1 /**********************************************************************
2  * Author: Cavium, Inc.
3  *
4  * Contact: support@cavium.com
5  *          Please include "LiquidIO" in the subject.
6  *
7  * Copyright (c) 2003-2016 Cavium, Inc.
8  *
9  * This file is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License, Version 2, as
11  * published by the Free Software Foundation.
12  *
13  * This file is distributed in the hope that it will be useful, but
14  * AS-IS and WITHOUT ANY WARRANTY; without even the implied warranty
15  * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE, TITLE, or
16  * NONINFRINGEMENT.  See the GNU General Public License for more details.
17  ***********************************************************************/
18 #include <linux/pci.h>
19 #include <linux/firmware.h>
20 #include <net/vxlan.h>
21 #include <linux/kthread.h>
22 #include "liquidio_common.h"
23 #include "octeon_droq.h"
24 #include "octeon_iq.h"
25 #include "response_manager.h"
26 #include "octeon_device.h"
27 #include "octeon_nic.h"
28 #include "octeon_main.h"
29 #include "octeon_network.h"
30 #include "cn66xx_regs.h"
31 #include "cn66xx_device.h"
32 #include "cn68xx_device.h"
33 #include "cn23xx_pf_device.h"
34 #include "liquidio_image.h"
35 
36 MODULE_AUTHOR("Cavium Networks, <support@cavium.com>");
37 MODULE_DESCRIPTION("Cavium LiquidIO Intelligent Server Adapter Driver");
38 MODULE_LICENSE("GPL");
39 MODULE_VERSION(LIQUIDIO_VERSION);
40 MODULE_FIRMWARE(LIO_FW_DIR LIO_FW_BASE_NAME LIO_210SV_NAME LIO_FW_NAME_SUFFIX);
41 MODULE_FIRMWARE(LIO_FW_DIR LIO_FW_BASE_NAME LIO_210NV_NAME LIO_FW_NAME_SUFFIX);
42 MODULE_FIRMWARE(LIO_FW_DIR LIO_FW_BASE_NAME LIO_410NV_NAME LIO_FW_NAME_SUFFIX);
43 MODULE_FIRMWARE(LIO_FW_DIR LIO_FW_BASE_NAME LIO_23XX_NAME LIO_FW_NAME_SUFFIX);
44 
45 static int ddr_timeout = 10000;
46 module_param(ddr_timeout, int, 0644);
47 MODULE_PARM_DESC(ddr_timeout,
48 		 "Number of milliseconds to wait for DDR initialization. 0 waits for ddr_timeout to be set to non-zero value before starting to check");
49 
50 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK)
51 
52 static int debug = -1;
53 module_param(debug, int, 0644);
54 MODULE_PARM_DESC(debug, "NETIF_MSG debug bits");
55 
56 static char fw_type[LIO_MAX_FW_TYPE_LEN];
57 module_param_string(fw_type, fw_type, sizeof(fw_type), 0000);
58 MODULE_PARM_DESC(fw_type, "Type of firmware to be loaded. Default \"nic\"");
59 
60 static int ptp_enable = 1;
61 
62 /* Bit mask values for lio->ifstate */
63 #define   LIO_IFSTATE_DROQ_OPS             0x01
64 #define   LIO_IFSTATE_REGISTERED           0x02
65 #define   LIO_IFSTATE_RUNNING              0x04
66 #define   LIO_IFSTATE_RX_TIMESTAMP_ENABLED 0x08
67 
68 /* Polling interval for determining when NIC application is alive */
69 #define LIQUIDIO_STARTER_POLL_INTERVAL_MS 100
70 
71 /* runtime link query interval */
72 #define LIQUIDIO_LINK_QUERY_INTERVAL_MS         1000
73 
74 struct liquidio_if_cfg_context {
75 	int octeon_id;
76 
77 	wait_queue_head_t wc;
78 
79 	int cond;
80 };
81 
82 struct liquidio_if_cfg_resp {
83 	u64 rh;
84 	struct liquidio_if_cfg_info cfg_info;
85 	u64 status;
86 };
87 
88 struct liquidio_rx_ctl_context {
89 	int octeon_id;
90 
91 	wait_queue_head_t wc;
92 
93 	int cond;
94 };
95 
96 struct oct_link_status_resp {
97 	u64 rh;
98 	struct oct_link_info link_info;
99 	u64 status;
100 };
101 
102 struct oct_timestamp_resp {
103 	u64 rh;
104 	u64 timestamp;
105 	u64 status;
106 };
107 
108 #define OCT_TIMESTAMP_RESP_SIZE (sizeof(struct oct_timestamp_resp))
109 
110 union tx_info {
111 	u64 u64;
112 	struct {
113 #ifdef __BIG_ENDIAN_BITFIELD
114 		u16 gso_size;
115 		u16 gso_segs;
116 		u32 reserved;
117 #else
118 		u32 reserved;
119 		u16 gso_segs;
120 		u16 gso_size;
121 #endif
122 	} s;
123 };
124 
125 /** Octeon device properties to be used by the NIC module.
126  * Each octeon device in the system will be represented
127  * by this structure in the NIC module.
128  */
129 
130 #define OCTNIC_MAX_SG  (MAX_SKB_FRAGS)
131 
132 #define OCTNIC_GSO_MAX_HEADER_SIZE 128
133 #define OCTNIC_GSO_MAX_SIZE                                                    \
134 	(CN23XX_DEFAULT_INPUT_JABBER - OCTNIC_GSO_MAX_HEADER_SIZE)
135 
136 /** Structure of a node in list of gather components maintained by
137  * NIC driver for each network device.
138  */
139 struct octnic_gather {
140 	/** List manipulation. Next and prev pointers. */
141 	struct list_head list;
142 
143 	/** Size of the gather component at sg in bytes. */
144 	int sg_size;
145 
146 	/** Number of bytes that sg was adjusted to make it 8B-aligned. */
147 	int adjust;
148 
149 	/** Gather component that can accommodate max sized fragment list
150 	 *  received from the IP layer.
151 	 */
152 	struct octeon_sg_entry *sg;
153 
154 	u64 sg_dma_ptr;
155 };
156 
157 struct handshake {
158 	struct completion init;
159 	struct completion started;
160 	struct pci_dev *pci_dev;
161 	int init_ok;
162 	int started_ok;
163 };
164 
165 struct octeon_device_priv {
166 	/** Tasklet structures for this device. */
167 	struct tasklet_struct droq_tasklet;
168 	unsigned long napi_mask;
169 };
170 
171 #ifdef CONFIG_PCI_IOV
172 static int liquidio_enable_sriov(struct pci_dev *dev, int num_vfs);
173 #endif
174 
175 static int octeon_device_init(struct octeon_device *);
176 static int liquidio_stop(struct net_device *netdev);
177 static void liquidio_remove(struct pci_dev *pdev);
178 static int liquidio_probe(struct pci_dev *pdev,
179 			  const struct pci_device_id *ent);
180 
181 static struct handshake handshake[MAX_OCTEON_DEVICES];
182 static struct completion first_stage;
183 
184 static void octeon_droq_bh(unsigned long pdev)
185 {
186 	int q_no;
187 	int reschedule = 0;
188 	struct octeon_device *oct = (struct octeon_device *)pdev;
189 	struct octeon_device_priv *oct_priv =
190 		(struct octeon_device_priv *)oct->priv;
191 
192 	for (q_no = 0; q_no < MAX_OCTEON_OUTPUT_QUEUES(oct); q_no++) {
193 		if (!(oct->io_qmask.oq & BIT_ULL(q_no)))
194 			continue;
195 		reschedule |= octeon_droq_process_packets(oct, oct->droq[q_no],
196 							  MAX_PACKET_BUDGET);
197 		lio_enable_irq(oct->droq[q_no], NULL);
198 
199 		if (OCTEON_CN23XX_PF(oct) && oct->msix_on) {
200 			/* set time and cnt interrupt thresholds for this DROQ
201 			 * for NAPI
202 			 */
203 			int adjusted_q_no = q_no + oct->sriov_info.pf_srn;
204 
205 			octeon_write_csr64(
206 			    oct, CN23XX_SLI_OQ_PKT_INT_LEVELS(adjusted_q_no),
207 			    0x5700000040ULL);
208 			octeon_write_csr64(
209 			    oct, CN23XX_SLI_OQ_PKTS_SENT(adjusted_q_no), 0);
210 		}
211 	}
212 
213 	if (reschedule)
214 		tasklet_schedule(&oct_priv->droq_tasklet);
215 }
216 
217 static int lio_wait_for_oq_pkts(struct octeon_device *oct)
218 {
219 	struct octeon_device_priv *oct_priv =
220 		(struct octeon_device_priv *)oct->priv;
221 	int retry = 100, pkt_cnt = 0, pending_pkts = 0;
222 	int i;
223 
224 	do {
225 		pending_pkts = 0;
226 
227 		for (i = 0; i < MAX_OCTEON_OUTPUT_QUEUES(oct); i++) {
228 			if (!(oct->io_qmask.oq & BIT_ULL(i)))
229 				continue;
230 			pkt_cnt += octeon_droq_check_hw_for_pkts(oct->droq[i]);
231 		}
232 		if (pkt_cnt > 0) {
233 			pending_pkts += pkt_cnt;
234 			tasklet_schedule(&oct_priv->droq_tasklet);
235 		}
236 		pkt_cnt = 0;
237 		schedule_timeout_uninterruptible(1);
238 
239 	} while (retry-- && pending_pkts);
240 
241 	return pkt_cnt;
242 }
243 
244 /**
245  * \brief Forces all IO queues off on a given device
246  * @param oct Pointer to Octeon device
247  */
248 static void force_io_queues_off(struct octeon_device *oct)
249 {
250 	if ((oct->chip_id == OCTEON_CN66XX) ||
251 	    (oct->chip_id == OCTEON_CN68XX)) {
252 		/* Reset the Enable bits for Input Queues. */
253 		octeon_write_csr(oct, CN6XXX_SLI_PKT_INSTR_ENB, 0);
254 
255 		/* Reset the Enable bits for Output Queues. */
256 		octeon_write_csr(oct, CN6XXX_SLI_PKT_OUT_ENB, 0);
257 	}
258 }
259 
260 /**
261  * \brief wait for all pending requests to complete
262  * @param oct Pointer to Octeon device
263  *
264  * Called during shutdown sequence
265  */
266 static int wait_for_pending_requests(struct octeon_device *oct)
267 {
268 	int i, pcount = 0;
269 
270 	for (i = 0; i < 100; i++) {
271 		pcount =
272 			atomic_read(&oct->response_list
273 				[OCTEON_ORDERED_SC_LIST].pending_req_count);
274 		if (pcount)
275 			schedule_timeout_uninterruptible(HZ / 10);
276 		else
277 			break;
278 	}
279 
280 	if (pcount)
281 		return 1;
282 
283 	return 0;
284 }
285 
286 /**
287  * \brief Cause device to go quiet so it can be safely removed/reset/etc
288  * @param oct Pointer to Octeon device
289  */
290 static inline void pcierror_quiesce_device(struct octeon_device *oct)
291 {
292 	int i;
293 
294 	/* Disable the input and output queues now. No more packets will
295 	 * arrive from Octeon, but we should wait for all packet processing
296 	 * to finish.
297 	 */
298 	force_io_queues_off(oct);
299 
300 	/* To allow for in-flight requests */
301 	schedule_timeout_uninterruptible(100);
302 
303 	if (wait_for_pending_requests(oct))
304 		dev_err(&oct->pci_dev->dev, "There were pending requests\n");
305 
306 	/* Force all requests waiting to be fetched by OCTEON to complete. */
307 	for (i = 0; i < MAX_OCTEON_INSTR_QUEUES(oct); i++) {
308 		struct octeon_instr_queue *iq;
309 
310 		if (!(oct->io_qmask.iq & BIT_ULL(i)))
311 			continue;
312 		iq = oct->instr_queue[i];
313 
314 		if (atomic_read(&iq->instr_pending)) {
315 			spin_lock_bh(&iq->lock);
316 			iq->fill_cnt = 0;
317 			iq->octeon_read_index = iq->host_write_index;
318 			iq->stats.instr_processed +=
319 				atomic_read(&iq->instr_pending);
320 			lio_process_iq_request_list(oct, iq, 0);
321 			spin_unlock_bh(&iq->lock);
322 		}
323 	}
324 
325 	/* Force all pending ordered list requests to time out. */
326 	lio_process_ordered_list(oct, 1);
327 
328 	/* We do not need to wait for output queue packets to be processed. */
329 }
330 
331 /**
332  * \brief Cleanup PCI AER uncorrectable error status
333  * @param dev Pointer to PCI device
334  */
335 static void cleanup_aer_uncorrect_error_status(struct pci_dev *dev)
336 {
337 	int pos = 0x100;
338 	u32 status, mask;
339 
340 	pr_info("%s :\n", __func__);
341 
342 	pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
343 	pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, &mask);
344 	if (dev->error_state == pci_channel_io_normal)
345 		status &= ~mask;        /* Clear corresponding nonfatal bits */
346 	else
347 		status &= mask;         /* Clear corresponding fatal bits */
348 	pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
349 }
350 
351 /**
352  * \brief Stop all PCI IO to a given device
353  * @param dev Pointer to Octeon device
354  */
355 static void stop_pci_io(struct octeon_device *oct)
356 {
357 	/* No more instructions will be forwarded. */
358 	atomic_set(&oct->status, OCT_DEV_IN_RESET);
359 
360 	pci_disable_device(oct->pci_dev);
361 
362 	/* Disable interrupts  */
363 	oct->fn_list.disable_interrupt(oct, OCTEON_ALL_INTR);
364 
365 	pcierror_quiesce_device(oct);
366 
367 	/* Release the interrupt line */
368 	free_irq(oct->pci_dev->irq, oct);
369 
370 	if (oct->flags & LIO_FLAG_MSI_ENABLED)
371 		pci_disable_msi(oct->pci_dev);
372 
373 	dev_dbg(&oct->pci_dev->dev, "Device state is now %s\n",
374 		lio_get_state_string(&oct->status));
375 
376 	/* making it a common function for all OCTEON models */
377 	cleanup_aer_uncorrect_error_status(oct->pci_dev);
378 }
379 
380 /**
381  * \brief called when PCI error is detected
382  * @param pdev Pointer to PCI device
383  * @param state The current pci connection state
384  *
385  * This function is called after a PCI bus error affecting
386  * this device has been detected.
387  */
388 static pci_ers_result_t liquidio_pcie_error_detected(struct pci_dev *pdev,
389 						     pci_channel_state_t state)
390 {
391 	struct octeon_device *oct = pci_get_drvdata(pdev);
392 
393 	/* Non-correctable Non-fatal errors */
394 	if (state == pci_channel_io_normal) {
395 		dev_err(&oct->pci_dev->dev, "Non-correctable non-fatal error reported:\n");
396 		cleanup_aer_uncorrect_error_status(oct->pci_dev);
397 		return PCI_ERS_RESULT_CAN_RECOVER;
398 	}
399 
400 	/* Non-correctable Fatal errors */
401 	dev_err(&oct->pci_dev->dev, "Non-correctable FATAL reported by PCI AER driver\n");
402 	stop_pci_io(oct);
403 
404 	/* Always return a DISCONNECT. There is no support for recovery but only
405 	 * for a clean shutdown.
406 	 */
407 	return PCI_ERS_RESULT_DISCONNECT;
408 }
409 
410 /**
411  * \brief mmio handler
412  * @param pdev Pointer to PCI device
413  */
414 static pci_ers_result_t liquidio_pcie_mmio_enabled(
415 				struct pci_dev *pdev __attribute__((unused)))
416 {
417 	/* We should never hit this since we never ask for a reset for a Fatal
418 	 * Error. We always return DISCONNECT in io_error above.
419 	 * But play safe and return RECOVERED for now.
420 	 */
421 	return PCI_ERS_RESULT_RECOVERED;
422 }
423 
424 /**
425  * \brief called after the pci bus has been reset.
426  * @param pdev Pointer to PCI device
427  *
428  * Restart the card from scratch, as if from a cold-boot. Implementation
429  * resembles the first-half of the octeon_resume routine.
430  */
431 static pci_ers_result_t liquidio_pcie_slot_reset(
432 				struct pci_dev *pdev __attribute__((unused)))
433 {
434 	/* We should never hit this since we never ask for a reset for a Fatal
435 	 * Error. We always return DISCONNECT in io_error above.
436 	 * But play safe and return RECOVERED for now.
437 	 */
438 	return PCI_ERS_RESULT_RECOVERED;
439 }
440 
441 /**
442  * \brief called when traffic can start flowing again.
443  * @param pdev Pointer to PCI device
444  *
445  * This callback is called when the error recovery driver tells us that
446  * its OK to resume normal operation. Implementation resembles the
447  * second-half of the octeon_resume routine.
448  */
449 static void liquidio_pcie_resume(struct pci_dev *pdev __attribute__((unused)))
450 {
451 	/* Nothing to be done here. */
452 }
453 
454 #ifdef CONFIG_PM
455 /**
456  * \brief called when suspending
457  * @param pdev Pointer to PCI device
458  * @param state state to suspend to
459  */
460 static int liquidio_suspend(struct pci_dev *pdev __attribute__((unused)),
461 			    pm_message_t state __attribute__((unused)))
462 {
463 	return 0;
464 }
465 
466 /**
467  * \brief called when resuming
468  * @param pdev Pointer to PCI device
469  */
470 static int liquidio_resume(struct pci_dev *pdev __attribute__((unused)))
471 {
472 	return 0;
473 }
474 #endif
475 
476 /* For PCI-E Advanced Error Recovery (AER) Interface */
477 static const struct pci_error_handlers liquidio_err_handler = {
478 	.error_detected = liquidio_pcie_error_detected,
479 	.mmio_enabled	= liquidio_pcie_mmio_enabled,
480 	.slot_reset	= liquidio_pcie_slot_reset,
481 	.resume		= liquidio_pcie_resume,
482 };
483 
484 static const struct pci_device_id liquidio_pci_tbl[] = {
485 	{       /* 68xx */
486 		PCI_VENDOR_ID_CAVIUM, 0x91, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0
487 	},
488 	{       /* 66xx */
489 		PCI_VENDOR_ID_CAVIUM, 0x92, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0
490 	},
491 	{       /* 23xx pf */
492 		PCI_VENDOR_ID_CAVIUM, 0x9702, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0
493 	},
494 	{
495 		0, 0, 0, 0, 0, 0, 0
496 	}
497 };
498 MODULE_DEVICE_TABLE(pci, liquidio_pci_tbl);
499 
500 static struct pci_driver liquidio_pci_driver = {
501 	.name		= "LiquidIO",
502 	.id_table	= liquidio_pci_tbl,
503 	.probe		= liquidio_probe,
504 	.remove		= liquidio_remove,
505 	.err_handler	= &liquidio_err_handler,    /* For AER */
506 
507 #ifdef CONFIG_PM
508 	.suspend	= liquidio_suspend,
509 	.resume		= liquidio_resume,
510 #endif
511 #ifdef CONFIG_PCI_IOV
512 	.sriov_configure = liquidio_enable_sriov,
513 #endif
514 };
515 
516 /**
517  * \brief register PCI driver
518  */
519 static int liquidio_init_pci(void)
520 {
521 	return pci_register_driver(&liquidio_pci_driver);
522 }
523 
524 /**
525  * \brief unregister PCI driver
526  */
527 static void liquidio_deinit_pci(void)
528 {
529 	pci_unregister_driver(&liquidio_pci_driver);
530 }
531 
532 /**
533  * \brief check interface state
534  * @param lio per-network private data
535  * @param state_flag flag state to check
536  */
537 static inline int ifstate_check(struct lio *lio, int state_flag)
538 {
539 	return atomic_read(&lio->ifstate) & state_flag;
540 }
541 
542 /**
543  * \brief set interface state
544  * @param lio per-network private data
545  * @param state_flag flag state to set
546  */
547 static inline void ifstate_set(struct lio *lio, int state_flag)
548 {
549 	atomic_set(&lio->ifstate, (atomic_read(&lio->ifstate) | state_flag));
550 }
551 
552 /**
553  * \brief clear interface state
554  * @param lio per-network private data
555  * @param state_flag flag state to clear
556  */
557 static inline void ifstate_reset(struct lio *lio, int state_flag)
558 {
559 	atomic_set(&lio->ifstate, (atomic_read(&lio->ifstate) & ~(state_flag)));
560 }
561 
562 /**
563  * \brief Stop Tx queues
564  * @param netdev network device
565  */
566 static inline void txqs_stop(struct net_device *netdev)
567 {
568 	if (netif_is_multiqueue(netdev)) {
569 		int i;
570 
571 		for (i = 0; i < netdev->num_tx_queues; i++)
572 			netif_stop_subqueue(netdev, i);
573 	} else {
574 		netif_stop_queue(netdev);
575 	}
576 }
577 
578 /**
579  * \brief Start Tx queues
580  * @param netdev network device
581  */
582 static inline void txqs_start(struct net_device *netdev)
583 {
584 	if (netif_is_multiqueue(netdev)) {
585 		int i;
586 
587 		for (i = 0; i < netdev->num_tx_queues; i++)
588 			netif_start_subqueue(netdev, i);
589 	} else {
590 		netif_start_queue(netdev);
591 	}
592 }
593 
594 /**
595  * \brief Wake Tx queues
596  * @param netdev network device
597  */
598 static inline void txqs_wake(struct net_device *netdev)
599 {
600 	struct lio *lio = GET_LIO(netdev);
601 
602 	if (netif_is_multiqueue(netdev)) {
603 		int i;
604 
605 		for (i = 0; i < netdev->num_tx_queues; i++) {
606 			int qno = lio->linfo.txpciq[i %
607 				(lio->linfo.num_txpciq)].s.q_no;
608 
609 			if (__netif_subqueue_stopped(netdev, i)) {
610 				INCR_INSTRQUEUE_PKT_COUNT(lio->oct_dev, qno,
611 							  tx_restart, 1);
612 				netif_wake_subqueue(netdev, i);
613 			}
614 		}
615 	} else {
616 		INCR_INSTRQUEUE_PKT_COUNT(lio->oct_dev, lio->txq,
617 					  tx_restart, 1);
618 		netif_wake_queue(netdev);
619 	}
620 }
621 
622 /**
623  * \brief Stop Tx queue
624  * @param netdev network device
625  */
626 static void stop_txq(struct net_device *netdev)
627 {
628 	txqs_stop(netdev);
629 }
630 
631 /**
632  * \brief Start Tx queue
633  * @param netdev network device
634  */
635 static void start_txq(struct net_device *netdev)
636 {
637 	struct lio *lio = GET_LIO(netdev);
638 
639 	if (lio->linfo.link.s.link_up) {
640 		txqs_start(netdev);
641 		return;
642 	}
643 }
644 
645 /**
646  * \brief Wake a queue
647  * @param netdev network device
648  * @param q which queue to wake
649  */
650 static inline void wake_q(struct net_device *netdev, int q)
651 {
652 	if (netif_is_multiqueue(netdev))
653 		netif_wake_subqueue(netdev, q);
654 	else
655 		netif_wake_queue(netdev);
656 }
657 
658 /**
659  * \brief Stop a queue
660  * @param netdev network device
661  * @param q which queue to stop
662  */
663 static inline void stop_q(struct net_device *netdev, int q)
664 {
665 	if (netif_is_multiqueue(netdev))
666 		netif_stop_subqueue(netdev, q);
667 	else
668 		netif_stop_queue(netdev);
669 }
670 
671 /**
672  * \brief Check Tx queue status, and take appropriate action
673  * @param lio per-network private data
674  * @returns 0 if full, number of queues woken up otherwise
675  */
676 static inline int check_txq_status(struct lio *lio)
677 {
678 	int ret_val = 0;
679 
680 	if (netif_is_multiqueue(lio->netdev)) {
681 		int numqs = lio->netdev->num_tx_queues;
682 		int q, iq = 0;
683 
684 		/* check each sub-queue state */
685 		for (q = 0; q < numqs; q++) {
686 			iq = lio->linfo.txpciq[q %
687 				(lio->linfo.num_txpciq)].s.q_no;
688 			if (octnet_iq_is_full(lio->oct_dev, iq))
689 				continue;
690 			if (__netif_subqueue_stopped(lio->netdev, q)) {
691 				wake_q(lio->netdev, q);
692 				INCR_INSTRQUEUE_PKT_COUNT(lio->oct_dev, iq,
693 							  tx_restart, 1);
694 				ret_val++;
695 			}
696 		}
697 	} else {
698 		if (octnet_iq_is_full(lio->oct_dev, lio->txq))
699 			return 0;
700 		wake_q(lio->netdev, lio->txq);
701 		INCR_INSTRQUEUE_PKT_COUNT(lio->oct_dev, lio->txq,
702 					  tx_restart, 1);
703 		ret_val = 1;
704 	}
705 	return ret_val;
706 }
707 
708 /**
709  * Remove the node at the head of the list. The list would be empty at
710  * the end of this call if there are no more nodes in the list.
711  */
712 static inline struct list_head *list_delete_head(struct list_head *root)
713 {
714 	struct list_head *node;
715 
716 	if ((root->prev == root) && (root->next == root))
717 		node = NULL;
718 	else
719 		node = root->next;
720 
721 	if (node)
722 		list_del(node);
723 
724 	return node;
725 }
726 
727 /**
728  * \brief Delete gather lists
729  * @param lio per-network private data
730  */
731 static void delete_glists(struct lio *lio)
732 {
733 	struct octnic_gather *g;
734 	int i;
735 
736 	if (!lio->glist)
737 		return;
738 
739 	for (i = 0; i < lio->linfo.num_txpciq; i++) {
740 		do {
741 			g = (struct octnic_gather *)
742 				list_delete_head(&lio->glist[i]);
743 			if (g) {
744 				if (g->sg) {
745 					dma_unmap_single(&lio->oct_dev->
746 							 pci_dev->dev,
747 							 g->sg_dma_ptr,
748 							 g->sg_size,
749 							 DMA_TO_DEVICE);
750 					kfree((void *)((unsigned long)g->sg -
751 						       g->adjust));
752 				}
753 				kfree(g);
754 			}
755 		} while (g);
756 	}
757 
758 	kfree((void *)lio->glist);
759 	kfree((void *)lio->glist_lock);
760 }
761 
762 /**
763  * \brief Setup gather lists
764  * @param lio per-network private data
765  */
766 static int setup_glists(struct octeon_device *oct, struct lio *lio, int num_iqs)
767 {
768 	int i, j;
769 	struct octnic_gather *g;
770 
771 	lio->glist_lock = kcalloc(num_iqs, sizeof(*lio->glist_lock),
772 				  GFP_KERNEL);
773 	if (!lio->glist_lock)
774 		return 1;
775 
776 	lio->glist = kcalloc(num_iqs, sizeof(*lio->glist),
777 			     GFP_KERNEL);
778 	if (!lio->glist) {
779 		kfree((void *)lio->glist_lock);
780 		return 1;
781 	}
782 
783 	for (i = 0; i < num_iqs; i++) {
784 		int numa_node = cpu_to_node(i % num_online_cpus());
785 
786 		spin_lock_init(&lio->glist_lock[i]);
787 
788 		INIT_LIST_HEAD(&lio->glist[i]);
789 
790 		for (j = 0; j < lio->tx_qsize; j++) {
791 			g = kzalloc_node(sizeof(*g), GFP_KERNEL,
792 					 numa_node);
793 			if (!g)
794 				g = kzalloc(sizeof(*g), GFP_KERNEL);
795 			if (!g)
796 				break;
797 
798 			g->sg_size = ((ROUNDUP4(OCTNIC_MAX_SG) >> 2) *
799 				      OCT_SG_ENTRY_SIZE);
800 
801 			g->sg = kmalloc_node(g->sg_size + 8,
802 					     GFP_KERNEL, numa_node);
803 			if (!g->sg)
804 				g->sg = kmalloc(g->sg_size + 8, GFP_KERNEL);
805 			if (!g->sg) {
806 				kfree(g);
807 				break;
808 			}
809 
810 			/* The gather component should be aligned on 64-bit
811 			 * boundary
812 			 */
813 			if (((unsigned long)g->sg) & 7) {
814 				g->adjust = 8 - (((unsigned long)g->sg) & 7);
815 				g->sg = (struct octeon_sg_entry *)
816 					((unsigned long)g->sg + g->adjust);
817 			}
818 			g->sg_dma_ptr = dma_map_single(&oct->pci_dev->dev,
819 						       g->sg, g->sg_size,
820 						       DMA_TO_DEVICE);
821 			if (dma_mapping_error(&oct->pci_dev->dev,
822 					      g->sg_dma_ptr)) {
823 				kfree((void *)((unsigned long)g->sg -
824 					       g->adjust));
825 				kfree(g);
826 				break;
827 			}
828 
829 			list_add_tail(&g->list, &lio->glist[i]);
830 		}
831 
832 		if (j != lio->tx_qsize) {
833 			delete_glists(lio);
834 			return 1;
835 		}
836 	}
837 
838 	return 0;
839 }
840 
841 /**
842  * \brief Print link information
843  * @param netdev network device
844  */
845 static void print_link_info(struct net_device *netdev)
846 {
847 	struct lio *lio = GET_LIO(netdev);
848 
849 	if (atomic_read(&lio->ifstate) & LIO_IFSTATE_REGISTERED) {
850 		struct oct_link_info *linfo = &lio->linfo;
851 
852 		if (linfo->link.s.link_up) {
853 			netif_info(lio, link, lio->netdev, "%d Mbps %s Duplex UP\n",
854 				   linfo->link.s.speed,
855 				   (linfo->link.s.duplex) ? "Full" : "Half");
856 		} else {
857 			netif_info(lio, link, lio->netdev, "Link Down\n");
858 		}
859 	}
860 }
861 
862 /**
863  * \brief Routine to notify MTU change
864  * @param work work_struct data structure
865  */
866 static void octnet_link_status_change(struct work_struct *work)
867 {
868 	struct cavium_wk *wk = (struct cavium_wk *)work;
869 	struct lio *lio = (struct lio *)wk->ctxptr;
870 
871 	rtnl_lock();
872 	call_netdevice_notifiers(NETDEV_CHANGEMTU, lio->netdev);
873 	rtnl_unlock();
874 }
875 
876 /**
877  * \brief Sets up the mtu status change work
878  * @param netdev network device
879  */
880 static inline int setup_link_status_change_wq(struct net_device *netdev)
881 {
882 	struct lio *lio = GET_LIO(netdev);
883 	struct octeon_device *oct = lio->oct_dev;
884 
885 	lio->link_status_wq.wq = alloc_workqueue("link-status",
886 						 WQ_MEM_RECLAIM, 0);
887 	if (!lio->link_status_wq.wq) {
888 		dev_err(&oct->pci_dev->dev, "unable to create cavium link status wq\n");
889 		return -1;
890 	}
891 	INIT_DELAYED_WORK(&lio->link_status_wq.wk.work,
892 			  octnet_link_status_change);
893 	lio->link_status_wq.wk.ctxptr = lio;
894 
895 	return 0;
896 }
897 
898 static inline void cleanup_link_status_change_wq(struct net_device *netdev)
899 {
900 	struct lio *lio = GET_LIO(netdev);
901 
902 	if (lio->link_status_wq.wq) {
903 		cancel_delayed_work_sync(&lio->link_status_wq.wk.work);
904 		destroy_workqueue(lio->link_status_wq.wq);
905 	}
906 }
907 
908 /**
909  * \brief Update link status
910  * @param netdev network device
911  * @param ls link status structure
912  *
913  * Called on receipt of a link status response from the core application to
914  * update each interface's link status.
915  */
916 static inline void update_link_status(struct net_device *netdev,
917 				      union oct_link_status *ls)
918 {
919 	struct lio *lio = GET_LIO(netdev);
920 	int changed = (lio->linfo.link.u64 != ls->u64);
921 
922 	lio->linfo.link.u64 = ls->u64;
923 
924 	if ((lio->intf_open) && (changed)) {
925 		print_link_info(netdev);
926 		lio->link_changes++;
927 
928 		if (lio->linfo.link.s.link_up) {
929 			netif_carrier_on(netdev);
930 			txqs_wake(netdev);
931 		} else {
932 			netif_carrier_off(netdev);
933 			stop_txq(netdev);
934 		}
935 	}
936 }
937 
938 /* Runs in interrupt context. */
939 static void update_txq_status(struct octeon_device *oct, int iq_num)
940 {
941 	struct net_device *netdev;
942 	struct lio *lio;
943 	struct octeon_instr_queue *iq = oct->instr_queue[iq_num];
944 
945 	netdev = oct->props[iq->ifidx].netdev;
946 
947 	/* This is needed because the first IQ does not have
948 	 * a netdev associated with it.
949 	 */
950 	if (!netdev)
951 		return;
952 
953 	lio = GET_LIO(netdev);
954 	if (netif_is_multiqueue(netdev)) {
955 		if (__netif_subqueue_stopped(netdev, iq->q_index) &&
956 		    lio->linfo.link.s.link_up &&
957 		    (!octnet_iq_is_full(oct, iq_num))) {
958 			INCR_INSTRQUEUE_PKT_COUNT(lio->oct_dev, iq_num,
959 						  tx_restart, 1);
960 			netif_wake_subqueue(netdev, iq->q_index);
961 		} else {
962 			if (!octnet_iq_is_full(oct, lio->txq)) {
963 				INCR_INSTRQUEUE_PKT_COUNT(lio->oct_dev,
964 							  lio->txq,
965 							  tx_restart, 1);
966 				wake_q(netdev, lio->txq);
967 			}
968 		}
969 	}
970 }
971 
972 static
973 int liquidio_schedule_msix_droq_pkt_handler(struct octeon_droq *droq, u64 ret)
974 {
975 	struct octeon_device *oct = droq->oct_dev;
976 	struct octeon_device_priv *oct_priv =
977 	    (struct octeon_device_priv *)oct->priv;
978 
979 	if (droq->ops.poll_mode) {
980 		droq->ops.napi_fn(droq);
981 	} else {
982 		if (ret & MSIX_PO_INT) {
983 			tasklet_schedule(&oct_priv->droq_tasklet);
984 			return 1;
985 		}
986 		/* this will be flushed periodically by check iq db */
987 		if (ret & MSIX_PI_INT)
988 			return 0;
989 	}
990 	return 0;
991 }
992 
993 /**
994  * \brief Droq packet processor sceduler
995  * @param oct octeon device
996  */
997 static void liquidio_schedule_droq_pkt_handlers(struct octeon_device *oct)
998 {
999 	struct octeon_device_priv *oct_priv =
1000 		(struct octeon_device_priv *)oct->priv;
1001 	u64 oq_no;
1002 	struct octeon_droq *droq;
1003 
1004 	if (oct->int_status & OCT_DEV_INTR_PKT_DATA) {
1005 		for (oq_no = 0; oq_no < MAX_OCTEON_OUTPUT_QUEUES(oct);
1006 		     oq_no++) {
1007 			if (!(oct->droq_intr & BIT_ULL(oq_no)))
1008 				continue;
1009 
1010 			droq = oct->droq[oq_no];
1011 
1012 			if (droq->ops.poll_mode) {
1013 				droq->ops.napi_fn(droq);
1014 				oct_priv->napi_mask |= (1 << oq_no);
1015 			} else {
1016 				tasklet_schedule(&oct_priv->droq_tasklet);
1017 			}
1018 		}
1019 	}
1020 }
1021 
1022 static irqreturn_t
1023 liquidio_msix_intr_handler(int irq __attribute__((unused)), void *dev)
1024 {
1025 	u64 ret;
1026 	struct octeon_ioq_vector *ioq_vector = (struct octeon_ioq_vector *)dev;
1027 	struct octeon_device *oct = ioq_vector->oct_dev;
1028 	struct octeon_droq *droq = oct->droq[ioq_vector->droq_index];
1029 
1030 	ret = oct->fn_list.msix_interrupt_handler(ioq_vector);
1031 
1032 	if ((ret & MSIX_PO_INT) || (ret & MSIX_PI_INT))
1033 		liquidio_schedule_msix_droq_pkt_handler(droq, ret);
1034 
1035 	return IRQ_HANDLED;
1036 }
1037 
1038 /**
1039  * \brief Interrupt handler for octeon
1040  * @param irq unused
1041  * @param dev octeon device
1042  */
1043 static
1044 irqreturn_t liquidio_legacy_intr_handler(int irq __attribute__((unused)),
1045 					 void *dev)
1046 {
1047 	struct octeon_device *oct = (struct octeon_device *)dev;
1048 	irqreturn_t ret;
1049 
1050 	/* Disable our interrupts for the duration of ISR */
1051 	oct->fn_list.disable_interrupt(oct, OCTEON_ALL_INTR);
1052 
1053 	ret = oct->fn_list.process_interrupt_regs(oct);
1054 
1055 	if (ret == IRQ_HANDLED)
1056 		liquidio_schedule_droq_pkt_handlers(oct);
1057 
1058 	/* Re-enable our interrupts  */
1059 	if (!(atomic_read(&oct->status) == OCT_DEV_IN_RESET))
1060 		oct->fn_list.enable_interrupt(oct, OCTEON_ALL_INTR);
1061 
1062 	return ret;
1063 }
1064 
1065 /**
1066  * \brief Setup interrupt for octeon device
1067  * @param oct octeon device
1068  *
1069  *  Enable interrupt in Octeon device as given in the PCI interrupt mask.
1070  */
1071 static int octeon_setup_interrupt(struct octeon_device *oct)
1072 {
1073 	int irqret, err;
1074 	struct msix_entry *msix_entries;
1075 	int i;
1076 	int num_ioq_vectors;
1077 	int num_alloc_ioq_vectors;
1078 
1079 	if (OCTEON_CN23XX_PF(oct) && oct->msix_on) {
1080 		oct->num_msix_irqs = oct->sriov_info.num_pf_rings;
1081 		/* one non ioq interrupt for handling sli_mac_pf_int_sum */
1082 		oct->num_msix_irqs += 1;
1083 
1084 		oct->msix_entries = kcalloc(
1085 		    oct->num_msix_irqs, sizeof(struct msix_entry), GFP_KERNEL);
1086 		if (!oct->msix_entries)
1087 			return 1;
1088 
1089 		msix_entries = (struct msix_entry *)oct->msix_entries;
1090 		/*Assumption is that pf msix vectors start from pf srn to pf to
1091 		 * trs and not from 0. if not change this code
1092 		 */
1093 		for (i = 0; i < oct->num_msix_irqs - 1; i++)
1094 			msix_entries[i].entry = oct->sriov_info.pf_srn + i;
1095 		msix_entries[oct->num_msix_irqs - 1].entry =
1096 		    oct->sriov_info.trs;
1097 		num_alloc_ioq_vectors = pci_enable_msix_range(
1098 						oct->pci_dev, msix_entries,
1099 						oct->num_msix_irqs,
1100 						oct->num_msix_irqs);
1101 		if (num_alloc_ioq_vectors < 0) {
1102 			dev_err(&oct->pci_dev->dev, "unable to Allocate MSI-X interrupts\n");
1103 			kfree(oct->msix_entries);
1104 			oct->msix_entries = NULL;
1105 			return 1;
1106 		}
1107 		dev_dbg(&oct->pci_dev->dev, "OCTEON: Enough MSI-X interrupts are allocated...\n");
1108 
1109 		num_ioq_vectors = oct->num_msix_irqs;
1110 
1111 		/** For PF, there is one non-ioq interrupt handler */
1112 		num_ioq_vectors -= 1;
1113 		irqret = request_irq(msix_entries[num_ioq_vectors].vector,
1114 				     liquidio_legacy_intr_handler, 0, "octeon",
1115 				     oct);
1116 		if (irqret) {
1117 			dev_err(&oct->pci_dev->dev,
1118 				"OCTEON: Request_irq failed for MSIX interrupt Error: %d\n",
1119 				irqret);
1120 			pci_disable_msix(oct->pci_dev);
1121 			kfree(oct->msix_entries);
1122 			oct->msix_entries = NULL;
1123 			return 1;
1124 		}
1125 
1126 		for (i = 0; i < num_ioq_vectors; i++) {
1127 			irqret = request_irq(msix_entries[i].vector,
1128 					     liquidio_msix_intr_handler, 0,
1129 					     "octeon", &oct->ioq_vector[i]);
1130 			if (irqret) {
1131 				dev_err(&oct->pci_dev->dev,
1132 					"OCTEON: Request_irq failed for MSIX interrupt Error: %d\n",
1133 					irqret);
1134 				/** Freeing the non-ioq irq vector here . */
1135 				free_irq(msix_entries[num_ioq_vectors].vector,
1136 					 oct);
1137 
1138 				while (i) {
1139 					i--;
1140 					/** clearing affinity mask. */
1141 					irq_set_affinity_hint(
1142 						msix_entries[i].vector, NULL);
1143 					free_irq(msix_entries[i].vector,
1144 						 &oct->ioq_vector[i]);
1145 				}
1146 				pci_disable_msix(oct->pci_dev);
1147 				kfree(oct->msix_entries);
1148 				oct->msix_entries = NULL;
1149 				return 1;
1150 			}
1151 			oct->ioq_vector[i].vector = msix_entries[i].vector;
1152 			/* assign the cpu mask for this msix interrupt vector */
1153 			irq_set_affinity_hint(
1154 					msix_entries[i].vector,
1155 					(&oct->ioq_vector[i].affinity_mask));
1156 		}
1157 		dev_dbg(&oct->pci_dev->dev, "OCTEON[%d]: MSI-X enabled\n",
1158 			oct->octeon_id);
1159 	} else {
1160 		err = pci_enable_msi(oct->pci_dev);
1161 		if (err)
1162 			dev_warn(&oct->pci_dev->dev, "Reverting to legacy interrupts. Error: %d\n",
1163 				 err);
1164 		else
1165 			oct->flags |= LIO_FLAG_MSI_ENABLED;
1166 
1167 		irqret = request_irq(oct->pci_dev->irq,
1168 				     liquidio_legacy_intr_handler, IRQF_SHARED,
1169 				     "octeon", oct);
1170 		if (irqret) {
1171 			if (oct->flags & LIO_FLAG_MSI_ENABLED)
1172 				pci_disable_msi(oct->pci_dev);
1173 			dev_err(&oct->pci_dev->dev, "Request IRQ failed with code: %d\n",
1174 				irqret);
1175 			return 1;
1176 		}
1177 	}
1178 	return 0;
1179 }
1180 
1181 static int liquidio_watchdog(void *param)
1182 {
1183 	u64 wdog;
1184 	u16 mask_of_stuck_cores = 0;
1185 	u16 mask_of_crashed_cores = 0;
1186 	int core_num;
1187 	u8 core_is_stuck[LIO_MAX_CORES];
1188 	u8 core_crashed[LIO_MAX_CORES];
1189 	struct octeon_device *oct = param;
1190 
1191 	memset(core_is_stuck, 0, sizeof(core_is_stuck));
1192 	memset(core_crashed, 0, sizeof(core_crashed));
1193 
1194 	while (!kthread_should_stop()) {
1195 		mask_of_crashed_cores =
1196 		    (u16)octeon_read_csr64(oct, CN23XX_SLI_SCRATCH2);
1197 
1198 		for (core_num = 0; core_num < LIO_MAX_CORES; core_num++) {
1199 			if (!core_is_stuck[core_num]) {
1200 				wdog = lio_pci_readq(oct, CIU3_WDOG(core_num));
1201 
1202 				/* look at watchdog state field */
1203 				wdog &= CIU3_WDOG_MASK;
1204 				if (wdog) {
1205 					/* this watchdog timer has expired */
1206 					core_is_stuck[core_num] =
1207 						LIO_MONITOR_WDOG_EXPIRE;
1208 					mask_of_stuck_cores |= (1 << core_num);
1209 				}
1210 			}
1211 
1212 			if (!core_crashed[core_num])
1213 				core_crashed[core_num] =
1214 				    (mask_of_crashed_cores >> core_num) & 1;
1215 		}
1216 
1217 		if (mask_of_stuck_cores) {
1218 			for (core_num = 0; core_num < LIO_MAX_CORES;
1219 			     core_num++) {
1220 				if (core_is_stuck[core_num] == 1) {
1221 					dev_err(&oct->pci_dev->dev,
1222 						"ERROR: Octeon core %d is stuck!\n",
1223 						core_num);
1224 					/* 2 means we have printk'd  an error
1225 					 * so no need to repeat the same printk
1226 					 */
1227 					core_is_stuck[core_num] =
1228 						LIO_MONITOR_CORE_STUCK_MSGD;
1229 				}
1230 			}
1231 		}
1232 
1233 		if (mask_of_crashed_cores) {
1234 			for (core_num = 0; core_num < LIO_MAX_CORES;
1235 			     core_num++) {
1236 				if (core_crashed[core_num] == 1) {
1237 					dev_err(&oct->pci_dev->dev,
1238 						"ERROR: Octeon core %d crashed!  See oct-fwdump for details.\n",
1239 						core_num);
1240 					/* 2 means we have printk'd  an error
1241 					 * so no need to repeat the same printk
1242 					 */
1243 					core_crashed[core_num] =
1244 						LIO_MONITOR_CORE_STUCK_MSGD;
1245 				}
1246 			}
1247 		}
1248 #ifdef CONFIG_MODULE_UNLOAD
1249 		if (mask_of_stuck_cores || mask_of_crashed_cores) {
1250 			/* make module refcount=0 so that rmmod will work */
1251 			long refcount;
1252 
1253 			refcount = module_refcount(THIS_MODULE);
1254 
1255 			while (refcount > 0) {
1256 				module_put(THIS_MODULE);
1257 				refcount = module_refcount(THIS_MODULE);
1258 			}
1259 
1260 			/* compensate for and withstand an unlikely (but still
1261 			 * possible) race condition
1262 			 */
1263 			while (refcount < 0) {
1264 				try_module_get(THIS_MODULE);
1265 				refcount = module_refcount(THIS_MODULE);
1266 			}
1267 		}
1268 #endif
1269 		/* sleep for two seconds */
1270 		set_current_state(TASK_INTERRUPTIBLE);
1271 		schedule_timeout(2 * HZ);
1272 	}
1273 
1274 	return 0;
1275 }
1276 
1277 /**
1278  * \brief PCI probe handler
1279  * @param pdev PCI device structure
1280  * @param ent unused
1281  */
1282 static int
1283 liquidio_probe(struct pci_dev *pdev,
1284 	       const struct pci_device_id *ent __attribute__((unused)))
1285 {
1286 	struct octeon_device *oct_dev = NULL;
1287 	struct handshake *hs;
1288 
1289 	oct_dev = octeon_allocate_device(pdev->device,
1290 					 sizeof(struct octeon_device_priv));
1291 	if (!oct_dev) {
1292 		dev_err(&pdev->dev, "Unable to allocate device\n");
1293 		return -ENOMEM;
1294 	}
1295 
1296 	if (pdev->device == OCTEON_CN23XX_PF_VID)
1297 		oct_dev->msix_on = LIO_FLAG_MSIX_ENABLED;
1298 
1299 	dev_info(&pdev->dev, "Initializing device %x:%x.\n",
1300 		 (u32)pdev->vendor, (u32)pdev->device);
1301 
1302 	/* Assign octeon_device for this device to the private data area. */
1303 	pci_set_drvdata(pdev, oct_dev);
1304 
1305 	/* set linux specific device pointer */
1306 	oct_dev->pci_dev = (void *)pdev;
1307 
1308 	hs = &handshake[oct_dev->octeon_id];
1309 	init_completion(&hs->init);
1310 	init_completion(&hs->started);
1311 	hs->pci_dev = pdev;
1312 
1313 	if (oct_dev->octeon_id == 0)
1314 		/* first LiquidIO NIC is detected */
1315 		complete(&first_stage);
1316 
1317 	if (octeon_device_init(oct_dev)) {
1318 		complete(&hs->init);
1319 		liquidio_remove(pdev);
1320 		return -ENOMEM;
1321 	}
1322 
1323 	if (OCTEON_CN23XX_PF(oct_dev)) {
1324 		u64 scratch1;
1325 		u8 bus, device, function;
1326 
1327 		scratch1 = octeon_read_csr64(oct_dev, CN23XX_SLI_SCRATCH1);
1328 		if (!(scratch1 & 4ULL)) {
1329 			/* Bit 2 of SLI_SCRATCH_1 is a flag that indicates that
1330 			 * the lio watchdog kernel thread is running for this
1331 			 * NIC.  Each NIC gets one watchdog kernel thread.
1332 			 */
1333 			scratch1 |= 4ULL;
1334 			octeon_write_csr64(oct_dev, CN23XX_SLI_SCRATCH1,
1335 					   scratch1);
1336 
1337 			bus = pdev->bus->number;
1338 			device = PCI_SLOT(pdev->devfn);
1339 			function = PCI_FUNC(pdev->devfn);
1340 			oct_dev->watchdog_task = kthread_create(
1341 			    liquidio_watchdog, oct_dev,
1342 			    "liowd/%02hhx:%02hhx.%hhx", bus, device, function);
1343 			if (!IS_ERR(oct_dev->watchdog_task)) {
1344 				wake_up_process(oct_dev->watchdog_task);
1345 			} else {
1346 				oct_dev->watchdog_task = NULL;
1347 				dev_err(&oct_dev->pci_dev->dev,
1348 					"failed to create kernel_thread\n");
1349 				liquidio_remove(pdev);
1350 				return -1;
1351 			}
1352 		}
1353 	}
1354 
1355 	oct_dev->rx_pause = 1;
1356 	oct_dev->tx_pause = 1;
1357 
1358 	dev_dbg(&oct_dev->pci_dev->dev, "Device is ready\n");
1359 
1360 	return 0;
1361 }
1362 
1363 /**
1364  *\brief Destroy resources associated with octeon device
1365  * @param pdev PCI device structure
1366  * @param ent unused
1367  */
1368 static void octeon_destroy_resources(struct octeon_device *oct)
1369 {
1370 	int i;
1371 	struct msix_entry *msix_entries;
1372 	struct octeon_device_priv *oct_priv =
1373 		(struct octeon_device_priv *)oct->priv;
1374 
1375 	struct handshake *hs;
1376 
1377 	switch (atomic_read(&oct->status)) {
1378 	case OCT_DEV_RUNNING:
1379 	case OCT_DEV_CORE_OK:
1380 
1381 		/* No more instructions will be forwarded. */
1382 		atomic_set(&oct->status, OCT_DEV_IN_RESET);
1383 
1384 		oct->app_mode = CVM_DRV_INVALID_APP;
1385 		dev_dbg(&oct->pci_dev->dev, "Device state is now %s\n",
1386 			lio_get_state_string(&oct->status));
1387 
1388 		schedule_timeout_uninterruptible(HZ / 10);
1389 
1390 		/* fallthrough */
1391 	case OCT_DEV_HOST_OK:
1392 
1393 		/* fallthrough */
1394 	case OCT_DEV_CONSOLE_INIT_DONE:
1395 		/* Remove any consoles */
1396 		octeon_remove_consoles(oct);
1397 
1398 		/* fallthrough */
1399 	case OCT_DEV_IO_QUEUES_DONE:
1400 		if (wait_for_pending_requests(oct))
1401 			dev_err(&oct->pci_dev->dev, "There were pending requests\n");
1402 
1403 		if (lio_wait_for_instr_fetch(oct))
1404 			dev_err(&oct->pci_dev->dev, "IQ had pending instructions\n");
1405 
1406 		/* Disable the input and output queues now. No more packets will
1407 		 * arrive from Octeon, but we should wait for all packet
1408 		 * processing to finish.
1409 		 */
1410 		oct->fn_list.disable_io_queues(oct);
1411 
1412 		if (lio_wait_for_oq_pkts(oct))
1413 			dev_err(&oct->pci_dev->dev, "OQ had pending packets\n");
1414 
1415 	/* fallthrough */
1416 	case OCT_DEV_INTR_SET_DONE:
1417 		/* Disable interrupts  */
1418 		oct->fn_list.disable_interrupt(oct, OCTEON_ALL_INTR);
1419 
1420 		if (oct->msix_on) {
1421 			msix_entries = (struct msix_entry *)oct->msix_entries;
1422 			for (i = 0; i < oct->num_msix_irqs - 1; i++) {
1423 				/* clear the affinity_cpumask */
1424 				irq_set_affinity_hint(msix_entries[i].vector,
1425 						      NULL);
1426 				free_irq(msix_entries[i].vector,
1427 					 &oct->ioq_vector[i]);
1428 			}
1429 			/* non-iov vector's argument is oct struct */
1430 			free_irq(msix_entries[i].vector, oct);
1431 
1432 			pci_disable_msix(oct->pci_dev);
1433 			kfree(oct->msix_entries);
1434 			oct->msix_entries = NULL;
1435 		} else {
1436 			/* Release the interrupt line */
1437 			free_irq(oct->pci_dev->irq, oct);
1438 
1439 			if (oct->flags & LIO_FLAG_MSI_ENABLED)
1440 				pci_disable_msi(oct->pci_dev);
1441 		}
1442 
1443 	/* fallthrough */
1444 	case OCT_DEV_MSIX_ALLOC_VECTOR_DONE:
1445 		if (OCTEON_CN23XX_PF(oct))
1446 			octeon_free_ioq_vector(oct);
1447 
1448 	/* fallthrough */
1449 	case OCT_DEV_MBOX_SETUP_DONE:
1450 		if (OCTEON_CN23XX_PF(oct))
1451 			oct->fn_list.free_mbox(oct);
1452 
1453 	/* fallthrough */
1454 	case OCT_DEV_IN_RESET:
1455 	case OCT_DEV_DROQ_INIT_DONE:
1456 		/* Wait for any pending operations */
1457 		mdelay(100);
1458 		for (i = 0; i < MAX_OCTEON_OUTPUT_QUEUES(oct); i++) {
1459 			if (!(oct->io_qmask.oq & BIT_ULL(i)))
1460 				continue;
1461 			octeon_delete_droq(oct, i);
1462 		}
1463 
1464 		/* Force any pending handshakes to complete */
1465 		for (i = 0; i < MAX_OCTEON_DEVICES; i++) {
1466 			hs = &handshake[i];
1467 
1468 			if (hs->pci_dev) {
1469 				handshake[oct->octeon_id].init_ok = 0;
1470 				complete(&handshake[oct->octeon_id].init);
1471 				handshake[oct->octeon_id].started_ok = 0;
1472 				complete(&handshake[oct->octeon_id].started);
1473 			}
1474 		}
1475 
1476 		/* fallthrough */
1477 	case OCT_DEV_RESP_LIST_INIT_DONE:
1478 		octeon_delete_response_list(oct);
1479 
1480 		/* fallthrough */
1481 	case OCT_DEV_INSTR_QUEUE_INIT_DONE:
1482 		for (i = 0; i < MAX_OCTEON_INSTR_QUEUES(oct); i++) {
1483 			if (!(oct->io_qmask.iq & BIT_ULL(i)))
1484 				continue;
1485 			octeon_delete_instr_queue(oct, i);
1486 		}
1487 #ifdef CONFIG_PCI_IOV
1488 		if (oct->sriov_info.sriov_enabled)
1489 			pci_disable_sriov(oct->pci_dev);
1490 #endif
1491 		/* fallthrough */
1492 	case OCT_DEV_SC_BUFF_POOL_INIT_DONE:
1493 		octeon_free_sc_buffer_pool(oct);
1494 
1495 		/* fallthrough */
1496 	case OCT_DEV_DISPATCH_INIT_DONE:
1497 		octeon_delete_dispatch_list(oct);
1498 		cancel_delayed_work_sync(&oct->nic_poll_work.work);
1499 
1500 		/* fallthrough */
1501 	case OCT_DEV_PCI_MAP_DONE:
1502 		/* Soft reset the octeon device before exiting */
1503 		if ((!OCTEON_CN23XX_PF(oct)) || !oct->octeon_id)
1504 			oct->fn_list.soft_reset(oct);
1505 
1506 		octeon_unmap_pci_barx(oct, 0);
1507 		octeon_unmap_pci_barx(oct, 1);
1508 
1509 		/* fallthrough */
1510 	case OCT_DEV_PCI_ENABLE_DONE:
1511 		pci_clear_master(oct->pci_dev);
1512 		/* Disable the device, releasing the PCI INT */
1513 		pci_disable_device(oct->pci_dev);
1514 
1515 		/* fallthrough */
1516 	case OCT_DEV_BEGIN_STATE:
1517 		/* Nothing to be done here either */
1518 		break;
1519 	}                       /* end switch (oct->status) */
1520 
1521 	tasklet_kill(&oct_priv->droq_tasklet);
1522 }
1523 
1524 /**
1525  * \brief Callback for rx ctrl
1526  * @param status status of request
1527  * @param buf pointer to resp structure
1528  */
1529 static void rx_ctl_callback(struct octeon_device *oct,
1530 			    u32 status,
1531 			    void *buf)
1532 {
1533 	struct octeon_soft_command *sc = (struct octeon_soft_command *)buf;
1534 	struct liquidio_rx_ctl_context *ctx;
1535 
1536 	ctx  = (struct liquidio_rx_ctl_context *)sc->ctxptr;
1537 
1538 	oct = lio_get_device(ctx->octeon_id);
1539 	if (status)
1540 		dev_err(&oct->pci_dev->dev, "rx ctl instruction failed. Status: %llx\n",
1541 			CVM_CAST64(status));
1542 	WRITE_ONCE(ctx->cond, 1);
1543 
1544 	/* This barrier is required to be sure that the response has been
1545 	 * written fully before waking up the handler
1546 	 */
1547 	wmb();
1548 
1549 	wake_up_interruptible(&ctx->wc);
1550 }
1551 
1552 /**
1553  * \brief Send Rx control command
1554  * @param lio per-network private data
1555  * @param start_stop whether to start or stop
1556  */
1557 static void send_rx_ctrl_cmd(struct lio *lio, int start_stop)
1558 {
1559 	struct octeon_soft_command *sc;
1560 	struct liquidio_rx_ctl_context *ctx;
1561 	union octnet_cmd *ncmd;
1562 	int ctx_size = sizeof(struct liquidio_rx_ctl_context);
1563 	struct octeon_device *oct = (struct octeon_device *)lio->oct_dev;
1564 	int retval;
1565 
1566 	if (oct->props[lio->ifidx].rx_on == start_stop)
1567 		return;
1568 
1569 	sc = (struct octeon_soft_command *)
1570 		octeon_alloc_soft_command(oct, OCTNET_CMD_SIZE,
1571 					  16, ctx_size);
1572 
1573 	ncmd = (union octnet_cmd *)sc->virtdptr;
1574 	ctx  = (struct liquidio_rx_ctl_context *)sc->ctxptr;
1575 
1576 	WRITE_ONCE(ctx->cond, 0);
1577 	ctx->octeon_id = lio_get_device_id(oct);
1578 	init_waitqueue_head(&ctx->wc);
1579 
1580 	ncmd->u64 = 0;
1581 	ncmd->s.cmd = OCTNET_CMD_RX_CTL;
1582 	ncmd->s.param1 = start_stop;
1583 
1584 	octeon_swap_8B_data((u64 *)ncmd, (OCTNET_CMD_SIZE >> 3));
1585 
1586 	sc->iq_no = lio->linfo.txpciq[0].s.q_no;
1587 
1588 	octeon_prepare_soft_command(oct, sc, OPCODE_NIC,
1589 				    OPCODE_NIC_CMD, 0, 0, 0);
1590 
1591 	sc->callback = rx_ctl_callback;
1592 	sc->callback_arg = sc;
1593 	sc->wait_time = 5000;
1594 
1595 	retval = octeon_send_soft_command(oct, sc);
1596 	if (retval == IQ_SEND_FAILED) {
1597 		netif_info(lio, rx_err, lio->netdev, "Failed to send RX Control message\n");
1598 	} else {
1599 		/* Sleep on a wait queue till the cond flag indicates that the
1600 		 * response arrived or timed-out.
1601 		 */
1602 		if (sleep_cond(&ctx->wc, &ctx->cond) == -EINTR)
1603 			return;
1604 		oct->props[lio->ifidx].rx_on = start_stop;
1605 	}
1606 
1607 	octeon_free_soft_command(oct, sc);
1608 }
1609 
1610 /**
1611  * \brief Destroy NIC device interface
1612  * @param oct octeon device
1613  * @param ifidx which interface to destroy
1614  *
1615  * Cleanup associated with each interface for an Octeon device  when NIC
1616  * module is being unloaded or if initialization fails during load.
1617  */
1618 static void liquidio_destroy_nic_device(struct octeon_device *oct, int ifidx)
1619 {
1620 	struct net_device *netdev = oct->props[ifidx].netdev;
1621 	struct lio *lio;
1622 	struct napi_struct *napi, *n;
1623 
1624 	if (!netdev) {
1625 		dev_err(&oct->pci_dev->dev, "%s No netdevice ptr for index %d\n",
1626 			__func__, ifidx);
1627 		return;
1628 	}
1629 
1630 	lio = GET_LIO(netdev);
1631 
1632 	dev_dbg(&oct->pci_dev->dev, "NIC device cleanup\n");
1633 
1634 	if (atomic_read(&lio->ifstate) & LIO_IFSTATE_RUNNING)
1635 		liquidio_stop(netdev);
1636 
1637 	if (oct->props[lio->ifidx].napi_enabled == 1) {
1638 		list_for_each_entry_safe(napi, n, &netdev->napi_list, dev_list)
1639 			napi_disable(napi);
1640 
1641 		oct->props[lio->ifidx].napi_enabled = 0;
1642 
1643 		if (OCTEON_CN23XX_PF(oct))
1644 			oct->droq[0]->ops.poll_mode = 0;
1645 	}
1646 
1647 	if (atomic_read(&lio->ifstate) & LIO_IFSTATE_REGISTERED)
1648 		unregister_netdev(netdev);
1649 
1650 	cleanup_link_status_change_wq(netdev);
1651 
1652 	delete_glists(lio);
1653 
1654 	free_netdev(netdev);
1655 
1656 	oct->props[ifidx].gmxport = -1;
1657 
1658 	oct->props[ifidx].netdev = NULL;
1659 }
1660 
1661 /**
1662  * \brief Stop complete NIC functionality
1663  * @param oct octeon device
1664  */
1665 static int liquidio_stop_nic_module(struct octeon_device *oct)
1666 {
1667 	int i, j;
1668 	struct lio *lio;
1669 
1670 	dev_dbg(&oct->pci_dev->dev, "Stopping network interfaces\n");
1671 	if (!oct->ifcount) {
1672 		dev_err(&oct->pci_dev->dev, "Init for Octeon was not completed\n");
1673 		return 1;
1674 	}
1675 
1676 	spin_lock_bh(&oct->cmd_resp_wqlock);
1677 	oct->cmd_resp_state = OCT_DRV_OFFLINE;
1678 	spin_unlock_bh(&oct->cmd_resp_wqlock);
1679 
1680 	for (i = 0; i < oct->ifcount; i++) {
1681 		lio = GET_LIO(oct->props[i].netdev);
1682 		for (j = 0; j < lio->linfo.num_rxpciq; j++)
1683 			octeon_unregister_droq_ops(oct,
1684 						   lio->linfo.rxpciq[j].s.q_no);
1685 	}
1686 
1687 	for (i = 0; i < oct->ifcount; i++)
1688 		liquidio_destroy_nic_device(oct, i);
1689 
1690 	dev_dbg(&oct->pci_dev->dev, "Network interfaces stopped\n");
1691 	return 0;
1692 }
1693 
1694 /**
1695  * \brief Cleans up resources at unload time
1696  * @param pdev PCI device structure
1697  */
1698 static void liquidio_remove(struct pci_dev *pdev)
1699 {
1700 	struct octeon_device *oct_dev = pci_get_drvdata(pdev);
1701 
1702 	dev_dbg(&oct_dev->pci_dev->dev, "Stopping device\n");
1703 
1704 	if (oct_dev->watchdog_task)
1705 		kthread_stop(oct_dev->watchdog_task);
1706 
1707 	if (oct_dev->app_mode && (oct_dev->app_mode == CVM_DRV_NIC_APP))
1708 		liquidio_stop_nic_module(oct_dev);
1709 
1710 	/* Reset the octeon device and cleanup all memory allocated for
1711 	 * the octeon device by driver.
1712 	 */
1713 	octeon_destroy_resources(oct_dev);
1714 
1715 	dev_info(&oct_dev->pci_dev->dev, "Device removed\n");
1716 
1717 	/* This octeon device has been removed. Update the global
1718 	 * data structure to reflect this. Free the device structure.
1719 	 */
1720 	octeon_free_device_mem(oct_dev);
1721 }
1722 
1723 /**
1724  * \brief Identify the Octeon device and to map the BAR address space
1725  * @param oct octeon device
1726  */
1727 static int octeon_chip_specific_setup(struct octeon_device *oct)
1728 {
1729 	u32 dev_id, rev_id;
1730 	int ret = 1;
1731 	char *s;
1732 
1733 	pci_read_config_dword(oct->pci_dev, 0, &dev_id);
1734 	pci_read_config_dword(oct->pci_dev, 8, &rev_id);
1735 	oct->rev_id = rev_id & 0xff;
1736 
1737 	switch (dev_id) {
1738 	case OCTEON_CN68XX_PCIID:
1739 		oct->chip_id = OCTEON_CN68XX;
1740 		ret = lio_setup_cn68xx_octeon_device(oct);
1741 		s = "CN68XX";
1742 		break;
1743 
1744 	case OCTEON_CN66XX_PCIID:
1745 		oct->chip_id = OCTEON_CN66XX;
1746 		ret = lio_setup_cn66xx_octeon_device(oct);
1747 		s = "CN66XX";
1748 		break;
1749 
1750 	case OCTEON_CN23XX_PCIID_PF:
1751 		oct->chip_id = OCTEON_CN23XX_PF_VID;
1752 		ret = setup_cn23xx_octeon_pf_device(oct);
1753 		s = "CN23XX";
1754 		break;
1755 
1756 	default:
1757 		s = "?";
1758 		dev_err(&oct->pci_dev->dev, "Unknown device found (dev_id: %x)\n",
1759 			dev_id);
1760 	}
1761 
1762 	if (!ret)
1763 		dev_info(&oct->pci_dev->dev, "%s PASS%d.%d %s Version: %s\n", s,
1764 			 OCTEON_MAJOR_REV(oct),
1765 			 OCTEON_MINOR_REV(oct),
1766 			 octeon_get_conf(oct)->card_name,
1767 			 LIQUIDIO_VERSION);
1768 
1769 	return ret;
1770 }
1771 
1772 /**
1773  * \brief PCI initialization for each Octeon device.
1774  * @param oct octeon device
1775  */
1776 static int octeon_pci_os_setup(struct octeon_device *oct)
1777 {
1778 	/* setup PCI stuff first */
1779 	if (pci_enable_device(oct->pci_dev)) {
1780 		dev_err(&oct->pci_dev->dev, "pci_enable_device failed\n");
1781 		return 1;
1782 	}
1783 
1784 	if (dma_set_mask_and_coherent(&oct->pci_dev->dev, DMA_BIT_MASK(64))) {
1785 		dev_err(&oct->pci_dev->dev, "Unexpected DMA device capability\n");
1786 		pci_disable_device(oct->pci_dev);
1787 		return 1;
1788 	}
1789 
1790 	/* Enable PCI DMA Master. */
1791 	pci_set_master(oct->pci_dev);
1792 
1793 	return 0;
1794 }
1795 
1796 static inline int skb_iq(struct lio *lio, struct sk_buff *skb)
1797 {
1798 	int q = 0;
1799 
1800 	if (netif_is_multiqueue(lio->netdev))
1801 		q = skb->queue_mapping % lio->linfo.num_txpciq;
1802 
1803 	return q;
1804 }
1805 
1806 /**
1807  * \brief Check Tx queue state for a given network buffer
1808  * @param lio per-network private data
1809  * @param skb network buffer
1810  */
1811 static inline int check_txq_state(struct lio *lio, struct sk_buff *skb)
1812 {
1813 	int q = 0, iq = 0;
1814 
1815 	if (netif_is_multiqueue(lio->netdev)) {
1816 		q = skb->queue_mapping;
1817 		iq = lio->linfo.txpciq[(q % (lio->linfo.num_txpciq))].s.q_no;
1818 	} else {
1819 		iq = lio->txq;
1820 		q = iq;
1821 	}
1822 
1823 	if (octnet_iq_is_full(lio->oct_dev, iq))
1824 		return 0;
1825 
1826 	if (__netif_subqueue_stopped(lio->netdev, q)) {
1827 		INCR_INSTRQUEUE_PKT_COUNT(lio->oct_dev, iq, tx_restart, 1);
1828 		wake_q(lio->netdev, q);
1829 	}
1830 	return 1;
1831 }
1832 
1833 /**
1834  * \brief Unmap and free network buffer
1835  * @param buf buffer
1836  */
1837 static void free_netbuf(void *buf)
1838 {
1839 	struct sk_buff *skb;
1840 	struct octnet_buf_free_info *finfo;
1841 	struct lio *lio;
1842 
1843 	finfo = (struct octnet_buf_free_info *)buf;
1844 	skb = finfo->skb;
1845 	lio = finfo->lio;
1846 
1847 	dma_unmap_single(&lio->oct_dev->pci_dev->dev, finfo->dptr, skb->len,
1848 			 DMA_TO_DEVICE);
1849 
1850 	check_txq_state(lio, skb);
1851 
1852 	tx_buffer_free(skb);
1853 }
1854 
1855 /**
1856  * \brief Unmap and free gather buffer
1857  * @param buf buffer
1858  */
1859 static void free_netsgbuf(void *buf)
1860 {
1861 	struct octnet_buf_free_info *finfo;
1862 	struct sk_buff *skb;
1863 	struct lio *lio;
1864 	struct octnic_gather *g;
1865 	int i, frags, iq;
1866 
1867 	finfo = (struct octnet_buf_free_info *)buf;
1868 	skb = finfo->skb;
1869 	lio = finfo->lio;
1870 	g = finfo->g;
1871 	frags = skb_shinfo(skb)->nr_frags;
1872 
1873 	dma_unmap_single(&lio->oct_dev->pci_dev->dev,
1874 			 g->sg[0].ptr[0], (skb->len - skb->data_len),
1875 			 DMA_TO_DEVICE);
1876 
1877 	i = 1;
1878 	while (frags--) {
1879 		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
1880 
1881 		pci_unmap_page((lio->oct_dev)->pci_dev,
1882 			       g->sg[(i >> 2)].ptr[(i & 3)],
1883 			       frag->size, DMA_TO_DEVICE);
1884 		i++;
1885 	}
1886 
1887 	dma_sync_single_for_cpu(&lio->oct_dev->pci_dev->dev,
1888 				g->sg_dma_ptr, g->sg_size, DMA_TO_DEVICE);
1889 
1890 	iq = skb_iq(lio, skb);
1891 	spin_lock(&lio->glist_lock[iq]);
1892 	list_add_tail(&g->list, &lio->glist[iq]);
1893 	spin_unlock(&lio->glist_lock[iq]);
1894 
1895 	check_txq_state(lio, skb);     /* mq support: sub-queue state check */
1896 
1897 	tx_buffer_free(skb);
1898 }
1899 
1900 /**
1901  * \brief Unmap and free gather buffer with response
1902  * @param buf buffer
1903  */
1904 static void free_netsgbuf_with_resp(void *buf)
1905 {
1906 	struct octeon_soft_command *sc;
1907 	struct octnet_buf_free_info *finfo;
1908 	struct sk_buff *skb;
1909 	struct lio *lio;
1910 	struct octnic_gather *g;
1911 	int i, frags, iq;
1912 
1913 	sc = (struct octeon_soft_command *)buf;
1914 	skb = (struct sk_buff *)sc->callback_arg;
1915 	finfo = (struct octnet_buf_free_info *)&skb->cb;
1916 
1917 	lio = finfo->lio;
1918 	g = finfo->g;
1919 	frags = skb_shinfo(skb)->nr_frags;
1920 
1921 	dma_unmap_single(&lio->oct_dev->pci_dev->dev,
1922 			 g->sg[0].ptr[0], (skb->len - skb->data_len),
1923 			 DMA_TO_DEVICE);
1924 
1925 	i = 1;
1926 	while (frags--) {
1927 		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
1928 
1929 		pci_unmap_page((lio->oct_dev)->pci_dev,
1930 			       g->sg[(i >> 2)].ptr[(i & 3)],
1931 			       frag->size, DMA_TO_DEVICE);
1932 		i++;
1933 	}
1934 
1935 	dma_sync_single_for_cpu(&lio->oct_dev->pci_dev->dev,
1936 				g->sg_dma_ptr, g->sg_size, DMA_TO_DEVICE);
1937 
1938 	iq = skb_iq(lio, skb);
1939 
1940 	spin_lock(&lio->glist_lock[iq]);
1941 	list_add_tail(&g->list, &lio->glist[iq]);
1942 	spin_unlock(&lio->glist_lock[iq]);
1943 
1944 	/* Don't free the skb yet */
1945 
1946 	check_txq_state(lio, skb);
1947 }
1948 
1949 /**
1950  * \brief Adjust ptp frequency
1951  * @param ptp PTP clock info
1952  * @param ppb how much to adjust by, in parts-per-billion
1953  */
1954 static int liquidio_ptp_adjfreq(struct ptp_clock_info *ptp, s32 ppb)
1955 {
1956 	struct lio *lio = container_of(ptp, struct lio, ptp_info);
1957 	struct octeon_device *oct = (struct octeon_device *)lio->oct_dev;
1958 	u64 comp, delta;
1959 	unsigned long flags;
1960 	bool neg_adj = false;
1961 
1962 	if (ppb < 0) {
1963 		neg_adj = true;
1964 		ppb = -ppb;
1965 	}
1966 
1967 	/* The hardware adds the clock compensation value to the
1968 	 * PTP clock on every coprocessor clock cycle, so we
1969 	 * compute the delta in terms of coprocessor clocks.
1970 	 */
1971 	delta = (u64)ppb << 32;
1972 	do_div(delta, oct->coproc_clock_rate);
1973 
1974 	spin_lock_irqsave(&lio->ptp_lock, flags);
1975 	comp = lio_pci_readq(oct, CN6XXX_MIO_PTP_CLOCK_COMP);
1976 	if (neg_adj)
1977 		comp -= delta;
1978 	else
1979 		comp += delta;
1980 	lio_pci_writeq(oct, comp, CN6XXX_MIO_PTP_CLOCK_COMP);
1981 	spin_unlock_irqrestore(&lio->ptp_lock, flags);
1982 
1983 	return 0;
1984 }
1985 
1986 /**
1987  * \brief Adjust ptp time
1988  * @param ptp PTP clock info
1989  * @param delta how much to adjust by, in nanosecs
1990  */
1991 static int liquidio_ptp_adjtime(struct ptp_clock_info *ptp, s64 delta)
1992 {
1993 	unsigned long flags;
1994 	struct lio *lio = container_of(ptp, struct lio, ptp_info);
1995 
1996 	spin_lock_irqsave(&lio->ptp_lock, flags);
1997 	lio->ptp_adjust += delta;
1998 	spin_unlock_irqrestore(&lio->ptp_lock, flags);
1999 
2000 	return 0;
2001 }
2002 
2003 /**
2004  * \brief Get hardware clock time, including any adjustment
2005  * @param ptp PTP clock info
2006  * @param ts timespec
2007  */
2008 static int liquidio_ptp_gettime(struct ptp_clock_info *ptp,
2009 				struct timespec64 *ts)
2010 {
2011 	u64 ns;
2012 	unsigned long flags;
2013 	struct lio *lio = container_of(ptp, struct lio, ptp_info);
2014 	struct octeon_device *oct = (struct octeon_device *)lio->oct_dev;
2015 
2016 	spin_lock_irqsave(&lio->ptp_lock, flags);
2017 	ns = lio_pci_readq(oct, CN6XXX_MIO_PTP_CLOCK_HI);
2018 	ns += lio->ptp_adjust;
2019 	spin_unlock_irqrestore(&lio->ptp_lock, flags);
2020 
2021 	*ts = ns_to_timespec64(ns);
2022 
2023 	return 0;
2024 }
2025 
2026 /**
2027  * \brief Set hardware clock time. Reset adjustment
2028  * @param ptp PTP clock info
2029  * @param ts timespec
2030  */
2031 static int liquidio_ptp_settime(struct ptp_clock_info *ptp,
2032 				const struct timespec64 *ts)
2033 {
2034 	u64 ns;
2035 	unsigned long flags;
2036 	struct lio *lio = container_of(ptp, struct lio, ptp_info);
2037 	struct octeon_device *oct = (struct octeon_device *)lio->oct_dev;
2038 
2039 	ns = timespec_to_ns(ts);
2040 
2041 	spin_lock_irqsave(&lio->ptp_lock, flags);
2042 	lio_pci_writeq(oct, ns, CN6XXX_MIO_PTP_CLOCK_HI);
2043 	lio->ptp_adjust = 0;
2044 	spin_unlock_irqrestore(&lio->ptp_lock, flags);
2045 
2046 	return 0;
2047 }
2048 
2049 /**
2050  * \brief Check if PTP is enabled
2051  * @param ptp PTP clock info
2052  * @param rq request
2053  * @param on is it on
2054  */
2055 static int
2056 liquidio_ptp_enable(struct ptp_clock_info *ptp __attribute__((unused)),
2057 		    struct ptp_clock_request *rq __attribute__((unused)),
2058 		    int on __attribute__((unused)))
2059 {
2060 	return -EOPNOTSUPP;
2061 }
2062 
2063 /**
2064  * \brief Open PTP clock source
2065  * @param netdev network device
2066  */
2067 static void oct_ptp_open(struct net_device *netdev)
2068 {
2069 	struct lio *lio = GET_LIO(netdev);
2070 	struct octeon_device *oct = (struct octeon_device *)lio->oct_dev;
2071 
2072 	spin_lock_init(&lio->ptp_lock);
2073 
2074 	snprintf(lio->ptp_info.name, 16, "%s", netdev->name);
2075 	lio->ptp_info.owner = THIS_MODULE;
2076 	lio->ptp_info.max_adj = 250000000;
2077 	lio->ptp_info.n_alarm = 0;
2078 	lio->ptp_info.n_ext_ts = 0;
2079 	lio->ptp_info.n_per_out = 0;
2080 	lio->ptp_info.pps = 0;
2081 	lio->ptp_info.adjfreq = liquidio_ptp_adjfreq;
2082 	lio->ptp_info.adjtime = liquidio_ptp_adjtime;
2083 	lio->ptp_info.gettime64 = liquidio_ptp_gettime;
2084 	lio->ptp_info.settime64 = liquidio_ptp_settime;
2085 	lio->ptp_info.enable = liquidio_ptp_enable;
2086 
2087 	lio->ptp_adjust = 0;
2088 
2089 	lio->ptp_clock = ptp_clock_register(&lio->ptp_info,
2090 					     &oct->pci_dev->dev);
2091 
2092 	if (IS_ERR(lio->ptp_clock))
2093 		lio->ptp_clock = NULL;
2094 }
2095 
2096 /**
2097  * \brief Init PTP clock
2098  * @param oct octeon device
2099  */
2100 static void liquidio_ptp_init(struct octeon_device *oct)
2101 {
2102 	u64 clock_comp, cfg;
2103 
2104 	clock_comp = (u64)NSEC_PER_SEC << 32;
2105 	do_div(clock_comp, oct->coproc_clock_rate);
2106 	lio_pci_writeq(oct, clock_comp, CN6XXX_MIO_PTP_CLOCK_COMP);
2107 
2108 	/* Enable */
2109 	cfg = lio_pci_readq(oct, CN6XXX_MIO_PTP_CLOCK_CFG);
2110 	lio_pci_writeq(oct, cfg | 0x01, CN6XXX_MIO_PTP_CLOCK_CFG);
2111 }
2112 
2113 /**
2114  * \brief Load firmware to device
2115  * @param oct octeon device
2116  *
2117  * Maps device to firmware filename, requests firmware, and downloads it
2118  */
2119 static int load_firmware(struct octeon_device *oct)
2120 {
2121 	int ret = 0;
2122 	const struct firmware *fw;
2123 	char fw_name[LIO_MAX_FW_FILENAME_LEN];
2124 	char *tmp_fw_type;
2125 
2126 	if (strncmp(fw_type, LIO_FW_NAME_TYPE_NONE,
2127 		    sizeof(LIO_FW_NAME_TYPE_NONE)) == 0) {
2128 		dev_info(&oct->pci_dev->dev, "Skipping firmware load\n");
2129 		return ret;
2130 	}
2131 
2132 	if (fw_type[0] == '\0')
2133 		tmp_fw_type = LIO_FW_NAME_TYPE_NIC;
2134 	else
2135 		tmp_fw_type = fw_type;
2136 
2137 	sprintf(fw_name, "%s%s%s_%s%s", LIO_FW_DIR, LIO_FW_BASE_NAME,
2138 		octeon_get_conf(oct)->card_name, tmp_fw_type,
2139 		LIO_FW_NAME_SUFFIX);
2140 
2141 	ret = request_firmware(&fw, fw_name, &oct->pci_dev->dev);
2142 	if (ret) {
2143 		dev_err(&oct->pci_dev->dev, "Request firmware failed. Could not find file %s.\n.",
2144 			fw_name);
2145 		release_firmware(fw);
2146 		return ret;
2147 	}
2148 
2149 	ret = octeon_download_firmware(oct, fw->data, fw->size);
2150 
2151 	release_firmware(fw);
2152 
2153 	return ret;
2154 }
2155 
2156 /**
2157  * \brief Setup output queue
2158  * @param oct octeon device
2159  * @param q_no which queue
2160  * @param num_descs how many descriptors
2161  * @param desc_size size of each descriptor
2162  * @param app_ctx application context
2163  */
2164 static int octeon_setup_droq(struct octeon_device *oct, int q_no, int num_descs,
2165 			     int desc_size, void *app_ctx)
2166 {
2167 	int ret_val = 0;
2168 
2169 	dev_dbg(&oct->pci_dev->dev, "Creating Droq: %d\n", q_no);
2170 	/* droq creation and local register settings. */
2171 	ret_val = octeon_create_droq(oct, q_no, num_descs, desc_size, app_ctx);
2172 	if (ret_val < 0)
2173 		return ret_val;
2174 
2175 	if (ret_val == 1) {
2176 		dev_dbg(&oct->pci_dev->dev, "Using default droq %d\n", q_no);
2177 		return 0;
2178 	}
2179 	/* tasklet creation for the droq */
2180 
2181 	/* Enable the droq queues */
2182 	octeon_set_droq_pkt_op(oct, q_no, 1);
2183 
2184 	/* Send Credit for Octeon Output queues. Credits are always
2185 	 * sent after the output queue is enabled.
2186 	 */
2187 	writel(oct->droq[q_no]->max_count,
2188 	       oct->droq[q_no]->pkts_credit_reg);
2189 
2190 	return ret_val;
2191 }
2192 
2193 /**
2194  * \brief Callback for getting interface configuration
2195  * @param status status of request
2196  * @param buf pointer to resp structure
2197  */
2198 static void if_cfg_callback(struct octeon_device *oct,
2199 			    u32 status __attribute__((unused)),
2200 			    void *buf)
2201 {
2202 	struct octeon_soft_command *sc = (struct octeon_soft_command *)buf;
2203 	struct liquidio_if_cfg_resp *resp;
2204 	struct liquidio_if_cfg_context *ctx;
2205 
2206 	resp = (struct liquidio_if_cfg_resp *)sc->virtrptr;
2207 	ctx = (struct liquidio_if_cfg_context *)sc->ctxptr;
2208 
2209 	oct = lio_get_device(ctx->octeon_id);
2210 	if (resp->status)
2211 		dev_err(&oct->pci_dev->dev, "nic if cfg instruction failed. Status: %llx\n",
2212 			CVM_CAST64(resp->status));
2213 	WRITE_ONCE(ctx->cond, 1);
2214 
2215 	snprintf(oct->fw_info.liquidio_firmware_version, 32, "%s",
2216 		 resp->cfg_info.liquidio_firmware_version);
2217 
2218 	/* This barrier is required to be sure that the response has been
2219 	 * written fully before waking up the handler
2220 	 */
2221 	wmb();
2222 
2223 	wake_up_interruptible(&ctx->wc);
2224 }
2225 
2226 /**
2227  * \brief Select queue based on hash
2228  * @param dev Net device
2229  * @param skb sk_buff structure
2230  * @returns selected queue number
2231  */
2232 static u16 select_q(struct net_device *dev, struct sk_buff *skb,
2233 		    void *accel_priv __attribute__((unused)),
2234 		    select_queue_fallback_t fallback __attribute__((unused)))
2235 {
2236 	u32 qindex = 0;
2237 	struct lio *lio;
2238 
2239 	lio = GET_LIO(dev);
2240 	qindex = skb_tx_hash(dev, skb);
2241 
2242 	return (u16)(qindex % (lio->linfo.num_txpciq));
2243 }
2244 
2245 /** Routine to push packets arriving on Octeon interface upto network layer.
2246  * @param oct_id   - octeon device id.
2247  * @param skbuff   - skbuff struct to be passed to network layer.
2248  * @param len      - size of total data received.
2249  * @param rh       - Control header associated with the packet
2250  * @param param    - additional control data with the packet
2251  * @param arg	   - farg registered in droq_ops
2252  */
2253 static void
2254 liquidio_push_packet(u32 octeon_id __attribute__((unused)),
2255 		     void *skbuff,
2256 		     u32 len,
2257 		     union octeon_rh *rh,
2258 		     void *param,
2259 		     void *arg)
2260 {
2261 	struct napi_struct *napi = param;
2262 	struct sk_buff *skb = (struct sk_buff *)skbuff;
2263 	struct skb_shared_hwtstamps *shhwtstamps;
2264 	u64 ns;
2265 	u16 vtag = 0;
2266 	struct net_device *netdev = (struct net_device *)arg;
2267 	struct octeon_droq *droq = container_of(param, struct octeon_droq,
2268 						napi);
2269 	if (netdev) {
2270 		int packet_was_received;
2271 		struct lio *lio = GET_LIO(netdev);
2272 		struct octeon_device *oct = lio->oct_dev;
2273 
2274 		/* Do not proceed if the interface is not in RUNNING state. */
2275 		if (!ifstate_check(lio, LIO_IFSTATE_RUNNING)) {
2276 			recv_buffer_free(skb);
2277 			droq->stats.rx_dropped++;
2278 			return;
2279 		}
2280 
2281 		skb->dev = netdev;
2282 
2283 		skb_record_rx_queue(skb, droq->q_no);
2284 		if (likely(len > MIN_SKB_SIZE)) {
2285 			struct octeon_skb_page_info *pg_info;
2286 			unsigned char *va;
2287 
2288 			pg_info = ((struct octeon_skb_page_info *)(skb->cb));
2289 			if (pg_info->page) {
2290 				/* For Paged allocation use the frags */
2291 				va = page_address(pg_info->page) +
2292 					pg_info->page_offset;
2293 				memcpy(skb->data, va, MIN_SKB_SIZE);
2294 				skb_put(skb, MIN_SKB_SIZE);
2295 				skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
2296 						pg_info->page,
2297 						pg_info->page_offset +
2298 						MIN_SKB_SIZE,
2299 						len - MIN_SKB_SIZE,
2300 						LIO_RXBUFFER_SZ);
2301 			}
2302 		} else {
2303 			struct octeon_skb_page_info *pg_info =
2304 				((struct octeon_skb_page_info *)(skb->cb));
2305 			skb_copy_to_linear_data(skb, page_address(pg_info->page)
2306 						+ pg_info->page_offset, len);
2307 			skb_put(skb, len);
2308 			put_page(pg_info->page);
2309 		}
2310 
2311 		if (((oct->chip_id == OCTEON_CN66XX) ||
2312 		     (oct->chip_id == OCTEON_CN68XX)) &&
2313 		    ptp_enable) {
2314 			if (rh->r_dh.has_hwtstamp) {
2315 				/* timestamp is included from the hardware at
2316 				 * the beginning of the packet.
2317 				 */
2318 				if (ifstate_check
2319 				    (lio, LIO_IFSTATE_RX_TIMESTAMP_ENABLED)) {
2320 					/* Nanoseconds are in the first 64-bits
2321 					 * of the packet.
2322 					 */
2323 					memcpy(&ns, (skb->data), sizeof(ns));
2324 					shhwtstamps = skb_hwtstamps(skb);
2325 					shhwtstamps->hwtstamp =
2326 						ns_to_ktime(ns +
2327 							    lio->ptp_adjust);
2328 				}
2329 				skb_pull(skb, sizeof(ns));
2330 			}
2331 		}
2332 
2333 		skb->protocol = eth_type_trans(skb, skb->dev);
2334 		if ((netdev->features & NETIF_F_RXCSUM) &&
2335 		    (((rh->r_dh.encap_on) &&
2336 		      (rh->r_dh.csum_verified & CNNIC_TUN_CSUM_VERIFIED)) ||
2337 		     (!(rh->r_dh.encap_on) &&
2338 		      (rh->r_dh.csum_verified & CNNIC_CSUM_VERIFIED))))
2339 			/* checksum has already been verified */
2340 			skb->ip_summed = CHECKSUM_UNNECESSARY;
2341 		else
2342 			skb->ip_summed = CHECKSUM_NONE;
2343 
2344 		/* Setting Encapsulation field on basis of status received
2345 		 * from the firmware
2346 		 */
2347 		if (rh->r_dh.encap_on) {
2348 			skb->encapsulation = 1;
2349 			skb->csum_level = 1;
2350 			droq->stats.rx_vxlan++;
2351 		}
2352 
2353 		/* inbound VLAN tag */
2354 		if ((netdev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
2355 		    (rh->r_dh.vlan != 0)) {
2356 			u16 vid = rh->r_dh.vlan;
2357 			u16 priority = rh->r_dh.priority;
2358 
2359 			vtag = priority << 13 | vid;
2360 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vtag);
2361 		}
2362 
2363 		packet_was_received = napi_gro_receive(napi, skb) != GRO_DROP;
2364 
2365 		if (packet_was_received) {
2366 			droq->stats.rx_bytes_received += len;
2367 			droq->stats.rx_pkts_received++;
2368 			netdev->last_rx = jiffies;
2369 		} else {
2370 			droq->stats.rx_dropped++;
2371 			netif_info(lio, rx_err, lio->netdev,
2372 				   "droq:%d  error rx_dropped:%llu\n",
2373 				   droq->q_no, droq->stats.rx_dropped);
2374 		}
2375 
2376 	} else {
2377 		recv_buffer_free(skb);
2378 	}
2379 }
2380 
2381 /**
2382  * \brief wrapper for calling napi_schedule
2383  * @param param parameters to pass to napi_schedule
2384  *
2385  * Used when scheduling on different CPUs
2386  */
2387 static void napi_schedule_wrapper(void *param)
2388 {
2389 	struct napi_struct *napi = param;
2390 
2391 	napi_schedule(napi);
2392 }
2393 
2394 /**
2395  * \brief callback when receive interrupt occurs and we are in NAPI mode
2396  * @param arg pointer to octeon output queue
2397  */
2398 static void liquidio_napi_drv_callback(void *arg)
2399 {
2400 	struct octeon_device *oct;
2401 	struct octeon_droq *droq = arg;
2402 	int this_cpu = smp_processor_id();
2403 
2404 	oct = droq->oct_dev;
2405 
2406 	if (OCTEON_CN23XX_PF(oct) || droq->cpu_id == this_cpu) {
2407 		napi_schedule_irqoff(&droq->napi);
2408 	} else {
2409 		struct call_single_data *csd = &droq->csd;
2410 
2411 		csd->func = napi_schedule_wrapper;
2412 		csd->info = &droq->napi;
2413 		csd->flags = 0;
2414 
2415 		smp_call_function_single_async(droq->cpu_id, csd);
2416 	}
2417 }
2418 
2419 /**
2420  * \brief Entry point for NAPI polling
2421  * @param napi NAPI structure
2422  * @param budget maximum number of items to process
2423  */
2424 static int liquidio_napi_poll(struct napi_struct *napi, int budget)
2425 {
2426 	struct octeon_droq *droq;
2427 	int work_done;
2428 	int tx_done = 0, iq_no;
2429 	struct octeon_instr_queue *iq;
2430 	struct octeon_device *oct;
2431 
2432 	droq = container_of(napi, struct octeon_droq, napi);
2433 	oct = droq->oct_dev;
2434 	iq_no = droq->q_no;
2435 	/* Handle Droq descriptors */
2436 	work_done = octeon_process_droq_poll_cmd(oct, droq->q_no,
2437 						 POLL_EVENT_PROCESS_PKTS,
2438 						 budget);
2439 
2440 	/* Flush the instruction queue */
2441 	iq = oct->instr_queue[iq_no];
2442 	if (iq) {
2443 		/* Process iq buffers with in the budget limits */
2444 		tx_done = octeon_flush_iq(oct, iq, 1, budget);
2445 		/* Update iq read-index rather than waiting for next interrupt.
2446 		 * Return back if tx_done is false.
2447 		 */
2448 		update_txq_status(oct, iq_no);
2449 	} else {
2450 		dev_err(&oct->pci_dev->dev, "%s:  iq (%d) num invalid\n",
2451 			__func__, iq_no);
2452 	}
2453 
2454 	if ((work_done < budget) && (tx_done)) {
2455 		napi_complete(napi);
2456 		octeon_process_droq_poll_cmd(droq->oct_dev, droq->q_no,
2457 					     POLL_EVENT_ENABLE_INTR, 0);
2458 		return 0;
2459 	}
2460 
2461 	return (!tx_done) ? (budget) : (work_done);
2462 }
2463 
2464 /**
2465  * \brief Setup input and output queues
2466  * @param octeon_dev octeon device
2467  * @param ifidx  Interface Index
2468  *
2469  * Note: Queues are with respect to the octeon device. Thus
2470  * an input queue is for egress packets, and output queues
2471  * are for ingress packets.
2472  */
2473 static inline int setup_io_queues(struct octeon_device *octeon_dev,
2474 				  int ifidx)
2475 {
2476 	struct octeon_droq_ops droq_ops;
2477 	struct net_device *netdev;
2478 	static int cpu_id;
2479 	static int cpu_id_modulus;
2480 	struct octeon_droq *droq;
2481 	struct napi_struct *napi;
2482 	int q, q_no, retval = 0;
2483 	struct lio *lio;
2484 	int num_tx_descs;
2485 
2486 	netdev = octeon_dev->props[ifidx].netdev;
2487 
2488 	lio = GET_LIO(netdev);
2489 
2490 	memset(&droq_ops, 0, sizeof(struct octeon_droq_ops));
2491 
2492 	droq_ops.fptr = liquidio_push_packet;
2493 	droq_ops.farg = (void *)netdev;
2494 
2495 	droq_ops.poll_mode = 1;
2496 	droq_ops.napi_fn = liquidio_napi_drv_callback;
2497 	cpu_id = 0;
2498 	cpu_id_modulus = num_present_cpus();
2499 
2500 	/* set up DROQs. */
2501 	for (q = 0; q < lio->linfo.num_rxpciq; q++) {
2502 		q_no = lio->linfo.rxpciq[q].s.q_no;
2503 		dev_dbg(&octeon_dev->pci_dev->dev,
2504 			"setup_io_queues index:%d linfo.rxpciq.s.q_no:%d\n",
2505 			q, q_no);
2506 		retval = octeon_setup_droq(octeon_dev, q_no,
2507 					   CFG_GET_NUM_RX_DESCS_NIC_IF
2508 						   (octeon_get_conf(octeon_dev),
2509 						   lio->ifidx),
2510 					   CFG_GET_NUM_RX_BUF_SIZE_NIC_IF
2511 						   (octeon_get_conf(octeon_dev),
2512 						   lio->ifidx), NULL);
2513 		if (retval) {
2514 			dev_err(&octeon_dev->pci_dev->dev,
2515 				"%s : Runtime DROQ(RxQ) creation failed.\n",
2516 				__func__);
2517 			return 1;
2518 		}
2519 
2520 		droq = octeon_dev->droq[q_no];
2521 		napi = &droq->napi;
2522 		dev_dbg(&octeon_dev->pci_dev->dev, "netif_napi_add netdev:%llx oct:%llx pf_num:%d\n",
2523 			(u64)netdev, (u64)octeon_dev, octeon_dev->pf_num);
2524 		netif_napi_add(netdev, napi, liquidio_napi_poll, 64);
2525 
2526 		/* designate a CPU for this droq */
2527 		droq->cpu_id = cpu_id;
2528 		cpu_id++;
2529 		if (cpu_id >= cpu_id_modulus)
2530 			cpu_id = 0;
2531 
2532 		octeon_register_droq_ops(octeon_dev, q_no, &droq_ops);
2533 	}
2534 
2535 	if (OCTEON_CN23XX_PF(octeon_dev)) {
2536 		/* 23XX PF can receive control messages (via the first PF-owned
2537 		 * droq) from the firmware even if the ethX interface is down,
2538 		 * so that's why poll_mode must be off for the first droq.
2539 		 */
2540 		octeon_dev->droq[0]->ops.poll_mode = 0;
2541 	}
2542 
2543 	/* set up IQs. */
2544 	for (q = 0; q < lio->linfo.num_txpciq; q++) {
2545 		num_tx_descs = CFG_GET_NUM_TX_DESCS_NIC_IF(octeon_get_conf
2546 							   (octeon_dev),
2547 							   lio->ifidx);
2548 		retval = octeon_setup_iq(octeon_dev, ifidx, q,
2549 					 lio->linfo.txpciq[q], num_tx_descs,
2550 					 netdev_get_tx_queue(netdev, q));
2551 		if (retval) {
2552 			dev_err(&octeon_dev->pci_dev->dev,
2553 				" %s : Runtime IQ(TxQ) creation failed.\n",
2554 				__func__);
2555 			return 1;
2556 		}
2557 	}
2558 
2559 	return 0;
2560 }
2561 
2562 /**
2563  * \brief Poll routine for checking transmit queue status
2564  * @param work work_struct data structure
2565  */
2566 static void octnet_poll_check_txq_status(struct work_struct *work)
2567 {
2568 	struct cavium_wk *wk = (struct cavium_wk *)work;
2569 	struct lio *lio = (struct lio *)wk->ctxptr;
2570 
2571 	if (!ifstate_check(lio, LIO_IFSTATE_RUNNING))
2572 		return;
2573 
2574 	check_txq_status(lio);
2575 	queue_delayed_work(lio->txq_status_wq.wq,
2576 			   &lio->txq_status_wq.wk.work, msecs_to_jiffies(1));
2577 }
2578 
2579 /**
2580  * \brief Sets up the txq poll check
2581  * @param netdev network device
2582  */
2583 static inline int setup_tx_poll_fn(struct net_device *netdev)
2584 {
2585 	struct lio *lio = GET_LIO(netdev);
2586 	struct octeon_device *oct = lio->oct_dev;
2587 
2588 	lio->txq_status_wq.wq = alloc_workqueue("txq-status",
2589 						WQ_MEM_RECLAIM, 0);
2590 	if (!lio->txq_status_wq.wq) {
2591 		dev_err(&oct->pci_dev->dev, "unable to create cavium txq status wq\n");
2592 		return -1;
2593 	}
2594 	INIT_DELAYED_WORK(&lio->txq_status_wq.wk.work,
2595 			  octnet_poll_check_txq_status);
2596 	lio->txq_status_wq.wk.ctxptr = lio;
2597 	queue_delayed_work(lio->txq_status_wq.wq,
2598 			   &lio->txq_status_wq.wk.work, msecs_to_jiffies(1));
2599 	return 0;
2600 }
2601 
2602 static inline void cleanup_tx_poll_fn(struct net_device *netdev)
2603 {
2604 	struct lio *lio = GET_LIO(netdev);
2605 
2606 	if (lio->txq_status_wq.wq) {
2607 		cancel_delayed_work_sync(&lio->txq_status_wq.wk.work);
2608 		destroy_workqueue(lio->txq_status_wq.wq);
2609 	}
2610 }
2611 
2612 /**
2613  * \brief Net device open for LiquidIO
2614  * @param netdev network device
2615  */
2616 static int liquidio_open(struct net_device *netdev)
2617 {
2618 	struct lio *lio = GET_LIO(netdev);
2619 	struct octeon_device *oct = lio->oct_dev;
2620 	struct napi_struct *napi, *n;
2621 
2622 	if (oct->props[lio->ifidx].napi_enabled == 0) {
2623 		list_for_each_entry_safe(napi, n, &netdev->napi_list, dev_list)
2624 			napi_enable(napi);
2625 
2626 		oct->props[lio->ifidx].napi_enabled = 1;
2627 
2628 		if (OCTEON_CN23XX_PF(oct))
2629 			oct->droq[0]->ops.poll_mode = 1;
2630 	}
2631 
2632 	oct_ptp_open(netdev);
2633 
2634 	ifstate_set(lio, LIO_IFSTATE_RUNNING);
2635 
2636 	/* Ready for link status updates */
2637 	lio->intf_open = 1;
2638 
2639 	netif_info(lio, ifup, lio->netdev, "Interface Open, ready for traffic\n");
2640 
2641 	if (OCTEON_CN23XX_PF(oct)) {
2642 		if (!oct->msix_on)
2643 			if (setup_tx_poll_fn(netdev))
2644 				return -1;
2645 	} else {
2646 		if (setup_tx_poll_fn(netdev))
2647 			return -1;
2648 	}
2649 
2650 	start_txq(netdev);
2651 
2652 	/* tell Octeon to start forwarding packets to host */
2653 	send_rx_ctrl_cmd(lio, 1);
2654 
2655 	dev_info(&oct->pci_dev->dev, "%s interface is opened\n",
2656 		 netdev->name);
2657 
2658 	return 0;
2659 }
2660 
2661 /**
2662  * \brief Net device stop for LiquidIO
2663  * @param netdev network device
2664  */
2665 static int liquidio_stop(struct net_device *netdev)
2666 {
2667 	struct lio *lio = GET_LIO(netdev);
2668 	struct octeon_device *oct = lio->oct_dev;
2669 
2670 	ifstate_reset(lio, LIO_IFSTATE_RUNNING);
2671 
2672 	netif_tx_disable(netdev);
2673 
2674 	/* Inform that netif carrier is down */
2675 	netif_carrier_off(netdev);
2676 	lio->intf_open = 0;
2677 	lio->linfo.link.s.link_up = 0;
2678 	lio->link_changes++;
2679 
2680 	/* Pause for a moment and wait for Octeon to flush out (to the wire) any
2681 	 * egress packets that are in-flight.
2682 	 */
2683 	set_current_state(TASK_INTERRUPTIBLE);
2684 	schedule_timeout(msecs_to_jiffies(100));
2685 
2686 	/* Now it should be safe to tell Octeon that nic interface is down. */
2687 	send_rx_ctrl_cmd(lio, 0);
2688 
2689 	if (OCTEON_CN23XX_PF(oct)) {
2690 		if (!oct->msix_on)
2691 			cleanup_tx_poll_fn(netdev);
2692 	} else {
2693 		cleanup_tx_poll_fn(netdev);
2694 	}
2695 
2696 	if (lio->ptp_clock) {
2697 		ptp_clock_unregister(lio->ptp_clock);
2698 		lio->ptp_clock = NULL;
2699 	}
2700 
2701 	dev_info(&oct->pci_dev->dev, "%s interface is stopped\n", netdev->name);
2702 
2703 	return 0;
2704 }
2705 
2706 /**
2707  * \brief Converts a mask based on net device flags
2708  * @param netdev network device
2709  *
2710  * This routine generates a octnet_ifflags mask from the net device flags
2711  * received from the OS.
2712  */
2713 static inline enum octnet_ifflags get_new_flags(struct net_device *netdev)
2714 {
2715 	enum octnet_ifflags f = OCTNET_IFFLAG_UNICAST;
2716 
2717 	if (netdev->flags & IFF_PROMISC)
2718 		f |= OCTNET_IFFLAG_PROMISC;
2719 
2720 	if (netdev->flags & IFF_ALLMULTI)
2721 		f |= OCTNET_IFFLAG_ALLMULTI;
2722 
2723 	if (netdev->flags & IFF_MULTICAST) {
2724 		f |= OCTNET_IFFLAG_MULTICAST;
2725 
2726 		/* Accept all multicast addresses if there are more than we
2727 		 * can handle
2728 		 */
2729 		if (netdev_mc_count(netdev) > MAX_OCTEON_MULTICAST_ADDR)
2730 			f |= OCTNET_IFFLAG_ALLMULTI;
2731 	}
2732 
2733 	if (netdev->flags & IFF_BROADCAST)
2734 		f |= OCTNET_IFFLAG_BROADCAST;
2735 
2736 	return f;
2737 }
2738 
2739 /**
2740  * \brief Net device set_multicast_list
2741  * @param netdev network device
2742  */
2743 static void liquidio_set_mcast_list(struct net_device *netdev)
2744 {
2745 	struct lio *lio = GET_LIO(netdev);
2746 	struct octeon_device *oct = lio->oct_dev;
2747 	struct octnic_ctrl_pkt nctrl;
2748 	struct netdev_hw_addr *ha;
2749 	u64 *mc;
2750 	int ret;
2751 	int mc_count = min(netdev_mc_count(netdev), MAX_OCTEON_MULTICAST_ADDR);
2752 
2753 	memset(&nctrl, 0, sizeof(struct octnic_ctrl_pkt));
2754 
2755 	/* Create a ctrl pkt command to be sent to core app. */
2756 	nctrl.ncmd.u64 = 0;
2757 	nctrl.ncmd.s.cmd = OCTNET_CMD_SET_MULTI_LIST;
2758 	nctrl.ncmd.s.param1 = get_new_flags(netdev);
2759 	nctrl.ncmd.s.param2 = mc_count;
2760 	nctrl.ncmd.s.more = mc_count;
2761 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
2762 	nctrl.netpndev = (u64)netdev;
2763 	nctrl.cb_fn = liquidio_link_ctrl_cmd_completion;
2764 
2765 	/* copy all the addresses into the udd */
2766 	mc = &nctrl.udd[0];
2767 	netdev_for_each_mc_addr(ha, netdev) {
2768 		*mc = 0;
2769 		memcpy(((u8 *)mc) + 2, ha->addr, ETH_ALEN);
2770 		/* no need to swap bytes */
2771 
2772 		if (++mc > &nctrl.udd[mc_count])
2773 			break;
2774 	}
2775 
2776 	/* Apparently, any activity in this call from the kernel has to
2777 	 * be atomic. So we won't wait for response.
2778 	 */
2779 	nctrl.wait_time = 0;
2780 
2781 	ret = octnet_send_nic_ctrl_pkt(lio->oct_dev, &nctrl);
2782 	if (ret < 0) {
2783 		dev_err(&oct->pci_dev->dev, "DEVFLAGS change failed in core (ret: 0x%x)\n",
2784 			ret);
2785 	}
2786 }
2787 
2788 /**
2789  * \brief Net device set_mac_address
2790  * @param netdev network device
2791  */
2792 static int liquidio_set_mac(struct net_device *netdev, void *p)
2793 {
2794 	int ret = 0;
2795 	struct lio *lio = GET_LIO(netdev);
2796 	struct octeon_device *oct = lio->oct_dev;
2797 	struct sockaddr *addr = (struct sockaddr *)p;
2798 	struct octnic_ctrl_pkt nctrl;
2799 
2800 	if (!is_valid_ether_addr(addr->sa_data))
2801 		return -EADDRNOTAVAIL;
2802 
2803 	memset(&nctrl, 0, sizeof(struct octnic_ctrl_pkt));
2804 
2805 	nctrl.ncmd.u64 = 0;
2806 	nctrl.ncmd.s.cmd = OCTNET_CMD_CHANGE_MACADDR;
2807 	nctrl.ncmd.s.param1 = 0;
2808 	nctrl.ncmd.s.more = 1;
2809 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
2810 	nctrl.netpndev = (u64)netdev;
2811 	nctrl.cb_fn = liquidio_link_ctrl_cmd_completion;
2812 	nctrl.wait_time = 100;
2813 
2814 	nctrl.udd[0] = 0;
2815 	/* The MAC Address is presented in network byte order. */
2816 	memcpy((u8 *)&nctrl.udd[0] + 2, addr->sa_data, ETH_ALEN);
2817 
2818 	ret = octnet_send_nic_ctrl_pkt(lio->oct_dev, &nctrl);
2819 	if (ret < 0) {
2820 		dev_err(&oct->pci_dev->dev, "MAC Address change failed\n");
2821 		return -ENOMEM;
2822 	}
2823 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2824 	memcpy(((u8 *)&lio->linfo.hw_addr) + 2, addr->sa_data, ETH_ALEN);
2825 
2826 	return 0;
2827 }
2828 
2829 /**
2830  * \brief Net device get_stats
2831  * @param netdev network device
2832  */
2833 static struct net_device_stats *liquidio_get_stats(struct net_device *netdev)
2834 {
2835 	struct lio *lio = GET_LIO(netdev);
2836 	struct net_device_stats *stats = &netdev->stats;
2837 	struct octeon_device *oct;
2838 	u64 pkts = 0, drop = 0, bytes = 0;
2839 	struct oct_droq_stats *oq_stats;
2840 	struct oct_iq_stats *iq_stats;
2841 	int i, iq_no, oq_no;
2842 
2843 	oct = lio->oct_dev;
2844 
2845 	for (i = 0; i < lio->linfo.num_txpciq; i++) {
2846 		iq_no = lio->linfo.txpciq[i].s.q_no;
2847 		iq_stats = &oct->instr_queue[iq_no]->stats;
2848 		pkts += iq_stats->tx_done;
2849 		drop += iq_stats->tx_dropped;
2850 		bytes += iq_stats->tx_tot_bytes;
2851 	}
2852 
2853 	stats->tx_packets = pkts;
2854 	stats->tx_bytes = bytes;
2855 	stats->tx_dropped = drop;
2856 
2857 	pkts = 0;
2858 	drop = 0;
2859 	bytes = 0;
2860 
2861 	for (i = 0; i < lio->linfo.num_rxpciq; i++) {
2862 		oq_no = lio->linfo.rxpciq[i].s.q_no;
2863 		oq_stats = &oct->droq[oq_no]->stats;
2864 		pkts += oq_stats->rx_pkts_received;
2865 		drop += (oq_stats->rx_dropped +
2866 			 oq_stats->dropped_nodispatch +
2867 			 oq_stats->dropped_toomany +
2868 			 oq_stats->dropped_nomem);
2869 		bytes += oq_stats->rx_bytes_received;
2870 	}
2871 
2872 	stats->rx_bytes = bytes;
2873 	stats->rx_packets = pkts;
2874 	stats->rx_dropped = drop;
2875 
2876 	return stats;
2877 }
2878 
2879 /**
2880  * \brief Net device change_mtu
2881  * @param netdev network device
2882  */
2883 static int liquidio_change_mtu(struct net_device *netdev, int new_mtu)
2884 {
2885 	struct lio *lio = GET_LIO(netdev);
2886 	struct octeon_device *oct = lio->oct_dev;
2887 	struct octnic_ctrl_pkt nctrl;
2888 	int ret = 0;
2889 
2890 	memset(&nctrl, 0, sizeof(struct octnic_ctrl_pkt));
2891 
2892 	nctrl.ncmd.u64 = 0;
2893 	nctrl.ncmd.s.cmd = OCTNET_CMD_CHANGE_MTU;
2894 	nctrl.ncmd.s.param1 = new_mtu;
2895 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
2896 	nctrl.wait_time = 100;
2897 	nctrl.netpndev = (u64)netdev;
2898 	nctrl.cb_fn = liquidio_link_ctrl_cmd_completion;
2899 
2900 	ret = octnet_send_nic_ctrl_pkt(lio->oct_dev, &nctrl);
2901 	if (ret < 0) {
2902 		dev_err(&oct->pci_dev->dev, "Failed to set MTU\n");
2903 		return -1;
2904 	}
2905 
2906 	lio->mtu = new_mtu;
2907 
2908 	return 0;
2909 }
2910 
2911 /**
2912  * \brief Handler for SIOCSHWTSTAMP ioctl
2913  * @param netdev network device
2914  * @param ifr interface request
2915  * @param cmd command
2916  */
2917 static int hwtstamp_ioctl(struct net_device *netdev, struct ifreq *ifr)
2918 {
2919 	struct hwtstamp_config conf;
2920 	struct lio *lio = GET_LIO(netdev);
2921 
2922 	if (copy_from_user(&conf, ifr->ifr_data, sizeof(conf)))
2923 		return -EFAULT;
2924 
2925 	if (conf.flags)
2926 		return -EINVAL;
2927 
2928 	switch (conf.tx_type) {
2929 	case HWTSTAMP_TX_ON:
2930 	case HWTSTAMP_TX_OFF:
2931 		break;
2932 	default:
2933 		return -ERANGE;
2934 	}
2935 
2936 	switch (conf.rx_filter) {
2937 	case HWTSTAMP_FILTER_NONE:
2938 		break;
2939 	case HWTSTAMP_FILTER_ALL:
2940 	case HWTSTAMP_FILTER_SOME:
2941 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2942 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2943 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2944 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2945 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2946 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2947 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2948 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2949 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2950 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
2951 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
2952 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2953 		conf.rx_filter = HWTSTAMP_FILTER_ALL;
2954 		break;
2955 	default:
2956 		return -ERANGE;
2957 	}
2958 
2959 	if (conf.rx_filter == HWTSTAMP_FILTER_ALL)
2960 		ifstate_set(lio, LIO_IFSTATE_RX_TIMESTAMP_ENABLED);
2961 
2962 	else
2963 		ifstate_reset(lio, LIO_IFSTATE_RX_TIMESTAMP_ENABLED);
2964 
2965 	return copy_to_user(ifr->ifr_data, &conf, sizeof(conf)) ? -EFAULT : 0;
2966 }
2967 
2968 /**
2969  * \brief ioctl handler
2970  * @param netdev network device
2971  * @param ifr interface request
2972  * @param cmd command
2973  */
2974 static int liquidio_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2975 {
2976 	switch (cmd) {
2977 	case SIOCSHWTSTAMP:
2978 		return hwtstamp_ioctl(netdev, ifr);
2979 	default:
2980 		return -EOPNOTSUPP;
2981 	}
2982 }
2983 
2984 /**
2985  * \brief handle a Tx timestamp response
2986  * @param status response status
2987  * @param buf pointer to skb
2988  */
2989 static void handle_timestamp(struct octeon_device *oct,
2990 			     u32 status,
2991 			     void *buf)
2992 {
2993 	struct octnet_buf_free_info *finfo;
2994 	struct octeon_soft_command *sc;
2995 	struct oct_timestamp_resp *resp;
2996 	struct lio *lio;
2997 	struct sk_buff *skb = (struct sk_buff *)buf;
2998 
2999 	finfo = (struct octnet_buf_free_info *)skb->cb;
3000 	lio = finfo->lio;
3001 	sc = finfo->sc;
3002 	oct = lio->oct_dev;
3003 	resp = (struct oct_timestamp_resp *)sc->virtrptr;
3004 
3005 	if (status != OCTEON_REQUEST_DONE) {
3006 		dev_err(&oct->pci_dev->dev, "Tx timestamp instruction failed. Status: %llx\n",
3007 			CVM_CAST64(status));
3008 		resp->timestamp = 0;
3009 	}
3010 
3011 	octeon_swap_8B_data(&resp->timestamp, 1);
3012 
3013 	if (unlikely((skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS) != 0)) {
3014 		struct skb_shared_hwtstamps ts;
3015 		u64 ns = resp->timestamp;
3016 
3017 		netif_info(lio, tx_done, lio->netdev,
3018 			   "Got resulting SKBTX_HW_TSTAMP skb=%p ns=%016llu\n",
3019 			   skb, (unsigned long long)ns);
3020 		ts.hwtstamp = ns_to_ktime(ns + lio->ptp_adjust);
3021 		skb_tstamp_tx(skb, &ts);
3022 	}
3023 
3024 	octeon_free_soft_command(oct, sc);
3025 	tx_buffer_free(skb);
3026 }
3027 
3028 /* \brief Send a data packet that will be timestamped
3029  * @param oct octeon device
3030  * @param ndata pointer to network data
3031  * @param finfo pointer to private network data
3032  */
3033 static inline int send_nic_timestamp_pkt(struct octeon_device *oct,
3034 					 struct octnic_data_pkt *ndata,
3035 					 struct octnet_buf_free_info *finfo)
3036 {
3037 	int retval;
3038 	struct octeon_soft_command *sc;
3039 	struct lio *lio;
3040 	int ring_doorbell;
3041 	u32 len;
3042 
3043 	lio = finfo->lio;
3044 
3045 	sc = octeon_alloc_soft_command_resp(oct, &ndata->cmd,
3046 					    sizeof(struct oct_timestamp_resp));
3047 	finfo->sc = sc;
3048 
3049 	if (!sc) {
3050 		dev_err(&oct->pci_dev->dev, "No memory for timestamped data packet\n");
3051 		return IQ_SEND_FAILED;
3052 	}
3053 
3054 	if (ndata->reqtype == REQTYPE_NORESP_NET)
3055 		ndata->reqtype = REQTYPE_RESP_NET;
3056 	else if (ndata->reqtype == REQTYPE_NORESP_NET_SG)
3057 		ndata->reqtype = REQTYPE_RESP_NET_SG;
3058 
3059 	sc->callback = handle_timestamp;
3060 	sc->callback_arg = finfo->skb;
3061 	sc->iq_no = ndata->q_no;
3062 
3063 	if (OCTEON_CN23XX_PF(oct))
3064 		len = (u32)((struct octeon_instr_ih3 *)
3065 			    (&sc->cmd.cmd3.ih3))->dlengsz;
3066 	else
3067 		len = (u32)((struct octeon_instr_ih2 *)
3068 			    (&sc->cmd.cmd2.ih2))->dlengsz;
3069 
3070 	ring_doorbell = 1;
3071 
3072 	retval = octeon_send_command(oct, sc->iq_no, ring_doorbell, &sc->cmd,
3073 				     sc, len, ndata->reqtype);
3074 
3075 	if (retval == IQ_SEND_FAILED) {
3076 		dev_err(&oct->pci_dev->dev, "timestamp data packet failed status: %x\n",
3077 			retval);
3078 		octeon_free_soft_command(oct, sc);
3079 	} else {
3080 		netif_info(lio, tx_queued, lio->netdev, "Queued timestamp packet\n");
3081 	}
3082 
3083 	return retval;
3084 }
3085 
3086 /** \brief Transmit networks packets to the Octeon interface
3087  * @param skbuff   skbuff struct to be passed to network layer.
3088  * @param netdev    pointer to network device
3089  * @returns whether the packet was transmitted to the device okay or not
3090  *             (NETDEV_TX_OK or NETDEV_TX_BUSY)
3091  */
3092 static int liquidio_xmit(struct sk_buff *skb, struct net_device *netdev)
3093 {
3094 	struct lio *lio;
3095 	struct octnet_buf_free_info *finfo;
3096 	union octnic_cmd_setup cmdsetup;
3097 	struct octnic_data_pkt ndata;
3098 	struct octeon_device *oct;
3099 	struct oct_iq_stats *stats;
3100 	struct octeon_instr_irh *irh;
3101 	union tx_info *tx_info;
3102 	int status = 0;
3103 	int q_idx = 0, iq_no = 0;
3104 	int j;
3105 	u64 dptr = 0;
3106 	u32 tag = 0;
3107 
3108 	lio = GET_LIO(netdev);
3109 	oct = lio->oct_dev;
3110 
3111 	if (netif_is_multiqueue(netdev)) {
3112 		q_idx = skb->queue_mapping;
3113 		q_idx = (q_idx % (lio->linfo.num_txpciq));
3114 		tag = q_idx;
3115 		iq_no = lio->linfo.txpciq[q_idx].s.q_no;
3116 	} else {
3117 		iq_no = lio->txq;
3118 	}
3119 
3120 	stats = &oct->instr_queue[iq_no]->stats;
3121 
3122 	/* Check for all conditions in which the current packet cannot be
3123 	 * transmitted.
3124 	 */
3125 	if (!(atomic_read(&lio->ifstate) & LIO_IFSTATE_RUNNING) ||
3126 	    (!lio->linfo.link.s.link_up) ||
3127 	    (skb->len <= 0)) {
3128 		netif_info(lio, tx_err, lio->netdev,
3129 			   "Transmit failed link_status : %d\n",
3130 			   lio->linfo.link.s.link_up);
3131 		goto lio_xmit_failed;
3132 	}
3133 
3134 	/* Use space in skb->cb to store info used to unmap and
3135 	 * free the buffers.
3136 	 */
3137 	finfo = (struct octnet_buf_free_info *)skb->cb;
3138 	finfo->lio = lio;
3139 	finfo->skb = skb;
3140 	finfo->sc = NULL;
3141 
3142 	/* Prepare the attributes for the data to be passed to OSI. */
3143 	memset(&ndata, 0, sizeof(struct octnic_data_pkt));
3144 
3145 	ndata.buf = (void *)finfo;
3146 
3147 	ndata.q_no = iq_no;
3148 
3149 	if (netif_is_multiqueue(netdev)) {
3150 		if (octnet_iq_is_full(oct, ndata.q_no)) {
3151 			/* defer sending if queue is full */
3152 			netif_info(lio, tx_err, lio->netdev, "Transmit failed iq:%d full\n",
3153 				   ndata.q_no);
3154 			stats->tx_iq_busy++;
3155 			return NETDEV_TX_BUSY;
3156 		}
3157 	} else {
3158 		if (octnet_iq_is_full(oct, lio->txq)) {
3159 			/* defer sending if queue is full */
3160 			stats->tx_iq_busy++;
3161 			netif_info(lio, tx_err, lio->netdev, "Transmit failed iq:%d full\n",
3162 				   lio->txq);
3163 			return NETDEV_TX_BUSY;
3164 		}
3165 	}
3166 	/* pr_info(" XMIT - valid Qs: %d, 1st Q no: %d, cpu:  %d, q_no:%d\n",
3167 	 *	lio->linfo.num_txpciq, lio->txq, cpu, ndata.q_no);
3168 	 */
3169 
3170 	ndata.datasize = skb->len;
3171 
3172 	cmdsetup.u64 = 0;
3173 	cmdsetup.s.iq_no = iq_no;
3174 
3175 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
3176 		if (skb->encapsulation) {
3177 			cmdsetup.s.tnl_csum = 1;
3178 			stats->tx_vxlan++;
3179 		} else {
3180 			cmdsetup.s.transport_csum = 1;
3181 		}
3182 	}
3183 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
3184 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
3185 		cmdsetup.s.timestamp = 1;
3186 	}
3187 
3188 	if (skb_shinfo(skb)->nr_frags == 0) {
3189 		cmdsetup.s.u.datasize = skb->len;
3190 		octnet_prepare_pci_cmd(oct, &ndata.cmd, &cmdsetup, tag);
3191 
3192 		/* Offload checksum calculation for TCP/UDP packets */
3193 		dptr = dma_map_single(&oct->pci_dev->dev,
3194 				      skb->data,
3195 				      skb->len,
3196 				      DMA_TO_DEVICE);
3197 		if (dma_mapping_error(&oct->pci_dev->dev, dptr)) {
3198 			dev_err(&oct->pci_dev->dev, "%s DMA mapping error 1\n",
3199 				__func__);
3200 			return NETDEV_TX_BUSY;
3201 		}
3202 
3203 		if (OCTEON_CN23XX_PF(oct))
3204 			ndata.cmd.cmd3.dptr = dptr;
3205 		else
3206 			ndata.cmd.cmd2.dptr = dptr;
3207 		finfo->dptr = dptr;
3208 		ndata.reqtype = REQTYPE_NORESP_NET;
3209 
3210 	} else {
3211 		int i, frags;
3212 		struct skb_frag_struct *frag;
3213 		struct octnic_gather *g;
3214 
3215 		spin_lock(&lio->glist_lock[q_idx]);
3216 		g = (struct octnic_gather *)
3217 			list_delete_head(&lio->glist[q_idx]);
3218 		spin_unlock(&lio->glist_lock[q_idx]);
3219 
3220 		if (!g) {
3221 			netif_info(lio, tx_err, lio->netdev,
3222 				   "Transmit scatter gather: glist null!\n");
3223 			goto lio_xmit_failed;
3224 		}
3225 
3226 		cmdsetup.s.gather = 1;
3227 		cmdsetup.s.u.gatherptrs = (skb_shinfo(skb)->nr_frags + 1);
3228 		octnet_prepare_pci_cmd(oct, &ndata.cmd, &cmdsetup, tag);
3229 
3230 		memset(g->sg, 0, g->sg_size);
3231 
3232 		g->sg[0].ptr[0] = dma_map_single(&oct->pci_dev->dev,
3233 						 skb->data,
3234 						 (skb->len - skb->data_len),
3235 						 DMA_TO_DEVICE);
3236 		if (dma_mapping_error(&oct->pci_dev->dev, g->sg[0].ptr[0])) {
3237 			dev_err(&oct->pci_dev->dev, "%s DMA mapping error 2\n",
3238 				__func__);
3239 			return NETDEV_TX_BUSY;
3240 		}
3241 		add_sg_size(&g->sg[0], (skb->len - skb->data_len), 0);
3242 
3243 		frags = skb_shinfo(skb)->nr_frags;
3244 		i = 1;
3245 		while (frags--) {
3246 			frag = &skb_shinfo(skb)->frags[i - 1];
3247 
3248 			g->sg[(i >> 2)].ptr[(i & 3)] =
3249 				dma_map_page(&oct->pci_dev->dev,
3250 					     frag->page.p,
3251 					     frag->page_offset,
3252 					     frag->size,
3253 					     DMA_TO_DEVICE);
3254 
3255 			if (dma_mapping_error(&oct->pci_dev->dev,
3256 					      g->sg[i >> 2].ptr[i & 3])) {
3257 				dma_unmap_single(&oct->pci_dev->dev,
3258 						 g->sg[0].ptr[0],
3259 						 skb->len - skb->data_len,
3260 						 DMA_TO_DEVICE);
3261 				for (j = 1; j < i; j++) {
3262 					frag = &skb_shinfo(skb)->frags[j - 1];
3263 					dma_unmap_page(&oct->pci_dev->dev,
3264 						       g->sg[j >> 2].ptr[j & 3],
3265 						       frag->size,
3266 						       DMA_TO_DEVICE);
3267 				}
3268 				dev_err(&oct->pci_dev->dev, "%s DMA mapping error 3\n",
3269 					__func__);
3270 				return NETDEV_TX_BUSY;
3271 			}
3272 
3273 			add_sg_size(&g->sg[(i >> 2)], frag->size, (i & 3));
3274 			i++;
3275 		}
3276 
3277 		dma_sync_single_for_device(&oct->pci_dev->dev, g->sg_dma_ptr,
3278 					   g->sg_size, DMA_TO_DEVICE);
3279 		dptr = g->sg_dma_ptr;
3280 
3281 		if (OCTEON_CN23XX_PF(oct))
3282 			ndata.cmd.cmd3.dptr = dptr;
3283 		else
3284 			ndata.cmd.cmd2.dptr = dptr;
3285 		finfo->dptr = dptr;
3286 		finfo->g = g;
3287 
3288 		ndata.reqtype = REQTYPE_NORESP_NET_SG;
3289 	}
3290 
3291 	if (OCTEON_CN23XX_PF(oct)) {
3292 		irh = (struct octeon_instr_irh *)&ndata.cmd.cmd3.irh;
3293 		tx_info = (union tx_info *)&ndata.cmd.cmd3.ossp[0];
3294 	} else {
3295 		irh = (struct octeon_instr_irh *)&ndata.cmd.cmd2.irh;
3296 		tx_info = (union tx_info *)&ndata.cmd.cmd2.ossp[0];
3297 	}
3298 
3299 	if (skb_shinfo(skb)->gso_size) {
3300 		tx_info->s.gso_size = skb_shinfo(skb)->gso_size;
3301 		tx_info->s.gso_segs = skb_shinfo(skb)->gso_segs;
3302 		stats->tx_gso++;
3303 	}
3304 
3305 	/* HW insert VLAN tag */
3306 	if (skb_vlan_tag_present(skb)) {
3307 		irh->priority = skb_vlan_tag_get(skb) >> 13;
3308 		irh->vlan = skb_vlan_tag_get(skb) & 0xfff;
3309 	}
3310 
3311 	if (unlikely(cmdsetup.s.timestamp))
3312 		status = send_nic_timestamp_pkt(oct, &ndata, finfo);
3313 	else
3314 		status = octnet_send_nic_data_pkt(oct, &ndata);
3315 	if (status == IQ_SEND_FAILED)
3316 		goto lio_xmit_failed;
3317 
3318 	netif_info(lio, tx_queued, lio->netdev, "Transmit queued successfully\n");
3319 
3320 	if (status == IQ_SEND_STOP)
3321 		stop_q(lio->netdev, q_idx);
3322 
3323 	netif_trans_update(netdev);
3324 
3325 	if (skb_shinfo(skb)->gso_size)
3326 		stats->tx_done += skb_shinfo(skb)->gso_segs;
3327 	else
3328 		stats->tx_done++;
3329 	stats->tx_tot_bytes += skb->len;
3330 
3331 	return NETDEV_TX_OK;
3332 
3333 lio_xmit_failed:
3334 	stats->tx_dropped++;
3335 	netif_info(lio, tx_err, lio->netdev, "IQ%d Transmit dropped:%llu\n",
3336 		   iq_no, stats->tx_dropped);
3337 	if (dptr)
3338 		dma_unmap_single(&oct->pci_dev->dev, dptr,
3339 				 ndata.datasize, DMA_TO_DEVICE);
3340 	tx_buffer_free(skb);
3341 	return NETDEV_TX_OK;
3342 }
3343 
3344 /** \brief Network device Tx timeout
3345  * @param netdev    pointer to network device
3346  */
3347 static void liquidio_tx_timeout(struct net_device *netdev)
3348 {
3349 	struct lio *lio;
3350 
3351 	lio = GET_LIO(netdev);
3352 
3353 	netif_info(lio, tx_err, lio->netdev,
3354 		   "Transmit timeout tx_dropped:%ld, waking up queues now!!\n",
3355 		   netdev->stats.tx_dropped);
3356 	netif_trans_update(netdev);
3357 	txqs_wake(netdev);
3358 }
3359 
3360 static int liquidio_vlan_rx_add_vid(struct net_device *netdev,
3361 				    __be16 proto __attribute__((unused)),
3362 				    u16 vid)
3363 {
3364 	struct lio *lio = GET_LIO(netdev);
3365 	struct octeon_device *oct = lio->oct_dev;
3366 	struct octnic_ctrl_pkt nctrl;
3367 	int ret = 0;
3368 
3369 	memset(&nctrl, 0, sizeof(struct octnic_ctrl_pkt));
3370 
3371 	nctrl.ncmd.u64 = 0;
3372 	nctrl.ncmd.s.cmd = OCTNET_CMD_ADD_VLAN_FILTER;
3373 	nctrl.ncmd.s.param1 = vid;
3374 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
3375 	nctrl.wait_time = 100;
3376 	nctrl.netpndev = (u64)netdev;
3377 	nctrl.cb_fn = liquidio_link_ctrl_cmd_completion;
3378 
3379 	ret = octnet_send_nic_ctrl_pkt(lio->oct_dev, &nctrl);
3380 	if (ret < 0) {
3381 		dev_err(&oct->pci_dev->dev, "Add VLAN filter failed in core (ret: 0x%x)\n",
3382 			ret);
3383 	}
3384 
3385 	return ret;
3386 }
3387 
3388 static int liquidio_vlan_rx_kill_vid(struct net_device *netdev,
3389 				     __be16 proto __attribute__((unused)),
3390 				     u16 vid)
3391 {
3392 	struct lio *lio = GET_LIO(netdev);
3393 	struct octeon_device *oct = lio->oct_dev;
3394 	struct octnic_ctrl_pkt nctrl;
3395 	int ret = 0;
3396 
3397 	memset(&nctrl, 0, sizeof(struct octnic_ctrl_pkt));
3398 
3399 	nctrl.ncmd.u64 = 0;
3400 	nctrl.ncmd.s.cmd = OCTNET_CMD_DEL_VLAN_FILTER;
3401 	nctrl.ncmd.s.param1 = vid;
3402 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
3403 	nctrl.wait_time = 100;
3404 	nctrl.netpndev = (u64)netdev;
3405 	nctrl.cb_fn = liquidio_link_ctrl_cmd_completion;
3406 
3407 	ret = octnet_send_nic_ctrl_pkt(lio->oct_dev, &nctrl);
3408 	if (ret < 0) {
3409 		dev_err(&oct->pci_dev->dev, "Add VLAN filter failed in core (ret: 0x%x)\n",
3410 			ret);
3411 	}
3412 	return ret;
3413 }
3414 
3415 /** Sending command to enable/disable RX checksum offload
3416  * @param netdev                pointer to network device
3417  * @param command               OCTNET_CMD_TNL_RX_CSUM_CTL
3418  * @param rx_cmd_bit            OCTNET_CMD_RXCSUM_ENABLE/
3419  *                              OCTNET_CMD_RXCSUM_DISABLE
3420  * @returns                     SUCCESS or FAILURE
3421  */
3422 static int liquidio_set_rxcsum_command(struct net_device *netdev, int command,
3423 				       u8 rx_cmd)
3424 {
3425 	struct lio *lio = GET_LIO(netdev);
3426 	struct octeon_device *oct = lio->oct_dev;
3427 	struct octnic_ctrl_pkt nctrl;
3428 	int ret = 0;
3429 
3430 	nctrl.ncmd.u64 = 0;
3431 	nctrl.ncmd.s.cmd = command;
3432 	nctrl.ncmd.s.param1 = rx_cmd;
3433 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
3434 	nctrl.wait_time = 100;
3435 	nctrl.netpndev = (u64)netdev;
3436 	nctrl.cb_fn = liquidio_link_ctrl_cmd_completion;
3437 
3438 	ret = octnet_send_nic_ctrl_pkt(lio->oct_dev, &nctrl);
3439 	if (ret < 0) {
3440 		dev_err(&oct->pci_dev->dev,
3441 			"DEVFLAGS RXCSUM change failed in core(ret:0x%x)\n",
3442 			ret);
3443 	}
3444 	return ret;
3445 }
3446 
3447 /** Sending command to add/delete VxLAN UDP port to firmware
3448  * @param netdev                pointer to network device
3449  * @param command               OCTNET_CMD_VXLAN_PORT_CONFIG
3450  * @param vxlan_port            VxLAN port to be added or deleted
3451  * @param vxlan_cmd_bit         OCTNET_CMD_VXLAN_PORT_ADD,
3452  *                              OCTNET_CMD_VXLAN_PORT_DEL
3453  * @returns                     SUCCESS or FAILURE
3454  */
3455 static int liquidio_vxlan_port_command(struct net_device *netdev, int command,
3456 				       u16 vxlan_port, u8 vxlan_cmd_bit)
3457 {
3458 	struct lio *lio = GET_LIO(netdev);
3459 	struct octeon_device *oct = lio->oct_dev;
3460 	struct octnic_ctrl_pkt nctrl;
3461 	int ret = 0;
3462 
3463 	nctrl.ncmd.u64 = 0;
3464 	nctrl.ncmd.s.cmd = command;
3465 	nctrl.ncmd.s.more = vxlan_cmd_bit;
3466 	nctrl.ncmd.s.param1 = vxlan_port;
3467 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
3468 	nctrl.wait_time = 100;
3469 	nctrl.netpndev = (u64)netdev;
3470 	nctrl.cb_fn = liquidio_link_ctrl_cmd_completion;
3471 
3472 	ret = octnet_send_nic_ctrl_pkt(lio->oct_dev, &nctrl);
3473 	if (ret < 0) {
3474 		dev_err(&oct->pci_dev->dev,
3475 			"VxLAN port add/delete failed in core (ret:0x%x)\n",
3476 			ret);
3477 	}
3478 	return ret;
3479 }
3480 
3481 /** \brief Net device fix features
3482  * @param netdev  pointer to network device
3483  * @param request features requested
3484  * @returns updated features list
3485  */
3486 static netdev_features_t liquidio_fix_features(struct net_device *netdev,
3487 					       netdev_features_t request)
3488 {
3489 	struct lio *lio = netdev_priv(netdev);
3490 
3491 	if ((request & NETIF_F_RXCSUM) &&
3492 	    !(lio->dev_capability & NETIF_F_RXCSUM))
3493 		request &= ~NETIF_F_RXCSUM;
3494 
3495 	if ((request & NETIF_F_HW_CSUM) &&
3496 	    !(lio->dev_capability & NETIF_F_HW_CSUM))
3497 		request &= ~NETIF_F_HW_CSUM;
3498 
3499 	if ((request & NETIF_F_TSO) && !(lio->dev_capability & NETIF_F_TSO))
3500 		request &= ~NETIF_F_TSO;
3501 
3502 	if ((request & NETIF_F_TSO6) && !(lio->dev_capability & NETIF_F_TSO6))
3503 		request &= ~NETIF_F_TSO6;
3504 
3505 	if ((request & NETIF_F_LRO) && !(lio->dev_capability & NETIF_F_LRO))
3506 		request &= ~NETIF_F_LRO;
3507 
3508 	/*Disable LRO if RXCSUM is off */
3509 	if (!(request & NETIF_F_RXCSUM) && (netdev->features & NETIF_F_LRO) &&
3510 	    (lio->dev_capability & NETIF_F_LRO))
3511 		request &= ~NETIF_F_LRO;
3512 
3513 	return request;
3514 }
3515 
3516 /** \brief Net device set features
3517  * @param netdev  pointer to network device
3518  * @param features features to enable/disable
3519  */
3520 static int liquidio_set_features(struct net_device *netdev,
3521 				 netdev_features_t features)
3522 {
3523 	struct lio *lio = netdev_priv(netdev);
3524 
3525 	if (!((netdev->features ^ features) & NETIF_F_LRO))
3526 		return 0;
3527 
3528 	if ((features & NETIF_F_LRO) && (lio->dev_capability & NETIF_F_LRO))
3529 		liquidio_set_feature(netdev, OCTNET_CMD_LRO_ENABLE,
3530 				     OCTNIC_LROIPV4 | OCTNIC_LROIPV6);
3531 	else if (!(features & NETIF_F_LRO) &&
3532 		 (lio->dev_capability & NETIF_F_LRO))
3533 		liquidio_set_feature(netdev, OCTNET_CMD_LRO_DISABLE,
3534 				     OCTNIC_LROIPV4 | OCTNIC_LROIPV6);
3535 
3536 	/* Sending command to firmware to enable/disable RX checksum
3537 	 * offload settings using ethtool
3538 	 */
3539 	if (!(netdev->features & NETIF_F_RXCSUM) &&
3540 	    (lio->enc_dev_capability & NETIF_F_RXCSUM) &&
3541 	    (features & NETIF_F_RXCSUM))
3542 		liquidio_set_rxcsum_command(netdev,
3543 					    OCTNET_CMD_TNL_RX_CSUM_CTL,
3544 					    OCTNET_CMD_RXCSUM_ENABLE);
3545 	else if ((netdev->features & NETIF_F_RXCSUM) &&
3546 		 (lio->enc_dev_capability & NETIF_F_RXCSUM) &&
3547 		 !(features & NETIF_F_RXCSUM))
3548 		liquidio_set_rxcsum_command(netdev, OCTNET_CMD_TNL_RX_CSUM_CTL,
3549 					    OCTNET_CMD_RXCSUM_DISABLE);
3550 
3551 	return 0;
3552 }
3553 
3554 static void liquidio_add_vxlan_port(struct net_device *netdev,
3555 				    struct udp_tunnel_info *ti)
3556 {
3557 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
3558 		return;
3559 
3560 	liquidio_vxlan_port_command(netdev,
3561 				    OCTNET_CMD_VXLAN_PORT_CONFIG,
3562 				    htons(ti->port),
3563 				    OCTNET_CMD_VXLAN_PORT_ADD);
3564 }
3565 
3566 static void liquidio_del_vxlan_port(struct net_device *netdev,
3567 				    struct udp_tunnel_info *ti)
3568 {
3569 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
3570 		return;
3571 
3572 	liquidio_vxlan_port_command(netdev,
3573 				    OCTNET_CMD_VXLAN_PORT_CONFIG,
3574 				    htons(ti->port),
3575 				    OCTNET_CMD_VXLAN_PORT_DEL);
3576 }
3577 
3578 static int __liquidio_set_vf_mac(struct net_device *netdev, int vfidx,
3579 				 u8 *mac, bool is_admin_assigned)
3580 {
3581 	struct lio *lio = GET_LIO(netdev);
3582 	struct octeon_device *oct = lio->oct_dev;
3583 	struct octnic_ctrl_pkt nctrl;
3584 
3585 	if (!is_valid_ether_addr(mac))
3586 		return -EINVAL;
3587 
3588 	if (vfidx < 0 || vfidx >= oct->sriov_info.max_vfs)
3589 		return -EINVAL;
3590 
3591 	memset(&nctrl, 0, sizeof(struct octnic_ctrl_pkt));
3592 
3593 	nctrl.ncmd.u64 = 0;
3594 	nctrl.ncmd.s.cmd = OCTNET_CMD_CHANGE_MACADDR;
3595 	/* vfidx is 0 based, but vf_num (param1) is 1 based */
3596 	nctrl.ncmd.s.param1 = vfidx + 1;
3597 	nctrl.ncmd.s.param2 = (is_admin_assigned ? 1 : 0);
3598 	nctrl.ncmd.s.more = 1;
3599 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
3600 	nctrl.cb_fn = 0;
3601 	nctrl.wait_time = LIO_CMD_WAIT_TM;
3602 
3603 	nctrl.udd[0] = 0;
3604 	/* The MAC Address is presented in network byte order. */
3605 	ether_addr_copy((u8 *)&nctrl.udd[0] + 2, mac);
3606 
3607 	oct->sriov_info.vf_macaddr[vfidx] = nctrl.udd[0];
3608 
3609 	octnet_send_nic_ctrl_pkt(oct, &nctrl);
3610 
3611 	return 0;
3612 }
3613 
3614 static int liquidio_set_vf_mac(struct net_device *netdev, int vfidx, u8 *mac)
3615 {
3616 	struct lio *lio = GET_LIO(netdev);
3617 	struct octeon_device *oct = lio->oct_dev;
3618 	int retval;
3619 
3620 	retval = __liquidio_set_vf_mac(netdev, vfidx, mac, true);
3621 	if (!retval)
3622 		cn23xx_tell_vf_its_macaddr_changed(oct, vfidx, mac);
3623 
3624 	return retval;
3625 }
3626 
3627 static int liquidio_set_vf_vlan(struct net_device *netdev, int vfidx,
3628 				u16 vlan, u8 qos, __be16 vlan_proto)
3629 {
3630 	struct lio *lio = GET_LIO(netdev);
3631 	struct octeon_device *oct = lio->oct_dev;
3632 	struct octnic_ctrl_pkt nctrl;
3633 	u16 vlantci;
3634 
3635 	if (vfidx < 0 || vfidx >= oct->sriov_info.num_vfs_alloced)
3636 		return -EINVAL;
3637 
3638 	if (vlan_proto != htons(ETH_P_8021Q))
3639 		return -EPROTONOSUPPORT;
3640 
3641 	if (vlan >= VLAN_N_VID || qos > 7)
3642 		return -EINVAL;
3643 
3644 	if (vlan)
3645 		vlantci = vlan | (u16)qos << VLAN_PRIO_SHIFT;
3646 	else
3647 		vlantci = 0;
3648 
3649 	if (oct->sriov_info.vf_vlantci[vfidx] == vlantci)
3650 		return 0;
3651 
3652 	memset(&nctrl, 0, sizeof(struct octnic_ctrl_pkt));
3653 
3654 	if (vlan)
3655 		nctrl.ncmd.s.cmd = OCTNET_CMD_ADD_VLAN_FILTER;
3656 	else
3657 		nctrl.ncmd.s.cmd = OCTNET_CMD_DEL_VLAN_FILTER;
3658 
3659 	nctrl.ncmd.s.param1 = vlantci;
3660 	nctrl.ncmd.s.param2 =
3661 	    vfidx + 1; /* vfidx is 0 based, but vf_num (param2) is 1 based */
3662 	nctrl.ncmd.s.more = 0;
3663 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
3664 	nctrl.cb_fn = 0;
3665 	nctrl.wait_time = LIO_CMD_WAIT_TM;
3666 
3667 	octnet_send_nic_ctrl_pkt(oct, &nctrl);
3668 
3669 	oct->sriov_info.vf_vlantci[vfidx] = vlantci;
3670 
3671 	return 0;
3672 }
3673 
3674 static int liquidio_get_vf_config(struct net_device *netdev, int vfidx,
3675 				  struct ifla_vf_info *ivi)
3676 {
3677 	struct lio *lio = GET_LIO(netdev);
3678 	struct octeon_device *oct = lio->oct_dev;
3679 	u8 *macaddr;
3680 
3681 	if (vfidx < 0 || vfidx >= oct->sriov_info.num_vfs_alloced)
3682 		return -EINVAL;
3683 
3684 	ivi->vf = vfidx;
3685 	macaddr = 2 + (u8 *)&oct->sriov_info.vf_macaddr[vfidx];
3686 	ether_addr_copy(&ivi->mac[0], macaddr);
3687 	ivi->vlan = oct->sriov_info.vf_vlantci[vfidx] & VLAN_VID_MASK;
3688 	ivi->qos = oct->sriov_info.vf_vlantci[vfidx] >> VLAN_PRIO_SHIFT;
3689 	ivi->linkstate = oct->sriov_info.vf_linkstate[vfidx];
3690 	return 0;
3691 }
3692 
3693 static int liquidio_set_vf_link_state(struct net_device *netdev, int vfidx,
3694 				      int linkstate)
3695 {
3696 	struct lio *lio = GET_LIO(netdev);
3697 	struct octeon_device *oct = lio->oct_dev;
3698 	struct octnic_ctrl_pkt nctrl;
3699 
3700 	if (vfidx < 0 || vfidx >= oct->sriov_info.num_vfs_alloced)
3701 		return -EINVAL;
3702 
3703 	if (oct->sriov_info.vf_linkstate[vfidx] == linkstate)
3704 		return 0;
3705 
3706 	memset(&nctrl, 0, sizeof(struct octnic_ctrl_pkt));
3707 	nctrl.ncmd.s.cmd = OCTNET_CMD_SET_VF_LINKSTATE;
3708 	nctrl.ncmd.s.param1 =
3709 	    vfidx + 1; /* vfidx is 0 based, but vf_num (param1) is 1 based */
3710 	nctrl.ncmd.s.param2 = linkstate;
3711 	nctrl.ncmd.s.more = 0;
3712 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
3713 	nctrl.cb_fn = 0;
3714 	nctrl.wait_time = LIO_CMD_WAIT_TM;
3715 
3716 	octnet_send_nic_ctrl_pkt(oct, &nctrl);
3717 
3718 	oct->sriov_info.vf_linkstate[vfidx] = linkstate;
3719 
3720 	return 0;
3721 }
3722 
3723 static const struct net_device_ops lionetdevops = {
3724 	.ndo_open		= liquidio_open,
3725 	.ndo_stop		= liquidio_stop,
3726 	.ndo_start_xmit		= liquidio_xmit,
3727 	.ndo_get_stats		= liquidio_get_stats,
3728 	.ndo_set_mac_address	= liquidio_set_mac,
3729 	.ndo_set_rx_mode	= liquidio_set_mcast_list,
3730 	.ndo_tx_timeout		= liquidio_tx_timeout,
3731 
3732 	.ndo_vlan_rx_add_vid    = liquidio_vlan_rx_add_vid,
3733 	.ndo_vlan_rx_kill_vid   = liquidio_vlan_rx_kill_vid,
3734 	.ndo_change_mtu		= liquidio_change_mtu,
3735 	.ndo_do_ioctl		= liquidio_ioctl,
3736 	.ndo_fix_features	= liquidio_fix_features,
3737 	.ndo_set_features	= liquidio_set_features,
3738 	.ndo_udp_tunnel_add	= liquidio_add_vxlan_port,
3739 	.ndo_udp_tunnel_del	= liquidio_del_vxlan_port,
3740 	.ndo_set_vf_mac		= liquidio_set_vf_mac,
3741 	.ndo_set_vf_vlan	= liquidio_set_vf_vlan,
3742 	.ndo_get_vf_config	= liquidio_get_vf_config,
3743 	.ndo_set_vf_link_state  = liquidio_set_vf_link_state,
3744 	.ndo_select_queue	= select_q
3745 };
3746 
3747 /** \brief Entry point for the liquidio module
3748  */
3749 static int __init liquidio_init(void)
3750 {
3751 	int i;
3752 	struct handshake *hs;
3753 
3754 	init_completion(&first_stage);
3755 
3756 	octeon_init_device_list(OCTEON_CONFIG_TYPE_DEFAULT);
3757 
3758 	if (liquidio_init_pci())
3759 		return -EINVAL;
3760 
3761 	wait_for_completion_timeout(&first_stage, msecs_to_jiffies(1000));
3762 
3763 	for (i = 0; i < MAX_OCTEON_DEVICES; i++) {
3764 		hs = &handshake[i];
3765 		if (hs->pci_dev) {
3766 			wait_for_completion(&hs->init);
3767 			if (!hs->init_ok) {
3768 				/* init handshake failed */
3769 				dev_err(&hs->pci_dev->dev,
3770 					"Failed to init device\n");
3771 				liquidio_deinit_pci();
3772 				return -EIO;
3773 			}
3774 		}
3775 	}
3776 
3777 	for (i = 0; i < MAX_OCTEON_DEVICES; i++) {
3778 		hs = &handshake[i];
3779 		if (hs->pci_dev) {
3780 			wait_for_completion_timeout(&hs->started,
3781 						    msecs_to_jiffies(30000));
3782 			if (!hs->started_ok) {
3783 				/* starter handshake failed */
3784 				dev_err(&hs->pci_dev->dev,
3785 					"Firmware failed to start\n");
3786 				liquidio_deinit_pci();
3787 				return -EIO;
3788 			}
3789 		}
3790 	}
3791 
3792 	return 0;
3793 }
3794 
3795 static int lio_nic_info(struct octeon_recv_info *recv_info, void *buf)
3796 {
3797 	struct octeon_device *oct = (struct octeon_device *)buf;
3798 	struct octeon_recv_pkt *recv_pkt = recv_info->recv_pkt;
3799 	int gmxport = 0;
3800 	union oct_link_status *ls;
3801 	int i;
3802 
3803 	if (recv_pkt->buffer_size[0] != sizeof(*ls)) {
3804 		dev_err(&oct->pci_dev->dev, "Malformed NIC_INFO, len=%d, ifidx=%d\n",
3805 			recv_pkt->buffer_size[0],
3806 			recv_pkt->rh.r_nic_info.gmxport);
3807 		goto nic_info_err;
3808 	}
3809 
3810 	gmxport = recv_pkt->rh.r_nic_info.gmxport;
3811 	ls = (union oct_link_status *)get_rbd(recv_pkt->buffer_ptr[0]);
3812 
3813 	octeon_swap_8B_data((u64 *)ls, (sizeof(union oct_link_status)) >> 3);
3814 	for (i = 0; i < oct->ifcount; i++) {
3815 		if (oct->props[i].gmxport == gmxport) {
3816 			update_link_status(oct->props[i].netdev, ls);
3817 			break;
3818 		}
3819 	}
3820 
3821 nic_info_err:
3822 	for (i = 0; i < recv_pkt->buffer_count; i++)
3823 		recv_buffer_free(recv_pkt->buffer_ptr[i]);
3824 	octeon_free_recv_info(recv_info);
3825 	return 0;
3826 }
3827 
3828 /**
3829  * \brief Setup network interfaces
3830  * @param octeon_dev  octeon device
3831  *
3832  * Called during init time for each device. It assumes the NIC
3833  * is already up and running.  The link information for each
3834  * interface is passed in link_info.
3835  */
3836 static int setup_nic_devices(struct octeon_device *octeon_dev)
3837 {
3838 	struct lio *lio = NULL;
3839 	struct net_device *netdev;
3840 	u8 mac[6], i, j;
3841 	struct octeon_soft_command *sc;
3842 	struct liquidio_if_cfg_context *ctx;
3843 	struct liquidio_if_cfg_resp *resp;
3844 	struct octdev_props *props;
3845 	int retval, num_iqueues, num_oqueues;
3846 	union oct_nic_if_cfg if_cfg;
3847 	unsigned int base_queue;
3848 	unsigned int gmx_port_id;
3849 	u32 resp_size, ctx_size, data_size;
3850 	u32 ifidx_or_pfnum;
3851 	struct lio_version *vdata;
3852 
3853 	/* This is to handle link status changes */
3854 	octeon_register_dispatch_fn(octeon_dev, OPCODE_NIC,
3855 				    OPCODE_NIC_INFO,
3856 				    lio_nic_info, octeon_dev);
3857 
3858 	/* REQTYPE_RESP_NET and REQTYPE_SOFT_COMMAND do not have free functions.
3859 	 * They are handled directly.
3860 	 */
3861 	octeon_register_reqtype_free_fn(octeon_dev, REQTYPE_NORESP_NET,
3862 					free_netbuf);
3863 
3864 	octeon_register_reqtype_free_fn(octeon_dev, REQTYPE_NORESP_NET_SG,
3865 					free_netsgbuf);
3866 
3867 	octeon_register_reqtype_free_fn(octeon_dev, REQTYPE_RESP_NET_SG,
3868 					free_netsgbuf_with_resp);
3869 
3870 	for (i = 0; i < octeon_dev->ifcount; i++) {
3871 		resp_size = sizeof(struct liquidio_if_cfg_resp);
3872 		ctx_size = sizeof(struct liquidio_if_cfg_context);
3873 		data_size = sizeof(struct lio_version);
3874 		sc = (struct octeon_soft_command *)
3875 			octeon_alloc_soft_command(octeon_dev, data_size,
3876 						  resp_size, ctx_size);
3877 		resp = (struct liquidio_if_cfg_resp *)sc->virtrptr;
3878 		ctx  = (struct liquidio_if_cfg_context *)sc->ctxptr;
3879 		vdata = (struct lio_version *)sc->virtdptr;
3880 
3881 		*((u64 *)vdata) = 0;
3882 		vdata->major = cpu_to_be16(LIQUIDIO_BASE_MAJOR_VERSION);
3883 		vdata->minor = cpu_to_be16(LIQUIDIO_BASE_MINOR_VERSION);
3884 		vdata->micro = cpu_to_be16(LIQUIDIO_BASE_MICRO_VERSION);
3885 
3886 		if (OCTEON_CN23XX_PF(octeon_dev)) {
3887 			num_iqueues = octeon_dev->sriov_info.num_pf_rings;
3888 			num_oqueues = octeon_dev->sriov_info.num_pf_rings;
3889 			base_queue = octeon_dev->sriov_info.pf_srn;
3890 
3891 			gmx_port_id = octeon_dev->pf_num;
3892 			ifidx_or_pfnum = octeon_dev->pf_num;
3893 		} else {
3894 			num_iqueues = CFG_GET_NUM_TXQS_NIC_IF(
3895 						octeon_get_conf(octeon_dev), i);
3896 			num_oqueues = CFG_GET_NUM_RXQS_NIC_IF(
3897 						octeon_get_conf(octeon_dev), i);
3898 			base_queue = CFG_GET_BASE_QUE_NIC_IF(
3899 						octeon_get_conf(octeon_dev), i);
3900 			gmx_port_id = CFG_GET_GMXID_NIC_IF(
3901 						octeon_get_conf(octeon_dev), i);
3902 			ifidx_or_pfnum = i;
3903 		}
3904 
3905 		dev_dbg(&octeon_dev->pci_dev->dev,
3906 			"requesting config for interface %d, iqs %d, oqs %d\n",
3907 			ifidx_or_pfnum, num_iqueues, num_oqueues);
3908 		WRITE_ONCE(ctx->cond, 0);
3909 		ctx->octeon_id = lio_get_device_id(octeon_dev);
3910 		init_waitqueue_head(&ctx->wc);
3911 
3912 		if_cfg.u64 = 0;
3913 		if_cfg.s.num_iqueues = num_iqueues;
3914 		if_cfg.s.num_oqueues = num_oqueues;
3915 		if_cfg.s.base_queue = base_queue;
3916 		if_cfg.s.gmx_port_id = gmx_port_id;
3917 
3918 		sc->iq_no = 0;
3919 
3920 		octeon_prepare_soft_command(octeon_dev, sc, OPCODE_NIC,
3921 					    OPCODE_NIC_IF_CFG, 0,
3922 					    if_cfg.u64, 0);
3923 
3924 		sc->callback = if_cfg_callback;
3925 		sc->callback_arg = sc;
3926 		sc->wait_time = 3000;
3927 
3928 		retval = octeon_send_soft_command(octeon_dev, sc);
3929 		if (retval == IQ_SEND_FAILED) {
3930 			dev_err(&octeon_dev->pci_dev->dev,
3931 				"iq/oq config failed status: %x\n",
3932 				retval);
3933 			/* Soft instr is freed by driver in case of failure. */
3934 			goto setup_nic_dev_fail;
3935 		}
3936 
3937 		/* Sleep on a wait queue till the cond flag indicates that the
3938 		 * response arrived or timed-out.
3939 		 */
3940 		if (sleep_cond(&ctx->wc, &ctx->cond) == -EINTR) {
3941 			dev_err(&octeon_dev->pci_dev->dev, "Wait interrupted\n");
3942 			goto setup_nic_wait_intr;
3943 		}
3944 
3945 		retval = resp->status;
3946 		if (retval) {
3947 			dev_err(&octeon_dev->pci_dev->dev, "iq/oq config failed\n");
3948 			goto setup_nic_dev_fail;
3949 		}
3950 
3951 		octeon_swap_8B_data((u64 *)(&resp->cfg_info),
3952 				    (sizeof(struct liquidio_if_cfg_info)) >> 3);
3953 
3954 		num_iqueues = hweight64(resp->cfg_info.iqmask);
3955 		num_oqueues = hweight64(resp->cfg_info.oqmask);
3956 
3957 		if (!(num_iqueues) || !(num_oqueues)) {
3958 			dev_err(&octeon_dev->pci_dev->dev,
3959 				"Got bad iqueues (%016llx) or oqueues (%016llx) from firmware.\n",
3960 				resp->cfg_info.iqmask,
3961 				resp->cfg_info.oqmask);
3962 			goto setup_nic_dev_fail;
3963 		}
3964 		dev_dbg(&octeon_dev->pci_dev->dev,
3965 			"interface %d, iqmask %016llx, oqmask %016llx, numiqueues %d, numoqueues %d\n",
3966 			i, resp->cfg_info.iqmask, resp->cfg_info.oqmask,
3967 			num_iqueues, num_oqueues);
3968 		netdev = alloc_etherdev_mq(LIO_SIZE, num_iqueues);
3969 
3970 		if (!netdev) {
3971 			dev_err(&octeon_dev->pci_dev->dev, "Device allocation failed\n");
3972 			goto setup_nic_dev_fail;
3973 		}
3974 
3975 		SET_NETDEV_DEV(netdev, &octeon_dev->pci_dev->dev);
3976 
3977 		/* Associate the routines that will handle different
3978 		 * netdev tasks.
3979 		 */
3980 		netdev->netdev_ops = &lionetdevops;
3981 
3982 		lio = GET_LIO(netdev);
3983 
3984 		memset(lio, 0, sizeof(struct lio));
3985 
3986 		lio->ifidx = ifidx_or_pfnum;
3987 
3988 		props = &octeon_dev->props[i];
3989 		props->gmxport = resp->cfg_info.linfo.gmxport;
3990 		props->netdev = netdev;
3991 
3992 		lio->linfo.num_rxpciq = num_oqueues;
3993 		lio->linfo.num_txpciq = num_iqueues;
3994 		for (j = 0; j < num_oqueues; j++) {
3995 			lio->linfo.rxpciq[j].u64 =
3996 				resp->cfg_info.linfo.rxpciq[j].u64;
3997 		}
3998 		for (j = 0; j < num_iqueues; j++) {
3999 			lio->linfo.txpciq[j].u64 =
4000 				resp->cfg_info.linfo.txpciq[j].u64;
4001 		}
4002 		lio->linfo.hw_addr = resp->cfg_info.linfo.hw_addr;
4003 		lio->linfo.gmxport = resp->cfg_info.linfo.gmxport;
4004 		lio->linfo.link.u64 = resp->cfg_info.linfo.link.u64;
4005 
4006 		lio->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4007 
4008 		if (OCTEON_CN23XX_PF(octeon_dev) ||
4009 		    OCTEON_CN6XXX(octeon_dev)) {
4010 			lio->dev_capability = NETIF_F_HIGHDMA
4011 					      | NETIF_F_IP_CSUM
4012 					      | NETIF_F_IPV6_CSUM
4013 					      | NETIF_F_SG | NETIF_F_RXCSUM
4014 					      | NETIF_F_GRO
4015 					      | NETIF_F_TSO | NETIF_F_TSO6
4016 					      | NETIF_F_LRO;
4017 		}
4018 		netif_set_gso_max_size(netdev, OCTNIC_GSO_MAX_SIZE);
4019 
4020 		/*  Copy of transmit encapsulation capabilities:
4021 		 *  TSO, TSO6, Checksums for this device
4022 		 */
4023 		lio->enc_dev_capability = NETIF_F_IP_CSUM
4024 					  | NETIF_F_IPV6_CSUM
4025 					  | NETIF_F_GSO_UDP_TUNNEL
4026 					  | NETIF_F_HW_CSUM | NETIF_F_SG
4027 					  | NETIF_F_RXCSUM
4028 					  | NETIF_F_TSO | NETIF_F_TSO6
4029 					  | NETIF_F_LRO;
4030 
4031 		netdev->hw_enc_features = (lio->enc_dev_capability &
4032 					   ~NETIF_F_LRO);
4033 
4034 		lio->dev_capability |= NETIF_F_GSO_UDP_TUNNEL;
4035 
4036 		netdev->vlan_features = lio->dev_capability;
4037 		/* Add any unchangeable hw features */
4038 		lio->dev_capability |=  NETIF_F_HW_VLAN_CTAG_FILTER |
4039 					NETIF_F_HW_VLAN_CTAG_RX |
4040 					NETIF_F_HW_VLAN_CTAG_TX;
4041 
4042 		netdev->features = (lio->dev_capability & ~NETIF_F_LRO);
4043 
4044 		netdev->hw_features = lio->dev_capability;
4045 		/*HW_VLAN_RX and HW_VLAN_FILTER is always on*/
4046 		netdev->hw_features = netdev->hw_features &
4047 			~NETIF_F_HW_VLAN_CTAG_RX;
4048 
4049 		/* MTU range: 68 - 16000 */
4050 		netdev->min_mtu = LIO_MIN_MTU_SIZE;
4051 		netdev->max_mtu = LIO_MAX_MTU_SIZE;
4052 
4053 		/* Point to the  properties for octeon device to which this
4054 		 * interface belongs.
4055 		 */
4056 		lio->oct_dev = octeon_dev;
4057 		lio->octprops = props;
4058 		lio->netdev = netdev;
4059 
4060 		dev_dbg(&octeon_dev->pci_dev->dev,
4061 			"if%d gmx: %d hw_addr: 0x%llx\n", i,
4062 			lio->linfo.gmxport, CVM_CAST64(lio->linfo.hw_addr));
4063 
4064 		for (j = 0; j < octeon_dev->sriov_info.max_vfs; j++) {
4065 			u8 vfmac[ETH_ALEN];
4066 
4067 			random_ether_addr(&vfmac[0]);
4068 			if (__liquidio_set_vf_mac(netdev, j,
4069 						  &vfmac[0], false)) {
4070 				dev_err(&octeon_dev->pci_dev->dev,
4071 					"Error setting VF%d MAC address\n",
4072 					j);
4073 				goto setup_nic_dev_fail;
4074 			}
4075 		}
4076 
4077 		/* 64-bit swap required on LE machines */
4078 		octeon_swap_8B_data(&lio->linfo.hw_addr, 1);
4079 		for (j = 0; j < 6; j++)
4080 			mac[j] = *((u8 *)(((u8 *)&lio->linfo.hw_addr) + 2 + j));
4081 
4082 		/* Copy MAC Address to OS network device structure */
4083 
4084 		ether_addr_copy(netdev->dev_addr, mac);
4085 
4086 		/* By default all interfaces on a single Octeon uses the same
4087 		 * tx and rx queues
4088 		 */
4089 		lio->txq = lio->linfo.txpciq[0].s.q_no;
4090 		lio->rxq = lio->linfo.rxpciq[0].s.q_no;
4091 		if (setup_io_queues(octeon_dev, i)) {
4092 			dev_err(&octeon_dev->pci_dev->dev, "I/O queues creation failed\n");
4093 			goto setup_nic_dev_fail;
4094 		}
4095 
4096 		ifstate_set(lio, LIO_IFSTATE_DROQ_OPS);
4097 
4098 		lio->tx_qsize = octeon_get_tx_qsize(octeon_dev, lio->txq);
4099 		lio->rx_qsize = octeon_get_rx_qsize(octeon_dev, lio->rxq);
4100 
4101 		if (setup_glists(octeon_dev, lio, num_iqueues)) {
4102 			dev_err(&octeon_dev->pci_dev->dev,
4103 				"Gather list allocation failed\n");
4104 			goto setup_nic_dev_fail;
4105 		}
4106 
4107 		/* Register ethtool support */
4108 		liquidio_set_ethtool_ops(netdev);
4109 		if (lio->oct_dev->chip_id == OCTEON_CN23XX_PF_VID)
4110 			octeon_dev->priv_flags = OCT_PRIV_FLAG_DEFAULT;
4111 		else
4112 			octeon_dev->priv_flags = 0x0;
4113 
4114 		if (netdev->features & NETIF_F_LRO)
4115 			liquidio_set_feature(netdev, OCTNET_CMD_LRO_ENABLE,
4116 					     OCTNIC_LROIPV4 | OCTNIC_LROIPV6);
4117 
4118 		liquidio_set_feature(netdev, OCTNET_CMD_ENABLE_VLAN_FILTER, 0);
4119 
4120 		if ((debug != -1) && (debug & NETIF_MSG_HW))
4121 			liquidio_set_feature(netdev,
4122 					     OCTNET_CMD_VERBOSE_ENABLE, 0);
4123 
4124 		if (setup_link_status_change_wq(netdev))
4125 			goto setup_nic_dev_fail;
4126 
4127 		/* Register the network device with the OS */
4128 		if (register_netdev(netdev)) {
4129 			dev_err(&octeon_dev->pci_dev->dev, "Device registration failed\n");
4130 			goto setup_nic_dev_fail;
4131 		}
4132 
4133 		dev_dbg(&octeon_dev->pci_dev->dev,
4134 			"Setup NIC ifidx:%d mac:%02x%02x%02x%02x%02x%02x\n",
4135 			i, mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
4136 		netif_carrier_off(netdev);
4137 		lio->link_changes++;
4138 
4139 		ifstate_set(lio, LIO_IFSTATE_REGISTERED);
4140 
4141 		/* Sending command to firmware to enable Rx checksum offload
4142 		 * by default at the time of setup of Liquidio driver for
4143 		 * this device
4144 		 */
4145 		liquidio_set_rxcsum_command(netdev, OCTNET_CMD_TNL_RX_CSUM_CTL,
4146 					    OCTNET_CMD_RXCSUM_ENABLE);
4147 		liquidio_set_feature(netdev, OCTNET_CMD_TNL_TX_CSUM_CTL,
4148 				     OCTNET_CMD_TXCSUM_ENABLE);
4149 
4150 		dev_dbg(&octeon_dev->pci_dev->dev,
4151 			"NIC ifidx:%d Setup successful\n", i);
4152 
4153 		octeon_free_soft_command(octeon_dev, sc);
4154 	}
4155 
4156 	return 0;
4157 
4158 setup_nic_dev_fail:
4159 
4160 	octeon_free_soft_command(octeon_dev, sc);
4161 
4162 setup_nic_wait_intr:
4163 
4164 	while (i--) {
4165 		dev_err(&octeon_dev->pci_dev->dev,
4166 			"NIC ifidx:%d Setup failed\n", i);
4167 		liquidio_destroy_nic_device(octeon_dev, i);
4168 	}
4169 	return -ENODEV;
4170 }
4171 
4172 #ifdef CONFIG_PCI_IOV
4173 static int octeon_enable_sriov(struct octeon_device *oct)
4174 {
4175 	unsigned int num_vfs_alloced = oct->sriov_info.num_vfs_alloced;
4176 	struct pci_dev *vfdev;
4177 	int err;
4178 	u32 u;
4179 
4180 	if (OCTEON_CN23XX_PF(oct) && num_vfs_alloced) {
4181 		err = pci_enable_sriov(oct->pci_dev,
4182 				       oct->sriov_info.num_vfs_alloced);
4183 		if (err) {
4184 			dev_err(&oct->pci_dev->dev,
4185 				"OCTEON: Failed to enable PCI sriov: %d\n",
4186 				err);
4187 			oct->sriov_info.num_vfs_alloced = 0;
4188 			return err;
4189 		}
4190 		oct->sriov_info.sriov_enabled = 1;
4191 
4192 		/* init lookup table that maps DPI ring number to VF pci_dev
4193 		 * struct pointer
4194 		 */
4195 		u = 0;
4196 		vfdev = pci_get_device(PCI_VENDOR_ID_CAVIUM,
4197 				       OCTEON_CN23XX_VF_VID, NULL);
4198 		while (vfdev) {
4199 			if (vfdev->is_virtfn &&
4200 			    (vfdev->physfn == oct->pci_dev)) {
4201 				oct->sriov_info.dpiring_to_vfpcidev_lut[u] =
4202 					vfdev;
4203 				u += oct->sriov_info.rings_per_vf;
4204 			}
4205 			vfdev = pci_get_device(PCI_VENDOR_ID_CAVIUM,
4206 					       OCTEON_CN23XX_VF_VID, vfdev);
4207 		}
4208 	}
4209 
4210 	return num_vfs_alloced;
4211 }
4212 
4213 static int lio_pci_sriov_disable(struct octeon_device *oct)
4214 {
4215 	int u;
4216 
4217 	if (pci_vfs_assigned(oct->pci_dev)) {
4218 		dev_err(&oct->pci_dev->dev, "VFs are still assigned to VMs.\n");
4219 		return -EPERM;
4220 	}
4221 
4222 	pci_disable_sriov(oct->pci_dev);
4223 
4224 	u = 0;
4225 	while (u < MAX_POSSIBLE_VFS) {
4226 		oct->sriov_info.dpiring_to_vfpcidev_lut[u] = NULL;
4227 		u += oct->sriov_info.rings_per_vf;
4228 	}
4229 
4230 	oct->sriov_info.num_vfs_alloced = 0;
4231 	dev_info(&oct->pci_dev->dev, "oct->pf_num:%d disabled VFs\n",
4232 		 oct->pf_num);
4233 
4234 	return 0;
4235 }
4236 
4237 static int liquidio_enable_sriov(struct pci_dev *dev, int num_vfs)
4238 {
4239 	struct octeon_device *oct = pci_get_drvdata(dev);
4240 	int ret = 0;
4241 
4242 	if ((num_vfs == oct->sriov_info.num_vfs_alloced) &&
4243 	    (oct->sriov_info.sriov_enabled)) {
4244 		dev_info(&oct->pci_dev->dev, "oct->pf_num:%d already enabled num_vfs:%d\n",
4245 			 oct->pf_num, num_vfs);
4246 		return 0;
4247 	}
4248 
4249 	if (!num_vfs) {
4250 		ret = lio_pci_sriov_disable(oct);
4251 	} else if (num_vfs > oct->sriov_info.max_vfs) {
4252 		dev_err(&oct->pci_dev->dev,
4253 			"OCTEON: Max allowed VFs:%d user requested:%d",
4254 			oct->sriov_info.max_vfs, num_vfs);
4255 		ret = -EPERM;
4256 	} else {
4257 		oct->sriov_info.num_vfs_alloced = num_vfs;
4258 		ret = octeon_enable_sriov(oct);
4259 		dev_info(&oct->pci_dev->dev, "oct->pf_num:%d num_vfs:%d\n",
4260 			 oct->pf_num, num_vfs);
4261 	}
4262 
4263 	return ret;
4264 }
4265 #endif
4266 
4267 /**
4268  * \brief initialize the NIC
4269  * @param oct octeon device
4270  *
4271  * This initialization routine is called once the Octeon device application is
4272  * up and running
4273  */
4274 static int liquidio_init_nic_module(struct octeon_device *oct)
4275 {
4276 	struct oct_intrmod_cfg *intrmod_cfg;
4277 	int i, retval = 0;
4278 	int num_nic_ports = CFG_GET_NUM_NIC_PORTS(octeon_get_conf(oct));
4279 
4280 	dev_dbg(&oct->pci_dev->dev, "Initializing network interfaces\n");
4281 
4282 	/* only default iq and oq were initialized
4283 	 * initialize the rest as well
4284 	 */
4285 	/* run port_config command for each port */
4286 	oct->ifcount = num_nic_ports;
4287 
4288 	memset(oct->props, 0, sizeof(struct octdev_props) * num_nic_ports);
4289 
4290 	for (i = 0; i < MAX_OCTEON_LINKS; i++)
4291 		oct->props[i].gmxport = -1;
4292 
4293 	retval = setup_nic_devices(oct);
4294 	if (retval) {
4295 		dev_err(&oct->pci_dev->dev, "Setup NIC devices failed\n");
4296 		goto octnet_init_failure;
4297 	}
4298 
4299 	liquidio_ptp_init(oct);
4300 
4301 	/* Initialize interrupt moderation params */
4302 	intrmod_cfg = &((struct octeon_device *)oct)->intrmod;
4303 	intrmod_cfg->rx_enable = 1;
4304 	intrmod_cfg->check_intrvl = LIO_INTRMOD_CHECK_INTERVAL;
4305 	intrmod_cfg->maxpkt_ratethr = LIO_INTRMOD_MAXPKT_RATETHR;
4306 	intrmod_cfg->minpkt_ratethr = LIO_INTRMOD_MINPKT_RATETHR;
4307 	intrmod_cfg->rx_maxcnt_trigger = LIO_INTRMOD_RXMAXCNT_TRIGGER;
4308 	intrmod_cfg->rx_maxtmr_trigger = LIO_INTRMOD_RXMAXTMR_TRIGGER;
4309 	intrmod_cfg->rx_mintmr_trigger = LIO_INTRMOD_RXMINTMR_TRIGGER;
4310 	intrmod_cfg->rx_mincnt_trigger = LIO_INTRMOD_RXMINCNT_TRIGGER;
4311 	intrmod_cfg->tx_enable = 1;
4312 	intrmod_cfg->tx_maxcnt_trigger = LIO_INTRMOD_TXMAXCNT_TRIGGER;
4313 	intrmod_cfg->tx_mincnt_trigger = LIO_INTRMOD_TXMINCNT_TRIGGER;
4314 	intrmod_cfg->rx_frames = CFG_GET_OQ_INTR_PKT(octeon_get_conf(oct));
4315 	intrmod_cfg->rx_usecs = CFG_GET_OQ_INTR_TIME(octeon_get_conf(oct));
4316 	intrmod_cfg->tx_frames = CFG_GET_IQ_INTR_PKT(octeon_get_conf(oct));
4317 	dev_dbg(&oct->pci_dev->dev, "Network interfaces ready\n");
4318 
4319 	return retval;
4320 
4321 octnet_init_failure:
4322 
4323 	oct->ifcount = 0;
4324 
4325 	return retval;
4326 }
4327 
4328 /**
4329  * \brief starter callback that invokes the remaining initialization work after
4330  * the NIC is up and running.
4331  * @param octptr  work struct work_struct
4332  */
4333 static void nic_starter(struct work_struct *work)
4334 {
4335 	struct octeon_device *oct;
4336 	struct cavium_wk *wk = (struct cavium_wk *)work;
4337 
4338 	oct = (struct octeon_device *)wk->ctxptr;
4339 
4340 	if (atomic_read(&oct->status) == OCT_DEV_RUNNING)
4341 		return;
4342 
4343 	/* If the status of the device is CORE_OK, the core
4344 	 * application has reported its application type. Call
4345 	 * any registered handlers now and move to the RUNNING
4346 	 * state.
4347 	 */
4348 	if (atomic_read(&oct->status) != OCT_DEV_CORE_OK) {
4349 		schedule_delayed_work(&oct->nic_poll_work.work,
4350 				      LIQUIDIO_STARTER_POLL_INTERVAL_MS);
4351 		return;
4352 	}
4353 
4354 	atomic_set(&oct->status, OCT_DEV_RUNNING);
4355 
4356 	if (oct->app_mode && oct->app_mode == CVM_DRV_NIC_APP) {
4357 		dev_dbg(&oct->pci_dev->dev, "Starting NIC module\n");
4358 
4359 		if (liquidio_init_nic_module(oct))
4360 			dev_err(&oct->pci_dev->dev, "NIC initialization failed\n");
4361 		else
4362 			handshake[oct->octeon_id].started_ok = 1;
4363 	} else {
4364 		dev_err(&oct->pci_dev->dev,
4365 			"Unexpected application running on NIC (%d). Check firmware.\n",
4366 			oct->app_mode);
4367 	}
4368 
4369 	complete(&handshake[oct->octeon_id].started);
4370 }
4371 
4372 static int
4373 octeon_recv_vf_drv_notice(struct octeon_recv_info *recv_info, void *buf)
4374 {
4375 	struct octeon_device *oct = (struct octeon_device *)buf;
4376 	struct octeon_recv_pkt *recv_pkt = recv_info->recv_pkt;
4377 	int i, notice, vf_idx;
4378 	u64 *data, vf_num;
4379 
4380 	notice = recv_pkt->rh.r.ossp;
4381 	data = (u64 *)get_rbd(recv_pkt->buffer_ptr[0]);
4382 
4383 	/* the first 64-bit word of data is the vf_num */
4384 	vf_num = data[0];
4385 	octeon_swap_8B_data(&vf_num, 1);
4386 	vf_idx = (int)vf_num - 1;
4387 
4388 	if (notice == VF_DRV_LOADED) {
4389 		if (!(oct->sriov_info.vf_drv_loaded_mask & BIT_ULL(vf_idx))) {
4390 			oct->sriov_info.vf_drv_loaded_mask |= BIT_ULL(vf_idx);
4391 			dev_info(&oct->pci_dev->dev,
4392 				 "driver for VF%d was loaded\n", vf_idx);
4393 			try_module_get(THIS_MODULE);
4394 		}
4395 	} else if (notice == VF_DRV_REMOVED) {
4396 		if (oct->sriov_info.vf_drv_loaded_mask & BIT_ULL(vf_idx)) {
4397 			oct->sriov_info.vf_drv_loaded_mask &= ~BIT_ULL(vf_idx);
4398 			dev_info(&oct->pci_dev->dev,
4399 				 "driver for VF%d was removed\n", vf_idx);
4400 			module_put(THIS_MODULE);
4401 		}
4402 	} else if (notice == VF_DRV_MACADDR_CHANGED) {
4403 		u8 *b = (u8 *)&data[1];
4404 
4405 		oct->sriov_info.vf_macaddr[vf_idx] = data[1];
4406 		dev_info(&oct->pci_dev->dev,
4407 			 "VF driver changed VF%d's MAC address to %pM\n",
4408 			 vf_idx, b + 2);
4409 	}
4410 
4411 	for (i = 0; i < recv_pkt->buffer_count; i++)
4412 		recv_buffer_free(recv_pkt->buffer_ptr[i]);
4413 	octeon_free_recv_info(recv_info);
4414 
4415 	return 0;
4416 }
4417 
4418 /**
4419  * \brief Device initialization for each Octeon device that is probed
4420  * @param octeon_dev  octeon device
4421  */
4422 static int octeon_device_init(struct octeon_device *octeon_dev)
4423 {
4424 	int j, ret;
4425 	int fw_loaded = 0;
4426 	char bootcmd[] = "\n";
4427 	struct octeon_device_priv *oct_priv =
4428 		(struct octeon_device_priv *)octeon_dev->priv;
4429 	atomic_set(&octeon_dev->status, OCT_DEV_BEGIN_STATE);
4430 
4431 	/* Enable access to the octeon device and make its DMA capability
4432 	 * known to the OS.
4433 	 */
4434 	if (octeon_pci_os_setup(octeon_dev))
4435 		return 1;
4436 
4437 	atomic_set(&octeon_dev->status, OCT_DEV_PCI_ENABLE_DONE);
4438 
4439 	/* Identify the Octeon type and map the BAR address space. */
4440 	if (octeon_chip_specific_setup(octeon_dev)) {
4441 		dev_err(&octeon_dev->pci_dev->dev, "Chip specific setup failed\n");
4442 		return 1;
4443 	}
4444 
4445 	atomic_set(&octeon_dev->status, OCT_DEV_PCI_MAP_DONE);
4446 
4447 	octeon_dev->app_mode = CVM_DRV_INVALID_APP;
4448 
4449 	if (OCTEON_CN23XX_PF(octeon_dev)) {
4450 		if (!cn23xx_fw_loaded(octeon_dev)) {
4451 			fw_loaded = 0;
4452 			/* Do a soft reset of the Octeon device. */
4453 			if (octeon_dev->fn_list.soft_reset(octeon_dev))
4454 				return 1;
4455 			/* things might have changed */
4456 			if (!cn23xx_fw_loaded(octeon_dev))
4457 				fw_loaded = 0;
4458 			else
4459 				fw_loaded = 1;
4460 		} else {
4461 			fw_loaded = 1;
4462 		}
4463 	} else if (octeon_dev->fn_list.soft_reset(octeon_dev)) {
4464 		return 1;
4465 	}
4466 
4467 	/* Initialize the dispatch mechanism used to push packets arriving on
4468 	 * Octeon Output queues.
4469 	 */
4470 	if (octeon_init_dispatch_list(octeon_dev))
4471 		return 1;
4472 
4473 	octeon_register_dispatch_fn(octeon_dev, OPCODE_NIC,
4474 				    OPCODE_NIC_CORE_DRV_ACTIVE,
4475 				    octeon_core_drv_init,
4476 				    octeon_dev);
4477 
4478 	octeon_register_dispatch_fn(octeon_dev, OPCODE_NIC,
4479 				    OPCODE_NIC_VF_DRV_NOTICE,
4480 				    octeon_recv_vf_drv_notice, octeon_dev);
4481 	INIT_DELAYED_WORK(&octeon_dev->nic_poll_work.work, nic_starter);
4482 	octeon_dev->nic_poll_work.ctxptr = (void *)octeon_dev;
4483 	schedule_delayed_work(&octeon_dev->nic_poll_work.work,
4484 			      LIQUIDIO_STARTER_POLL_INTERVAL_MS);
4485 
4486 	atomic_set(&octeon_dev->status, OCT_DEV_DISPATCH_INIT_DONE);
4487 
4488 	if (octeon_set_io_queues_off(octeon_dev)) {
4489 		dev_err(&octeon_dev->pci_dev->dev, "setting io queues off failed\n");
4490 		return 1;
4491 	}
4492 
4493 	if (OCTEON_CN23XX_PF(octeon_dev)) {
4494 		ret = octeon_dev->fn_list.setup_device_regs(octeon_dev);
4495 		if (ret) {
4496 			dev_err(&octeon_dev->pci_dev->dev, "OCTEON: Failed to configure device registers\n");
4497 			return ret;
4498 		}
4499 	}
4500 
4501 	/* Initialize soft command buffer pool
4502 	 */
4503 	if (octeon_setup_sc_buffer_pool(octeon_dev)) {
4504 		dev_err(&octeon_dev->pci_dev->dev, "sc buffer pool allocation failed\n");
4505 		return 1;
4506 	}
4507 	atomic_set(&octeon_dev->status, OCT_DEV_SC_BUFF_POOL_INIT_DONE);
4508 
4509 	/*  Setup the data structures that manage this Octeon's Input queues. */
4510 	if (octeon_setup_instr_queues(octeon_dev)) {
4511 		dev_err(&octeon_dev->pci_dev->dev,
4512 			"instruction queue initialization failed\n");
4513 		return 1;
4514 	}
4515 	atomic_set(&octeon_dev->status, OCT_DEV_INSTR_QUEUE_INIT_DONE);
4516 
4517 	/* Initialize lists to manage the requests of different types that
4518 	 * arrive from user & kernel applications for this octeon device.
4519 	 */
4520 	if (octeon_setup_response_list(octeon_dev)) {
4521 		dev_err(&octeon_dev->pci_dev->dev, "Response list allocation failed\n");
4522 		return 1;
4523 	}
4524 	atomic_set(&octeon_dev->status, OCT_DEV_RESP_LIST_INIT_DONE);
4525 
4526 	if (octeon_setup_output_queues(octeon_dev)) {
4527 		dev_err(&octeon_dev->pci_dev->dev, "Output queue initialization failed\n");
4528 		return 1;
4529 	}
4530 
4531 	atomic_set(&octeon_dev->status, OCT_DEV_DROQ_INIT_DONE);
4532 
4533 	if (OCTEON_CN23XX_PF(octeon_dev)) {
4534 		if (octeon_dev->fn_list.setup_mbox(octeon_dev)) {
4535 			dev_err(&octeon_dev->pci_dev->dev, "OCTEON: Mailbox setup failed\n");
4536 			return 1;
4537 		}
4538 		atomic_set(&octeon_dev->status, OCT_DEV_MBOX_SETUP_DONE);
4539 
4540 		if (octeon_allocate_ioq_vector(octeon_dev)) {
4541 			dev_err(&octeon_dev->pci_dev->dev, "OCTEON: ioq vector allocation failed\n");
4542 			return 1;
4543 		}
4544 		atomic_set(&octeon_dev->status, OCT_DEV_MSIX_ALLOC_VECTOR_DONE);
4545 
4546 	} else {
4547 		/* The input and output queue registers were setup earlier (the
4548 		 * queues were not enabled). Any additional registers
4549 		 * that need to be programmed should be done now.
4550 		 */
4551 		ret = octeon_dev->fn_list.setup_device_regs(octeon_dev);
4552 		if (ret) {
4553 			dev_err(&octeon_dev->pci_dev->dev,
4554 				"Failed to configure device registers\n");
4555 			return ret;
4556 		}
4557 	}
4558 
4559 	/* Initialize the tasklet that handles output queue packet processing.*/
4560 	dev_dbg(&octeon_dev->pci_dev->dev, "Initializing droq tasklet\n");
4561 	tasklet_init(&oct_priv->droq_tasklet, octeon_droq_bh,
4562 		     (unsigned long)octeon_dev);
4563 
4564 	/* Setup the interrupt handler and record the INT SUM register address
4565 	 */
4566 	if (octeon_setup_interrupt(octeon_dev))
4567 		return 1;
4568 
4569 	/* Enable Octeon device interrupts */
4570 	octeon_dev->fn_list.enable_interrupt(octeon_dev, OCTEON_ALL_INTR);
4571 
4572 	atomic_set(&octeon_dev->status, OCT_DEV_INTR_SET_DONE);
4573 
4574 	/* Enable the input and output queues for this Octeon device */
4575 	ret = octeon_dev->fn_list.enable_io_queues(octeon_dev);
4576 	if (ret) {
4577 		dev_err(&octeon_dev->pci_dev->dev, "Failed to enable input/output queues");
4578 		return ret;
4579 	}
4580 
4581 	atomic_set(&octeon_dev->status, OCT_DEV_IO_QUEUES_DONE);
4582 
4583 	if ((!OCTEON_CN23XX_PF(octeon_dev)) || !fw_loaded) {
4584 		dev_dbg(&octeon_dev->pci_dev->dev, "Waiting for DDR initialization...\n");
4585 		if (!ddr_timeout) {
4586 			dev_info(&octeon_dev->pci_dev->dev,
4587 				 "WAITING. Set ddr_timeout to non-zero value to proceed with initialization.\n");
4588 		}
4589 
4590 		schedule_timeout_uninterruptible(HZ * LIO_RESET_SECS);
4591 
4592 		/* Wait for the octeon to initialize DDR after the soft-reset.*/
4593 		while (!ddr_timeout) {
4594 			set_current_state(TASK_INTERRUPTIBLE);
4595 			if (schedule_timeout(HZ / 10)) {
4596 				/* user probably pressed Control-C */
4597 				return 1;
4598 			}
4599 		}
4600 		ret = octeon_wait_for_ddr_init(octeon_dev, &ddr_timeout);
4601 		if (ret) {
4602 			dev_err(&octeon_dev->pci_dev->dev,
4603 				"DDR not initialized. Please confirm that board is configured to boot from Flash, ret: %d\n",
4604 				ret);
4605 			return 1;
4606 		}
4607 
4608 		if (octeon_wait_for_bootloader(octeon_dev, 1000)) {
4609 			dev_err(&octeon_dev->pci_dev->dev, "Board not responding\n");
4610 			return 1;
4611 		}
4612 
4613 		/* Divert uboot to take commands from host instead. */
4614 		ret = octeon_console_send_cmd(octeon_dev, bootcmd, 50);
4615 
4616 		dev_dbg(&octeon_dev->pci_dev->dev, "Initializing consoles\n");
4617 		ret = octeon_init_consoles(octeon_dev);
4618 		if (ret) {
4619 			dev_err(&octeon_dev->pci_dev->dev, "Could not access board consoles\n");
4620 			return 1;
4621 		}
4622 		ret = octeon_add_console(octeon_dev, 0);
4623 		if (ret) {
4624 			dev_err(&octeon_dev->pci_dev->dev, "Could not access board console\n");
4625 			return 1;
4626 		}
4627 
4628 		atomic_set(&octeon_dev->status, OCT_DEV_CONSOLE_INIT_DONE);
4629 
4630 		dev_dbg(&octeon_dev->pci_dev->dev, "Loading firmware\n");
4631 		ret = load_firmware(octeon_dev);
4632 		if (ret) {
4633 			dev_err(&octeon_dev->pci_dev->dev, "Could not load firmware to board\n");
4634 			return 1;
4635 		}
4636 		/* set bit 1 of SLI_SCRATCH_1 to indicate that firmware is
4637 		 * loaded
4638 		 */
4639 		if (OCTEON_CN23XX_PF(octeon_dev))
4640 			octeon_write_csr64(octeon_dev, CN23XX_SLI_SCRATCH1,
4641 					   2ULL);
4642 	}
4643 
4644 	handshake[octeon_dev->octeon_id].init_ok = 1;
4645 	complete(&handshake[octeon_dev->octeon_id].init);
4646 
4647 	atomic_set(&octeon_dev->status, OCT_DEV_HOST_OK);
4648 
4649 	/* Send Credit for Octeon Output queues. Credits are always sent after
4650 	 * the output queue is enabled.
4651 	 */
4652 	for (j = 0; j < octeon_dev->num_oqs; j++)
4653 		writel(octeon_dev->droq[j]->max_count,
4654 		       octeon_dev->droq[j]->pkts_credit_reg);
4655 
4656 	/* Packets can start arriving on the output queues from this point. */
4657 	return 0;
4658 }
4659 
4660 /**
4661  * \brief Exits the module
4662  */
4663 static void __exit liquidio_exit(void)
4664 {
4665 	liquidio_deinit_pci();
4666 
4667 	pr_info("LiquidIO network module is now unloaded\n");
4668 }
4669 
4670 module_init(liquidio_init);
4671 module_exit(liquidio_exit);
4672