xref: /openbmc/linux/drivers/net/ethernet/qlogic/qede/qede_main.c (revision e983940270f10fe8551baf0098be76ea478294a3)
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/udp_tunnel.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 #include <linux/qed/qede_roce.h>
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 #define CHIP_NUM_57980S_IOV		0x1664
62 
63 #ifndef PCI_DEVICE_ID_NX2_57980E
64 #define PCI_DEVICE_ID_57980S_40		CHIP_NUM_57980S_40
65 #define PCI_DEVICE_ID_57980S_10		CHIP_NUM_57980S_10
66 #define PCI_DEVICE_ID_57980S_MF		CHIP_NUM_57980S_MF
67 #define PCI_DEVICE_ID_57980S_100	CHIP_NUM_57980S_100
68 #define PCI_DEVICE_ID_57980S_50		CHIP_NUM_57980S_50
69 #define PCI_DEVICE_ID_57980S_25		CHIP_NUM_57980S_25
70 #define PCI_DEVICE_ID_57980S_IOV	CHIP_NUM_57980S_IOV
71 #endif
72 
73 enum qede_pci_private {
74 	QEDE_PRIVATE_PF,
75 	QEDE_PRIVATE_VF
76 };
77 
78 static const struct pci_device_id qede_pci_tbl[] = {
79 	{PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_40), QEDE_PRIVATE_PF},
80 	{PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_10), QEDE_PRIVATE_PF},
81 	{PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_MF), QEDE_PRIVATE_PF},
82 	{PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_100), QEDE_PRIVATE_PF},
83 	{PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_50), QEDE_PRIVATE_PF},
84 	{PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_25), QEDE_PRIVATE_PF},
85 #ifdef CONFIG_QED_SRIOV
86 	{PCI_VDEVICE(QLOGIC, PCI_DEVICE_ID_57980S_IOV), QEDE_PRIVATE_VF},
87 #endif
88 	{ 0 }
89 };
90 
91 MODULE_DEVICE_TABLE(pci, qede_pci_tbl);
92 
93 static int qede_probe(struct pci_dev *pdev, const struct pci_device_id *id);
94 
95 #define TX_TIMEOUT		(5 * HZ)
96 
97 static void qede_remove(struct pci_dev *pdev);
98 static int qede_alloc_rx_buffer(struct qede_dev *edev,
99 				struct qede_rx_queue *rxq);
100 static void qede_link_update(void *dev, struct qed_link_output *link);
101 
102 #ifdef CONFIG_QED_SRIOV
103 static int qede_set_vf_vlan(struct net_device *ndev, int vf, u16 vlan, u8 qos,
104 			    __be16 vlan_proto)
105 {
106 	struct qede_dev *edev = netdev_priv(ndev);
107 
108 	if (vlan > 4095) {
109 		DP_NOTICE(edev, "Illegal vlan value %d\n", vlan);
110 		return -EINVAL;
111 	}
112 
113 	if (vlan_proto != htons(ETH_P_8021Q))
114 		return -EPROTONOSUPPORT;
115 
116 	DP_VERBOSE(edev, QED_MSG_IOV, "Setting Vlan 0x%04x to VF [%d]\n",
117 		   vlan, vf);
118 
119 	return edev->ops->iov->set_vlan(edev->cdev, vlan, vf);
120 }
121 
122 static int qede_set_vf_mac(struct net_device *ndev, int vfidx, u8 *mac)
123 {
124 	struct qede_dev *edev = netdev_priv(ndev);
125 
126 	DP_VERBOSE(edev, QED_MSG_IOV,
127 		   "Setting MAC %02x:%02x:%02x:%02x:%02x:%02x to VF [%d]\n",
128 		   mac[0], mac[1], mac[2], mac[3], mac[4], mac[5], vfidx);
129 
130 	if (!is_valid_ether_addr(mac)) {
131 		DP_VERBOSE(edev, QED_MSG_IOV, "MAC address isn't valid\n");
132 		return -EINVAL;
133 	}
134 
135 	return edev->ops->iov->set_mac(edev->cdev, mac, vfidx);
136 }
137 
138 static int qede_sriov_configure(struct pci_dev *pdev, int num_vfs_param)
139 {
140 	struct qede_dev *edev = netdev_priv(pci_get_drvdata(pdev));
141 	struct qed_dev_info *qed_info = &edev->dev_info.common;
142 	int rc;
143 
144 	DP_VERBOSE(edev, QED_MSG_IOV, "Requested %d VFs\n", num_vfs_param);
145 
146 	rc = edev->ops->iov->configure(edev->cdev, num_vfs_param);
147 
148 	/* Enable/Disable Tx switching for PF */
149 	if ((rc == num_vfs_param) && netif_running(edev->ndev) &&
150 	    qed_info->mf_mode != QED_MF_NPAR && qed_info->tx_switching) {
151 		struct qed_update_vport_params params;
152 
153 		memset(&params, 0, sizeof(params));
154 		params.vport_id = 0;
155 		params.update_tx_switching_flg = 1;
156 		params.tx_switching_flg = num_vfs_param ? 1 : 0;
157 		edev->ops->vport_update(edev->cdev, &params);
158 	}
159 
160 	return rc;
161 }
162 #endif
163 
164 static struct pci_driver qede_pci_driver = {
165 	.name = "qede",
166 	.id_table = qede_pci_tbl,
167 	.probe = qede_probe,
168 	.remove = qede_remove,
169 #ifdef CONFIG_QED_SRIOV
170 	.sriov_configure = qede_sriov_configure,
171 #endif
172 };
173 
174 static void qede_force_mac(void *dev, u8 *mac)
175 {
176 	struct qede_dev *edev = dev;
177 
178 	ether_addr_copy(edev->ndev->dev_addr, mac);
179 	ether_addr_copy(edev->primary_mac, mac);
180 }
181 
182 static struct qed_eth_cb_ops qede_ll_ops = {
183 	{
184 		.link_update = qede_link_update,
185 	},
186 	.force_mac = qede_force_mac,
187 };
188 
189 static int qede_netdev_event(struct notifier_block *this, unsigned long event,
190 			     void *ptr)
191 {
192 	struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
193 	struct ethtool_drvinfo drvinfo;
194 	struct qede_dev *edev;
195 
196 	if (event != NETDEV_CHANGENAME && event != NETDEV_CHANGEADDR)
197 		goto done;
198 
199 	/* Check whether this is a qede device */
200 	if (!ndev || !ndev->ethtool_ops || !ndev->ethtool_ops->get_drvinfo)
201 		goto done;
202 
203 	memset(&drvinfo, 0, sizeof(drvinfo));
204 	ndev->ethtool_ops->get_drvinfo(ndev, &drvinfo);
205 	if (strcmp(drvinfo.driver, "qede"))
206 		goto done;
207 	edev = netdev_priv(ndev);
208 
209 	switch (event) {
210 	case NETDEV_CHANGENAME:
211 		/* Notify qed of the name change */
212 		if (!edev->ops || !edev->ops->common)
213 			goto done;
214 		edev->ops->common->set_id(edev->cdev, edev->ndev->name, "qede");
215 		break;
216 	case NETDEV_CHANGEADDR:
217 		edev = netdev_priv(ndev);
218 		qede_roce_event_changeaddr(edev);
219 		break;
220 	}
221 
222 done:
223 	return NOTIFY_DONE;
224 }
225 
226 static struct notifier_block qede_netdev_notifier = {
227 	.notifier_call = qede_netdev_event,
228 };
229 
230 static
231 int __init qede_init(void)
232 {
233 	int ret;
234 
235 	pr_info("qede_init: %s\n", version);
236 
237 	qed_ops = qed_get_eth_ops();
238 	if (!qed_ops) {
239 		pr_notice("Failed to get qed ethtool operations\n");
240 		return -EINVAL;
241 	}
242 
243 	/* Must register notifier before pci ops, since we might miss
244 	 * interface rename after pci probe and netdev registeration.
245 	 */
246 	ret = register_netdevice_notifier(&qede_netdev_notifier);
247 	if (ret) {
248 		pr_notice("Failed to register netdevice_notifier\n");
249 		qed_put_eth_ops();
250 		return -EINVAL;
251 	}
252 
253 	ret = pci_register_driver(&qede_pci_driver);
254 	if (ret) {
255 		pr_notice("Failed to register driver\n");
256 		unregister_netdevice_notifier(&qede_netdev_notifier);
257 		qed_put_eth_ops();
258 		return -EINVAL;
259 	}
260 
261 	return 0;
262 }
263 
264 static void __exit qede_cleanup(void)
265 {
266 	if (debug & QED_LOG_INFO_MASK)
267 		pr_info("qede_cleanup called\n");
268 
269 	unregister_netdevice_notifier(&qede_netdev_notifier);
270 	pci_unregister_driver(&qede_pci_driver);
271 	qed_put_eth_ops();
272 }
273 
274 module_init(qede_init);
275 module_exit(qede_cleanup);
276 
277 /* -------------------------------------------------------------------------
278  * START OF FAST-PATH
279  * -------------------------------------------------------------------------
280  */
281 
282 /* Unmap the data and free skb */
283 static int qede_free_tx_pkt(struct qede_dev *edev,
284 			    struct qede_tx_queue *txq, int *len)
285 {
286 	u16 idx = txq->sw_tx_cons & NUM_TX_BDS_MAX;
287 	struct sk_buff *skb = txq->sw_tx_ring[idx].skb;
288 	struct eth_tx_1st_bd *first_bd;
289 	struct eth_tx_bd *tx_data_bd;
290 	int bds_consumed = 0;
291 	int nbds;
292 	bool data_split = txq->sw_tx_ring[idx].flags & QEDE_TSO_SPLIT_BD;
293 	int i, split_bd_len = 0;
294 
295 	if (unlikely(!skb)) {
296 		DP_ERR(edev,
297 		       "skb is null for txq idx=%d txq->sw_tx_cons=%d txq->sw_tx_prod=%d\n",
298 		       idx, txq->sw_tx_cons, txq->sw_tx_prod);
299 		return -1;
300 	}
301 
302 	*len = skb->len;
303 
304 	first_bd = (struct eth_tx_1st_bd *)qed_chain_consume(&txq->tx_pbl);
305 
306 	bds_consumed++;
307 
308 	nbds = first_bd->data.nbds;
309 
310 	if (data_split) {
311 		struct eth_tx_bd *split = (struct eth_tx_bd *)
312 			qed_chain_consume(&txq->tx_pbl);
313 		split_bd_len = BD_UNMAP_LEN(split);
314 		bds_consumed++;
315 	}
316 	dma_unmap_page(&edev->pdev->dev, BD_UNMAP_ADDR(first_bd),
317 		       BD_UNMAP_LEN(first_bd) + split_bd_len, DMA_TO_DEVICE);
318 
319 	/* Unmap the data of the skb frags */
320 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++, bds_consumed++) {
321 		tx_data_bd = (struct eth_tx_bd *)
322 			qed_chain_consume(&txq->tx_pbl);
323 		dma_unmap_page(&edev->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
324 			       BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
325 	}
326 
327 	while (bds_consumed++ < nbds)
328 		qed_chain_consume(&txq->tx_pbl);
329 
330 	/* Free skb */
331 	dev_kfree_skb_any(skb);
332 	txq->sw_tx_ring[idx].skb = NULL;
333 	txq->sw_tx_ring[idx].flags = 0;
334 
335 	return 0;
336 }
337 
338 /* Unmap the data and free skb when mapping failed during start_xmit */
339 static void qede_free_failed_tx_pkt(struct qede_dev *edev,
340 				    struct qede_tx_queue *txq,
341 				    struct eth_tx_1st_bd *first_bd,
342 				    int nbd, bool data_split)
343 {
344 	u16 idx = txq->sw_tx_prod & NUM_TX_BDS_MAX;
345 	struct sk_buff *skb = txq->sw_tx_ring[idx].skb;
346 	struct eth_tx_bd *tx_data_bd;
347 	int i, split_bd_len = 0;
348 
349 	/* Return prod to its position before this skb was handled */
350 	qed_chain_set_prod(&txq->tx_pbl,
351 			   le16_to_cpu(txq->tx_db.data.bd_prod), first_bd);
352 
353 	first_bd = (struct eth_tx_1st_bd *)qed_chain_produce(&txq->tx_pbl);
354 
355 	if (data_split) {
356 		struct eth_tx_bd *split = (struct eth_tx_bd *)
357 					  qed_chain_produce(&txq->tx_pbl);
358 		split_bd_len = BD_UNMAP_LEN(split);
359 		nbd--;
360 	}
361 
362 	dma_unmap_page(&edev->pdev->dev, BD_UNMAP_ADDR(first_bd),
363 		       BD_UNMAP_LEN(first_bd) + split_bd_len, DMA_TO_DEVICE);
364 
365 	/* Unmap the data of the skb frags */
366 	for (i = 0; i < nbd; i++) {
367 		tx_data_bd = (struct eth_tx_bd *)
368 			qed_chain_produce(&txq->tx_pbl);
369 		if (tx_data_bd->nbytes)
370 			dma_unmap_page(&edev->pdev->dev,
371 				       BD_UNMAP_ADDR(tx_data_bd),
372 				       BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
373 	}
374 
375 	/* Return again prod to its position before this skb was handled */
376 	qed_chain_set_prod(&txq->tx_pbl,
377 			   le16_to_cpu(txq->tx_db.data.bd_prod), first_bd);
378 
379 	/* Free skb */
380 	dev_kfree_skb_any(skb);
381 	txq->sw_tx_ring[idx].skb = NULL;
382 	txq->sw_tx_ring[idx].flags = 0;
383 }
384 
385 static u32 qede_xmit_type(struct qede_dev *edev,
386 			  struct sk_buff *skb, int *ipv6_ext)
387 {
388 	u32 rc = XMIT_L4_CSUM;
389 	__be16 l3_proto;
390 
391 	if (skb->ip_summed != CHECKSUM_PARTIAL)
392 		return XMIT_PLAIN;
393 
394 	l3_proto = vlan_get_protocol(skb);
395 	if (l3_proto == htons(ETH_P_IPV6) &&
396 	    (ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
397 		*ipv6_ext = 1;
398 
399 	if (skb->encapsulation)
400 		rc |= XMIT_ENC;
401 
402 	if (skb_is_gso(skb))
403 		rc |= XMIT_LSO;
404 
405 	return rc;
406 }
407 
408 static void qede_set_params_for_ipv6_ext(struct sk_buff *skb,
409 					 struct eth_tx_2nd_bd *second_bd,
410 					 struct eth_tx_3rd_bd *third_bd)
411 {
412 	u8 l4_proto;
413 	u16 bd2_bits1 = 0, bd2_bits2 = 0;
414 
415 	bd2_bits1 |= (1 << ETH_TX_DATA_2ND_BD_IPV6_EXT_SHIFT);
416 
417 	bd2_bits2 |= ((((u8 *)skb_transport_header(skb) - skb->data) >> 1) &
418 		     ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_MASK)
419 		    << ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_SHIFT;
420 
421 	bd2_bits1 |= (ETH_L4_PSEUDO_CSUM_CORRECT_LENGTH <<
422 		      ETH_TX_DATA_2ND_BD_L4_PSEUDO_CSUM_MODE_SHIFT);
423 
424 	if (vlan_get_protocol(skb) == htons(ETH_P_IPV6))
425 		l4_proto = ipv6_hdr(skb)->nexthdr;
426 	else
427 		l4_proto = ip_hdr(skb)->protocol;
428 
429 	if (l4_proto == IPPROTO_UDP)
430 		bd2_bits1 |= 1 << ETH_TX_DATA_2ND_BD_L4_UDP_SHIFT;
431 
432 	if (third_bd)
433 		third_bd->data.bitfields |=
434 			cpu_to_le16(((tcp_hdrlen(skb) / 4) &
435 				ETH_TX_DATA_3RD_BD_TCP_HDR_LEN_DW_MASK) <<
436 				ETH_TX_DATA_3RD_BD_TCP_HDR_LEN_DW_SHIFT);
437 
438 	second_bd->data.bitfields1 = cpu_to_le16(bd2_bits1);
439 	second_bd->data.bitfields2 = cpu_to_le16(bd2_bits2);
440 }
441 
442 static int map_frag_to_bd(struct qede_dev *edev,
443 			  skb_frag_t *frag, struct eth_tx_bd *bd)
444 {
445 	dma_addr_t mapping;
446 
447 	/* Map skb non-linear frag data for DMA */
448 	mapping = skb_frag_dma_map(&edev->pdev->dev, frag, 0,
449 				   skb_frag_size(frag), DMA_TO_DEVICE);
450 	if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
451 		DP_NOTICE(edev, "Unable to map frag - dropping packet\n");
452 		return -ENOMEM;
453 	}
454 
455 	/* Setup the data pointer of the frag data */
456 	BD_SET_UNMAP_ADDR_LEN(bd, mapping, skb_frag_size(frag));
457 
458 	return 0;
459 }
460 
461 static u16 qede_get_skb_hlen(struct sk_buff *skb, bool is_encap_pkt)
462 {
463 	if (is_encap_pkt)
464 		return (skb_inner_transport_header(skb) +
465 			inner_tcp_hdrlen(skb) - skb->data);
466 	else
467 		return (skb_transport_header(skb) +
468 			tcp_hdrlen(skb) - skb->data);
469 }
470 
471 /* +2 for 1st BD for headers and 2nd BD for headlen (if required) */
472 #if ((MAX_SKB_FRAGS + 2) > ETH_TX_MAX_BDS_PER_NON_LSO_PACKET)
473 static bool qede_pkt_req_lin(struct qede_dev *edev, struct sk_buff *skb,
474 			     u8 xmit_type)
475 {
476 	int allowed_frags = ETH_TX_MAX_BDS_PER_NON_LSO_PACKET - 1;
477 
478 	if (xmit_type & XMIT_LSO) {
479 		int hlen;
480 
481 		hlen = qede_get_skb_hlen(skb, xmit_type & XMIT_ENC);
482 
483 		/* linear payload would require its own BD */
484 		if (skb_headlen(skb) > hlen)
485 			allowed_frags--;
486 	}
487 
488 	return (skb_shinfo(skb)->nr_frags > allowed_frags);
489 }
490 #endif
491 
492 static inline void qede_update_tx_producer(struct qede_tx_queue *txq)
493 {
494 	/* wmb makes sure that the BDs data is updated before updating the
495 	 * producer, otherwise FW may read old data from the BDs.
496 	 */
497 	wmb();
498 	barrier();
499 	writel(txq->tx_db.raw, txq->doorbell_addr);
500 
501 	/* mmiowb is needed to synchronize doorbell writes from more than one
502 	 * processor. It guarantees that the write arrives to the device before
503 	 * the queue lock is released and another start_xmit is called (possibly
504 	 * on another CPU). Without this barrier, the next doorbell can bypass
505 	 * this doorbell. This is applicable to IA64/Altix systems.
506 	 */
507 	mmiowb();
508 }
509 
510 /* Main transmit function */
511 static netdev_tx_t qede_start_xmit(struct sk_buff *skb,
512 				   struct net_device *ndev)
513 {
514 	struct qede_dev *edev = netdev_priv(ndev);
515 	struct netdev_queue *netdev_txq;
516 	struct qede_tx_queue *txq;
517 	struct eth_tx_1st_bd *first_bd;
518 	struct eth_tx_2nd_bd *second_bd = NULL;
519 	struct eth_tx_3rd_bd *third_bd = NULL;
520 	struct eth_tx_bd *tx_data_bd = NULL;
521 	u16 txq_index;
522 	u8 nbd = 0;
523 	dma_addr_t mapping;
524 	int rc, frag_idx = 0, ipv6_ext = 0;
525 	u8 xmit_type;
526 	u16 idx;
527 	u16 hlen;
528 	bool data_split = false;
529 
530 	/* Get tx-queue context and netdev index */
531 	txq_index = skb_get_queue_mapping(skb);
532 	WARN_ON(txq_index >= QEDE_TSS_COUNT(edev));
533 	txq = QEDE_TX_QUEUE(edev, txq_index);
534 	netdev_txq = netdev_get_tx_queue(ndev, txq_index);
535 
536 	WARN_ON(qed_chain_get_elem_left(&txq->tx_pbl) < (MAX_SKB_FRAGS + 1));
537 
538 	xmit_type = qede_xmit_type(edev, skb, &ipv6_ext);
539 
540 #if ((MAX_SKB_FRAGS + 2) > ETH_TX_MAX_BDS_PER_NON_LSO_PACKET)
541 	if (qede_pkt_req_lin(edev, skb, xmit_type)) {
542 		if (skb_linearize(skb)) {
543 			DP_NOTICE(edev,
544 				  "SKB linearization failed - silently dropping this SKB\n");
545 			dev_kfree_skb_any(skb);
546 			return NETDEV_TX_OK;
547 		}
548 	}
549 #endif
550 
551 	/* Fill the entry in the SW ring and the BDs in the FW ring */
552 	idx = txq->sw_tx_prod & NUM_TX_BDS_MAX;
553 	txq->sw_tx_ring[idx].skb = skb;
554 	first_bd = (struct eth_tx_1st_bd *)
555 		   qed_chain_produce(&txq->tx_pbl);
556 	memset(first_bd, 0, sizeof(*first_bd));
557 	first_bd->data.bd_flags.bitfields =
558 		1 << ETH_TX_1ST_BD_FLAGS_START_BD_SHIFT;
559 
560 	/* Map skb linear data for DMA and set in the first BD */
561 	mapping = dma_map_single(&edev->pdev->dev, skb->data,
562 				 skb_headlen(skb), DMA_TO_DEVICE);
563 	if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
564 		DP_NOTICE(edev, "SKB mapping failed\n");
565 		qede_free_failed_tx_pkt(edev, txq, first_bd, 0, false);
566 		qede_update_tx_producer(txq);
567 		return NETDEV_TX_OK;
568 	}
569 	nbd++;
570 	BD_SET_UNMAP_ADDR_LEN(first_bd, mapping, skb_headlen(skb));
571 
572 	/* In case there is IPv6 with extension headers or LSO we need 2nd and
573 	 * 3rd BDs.
574 	 */
575 	if (unlikely((xmit_type & XMIT_LSO) | ipv6_ext)) {
576 		second_bd = (struct eth_tx_2nd_bd *)
577 			qed_chain_produce(&txq->tx_pbl);
578 		memset(second_bd, 0, sizeof(*second_bd));
579 
580 		nbd++;
581 		third_bd = (struct eth_tx_3rd_bd *)
582 			qed_chain_produce(&txq->tx_pbl);
583 		memset(third_bd, 0, sizeof(*third_bd));
584 
585 		nbd++;
586 		/* We need to fill in additional data in second_bd... */
587 		tx_data_bd = (struct eth_tx_bd *)second_bd;
588 	}
589 
590 	if (skb_vlan_tag_present(skb)) {
591 		first_bd->data.vlan = cpu_to_le16(skb_vlan_tag_get(skb));
592 		first_bd->data.bd_flags.bitfields |=
593 			1 << ETH_TX_1ST_BD_FLAGS_VLAN_INSERTION_SHIFT;
594 	}
595 
596 	/* Fill the parsing flags & params according to the requested offload */
597 	if (xmit_type & XMIT_L4_CSUM) {
598 		/* We don't re-calculate IP checksum as it is already done by
599 		 * the upper stack
600 		 */
601 		first_bd->data.bd_flags.bitfields |=
602 			1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
603 
604 		if (xmit_type & XMIT_ENC) {
605 			first_bd->data.bd_flags.bitfields |=
606 				1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
607 			first_bd->data.bitfields |=
608 			    1 << ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT;
609 		}
610 
611 		/* Legacy FW had flipped behavior in regard to this bit -
612 		 * I.e., needed to set to prevent FW from touching encapsulated
613 		 * packets when it didn't need to.
614 		 */
615 		if (unlikely(txq->is_legacy))
616 			first_bd->data.bitfields ^=
617 			    1 << ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT;
618 
619 		/* If the packet is IPv6 with extension header, indicate that
620 		 * to FW and pass few params, since the device cracker doesn't
621 		 * support parsing IPv6 with extension header/s.
622 		 */
623 		if (unlikely(ipv6_ext))
624 			qede_set_params_for_ipv6_ext(skb, second_bd, third_bd);
625 	}
626 
627 	if (xmit_type & XMIT_LSO) {
628 		first_bd->data.bd_flags.bitfields |=
629 			(1 << ETH_TX_1ST_BD_FLAGS_LSO_SHIFT);
630 		third_bd->data.lso_mss =
631 			cpu_to_le16(skb_shinfo(skb)->gso_size);
632 
633 		if (unlikely(xmit_type & XMIT_ENC)) {
634 			first_bd->data.bd_flags.bitfields |=
635 				1 << ETH_TX_1ST_BD_FLAGS_TUNN_IP_CSUM_SHIFT;
636 			hlen = qede_get_skb_hlen(skb, true);
637 		} else {
638 			first_bd->data.bd_flags.bitfields |=
639 				1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
640 			hlen = qede_get_skb_hlen(skb, false);
641 		}
642 
643 		/* @@@TBD - if will not be removed need to check */
644 		third_bd->data.bitfields |=
645 			cpu_to_le16((1 << ETH_TX_DATA_3RD_BD_HDR_NBD_SHIFT));
646 
647 		/* Make life easier for FW guys who can't deal with header and
648 		 * data on same BD. If we need to split, use the second bd...
649 		 */
650 		if (unlikely(skb_headlen(skb) > hlen)) {
651 			DP_VERBOSE(edev, NETIF_MSG_TX_QUEUED,
652 				   "TSO split header size is %d (%x:%x)\n",
653 				   first_bd->nbytes, first_bd->addr.hi,
654 				   first_bd->addr.lo);
655 
656 			mapping = HILO_U64(le32_to_cpu(first_bd->addr.hi),
657 					   le32_to_cpu(first_bd->addr.lo)) +
658 					   hlen;
659 
660 			BD_SET_UNMAP_ADDR_LEN(tx_data_bd, mapping,
661 					      le16_to_cpu(first_bd->nbytes) -
662 					      hlen);
663 
664 			/* this marks the BD as one that has no
665 			 * individual mapping
666 			 */
667 			txq->sw_tx_ring[idx].flags |= QEDE_TSO_SPLIT_BD;
668 
669 			first_bd->nbytes = cpu_to_le16(hlen);
670 
671 			tx_data_bd = (struct eth_tx_bd *)third_bd;
672 			data_split = true;
673 		}
674 	} else {
675 		first_bd->data.bitfields |=
676 		    (skb->len & ETH_TX_DATA_1ST_BD_PKT_LEN_MASK) <<
677 		    ETH_TX_DATA_1ST_BD_PKT_LEN_SHIFT;
678 	}
679 
680 	/* Handle fragmented skb */
681 	/* special handle for frags inside 2nd and 3rd bds.. */
682 	while (tx_data_bd && frag_idx < skb_shinfo(skb)->nr_frags) {
683 		rc = map_frag_to_bd(edev,
684 				    &skb_shinfo(skb)->frags[frag_idx],
685 				    tx_data_bd);
686 		if (rc) {
687 			qede_free_failed_tx_pkt(edev, txq, first_bd, nbd,
688 						data_split);
689 			qede_update_tx_producer(txq);
690 			return NETDEV_TX_OK;
691 		}
692 
693 		if (tx_data_bd == (struct eth_tx_bd *)second_bd)
694 			tx_data_bd = (struct eth_tx_bd *)third_bd;
695 		else
696 			tx_data_bd = NULL;
697 
698 		frag_idx++;
699 	}
700 
701 	/* map last frags into 4th, 5th .... */
702 	for (; frag_idx < skb_shinfo(skb)->nr_frags; frag_idx++, nbd++) {
703 		tx_data_bd = (struct eth_tx_bd *)
704 			     qed_chain_produce(&txq->tx_pbl);
705 
706 		memset(tx_data_bd, 0, sizeof(*tx_data_bd));
707 
708 		rc = map_frag_to_bd(edev,
709 				    &skb_shinfo(skb)->frags[frag_idx],
710 				    tx_data_bd);
711 		if (rc) {
712 			qede_free_failed_tx_pkt(edev, txq, first_bd, nbd,
713 						data_split);
714 			qede_update_tx_producer(txq);
715 			return NETDEV_TX_OK;
716 		}
717 	}
718 
719 	/* update the first BD with the actual num BDs */
720 	first_bd->data.nbds = nbd;
721 
722 	netdev_tx_sent_queue(netdev_txq, skb->len);
723 
724 	skb_tx_timestamp(skb);
725 
726 	/* Advance packet producer only before sending the packet since mapping
727 	 * of pages may fail.
728 	 */
729 	txq->sw_tx_prod++;
730 
731 	/* 'next page' entries are counted in the producer value */
732 	txq->tx_db.data.bd_prod =
733 		cpu_to_le16(qed_chain_get_prod_idx(&txq->tx_pbl));
734 
735 	if (!skb->xmit_more || netif_xmit_stopped(netdev_txq))
736 		qede_update_tx_producer(txq);
737 
738 	if (unlikely(qed_chain_get_elem_left(&txq->tx_pbl)
739 		      < (MAX_SKB_FRAGS + 1))) {
740 		if (skb->xmit_more)
741 			qede_update_tx_producer(txq);
742 
743 		netif_tx_stop_queue(netdev_txq);
744 		txq->stopped_cnt++;
745 		DP_VERBOSE(edev, NETIF_MSG_TX_QUEUED,
746 			   "Stop queue was called\n");
747 		/* paired memory barrier is in qede_tx_int(), we have to keep
748 		 * ordering of set_bit() in netif_tx_stop_queue() and read of
749 		 * fp->bd_tx_cons
750 		 */
751 		smp_mb();
752 
753 		if (qed_chain_get_elem_left(&txq->tx_pbl)
754 		     >= (MAX_SKB_FRAGS + 1) &&
755 		    (edev->state == QEDE_STATE_OPEN)) {
756 			netif_tx_wake_queue(netdev_txq);
757 			DP_VERBOSE(edev, NETIF_MSG_TX_QUEUED,
758 				   "Wake queue was called\n");
759 		}
760 	}
761 
762 	return NETDEV_TX_OK;
763 }
764 
765 int qede_txq_has_work(struct qede_tx_queue *txq)
766 {
767 	u16 hw_bd_cons;
768 
769 	/* Tell compiler that consumer and producer can change */
770 	barrier();
771 	hw_bd_cons = le16_to_cpu(*txq->hw_cons_ptr);
772 	if (qed_chain_get_cons_idx(&txq->tx_pbl) == hw_bd_cons + 1)
773 		return 0;
774 
775 	return hw_bd_cons != qed_chain_get_cons_idx(&txq->tx_pbl);
776 }
777 
778 static int qede_tx_int(struct qede_dev *edev, struct qede_tx_queue *txq)
779 {
780 	struct netdev_queue *netdev_txq;
781 	u16 hw_bd_cons;
782 	unsigned int pkts_compl = 0, bytes_compl = 0;
783 	int rc;
784 
785 	netdev_txq = netdev_get_tx_queue(edev->ndev, txq->index);
786 
787 	hw_bd_cons = le16_to_cpu(*txq->hw_cons_ptr);
788 	barrier();
789 
790 	while (hw_bd_cons != qed_chain_get_cons_idx(&txq->tx_pbl)) {
791 		int len = 0;
792 
793 		rc = qede_free_tx_pkt(edev, txq, &len);
794 		if (rc) {
795 			DP_NOTICE(edev, "hw_bd_cons = %d, chain_cons=%d\n",
796 				  hw_bd_cons,
797 				  qed_chain_get_cons_idx(&txq->tx_pbl));
798 			break;
799 		}
800 
801 		bytes_compl += len;
802 		pkts_compl++;
803 		txq->sw_tx_cons++;
804 		txq->xmit_pkts++;
805 	}
806 
807 	netdev_tx_completed_queue(netdev_txq, pkts_compl, bytes_compl);
808 
809 	/* Need to make the tx_bd_cons update visible to start_xmit()
810 	 * before checking for netif_tx_queue_stopped().  Without the
811 	 * memory barrier, there is a small possibility that
812 	 * start_xmit() will miss it and cause the queue to be stopped
813 	 * forever.
814 	 * On the other hand we need an rmb() here to ensure the proper
815 	 * ordering of bit testing in the following
816 	 * netif_tx_queue_stopped(txq) call.
817 	 */
818 	smp_mb();
819 
820 	if (unlikely(netif_tx_queue_stopped(netdev_txq))) {
821 		/* Taking tx_lock is needed to prevent reenabling the queue
822 		 * while it's empty. This could have happen if rx_action() gets
823 		 * suspended in qede_tx_int() after the condition before
824 		 * netif_tx_wake_queue(), while tx_action (qede_start_xmit()):
825 		 *
826 		 * stops the queue->sees fresh tx_bd_cons->releases the queue->
827 		 * sends some packets consuming the whole queue again->
828 		 * stops the queue
829 		 */
830 
831 		__netif_tx_lock(netdev_txq, smp_processor_id());
832 
833 		if ((netif_tx_queue_stopped(netdev_txq)) &&
834 		    (edev->state == QEDE_STATE_OPEN) &&
835 		    (qed_chain_get_elem_left(&txq->tx_pbl)
836 		      >= (MAX_SKB_FRAGS + 1))) {
837 			netif_tx_wake_queue(netdev_txq);
838 			DP_VERBOSE(edev, NETIF_MSG_TX_DONE,
839 				   "Wake queue was called\n");
840 		}
841 
842 		__netif_tx_unlock(netdev_txq);
843 	}
844 
845 	return 0;
846 }
847 
848 bool qede_has_rx_work(struct qede_rx_queue *rxq)
849 {
850 	u16 hw_comp_cons, sw_comp_cons;
851 
852 	/* Tell compiler that status block fields can change */
853 	barrier();
854 
855 	hw_comp_cons = le16_to_cpu(*rxq->hw_cons_ptr);
856 	sw_comp_cons = qed_chain_get_cons_idx(&rxq->rx_comp_ring);
857 
858 	return hw_comp_cons != sw_comp_cons;
859 }
860 
861 static bool qede_has_tx_work(struct qede_fastpath *fp)
862 {
863 	u8 tc;
864 
865 	for (tc = 0; tc < fp->edev->num_tc; tc++)
866 		if (qede_txq_has_work(&fp->txqs[tc]))
867 			return true;
868 	return false;
869 }
870 
871 static inline void qede_rx_bd_ring_consume(struct qede_rx_queue *rxq)
872 {
873 	qed_chain_consume(&rxq->rx_bd_ring);
874 	rxq->sw_rx_cons++;
875 }
876 
877 /* This function reuses the buffer(from an offset) from
878  * consumer index to producer index in the bd ring
879  */
880 static inline void qede_reuse_page(struct qede_dev *edev,
881 				   struct qede_rx_queue *rxq,
882 				   struct sw_rx_data *curr_cons)
883 {
884 	struct eth_rx_bd *rx_bd_prod = qed_chain_produce(&rxq->rx_bd_ring);
885 	struct sw_rx_data *curr_prod;
886 	dma_addr_t new_mapping;
887 
888 	curr_prod = &rxq->sw_rx_ring[rxq->sw_rx_prod & NUM_RX_BDS_MAX];
889 	*curr_prod = *curr_cons;
890 
891 	new_mapping = curr_prod->mapping + curr_prod->page_offset;
892 
893 	rx_bd_prod->addr.hi = cpu_to_le32(upper_32_bits(new_mapping));
894 	rx_bd_prod->addr.lo = cpu_to_le32(lower_32_bits(new_mapping));
895 
896 	rxq->sw_rx_prod++;
897 	curr_cons->data = NULL;
898 }
899 
900 /* In case of allocation failures reuse buffers
901  * from consumer index to produce buffers for firmware
902  */
903 void qede_recycle_rx_bd_ring(struct qede_rx_queue *rxq,
904 			     struct qede_dev *edev, u8 count)
905 {
906 	struct sw_rx_data *curr_cons;
907 
908 	for (; count > 0; count--) {
909 		curr_cons = &rxq->sw_rx_ring[rxq->sw_rx_cons & NUM_RX_BDS_MAX];
910 		qede_reuse_page(edev, rxq, curr_cons);
911 		qede_rx_bd_ring_consume(rxq);
912 	}
913 }
914 
915 static inline int qede_realloc_rx_buffer(struct qede_dev *edev,
916 					 struct qede_rx_queue *rxq,
917 					 struct sw_rx_data *curr_cons)
918 {
919 	/* Move to the next segment in the page */
920 	curr_cons->page_offset += rxq->rx_buf_seg_size;
921 
922 	if (curr_cons->page_offset == PAGE_SIZE) {
923 		if (unlikely(qede_alloc_rx_buffer(edev, rxq))) {
924 			/* Since we failed to allocate new buffer
925 			 * current buffer can be used again.
926 			 */
927 			curr_cons->page_offset -= rxq->rx_buf_seg_size;
928 
929 			return -ENOMEM;
930 		}
931 
932 		dma_unmap_page(&edev->pdev->dev, curr_cons->mapping,
933 			       PAGE_SIZE, DMA_FROM_DEVICE);
934 	} else {
935 		/* Increment refcount of the page as we don't want
936 		 * network stack to take the ownership of the page
937 		 * which can be recycled multiple times by the driver.
938 		 */
939 		page_ref_inc(curr_cons->data);
940 		qede_reuse_page(edev, rxq, curr_cons);
941 	}
942 
943 	return 0;
944 }
945 
946 static inline void qede_update_rx_prod(struct qede_dev *edev,
947 				       struct qede_rx_queue *rxq)
948 {
949 	u16 bd_prod = qed_chain_get_prod_idx(&rxq->rx_bd_ring);
950 	u16 cqe_prod = qed_chain_get_prod_idx(&rxq->rx_comp_ring);
951 	struct eth_rx_prod_data rx_prods = {0};
952 
953 	/* Update producers */
954 	rx_prods.bd_prod = cpu_to_le16(bd_prod);
955 	rx_prods.cqe_prod = cpu_to_le16(cqe_prod);
956 
957 	/* Make sure that the BD and SGE data is updated before updating the
958 	 * producers since FW might read the BD/SGE right after the producer
959 	 * is updated.
960 	 */
961 	wmb();
962 
963 	internal_ram_wr(rxq->hw_rxq_prod_addr, sizeof(rx_prods),
964 			(u32 *)&rx_prods);
965 
966 	/* mmiowb is needed to synchronize doorbell writes from more than one
967 	 * processor. It guarantees that the write arrives to the device before
968 	 * the napi lock is released and another qede_poll is called (possibly
969 	 * on another CPU). Without this barrier, the next doorbell can bypass
970 	 * this doorbell. This is applicable to IA64/Altix systems.
971 	 */
972 	mmiowb();
973 }
974 
975 static u32 qede_get_rxhash(struct qede_dev *edev,
976 			   u8 bitfields,
977 			   __le32 rss_hash, enum pkt_hash_types *rxhash_type)
978 {
979 	enum rss_hash_type htype;
980 
981 	htype = GET_FIELD(bitfields, ETH_FAST_PATH_RX_REG_CQE_RSS_HASH_TYPE);
982 
983 	if ((edev->ndev->features & NETIF_F_RXHASH) && htype) {
984 		*rxhash_type = ((htype == RSS_HASH_TYPE_IPV4) ||
985 				(htype == RSS_HASH_TYPE_IPV6)) ?
986 				PKT_HASH_TYPE_L3 : PKT_HASH_TYPE_L4;
987 		return le32_to_cpu(rss_hash);
988 	}
989 	*rxhash_type = PKT_HASH_TYPE_NONE;
990 	return 0;
991 }
992 
993 static void qede_set_skb_csum(struct sk_buff *skb, u8 csum_flag)
994 {
995 	skb_checksum_none_assert(skb);
996 
997 	if (csum_flag & QEDE_CSUM_UNNECESSARY)
998 		skb->ip_summed = CHECKSUM_UNNECESSARY;
999 
1000 	if (csum_flag & QEDE_TUNN_CSUM_UNNECESSARY)
1001 		skb->csum_level = 1;
1002 }
1003 
1004 static inline void qede_skb_receive(struct qede_dev *edev,
1005 				    struct qede_fastpath *fp,
1006 				    struct sk_buff *skb, u16 vlan_tag)
1007 {
1008 	if (vlan_tag)
1009 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
1010 
1011 	napi_gro_receive(&fp->napi, skb);
1012 }
1013 
1014 static void qede_set_gro_params(struct qede_dev *edev,
1015 				struct sk_buff *skb,
1016 				struct eth_fast_path_rx_tpa_start_cqe *cqe)
1017 {
1018 	u16 parsing_flags = le16_to_cpu(cqe->pars_flags.flags);
1019 
1020 	if (((parsing_flags >> PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) &
1021 	    PARSING_AND_ERR_FLAGS_L3TYPE_MASK) == 2)
1022 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1023 	else
1024 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1025 
1026 	skb_shinfo(skb)->gso_size = __le16_to_cpu(cqe->len_on_first_bd) -
1027 					cqe->header_len;
1028 }
1029 
1030 static int qede_fill_frag_skb(struct qede_dev *edev,
1031 			      struct qede_rx_queue *rxq,
1032 			      u8 tpa_agg_index, u16 len_on_bd)
1033 {
1034 	struct sw_rx_data *current_bd = &rxq->sw_rx_ring[rxq->sw_rx_cons &
1035 							 NUM_RX_BDS_MAX];
1036 	struct qede_agg_info *tpa_info = &rxq->tpa_info[tpa_agg_index];
1037 	struct sk_buff *skb = tpa_info->skb;
1038 
1039 	if (unlikely(tpa_info->agg_state != QEDE_AGG_STATE_START))
1040 		goto out;
1041 
1042 	/* Add one frag and update the appropriate fields in the skb */
1043 	skb_fill_page_desc(skb, tpa_info->frag_id++,
1044 			   current_bd->data, current_bd->page_offset,
1045 			   len_on_bd);
1046 
1047 	if (unlikely(qede_realloc_rx_buffer(edev, rxq, current_bd))) {
1048 		/* Incr page ref count to reuse on allocation failure
1049 		 * so that it doesn't get freed while freeing SKB.
1050 		 */
1051 		page_ref_inc(current_bd->data);
1052 		goto out;
1053 	}
1054 
1055 	qed_chain_consume(&rxq->rx_bd_ring);
1056 	rxq->sw_rx_cons++;
1057 
1058 	skb->data_len += len_on_bd;
1059 	skb->truesize += rxq->rx_buf_seg_size;
1060 	skb->len += len_on_bd;
1061 
1062 	return 0;
1063 
1064 out:
1065 	tpa_info->agg_state = QEDE_AGG_STATE_ERROR;
1066 	qede_recycle_rx_bd_ring(rxq, edev, 1);
1067 	return -ENOMEM;
1068 }
1069 
1070 static void qede_tpa_start(struct qede_dev *edev,
1071 			   struct qede_rx_queue *rxq,
1072 			   struct eth_fast_path_rx_tpa_start_cqe *cqe)
1073 {
1074 	struct qede_agg_info *tpa_info = &rxq->tpa_info[cqe->tpa_agg_index];
1075 	struct eth_rx_bd *rx_bd_cons = qed_chain_consume(&rxq->rx_bd_ring);
1076 	struct eth_rx_bd *rx_bd_prod = qed_chain_produce(&rxq->rx_bd_ring);
1077 	struct sw_rx_data *replace_buf = &tpa_info->replace_buf;
1078 	dma_addr_t mapping = tpa_info->replace_buf_mapping;
1079 	struct sw_rx_data *sw_rx_data_cons;
1080 	struct sw_rx_data *sw_rx_data_prod;
1081 	enum pkt_hash_types rxhash_type;
1082 	u32 rxhash;
1083 
1084 	sw_rx_data_cons = &rxq->sw_rx_ring[rxq->sw_rx_cons & NUM_RX_BDS_MAX];
1085 	sw_rx_data_prod = &rxq->sw_rx_ring[rxq->sw_rx_prod & NUM_RX_BDS_MAX];
1086 
1087 	/* Use pre-allocated replacement buffer - we can't release the agg.
1088 	 * start until its over and we don't want to risk allocation failing
1089 	 * here, so re-allocate when aggregation will be over.
1090 	 */
1091 	sw_rx_data_prod->mapping = replace_buf->mapping;
1092 
1093 	sw_rx_data_prod->data = replace_buf->data;
1094 	rx_bd_prod->addr.hi = cpu_to_le32(upper_32_bits(mapping));
1095 	rx_bd_prod->addr.lo = cpu_to_le32(lower_32_bits(mapping));
1096 	sw_rx_data_prod->page_offset = replace_buf->page_offset;
1097 
1098 	rxq->sw_rx_prod++;
1099 
1100 	/* move partial skb from cons to pool (don't unmap yet)
1101 	 * save mapping, incase we drop the packet later on.
1102 	 */
1103 	tpa_info->start_buf = *sw_rx_data_cons;
1104 	mapping = HILO_U64(le32_to_cpu(rx_bd_cons->addr.hi),
1105 			   le32_to_cpu(rx_bd_cons->addr.lo));
1106 
1107 	tpa_info->start_buf_mapping = mapping;
1108 	rxq->sw_rx_cons++;
1109 
1110 	/* set tpa state to start only if we are able to allocate skb
1111 	 * for this aggregation, otherwise mark as error and aggregation will
1112 	 * be dropped
1113 	 */
1114 	tpa_info->skb = netdev_alloc_skb(edev->ndev,
1115 					 le16_to_cpu(cqe->len_on_first_bd));
1116 	if (unlikely(!tpa_info->skb)) {
1117 		DP_NOTICE(edev, "Failed to allocate SKB for gro\n");
1118 		tpa_info->agg_state = QEDE_AGG_STATE_ERROR;
1119 		goto cons_buf;
1120 	}
1121 
1122 	skb_put(tpa_info->skb, le16_to_cpu(cqe->len_on_first_bd));
1123 	memcpy(&tpa_info->start_cqe, cqe, sizeof(tpa_info->start_cqe));
1124 
1125 	/* Start filling in the aggregation info */
1126 	tpa_info->frag_id = 0;
1127 	tpa_info->agg_state = QEDE_AGG_STATE_START;
1128 
1129 	rxhash = qede_get_rxhash(edev, cqe->bitfields,
1130 				 cqe->rss_hash, &rxhash_type);
1131 	skb_set_hash(tpa_info->skb, rxhash, rxhash_type);
1132 	if ((le16_to_cpu(cqe->pars_flags.flags) >>
1133 	     PARSING_AND_ERR_FLAGS_TAG8021QEXIST_SHIFT) &
1134 		    PARSING_AND_ERR_FLAGS_TAG8021QEXIST_MASK)
1135 		tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
1136 	else
1137 		tpa_info->vlan_tag = 0;
1138 
1139 	/* This is needed in order to enable forwarding support */
1140 	qede_set_gro_params(edev, tpa_info->skb, cqe);
1141 
1142 cons_buf: /* We still need to handle bd_len_list to consume buffers */
1143 	if (likely(cqe->ext_bd_len_list[0]))
1144 		qede_fill_frag_skb(edev, rxq, cqe->tpa_agg_index,
1145 				   le16_to_cpu(cqe->ext_bd_len_list[0]));
1146 
1147 	if (unlikely(cqe->ext_bd_len_list[1])) {
1148 		DP_ERR(edev,
1149 		       "Unlikely - got a TPA aggregation with more than one ext_bd_len_list entry in the TPA start\n");
1150 		tpa_info->agg_state = QEDE_AGG_STATE_ERROR;
1151 	}
1152 }
1153 
1154 #ifdef CONFIG_INET
1155 static void qede_gro_ip_csum(struct sk_buff *skb)
1156 {
1157 	const struct iphdr *iph = ip_hdr(skb);
1158 	struct tcphdr *th;
1159 
1160 	skb_set_transport_header(skb, sizeof(struct iphdr));
1161 	th = tcp_hdr(skb);
1162 
1163 	th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
1164 				  iph->saddr, iph->daddr, 0);
1165 
1166 	tcp_gro_complete(skb);
1167 }
1168 
1169 static void qede_gro_ipv6_csum(struct sk_buff *skb)
1170 {
1171 	struct ipv6hdr *iph = ipv6_hdr(skb);
1172 	struct tcphdr *th;
1173 
1174 	skb_set_transport_header(skb, sizeof(struct ipv6hdr));
1175 	th = tcp_hdr(skb);
1176 
1177 	th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
1178 				  &iph->saddr, &iph->daddr, 0);
1179 	tcp_gro_complete(skb);
1180 }
1181 #endif
1182 
1183 static void qede_gro_receive(struct qede_dev *edev,
1184 			     struct qede_fastpath *fp,
1185 			     struct sk_buff *skb,
1186 			     u16 vlan_tag)
1187 {
1188 	/* FW can send a single MTU sized packet from gro flow
1189 	 * due to aggregation timeout/last segment etc. which
1190 	 * is not expected to be a gro packet. If a skb has zero
1191 	 * frags then simply push it in the stack as non gso skb.
1192 	 */
1193 	if (unlikely(!skb->data_len)) {
1194 		skb_shinfo(skb)->gso_type = 0;
1195 		skb_shinfo(skb)->gso_size = 0;
1196 		goto send_skb;
1197 	}
1198 
1199 #ifdef CONFIG_INET
1200 	if (skb_shinfo(skb)->gso_size) {
1201 		skb_set_network_header(skb, 0);
1202 
1203 		switch (skb->protocol) {
1204 		case htons(ETH_P_IP):
1205 			qede_gro_ip_csum(skb);
1206 			break;
1207 		case htons(ETH_P_IPV6):
1208 			qede_gro_ipv6_csum(skb);
1209 			break;
1210 		default:
1211 			DP_ERR(edev,
1212 			       "Error: FW GRO supports only IPv4/IPv6, not 0x%04x\n",
1213 			       ntohs(skb->protocol));
1214 		}
1215 	}
1216 #endif
1217 
1218 send_skb:
1219 	skb_record_rx_queue(skb, fp->rxq->rxq_id);
1220 	qede_skb_receive(edev, fp, skb, vlan_tag);
1221 }
1222 
1223 static inline void qede_tpa_cont(struct qede_dev *edev,
1224 				 struct qede_rx_queue *rxq,
1225 				 struct eth_fast_path_rx_tpa_cont_cqe *cqe)
1226 {
1227 	int i;
1228 
1229 	for (i = 0; cqe->len_list[i]; i++)
1230 		qede_fill_frag_skb(edev, rxq, cqe->tpa_agg_index,
1231 				   le16_to_cpu(cqe->len_list[i]));
1232 
1233 	if (unlikely(i > 1))
1234 		DP_ERR(edev,
1235 		       "Strange - TPA cont with more than a single len_list entry\n");
1236 }
1237 
1238 static void qede_tpa_end(struct qede_dev *edev,
1239 			 struct qede_fastpath *fp,
1240 			 struct eth_fast_path_rx_tpa_end_cqe *cqe)
1241 {
1242 	struct qede_rx_queue *rxq = fp->rxq;
1243 	struct qede_agg_info *tpa_info;
1244 	struct sk_buff *skb;
1245 	int i;
1246 
1247 	tpa_info = &rxq->tpa_info[cqe->tpa_agg_index];
1248 	skb = tpa_info->skb;
1249 
1250 	for (i = 0; cqe->len_list[i]; i++)
1251 		qede_fill_frag_skb(edev, rxq, cqe->tpa_agg_index,
1252 				   le16_to_cpu(cqe->len_list[i]));
1253 	if (unlikely(i > 1))
1254 		DP_ERR(edev,
1255 		       "Strange - TPA emd with more than a single len_list entry\n");
1256 
1257 	if (unlikely(tpa_info->agg_state != QEDE_AGG_STATE_START))
1258 		goto err;
1259 
1260 	/* Sanity */
1261 	if (unlikely(cqe->num_of_bds != tpa_info->frag_id + 1))
1262 		DP_ERR(edev,
1263 		       "Strange - TPA had %02x BDs, but SKB has only %d frags\n",
1264 		       cqe->num_of_bds, tpa_info->frag_id);
1265 	if (unlikely(skb->len != le16_to_cpu(cqe->total_packet_len)))
1266 		DP_ERR(edev,
1267 		       "Strange - total packet len [cqe] is %4x but SKB has len %04x\n",
1268 		       le16_to_cpu(cqe->total_packet_len), skb->len);
1269 
1270 	memcpy(skb->data,
1271 	       page_address(tpa_info->start_buf.data) +
1272 		tpa_info->start_cqe.placement_offset +
1273 		tpa_info->start_buf.page_offset,
1274 	       le16_to_cpu(tpa_info->start_cqe.len_on_first_bd));
1275 
1276 	/* Recycle [mapped] start buffer for the next replacement */
1277 	tpa_info->replace_buf = tpa_info->start_buf;
1278 	tpa_info->replace_buf_mapping = tpa_info->start_buf_mapping;
1279 
1280 	/* Finalize the SKB */
1281 	skb->protocol = eth_type_trans(skb, edev->ndev);
1282 	skb->ip_summed = CHECKSUM_UNNECESSARY;
1283 
1284 	/* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
1285 	 * to skb_shinfo(skb)->gso_segs
1286 	 */
1287 	NAPI_GRO_CB(skb)->count = le16_to_cpu(cqe->num_of_coalesced_segs);
1288 
1289 	qede_gro_receive(edev, fp, skb, tpa_info->vlan_tag);
1290 
1291 	tpa_info->agg_state = QEDE_AGG_STATE_NONE;
1292 
1293 	return;
1294 err:
1295 	/* The BD starting the aggregation is still mapped; Re-use it for
1296 	 * future aggregations [as replacement buffer]
1297 	 */
1298 	memcpy(&tpa_info->replace_buf, &tpa_info->start_buf,
1299 	       sizeof(struct sw_rx_data));
1300 	tpa_info->replace_buf_mapping = tpa_info->start_buf_mapping;
1301 	tpa_info->start_buf.data = NULL;
1302 	tpa_info->agg_state = QEDE_AGG_STATE_NONE;
1303 	dev_kfree_skb_any(tpa_info->skb);
1304 	tpa_info->skb = NULL;
1305 }
1306 
1307 static bool qede_tunn_exist(u16 flag)
1308 {
1309 	return !!(flag & (PARSING_AND_ERR_FLAGS_TUNNELEXIST_MASK <<
1310 			  PARSING_AND_ERR_FLAGS_TUNNELEXIST_SHIFT));
1311 }
1312 
1313 static u8 qede_check_tunn_csum(u16 flag)
1314 {
1315 	u16 csum_flag = 0;
1316 	u8 tcsum = 0;
1317 
1318 	if (flag & (PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMWASCALCULATED_MASK <<
1319 		    PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMWASCALCULATED_SHIFT))
1320 		csum_flag |= PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_MASK <<
1321 			     PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_SHIFT;
1322 
1323 	if (flag & (PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_MASK <<
1324 		    PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_SHIFT)) {
1325 		csum_flag |= PARSING_AND_ERR_FLAGS_L4CHKSMERROR_MASK <<
1326 			     PARSING_AND_ERR_FLAGS_L4CHKSMERROR_SHIFT;
1327 		tcsum = QEDE_TUNN_CSUM_UNNECESSARY;
1328 	}
1329 
1330 	csum_flag |= PARSING_AND_ERR_FLAGS_TUNNELIPHDRERROR_MASK <<
1331 		     PARSING_AND_ERR_FLAGS_TUNNELIPHDRERROR_SHIFT |
1332 		     PARSING_AND_ERR_FLAGS_IPHDRERROR_MASK <<
1333 		     PARSING_AND_ERR_FLAGS_IPHDRERROR_SHIFT;
1334 
1335 	if (csum_flag & flag)
1336 		return QEDE_CSUM_ERROR;
1337 
1338 	return QEDE_CSUM_UNNECESSARY | tcsum;
1339 }
1340 
1341 static u8 qede_check_notunn_csum(u16 flag)
1342 {
1343 	u16 csum_flag = 0;
1344 	u8 csum = 0;
1345 
1346 	if (flag & (PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_MASK <<
1347 		    PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_SHIFT)) {
1348 		csum_flag |= PARSING_AND_ERR_FLAGS_L4CHKSMERROR_MASK <<
1349 			     PARSING_AND_ERR_FLAGS_L4CHKSMERROR_SHIFT;
1350 		csum = QEDE_CSUM_UNNECESSARY;
1351 	}
1352 
1353 	csum_flag |= PARSING_AND_ERR_FLAGS_IPHDRERROR_MASK <<
1354 		     PARSING_AND_ERR_FLAGS_IPHDRERROR_SHIFT;
1355 
1356 	if (csum_flag & flag)
1357 		return QEDE_CSUM_ERROR;
1358 
1359 	return csum;
1360 }
1361 
1362 static u8 qede_check_csum(u16 flag)
1363 {
1364 	if (!qede_tunn_exist(flag))
1365 		return qede_check_notunn_csum(flag);
1366 	else
1367 		return qede_check_tunn_csum(flag);
1368 }
1369 
1370 static bool qede_pkt_is_ip_fragmented(struct eth_fast_path_rx_reg_cqe *cqe,
1371 				      u16 flag)
1372 {
1373 	u8 tun_pars_flg = cqe->tunnel_pars_flags.flags;
1374 
1375 	if ((tun_pars_flg & (ETH_TUNNEL_PARSING_FLAGS_IPV4_FRAGMENT_MASK <<
1376 			     ETH_TUNNEL_PARSING_FLAGS_IPV4_FRAGMENT_SHIFT)) ||
1377 	    (flag & (PARSING_AND_ERR_FLAGS_IPV4FRAG_MASK <<
1378 		     PARSING_AND_ERR_FLAGS_IPV4FRAG_SHIFT)))
1379 		return true;
1380 
1381 	return false;
1382 }
1383 
1384 static int qede_rx_int(struct qede_fastpath *fp, int budget)
1385 {
1386 	struct qede_dev *edev = fp->edev;
1387 	struct qede_rx_queue *rxq = fp->rxq;
1388 
1389 	u16 hw_comp_cons, sw_comp_cons, sw_rx_index, parse_flag;
1390 	int rx_pkt = 0;
1391 	u8 csum_flag;
1392 
1393 	hw_comp_cons = le16_to_cpu(*rxq->hw_cons_ptr);
1394 	sw_comp_cons = qed_chain_get_cons_idx(&rxq->rx_comp_ring);
1395 
1396 	/* Memory barrier to prevent the CPU from doing speculative reads of CQE
1397 	 * / BD in the while-loop before reading hw_comp_cons. If the CQE is
1398 	 * read before it is written by FW, then FW writes CQE and SB, and then
1399 	 * the CPU reads the hw_comp_cons, it will use an old CQE.
1400 	 */
1401 	rmb();
1402 
1403 	/* Loop to complete all indicated BDs */
1404 	while (sw_comp_cons != hw_comp_cons) {
1405 		struct eth_fast_path_rx_reg_cqe *fp_cqe;
1406 		enum pkt_hash_types rxhash_type;
1407 		enum eth_rx_cqe_type cqe_type;
1408 		struct sw_rx_data *sw_rx_data;
1409 		union eth_rx_cqe *cqe;
1410 		struct sk_buff *skb;
1411 		struct page *data;
1412 		__le16 flags;
1413 		u16 len, pad;
1414 		u32 rx_hash;
1415 
1416 		/* Get the CQE from the completion ring */
1417 		cqe = (union eth_rx_cqe *)
1418 			qed_chain_consume(&rxq->rx_comp_ring);
1419 		cqe_type = cqe->fast_path_regular.type;
1420 
1421 		if (unlikely(cqe_type == ETH_RX_CQE_TYPE_SLOW_PATH)) {
1422 			edev->ops->eth_cqe_completion(
1423 					edev->cdev, fp->id,
1424 					(struct eth_slow_path_rx_cqe *)cqe);
1425 			goto next_cqe;
1426 		}
1427 
1428 		if (cqe_type != ETH_RX_CQE_TYPE_REGULAR) {
1429 			switch (cqe_type) {
1430 			case ETH_RX_CQE_TYPE_TPA_START:
1431 				qede_tpa_start(edev, rxq,
1432 					       &cqe->fast_path_tpa_start);
1433 				goto next_cqe;
1434 			case ETH_RX_CQE_TYPE_TPA_CONT:
1435 				qede_tpa_cont(edev, rxq,
1436 					      &cqe->fast_path_tpa_cont);
1437 				goto next_cqe;
1438 			case ETH_RX_CQE_TYPE_TPA_END:
1439 				qede_tpa_end(edev, fp,
1440 					     &cqe->fast_path_tpa_end);
1441 				goto next_rx_only;
1442 			default:
1443 				break;
1444 			}
1445 		}
1446 
1447 		/* Get the data from the SW ring */
1448 		sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS_MAX;
1449 		sw_rx_data = &rxq->sw_rx_ring[sw_rx_index];
1450 		data = sw_rx_data->data;
1451 
1452 		fp_cqe = &cqe->fast_path_regular;
1453 		len =  le16_to_cpu(fp_cqe->len_on_first_bd);
1454 		pad = fp_cqe->placement_offset;
1455 		flags = cqe->fast_path_regular.pars_flags.flags;
1456 
1457 		/* If this is an error packet then drop it */
1458 		parse_flag = le16_to_cpu(flags);
1459 
1460 		csum_flag = qede_check_csum(parse_flag);
1461 		if (unlikely(csum_flag == QEDE_CSUM_ERROR)) {
1462 			if (qede_pkt_is_ip_fragmented(&cqe->fast_path_regular,
1463 						      parse_flag)) {
1464 				rxq->rx_ip_frags++;
1465 				goto alloc_skb;
1466 			}
1467 
1468 			DP_NOTICE(edev,
1469 				  "CQE in CONS = %u has error, flags = %x, dropping incoming packet\n",
1470 				  sw_comp_cons, parse_flag);
1471 			rxq->rx_hw_errors++;
1472 			qede_recycle_rx_bd_ring(rxq, edev, fp_cqe->bd_num);
1473 			goto next_cqe;
1474 		}
1475 
1476 alloc_skb:
1477 		skb = netdev_alloc_skb(edev->ndev, QEDE_RX_HDR_SIZE);
1478 		if (unlikely(!skb)) {
1479 			DP_NOTICE(edev,
1480 				  "skb allocation failed, dropping incoming packet\n");
1481 			qede_recycle_rx_bd_ring(rxq, edev, fp_cqe->bd_num);
1482 			rxq->rx_alloc_errors++;
1483 			goto next_cqe;
1484 		}
1485 
1486 		/* Copy data into SKB */
1487 		if (len + pad <= edev->rx_copybreak) {
1488 			memcpy(skb_put(skb, len),
1489 			       page_address(data) + pad +
1490 				sw_rx_data->page_offset, len);
1491 			qede_reuse_page(edev, rxq, sw_rx_data);
1492 		} else {
1493 			struct skb_frag_struct *frag;
1494 			unsigned int pull_len;
1495 			unsigned char *va;
1496 
1497 			frag = &skb_shinfo(skb)->frags[0];
1498 
1499 			skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, data,
1500 					pad + sw_rx_data->page_offset,
1501 					len, rxq->rx_buf_seg_size);
1502 
1503 			va = skb_frag_address(frag);
1504 			pull_len = eth_get_headlen(va, QEDE_RX_HDR_SIZE);
1505 
1506 			/* Align the pull_len to optimize memcpy */
1507 			memcpy(skb->data, va, ALIGN(pull_len, sizeof(long)));
1508 
1509 			skb_frag_size_sub(frag, pull_len);
1510 			frag->page_offset += pull_len;
1511 			skb->data_len -= pull_len;
1512 			skb->tail += pull_len;
1513 
1514 			if (unlikely(qede_realloc_rx_buffer(edev, rxq,
1515 							    sw_rx_data))) {
1516 				DP_ERR(edev, "Failed to allocate rx buffer\n");
1517 				/* Incr page ref count to reuse on allocation
1518 				 * failure so that it doesn't get freed while
1519 				 * freeing SKB.
1520 				 */
1521 
1522 				page_ref_inc(sw_rx_data->data);
1523 				rxq->rx_alloc_errors++;
1524 				qede_recycle_rx_bd_ring(rxq, edev,
1525 							fp_cqe->bd_num);
1526 				dev_kfree_skb_any(skb);
1527 				goto next_cqe;
1528 			}
1529 		}
1530 
1531 		qede_rx_bd_ring_consume(rxq);
1532 
1533 		if (fp_cqe->bd_num != 1) {
1534 			u16 pkt_len = le16_to_cpu(fp_cqe->pkt_len);
1535 			u8 num_frags;
1536 
1537 			pkt_len -= len;
1538 
1539 			for (num_frags = fp_cqe->bd_num - 1; num_frags > 0;
1540 			     num_frags--) {
1541 				u16 cur_size = pkt_len > rxq->rx_buf_size ?
1542 						rxq->rx_buf_size : pkt_len;
1543 				if (unlikely(!cur_size)) {
1544 					DP_ERR(edev,
1545 					       "Still got %d BDs for mapping jumbo, but length became 0\n",
1546 					       num_frags);
1547 					qede_recycle_rx_bd_ring(rxq, edev,
1548 								num_frags);
1549 					dev_kfree_skb_any(skb);
1550 					goto next_cqe;
1551 				}
1552 
1553 				if (unlikely(qede_alloc_rx_buffer(edev, rxq))) {
1554 					qede_recycle_rx_bd_ring(rxq, edev,
1555 								num_frags);
1556 					dev_kfree_skb_any(skb);
1557 					goto next_cqe;
1558 				}
1559 
1560 				sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS_MAX;
1561 				sw_rx_data = &rxq->sw_rx_ring[sw_rx_index];
1562 				qede_rx_bd_ring_consume(rxq);
1563 
1564 				dma_unmap_page(&edev->pdev->dev,
1565 					       sw_rx_data->mapping,
1566 					       PAGE_SIZE, DMA_FROM_DEVICE);
1567 
1568 				skb_fill_page_desc(skb,
1569 						   skb_shinfo(skb)->nr_frags++,
1570 						   sw_rx_data->data, 0,
1571 						   cur_size);
1572 
1573 				skb->truesize += PAGE_SIZE;
1574 				skb->data_len += cur_size;
1575 				skb->len += cur_size;
1576 				pkt_len -= cur_size;
1577 			}
1578 
1579 			if (unlikely(pkt_len))
1580 				DP_ERR(edev,
1581 				       "Mapped all BDs of jumbo, but still have %d bytes\n",
1582 				       pkt_len);
1583 		}
1584 
1585 		skb->protocol = eth_type_trans(skb, edev->ndev);
1586 
1587 		rx_hash = qede_get_rxhash(edev, fp_cqe->bitfields,
1588 					  fp_cqe->rss_hash, &rxhash_type);
1589 
1590 		skb_set_hash(skb, rx_hash, rxhash_type);
1591 
1592 		qede_set_skb_csum(skb, csum_flag);
1593 
1594 		skb_record_rx_queue(skb, fp->rxq->rxq_id);
1595 
1596 		qede_skb_receive(edev, fp, skb, le16_to_cpu(fp_cqe->vlan_tag));
1597 next_rx_only:
1598 		rx_pkt++;
1599 
1600 next_cqe: /* don't consume bd rx buffer */
1601 		qed_chain_recycle_consumed(&rxq->rx_comp_ring);
1602 		sw_comp_cons = qed_chain_get_cons_idx(&rxq->rx_comp_ring);
1603 		/* CR TPA - revisit how to handle budget in TPA perhaps
1604 		 * increase on "end"
1605 		 */
1606 		if (rx_pkt == budget)
1607 			break;
1608 	} /* repeat while sw_comp_cons != hw_comp_cons... */
1609 
1610 	/* Update producers */
1611 	qede_update_rx_prod(edev, rxq);
1612 
1613 	rxq->rcv_pkts += rx_pkt;
1614 
1615 	return rx_pkt;
1616 }
1617 
1618 static int qede_poll(struct napi_struct *napi, int budget)
1619 {
1620 	struct qede_fastpath *fp = container_of(napi, struct qede_fastpath,
1621 						napi);
1622 	struct qede_dev *edev = fp->edev;
1623 	int rx_work_done = 0;
1624 	u8 tc;
1625 
1626 	for (tc = 0; tc < edev->num_tc; tc++)
1627 		if (likely(fp->type & QEDE_FASTPATH_TX) &&
1628 		    qede_txq_has_work(&fp->txqs[tc]))
1629 			qede_tx_int(edev, &fp->txqs[tc]);
1630 
1631 	rx_work_done = (likely(fp->type & QEDE_FASTPATH_RX) &&
1632 			qede_has_rx_work(fp->rxq)) ?
1633 			qede_rx_int(fp, budget) : 0;
1634 	if (rx_work_done < budget) {
1635 		qed_sb_update_sb_idx(fp->sb_info);
1636 		/* *_has_*_work() reads the status block,
1637 		 * thus we need to ensure that status block indices
1638 		 * have been actually read (qed_sb_update_sb_idx)
1639 		 * prior to this check (*_has_*_work) so that
1640 		 * we won't write the "newer" value of the status block
1641 		 * to HW (if there was a DMA right after
1642 		 * qede_has_rx_work and if there is no rmb, the memory
1643 		 * reading (qed_sb_update_sb_idx) may be postponed
1644 		 * to right before *_ack_sb). In this case there
1645 		 * will never be another interrupt until there is
1646 		 * another update of the status block, while there
1647 		 * is still unhandled work.
1648 		 */
1649 		rmb();
1650 
1651 		/* Fall out from the NAPI loop if needed */
1652 		if (!((likely(fp->type & QEDE_FASTPATH_RX) &&
1653 		       qede_has_rx_work(fp->rxq)) ||
1654 		      (likely(fp->type & QEDE_FASTPATH_TX) &&
1655 		       qede_has_tx_work(fp)))) {
1656 			napi_complete(napi);
1657 
1658 			/* Update and reenable interrupts */
1659 			qed_sb_ack(fp->sb_info, IGU_INT_ENABLE,
1660 				   1 /*update*/);
1661 		} else {
1662 			rx_work_done = budget;
1663 		}
1664 	}
1665 
1666 	return rx_work_done;
1667 }
1668 
1669 static irqreturn_t qede_msix_fp_int(int irq, void *fp_cookie)
1670 {
1671 	struct qede_fastpath *fp = fp_cookie;
1672 
1673 	qed_sb_ack(fp->sb_info, IGU_INT_DISABLE, 0 /*do not update*/);
1674 
1675 	napi_schedule_irqoff(&fp->napi);
1676 	return IRQ_HANDLED;
1677 }
1678 
1679 /* -------------------------------------------------------------------------
1680  * END OF FAST-PATH
1681  * -------------------------------------------------------------------------
1682  */
1683 
1684 static int qede_open(struct net_device *ndev);
1685 static int qede_close(struct net_device *ndev);
1686 static int qede_set_mac_addr(struct net_device *ndev, void *p);
1687 static void qede_set_rx_mode(struct net_device *ndev);
1688 static void qede_config_rx_mode(struct net_device *ndev);
1689 
1690 static int qede_set_ucast_rx_mac(struct qede_dev *edev,
1691 				 enum qed_filter_xcast_params_type opcode,
1692 				 unsigned char mac[ETH_ALEN])
1693 {
1694 	struct qed_filter_params filter_cmd;
1695 
1696 	memset(&filter_cmd, 0, sizeof(filter_cmd));
1697 	filter_cmd.type = QED_FILTER_TYPE_UCAST;
1698 	filter_cmd.filter.ucast.type = opcode;
1699 	filter_cmd.filter.ucast.mac_valid = 1;
1700 	ether_addr_copy(filter_cmd.filter.ucast.mac, mac);
1701 
1702 	return edev->ops->filter_config(edev->cdev, &filter_cmd);
1703 }
1704 
1705 static int qede_set_ucast_rx_vlan(struct qede_dev *edev,
1706 				  enum qed_filter_xcast_params_type opcode,
1707 				  u16 vid)
1708 {
1709 	struct qed_filter_params filter_cmd;
1710 
1711 	memset(&filter_cmd, 0, sizeof(filter_cmd));
1712 	filter_cmd.type = QED_FILTER_TYPE_UCAST;
1713 	filter_cmd.filter.ucast.type = opcode;
1714 	filter_cmd.filter.ucast.vlan_valid = 1;
1715 	filter_cmd.filter.ucast.vlan = vid;
1716 
1717 	return edev->ops->filter_config(edev->cdev, &filter_cmd);
1718 }
1719 
1720 void qede_fill_by_demand_stats(struct qede_dev *edev)
1721 {
1722 	struct qed_eth_stats stats;
1723 
1724 	edev->ops->get_vport_stats(edev->cdev, &stats);
1725 	edev->stats.no_buff_discards = stats.no_buff_discards;
1726 	edev->stats.packet_too_big_discard = stats.packet_too_big_discard;
1727 	edev->stats.ttl0_discard = stats.ttl0_discard;
1728 	edev->stats.rx_ucast_bytes = stats.rx_ucast_bytes;
1729 	edev->stats.rx_mcast_bytes = stats.rx_mcast_bytes;
1730 	edev->stats.rx_bcast_bytes = stats.rx_bcast_bytes;
1731 	edev->stats.rx_ucast_pkts = stats.rx_ucast_pkts;
1732 	edev->stats.rx_mcast_pkts = stats.rx_mcast_pkts;
1733 	edev->stats.rx_bcast_pkts = stats.rx_bcast_pkts;
1734 	edev->stats.mftag_filter_discards = stats.mftag_filter_discards;
1735 	edev->stats.mac_filter_discards = stats.mac_filter_discards;
1736 
1737 	edev->stats.tx_ucast_bytes = stats.tx_ucast_bytes;
1738 	edev->stats.tx_mcast_bytes = stats.tx_mcast_bytes;
1739 	edev->stats.tx_bcast_bytes = stats.tx_bcast_bytes;
1740 	edev->stats.tx_ucast_pkts = stats.tx_ucast_pkts;
1741 	edev->stats.tx_mcast_pkts = stats.tx_mcast_pkts;
1742 	edev->stats.tx_bcast_pkts = stats.tx_bcast_pkts;
1743 	edev->stats.tx_err_drop_pkts = stats.tx_err_drop_pkts;
1744 	edev->stats.coalesced_pkts = stats.tpa_coalesced_pkts;
1745 	edev->stats.coalesced_events = stats.tpa_coalesced_events;
1746 	edev->stats.coalesced_aborts_num = stats.tpa_aborts_num;
1747 	edev->stats.non_coalesced_pkts = stats.tpa_not_coalesced_pkts;
1748 	edev->stats.coalesced_bytes = stats.tpa_coalesced_bytes;
1749 
1750 	edev->stats.rx_64_byte_packets = stats.rx_64_byte_packets;
1751 	edev->stats.rx_65_to_127_byte_packets = stats.rx_65_to_127_byte_packets;
1752 	edev->stats.rx_128_to_255_byte_packets =
1753 				stats.rx_128_to_255_byte_packets;
1754 	edev->stats.rx_256_to_511_byte_packets =
1755 				stats.rx_256_to_511_byte_packets;
1756 	edev->stats.rx_512_to_1023_byte_packets =
1757 				stats.rx_512_to_1023_byte_packets;
1758 	edev->stats.rx_1024_to_1518_byte_packets =
1759 				stats.rx_1024_to_1518_byte_packets;
1760 	edev->stats.rx_1519_to_1522_byte_packets =
1761 				stats.rx_1519_to_1522_byte_packets;
1762 	edev->stats.rx_1519_to_2047_byte_packets =
1763 				stats.rx_1519_to_2047_byte_packets;
1764 	edev->stats.rx_2048_to_4095_byte_packets =
1765 				stats.rx_2048_to_4095_byte_packets;
1766 	edev->stats.rx_4096_to_9216_byte_packets =
1767 				stats.rx_4096_to_9216_byte_packets;
1768 	edev->stats.rx_9217_to_16383_byte_packets =
1769 				stats.rx_9217_to_16383_byte_packets;
1770 	edev->stats.rx_crc_errors = stats.rx_crc_errors;
1771 	edev->stats.rx_mac_crtl_frames = stats.rx_mac_crtl_frames;
1772 	edev->stats.rx_pause_frames = stats.rx_pause_frames;
1773 	edev->stats.rx_pfc_frames = stats.rx_pfc_frames;
1774 	edev->stats.rx_align_errors = stats.rx_align_errors;
1775 	edev->stats.rx_carrier_errors = stats.rx_carrier_errors;
1776 	edev->stats.rx_oversize_packets = stats.rx_oversize_packets;
1777 	edev->stats.rx_jabbers = stats.rx_jabbers;
1778 	edev->stats.rx_undersize_packets = stats.rx_undersize_packets;
1779 	edev->stats.rx_fragments = stats.rx_fragments;
1780 	edev->stats.tx_64_byte_packets = stats.tx_64_byte_packets;
1781 	edev->stats.tx_65_to_127_byte_packets = stats.tx_65_to_127_byte_packets;
1782 	edev->stats.tx_128_to_255_byte_packets =
1783 				stats.tx_128_to_255_byte_packets;
1784 	edev->stats.tx_256_to_511_byte_packets =
1785 				stats.tx_256_to_511_byte_packets;
1786 	edev->stats.tx_512_to_1023_byte_packets =
1787 				stats.tx_512_to_1023_byte_packets;
1788 	edev->stats.tx_1024_to_1518_byte_packets =
1789 				stats.tx_1024_to_1518_byte_packets;
1790 	edev->stats.tx_1519_to_2047_byte_packets =
1791 				stats.tx_1519_to_2047_byte_packets;
1792 	edev->stats.tx_2048_to_4095_byte_packets =
1793 				stats.tx_2048_to_4095_byte_packets;
1794 	edev->stats.tx_4096_to_9216_byte_packets =
1795 				stats.tx_4096_to_9216_byte_packets;
1796 	edev->stats.tx_9217_to_16383_byte_packets =
1797 				stats.tx_9217_to_16383_byte_packets;
1798 	edev->stats.tx_pause_frames = stats.tx_pause_frames;
1799 	edev->stats.tx_pfc_frames = stats.tx_pfc_frames;
1800 	edev->stats.tx_lpi_entry_count = stats.tx_lpi_entry_count;
1801 	edev->stats.tx_total_collisions = stats.tx_total_collisions;
1802 	edev->stats.brb_truncates = stats.brb_truncates;
1803 	edev->stats.brb_discards = stats.brb_discards;
1804 	edev->stats.tx_mac_ctrl_frames = stats.tx_mac_ctrl_frames;
1805 }
1806 
1807 static
1808 struct rtnl_link_stats64 *qede_get_stats64(struct net_device *dev,
1809 					   struct rtnl_link_stats64 *stats)
1810 {
1811 	struct qede_dev *edev = netdev_priv(dev);
1812 
1813 	qede_fill_by_demand_stats(edev);
1814 
1815 	stats->rx_packets = edev->stats.rx_ucast_pkts +
1816 			    edev->stats.rx_mcast_pkts +
1817 			    edev->stats.rx_bcast_pkts;
1818 	stats->tx_packets = edev->stats.tx_ucast_pkts +
1819 			    edev->stats.tx_mcast_pkts +
1820 			    edev->stats.tx_bcast_pkts;
1821 
1822 	stats->rx_bytes = edev->stats.rx_ucast_bytes +
1823 			  edev->stats.rx_mcast_bytes +
1824 			  edev->stats.rx_bcast_bytes;
1825 
1826 	stats->tx_bytes = edev->stats.tx_ucast_bytes +
1827 			  edev->stats.tx_mcast_bytes +
1828 			  edev->stats.tx_bcast_bytes;
1829 
1830 	stats->tx_errors = edev->stats.tx_err_drop_pkts;
1831 	stats->multicast = edev->stats.rx_mcast_pkts +
1832 			   edev->stats.rx_bcast_pkts;
1833 
1834 	stats->rx_fifo_errors = edev->stats.no_buff_discards;
1835 
1836 	stats->collisions = edev->stats.tx_total_collisions;
1837 	stats->rx_crc_errors = edev->stats.rx_crc_errors;
1838 	stats->rx_frame_errors = edev->stats.rx_align_errors;
1839 
1840 	return stats;
1841 }
1842 
1843 #ifdef CONFIG_QED_SRIOV
1844 static int qede_get_vf_config(struct net_device *dev, int vfidx,
1845 			      struct ifla_vf_info *ivi)
1846 {
1847 	struct qede_dev *edev = netdev_priv(dev);
1848 
1849 	if (!edev->ops)
1850 		return -EINVAL;
1851 
1852 	return edev->ops->iov->get_config(edev->cdev, vfidx, ivi);
1853 }
1854 
1855 static int qede_set_vf_rate(struct net_device *dev, int vfidx,
1856 			    int min_tx_rate, int max_tx_rate)
1857 {
1858 	struct qede_dev *edev = netdev_priv(dev);
1859 
1860 	return edev->ops->iov->set_rate(edev->cdev, vfidx, min_tx_rate,
1861 					max_tx_rate);
1862 }
1863 
1864 static int qede_set_vf_spoofchk(struct net_device *dev, int vfidx, bool val)
1865 {
1866 	struct qede_dev *edev = netdev_priv(dev);
1867 
1868 	if (!edev->ops)
1869 		return -EINVAL;
1870 
1871 	return edev->ops->iov->set_spoof(edev->cdev, vfidx, val);
1872 }
1873 
1874 static int qede_set_vf_link_state(struct net_device *dev, int vfidx,
1875 				  int link_state)
1876 {
1877 	struct qede_dev *edev = netdev_priv(dev);
1878 
1879 	if (!edev->ops)
1880 		return -EINVAL;
1881 
1882 	return edev->ops->iov->set_link_state(edev->cdev, vfidx, link_state);
1883 }
1884 #endif
1885 
1886 static void qede_config_accept_any_vlan(struct qede_dev *edev, bool action)
1887 {
1888 	struct qed_update_vport_params params;
1889 	int rc;
1890 
1891 	/* Proceed only if action actually needs to be performed */
1892 	if (edev->accept_any_vlan == action)
1893 		return;
1894 
1895 	memset(&params, 0, sizeof(params));
1896 
1897 	params.vport_id = 0;
1898 	params.accept_any_vlan = action;
1899 	params.update_accept_any_vlan_flg = 1;
1900 
1901 	rc = edev->ops->vport_update(edev->cdev, &params);
1902 	if (rc) {
1903 		DP_ERR(edev, "Failed to %s accept-any-vlan\n",
1904 		       action ? "enable" : "disable");
1905 	} else {
1906 		DP_INFO(edev, "%s accept-any-vlan\n",
1907 			action ? "enabled" : "disabled");
1908 		edev->accept_any_vlan = action;
1909 	}
1910 }
1911 
1912 static int qede_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
1913 {
1914 	struct qede_dev *edev = netdev_priv(dev);
1915 	struct qede_vlan *vlan, *tmp;
1916 	int rc;
1917 
1918 	DP_VERBOSE(edev, NETIF_MSG_IFUP, "Adding vlan 0x%04x\n", vid);
1919 
1920 	vlan = kzalloc(sizeof(*vlan), GFP_KERNEL);
1921 	if (!vlan) {
1922 		DP_INFO(edev, "Failed to allocate struct for vlan\n");
1923 		return -ENOMEM;
1924 	}
1925 	INIT_LIST_HEAD(&vlan->list);
1926 	vlan->vid = vid;
1927 	vlan->configured = false;
1928 
1929 	/* Verify vlan isn't already configured */
1930 	list_for_each_entry(tmp, &edev->vlan_list, list) {
1931 		if (tmp->vid == vlan->vid) {
1932 			DP_VERBOSE(edev, (NETIF_MSG_IFUP | NETIF_MSG_IFDOWN),
1933 				   "vlan already configured\n");
1934 			kfree(vlan);
1935 			return -EEXIST;
1936 		}
1937 	}
1938 
1939 	/* If interface is down, cache this VLAN ID and return */
1940 	if (edev->state != QEDE_STATE_OPEN) {
1941 		DP_VERBOSE(edev, NETIF_MSG_IFDOWN,
1942 			   "Interface is down, VLAN %d will be configured when interface is up\n",
1943 			   vid);
1944 		if (vid != 0)
1945 			edev->non_configured_vlans++;
1946 		list_add(&vlan->list, &edev->vlan_list);
1947 
1948 		return 0;
1949 	}
1950 
1951 	/* Check for the filter limit.
1952 	 * Note - vlan0 has a reserved filter and can be added without
1953 	 * worrying about quota
1954 	 */
1955 	if ((edev->configured_vlans < edev->dev_info.num_vlan_filters) ||
1956 	    (vlan->vid == 0)) {
1957 		rc = qede_set_ucast_rx_vlan(edev,
1958 					    QED_FILTER_XCAST_TYPE_ADD,
1959 					    vlan->vid);
1960 		if (rc) {
1961 			DP_ERR(edev, "Failed to configure VLAN %d\n",
1962 			       vlan->vid);
1963 			kfree(vlan);
1964 			return -EINVAL;
1965 		}
1966 		vlan->configured = true;
1967 
1968 		/* vlan0 filter isn't consuming out of our quota */
1969 		if (vlan->vid != 0)
1970 			edev->configured_vlans++;
1971 	} else {
1972 		/* Out of quota; Activate accept-any-VLAN mode */
1973 		if (!edev->non_configured_vlans)
1974 			qede_config_accept_any_vlan(edev, true);
1975 
1976 		edev->non_configured_vlans++;
1977 	}
1978 
1979 	list_add(&vlan->list, &edev->vlan_list);
1980 
1981 	return 0;
1982 }
1983 
1984 static void qede_del_vlan_from_list(struct qede_dev *edev,
1985 				    struct qede_vlan *vlan)
1986 {
1987 	/* vlan0 filter isn't consuming out of our quota */
1988 	if (vlan->vid != 0) {
1989 		if (vlan->configured)
1990 			edev->configured_vlans--;
1991 		else
1992 			edev->non_configured_vlans--;
1993 	}
1994 
1995 	list_del(&vlan->list);
1996 	kfree(vlan);
1997 }
1998 
1999 static int qede_configure_vlan_filters(struct qede_dev *edev)
2000 {
2001 	int rc = 0, real_rc = 0, accept_any_vlan = 0;
2002 	struct qed_dev_eth_info *dev_info;
2003 	struct qede_vlan *vlan = NULL;
2004 
2005 	if (list_empty(&edev->vlan_list))
2006 		return 0;
2007 
2008 	dev_info = &edev->dev_info;
2009 
2010 	/* Configure non-configured vlans */
2011 	list_for_each_entry(vlan, &edev->vlan_list, list) {
2012 		if (vlan->configured)
2013 			continue;
2014 
2015 		/* We have used all our credits, now enable accept_any_vlan */
2016 		if ((vlan->vid != 0) &&
2017 		    (edev->configured_vlans == dev_info->num_vlan_filters)) {
2018 			accept_any_vlan = 1;
2019 			continue;
2020 		}
2021 
2022 		DP_VERBOSE(edev, NETIF_MSG_IFUP, "Adding vlan %d\n", vlan->vid);
2023 
2024 		rc = qede_set_ucast_rx_vlan(edev, QED_FILTER_XCAST_TYPE_ADD,
2025 					    vlan->vid);
2026 		if (rc) {
2027 			DP_ERR(edev, "Failed to configure VLAN %u\n",
2028 			       vlan->vid);
2029 			real_rc = rc;
2030 			continue;
2031 		}
2032 
2033 		vlan->configured = true;
2034 		/* vlan0 filter doesn't consume our VLAN filter's quota */
2035 		if (vlan->vid != 0) {
2036 			edev->non_configured_vlans--;
2037 			edev->configured_vlans++;
2038 		}
2039 	}
2040 
2041 	/* enable accept_any_vlan mode if we have more VLANs than credits,
2042 	 * or remove accept_any_vlan mode if we've actually removed
2043 	 * a non-configured vlan, and all remaining vlans are truly configured.
2044 	 */
2045 
2046 	if (accept_any_vlan)
2047 		qede_config_accept_any_vlan(edev, true);
2048 	else if (!edev->non_configured_vlans)
2049 		qede_config_accept_any_vlan(edev, false);
2050 
2051 	return real_rc;
2052 }
2053 
2054 static int qede_vlan_rx_kill_vid(struct net_device *dev, __be16 proto, u16 vid)
2055 {
2056 	struct qede_dev *edev = netdev_priv(dev);
2057 	struct qede_vlan *vlan = NULL;
2058 	int rc;
2059 
2060 	DP_VERBOSE(edev, NETIF_MSG_IFDOWN, "Removing vlan 0x%04x\n", vid);
2061 
2062 	/* Find whether entry exists */
2063 	list_for_each_entry(vlan, &edev->vlan_list, list)
2064 		if (vlan->vid == vid)
2065 			break;
2066 
2067 	if (!vlan || (vlan->vid != vid)) {
2068 		DP_VERBOSE(edev, (NETIF_MSG_IFUP | NETIF_MSG_IFDOWN),
2069 			   "Vlan isn't configured\n");
2070 		return 0;
2071 	}
2072 
2073 	if (edev->state != QEDE_STATE_OPEN) {
2074 		/* As interface is already down, we don't have a VPORT
2075 		 * instance to remove vlan filter. So just update vlan list
2076 		 */
2077 		DP_VERBOSE(edev, NETIF_MSG_IFDOWN,
2078 			   "Interface is down, removing VLAN from list only\n");
2079 		qede_del_vlan_from_list(edev, vlan);
2080 		return 0;
2081 	}
2082 
2083 	/* Remove vlan */
2084 	if (vlan->configured) {
2085 		rc = qede_set_ucast_rx_vlan(edev, QED_FILTER_XCAST_TYPE_DEL,
2086 					    vid);
2087 		if (rc) {
2088 			DP_ERR(edev, "Failed to remove VLAN %d\n", vid);
2089 			return -EINVAL;
2090 		}
2091 	}
2092 
2093 	qede_del_vlan_from_list(edev, vlan);
2094 
2095 	/* We have removed a VLAN - try to see if we can
2096 	 * configure non-configured VLAN from the list.
2097 	 */
2098 	rc = qede_configure_vlan_filters(edev);
2099 
2100 	return rc;
2101 }
2102 
2103 static void qede_vlan_mark_nonconfigured(struct qede_dev *edev)
2104 {
2105 	struct qede_vlan *vlan = NULL;
2106 
2107 	if (list_empty(&edev->vlan_list))
2108 		return;
2109 
2110 	list_for_each_entry(vlan, &edev->vlan_list, list) {
2111 		if (!vlan->configured)
2112 			continue;
2113 
2114 		vlan->configured = false;
2115 
2116 		/* vlan0 filter isn't consuming out of our quota */
2117 		if (vlan->vid != 0) {
2118 			edev->non_configured_vlans++;
2119 			edev->configured_vlans--;
2120 		}
2121 
2122 		DP_VERBOSE(edev, NETIF_MSG_IFDOWN,
2123 			   "marked vlan %d as non-configured\n", vlan->vid);
2124 	}
2125 
2126 	edev->accept_any_vlan = false;
2127 }
2128 
2129 static int qede_set_features(struct net_device *dev, netdev_features_t features)
2130 {
2131 	struct qede_dev *edev = netdev_priv(dev);
2132 	netdev_features_t changes = features ^ dev->features;
2133 	bool need_reload = false;
2134 
2135 	/* No action needed if hardware GRO is disabled during driver load */
2136 	if (changes & NETIF_F_GRO) {
2137 		if (dev->features & NETIF_F_GRO)
2138 			need_reload = !edev->gro_disable;
2139 		else
2140 			need_reload = edev->gro_disable;
2141 	}
2142 
2143 	if (need_reload && netif_running(edev->ndev)) {
2144 		dev->features = features;
2145 		qede_reload(edev, NULL, NULL);
2146 		return 1;
2147 	}
2148 
2149 	return 0;
2150 }
2151 
2152 static void qede_udp_tunnel_add(struct net_device *dev,
2153 				struct udp_tunnel_info *ti)
2154 {
2155 	struct qede_dev *edev = netdev_priv(dev);
2156 	u16 t_port = ntohs(ti->port);
2157 
2158 	switch (ti->type) {
2159 	case UDP_TUNNEL_TYPE_VXLAN:
2160 		if (edev->vxlan_dst_port)
2161 			return;
2162 
2163 		edev->vxlan_dst_port = t_port;
2164 
2165 		DP_VERBOSE(edev, QED_MSG_DEBUG, "Added vxlan port=%d\n",
2166 			   t_port);
2167 
2168 		set_bit(QEDE_SP_VXLAN_PORT_CONFIG, &edev->sp_flags);
2169 		break;
2170 	case UDP_TUNNEL_TYPE_GENEVE:
2171 		if (edev->geneve_dst_port)
2172 			return;
2173 
2174 		edev->geneve_dst_port = t_port;
2175 
2176 		DP_VERBOSE(edev, QED_MSG_DEBUG, "Added geneve port=%d\n",
2177 			   t_port);
2178 		set_bit(QEDE_SP_GENEVE_PORT_CONFIG, &edev->sp_flags);
2179 		break;
2180 	default:
2181 		return;
2182 	}
2183 
2184 	schedule_delayed_work(&edev->sp_task, 0);
2185 }
2186 
2187 static void qede_udp_tunnel_del(struct net_device *dev,
2188 				struct udp_tunnel_info *ti)
2189 {
2190 	struct qede_dev *edev = netdev_priv(dev);
2191 	u16 t_port = ntohs(ti->port);
2192 
2193 	switch (ti->type) {
2194 	case UDP_TUNNEL_TYPE_VXLAN:
2195 		if (t_port != edev->vxlan_dst_port)
2196 			return;
2197 
2198 		edev->vxlan_dst_port = 0;
2199 
2200 		DP_VERBOSE(edev, QED_MSG_DEBUG, "Deleted vxlan port=%d\n",
2201 			   t_port);
2202 
2203 		set_bit(QEDE_SP_VXLAN_PORT_CONFIG, &edev->sp_flags);
2204 		break;
2205 	case UDP_TUNNEL_TYPE_GENEVE:
2206 		if (t_port != edev->geneve_dst_port)
2207 			return;
2208 
2209 		edev->geneve_dst_port = 0;
2210 
2211 		DP_VERBOSE(edev, QED_MSG_DEBUG, "Deleted geneve port=%d\n",
2212 			   t_port);
2213 		set_bit(QEDE_SP_GENEVE_PORT_CONFIG, &edev->sp_flags);
2214 		break;
2215 	default:
2216 		return;
2217 	}
2218 
2219 	schedule_delayed_work(&edev->sp_task, 0);
2220 }
2221 
2222 static const struct net_device_ops qede_netdev_ops = {
2223 	.ndo_open = qede_open,
2224 	.ndo_stop = qede_close,
2225 	.ndo_start_xmit = qede_start_xmit,
2226 	.ndo_set_rx_mode = qede_set_rx_mode,
2227 	.ndo_set_mac_address = qede_set_mac_addr,
2228 	.ndo_validate_addr = eth_validate_addr,
2229 	.ndo_change_mtu = qede_change_mtu,
2230 #ifdef CONFIG_QED_SRIOV
2231 	.ndo_set_vf_mac = qede_set_vf_mac,
2232 	.ndo_set_vf_vlan = qede_set_vf_vlan,
2233 #endif
2234 	.ndo_vlan_rx_add_vid = qede_vlan_rx_add_vid,
2235 	.ndo_vlan_rx_kill_vid = qede_vlan_rx_kill_vid,
2236 	.ndo_set_features = qede_set_features,
2237 	.ndo_get_stats64 = qede_get_stats64,
2238 #ifdef CONFIG_QED_SRIOV
2239 	.ndo_set_vf_link_state = qede_set_vf_link_state,
2240 	.ndo_set_vf_spoofchk = qede_set_vf_spoofchk,
2241 	.ndo_get_vf_config = qede_get_vf_config,
2242 	.ndo_set_vf_rate = qede_set_vf_rate,
2243 #endif
2244 	.ndo_udp_tunnel_add = qede_udp_tunnel_add,
2245 	.ndo_udp_tunnel_del = qede_udp_tunnel_del,
2246 };
2247 
2248 /* -------------------------------------------------------------------------
2249  * START OF PROBE / REMOVE
2250  * -------------------------------------------------------------------------
2251  */
2252 
2253 static struct qede_dev *qede_alloc_etherdev(struct qed_dev *cdev,
2254 					    struct pci_dev *pdev,
2255 					    struct qed_dev_eth_info *info,
2256 					    u32 dp_module, u8 dp_level)
2257 {
2258 	struct net_device *ndev;
2259 	struct qede_dev *edev;
2260 
2261 	ndev = alloc_etherdev_mqs(sizeof(*edev),
2262 				  info->num_queues, info->num_queues);
2263 	if (!ndev) {
2264 		pr_err("etherdev allocation failed\n");
2265 		return NULL;
2266 	}
2267 
2268 	edev = netdev_priv(ndev);
2269 	edev->ndev = ndev;
2270 	edev->cdev = cdev;
2271 	edev->pdev = pdev;
2272 	edev->dp_module = dp_module;
2273 	edev->dp_level = dp_level;
2274 	edev->ops = qed_ops;
2275 	edev->q_num_rx_buffers = NUM_RX_BDS_DEF;
2276 	edev->q_num_tx_buffers = NUM_TX_BDS_DEF;
2277 
2278 	DP_INFO(edev, "Allocated netdev with %d tx queues and %d rx queues\n",
2279 		info->num_queues, info->num_queues);
2280 
2281 	SET_NETDEV_DEV(ndev, &pdev->dev);
2282 
2283 	memset(&edev->stats, 0, sizeof(edev->stats));
2284 	memcpy(&edev->dev_info, info, sizeof(*info));
2285 
2286 	edev->num_tc = edev->dev_info.num_tc;
2287 
2288 	INIT_LIST_HEAD(&edev->vlan_list);
2289 
2290 	return edev;
2291 }
2292 
2293 static void qede_init_ndev(struct qede_dev *edev)
2294 {
2295 	struct net_device *ndev = edev->ndev;
2296 	struct pci_dev *pdev = edev->pdev;
2297 	u32 hw_features;
2298 
2299 	pci_set_drvdata(pdev, ndev);
2300 
2301 	ndev->mem_start = edev->dev_info.common.pci_mem_start;
2302 	ndev->base_addr = ndev->mem_start;
2303 	ndev->mem_end = edev->dev_info.common.pci_mem_end;
2304 	ndev->irq = edev->dev_info.common.pci_irq;
2305 
2306 	ndev->watchdog_timeo = TX_TIMEOUT;
2307 
2308 	ndev->netdev_ops = &qede_netdev_ops;
2309 
2310 	qede_set_ethtool_ops(ndev);
2311 
2312 	/* user-changeble features */
2313 	hw_features = NETIF_F_GRO | NETIF_F_SG |
2314 		      NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2315 		      NETIF_F_TSO | NETIF_F_TSO6;
2316 
2317 	/* Encap features*/
2318 	hw_features |= NETIF_F_GSO_GRE | NETIF_F_GSO_UDP_TUNNEL |
2319 		       NETIF_F_TSO_ECN;
2320 	ndev->hw_enc_features = NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2321 				NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO_ECN |
2322 				NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2323 				NETIF_F_GSO_UDP_TUNNEL | NETIF_F_RXCSUM;
2324 
2325 	ndev->vlan_features = hw_features | NETIF_F_RXHASH | NETIF_F_RXCSUM |
2326 			      NETIF_F_HIGHDMA;
2327 	ndev->features = hw_features | NETIF_F_RXHASH | NETIF_F_RXCSUM |
2328 			 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HIGHDMA |
2329 			 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_CTAG_TX;
2330 
2331 	ndev->hw_features = hw_features;
2332 
2333 	/* Set network device HW mac */
2334 	ether_addr_copy(edev->ndev->dev_addr, edev->dev_info.common.hw_mac);
2335 }
2336 
2337 /* This function converts from 32b param to two params of level and module
2338  * Input 32b decoding:
2339  * b31 - enable all NOTICE prints. NOTICE prints are for deviation from the
2340  * 'happy' flow, e.g. memory allocation failed.
2341  * b30 - enable all INFO prints. INFO prints are for major steps in the flow
2342  * and provide important parameters.
2343  * b29-b0 - per-module bitmap, where each bit enables VERBOSE prints of that
2344  * module. VERBOSE prints are for tracking the specific flow in low level.
2345  *
2346  * Notice that the level should be that of the lowest required logs.
2347  */
2348 void qede_config_debug(uint debug, u32 *p_dp_module, u8 *p_dp_level)
2349 {
2350 	*p_dp_level = QED_LEVEL_NOTICE;
2351 	*p_dp_module = 0;
2352 
2353 	if (debug & QED_LOG_VERBOSE_MASK) {
2354 		*p_dp_level = QED_LEVEL_VERBOSE;
2355 		*p_dp_module = (debug & 0x3FFFFFFF);
2356 	} else if (debug & QED_LOG_INFO_MASK) {
2357 		*p_dp_level = QED_LEVEL_INFO;
2358 	} else if (debug & QED_LOG_NOTICE_MASK) {
2359 		*p_dp_level = QED_LEVEL_NOTICE;
2360 	}
2361 }
2362 
2363 static void qede_free_fp_array(struct qede_dev *edev)
2364 {
2365 	if (edev->fp_array) {
2366 		struct qede_fastpath *fp;
2367 		int i;
2368 
2369 		for_each_queue(i) {
2370 			fp = &edev->fp_array[i];
2371 
2372 			kfree(fp->sb_info);
2373 			kfree(fp->rxq);
2374 			kfree(fp->txqs);
2375 		}
2376 		kfree(edev->fp_array);
2377 	}
2378 
2379 	edev->num_queues = 0;
2380 	edev->fp_num_tx = 0;
2381 	edev->fp_num_rx = 0;
2382 }
2383 
2384 static int qede_alloc_fp_array(struct qede_dev *edev)
2385 {
2386 	u8 fp_combined, fp_rx = edev->fp_num_rx;
2387 	struct qede_fastpath *fp;
2388 	int i;
2389 
2390 	edev->fp_array = kcalloc(QEDE_QUEUE_CNT(edev),
2391 				 sizeof(*edev->fp_array), GFP_KERNEL);
2392 	if (!edev->fp_array) {
2393 		DP_NOTICE(edev, "fp array allocation failed\n");
2394 		goto err;
2395 	}
2396 
2397 	fp_combined = QEDE_QUEUE_CNT(edev) - fp_rx - edev->fp_num_tx;
2398 
2399 	/* Allocate the FP elements for Rx queues followed by combined and then
2400 	 * the Tx. This ordering should be maintained so that the respective
2401 	 * queues (Rx or Tx) will be together in the fastpath array and the
2402 	 * associated ids will be sequential.
2403 	 */
2404 	for_each_queue(i) {
2405 		fp = &edev->fp_array[i];
2406 
2407 		fp->sb_info = kcalloc(1, sizeof(*fp->sb_info), GFP_KERNEL);
2408 		if (!fp->sb_info) {
2409 			DP_NOTICE(edev, "sb info struct allocation failed\n");
2410 			goto err;
2411 		}
2412 
2413 		if (fp_rx) {
2414 			fp->type = QEDE_FASTPATH_RX;
2415 			fp_rx--;
2416 		} else if (fp_combined) {
2417 			fp->type = QEDE_FASTPATH_COMBINED;
2418 			fp_combined--;
2419 		} else {
2420 			fp->type = QEDE_FASTPATH_TX;
2421 		}
2422 
2423 		if (fp->type & QEDE_FASTPATH_TX) {
2424 			fp->txqs = kcalloc(edev->num_tc, sizeof(*fp->txqs),
2425 					   GFP_KERNEL);
2426 			if (!fp->txqs) {
2427 				DP_NOTICE(edev,
2428 					  "TXQ array allocation failed\n");
2429 				goto err;
2430 			}
2431 		}
2432 
2433 		if (fp->type & QEDE_FASTPATH_RX) {
2434 			fp->rxq = kcalloc(1, sizeof(*fp->rxq), GFP_KERNEL);
2435 			if (!fp->rxq) {
2436 				DP_NOTICE(edev,
2437 					  "RXQ struct allocation failed\n");
2438 				goto err;
2439 			}
2440 		}
2441 	}
2442 
2443 	return 0;
2444 err:
2445 	qede_free_fp_array(edev);
2446 	return -ENOMEM;
2447 }
2448 
2449 static void qede_sp_task(struct work_struct *work)
2450 {
2451 	struct qede_dev *edev = container_of(work, struct qede_dev,
2452 					     sp_task.work);
2453 	struct qed_dev *cdev = edev->cdev;
2454 
2455 	mutex_lock(&edev->qede_lock);
2456 
2457 	if (edev->state == QEDE_STATE_OPEN) {
2458 		if (test_and_clear_bit(QEDE_SP_RX_MODE, &edev->sp_flags))
2459 			qede_config_rx_mode(edev->ndev);
2460 	}
2461 
2462 	if (test_and_clear_bit(QEDE_SP_VXLAN_PORT_CONFIG, &edev->sp_flags)) {
2463 		struct qed_tunn_params tunn_params;
2464 
2465 		memset(&tunn_params, 0, sizeof(tunn_params));
2466 		tunn_params.update_vxlan_port = 1;
2467 		tunn_params.vxlan_port = edev->vxlan_dst_port;
2468 		qed_ops->tunn_config(cdev, &tunn_params);
2469 	}
2470 
2471 	if (test_and_clear_bit(QEDE_SP_GENEVE_PORT_CONFIG, &edev->sp_flags)) {
2472 		struct qed_tunn_params tunn_params;
2473 
2474 		memset(&tunn_params, 0, sizeof(tunn_params));
2475 		tunn_params.update_geneve_port = 1;
2476 		tunn_params.geneve_port = edev->geneve_dst_port;
2477 		qed_ops->tunn_config(cdev, &tunn_params);
2478 	}
2479 
2480 	mutex_unlock(&edev->qede_lock);
2481 }
2482 
2483 static void qede_update_pf_params(struct qed_dev *cdev)
2484 {
2485 	struct qed_pf_params pf_params;
2486 
2487 	/* 64 rx + 64 tx */
2488 	memset(&pf_params, 0, sizeof(struct qed_pf_params));
2489 	pf_params.eth_pf_params.num_cons = 128;
2490 	qed_ops->common->update_pf_params(cdev, &pf_params);
2491 }
2492 
2493 enum qede_probe_mode {
2494 	QEDE_PROBE_NORMAL,
2495 };
2496 
2497 static int __qede_probe(struct pci_dev *pdev, u32 dp_module, u8 dp_level,
2498 			bool is_vf, enum qede_probe_mode mode)
2499 {
2500 	struct qed_probe_params probe_params;
2501 	struct qed_slowpath_params sp_params;
2502 	struct qed_dev_eth_info dev_info;
2503 	struct qede_dev *edev;
2504 	struct qed_dev *cdev;
2505 	int rc;
2506 
2507 	if (unlikely(dp_level & QED_LEVEL_INFO))
2508 		pr_notice("Starting qede probe\n");
2509 
2510 	memset(&probe_params, 0, sizeof(probe_params));
2511 	probe_params.protocol = QED_PROTOCOL_ETH;
2512 	probe_params.dp_module = dp_module;
2513 	probe_params.dp_level = dp_level;
2514 	probe_params.is_vf = is_vf;
2515 	cdev = qed_ops->common->probe(pdev, &probe_params);
2516 	if (!cdev) {
2517 		rc = -ENODEV;
2518 		goto err0;
2519 	}
2520 
2521 	qede_update_pf_params(cdev);
2522 
2523 	/* Start the Slowpath-process */
2524 	memset(&sp_params, 0, sizeof(sp_params));
2525 	sp_params.int_mode = QED_INT_MODE_MSIX;
2526 	sp_params.drv_major = QEDE_MAJOR_VERSION;
2527 	sp_params.drv_minor = QEDE_MINOR_VERSION;
2528 	sp_params.drv_rev = QEDE_REVISION_VERSION;
2529 	sp_params.drv_eng = QEDE_ENGINEERING_VERSION;
2530 	strlcpy(sp_params.name, "qede LAN", QED_DRV_VER_STR_SIZE);
2531 	rc = qed_ops->common->slowpath_start(cdev, &sp_params);
2532 	if (rc) {
2533 		pr_notice("Cannot start slowpath\n");
2534 		goto err1;
2535 	}
2536 
2537 	/* Learn information crucial for qede to progress */
2538 	rc = qed_ops->fill_dev_info(cdev, &dev_info);
2539 	if (rc)
2540 		goto err2;
2541 
2542 	edev = qede_alloc_etherdev(cdev, pdev, &dev_info, dp_module,
2543 				   dp_level);
2544 	if (!edev) {
2545 		rc = -ENOMEM;
2546 		goto err2;
2547 	}
2548 
2549 	if (is_vf)
2550 		edev->flags |= QEDE_FLAG_IS_VF;
2551 
2552 	qede_init_ndev(edev);
2553 
2554 	rc = qede_roce_dev_add(edev);
2555 	if (rc)
2556 		goto err3;
2557 
2558 	rc = register_netdev(edev->ndev);
2559 	if (rc) {
2560 		DP_NOTICE(edev, "Cannot register net-device\n");
2561 		goto err4;
2562 	}
2563 
2564 	edev->ops->common->set_id(cdev, edev->ndev->name, DRV_MODULE_VERSION);
2565 
2566 	edev->ops->register_ops(cdev, &qede_ll_ops, edev);
2567 
2568 #ifdef CONFIG_DCB
2569 	if (!IS_VF(edev))
2570 		qede_set_dcbnl_ops(edev->ndev);
2571 #endif
2572 
2573 	INIT_DELAYED_WORK(&edev->sp_task, qede_sp_task);
2574 	mutex_init(&edev->qede_lock);
2575 	edev->rx_copybreak = QEDE_RX_HDR_SIZE;
2576 
2577 	DP_INFO(edev, "Ending successfully qede probe\n");
2578 
2579 	return 0;
2580 
2581 err4:
2582 	qede_roce_dev_remove(edev);
2583 err3:
2584 	free_netdev(edev->ndev);
2585 err2:
2586 	qed_ops->common->slowpath_stop(cdev);
2587 err1:
2588 	qed_ops->common->remove(cdev);
2589 err0:
2590 	return rc;
2591 }
2592 
2593 static int qede_probe(struct pci_dev *pdev, const struct pci_device_id *id)
2594 {
2595 	bool is_vf = false;
2596 	u32 dp_module = 0;
2597 	u8 dp_level = 0;
2598 
2599 	switch ((enum qede_pci_private)id->driver_data) {
2600 	case QEDE_PRIVATE_VF:
2601 		if (debug & QED_LOG_VERBOSE_MASK)
2602 			dev_err(&pdev->dev, "Probing a VF\n");
2603 		is_vf = true;
2604 		break;
2605 	default:
2606 		if (debug & QED_LOG_VERBOSE_MASK)
2607 			dev_err(&pdev->dev, "Probing a PF\n");
2608 	}
2609 
2610 	qede_config_debug(debug, &dp_module, &dp_level);
2611 
2612 	return __qede_probe(pdev, dp_module, dp_level, is_vf,
2613 			    QEDE_PROBE_NORMAL);
2614 }
2615 
2616 enum qede_remove_mode {
2617 	QEDE_REMOVE_NORMAL,
2618 };
2619 
2620 static void __qede_remove(struct pci_dev *pdev, enum qede_remove_mode mode)
2621 {
2622 	struct net_device *ndev = pci_get_drvdata(pdev);
2623 	struct qede_dev *edev = netdev_priv(ndev);
2624 	struct qed_dev *cdev = edev->cdev;
2625 
2626 	DP_INFO(edev, "Starting qede_remove\n");
2627 
2628 	cancel_delayed_work_sync(&edev->sp_task);
2629 
2630 	unregister_netdev(ndev);
2631 
2632 	qede_roce_dev_remove(edev);
2633 
2634 	edev->ops->common->set_power_state(cdev, PCI_D0);
2635 
2636 	pci_set_drvdata(pdev, NULL);
2637 
2638 	free_netdev(ndev);
2639 
2640 	/* Use global ops since we've freed edev */
2641 	qed_ops->common->slowpath_stop(cdev);
2642 	qed_ops->common->remove(cdev);
2643 
2644 	dev_info(&pdev->dev, "Ending qede_remove successfully\n");
2645 }
2646 
2647 static void qede_remove(struct pci_dev *pdev)
2648 {
2649 	__qede_remove(pdev, QEDE_REMOVE_NORMAL);
2650 }
2651 
2652 /* -------------------------------------------------------------------------
2653  * START OF LOAD / UNLOAD
2654  * -------------------------------------------------------------------------
2655  */
2656 
2657 static int qede_set_num_queues(struct qede_dev *edev)
2658 {
2659 	int rc;
2660 	u16 rss_num;
2661 
2662 	/* Setup queues according to possible resources*/
2663 	if (edev->req_queues)
2664 		rss_num = edev->req_queues;
2665 	else
2666 		rss_num = netif_get_num_default_rss_queues() *
2667 			  edev->dev_info.common.num_hwfns;
2668 
2669 	rss_num = min_t(u16, QEDE_MAX_RSS_CNT(edev), rss_num);
2670 
2671 	rc = edev->ops->common->set_fp_int(edev->cdev, rss_num);
2672 	if (rc > 0) {
2673 		/* Managed to request interrupts for our queues */
2674 		edev->num_queues = rc;
2675 		DP_INFO(edev, "Managed %d [of %d] RSS queues\n",
2676 			QEDE_QUEUE_CNT(edev), rss_num);
2677 		rc = 0;
2678 	}
2679 
2680 	edev->fp_num_tx = edev->req_num_tx;
2681 	edev->fp_num_rx = edev->req_num_rx;
2682 
2683 	return rc;
2684 }
2685 
2686 static void qede_free_mem_sb(struct qede_dev *edev,
2687 			     struct qed_sb_info *sb_info)
2688 {
2689 	if (sb_info->sb_virt)
2690 		dma_free_coherent(&edev->pdev->dev, sizeof(*sb_info->sb_virt),
2691 				  (void *)sb_info->sb_virt, sb_info->sb_phys);
2692 }
2693 
2694 /* This function allocates fast-path status block memory */
2695 static int qede_alloc_mem_sb(struct qede_dev *edev,
2696 			     struct qed_sb_info *sb_info, u16 sb_id)
2697 {
2698 	struct status_block *sb_virt;
2699 	dma_addr_t sb_phys;
2700 	int rc;
2701 
2702 	sb_virt = dma_alloc_coherent(&edev->pdev->dev,
2703 				     sizeof(*sb_virt), &sb_phys, GFP_KERNEL);
2704 	if (!sb_virt) {
2705 		DP_ERR(edev, "Status block allocation failed\n");
2706 		return -ENOMEM;
2707 	}
2708 
2709 	rc = edev->ops->common->sb_init(edev->cdev, sb_info,
2710 					sb_virt, sb_phys, sb_id,
2711 					QED_SB_TYPE_L2_QUEUE);
2712 	if (rc) {
2713 		DP_ERR(edev, "Status block initialization failed\n");
2714 		dma_free_coherent(&edev->pdev->dev, sizeof(*sb_virt),
2715 				  sb_virt, sb_phys);
2716 		return rc;
2717 	}
2718 
2719 	return 0;
2720 }
2721 
2722 static void qede_free_rx_buffers(struct qede_dev *edev,
2723 				 struct qede_rx_queue *rxq)
2724 {
2725 	u16 i;
2726 
2727 	for (i = rxq->sw_rx_cons; i != rxq->sw_rx_prod; i++) {
2728 		struct sw_rx_data *rx_buf;
2729 		struct page *data;
2730 
2731 		rx_buf = &rxq->sw_rx_ring[i & NUM_RX_BDS_MAX];
2732 		data = rx_buf->data;
2733 
2734 		dma_unmap_page(&edev->pdev->dev,
2735 			       rx_buf->mapping, PAGE_SIZE, DMA_FROM_DEVICE);
2736 
2737 		rx_buf->data = NULL;
2738 		__free_page(data);
2739 	}
2740 }
2741 
2742 static void qede_free_sge_mem(struct qede_dev *edev, struct qede_rx_queue *rxq)
2743 {
2744 	int i;
2745 
2746 	if (edev->gro_disable)
2747 		return;
2748 
2749 	for (i = 0; i < ETH_TPA_MAX_AGGS_NUM; i++) {
2750 		struct qede_agg_info *tpa_info = &rxq->tpa_info[i];
2751 		struct sw_rx_data *replace_buf = &tpa_info->replace_buf;
2752 
2753 		if (replace_buf->data) {
2754 			dma_unmap_page(&edev->pdev->dev,
2755 				       replace_buf->mapping,
2756 				       PAGE_SIZE, DMA_FROM_DEVICE);
2757 			__free_page(replace_buf->data);
2758 		}
2759 	}
2760 }
2761 
2762 static void qede_free_mem_rxq(struct qede_dev *edev, struct qede_rx_queue *rxq)
2763 {
2764 	qede_free_sge_mem(edev, rxq);
2765 
2766 	/* Free rx buffers */
2767 	qede_free_rx_buffers(edev, rxq);
2768 
2769 	/* Free the parallel SW ring */
2770 	kfree(rxq->sw_rx_ring);
2771 
2772 	/* Free the real RQ ring used by FW */
2773 	edev->ops->common->chain_free(edev->cdev, &rxq->rx_bd_ring);
2774 	edev->ops->common->chain_free(edev->cdev, &rxq->rx_comp_ring);
2775 }
2776 
2777 static int qede_alloc_rx_buffer(struct qede_dev *edev,
2778 				struct qede_rx_queue *rxq)
2779 {
2780 	struct sw_rx_data *sw_rx_data;
2781 	struct eth_rx_bd *rx_bd;
2782 	dma_addr_t mapping;
2783 	struct page *data;
2784 
2785 	data = alloc_pages(GFP_ATOMIC, 0);
2786 	if (unlikely(!data)) {
2787 		DP_NOTICE(edev, "Failed to allocate Rx data [page]\n");
2788 		return -ENOMEM;
2789 	}
2790 
2791 	/* Map the entire page as it would be used
2792 	 * for multiple RX buffer segment size mapping.
2793 	 */
2794 	mapping = dma_map_page(&edev->pdev->dev, data, 0,
2795 			       PAGE_SIZE, DMA_FROM_DEVICE);
2796 	if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
2797 		__free_page(data);
2798 		DP_NOTICE(edev, "Failed to map Rx buffer\n");
2799 		return -ENOMEM;
2800 	}
2801 
2802 	sw_rx_data = &rxq->sw_rx_ring[rxq->sw_rx_prod & NUM_RX_BDS_MAX];
2803 	sw_rx_data->page_offset = 0;
2804 	sw_rx_data->data = data;
2805 	sw_rx_data->mapping = mapping;
2806 
2807 	/* Advance PROD and get BD pointer */
2808 	rx_bd = (struct eth_rx_bd *)qed_chain_produce(&rxq->rx_bd_ring);
2809 	WARN_ON(!rx_bd);
2810 	rx_bd->addr.hi = cpu_to_le32(upper_32_bits(mapping));
2811 	rx_bd->addr.lo = cpu_to_le32(lower_32_bits(mapping));
2812 
2813 	rxq->sw_rx_prod++;
2814 
2815 	return 0;
2816 }
2817 
2818 static int qede_alloc_sge_mem(struct qede_dev *edev, struct qede_rx_queue *rxq)
2819 {
2820 	dma_addr_t mapping;
2821 	int i;
2822 
2823 	if (edev->gro_disable)
2824 		return 0;
2825 
2826 	if (edev->ndev->mtu > PAGE_SIZE) {
2827 		edev->gro_disable = 1;
2828 		return 0;
2829 	}
2830 
2831 	for (i = 0; i < ETH_TPA_MAX_AGGS_NUM; i++) {
2832 		struct qede_agg_info *tpa_info = &rxq->tpa_info[i];
2833 		struct sw_rx_data *replace_buf = &tpa_info->replace_buf;
2834 
2835 		replace_buf->data = alloc_pages(GFP_ATOMIC, 0);
2836 		if (unlikely(!replace_buf->data)) {
2837 			DP_NOTICE(edev,
2838 				  "Failed to allocate TPA skb pool [replacement buffer]\n");
2839 			goto err;
2840 		}
2841 
2842 		mapping = dma_map_page(&edev->pdev->dev, replace_buf->data, 0,
2843 				       rxq->rx_buf_size, DMA_FROM_DEVICE);
2844 		if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
2845 			DP_NOTICE(edev,
2846 				  "Failed to map TPA replacement buffer\n");
2847 			goto err;
2848 		}
2849 
2850 		replace_buf->mapping = mapping;
2851 		tpa_info->replace_buf.page_offset = 0;
2852 
2853 		tpa_info->replace_buf_mapping = mapping;
2854 		tpa_info->agg_state = QEDE_AGG_STATE_NONE;
2855 	}
2856 
2857 	return 0;
2858 err:
2859 	qede_free_sge_mem(edev, rxq);
2860 	edev->gro_disable = 1;
2861 	return -ENOMEM;
2862 }
2863 
2864 /* This function allocates all memory needed per Rx queue */
2865 static int qede_alloc_mem_rxq(struct qede_dev *edev, struct qede_rx_queue *rxq)
2866 {
2867 	int i, rc, size;
2868 
2869 	rxq->num_rx_buffers = edev->q_num_rx_buffers;
2870 
2871 	rxq->rx_buf_size = NET_IP_ALIGN + ETH_OVERHEAD + edev->ndev->mtu;
2872 
2873 	if (rxq->rx_buf_size > PAGE_SIZE)
2874 		rxq->rx_buf_size = PAGE_SIZE;
2875 
2876 	/* Segment size to spilt a page in multiple equal parts */
2877 	rxq->rx_buf_seg_size = roundup_pow_of_two(rxq->rx_buf_size);
2878 
2879 	/* Allocate the parallel driver ring for Rx buffers */
2880 	size = sizeof(*rxq->sw_rx_ring) * RX_RING_SIZE;
2881 	rxq->sw_rx_ring = kzalloc(size, GFP_KERNEL);
2882 	if (!rxq->sw_rx_ring) {
2883 		DP_ERR(edev, "Rx buffers ring allocation failed\n");
2884 		rc = -ENOMEM;
2885 		goto err;
2886 	}
2887 
2888 	/* Allocate FW Rx ring  */
2889 	rc = edev->ops->common->chain_alloc(edev->cdev,
2890 					    QED_CHAIN_USE_TO_CONSUME_PRODUCE,
2891 					    QED_CHAIN_MODE_NEXT_PTR,
2892 					    QED_CHAIN_CNT_TYPE_U16,
2893 					    RX_RING_SIZE,
2894 					    sizeof(struct eth_rx_bd),
2895 					    &rxq->rx_bd_ring);
2896 
2897 	if (rc)
2898 		goto err;
2899 
2900 	/* Allocate FW completion ring */
2901 	rc = edev->ops->common->chain_alloc(edev->cdev,
2902 					    QED_CHAIN_USE_TO_CONSUME,
2903 					    QED_CHAIN_MODE_PBL,
2904 					    QED_CHAIN_CNT_TYPE_U16,
2905 					    RX_RING_SIZE,
2906 					    sizeof(union eth_rx_cqe),
2907 					    &rxq->rx_comp_ring);
2908 	if (rc)
2909 		goto err;
2910 
2911 	/* Allocate buffers for the Rx ring */
2912 	for (i = 0; i < rxq->num_rx_buffers; i++) {
2913 		rc = qede_alloc_rx_buffer(edev, rxq);
2914 		if (rc) {
2915 			DP_ERR(edev,
2916 			       "Rx buffers allocation failed at index %d\n", i);
2917 			goto err;
2918 		}
2919 	}
2920 
2921 	rc = qede_alloc_sge_mem(edev, rxq);
2922 err:
2923 	return rc;
2924 }
2925 
2926 static void qede_free_mem_txq(struct qede_dev *edev, struct qede_tx_queue *txq)
2927 {
2928 	/* Free the parallel SW ring */
2929 	kfree(txq->sw_tx_ring);
2930 
2931 	/* Free the real RQ ring used by FW */
2932 	edev->ops->common->chain_free(edev->cdev, &txq->tx_pbl);
2933 }
2934 
2935 /* This function allocates all memory needed per Tx queue */
2936 static int qede_alloc_mem_txq(struct qede_dev *edev, struct qede_tx_queue *txq)
2937 {
2938 	int size, rc;
2939 	union eth_tx_bd_types *p_virt;
2940 
2941 	txq->num_tx_buffers = edev->q_num_tx_buffers;
2942 
2943 	/* Allocate the parallel driver ring for Tx buffers */
2944 	size = sizeof(*txq->sw_tx_ring) * NUM_TX_BDS_MAX;
2945 	txq->sw_tx_ring = kzalloc(size, GFP_KERNEL);
2946 	if (!txq->sw_tx_ring) {
2947 		DP_NOTICE(edev, "Tx buffers ring allocation failed\n");
2948 		goto err;
2949 	}
2950 
2951 	rc = edev->ops->common->chain_alloc(edev->cdev,
2952 					    QED_CHAIN_USE_TO_CONSUME_PRODUCE,
2953 					    QED_CHAIN_MODE_PBL,
2954 					    QED_CHAIN_CNT_TYPE_U16,
2955 					    NUM_TX_BDS_MAX,
2956 					    sizeof(*p_virt), &txq->tx_pbl);
2957 	if (rc)
2958 		goto err;
2959 
2960 	return 0;
2961 
2962 err:
2963 	qede_free_mem_txq(edev, txq);
2964 	return -ENOMEM;
2965 }
2966 
2967 /* This function frees all memory of a single fp */
2968 static void qede_free_mem_fp(struct qede_dev *edev, struct qede_fastpath *fp)
2969 {
2970 	int tc;
2971 
2972 	qede_free_mem_sb(edev, fp->sb_info);
2973 
2974 	if (fp->type & QEDE_FASTPATH_RX)
2975 		qede_free_mem_rxq(edev, fp->rxq);
2976 
2977 	if (fp->type & QEDE_FASTPATH_TX)
2978 		for (tc = 0; tc < edev->num_tc; tc++)
2979 			qede_free_mem_txq(edev, &fp->txqs[tc]);
2980 }
2981 
2982 /* This function allocates all memory needed for a single fp (i.e. an entity
2983  * which contains status block, one rx queue and/or multiple per-TC tx queues.
2984  */
2985 static int qede_alloc_mem_fp(struct qede_dev *edev, struct qede_fastpath *fp)
2986 {
2987 	int rc, tc;
2988 
2989 	rc = qede_alloc_mem_sb(edev, fp->sb_info, fp->id);
2990 	if (rc)
2991 		goto err;
2992 
2993 	if (fp->type & QEDE_FASTPATH_RX) {
2994 		rc = qede_alloc_mem_rxq(edev, fp->rxq);
2995 		if (rc)
2996 			goto err;
2997 	}
2998 
2999 	if (fp->type & QEDE_FASTPATH_TX) {
3000 		for (tc = 0; tc < edev->num_tc; tc++) {
3001 			rc = qede_alloc_mem_txq(edev, &fp->txqs[tc]);
3002 			if (rc)
3003 				goto err;
3004 		}
3005 	}
3006 
3007 	return 0;
3008 err:
3009 	return rc;
3010 }
3011 
3012 static void qede_free_mem_load(struct qede_dev *edev)
3013 {
3014 	int i;
3015 
3016 	for_each_queue(i) {
3017 		struct qede_fastpath *fp = &edev->fp_array[i];
3018 
3019 		qede_free_mem_fp(edev, fp);
3020 	}
3021 }
3022 
3023 /* This function allocates all qede memory at NIC load. */
3024 static int qede_alloc_mem_load(struct qede_dev *edev)
3025 {
3026 	int rc = 0, queue_id;
3027 
3028 	for (queue_id = 0; queue_id < QEDE_QUEUE_CNT(edev); queue_id++) {
3029 		struct qede_fastpath *fp = &edev->fp_array[queue_id];
3030 
3031 		rc = qede_alloc_mem_fp(edev, fp);
3032 		if (rc) {
3033 			DP_ERR(edev,
3034 			       "Failed to allocate memory for fastpath - rss id = %d\n",
3035 			       queue_id);
3036 			qede_free_mem_load(edev);
3037 			return rc;
3038 		}
3039 	}
3040 
3041 	return 0;
3042 }
3043 
3044 /* This function inits fp content and resets the SB, RXQ and TXQ structures */
3045 static void qede_init_fp(struct qede_dev *edev)
3046 {
3047 	int queue_id, rxq_index = 0, txq_index = 0, tc;
3048 	struct qede_fastpath *fp;
3049 
3050 	for_each_queue(queue_id) {
3051 		fp = &edev->fp_array[queue_id];
3052 
3053 		fp->edev = edev;
3054 		fp->id = queue_id;
3055 
3056 		memset((void *)&fp->napi, 0, sizeof(fp->napi));
3057 
3058 		memset((void *)fp->sb_info, 0, sizeof(*fp->sb_info));
3059 
3060 		if (fp->type & QEDE_FASTPATH_RX) {
3061 			memset((void *)fp->rxq, 0, sizeof(*fp->rxq));
3062 			fp->rxq->rxq_id = rxq_index++;
3063 		}
3064 
3065 		if (fp->type & QEDE_FASTPATH_TX) {
3066 			memset((void *)fp->txqs, 0,
3067 			       (edev->num_tc * sizeof(*fp->txqs)));
3068 			for (tc = 0; tc < edev->num_tc; tc++) {
3069 				fp->txqs[tc].index = txq_index +
3070 				    tc * QEDE_TSS_COUNT(edev);
3071 				if (edev->dev_info.is_legacy)
3072 					fp->txqs[tc].is_legacy = true;
3073 			}
3074 			txq_index++;
3075 		}
3076 
3077 		snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
3078 			 edev->ndev->name, queue_id);
3079 	}
3080 
3081 	edev->gro_disable = !(edev->ndev->features & NETIF_F_GRO);
3082 }
3083 
3084 static int qede_set_real_num_queues(struct qede_dev *edev)
3085 {
3086 	int rc = 0;
3087 
3088 	rc = netif_set_real_num_tx_queues(edev->ndev, QEDE_TSS_COUNT(edev));
3089 	if (rc) {
3090 		DP_NOTICE(edev, "Failed to set real number of Tx queues\n");
3091 		return rc;
3092 	}
3093 
3094 	rc = netif_set_real_num_rx_queues(edev->ndev, QEDE_RSS_COUNT(edev));
3095 	if (rc) {
3096 		DP_NOTICE(edev, "Failed to set real number of Rx queues\n");
3097 		return rc;
3098 	}
3099 
3100 	return 0;
3101 }
3102 
3103 static void qede_napi_disable_remove(struct qede_dev *edev)
3104 {
3105 	int i;
3106 
3107 	for_each_queue(i) {
3108 		napi_disable(&edev->fp_array[i].napi);
3109 
3110 		netif_napi_del(&edev->fp_array[i].napi);
3111 	}
3112 }
3113 
3114 static void qede_napi_add_enable(struct qede_dev *edev)
3115 {
3116 	int i;
3117 
3118 	/* Add NAPI objects */
3119 	for_each_queue(i) {
3120 		netif_napi_add(edev->ndev, &edev->fp_array[i].napi,
3121 			       qede_poll, NAPI_POLL_WEIGHT);
3122 		napi_enable(&edev->fp_array[i].napi);
3123 	}
3124 }
3125 
3126 static void qede_sync_free_irqs(struct qede_dev *edev)
3127 {
3128 	int i;
3129 
3130 	for (i = 0; i < edev->int_info.used_cnt; i++) {
3131 		if (edev->int_info.msix_cnt) {
3132 			synchronize_irq(edev->int_info.msix[i].vector);
3133 			free_irq(edev->int_info.msix[i].vector,
3134 				 &edev->fp_array[i]);
3135 		} else {
3136 			edev->ops->common->simd_handler_clean(edev->cdev, i);
3137 		}
3138 	}
3139 
3140 	edev->int_info.used_cnt = 0;
3141 }
3142 
3143 static int qede_req_msix_irqs(struct qede_dev *edev)
3144 {
3145 	int i, rc;
3146 
3147 	/* Sanitize number of interrupts == number of prepared RSS queues */
3148 	if (QEDE_QUEUE_CNT(edev) > edev->int_info.msix_cnt) {
3149 		DP_ERR(edev,
3150 		       "Interrupt mismatch: %d RSS queues > %d MSI-x vectors\n",
3151 		       QEDE_QUEUE_CNT(edev), edev->int_info.msix_cnt);
3152 		return -EINVAL;
3153 	}
3154 
3155 	for (i = 0; i < QEDE_QUEUE_CNT(edev); i++) {
3156 		rc = request_irq(edev->int_info.msix[i].vector,
3157 				 qede_msix_fp_int, 0, edev->fp_array[i].name,
3158 				 &edev->fp_array[i]);
3159 		if (rc) {
3160 			DP_ERR(edev, "Request fp %d irq failed\n", i);
3161 			qede_sync_free_irqs(edev);
3162 			return rc;
3163 		}
3164 		DP_VERBOSE(edev, NETIF_MSG_INTR,
3165 			   "Requested fp irq for %s [entry %d]. Cookie is at %p\n",
3166 			   edev->fp_array[i].name, i,
3167 			   &edev->fp_array[i]);
3168 		edev->int_info.used_cnt++;
3169 	}
3170 
3171 	return 0;
3172 }
3173 
3174 static void qede_simd_fp_handler(void *cookie)
3175 {
3176 	struct qede_fastpath *fp = (struct qede_fastpath *)cookie;
3177 
3178 	napi_schedule_irqoff(&fp->napi);
3179 }
3180 
3181 static int qede_setup_irqs(struct qede_dev *edev)
3182 {
3183 	int i, rc = 0;
3184 
3185 	/* Learn Interrupt configuration */
3186 	rc = edev->ops->common->get_fp_int(edev->cdev, &edev->int_info);
3187 	if (rc)
3188 		return rc;
3189 
3190 	if (edev->int_info.msix_cnt) {
3191 		rc = qede_req_msix_irqs(edev);
3192 		if (rc)
3193 			return rc;
3194 		edev->ndev->irq = edev->int_info.msix[0].vector;
3195 	} else {
3196 		const struct qed_common_ops *ops;
3197 
3198 		/* qed should learn receive the RSS ids and callbacks */
3199 		ops = edev->ops->common;
3200 		for (i = 0; i < QEDE_QUEUE_CNT(edev); i++)
3201 			ops->simd_handler_config(edev->cdev,
3202 						 &edev->fp_array[i], i,
3203 						 qede_simd_fp_handler);
3204 		edev->int_info.used_cnt = QEDE_QUEUE_CNT(edev);
3205 	}
3206 	return 0;
3207 }
3208 
3209 static int qede_drain_txq(struct qede_dev *edev,
3210 			  struct qede_tx_queue *txq, bool allow_drain)
3211 {
3212 	int rc, cnt = 1000;
3213 
3214 	while (txq->sw_tx_cons != txq->sw_tx_prod) {
3215 		if (!cnt) {
3216 			if (allow_drain) {
3217 				DP_NOTICE(edev,
3218 					  "Tx queue[%d] is stuck, requesting MCP to drain\n",
3219 					  txq->index);
3220 				rc = edev->ops->common->drain(edev->cdev);
3221 				if (rc)
3222 					return rc;
3223 				return qede_drain_txq(edev, txq, false);
3224 			}
3225 			DP_NOTICE(edev,
3226 				  "Timeout waiting for tx queue[%d]: PROD=%d, CONS=%d\n",
3227 				  txq->index, txq->sw_tx_prod,
3228 				  txq->sw_tx_cons);
3229 			return -ENODEV;
3230 		}
3231 		cnt--;
3232 		usleep_range(1000, 2000);
3233 		barrier();
3234 	}
3235 
3236 	/* FW finished processing, wait for HW to transmit all tx packets */
3237 	usleep_range(1000, 2000);
3238 
3239 	return 0;
3240 }
3241 
3242 static int qede_stop_queues(struct qede_dev *edev)
3243 {
3244 	struct qed_update_vport_params vport_update_params;
3245 	struct qed_dev *cdev = edev->cdev;
3246 	int rc, tc, i;
3247 
3248 	/* Disable the vport */
3249 	memset(&vport_update_params, 0, sizeof(vport_update_params));
3250 	vport_update_params.vport_id = 0;
3251 	vport_update_params.update_vport_active_flg = 1;
3252 	vport_update_params.vport_active_flg = 0;
3253 	vport_update_params.update_rss_flg = 0;
3254 
3255 	rc = edev->ops->vport_update(cdev, &vport_update_params);
3256 	if (rc) {
3257 		DP_ERR(edev, "Failed to update vport\n");
3258 		return rc;
3259 	}
3260 
3261 	/* Flush Tx queues. If needed, request drain from MCP */
3262 	for_each_queue(i) {
3263 		struct qede_fastpath *fp = &edev->fp_array[i];
3264 
3265 		if (fp->type & QEDE_FASTPATH_TX) {
3266 			for (tc = 0; tc < edev->num_tc; tc++) {
3267 				struct qede_tx_queue *txq = &fp->txqs[tc];
3268 
3269 				rc = qede_drain_txq(edev, txq, true);
3270 				if (rc)
3271 					return rc;
3272 			}
3273 		}
3274 	}
3275 
3276 	/* Stop all Queues in reverse order */
3277 	for (i = QEDE_QUEUE_CNT(edev) - 1; i >= 0; i--) {
3278 		struct qed_stop_rxq_params rx_params;
3279 
3280 		/* Stop the Tx Queue(s) */
3281 		if (edev->fp_array[i].type & QEDE_FASTPATH_TX) {
3282 			for (tc = 0; tc < edev->num_tc; tc++) {
3283 				struct qed_stop_txq_params tx_params;
3284 				u8 val;
3285 
3286 				tx_params.rss_id = i;
3287 				val = edev->fp_array[i].txqs[tc].index;
3288 				tx_params.tx_queue_id = val;
3289 				rc = edev->ops->q_tx_stop(cdev, &tx_params);
3290 				if (rc) {
3291 					DP_ERR(edev, "Failed to stop TXQ #%d\n",
3292 					       tx_params.tx_queue_id);
3293 					return rc;
3294 				}
3295 			}
3296 		}
3297 
3298 		/* Stop the Rx Queue */
3299 		if (edev->fp_array[i].type & QEDE_FASTPATH_RX) {
3300 			memset(&rx_params, 0, sizeof(rx_params));
3301 			rx_params.rss_id = i;
3302 			rx_params.rx_queue_id = edev->fp_array[i].rxq->rxq_id;
3303 
3304 			rc = edev->ops->q_rx_stop(cdev, &rx_params);
3305 			if (rc) {
3306 				DP_ERR(edev, "Failed to stop RXQ #%d\n", i);
3307 				return rc;
3308 			}
3309 		}
3310 	}
3311 
3312 	/* Stop the vport */
3313 	rc = edev->ops->vport_stop(cdev, 0);
3314 	if (rc)
3315 		DP_ERR(edev, "Failed to stop VPORT\n");
3316 
3317 	return rc;
3318 }
3319 
3320 static int qede_start_queues(struct qede_dev *edev, bool clear_stats)
3321 {
3322 	int rc, tc, i;
3323 	int vlan_removal_en = 1;
3324 	struct qed_dev *cdev = edev->cdev;
3325 	struct qed_update_vport_params vport_update_params;
3326 	struct qed_queue_start_common_params q_params;
3327 	struct qed_dev_info *qed_info = &edev->dev_info.common;
3328 	struct qed_start_vport_params start = {0};
3329 	bool reset_rss_indir = false;
3330 
3331 	if (!edev->num_queues) {
3332 		DP_ERR(edev,
3333 		       "Cannot update V-VPORT as active as there are no Rx queues\n");
3334 		return -EINVAL;
3335 	}
3336 
3337 	start.gro_enable = !edev->gro_disable;
3338 	start.mtu = edev->ndev->mtu;
3339 	start.vport_id = 0;
3340 	start.drop_ttl0 = true;
3341 	start.remove_inner_vlan = vlan_removal_en;
3342 	start.clear_stats = clear_stats;
3343 
3344 	rc = edev->ops->vport_start(cdev, &start);
3345 
3346 	if (rc) {
3347 		DP_ERR(edev, "Start V-PORT failed %d\n", rc);
3348 		return rc;
3349 	}
3350 
3351 	DP_VERBOSE(edev, NETIF_MSG_IFUP,
3352 		   "Start vport ramrod passed, vport_id = %d, MTU = %d, vlan_removal_en = %d\n",
3353 		   start.vport_id, edev->ndev->mtu + 0xe, vlan_removal_en);
3354 
3355 	for_each_queue(i) {
3356 		struct qede_fastpath *fp = &edev->fp_array[i];
3357 		dma_addr_t p_phys_table;
3358 		u32 page_cnt;
3359 
3360 		if (fp->type & QEDE_FASTPATH_RX) {
3361 			struct qede_rx_queue *rxq = fp->rxq;
3362 			__le16 *val;
3363 
3364 			memset(&q_params, 0, sizeof(q_params));
3365 			q_params.rss_id = i;
3366 			q_params.queue_id = rxq->rxq_id;
3367 			q_params.vport_id = 0;
3368 			q_params.sb = fp->sb_info->igu_sb_id;
3369 			q_params.sb_idx = RX_PI;
3370 
3371 			p_phys_table =
3372 			    qed_chain_get_pbl_phys(&rxq->rx_comp_ring);
3373 			page_cnt = qed_chain_get_page_cnt(&rxq->rx_comp_ring);
3374 
3375 			rc = edev->ops->q_rx_start(cdev, &q_params,
3376 						   rxq->rx_buf_size,
3377 						   rxq->rx_bd_ring.p_phys_addr,
3378 						   p_phys_table,
3379 						   page_cnt,
3380 						   &rxq->hw_rxq_prod_addr);
3381 			if (rc) {
3382 				DP_ERR(edev, "Start RXQ #%d failed %d\n", i,
3383 				       rc);
3384 				return rc;
3385 			}
3386 
3387 			val = &fp->sb_info->sb_virt->pi_array[RX_PI];
3388 			rxq->hw_cons_ptr = val;
3389 
3390 			qede_update_rx_prod(edev, rxq);
3391 		}
3392 
3393 		if (!(fp->type & QEDE_FASTPATH_TX))
3394 			continue;
3395 
3396 		for (tc = 0; tc < edev->num_tc; tc++) {
3397 			struct qede_tx_queue *txq = &fp->txqs[tc];
3398 
3399 			p_phys_table = qed_chain_get_pbl_phys(&txq->tx_pbl);
3400 			page_cnt = qed_chain_get_page_cnt(&txq->tx_pbl);
3401 
3402 			memset(&q_params, 0, sizeof(q_params));
3403 			q_params.rss_id = i;
3404 			q_params.queue_id = txq->index;
3405 			q_params.vport_id = 0;
3406 			q_params.sb = fp->sb_info->igu_sb_id;
3407 			q_params.sb_idx = TX_PI(tc);
3408 
3409 			rc = edev->ops->q_tx_start(cdev, &q_params,
3410 						   p_phys_table, page_cnt,
3411 						   &txq->doorbell_addr);
3412 			if (rc) {
3413 				DP_ERR(edev, "Start TXQ #%d failed %d\n",
3414 				       txq->index, rc);
3415 				return rc;
3416 			}
3417 
3418 			txq->hw_cons_ptr =
3419 				&fp->sb_info->sb_virt->pi_array[TX_PI(tc)];
3420 			SET_FIELD(txq->tx_db.data.params,
3421 				  ETH_DB_DATA_DEST, DB_DEST_XCM);
3422 			SET_FIELD(txq->tx_db.data.params, ETH_DB_DATA_AGG_CMD,
3423 				  DB_AGG_CMD_SET);
3424 			SET_FIELD(txq->tx_db.data.params,
3425 				  ETH_DB_DATA_AGG_VAL_SEL,
3426 				  DQ_XCM_ETH_TX_BD_PROD_CMD);
3427 
3428 			txq->tx_db.data.agg_flags = DQ_XCM_ETH_DQ_CF_CMD;
3429 		}
3430 	}
3431 
3432 	/* Prepare and send the vport enable */
3433 	memset(&vport_update_params, 0, sizeof(vport_update_params));
3434 	vport_update_params.vport_id = start.vport_id;
3435 	vport_update_params.update_vport_active_flg = 1;
3436 	vport_update_params.vport_active_flg = 1;
3437 
3438 	if ((qed_info->mf_mode == QED_MF_NPAR || pci_num_vf(edev->pdev)) &&
3439 	    qed_info->tx_switching) {
3440 		vport_update_params.update_tx_switching_flg = 1;
3441 		vport_update_params.tx_switching_flg = 1;
3442 	}
3443 
3444 	/* Fill struct with RSS params */
3445 	if (QEDE_RSS_COUNT(edev) > 1) {
3446 		vport_update_params.update_rss_flg = 1;
3447 
3448 		/* Need to validate current RSS config uses valid entries */
3449 		for (i = 0; i < QED_RSS_IND_TABLE_SIZE; i++) {
3450 			if (edev->rss_params.rss_ind_table[i] >=
3451 			    QEDE_RSS_COUNT(edev)) {
3452 				reset_rss_indir = true;
3453 				break;
3454 			}
3455 		}
3456 
3457 		if (!(edev->rss_params_inited & QEDE_RSS_INDIR_INITED) ||
3458 		    reset_rss_indir) {
3459 			u16 val;
3460 
3461 			for (i = 0; i < QED_RSS_IND_TABLE_SIZE; i++) {
3462 				u16 indir_val;
3463 
3464 				val = QEDE_RSS_COUNT(edev);
3465 				indir_val = ethtool_rxfh_indir_default(i, val);
3466 				edev->rss_params.rss_ind_table[i] = indir_val;
3467 			}
3468 			edev->rss_params_inited |= QEDE_RSS_INDIR_INITED;
3469 		}
3470 
3471 		if (!(edev->rss_params_inited & QEDE_RSS_KEY_INITED)) {
3472 			netdev_rss_key_fill(edev->rss_params.rss_key,
3473 					    sizeof(edev->rss_params.rss_key));
3474 			edev->rss_params_inited |= QEDE_RSS_KEY_INITED;
3475 		}
3476 
3477 		if (!(edev->rss_params_inited & QEDE_RSS_CAPS_INITED)) {
3478 			edev->rss_params.rss_caps = QED_RSS_IPV4 |
3479 						    QED_RSS_IPV6 |
3480 						    QED_RSS_IPV4_TCP |
3481 						    QED_RSS_IPV6_TCP;
3482 			edev->rss_params_inited |= QEDE_RSS_CAPS_INITED;
3483 		}
3484 
3485 		memcpy(&vport_update_params.rss_params, &edev->rss_params,
3486 		       sizeof(vport_update_params.rss_params));
3487 	} else {
3488 		memset(&vport_update_params.rss_params, 0,
3489 		       sizeof(vport_update_params.rss_params));
3490 	}
3491 
3492 	rc = edev->ops->vport_update(cdev, &vport_update_params);
3493 	if (rc) {
3494 		DP_ERR(edev, "Update V-PORT failed %d\n", rc);
3495 		return rc;
3496 	}
3497 
3498 	return 0;
3499 }
3500 
3501 static int qede_set_mcast_rx_mac(struct qede_dev *edev,
3502 				 enum qed_filter_xcast_params_type opcode,
3503 				 unsigned char *mac, int num_macs)
3504 {
3505 	struct qed_filter_params filter_cmd;
3506 	int i;
3507 
3508 	memset(&filter_cmd, 0, sizeof(filter_cmd));
3509 	filter_cmd.type = QED_FILTER_TYPE_MCAST;
3510 	filter_cmd.filter.mcast.type = opcode;
3511 	filter_cmd.filter.mcast.num = num_macs;
3512 
3513 	for (i = 0; i < num_macs; i++, mac += ETH_ALEN)
3514 		ether_addr_copy(filter_cmd.filter.mcast.mac[i], mac);
3515 
3516 	return edev->ops->filter_config(edev->cdev, &filter_cmd);
3517 }
3518 
3519 enum qede_unload_mode {
3520 	QEDE_UNLOAD_NORMAL,
3521 };
3522 
3523 static void qede_unload(struct qede_dev *edev, enum qede_unload_mode mode)
3524 {
3525 	struct qed_link_params link_params;
3526 	int rc;
3527 
3528 	DP_INFO(edev, "Starting qede unload\n");
3529 
3530 	qede_roce_dev_event_close(edev);
3531 	mutex_lock(&edev->qede_lock);
3532 	edev->state = QEDE_STATE_CLOSED;
3533 
3534 	/* Close OS Tx */
3535 	netif_tx_disable(edev->ndev);
3536 	netif_carrier_off(edev->ndev);
3537 
3538 	/* Reset the link */
3539 	memset(&link_params, 0, sizeof(link_params));
3540 	link_params.link_up = false;
3541 	edev->ops->common->set_link(edev->cdev, &link_params);
3542 	rc = qede_stop_queues(edev);
3543 	if (rc) {
3544 		qede_sync_free_irqs(edev);
3545 		goto out;
3546 	}
3547 
3548 	DP_INFO(edev, "Stopped Queues\n");
3549 
3550 	qede_vlan_mark_nonconfigured(edev);
3551 	edev->ops->fastpath_stop(edev->cdev);
3552 
3553 	/* Release the interrupts */
3554 	qede_sync_free_irqs(edev);
3555 	edev->ops->common->set_fp_int(edev->cdev, 0);
3556 
3557 	qede_napi_disable_remove(edev);
3558 
3559 	qede_free_mem_load(edev);
3560 	qede_free_fp_array(edev);
3561 
3562 out:
3563 	mutex_unlock(&edev->qede_lock);
3564 	DP_INFO(edev, "Ending qede unload\n");
3565 }
3566 
3567 enum qede_load_mode {
3568 	QEDE_LOAD_NORMAL,
3569 	QEDE_LOAD_RELOAD,
3570 };
3571 
3572 static int qede_load(struct qede_dev *edev, enum qede_load_mode mode)
3573 {
3574 	struct qed_link_params link_params;
3575 	struct qed_link_output link_output;
3576 	int rc;
3577 
3578 	DP_INFO(edev, "Starting qede load\n");
3579 
3580 	rc = qede_set_num_queues(edev);
3581 	if (rc)
3582 		goto err0;
3583 
3584 	rc = qede_alloc_fp_array(edev);
3585 	if (rc)
3586 		goto err0;
3587 
3588 	qede_init_fp(edev);
3589 
3590 	rc = qede_alloc_mem_load(edev);
3591 	if (rc)
3592 		goto err1;
3593 	DP_INFO(edev, "Allocated %d RSS queues on %d TC/s\n",
3594 		QEDE_QUEUE_CNT(edev), edev->num_tc);
3595 
3596 	rc = qede_set_real_num_queues(edev);
3597 	if (rc)
3598 		goto err2;
3599 
3600 	qede_napi_add_enable(edev);
3601 	DP_INFO(edev, "Napi added and enabled\n");
3602 
3603 	rc = qede_setup_irqs(edev);
3604 	if (rc)
3605 		goto err3;
3606 	DP_INFO(edev, "Setup IRQs succeeded\n");
3607 
3608 	rc = qede_start_queues(edev, mode != QEDE_LOAD_RELOAD);
3609 	if (rc)
3610 		goto err4;
3611 	DP_INFO(edev, "Start VPORT, RXQ and TXQ succeeded\n");
3612 
3613 	/* Add primary mac and set Rx filters */
3614 	ether_addr_copy(edev->primary_mac, edev->ndev->dev_addr);
3615 
3616 	mutex_lock(&edev->qede_lock);
3617 	edev->state = QEDE_STATE_OPEN;
3618 	mutex_unlock(&edev->qede_lock);
3619 
3620 	/* Program un-configured VLANs */
3621 	qede_configure_vlan_filters(edev);
3622 
3623 	/* Ask for link-up using current configuration */
3624 	memset(&link_params, 0, sizeof(link_params));
3625 	link_params.link_up = true;
3626 	edev->ops->common->set_link(edev->cdev, &link_params);
3627 
3628 	/* Query whether link is already-up */
3629 	memset(&link_output, 0, sizeof(link_output));
3630 	edev->ops->common->get_link(edev->cdev, &link_output);
3631 	qede_roce_dev_event_open(edev);
3632 	qede_link_update(edev, &link_output);
3633 
3634 	DP_INFO(edev, "Ending successfully qede load\n");
3635 
3636 	return 0;
3637 
3638 err4:
3639 	qede_sync_free_irqs(edev);
3640 	memset(&edev->int_info.msix_cnt, 0, sizeof(struct qed_int_info));
3641 err3:
3642 	qede_napi_disable_remove(edev);
3643 err2:
3644 	qede_free_mem_load(edev);
3645 err1:
3646 	edev->ops->common->set_fp_int(edev->cdev, 0);
3647 	qede_free_fp_array(edev);
3648 	edev->num_queues = 0;
3649 	edev->fp_num_tx = 0;
3650 	edev->fp_num_rx = 0;
3651 err0:
3652 	return rc;
3653 }
3654 
3655 void qede_reload(struct qede_dev *edev,
3656 		 void (*func)(struct qede_dev *, union qede_reload_args *),
3657 		 union qede_reload_args *args)
3658 {
3659 	qede_unload(edev, QEDE_UNLOAD_NORMAL);
3660 	/* Call function handler to update parameters
3661 	 * needed for function load.
3662 	 */
3663 	if (func)
3664 		func(edev, args);
3665 
3666 	qede_load(edev, QEDE_LOAD_RELOAD);
3667 
3668 	mutex_lock(&edev->qede_lock);
3669 	qede_config_rx_mode(edev->ndev);
3670 	mutex_unlock(&edev->qede_lock);
3671 }
3672 
3673 /* called with rtnl_lock */
3674 static int qede_open(struct net_device *ndev)
3675 {
3676 	struct qede_dev *edev = netdev_priv(ndev);
3677 	int rc;
3678 
3679 	netif_carrier_off(ndev);
3680 
3681 	edev->ops->common->set_power_state(edev->cdev, PCI_D0);
3682 
3683 	rc = qede_load(edev, QEDE_LOAD_NORMAL);
3684 
3685 	if (rc)
3686 		return rc;
3687 
3688 	udp_tunnel_get_rx_info(ndev);
3689 
3690 	return 0;
3691 }
3692 
3693 static int qede_close(struct net_device *ndev)
3694 {
3695 	struct qede_dev *edev = netdev_priv(ndev);
3696 
3697 	qede_unload(edev, QEDE_UNLOAD_NORMAL);
3698 
3699 	return 0;
3700 }
3701 
3702 static void qede_link_update(void *dev, struct qed_link_output *link)
3703 {
3704 	struct qede_dev *edev = dev;
3705 
3706 	if (!netif_running(edev->ndev)) {
3707 		DP_VERBOSE(edev, NETIF_MSG_LINK, "Interface is not running\n");
3708 		return;
3709 	}
3710 
3711 	if (link->link_up) {
3712 		if (!netif_carrier_ok(edev->ndev)) {
3713 			DP_NOTICE(edev, "Link is up\n");
3714 			netif_tx_start_all_queues(edev->ndev);
3715 			netif_carrier_on(edev->ndev);
3716 		}
3717 	} else {
3718 		if (netif_carrier_ok(edev->ndev)) {
3719 			DP_NOTICE(edev, "Link is down\n");
3720 			netif_tx_disable(edev->ndev);
3721 			netif_carrier_off(edev->ndev);
3722 		}
3723 	}
3724 }
3725 
3726 static int qede_set_mac_addr(struct net_device *ndev, void *p)
3727 {
3728 	struct qede_dev *edev = netdev_priv(ndev);
3729 	struct sockaddr *addr = p;
3730 	int rc;
3731 
3732 	ASSERT_RTNL(); /* @@@TBD To be removed */
3733 
3734 	DP_INFO(edev, "Set_mac_addr called\n");
3735 
3736 	if (!is_valid_ether_addr(addr->sa_data)) {
3737 		DP_NOTICE(edev, "The MAC address is not valid\n");
3738 		return -EFAULT;
3739 	}
3740 
3741 	if (!edev->ops->check_mac(edev->cdev, addr->sa_data)) {
3742 		DP_NOTICE(edev, "qed prevents setting MAC\n");
3743 		return -EINVAL;
3744 	}
3745 
3746 	ether_addr_copy(ndev->dev_addr, addr->sa_data);
3747 
3748 	if (!netif_running(ndev))  {
3749 		DP_NOTICE(edev, "The device is currently down\n");
3750 		return 0;
3751 	}
3752 
3753 	/* Remove the previous primary mac */
3754 	rc = qede_set_ucast_rx_mac(edev, QED_FILTER_XCAST_TYPE_DEL,
3755 				   edev->primary_mac);
3756 	if (rc)
3757 		return rc;
3758 
3759 	/* Add MAC filter according to the new unicast HW MAC address */
3760 	ether_addr_copy(edev->primary_mac, ndev->dev_addr);
3761 	return qede_set_ucast_rx_mac(edev, QED_FILTER_XCAST_TYPE_ADD,
3762 				      edev->primary_mac);
3763 }
3764 
3765 static int
3766 qede_configure_mcast_filtering(struct net_device *ndev,
3767 			       enum qed_filter_rx_mode_type *accept_flags)
3768 {
3769 	struct qede_dev *edev = netdev_priv(ndev);
3770 	unsigned char *mc_macs, *temp;
3771 	struct netdev_hw_addr *ha;
3772 	int rc = 0, mc_count;
3773 	size_t size;
3774 
3775 	size = 64 * ETH_ALEN;
3776 
3777 	mc_macs = kzalloc(size, GFP_KERNEL);
3778 	if (!mc_macs) {
3779 		DP_NOTICE(edev,
3780 			  "Failed to allocate memory for multicast MACs\n");
3781 		rc = -ENOMEM;
3782 		goto exit;
3783 	}
3784 
3785 	temp = mc_macs;
3786 
3787 	/* Remove all previously configured MAC filters */
3788 	rc = qede_set_mcast_rx_mac(edev, QED_FILTER_XCAST_TYPE_DEL,
3789 				   mc_macs, 1);
3790 	if (rc)
3791 		goto exit;
3792 
3793 	netif_addr_lock_bh(ndev);
3794 
3795 	mc_count = netdev_mc_count(ndev);
3796 	if (mc_count < 64) {
3797 		netdev_for_each_mc_addr(ha, ndev) {
3798 			ether_addr_copy(temp, ha->addr);
3799 			temp += ETH_ALEN;
3800 		}
3801 	}
3802 
3803 	netif_addr_unlock_bh(ndev);
3804 
3805 	/* Check for all multicast @@@TBD resource allocation */
3806 	if ((ndev->flags & IFF_ALLMULTI) ||
3807 	    (mc_count > 64)) {
3808 		if (*accept_flags == QED_FILTER_RX_MODE_TYPE_REGULAR)
3809 			*accept_flags = QED_FILTER_RX_MODE_TYPE_MULTI_PROMISC;
3810 	} else {
3811 		/* Add all multicast MAC filters */
3812 		rc = qede_set_mcast_rx_mac(edev, QED_FILTER_XCAST_TYPE_ADD,
3813 					   mc_macs, mc_count);
3814 	}
3815 
3816 exit:
3817 	kfree(mc_macs);
3818 	return rc;
3819 }
3820 
3821 static void qede_set_rx_mode(struct net_device *ndev)
3822 {
3823 	struct qede_dev *edev = netdev_priv(ndev);
3824 
3825 	DP_INFO(edev, "qede_set_rx_mode called\n");
3826 
3827 	if (edev->state != QEDE_STATE_OPEN) {
3828 		DP_INFO(edev,
3829 			"qede_set_rx_mode called while interface is down\n");
3830 	} else {
3831 		set_bit(QEDE_SP_RX_MODE, &edev->sp_flags);
3832 		schedule_delayed_work(&edev->sp_task, 0);
3833 	}
3834 }
3835 
3836 /* Must be called with qede_lock held */
3837 static void qede_config_rx_mode(struct net_device *ndev)
3838 {
3839 	enum qed_filter_rx_mode_type accept_flags = QED_FILTER_TYPE_UCAST;
3840 	struct qede_dev *edev = netdev_priv(ndev);
3841 	struct qed_filter_params rx_mode;
3842 	unsigned char *uc_macs, *temp;
3843 	struct netdev_hw_addr *ha;
3844 	int rc, uc_count;
3845 	size_t size;
3846 
3847 	netif_addr_lock_bh(ndev);
3848 
3849 	uc_count = netdev_uc_count(ndev);
3850 	size = uc_count * ETH_ALEN;
3851 
3852 	uc_macs = kzalloc(size, GFP_ATOMIC);
3853 	if (!uc_macs) {
3854 		DP_NOTICE(edev, "Failed to allocate memory for unicast MACs\n");
3855 		netif_addr_unlock_bh(ndev);
3856 		return;
3857 	}
3858 
3859 	temp = uc_macs;
3860 	netdev_for_each_uc_addr(ha, ndev) {
3861 		ether_addr_copy(temp, ha->addr);
3862 		temp += ETH_ALEN;
3863 	}
3864 
3865 	netif_addr_unlock_bh(ndev);
3866 
3867 	/* Configure the struct for the Rx mode */
3868 	memset(&rx_mode, 0, sizeof(struct qed_filter_params));
3869 	rx_mode.type = QED_FILTER_TYPE_RX_MODE;
3870 
3871 	/* Remove all previous unicast secondary macs and multicast macs
3872 	 * (configrue / leave the primary mac)
3873 	 */
3874 	rc = qede_set_ucast_rx_mac(edev, QED_FILTER_XCAST_TYPE_REPLACE,
3875 				   edev->primary_mac);
3876 	if (rc)
3877 		goto out;
3878 
3879 	/* Check for promiscuous */
3880 	if ((ndev->flags & IFF_PROMISC) ||
3881 	    (uc_count > 15)) { /* @@@TBD resource allocation - 1 */
3882 		accept_flags = QED_FILTER_RX_MODE_TYPE_PROMISC;
3883 	} else {
3884 		/* Add MAC filters according to the unicast secondary macs */
3885 		int i;
3886 
3887 		temp = uc_macs;
3888 		for (i = 0; i < uc_count; i++) {
3889 			rc = qede_set_ucast_rx_mac(edev,
3890 						   QED_FILTER_XCAST_TYPE_ADD,
3891 						   temp);
3892 			if (rc)
3893 				goto out;
3894 
3895 			temp += ETH_ALEN;
3896 		}
3897 
3898 		rc = qede_configure_mcast_filtering(ndev, &accept_flags);
3899 		if (rc)
3900 			goto out;
3901 	}
3902 
3903 	/* take care of VLAN mode */
3904 	if (ndev->flags & IFF_PROMISC) {
3905 		qede_config_accept_any_vlan(edev, true);
3906 	} else if (!edev->non_configured_vlans) {
3907 		/* It's possible that accept_any_vlan mode is set due to a
3908 		 * previous setting of IFF_PROMISC. If vlan credits are
3909 		 * sufficient, disable accept_any_vlan.
3910 		 */
3911 		qede_config_accept_any_vlan(edev, false);
3912 	}
3913 
3914 	rx_mode.filter.accept_flags = accept_flags;
3915 	edev->ops->filter_config(edev->cdev, &rx_mode);
3916 out:
3917 	kfree(uc_macs);
3918 }
3919