xref: /openbmc/linux/drivers/net/ethernet/emulex/benet/be_main.c (revision 5ef12cb4a3a78ffb331c03a795a15eea4ae35155)
1 /*
2  * Copyright (C) 2005 - 2016 Broadcom
3  * All rights reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License version 2
7  * as published by the Free Software Foundation.  The full GNU General
8  * Public License is included in this distribution in the file called COPYING.
9  *
10  * Contact Information:
11  * linux-drivers@emulex.com
12  *
13  * Emulex
14  * 3333 Susan Street
15  * Costa Mesa, CA 92626
16  */
17 
18 #include <linux/prefetch.h>
19 #include <linux/module.h>
20 #include "be.h"
21 #include "be_cmds.h"
22 #include <asm/div64.h>
23 #include <linux/aer.h>
24 #include <linux/if_bridge.h>
25 #include <net/busy_poll.h>
26 #include <net/vxlan.h>
27 
28 MODULE_VERSION(DRV_VER);
29 MODULE_DESCRIPTION(DRV_DESC " " DRV_VER);
30 MODULE_AUTHOR("Emulex Corporation");
31 MODULE_LICENSE("GPL");
32 
33 /* num_vfs module param is obsolete.
34  * Use sysfs method to enable/disable VFs.
35  */
36 static unsigned int num_vfs;
37 module_param(num_vfs, uint, 0444);
38 MODULE_PARM_DESC(num_vfs, "Number of PCI VFs to initialize");
39 
40 static ushort rx_frag_size = 2048;
41 module_param(rx_frag_size, ushort, 0444);
42 MODULE_PARM_DESC(rx_frag_size, "Size of a fragment that holds rcvd data.");
43 
44 /* Per-module error detection/recovery workq shared across all functions.
45  * Each function schedules its own work request on this shared workq.
46  */
47 static struct workqueue_struct *be_err_recovery_workq;
48 
49 static const struct pci_device_id be_dev_ids[] = {
50 	{ PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID1) },
51 	{ PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID2) },
52 	{ PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID1) },
53 	{ PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID2) },
54 	{ PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID3)},
55 	{ PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID4)},
56 	{ PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID5)},
57 	{ PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID6)},
58 	{ 0 }
59 };
60 MODULE_DEVICE_TABLE(pci, be_dev_ids);
61 
62 /* Workqueue used by all functions for defering cmd calls to the adapter */
63 static struct workqueue_struct *be_wq;
64 
65 /* UE Status Low CSR */
66 static const char * const ue_status_low_desc[] = {
67 	"CEV",
68 	"CTX",
69 	"DBUF",
70 	"ERX",
71 	"Host",
72 	"MPU",
73 	"NDMA",
74 	"PTC ",
75 	"RDMA ",
76 	"RXF ",
77 	"RXIPS ",
78 	"RXULP0 ",
79 	"RXULP1 ",
80 	"RXULP2 ",
81 	"TIM ",
82 	"TPOST ",
83 	"TPRE ",
84 	"TXIPS ",
85 	"TXULP0 ",
86 	"TXULP1 ",
87 	"UC ",
88 	"WDMA ",
89 	"TXULP2 ",
90 	"HOST1 ",
91 	"P0_OB_LINK ",
92 	"P1_OB_LINK ",
93 	"HOST_GPIO ",
94 	"MBOX ",
95 	"ERX2 ",
96 	"SPARE ",
97 	"JTAG ",
98 	"MPU_INTPEND "
99 };
100 
101 /* UE Status High CSR */
102 static const char * const ue_status_hi_desc[] = {
103 	"LPCMEMHOST",
104 	"MGMT_MAC",
105 	"PCS0ONLINE",
106 	"MPU_IRAM",
107 	"PCS1ONLINE",
108 	"PCTL0",
109 	"PCTL1",
110 	"PMEM",
111 	"RR",
112 	"TXPB",
113 	"RXPP",
114 	"XAUI",
115 	"TXP",
116 	"ARM",
117 	"IPC",
118 	"HOST2",
119 	"HOST3",
120 	"HOST4",
121 	"HOST5",
122 	"HOST6",
123 	"HOST7",
124 	"ECRC",
125 	"Poison TLP",
126 	"NETC",
127 	"PERIPH",
128 	"LLTXULP",
129 	"D2P",
130 	"RCON",
131 	"LDMA",
132 	"LLTXP",
133 	"LLTXPB",
134 	"Unknown"
135 };
136 
137 #define BE_VF_IF_EN_FLAGS	(BE_IF_FLAGS_UNTAGGED | \
138 				 BE_IF_FLAGS_BROADCAST | \
139 				 BE_IF_FLAGS_MULTICAST | \
140 				 BE_IF_FLAGS_PASS_L3L4_ERRORS)
141 
142 static void be_queue_free(struct be_adapter *adapter, struct be_queue_info *q)
143 {
144 	struct be_dma_mem *mem = &q->dma_mem;
145 
146 	if (mem->va) {
147 		dma_free_coherent(&adapter->pdev->dev, mem->size, mem->va,
148 				  mem->dma);
149 		mem->va = NULL;
150 	}
151 }
152 
153 static int be_queue_alloc(struct be_adapter *adapter, struct be_queue_info *q,
154 			  u16 len, u16 entry_size)
155 {
156 	struct be_dma_mem *mem = &q->dma_mem;
157 
158 	memset(q, 0, sizeof(*q));
159 	q->len = len;
160 	q->entry_size = entry_size;
161 	mem->size = len * entry_size;
162 	mem->va = dma_zalloc_coherent(&adapter->pdev->dev, mem->size, &mem->dma,
163 				      GFP_KERNEL);
164 	if (!mem->va)
165 		return -ENOMEM;
166 	return 0;
167 }
168 
169 static void be_reg_intr_set(struct be_adapter *adapter, bool enable)
170 {
171 	u32 reg, enabled;
172 
173 	pci_read_config_dword(adapter->pdev, PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET,
174 			      &reg);
175 	enabled = reg & MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
176 
177 	if (!enabled && enable)
178 		reg |= MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
179 	else if (enabled && !enable)
180 		reg &= ~MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
181 	else
182 		return;
183 
184 	pci_write_config_dword(adapter->pdev,
185 			       PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET, reg);
186 }
187 
188 static void be_intr_set(struct be_adapter *adapter, bool enable)
189 {
190 	int status = 0;
191 
192 	/* On lancer interrupts can't be controlled via this register */
193 	if (lancer_chip(adapter))
194 		return;
195 
196 	if (be_check_error(adapter, BE_ERROR_EEH))
197 		return;
198 
199 	status = be_cmd_intr_set(adapter, enable);
200 	if (status)
201 		be_reg_intr_set(adapter, enable);
202 }
203 
204 static void be_rxq_notify(struct be_adapter *adapter, u16 qid, u16 posted)
205 {
206 	u32 val = 0;
207 
208 	if (be_check_error(adapter, BE_ERROR_HW))
209 		return;
210 
211 	val |= qid & DB_RQ_RING_ID_MASK;
212 	val |= posted << DB_RQ_NUM_POSTED_SHIFT;
213 
214 	wmb();
215 	iowrite32(val, adapter->db + DB_RQ_OFFSET);
216 }
217 
218 static void be_txq_notify(struct be_adapter *adapter, struct be_tx_obj *txo,
219 			  u16 posted)
220 {
221 	u32 val = 0;
222 
223 	if (be_check_error(adapter, BE_ERROR_HW))
224 		return;
225 
226 	val |= txo->q.id & DB_TXULP_RING_ID_MASK;
227 	val |= (posted & DB_TXULP_NUM_POSTED_MASK) << DB_TXULP_NUM_POSTED_SHIFT;
228 
229 	wmb();
230 	iowrite32(val, adapter->db + txo->db_offset);
231 }
232 
233 static void be_eq_notify(struct be_adapter *adapter, u16 qid,
234 			 bool arm, bool clear_int, u16 num_popped,
235 			 u32 eq_delay_mult_enc)
236 {
237 	u32 val = 0;
238 
239 	val |= qid & DB_EQ_RING_ID_MASK;
240 	val |= ((qid & DB_EQ_RING_ID_EXT_MASK) << DB_EQ_RING_ID_EXT_MASK_SHIFT);
241 
242 	if (be_check_error(adapter, BE_ERROR_HW))
243 		return;
244 
245 	if (arm)
246 		val |= 1 << DB_EQ_REARM_SHIFT;
247 	if (clear_int)
248 		val |= 1 << DB_EQ_CLR_SHIFT;
249 	val |= 1 << DB_EQ_EVNT_SHIFT;
250 	val |= num_popped << DB_EQ_NUM_POPPED_SHIFT;
251 	val |= eq_delay_mult_enc << DB_EQ_R2I_DLY_SHIFT;
252 	iowrite32(val, adapter->db + DB_EQ_OFFSET);
253 }
254 
255 void be_cq_notify(struct be_adapter *adapter, u16 qid, bool arm, u16 num_popped)
256 {
257 	u32 val = 0;
258 
259 	val |= qid & DB_CQ_RING_ID_MASK;
260 	val |= ((qid & DB_CQ_RING_ID_EXT_MASK) <<
261 			DB_CQ_RING_ID_EXT_MASK_SHIFT);
262 
263 	if (be_check_error(adapter, BE_ERROR_HW))
264 		return;
265 
266 	if (arm)
267 		val |= 1 << DB_CQ_REARM_SHIFT;
268 	val |= num_popped << DB_CQ_NUM_POPPED_SHIFT;
269 	iowrite32(val, adapter->db + DB_CQ_OFFSET);
270 }
271 
272 static int be_dev_mac_add(struct be_adapter *adapter, u8 *mac)
273 {
274 	int i;
275 
276 	/* Check if mac has already been added as part of uc-list */
277 	for (i = 0; i < adapter->uc_macs; i++) {
278 		if (ether_addr_equal(adapter->uc_list[i].mac, mac)) {
279 			/* mac already added, skip addition */
280 			adapter->pmac_id[0] = adapter->pmac_id[i + 1];
281 			return 0;
282 		}
283 	}
284 
285 	return be_cmd_pmac_add(adapter, mac, adapter->if_handle,
286 			       &adapter->pmac_id[0], 0);
287 }
288 
289 static void be_dev_mac_del(struct be_adapter *adapter, int pmac_id)
290 {
291 	int i;
292 
293 	/* Skip deletion if the programmed mac is
294 	 * being used in uc-list
295 	 */
296 	for (i = 0; i < adapter->uc_macs; i++) {
297 		if (adapter->pmac_id[i + 1] == pmac_id)
298 			return;
299 	}
300 	be_cmd_pmac_del(adapter, adapter->if_handle, pmac_id, 0);
301 }
302 
303 static int be_mac_addr_set(struct net_device *netdev, void *p)
304 {
305 	struct be_adapter *adapter = netdev_priv(netdev);
306 	struct device *dev = &adapter->pdev->dev;
307 	struct sockaddr *addr = p;
308 	int status;
309 	u8 mac[ETH_ALEN];
310 	u32 old_pmac_id = adapter->pmac_id[0];
311 
312 	if (!is_valid_ether_addr(addr->sa_data))
313 		return -EADDRNOTAVAIL;
314 
315 	/* Proceed further only if, User provided MAC is different
316 	 * from active MAC
317 	 */
318 	if (ether_addr_equal(addr->sa_data, adapter->dev_mac))
319 		return 0;
320 
321 	/* BE3 VFs without FILTMGMT privilege are not allowed to set its MAC
322 	 * address
323 	 */
324 	if (BEx_chip(adapter) && be_virtfn(adapter) &&
325 	    !check_privilege(adapter, BE_PRIV_FILTMGMT))
326 		return -EPERM;
327 
328 	/* if device is not running, copy MAC to netdev->dev_addr */
329 	if (!netif_running(netdev))
330 		goto done;
331 
332 	/* The PMAC_ADD cmd may fail if the VF doesn't have FILTMGMT
333 	 * privilege or if PF did not provision the new MAC address.
334 	 * On BE3, this cmd will always fail if the VF doesn't have the
335 	 * FILTMGMT privilege. This failure is OK, only if the PF programmed
336 	 * the MAC for the VF.
337 	 */
338 	mutex_lock(&adapter->rx_filter_lock);
339 	status = be_dev_mac_add(adapter, (u8 *)addr->sa_data);
340 	if (!status) {
341 
342 		/* Delete the old programmed MAC. This call may fail if the
343 		 * old MAC was already deleted by the PF driver.
344 		 */
345 		if (adapter->pmac_id[0] != old_pmac_id)
346 			be_dev_mac_del(adapter, old_pmac_id);
347 	}
348 
349 	mutex_unlock(&adapter->rx_filter_lock);
350 	/* Decide if the new MAC is successfully activated only after
351 	 * querying the FW
352 	 */
353 	status = be_cmd_get_active_mac(adapter, adapter->pmac_id[0], mac,
354 				       adapter->if_handle, true, 0);
355 	if (status)
356 		goto err;
357 
358 	/* The MAC change did not happen, either due to lack of privilege
359 	 * or PF didn't pre-provision.
360 	 */
361 	if (!ether_addr_equal(addr->sa_data, mac)) {
362 		status = -EPERM;
363 		goto err;
364 	}
365 
366 	/* Remember currently programmed MAC */
367 	ether_addr_copy(adapter->dev_mac, addr->sa_data);
368 done:
369 	ether_addr_copy(netdev->dev_addr, addr->sa_data);
370 	dev_info(dev, "MAC address changed to %pM\n", addr->sa_data);
371 	return 0;
372 err:
373 	dev_warn(dev, "MAC address change to %pM failed\n", addr->sa_data);
374 	return status;
375 }
376 
377 /* BE2 supports only v0 cmd */
378 static void *hw_stats_from_cmd(struct be_adapter *adapter)
379 {
380 	if (BE2_chip(adapter)) {
381 		struct be_cmd_resp_get_stats_v0 *cmd = adapter->stats_cmd.va;
382 
383 		return &cmd->hw_stats;
384 	} else if (BE3_chip(adapter)) {
385 		struct be_cmd_resp_get_stats_v1 *cmd = adapter->stats_cmd.va;
386 
387 		return &cmd->hw_stats;
388 	} else {
389 		struct be_cmd_resp_get_stats_v2 *cmd = adapter->stats_cmd.va;
390 
391 		return &cmd->hw_stats;
392 	}
393 }
394 
395 /* BE2 supports only v0 cmd */
396 static void *be_erx_stats_from_cmd(struct be_adapter *adapter)
397 {
398 	if (BE2_chip(adapter)) {
399 		struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
400 
401 		return &hw_stats->erx;
402 	} else if (BE3_chip(adapter)) {
403 		struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
404 
405 		return &hw_stats->erx;
406 	} else {
407 		struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
408 
409 		return &hw_stats->erx;
410 	}
411 }
412 
413 static void populate_be_v0_stats(struct be_adapter *adapter)
414 {
415 	struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
416 	struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
417 	struct be_rxf_stats_v0 *rxf_stats = &hw_stats->rxf;
418 	struct be_port_rxf_stats_v0 *port_stats =
419 					&rxf_stats->port[adapter->port_num];
420 	struct be_drv_stats *drvs = &adapter->drv_stats;
421 
422 	be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
423 	drvs->rx_pause_frames = port_stats->rx_pause_frames;
424 	drvs->rx_crc_errors = port_stats->rx_crc_errors;
425 	drvs->rx_control_frames = port_stats->rx_control_frames;
426 	drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
427 	drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
428 	drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
429 	drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
430 	drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
431 	drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
432 	drvs->rxpp_fifo_overflow_drop = port_stats->rx_fifo_overflow;
433 	drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
434 	drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
435 	drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
436 	drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
437 	drvs->rx_input_fifo_overflow_drop = port_stats->rx_input_fifo_overflow;
438 	drvs->rx_dropped_header_too_small =
439 		port_stats->rx_dropped_header_too_small;
440 	drvs->rx_address_filtered =
441 					port_stats->rx_address_filtered +
442 					port_stats->rx_vlan_filtered;
443 	drvs->rx_alignment_symbol_errors =
444 		port_stats->rx_alignment_symbol_errors;
445 
446 	drvs->tx_pauseframes = port_stats->tx_pauseframes;
447 	drvs->tx_controlframes = port_stats->tx_controlframes;
448 
449 	if (adapter->port_num)
450 		drvs->jabber_events = rxf_stats->port1_jabber_events;
451 	else
452 		drvs->jabber_events = rxf_stats->port0_jabber_events;
453 	drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
454 	drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
455 	drvs->forwarded_packets = rxf_stats->forwarded_packets;
456 	drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
457 	drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
458 	drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
459 	adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
460 }
461 
462 static void populate_be_v1_stats(struct be_adapter *adapter)
463 {
464 	struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
465 	struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
466 	struct be_rxf_stats_v1 *rxf_stats = &hw_stats->rxf;
467 	struct be_port_rxf_stats_v1 *port_stats =
468 					&rxf_stats->port[adapter->port_num];
469 	struct be_drv_stats *drvs = &adapter->drv_stats;
470 
471 	be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
472 	drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
473 	drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
474 	drvs->rx_pause_frames = port_stats->rx_pause_frames;
475 	drvs->rx_crc_errors = port_stats->rx_crc_errors;
476 	drvs->rx_control_frames = port_stats->rx_control_frames;
477 	drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
478 	drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
479 	drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
480 	drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
481 	drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
482 	drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
483 	drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
484 	drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
485 	drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
486 	drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
487 	drvs->rx_dropped_header_too_small =
488 		port_stats->rx_dropped_header_too_small;
489 	drvs->rx_input_fifo_overflow_drop =
490 		port_stats->rx_input_fifo_overflow_drop;
491 	drvs->rx_address_filtered = port_stats->rx_address_filtered;
492 	drvs->rx_alignment_symbol_errors =
493 		port_stats->rx_alignment_symbol_errors;
494 	drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
495 	drvs->tx_pauseframes = port_stats->tx_pauseframes;
496 	drvs->tx_controlframes = port_stats->tx_controlframes;
497 	drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
498 	drvs->jabber_events = port_stats->jabber_events;
499 	drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
500 	drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
501 	drvs->forwarded_packets = rxf_stats->forwarded_packets;
502 	drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
503 	drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
504 	drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
505 	adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
506 }
507 
508 static void populate_be_v2_stats(struct be_adapter *adapter)
509 {
510 	struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
511 	struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
512 	struct be_rxf_stats_v2 *rxf_stats = &hw_stats->rxf;
513 	struct be_port_rxf_stats_v2 *port_stats =
514 					&rxf_stats->port[adapter->port_num];
515 	struct be_drv_stats *drvs = &adapter->drv_stats;
516 
517 	be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
518 	drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
519 	drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
520 	drvs->rx_pause_frames = port_stats->rx_pause_frames;
521 	drvs->rx_crc_errors = port_stats->rx_crc_errors;
522 	drvs->rx_control_frames = port_stats->rx_control_frames;
523 	drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
524 	drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
525 	drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
526 	drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
527 	drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
528 	drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
529 	drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
530 	drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
531 	drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
532 	drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
533 	drvs->rx_dropped_header_too_small =
534 		port_stats->rx_dropped_header_too_small;
535 	drvs->rx_input_fifo_overflow_drop =
536 		port_stats->rx_input_fifo_overflow_drop;
537 	drvs->rx_address_filtered = port_stats->rx_address_filtered;
538 	drvs->rx_alignment_symbol_errors =
539 		port_stats->rx_alignment_symbol_errors;
540 	drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
541 	drvs->tx_pauseframes = port_stats->tx_pauseframes;
542 	drvs->tx_controlframes = port_stats->tx_controlframes;
543 	drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
544 	drvs->jabber_events = port_stats->jabber_events;
545 	drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
546 	drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
547 	drvs->forwarded_packets = rxf_stats->forwarded_packets;
548 	drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
549 	drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
550 	drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
551 	adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
552 	if (be_roce_supported(adapter)) {
553 		drvs->rx_roce_bytes_lsd = port_stats->roce_bytes_received_lsd;
554 		drvs->rx_roce_bytes_msd = port_stats->roce_bytes_received_msd;
555 		drvs->rx_roce_frames = port_stats->roce_frames_received;
556 		drvs->roce_drops_crc = port_stats->roce_drops_crc;
557 		drvs->roce_drops_payload_len =
558 			port_stats->roce_drops_payload_len;
559 	}
560 }
561 
562 static void populate_lancer_stats(struct be_adapter *adapter)
563 {
564 	struct be_drv_stats *drvs = &adapter->drv_stats;
565 	struct lancer_pport_stats *pport_stats = pport_stats_from_cmd(adapter);
566 
567 	be_dws_le_to_cpu(pport_stats, sizeof(*pport_stats));
568 	drvs->rx_pause_frames = pport_stats->rx_pause_frames_lo;
569 	drvs->rx_crc_errors = pport_stats->rx_crc_errors_lo;
570 	drvs->rx_control_frames = pport_stats->rx_control_frames_lo;
571 	drvs->rx_in_range_errors = pport_stats->rx_in_range_errors;
572 	drvs->rx_frame_too_long = pport_stats->rx_frames_too_long_lo;
573 	drvs->rx_dropped_runt = pport_stats->rx_dropped_runt;
574 	drvs->rx_ip_checksum_errs = pport_stats->rx_ip_checksum_errors;
575 	drvs->rx_tcp_checksum_errs = pport_stats->rx_tcp_checksum_errors;
576 	drvs->rx_udp_checksum_errs = pport_stats->rx_udp_checksum_errors;
577 	drvs->rx_dropped_tcp_length =
578 				pport_stats->rx_dropped_invalid_tcp_length;
579 	drvs->rx_dropped_too_small = pport_stats->rx_dropped_too_small;
580 	drvs->rx_dropped_too_short = pport_stats->rx_dropped_too_short;
581 	drvs->rx_out_range_errors = pport_stats->rx_out_of_range_errors;
582 	drvs->rx_dropped_header_too_small =
583 				pport_stats->rx_dropped_header_too_small;
584 	drvs->rx_input_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
585 	drvs->rx_address_filtered =
586 					pport_stats->rx_address_filtered +
587 					pport_stats->rx_vlan_filtered;
588 	drvs->rx_alignment_symbol_errors = pport_stats->rx_symbol_errors_lo;
589 	drvs->rxpp_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
590 	drvs->tx_pauseframes = pport_stats->tx_pause_frames_lo;
591 	drvs->tx_controlframes = pport_stats->tx_control_frames_lo;
592 	drvs->jabber_events = pport_stats->rx_jabbers;
593 	drvs->forwarded_packets = pport_stats->num_forwards_lo;
594 	drvs->rx_drops_mtu = pport_stats->rx_drops_mtu_lo;
595 	drvs->rx_drops_too_many_frags =
596 				pport_stats->rx_drops_too_many_frags_lo;
597 }
598 
599 static void accumulate_16bit_val(u32 *acc, u16 val)
600 {
601 #define lo(x)			(x & 0xFFFF)
602 #define hi(x)			(x & 0xFFFF0000)
603 	bool wrapped = val < lo(*acc);
604 	u32 newacc = hi(*acc) + val;
605 
606 	if (wrapped)
607 		newacc += 65536;
608 	WRITE_ONCE(*acc, newacc);
609 }
610 
611 static void populate_erx_stats(struct be_adapter *adapter,
612 			       struct be_rx_obj *rxo, u32 erx_stat)
613 {
614 	if (!BEx_chip(adapter))
615 		rx_stats(rxo)->rx_drops_no_frags = erx_stat;
616 	else
617 		/* below erx HW counter can actually wrap around after
618 		 * 65535. Driver accumulates a 32-bit value
619 		 */
620 		accumulate_16bit_val(&rx_stats(rxo)->rx_drops_no_frags,
621 				     (u16)erx_stat);
622 }
623 
624 void be_parse_stats(struct be_adapter *adapter)
625 {
626 	struct be_erx_stats_v2 *erx = be_erx_stats_from_cmd(adapter);
627 	struct be_rx_obj *rxo;
628 	int i;
629 	u32 erx_stat;
630 
631 	if (lancer_chip(adapter)) {
632 		populate_lancer_stats(adapter);
633 	} else {
634 		if (BE2_chip(adapter))
635 			populate_be_v0_stats(adapter);
636 		else if (BE3_chip(adapter))
637 			/* for BE3 */
638 			populate_be_v1_stats(adapter);
639 		else
640 			populate_be_v2_stats(adapter);
641 
642 		/* erx_v2 is longer than v0, v1. use v2 for v0, v1 access */
643 		for_all_rx_queues(adapter, rxo, i) {
644 			erx_stat = erx->rx_drops_no_fragments[rxo->q.id];
645 			populate_erx_stats(adapter, rxo, erx_stat);
646 		}
647 	}
648 }
649 
650 static void be_get_stats64(struct net_device *netdev,
651 			   struct rtnl_link_stats64 *stats)
652 {
653 	struct be_adapter *adapter = netdev_priv(netdev);
654 	struct be_drv_stats *drvs = &adapter->drv_stats;
655 	struct be_rx_obj *rxo;
656 	struct be_tx_obj *txo;
657 	u64 pkts, bytes;
658 	unsigned int start;
659 	int i;
660 
661 	for_all_rx_queues(adapter, rxo, i) {
662 		const struct be_rx_stats *rx_stats = rx_stats(rxo);
663 
664 		do {
665 			start = u64_stats_fetch_begin_irq(&rx_stats->sync);
666 			pkts = rx_stats(rxo)->rx_pkts;
667 			bytes = rx_stats(rxo)->rx_bytes;
668 		} while (u64_stats_fetch_retry_irq(&rx_stats->sync, start));
669 		stats->rx_packets += pkts;
670 		stats->rx_bytes += bytes;
671 		stats->multicast += rx_stats(rxo)->rx_mcast_pkts;
672 		stats->rx_dropped += rx_stats(rxo)->rx_drops_no_skbs +
673 					rx_stats(rxo)->rx_drops_no_frags;
674 	}
675 
676 	for_all_tx_queues(adapter, txo, i) {
677 		const struct be_tx_stats *tx_stats = tx_stats(txo);
678 
679 		do {
680 			start = u64_stats_fetch_begin_irq(&tx_stats->sync);
681 			pkts = tx_stats(txo)->tx_pkts;
682 			bytes = tx_stats(txo)->tx_bytes;
683 		} while (u64_stats_fetch_retry_irq(&tx_stats->sync, start));
684 		stats->tx_packets += pkts;
685 		stats->tx_bytes += bytes;
686 	}
687 
688 	/* bad pkts received */
689 	stats->rx_errors = drvs->rx_crc_errors +
690 		drvs->rx_alignment_symbol_errors +
691 		drvs->rx_in_range_errors +
692 		drvs->rx_out_range_errors +
693 		drvs->rx_frame_too_long +
694 		drvs->rx_dropped_too_small +
695 		drvs->rx_dropped_too_short +
696 		drvs->rx_dropped_header_too_small +
697 		drvs->rx_dropped_tcp_length +
698 		drvs->rx_dropped_runt;
699 
700 	/* detailed rx errors */
701 	stats->rx_length_errors = drvs->rx_in_range_errors +
702 		drvs->rx_out_range_errors +
703 		drvs->rx_frame_too_long;
704 
705 	stats->rx_crc_errors = drvs->rx_crc_errors;
706 
707 	/* frame alignment errors */
708 	stats->rx_frame_errors = drvs->rx_alignment_symbol_errors;
709 
710 	/* receiver fifo overrun */
711 	/* drops_no_pbuf is no per i/f, it's per BE card */
712 	stats->rx_fifo_errors = drvs->rxpp_fifo_overflow_drop +
713 				drvs->rx_input_fifo_overflow_drop +
714 				drvs->rx_drops_no_pbuf;
715 }
716 
717 void be_link_status_update(struct be_adapter *adapter, u8 link_status)
718 {
719 	struct net_device *netdev = adapter->netdev;
720 
721 	if (!(adapter->flags & BE_FLAGS_LINK_STATUS_INIT)) {
722 		netif_carrier_off(netdev);
723 		adapter->flags |= BE_FLAGS_LINK_STATUS_INIT;
724 	}
725 
726 	if (link_status)
727 		netif_carrier_on(netdev);
728 	else
729 		netif_carrier_off(netdev);
730 
731 	netdev_info(netdev, "Link is %s\n", link_status ? "Up" : "Down");
732 }
733 
734 static int be_gso_hdr_len(struct sk_buff *skb)
735 {
736 	if (skb->encapsulation)
737 		return skb_inner_transport_offset(skb) +
738 		       inner_tcp_hdrlen(skb);
739 	return skb_transport_offset(skb) + tcp_hdrlen(skb);
740 }
741 
742 static void be_tx_stats_update(struct be_tx_obj *txo, struct sk_buff *skb)
743 {
744 	struct be_tx_stats *stats = tx_stats(txo);
745 	u32 tx_pkts = skb_shinfo(skb)->gso_segs ? : 1;
746 	/* Account for headers which get duplicated in TSO pkt */
747 	u32 dup_hdr_len = tx_pkts > 1 ? be_gso_hdr_len(skb) * (tx_pkts - 1) : 0;
748 
749 	u64_stats_update_begin(&stats->sync);
750 	stats->tx_reqs++;
751 	stats->tx_bytes += skb->len + dup_hdr_len;
752 	stats->tx_pkts += tx_pkts;
753 	if (skb->encapsulation && skb->ip_summed == CHECKSUM_PARTIAL)
754 		stats->tx_vxlan_offload_pkts += tx_pkts;
755 	u64_stats_update_end(&stats->sync);
756 }
757 
758 /* Returns number of WRBs needed for the skb */
759 static u32 skb_wrb_cnt(struct sk_buff *skb)
760 {
761 	/* +1 for the header wrb */
762 	return 1 + (skb_headlen(skb) ? 1 : 0) + skb_shinfo(skb)->nr_frags;
763 }
764 
765 static inline void wrb_fill(struct be_eth_wrb *wrb, u64 addr, int len)
766 {
767 	wrb->frag_pa_hi = cpu_to_le32(upper_32_bits(addr));
768 	wrb->frag_pa_lo = cpu_to_le32(lower_32_bits(addr));
769 	wrb->frag_len = cpu_to_le32(len & ETH_WRB_FRAG_LEN_MASK);
770 	wrb->rsvd0 = 0;
771 }
772 
773 /* A dummy wrb is just all zeros. Using a separate routine for dummy-wrb
774  * to avoid the swap and shift/mask operations in wrb_fill().
775  */
776 static inline void wrb_fill_dummy(struct be_eth_wrb *wrb)
777 {
778 	wrb->frag_pa_hi = 0;
779 	wrb->frag_pa_lo = 0;
780 	wrb->frag_len = 0;
781 	wrb->rsvd0 = 0;
782 }
783 
784 static inline u16 be_get_tx_vlan_tag(struct be_adapter *adapter,
785 				     struct sk_buff *skb)
786 {
787 	u8 vlan_prio;
788 	u16 vlan_tag;
789 
790 	vlan_tag = skb_vlan_tag_get(skb);
791 	vlan_prio = (vlan_tag & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
792 	/* If vlan priority provided by OS is NOT in available bmap */
793 	if (!(adapter->vlan_prio_bmap & (1 << vlan_prio)))
794 		vlan_tag = (vlan_tag & ~VLAN_PRIO_MASK) |
795 				adapter->recommended_prio_bits;
796 
797 	return vlan_tag;
798 }
799 
800 /* Used only for IP tunnel packets */
801 static u16 skb_inner_ip_proto(struct sk_buff *skb)
802 {
803 	return (inner_ip_hdr(skb)->version == 4) ?
804 		inner_ip_hdr(skb)->protocol : inner_ipv6_hdr(skb)->nexthdr;
805 }
806 
807 static u16 skb_ip_proto(struct sk_buff *skb)
808 {
809 	return (ip_hdr(skb)->version == 4) ?
810 		ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
811 }
812 
813 static inline bool be_is_txq_full(struct be_tx_obj *txo)
814 {
815 	return atomic_read(&txo->q.used) + BE_MAX_TX_FRAG_COUNT >= txo->q.len;
816 }
817 
818 static inline bool be_can_txq_wake(struct be_tx_obj *txo)
819 {
820 	return atomic_read(&txo->q.used) < txo->q.len / 2;
821 }
822 
823 static inline bool be_is_tx_compl_pending(struct be_tx_obj *txo)
824 {
825 	return atomic_read(&txo->q.used) > txo->pend_wrb_cnt;
826 }
827 
828 static void be_get_wrb_params_from_skb(struct be_adapter *adapter,
829 				       struct sk_buff *skb,
830 				       struct be_wrb_params *wrb_params)
831 {
832 	u16 proto;
833 
834 	if (skb_is_gso(skb)) {
835 		BE_WRB_F_SET(wrb_params->features, LSO, 1);
836 		wrb_params->lso_mss = skb_shinfo(skb)->gso_size;
837 		if (skb_is_gso_v6(skb) && !lancer_chip(adapter))
838 			BE_WRB_F_SET(wrb_params->features, LSO6, 1);
839 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
840 		if (skb->encapsulation) {
841 			BE_WRB_F_SET(wrb_params->features, IPCS, 1);
842 			proto = skb_inner_ip_proto(skb);
843 		} else {
844 			proto = skb_ip_proto(skb);
845 		}
846 		if (proto == IPPROTO_TCP)
847 			BE_WRB_F_SET(wrb_params->features, TCPCS, 1);
848 		else if (proto == IPPROTO_UDP)
849 			BE_WRB_F_SET(wrb_params->features, UDPCS, 1);
850 	}
851 
852 	if (skb_vlan_tag_present(skb)) {
853 		BE_WRB_F_SET(wrb_params->features, VLAN, 1);
854 		wrb_params->vlan_tag = be_get_tx_vlan_tag(adapter, skb);
855 	}
856 
857 	BE_WRB_F_SET(wrb_params->features, CRC, 1);
858 }
859 
860 static void wrb_fill_hdr(struct be_adapter *adapter,
861 			 struct be_eth_hdr_wrb *hdr,
862 			 struct be_wrb_params *wrb_params,
863 			 struct sk_buff *skb)
864 {
865 	memset(hdr, 0, sizeof(*hdr));
866 
867 	SET_TX_WRB_HDR_BITS(crc, hdr,
868 			    BE_WRB_F_GET(wrb_params->features, CRC));
869 	SET_TX_WRB_HDR_BITS(ipcs, hdr,
870 			    BE_WRB_F_GET(wrb_params->features, IPCS));
871 	SET_TX_WRB_HDR_BITS(tcpcs, hdr,
872 			    BE_WRB_F_GET(wrb_params->features, TCPCS));
873 	SET_TX_WRB_HDR_BITS(udpcs, hdr,
874 			    BE_WRB_F_GET(wrb_params->features, UDPCS));
875 
876 	SET_TX_WRB_HDR_BITS(lso, hdr,
877 			    BE_WRB_F_GET(wrb_params->features, LSO));
878 	SET_TX_WRB_HDR_BITS(lso6, hdr,
879 			    BE_WRB_F_GET(wrb_params->features, LSO6));
880 	SET_TX_WRB_HDR_BITS(lso_mss, hdr, wrb_params->lso_mss);
881 
882 	/* Hack to skip HW VLAN tagging needs evt = 1, compl = 0. When this
883 	 * hack is not needed, the evt bit is set while ringing DB.
884 	 */
885 	SET_TX_WRB_HDR_BITS(event, hdr,
886 			    BE_WRB_F_GET(wrb_params->features, VLAN_SKIP_HW));
887 	SET_TX_WRB_HDR_BITS(vlan, hdr,
888 			    BE_WRB_F_GET(wrb_params->features, VLAN));
889 	SET_TX_WRB_HDR_BITS(vlan_tag, hdr, wrb_params->vlan_tag);
890 
891 	SET_TX_WRB_HDR_BITS(num_wrb, hdr, skb_wrb_cnt(skb));
892 	SET_TX_WRB_HDR_BITS(len, hdr, skb->len);
893 	SET_TX_WRB_HDR_BITS(mgmt, hdr,
894 			    BE_WRB_F_GET(wrb_params->features, OS2BMC));
895 }
896 
897 static void unmap_tx_frag(struct device *dev, struct be_eth_wrb *wrb,
898 			  bool unmap_single)
899 {
900 	dma_addr_t dma;
901 	u32 frag_len = le32_to_cpu(wrb->frag_len);
902 
903 
904 	dma = (u64)le32_to_cpu(wrb->frag_pa_hi) << 32 |
905 		(u64)le32_to_cpu(wrb->frag_pa_lo);
906 	if (frag_len) {
907 		if (unmap_single)
908 			dma_unmap_single(dev, dma, frag_len, DMA_TO_DEVICE);
909 		else
910 			dma_unmap_page(dev, dma, frag_len, DMA_TO_DEVICE);
911 	}
912 }
913 
914 /* Grab a WRB header for xmit */
915 static u32 be_tx_get_wrb_hdr(struct be_tx_obj *txo)
916 {
917 	u32 head = txo->q.head;
918 
919 	queue_head_inc(&txo->q);
920 	return head;
921 }
922 
923 /* Set up the WRB header for xmit */
924 static void be_tx_setup_wrb_hdr(struct be_adapter *adapter,
925 				struct be_tx_obj *txo,
926 				struct be_wrb_params *wrb_params,
927 				struct sk_buff *skb, u16 head)
928 {
929 	u32 num_frags = skb_wrb_cnt(skb);
930 	struct be_queue_info *txq = &txo->q;
931 	struct be_eth_hdr_wrb *hdr = queue_index_node(txq, head);
932 
933 	wrb_fill_hdr(adapter, hdr, wrb_params, skb);
934 	be_dws_cpu_to_le(hdr, sizeof(*hdr));
935 
936 	BUG_ON(txo->sent_skb_list[head]);
937 	txo->sent_skb_list[head] = skb;
938 	txo->last_req_hdr = head;
939 	atomic_add(num_frags, &txq->used);
940 	txo->last_req_wrb_cnt = num_frags;
941 	txo->pend_wrb_cnt += num_frags;
942 }
943 
944 /* Setup a WRB fragment (buffer descriptor) for xmit */
945 static void be_tx_setup_wrb_frag(struct be_tx_obj *txo, dma_addr_t busaddr,
946 				 int len)
947 {
948 	struct be_eth_wrb *wrb;
949 	struct be_queue_info *txq = &txo->q;
950 
951 	wrb = queue_head_node(txq);
952 	wrb_fill(wrb, busaddr, len);
953 	queue_head_inc(txq);
954 }
955 
956 /* Bring the queue back to the state it was in before be_xmit_enqueue() routine
957  * was invoked. The producer index is restored to the previous packet and the
958  * WRBs of the current packet are unmapped. Invoked to handle tx setup errors.
959  */
960 static void be_xmit_restore(struct be_adapter *adapter,
961 			    struct be_tx_obj *txo, u32 head, bool map_single,
962 			    u32 copied)
963 {
964 	struct device *dev;
965 	struct be_eth_wrb *wrb;
966 	struct be_queue_info *txq = &txo->q;
967 
968 	dev = &adapter->pdev->dev;
969 	txq->head = head;
970 
971 	/* skip the first wrb (hdr); it's not mapped */
972 	queue_head_inc(txq);
973 	while (copied) {
974 		wrb = queue_head_node(txq);
975 		unmap_tx_frag(dev, wrb, map_single);
976 		map_single = false;
977 		copied -= le32_to_cpu(wrb->frag_len);
978 		queue_head_inc(txq);
979 	}
980 
981 	txq->head = head;
982 }
983 
984 /* Enqueue the given packet for transmit. This routine allocates WRBs for the
985  * packet, dma maps the packet buffers and sets up the WRBs. Returns the number
986  * of WRBs used up by the packet.
987  */
988 static u32 be_xmit_enqueue(struct be_adapter *adapter, struct be_tx_obj *txo,
989 			   struct sk_buff *skb,
990 			   struct be_wrb_params *wrb_params)
991 {
992 	u32 i, copied = 0, wrb_cnt = skb_wrb_cnt(skb);
993 	struct device *dev = &adapter->pdev->dev;
994 	bool map_single = false;
995 	u32 head;
996 	dma_addr_t busaddr;
997 	int len;
998 
999 	head = be_tx_get_wrb_hdr(txo);
1000 
1001 	if (skb->len > skb->data_len) {
1002 		len = skb_headlen(skb);
1003 
1004 		busaddr = dma_map_single(dev, skb->data, len, DMA_TO_DEVICE);
1005 		if (dma_mapping_error(dev, busaddr))
1006 			goto dma_err;
1007 		map_single = true;
1008 		be_tx_setup_wrb_frag(txo, busaddr, len);
1009 		copied += len;
1010 	}
1011 
1012 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1013 		const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
1014 		len = skb_frag_size(frag);
1015 
1016 		busaddr = skb_frag_dma_map(dev, frag, 0, len, DMA_TO_DEVICE);
1017 		if (dma_mapping_error(dev, busaddr))
1018 			goto dma_err;
1019 		be_tx_setup_wrb_frag(txo, busaddr, len);
1020 		copied += len;
1021 	}
1022 
1023 	be_tx_setup_wrb_hdr(adapter, txo, wrb_params, skb, head);
1024 
1025 	be_tx_stats_update(txo, skb);
1026 	return wrb_cnt;
1027 
1028 dma_err:
1029 	adapter->drv_stats.dma_map_errors++;
1030 	be_xmit_restore(adapter, txo, head, map_single, copied);
1031 	return 0;
1032 }
1033 
1034 static inline int qnq_async_evt_rcvd(struct be_adapter *adapter)
1035 {
1036 	return adapter->flags & BE_FLAGS_QNQ_ASYNC_EVT_RCVD;
1037 }
1038 
1039 static struct sk_buff *be_insert_vlan_in_pkt(struct be_adapter *adapter,
1040 					     struct sk_buff *skb,
1041 					     struct be_wrb_params
1042 					     *wrb_params)
1043 {
1044 	u16 vlan_tag = 0;
1045 
1046 	skb = skb_share_check(skb, GFP_ATOMIC);
1047 	if (unlikely(!skb))
1048 		return skb;
1049 
1050 	if (skb_vlan_tag_present(skb))
1051 		vlan_tag = be_get_tx_vlan_tag(adapter, skb);
1052 
1053 	if (qnq_async_evt_rcvd(adapter) && adapter->pvid) {
1054 		if (!vlan_tag)
1055 			vlan_tag = adapter->pvid;
1056 		/* f/w workaround to set skip_hw_vlan = 1, informs the F/W to
1057 		 * skip VLAN insertion
1058 		 */
1059 		BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1060 	}
1061 
1062 	if (vlan_tag) {
1063 		skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1064 						vlan_tag);
1065 		if (unlikely(!skb))
1066 			return skb;
1067 		skb->vlan_tci = 0;
1068 	}
1069 
1070 	/* Insert the outer VLAN, if any */
1071 	if (adapter->qnq_vid) {
1072 		vlan_tag = adapter->qnq_vid;
1073 		skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1074 						vlan_tag);
1075 		if (unlikely(!skb))
1076 			return skb;
1077 		BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1078 	}
1079 
1080 	return skb;
1081 }
1082 
1083 static bool be_ipv6_exthdr_check(struct sk_buff *skb)
1084 {
1085 	struct ethhdr *eh = (struct ethhdr *)skb->data;
1086 	u16 offset = ETH_HLEN;
1087 
1088 	if (eh->h_proto == htons(ETH_P_IPV6)) {
1089 		struct ipv6hdr *ip6h = (struct ipv6hdr *)(skb->data + offset);
1090 
1091 		offset += sizeof(struct ipv6hdr);
1092 		if (ip6h->nexthdr != NEXTHDR_TCP &&
1093 		    ip6h->nexthdr != NEXTHDR_UDP) {
1094 			struct ipv6_opt_hdr *ehdr =
1095 				(struct ipv6_opt_hdr *)(skb->data + offset);
1096 
1097 			/* offending pkt: 2nd byte following IPv6 hdr is 0xff */
1098 			if (ehdr->hdrlen == 0xff)
1099 				return true;
1100 		}
1101 	}
1102 	return false;
1103 }
1104 
1105 static int be_vlan_tag_tx_chk(struct be_adapter *adapter, struct sk_buff *skb)
1106 {
1107 	return skb_vlan_tag_present(skb) || adapter->pvid || adapter->qnq_vid;
1108 }
1109 
1110 static int be_ipv6_tx_stall_chk(struct be_adapter *adapter, struct sk_buff *skb)
1111 {
1112 	return BE3_chip(adapter) && be_ipv6_exthdr_check(skb);
1113 }
1114 
1115 static struct sk_buff *be_lancer_xmit_workarounds(struct be_adapter *adapter,
1116 						  struct sk_buff *skb,
1117 						  struct be_wrb_params
1118 						  *wrb_params)
1119 {
1120 	struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1121 	unsigned int eth_hdr_len;
1122 	struct iphdr *ip;
1123 
1124 	/* For padded packets, BE HW modifies tot_len field in IP header
1125 	 * incorrecly when VLAN tag is inserted by HW.
1126 	 * For padded packets, Lancer computes incorrect checksum.
1127 	 */
1128 	eth_hdr_len = ntohs(skb->protocol) == ETH_P_8021Q ?
1129 						VLAN_ETH_HLEN : ETH_HLEN;
1130 	if (skb->len <= 60 &&
1131 	    (lancer_chip(adapter) || skb_vlan_tag_present(skb)) &&
1132 	    is_ipv4_pkt(skb)) {
1133 		ip = (struct iphdr *)ip_hdr(skb);
1134 		pskb_trim(skb, eth_hdr_len + ntohs(ip->tot_len));
1135 	}
1136 
1137 	/* If vlan tag is already inlined in the packet, skip HW VLAN
1138 	 * tagging in pvid-tagging mode
1139 	 */
1140 	if (be_pvid_tagging_enabled(adapter) &&
1141 	    veh->h_vlan_proto == htons(ETH_P_8021Q))
1142 		BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1143 
1144 	/* HW has a bug wherein it will calculate CSUM for VLAN
1145 	 * pkts even though it is disabled.
1146 	 * Manually insert VLAN in pkt.
1147 	 */
1148 	if (skb->ip_summed != CHECKSUM_PARTIAL &&
1149 	    skb_vlan_tag_present(skb)) {
1150 		skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1151 		if (unlikely(!skb))
1152 			goto err;
1153 	}
1154 
1155 	/* HW may lockup when VLAN HW tagging is requested on
1156 	 * certain ipv6 packets. Drop such pkts if the HW workaround to
1157 	 * skip HW tagging is not enabled by FW.
1158 	 */
1159 	if (unlikely(be_ipv6_tx_stall_chk(adapter, skb) &&
1160 		     (adapter->pvid || adapter->qnq_vid) &&
1161 		     !qnq_async_evt_rcvd(adapter)))
1162 		goto tx_drop;
1163 
1164 	/* Manual VLAN tag insertion to prevent:
1165 	 * ASIC lockup when the ASIC inserts VLAN tag into
1166 	 * certain ipv6 packets. Insert VLAN tags in driver,
1167 	 * and set event, completion, vlan bits accordingly
1168 	 * in the Tx WRB.
1169 	 */
1170 	if (be_ipv6_tx_stall_chk(adapter, skb) &&
1171 	    be_vlan_tag_tx_chk(adapter, skb)) {
1172 		skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1173 		if (unlikely(!skb))
1174 			goto err;
1175 	}
1176 
1177 	return skb;
1178 tx_drop:
1179 	dev_kfree_skb_any(skb);
1180 err:
1181 	return NULL;
1182 }
1183 
1184 static struct sk_buff *be_xmit_workarounds(struct be_adapter *adapter,
1185 					   struct sk_buff *skb,
1186 					   struct be_wrb_params *wrb_params)
1187 {
1188 	int err;
1189 
1190 	/* Lancer, SH and BE3 in SRIOV mode have a bug wherein
1191 	 * packets that are 32b or less may cause a transmit stall
1192 	 * on that port. The workaround is to pad such packets
1193 	 * (len <= 32 bytes) to a minimum length of 36b.
1194 	 */
1195 	if (skb->len <= 32) {
1196 		if (skb_put_padto(skb, 36))
1197 			return NULL;
1198 	}
1199 
1200 	if (BEx_chip(adapter) || lancer_chip(adapter)) {
1201 		skb = be_lancer_xmit_workarounds(adapter, skb, wrb_params);
1202 		if (!skb)
1203 			return NULL;
1204 	}
1205 
1206 	/* The stack can send us skbs with length greater than
1207 	 * what the HW can handle. Trim the extra bytes.
1208 	 */
1209 	WARN_ON_ONCE(skb->len > BE_MAX_GSO_SIZE);
1210 	err = pskb_trim(skb, BE_MAX_GSO_SIZE);
1211 	WARN_ON(err);
1212 
1213 	return skb;
1214 }
1215 
1216 static void be_xmit_flush(struct be_adapter *adapter, struct be_tx_obj *txo)
1217 {
1218 	struct be_queue_info *txq = &txo->q;
1219 	struct be_eth_hdr_wrb *hdr = queue_index_node(txq, txo->last_req_hdr);
1220 
1221 	/* Mark the last request eventable if it hasn't been marked already */
1222 	if (!(hdr->dw[2] & cpu_to_le32(TX_HDR_WRB_EVT)))
1223 		hdr->dw[2] |= cpu_to_le32(TX_HDR_WRB_EVT | TX_HDR_WRB_COMPL);
1224 
1225 	/* compose a dummy wrb if there are odd set of wrbs to notify */
1226 	if (!lancer_chip(adapter) && (txo->pend_wrb_cnt & 1)) {
1227 		wrb_fill_dummy(queue_head_node(txq));
1228 		queue_head_inc(txq);
1229 		atomic_inc(&txq->used);
1230 		txo->pend_wrb_cnt++;
1231 		hdr->dw[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK <<
1232 					   TX_HDR_WRB_NUM_SHIFT);
1233 		hdr->dw[2] |= cpu_to_le32((txo->last_req_wrb_cnt + 1) <<
1234 					  TX_HDR_WRB_NUM_SHIFT);
1235 	}
1236 	be_txq_notify(adapter, txo, txo->pend_wrb_cnt);
1237 	txo->pend_wrb_cnt = 0;
1238 }
1239 
1240 /* OS2BMC related */
1241 
1242 #define DHCP_CLIENT_PORT	68
1243 #define DHCP_SERVER_PORT	67
1244 #define NET_BIOS_PORT1		137
1245 #define NET_BIOS_PORT2		138
1246 #define DHCPV6_RAS_PORT		547
1247 
1248 #define is_mc_allowed_on_bmc(adapter, eh)	\
1249 	(!is_multicast_filt_enabled(adapter) &&	\
1250 	 is_multicast_ether_addr(eh->h_dest) &&	\
1251 	 !is_broadcast_ether_addr(eh->h_dest))
1252 
1253 #define is_bc_allowed_on_bmc(adapter, eh)	\
1254 	(!is_broadcast_filt_enabled(adapter) &&	\
1255 	 is_broadcast_ether_addr(eh->h_dest))
1256 
1257 #define is_arp_allowed_on_bmc(adapter, skb)	\
1258 	(is_arp(skb) && is_arp_filt_enabled(adapter))
1259 
1260 #define is_broadcast_packet(eh, adapter)	\
1261 		(is_multicast_ether_addr(eh->h_dest) && \
1262 		!compare_ether_addr(eh->h_dest, adapter->netdev->broadcast))
1263 
1264 #define is_arp(skb)	(skb->protocol == htons(ETH_P_ARP))
1265 
1266 #define is_arp_filt_enabled(adapter)	\
1267 		(adapter->bmc_filt_mask & (BMC_FILT_BROADCAST_ARP))
1268 
1269 #define is_dhcp_client_filt_enabled(adapter)	\
1270 		(adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_CLIENT)
1271 
1272 #define is_dhcp_srvr_filt_enabled(adapter)	\
1273 		(adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_SERVER)
1274 
1275 #define is_nbios_filt_enabled(adapter)	\
1276 		(adapter->bmc_filt_mask & BMC_FILT_BROADCAST_NET_BIOS)
1277 
1278 #define is_ipv6_na_filt_enabled(adapter)	\
1279 		(adapter->bmc_filt_mask &	\
1280 			BMC_FILT_MULTICAST_IPV6_NEIGH_ADVER)
1281 
1282 #define is_ipv6_ra_filt_enabled(adapter)	\
1283 		(adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RA)
1284 
1285 #define is_ipv6_ras_filt_enabled(adapter)	\
1286 		(adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RAS)
1287 
1288 #define is_broadcast_filt_enabled(adapter)	\
1289 		(adapter->bmc_filt_mask & BMC_FILT_BROADCAST)
1290 
1291 #define is_multicast_filt_enabled(adapter)	\
1292 		(adapter->bmc_filt_mask & BMC_FILT_MULTICAST)
1293 
1294 static bool be_send_pkt_to_bmc(struct be_adapter *adapter,
1295 			       struct sk_buff **skb)
1296 {
1297 	struct ethhdr *eh = (struct ethhdr *)(*skb)->data;
1298 	bool os2bmc = false;
1299 
1300 	if (!be_is_os2bmc_enabled(adapter))
1301 		goto done;
1302 
1303 	if (!is_multicast_ether_addr(eh->h_dest))
1304 		goto done;
1305 
1306 	if (is_mc_allowed_on_bmc(adapter, eh) ||
1307 	    is_bc_allowed_on_bmc(adapter, eh) ||
1308 	    is_arp_allowed_on_bmc(adapter, (*skb))) {
1309 		os2bmc = true;
1310 		goto done;
1311 	}
1312 
1313 	if ((*skb)->protocol == htons(ETH_P_IPV6)) {
1314 		struct ipv6hdr *hdr = ipv6_hdr((*skb));
1315 		u8 nexthdr = hdr->nexthdr;
1316 
1317 		if (nexthdr == IPPROTO_ICMPV6) {
1318 			struct icmp6hdr *icmp6 = icmp6_hdr((*skb));
1319 
1320 			switch (icmp6->icmp6_type) {
1321 			case NDISC_ROUTER_ADVERTISEMENT:
1322 				os2bmc = is_ipv6_ra_filt_enabled(adapter);
1323 				goto done;
1324 			case NDISC_NEIGHBOUR_ADVERTISEMENT:
1325 				os2bmc = is_ipv6_na_filt_enabled(adapter);
1326 				goto done;
1327 			default:
1328 				break;
1329 			}
1330 		}
1331 	}
1332 
1333 	if (is_udp_pkt((*skb))) {
1334 		struct udphdr *udp = udp_hdr((*skb));
1335 
1336 		switch (ntohs(udp->dest)) {
1337 		case DHCP_CLIENT_PORT:
1338 			os2bmc = is_dhcp_client_filt_enabled(adapter);
1339 			goto done;
1340 		case DHCP_SERVER_PORT:
1341 			os2bmc = is_dhcp_srvr_filt_enabled(adapter);
1342 			goto done;
1343 		case NET_BIOS_PORT1:
1344 		case NET_BIOS_PORT2:
1345 			os2bmc = is_nbios_filt_enabled(adapter);
1346 			goto done;
1347 		case DHCPV6_RAS_PORT:
1348 			os2bmc = is_ipv6_ras_filt_enabled(adapter);
1349 			goto done;
1350 		default:
1351 			break;
1352 		}
1353 	}
1354 done:
1355 	/* For packets over a vlan, which are destined
1356 	 * to BMC, asic expects the vlan to be inline in the packet.
1357 	 */
1358 	if (os2bmc)
1359 		*skb = be_insert_vlan_in_pkt(adapter, *skb, NULL);
1360 
1361 	return os2bmc;
1362 }
1363 
1364 static netdev_tx_t be_xmit(struct sk_buff *skb, struct net_device *netdev)
1365 {
1366 	struct be_adapter *adapter = netdev_priv(netdev);
1367 	u16 q_idx = skb_get_queue_mapping(skb);
1368 	struct be_tx_obj *txo = &adapter->tx_obj[q_idx];
1369 	struct be_wrb_params wrb_params = { 0 };
1370 	bool flush = !skb->xmit_more;
1371 	u16 wrb_cnt;
1372 
1373 	skb = be_xmit_workarounds(adapter, skb, &wrb_params);
1374 	if (unlikely(!skb))
1375 		goto drop;
1376 
1377 	be_get_wrb_params_from_skb(adapter, skb, &wrb_params);
1378 
1379 	wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1380 	if (unlikely(!wrb_cnt)) {
1381 		dev_kfree_skb_any(skb);
1382 		goto drop;
1383 	}
1384 
1385 	/* if os2bmc is enabled and if the pkt is destined to bmc,
1386 	 * enqueue the pkt a 2nd time with mgmt bit set.
1387 	 */
1388 	if (be_send_pkt_to_bmc(adapter, &skb)) {
1389 		BE_WRB_F_SET(wrb_params.features, OS2BMC, 1);
1390 		wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1391 		if (unlikely(!wrb_cnt))
1392 			goto drop;
1393 		else
1394 			skb_get(skb);
1395 	}
1396 
1397 	if (be_is_txq_full(txo)) {
1398 		netif_stop_subqueue(netdev, q_idx);
1399 		tx_stats(txo)->tx_stops++;
1400 	}
1401 
1402 	if (flush || __netif_subqueue_stopped(netdev, q_idx))
1403 		be_xmit_flush(adapter, txo);
1404 
1405 	return NETDEV_TX_OK;
1406 drop:
1407 	tx_stats(txo)->tx_drv_drops++;
1408 	/* Flush the already enqueued tx requests */
1409 	if (flush && txo->pend_wrb_cnt)
1410 		be_xmit_flush(adapter, txo);
1411 
1412 	return NETDEV_TX_OK;
1413 }
1414 
1415 static inline bool be_in_all_promisc(struct be_adapter *adapter)
1416 {
1417 	return (adapter->if_flags & BE_IF_FLAGS_ALL_PROMISCUOUS) ==
1418 			BE_IF_FLAGS_ALL_PROMISCUOUS;
1419 }
1420 
1421 static int be_set_vlan_promisc(struct be_adapter *adapter)
1422 {
1423 	struct device *dev = &adapter->pdev->dev;
1424 	int status;
1425 
1426 	if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS)
1427 		return 0;
1428 
1429 	status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, ON);
1430 	if (!status) {
1431 		dev_info(dev, "Enabled VLAN promiscuous mode\n");
1432 		adapter->if_flags |= BE_IF_FLAGS_VLAN_PROMISCUOUS;
1433 	} else {
1434 		dev_err(dev, "Failed to enable VLAN promiscuous mode\n");
1435 	}
1436 	return status;
1437 }
1438 
1439 static int be_clear_vlan_promisc(struct be_adapter *adapter)
1440 {
1441 	struct device *dev = &adapter->pdev->dev;
1442 	int status;
1443 
1444 	status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, OFF);
1445 	if (!status) {
1446 		dev_info(dev, "Disabling VLAN promiscuous mode\n");
1447 		adapter->if_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
1448 	}
1449 	return status;
1450 }
1451 
1452 /*
1453  * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE.
1454  * If the user configures more, place BE in vlan promiscuous mode.
1455  */
1456 static int be_vid_config(struct be_adapter *adapter)
1457 {
1458 	struct device *dev = &adapter->pdev->dev;
1459 	u16 vids[BE_NUM_VLANS_SUPPORTED];
1460 	u16 num = 0, i = 0;
1461 	int status = 0;
1462 
1463 	/* No need to change the VLAN state if the I/F is in promiscuous */
1464 	if (adapter->netdev->flags & IFF_PROMISC)
1465 		return 0;
1466 
1467 	if (adapter->vlans_added > be_max_vlans(adapter))
1468 		return be_set_vlan_promisc(adapter);
1469 
1470 	if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
1471 		status = be_clear_vlan_promisc(adapter);
1472 		if (status)
1473 			return status;
1474 	}
1475 	/* Construct VLAN Table to give to HW */
1476 	for_each_set_bit(i, adapter->vids, VLAN_N_VID)
1477 		vids[num++] = cpu_to_le16(i);
1478 
1479 	status = be_cmd_vlan_config(adapter, adapter->if_handle, vids, num, 0);
1480 	if (status) {
1481 		dev_err(dev, "Setting HW VLAN filtering failed\n");
1482 		/* Set to VLAN promisc mode as setting VLAN filter failed */
1483 		if (addl_status(status) == MCC_ADDL_STATUS_INSUFFICIENT_VLANS ||
1484 		    addl_status(status) ==
1485 				MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES)
1486 			return be_set_vlan_promisc(adapter);
1487 	}
1488 	return status;
1489 }
1490 
1491 static int be_vlan_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1492 {
1493 	struct be_adapter *adapter = netdev_priv(netdev);
1494 	int status = 0;
1495 
1496 	mutex_lock(&adapter->rx_filter_lock);
1497 
1498 	/* Packets with VID 0 are always received by Lancer by default */
1499 	if (lancer_chip(adapter) && vid == 0)
1500 		goto done;
1501 
1502 	if (test_bit(vid, adapter->vids))
1503 		goto done;
1504 
1505 	set_bit(vid, adapter->vids);
1506 	adapter->vlans_added++;
1507 
1508 	status = be_vid_config(adapter);
1509 done:
1510 	mutex_unlock(&adapter->rx_filter_lock);
1511 	return status;
1512 }
1513 
1514 static int be_vlan_rem_vid(struct net_device *netdev, __be16 proto, u16 vid)
1515 {
1516 	struct be_adapter *adapter = netdev_priv(netdev);
1517 	int status = 0;
1518 
1519 	mutex_lock(&adapter->rx_filter_lock);
1520 
1521 	/* Packets with VID 0 are always received by Lancer by default */
1522 	if (lancer_chip(adapter) && vid == 0)
1523 		goto done;
1524 
1525 	if (!test_bit(vid, adapter->vids))
1526 		goto done;
1527 
1528 	clear_bit(vid, adapter->vids);
1529 	adapter->vlans_added--;
1530 
1531 	status = be_vid_config(adapter);
1532 done:
1533 	mutex_unlock(&adapter->rx_filter_lock);
1534 	return status;
1535 }
1536 
1537 static void be_set_all_promisc(struct be_adapter *adapter)
1538 {
1539 	be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, ON);
1540 	adapter->if_flags |= BE_IF_FLAGS_ALL_PROMISCUOUS;
1541 }
1542 
1543 static void be_set_mc_promisc(struct be_adapter *adapter)
1544 {
1545 	int status;
1546 
1547 	if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS)
1548 		return;
1549 
1550 	status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MCAST_PROMISCUOUS, ON);
1551 	if (!status)
1552 		adapter->if_flags |= BE_IF_FLAGS_MCAST_PROMISCUOUS;
1553 }
1554 
1555 static void be_set_uc_promisc(struct be_adapter *adapter)
1556 {
1557 	int status;
1558 
1559 	if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS)
1560 		return;
1561 
1562 	status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, ON);
1563 	if (!status)
1564 		adapter->if_flags |= BE_IF_FLAGS_PROMISCUOUS;
1565 }
1566 
1567 static void be_clear_uc_promisc(struct be_adapter *adapter)
1568 {
1569 	int status;
1570 
1571 	if (!(adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS))
1572 		return;
1573 
1574 	status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, OFF);
1575 	if (!status)
1576 		adapter->if_flags &= ~BE_IF_FLAGS_PROMISCUOUS;
1577 }
1578 
1579 /* The below 2 functions are the callback args for __dev_mc_sync/dev_uc_sync().
1580  * We use a single callback function for both sync and unsync. We really don't
1581  * add/remove addresses through this callback. But, we use it to detect changes
1582  * to the uc/mc lists. The entire uc/mc list is programmed in be_set_rx_mode().
1583  */
1584 static int be_uc_list_update(struct net_device *netdev,
1585 			     const unsigned char *addr)
1586 {
1587 	struct be_adapter *adapter = netdev_priv(netdev);
1588 
1589 	adapter->update_uc_list = true;
1590 	return 0;
1591 }
1592 
1593 static int be_mc_list_update(struct net_device *netdev,
1594 			     const unsigned char *addr)
1595 {
1596 	struct be_adapter *adapter = netdev_priv(netdev);
1597 
1598 	adapter->update_mc_list = true;
1599 	return 0;
1600 }
1601 
1602 static void be_set_mc_list(struct be_adapter *adapter)
1603 {
1604 	struct net_device *netdev = adapter->netdev;
1605 	struct netdev_hw_addr *ha;
1606 	bool mc_promisc = false;
1607 	int status;
1608 
1609 	netif_addr_lock_bh(netdev);
1610 	__dev_mc_sync(netdev, be_mc_list_update, be_mc_list_update);
1611 
1612 	if (netdev->flags & IFF_PROMISC) {
1613 		adapter->update_mc_list = false;
1614 	} else if (netdev->flags & IFF_ALLMULTI ||
1615 		   netdev_mc_count(netdev) > be_max_mc(adapter)) {
1616 		/* Enable multicast promisc if num configured exceeds
1617 		 * what we support
1618 		 */
1619 		mc_promisc = true;
1620 		adapter->update_mc_list = false;
1621 	} else if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS) {
1622 		/* Update mc-list unconditionally if the iface was previously
1623 		 * in mc-promisc mode and now is out of that mode.
1624 		 */
1625 		adapter->update_mc_list = true;
1626 	}
1627 
1628 	if (adapter->update_mc_list) {
1629 		int i = 0;
1630 
1631 		/* cache the mc-list in adapter */
1632 		netdev_for_each_mc_addr(ha, netdev) {
1633 			ether_addr_copy(adapter->mc_list[i].mac, ha->addr);
1634 			i++;
1635 		}
1636 		adapter->mc_count = netdev_mc_count(netdev);
1637 	}
1638 	netif_addr_unlock_bh(netdev);
1639 
1640 	if (mc_promisc) {
1641 		be_set_mc_promisc(adapter);
1642 	} else if (adapter->update_mc_list) {
1643 		status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, ON);
1644 		if (!status)
1645 			adapter->if_flags &= ~BE_IF_FLAGS_MCAST_PROMISCUOUS;
1646 		else
1647 			be_set_mc_promisc(adapter);
1648 
1649 		adapter->update_mc_list = false;
1650 	}
1651 }
1652 
1653 static void be_clear_mc_list(struct be_adapter *adapter)
1654 {
1655 	struct net_device *netdev = adapter->netdev;
1656 
1657 	__dev_mc_unsync(netdev, NULL);
1658 	be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, OFF);
1659 	adapter->mc_count = 0;
1660 }
1661 
1662 static int be_uc_mac_add(struct be_adapter *adapter, int uc_idx)
1663 {
1664 	if (ether_addr_equal(adapter->uc_list[uc_idx].mac, adapter->dev_mac)) {
1665 		adapter->pmac_id[uc_idx + 1] = adapter->pmac_id[0];
1666 		return 0;
1667 	}
1668 
1669 	return be_cmd_pmac_add(adapter, adapter->uc_list[uc_idx].mac,
1670 			       adapter->if_handle,
1671 			       &adapter->pmac_id[uc_idx + 1], 0);
1672 }
1673 
1674 static void be_uc_mac_del(struct be_adapter *adapter, int pmac_id)
1675 {
1676 	if (pmac_id == adapter->pmac_id[0])
1677 		return;
1678 
1679 	be_cmd_pmac_del(adapter, adapter->if_handle, pmac_id, 0);
1680 }
1681 
1682 static void be_set_uc_list(struct be_adapter *adapter)
1683 {
1684 	struct net_device *netdev = adapter->netdev;
1685 	struct netdev_hw_addr *ha;
1686 	bool uc_promisc = false;
1687 	int curr_uc_macs = 0, i;
1688 
1689 	netif_addr_lock_bh(netdev);
1690 	__dev_uc_sync(netdev, be_uc_list_update, be_uc_list_update);
1691 
1692 	if (netdev->flags & IFF_PROMISC) {
1693 		adapter->update_uc_list = false;
1694 	} else if (netdev_uc_count(netdev) > (be_max_uc(adapter) - 1)) {
1695 		uc_promisc = true;
1696 		adapter->update_uc_list = false;
1697 	}  else if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS) {
1698 		/* Update uc-list unconditionally if the iface was previously
1699 		 * in uc-promisc mode and now is out of that mode.
1700 		 */
1701 		adapter->update_uc_list = true;
1702 	}
1703 
1704 	if (adapter->update_uc_list) {
1705 		/* cache the uc-list in adapter array */
1706 		i = 0;
1707 		netdev_for_each_uc_addr(ha, netdev) {
1708 			ether_addr_copy(adapter->uc_list[i].mac, ha->addr);
1709 			i++;
1710 		}
1711 		curr_uc_macs = netdev_uc_count(netdev);
1712 	}
1713 	netif_addr_unlock_bh(netdev);
1714 
1715 	if (uc_promisc) {
1716 		be_set_uc_promisc(adapter);
1717 	} else if (adapter->update_uc_list) {
1718 		be_clear_uc_promisc(adapter);
1719 
1720 		for (i = 0; i < adapter->uc_macs; i++)
1721 			be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1722 
1723 		for (i = 0; i < curr_uc_macs; i++)
1724 			be_uc_mac_add(adapter, i);
1725 		adapter->uc_macs = curr_uc_macs;
1726 		adapter->update_uc_list = false;
1727 	}
1728 }
1729 
1730 static void be_clear_uc_list(struct be_adapter *adapter)
1731 {
1732 	struct net_device *netdev = adapter->netdev;
1733 	int i;
1734 
1735 	__dev_uc_unsync(netdev, NULL);
1736 	for (i = 0; i < adapter->uc_macs; i++)
1737 		be_uc_mac_del(adapter, adapter->pmac_id[i + 1]);
1738 
1739 	adapter->uc_macs = 0;
1740 }
1741 
1742 static void __be_set_rx_mode(struct be_adapter *adapter)
1743 {
1744 	struct net_device *netdev = adapter->netdev;
1745 
1746 	mutex_lock(&adapter->rx_filter_lock);
1747 
1748 	if (netdev->flags & IFF_PROMISC) {
1749 		if (!be_in_all_promisc(adapter))
1750 			be_set_all_promisc(adapter);
1751 	} else if (be_in_all_promisc(adapter)) {
1752 		/* We need to re-program the vlan-list or clear
1753 		 * vlan-promisc mode (if needed) when the interface
1754 		 * comes out of promisc mode.
1755 		 */
1756 		be_vid_config(adapter);
1757 	}
1758 
1759 	be_set_uc_list(adapter);
1760 	be_set_mc_list(adapter);
1761 
1762 	mutex_unlock(&adapter->rx_filter_lock);
1763 }
1764 
1765 static void be_work_set_rx_mode(struct work_struct *work)
1766 {
1767 	struct be_cmd_work *cmd_work =
1768 				container_of(work, struct be_cmd_work, work);
1769 
1770 	__be_set_rx_mode(cmd_work->adapter);
1771 	kfree(cmd_work);
1772 }
1773 
1774 static int be_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1775 {
1776 	struct be_adapter *adapter = netdev_priv(netdev);
1777 	struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1778 	int status;
1779 
1780 	if (!sriov_enabled(adapter))
1781 		return -EPERM;
1782 
1783 	if (!is_valid_ether_addr(mac) || vf >= adapter->num_vfs)
1784 		return -EINVAL;
1785 
1786 	/* Proceed further only if user provided MAC is different
1787 	 * from active MAC
1788 	 */
1789 	if (ether_addr_equal(mac, vf_cfg->mac_addr))
1790 		return 0;
1791 
1792 	if (BEx_chip(adapter)) {
1793 		be_cmd_pmac_del(adapter, vf_cfg->if_handle, vf_cfg->pmac_id,
1794 				vf + 1);
1795 
1796 		status = be_cmd_pmac_add(adapter, mac, vf_cfg->if_handle,
1797 					 &vf_cfg->pmac_id, vf + 1);
1798 	} else {
1799 		status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
1800 					vf + 1);
1801 	}
1802 
1803 	if (status) {
1804 		dev_err(&adapter->pdev->dev, "MAC %pM set on VF %d Failed: %#x",
1805 			mac, vf, status);
1806 		return be_cmd_status(status);
1807 	}
1808 
1809 	ether_addr_copy(vf_cfg->mac_addr, mac);
1810 
1811 	return 0;
1812 }
1813 
1814 static int be_get_vf_config(struct net_device *netdev, int vf,
1815 			    struct ifla_vf_info *vi)
1816 {
1817 	struct be_adapter *adapter = netdev_priv(netdev);
1818 	struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1819 
1820 	if (!sriov_enabled(adapter))
1821 		return -EPERM;
1822 
1823 	if (vf >= adapter->num_vfs)
1824 		return -EINVAL;
1825 
1826 	vi->vf = vf;
1827 	vi->max_tx_rate = vf_cfg->tx_rate;
1828 	vi->min_tx_rate = 0;
1829 	vi->vlan = vf_cfg->vlan_tag & VLAN_VID_MASK;
1830 	vi->qos = vf_cfg->vlan_tag >> VLAN_PRIO_SHIFT;
1831 	memcpy(&vi->mac, vf_cfg->mac_addr, ETH_ALEN);
1832 	vi->linkstate = adapter->vf_cfg[vf].plink_tracking;
1833 	vi->spoofchk = adapter->vf_cfg[vf].spoofchk;
1834 
1835 	return 0;
1836 }
1837 
1838 static int be_set_vf_tvt(struct be_adapter *adapter, int vf, u16 vlan)
1839 {
1840 	struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1841 	u16 vids[BE_NUM_VLANS_SUPPORTED];
1842 	int vf_if_id = vf_cfg->if_handle;
1843 	int status;
1844 
1845 	/* Enable Transparent VLAN Tagging */
1846 	status = be_cmd_set_hsw_config(adapter, vlan, vf + 1, vf_if_id, 0, 0);
1847 	if (status)
1848 		return status;
1849 
1850 	/* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */
1851 	vids[0] = 0;
1852 	status = be_cmd_vlan_config(adapter, vf_if_id, vids, 1, vf + 1);
1853 	if (!status)
1854 		dev_info(&adapter->pdev->dev,
1855 			 "Cleared guest VLANs on VF%d", vf);
1856 
1857 	/* After TVT is enabled, disallow VFs to program VLAN filters */
1858 	if (vf_cfg->privileges & BE_PRIV_FILTMGMT) {
1859 		status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges &
1860 						  ~BE_PRIV_FILTMGMT, vf + 1);
1861 		if (!status)
1862 			vf_cfg->privileges &= ~BE_PRIV_FILTMGMT;
1863 	}
1864 	return 0;
1865 }
1866 
1867 static int be_clear_vf_tvt(struct be_adapter *adapter, int vf)
1868 {
1869 	struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1870 	struct device *dev = &adapter->pdev->dev;
1871 	int status;
1872 
1873 	/* Reset Transparent VLAN Tagging. */
1874 	status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID, vf + 1,
1875 				       vf_cfg->if_handle, 0, 0);
1876 	if (status)
1877 		return status;
1878 
1879 	/* Allow VFs to program VLAN filtering */
1880 	if (!(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
1881 		status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges |
1882 						  BE_PRIV_FILTMGMT, vf + 1);
1883 		if (!status) {
1884 			vf_cfg->privileges |= BE_PRIV_FILTMGMT;
1885 			dev_info(dev, "VF%d: FILTMGMT priv enabled", vf);
1886 		}
1887 	}
1888 
1889 	dev_info(dev,
1890 		 "Disable/re-enable i/f in VM to clear Transparent VLAN tag");
1891 	return 0;
1892 }
1893 
1894 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos,
1895 			  __be16 vlan_proto)
1896 {
1897 	struct be_adapter *adapter = netdev_priv(netdev);
1898 	struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1899 	int status;
1900 
1901 	if (!sriov_enabled(adapter))
1902 		return -EPERM;
1903 
1904 	if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7)
1905 		return -EINVAL;
1906 
1907 	if (vlan_proto != htons(ETH_P_8021Q))
1908 		return -EPROTONOSUPPORT;
1909 
1910 	if (vlan || qos) {
1911 		vlan |= qos << VLAN_PRIO_SHIFT;
1912 		status = be_set_vf_tvt(adapter, vf, vlan);
1913 	} else {
1914 		status = be_clear_vf_tvt(adapter, vf);
1915 	}
1916 
1917 	if (status) {
1918 		dev_err(&adapter->pdev->dev,
1919 			"VLAN %d config on VF %d failed : %#x\n", vlan, vf,
1920 			status);
1921 		return be_cmd_status(status);
1922 	}
1923 
1924 	vf_cfg->vlan_tag = vlan;
1925 	return 0;
1926 }
1927 
1928 static int be_set_vf_tx_rate(struct net_device *netdev, int vf,
1929 			     int min_tx_rate, int max_tx_rate)
1930 {
1931 	struct be_adapter *adapter = netdev_priv(netdev);
1932 	struct device *dev = &adapter->pdev->dev;
1933 	int percent_rate, status = 0;
1934 	u16 link_speed = 0;
1935 	u8 link_status;
1936 
1937 	if (!sriov_enabled(adapter))
1938 		return -EPERM;
1939 
1940 	if (vf >= adapter->num_vfs)
1941 		return -EINVAL;
1942 
1943 	if (min_tx_rate)
1944 		return -EINVAL;
1945 
1946 	if (!max_tx_rate)
1947 		goto config_qos;
1948 
1949 	status = be_cmd_link_status_query(adapter, &link_speed,
1950 					  &link_status, 0);
1951 	if (status)
1952 		goto err;
1953 
1954 	if (!link_status) {
1955 		dev_err(dev, "TX-rate setting not allowed when link is down\n");
1956 		status = -ENETDOWN;
1957 		goto err;
1958 	}
1959 
1960 	if (max_tx_rate < 100 || max_tx_rate > link_speed) {
1961 		dev_err(dev, "TX-rate must be between 100 and %d Mbps\n",
1962 			link_speed);
1963 		status = -EINVAL;
1964 		goto err;
1965 	}
1966 
1967 	/* On Skyhawk the QOS setting must be done only as a % value */
1968 	percent_rate = link_speed / 100;
1969 	if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) {
1970 		dev_err(dev, "TX-rate must be a multiple of %d Mbps\n",
1971 			percent_rate);
1972 		status = -EINVAL;
1973 		goto err;
1974 	}
1975 
1976 config_qos:
1977 	status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1);
1978 	if (status)
1979 		goto err;
1980 
1981 	adapter->vf_cfg[vf].tx_rate = max_tx_rate;
1982 	return 0;
1983 
1984 err:
1985 	dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n",
1986 		max_tx_rate, vf);
1987 	return be_cmd_status(status);
1988 }
1989 
1990 static int be_set_vf_link_state(struct net_device *netdev, int vf,
1991 				int link_state)
1992 {
1993 	struct be_adapter *adapter = netdev_priv(netdev);
1994 	int status;
1995 
1996 	if (!sriov_enabled(adapter))
1997 		return -EPERM;
1998 
1999 	if (vf >= adapter->num_vfs)
2000 		return -EINVAL;
2001 
2002 	status = be_cmd_set_logical_link_config(adapter, link_state, vf+1);
2003 	if (status) {
2004 		dev_err(&adapter->pdev->dev,
2005 			"Link state change on VF %d failed: %#x\n", vf, status);
2006 		return be_cmd_status(status);
2007 	}
2008 
2009 	adapter->vf_cfg[vf].plink_tracking = link_state;
2010 
2011 	return 0;
2012 }
2013 
2014 static int be_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
2015 {
2016 	struct be_adapter *adapter = netdev_priv(netdev);
2017 	struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
2018 	u8 spoofchk;
2019 	int status;
2020 
2021 	if (!sriov_enabled(adapter))
2022 		return -EPERM;
2023 
2024 	if (vf >= adapter->num_vfs)
2025 		return -EINVAL;
2026 
2027 	if (BEx_chip(adapter))
2028 		return -EOPNOTSUPP;
2029 
2030 	if (enable == vf_cfg->spoofchk)
2031 		return 0;
2032 
2033 	spoofchk = enable ? ENABLE_MAC_SPOOFCHK : DISABLE_MAC_SPOOFCHK;
2034 
2035 	status = be_cmd_set_hsw_config(adapter, 0, vf + 1, vf_cfg->if_handle,
2036 				       0, spoofchk);
2037 	if (status) {
2038 		dev_err(&adapter->pdev->dev,
2039 			"Spoofchk change on VF %d failed: %#x\n", vf, status);
2040 		return be_cmd_status(status);
2041 	}
2042 
2043 	vf_cfg->spoofchk = enable;
2044 	return 0;
2045 }
2046 
2047 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts,
2048 			  ulong now)
2049 {
2050 	aic->rx_pkts_prev = rx_pkts;
2051 	aic->tx_reqs_prev = tx_pkts;
2052 	aic->jiffies = now;
2053 }
2054 
2055 static int be_get_new_eqd(struct be_eq_obj *eqo)
2056 {
2057 	struct be_adapter *adapter = eqo->adapter;
2058 	int eqd, start;
2059 	struct be_aic_obj *aic;
2060 	struct be_rx_obj *rxo;
2061 	struct be_tx_obj *txo;
2062 	u64 rx_pkts = 0, tx_pkts = 0;
2063 	ulong now;
2064 	u32 pps, delta;
2065 	int i;
2066 
2067 	aic = &adapter->aic_obj[eqo->idx];
2068 	if (!aic->enable) {
2069 		if (aic->jiffies)
2070 			aic->jiffies = 0;
2071 		eqd = aic->et_eqd;
2072 		return eqd;
2073 	}
2074 
2075 	for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
2076 		do {
2077 			start = u64_stats_fetch_begin_irq(&rxo->stats.sync);
2078 			rx_pkts += rxo->stats.rx_pkts;
2079 		} while (u64_stats_fetch_retry_irq(&rxo->stats.sync, start));
2080 	}
2081 
2082 	for_all_tx_queues_on_eq(adapter, eqo, txo, i) {
2083 		do {
2084 			start = u64_stats_fetch_begin_irq(&txo->stats.sync);
2085 			tx_pkts += txo->stats.tx_reqs;
2086 		} while (u64_stats_fetch_retry_irq(&txo->stats.sync, start));
2087 	}
2088 
2089 	/* Skip, if wrapped around or first calculation */
2090 	now = jiffies;
2091 	if (!aic->jiffies || time_before(now, aic->jiffies) ||
2092 	    rx_pkts < aic->rx_pkts_prev ||
2093 	    tx_pkts < aic->tx_reqs_prev) {
2094 		be_aic_update(aic, rx_pkts, tx_pkts, now);
2095 		return aic->prev_eqd;
2096 	}
2097 
2098 	delta = jiffies_to_msecs(now - aic->jiffies);
2099 	if (delta == 0)
2100 		return aic->prev_eqd;
2101 
2102 	pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) +
2103 		(((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta);
2104 	eqd = (pps / 15000) << 2;
2105 
2106 	if (eqd < 8)
2107 		eqd = 0;
2108 	eqd = min_t(u32, eqd, aic->max_eqd);
2109 	eqd = max_t(u32, eqd, aic->min_eqd);
2110 
2111 	be_aic_update(aic, rx_pkts, tx_pkts, now);
2112 
2113 	return eqd;
2114 }
2115 
2116 /* For Skyhawk-R only */
2117 static u32 be_get_eq_delay_mult_enc(struct be_eq_obj *eqo)
2118 {
2119 	struct be_adapter *adapter = eqo->adapter;
2120 	struct be_aic_obj *aic = &adapter->aic_obj[eqo->idx];
2121 	ulong now = jiffies;
2122 	int eqd;
2123 	u32 mult_enc;
2124 
2125 	if (!aic->enable)
2126 		return 0;
2127 
2128 	if (jiffies_to_msecs(now - aic->jiffies) < 1)
2129 		eqd = aic->prev_eqd;
2130 	else
2131 		eqd = be_get_new_eqd(eqo);
2132 
2133 	if (eqd > 100)
2134 		mult_enc = R2I_DLY_ENC_1;
2135 	else if (eqd > 60)
2136 		mult_enc = R2I_DLY_ENC_2;
2137 	else if (eqd > 20)
2138 		mult_enc = R2I_DLY_ENC_3;
2139 	else
2140 		mult_enc = R2I_DLY_ENC_0;
2141 
2142 	aic->prev_eqd = eqd;
2143 
2144 	return mult_enc;
2145 }
2146 
2147 void be_eqd_update(struct be_adapter *adapter, bool force_update)
2148 {
2149 	struct be_set_eqd set_eqd[MAX_EVT_QS];
2150 	struct be_aic_obj *aic;
2151 	struct be_eq_obj *eqo;
2152 	int i, num = 0, eqd;
2153 
2154 	for_all_evt_queues(adapter, eqo, i) {
2155 		aic = &adapter->aic_obj[eqo->idx];
2156 		eqd = be_get_new_eqd(eqo);
2157 		if (force_update || eqd != aic->prev_eqd) {
2158 			set_eqd[num].delay_multiplier = (eqd * 65)/100;
2159 			set_eqd[num].eq_id = eqo->q.id;
2160 			aic->prev_eqd = eqd;
2161 			num++;
2162 		}
2163 	}
2164 
2165 	if (num)
2166 		be_cmd_modify_eqd(adapter, set_eqd, num);
2167 }
2168 
2169 static void be_rx_stats_update(struct be_rx_obj *rxo,
2170 			       struct be_rx_compl_info *rxcp)
2171 {
2172 	struct be_rx_stats *stats = rx_stats(rxo);
2173 
2174 	u64_stats_update_begin(&stats->sync);
2175 	stats->rx_compl++;
2176 	stats->rx_bytes += rxcp->pkt_size;
2177 	stats->rx_pkts++;
2178 	if (rxcp->tunneled)
2179 		stats->rx_vxlan_offload_pkts++;
2180 	if (rxcp->pkt_type == BE_MULTICAST_PACKET)
2181 		stats->rx_mcast_pkts++;
2182 	if (rxcp->err)
2183 		stats->rx_compl_err++;
2184 	u64_stats_update_end(&stats->sync);
2185 }
2186 
2187 static inline bool csum_passed(struct be_rx_compl_info *rxcp)
2188 {
2189 	/* L4 checksum is not reliable for non TCP/UDP packets.
2190 	 * Also ignore ipcksm for ipv6 pkts
2191 	 */
2192 	return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum &&
2193 		(rxcp->ip_csum || rxcp->ipv6) && !rxcp->err;
2194 }
2195 
2196 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo)
2197 {
2198 	struct be_adapter *adapter = rxo->adapter;
2199 	struct be_rx_page_info *rx_page_info;
2200 	struct be_queue_info *rxq = &rxo->q;
2201 	u32 frag_idx = rxq->tail;
2202 
2203 	rx_page_info = &rxo->page_info_tbl[frag_idx];
2204 	BUG_ON(!rx_page_info->page);
2205 
2206 	if (rx_page_info->last_frag) {
2207 		dma_unmap_page(&adapter->pdev->dev,
2208 			       dma_unmap_addr(rx_page_info, bus),
2209 			       adapter->big_page_size, DMA_FROM_DEVICE);
2210 		rx_page_info->last_frag = false;
2211 	} else {
2212 		dma_sync_single_for_cpu(&adapter->pdev->dev,
2213 					dma_unmap_addr(rx_page_info, bus),
2214 					rx_frag_size, DMA_FROM_DEVICE);
2215 	}
2216 
2217 	queue_tail_inc(rxq);
2218 	atomic_dec(&rxq->used);
2219 	return rx_page_info;
2220 }
2221 
2222 /* Throwaway the data in the Rx completion */
2223 static void be_rx_compl_discard(struct be_rx_obj *rxo,
2224 				struct be_rx_compl_info *rxcp)
2225 {
2226 	struct be_rx_page_info *page_info;
2227 	u16 i, num_rcvd = rxcp->num_rcvd;
2228 
2229 	for (i = 0; i < num_rcvd; i++) {
2230 		page_info = get_rx_page_info(rxo);
2231 		put_page(page_info->page);
2232 		memset(page_info, 0, sizeof(*page_info));
2233 	}
2234 }
2235 
2236 /*
2237  * skb_fill_rx_data forms a complete skb for an ether frame
2238  * indicated by rxcp.
2239  */
2240 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
2241 			     struct be_rx_compl_info *rxcp)
2242 {
2243 	struct be_rx_page_info *page_info;
2244 	u16 i, j;
2245 	u16 hdr_len, curr_frag_len, remaining;
2246 	u8 *start;
2247 
2248 	page_info = get_rx_page_info(rxo);
2249 	start = page_address(page_info->page) + page_info->page_offset;
2250 	prefetch(start);
2251 
2252 	/* Copy data in the first descriptor of this completion */
2253 	curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
2254 
2255 	skb->len = curr_frag_len;
2256 	if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
2257 		memcpy(skb->data, start, curr_frag_len);
2258 		/* Complete packet has now been moved to data */
2259 		put_page(page_info->page);
2260 		skb->data_len = 0;
2261 		skb->tail += curr_frag_len;
2262 	} else {
2263 		hdr_len = ETH_HLEN;
2264 		memcpy(skb->data, start, hdr_len);
2265 		skb_shinfo(skb)->nr_frags = 1;
2266 		skb_frag_set_page(skb, 0, page_info->page);
2267 		skb_shinfo(skb)->frags[0].page_offset =
2268 					page_info->page_offset + hdr_len;
2269 		skb_frag_size_set(&skb_shinfo(skb)->frags[0],
2270 				  curr_frag_len - hdr_len);
2271 		skb->data_len = curr_frag_len - hdr_len;
2272 		skb->truesize += rx_frag_size;
2273 		skb->tail += hdr_len;
2274 	}
2275 	page_info->page = NULL;
2276 
2277 	if (rxcp->pkt_size <= rx_frag_size) {
2278 		BUG_ON(rxcp->num_rcvd != 1);
2279 		return;
2280 	}
2281 
2282 	/* More frags present for this completion */
2283 	remaining = rxcp->pkt_size - curr_frag_len;
2284 	for (i = 1, j = 0; i < rxcp->num_rcvd; i++) {
2285 		page_info = get_rx_page_info(rxo);
2286 		curr_frag_len = min(remaining, rx_frag_size);
2287 
2288 		/* Coalesce all frags from the same physical page in one slot */
2289 		if (page_info->page_offset == 0) {
2290 			/* Fresh page */
2291 			j++;
2292 			skb_frag_set_page(skb, j, page_info->page);
2293 			skb_shinfo(skb)->frags[j].page_offset =
2294 							page_info->page_offset;
2295 			skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2296 			skb_shinfo(skb)->nr_frags++;
2297 		} else {
2298 			put_page(page_info->page);
2299 		}
2300 
2301 		skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2302 		skb->len += curr_frag_len;
2303 		skb->data_len += curr_frag_len;
2304 		skb->truesize += rx_frag_size;
2305 		remaining -= curr_frag_len;
2306 		page_info->page = NULL;
2307 	}
2308 	BUG_ON(j > MAX_SKB_FRAGS);
2309 }
2310 
2311 /* Process the RX completion indicated by rxcp when GRO is disabled */
2312 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi,
2313 				struct be_rx_compl_info *rxcp)
2314 {
2315 	struct be_adapter *adapter = rxo->adapter;
2316 	struct net_device *netdev = adapter->netdev;
2317 	struct sk_buff *skb;
2318 
2319 	skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE);
2320 	if (unlikely(!skb)) {
2321 		rx_stats(rxo)->rx_drops_no_skbs++;
2322 		be_rx_compl_discard(rxo, rxcp);
2323 		return;
2324 	}
2325 
2326 	skb_fill_rx_data(rxo, skb, rxcp);
2327 
2328 	if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp)))
2329 		skb->ip_summed = CHECKSUM_UNNECESSARY;
2330 	else
2331 		skb_checksum_none_assert(skb);
2332 
2333 	skb->protocol = eth_type_trans(skb, netdev);
2334 	skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2335 	if (netdev->features & NETIF_F_RXHASH)
2336 		skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2337 
2338 	skb->csum_level = rxcp->tunneled;
2339 	skb_mark_napi_id(skb, napi);
2340 
2341 	if (rxcp->vlanf)
2342 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2343 
2344 	netif_receive_skb(skb);
2345 }
2346 
2347 /* Process the RX completion indicated by rxcp when GRO is enabled */
2348 static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
2349 				    struct napi_struct *napi,
2350 				    struct be_rx_compl_info *rxcp)
2351 {
2352 	struct be_adapter *adapter = rxo->adapter;
2353 	struct be_rx_page_info *page_info;
2354 	struct sk_buff *skb = NULL;
2355 	u16 remaining, curr_frag_len;
2356 	u16 i, j;
2357 
2358 	skb = napi_get_frags(napi);
2359 	if (!skb) {
2360 		be_rx_compl_discard(rxo, rxcp);
2361 		return;
2362 	}
2363 
2364 	remaining = rxcp->pkt_size;
2365 	for (i = 0, j = -1; i < rxcp->num_rcvd; i++) {
2366 		page_info = get_rx_page_info(rxo);
2367 
2368 		curr_frag_len = min(remaining, rx_frag_size);
2369 
2370 		/* Coalesce all frags from the same physical page in one slot */
2371 		if (i == 0 || page_info->page_offset == 0) {
2372 			/* First frag or Fresh page */
2373 			j++;
2374 			skb_frag_set_page(skb, j, page_info->page);
2375 			skb_shinfo(skb)->frags[j].page_offset =
2376 							page_info->page_offset;
2377 			skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2378 		} else {
2379 			put_page(page_info->page);
2380 		}
2381 		skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2382 		skb->truesize += rx_frag_size;
2383 		remaining -= curr_frag_len;
2384 		memset(page_info, 0, sizeof(*page_info));
2385 	}
2386 	BUG_ON(j > MAX_SKB_FRAGS);
2387 
2388 	skb_shinfo(skb)->nr_frags = j + 1;
2389 	skb->len = rxcp->pkt_size;
2390 	skb->data_len = rxcp->pkt_size;
2391 	skb->ip_summed = CHECKSUM_UNNECESSARY;
2392 	skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2393 	if (adapter->netdev->features & NETIF_F_RXHASH)
2394 		skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2395 
2396 	skb->csum_level = rxcp->tunneled;
2397 
2398 	if (rxcp->vlanf)
2399 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2400 
2401 	napi_gro_frags(napi);
2402 }
2403 
2404 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl,
2405 				 struct be_rx_compl_info *rxcp)
2406 {
2407 	rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl);
2408 	rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl);
2409 	rxcp->err = GET_RX_COMPL_V1_BITS(err, compl);
2410 	rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl);
2411 	rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl);
2412 	rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl);
2413 	rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl);
2414 	rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl);
2415 	rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl);
2416 	rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl);
2417 	rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl);
2418 	if (rxcp->vlanf) {
2419 		rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl);
2420 		rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl);
2421 	}
2422 	rxcp->port = GET_RX_COMPL_V1_BITS(port, compl);
2423 	rxcp->tunneled =
2424 		GET_RX_COMPL_V1_BITS(tunneled, compl);
2425 }
2426 
2427 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl,
2428 				 struct be_rx_compl_info *rxcp)
2429 {
2430 	rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl);
2431 	rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl);
2432 	rxcp->err = GET_RX_COMPL_V0_BITS(err, compl);
2433 	rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl);
2434 	rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl);
2435 	rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl);
2436 	rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl);
2437 	rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl);
2438 	rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl);
2439 	rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl);
2440 	rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl);
2441 	if (rxcp->vlanf) {
2442 		rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl);
2443 		rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl);
2444 	}
2445 	rxcp->port = GET_RX_COMPL_V0_BITS(port, compl);
2446 	rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl);
2447 }
2448 
2449 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo)
2450 {
2451 	struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq);
2452 	struct be_rx_compl_info *rxcp = &rxo->rxcp;
2453 	struct be_adapter *adapter = rxo->adapter;
2454 
2455 	/* For checking the valid bit it is Ok to use either definition as the
2456 	 * valid bit is at the same position in both v0 and v1 Rx compl */
2457 	if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0)
2458 		return NULL;
2459 
2460 	rmb();
2461 	be_dws_le_to_cpu(compl, sizeof(*compl));
2462 
2463 	if (adapter->be3_native)
2464 		be_parse_rx_compl_v1(compl, rxcp);
2465 	else
2466 		be_parse_rx_compl_v0(compl, rxcp);
2467 
2468 	if (rxcp->ip_frag)
2469 		rxcp->l4_csum = 0;
2470 
2471 	if (rxcp->vlanf) {
2472 		/* In QNQ modes, if qnq bit is not set, then the packet was
2473 		 * tagged only with the transparent outer vlan-tag and must
2474 		 * not be treated as a vlan packet by host
2475 		 */
2476 		if (be_is_qnq_mode(adapter) && !rxcp->qnq)
2477 			rxcp->vlanf = 0;
2478 
2479 		if (!lancer_chip(adapter))
2480 			rxcp->vlan_tag = swab16(rxcp->vlan_tag);
2481 
2482 		if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) &&
2483 		    !test_bit(rxcp->vlan_tag, adapter->vids))
2484 			rxcp->vlanf = 0;
2485 	}
2486 
2487 	/* As the compl has been parsed, reset it; we wont touch it again */
2488 	compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0;
2489 
2490 	queue_tail_inc(&rxo->cq);
2491 	return rxcp;
2492 }
2493 
2494 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp)
2495 {
2496 	u32 order = get_order(size);
2497 
2498 	if (order > 0)
2499 		gfp |= __GFP_COMP;
2500 	return  alloc_pages(gfp, order);
2501 }
2502 
2503 /*
2504  * Allocate a page, split it to fragments of size rx_frag_size and post as
2505  * receive buffers to BE
2506  */
2507 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed)
2508 {
2509 	struct be_adapter *adapter = rxo->adapter;
2510 	struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL;
2511 	struct be_queue_info *rxq = &rxo->q;
2512 	struct page *pagep = NULL;
2513 	struct device *dev = &adapter->pdev->dev;
2514 	struct be_eth_rx_d *rxd;
2515 	u64 page_dmaaddr = 0, frag_dmaaddr;
2516 	u32 posted, page_offset = 0, notify = 0;
2517 
2518 	page_info = &rxo->page_info_tbl[rxq->head];
2519 	for (posted = 0; posted < frags_needed && !page_info->page; posted++) {
2520 		if (!pagep) {
2521 			pagep = be_alloc_pages(adapter->big_page_size, gfp);
2522 			if (unlikely(!pagep)) {
2523 				rx_stats(rxo)->rx_post_fail++;
2524 				break;
2525 			}
2526 			page_dmaaddr = dma_map_page(dev, pagep, 0,
2527 						    adapter->big_page_size,
2528 						    DMA_FROM_DEVICE);
2529 			if (dma_mapping_error(dev, page_dmaaddr)) {
2530 				put_page(pagep);
2531 				pagep = NULL;
2532 				adapter->drv_stats.dma_map_errors++;
2533 				break;
2534 			}
2535 			page_offset = 0;
2536 		} else {
2537 			get_page(pagep);
2538 			page_offset += rx_frag_size;
2539 		}
2540 		page_info->page_offset = page_offset;
2541 		page_info->page = pagep;
2542 
2543 		rxd = queue_head_node(rxq);
2544 		frag_dmaaddr = page_dmaaddr + page_info->page_offset;
2545 		rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
2546 		rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
2547 
2548 		/* Any space left in the current big page for another frag? */
2549 		if ((page_offset + rx_frag_size + rx_frag_size) >
2550 					adapter->big_page_size) {
2551 			pagep = NULL;
2552 			page_info->last_frag = true;
2553 			dma_unmap_addr_set(page_info, bus, page_dmaaddr);
2554 		} else {
2555 			dma_unmap_addr_set(page_info, bus, frag_dmaaddr);
2556 		}
2557 
2558 		prev_page_info = page_info;
2559 		queue_head_inc(rxq);
2560 		page_info = &rxo->page_info_tbl[rxq->head];
2561 	}
2562 
2563 	/* Mark the last frag of a page when we break out of the above loop
2564 	 * with no more slots available in the RXQ
2565 	 */
2566 	if (pagep) {
2567 		prev_page_info->last_frag = true;
2568 		dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr);
2569 	}
2570 
2571 	if (posted) {
2572 		atomic_add(posted, &rxq->used);
2573 		if (rxo->rx_post_starved)
2574 			rxo->rx_post_starved = false;
2575 		do {
2576 			notify = min(MAX_NUM_POST_ERX_DB, posted);
2577 			be_rxq_notify(adapter, rxq->id, notify);
2578 			posted -= notify;
2579 		} while (posted);
2580 	} else if (atomic_read(&rxq->used) == 0) {
2581 		/* Let be_worker replenish when memory is available */
2582 		rxo->rx_post_starved = true;
2583 	}
2584 }
2585 
2586 static inline void be_update_tx_err(struct be_tx_obj *txo, u8 status)
2587 {
2588 	switch (status) {
2589 	case BE_TX_COMP_HDR_PARSE_ERR:
2590 		tx_stats(txo)->tx_hdr_parse_err++;
2591 		break;
2592 	case BE_TX_COMP_NDMA_ERR:
2593 		tx_stats(txo)->tx_dma_err++;
2594 		break;
2595 	case BE_TX_COMP_ACL_ERR:
2596 		tx_stats(txo)->tx_spoof_check_err++;
2597 		break;
2598 	}
2599 }
2600 
2601 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u8 status)
2602 {
2603 	switch (status) {
2604 	case LANCER_TX_COMP_LSO_ERR:
2605 		tx_stats(txo)->tx_tso_err++;
2606 		break;
2607 	case LANCER_TX_COMP_HSW_DROP_MAC_ERR:
2608 	case LANCER_TX_COMP_HSW_DROP_VLAN_ERR:
2609 		tx_stats(txo)->tx_spoof_check_err++;
2610 		break;
2611 	case LANCER_TX_COMP_QINQ_ERR:
2612 		tx_stats(txo)->tx_qinq_err++;
2613 		break;
2614 	case LANCER_TX_COMP_PARITY_ERR:
2615 		tx_stats(txo)->tx_internal_parity_err++;
2616 		break;
2617 	case LANCER_TX_COMP_DMA_ERR:
2618 		tx_stats(txo)->tx_dma_err++;
2619 		break;
2620 	case LANCER_TX_COMP_SGE_ERR:
2621 		tx_stats(txo)->tx_sge_err++;
2622 		break;
2623 	}
2624 }
2625 
2626 static struct be_tx_compl_info *be_tx_compl_get(struct be_adapter *adapter,
2627 						struct be_tx_obj *txo)
2628 {
2629 	struct be_queue_info *tx_cq = &txo->cq;
2630 	struct be_tx_compl_info *txcp = &txo->txcp;
2631 	struct be_eth_tx_compl *compl = queue_tail_node(tx_cq);
2632 
2633 	if (compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
2634 		return NULL;
2635 
2636 	/* Ensure load ordering of valid bit dword and other dwords below */
2637 	rmb();
2638 	be_dws_le_to_cpu(compl, sizeof(*compl));
2639 
2640 	txcp->status = GET_TX_COMPL_BITS(status, compl);
2641 	txcp->end_index = GET_TX_COMPL_BITS(wrb_index, compl);
2642 
2643 	if (txcp->status) {
2644 		if (lancer_chip(adapter)) {
2645 			lancer_update_tx_err(txo, txcp->status);
2646 			/* Reset the adapter incase of TSO,
2647 			 * SGE or Parity error
2648 			 */
2649 			if (txcp->status == LANCER_TX_COMP_LSO_ERR ||
2650 			    txcp->status == LANCER_TX_COMP_PARITY_ERR ||
2651 			    txcp->status == LANCER_TX_COMP_SGE_ERR)
2652 				be_set_error(adapter, BE_ERROR_TX);
2653 		} else {
2654 			be_update_tx_err(txo, txcp->status);
2655 		}
2656 	}
2657 
2658 	if (be_check_error(adapter, BE_ERROR_TX))
2659 		return NULL;
2660 
2661 	compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
2662 	queue_tail_inc(tx_cq);
2663 	return txcp;
2664 }
2665 
2666 static u16 be_tx_compl_process(struct be_adapter *adapter,
2667 			       struct be_tx_obj *txo, u16 last_index)
2668 {
2669 	struct sk_buff **sent_skbs = txo->sent_skb_list;
2670 	struct be_queue_info *txq = &txo->q;
2671 	struct sk_buff *skb = NULL;
2672 	bool unmap_skb_hdr = false;
2673 	struct be_eth_wrb *wrb;
2674 	u16 num_wrbs = 0;
2675 	u32 frag_index;
2676 
2677 	do {
2678 		if (sent_skbs[txq->tail]) {
2679 			/* Free skb from prev req */
2680 			if (skb)
2681 				dev_consume_skb_any(skb);
2682 			skb = sent_skbs[txq->tail];
2683 			sent_skbs[txq->tail] = NULL;
2684 			queue_tail_inc(txq);  /* skip hdr wrb */
2685 			num_wrbs++;
2686 			unmap_skb_hdr = true;
2687 		}
2688 		wrb = queue_tail_node(txq);
2689 		frag_index = txq->tail;
2690 		unmap_tx_frag(&adapter->pdev->dev, wrb,
2691 			      (unmap_skb_hdr && skb_headlen(skb)));
2692 		unmap_skb_hdr = false;
2693 		queue_tail_inc(txq);
2694 		num_wrbs++;
2695 	} while (frag_index != last_index);
2696 	dev_consume_skb_any(skb);
2697 
2698 	return num_wrbs;
2699 }
2700 
2701 /* Return the number of events in the event queue */
2702 static inline int events_get(struct be_eq_obj *eqo)
2703 {
2704 	struct be_eq_entry *eqe;
2705 	int num = 0;
2706 
2707 	do {
2708 		eqe = queue_tail_node(&eqo->q);
2709 		if (eqe->evt == 0)
2710 			break;
2711 
2712 		rmb();
2713 		eqe->evt = 0;
2714 		num++;
2715 		queue_tail_inc(&eqo->q);
2716 	} while (true);
2717 
2718 	return num;
2719 }
2720 
2721 /* Leaves the EQ is disarmed state */
2722 static void be_eq_clean(struct be_eq_obj *eqo)
2723 {
2724 	int num = events_get(eqo);
2725 
2726 	be_eq_notify(eqo->adapter, eqo->q.id, false, true, num, 0);
2727 }
2728 
2729 /* Free posted rx buffers that were not used */
2730 static void be_rxq_clean(struct be_rx_obj *rxo)
2731 {
2732 	struct be_queue_info *rxq = &rxo->q;
2733 	struct be_rx_page_info *page_info;
2734 
2735 	while (atomic_read(&rxq->used) > 0) {
2736 		page_info = get_rx_page_info(rxo);
2737 		put_page(page_info->page);
2738 		memset(page_info, 0, sizeof(*page_info));
2739 	}
2740 	BUG_ON(atomic_read(&rxq->used));
2741 	rxq->tail = 0;
2742 	rxq->head = 0;
2743 }
2744 
2745 static void be_rx_cq_clean(struct be_rx_obj *rxo)
2746 {
2747 	struct be_queue_info *rx_cq = &rxo->cq;
2748 	struct be_rx_compl_info *rxcp;
2749 	struct be_adapter *adapter = rxo->adapter;
2750 	int flush_wait = 0;
2751 
2752 	/* Consume pending rx completions.
2753 	 * Wait for the flush completion (identified by zero num_rcvd)
2754 	 * to arrive. Notify CQ even when there are no more CQ entries
2755 	 * for HW to flush partially coalesced CQ entries.
2756 	 * In Lancer, there is no need to wait for flush compl.
2757 	 */
2758 	for (;;) {
2759 		rxcp = be_rx_compl_get(rxo);
2760 		if (!rxcp) {
2761 			if (lancer_chip(adapter))
2762 				break;
2763 
2764 			if (flush_wait++ > 50 ||
2765 			    be_check_error(adapter,
2766 					   BE_ERROR_HW)) {
2767 				dev_warn(&adapter->pdev->dev,
2768 					 "did not receive flush compl\n");
2769 				break;
2770 			}
2771 			be_cq_notify(adapter, rx_cq->id, true, 0);
2772 			mdelay(1);
2773 		} else {
2774 			be_rx_compl_discard(rxo, rxcp);
2775 			be_cq_notify(adapter, rx_cq->id, false, 1);
2776 			if (rxcp->num_rcvd == 0)
2777 				break;
2778 		}
2779 	}
2780 
2781 	/* After cleanup, leave the CQ in unarmed state */
2782 	be_cq_notify(adapter, rx_cq->id, false, 0);
2783 }
2784 
2785 static void be_tx_compl_clean(struct be_adapter *adapter)
2786 {
2787 	struct device *dev = &adapter->pdev->dev;
2788 	u16 cmpl = 0, timeo = 0, num_wrbs = 0;
2789 	struct be_tx_compl_info *txcp;
2790 	struct be_queue_info *txq;
2791 	u32 end_idx, notified_idx;
2792 	struct be_tx_obj *txo;
2793 	int i, pending_txqs;
2794 
2795 	/* Stop polling for compls when HW has been silent for 10ms */
2796 	do {
2797 		pending_txqs = adapter->num_tx_qs;
2798 
2799 		for_all_tx_queues(adapter, txo, i) {
2800 			cmpl = 0;
2801 			num_wrbs = 0;
2802 			txq = &txo->q;
2803 			while ((txcp = be_tx_compl_get(adapter, txo))) {
2804 				num_wrbs +=
2805 					be_tx_compl_process(adapter, txo,
2806 							    txcp->end_index);
2807 				cmpl++;
2808 			}
2809 			if (cmpl) {
2810 				be_cq_notify(adapter, txo->cq.id, false, cmpl);
2811 				atomic_sub(num_wrbs, &txq->used);
2812 				timeo = 0;
2813 			}
2814 			if (!be_is_tx_compl_pending(txo))
2815 				pending_txqs--;
2816 		}
2817 
2818 		if (pending_txqs == 0 || ++timeo > 10 ||
2819 		    be_check_error(adapter, BE_ERROR_HW))
2820 			break;
2821 
2822 		mdelay(1);
2823 	} while (true);
2824 
2825 	/* Free enqueued TX that was never notified to HW */
2826 	for_all_tx_queues(adapter, txo, i) {
2827 		txq = &txo->q;
2828 
2829 		if (atomic_read(&txq->used)) {
2830 			dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n",
2831 				 i, atomic_read(&txq->used));
2832 			notified_idx = txq->tail;
2833 			end_idx = txq->tail;
2834 			index_adv(&end_idx, atomic_read(&txq->used) - 1,
2835 				  txq->len);
2836 			/* Use the tx-compl process logic to handle requests
2837 			 * that were not sent to the HW.
2838 			 */
2839 			num_wrbs = be_tx_compl_process(adapter, txo, end_idx);
2840 			atomic_sub(num_wrbs, &txq->used);
2841 			BUG_ON(atomic_read(&txq->used));
2842 			txo->pend_wrb_cnt = 0;
2843 			/* Since hw was never notified of these requests,
2844 			 * reset TXQ indices
2845 			 */
2846 			txq->head = notified_idx;
2847 			txq->tail = notified_idx;
2848 		}
2849 	}
2850 }
2851 
2852 static void be_evt_queues_destroy(struct be_adapter *adapter)
2853 {
2854 	struct be_eq_obj *eqo;
2855 	int i;
2856 
2857 	for_all_evt_queues(adapter, eqo, i) {
2858 		if (eqo->q.created) {
2859 			be_eq_clean(eqo);
2860 			be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ);
2861 			netif_napi_del(&eqo->napi);
2862 			free_cpumask_var(eqo->affinity_mask);
2863 		}
2864 		be_queue_free(adapter, &eqo->q);
2865 	}
2866 }
2867 
2868 static int be_evt_queues_create(struct be_adapter *adapter)
2869 {
2870 	struct be_queue_info *eq;
2871 	struct be_eq_obj *eqo;
2872 	struct be_aic_obj *aic;
2873 	int i, rc;
2874 
2875 	/* need enough EQs to service both RX and TX queues */
2876 	adapter->num_evt_qs = min_t(u16, num_irqs(adapter),
2877 				    max(adapter->cfg_num_rx_irqs,
2878 					adapter->cfg_num_tx_irqs));
2879 
2880 	for_all_evt_queues(adapter, eqo, i) {
2881 		int numa_node = dev_to_node(&adapter->pdev->dev);
2882 
2883 		aic = &adapter->aic_obj[i];
2884 		eqo->adapter = adapter;
2885 		eqo->idx = i;
2886 		aic->max_eqd = BE_MAX_EQD;
2887 		aic->enable = true;
2888 
2889 		eq = &eqo->q;
2890 		rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
2891 				    sizeof(struct be_eq_entry));
2892 		if (rc)
2893 			return rc;
2894 
2895 		rc = be_cmd_eq_create(adapter, eqo);
2896 		if (rc)
2897 			return rc;
2898 
2899 		if (!zalloc_cpumask_var(&eqo->affinity_mask, GFP_KERNEL))
2900 			return -ENOMEM;
2901 		cpumask_set_cpu(cpumask_local_spread(i, numa_node),
2902 				eqo->affinity_mask);
2903 		netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
2904 			       BE_NAPI_WEIGHT);
2905 	}
2906 	return 0;
2907 }
2908 
2909 static void be_mcc_queues_destroy(struct be_adapter *adapter)
2910 {
2911 	struct be_queue_info *q;
2912 
2913 	q = &adapter->mcc_obj.q;
2914 	if (q->created)
2915 		be_cmd_q_destroy(adapter, q, QTYPE_MCCQ);
2916 	be_queue_free(adapter, q);
2917 
2918 	q = &adapter->mcc_obj.cq;
2919 	if (q->created)
2920 		be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2921 	be_queue_free(adapter, q);
2922 }
2923 
2924 /* Must be called only after TX qs are created as MCC shares TX EQ */
2925 static int be_mcc_queues_create(struct be_adapter *adapter)
2926 {
2927 	struct be_queue_info *q, *cq;
2928 
2929 	cq = &adapter->mcc_obj.cq;
2930 	if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
2931 			   sizeof(struct be_mcc_compl)))
2932 		goto err;
2933 
2934 	/* Use the default EQ for MCC completions */
2935 	if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0))
2936 		goto mcc_cq_free;
2937 
2938 	q = &adapter->mcc_obj.q;
2939 	if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
2940 		goto mcc_cq_destroy;
2941 
2942 	if (be_cmd_mccq_create(adapter, q, cq))
2943 		goto mcc_q_free;
2944 
2945 	return 0;
2946 
2947 mcc_q_free:
2948 	be_queue_free(adapter, q);
2949 mcc_cq_destroy:
2950 	be_cmd_q_destroy(adapter, cq, QTYPE_CQ);
2951 mcc_cq_free:
2952 	be_queue_free(adapter, cq);
2953 err:
2954 	return -1;
2955 }
2956 
2957 static void be_tx_queues_destroy(struct be_adapter *adapter)
2958 {
2959 	struct be_queue_info *q;
2960 	struct be_tx_obj *txo;
2961 	u8 i;
2962 
2963 	for_all_tx_queues(adapter, txo, i) {
2964 		q = &txo->q;
2965 		if (q->created)
2966 			be_cmd_q_destroy(adapter, q, QTYPE_TXQ);
2967 		be_queue_free(adapter, q);
2968 
2969 		q = &txo->cq;
2970 		if (q->created)
2971 			be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2972 		be_queue_free(adapter, q);
2973 	}
2974 }
2975 
2976 static int be_tx_qs_create(struct be_adapter *adapter)
2977 {
2978 	struct be_queue_info *cq;
2979 	struct be_tx_obj *txo;
2980 	struct be_eq_obj *eqo;
2981 	int status, i;
2982 
2983 	adapter->num_tx_qs = min(adapter->num_evt_qs, adapter->cfg_num_tx_irqs);
2984 
2985 	for_all_tx_queues(adapter, txo, i) {
2986 		cq = &txo->cq;
2987 		status = be_queue_alloc(adapter, cq, TX_CQ_LEN,
2988 					sizeof(struct be_eth_tx_compl));
2989 		if (status)
2990 			return status;
2991 
2992 		u64_stats_init(&txo->stats.sync);
2993 		u64_stats_init(&txo->stats.sync_compl);
2994 
2995 		/* If num_evt_qs is less than num_tx_qs, then more than
2996 		 * one txq share an eq
2997 		 */
2998 		eqo = &adapter->eq_obj[i % adapter->num_evt_qs];
2999 		status = be_cmd_cq_create(adapter, cq, &eqo->q, false, 3);
3000 		if (status)
3001 			return status;
3002 
3003 		status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN,
3004 					sizeof(struct be_eth_wrb));
3005 		if (status)
3006 			return status;
3007 
3008 		status = be_cmd_txq_create(adapter, txo);
3009 		if (status)
3010 			return status;
3011 
3012 		netif_set_xps_queue(adapter->netdev, eqo->affinity_mask,
3013 				    eqo->idx);
3014 	}
3015 
3016 	dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
3017 		 adapter->num_tx_qs);
3018 	return 0;
3019 }
3020 
3021 static void be_rx_cqs_destroy(struct be_adapter *adapter)
3022 {
3023 	struct be_queue_info *q;
3024 	struct be_rx_obj *rxo;
3025 	int i;
3026 
3027 	for_all_rx_queues(adapter, rxo, i) {
3028 		q = &rxo->cq;
3029 		if (q->created)
3030 			be_cmd_q_destroy(adapter, q, QTYPE_CQ);
3031 		be_queue_free(adapter, q);
3032 	}
3033 }
3034 
3035 static int be_rx_cqs_create(struct be_adapter *adapter)
3036 {
3037 	struct be_queue_info *eq, *cq;
3038 	struct be_rx_obj *rxo;
3039 	int rc, i;
3040 
3041 	adapter->num_rss_qs =
3042 			min(adapter->num_evt_qs, adapter->cfg_num_rx_irqs);
3043 
3044 	/* We'll use RSS only if atleast 2 RSS rings are supported. */
3045 	if (adapter->num_rss_qs < 2)
3046 		adapter->num_rss_qs = 0;
3047 
3048 	adapter->num_rx_qs = adapter->num_rss_qs + adapter->need_def_rxq;
3049 
3050 	/* When the interface is not capable of RSS rings (and there is no
3051 	 * need to create a default RXQ) we'll still need one RXQ
3052 	 */
3053 	if (adapter->num_rx_qs == 0)
3054 		adapter->num_rx_qs = 1;
3055 
3056 	adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
3057 	for_all_rx_queues(adapter, rxo, i) {
3058 		rxo->adapter = adapter;
3059 		cq = &rxo->cq;
3060 		rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
3061 				    sizeof(struct be_eth_rx_compl));
3062 		if (rc)
3063 			return rc;
3064 
3065 		u64_stats_init(&rxo->stats.sync);
3066 		eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
3067 		rc = be_cmd_cq_create(adapter, cq, eq, false, 3);
3068 		if (rc)
3069 			return rc;
3070 	}
3071 
3072 	dev_info(&adapter->pdev->dev,
3073 		 "created %d RX queue(s)\n", adapter->num_rx_qs);
3074 	return 0;
3075 }
3076 
3077 static irqreturn_t be_intx(int irq, void *dev)
3078 {
3079 	struct be_eq_obj *eqo = dev;
3080 	struct be_adapter *adapter = eqo->adapter;
3081 	int num_evts = 0;
3082 
3083 	/* IRQ is not expected when NAPI is scheduled as the EQ
3084 	 * will not be armed.
3085 	 * But, this can happen on Lancer INTx where it takes
3086 	 * a while to de-assert INTx or in BE2 where occasionaly
3087 	 * an interrupt may be raised even when EQ is unarmed.
3088 	 * If NAPI is already scheduled, then counting & notifying
3089 	 * events will orphan them.
3090 	 */
3091 	if (napi_schedule_prep(&eqo->napi)) {
3092 		num_evts = events_get(eqo);
3093 		__napi_schedule(&eqo->napi);
3094 		if (num_evts)
3095 			eqo->spurious_intr = 0;
3096 	}
3097 	be_eq_notify(adapter, eqo->q.id, false, true, num_evts, 0);
3098 
3099 	/* Return IRQ_HANDLED only for the the first spurious intr
3100 	 * after a valid intr to stop the kernel from branding
3101 	 * this irq as a bad one!
3102 	 */
3103 	if (num_evts || eqo->spurious_intr++ == 0)
3104 		return IRQ_HANDLED;
3105 	else
3106 		return IRQ_NONE;
3107 }
3108 
3109 static irqreturn_t be_msix(int irq, void *dev)
3110 {
3111 	struct be_eq_obj *eqo = dev;
3112 
3113 	be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
3114 	napi_schedule(&eqo->napi);
3115 	return IRQ_HANDLED;
3116 }
3117 
3118 static inline bool do_gro(struct be_rx_compl_info *rxcp)
3119 {
3120 	return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false;
3121 }
3122 
3123 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi,
3124 			 int budget)
3125 {
3126 	struct be_adapter *adapter = rxo->adapter;
3127 	struct be_queue_info *rx_cq = &rxo->cq;
3128 	struct be_rx_compl_info *rxcp;
3129 	u32 work_done;
3130 	u32 frags_consumed = 0;
3131 
3132 	for (work_done = 0; work_done < budget; work_done++) {
3133 		rxcp = be_rx_compl_get(rxo);
3134 		if (!rxcp)
3135 			break;
3136 
3137 		/* Is it a flush compl that has no data */
3138 		if (unlikely(rxcp->num_rcvd == 0))
3139 			goto loop_continue;
3140 
3141 		/* Discard compl with partial DMA Lancer B0 */
3142 		if (unlikely(!rxcp->pkt_size)) {
3143 			be_rx_compl_discard(rxo, rxcp);
3144 			goto loop_continue;
3145 		}
3146 
3147 		/* On BE drop pkts that arrive due to imperfect filtering in
3148 		 * promiscuous mode on some skews
3149 		 */
3150 		if (unlikely(rxcp->port != adapter->port_num &&
3151 			     !lancer_chip(adapter))) {
3152 			be_rx_compl_discard(rxo, rxcp);
3153 			goto loop_continue;
3154 		}
3155 
3156 		if (do_gro(rxcp))
3157 			be_rx_compl_process_gro(rxo, napi, rxcp);
3158 		else
3159 			be_rx_compl_process(rxo, napi, rxcp);
3160 
3161 loop_continue:
3162 		frags_consumed += rxcp->num_rcvd;
3163 		be_rx_stats_update(rxo, rxcp);
3164 	}
3165 
3166 	if (work_done) {
3167 		be_cq_notify(adapter, rx_cq->id, true, work_done);
3168 
3169 		/* When an rx-obj gets into post_starved state, just
3170 		 * let be_worker do the posting.
3171 		 */
3172 		if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM &&
3173 		    !rxo->rx_post_starved)
3174 			be_post_rx_frags(rxo, GFP_ATOMIC,
3175 					 max_t(u32, MAX_RX_POST,
3176 					       frags_consumed));
3177 	}
3178 
3179 	return work_done;
3180 }
3181 
3182 
3183 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo,
3184 			  int idx)
3185 {
3186 	int num_wrbs = 0, work_done = 0;
3187 	struct be_tx_compl_info *txcp;
3188 
3189 	while ((txcp = be_tx_compl_get(adapter, txo))) {
3190 		num_wrbs += be_tx_compl_process(adapter, txo, txcp->end_index);
3191 		work_done++;
3192 	}
3193 
3194 	if (work_done) {
3195 		be_cq_notify(adapter, txo->cq.id, true, work_done);
3196 		atomic_sub(num_wrbs, &txo->q.used);
3197 
3198 		/* As Tx wrbs have been freed up, wake up netdev queue
3199 		 * if it was stopped due to lack of tx wrbs.  */
3200 		if (__netif_subqueue_stopped(adapter->netdev, idx) &&
3201 		    be_can_txq_wake(txo)) {
3202 			netif_wake_subqueue(adapter->netdev, idx);
3203 		}
3204 
3205 		u64_stats_update_begin(&tx_stats(txo)->sync_compl);
3206 		tx_stats(txo)->tx_compl += work_done;
3207 		u64_stats_update_end(&tx_stats(txo)->sync_compl);
3208 	}
3209 }
3210 
3211 int be_poll(struct napi_struct *napi, int budget)
3212 {
3213 	struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3214 	struct be_adapter *adapter = eqo->adapter;
3215 	int max_work = 0, work, i, num_evts;
3216 	struct be_rx_obj *rxo;
3217 	struct be_tx_obj *txo;
3218 	u32 mult_enc = 0;
3219 
3220 	num_evts = events_get(eqo);
3221 
3222 	for_all_tx_queues_on_eq(adapter, eqo, txo, i)
3223 		be_process_tx(adapter, txo, i);
3224 
3225 	/* This loop will iterate twice for EQ0 in which
3226 	 * completions of the last RXQ (default one) are also processed
3227 	 * For other EQs the loop iterates only once
3228 	 */
3229 	for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3230 		work = be_process_rx(rxo, napi, budget);
3231 		max_work = max(work, max_work);
3232 	}
3233 
3234 	if (is_mcc_eqo(eqo))
3235 		be_process_mcc(adapter);
3236 
3237 	if (max_work < budget) {
3238 		napi_complete_done(napi, max_work);
3239 
3240 		/* Skyhawk EQ_DB has a provision to set the rearm to interrupt
3241 		 * delay via a delay multiplier encoding value
3242 		 */
3243 		if (skyhawk_chip(adapter))
3244 			mult_enc = be_get_eq_delay_mult_enc(eqo);
3245 
3246 		be_eq_notify(adapter, eqo->q.id, true, false, num_evts,
3247 			     mult_enc);
3248 	} else {
3249 		/* As we'll continue in polling mode, count and clear events */
3250 		be_eq_notify(adapter, eqo->q.id, false, false, num_evts, 0);
3251 	}
3252 	return max_work;
3253 }
3254 
3255 void be_detect_error(struct be_adapter *adapter)
3256 {
3257 	u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0;
3258 	u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
3259 	struct device *dev = &adapter->pdev->dev;
3260 	u16 val;
3261 	u32 i;
3262 
3263 	if (be_check_error(adapter, BE_ERROR_HW))
3264 		return;
3265 
3266 	if (lancer_chip(adapter)) {
3267 		sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET);
3268 		if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
3269 			be_set_error(adapter, BE_ERROR_UE);
3270 			sliport_err1 = ioread32(adapter->db +
3271 						SLIPORT_ERROR1_OFFSET);
3272 			sliport_err2 = ioread32(adapter->db +
3273 						SLIPORT_ERROR2_OFFSET);
3274 			/* Do not log error messages if its a FW reset */
3275 			if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 &&
3276 			    sliport_err2 == SLIPORT_ERROR_FW_RESET2) {
3277 				dev_info(dev, "Firmware update in progress\n");
3278 			} else {
3279 				dev_err(dev, "Error detected in the card\n");
3280 				dev_err(dev, "ERR: sliport status 0x%x\n",
3281 					sliport_status);
3282 				dev_err(dev, "ERR: sliport error1 0x%x\n",
3283 					sliport_err1);
3284 				dev_err(dev, "ERR: sliport error2 0x%x\n",
3285 					sliport_err2);
3286 			}
3287 		}
3288 	} else {
3289 		ue_lo = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_LOW);
3290 		ue_hi = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_HIGH);
3291 		ue_lo_mask = ioread32(adapter->pcicfg +
3292 				      PCICFG_UE_STATUS_LOW_MASK);
3293 		ue_hi_mask = ioread32(adapter->pcicfg +
3294 				      PCICFG_UE_STATUS_HI_MASK);
3295 
3296 		ue_lo = (ue_lo & ~ue_lo_mask);
3297 		ue_hi = (ue_hi & ~ue_hi_mask);
3298 
3299 		if (ue_lo || ue_hi) {
3300 			/* On certain platforms BE3 hardware can indicate
3301 			 * spurious UEs. In case of a UE in the chip,
3302 			 * the POST register correctly reports either a
3303 			 * FAT_LOG_START state (FW is currently dumping
3304 			 * FAT log data) or a ARMFW_UE state. Check for the
3305 			 * above states to ascertain if the UE is valid or not.
3306 			 */
3307 			if (BE3_chip(adapter)) {
3308 				val = be_POST_stage_get(adapter);
3309 				if ((val & POST_STAGE_FAT_LOG_START)
3310 				     != POST_STAGE_FAT_LOG_START &&
3311 				    (val & POST_STAGE_ARMFW_UE)
3312 				     != POST_STAGE_ARMFW_UE)
3313 					return;
3314 			}
3315 
3316 			dev_err(dev, "Error detected in the adapter");
3317 			be_set_error(adapter, BE_ERROR_UE);
3318 
3319 			for (i = 0; ue_lo; ue_lo >>= 1, i++) {
3320 				if (ue_lo & 1)
3321 					dev_err(dev, "UE: %s bit set\n",
3322 						ue_status_low_desc[i]);
3323 			}
3324 			for (i = 0; ue_hi; ue_hi >>= 1, i++) {
3325 				if (ue_hi & 1)
3326 					dev_err(dev, "UE: %s bit set\n",
3327 						ue_status_hi_desc[i]);
3328 			}
3329 		}
3330 	}
3331 }
3332 
3333 static void be_msix_disable(struct be_adapter *adapter)
3334 {
3335 	if (msix_enabled(adapter)) {
3336 		pci_disable_msix(adapter->pdev);
3337 		adapter->num_msix_vec = 0;
3338 		adapter->num_msix_roce_vec = 0;
3339 	}
3340 }
3341 
3342 static int be_msix_enable(struct be_adapter *adapter)
3343 {
3344 	unsigned int i, max_roce_eqs;
3345 	struct device *dev = &adapter->pdev->dev;
3346 	int num_vec;
3347 
3348 	/* If RoCE is supported, program the max number of vectors that
3349 	 * could be used for NIC and RoCE, else, just program the number
3350 	 * we'll use initially.
3351 	 */
3352 	if (be_roce_supported(adapter)) {
3353 		max_roce_eqs =
3354 			be_max_func_eqs(adapter) - be_max_nic_eqs(adapter);
3355 		max_roce_eqs = min(max_roce_eqs, num_online_cpus());
3356 		num_vec = be_max_any_irqs(adapter) + max_roce_eqs;
3357 	} else {
3358 		num_vec = max(adapter->cfg_num_rx_irqs,
3359 			      adapter->cfg_num_tx_irqs);
3360 	}
3361 
3362 	for (i = 0; i < num_vec; i++)
3363 		adapter->msix_entries[i].entry = i;
3364 
3365 	num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
3366 					MIN_MSIX_VECTORS, num_vec);
3367 	if (num_vec < 0)
3368 		goto fail;
3369 
3370 	if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) {
3371 		adapter->num_msix_roce_vec = num_vec / 2;
3372 		dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n",
3373 			 adapter->num_msix_roce_vec);
3374 	}
3375 
3376 	adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec;
3377 
3378 	dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n",
3379 		 adapter->num_msix_vec);
3380 	return 0;
3381 
3382 fail:
3383 	dev_warn(dev, "MSIx enable failed\n");
3384 
3385 	/* INTx is not supported in VFs, so fail probe if enable_msix fails */
3386 	if (be_virtfn(adapter))
3387 		return num_vec;
3388 	return 0;
3389 }
3390 
3391 static inline int be_msix_vec_get(struct be_adapter *adapter,
3392 				  struct be_eq_obj *eqo)
3393 {
3394 	return adapter->msix_entries[eqo->msix_idx].vector;
3395 }
3396 
3397 static int be_msix_register(struct be_adapter *adapter)
3398 {
3399 	struct net_device *netdev = adapter->netdev;
3400 	struct be_eq_obj *eqo;
3401 	int status, i, vec;
3402 
3403 	for_all_evt_queues(adapter, eqo, i) {
3404 		sprintf(eqo->desc, "%s-q%d", netdev->name, i);
3405 		vec = be_msix_vec_get(adapter, eqo);
3406 		status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
3407 		if (status)
3408 			goto err_msix;
3409 
3410 		irq_set_affinity_hint(vec, eqo->affinity_mask);
3411 	}
3412 
3413 	return 0;
3414 err_msix:
3415 	for (i--; i >= 0; i--) {
3416 		eqo = &adapter->eq_obj[i];
3417 		free_irq(be_msix_vec_get(adapter, eqo), eqo);
3418 	}
3419 	dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n",
3420 		 status);
3421 	be_msix_disable(adapter);
3422 	return status;
3423 }
3424 
3425 static int be_irq_register(struct be_adapter *adapter)
3426 {
3427 	struct net_device *netdev = adapter->netdev;
3428 	int status;
3429 
3430 	if (msix_enabled(adapter)) {
3431 		status = be_msix_register(adapter);
3432 		if (status == 0)
3433 			goto done;
3434 		/* INTx is not supported for VF */
3435 		if (be_virtfn(adapter))
3436 			return status;
3437 	}
3438 
3439 	/* INTx: only the first EQ is used */
3440 	netdev->irq = adapter->pdev->irq;
3441 	status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
3442 			     &adapter->eq_obj[0]);
3443 	if (status) {
3444 		dev_err(&adapter->pdev->dev,
3445 			"INTx request IRQ failed - err %d\n", status);
3446 		return status;
3447 	}
3448 done:
3449 	adapter->isr_registered = true;
3450 	return 0;
3451 }
3452 
3453 static void be_irq_unregister(struct be_adapter *adapter)
3454 {
3455 	struct net_device *netdev = adapter->netdev;
3456 	struct be_eq_obj *eqo;
3457 	int i, vec;
3458 
3459 	if (!adapter->isr_registered)
3460 		return;
3461 
3462 	/* INTx */
3463 	if (!msix_enabled(adapter)) {
3464 		free_irq(netdev->irq, &adapter->eq_obj[0]);
3465 		goto done;
3466 	}
3467 
3468 	/* MSIx */
3469 	for_all_evt_queues(adapter, eqo, i) {
3470 		vec = be_msix_vec_get(adapter, eqo);
3471 		irq_set_affinity_hint(vec, NULL);
3472 		free_irq(vec, eqo);
3473 	}
3474 
3475 done:
3476 	adapter->isr_registered = false;
3477 }
3478 
3479 static void be_rx_qs_destroy(struct be_adapter *adapter)
3480 {
3481 	struct rss_info *rss = &adapter->rss_info;
3482 	struct be_queue_info *q;
3483 	struct be_rx_obj *rxo;
3484 	int i;
3485 
3486 	for_all_rx_queues(adapter, rxo, i) {
3487 		q = &rxo->q;
3488 		if (q->created) {
3489 			/* If RXQs are destroyed while in an "out of buffer"
3490 			 * state, there is a possibility of an HW stall on
3491 			 * Lancer. So, post 64 buffers to each queue to relieve
3492 			 * the "out of buffer" condition.
3493 			 * Make sure there's space in the RXQ before posting.
3494 			 */
3495 			if (lancer_chip(adapter)) {
3496 				be_rx_cq_clean(rxo);
3497 				if (atomic_read(&q->used) == 0)
3498 					be_post_rx_frags(rxo, GFP_KERNEL,
3499 							 MAX_RX_POST);
3500 			}
3501 
3502 			be_cmd_rxq_destroy(adapter, q);
3503 			be_rx_cq_clean(rxo);
3504 			be_rxq_clean(rxo);
3505 		}
3506 		be_queue_free(adapter, q);
3507 	}
3508 
3509 	if (rss->rss_flags) {
3510 		rss->rss_flags = RSS_ENABLE_NONE;
3511 		be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3512 				  128, rss->rss_hkey);
3513 	}
3514 }
3515 
3516 static void be_disable_if_filters(struct be_adapter *adapter)
3517 {
3518 	/* Don't delete MAC on BE3 VFs without FILTMGMT privilege  */
3519 	if (!BEx_chip(adapter) || !be_virtfn(adapter) ||
3520 	    check_privilege(adapter, BE_PRIV_FILTMGMT)) {
3521 		be_dev_mac_del(adapter, adapter->pmac_id[0]);
3522 		eth_zero_addr(adapter->dev_mac);
3523 	}
3524 
3525 	be_clear_uc_list(adapter);
3526 	be_clear_mc_list(adapter);
3527 
3528 	/* The IFACE flags are enabled in the open path and cleared
3529 	 * in the close path. When a VF gets detached from the host and
3530 	 * assigned to a VM the following happens:
3531 	 *	- VF's IFACE flags get cleared in the detach path
3532 	 *	- IFACE create is issued by the VF in the attach path
3533 	 * Due to a bug in the BE3/Skyhawk-R FW
3534 	 * (Lancer FW doesn't have the bug), the IFACE capability flags
3535 	 * specified along with the IFACE create cmd issued by a VF are not
3536 	 * honoured by FW.  As a consequence, if a *new* driver
3537 	 * (that enables/disables IFACE flags in open/close)
3538 	 * is loaded in the host and an *old* driver is * used by a VM/VF,
3539 	 * the IFACE gets created *without* the needed flags.
3540 	 * To avoid this, disable RX-filter flags only for Lancer.
3541 	 */
3542 	if (lancer_chip(adapter)) {
3543 		be_cmd_rx_filter(adapter, BE_IF_ALL_FILT_FLAGS, OFF);
3544 		adapter->if_flags &= ~BE_IF_ALL_FILT_FLAGS;
3545 	}
3546 }
3547 
3548 static int be_close(struct net_device *netdev)
3549 {
3550 	struct be_adapter *adapter = netdev_priv(netdev);
3551 	struct be_eq_obj *eqo;
3552 	int i;
3553 
3554 	/* This protection is needed as be_close() may be called even when the
3555 	 * adapter is in cleared state (after eeh perm failure)
3556 	 */
3557 	if (!(adapter->flags & BE_FLAGS_SETUP_DONE))
3558 		return 0;
3559 
3560 	/* Before attempting cleanup ensure all the pending cmds in the
3561 	 * config_wq have finished execution
3562 	 */
3563 	flush_workqueue(be_wq);
3564 
3565 	be_disable_if_filters(adapter);
3566 
3567 	if (adapter->flags & BE_FLAGS_NAPI_ENABLED) {
3568 		for_all_evt_queues(adapter, eqo, i) {
3569 			napi_disable(&eqo->napi);
3570 		}
3571 		adapter->flags &= ~BE_FLAGS_NAPI_ENABLED;
3572 	}
3573 
3574 	be_async_mcc_disable(adapter);
3575 
3576 	/* Wait for all pending tx completions to arrive so that
3577 	 * all tx skbs are freed.
3578 	 */
3579 	netif_tx_disable(netdev);
3580 	be_tx_compl_clean(adapter);
3581 
3582 	be_rx_qs_destroy(adapter);
3583 
3584 	for_all_evt_queues(adapter, eqo, i) {
3585 		if (msix_enabled(adapter))
3586 			synchronize_irq(be_msix_vec_get(adapter, eqo));
3587 		else
3588 			synchronize_irq(netdev->irq);
3589 		be_eq_clean(eqo);
3590 	}
3591 
3592 	be_irq_unregister(adapter);
3593 
3594 	return 0;
3595 }
3596 
3597 static int be_rx_qs_create(struct be_adapter *adapter)
3598 {
3599 	struct rss_info *rss = &adapter->rss_info;
3600 	u8 rss_key[RSS_HASH_KEY_LEN];
3601 	struct be_rx_obj *rxo;
3602 	int rc, i, j;
3603 
3604 	for_all_rx_queues(adapter, rxo, i) {
3605 		rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN,
3606 				    sizeof(struct be_eth_rx_d));
3607 		if (rc)
3608 			return rc;
3609 	}
3610 
3611 	if (adapter->need_def_rxq || !adapter->num_rss_qs) {
3612 		rxo = default_rxo(adapter);
3613 		rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3614 				       rx_frag_size, adapter->if_handle,
3615 				       false, &rxo->rss_id);
3616 		if (rc)
3617 			return rc;
3618 	}
3619 
3620 	for_all_rss_queues(adapter, rxo, i) {
3621 		rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3622 				       rx_frag_size, adapter->if_handle,
3623 				       true, &rxo->rss_id);
3624 		if (rc)
3625 			return rc;
3626 	}
3627 
3628 	if (be_multi_rxq(adapter)) {
3629 		for (j = 0; j < RSS_INDIR_TABLE_LEN; j += adapter->num_rss_qs) {
3630 			for_all_rss_queues(adapter, rxo, i) {
3631 				if ((j + i) >= RSS_INDIR_TABLE_LEN)
3632 					break;
3633 				rss->rsstable[j + i] = rxo->rss_id;
3634 				rss->rss_queue[j + i] = i;
3635 			}
3636 		}
3637 		rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 |
3638 			RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6;
3639 
3640 		if (!BEx_chip(adapter))
3641 			rss->rss_flags |= RSS_ENABLE_UDP_IPV4 |
3642 				RSS_ENABLE_UDP_IPV6;
3643 
3644 		netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN);
3645 		rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3646 				       RSS_INDIR_TABLE_LEN, rss_key);
3647 		if (rc) {
3648 			rss->rss_flags = RSS_ENABLE_NONE;
3649 			return rc;
3650 		}
3651 
3652 		memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN);
3653 	} else {
3654 		/* Disable RSS, if only default RX Q is created */
3655 		rss->rss_flags = RSS_ENABLE_NONE;
3656 	}
3657 
3658 
3659 	/* Post 1 less than RXQ-len to avoid head being equal to tail,
3660 	 * which is a queue empty condition
3661 	 */
3662 	for_all_rx_queues(adapter, rxo, i)
3663 		be_post_rx_frags(rxo, GFP_KERNEL, RX_Q_LEN - 1);
3664 
3665 	return 0;
3666 }
3667 
3668 static int be_enable_if_filters(struct be_adapter *adapter)
3669 {
3670 	int status;
3671 
3672 	status = be_cmd_rx_filter(adapter, BE_IF_FILT_FLAGS_BASIC, ON);
3673 	if (status)
3674 		return status;
3675 
3676 	/* Normally this condition usually true as the ->dev_mac is zeroed.
3677 	 * But on BE3 VFs the initial MAC is pre-programmed by PF and
3678 	 * subsequent be_dev_mac_add() can fail (after fresh boot)
3679 	 */
3680 	if (!ether_addr_equal(adapter->dev_mac, adapter->netdev->dev_addr)) {
3681 		int old_pmac_id = -1;
3682 
3683 		/* Remember old programmed MAC if any - can happen on BE3 VF */
3684 		if (!is_zero_ether_addr(adapter->dev_mac))
3685 			old_pmac_id = adapter->pmac_id[0];
3686 
3687 		status = be_dev_mac_add(adapter, adapter->netdev->dev_addr);
3688 		if (status)
3689 			return status;
3690 
3691 		/* Delete the old programmed MAC as we successfully programmed
3692 		 * a new MAC
3693 		 */
3694 		if (old_pmac_id >= 0 && old_pmac_id != adapter->pmac_id[0])
3695 			be_dev_mac_del(adapter, old_pmac_id);
3696 
3697 		ether_addr_copy(adapter->dev_mac, adapter->netdev->dev_addr);
3698 	}
3699 
3700 	if (adapter->vlans_added)
3701 		be_vid_config(adapter);
3702 
3703 	__be_set_rx_mode(adapter);
3704 
3705 	return 0;
3706 }
3707 
3708 static int be_open(struct net_device *netdev)
3709 {
3710 	struct be_adapter *adapter = netdev_priv(netdev);
3711 	struct be_eq_obj *eqo;
3712 	struct be_rx_obj *rxo;
3713 	struct be_tx_obj *txo;
3714 	u8 link_status;
3715 	int status, i;
3716 
3717 	status = be_rx_qs_create(adapter);
3718 	if (status)
3719 		goto err;
3720 
3721 	status = be_enable_if_filters(adapter);
3722 	if (status)
3723 		goto err;
3724 
3725 	status = be_irq_register(adapter);
3726 	if (status)
3727 		goto err;
3728 
3729 	for_all_rx_queues(adapter, rxo, i)
3730 		be_cq_notify(adapter, rxo->cq.id, true, 0);
3731 
3732 	for_all_tx_queues(adapter, txo, i)
3733 		be_cq_notify(adapter, txo->cq.id, true, 0);
3734 
3735 	be_async_mcc_enable(adapter);
3736 
3737 	for_all_evt_queues(adapter, eqo, i) {
3738 		napi_enable(&eqo->napi);
3739 		be_eq_notify(adapter, eqo->q.id, true, true, 0, 0);
3740 	}
3741 	adapter->flags |= BE_FLAGS_NAPI_ENABLED;
3742 
3743 	status = be_cmd_link_status_query(adapter, NULL, &link_status, 0);
3744 	if (!status)
3745 		be_link_status_update(adapter, link_status);
3746 
3747 	netif_tx_start_all_queues(netdev);
3748 	if (skyhawk_chip(adapter))
3749 		udp_tunnel_get_rx_info(netdev);
3750 
3751 	return 0;
3752 err:
3753 	be_close(adapter->netdev);
3754 	return -EIO;
3755 }
3756 
3757 static void be_vf_eth_addr_generate(struct be_adapter *adapter, u8 *mac)
3758 {
3759 	u32 addr;
3760 
3761 	addr = jhash(adapter->netdev->dev_addr, ETH_ALEN, 0);
3762 
3763 	mac[5] = (u8)(addr & 0xFF);
3764 	mac[4] = (u8)((addr >> 8) & 0xFF);
3765 	mac[3] = (u8)((addr >> 16) & 0xFF);
3766 	/* Use the OUI from the current MAC address */
3767 	memcpy(mac, adapter->netdev->dev_addr, 3);
3768 }
3769 
3770 /*
3771  * Generate a seed MAC address from the PF MAC Address using jhash.
3772  * MAC Address for VFs are assigned incrementally starting from the seed.
3773  * These addresses are programmed in the ASIC by the PF and the VF driver
3774  * queries for the MAC address during its probe.
3775  */
3776 static int be_vf_eth_addr_config(struct be_adapter *adapter)
3777 {
3778 	u32 vf;
3779 	int status = 0;
3780 	u8 mac[ETH_ALEN];
3781 	struct be_vf_cfg *vf_cfg;
3782 
3783 	be_vf_eth_addr_generate(adapter, mac);
3784 
3785 	for_all_vfs(adapter, vf_cfg, vf) {
3786 		if (BEx_chip(adapter))
3787 			status = be_cmd_pmac_add(adapter, mac,
3788 						 vf_cfg->if_handle,
3789 						 &vf_cfg->pmac_id, vf + 1);
3790 		else
3791 			status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
3792 						vf + 1);
3793 
3794 		if (status)
3795 			dev_err(&adapter->pdev->dev,
3796 				"Mac address assignment failed for VF %d\n",
3797 				vf);
3798 		else
3799 			memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3800 
3801 		mac[5] += 1;
3802 	}
3803 	return status;
3804 }
3805 
3806 static int be_vfs_mac_query(struct be_adapter *adapter)
3807 {
3808 	int status, vf;
3809 	u8 mac[ETH_ALEN];
3810 	struct be_vf_cfg *vf_cfg;
3811 
3812 	for_all_vfs(adapter, vf_cfg, vf) {
3813 		status = be_cmd_get_active_mac(adapter, vf_cfg->pmac_id,
3814 					       mac, vf_cfg->if_handle,
3815 					       false, vf+1);
3816 		if (status)
3817 			return status;
3818 		memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3819 	}
3820 	return 0;
3821 }
3822 
3823 static void be_vf_clear(struct be_adapter *adapter)
3824 {
3825 	struct be_vf_cfg *vf_cfg;
3826 	u32 vf;
3827 
3828 	if (pci_vfs_assigned(adapter->pdev)) {
3829 		dev_warn(&adapter->pdev->dev,
3830 			 "VFs are assigned to VMs: not disabling VFs\n");
3831 		goto done;
3832 	}
3833 
3834 	pci_disable_sriov(adapter->pdev);
3835 
3836 	for_all_vfs(adapter, vf_cfg, vf) {
3837 		if (BEx_chip(adapter))
3838 			be_cmd_pmac_del(adapter, vf_cfg->if_handle,
3839 					vf_cfg->pmac_id, vf + 1);
3840 		else
3841 			be_cmd_set_mac(adapter, NULL, vf_cfg->if_handle,
3842 				       vf + 1);
3843 
3844 		be_cmd_if_destroy(adapter, vf_cfg->if_handle, vf + 1);
3845 	}
3846 
3847 	if (BE3_chip(adapter))
3848 		be_cmd_set_hsw_config(adapter, 0, 0,
3849 				      adapter->if_handle,
3850 				      PORT_FWD_TYPE_PASSTHRU, 0);
3851 done:
3852 	kfree(adapter->vf_cfg);
3853 	adapter->num_vfs = 0;
3854 	adapter->flags &= ~BE_FLAGS_SRIOV_ENABLED;
3855 }
3856 
3857 static void be_clear_queues(struct be_adapter *adapter)
3858 {
3859 	be_mcc_queues_destroy(adapter);
3860 	be_rx_cqs_destroy(adapter);
3861 	be_tx_queues_destroy(adapter);
3862 	be_evt_queues_destroy(adapter);
3863 }
3864 
3865 static void be_cancel_worker(struct be_adapter *adapter)
3866 {
3867 	if (adapter->flags & BE_FLAGS_WORKER_SCHEDULED) {
3868 		cancel_delayed_work_sync(&adapter->work);
3869 		adapter->flags &= ~BE_FLAGS_WORKER_SCHEDULED;
3870 	}
3871 }
3872 
3873 static void be_cancel_err_detection(struct be_adapter *adapter)
3874 {
3875 	struct be_error_recovery *err_rec = &adapter->error_recovery;
3876 
3877 	if (!be_err_recovery_workq)
3878 		return;
3879 
3880 	if (adapter->flags & BE_FLAGS_ERR_DETECTION_SCHEDULED) {
3881 		cancel_delayed_work_sync(&err_rec->err_detection_work);
3882 		adapter->flags &= ~BE_FLAGS_ERR_DETECTION_SCHEDULED;
3883 	}
3884 }
3885 
3886 static int be_enable_vxlan_offloads(struct be_adapter *adapter)
3887 {
3888 	struct net_device *netdev = adapter->netdev;
3889 	struct device *dev = &adapter->pdev->dev;
3890 	struct be_vxlan_port *vxlan_port;
3891 	__be16 port;
3892 	int status;
3893 
3894 	vxlan_port = list_first_entry(&adapter->vxlan_port_list,
3895 				      struct be_vxlan_port, list);
3896 	port = vxlan_port->port;
3897 
3898 	status = be_cmd_manage_iface(adapter, adapter->if_handle,
3899 				     OP_CONVERT_NORMAL_TO_TUNNEL);
3900 	if (status) {
3901 		dev_warn(dev, "Failed to convert normal interface to tunnel\n");
3902 		return status;
3903 	}
3904 	adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS;
3905 
3906 	status = be_cmd_set_vxlan_port(adapter, port);
3907 	if (status) {
3908 		dev_warn(dev, "Failed to add VxLAN port\n");
3909 		return status;
3910 	}
3911 	adapter->vxlan_port = port;
3912 
3913 	netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
3914 				   NETIF_F_TSO | NETIF_F_TSO6 |
3915 				   NETIF_F_GSO_UDP_TUNNEL;
3916 	netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
3917 	netdev->features |= NETIF_F_GSO_UDP_TUNNEL;
3918 
3919 	dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n",
3920 		 be16_to_cpu(port));
3921 	return 0;
3922 }
3923 
3924 static void be_disable_vxlan_offloads(struct be_adapter *adapter)
3925 {
3926 	struct net_device *netdev = adapter->netdev;
3927 
3928 	if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS)
3929 		be_cmd_manage_iface(adapter, adapter->if_handle,
3930 				    OP_CONVERT_TUNNEL_TO_NORMAL);
3931 
3932 	if (adapter->vxlan_port)
3933 		be_cmd_set_vxlan_port(adapter, 0);
3934 
3935 	adapter->flags &= ~BE_FLAGS_VXLAN_OFFLOADS;
3936 	adapter->vxlan_port = 0;
3937 
3938 	netdev->hw_enc_features = 0;
3939 	netdev->hw_features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3940 	netdev->features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3941 }
3942 
3943 static void be_calculate_vf_res(struct be_adapter *adapter, u16 num_vfs,
3944 				struct be_resources *vft_res)
3945 {
3946 	struct be_resources res = adapter->pool_res;
3947 	u32 vf_if_cap_flags = res.vf_if_cap_flags;
3948 	struct be_resources res_mod = {0};
3949 	u16 num_vf_qs = 1;
3950 
3951 	/* Distribute the queue resources among the PF and it's VFs */
3952 	if (num_vfs) {
3953 		/* Divide the rx queues evenly among the VFs and the PF, capped
3954 		 * at VF-EQ-count. Any remainder queues belong to the PF.
3955 		 */
3956 		num_vf_qs = min(SH_VF_MAX_NIC_EQS,
3957 				res.max_rss_qs / (num_vfs + 1));
3958 
3959 		/* Skyhawk-R chip supports only MAX_PORT_RSS_TABLES
3960 		 * RSS Tables per port. Provide RSS on VFs, only if number of
3961 		 * VFs requested is less than it's PF Pool's RSS Tables limit.
3962 		 */
3963 		if (num_vfs >= be_max_pf_pool_rss_tables(adapter))
3964 			num_vf_qs = 1;
3965 	}
3966 
3967 	/* Resource with fields set to all '1's by GET_PROFILE_CONFIG cmd,
3968 	 * which are modifiable using SET_PROFILE_CONFIG cmd.
3969 	 */
3970 	be_cmd_get_profile_config(adapter, &res_mod, NULL, ACTIVE_PROFILE_TYPE,
3971 				  RESOURCE_MODIFIABLE, 0);
3972 
3973 	/* If RSS IFACE capability flags are modifiable for a VF, set the
3974 	 * capability flag as valid and set RSS and DEFQ_RSS IFACE flags if
3975 	 * more than 1 RSSQ is available for a VF.
3976 	 * Otherwise, provision only 1 queue pair for VF.
3977 	 */
3978 	if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_RSS) {
3979 		vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
3980 		if (num_vf_qs > 1) {
3981 			vf_if_cap_flags |= BE_IF_FLAGS_RSS;
3982 			if (res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS)
3983 				vf_if_cap_flags |= BE_IF_FLAGS_DEFQ_RSS;
3984 		} else {
3985 			vf_if_cap_flags &= ~(BE_IF_FLAGS_RSS |
3986 					     BE_IF_FLAGS_DEFQ_RSS);
3987 		}
3988 	} else {
3989 		num_vf_qs = 1;
3990 	}
3991 
3992 	if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
3993 		vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT);
3994 		vf_if_cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
3995 	}
3996 
3997 	vft_res->vf_if_cap_flags = vf_if_cap_flags;
3998 	vft_res->max_rx_qs = num_vf_qs;
3999 	vft_res->max_rss_qs = num_vf_qs;
4000 	vft_res->max_tx_qs = res.max_tx_qs / (num_vfs + 1);
4001 	vft_res->max_cq_count = res.max_cq_count / (num_vfs + 1);
4002 
4003 	/* Distribute unicast MACs, VLANs, IFACE count and MCCQ count equally
4004 	 * among the PF and it's VFs, if the fields are changeable
4005 	 */
4006 	if (res_mod.max_uc_mac == FIELD_MODIFIABLE)
4007 		vft_res->max_uc_mac = res.max_uc_mac / (num_vfs + 1);
4008 
4009 	if (res_mod.max_vlans == FIELD_MODIFIABLE)
4010 		vft_res->max_vlans = res.max_vlans / (num_vfs + 1);
4011 
4012 	if (res_mod.max_iface_count == FIELD_MODIFIABLE)
4013 		vft_res->max_iface_count = res.max_iface_count / (num_vfs + 1);
4014 
4015 	if (res_mod.max_mcc_count == FIELD_MODIFIABLE)
4016 		vft_res->max_mcc_count = res.max_mcc_count / (num_vfs + 1);
4017 }
4018 
4019 static void be_if_destroy(struct be_adapter *adapter)
4020 {
4021 	be_cmd_if_destroy(adapter, adapter->if_handle,  0);
4022 
4023 	kfree(adapter->pmac_id);
4024 	adapter->pmac_id = NULL;
4025 
4026 	kfree(adapter->mc_list);
4027 	adapter->mc_list = NULL;
4028 
4029 	kfree(adapter->uc_list);
4030 	adapter->uc_list = NULL;
4031 }
4032 
4033 static int be_clear(struct be_adapter *adapter)
4034 {
4035 	struct pci_dev *pdev = adapter->pdev;
4036 	struct  be_resources vft_res = {0};
4037 
4038 	be_cancel_worker(adapter);
4039 
4040 	flush_workqueue(be_wq);
4041 
4042 	if (sriov_enabled(adapter))
4043 		be_vf_clear(adapter);
4044 
4045 	/* Re-configure FW to distribute resources evenly across max-supported
4046 	 * number of VFs, only when VFs are not already enabled.
4047 	 */
4048 	if (skyhawk_chip(adapter) && be_physfn(adapter) &&
4049 	    !pci_vfs_assigned(pdev)) {
4050 		be_calculate_vf_res(adapter,
4051 				    pci_sriov_get_totalvfs(pdev),
4052 				    &vft_res);
4053 		be_cmd_set_sriov_config(adapter, adapter->pool_res,
4054 					pci_sriov_get_totalvfs(pdev),
4055 					&vft_res);
4056 	}
4057 
4058 	be_disable_vxlan_offloads(adapter);
4059 
4060 	be_if_destroy(adapter);
4061 
4062 	be_clear_queues(adapter);
4063 
4064 	be_msix_disable(adapter);
4065 	adapter->flags &= ~BE_FLAGS_SETUP_DONE;
4066 	return 0;
4067 }
4068 
4069 static int be_vfs_if_create(struct be_adapter *adapter)
4070 {
4071 	struct be_resources res = {0};
4072 	u32 cap_flags, en_flags, vf;
4073 	struct be_vf_cfg *vf_cfg;
4074 	int status;
4075 
4076 	/* If a FW profile exists, then cap_flags are updated */
4077 	cap_flags = BE_VF_IF_EN_FLAGS;
4078 
4079 	for_all_vfs(adapter, vf_cfg, vf) {
4080 		if (!BE3_chip(adapter)) {
4081 			status = be_cmd_get_profile_config(adapter, &res, NULL,
4082 							   ACTIVE_PROFILE_TYPE,
4083 							   RESOURCE_LIMITS,
4084 							   vf + 1);
4085 			if (!status) {
4086 				cap_flags = res.if_cap_flags;
4087 				/* Prevent VFs from enabling VLAN promiscuous
4088 				 * mode
4089 				 */
4090 				cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
4091 			}
4092 		}
4093 
4094 		/* PF should enable IF flags during proxy if_create call */
4095 		en_flags = cap_flags & BE_VF_IF_EN_FLAGS;
4096 		status = be_cmd_if_create(adapter, cap_flags, en_flags,
4097 					  &vf_cfg->if_handle, vf + 1);
4098 		if (status)
4099 			return status;
4100 	}
4101 
4102 	return 0;
4103 }
4104 
4105 static int be_vf_setup_init(struct be_adapter *adapter)
4106 {
4107 	struct be_vf_cfg *vf_cfg;
4108 	int vf;
4109 
4110 	adapter->vf_cfg = kcalloc(adapter->num_vfs, sizeof(*vf_cfg),
4111 				  GFP_KERNEL);
4112 	if (!adapter->vf_cfg)
4113 		return -ENOMEM;
4114 
4115 	for_all_vfs(adapter, vf_cfg, vf) {
4116 		vf_cfg->if_handle = -1;
4117 		vf_cfg->pmac_id = -1;
4118 	}
4119 	return 0;
4120 }
4121 
4122 static int be_vf_setup(struct be_adapter *adapter)
4123 {
4124 	struct device *dev = &adapter->pdev->dev;
4125 	struct be_vf_cfg *vf_cfg;
4126 	int status, old_vfs, vf;
4127 	bool spoofchk;
4128 
4129 	old_vfs = pci_num_vf(adapter->pdev);
4130 
4131 	status = be_vf_setup_init(adapter);
4132 	if (status)
4133 		goto err;
4134 
4135 	if (old_vfs) {
4136 		for_all_vfs(adapter, vf_cfg, vf) {
4137 			status = be_cmd_get_if_id(adapter, vf_cfg, vf);
4138 			if (status)
4139 				goto err;
4140 		}
4141 
4142 		status = be_vfs_mac_query(adapter);
4143 		if (status)
4144 			goto err;
4145 	} else {
4146 		status = be_vfs_if_create(adapter);
4147 		if (status)
4148 			goto err;
4149 
4150 		status = be_vf_eth_addr_config(adapter);
4151 		if (status)
4152 			goto err;
4153 	}
4154 
4155 	for_all_vfs(adapter, vf_cfg, vf) {
4156 		/* Allow VFs to programs MAC/VLAN filters */
4157 		status = be_cmd_get_fn_privileges(adapter, &vf_cfg->privileges,
4158 						  vf + 1);
4159 		if (!status && !(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
4160 			status = be_cmd_set_fn_privileges(adapter,
4161 							  vf_cfg->privileges |
4162 							  BE_PRIV_FILTMGMT,
4163 							  vf + 1);
4164 			if (!status) {
4165 				vf_cfg->privileges |= BE_PRIV_FILTMGMT;
4166 				dev_info(dev, "VF%d has FILTMGMT privilege\n",
4167 					 vf);
4168 			}
4169 		}
4170 
4171 		/* Allow full available bandwidth */
4172 		if (!old_vfs)
4173 			be_cmd_config_qos(adapter, 0, 0, vf + 1);
4174 
4175 		status = be_cmd_get_hsw_config(adapter, NULL, vf + 1,
4176 					       vf_cfg->if_handle, NULL,
4177 					       &spoofchk);
4178 		if (!status)
4179 			vf_cfg->spoofchk = spoofchk;
4180 
4181 		if (!old_vfs) {
4182 			be_cmd_enable_vf(adapter, vf + 1);
4183 			be_cmd_set_logical_link_config(adapter,
4184 						       IFLA_VF_LINK_STATE_AUTO,
4185 						       vf+1);
4186 		}
4187 	}
4188 
4189 	if (!old_vfs) {
4190 		status = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
4191 		if (status) {
4192 			dev_err(dev, "SRIOV enable failed\n");
4193 			adapter->num_vfs = 0;
4194 			goto err;
4195 		}
4196 	}
4197 
4198 	if (BE3_chip(adapter)) {
4199 		/* On BE3, enable VEB only when SRIOV is enabled */
4200 		status = be_cmd_set_hsw_config(adapter, 0, 0,
4201 					       adapter->if_handle,
4202 					       PORT_FWD_TYPE_VEB, 0);
4203 		if (status)
4204 			goto err;
4205 	}
4206 
4207 	adapter->flags |= BE_FLAGS_SRIOV_ENABLED;
4208 	return 0;
4209 err:
4210 	dev_err(dev, "VF setup failed\n");
4211 	be_vf_clear(adapter);
4212 	return status;
4213 }
4214 
4215 /* Converting function_mode bits on BE3 to SH mc_type enums */
4216 
4217 static u8 be_convert_mc_type(u32 function_mode)
4218 {
4219 	if (function_mode & VNIC_MODE && function_mode & QNQ_MODE)
4220 		return vNIC1;
4221 	else if (function_mode & QNQ_MODE)
4222 		return FLEX10;
4223 	else if (function_mode & VNIC_MODE)
4224 		return vNIC2;
4225 	else if (function_mode & UMC_ENABLED)
4226 		return UMC;
4227 	else
4228 		return MC_NONE;
4229 }
4230 
4231 /* On BE2/BE3 FW does not suggest the supported limits */
4232 static void BEx_get_resources(struct be_adapter *adapter,
4233 			      struct be_resources *res)
4234 {
4235 	bool use_sriov = adapter->num_vfs ? 1 : 0;
4236 
4237 	if (be_physfn(adapter))
4238 		res->max_uc_mac = BE_UC_PMAC_COUNT;
4239 	else
4240 		res->max_uc_mac = BE_VF_UC_PMAC_COUNT;
4241 
4242 	adapter->mc_type = be_convert_mc_type(adapter->function_mode);
4243 
4244 	if (be_is_mc(adapter)) {
4245 		/* Assuming that there are 4 channels per port,
4246 		 * when multi-channel is enabled
4247 		 */
4248 		if (be_is_qnq_mode(adapter))
4249 			res->max_vlans = BE_NUM_VLANS_SUPPORTED/8;
4250 		else
4251 			/* In a non-qnq multichannel mode, the pvid
4252 			 * takes up one vlan entry
4253 			 */
4254 			res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1;
4255 	} else {
4256 		res->max_vlans = BE_NUM_VLANS_SUPPORTED;
4257 	}
4258 
4259 	res->max_mcast_mac = BE_MAX_MC;
4260 
4261 	/* 1) For BE3 1Gb ports, FW does not support multiple TXQs
4262 	 * 2) Create multiple TX rings on a BE3-R multi-channel interface
4263 	 *    *only* if it is RSS-capable.
4264 	 */
4265 	if (BE2_chip(adapter) || use_sriov ||  (adapter->port_num > 1) ||
4266 	    be_virtfn(adapter) ||
4267 	    (be_is_mc(adapter) &&
4268 	     !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) {
4269 		res->max_tx_qs = 1;
4270 	} else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) {
4271 		struct be_resources super_nic_res = {0};
4272 
4273 		/* On a SuperNIC profile, the driver needs to use the
4274 		 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
4275 		 */
4276 		be_cmd_get_profile_config(adapter, &super_nic_res, NULL,
4277 					  ACTIVE_PROFILE_TYPE, RESOURCE_LIMITS,
4278 					  0);
4279 		/* Some old versions of BE3 FW don't report max_tx_qs value */
4280 		res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS;
4281 	} else {
4282 		res->max_tx_qs = BE3_MAX_TX_QS;
4283 	}
4284 
4285 	if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) &&
4286 	    !use_sriov && be_physfn(adapter))
4287 		res->max_rss_qs = (adapter->be3_native) ?
4288 					   BE3_MAX_RSS_QS : BE2_MAX_RSS_QS;
4289 	res->max_rx_qs = res->max_rss_qs + 1;
4290 
4291 	if (be_physfn(adapter))
4292 		res->max_evt_qs = (be_max_vfs(adapter) > 0) ?
4293 					BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS;
4294 	else
4295 		res->max_evt_qs = 1;
4296 
4297 	res->if_cap_flags = BE_IF_CAP_FLAGS_WANT;
4298 	res->if_cap_flags &= ~BE_IF_FLAGS_DEFQ_RSS;
4299 	if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS))
4300 		res->if_cap_flags &= ~BE_IF_FLAGS_RSS;
4301 }
4302 
4303 static void be_setup_init(struct be_adapter *adapter)
4304 {
4305 	adapter->vlan_prio_bmap = 0xff;
4306 	adapter->phy.link_speed = -1;
4307 	adapter->if_handle = -1;
4308 	adapter->be3_native = false;
4309 	adapter->if_flags = 0;
4310 	adapter->phy_state = BE_UNKNOWN_PHY_STATE;
4311 	if (be_physfn(adapter))
4312 		adapter->cmd_privileges = MAX_PRIVILEGES;
4313 	else
4314 		adapter->cmd_privileges = MIN_PRIVILEGES;
4315 }
4316 
4317 /* HW supports only MAX_PORT_RSS_TABLES RSS Policy Tables per port.
4318  * However, this HW limitation is not exposed to the host via any SLI cmd.
4319  * As a result, in the case of SRIOV and in particular multi-partition configs
4320  * the driver needs to calcuate a proportional share of RSS Tables per PF-pool
4321  * for distribution between the VFs. This self-imposed limit will determine the
4322  * no: of VFs for which RSS can be enabled.
4323  */
4324 static void be_calculate_pf_pool_rss_tables(struct be_adapter *adapter)
4325 {
4326 	struct be_port_resources port_res = {0};
4327 	u8 rss_tables_on_port;
4328 	u16 max_vfs = be_max_vfs(adapter);
4329 
4330 	be_cmd_get_profile_config(adapter, NULL, &port_res, SAVED_PROFILE_TYPE,
4331 				  RESOURCE_LIMITS, 0);
4332 
4333 	rss_tables_on_port = MAX_PORT_RSS_TABLES - port_res.nic_pfs;
4334 
4335 	/* Each PF Pool's RSS Tables limit =
4336 	 * PF's Max VFs / Total_Max_VFs on Port * RSS Tables on Port
4337 	 */
4338 	adapter->pool_res.max_rss_tables =
4339 		max_vfs * rss_tables_on_port / port_res.max_vfs;
4340 }
4341 
4342 static int be_get_sriov_config(struct be_adapter *adapter)
4343 {
4344 	struct be_resources res = {0};
4345 	int max_vfs, old_vfs;
4346 
4347 	be_cmd_get_profile_config(adapter, &res, NULL, ACTIVE_PROFILE_TYPE,
4348 				  RESOURCE_LIMITS, 0);
4349 
4350 	/* Some old versions of BE3 FW don't report max_vfs value */
4351 	if (BE3_chip(adapter) && !res.max_vfs) {
4352 		max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
4353 		res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
4354 	}
4355 
4356 	adapter->pool_res = res;
4357 
4358 	/* If during previous unload of the driver, the VFs were not disabled,
4359 	 * then we cannot rely on the PF POOL limits for the TotalVFs value.
4360 	 * Instead use the TotalVFs value stored in the pci-dev struct.
4361 	 */
4362 	old_vfs = pci_num_vf(adapter->pdev);
4363 	if (old_vfs) {
4364 		dev_info(&adapter->pdev->dev, "%d VFs are already enabled\n",
4365 			 old_vfs);
4366 
4367 		adapter->pool_res.max_vfs =
4368 			pci_sriov_get_totalvfs(adapter->pdev);
4369 		adapter->num_vfs = old_vfs;
4370 	}
4371 
4372 	if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4373 		be_calculate_pf_pool_rss_tables(adapter);
4374 		dev_info(&adapter->pdev->dev,
4375 			 "RSS can be enabled for all VFs if num_vfs <= %d\n",
4376 			 be_max_pf_pool_rss_tables(adapter));
4377 	}
4378 	return 0;
4379 }
4380 
4381 static void be_alloc_sriov_res(struct be_adapter *adapter)
4382 {
4383 	int old_vfs = pci_num_vf(adapter->pdev);
4384 	struct  be_resources vft_res = {0};
4385 	int status;
4386 
4387 	be_get_sriov_config(adapter);
4388 
4389 	if (!old_vfs)
4390 		pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
4391 
4392 	/* When the HW is in SRIOV capable configuration, the PF-pool
4393 	 * resources are given to PF during driver load, if there are no
4394 	 * old VFs. This facility is not available in BE3 FW.
4395 	 * Also, this is done by FW in Lancer chip.
4396 	 */
4397 	if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4398 		be_calculate_vf_res(adapter, 0, &vft_res);
4399 		status = be_cmd_set_sriov_config(adapter, adapter->pool_res, 0,
4400 						 &vft_res);
4401 		if (status)
4402 			dev_err(&adapter->pdev->dev,
4403 				"Failed to optimize SRIOV resources\n");
4404 	}
4405 }
4406 
4407 static int be_get_resources(struct be_adapter *adapter)
4408 {
4409 	struct device *dev = &adapter->pdev->dev;
4410 	struct be_resources res = {0};
4411 	int status;
4412 
4413 	/* For Lancer, SH etc read per-function resource limits from FW.
4414 	 * GET_FUNC_CONFIG returns per function guaranteed limits.
4415 	 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
4416 	 */
4417 	if (BEx_chip(adapter)) {
4418 		BEx_get_resources(adapter, &res);
4419 	} else {
4420 		status = be_cmd_get_func_config(adapter, &res);
4421 		if (status)
4422 			return status;
4423 
4424 		/* If a deafault RXQ must be created, we'll use up one RSSQ*/
4425 		if (res.max_rss_qs && res.max_rss_qs == res.max_rx_qs &&
4426 		    !(res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS))
4427 			res.max_rss_qs -= 1;
4428 	}
4429 
4430 	/* If RoCE is supported stash away half the EQs for RoCE */
4431 	res.max_nic_evt_qs = be_roce_supported(adapter) ?
4432 				res.max_evt_qs / 2 : res.max_evt_qs;
4433 	adapter->res = res;
4434 
4435 	/* If FW supports RSS default queue, then skip creating non-RSS
4436 	 * queue for non-IP traffic.
4437 	 */
4438 	adapter->need_def_rxq = (be_if_cap_flags(adapter) &
4439 				 BE_IF_FLAGS_DEFQ_RSS) ? 0 : 1;
4440 
4441 	dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
4442 		 be_max_txqs(adapter), be_max_rxqs(adapter),
4443 		 be_max_rss(adapter), be_max_nic_eqs(adapter),
4444 		 be_max_vfs(adapter));
4445 	dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
4446 		 be_max_uc(adapter), be_max_mc(adapter),
4447 		 be_max_vlans(adapter));
4448 
4449 	/* Ensure RX and TX queues are created in pairs at init time */
4450 	adapter->cfg_num_rx_irqs =
4451 				min_t(u16, netif_get_num_default_rss_queues(),
4452 				      be_max_qp_irqs(adapter));
4453 	adapter->cfg_num_tx_irqs = adapter->cfg_num_rx_irqs;
4454 	return 0;
4455 }
4456 
4457 static int be_get_config(struct be_adapter *adapter)
4458 {
4459 	int status, level;
4460 	u16 profile_id;
4461 
4462 	status = be_cmd_get_cntl_attributes(adapter);
4463 	if (status)
4464 		return status;
4465 
4466 	status = be_cmd_query_fw_cfg(adapter);
4467 	if (status)
4468 		return status;
4469 
4470 	if (!lancer_chip(adapter) && be_physfn(adapter))
4471 		be_cmd_get_fat_dump_len(adapter, &adapter->fat_dump_len);
4472 
4473 	if (BEx_chip(adapter)) {
4474 		level = be_cmd_get_fw_log_level(adapter);
4475 		adapter->msg_enable =
4476 			level <= FW_LOG_LEVEL_DEFAULT ? NETIF_MSG_HW : 0;
4477 	}
4478 
4479 	be_cmd_get_acpi_wol_cap(adapter);
4480 	pci_enable_wake(adapter->pdev, PCI_D3hot, adapter->wol_en);
4481 	pci_enable_wake(adapter->pdev, PCI_D3cold, adapter->wol_en);
4482 
4483 	be_cmd_query_port_name(adapter);
4484 
4485 	if (be_physfn(adapter)) {
4486 		status = be_cmd_get_active_profile(adapter, &profile_id);
4487 		if (!status)
4488 			dev_info(&adapter->pdev->dev,
4489 				 "Using profile 0x%x\n", profile_id);
4490 	}
4491 
4492 	return 0;
4493 }
4494 
4495 static int be_mac_setup(struct be_adapter *adapter)
4496 {
4497 	u8 mac[ETH_ALEN];
4498 	int status;
4499 
4500 	if (is_zero_ether_addr(adapter->netdev->dev_addr)) {
4501 		status = be_cmd_get_perm_mac(adapter, mac);
4502 		if (status)
4503 			return status;
4504 
4505 		memcpy(adapter->netdev->dev_addr, mac, ETH_ALEN);
4506 		memcpy(adapter->netdev->perm_addr, mac, ETH_ALEN);
4507 
4508 		/* Initial MAC for BE3 VFs is already programmed by PF */
4509 		if (BEx_chip(adapter) && be_virtfn(adapter))
4510 			memcpy(adapter->dev_mac, mac, ETH_ALEN);
4511 	}
4512 
4513 	return 0;
4514 }
4515 
4516 static void be_schedule_worker(struct be_adapter *adapter)
4517 {
4518 	queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
4519 	adapter->flags |= BE_FLAGS_WORKER_SCHEDULED;
4520 }
4521 
4522 static void be_destroy_err_recovery_workq(void)
4523 {
4524 	if (!be_err_recovery_workq)
4525 		return;
4526 
4527 	flush_workqueue(be_err_recovery_workq);
4528 	destroy_workqueue(be_err_recovery_workq);
4529 	be_err_recovery_workq = NULL;
4530 }
4531 
4532 static void be_schedule_err_detection(struct be_adapter *adapter, u32 delay)
4533 {
4534 	struct be_error_recovery *err_rec = &adapter->error_recovery;
4535 
4536 	if (!be_err_recovery_workq)
4537 		return;
4538 
4539 	queue_delayed_work(be_err_recovery_workq, &err_rec->err_detection_work,
4540 			   msecs_to_jiffies(delay));
4541 	adapter->flags |= BE_FLAGS_ERR_DETECTION_SCHEDULED;
4542 }
4543 
4544 static int be_setup_queues(struct be_adapter *adapter)
4545 {
4546 	struct net_device *netdev = adapter->netdev;
4547 	int status;
4548 
4549 	status = be_evt_queues_create(adapter);
4550 	if (status)
4551 		goto err;
4552 
4553 	status = be_tx_qs_create(adapter);
4554 	if (status)
4555 		goto err;
4556 
4557 	status = be_rx_cqs_create(adapter);
4558 	if (status)
4559 		goto err;
4560 
4561 	status = be_mcc_queues_create(adapter);
4562 	if (status)
4563 		goto err;
4564 
4565 	status = netif_set_real_num_rx_queues(netdev, adapter->num_rx_qs);
4566 	if (status)
4567 		goto err;
4568 
4569 	status = netif_set_real_num_tx_queues(netdev, adapter->num_tx_qs);
4570 	if (status)
4571 		goto err;
4572 
4573 	return 0;
4574 err:
4575 	dev_err(&adapter->pdev->dev, "queue_setup failed\n");
4576 	return status;
4577 }
4578 
4579 static int be_if_create(struct be_adapter *adapter)
4580 {
4581 	u32 en_flags = BE_IF_FLAGS_RSS | BE_IF_FLAGS_DEFQ_RSS;
4582 	u32 cap_flags = be_if_cap_flags(adapter);
4583 	int status;
4584 
4585 	/* alloc required memory for other filtering fields */
4586 	adapter->pmac_id = kcalloc(be_max_uc(adapter),
4587 				   sizeof(*adapter->pmac_id), GFP_KERNEL);
4588 	if (!adapter->pmac_id)
4589 		return -ENOMEM;
4590 
4591 	adapter->mc_list = kcalloc(be_max_mc(adapter),
4592 				   sizeof(*adapter->mc_list), GFP_KERNEL);
4593 	if (!adapter->mc_list)
4594 		return -ENOMEM;
4595 
4596 	adapter->uc_list = kcalloc(be_max_uc(adapter),
4597 				   sizeof(*adapter->uc_list), GFP_KERNEL);
4598 	if (!adapter->uc_list)
4599 		return -ENOMEM;
4600 
4601 	if (adapter->cfg_num_rx_irqs == 1)
4602 		cap_flags &= ~(BE_IF_FLAGS_DEFQ_RSS | BE_IF_FLAGS_RSS);
4603 
4604 	en_flags &= cap_flags;
4605 	/* will enable all the needed filter flags in be_open() */
4606 	status = be_cmd_if_create(adapter, be_if_cap_flags(adapter), en_flags,
4607 				  &adapter->if_handle, 0);
4608 
4609 	if (status)
4610 		return status;
4611 
4612 	return 0;
4613 }
4614 
4615 int be_update_queues(struct be_adapter *adapter)
4616 {
4617 	struct net_device *netdev = adapter->netdev;
4618 	int status;
4619 
4620 	if (netif_running(netdev))
4621 		be_close(netdev);
4622 
4623 	be_cancel_worker(adapter);
4624 
4625 	/* If any vectors have been shared with RoCE we cannot re-program
4626 	 * the MSIx table.
4627 	 */
4628 	if (!adapter->num_msix_roce_vec)
4629 		be_msix_disable(adapter);
4630 
4631 	be_clear_queues(adapter);
4632 	status = be_cmd_if_destroy(adapter, adapter->if_handle,  0);
4633 	if (status)
4634 		return status;
4635 
4636 	if (!msix_enabled(adapter)) {
4637 		status = be_msix_enable(adapter);
4638 		if (status)
4639 			return status;
4640 	}
4641 
4642 	status = be_if_create(adapter);
4643 	if (status)
4644 		return status;
4645 
4646 	status = be_setup_queues(adapter);
4647 	if (status)
4648 		return status;
4649 
4650 	be_schedule_worker(adapter);
4651 
4652 	/* The IF was destroyed and re-created. We need to clear
4653 	 * all promiscuous flags valid for the destroyed IF.
4654 	 * Without this promisc mode is not restored during
4655 	 * be_open() because the driver thinks that it is
4656 	 * already enabled in HW.
4657 	 */
4658 	adapter->if_flags &= ~BE_IF_FLAGS_ALL_PROMISCUOUS;
4659 
4660 	if (netif_running(netdev))
4661 		status = be_open(netdev);
4662 
4663 	return status;
4664 }
4665 
4666 static inline int fw_major_num(const char *fw_ver)
4667 {
4668 	int fw_major = 0, i;
4669 
4670 	i = sscanf(fw_ver, "%d.", &fw_major);
4671 	if (i != 1)
4672 		return 0;
4673 
4674 	return fw_major;
4675 }
4676 
4677 /* If it is error recovery, FLR the PF
4678  * Else if any VFs are already enabled don't FLR the PF
4679  */
4680 static bool be_reset_required(struct be_adapter *adapter)
4681 {
4682 	if (be_error_recovering(adapter))
4683 		return true;
4684 	else
4685 		return pci_num_vf(adapter->pdev) == 0;
4686 }
4687 
4688 /* Wait for the FW to be ready and perform the required initialization */
4689 static int be_func_init(struct be_adapter *adapter)
4690 {
4691 	int status;
4692 
4693 	status = be_fw_wait_ready(adapter);
4694 	if (status)
4695 		return status;
4696 
4697 	/* FW is now ready; clear errors to allow cmds/doorbell */
4698 	be_clear_error(adapter, BE_CLEAR_ALL);
4699 
4700 	if (be_reset_required(adapter)) {
4701 		status = be_cmd_reset_function(adapter);
4702 		if (status)
4703 			return status;
4704 
4705 		/* Wait for interrupts to quiesce after an FLR */
4706 		msleep(100);
4707 	}
4708 
4709 	/* Tell FW we're ready to fire cmds */
4710 	status = be_cmd_fw_init(adapter);
4711 	if (status)
4712 		return status;
4713 
4714 	/* Allow interrupts for other ULPs running on NIC function */
4715 	be_intr_set(adapter, true);
4716 
4717 	return 0;
4718 }
4719 
4720 static int be_setup(struct be_adapter *adapter)
4721 {
4722 	struct device *dev = &adapter->pdev->dev;
4723 	int status;
4724 
4725 	status = be_func_init(adapter);
4726 	if (status)
4727 		return status;
4728 
4729 	be_setup_init(adapter);
4730 
4731 	if (!lancer_chip(adapter))
4732 		be_cmd_req_native_mode(adapter);
4733 
4734 	/* invoke this cmd first to get pf_num and vf_num which are needed
4735 	 * for issuing profile related cmds
4736 	 */
4737 	if (!BEx_chip(adapter)) {
4738 		status = be_cmd_get_func_config(adapter, NULL);
4739 		if (status)
4740 			return status;
4741 	}
4742 
4743 	status = be_get_config(adapter);
4744 	if (status)
4745 		goto err;
4746 
4747 	if (!BE2_chip(adapter) && be_physfn(adapter))
4748 		be_alloc_sriov_res(adapter);
4749 
4750 	status = be_get_resources(adapter);
4751 	if (status)
4752 		goto err;
4753 
4754 	status = be_msix_enable(adapter);
4755 	if (status)
4756 		goto err;
4757 
4758 	/* will enable all the needed filter flags in be_open() */
4759 	status = be_if_create(adapter);
4760 	if (status)
4761 		goto err;
4762 
4763 	/* Updating real_num_tx/rx_queues() requires rtnl_lock() */
4764 	rtnl_lock();
4765 	status = be_setup_queues(adapter);
4766 	rtnl_unlock();
4767 	if (status)
4768 		goto err;
4769 
4770 	be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0);
4771 
4772 	status = be_mac_setup(adapter);
4773 	if (status)
4774 		goto err;
4775 
4776 	be_cmd_get_fw_ver(adapter);
4777 	dev_info(dev, "FW version is %s\n", adapter->fw_ver);
4778 
4779 	if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) {
4780 		dev_err(dev, "Firmware on card is old(%s), IRQs may not work",
4781 			adapter->fw_ver);
4782 		dev_err(dev, "Please upgrade firmware to version >= 4.0\n");
4783 	}
4784 
4785 	status = be_cmd_set_flow_control(adapter, adapter->tx_fc,
4786 					 adapter->rx_fc);
4787 	if (status)
4788 		be_cmd_get_flow_control(adapter, &adapter->tx_fc,
4789 					&adapter->rx_fc);
4790 
4791 	dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n",
4792 		 adapter->tx_fc, adapter->rx_fc);
4793 
4794 	if (be_physfn(adapter))
4795 		be_cmd_set_logical_link_config(adapter,
4796 					       IFLA_VF_LINK_STATE_AUTO, 0);
4797 
4798 	/* BE3 EVB echoes broadcast/multicast packets back to PF's vport
4799 	 * confusing a linux bridge or OVS that it might be connected to.
4800 	 * Set the EVB to PASSTHRU mode which effectively disables the EVB
4801 	 * when SRIOV is not enabled.
4802 	 */
4803 	if (BE3_chip(adapter))
4804 		be_cmd_set_hsw_config(adapter, 0, 0, adapter->if_handle,
4805 				      PORT_FWD_TYPE_PASSTHRU, 0);
4806 
4807 	if (adapter->num_vfs)
4808 		be_vf_setup(adapter);
4809 
4810 	status = be_cmd_get_phy_info(adapter);
4811 	if (!status && be_pause_supported(adapter))
4812 		adapter->phy.fc_autoneg = 1;
4813 
4814 	if (be_physfn(adapter) && !lancer_chip(adapter))
4815 		be_cmd_set_features(adapter);
4816 
4817 	be_schedule_worker(adapter);
4818 	adapter->flags |= BE_FLAGS_SETUP_DONE;
4819 	return 0;
4820 err:
4821 	be_clear(adapter);
4822 	return status;
4823 }
4824 
4825 #ifdef CONFIG_NET_POLL_CONTROLLER
4826 static void be_netpoll(struct net_device *netdev)
4827 {
4828 	struct be_adapter *adapter = netdev_priv(netdev);
4829 	struct be_eq_obj *eqo;
4830 	int i;
4831 
4832 	for_all_evt_queues(adapter, eqo, i) {
4833 		be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
4834 		napi_schedule(&eqo->napi);
4835 	}
4836 }
4837 #endif
4838 
4839 int be_load_fw(struct be_adapter *adapter, u8 *fw_file)
4840 {
4841 	const struct firmware *fw;
4842 	int status;
4843 
4844 	if (!netif_running(adapter->netdev)) {
4845 		dev_err(&adapter->pdev->dev,
4846 			"Firmware load not allowed (interface is down)\n");
4847 		return -ENETDOWN;
4848 	}
4849 
4850 	status = request_firmware(&fw, fw_file, &adapter->pdev->dev);
4851 	if (status)
4852 		goto fw_exit;
4853 
4854 	dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file);
4855 
4856 	if (lancer_chip(adapter))
4857 		status = lancer_fw_download(adapter, fw);
4858 	else
4859 		status = be_fw_download(adapter, fw);
4860 
4861 	if (!status)
4862 		be_cmd_get_fw_ver(adapter);
4863 
4864 fw_exit:
4865 	release_firmware(fw);
4866 	return status;
4867 }
4868 
4869 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4870 				 u16 flags)
4871 {
4872 	struct be_adapter *adapter = netdev_priv(dev);
4873 	struct nlattr *attr, *br_spec;
4874 	int rem;
4875 	int status = 0;
4876 	u16 mode = 0;
4877 
4878 	if (!sriov_enabled(adapter))
4879 		return -EOPNOTSUPP;
4880 
4881 	br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4882 	if (!br_spec)
4883 		return -EINVAL;
4884 
4885 	nla_for_each_nested(attr, br_spec, rem) {
4886 		if (nla_type(attr) != IFLA_BRIDGE_MODE)
4887 			continue;
4888 
4889 		if (nla_len(attr) < sizeof(mode))
4890 			return -EINVAL;
4891 
4892 		mode = nla_get_u16(attr);
4893 		if (BE3_chip(adapter) && mode == BRIDGE_MODE_VEPA)
4894 			return -EOPNOTSUPP;
4895 
4896 		if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB)
4897 			return -EINVAL;
4898 
4899 		status = be_cmd_set_hsw_config(adapter, 0, 0,
4900 					       adapter->if_handle,
4901 					       mode == BRIDGE_MODE_VEPA ?
4902 					       PORT_FWD_TYPE_VEPA :
4903 					       PORT_FWD_TYPE_VEB, 0);
4904 		if (status)
4905 			goto err;
4906 
4907 		dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n",
4908 			 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4909 
4910 		return status;
4911 	}
4912 err:
4913 	dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n",
4914 		mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4915 
4916 	return status;
4917 }
4918 
4919 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
4920 				 struct net_device *dev, u32 filter_mask,
4921 				 int nlflags)
4922 {
4923 	struct be_adapter *adapter = netdev_priv(dev);
4924 	int status = 0;
4925 	u8 hsw_mode;
4926 
4927 	/* BE and Lancer chips support VEB mode only */
4928 	if (BEx_chip(adapter) || lancer_chip(adapter)) {
4929 		/* VEB is disabled in non-SR-IOV profiles on BE3/Lancer */
4930 		if (!pci_sriov_get_totalvfs(adapter->pdev))
4931 			return 0;
4932 		hsw_mode = PORT_FWD_TYPE_VEB;
4933 	} else {
4934 		status = be_cmd_get_hsw_config(adapter, NULL, 0,
4935 					       adapter->if_handle, &hsw_mode,
4936 					       NULL);
4937 		if (status)
4938 			return 0;
4939 
4940 		if (hsw_mode == PORT_FWD_TYPE_PASSTHRU)
4941 			return 0;
4942 	}
4943 
4944 	return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
4945 				       hsw_mode == PORT_FWD_TYPE_VEPA ?
4946 				       BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB,
4947 				       0, 0, nlflags, filter_mask, NULL);
4948 }
4949 
4950 static struct be_cmd_work *be_alloc_work(struct be_adapter *adapter,
4951 					 void (*func)(struct work_struct *))
4952 {
4953 	struct be_cmd_work *work;
4954 
4955 	work = kzalloc(sizeof(*work), GFP_ATOMIC);
4956 	if (!work) {
4957 		dev_err(&adapter->pdev->dev,
4958 			"be_work memory allocation failed\n");
4959 		return NULL;
4960 	}
4961 
4962 	INIT_WORK(&work->work, func);
4963 	work->adapter = adapter;
4964 	return work;
4965 }
4966 
4967 /* VxLAN offload Notes:
4968  *
4969  * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
4970  * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
4971  * is expected to work across all types of IP tunnels once exported. Skyhawk
4972  * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
4973  * offloads in hw_enc_features only when a VxLAN port is added. If other (non
4974  * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
4975  * those other tunnels are unexported on the fly through ndo_features_check().
4976  *
4977  * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
4978  * adds more than one port, disable offloads and re-enable them again when
4979  * there's only one port left. We maintain a list of ports for this purpose.
4980  */
4981 static void be_work_add_vxlan_port(struct work_struct *work)
4982 {
4983 	struct be_cmd_work *cmd_work =
4984 				container_of(work, struct be_cmd_work, work);
4985 	struct be_adapter *adapter = cmd_work->adapter;
4986 	struct device *dev = &adapter->pdev->dev;
4987 	__be16 port = cmd_work->info.vxlan_port;
4988 	struct be_vxlan_port *vxlan_port;
4989 	int status;
4990 
4991 	/* Bump up the alias count if it is an existing port */
4992 	list_for_each_entry(vxlan_port, &adapter->vxlan_port_list, list) {
4993 		if (vxlan_port->port == port) {
4994 			vxlan_port->port_aliases++;
4995 			goto done;
4996 		}
4997 	}
4998 
4999 	/* Add a new port to our list. We don't need a lock here since port
5000 	 * add/delete are done only in the context of a single-threaded work
5001 	 * queue (be_wq).
5002 	 */
5003 	vxlan_port = kzalloc(sizeof(*vxlan_port), GFP_KERNEL);
5004 	if (!vxlan_port)
5005 		goto done;
5006 
5007 	vxlan_port->port = port;
5008 	INIT_LIST_HEAD(&vxlan_port->list);
5009 	list_add_tail(&vxlan_port->list, &adapter->vxlan_port_list);
5010 	adapter->vxlan_port_count++;
5011 
5012 	if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) {
5013 		dev_info(dev,
5014 			 "Only one UDP port supported for VxLAN offloads\n");
5015 		dev_info(dev, "Disabling VxLAN offloads\n");
5016 		goto err;
5017 	}
5018 
5019 	if (adapter->vxlan_port_count > 1)
5020 		goto done;
5021 
5022 	status = be_enable_vxlan_offloads(adapter);
5023 	if (!status)
5024 		goto done;
5025 
5026 err:
5027 	be_disable_vxlan_offloads(adapter);
5028 done:
5029 	kfree(cmd_work);
5030 	return;
5031 }
5032 
5033 static void be_work_del_vxlan_port(struct work_struct *work)
5034 {
5035 	struct be_cmd_work *cmd_work =
5036 				container_of(work, struct be_cmd_work, work);
5037 	struct be_adapter *adapter = cmd_work->adapter;
5038 	__be16 port = cmd_work->info.vxlan_port;
5039 	struct be_vxlan_port *vxlan_port;
5040 
5041 	/* Nothing to be done if a port alias is being deleted */
5042 	list_for_each_entry(vxlan_port, &adapter->vxlan_port_list, list) {
5043 		if (vxlan_port->port == port) {
5044 			if (vxlan_port->port_aliases) {
5045 				vxlan_port->port_aliases--;
5046 				goto done;
5047 			}
5048 			break;
5049 		}
5050 	}
5051 
5052 	/* No port aliases left; delete the port from the list */
5053 	list_del(&vxlan_port->list);
5054 	adapter->vxlan_port_count--;
5055 
5056 	/* Disable VxLAN offload if this is the offloaded port */
5057 	if (adapter->vxlan_port == vxlan_port->port) {
5058 		WARN_ON(adapter->vxlan_port_count);
5059 		be_disable_vxlan_offloads(adapter);
5060 		dev_info(&adapter->pdev->dev,
5061 			 "Disabled VxLAN offloads for UDP port %d\n",
5062 			 be16_to_cpu(port));
5063 		goto out;
5064 	}
5065 
5066 	/* If only 1 port is left, re-enable VxLAN offload */
5067 	if (adapter->vxlan_port_count == 1)
5068 		be_enable_vxlan_offloads(adapter);
5069 
5070 out:
5071 	kfree(vxlan_port);
5072 done:
5073 	kfree(cmd_work);
5074 }
5075 
5076 static void be_cfg_vxlan_port(struct net_device *netdev,
5077 			      struct udp_tunnel_info *ti,
5078 			      void (*func)(struct work_struct *))
5079 {
5080 	struct be_adapter *adapter = netdev_priv(netdev);
5081 	struct be_cmd_work *cmd_work;
5082 
5083 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
5084 		return;
5085 
5086 	if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
5087 		return;
5088 
5089 	cmd_work = be_alloc_work(adapter, func);
5090 	if (cmd_work) {
5091 		cmd_work->info.vxlan_port = ti->port;
5092 		queue_work(be_wq, &cmd_work->work);
5093 	}
5094 }
5095 
5096 static void be_del_vxlan_port(struct net_device *netdev,
5097 			      struct udp_tunnel_info *ti)
5098 {
5099 	be_cfg_vxlan_port(netdev, ti, be_work_del_vxlan_port);
5100 }
5101 
5102 static void be_add_vxlan_port(struct net_device *netdev,
5103 			      struct udp_tunnel_info *ti)
5104 {
5105 	be_cfg_vxlan_port(netdev, ti, be_work_add_vxlan_port);
5106 }
5107 
5108 static netdev_features_t be_features_check(struct sk_buff *skb,
5109 					   struct net_device *dev,
5110 					   netdev_features_t features)
5111 {
5112 	struct be_adapter *adapter = netdev_priv(dev);
5113 	u8 l4_hdr = 0;
5114 
5115 	if (skb_is_gso(skb)) {
5116 		/* IPv6 TSO requests with extension hdrs are a problem
5117 		 * to Lancer and BE3 HW. Disable TSO6 feature.
5118 		 */
5119 		if (!skyhawk_chip(adapter) && is_ipv6_ext_hdr(skb))
5120 			features &= ~NETIF_F_TSO6;
5121 
5122 		/* Lancer cannot handle the packet with MSS less than 256.
5123 		 * Also it can't handle a TSO packet with a single segment
5124 		 * Disable the GSO support in such cases
5125 		 */
5126 		if (lancer_chip(adapter) &&
5127 		    (skb_shinfo(skb)->gso_size < 256 ||
5128 		     skb_shinfo(skb)->gso_segs == 1))
5129 			features &= ~NETIF_F_GSO_MASK;
5130 	}
5131 
5132 	/* The code below restricts offload features for some tunneled and
5133 	 * Q-in-Q packets.
5134 	 * Offload features for normal (non tunnel) packets are unchanged.
5135 	 */
5136 	features = vlan_features_check(skb, features);
5137 	if (!skb->encapsulation ||
5138 	    !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS))
5139 		return features;
5140 
5141 	/* It's an encapsulated packet and VxLAN offloads are enabled. We
5142 	 * should disable tunnel offload features if it's not a VxLAN packet,
5143 	 * as tunnel offloads have been enabled only for VxLAN. This is done to
5144 	 * allow other tunneled traffic like GRE work fine while VxLAN
5145 	 * offloads are configured in Skyhawk-R.
5146 	 */
5147 	switch (vlan_get_protocol(skb)) {
5148 	case htons(ETH_P_IP):
5149 		l4_hdr = ip_hdr(skb)->protocol;
5150 		break;
5151 	case htons(ETH_P_IPV6):
5152 		l4_hdr = ipv6_hdr(skb)->nexthdr;
5153 		break;
5154 	default:
5155 		return features;
5156 	}
5157 
5158 	if (l4_hdr != IPPROTO_UDP ||
5159 	    skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
5160 	    skb->inner_protocol != htons(ETH_P_TEB) ||
5161 	    skb_inner_mac_header(skb) - skb_transport_header(skb) !=
5162 		sizeof(struct udphdr) + sizeof(struct vxlanhdr) ||
5163 	    !adapter->vxlan_port ||
5164 	    udp_hdr(skb)->dest != adapter->vxlan_port)
5165 		return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
5166 
5167 	return features;
5168 }
5169 
5170 static int be_get_phys_port_id(struct net_device *dev,
5171 			       struct netdev_phys_item_id *ppid)
5172 {
5173 	int i, id_len = CNTL_SERIAL_NUM_WORDS * CNTL_SERIAL_NUM_WORD_SZ + 1;
5174 	struct be_adapter *adapter = netdev_priv(dev);
5175 	u8 *id;
5176 
5177 	if (MAX_PHYS_ITEM_ID_LEN < id_len)
5178 		return -ENOSPC;
5179 
5180 	ppid->id[0] = adapter->hba_port_num + 1;
5181 	id = &ppid->id[1];
5182 	for (i = CNTL_SERIAL_NUM_WORDS - 1; i >= 0;
5183 	     i--, id += CNTL_SERIAL_NUM_WORD_SZ)
5184 		memcpy(id, &adapter->serial_num[i], CNTL_SERIAL_NUM_WORD_SZ);
5185 
5186 	ppid->id_len = id_len;
5187 
5188 	return 0;
5189 }
5190 
5191 static void be_set_rx_mode(struct net_device *dev)
5192 {
5193 	struct be_adapter *adapter = netdev_priv(dev);
5194 	struct be_cmd_work *work;
5195 
5196 	work = be_alloc_work(adapter, be_work_set_rx_mode);
5197 	if (work)
5198 		queue_work(be_wq, &work->work);
5199 }
5200 
5201 static const struct net_device_ops be_netdev_ops = {
5202 	.ndo_open		= be_open,
5203 	.ndo_stop		= be_close,
5204 	.ndo_start_xmit		= be_xmit,
5205 	.ndo_set_rx_mode	= be_set_rx_mode,
5206 	.ndo_set_mac_address	= be_mac_addr_set,
5207 	.ndo_get_stats64	= be_get_stats64,
5208 	.ndo_validate_addr	= eth_validate_addr,
5209 	.ndo_vlan_rx_add_vid	= be_vlan_add_vid,
5210 	.ndo_vlan_rx_kill_vid	= be_vlan_rem_vid,
5211 	.ndo_set_vf_mac		= be_set_vf_mac,
5212 	.ndo_set_vf_vlan	= be_set_vf_vlan,
5213 	.ndo_set_vf_rate	= be_set_vf_tx_rate,
5214 	.ndo_get_vf_config	= be_get_vf_config,
5215 	.ndo_set_vf_link_state  = be_set_vf_link_state,
5216 	.ndo_set_vf_spoofchk    = be_set_vf_spoofchk,
5217 #ifdef CONFIG_NET_POLL_CONTROLLER
5218 	.ndo_poll_controller	= be_netpoll,
5219 #endif
5220 	.ndo_bridge_setlink	= be_ndo_bridge_setlink,
5221 	.ndo_bridge_getlink	= be_ndo_bridge_getlink,
5222 	.ndo_udp_tunnel_add	= be_add_vxlan_port,
5223 	.ndo_udp_tunnel_del	= be_del_vxlan_port,
5224 	.ndo_features_check	= be_features_check,
5225 	.ndo_get_phys_port_id   = be_get_phys_port_id,
5226 };
5227 
5228 static void be_netdev_init(struct net_device *netdev)
5229 {
5230 	struct be_adapter *adapter = netdev_priv(netdev);
5231 
5232 	netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5233 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
5234 		NETIF_F_HW_VLAN_CTAG_TX;
5235 	if ((be_if_cap_flags(adapter) & BE_IF_FLAGS_RSS))
5236 		netdev->hw_features |= NETIF_F_RXHASH;
5237 
5238 	netdev->features |= netdev->hw_features |
5239 		NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
5240 
5241 	netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
5242 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
5243 
5244 	netdev->priv_flags |= IFF_UNICAST_FLT;
5245 
5246 	netdev->flags |= IFF_MULTICAST;
5247 
5248 	netif_set_gso_max_size(netdev, BE_MAX_GSO_SIZE - ETH_HLEN);
5249 
5250 	netdev->netdev_ops = &be_netdev_ops;
5251 
5252 	netdev->ethtool_ops = &be_ethtool_ops;
5253 
5254 	/* MTU range: 256 - 9000 */
5255 	netdev->min_mtu = BE_MIN_MTU;
5256 	netdev->max_mtu = BE_MAX_MTU;
5257 }
5258 
5259 static void be_cleanup(struct be_adapter *adapter)
5260 {
5261 	struct net_device *netdev = adapter->netdev;
5262 
5263 	rtnl_lock();
5264 	netif_device_detach(netdev);
5265 	if (netif_running(netdev))
5266 		be_close(netdev);
5267 	rtnl_unlock();
5268 
5269 	be_clear(adapter);
5270 }
5271 
5272 static int be_resume(struct be_adapter *adapter)
5273 {
5274 	struct net_device *netdev = adapter->netdev;
5275 	int status;
5276 
5277 	status = be_setup(adapter);
5278 	if (status)
5279 		return status;
5280 
5281 	rtnl_lock();
5282 	if (netif_running(netdev))
5283 		status = be_open(netdev);
5284 	rtnl_unlock();
5285 
5286 	if (status)
5287 		return status;
5288 
5289 	netif_device_attach(netdev);
5290 
5291 	return 0;
5292 }
5293 
5294 static void be_soft_reset(struct be_adapter *adapter)
5295 {
5296 	u32 val;
5297 
5298 	dev_info(&adapter->pdev->dev, "Initiating chip soft reset\n");
5299 	val = ioread32(adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5300 	val |= SLIPORT_SOFTRESET_SR_MASK;
5301 	iowrite32(val, adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET);
5302 }
5303 
5304 static bool be_err_is_recoverable(struct be_adapter *adapter)
5305 {
5306 	struct be_error_recovery *err_rec = &adapter->error_recovery;
5307 	unsigned long initial_idle_time =
5308 		msecs_to_jiffies(ERR_RECOVERY_IDLE_TIME);
5309 	unsigned long recovery_interval =
5310 		msecs_to_jiffies(ERR_RECOVERY_INTERVAL);
5311 	u16 ue_err_code;
5312 	u32 val;
5313 
5314 	val = be_POST_stage_get(adapter);
5315 	if ((val & POST_STAGE_RECOVERABLE_ERR) != POST_STAGE_RECOVERABLE_ERR)
5316 		return false;
5317 	ue_err_code = val & POST_ERR_RECOVERY_CODE_MASK;
5318 	if (ue_err_code == 0)
5319 		return false;
5320 
5321 	dev_err(&adapter->pdev->dev, "Recoverable HW error code: 0x%x\n",
5322 		ue_err_code);
5323 
5324 	if (time_before_eq(jiffies - err_rec->probe_time, initial_idle_time)) {
5325 		dev_err(&adapter->pdev->dev,
5326 			"Cannot recover within %lu sec from driver load\n",
5327 			jiffies_to_msecs(initial_idle_time) / MSEC_PER_SEC);
5328 		return false;
5329 	}
5330 
5331 	if (err_rec->last_recovery_time && time_before_eq(
5332 		jiffies - err_rec->last_recovery_time, recovery_interval)) {
5333 		dev_err(&adapter->pdev->dev,
5334 			"Cannot recover within %lu sec from last recovery\n",
5335 			jiffies_to_msecs(recovery_interval) / MSEC_PER_SEC);
5336 		return false;
5337 	}
5338 
5339 	if (ue_err_code == err_rec->last_err_code) {
5340 		dev_err(&adapter->pdev->dev,
5341 			"Cannot recover from a consecutive TPE error\n");
5342 		return false;
5343 	}
5344 
5345 	err_rec->last_recovery_time = jiffies;
5346 	err_rec->last_err_code = ue_err_code;
5347 	return true;
5348 }
5349 
5350 static int be_tpe_recover(struct be_adapter *adapter)
5351 {
5352 	struct be_error_recovery *err_rec = &adapter->error_recovery;
5353 	int status = -EAGAIN;
5354 	u32 val;
5355 
5356 	switch (err_rec->recovery_state) {
5357 	case ERR_RECOVERY_ST_NONE:
5358 		err_rec->recovery_state = ERR_RECOVERY_ST_DETECT;
5359 		err_rec->resched_delay = ERR_RECOVERY_UE_DETECT_DURATION;
5360 		break;
5361 
5362 	case ERR_RECOVERY_ST_DETECT:
5363 		val = be_POST_stage_get(adapter);
5364 		if ((val & POST_STAGE_RECOVERABLE_ERR) !=
5365 		    POST_STAGE_RECOVERABLE_ERR) {
5366 			dev_err(&adapter->pdev->dev,
5367 				"Unrecoverable HW error detected: 0x%x\n", val);
5368 			status = -EINVAL;
5369 			err_rec->resched_delay = 0;
5370 			break;
5371 		}
5372 
5373 		dev_err(&adapter->pdev->dev, "Recoverable HW error detected\n");
5374 
5375 		/* Only PF0 initiates Chip Soft Reset. But PF0 must wait UE2SR
5376 		 * milliseconds before it checks for final error status in
5377 		 * SLIPORT_SEMAPHORE to determine if recovery criteria is met.
5378 		 * If it does, then PF0 initiates a Soft Reset.
5379 		 */
5380 		if (adapter->pf_num == 0) {
5381 			err_rec->recovery_state = ERR_RECOVERY_ST_RESET;
5382 			err_rec->resched_delay = err_rec->ue_to_reset_time -
5383 					ERR_RECOVERY_UE_DETECT_DURATION;
5384 			break;
5385 		}
5386 
5387 		err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5388 		err_rec->resched_delay = err_rec->ue_to_poll_time -
5389 					ERR_RECOVERY_UE_DETECT_DURATION;
5390 		break;
5391 
5392 	case ERR_RECOVERY_ST_RESET:
5393 		if (!be_err_is_recoverable(adapter)) {
5394 			dev_err(&adapter->pdev->dev,
5395 				"Failed to meet recovery criteria\n");
5396 			status = -EIO;
5397 			err_rec->resched_delay = 0;
5398 			break;
5399 		}
5400 		be_soft_reset(adapter);
5401 		err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL;
5402 		err_rec->resched_delay = err_rec->ue_to_poll_time -
5403 					err_rec->ue_to_reset_time;
5404 		break;
5405 
5406 	case ERR_RECOVERY_ST_PRE_POLL:
5407 		err_rec->recovery_state = ERR_RECOVERY_ST_REINIT;
5408 		err_rec->resched_delay = 0;
5409 		status = 0;			/* done */
5410 		break;
5411 
5412 	default:
5413 		status = -EINVAL;
5414 		err_rec->resched_delay = 0;
5415 		break;
5416 	}
5417 
5418 	return status;
5419 }
5420 
5421 static int be_err_recover(struct be_adapter *adapter)
5422 {
5423 	int status;
5424 
5425 	if (!lancer_chip(adapter)) {
5426 		if (!adapter->error_recovery.recovery_supported ||
5427 		    adapter->priv_flags & BE_DISABLE_TPE_RECOVERY)
5428 			return -EIO;
5429 		status = be_tpe_recover(adapter);
5430 		if (status)
5431 			goto err;
5432 	}
5433 
5434 	/* Wait for adapter to reach quiescent state before
5435 	 * destroying queues
5436 	 */
5437 	status = be_fw_wait_ready(adapter);
5438 	if (status)
5439 		goto err;
5440 
5441 	adapter->flags |= BE_FLAGS_TRY_RECOVERY;
5442 
5443 	be_cleanup(adapter);
5444 
5445 	status = be_resume(adapter);
5446 	if (status)
5447 		goto err;
5448 
5449 	adapter->flags &= ~BE_FLAGS_TRY_RECOVERY;
5450 
5451 err:
5452 	return status;
5453 }
5454 
5455 static void be_err_detection_task(struct work_struct *work)
5456 {
5457 	struct be_error_recovery *err_rec =
5458 			container_of(work, struct be_error_recovery,
5459 				     err_detection_work.work);
5460 	struct be_adapter *adapter =
5461 			container_of(err_rec, struct be_adapter,
5462 				     error_recovery);
5463 	u32 resched_delay = ERR_RECOVERY_DETECTION_DELAY;
5464 	struct device *dev = &adapter->pdev->dev;
5465 	int recovery_status;
5466 
5467 	be_detect_error(adapter);
5468 	if (!be_check_error(adapter, BE_ERROR_HW))
5469 		goto reschedule_task;
5470 
5471 	recovery_status = be_err_recover(adapter);
5472 	if (!recovery_status) {
5473 		err_rec->recovery_retries = 0;
5474 		err_rec->recovery_state = ERR_RECOVERY_ST_NONE;
5475 		dev_info(dev, "Adapter recovery successful\n");
5476 		goto reschedule_task;
5477 	} else if (!lancer_chip(adapter) && err_rec->resched_delay) {
5478 		/* BEx/SH recovery state machine */
5479 		if (adapter->pf_num == 0 &&
5480 		    err_rec->recovery_state > ERR_RECOVERY_ST_DETECT)
5481 			dev_err(&adapter->pdev->dev,
5482 				"Adapter recovery in progress\n");
5483 		resched_delay = err_rec->resched_delay;
5484 		goto reschedule_task;
5485 	} else if (lancer_chip(adapter) && be_virtfn(adapter)) {
5486 		/* For VFs, check if PF have allocated resources
5487 		 * every second.
5488 		 */
5489 		dev_err(dev, "Re-trying adapter recovery\n");
5490 		goto reschedule_task;
5491 	} else if (lancer_chip(adapter) && err_rec->recovery_retries++ <
5492 		   ERR_RECOVERY_MAX_RETRY_COUNT) {
5493 		/* In case of another error during recovery, it takes 30 sec
5494 		 * for adapter to come out of error. Retry error recovery after
5495 		 * this time interval.
5496 		 */
5497 		dev_err(&adapter->pdev->dev, "Re-trying adapter recovery\n");
5498 		resched_delay = ERR_RECOVERY_RETRY_DELAY;
5499 		goto reschedule_task;
5500 	} else {
5501 		dev_err(dev, "Adapter recovery failed\n");
5502 		dev_err(dev, "Please reboot server to recover\n");
5503 	}
5504 
5505 	return;
5506 
5507 reschedule_task:
5508 	be_schedule_err_detection(adapter, resched_delay);
5509 }
5510 
5511 static void be_log_sfp_info(struct be_adapter *adapter)
5512 {
5513 	int status;
5514 
5515 	status = be_cmd_query_sfp_info(adapter);
5516 	if (!status) {
5517 		dev_err(&adapter->pdev->dev,
5518 			"Port %c: %s Vendor: %s part no: %s",
5519 			adapter->port_name,
5520 			be_misconfig_evt_port_state[adapter->phy_state],
5521 			adapter->phy.vendor_name,
5522 			adapter->phy.vendor_pn);
5523 	}
5524 	adapter->flags &= ~BE_FLAGS_PHY_MISCONFIGURED;
5525 }
5526 
5527 static void be_worker(struct work_struct *work)
5528 {
5529 	struct be_adapter *adapter =
5530 		container_of(work, struct be_adapter, work.work);
5531 	struct be_rx_obj *rxo;
5532 	int i;
5533 
5534 	if (be_physfn(adapter) &&
5535 	    MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0)
5536 		be_cmd_get_die_temperature(adapter);
5537 
5538 	/* when interrupts are not yet enabled, just reap any pending
5539 	 * mcc completions
5540 	 */
5541 	if (!netif_running(adapter->netdev)) {
5542 		local_bh_disable();
5543 		be_process_mcc(adapter);
5544 		local_bh_enable();
5545 		goto reschedule;
5546 	}
5547 
5548 	if (!adapter->stats_cmd_sent) {
5549 		if (lancer_chip(adapter))
5550 			lancer_cmd_get_pport_stats(adapter,
5551 						   &adapter->stats_cmd);
5552 		else
5553 			be_cmd_get_stats(adapter, &adapter->stats_cmd);
5554 	}
5555 
5556 	for_all_rx_queues(adapter, rxo, i) {
5557 		/* Replenish RX-queues starved due to memory
5558 		 * allocation failures.
5559 		 */
5560 		if (rxo->rx_post_starved)
5561 			be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
5562 	}
5563 
5564 	/* EQ-delay update for Skyhawk is done while notifying EQ */
5565 	if (!skyhawk_chip(adapter))
5566 		be_eqd_update(adapter, false);
5567 
5568 	if (adapter->flags & BE_FLAGS_PHY_MISCONFIGURED)
5569 		be_log_sfp_info(adapter);
5570 
5571 reschedule:
5572 	adapter->work_counter++;
5573 	queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000));
5574 }
5575 
5576 static void be_unmap_pci_bars(struct be_adapter *adapter)
5577 {
5578 	if (adapter->csr)
5579 		pci_iounmap(adapter->pdev, adapter->csr);
5580 	if (adapter->db)
5581 		pci_iounmap(adapter->pdev, adapter->db);
5582 	if (adapter->pcicfg && adapter->pcicfg_mapped)
5583 		pci_iounmap(adapter->pdev, adapter->pcicfg);
5584 }
5585 
5586 static int db_bar(struct be_adapter *adapter)
5587 {
5588 	if (lancer_chip(adapter) || be_virtfn(adapter))
5589 		return 0;
5590 	else
5591 		return 4;
5592 }
5593 
5594 static int be_roce_map_pci_bars(struct be_adapter *adapter)
5595 {
5596 	if (skyhawk_chip(adapter)) {
5597 		adapter->roce_db.size = 4096;
5598 		adapter->roce_db.io_addr = pci_resource_start(adapter->pdev,
5599 							      db_bar(adapter));
5600 		adapter->roce_db.total_size = pci_resource_len(adapter->pdev,
5601 							       db_bar(adapter));
5602 	}
5603 	return 0;
5604 }
5605 
5606 static int be_map_pci_bars(struct be_adapter *adapter)
5607 {
5608 	struct pci_dev *pdev = adapter->pdev;
5609 	u8 __iomem *addr;
5610 	u32 sli_intf;
5611 
5612 	pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf);
5613 	adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >>
5614 				SLI_INTF_FAMILY_SHIFT;
5615 	adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0;
5616 
5617 	if (BEx_chip(adapter) && be_physfn(adapter)) {
5618 		adapter->csr = pci_iomap(pdev, 2, 0);
5619 		if (!adapter->csr)
5620 			return -ENOMEM;
5621 	}
5622 
5623 	addr = pci_iomap(pdev, db_bar(adapter), 0);
5624 	if (!addr)
5625 		goto pci_map_err;
5626 	adapter->db = addr;
5627 
5628 	if (skyhawk_chip(adapter) || BEx_chip(adapter)) {
5629 		if (be_physfn(adapter)) {
5630 			/* PCICFG is the 2nd BAR in BE2 */
5631 			addr = pci_iomap(pdev, BE2_chip(adapter) ? 1 : 0, 0);
5632 			if (!addr)
5633 				goto pci_map_err;
5634 			adapter->pcicfg = addr;
5635 			adapter->pcicfg_mapped = true;
5636 		} else {
5637 			adapter->pcicfg = adapter->db + SRIOV_VF_PCICFG_OFFSET;
5638 			adapter->pcicfg_mapped = false;
5639 		}
5640 	}
5641 
5642 	be_roce_map_pci_bars(adapter);
5643 	return 0;
5644 
5645 pci_map_err:
5646 	dev_err(&pdev->dev, "Error in mapping PCI BARs\n");
5647 	be_unmap_pci_bars(adapter);
5648 	return -ENOMEM;
5649 }
5650 
5651 static void be_drv_cleanup(struct be_adapter *adapter)
5652 {
5653 	struct be_dma_mem *mem = &adapter->mbox_mem_alloced;
5654 	struct device *dev = &adapter->pdev->dev;
5655 
5656 	if (mem->va)
5657 		dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5658 
5659 	mem = &adapter->rx_filter;
5660 	if (mem->va)
5661 		dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5662 
5663 	mem = &adapter->stats_cmd;
5664 	if (mem->va)
5665 		dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5666 }
5667 
5668 /* Allocate and initialize various fields in be_adapter struct */
5669 static int be_drv_init(struct be_adapter *adapter)
5670 {
5671 	struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced;
5672 	struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem;
5673 	struct be_dma_mem *rx_filter = &adapter->rx_filter;
5674 	struct be_dma_mem *stats_cmd = &adapter->stats_cmd;
5675 	struct device *dev = &adapter->pdev->dev;
5676 	int status = 0;
5677 
5678 	mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
5679 	mbox_mem_alloc->va = dma_zalloc_coherent(dev, mbox_mem_alloc->size,
5680 						 &mbox_mem_alloc->dma,
5681 						 GFP_KERNEL);
5682 	if (!mbox_mem_alloc->va)
5683 		return -ENOMEM;
5684 
5685 	mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
5686 	mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
5687 	mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
5688 
5689 	rx_filter->size = sizeof(struct be_cmd_req_rx_filter);
5690 	rx_filter->va = dma_zalloc_coherent(dev, rx_filter->size,
5691 					    &rx_filter->dma, GFP_KERNEL);
5692 	if (!rx_filter->va) {
5693 		status = -ENOMEM;
5694 		goto free_mbox;
5695 	}
5696 
5697 	if (lancer_chip(adapter))
5698 		stats_cmd->size = sizeof(struct lancer_cmd_req_pport_stats);
5699 	else if (BE2_chip(adapter))
5700 		stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v0);
5701 	else if (BE3_chip(adapter))
5702 		stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v1);
5703 	else
5704 		stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v2);
5705 	stats_cmd->va = dma_zalloc_coherent(dev, stats_cmd->size,
5706 					    &stats_cmd->dma, GFP_KERNEL);
5707 	if (!stats_cmd->va) {
5708 		status = -ENOMEM;
5709 		goto free_rx_filter;
5710 	}
5711 
5712 	mutex_init(&adapter->mbox_lock);
5713 	mutex_init(&adapter->mcc_lock);
5714 	mutex_init(&adapter->rx_filter_lock);
5715 	spin_lock_init(&adapter->mcc_cq_lock);
5716 	init_completion(&adapter->et_cmd_compl);
5717 
5718 	pci_save_state(adapter->pdev);
5719 
5720 	INIT_DELAYED_WORK(&adapter->work, be_worker);
5721 
5722 	adapter->error_recovery.recovery_state = ERR_RECOVERY_ST_NONE;
5723 	adapter->error_recovery.resched_delay = 0;
5724 	INIT_DELAYED_WORK(&adapter->error_recovery.err_detection_work,
5725 			  be_err_detection_task);
5726 
5727 	adapter->rx_fc = true;
5728 	adapter->tx_fc = true;
5729 
5730 	/* Must be a power of 2 or else MODULO will BUG_ON */
5731 	adapter->be_get_temp_freq = 64;
5732 
5733 	INIT_LIST_HEAD(&adapter->vxlan_port_list);
5734 	return 0;
5735 
5736 free_rx_filter:
5737 	dma_free_coherent(dev, rx_filter->size, rx_filter->va, rx_filter->dma);
5738 free_mbox:
5739 	dma_free_coherent(dev, mbox_mem_alloc->size, mbox_mem_alloc->va,
5740 			  mbox_mem_alloc->dma);
5741 	return status;
5742 }
5743 
5744 static void be_remove(struct pci_dev *pdev)
5745 {
5746 	struct be_adapter *adapter = pci_get_drvdata(pdev);
5747 
5748 	if (!adapter)
5749 		return;
5750 
5751 	be_roce_dev_remove(adapter);
5752 	be_intr_set(adapter, false);
5753 
5754 	be_cancel_err_detection(adapter);
5755 
5756 	unregister_netdev(adapter->netdev);
5757 
5758 	be_clear(adapter);
5759 
5760 	if (!pci_vfs_assigned(adapter->pdev))
5761 		be_cmd_reset_function(adapter);
5762 
5763 	/* tell fw we're done with firing cmds */
5764 	be_cmd_fw_clean(adapter);
5765 
5766 	be_unmap_pci_bars(adapter);
5767 	be_drv_cleanup(adapter);
5768 
5769 	pci_disable_pcie_error_reporting(pdev);
5770 
5771 	pci_release_regions(pdev);
5772 	pci_disable_device(pdev);
5773 
5774 	free_netdev(adapter->netdev);
5775 }
5776 
5777 static ssize_t be_hwmon_show_temp(struct device *dev,
5778 				  struct device_attribute *dev_attr,
5779 				  char *buf)
5780 {
5781 	struct be_adapter *adapter = dev_get_drvdata(dev);
5782 
5783 	/* Unit: millidegree Celsius */
5784 	if (adapter->hwmon_info.be_on_die_temp == BE_INVALID_DIE_TEMP)
5785 		return -EIO;
5786 	else
5787 		return sprintf(buf, "%u\n",
5788 			       adapter->hwmon_info.be_on_die_temp * 1000);
5789 }
5790 
5791 static SENSOR_DEVICE_ATTR(temp1_input, 0444,
5792 			  be_hwmon_show_temp, NULL, 1);
5793 
5794 static struct attribute *be_hwmon_attrs[] = {
5795 	&sensor_dev_attr_temp1_input.dev_attr.attr,
5796 	NULL
5797 };
5798 
5799 ATTRIBUTE_GROUPS(be_hwmon);
5800 
5801 static char *mc_name(struct be_adapter *adapter)
5802 {
5803 	char *str = "";	/* default */
5804 
5805 	switch (adapter->mc_type) {
5806 	case UMC:
5807 		str = "UMC";
5808 		break;
5809 	case FLEX10:
5810 		str = "FLEX10";
5811 		break;
5812 	case vNIC1:
5813 		str = "vNIC-1";
5814 		break;
5815 	case nPAR:
5816 		str = "nPAR";
5817 		break;
5818 	case UFP:
5819 		str = "UFP";
5820 		break;
5821 	case vNIC2:
5822 		str = "vNIC-2";
5823 		break;
5824 	default:
5825 		str = "";
5826 	}
5827 
5828 	return str;
5829 }
5830 
5831 static inline char *func_name(struct be_adapter *adapter)
5832 {
5833 	return be_physfn(adapter) ? "PF" : "VF";
5834 }
5835 
5836 static inline char *nic_name(struct pci_dev *pdev)
5837 {
5838 	switch (pdev->device) {
5839 	case OC_DEVICE_ID1:
5840 		return OC_NAME;
5841 	case OC_DEVICE_ID2:
5842 		return OC_NAME_BE;
5843 	case OC_DEVICE_ID3:
5844 	case OC_DEVICE_ID4:
5845 		return OC_NAME_LANCER;
5846 	case BE_DEVICE_ID2:
5847 		return BE3_NAME;
5848 	case OC_DEVICE_ID5:
5849 	case OC_DEVICE_ID6:
5850 		return OC_NAME_SH;
5851 	default:
5852 		return BE_NAME;
5853 	}
5854 }
5855 
5856 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id)
5857 {
5858 	struct be_adapter *adapter;
5859 	struct net_device *netdev;
5860 	int status = 0;
5861 
5862 	dev_info(&pdev->dev, "%s version is %s\n", DRV_NAME, DRV_VER);
5863 
5864 	status = pci_enable_device(pdev);
5865 	if (status)
5866 		goto do_none;
5867 
5868 	status = pci_request_regions(pdev, DRV_NAME);
5869 	if (status)
5870 		goto disable_dev;
5871 	pci_set_master(pdev);
5872 
5873 	netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS);
5874 	if (!netdev) {
5875 		status = -ENOMEM;
5876 		goto rel_reg;
5877 	}
5878 	adapter = netdev_priv(netdev);
5879 	adapter->pdev = pdev;
5880 	pci_set_drvdata(pdev, adapter);
5881 	adapter->netdev = netdev;
5882 	SET_NETDEV_DEV(netdev, &pdev->dev);
5883 
5884 	status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
5885 	if (!status) {
5886 		netdev->features |= NETIF_F_HIGHDMA;
5887 	} else {
5888 		status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
5889 		if (status) {
5890 			dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
5891 			goto free_netdev;
5892 		}
5893 	}
5894 
5895 	status = pci_enable_pcie_error_reporting(pdev);
5896 	if (!status)
5897 		dev_info(&pdev->dev, "PCIe error reporting enabled\n");
5898 
5899 	status = be_map_pci_bars(adapter);
5900 	if (status)
5901 		goto free_netdev;
5902 
5903 	status = be_drv_init(adapter);
5904 	if (status)
5905 		goto unmap_bars;
5906 
5907 	status = be_setup(adapter);
5908 	if (status)
5909 		goto drv_cleanup;
5910 
5911 	be_netdev_init(netdev);
5912 	status = register_netdev(netdev);
5913 	if (status != 0)
5914 		goto unsetup;
5915 
5916 	be_roce_dev_add(adapter);
5917 
5918 	be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5919 	adapter->error_recovery.probe_time = jiffies;
5920 
5921 	/* On Die temperature not supported for VF. */
5922 	if (be_physfn(adapter) && IS_ENABLED(CONFIG_BE2NET_HWMON)) {
5923 		adapter->hwmon_info.hwmon_dev =
5924 			devm_hwmon_device_register_with_groups(&pdev->dev,
5925 							       DRV_NAME,
5926 							       adapter,
5927 							       be_hwmon_groups);
5928 		adapter->hwmon_info.be_on_die_temp = BE_INVALID_DIE_TEMP;
5929 	}
5930 
5931 	dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev),
5932 		 func_name(adapter), mc_name(adapter), adapter->port_name);
5933 
5934 	return 0;
5935 
5936 unsetup:
5937 	be_clear(adapter);
5938 drv_cleanup:
5939 	be_drv_cleanup(adapter);
5940 unmap_bars:
5941 	be_unmap_pci_bars(adapter);
5942 free_netdev:
5943 	free_netdev(netdev);
5944 rel_reg:
5945 	pci_release_regions(pdev);
5946 disable_dev:
5947 	pci_disable_device(pdev);
5948 do_none:
5949 	dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
5950 	return status;
5951 }
5952 
5953 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
5954 {
5955 	struct be_adapter *adapter = pci_get_drvdata(pdev);
5956 
5957 	be_intr_set(adapter, false);
5958 	be_cancel_err_detection(adapter);
5959 
5960 	be_cleanup(adapter);
5961 
5962 	pci_save_state(pdev);
5963 	pci_disable_device(pdev);
5964 	pci_set_power_state(pdev, pci_choose_state(pdev, state));
5965 	return 0;
5966 }
5967 
5968 static int be_pci_resume(struct pci_dev *pdev)
5969 {
5970 	struct be_adapter *adapter = pci_get_drvdata(pdev);
5971 	int status = 0;
5972 
5973 	status = pci_enable_device(pdev);
5974 	if (status)
5975 		return status;
5976 
5977 	pci_restore_state(pdev);
5978 
5979 	status = be_resume(adapter);
5980 	if (status)
5981 		return status;
5982 
5983 	be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5984 
5985 	return 0;
5986 }
5987 
5988 /*
5989  * An FLR will stop BE from DMAing any data.
5990  */
5991 static void be_shutdown(struct pci_dev *pdev)
5992 {
5993 	struct be_adapter *adapter = pci_get_drvdata(pdev);
5994 
5995 	if (!adapter)
5996 		return;
5997 
5998 	be_roce_dev_shutdown(adapter);
5999 	cancel_delayed_work_sync(&adapter->work);
6000 	be_cancel_err_detection(adapter);
6001 
6002 	netif_device_detach(adapter->netdev);
6003 
6004 	be_cmd_reset_function(adapter);
6005 
6006 	pci_disable_device(pdev);
6007 }
6008 
6009 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev,
6010 					    pci_channel_state_t state)
6011 {
6012 	struct be_adapter *adapter = pci_get_drvdata(pdev);
6013 
6014 	dev_err(&adapter->pdev->dev, "EEH error detected\n");
6015 
6016 	be_roce_dev_remove(adapter);
6017 
6018 	if (!be_check_error(adapter, BE_ERROR_EEH)) {
6019 		be_set_error(adapter, BE_ERROR_EEH);
6020 
6021 		be_cancel_err_detection(adapter);
6022 
6023 		be_cleanup(adapter);
6024 	}
6025 
6026 	if (state == pci_channel_io_perm_failure)
6027 		return PCI_ERS_RESULT_DISCONNECT;
6028 
6029 	pci_disable_device(pdev);
6030 
6031 	/* The error could cause the FW to trigger a flash debug dump.
6032 	 * Resetting the card while flash dump is in progress
6033 	 * can cause it not to recover; wait for it to finish.
6034 	 * Wait only for first function as it is needed only once per
6035 	 * adapter.
6036 	 */
6037 	if (pdev->devfn == 0)
6038 		ssleep(30);
6039 
6040 	return PCI_ERS_RESULT_NEED_RESET;
6041 }
6042 
6043 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev)
6044 {
6045 	struct be_adapter *adapter = pci_get_drvdata(pdev);
6046 	int status;
6047 
6048 	dev_info(&adapter->pdev->dev, "EEH reset\n");
6049 
6050 	status = pci_enable_device(pdev);
6051 	if (status)
6052 		return PCI_ERS_RESULT_DISCONNECT;
6053 
6054 	pci_set_master(pdev);
6055 	pci_restore_state(pdev);
6056 
6057 	/* Check if card is ok and fw is ready */
6058 	dev_info(&adapter->pdev->dev,
6059 		 "Waiting for FW to be ready after EEH reset\n");
6060 	status = be_fw_wait_ready(adapter);
6061 	if (status)
6062 		return PCI_ERS_RESULT_DISCONNECT;
6063 
6064 	pci_cleanup_aer_uncorrect_error_status(pdev);
6065 	be_clear_error(adapter, BE_CLEAR_ALL);
6066 	return PCI_ERS_RESULT_RECOVERED;
6067 }
6068 
6069 static void be_eeh_resume(struct pci_dev *pdev)
6070 {
6071 	int status = 0;
6072 	struct be_adapter *adapter = pci_get_drvdata(pdev);
6073 
6074 	dev_info(&adapter->pdev->dev, "EEH resume\n");
6075 
6076 	pci_save_state(pdev);
6077 
6078 	status = be_resume(adapter);
6079 	if (status)
6080 		goto err;
6081 
6082 	be_roce_dev_add(adapter);
6083 
6084 	be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
6085 	return;
6086 err:
6087 	dev_err(&adapter->pdev->dev, "EEH resume failed\n");
6088 }
6089 
6090 static int be_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
6091 {
6092 	struct be_adapter *adapter = pci_get_drvdata(pdev);
6093 	struct be_resources vft_res = {0};
6094 	int status;
6095 
6096 	if (!num_vfs)
6097 		be_vf_clear(adapter);
6098 
6099 	adapter->num_vfs = num_vfs;
6100 
6101 	if (adapter->num_vfs == 0 && pci_vfs_assigned(pdev)) {
6102 		dev_warn(&pdev->dev,
6103 			 "Cannot disable VFs while they are assigned\n");
6104 		return -EBUSY;
6105 	}
6106 
6107 	/* When the HW is in SRIOV capable configuration, the PF-pool resources
6108 	 * are equally distributed across the max-number of VFs. The user may
6109 	 * request only a subset of the max-vfs to be enabled.
6110 	 * Based on num_vfs, redistribute the resources across num_vfs so that
6111 	 * each VF will have access to more number of resources.
6112 	 * This facility is not available in BE3 FW.
6113 	 * Also, this is done by FW in Lancer chip.
6114 	 */
6115 	if (skyhawk_chip(adapter) && !pci_num_vf(pdev)) {
6116 		be_calculate_vf_res(adapter, adapter->num_vfs,
6117 				    &vft_res);
6118 		status = be_cmd_set_sriov_config(adapter, adapter->pool_res,
6119 						 adapter->num_vfs, &vft_res);
6120 		if (status)
6121 			dev_err(&pdev->dev,
6122 				"Failed to optimize SR-IOV resources\n");
6123 	}
6124 
6125 	status = be_get_resources(adapter);
6126 	if (status)
6127 		return be_cmd_status(status);
6128 
6129 	/* Updating real_num_tx/rx_queues() requires rtnl_lock() */
6130 	rtnl_lock();
6131 	status = be_update_queues(adapter);
6132 	rtnl_unlock();
6133 	if (status)
6134 		return be_cmd_status(status);
6135 
6136 	if (adapter->num_vfs)
6137 		status = be_vf_setup(adapter);
6138 
6139 	if (!status)
6140 		return adapter->num_vfs;
6141 
6142 	return 0;
6143 }
6144 
6145 static const struct pci_error_handlers be_eeh_handlers = {
6146 	.error_detected = be_eeh_err_detected,
6147 	.slot_reset = be_eeh_reset,
6148 	.resume = be_eeh_resume,
6149 };
6150 
6151 static struct pci_driver be_driver = {
6152 	.name = DRV_NAME,
6153 	.id_table = be_dev_ids,
6154 	.probe = be_probe,
6155 	.remove = be_remove,
6156 	.suspend = be_suspend,
6157 	.resume = be_pci_resume,
6158 	.shutdown = be_shutdown,
6159 	.sriov_configure = be_pci_sriov_configure,
6160 	.err_handler = &be_eeh_handlers
6161 };
6162 
6163 static int __init be_init_module(void)
6164 {
6165 	int status;
6166 
6167 	if (rx_frag_size != 8192 && rx_frag_size != 4096 &&
6168 	    rx_frag_size != 2048) {
6169 		printk(KERN_WARNING DRV_NAME
6170 			" : Module param rx_frag_size must be 2048/4096/8192."
6171 			" Using 2048\n");
6172 		rx_frag_size = 2048;
6173 	}
6174 
6175 	if (num_vfs > 0) {
6176 		pr_info(DRV_NAME " : Module param num_vfs is obsolete.");
6177 		pr_info(DRV_NAME " : Use sysfs method to enable VFs\n");
6178 	}
6179 
6180 	be_wq = create_singlethread_workqueue("be_wq");
6181 	if (!be_wq) {
6182 		pr_warn(DRV_NAME "workqueue creation failed\n");
6183 		return -1;
6184 	}
6185 
6186 	be_err_recovery_workq =
6187 		create_singlethread_workqueue("be_err_recover");
6188 	if (!be_err_recovery_workq)
6189 		pr_warn(DRV_NAME "Could not create error recovery workqueue\n");
6190 
6191 	status = pci_register_driver(&be_driver);
6192 	if (status) {
6193 		destroy_workqueue(be_wq);
6194 		be_destroy_err_recovery_workq();
6195 	}
6196 	return status;
6197 }
6198 module_init(be_init_module);
6199 
6200 static void __exit be_exit_module(void)
6201 {
6202 	pci_unregister_driver(&be_driver);
6203 
6204 	be_destroy_err_recovery_workq();
6205 
6206 	if (be_wq)
6207 		destroy_workqueue(be_wq);
6208 }
6209 module_exit(be_exit_module);
6210