
396 		/* linear payload would require its own BD */
397 		if (skb_headlen(skb) > hlen)
398 			allowed_frags--;
399 	}
400 
401 	return (skb_shinfo(skb)->nr_frags > allowed_frags);
402 }
403 #endif
404 
405 /* Main transmit function */
406 static
407 netdev_tx_t qede_start_xmit(struct sk_buff *skb,
408 			    struct net_device *ndev)
409 {
410 	struct qede_dev *edev = netdev_priv(ndev);
411 	struct netdev_queue *netdev_txq;
412 	struct qede_tx_queue *txq;
413 	struct eth_tx_1st_bd *first_bd;
414 	struct eth_tx_2nd_bd *second_bd = NULL;
415 	struct eth_tx_3rd_bd *third_bd = NULL;
416 	struct eth_tx_bd *tx_data_bd = NULL;
417 	u16 txq_index;
418 	u8 nbd = 0;
419 	dma_addr_t mapping;
420 	int rc, frag_idx = 0, ipv6_ext = 0;
421 	u8 xmit_type;
422 	u16 idx;
423 	u16 hlen;
424 	bool data_split;
425 
426 	/* Get tx-queue context and netdev index */
427 	txq_index = skb_get_queue_mapping(skb);
428 	WARN_ON(txq_index >= QEDE_TSS_CNT(edev));
429 	txq = QEDE_TX_QUEUE(edev, txq_index);
430 	netdev_txq = netdev_get_tx_queue(ndev, txq_index);
431 
432 	WARN_ON(qed_chain_get_elem_left(&txq->tx_pbl) <
433 			       (MAX_SKB_FRAGS + 1));
434 
435 	xmit_type = qede_xmit_type(edev, skb, &ipv6_ext);
436 
437 #if ((MAX_SKB_FRAGS + 2) > ETH_TX_MAX_BDS_PER_NON_LSO_PACKET)
438 	if (qede_pkt_req_lin(edev, skb, xmit_type)) {
439 		if (skb_linearize(skb)) {
440 			DP_NOTICE(edev,
441 				  "SKB linearization failed - silently dropping this SKB\n");
442 			dev_kfree_skb_any(skb);
443 			return NETDEV_TX_OK;
444 		}
445 	}
446 #endif
447 
448 	/* Fill the entry in the SW ring and the BDs in the FW ring */
449 	idx = txq->sw_tx_prod & NUM_TX_BDS_MAX;
450 	txq->sw_tx_ring[idx].skb = skb;
451 	first_bd = (struct eth_tx_1st_bd *)
452 		   qed_chain_produce(&txq->tx_pbl);
453 	memset(first_bd, 0, sizeof(*first_bd));
454 	first_bd->data.bd_flags.bitfields =
455 		1 << ETH_TX_1ST_BD_FLAGS_START_BD_SHIFT;
456 
457 	/* Map skb linear data for DMA and set in the first BD */
458 	mapping = dma_map_single(&edev->pdev->dev, skb->data,
459 				 skb_headlen(skb), DMA_TO_DEVICE);
460 	if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
461 		DP_NOTICE(edev, "SKB mapping failed\n");
462 		qede_free_failed_tx_pkt(edev, txq, first_bd, 0, false);
463 		return NETDEV_TX_OK;
464 	}
465 	nbd++;
466 	BD_SET_UNMAP_ADDR_LEN(first_bd, mapping, skb_headlen(skb));
467 
468 	/* In case there is IPv6 with extension headers or LSO we need 2nd and
469 	 * 3rd BDs.
470 	 */
471 	if (unlikely((xmit_type & XMIT_LSO) | ipv6_ext)) {
472 		second_bd = (struct eth_tx_2nd_bd *)
473 			qed_chain_produce(&txq->tx_pbl);
474 		memset(second_bd, 0, sizeof(*second_bd));
475 
476 		nbd++;
477 		third_bd = (struct eth_tx_3rd_bd *)
478 			qed_chain_produce(&txq->tx_pbl);
479 		memset(third_bd, 0, sizeof(*third_bd));
480 
481 		nbd++;
482 		/* We need to fill in additional data in second_bd... */
483 		tx_data_bd = (struct eth_tx_bd *)second_bd;
484 	}
485 
486 	if (skb_vlan_tag_present(skb)) {
487 		first_bd->data.vlan = cpu_to_le16(skb_vlan_tag_get(skb));
488 		first_bd->data.bd_flags.bitfields |=
489 			1 << ETH_TX_1ST_BD_FLAGS_VLAN_INSERTION_SHIFT;
490 	}
491 
492 	/* Fill the parsing flags & params according to the requested offload */
493 	if (xmit_type & XMIT_L4_CSUM) {
494 		u16 temp = 1 << ETH_TX_DATA_1ST_BD_TUNN_CFG_OVERRIDE_SHIFT;
495 
496 		/* We don't re-calculate IP checksum as it is already done by
497 		 * the upper stack
498 		 */
499 		first_bd->data.bd_flags.bitfields |=
500 			1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
501 
502 		first_bd->data.bitfields |= cpu_to_le16(temp);
503 
504 		/* If the packet is IPv6 with extension header, indicate that
505 		 * to FW and pass few params, since the device cracker doesn't
506 		 * support parsing IPv6 with extension header/s.
507 		 */
508 		if (unlikely(ipv6_ext))
509 			qede_set_params_for_ipv6_ext(skb, second_bd, third_bd);
510 	}
511 
512 	if (xmit_type & XMIT_LSO) {
513 		first_bd->data.bd_flags.bitfields |=
514 			(1 << ETH_TX_1ST_BD_FLAGS_LSO_SHIFT);
515 		third_bd->data.lso_mss =
516 			cpu_to_le16(skb_shinfo(skb)->gso_size);
517 
518 		first_bd->data.bd_flags.bitfields |=
519 		1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
520 		hlen = skb_transport_header(skb) +
521 		       tcp_hdrlen(skb) - skb->data;
522 
523 		/* @@@TBD - if will not be removed need to check */
524 		third_bd->data.bitfields |=
525 			cpu_to_le16((1 << ETH_TX_DATA_3RD_BD_HDR_NBD_SHIFT));
526 
527 		/* Make life easier for FW guys who can't deal with header and
528 		 * data on same BD. If we need to split, use the second bd...
529 		 */
530 		if (unlikely(skb_headlen(skb) > hlen)) {
531 			DP_VERBOSE(edev, NETIF_MSG_TX_QUEUED,
532 				   "TSO split header size is %d (%x:%x)\n",
533 				   first_bd->nbytes, first_bd->addr.hi,
534 				   first_bd->addr.lo);
535 
536 			mapping = HILO_U64(le32_to_cpu(first_bd->addr.hi),
537 					   le32_to_cpu(first_bd->addr.lo)) +
538 					   hlen;
539 
540 			BD_SET_UNMAP_ADDR_LEN(tx_data_bd, mapping,
541 					      le16_to_cpu(first_bd->nbytes) -
542 					      hlen);
543 
544 			/* this marks the BD as one that has no
545 			 * individual mapping
546 			 */
547 			txq->sw_tx_ring[idx].flags |= QEDE_TSO_SPLIT_BD;
548 
549 			first_bd->nbytes = cpu_to_le16(hlen);
550 
551 			tx_data_bd = (struct eth_tx_bd *)third_bd;
552 			data_split = true;
553 		}
554 	}
555 
556 	/* Handle fragmented skb */
557 	/* special handle for frags inside 2nd and 3rd bds.. */
558 	while (tx_data_bd && frag_idx < skb_shinfo(skb)->nr_frags) {
559 		rc = map_frag_to_bd(edev,
560 				    &skb_shinfo(skb)->frags[frag_idx],
561 				    tx_data_bd);
562 		if (rc) {
563 			qede_free_failed_tx_pkt(edev, txq, first_bd, nbd,
564 						data_split);
565 			return NETDEV_TX_OK;
566 		}
567 
568 		if (tx_data_bd == (struct eth_tx_bd *)second_bd)
569 			tx_data_bd = (struct eth_tx_bd *)third_bd;
570 		else
571 			tx_data_bd = NULL;
572 
573 		frag_idx++;
574 	}
575 
576 	/* map last frags into 4th, 5th .... */
577 	for (; frag_idx < skb_shinfo(skb)->nr_frags; frag_idx++, nbd++) {
578 		tx_data_bd = (struct eth_tx_bd *)
579 			     qed_chain_produce(&txq->tx_pbl);
580 
581 		memset(tx_data_bd, 0, sizeof(*tx_data_bd));
582 
583 		rc = map_frag_to_bd(edev,
584 				    &skb_shinfo(skb)->frags[frag_idx],
585 				    tx_data_bd);
586 		if (rc) {
587 			qede_free_failed_tx_pkt(edev, txq, first_bd, nbd,
588 						data_split);
589 			return NETDEV_TX_OK;
590 		}
591 	}
592 
593 	/* update the first BD with the actual num BDs */
594 	first_bd->data.nbds = nbd;
595 
596 	netdev_tx_sent_queue(netdev_txq, skb->len);
597 
598 	skb_tx_timestamp(skb);
599 
600 	/* Advance packet producer only before sending the packet since mapping
601 	 * of pages may fail.
602 	 */
603 	txq->sw_tx_prod++;
604 
605 	/* 'next page' entries are counted in the producer value */
606 	txq->tx_db.data.bd_prod =
607 		cpu_to_le16(qed_chain_get_prod_idx(&txq->tx_pbl));
608 
609 	/* wmb makes sure that the BDs data is updated before updating the
610 	 * producer, otherwise FW may read old data from the BDs.
611 	 */
612 	wmb();
613 	barrier();
614 	writel(txq->tx_db.raw, txq->doorbell_addr);
615 
616 	/* mmiowb is needed to synchronize doorbell writes from more than one
617 	 * processor. It guarantees that the write arrives to the device before
618 	 * the queue lock is released and another start_xmit is called (possibly
619 	 * on another CPU). Without this barrier, the next doorbell can bypass
620 	 * this doorbell. This is applicable to IA64/Altix systems.
621 	 */
622 	mmiowb();
623 
624 	if (unlikely(qed_chain_get_elem_left(&txq->tx_pbl)
625 		      < (MAX_SKB_FRAGS + 1))) {
626 		netif_tx_stop_queue(netdev_txq);
627 		DP_VERBOSE(edev, NETIF_MSG_TX_QUEUED,
628 			   "Stop queue was called\n");
629 		/* paired memory barrier is in qede_tx_int(), we have to keep
630 		 * ordering of set_bit() in netif_tx_stop_queue() and read of
631 		 * fp->bd_tx_cons
632 		 */
633 		smp_mb();
634 
635 		if (qed_chain_get_elem_left(&txq->tx_pbl)
636 		     >= (MAX_SKB_FRAGS + 1) &&
637 		    (edev->state == QEDE_STATE_OPEN)) {
638 			netif_tx_wake_queue(netdev_txq);
639 			DP_VERBOSE(edev, NETIF_MSG_TX_QUEUED,
640 				   "Wake queue was called\n");
641 		}
642 	}
643 
644 	return NETDEV_TX_OK;
645 }
646 
647 static int qede_txq_has_work(struct qede_tx_queue *txq)
648 {
649 	u16 hw_bd_cons;
650 
651 	/* Tell compiler that consumer and producer can change */
652 	barrier();
653 	hw_bd_cons = le16_to_cpu(*txq->hw_cons_ptr);
654 	if (qed_chain_get_cons_idx(&txq->tx_pbl) == hw_bd_cons + 1)
655 		return 0;
656 
657 	return hw_bd_cons != qed_chain_get_cons_idx(&txq->tx_pbl);
658 }
659 
660 static int qede_tx_int(struct qede_dev *edev,
661 		       struct qede_tx_queue *txq)
662 {
663 	struct netdev_queue *netdev_txq;
664 	u16 hw_bd_cons;
665 	unsigned int pkts_compl = 0, bytes_compl = 0;
666 	int rc;
667 
668 	netdev_txq = netdev_get_tx_queue(edev->ndev, txq->index);
669 
670 	hw_bd_cons = le16_to_cpu(*txq->hw_cons_ptr);
671 	barrier();
672 
673 	while (hw_bd_cons != qed_chain_get_cons_idx(&txq->tx_pbl)) {
674 		int len = 0;
675 
676 		rc = qede_free_tx_pkt(edev, txq, &len);
677 		if (rc) {
678 			DP_NOTICE(edev, "hw_bd_cons = %d, chain_cons=%d\n",
679 				  hw_bd_cons,
680 				  qed_chain_get_cons_idx(&txq->tx_pbl));
681 			break;
682 		}
683 
684 		bytes_compl += len;
685 		pkts_compl++;
686 		txq->sw_tx_cons++;
687 	}
688 
689 	netdev_tx_completed_queue(netdev_txq, pkts_compl, bytes_compl);
690 
691 	/* Need to make the tx_bd_cons update visible to start_xmit()
692 	 * before checking for netif_tx_queue_stopped().  Without the
693 	 * memory barrier, there is a small possibility that
694 	 * start_xmit() will miss it and cause the queue to be stopped
695 	 * forever.
696 	 * On the other hand we need an rmb() here to ensure the proper
697 	 * ordering of bit testing in the following
698 	 * netif_tx_queue_stopped(txq) call.
699 	 */
700 	smp_mb();
701 
702 	if (unlikely(netif_tx_queue_stopped(netdev_txq))) {
703 		/* Taking tx_lock is needed to prevent reenabling the queue
704 		 * while it's empty. This could have happen if rx_action() gets
705 		 * suspended in qede_tx_int() after the condition before
706 		 * netif_tx_wake_queue(), while tx_action (qede_start_xmit()):
707 		 *
708 		 * stops the queue->sees fresh tx_bd_cons->releases the queue->
709 		 * sends some packets consuming the whole queue again->
710 		 * stops the queue
711 		 */
712 
713 		__netif_tx_lock(netdev_txq, smp_processor_id());
714 
715 		if ((netif_tx_queue_stopped(netdev_txq)) &&
716 		    (edev->state == QEDE_STATE_OPEN) &&
717 		    (qed_chain_get_elem_left(&txq->tx_pbl)
718 		      >= (MAX_SKB_FRAGS + 1))) {
719 			netif_tx_wake_queue(netdev_txq);
720 			DP_VERBOSE(edev, NETIF_MSG_TX_DONE,
721 				   "Wake queue was called\n");
722 		}
723 
724 		__netif_tx_unlock(netdev_txq);
725 	}
726 
727 	return 0;
728 }
729 
730 static bool qede_has_rx_work(struct qede_rx_queue *rxq)
731 {
732 	u16 hw_comp_cons, sw_comp_cons;
733 
734 	/* Tell compiler that status block fields can change */
735 	barrier();
736 
737 	hw_comp_cons = le16_to_cpu(*rxq->hw_cons_ptr);
738 	sw_comp_cons = qed_chain_get_cons_idx(&rxq->rx_comp_ring);
739 
740 	return hw_comp_cons != sw_comp_cons;
741 }
742 
743 static bool qede_has_tx_work(struct qede_fastpath *fp)
744 {
745 	u8 tc;
746 
747 	for (tc = 0; tc < fp->edev->num_tc; tc++)
748 		if (qede_txq_has_work(&fp->txqs[tc]))
749 			return true;
750 	return false;
751 }
752 
753 /* This function reuses the buffer(from an offset) from
754  * consumer index to producer index in the bd ring
755  */
756 static inline void qede_reuse_page(struct qede_dev *edev,
757 				   struct qede_rx_queue *rxq,
758 				   struct sw_rx_data *curr_cons)
759 {
760 	struct eth_rx_bd *rx_bd_prod = qed_chain_produce(&rxq->rx_bd_ring);
761 	struct sw_rx_data *curr_prod;
762 	dma_addr_t new_mapping;
763 
764 	curr_prod = &rxq->sw_rx_ring[rxq->sw_rx_prod & NUM_RX_BDS_MAX];
765 	*curr_prod = *curr_cons;
766 
767 	new_mapping = curr_prod->mapping + curr_prod->page_offset;
768 
769 	rx_bd_prod->addr.hi = cpu_to_le32(upper_32_bits(new_mapping));
770 	rx_bd_prod->addr.lo = cpu_to_le32(lower_32_bits(new_mapping));
771 
772 	rxq->sw_rx_prod++;
773 	curr_cons->data = NULL;
774 }
775 
776 static inline int qede_realloc_rx_buffer(struct qede_dev *edev,
777 					 struct qede_rx_queue *rxq,
778 					 struct sw_rx_data *curr_cons)
779 {
780 	/* Move to the next segment in the page */
781 	curr_cons->page_offset += rxq->rx_buf_seg_size;
782 
783 	if (curr_cons->page_offset == PAGE_SIZE) {
784 		if (unlikely(qede_alloc_rx_buffer(edev, rxq)))
785 			return -ENOMEM;
786 
787 		dma_unmap_page(&edev->pdev->dev, curr_cons->mapping,
788 			       PAGE_SIZE, DMA_FROM_DEVICE);
789 	} else {
790 		/* Increment refcount of the page as we don't want
791 		 * network stack to take the ownership of the page
792 		 * which can be recycled multiple times by the driver.
793 		 */
794 		atomic_inc(&curr_cons->data->_count);
795 		qede_reuse_page(edev, rxq, curr_cons);
796 	}
797 
798 	return 0;
799 }
800 
801 static inline void qede_update_rx_prod(struct qede_dev *edev,
802 				       struct qede_rx_queue *rxq)
803 {
804 	u16 bd_prod = qed_chain_get_prod_idx(&rxq->rx_bd_ring);
805 	u16 cqe_prod = qed_chain_get_prod_idx(&rxq->rx_comp_ring);
806 	struct eth_rx_prod_data rx_prods = {0};
807 
808 	/* Update producers */
809 	rx_prods.bd_prod = cpu_to_le16(bd_prod);
810 	rx_prods.cqe_prod = cpu_to_le16(cqe_prod);
811 
812 	/* Make sure that the BD and SGE data is updated before updating the
813 	 * producers since FW might read the BD/SGE right after the producer
814 	 * is updated.
815 	 */
816 	wmb();
817 
818 	internal_ram_wr(rxq->hw_rxq_prod_addr, sizeof(rx_prods),
819 			(u32 *)&rx_prods);
820 
821 	/* mmiowb is needed to synchronize doorbell writes from more than one
822 	 * processor. It guarantees that the write arrives to the device before
823 	 * the napi lock is released and another qede_poll is called (possibly
824 	 * on another CPU). Without this barrier, the next doorbell can bypass
825 	 * this doorbell. This is applicable to IA64/Altix systems.
826 	 */
827 	mmiowb();
828 }
829 
830 static u32 qede_get_rxhash(struct qede_dev *edev,
831 			   u8 bitfields,
832 			   __le32 rss_hash,
833 			   enum pkt_hash_types *rxhash_type)
834 {
835 	enum rss_hash_type htype;
836 
837 	htype = GET_FIELD(bitfields, ETH_FAST_PATH_RX_REG_CQE_RSS_HASH_TYPE);
838 
839 	if ((edev->ndev->features & NETIF_F_RXHASH) && htype) {
840 		*rxhash_type = ((htype == RSS_HASH_TYPE_IPV4) ||
841 				(htype == RSS_HASH_TYPE_IPV6)) ?
842 				PKT_HASH_TYPE_L3 : PKT_HASH_TYPE_L4;
843 		return le32_to_cpu(rss_hash);
844 	}
845 	*rxhash_type = PKT_HASH_TYPE_NONE;
846 	return 0;
847 }
848 
849 static void qede_set_skb_csum(struct sk_buff *skb, u8 csum_flag)
850 {
851 	skb_checksum_none_assert(skb);
852 
853 	if (csum_flag & QEDE_CSUM_UNNECESSARY)
854 		skb->ip_summed = CHECKSUM_UNNECESSARY;
855 }
856 
857 static inline void qede_skb_receive(struct qede_dev *edev,
858 				    struct qede_fastpath *fp,
859 				    struct sk_buff *skb,
860 				    u16 vlan_tag)
861 {
862 	if (vlan_tag)
863 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
864 				       vlan_tag);
865 
866 	napi_gro_receive(&fp->napi, skb);
867 }
868 
869 static void qede_set_gro_params(struct qede_dev *edev,
870 				struct sk_buff *skb,
871 				struct eth_fast_path_rx_tpa_start_cqe *cqe)
872 {
873 	u16 parsing_flags = le16_to_cpu(cqe->pars_flags.flags);
874 
875 	if (((parsing_flags >> PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) &
876 	    PARSING_AND_ERR_FLAGS_L3TYPE_MASK) == 2)
877 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
878 	else
879 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
880 
881 	skb_shinfo(skb)->gso_size = __le16_to_cpu(cqe->len_on_first_bd) -
882 					cqe->header_len;
883 }
884 
885 static int qede_fill_frag_skb(struct qede_dev *edev,
886 			      struct qede_rx_queue *rxq,
887 			      u8 tpa_agg_index,
888 			      u16 len_on_bd)
889 {
890 	struct sw_rx_data *current_bd = &rxq->sw_rx_ring[rxq->sw_rx_cons &
891 							 NUM_RX_BDS_MAX];
892 	struct qede_agg_info *tpa_info = &rxq->tpa_info[tpa_agg_index];
893 	struct sk_buff *skb = tpa_info->skb;
894 
895 	if (unlikely(tpa_info->agg_state != QEDE_AGG_STATE_START))
896 		goto out;
897 
898 	/* Add one frag and update the appropriate fields in the skb */
899 	skb_fill_page_desc(skb, tpa_info->frag_id++,
900 			   current_bd->data, current_bd->page_offset,
901 			   len_on_bd);
902 
903 	if (unlikely(qede_realloc_rx_buffer(edev, rxq, current_bd))) {
904 		tpa_info->agg_state = QEDE_AGG_STATE_ERROR;
905 		goto out;
906 	}
907 
908 	qed_chain_consume(&rxq->rx_bd_ring);
909 	rxq->sw_rx_cons++;
910 
911 	skb->data_len += len_on_bd;
912 	skb->truesize += rxq->rx_buf_seg_size;
913 	skb->len += len_on_bd;
914 
915 	return 0;
916 
917 out:
918 	return -ENOMEM;
919 }
920 
921 static void qede_tpa_start(struct qede_dev *edev,
922 			   struct qede_rx_queue *rxq,
923 			   struct eth_fast_path_rx_tpa_start_cqe *cqe)
924 {
925 	struct qede_agg_info *tpa_info = &rxq->tpa_info[cqe->tpa_agg_index];
926 	struct eth_rx_bd *rx_bd_cons = qed_chain_consume(&rxq->rx_bd_ring);
927 	struct eth_rx_bd *rx_bd_prod = qed_chain_produce(&rxq->rx_bd_ring);
928 	struct sw_rx_data *replace_buf = &tpa_info->replace_buf;
929 	dma_addr_t mapping = tpa_info->replace_buf_mapping;
930 	struct sw_rx_data *sw_rx_data_cons;
931 	struct sw_rx_data *sw_rx_data_prod;
932 	enum pkt_hash_types rxhash_type;
933 	u32 rxhash;
934 
935 	sw_rx_data_cons = &rxq->sw_rx_ring[rxq->sw_rx_cons & NUM_RX_BDS_MAX];
936 	sw_rx_data_prod = &rxq->sw_rx_ring[rxq->sw_rx_prod & NUM_RX_BDS_MAX];
937 
938 	/* Use pre-allocated replacement buffer - we can't release the agg.
939 	 * start until its over and we don't want to risk allocation failing
940 	 * here, so re-allocate when aggregation will be over.
941 	 */
942 	dma_unmap_addr_set(sw_rx_data_prod, mapping,
943 			   dma_unmap_addr(replace_buf, mapping));
944 
945 	sw_rx_data_prod->data = replace_buf->data;
946 	rx_bd_prod->addr.hi = cpu_to_le32(upper_32_bits(mapping));
947 	rx_bd_prod->addr.lo = cpu_to_le32(lower_32_bits(mapping));
948 	sw_rx_data_prod->page_offset = replace_buf->page_offset;
949 
950 	rxq->sw_rx_prod++;
951 
952 	/* move partial skb from cons to pool (don't unmap yet)
953 	 * save mapping, incase we drop the packet later on.
954 	 */
955 	tpa_info->start_buf = *sw_rx_data_cons;
956 	mapping = HILO_U64(le32_to_cpu(rx_bd_cons->addr.hi),
957 			   le32_to_cpu(rx_bd_cons->addr.lo));
958 
959 	tpa_info->start_buf_mapping = mapping;
960 	rxq->sw_rx_cons++;
961 
962 	/* set tpa state to start only if we are able to allocate skb
963 	 * for this aggregation, otherwise mark as error and aggregation will
964 	 * be dropped
965 	 */
966 	tpa_info->skb = netdev_alloc_skb(edev->ndev,
967 					 le16_to_cpu(cqe->len_on_first_bd));
968 	if (unlikely(!tpa_info->skb)) {
969 		tpa_info->agg_state = QEDE_AGG_STATE_ERROR;
970 		return;
971 	}
972 
973 	skb_put(tpa_info->skb, le16_to_cpu(cqe->len_on_first_bd));
974 	memcpy(&tpa_info->start_cqe, cqe, sizeof(tpa_info->start_cqe));
975 
976 	/* Start filling in the aggregation info */
977 	tpa_info->frag_id = 0;
978 	tpa_info->agg_state = QEDE_AGG_STATE_START;
979 
980 	rxhash = qede_get_rxhash(edev, cqe->bitfields,
981 				 cqe->rss_hash, &rxhash_type);
982 	skb_set_hash(tpa_info->skb, rxhash, rxhash_type);
983 	if ((le16_to_cpu(cqe->pars_flags.flags) >>
984 	     PARSING_AND_ERR_FLAGS_TAG8021QEXIST_SHIFT) &
985 		    PARSING_AND_ERR_FLAGS_TAG8021QEXIST_MASK)
986 		tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
987 	else
988 		tpa_info->vlan_tag = 0;
989 
990 	/* This is needed in order to enable forwarding support */
991 	qede_set_gro_params(edev, tpa_info->skb, cqe);
992 
993 	if (likely(cqe->ext_bd_len_list[0]))
994 		qede_fill_frag_skb(edev, rxq, cqe->tpa_agg_index,
995 				   le16_to_cpu(cqe->ext_bd_len_list[0]));
996 
997 	if (unlikely(cqe->ext_bd_len_list[1])) {
998 		DP_ERR(edev,
999 		       "Unlikely - got a TPA aggregation with more than one ext_bd_len_list entry in the TPA start\n");
1000 		tpa_info->agg_state = QEDE_AGG_STATE_ERROR;
1001 	}
1002 }
1003 
1004 #ifdef CONFIG_INET
1005 static void qede_gro_ip_csum(struct sk_buff *skb)
1006 {
1007 	const struct iphdr *iph = ip_hdr(skb);
1008 	struct tcphdr *th;
1009 
1010 	skb_set_network_header(skb, 0);
1011 	skb_set_transport_header(skb, sizeof(struct iphdr));
1012 	th = tcp_hdr(skb);
1013 
1014 	th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
1015 				  iph->saddr, iph->daddr, 0);
1016 
1017 	tcp_gro_complete(skb);
1018 }
1019 
1020 static void qede_gro_ipv6_csum(struct sk_buff *skb)
1021 {
1022 	struct ipv6hdr *iph = ipv6_hdr(skb);
1023 	struct tcphdr *th;
1024 
1025 	skb_set_network_header(skb, 0);
1026 	skb_set_transport_header(skb, sizeof(struct ipv6hdr));
1027 	th = tcp_hdr(skb);
1028 
1029 	th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
1030 				  &iph->saddr, &iph->daddr, 0);
1031 	tcp_gro_complete(skb);
1032 }
1033 #endif
1034 
1035 static void qede_gro_receive(struct qede_dev *edev,
1036 			     struct qede_fastpath *fp,
1037 			     struct sk_buff *skb,
1038 			     u16 vlan_tag)
1039 {
1040 #ifdef CONFIG_INET
1041 	if (skb_shinfo(skb)->gso_size) {
1042 		switch (skb->protocol) {
1043 		case htons(ETH_P_IP):
1044 			qede_gro_ip_csum(skb);
1045 			break;
1046 		case htons(ETH_P_IPV6):
1047 			qede_gro_ipv6_csum(skb);
1048 			break;
1049 		default:
1050 			DP_ERR(edev,
1051 			       "Error: FW GRO supports only IPv4/IPv6, not 0x%04x\n",
1052 			       ntohs(skb->protocol));
1053 		}
1054 	}
1055 #endif
1056 	skb_record_rx_queue(skb, fp->rss_id);
1057 	qede_skb_receive(edev, fp, skb, vlan_tag);
1058 }
1059 
1060 static inline void qede_tpa_cont(struct qede_dev *edev,
1061 				 struct qede_rx_queue *rxq,
1062 				 struct eth_fast_path_rx_tpa_cont_cqe *cqe)
1063 {
1064 	int i;
1065 
1066 	for (i = 0; cqe->len_list[i]; i++)
1067 		qede_fill_frag_skb(edev, rxq, cqe->tpa_agg_index,
1068 				   le16_to_cpu(cqe->len_list[i]));
1069 
1070 	if (unlikely(i > 1))
1071 		DP_ERR(edev,
1072 		       "Strange - TPA cont with more than a single len_list entry\n");
1073 }
1074 
1075 static void qede_tpa_end(struct qede_dev *edev,
1076 			 struct qede_fastpath *fp,
1077 			 struct eth_fast_path_rx_tpa_end_cqe *cqe)
1078 {
1079 	struct qede_rx_queue *rxq = fp->rxq;
1080 	struct qede_agg_info *tpa_info;
1081 	struct sk_buff *skb;
1082 	int i;
1083 
1084 	tpa_info = &rxq->tpa_info[cqe->tpa_agg_index];
1085 	skb = tpa_info->skb;
1086 
1087 	for (i = 0; cqe->len_list[i]; i++)
1088 		qede_fill_frag_skb(edev, rxq, cqe->tpa_agg_index,
1089 				   le16_to_cpu(cqe->len_list[i]));
1090 	if (unlikely(i > 1))
1091 		DP_ERR(edev,
1092 		       "Strange - TPA emd with more than a single len_list entry\n");
1093 
1094 	if (unlikely(tpa_info->agg_state != QEDE_AGG_STATE_START))
1095 		goto err;
1096 
1097 	/* Sanity */
1098 	if (unlikely(cqe->num_of_bds != tpa_info->frag_id + 1))
1099 		DP_ERR(edev,
1100 		       "Strange - TPA had %02x BDs, but SKB has only %d frags\n",
1101 		       cqe->num_of_bds, tpa_info->frag_id);
1102 	if (unlikely(skb->len != le16_to_cpu(cqe->total_packet_len)))
1103 		DP_ERR(edev,
1104 		       "Strange - total packet len [cqe] is %4x but SKB has len %04x\n",
1105 		       le16_to_cpu(cqe->total_packet_len), skb->len);
1106 
1107 	memcpy(skb->data,
1108 	       page_address(tpa_info->start_buf.data) +
1109 		tpa_info->start_cqe.placement_offset +
1110 		tpa_info->start_buf.page_offset,
1111 	       le16_to_cpu(tpa_info->start_cqe.len_on_first_bd));
1112 
1113 	/* Recycle [mapped] start buffer for the next replacement */
1114 	tpa_info->replace_buf = tpa_info->start_buf;
1115 	tpa_info->replace_buf_mapping = tpa_info->start_buf_mapping;
1116 
1117 	/* Finalize the SKB */
1118 	skb->protocol = eth_type_trans(skb, edev->ndev);
1119 	skb->ip_summed = CHECKSUM_UNNECESSARY;
1120 
1121 	/* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
1122 	 * to skb_shinfo(skb)->gso_segs
1123 	 */
1124 	NAPI_GRO_CB(skb)->count = le16_to_cpu(cqe->num_of_coalesced_segs);
1125 
1126 	qede_gro_receive(edev, fp, skb, tpa_info->vlan_tag);
1127 
1128 	tpa_info->agg_state = QEDE_AGG_STATE_NONE;
1129 
1130 	return;
1131 err:
1132 	/* The BD starting the aggregation is still mapped; Re-use it for
1133 	 * future aggregations [as replacement buffer]
1134 	 */
1135 	memcpy(&tpa_info->replace_buf, &tpa_info->start_buf,
1136 	       sizeof(struct sw_rx_data));
1137 	tpa_info->replace_buf_mapping = tpa_info->start_buf_mapping;
1138 	tpa_info->start_buf.data = NULL;
1139 	tpa_info->agg_state = QEDE_AGG_STATE_NONE;
1140 	dev_kfree_skb_any(tpa_info->skb);
1141 	tpa_info->skb = NULL;
1142 }
1143 
1144 static u8 qede_check_csum(u16 flag)
1145 {
1146 	u16 csum_flag = 0;
1147 	u8 csum = 0;
1148 
1149 	if ((PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_MASK <<
1150 	     PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_SHIFT) & flag) {
1151 		csum_flag |= PARSING_AND_ERR_FLAGS_L4CHKSMERROR_MASK <<
1152 			     PARSING_AND_ERR_FLAGS_L4CHKSMERROR_SHIFT;
1153 		csum = QEDE_CSUM_UNNECESSARY;
1154 	}
1155 
1156 	csum_flag |= PARSING_AND_ERR_FLAGS_IPHDRERROR_MASK <<
1157 		     PARSING_AND_ERR_FLAGS_IPHDRERROR_SHIFT;
1158 
1159 	if (csum_flag & flag)
1160 		return QEDE_CSUM_ERROR;
1161 
1162 	return csum;
1163 }
1164 
1165 static int qede_rx_int(struct qede_fastpath *fp, int budget)
1166 {
1167 	struct qede_dev *edev = fp->edev;
1168 	struct qede_rx_queue *rxq = fp->rxq;
1169 
1170 	u16 hw_comp_cons, sw_comp_cons, sw_rx_index, parse_flag;
1171 	int rx_pkt = 0;
1172 	u8 csum_flag;
1173 
1174 	hw_comp_cons = le16_to_cpu(*rxq->hw_cons_ptr);
1175 	sw_comp_cons = qed_chain_get_cons_idx(&rxq->rx_comp_ring);
1176 
1177 	/* Memory barrier to prevent the CPU from doing speculative reads of CQE
1178 	 * / BD in the while-loop before reading hw_comp_cons. If the CQE is
1179 	 * read before it is written by FW, then FW writes CQE and SB, and then
1180 	 * the CPU reads the hw_comp_cons, it will use an old CQE.
1181 	 */
1182 	rmb();
1183 
1184 	/* Loop to complete all indicated BDs */
1185 	while (sw_comp_cons != hw_comp_cons) {
1186 		struct eth_fast_path_rx_reg_cqe *fp_cqe;
1187 		enum pkt_hash_types rxhash_type;
1188 		enum eth_rx_cqe_type cqe_type;
1189 		struct sw_rx_data *sw_rx_data;
1190 		union eth_rx_cqe *cqe;
1191 		struct sk_buff *skb;
1192 		struct page *data;
1193 		__le16 flags;
1194 		u16 len, pad;
1195 		u32 rx_hash;
1196 
1197 		/* Get the CQE from the completion ring */
1198 		cqe = (union eth_rx_cqe *)
1199 			qed_chain_consume(&rxq->rx_comp_ring);
1200 		cqe_type = cqe->fast_path_regular.type;
1201 
1202 		if (unlikely(cqe_type == ETH_RX_CQE_TYPE_SLOW_PATH)) {
1203 			edev->ops->eth_cqe_completion(
1204 					edev->cdev, fp->rss_id,
1205 					(struct eth_slow_path_rx_cqe *)cqe);
1206 			goto next_cqe;
1207 		}
1208 
1209 		if (cqe_type != ETH_RX_CQE_TYPE_REGULAR) {
1210 			switch (cqe_type) {
1211 			case ETH_RX_CQE_TYPE_TPA_START:
1212 				qede_tpa_start(edev, rxq,
1213 					       &cqe->fast_path_tpa_start);
1214 				goto next_cqe;
1215 			case ETH_RX_CQE_TYPE_TPA_CONT:
1216 				qede_tpa_cont(edev, rxq,
1217 					      &cqe->fast_path_tpa_cont);
1218 				goto next_cqe;
1219 			case ETH_RX_CQE_TYPE_TPA_END:
1220 				qede_tpa_end(edev, fp,
1221 					     &cqe->fast_path_tpa_end);
1222 				goto next_rx_only;
1223 			default:
1224 				break;
1225 			}
1226 		}
1227 
1228 		/* Get the data from the SW ring */
1229 		sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS_MAX;
1230 		sw_rx_data = &rxq->sw_rx_ring[sw_rx_index];
1231 		data = sw_rx_data->data;
1232 
1233 		fp_cqe = &cqe->fast_path_regular;
1234 		len =  le16_to_cpu(fp_cqe->len_on_first_bd);
1235 		pad = fp_cqe->placement_offset;
1236 		flags = cqe->fast_path_regular.pars_flags.flags;
1237 
1238 		/* If this is an error packet then drop it */
1239 		parse_flag = le16_to_cpu(flags);
1240 
1241 		csum_flag = qede_check_csum(parse_flag);
1242 		if (unlikely(csum_flag == QEDE_CSUM_ERROR)) {
1243 			DP_NOTICE(edev,
1244 				  "CQE in CONS = %u has error, flags = %x, dropping incoming packet\n",
1245 				  sw_comp_cons, parse_flag);
1246 			rxq->rx_hw_errors++;
1247 			qede_reuse_page(edev, rxq, sw_rx_data);
1248 			goto next_rx;
1249 		}
1250 
1251 		skb = netdev_alloc_skb(edev->ndev, QEDE_RX_HDR_SIZE);
1252 		if (unlikely(!skb)) {
1253 			DP_NOTICE(edev,
1254 				  "Build_skb failed, dropping incoming packet\n");
1255 			qede_reuse_page(edev, rxq, sw_rx_data);
1256 			rxq->rx_alloc_errors++;
1257 			goto next_rx;
1258 		}
1259 
1260 		/* Copy data into SKB */
1261 		if (len + pad <= QEDE_RX_HDR_SIZE) {
1262 			memcpy(skb_put(skb, len),
1263 			       page_address(data) + pad +
1264 				sw_rx_data->page_offset, len);
1265 			qede_reuse_page(edev, rxq, sw_rx_data);
1266 		} else {
1267 			struct skb_frag_struct *frag;
1268 			unsigned int pull_len;
1269 			unsigned char *va;
1270 
1271 			frag = &skb_shinfo(skb)->frags[0];
1272 
1273 			skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, data,
1274 					pad + sw_rx_data->page_offset,
1275 					len, rxq->rx_buf_seg_size);
1276 
1277 			va = skb_frag_address(frag);
1278 			pull_len = eth_get_headlen(va, QEDE_RX_HDR_SIZE);
1279 
1280 			/* Align the pull_len to optimize memcpy */
1281 			memcpy(skb->data, va, ALIGN(pull_len, sizeof(long)));
1282 
1283 			skb_frag_size_sub(frag, pull_len);
1284 			frag->page_offset += pull_len;
1285 			skb->data_len -= pull_len;
1286 			skb->tail += pull_len;
1287 
1288 			if (unlikely(qede_realloc_rx_buffer(edev, rxq,
1289 							    sw_rx_data))) {
1290 				DP_ERR(edev, "Failed to allocate rx buffer\n");
1291 				rxq->rx_alloc_errors++;
1292 				goto next_cqe;
1293 			}
1294 		}
1295 
1296 		if (fp_cqe->bd_num != 1) {
1297 			u16 pkt_len = le16_to_cpu(fp_cqe->pkt_len);
1298 			u8 num_frags;
1299 
1300 			pkt_len -= len;
1301 
1302 			for (num_frags = fp_cqe->bd_num - 1; num_frags > 0;
1303 			     num_frags--) {
1304 				u16 cur_size = pkt_len > rxq->rx_buf_size ?
1305 						rxq->rx_buf_size : pkt_len;
1306 
1307 				WARN_ONCE(!cur_size,
1308 					  "Still got %d BDs for mapping jumbo, but length became 0\n",
1309 					  num_frags);
1310 
1311 				if (unlikely(qede_alloc_rx_buffer(edev, rxq)))
1312 					goto next_cqe;
1313 
1314 				rxq->sw_rx_cons++;
1315 				sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS_MAX;
1316 				sw_rx_data = &rxq->sw_rx_ring[sw_rx_index];
1317 				qed_chain_consume(&rxq->rx_bd_ring);
1318 				dma_unmap_page(&edev->pdev->dev,
1319 					       sw_rx_data->mapping,
1320 					       PAGE_SIZE, DMA_FROM_DEVICE);
1321 
1322 				skb_fill_page_desc(skb,
1323 						   skb_shinfo(skb)->nr_frags++,
1324 						   sw_rx_data->data, 0,
1325 						   cur_size);
1326 
1327 				skb->truesize += PAGE_SIZE;
1328 				skb->data_len += cur_size;
1329 				skb->len += cur_size;
1330 				pkt_len -= cur_size;
1331 			}
1332 
1333 			if (pkt_len)
1334 				DP_ERR(edev,
1335 				       "Mapped all BDs of jumbo, but still have %d bytes\n",
1336 				       pkt_len);
1337 		}
1338 
1339 		skb->protocol = eth_type_trans(skb, edev->ndev);
1340 
1341 		rx_hash = qede_get_rxhash(edev, fp_cqe->bitfields,
1342 					  fp_cqe->rss_hash,
1343 					  &rxhash_type);
1344 
1345 		skb_set_hash(skb, rx_hash, rxhash_type);
1346 
1347 		qede_set_skb_csum(skb, csum_flag);
1348 
1349 		skb_record_rx_queue(skb, fp->rss_id);
1350 
1351 		qede_skb_receive(edev, fp, skb, le16_to_cpu(fp_cqe->vlan_tag));
1352 
1353 		qed_chain_consume(&rxq->rx_bd_ring);
1354 next_rx:
1355 		rxq->sw_rx_cons++;
1356 next_rx_only:
1357 		rx_pkt++;
1358 
1359 next_cqe: /* don't consume bd rx buffer */
1360 		qed_chain_recycle_consumed(&rxq->rx_comp_ring);
1361 		sw_comp_cons = qed_chain_get_cons_idx(&rxq->rx_comp_ring);
1362 		/* CR TPA - revisit how to handle budget in TPA perhaps
1363 		 * increase on "end"
1364 		 */
1365 		if (rx_pkt == budget)
1366 			break;
1367 	} /* repeat while sw_comp_cons != hw_comp_cons... */
1368 
1369 	/* Update producers */
1370 	qede_update_rx_prod(edev, rxq);
1371 
1372 	return rx_pkt;
1373 }
1374 
1375 static int qede_poll(struct napi_struct *napi, int budget)
1376 {
1377 	int work_done = 0;
1378 	struct qede_fastpath *fp = container_of(napi, struct qede_fastpath,
1379 						 napi);
1380 	struct qede_dev *edev = fp->edev;
1381 
1382 	while (1) {
1383 		u8 tc;
1384 
1385 		for (tc = 0; tc < edev->num_tc; tc++)
1386 			if (qede_txq_has_work(&fp->txqs[tc]))
1387 				qede_tx_int(edev, &fp->txqs[tc]);
1388 
1389 		if (qede_has_rx_work(fp->rxq)) {
1390 			work_done += qede_rx_int(fp, budget - work_done);
1391 
1392 			/* must not complete if we consumed full budget */
1393 			if (work_done >= budget)
1394 				break;
1395 		}
1396 
1397 		/* Fall out from the NAPI loop if needed */
1398 		if (!(qede_has_rx_work(fp->rxq) || qede_has_tx_work(fp))) {
1399 			qed_sb_update_sb_idx(fp->sb_info);
1400 			/* *_has_*_work() reads the status block,
1401 			 * thus we need to ensure that status block indices
1402 			 * have been actually read (qed_sb_update_sb_idx)
1403 			 * prior to this check (*_has_*_work) so that
1404 			 * we won't write the "newer" value of the status block
1405 			 * to HW (if there was a DMA right after
1406 			 * qede_has_rx_work and if there is no rmb, the memory
1407 			 * reading (qed_sb_update_sb_idx) may be postponed
1408 			 * to right before *_ack_sb). In this case there
1409 			 * will never be another interrupt until there is
1410 			 * another update of the status block, while there
1411 			 * is still unhandled work.
1412 			 */
1413 			rmb();
1414 
1415 			if (!(qede_has_rx_work(fp->rxq) ||
1416 			      qede_has_tx_work(fp))) {
1417 				napi_complete(napi);
1418 				/* Update and reenable interrupts */
1419 				qed_sb_ack(fp->sb_info, IGU_INT_ENABLE,
1420 					   1 /*update*/);
1421 				break;
1422 			}
1423 		}
1424 	}
1425 
1426 	return work_done;
1427 }
1428 
1429 static irqreturn_t qede_msix_fp_int(int irq, void *fp_cookie)
1430 {
1431 	struct qede_fastpath *fp = fp_cookie;
1432 
1433 	qed_sb_ack(fp->sb_info, IGU_INT_DISABLE, 0 /*do not update*/);
1434 
1435 	napi_schedule_irqoff(&fp->napi);
1436 	return IRQ_HANDLED;
1437 }
1438 
1439 /* -------------------------------------------------------------------------
1440  * END OF FAST-PATH
1441  * -------------------------------------------------------------------------
1442  */
1443 
1444 static int qede_open(struct net_device *ndev);
1445 static int qede_close(struct net_device *ndev);
1446 static int qede_set_mac_addr(struct net_device *ndev, void *p);
1447 static void qede_set_rx_mode(struct net_device *ndev);
1448 static void qede_config_rx_mode(struct net_device *ndev);
1449 
1450 static int qede_set_ucast_rx_mac(struct qede_dev *edev,
1451 				 enum qed_filter_xcast_params_type opcode,
1452 				 unsigned char mac[ETH_ALEN])
1453 {
1454 	struct qed_filter_params filter_cmd;
1455 
1456 	memset(&filter_cmd, 0, sizeof(filter_cmd));
1457 	filter_cmd.type = QED_FILTER_TYPE_UCAST;
1458 	filter_cmd.filter.ucast.type = opcode;
1459 	filter_cmd.filter.ucast.mac_valid = 1;
1460 	ether_addr_copy(filter_cmd.filter.ucast.mac, mac);
1461 
1462 	return edev->ops->filter_config(edev->cdev, &filter_cmd);
1463 }
1464 
1465 static int qede_set_ucast_rx_vlan(struct qede_dev *edev,
1466 				  enum qed_filter_xcast_params_type opcode,
1467 				  u16 vid)
1468 {
1469 	struct qed_filter_params filter_cmd;
1470 
1471 	memset(&filter_cmd, 0, sizeof(filter_cmd));
1472 	filter_cmd.type = QED_FILTER_TYPE_UCAST;
1473 	filter_cmd.filter.ucast.type = opcode;
1474 	filter_cmd.filter.ucast.vlan_valid = 1;
1475 	filter_cmd.filter.ucast.vlan = vid;
1476 
1477 	return edev->ops->filter_config(edev->cdev, &filter_cmd);
1478 }
1479 
1480 void qede_fill_by_demand_stats(struct qede_dev *edev)
1481 {
1482 	struct qed_eth_stats stats;
1483 
1484 	edev->ops->get_vport_stats(edev->cdev, &stats);
1485 	edev->stats.no_buff_discards = stats.no_buff_discards;
1486 	edev->stats.rx_ucast_bytes = stats.rx_ucast_bytes;
1487 	edev->stats.rx_mcast_bytes = stats.rx_mcast_bytes;
1488 	edev->stats.rx_bcast_bytes = stats.rx_bcast_bytes;
1489 	edev->stats.rx_ucast_pkts = stats.rx_ucast_pkts;
1490 	edev->stats.rx_mcast_pkts = stats.rx_mcast_pkts;
1491 	edev->stats.rx_bcast_pkts = stats.rx_bcast_pkts;
1492 	edev->stats.mftag_filter_discards = stats.mftag_filter_discards;
1493 	edev->stats.mac_filter_discards = stats.mac_filter_discards;
1494 
1495 	edev->stats.tx_ucast_bytes = stats.tx_ucast_bytes;
1496 	edev->stats.tx_mcast_bytes = stats.tx_mcast_bytes;
1497 	edev->stats.tx_bcast_bytes = stats.tx_bcast_bytes;
1498 	edev->stats.tx_ucast_pkts = stats.tx_ucast_pkts;
1499 	edev->stats.tx_mcast_pkts = stats.tx_mcast_pkts;
1500 	edev->stats.tx_bcast_pkts = stats.tx_bcast_pkts;
1501 	edev->stats.tx_err_drop_pkts = stats.tx_err_drop_pkts;
1502 	edev->stats.coalesced_pkts = stats.tpa_coalesced_pkts;
1503 	edev->stats.coalesced_events = stats.tpa_coalesced_events;
1504 	edev->stats.coalesced_aborts_num = stats.tpa_aborts_num;
1505 	edev->stats.non_coalesced_pkts = stats.tpa_not_coalesced_pkts;
1506 	edev->stats.coalesced_bytes = stats.tpa_coalesced_bytes;
1507 
1508 	edev->stats.rx_64_byte_packets = stats.rx_64_byte_packets;
1509 	edev->stats.rx_127_byte_packets = stats.rx_127_byte_packets;
1510 	edev->stats.rx_255_byte_packets = stats.rx_255_byte_packets;
1511 	edev->stats.rx_511_byte_packets = stats.rx_511_byte_packets;
1512 	edev->stats.rx_1023_byte_packets = stats.rx_1023_byte_packets;
1513 	edev->stats.rx_1518_byte_packets = stats.rx_1518_byte_packets;
1514 	edev->stats.rx_1522_byte_packets = stats.rx_1522_byte_packets;
1515 	edev->stats.rx_2047_byte_packets = stats.rx_2047_byte_packets;
1516 	edev->stats.rx_4095_byte_packets = stats.rx_4095_byte_packets;
1517 	edev->stats.rx_9216_byte_packets = stats.rx_9216_byte_packets;
1518 	edev->stats.rx_16383_byte_packets = stats.rx_16383_byte_packets;
1519 	edev->stats.rx_crc_errors = stats.rx_crc_errors;
1520 	edev->stats.rx_mac_crtl_frames = stats.rx_mac_crtl_frames;
1521 	edev->stats.rx_pause_frames = stats.rx_pause_frames;
1522 	edev->stats.rx_pfc_frames = stats.rx_pfc_frames;
1523 	edev->stats.rx_align_errors = stats.rx_align_errors;
1524 	edev->stats.rx_carrier_errors = stats.rx_carrier_errors;
1525 	edev->stats.rx_oversize_packets = stats.rx_oversize_packets;
1526 	edev->stats.rx_jabbers = stats.rx_jabbers;
1527 	edev->stats.rx_undersize_packets = stats.rx_undersize_packets;
1528 	edev->stats.rx_fragments = stats.rx_fragments;
1529 	edev->stats.tx_64_byte_packets = stats.tx_64_byte_packets;
1530 	edev->stats.tx_65_to_127_byte_packets = stats.tx_65_to_127_byte_packets;
1531 	edev->stats.tx_128_to_255_byte_packets =
1532 				stats.tx_128_to_255_byte_packets;
1533 	edev->stats.tx_256_to_511_byte_packets =
1534 				stats.tx_256_to_511_byte_packets;
1535 	edev->stats.tx_512_to_1023_byte_packets =
1536 				stats.tx_512_to_1023_byte_packets;
1537 	edev->stats.tx_1024_to_1518_byte_packets =
1538 				stats.tx_1024_to_1518_byte_packets;
1539 	edev->stats.tx_1519_to_2047_byte_packets =
1540 				stats.tx_1519_to_2047_byte_packets;
1541 	edev->stats.tx_2048_to_4095_byte_packets =
1542 				stats.tx_2048_to_4095_byte_packets;
1543 	edev->stats.tx_4096_to_9216_byte_packets =
1544 				stats.tx_4096_to_9216_byte_packets;
1545 	edev->stats.tx_9217_to_16383_byte_packets =
1546 				stats.tx_9217_to_16383_byte_packets;
1547 	edev->stats.tx_pause_frames = stats.tx_pause_frames;
1548 	edev->stats.tx_pfc_frames = stats.tx_pfc_frames;
1549 	edev->stats.tx_lpi_entry_count = stats.tx_lpi_entry_count;
1550 	edev->stats.tx_total_collisions = stats.tx_total_collisions;
1551 	edev->stats.brb_truncates = stats.brb_truncates;
1552 	edev->stats.brb_discards = stats.brb_discards;
1553 	edev->stats.tx_mac_ctrl_frames = stats.tx_mac_ctrl_frames;
1554 }
1555 
1556 static struct rtnl_link_stats64 *qede_get_stats64(
1557 			    struct net_device *dev,
1558 			    struct rtnl_link_stats64 *stats)
1559 {
1560 	struct qede_dev *edev = netdev_priv(dev);
1561 
1562 	qede_fill_by_demand_stats(edev);
1563 
1564 	stats->rx_packets = edev->stats.rx_ucast_pkts +
1565 			    edev->stats.rx_mcast_pkts +
1566 			    edev->stats.rx_bcast_pkts;
1567 	stats->tx_packets = edev->stats.tx_ucast_pkts +
1568 			    edev->stats.tx_mcast_pkts +
1569 			    edev->stats.tx_bcast_pkts;
1570 
1571 	stats->rx_bytes = edev->stats.rx_ucast_bytes +
1572 			  edev->stats.rx_mcast_bytes +
1573 			  edev->stats.rx_bcast_bytes;
1574 
1575 	stats->tx_bytes = edev->stats.tx_ucast_bytes +
1576 			  edev->stats.tx_mcast_bytes +
1577 			  edev->stats.tx_bcast_bytes;
1578 
1579 	stats->tx_errors = edev->stats.tx_err_drop_pkts;
1580 	stats->multicast = edev->stats.rx_mcast_pkts +
1581 			   edev->stats.rx_bcast_pkts;
1582 
1583 	stats->rx_fifo_errors = edev->stats.no_buff_discards;
1584 
1585 	stats->collisions = edev->stats.tx_total_collisions;
1586 	stats->rx_crc_errors = edev->stats.rx_crc_errors;
1587 	stats->rx_frame_errors = edev->stats.rx_align_errors;
1588 
1589 	return stats;
1590 }
1591 
1592 static void qede_config_accept_any_vlan(struct qede_dev *edev, bool action)
1593 {
1594 	struct qed_update_vport_params params;
1595 	int rc;
1596 
1597 	/* Proceed only if action actually needs to be performed */
1598 	if (edev->accept_any_vlan == action)
1599 		return;
1600 
1601 	memset(&params, 0, sizeof(params));
1602 
1603 	params.vport_id = 0;
1604 	params.accept_any_vlan = action;
1605 	params.update_accept_any_vlan_flg = 1;
1606 
1607 	rc = edev->ops->vport_update(edev->cdev, &params);
1608 	if (rc) {
1609 		DP_ERR(edev, "Failed to %s accept-any-vlan\n",
1610 		       action ? "enable" : "disable");
1611 	} else {
1612 		DP_INFO(edev, "%s accept-any-vlan\n",
1613 			action ? "enabled" : "disabled");
1614 		edev->accept_any_vlan = action;
1615 	}
1616 }
1617 
1618 static int qede_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
1619 {
1620 	struct qede_dev *edev = netdev_priv(dev);
1621 	struct qede_vlan *vlan, *tmp;
1622 	int rc;
1623 
1624 	DP_VERBOSE(edev, NETIF_MSG_IFUP, "Adding vlan 0x%04x\n", vid);
1625 
1626 	vlan = kzalloc(sizeof(*vlan), GFP_KERNEL);
1627 	if (!vlan) {
1628 		DP_INFO(edev, "Failed to allocate struct for vlan\n");
1629 		return -ENOMEM;
1630 	}
1631 	INIT_LIST_HEAD(&vlan->list);
1632 	vlan->vid = vid;
1633 	vlan->configured = false;
1634 
1635 	/* Verify vlan isn't already configured */
1636 	list_for_each_entry(tmp, &edev->vlan_list, list) {
1637 		if (tmp->vid == vlan->vid) {
1638 			DP_VERBOSE(edev, (NETIF_MSG_IFUP | NETIF_MSG_IFDOWN),
1639 				   "vlan already configured\n");
1640 			kfree(vlan);
1641 			return -EEXIST;
1642 		}
1643 	}
1644 
1645 	/* If interface is down, cache this VLAN ID and return */
1646 	if (edev->state != QEDE_STATE_OPEN) {
1647 		DP_VERBOSE(edev, NETIF_MSG_IFDOWN,
1648 			   "Interface is down, VLAN %d will be configured when interface is up\n",
1649 			   vid);
1650 		if (vid != 0)
1651 			edev->non_configured_vlans++;
1652 		list_add(&vlan->list, &edev->vlan_list);
1653 
1654 		return 0;
1655 	}
1656 
1657 	/* Check for the filter limit.
1658 	 * Note - vlan0 has a reserved filter and can be added without
1659 	 * worrying about quota
1660 	 */
1661 	if ((edev->configured_vlans < edev->dev_info.num_vlan_filters) ||
1662 	    (vlan->vid == 0)) {
1663 		rc = qede_set_ucast_rx_vlan(edev,
1664 					    QED_FILTER_XCAST_TYPE_ADD,
1665 					    vlan->vid);
1666 		if (rc) {
1667 			DP_ERR(edev, "Failed to configure VLAN %d\n",
1668 			       vlan->vid);
1669 			kfree(vlan);
1670 			return -EINVAL;
1671 		}
1672 		vlan->configured = true;
1673 
1674 		/* vlan0 filter isn't consuming out of our quota */
1675 		if (vlan->vid != 0)
1676 			edev->configured_vlans++;
1677 	} else {
1678 		/* Out of quota; Activate accept-any-VLAN mode */
1679 		if (!edev->non_configured_vlans)
1680 			qede_config_accept_any_vlan(edev, true);
1681 
1682 		edev->non_configured_vlans++;
1683 	}
1684 
1685 	list_add(&vlan->list, &edev->vlan_list);
1686 
1687 	return 0;
1688 }
1689 
1690 static void qede_del_vlan_from_list(struct qede_dev *edev,
1691 				    struct qede_vlan *vlan)
1692 {
1693 	/* vlan0 filter isn't consuming out of our quota */
1694 	if (vlan->vid != 0) {
1695 		if (vlan->configured)
1696 			edev->configured_vlans--;
1697 		else
1698 			edev->non_configured_vlans--;
1699 	}
1700 
1701 	list_del(&vlan->list);
1702 	kfree(vlan);
1703 }
1704 
1705 static int qede_configure_vlan_filters(struct qede_dev *edev)
1706 {
1707 	int rc = 0, real_rc = 0, accept_any_vlan = 0;
1708 	struct qed_dev_eth_info *dev_info;
1709 	struct qede_vlan *vlan = NULL;
1710 
1711 	if (list_empty(&edev->vlan_list))
1712 		return 0;
1713 
1714 	dev_info = &edev->dev_info;
1715 
1716 	/* Configure non-configured vlans */
1717 	list_for_each_entry(vlan, &edev->vlan_list, list) {
1718 		if (vlan->configured)
1719 			continue;
1720 
1721 		/* We have used all our credits, now enable accept_any_vlan */
1722 		if ((vlan->vid != 0) &&
1723 		    (edev->configured_vlans == dev_info->num_vlan_filters)) {
1724 			accept_any_vlan = 1;
1725 			continue;
1726 		}
1727 
1728 		DP_VERBOSE(edev, NETIF_MSG_IFUP, "Adding vlan %d\n", vlan->vid);
1729 
1730 		rc = qede_set_ucast_rx_vlan(edev, QED_FILTER_XCAST_TYPE_ADD,
1731 					    vlan->vid);
1732 		if (rc) {
1733 			DP_ERR(edev, "Failed to configure VLAN %u\n",
1734 			       vlan->vid);
1735 			real_rc = rc;
1736 			continue;
1737 		}
1738 
1739 		vlan->configured = true;
1740 		/* vlan0 filter doesn't consume our VLAN filter's quota */
1741 		if (vlan->vid != 0) {
1742 			edev->non_configured_vlans--;
1743 			edev->configured_vlans++;
1744 		}
1745 	}
1746 
1747 	/* enable accept_any_vlan mode if we have more VLANs than credits,
1748 	 * or remove accept_any_vlan mode if we've actually removed
1749 	 * a non-configured vlan, and all remaining vlans are truly configured.
1750 	 */
1751 
1752 	if (accept_any_vlan)
1753 		qede_config_accept_any_vlan(edev, true);
1754 	else if (!edev->non_configured_vlans)
1755 		qede_config_accept_any_vlan(edev, false);
1756 
1757 	return real_rc;
1758 }
1759 
1760 static int qede_vlan_rx_kill_vid(struct net_device *dev, __be16 proto, u16 vid)
1761 {
1762 	struct qede_dev *edev = netdev_priv(dev);
1763 	struct qede_vlan *vlan = NULL;
1764 	int rc;
1765 
1766 	DP_VERBOSE(edev, NETIF_MSG_IFDOWN, "Removing vlan 0x%04x\n", vid);
1767 
1768 	/* Find whether entry exists */
1769 	list_for_each_entry(vlan, &edev->vlan_list, list)
1770 		if (vlan->vid == vid)
1771 			break;
1772 
1773 	if (!vlan || (vlan->vid != vid)) {
1774 		DP_VERBOSE(edev, (NETIF_MSG_IFUP | NETIF_MSG_IFDOWN),
1775 			   "Vlan isn't configured\n");
1776 		return 0;
1777 	}
1778 
1779 	if (edev->state != QEDE_STATE_OPEN) {
1780 		/* As interface is already down, we don't have a VPORT
1781 		 * instance to remove vlan filter. So just update vlan list
1782 		 */
1783 		DP_VERBOSE(edev, NETIF_MSG_IFDOWN,
1784 			   "Interface is down, removing VLAN from list only\n");
1785 		qede_del_vlan_from_list(edev, vlan);
1786 		return 0;
1787 	}
1788 
1789 	/* Remove vlan */
1790 	rc = qede_set_ucast_rx_vlan(edev, QED_FILTER_XCAST_TYPE_DEL, vid);
1791 	if (rc) {
1792 		DP_ERR(edev, "Failed to remove VLAN %d\n", vid);
1793 		return -EINVAL;
1794 	}
1795 
1796 	qede_del_vlan_from_list(edev, vlan);
1797 
1798 	/* We have removed a VLAN - try to see if we can
1799 	 * configure non-configured VLAN from the list.
1800 	 */
1801 	rc = qede_configure_vlan_filters(edev);
1802 
1803 	return rc;
1804 }
1805 
1806 static void qede_vlan_mark_nonconfigured(struct qede_dev *edev)
1807 {
1808 	struct qede_vlan *vlan = NULL;
1809 
1810 	if (list_empty(&edev->vlan_list))
1811 		return;
1812 
1813 	list_for_each_entry(vlan, &edev->vlan_list, list) {
1814 		if (!vlan->configured)
1815 			continue;
1816 
1817 		vlan->configured = false;
1818 
1819 		/* vlan0 filter isn't consuming out of our quota */
1820 		if (vlan->vid != 0) {
1821 			edev->non_configured_vlans++;
1822 			edev->configured_vlans--;
1823 		}
1824 
1825 		DP_VERBOSE(edev, NETIF_MSG_IFDOWN,
1826 			   "marked vlan %d as non-configured\n",
1827 			   vlan->vid);
1828 	}
1829 
1830 	edev->accept_any_vlan = false;
1831 }
1832 
1833 static const struct net_device_ops qede_netdev_ops = {
1834 	.ndo_open = qede_open,
1835 	.ndo_stop = qede_close,
1836 	.ndo_start_xmit = qede_start_xmit,
1837 	.ndo_set_rx_mode = qede_set_rx_mode,
1838 	.ndo_set_mac_address = qede_set_mac_addr,
1839 	.ndo_validate_addr = eth_validate_addr,
1840 	.ndo_change_mtu = qede_change_mtu,
1841 	.ndo_vlan_rx_add_vid = qede_vlan_rx_add_vid,
1842 	.ndo_vlan_rx_kill_vid = qede_vlan_rx_kill_vid,
1843 	.ndo_get_stats64 = qede_get_stats64,
1844 };
1845 
1846 /* -------------------------------------------------------------------------
1847  * START OF PROBE / REMOVE
1848  * -------------------------------------------------------------------------
1849  */
1850 
1851 static struct qede_dev *qede_alloc_etherdev(struct qed_dev *cdev,
1852 					    struct pci_dev *pdev,
1853 					    struct qed_dev_eth_info *info,
1854 					    u32 dp_module,
1855 					    u8 dp_level)
1856 {
1857 	struct net_device *ndev;
1858 	struct qede_dev *edev;
1859 
1860 	ndev = alloc_etherdev_mqs(sizeof(*edev),
1861 				  info->num_queues,
1862 				  info->num_queues);
1863 	if (!ndev) {
1864 		pr_err("etherdev allocation failed\n");
1865 		return NULL;
1866 	}
1867 
1868 	edev = netdev_priv(ndev);
1869 	edev->ndev = ndev;
1870 	edev->cdev = cdev;
1871 	edev->pdev = pdev;
1872 	edev->dp_module = dp_module;
1873 	edev->dp_level = dp_level;
1874 	edev->ops = qed_ops;
1875 	edev->q_num_rx_buffers = NUM_RX_BDS_DEF;
1876 	edev->q_num_tx_buffers = NUM_TX_BDS_DEF;
1877 
1878 	DP_INFO(edev, "Allocated netdev with 64 tx queues and 64 rx queues\n");
1879 
1880 	SET_NETDEV_DEV(ndev, &pdev->dev);
1881 
1882 	memset(&edev->stats, 0, sizeof(edev->stats));
1883 	memcpy(&edev->dev_info, info, sizeof(*info));
1884 
1885 	edev->num_tc = edev->dev_info.num_tc;
1886 
1887 	INIT_LIST_HEAD(&edev->vlan_list);
1888 
1889 	return edev;
1890 }
1891 
1892 static void qede_init_ndev(struct qede_dev *edev)
1893 {
1894 	struct net_device *ndev = edev->ndev;
1895 	struct pci_dev *pdev = edev->pdev;
1896 	u32 hw_features;
1897 
1898 	pci_set_drvdata(pdev, ndev);
1899 
1900 	ndev->mem_start = edev->dev_info.common.pci_mem_start;
1901 	ndev->base_addr = ndev->mem_start;
1902 	ndev->mem_end = edev->dev_info.common.pci_mem_end;
1903 	ndev->irq = edev->dev_info.common.pci_irq;
1904 
1905 	ndev->watchdog_timeo = TX_TIMEOUT;
1906 
1907 	ndev->netdev_ops = &qede_netdev_ops;
1908 
1909 	qede_set_ethtool_ops(ndev);
1910 
1911 	/* user-changeble features */
1912 	hw_features = NETIF_F_GRO | NETIF_F_SG |
1913 		      NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1914 		      NETIF_F_TSO | NETIF_F_TSO6;
1915 
1916 	ndev->vlan_features = hw_features | NETIF_F_RXHASH | NETIF_F_RXCSUM |
1917 			      NETIF_F_HIGHDMA;
1918 	ndev->features = hw_features | NETIF_F_RXHASH | NETIF_F_RXCSUM |
1919 			 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HIGHDMA |
1920 			 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_CTAG_TX;
1921 
1922 	ndev->hw_features = hw_features;
1923 
1924 	/* Set network device HW mac */
1925 	ether_addr_copy(edev->ndev->dev_addr, edev->dev_info.common.hw_mac);
1926 }
1927 
1928 /* This function converts from 32b param to two params of level and module
1929  * Input 32b decoding:
1930  * b31 - enable all NOTICE prints. NOTICE prints are for deviation from the
1931  * 'happy' flow, e.g. memory allocation failed.
1932  * b30 - enable all INFO prints. INFO prints are for major steps in the flow
1933  * and provide important parameters.
1934  * b29-b0 - per-module bitmap, where each bit enables VERBOSE prints of that
1935  * module. VERBOSE prints are for tracking the specific flow in low level.
1936  *
1937  * Notice that the level should be that of the lowest required logs.
1938  */
1939 void qede_config_debug(uint debug, u32 *p_dp_module, u8 *p_dp_level)
1940 {
1941 	*p_dp_level = QED_LEVEL_NOTICE;
1942 	*p_dp_module = 0;
1943 
1944 	if (debug & QED_LOG_VERBOSE_MASK) {
1945 		*p_dp_level = QED_LEVEL_VERBOSE;
1946 		*p_dp_module = (debug & 0x3FFFFFFF);
1947 	} else if (debug & QED_LOG_INFO_MASK) {
1948 		*p_dp_level = QED_LEVEL_INFO;
1949 	} else if (debug & QED_LOG_NOTICE_MASK) {
1950 		*p_dp_level = QED_LEVEL_NOTICE;
1951 	}
1952 }
1953 
1954 static void qede_free_fp_array(struct qede_dev *edev)
1955 {
1956 	if (edev->fp_array) {
1957 		struct qede_fastpath *fp;
1958 		int i;
1959 
1960 		for_each_rss(i) {
1961 			fp = &edev->fp_array[i];
1962 
1963 			kfree(fp->sb_info);
1964 			kfree(fp->rxq);
1965 			kfree(fp->txqs);
1966 		}
1967 		kfree(edev->fp_array);
1968 	}
1969 	edev->num_rss = 0;
1970 }
1971 
1972 static int qede_alloc_fp_array(struct qede_dev *edev)
1973 {
1974 	struct qede_fastpath *fp;
1975 	int i;
1976 
1977 	edev->fp_array = kcalloc(QEDE_RSS_CNT(edev),
1978 				 sizeof(*edev->fp_array), GFP_KERNEL);
1979 	if (!edev->fp_array) {
1980 		DP_NOTICE(edev, "fp array allocation failed\n");
1981 		goto err;
1982 	}
1983 
1984 	for_each_rss(i) {
1985 		fp = &edev->fp_array[i];
1986 
1987 		fp->sb_info = kcalloc(1, sizeof(*fp->sb_info), GFP_KERNEL);
1988 		if (!fp->sb_info) {
1989 			DP_NOTICE(edev, "sb info struct allocation failed\n");
1990 			goto err;
1991 		}
1992 
1993 		fp->rxq = kcalloc(1, sizeof(*fp->rxq), GFP_KERNEL);
1994 		if (!fp->rxq) {
1995 			DP_NOTICE(edev, "RXQ struct allocation failed\n");
1996 			goto err;
1997 		}
1998 
1999 		fp->txqs = kcalloc(edev->num_tc, sizeof(*fp->txqs), GFP_KERNEL);
2000 		if (!fp->txqs) {
2001 			DP_NOTICE(edev, "TXQ array allocation failed\n");
2002 			goto err;
2003 		}
2004 	}
2005 
2006 	return 0;
2007 err:
2008 	qede_free_fp_array(edev);
2009 	return -ENOMEM;
2010 }
2011 
2012 static void qede_sp_task(struct work_struct *work)
2013 {
2014 	struct qede_dev *edev = container_of(work, struct qede_dev,
2015 					     sp_task.work);
2016 	mutex_lock(&edev->qede_lock);
2017 
2018 	if (edev->state == QEDE_STATE_OPEN) {
2019 		if (test_and_clear_bit(QEDE_SP_RX_MODE, &edev->sp_flags))
2020 			qede_config_rx_mode(edev->ndev);
2021 	}
2022 
2023 	mutex_unlock(&edev->qede_lock);
2024 }
2025 
2026 static void qede_update_pf_params(struct qed_dev *cdev)
2027 {
2028 	struct qed_pf_params pf_params;
2029 
2030 	/* 16 rx + 16 tx */
2031 	memset(&pf_params, 0, sizeof(struct qed_pf_params));
2032 	pf_params.eth_pf_params.num_cons = 32;
2033 	qed_ops->common->update_pf_params(cdev, &pf_params);
2034 }
2035 
2036 enum qede_probe_mode {
2037 	QEDE_PROBE_NORMAL,
2038 };
2039 
2040 static int __qede_probe(struct pci_dev *pdev, u32 dp_module, u8 dp_level,
2041 			enum qede_probe_mode mode)
2042 {
2043 	struct qed_slowpath_params params;
2044 	struct qed_dev_eth_info dev_info;
2045 	struct qede_dev *edev;
2046 	struct qed_dev *cdev;
2047 	int rc;
2048 
2049 	if (unlikely(dp_level & QED_LEVEL_INFO))
2050 		pr_notice("Starting qede probe\n");
2051 
2052 	cdev = qed_ops->common->probe(pdev, QED_PROTOCOL_ETH,
2053 				      dp_module, dp_level);
2054 	if (!cdev) {
2055 		rc = -ENODEV;
2056 		goto err0;
2057 	}
2058 
2059 	qede_update_pf_params(cdev);
2060 
2061 	/* Start the Slowpath-process */
2062 	memset(&params, 0, sizeof(struct qed_slowpath_params));
2063 	params.int_mode = QED_INT_MODE_MSIX;
2064 	params.drv_major = QEDE_MAJOR_VERSION;
2065 	params.drv_minor = QEDE_MINOR_VERSION;
2066 	params.drv_rev = QEDE_REVISION_VERSION;
2067 	params.drv_eng = QEDE_ENGINEERING_VERSION;
2068 	strlcpy(params.name, "qede LAN", QED_DRV_VER_STR_SIZE);
2069 	rc = qed_ops->common->slowpath_start(cdev, &params);
2070 	if (rc) {
2071 		pr_notice("Cannot start slowpath\n");
2072 		goto err1;
2073 	}
2074 
2075 	/* Learn information crucial for qede to progress */
2076 	rc = qed_ops->fill_dev_info(cdev, &dev_info);
2077 	if (rc)
2078 		goto err2;
2079 
2080 	edev = qede_alloc_etherdev(cdev, pdev, &dev_info, dp_module,
2081 				   dp_level);
2082 	if (!edev) {
2083 		rc = -ENOMEM;
2084 		goto err2;
2085 	}
2086 
2087 	qede_init_ndev(edev);
2088 
2089 	rc = register_netdev(edev->ndev);
2090 	if (rc) {
2091 		DP_NOTICE(edev, "Cannot register net-device\n");
2092 		goto err3;
2093 	}
2094 
2095 	edev->ops->common->set_id(cdev, edev->ndev->name, DRV_MODULE_VERSION);
2096 
2097 	edev->ops->register_ops(cdev, &qede_ll_ops, edev);
2098 
2099 	INIT_DELAYED_WORK(&edev->sp_task, qede_sp_task);
2100 	mutex_init(&edev->qede_lock);
2101 
2102 	DP_INFO(edev, "Ending successfully qede probe\n");
2103 
2104 	return 0;
2105 
2106 err3:
2107 	free_netdev(edev->ndev);
2108 err2:
2109 	qed_ops->common->slowpath_stop(cdev);
2110 err1:
2111 	qed_ops->common->remove(cdev);
2112 err0:
2113 	return rc;
2114 }
2115 
2116 static int qede_probe(struct pci_dev *pdev, const struct pci_device_id *id)
2117 {
2118 	u32 dp_module = 0;
2119 	u8 dp_level = 0;
2120 
2121 	qede_config_debug(debug, &dp_module, &dp_level);
2122 
2123 	return __qede_probe(pdev, dp_module, dp_level,
2124 			    QEDE_PROBE_NORMAL);
2125 }
2126 
2127 enum qede_remove_mode {
2128 	QEDE_REMOVE_NORMAL,
2129 };
2130 
2131 static void __qede_remove(struct pci_dev *pdev, enum qede_remove_mode mode)
2132 {
2133 	struct net_device *ndev = pci_get_drvdata(pdev);
2134 	struct qede_dev *edev = netdev_priv(ndev);
2135 	struct qed_dev *cdev = edev->cdev;
2136 
2137 	DP_INFO(edev, "Starting qede_remove\n");
2138 
2139 	cancel_delayed_work_sync(&edev->sp_task);
2140 	unregister_netdev(ndev);
2141 
2142 	edev->ops->common->set_power_state(cdev, PCI_D0);
2143 
2144 	pci_set_drvdata(pdev, NULL);
2145 
2146 	free_netdev(ndev);
2147 
2148 	/* Use global ops since we've freed edev */
2149 	qed_ops->common->slowpath_stop(cdev);
2150 	qed_ops->common->remove(cdev);
2151 
2152 	pr_notice("Ending successfully qede_remove\n");
2153 }
2154 
2155 static void qede_remove(struct pci_dev *pdev)
2156 {
2157 	__qede_remove(pdev, QEDE_REMOVE_NORMAL);
2158 }
2159 
2160 /* -------------------------------------------------------------------------
2161  * START OF LOAD / UNLOAD
2162  * -------------------------------------------------------------------------
2163  */
2164 
2165 static int qede_set_num_queues(struct qede_dev *edev)
2166 {
2167 	int rc;
2168 	u16 rss_num;
2169 
2170 	/* Setup queues according to possible resources*/
2171 	if (edev->req_rss)
2172 		rss_num = edev->req_rss;
2173 	else
2174 		rss_num = netif_get_num_default_rss_queues() *
2175 			  edev->dev_info.common.num_hwfns;
2176 
2177 	rss_num = min_t(u16, QEDE_MAX_RSS_CNT(edev), rss_num);
2178 
2179 	rc = edev->ops->common->set_fp_int(edev->cdev, rss_num);
2180 	if (rc > 0) {
2181 		/* Managed to request interrupts for our queues */
2182 		edev->num_rss = rc;
2183 		DP_INFO(edev, "Managed %d [of %d] RSS queues\n",
2184 			QEDE_RSS_CNT(edev), rss_num);
2185 		rc = 0;
2186 	}
2187 	return rc;
2188 }
2189 
2190 static void qede_free_mem_sb(struct qede_dev *edev,
2191 			     struct qed_sb_info *sb_info)
2192 {
2193 	if (sb_info->sb_virt)
2194 		dma_free_coherent(&edev->pdev->dev, sizeof(*sb_info->sb_virt),
2195 				  (void *)sb_info->sb_virt, sb_info->sb_phys);
2196 }
2197 
2198 /* This function allocates fast-path status block memory */
2199 static int qede_alloc_mem_sb(struct qede_dev *edev,
2200 			     struct qed_sb_info *sb_info,
2201 			     u16 sb_id)
2202 {
2203 	struct status_block *sb_virt;
2204 	dma_addr_t sb_phys;
2205 	int rc;
2206 
2207 	sb_virt = dma_alloc_coherent(&edev->pdev->dev,
2208 				     sizeof(*sb_virt),
2209 				     &sb_phys, GFP_KERNEL);
2210 	if (!sb_virt) {
2211 		DP_ERR(edev, "Status block allocation failed\n");
2212 		return -ENOMEM;
2213 	}
2214 
2215 	rc = edev->ops->common->sb_init(edev->cdev, sb_info,
2216 					sb_virt, sb_phys, sb_id,
2217 					QED_SB_TYPE_L2_QUEUE);
2218 	if (rc) {
2219 		DP_ERR(edev, "Status block initialization failed\n");
2220 		dma_free_coherent(&edev->pdev->dev, sizeof(*sb_virt),
2221 				  sb_virt, sb_phys);
2222 		return rc;
2223 	}
2224 
2225 	return 0;
2226 }
2227 
2228 static void qede_free_rx_buffers(struct qede_dev *edev,
2229 				 struct qede_rx_queue *rxq)
2230 {
2231 	u16 i;
2232 
2233 	for (i = rxq->sw_rx_cons; i != rxq->sw_rx_prod; i++) {
2234 		struct sw_rx_data *rx_buf;
2235 		struct page *data;
2236 
2237 		rx_buf = &rxq->sw_rx_ring[i & NUM_RX_BDS_MAX];
2238 		data = rx_buf->data;
2239 
2240 		dma_unmap_page(&edev->pdev->dev,
2241 			       rx_buf->mapping,
2242 			       PAGE_SIZE, DMA_FROM_DEVICE);
2243 
2244 		rx_buf->data = NULL;
2245 		__free_page(data);
2246 	}
2247 }
2248 
2249 static void qede_free_sge_mem(struct qede_dev *edev,
2250 			      struct qede_rx_queue *rxq) {
2251 	int i;
2252 
2253 	if (edev->gro_disable)
2254 		return;
2255 
2256 	for (i = 0; i < ETH_TPA_MAX_AGGS_NUM; i++) {
2257 		struct qede_agg_info *tpa_info = &rxq->tpa_info[i];
2258 		struct sw_rx_data *replace_buf = &tpa_info->replace_buf;
2259 
2260 		if (replace_buf) {
2261 			dma_unmap_page(&edev->pdev->dev,
2262 				       dma_unmap_addr(replace_buf, mapping),
2263 				       PAGE_SIZE, DMA_FROM_DEVICE);
2264 			__free_page(replace_buf->data);
2265 		}
2266 	}
2267 }
2268 
2269 static void qede_free_mem_rxq(struct qede_dev *edev,
2270 			      struct qede_rx_queue *rxq)
2271 {
2272 	qede_free_sge_mem(edev, rxq);
2273 
2274 	/* Free rx buffers */
2275 	qede_free_rx_buffers(edev, rxq);
2276 
2277 	/* Free the parallel SW ring */
2278 	kfree(rxq->sw_rx_ring);
2279 
2280 	/* Free the real RQ ring used by FW */
2281 	edev->ops->common->chain_free(edev->cdev, &rxq->rx_bd_ring);
2282 	edev->ops->common->chain_free(edev->cdev, &rxq->rx_comp_ring);
2283 }
2284 
2285 static int qede_alloc_rx_buffer(struct qede_dev *edev,
2286 				struct qede_rx_queue *rxq)
2287 {
2288 	struct sw_rx_data *sw_rx_data;
2289 	struct eth_rx_bd *rx_bd;
2290 	dma_addr_t mapping;
2291 	struct page *data;
2292 	u16 rx_buf_size;
2293 
2294 	rx_buf_size = rxq->rx_buf_size;
2295 
2296 	data = alloc_pages(GFP_ATOMIC, 0);
2297 	if (unlikely(!data)) {
2298 		DP_NOTICE(edev, "Failed to allocate Rx data [page]\n");
2299 		return -ENOMEM;
2300 	}
2301 
2302 	/* Map the entire page as it would be used
2303 	 * for multiple RX buffer segment size mapping.
2304 	 */
2305 	mapping = dma_map_page(&edev->pdev->dev, data, 0,
2306 			       PAGE_SIZE, DMA_FROM_DEVICE);
2307 	if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
2308 		__free_page(data);
2309 		DP_NOTICE(edev, "Failed to map Rx buffer\n");
2310 		return -ENOMEM;
2311 	}
2312 
2313 	sw_rx_data = &rxq->sw_rx_ring[rxq->sw_rx_prod & NUM_RX_BDS_MAX];
2314 	sw_rx_data->page_offset = 0;
2315 	sw_rx_data->data = data;
2316 	sw_rx_data->mapping = mapping;
2317 
2318 	/* Advance PROD and get BD pointer */
2319 	rx_bd = (struct eth_rx_bd *)qed_chain_produce(&rxq->rx_bd_ring);
2320 	WARN_ON(!rx_bd);
2321 	rx_bd->addr.hi = cpu_to_le32(upper_32_bits(mapping));
2322 	rx_bd->addr.lo = cpu_to_le32(lower_32_bits(mapping));
2323 
2324 	rxq->sw_rx_prod++;
2325 
2326 	return 0;
2327 }
2328 
2329 static int qede_alloc_sge_mem(struct qede_dev *edev,
2330 			      struct qede_rx_queue *rxq)
2331 {
2332 	dma_addr_t mapping;
2333 	int i;
2334 
2335 	if (edev->gro_disable)
2336 		return 0;
2337 
2338 	if (edev->ndev->mtu > PAGE_SIZE) {
2339 		edev->gro_disable = 1;
2340 		return 0;
2341 	}
2342 
2343 	for (i = 0; i < ETH_TPA_MAX_AGGS_NUM; i++) {
2344 		struct qede_agg_info *tpa_info = &rxq->tpa_info[i];
2345 		struct sw_rx_data *replace_buf = &tpa_info->replace_buf;
2346 
2347 		replace_buf->data = alloc_pages(GFP_ATOMIC, 0);
2348 		if (unlikely(!replace_buf->data)) {
2349 			DP_NOTICE(edev,
2350 				  "Failed to allocate TPA skb pool [replacement buffer]\n");
2351 			goto err;
2352 		}
2353 
2354 		mapping = dma_map_page(&edev->pdev->dev, replace_buf->data, 0,
2355 				       rxq->rx_buf_size, DMA_FROM_DEVICE);
2356 		if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
2357 			DP_NOTICE(edev,
2358 				  "Failed to map TPA replacement buffer\n");
2359 			goto err;
2360 		}
2361 
2362 		dma_unmap_addr_set(replace_buf, mapping, mapping);
2363 		tpa_info->replace_buf.page_offset = 0;
2364 
2365 		tpa_info->replace_buf_mapping = mapping;
2366 		tpa_info->agg_state = QEDE_AGG_STATE_NONE;
2367 	}
2368 
2369 	return 0;
2370 err:
2371 	qede_free_sge_mem(edev, rxq);
2372 	edev->gro_disable = 1;
2373 	return -ENOMEM;
2374 }
2375 
2376 /* This function allocates all memory needed per Rx queue */
2377 static int qede_alloc_mem_rxq(struct qede_dev *edev,
2378 			      struct qede_rx_queue *rxq)
2379 {
2380 	int i, rc, size, num_allocated;
2381 
2382 	rxq->num_rx_buffers = edev->q_num_rx_buffers;
2383 
2384 	rxq->rx_buf_size = NET_IP_ALIGN + ETH_OVERHEAD +
2385 			   edev->ndev->mtu;
2386 	if (rxq->rx_buf_size > PAGE_SIZE)
2387 		rxq->rx_buf_size = PAGE_SIZE;
2388 
2389 	/* Segment size to spilt a page in multiple equal parts */
2390 	rxq->rx_buf_seg_size = roundup_pow_of_two(rxq->rx_buf_size);
2391 
2392 	/* Allocate the parallel driver ring for Rx buffers */
2393 	size = sizeof(*rxq->sw_rx_ring) * RX_RING_SIZE;
2394 	rxq->sw_rx_ring = kzalloc(size, GFP_KERNEL);
2395 	if (!rxq->sw_rx_ring) {
2396 		DP_ERR(edev, "Rx buffers ring allocation failed\n");
2397 		goto err;
2398 	}
2399 
2400 	/* Allocate FW Rx ring  */
2401 	rc = edev->ops->common->chain_alloc(edev->cdev,
2402 					    QED_CHAIN_USE_TO_CONSUME_PRODUCE,
2403 					    QED_CHAIN_MODE_NEXT_PTR,
2404 					    RX_RING_SIZE,
2405 					    sizeof(struct eth_rx_bd),
2406 					    &rxq->rx_bd_ring);
2407 
2408 	if (rc)
2409 		goto err;
2410 
2411 	/* Allocate FW completion ring */
2412 	rc = edev->ops->common->chain_alloc(edev->cdev,
2413 					    QED_CHAIN_USE_TO_CONSUME,
2414 					    QED_CHAIN_MODE_PBL,
2415 					    RX_RING_SIZE,
2416 					    sizeof(union eth_rx_cqe),
2417 					    &rxq->rx_comp_ring);
2418 	if (rc)
2419 		goto err;
2420 
2421 	/* Allocate buffers for the Rx ring */
2422 	for (i = 0; i < rxq->num_rx_buffers; i++) {
2423 		rc = qede_alloc_rx_buffer(edev, rxq);
2424 		if (rc)
2425 			break;
2426 	}
2427 	num_allocated = i;
2428 	if (!num_allocated) {
2429 		DP_ERR(edev, "Rx buffers allocation failed\n");
2430 		goto err;
2431 	} else if (num_allocated < rxq->num_rx_buffers) {
2432 		DP_NOTICE(edev,
2433 			  "Allocated less buffers than desired (%d allocated)\n",
2434 			  num_allocated);
2435 	}
2436 
2437 	qede_alloc_sge_mem(edev, rxq);
2438 
2439 	return 0;
2440 
2441 err:
2442 	qede_free_mem_rxq(edev, rxq);
2443 	return -ENOMEM;
2444 }
2445 
2446 static void qede_free_mem_txq(struct qede_dev *edev,
2447 			      struct qede_tx_queue *txq)
2448 {
2449 	/* Free the parallel SW ring */
2450 	kfree(txq->sw_tx_ring);
2451 
2452 	/* Free the real RQ ring used by FW */
2453 	edev->ops->common->chain_free(edev->cdev, &txq->tx_pbl);
2454 }
2455 
2456 /* This function allocates all memory needed per Tx queue */
2457 static int qede_alloc_mem_txq(struct qede_dev *edev,
2458 			      struct qede_tx_queue *txq)
2459 {
2460 	int size, rc;
2461 	union eth_tx_bd_types *p_virt;
2462 
2463 	txq->num_tx_buffers = edev->q_num_tx_buffers;
2464 
2465 	/* Allocate the parallel driver ring for Tx buffers */
2466 	size = sizeof(*txq->sw_tx_ring) * NUM_TX_BDS_MAX;
2467 	txq->sw_tx_ring = kzalloc(size, GFP_KERNEL);
2468 	if (!txq->sw_tx_ring) {
2469 		DP_NOTICE(edev, "Tx buffers ring allocation failed\n");
2470 		goto err;
2471 	}
2472 
2473 	rc = edev->ops->common->chain_alloc(edev->cdev,
2474 					    QED_CHAIN_USE_TO_CONSUME_PRODUCE,
2475 					    QED_CHAIN_MODE_PBL,
2476 					    NUM_TX_BDS_MAX,
2477 					    sizeof(*p_virt),
2478 					    &txq->tx_pbl);
2479 	if (rc)
2480 		goto err;
2481 
2482 	return 0;
2483 
2484 err:
2485 	qede_free_mem_txq(edev, txq);
2486 	return -ENOMEM;
2487 }
2488 
2489 /* This function frees all memory of a single fp */
2490 static void qede_free_mem_fp(struct qede_dev *edev,
2491 			     struct qede_fastpath *fp)
2492 {
2493 	int tc;
2494 
2495 	qede_free_mem_sb(edev, fp->sb_info);
2496 
2497 	qede_free_mem_rxq(edev, fp->rxq);
2498 
2499 	for (tc = 0; tc < edev->num_tc; tc++)
2500 		qede_free_mem_txq(edev, &fp->txqs[tc]);
2501 }
2502 
2503 /* This function allocates all memory needed for a single fp (i.e. an entity
2504  * which contains status block, one rx queue and multiple per-TC tx queues.
2505  */
2506 static int qede_alloc_mem_fp(struct qede_dev *edev,
2507 			     struct qede_fastpath *fp)
2508 {
2509 	int rc, tc;
2510 
2511 	rc = qede_alloc_mem_sb(edev, fp->sb_info, fp->rss_id);
2512 	if (rc)
2513 		goto err;
2514 
2515 	rc = qede_alloc_mem_rxq(edev, fp->rxq);
2516 	if (rc)
2517 		goto err;
2518 
2519 	for (tc = 0; tc < edev->num_tc; tc++) {
2520 		rc = qede_alloc_mem_txq(edev, &fp->txqs[tc]);
2521 		if (rc)
2522 			goto err;
2523 	}
2524 
2525 	return 0;
2526 
2527 err:
2528 	qede_free_mem_fp(edev, fp);
2529 	return -ENOMEM;
2530 }
2531 
2532 static void qede_free_mem_load(struct qede_dev *edev)
2533 {
2534 	int i;
2535 
2536 	for_each_rss(i) {
2537 		struct qede_fastpath *fp = &edev->fp_array[i];
2538 
2539 		qede_free_mem_fp(edev, fp);
2540 	}
2541 }
2542 
2543 /* This function allocates all qede memory at NIC load. */
2544 static int qede_alloc_mem_load(struct qede_dev *edev)
2545 {
2546 	int rc = 0, rss_id;
2547 
2548 	for (rss_id = 0; rss_id < QEDE_RSS_CNT(edev); rss_id++) {
2549 		struct qede_fastpath *fp = &edev->fp_array[rss_id];
2550 
2551 		rc = qede_alloc_mem_fp(edev, fp);
2552 		if (rc)
2553 			break;
2554 	}
2555 
2556 	if (rss_id != QEDE_RSS_CNT(edev)) {
2557 		/* Failed allocating memory for all the queues */
2558 		if (!rss_id) {
2559 			DP_ERR(edev,
2560 			       "Failed to allocate memory for the leading queue\n");
2561 			rc = -ENOMEM;
2562 		} else {
2563 			DP_NOTICE(edev,
2564 				  "Failed to allocate memory for all of RSS queues\n Desired: %d queues, allocated: %d queues\n",
2565 				  QEDE_RSS_CNT(edev), rss_id);
2566 		}
2567 		edev->num_rss = rss_id;
2568 	}
2569 
2570 	return 0;
2571 }
2572 
2573 /* This function inits fp content and resets the SB, RXQ and TXQ structures */
2574 static void qede_init_fp(struct qede_dev *edev)
2575 {
2576 	int rss_id, txq_index, tc;
2577 	struct qede_fastpath *fp;
2578 
2579 	for_each_rss(rss_id) {
2580 		fp = &edev->fp_array[rss_id];
2581 
2582 		fp->edev = edev;
2583 		fp->rss_id = rss_id;
2584 
2585 		memset((void *)&fp->napi, 0, sizeof(fp->napi));
2586 
2587 		memset((void *)fp->sb_info, 0, sizeof(*fp->sb_info));
2588 
2589 		memset((void *)fp->rxq, 0, sizeof(*fp->rxq));
2590 		fp->rxq->rxq_id = rss_id;
2591 
2592 		memset((void *)fp->txqs, 0, (edev->num_tc * sizeof(*fp->txqs)));
2593 		for (tc = 0; tc < edev->num_tc; tc++) {
2594 			txq_index = tc * QEDE_RSS_CNT(edev) + rss_id;
2595 			fp->txqs[tc].index = txq_index;
2596 		}
2597 
2598 		snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
2599 			 edev->ndev->name, rss_id);
2600 	}
2601 
2602 	edev->gro_disable = !(edev->ndev->features & NETIF_F_GRO);
2603 }
2604 
2605 static int qede_set_real_num_queues(struct qede_dev *edev)
2606 {
2607 	int rc = 0;
2608 
2609 	rc = netif_set_real_num_tx_queues(edev->ndev, QEDE_TSS_CNT(edev));
2610 	if (rc) {
2611 		DP_NOTICE(edev, "Failed to set real number of Tx queues\n");
2612 		return rc;
2613 	}
2614 	rc = netif_set_real_num_rx_queues(edev->ndev, QEDE_RSS_CNT(edev));
2615 	if (rc) {
2616 		DP_NOTICE(edev, "Failed to set real number of Rx queues\n");
2617 		return rc;
2618 	}
2619 
2620 	return 0;
2621 }
2622 
2623 static void qede_napi_disable_remove(struct qede_dev *edev)
2624 {
2625 	int i;
2626 
2627 	for_each_rss(i) {
2628 		napi_disable(&edev->fp_array[i].napi);
2629 
2630 		netif_napi_del(&edev->fp_array[i].napi);
2631 	}
2632 }
2633 
2634 static void qede_napi_add_enable(struct qede_dev *edev)
2635 {
2636 	int i;
2637 
2638 	/* Add NAPI objects */
2639 	for_each_rss(i) {
2640 		netif_napi_add(edev->ndev, &edev->fp_array[i].napi,
2641 			       qede_poll, NAPI_POLL_WEIGHT);
2642 		napi_enable(&edev->fp_array[i].napi);
2643 	}
2644 }
2645 
2646 static void qede_sync_free_irqs(struct qede_dev *edev)
2647 {
2648 	int i;
2649 
2650 	for (i = 0; i < edev->int_info.used_cnt; i++) {
2651 		if (edev->int_info.msix_cnt) {
2652 			synchronize_irq(edev->int_info.msix[i].vector);
2653 			free_irq(edev->int_info.msix[i].vector,
2654 				 &edev->fp_array[i]);
2655 		} else {
2656 			edev->ops->common->simd_handler_clean(edev->cdev, i);
2657 		}
2658 	}
2659 
2660 	edev->int_info.used_cnt = 0;
2661 }
2662 
2663 static int qede_req_msix_irqs(struct qede_dev *edev)
2664 {
2665 	int i, rc;
2666 
2667 	/* Sanitize number of interrupts == number of prepared RSS queues */
2668 	if (QEDE_RSS_CNT(edev) > edev->int_info.msix_cnt) {
2669 		DP_ERR(edev,
2670 		       "Interrupt mismatch: %d RSS queues > %d MSI-x vectors\n",
2671 		       QEDE_RSS_CNT(edev), edev->int_info.msix_cnt);
2672 		return -EINVAL;
2673 	}
2674 
2675 	for (i = 0; i < QEDE_RSS_CNT(edev); i++) {
2676 		rc = request_irq(edev->int_info.msix[i].vector,
2677 				 qede_msix_fp_int, 0, edev->fp_array[i].name,
2678 				 &edev->fp_array[i]);
2679 		if (rc) {
2680 			DP_ERR(edev, "Request fp %d irq failed\n", i);
2681 			qede_sync_free_irqs(edev);
2682 			return rc;
2683 		}
2684 		DP_VERBOSE(edev, NETIF_MSG_INTR,
2685 			   "Requested fp irq for %s [entry %d]. Cookie is at %p\n",
2686 			   edev->fp_array[i].name, i,
2687 			   &edev->fp_array[i]);
2688 		edev->int_info.used_cnt++;
2689 	}
2690 
2691 	return 0;
2692 }
2693 
2694 static void qede_simd_fp_handler(void *cookie)
2695 {
2696 	struct qede_fastpath *fp = (struct qede_fastpath *)cookie;
2697 
2698 	napi_schedule_irqoff(&fp->napi);
2699 }
2700 
2701 static int qede_setup_irqs(struct qede_dev *edev)
2702 {
2703 	int i, rc = 0;
2704 
2705 	/* Learn Interrupt configuration */
2706 	rc = edev->ops->common->get_fp_int(edev->cdev, &edev->int_info);
2707 	if (rc)
2708 		return rc;
2709 
2710 	if (edev->int_info.msix_cnt) {
2711 		rc = qede_req_msix_irqs(edev);
2712 		if (rc)
2713 			return rc;
2714 		edev->ndev->irq = edev->int_info.msix[0].vector;
2715 	} else {
2716 		const struct qed_common_ops *ops;
2717 
2718 		/* qed should learn receive the RSS ids and callbacks */
2719 		ops = edev->ops->common;
2720 		for (i = 0; i < QEDE_RSS_CNT(edev); i++)
2721 			ops->simd_handler_config(edev->cdev,
2722 						 &edev->fp_array[i], i,
2723 						 qede_simd_fp_handler);
2724 		edev->int_info.used_cnt = QEDE_RSS_CNT(edev);
2725 	}
2726 	return 0;
2727 }
2728 
2729 static int qede_drain_txq(struct qede_dev *edev,
2730 			  struct qede_tx_queue *txq,
2731 			  bool allow_drain)
2732 {
2733 	int rc, cnt = 1000;
2734 
2735 	while (txq->sw_tx_cons != txq->sw_tx_prod) {
2736 		if (!cnt) {
2737 			if (allow_drain) {
2738 				DP_NOTICE(edev,
2739 					  "Tx queue[%d] is stuck, requesting MCP to drain\n",
2740 					  txq->index);
2741 				rc = edev->ops->common->drain(edev->cdev);
2742 				if (rc)
2743 					return rc;
2744 				return qede_drain_txq(edev, txq, false);
2745 			}
2746 			DP_NOTICE(edev,
2747 				  "Timeout waiting for tx queue[%d]: PROD=%d, CONS=%d\n",
2748 				  txq->index, txq->sw_tx_prod,
2749 				  txq->sw_tx_cons);
2750 			return -ENODEV;
2751 		}
2752 		cnt--;
2753 		usleep_range(1000, 2000);
2754 		barrier();
2755 	}
2756 
2757 	/* FW finished processing, wait for HW to transmit all tx packets */
2758 	usleep_range(1000, 2000);
2759 
2760 	return 0;
2761 }
2762 
2763 static int qede_stop_queues(struct qede_dev *edev)
2764 {
2765 	struct qed_update_vport_params vport_update_params;
2766 	struct qed_dev *cdev = edev->cdev;
2767 	int rc, tc, i;
2768 
2769 	/* Disable the vport */
2770 	memset(&vport_update_params, 0, sizeof(vport_update_params));
2771 	vport_update_params.vport_id = 0;
2772 	vport_update_params.update_vport_active_flg = 1;
2773 	vport_update_params.vport_active_flg = 0;
2774 	vport_update_params.update_rss_flg = 0;
2775 
2776 	rc = edev->ops->vport_update(cdev, &vport_update_params);
2777 	if (rc) {
2778 		DP_ERR(edev, "Failed to update vport\n");
2779 		return rc;
2780 	}
2781 
2782 	/* Flush Tx queues. If needed, request drain from MCP */
2783 	for_each_rss(i) {
2784 		struct qede_fastpath *fp = &edev->fp_array[i];
2785 
2786 		for (tc = 0; tc < edev->num_tc; tc++) {
2787 			struct qede_tx_queue *txq = &fp->txqs[tc];
2788 
2789 			rc = qede_drain_txq(edev, txq, true);
2790 			if (rc)
2791 				return rc;
2792 		}
2793 	}
2794 
2795 	/* Stop all Queues in reverse order*/
2796 	for (i = QEDE_RSS_CNT(edev) - 1; i >= 0; i--) {
2797 		struct qed_stop_rxq_params rx_params;
2798 
2799 		/* Stop the Tx Queue(s)*/
2800 		for (tc = 0; tc < edev->num_tc; tc++) {
2801 			struct qed_stop_txq_params tx_params;
2802 
2803 			tx_params.rss_id = i;
2804 			tx_params.tx_queue_id = tc * QEDE_RSS_CNT(edev) + i;
2805 			rc = edev->ops->q_tx_stop(cdev, &tx_params);
2806 			if (rc) {
2807 				DP_ERR(edev, "Failed to stop TXQ #%d\n",
2808 				       tx_params.tx_queue_id);
2809 				return rc;
2810 			}
2811 		}
2812 
2813 		/* Stop the Rx Queue*/
2814 		memset(&rx_params, 0, sizeof(rx_params));
2815 		rx_params.rss_id = i;
2816 		rx_params.rx_queue_id = i;
2817 
2818 		rc = edev->ops->q_rx_stop(cdev, &rx_params);
2819 		if (rc) {
2820 			DP_ERR(edev, "Failed to stop RXQ #%d\n", i);
2821 			return rc;
2822 		}
2823 	}
2824 
2825 	/* Stop the vport */
2826 	rc = edev->ops->vport_stop(cdev, 0);
2827 	if (rc)
2828 		DP_ERR(edev, "Failed to stop VPORT\n");
2829 
2830 	return rc;
2831 }
2832 
2833 static int qede_start_queues(struct qede_dev *edev)
2834 {
2835 	int rc, tc, i;
2836 	int vlan_removal_en = 1;
2837 	struct qed_dev *cdev = edev->cdev;
2838 	struct qed_update_vport_rss_params *rss_params = &edev->rss_params;
2839 	struct qed_update_vport_params vport_update_params;
2840 	struct qed_queue_start_common_params q_params;
2841 	struct qed_start_vport_params start = {0};
2842 
2843 	if (!edev->num_rss) {
2844 		DP_ERR(edev,
2845 		       "Cannot update V-VPORT as active as there are no Rx queues\n");
2846 		return -EINVAL;
2847 	}
2848 
2849 	start.gro_enable = !edev->gro_disable;
2850 	start.mtu = edev->ndev->mtu;
2851 	start.vport_id = 0;
2852 	start.drop_ttl0 = true;
2853 	start.remove_inner_vlan = vlan_removal_en;
2854 
2855 	rc = edev->ops->vport_start(cdev, &start);
2856 
2857 	if (rc) {
2858 		DP_ERR(edev, "Start V-PORT failed %d\n", rc);
2859 		return rc;
2860 	}
2861 
2862 	DP_VERBOSE(edev, NETIF_MSG_IFUP,
2863 		   "Start vport ramrod passed, vport_id = %d, MTU = %d, vlan_removal_en = %d\n",
2864 		   start.vport_id, edev->ndev->mtu + 0xe, vlan_removal_en);
2865 
2866 	for_each_rss(i) {
2867 		struct qede_fastpath *fp = &edev->fp_array[i];
2868 		dma_addr_t phys_table = fp->rxq->rx_comp_ring.pbl.p_phys_table;
2869 
2870 		memset(&q_params, 0, sizeof(q_params));
2871 		q_params.rss_id = i;
2872 		q_params.queue_id = i;
2873 		q_params.vport_id = 0;
2874 		q_params.sb = fp->sb_info->igu_sb_id;
2875 		q_params.sb_idx = RX_PI;
2876 
2877 		rc = edev->ops->q_rx_start(cdev, &q_params,
2878 					   fp->rxq->rx_buf_size,
2879 					   fp->rxq->rx_bd_ring.p_phys_addr,
2880 					   phys_table,
2881 					   fp->rxq->rx_comp_ring.page_cnt,
2882 					   &fp->rxq->hw_rxq_prod_addr);
2883 		if (rc) {
2884 			DP_ERR(edev, "Start RXQ #%d failed %d\n", i, rc);
2885 			return rc;
2886 		}
2887 
2888 		fp->rxq->hw_cons_ptr = &fp->sb_info->sb_virt->pi_array[RX_PI];
2889 
2890 		qede_update_rx_prod(edev, fp->rxq);
2891 
2892 		for (tc = 0; tc < edev->num_tc; tc++) {
2893 			struct qede_tx_queue *txq = &fp->txqs[tc];
2894 			int txq_index = tc * QEDE_RSS_CNT(edev) + i;
2895 
2896 			memset(&q_params, 0, sizeof(q_params));
2897 			q_params.rss_id = i;
2898 			q_params.queue_id = txq_index;
2899 			q_params.vport_id = 0;
2900 			q_params.sb = fp->sb_info->igu_sb_id;
2901 			q_params.sb_idx = TX_PI(tc);
2902 
2903 			rc = edev->ops->q_tx_start(cdev, &q_params,
2904 						   txq->tx_pbl.pbl.p_phys_table,
2905 						   txq->tx_pbl.page_cnt,
2906 						   &txq->doorbell_addr);
2907 			if (rc) {
2908 				DP_ERR(edev, "Start TXQ #%d failed %d\n",
2909 				       txq_index, rc);
2910 				return rc;
2911 			}
2912 
2913 			txq->hw_cons_ptr =
2914 				&fp->sb_info->sb_virt->pi_array[TX_PI(tc)];
2915 			SET_FIELD(txq->tx_db.data.params,
2916 				  ETH_DB_DATA_DEST, DB_DEST_XCM);
2917 			SET_FIELD(txq->tx_db.data.params, ETH_DB_DATA_AGG_CMD,
2918 				  DB_AGG_CMD_SET);
2919 			SET_FIELD(txq->tx_db.data.params,
2920 				  ETH_DB_DATA_AGG_VAL_SEL,
2921 				  DQ_XCM_ETH_TX_BD_PROD_CMD);
2922 
2923 			txq->tx_db.data.agg_flags = DQ_XCM_ETH_DQ_CF_CMD;
2924 		}
2925 	}
2926 
2927 	/* Prepare and send the vport enable */
2928 	memset(&vport_update_params, 0, sizeof(vport_update_params));
2929 	vport_update_params.vport_id = start.vport_id;
2930 	vport_update_params.update_vport_active_flg = 1;
2931 	vport_update_params.vport_active_flg = 1;
2932 
2933 	/* Fill struct with RSS params */
2934 	if (QEDE_RSS_CNT(edev) > 1) {
2935 		vport_update_params.update_rss_flg = 1;
2936 		for (i = 0; i < 128; i++)
2937 			rss_params->rss_ind_table[i] =
2938 			ethtool_rxfh_indir_default(i, QEDE_RSS_CNT(edev));
2939 		netdev_rss_key_fill(rss_params->rss_key,
2940 				    sizeof(rss_params->rss_key));
2941 	} else {
2942 		memset(rss_params, 0, sizeof(*rss_params));
2943 	}
2944 	memcpy(&vport_update_params.rss_params, rss_params,
2945 	       sizeof(*rss_params));
2946 
2947 	rc = edev->ops->vport_update(cdev, &vport_update_params);
2948 	if (rc) {
2949 		DP_ERR(edev, "Update V-PORT failed %d\n", rc);
2950 		return rc;
2951 	}
2952 
2953 	return 0;
2954 }
2955 
2956 static int qede_set_mcast_rx_mac(struct qede_dev *edev,
2957 				 enum qed_filter_xcast_params_type opcode,
2958 				 unsigned char *mac, int num_macs)
2959 {
2960 	struct qed_filter_params filter_cmd;
2961 	int i;
2962 
2963 	memset(&filter_cmd, 0, sizeof(filter_cmd));
2964 	filter_cmd.type = QED_FILTER_TYPE_MCAST;
2965 	filter_cmd.filter.mcast.type = opcode;
2966 	filter_cmd.filter.mcast.num = num_macs;
2967 
2968 	for (i = 0; i < num_macs; i++, mac += ETH_ALEN)
2969 		ether_addr_copy(filter_cmd.filter.mcast.mac[i], mac);
2970 
2971 	return edev->ops->filter_config(edev->cdev, &filter_cmd);
2972 }
2973 
2974 enum qede_unload_mode {
2975 	QEDE_UNLOAD_NORMAL,
2976 };
2977 
2978 static void qede_unload(struct qede_dev *edev, enum qede_unload_mode mode)
2979 {
2980 	struct qed_link_params link_params;
2981 	int rc;
2982 
2983 	DP_INFO(edev, "Starting qede unload\n");
2984 
2985 	mutex_lock(&edev->qede_lock);
2986 	edev->state = QEDE_STATE_CLOSED;
2987 
2988 	/* Close OS Tx */
2989 	netif_tx_disable(edev->ndev);
2990 	netif_carrier_off(edev->ndev);
2991 
2992 	/* Reset the link */
2993 	memset(&link_params, 0, sizeof(link_params));
2994 	link_params.link_up = false;
2995 	edev->ops->common->set_link(edev->cdev, &link_params);
2996 	rc = qede_stop_queues(edev);
2997 	if (rc) {
2998 		qede_sync_free_irqs(edev);
2999 		goto out;
3000 	}
3001 
3002 	DP_INFO(edev, "Stopped Queues\n");
3003 
3004 	qede_vlan_mark_nonconfigured(edev);
3005 	edev->ops->fastpath_stop(edev->cdev);
3006 
3007 	/* Release the interrupts */
3008 	qede_sync_free_irqs(edev);
3009 	edev->ops->common->set_fp_int(edev->cdev, 0);
3010 
3011 	qede_napi_disable_remove(edev);
3012 
3013 	qede_free_mem_load(edev);
3014 	qede_free_fp_array(edev);
3015 
3016 out:
3017 	mutex_unlock(&edev->qede_lock);
3018 	DP_INFO(edev, "Ending qede unload\n");
3019 }
3020 
3021 enum qede_load_mode {
3022 	QEDE_LOAD_NORMAL,
3023 };
3024 
3025 static int qede_load(struct qede_dev *edev, enum qede_load_mode mode)
3026 {
3027 	struct qed_link_params link_params;
3028 	struct qed_link_output link_output;
3029 	int rc;
3030 
3031 	DP_INFO(edev, "Starting qede load\n");
3032 
3033 	rc = qede_set_num_queues(edev);
3034 	if (rc)
3035 		goto err0;
3036 
3037 	rc = qede_alloc_fp_array(edev);
3038 	if (rc)
3039 		goto err0;
3040 
3041 	qede_init_fp(edev);
3042 
3043 	rc = qede_alloc_mem_load(edev);
3044 	if (rc)
3045 		goto err1;
3046 	DP_INFO(edev, "Allocated %d RSS queues on %d TC/s\n",
3047 		QEDE_RSS_CNT(edev), edev->num_tc);
3048 
3049 	rc = qede_set_real_num_queues(edev);
3050 	if (rc)
3051 		goto err2;
3052 
3053 	qede_napi_add_enable(edev);
3054 	DP_INFO(edev, "Napi added and enabled\n");
3055 
3056 	rc = qede_setup_irqs(edev);
3057 	if (rc)
3058 		goto err3;
3059 	DP_INFO(edev, "Setup IRQs succeeded\n");
3060 
3061 	rc = qede_start_queues(edev);
3062 	if (rc)
3063 		goto err4;
3064 	DP_INFO(edev, "Start VPORT, RXQ and TXQ succeeded\n");
3065 
3066 	/* Add primary mac and set Rx filters */
3067 	ether_addr_copy(edev->primary_mac, edev->ndev->dev_addr);
3068 
3069 	mutex_lock(&edev->qede_lock);
3070 	edev->state = QEDE_STATE_OPEN;
3071 	mutex_unlock(&edev->qede_lock);
3072 
3073 	/* Program un-configured VLANs */
3074 	qede_configure_vlan_filters(edev);
3075 
3076 	/* Ask for link-up using current configuration */
3077 	memset(&link_params, 0, sizeof(link_params));
3078 	link_params.link_up = true;
3079 	edev->ops->common->set_link(edev->cdev, &link_params);
3080 
3081 	/* Query whether link is already-up */
3082 	memset(&link_output, 0, sizeof(link_output));
3083 	edev->ops->common->get_link(edev->cdev, &link_output);
3084 	qede_link_update(edev, &link_output);
3085 
3086 	DP_INFO(edev, "Ending successfully qede load\n");
3087 
3088 	return 0;
3089 
3090 err4:
3091 	qede_sync_free_irqs(edev);
3092 	memset(&edev->int_info.msix_cnt, 0, sizeof(struct qed_int_info));
3093 err3:
3094 	qede_napi_disable_remove(edev);
3095 err2:
3096 	qede_free_mem_load(edev);
3097 err1:
3098 	edev->ops->common->set_fp_int(edev->cdev, 0);
3099 	qede_free_fp_array(edev);
3100 	edev->num_rss = 0;
3101 err0:
3102 	return rc;
3103 }
3104 
3105 void qede_reload(struct qede_dev *edev,
3106 		 void (*func)(struct qede_dev *, union qede_reload_args *),
3107 		 union qede_reload_args *args)
3108 {
3109 	qede_unload(edev, QEDE_UNLOAD_NORMAL);
3110 	/* Call function handler to update parameters
3111 	 * needed for function load.
3112 	 */
3113 	if (func)
3114 		func(edev, args);
3115 
3116 	qede_load(edev, QEDE_LOAD_NORMAL);
3117 
3118 	mutex_lock(&edev->qede_lock);
3119 	qede_config_rx_mode(edev->ndev);
3120 	mutex_unlock(&edev->qede_lock);
3121 }
3122 
3123 /* called with rtnl_lock */
3124 static int qede_open(struct net_device *ndev)
3125 {
3126 	struct qede_dev *edev = netdev_priv(ndev);
3127 
3128 	netif_carrier_off(ndev);
3129 
3130 	edev->ops->common->set_power_state(edev->cdev, PCI_D0);
3131 
3132 	return qede_load(edev, QEDE_LOAD_NORMAL);
3133 }
3134 
3135 static int qede_close(struct net_device *ndev)
3136 {
3137 	struct qede_dev *edev = netdev_priv(ndev);
3138 
3139 	qede_unload(edev, QEDE_UNLOAD_NORMAL);
3140 
3141 	return 0;
3142 }
3143 
3144 static void qede_link_update(void *dev, struct qed_link_output *link)
3145 {
3146 	struct qede_dev *edev = dev;
3147 
3148 	if (!netif_running(edev->ndev)) {
3149 		DP_VERBOSE(edev, NETIF_MSG_LINK, "Interface is not running\n");
3150 		return;
3151 	}
3152 
3153 	if (link->link_up) {
3154 		if (!netif_carrier_ok(edev->ndev)) {
3155 			DP_NOTICE(edev, "Link is up\n");
3156 			netif_tx_start_all_queues(edev->ndev);
3157 			netif_carrier_on(edev->ndev);
3158 		}
3159 	} else {
3160 		if (netif_carrier_ok(edev->ndev)) {
3161 			DP_NOTICE(edev, "Link is down\n");
3162 			netif_tx_disable(edev->ndev);
3163 			netif_carrier_off(edev->ndev);
3164 		}
3165 	}
3166 }
3167 
3168 static int qede_set_mac_addr(struct net_device *ndev, void *p)
3169 {
3170 	struct qede_dev *edev = netdev_priv(ndev);
3171 	struct sockaddr *addr = p;
3172 	int rc;
3173 
3174 	ASSERT_RTNL(); /* @@@TBD To be removed */
3175 
3176 	DP_INFO(edev, "Set_mac_addr called\n");
3177 
3178 	if (!is_valid_ether_addr(addr->sa_data)) {
3179 		DP_NOTICE(edev, "The MAC address is not valid\n");
3180 		return -EFAULT;
3181 	}
3182 
3183 	ether_addr_copy(ndev->dev_addr, addr->sa_data);
3184 
3185 	if (!netif_running(ndev))  {
3186 		DP_NOTICE(edev, "The device is currently down\n");
3187 		return 0;
3188 	}
3189 
3190 	/* Remove the previous primary mac */
3191 	rc = qede_set_ucast_rx_mac(edev, QED_FILTER_XCAST_TYPE_DEL,
3192 				   edev->primary_mac);
3193 	if (rc)
3194 		return rc;
3195 
3196 	/* Add MAC filter according to the new unicast HW MAC address */
3197 	ether_addr_copy(edev->primary_mac, ndev->dev_addr);
3198 	return qede_set_ucast_rx_mac(edev, QED_FILTER_XCAST_TYPE_ADD,
3199 				      edev->primary_mac);
3200 }
3201 
3202 static int
3203 qede_configure_mcast_filtering(struct net_device *ndev,
3204 			       enum qed_filter_rx_mode_type *accept_flags)
3205 {
3206 	struct qede_dev *edev = netdev_priv(ndev);
3207 	unsigned char *mc_macs, *temp;
3208 	struct netdev_hw_addr *ha;
3209 	int rc = 0, mc_count;
3210 	size_t size;
3211 
3212 	size = 64 * ETH_ALEN;
3213 
3214 	mc_macs = kzalloc(size, GFP_KERNEL);
3215 	if (!mc_macs) {
3216 		DP_NOTICE(edev,
3217 			  "Failed to allocate memory for multicast MACs\n");
3218 		rc = -ENOMEM;
3219 		goto exit;
3220 	}
3221 
3222 	temp = mc_macs;
3223 
3224 	/* Remove all previously configured MAC filters */
3225 	rc = qede_set_mcast_rx_mac(edev, QED_FILTER_XCAST_TYPE_DEL,
3226 				   mc_macs, 1);
3227 	if (rc)
3228 		goto exit;
3229 
3230 	netif_addr_lock_bh(ndev);
3231 
3232 	mc_count = netdev_mc_count(ndev);
3233 	if (mc_count < 64) {
3234 		netdev_for_each_mc_addr(ha, ndev) {
3235 			ether_addr_copy(temp, ha->addr);
3236 			temp += ETH_ALEN;
3237 		}
3238 	}
3239 
3240 	netif_addr_unlock_bh(ndev);
3241 
3242 	/* Check for all multicast @@@TBD resource allocation */
3243 	if ((ndev->flags & IFF_ALLMULTI) ||
3244 	    (mc_count > 64)) {
3245 		if (*accept_flags == QED_FILTER_RX_MODE_TYPE_REGULAR)
3246 			*accept_flags = QED_FILTER_RX_MODE_TYPE_MULTI_PROMISC;
3247 	} else {
3248 		/* Add all multicast MAC filters */
3249 		rc = qede_set_mcast_rx_mac(edev, QED_FILTER_XCAST_TYPE_ADD,
3250 					   mc_macs, mc_count);
3251 	}
3252 
3253 exit:
3254 	kfree(mc_macs);
3255 	return rc;
3256 }
3257 
3258 static void qede_set_rx_mode(struct net_device *ndev)
3259 {
3260 	struct qede_dev *edev = netdev_priv(ndev);
3261 
3262 	DP_INFO(edev, "qede_set_rx_mode called\n");
3263 
3264 	if (edev->state != QEDE_STATE_OPEN) {
3265 		DP_INFO(edev,
3266 			"qede_set_rx_mode called while interface is down\n");
3267 	} else {
3268 		set_bit(QEDE_SP_RX_MODE, &edev->sp_flags);
3269 		schedule_delayed_work(&edev->sp_task, 0);
3270 	}
3271 }
3272 
3273 /* Must be called with qede_lock held */
3274 static void qede_config_rx_mode(struct net_device *ndev)
3275 {
3276 	enum qed_filter_rx_mode_type accept_flags = QED_FILTER_TYPE_UCAST;
3277 	struct qede_dev *edev = netdev_priv(ndev);
3278 	struct qed_filter_params rx_mode;
3279 	unsigned char *uc_macs, *temp;
3280 	struct netdev_hw_addr *ha;
3281 	int rc, uc_count;
3282 	size_t size;
3283 
3284 	netif_addr_lock_bh(ndev);
3285 
3286 	uc_count = netdev_uc_count(ndev);
3287 	size = uc_count * ETH_ALEN;
3288 
3289 	uc_macs = kzalloc(size, GFP_ATOMIC);
3290 	if (!uc_macs) {
3291 		DP_NOTICE(edev, "Failed to allocate memory for unicast MACs\n");
3292 		netif_addr_unlock_bh(ndev);
3293 		return;
3294 	}
3295 
3296 	temp = uc_macs;
3297 	netdev_for_each_uc_addr(ha, ndev) {
3298 		ether_addr_copy(temp, ha->addr);
3299 		temp += ETH_ALEN;
3300 	}
3301 
3302 	netif_addr_unlock_bh(ndev);
3303 
3304 	/* Configure the struct for the Rx mode */
3305 	memset(&rx_mode, 0, sizeof(struct qed_filter_params));
3306 	rx_mode.type = QED_FILTER_TYPE_RX_MODE;
3307 
3308 	/* Remove all previous unicast secondary macs and multicast macs
3309 	 * (configrue / leave the primary mac)
3310 	 */
3311 	rc = qede_set_ucast_rx_mac(edev, QED_FILTER_XCAST_TYPE_REPLACE,
3312 				   edev->primary_mac);
3313 	if (rc)
3314 		goto out;
3315 
3316 	/* Check for promiscuous */
3317 	if ((ndev->flags & IFF_PROMISC) ||
3318 	    (uc_count > 15)) { /* @@@TBD resource allocation - 1 */
3319 		accept_flags = QED_FILTER_RX_MODE_TYPE_PROMISC;
3320 	} else {
3321 		/* Add MAC filters according to the unicast secondary macs */
3322 		int i;
3323 
3324 		temp = uc_macs;
3325 		for (i = 0; i < uc_count; i++) {
3326 			rc = qede_set_ucast_rx_mac(edev,
3327 						   QED_FILTER_XCAST_TYPE_ADD,
3328 						   temp);
3329 			if (rc)
3330 				goto out;
3331 
3332 			temp += ETH_ALEN;
3333 		}
3334 
3335 		rc = qede_configure_mcast_filtering(ndev, &accept_flags);
3336 		if (rc)
3337 			goto out;
3338 	}
3339 
3340 	/* take care of VLAN mode */
3341 	if (ndev->flags & IFF_PROMISC) {
3342 		qede_config_accept_any_vlan(edev, true);
3343 	} else if (!edev->non_configured_vlans) {
3344 		/* It's possible that accept_any_vlan mode is set due to a
3345 		 * previous setting of IFF_PROMISC. If vlan credits are
3346 		 * sufficient, disable accept_any_vlan.
3347 		 */
3348 		qede_config_accept_any_vlan(edev, false);
3349 	}
3350 
3351 	rx_mode.filter.accept_flags = accept_flags;
3352 	edev->ops->filter_config(edev->cdev, &rx_mode);
3353 out:
3354 	kfree(uc_macs);
3355 }
3356