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/module.h>
19 #include <linux/pci.h>
20 #include <linux/firmware.h>
21 #include <net/vxlan.h>
22 #include <linux/kthread.h>
23 #include "liquidio_common.h"
24 #include "octeon_droq.h"
25 #include "octeon_iq.h"
26 #include "response_manager.h"
27 #include "octeon_device.h"
28 #include "octeon_nic.h"
29 #include "octeon_main.h"
30 #include "octeon_network.h"
31 #include "cn66xx_regs.h"
32 #include "cn66xx_device.h"
33 #include "cn68xx_device.h"
34 #include "cn23xx_pf_device.h"
35 #include "liquidio_image.h"
36 
37 MODULE_AUTHOR("Cavium Networks, <support@cavium.com>");
38 MODULE_DESCRIPTION("Cavium LiquidIO Intelligent Server Adapter Driver");
39 MODULE_LICENSE("GPL");
40 MODULE_VERSION(LIQUIDIO_VERSION);
41 MODULE_FIRMWARE(LIO_FW_DIR LIO_FW_BASE_NAME LIO_210SV_NAME LIO_FW_NAME_SUFFIX);
42 MODULE_FIRMWARE(LIO_FW_DIR LIO_FW_BASE_NAME LIO_210NV_NAME LIO_FW_NAME_SUFFIX);
43 MODULE_FIRMWARE(LIO_FW_DIR LIO_FW_BASE_NAME LIO_410NV_NAME LIO_FW_NAME_SUFFIX);
44 MODULE_FIRMWARE(LIO_FW_DIR LIO_FW_BASE_NAME LIO_23XX_NAME LIO_FW_NAME_SUFFIX);
45 
46 static int ddr_timeout = 10000;
47 module_param(ddr_timeout, int, 0644);
48 MODULE_PARM_DESC(ddr_timeout,
49 		 "Number of milliseconds to wait for DDR initialization. 0 waits for ddr_timeout to be set to non-zero value before starting to check");
50 
51 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK)
52 
53 static int debug = -1;
54 module_param(debug, int, 0644);
55 MODULE_PARM_DESC(debug, "NETIF_MSG debug bits");
56 
57 static char fw_type[LIO_MAX_FW_TYPE_LEN];
58 module_param_string(fw_type, fw_type, sizeof(fw_type), 0000);
59 MODULE_PARM_DESC(fw_type, "Type of firmware to be loaded. Default \"nic\"");
60 
61 static int ptp_enable = 1;
62 
63 /* Bit mask values for lio->ifstate */
64 #define   LIO_IFSTATE_DROQ_OPS             0x01
65 #define   LIO_IFSTATE_REGISTERED           0x02
66 #define   LIO_IFSTATE_RUNNING              0x04
67 #define   LIO_IFSTATE_RX_TIMESTAMP_ENABLED 0x08
68 
69 /* Polling interval for determining when NIC application is alive */
70 #define LIQUIDIO_STARTER_POLL_INTERVAL_MS 100
71 
72 /* runtime link query interval */
73 #define LIQUIDIO_LINK_QUERY_INTERVAL_MS         1000
74 
75 struct liquidio_if_cfg_context {
76 	int octeon_id;
77 
78 	wait_queue_head_t wc;
79 
80 	int cond;
81 };
82 
83 struct liquidio_if_cfg_resp {
84 	u64 rh;
85 	struct liquidio_if_cfg_info cfg_info;
86 	u64 status;
87 };
88 
89 struct liquidio_rx_ctl_context {
90 	int octeon_id;
91 
92 	wait_queue_head_t wc;
93 
94 	int cond;
95 };
96 
97 struct oct_link_status_resp {
98 	u64 rh;
99 	struct oct_link_info link_info;
100 	u64 status;
101 };
102 
103 struct oct_timestamp_resp {
104 	u64 rh;
105 	u64 timestamp;
106 	u64 status;
107 };
108 
109 #define OCT_TIMESTAMP_RESP_SIZE (sizeof(struct oct_timestamp_resp))
110 
111 union tx_info {
112 	u64 u64;
113 	struct {
114 #ifdef __BIG_ENDIAN_BITFIELD
115 		u16 gso_size;
116 		u16 gso_segs;
117 		u32 reserved;
118 #else
119 		u32 reserved;
120 		u16 gso_segs;
121 		u16 gso_size;
122 #endif
123 	} s;
124 };
125 
126 /** Octeon device properties to be used by the NIC module.
127  * Each octeon device in the system will be represented
128  * by this structure in the NIC module.
129  */
130 
131 #define OCTNIC_MAX_SG  (MAX_SKB_FRAGS)
132 
133 #define OCTNIC_GSO_MAX_HEADER_SIZE 128
134 #define OCTNIC_GSO_MAX_SIZE                                                    \
135 	(CN23XX_DEFAULT_INPUT_JABBER - OCTNIC_GSO_MAX_HEADER_SIZE)
136 
137 /** Structure of a node in list of gather components maintained by
138  * NIC driver for each network device.
139  */
140 struct octnic_gather {
141 	/** List manipulation. Next and prev pointers. */
142 	struct list_head list;
143 
144 	/** Size of the gather component at sg in bytes. */
145 	int sg_size;
146 
147 	/** Number of bytes that sg was adjusted to make it 8B-aligned. */
148 	int adjust;
149 
150 	/** Gather component that can accommodate max sized fragment list
151 	 *  received from the IP layer.
152 	 */
153 	struct octeon_sg_entry *sg;
154 
155 	u64 sg_dma_ptr;
156 };
157 
158 struct handshake {
159 	struct completion init;
160 	struct completion started;
161 	struct pci_dev *pci_dev;
162 	int init_ok;
163 	int started_ok;
164 };
165 
166 struct octeon_device_priv {
167 	/** Tasklet structures for this device. */
168 	struct tasklet_struct droq_tasklet;
169 	unsigned long napi_mask;
170 };
171 
172 #ifdef CONFIG_PCI_IOV
173 static int liquidio_enable_sriov(struct pci_dev *dev, int num_vfs);
174 #endif
175 
176 static int octeon_device_init(struct octeon_device *);
177 static int liquidio_stop(struct net_device *netdev);
178 static void liquidio_remove(struct pci_dev *pdev);
179 static int liquidio_probe(struct pci_dev *pdev,
180 			  const struct pci_device_id *ent);
181 
182 static struct handshake handshake[MAX_OCTEON_DEVICES];
183 static struct completion first_stage;
184 
185 static void octeon_droq_bh(unsigned long pdev)
186 {
187 	int q_no;
188 	int reschedule = 0;
189 	struct octeon_device *oct = (struct octeon_device *)pdev;
190 	struct octeon_device_priv *oct_priv =
191 		(struct octeon_device_priv *)oct->priv;
192 
193 	for (q_no = 0; q_no < MAX_OCTEON_OUTPUT_QUEUES(oct); q_no++) {
194 		if (!(oct->io_qmask.oq & BIT_ULL(q_no)))
195 			continue;
196 		reschedule |= octeon_droq_process_packets(oct, oct->droq[q_no],
197 							  MAX_PACKET_BUDGET);
198 		lio_enable_irq(oct->droq[q_no], NULL);
199 
200 		if (OCTEON_CN23XX_PF(oct) && oct->msix_on) {
201 			/* set time and cnt interrupt thresholds for this DROQ
202 			 * for NAPI
203 			 */
204 			int adjusted_q_no = q_no + oct->sriov_info.pf_srn;
205 
206 			octeon_write_csr64(
207 			    oct, CN23XX_SLI_OQ_PKT_INT_LEVELS(adjusted_q_no),
208 			    0x5700000040ULL);
209 			octeon_write_csr64(
210 			    oct, CN23XX_SLI_OQ_PKTS_SENT(adjusted_q_no), 0);
211 		}
212 	}
213 
214 	if (reschedule)
215 		tasklet_schedule(&oct_priv->droq_tasklet);
216 }
217 
218 static int lio_wait_for_oq_pkts(struct octeon_device *oct)
219 {
220 	struct octeon_device_priv *oct_priv =
221 		(struct octeon_device_priv *)oct->priv;
222 	int retry = 100, pkt_cnt = 0, pending_pkts = 0;
223 	int i;
224 
225 	do {
226 		pending_pkts = 0;
227 
228 		for (i = 0; i < MAX_OCTEON_OUTPUT_QUEUES(oct); i++) {
229 			if (!(oct->io_qmask.oq & BIT_ULL(i)))
230 				continue;
231 			pkt_cnt += octeon_droq_check_hw_for_pkts(oct->droq[i]);
232 		}
233 		if (pkt_cnt > 0) {
234 			pending_pkts += pkt_cnt;
235 			tasklet_schedule(&oct_priv->droq_tasklet);
236 		}
237 		pkt_cnt = 0;
238 		schedule_timeout_uninterruptible(1);
239 
240 	} while (retry-- && pending_pkts);
241 
242 	return pkt_cnt;
243 }
244 
245 /**
246  * \brief Forces all IO queues off on a given device
247  * @param oct Pointer to Octeon device
248  */
249 static void force_io_queues_off(struct octeon_device *oct)
250 {
251 	if ((oct->chip_id == OCTEON_CN66XX) ||
252 	    (oct->chip_id == OCTEON_CN68XX)) {
253 		/* Reset the Enable bits for Input Queues. */
254 		octeon_write_csr(oct, CN6XXX_SLI_PKT_INSTR_ENB, 0);
255 
256 		/* Reset the Enable bits for Output Queues. */
257 		octeon_write_csr(oct, CN6XXX_SLI_PKT_OUT_ENB, 0);
258 	}
259 }
260 
261 /**
262  * \brief wait for all pending requests to complete
263  * @param oct Pointer to Octeon device
264  *
265  * Called during shutdown sequence
266  */
267 static int wait_for_pending_requests(struct octeon_device *oct)
268 {
269 	int i, pcount = 0;
270 
271 	for (i = 0; i < 100; i++) {
272 		pcount =
273 			atomic_read(&oct->response_list
274 				[OCTEON_ORDERED_SC_LIST].pending_req_count);
275 		if (pcount)
276 			schedule_timeout_uninterruptible(HZ / 10);
277 		else
278 			break;
279 	}
280 
281 	if (pcount)
282 		return 1;
283 
284 	return 0;
285 }
286 
287 /**
288  * \brief Cause device to go quiet so it can be safely removed/reset/etc
289  * @param oct Pointer to Octeon device
290  */
291 static inline void pcierror_quiesce_device(struct octeon_device *oct)
292 {
293 	int i;
294 
295 	/* Disable the input and output queues now. No more packets will
296 	 * arrive from Octeon, but we should wait for all packet processing
297 	 * to finish.
298 	 */
299 	force_io_queues_off(oct);
300 
301 	/* To allow for in-flight requests */
302 	schedule_timeout_uninterruptible(100);
303 
304 	if (wait_for_pending_requests(oct))
305 		dev_err(&oct->pci_dev->dev, "There were pending requests\n");
306 
307 	/* Force all requests waiting to be fetched by OCTEON to complete. */
308 	for (i = 0; i < MAX_OCTEON_INSTR_QUEUES(oct); i++) {
309 		struct octeon_instr_queue *iq;
310 
311 		if (!(oct->io_qmask.iq & BIT_ULL(i)))
312 			continue;
313 		iq = oct->instr_queue[i];
314 
315 		if (atomic_read(&iq->instr_pending)) {
316 			spin_lock_bh(&iq->lock);
317 			iq->fill_cnt = 0;
318 			iq->octeon_read_index = iq->host_write_index;
319 			iq->stats.instr_processed +=
320 				atomic_read(&iq->instr_pending);
321 			lio_process_iq_request_list(oct, iq, 0);
322 			spin_unlock_bh(&iq->lock);
323 		}
324 	}
325 
326 	/* Force all pending ordered list requests to time out. */
327 	lio_process_ordered_list(oct, 1);
328 
329 	/* We do not need to wait for output queue packets to be processed. */
330 }
331 
332 /**
333  * \brief Cleanup PCI AER uncorrectable error status
334  * @param dev Pointer to PCI device
335  */
336 static void cleanup_aer_uncorrect_error_status(struct pci_dev *dev)
337 {
338 	int pos = 0x100;
339 	u32 status, mask;
340 
341 	pr_info("%s :\n", __func__);
342 
343 	pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
344 	pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, &mask);
345 	if (dev->error_state == pci_channel_io_normal)
346 		status &= ~mask;        /* Clear corresponding nonfatal bits */
347 	else
348 		status &= mask;         /* Clear corresponding fatal bits */
349 	pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
350 }
351 
352 /**
353  * \brief Stop all PCI IO to a given device
354  * @param dev Pointer to Octeon device
355  */
356 static void stop_pci_io(struct octeon_device *oct)
357 {
358 	/* No more instructions will be forwarded. */
359 	atomic_set(&oct->status, OCT_DEV_IN_RESET);
360 
361 	pci_disable_device(oct->pci_dev);
362 
363 	/* Disable interrupts  */
364 	oct->fn_list.disable_interrupt(oct, OCTEON_ALL_INTR);
365 
366 	pcierror_quiesce_device(oct);
367 
368 	/* Release the interrupt line */
369 	free_irq(oct->pci_dev->irq, oct);
370 
371 	if (oct->flags & LIO_FLAG_MSI_ENABLED)
372 		pci_disable_msi(oct->pci_dev);
373 
374 	dev_dbg(&oct->pci_dev->dev, "Device state is now %s\n",
375 		lio_get_state_string(&oct->status));
376 
377 	/* making it a common function for all OCTEON models */
378 	cleanup_aer_uncorrect_error_status(oct->pci_dev);
379 }
380 
381 /**
382  * \brief called when PCI error is detected
383  * @param pdev Pointer to PCI device
384  * @param state The current pci connection state
385  *
386  * This function is called after a PCI bus error affecting
387  * this device has been detected.
388  */
389 static pci_ers_result_t liquidio_pcie_error_detected(struct pci_dev *pdev,
390 						     pci_channel_state_t state)
391 {
392 	struct octeon_device *oct = pci_get_drvdata(pdev);
393 
394 	/* Non-correctable Non-fatal errors */
395 	if (state == pci_channel_io_normal) {
396 		dev_err(&oct->pci_dev->dev, "Non-correctable non-fatal error reported:\n");
397 		cleanup_aer_uncorrect_error_status(oct->pci_dev);
398 		return PCI_ERS_RESULT_CAN_RECOVER;
399 	}
400 
401 	/* Non-correctable Fatal errors */
402 	dev_err(&oct->pci_dev->dev, "Non-correctable FATAL reported by PCI AER driver\n");
403 	stop_pci_io(oct);
404 
405 	/* Always return a DISCONNECT. There is no support for recovery but only
406 	 * for a clean shutdown.
407 	 */
408 	return PCI_ERS_RESULT_DISCONNECT;
409 }
410 
411 /**
412  * \brief mmio handler
413  * @param pdev Pointer to PCI device
414  */
415 static pci_ers_result_t liquidio_pcie_mmio_enabled(
416 				struct pci_dev *pdev __attribute__((unused)))
417 {
418 	/* We should never hit this since we never ask for a reset for a Fatal
419 	 * Error. We always return DISCONNECT in io_error above.
420 	 * But play safe and return RECOVERED for now.
421 	 */
422 	return PCI_ERS_RESULT_RECOVERED;
423 }
424 
425 /**
426  * \brief called after the pci bus has been reset.
427  * @param pdev Pointer to PCI device
428  *
429  * Restart the card from scratch, as if from a cold-boot. Implementation
430  * resembles the first-half of the octeon_resume routine.
431  */
432 static pci_ers_result_t liquidio_pcie_slot_reset(
433 				struct pci_dev *pdev __attribute__((unused)))
434 {
435 	/* We should never hit this since we never ask for a reset for a Fatal
436 	 * Error. We always return DISCONNECT in io_error above.
437 	 * But play safe and return RECOVERED for now.
438 	 */
439 	return PCI_ERS_RESULT_RECOVERED;
440 }
441 
442 /**
443  * \brief called when traffic can start flowing again.
444  * @param pdev Pointer to PCI device
445  *
446  * This callback is called when the error recovery driver tells us that
447  * its OK to resume normal operation. Implementation resembles the
448  * second-half of the octeon_resume routine.
449  */
450 static void liquidio_pcie_resume(struct pci_dev *pdev __attribute__((unused)))
451 {
452 	/* Nothing to be done here. */
453 }
454 
455 #ifdef CONFIG_PM
456 /**
457  * \brief called when suspending
458  * @param pdev Pointer to PCI device
459  * @param state state to suspend to
460  */
461 static int liquidio_suspend(struct pci_dev *pdev __attribute__((unused)),
462 			    pm_message_t state __attribute__((unused)))
463 {
464 	return 0;
465 }
466 
467 /**
468  * \brief called when resuming
469  * @param pdev Pointer to PCI device
470  */
471 static int liquidio_resume(struct pci_dev *pdev __attribute__((unused)))
472 {
473 	return 0;
474 }
475 #endif
476 
477 /* For PCI-E Advanced Error Recovery (AER) Interface */
478 static const struct pci_error_handlers liquidio_err_handler = {
479 	.error_detected = liquidio_pcie_error_detected,
480 	.mmio_enabled	= liquidio_pcie_mmio_enabled,
481 	.slot_reset	= liquidio_pcie_slot_reset,
482 	.resume		= liquidio_pcie_resume,
483 };
484 
485 static const struct pci_device_id liquidio_pci_tbl[] = {
486 	{       /* 68xx */
487 		PCI_VENDOR_ID_CAVIUM, 0x91, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0
488 	},
489 	{       /* 66xx */
490 		PCI_VENDOR_ID_CAVIUM, 0x92, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0
491 	},
492 	{       /* 23xx pf */
493 		PCI_VENDOR_ID_CAVIUM, 0x9702, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0
494 	},
495 	{
496 		0, 0, 0, 0, 0, 0, 0
497 	}
498 };
499 MODULE_DEVICE_TABLE(pci, liquidio_pci_tbl);
500 
501 static struct pci_driver liquidio_pci_driver = {
502 	.name		= "LiquidIO",
503 	.id_table	= liquidio_pci_tbl,
504 	.probe		= liquidio_probe,
505 	.remove		= liquidio_remove,
506 	.err_handler	= &liquidio_err_handler,    /* For AER */
507 
508 #ifdef CONFIG_PM
509 	.suspend	= liquidio_suspend,
510 	.resume		= liquidio_resume,
511 #endif
512 #ifdef CONFIG_PCI_IOV
513 	.sriov_configure = liquidio_enable_sriov,
514 #endif
515 };
516 
517 /**
518  * \brief register PCI driver
519  */
520 static int liquidio_init_pci(void)
521 {
522 	return pci_register_driver(&liquidio_pci_driver);
523 }
524 
525 /**
526  * \brief unregister PCI driver
527  */
528 static void liquidio_deinit_pci(void)
529 {
530 	pci_unregister_driver(&liquidio_pci_driver);
531 }
532 
533 /**
534  * \brief check interface state
535  * @param lio per-network private data
536  * @param state_flag flag state to check
537  */
538 static inline int ifstate_check(struct lio *lio, int state_flag)
539 {
540 	return atomic_read(&lio->ifstate) & state_flag;
541 }
542 
543 /**
544  * \brief set interface state
545  * @param lio per-network private data
546  * @param state_flag flag state to set
547  */
548 static inline void ifstate_set(struct lio *lio, int state_flag)
549 {
550 	atomic_set(&lio->ifstate, (atomic_read(&lio->ifstate) | state_flag));
551 }
552 
553 /**
554  * \brief clear interface state
555  * @param lio per-network private data
556  * @param state_flag flag state to clear
557  */
558 static inline void ifstate_reset(struct lio *lio, int state_flag)
559 {
560 	atomic_set(&lio->ifstate, (atomic_read(&lio->ifstate) & ~(state_flag)));
561 }
562 
563 /**
564  * \brief Stop Tx queues
565  * @param netdev network device
566  */
567 static inline void txqs_stop(struct net_device *netdev)
568 {
569 	if (netif_is_multiqueue(netdev)) {
570 		int i;
571 
572 		for (i = 0; i < netdev->num_tx_queues; i++)
573 			netif_stop_subqueue(netdev, i);
574 	} else {
575 		netif_stop_queue(netdev);
576 	}
577 }
578 
579 /**
580  * \brief Start Tx queues
581  * @param netdev network device
582  */
583 static inline void txqs_start(struct net_device *netdev)
584 {
585 	if (netif_is_multiqueue(netdev)) {
586 		int i;
587 
588 		for (i = 0; i < netdev->num_tx_queues; i++)
589 			netif_start_subqueue(netdev, i);
590 	} else {
591 		netif_start_queue(netdev);
592 	}
593 }
594 
595 /**
596  * \brief Wake Tx queues
597  * @param netdev network device
598  */
599 static inline void txqs_wake(struct net_device *netdev)
600 {
601 	struct lio *lio = GET_LIO(netdev);
602 
603 	if (netif_is_multiqueue(netdev)) {
604 		int i;
605 
606 		for (i = 0; i < netdev->num_tx_queues; i++) {
607 			int qno = lio->linfo.txpciq[i %
608 				(lio->linfo.num_txpciq)].s.q_no;
609 
610 			if (__netif_subqueue_stopped(netdev, i)) {
611 				INCR_INSTRQUEUE_PKT_COUNT(lio->oct_dev, qno,
612 							  tx_restart, 1);
613 				netif_wake_subqueue(netdev, i);
614 			}
615 		}
616 	} else {
617 		INCR_INSTRQUEUE_PKT_COUNT(lio->oct_dev, lio->txq,
618 					  tx_restart, 1);
619 		netif_wake_queue(netdev);
620 	}
621 }
622 
623 /**
624  * \brief Stop Tx queue
625  * @param netdev network device
626  */
627 static void stop_txq(struct net_device *netdev)
628 {
629 	txqs_stop(netdev);
630 }
631 
632 /**
633  * \brief Start Tx queue
634  * @param netdev network device
635  */
636 static void start_txq(struct net_device *netdev)
637 {
638 	struct lio *lio = GET_LIO(netdev);
639 
640 	if (lio->linfo.link.s.link_up) {
641 		txqs_start(netdev);
642 		return;
643 	}
644 }
645 
646 /**
647  * \brief Wake a queue
648  * @param netdev network device
649  * @param q which queue to wake
650  */
651 static inline void wake_q(struct net_device *netdev, int q)
652 {
653 	if (netif_is_multiqueue(netdev))
654 		netif_wake_subqueue(netdev, q);
655 	else
656 		netif_wake_queue(netdev);
657 }
658 
659 /**
660  * \brief Stop a queue
661  * @param netdev network device
662  * @param q which queue to stop
663  */
664 static inline void stop_q(struct net_device *netdev, int q)
665 {
666 	if (netif_is_multiqueue(netdev))
667 		netif_stop_subqueue(netdev, q);
668 	else
669 		netif_stop_queue(netdev);
670 }
671 
672 /**
673  * \brief Check Tx queue status, and take appropriate action
674  * @param lio per-network private data
675  * @returns 0 if full, number of queues woken up otherwise
676  */
677 static inline int check_txq_status(struct lio *lio)
678 {
679 	int ret_val = 0;
680 
681 	if (netif_is_multiqueue(lio->netdev)) {
682 		int numqs = lio->netdev->num_tx_queues;
683 		int q, iq = 0;
684 
685 		/* check each sub-queue state */
686 		for (q = 0; q < numqs; q++) {
687 			iq = lio->linfo.txpciq[q %
688 				(lio->linfo.num_txpciq)].s.q_no;
689 			if (octnet_iq_is_full(lio->oct_dev, iq))
690 				continue;
691 			if (__netif_subqueue_stopped(lio->netdev, q)) {
692 				wake_q(lio->netdev, q);
693 				INCR_INSTRQUEUE_PKT_COUNT(lio->oct_dev, iq,
694 							  tx_restart, 1);
695 				ret_val++;
696 			}
697 		}
698 	} else {
699 		if (octnet_iq_is_full(lio->oct_dev, lio->txq))
700 			return 0;
701 		wake_q(lio->netdev, lio->txq);
702 		INCR_INSTRQUEUE_PKT_COUNT(lio->oct_dev, lio->txq,
703 					  tx_restart, 1);
704 		ret_val = 1;
705 	}
706 	return ret_val;
707 }
708 
709 /**
710  * Remove the node at the head of the list. The list would be empty at
711  * the end of this call if there are no more nodes in the list.
712  */
713 static inline struct list_head *list_delete_head(struct list_head *root)
714 {
715 	struct list_head *node;
716 
717 	if ((root->prev == root) && (root->next == root))
718 		node = NULL;
719 	else
720 		node = root->next;
721 
722 	if (node)
723 		list_del(node);
724 
725 	return node;
726 }
727 
728 /**
729  * \brief Delete gather lists
730  * @param lio per-network private data
731  */
732 static void delete_glists(struct lio *lio)
733 {
734 	struct octnic_gather *g;
735 	int i;
736 
737 	if (!lio->glist)
738 		return;
739 
740 	for (i = 0; i < lio->linfo.num_txpciq; i++) {
741 		do {
742 			g = (struct octnic_gather *)
743 				list_delete_head(&lio->glist[i]);
744 			if (g) {
745 				if (g->sg) {
746 					dma_unmap_single(&lio->oct_dev->
747 							 pci_dev->dev,
748 							 g->sg_dma_ptr,
749 							 g->sg_size,
750 							 DMA_TO_DEVICE);
751 					kfree((void *)((unsigned long)g->sg -
752 						       g->adjust));
753 				}
754 				kfree(g);
755 			}
756 		} while (g);
757 	}
758 
759 	kfree((void *)lio->glist);
760 	kfree((void *)lio->glist_lock);
761 }
762 
763 /**
764  * \brief Setup gather lists
765  * @param lio per-network private data
766  */
767 static int setup_glists(struct octeon_device *oct, struct lio *lio, int num_iqs)
768 {
769 	int i, j;
770 	struct octnic_gather *g;
771 
772 	lio->glist_lock = kcalloc(num_iqs, sizeof(*lio->glist_lock),
773 				  GFP_KERNEL);
774 	if (!lio->glist_lock)
775 		return 1;
776 
777 	lio->glist = kcalloc(num_iqs, sizeof(*lio->glist),
778 			     GFP_KERNEL);
779 	if (!lio->glist) {
780 		kfree((void *)lio->glist_lock);
781 		return 1;
782 	}
783 
784 	for (i = 0; i < num_iqs; i++) {
785 		int numa_node = cpu_to_node(i % num_online_cpus());
786 
787 		spin_lock_init(&lio->glist_lock[i]);
788 
789 		INIT_LIST_HEAD(&lio->glist[i]);
790 
791 		for (j = 0; j < lio->tx_qsize; j++) {
792 			g = kzalloc_node(sizeof(*g), GFP_KERNEL,
793 					 numa_node);
794 			if (!g)
795 				g = kzalloc(sizeof(*g), GFP_KERNEL);
796 			if (!g)
797 				break;
798 
799 			g->sg_size = ((ROUNDUP4(OCTNIC_MAX_SG) >> 2) *
800 				      OCT_SG_ENTRY_SIZE);
801 
802 			g->sg = kmalloc_node(g->sg_size + 8,
803 					     GFP_KERNEL, numa_node);
804 			if (!g->sg)
805 				g->sg = kmalloc(g->sg_size + 8, GFP_KERNEL);
806 			if (!g->sg) {
807 				kfree(g);
808 				break;
809 			}
810 
811 			/* The gather component should be aligned on 64-bit
812 			 * boundary
813 			 */
814 			if (((unsigned long)g->sg) & 7) {
815 				g->adjust = 8 - (((unsigned long)g->sg) & 7);
816 				g->sg = (struct octeon_sg_entry *)
817 					((unsigned long)g->sg + g->adjust);
818 			}
819 			g->sg_dma_ptr = dma_map_single(&oct->pci_dev->dev,
820 						       g->sg, g->sg_size,
821 						       DMA_TO_DEVICE);
822 			if (dma_mapping_error(&oct->pci_dev->dev,
823 					      g->sg_dma_ptr)) {
824 				kfree((void *)((unsigned long)g->sg -
825 					       g->adjust));
826 				kfree(g);
827 				break;
828 			}
829 
830 			list_add_tail(&g->list, &lio->glist[i]);
831 		}
832 
833 		if (j != lio->tx_qsize) {
834 			delete_glists(lio);
835 			return 1;
836 		}
837 	}
838 
839 	return 0;
840 }
841 
842 /**
843  * \brief Print link information
844  * @param netdev network device
845  */
846 static void print_link_info(struct net_device *netdev)
847 {
848 	struct lio *lio = GET_LIO(netdev);
849 
850 	if (atomic_read(&lio->ifstate) & LIO_IFSTATE_REGISTERED) {
851 		struct oct_link_info *linfo = &lio->linfo;
852 
853 		if (linfo->link.s.link_up) {
854 			netif_info(lio, link, lio->netdev, "%d Mbps %s Duplex UP\n",
855 				   linfo->link.s.speed,
856 				   (linfo->link.s.duplex) ? "Full" : "Half");
857 		} else {
858 			netif_info(lio, link, lio->netdev, "Link Down\n");
859 		}
860 	}
861 }
862 
863 /**
864  * \brief Routine to notify MTU change
865  * @param work work_struct data structure
866  */
867 static void octnet_link_status_change(struct work_struct *work)
868 {
869 	struct cavium_wk *wk = (struct cavium_wk *)work;
870 	struct lio *lio = (struct lio *)wk->ctxptr;
871 
872 	rtnl_lock();
873 	call_netdevice_notifiers(NETDEV_CHANGEMTU, lio->netdev);
874 	rtnl_unlock();
875 }
876 
877 /**
878  * \brief Sets up the mtu status change work
879  * @param netdev network device
880  */
881 static inline int setup_link_status_change_wq(struct net_device *netdev)
882 {
883 	struct lio *lio = GET_LIO(netdev);
884 	struct octeon_device *oct = lio->oct_dev;
885 
886 	lio->link_status_wq.wq = alloc_workqueue("link-status",
887 						 WQ_MEM_RECLAIM, 0);
888 	if (!lio->link_status_wq.wq) {
889 		dev_err(&oct->pci_dev->dev, "unable to create cavium link status wq\n");
890 		return -1;
891 	}
892 	INIT_DELAYED_WORK(&lio->link_status_wq.wk.work,
893 			  octnet_link_status_change);
894 	lio->link_status_wq.wk.ctxptr = lio;
895 
896 	return 0;
897 }
898 
899 static inline void cleanup_link_status_change_wq(struct net_device *netdev)
900 {
901 	struct lio *lio = GET_LIO(netdev);
902 
903 	if (lio->link_status_wq.wq) {
904 		cancel_delayed_work_sync(&lio->link_status_wq.wk.work);
905 		destroy_workqueue(lio->link_status_wq.wq);
906 	}
907 }
908 
909 /**
910  * \brief Update link status
911  * @param netdev network device
912  * @param ls link status structure
913  *
914  * Called on receipt of a link status response from the core application to
915  * update each interface's link status.
916  */
917 static inline void update_link_status(struct net_device *netdev,
918 				      union oct_link_status *ls)
919 {
920 	struct lio *lio = GET_LIO(netdev);
921 	int changed = (lio->linfo.link.u64 != ls->u64);
922 
923 	lio->linfo.link.u64 = ls->u64;
924 
925 	if ((lio->intf_open) && (changed)) {
926 		print_link_info(netdev);
927 		lio->link_changes++;
928 
929 		if (lio->linfo.link.s.link_up) {
930 			netif_carrier_on(netdev);
931 			txqs_wake(netdev);
932 		} else {
933 			netif_carrier_off(netdev);
934 			stop_txq(netdev);
935 		}
936 	}
937 }
938 
939 /* Runs in interrupt context. */
940 static void update_txq_status(struct octeon_device *oct, int iq_num)
941 {
942 	struct net_device *netdev;
943 	struct lio *lio;
944 	struct octeon_instr_queue *iq = oct->instr_queue[iq_num];
945 
946 	netdev = oct->props[iq->ifidx].netdev;
947 
948 	/* This is needed because the first IQ does not have
949 	 * a netdev associated with it.
950 	 */
951 	if (!netdev)
952 		return;
953 
954 	lio = GET_LIO(netdev);
955 	if (netif_is_multiqueue(netdev)) {
956 		if (__netif_subqueue_stopped(netdev, iq->q_index) &&
957 		    lio->linfo.link.s.link_up &&
958 		    (!octnet_iq_is_full(oct, iq_num))) {
959 			INCR_INSTRQUEUE_PKT_COUNT(lio->oct_dev, iq_num,
960 						  tx_restart, 1);
961 			netif_wake_subqueue(netdev, iq->q_index);
962 		} else {
963 			if (!octnet_iq_is_full(oct, lio->txq)) {
964 				INCR_INSTRQUEUE_PKT_COUNT(lio->oct_dev,
965 							  lio->txq,
966 							  tx_restart, 1);
967 				wake_q(netdev, lio->txq);
968 			}
969 		}
970 	}
971 }
972 
973 static
974 int liquidio_schedule_msix_droq_pkt_handler(struct octeon_droq *droq, u64 ret)
975 {
976 	struct octeon_device *oct = droq->oct_dev;
977 	struct octeon_device_priv *oct_priv =
978 	    (struct octeon_device_priv *)oct->priv;
979 
980 	if (droq->ops.poll_mode) {
981 		droq->ops.napi_fn(droq);
982 	} else {
983 		if (ret & MSIX_PO_INT) {
984 			tasklet_schedule(&oct_priv->droq_tasklet);
985 			return 1;
986 		}
987 		/* this will be flushed periodically by check iq db */
988 		if (ret & MSIX_PI_INT)
989 			return 0;
990 	}
991 	return 0;
992 }
993 
994 /**
995  * \brief Droq packet processor sceduler
996  * @param oct octeon device
997  */
998 static void liquidio_schedule_droq_pkt_handlers(struct octeon_device *oct)
999 {
1000 	struct octeon_device_priv *oct_priv =
1001 		(struct octeon_device_priv *)oct->priv;
1002 	u64 oq_no;
1003 	struct octeon_droq *droq;
1004 
1005 	if (oct->int_status & OCT_DEV_INTR_PKT_DATA) {
1006 		for (oq_no = 0; oq_no < MAX_OCTEON_OUTPUT_QUEUES(oct);
1007 		     oq_no++) {
1008 			if (!(oct->droq_intr & BIT_ULL(oq_no)))
1009 				continue;
1010 
1011 			droq = oct->droq[oq_no];
1012 
1013 			if (droq->ops.poll_mode) {
1014 				droq->ops.napi_fn(droq);
1015 				oct_priv->napi_mask |= (1 << oq_no);
1016 			} else {
1017 				tasklet_schedule(&oct_priv->droq_tasklet);
1018 			}
1019 		}
1020 	}
1021 }
1022 
1023 static irqreturn_t
1024 liquidio_msix_intr_handler(int irq __attribute__((unused)), void *dev)
1025 {
1026 	u64 ret;
1027 	struct octeon_ioq_vector *ioq_vector = (struct octeon_ioq_vector *)dev;
1028 	struct octeon_device *oct = ioq_vector->oct_dev;
1029 	struct octeon_droq *droq = oct->droq[ioq_vector->droq_index];
1030 
1031 	ret = oct->fn_list.msix_interrupt_handler(ioq_vector);
1032 
1033 	if ((ret & MSIX_PO_INT) || (ret & MSIX_PI_INT))
1034 		liquidio_schedule_msix_droq_pkt_handler(droq, ret);
1035 
1036 	return IRQ_HANDLED;
1037 }
1038 
1039 /**
1040  * \brief Interrupt handler for octeon
1041  * @param irq unused
1042  * @param dev octeon device
1043  */
1044 static
1045 irqreturn_t liquidio_legacy_intr_handler(int irq __attribute__((unused)),
1046 					 void *dev)
1047 {
1048 	struct octeon_device *oct = (struct octeon_device *)dev;
1049 	irqreturn_t ret;
1050 
1051 	/* Disable our interrupts for the duration of ISR */
1052 	oct->fn_list.disable_interrupt(oct, OCTEON_ALL_INTR);
1053 
1054 	ret = oct->fn_list.process_interrupt_regs(oct);
1055 
1056 	if (ret == IRQ_HANDLED)
1057 		liquidio_schedule_droq_pkt_handlers(oct);
1058 
1059 	/* Re-enable our interrupts  */
1060 	if (!(atomic_read(&oct->status) == OCT_DEV_IN_RESET))
1061 		oct->fn_list.enable_interrupt(oct, OCTEON_ALL_INTR);
1062 
1063 	return ret;
1064 }
1065 
1066 /**
1067  * \brief Setup interrupt for octeon device
1068  * @param oct octeon device
1069  *
1070  *  Enable interrupt in Octeon device as given in the PCI interrupt mask.
1071  */
1072 static int octeon_setup_interrupt(struct octeon_device *oct)
1073 {
1074 	int irqret, err;
1075 	struct msix_entry *msix_entries;
1076 	int i;
1077 	int num_ioq_vectors;
1078 	int num_alloc_ioq_vectors;
1079 
1080 	if (OCTEON_CN23XX_PF(oct) && oct->msix_on) {
1081 		oct->num_msix_irqs = oct->sriov_info.num_pf_rings;
1082 		/* one non ioq interrupt for handling sli_mac_pf_int_sum */
1083 		oct->num_msix_irqs += 1;
1084 
1085 		oct->msix_entries = kcalloc(
1086 		    oct->num_msix_irqs, sizeof(struct msix_entry), GFP_KERNEL);
1087 		if (!oct->msix_entries)
1088 			return 1;
1089 
1090 		msix_entries = (struct msix_entry *)oct->msix_entries;
1091 		/*Assumption is that pf msix vectors start from pf srn to pf to
1092 		 * trs and not from 0. if not change this code
1093 		 */
1094 		for (i = 0; i < oct->num_msix_irqs - 1; i++)
1095 			msix_entries[i].entry = oct->sriov_info.pf_srn + i;
1096 		msix_entries[oct->num_msix_irqs - 1].entry =
1097 		    oct->sriov_info.trs;
1098 		num_alloc_ioq_vectors = pci_enable_msix_range(
1099 						oct->pci_dev, msix_entries,
1100 						oct->num_msix_irqs,
1101 						oct->num_msix_irqs);
1102 		if (num_alloc_ioq_vectors < 0) {
1103 			dev_err(&oct->pci_dev->dev, "unable to Allocate MSI-X interrupts\n");
1104 			kfree(oct->msix_entries);
1105 			oct->msix_entries = NULL;
1106 			return 1;
1107 		}
1108 		dev_dbg(&oct->pci_dev->dev, "OCTEON: Enough MSI-X interrupts are allocated...\n");
1109 
1110 		num_ioq_vectors = oct->num_msix_irqs;
1111 
1112 		/** For PF, there is one non-ioq interrupt handler */
1113 		num_ioq_vectors -= 1;
1114 		irqret = request_irq(msix_entries[num_ioq_vectors].vector,
1115 				     liquidio_legacy_intr_handler, 0, "octeon",
1116 				     oct);
1117 		if (irqret) {
1118 			dev_err(&oct->pci_dev->dev,
1119 				"OCTEON: Request_irq failed for MSIX interrupt Error: %d\n",
1120 				irqret);
1121 			pci_disable_msix(oct->pci_dev);
1122 			kfree(oct->msix_entries);
1123 			oct->msix_entries = NULL;
1124 			return 1;
1125 		}
1126 
1127 		for (i = 0; i < num_ioq_vectors; i++) {
1128 			irqret = request_irq(msix_entries[i].vector,
1129 					     liquidio_msix_intr_handler, 0,
1130 					     "octeon", &oct->ioq_vector[i]);
1131 			if (irqret) {
1132 				dev_err(&oct->pci_dev->dev,
1133 					"OCTEON: Request_irq failed for MSIX interrupt Error: %d\n",
1134 					irqret);
1135 				/** Freeing the non-ioq irq vector here . */
1136 				free_irq(msix_entries[num_ioq_vectors].vector,
1137 					 oct);
1138 
1139 				while (i) {
1140 					i--;
1141 					/** clearing affinity mask. */
1142 					irq_set_affinity_hint(
1143 						msix_entries[i].vector, NULL);
1144 					free_irq(msix_entries[i].vector,
1145 						 &oct->ioq_vector[i]);
1146 				}
1147 				pci_disable_msix(oct->pci_dev);
1148 				kfree(oct->msix_entries);
1149 				oct->msix_entries = NULL;
1150 				return 1;
1151 			}
1152 			oct->ioq_vector[i].vector = msix_entries[i].vector;
1153 			/* assign the cpu mask for this msix interrupt vector */
1154 			irq_set_affinity_hint(
1155 					msix_entries[i].vector,
1156 					(&oct->ioq_vector[i].affinity_mask));
1157 		}
1158 		dev_dbg(&oct->pci_dev->dev, "OCTEON[%d]: MSI-X enabled\n",
1159 			oct->octeon_id);
1160 	} else {
1161 		err = pci_enable_msi(oct->pci_dev);
1162 		if (err)
1163 			dev_warn(&oct->pci_dev->dev, "Reverting to legacy interrupts. Error: %d\n",
1164 				 err);
1165 		else
1166 			oct->flags |= LIO_FLAG_MSI_ENABLED;
1167 
1168 		irqret = request_irq(oct->pci_dev->irq,
1169 				     liquidio_legacy_intr_handler, IRQF_SHARED,
1170 				     "octeon", oct);
1171 		if (irqret) {
1172 			if (oct->flags & LIO_FLAG_MSI_ENABLED)
1173 				pci_disable_msi(oct->pci_dev);
1174 			dev_err(&oct->pci_dev->dev, "Request IRQ failed with code: %d\n",
1175 				irqret);
1176 			return 1;
1177 		}
1178 	}
1179 	return 0;
1180 }
1181 
1182 static int liquidio_watchdog(void *param)
1183 {
1184 	u64 wdog;
1185 	u16 mask_of_stuck_cores = 0;
1186 	u16 mask_of_crashed_cores = 0;
1187 	int core_num;
1188 	u8 core_is_stuck[LIO_MAX_CORES];
1189 	u8 core_crashed[LIO_MAX_CORES];
1190 	struct octeon_device *oct = param;
1191 
1192 	memset(core_is_stuck, 0, sizeof(core_is_stuck));
1193 	memset(core_crashed, 0, sizeof(core_crashed));
1194 
1195 	while (!kthread_should_stop()) {
1196 		mask_of_crashed_cores =
1197 		    (u16)octeon_read_csr64(oct, CN23XX_SLI_SCRATCH2);
1198 
1199 		for (core_num = 0; core_num < LIO_MAX_CORES; core_num++) {
1200 			if (!core_is_stuck[core_num]) {
1201 				wdog = lio_pci_readq(oct, CIU3_WDOG(core_num));
1202 
1203 				/* look at watchdog state field */
1204 				wdog &= CIU3_WDOG_MASK;
1205 				if (wdog) {
1206 					/* this watchdog timer has expired */
1207 					core_is_stuck[core_num] =
1208 						LIO_MONITOR_WDOG_EXPIRE;
1209 					mask_of_stuck_cores |= (1 << core_num);
1210 				}
1211 			}
1212 
1213 			if (!core_crashed[core_num])
1214 				core_crashed[core_num] =
1215 				    (mask_of_crashed_cores >> core_num) & 1;
1216 		}
1217 
1218 		if (mask_of_stuck_cores) {
1219 			for (core_num = 0; core_num < LIO_MAX_CORES;
1220 			     core_num++) {
1221 				if (core_is_stuck[core_num] == 1) {
1222 					dev_err(&oct->pci_dev->dev,
1223 						"ERROR: Octeon core %d is stuck!\n",
1224 						core_num);
1225 					/* 2 means we have printk'd  an error
1226 					 * so no need to repeat the same printk
1227 					 */
1228 					core_is_stuck[core_num] =
1229 						LIO_MONITOR_CORE_STUCK_MSGD;
1230 				}
1231 			}
1232 		}
1233 
1234 		if (mask_of_crashed_cores) {
1235 			for (core_num = 0; core_num < LIO_MAX_CORES;
1236 			     core_num++) {
1237 				if (core_crashed[core_num] == 1) {
1238 					dev_err(&oct->pci_dev->dev,
1239 						"ERROR: Octeon core %d crashed!  See oct-fwdump for details.\n",
1240 						core_num);
1241 					/* 2 means we have printk'd  an error
1242 					 * so no need to repeat the same printk
1243 					 */
1244 					core_crashed[core_num] =
1245 						LIO_MONITOR_CORE_STUCK_MSGD;
1246 				}
1247 			}
1248 		}
1249 #ifdef CONFIG_MODULE_UNLOAD
1250 		if (mask_of_stuck_cores || mask_of_crashed_cores) {
1251 			/* make module refcount=0 so that rmmod will work */
1252 			long refcount;
1253 
1254 			refcount = module_refcount(THIS_MODULE);
1255 
1256 			while (refcount > 0) {
1257 				module_put(THIS_MODULE);
1258 				refcount = module_refcount(THIS_MODULE);
1259 			}
1260 
1261 			/* compensate for and withstand an unlikely (but still
1262 			 * possible) race condition
1263 			 */
1264 			while (refcount < 0) {
1265 				try_module_get(THIS_MODULE);
1266 				refcount = module_refcount(THIS_MODULE);
1267 			}
1268 		}
1269 #endif
1270 		/* sleep for two seconds */
1271 		set_current_state(TASK_INTERRUPTIBLE);
1272 		schedule_timeout(2 * HZ);
1273 	}
1274 
1275 	return 0;
1276 }
1277 
1278 /**
1279  * \brief PCI probe handler
1280  * @param pdev PCI device structure
1281  * @param ent unused
1282  */
1283 static int
1284 liquidio_probe(struct pci_dev *pdev,
1285 	       const struct pci_device_id *ent __attribute__((unused)))
1286 {
1287 	struct octeon_device *oct_dev = NULL;
1288 	struct handshake *hs;
1289 
1290 	oct_dev = octeon_allocate_device(pdev->device,
1291 					 sizeof(struct octeon_device_priv));
1292 	if (!oct_dev) {
1293 		dev_err(&pdev->dev, "Unable to allocate device\n");
1294 		return -ENOMEM;
1295 	}
1296 
1297 	if (pdev->device == OCTEON_CN23XX_PF_VID)
1298 		oct_dev->msix_on = LIO_FLAG_MSIX_ENABLED;
1299 
1300 	dev_info(&pdev->dev, "Initializing device %x:%x.\n",
1301 		 (u32)pdev->vendor, (u32)pdev->device);
1302 
1303 	/* Assign octeon_device for this device to the private data area. */
1304 	pci_set_drvdata(pdev, oct_dev);
1305 
1306 	/* set linux specific device pointer */
1307 	oct_dev->pci_dev = (void *)pdev;
1308 
1309 	hs = &handshake[oct_dev->octeon_id];
1310 	init_completion(&hs->init);
1311 	init_completion(&hs->started);
1312 	hs->pci_dev = pdev;
1313 
1314 	if (oct_dev->octeon_id == 0)
1315 		/* first LiquidIO NIC is detected */
1316 		complete(&first_stage);
1317 
1318 	if (octeon_device_init(oct_dev)) {
1319 		complete(&hs->init);
1320 		liquidio_remove(pdev);
1321 		return -ENOMEM;
1322 	}
1323 
1324 	if (OCTEON_CN23XX_PF(oct_dev)) {
1325 		u64 scratch1;
1326 		u8 bus, device, function;
1327 
1328 		scratch1 = octeon_read_csr64(oct_dev, CN23XX_SLI_SCRATCH1);
1329 		if (!(scratch1 & 4ULL)) {
1330 			/* Bit 2 of SLI_SCRATCH_1 is a flag that indicates that
1331 			 * the lio watchdog kernel thread is running for this
1332 			 * NIC.  Each NIC gets one watchdog kernel thread.
1333 			 */
1334 			scratch1 |= 4ULL;
1335 			octeon_write_csr64(oct_dev, CN23XX_SLI_SCRATCH1,
1336 					   scratch1);
1337 
1338 			bus = pdev->bus->number;
1339 			device = PCI_SLOT(pdev->devfn);
1340 			function = PCI_FUNC(pdev->devfn);
1341 			oct_dev->watchdog_task = kthread_create(
1342 			    liquidio_watchdog, oct_dev,
1343 			    "liowd/%02hhx:%02hhx.%hhx", bus, device, function);
1344 			if (!IS_ERR(oct_dev->watchdog_task)) {
1345 				wake_up_process(oct_dev->watchdog_task);
1346 			} else {
1347 				oct_dev->watchdog_task = NULL;
1348 				dev_err(&oct_dev->pci_dev->dev,
1349 					"failed to create kernel_thread\n");
1350 				liquidio_remove(pdev);
1351 				return -1;
1352 			}
1353 		}
1354 	}
1355 
1356 	oct_dev->rx_pause = 1;
1357 	oct_dev->tx_pause = 1;
1358 
1359 	dev_dbg(&oct_dev->pci_dev->dev, "Device is ready\n");
1360 
1361 	return 0;
1362 }
1363 
1364 /**
1365  *\brief Destroy resources associated with octeon device
1366  * @param pdev PCI device structure
1367  * @param ent unused
1368  */
1369 static void octeon_destroy_resources(struct octeon_device *oct)
1370 {
1371 	int i;
1372 	struct msix_entry *msix_entries;
1373 	struct octeon_device_priv *oct_priv =
1374 		(struct octeon_device_priv *)oct->priv;
1375 
1376 	struct handshake *hs;
1377 
1378 	switch (atomic_read(&oct->status)) {
1379 	case OCT_DEV_RUNNING:
1380 	case OCT_DEV_CORE_OK:
1381 
1382 		/* No more instructions will be forwarded. */
1383 		atomic_set(&oct->status, OCT_DEV_IN_RESET);
1384 
1385 		oct->app_mode = CVM_DRV_INVALID_APP;
1386 		dev_dbg(&oct->pci_dev->dev, "Device state is now %s\n",
1387 			lio_get_state_string(&oct->status));
1388 
1389 		schedule_timeout_uninterruptible(HZ / 10);
1390 
1391 		/* fallthrough */
1392 	case OCT_DEV_HOST_OK:
1393 
1394 		/* fallthrough */
1395 	case OCT_DEV_CONSOLE_INIT_DONE:
1396 		/* Remove any consoles */
1397 		octeon_remove_consoles(oct);
1398 
1399 		/* fallthrough */
1400 	case OCT_DEV_IO_QUEUES_DONE:
1401 		if (wait_for_pending_requests(oct))
1402 			dev_err(&oct->pci_dev->dev, "There were pending requests\n");
1403 
1404 		if (lio_wait_for_instr_fetch(oct))
1405 			dev_err(&oct->pci_dev->dev, "IQ had pending instructions\n");
1406 
1407 		/* Disable the input and output queues now. No more packets will
1408 		 * arrive from Octeon, but we should wait for all packet
1409 		 * processing to finish.
1410 		 */
1411 		oct->fn_list.disable_io_queues(oct);
1412 
1413 		if (lio_wait_for_oq_pkts(oct))
1414 			dev_err(&oct->pci_dev->dev, "OQ had pending packets\n");
1415 
1416 	/* fallthrough */
1417 	case OCT_DEV_INTR_SET_DONE:
1418 		/* Disable interrupts  */
1419 		oct->fn_list.disable_interrupt(oct, OCTEON_ALL_INTR);
1420 
1421 		if (oct->msix_on) {
1422 			msix_entries = (struct msix_entry *)oct->msix_entries;
1423 			for (i = 0; i < oct->num_msix_irqs - 1; i++) {
1424 				/* clear the affinity_cpumask */
1425 				irq_set_affinity_hint(msix_entries[i].vector,
1426 						      NULL);
1427 				free_irq(msix_entries[i].vector,
1428 					 &oct->ioq_vector[i]);
1429 			}
1430 			/* non-iov vector's argument is oct struct */
1431 			free_irq(msix_entries[i].vector, oct);
1432 
1433 			pci_disable_msix(oct->pci_dev);
1434 			kfree(oct->msix_entries);
1435 			oct->msix_entries = NULL;
1436 		} else {
1437 			/* Release the interrupt line */
1438 			free_irq(oct->pci_dev->irq, oct);
1439 
1440 			if (oct->flags & LIO_FLAG_MSI_ENABLED)
1441 				pci_disable_msi(oct->pci_dev);
1442 		}
1443 
1444 	/* fallthrough */
1445 	case OCT_DEV_MSIX_ALLOC_VECTOR_DONE:
1446 		if (OCTEON_CN23XX_PF(oct))
1447 			octeon_free_ioq_vector(oct);
1448 
1449 	/* fallthrough */
1450 	case OCT_DEV_MBOX_SETUP_DONE:
1451 		if (OCTEON_CN23XX_PF(oct))
1452 			oct->fn_list.free_mbox(oct);
1453 
1454 	/* fallthrough */
1455 	case OCT_DEV_IN_RESET:
1456 	case OCT_DEV_DROQ_INIT_DONE:
1457 		/* Wait for any pending operations */
1458 		mdelay(100);
1459 		for (i = 0; i < MAX_OCTEON_OUTPUT_QUEUES(oct); i++) {
1460 			if (!(oct->io_qmask.oq & BIT_ULL(i)))
1461 				continue;
1462 			octeon_delete_droq(oct, i);
1463 		}
1464 
1465 		/* Force any pending handshakes to complete */
1466 		for (i = 0; i < MAX_OCTEON_DEVICES; i++) {
1467 			hs = &handshake[i];
1468 
1469 			if (hs->pci_dev) {
1470 				handshake[oct->octeon_id].init_ok = 0;
1471 				complete(&handshake[oct->octeon_id].init);
1472 				handshake[oct->octeon_id].started_ok = 0;
1473 				complete(&handshake[oct->octeon_id].started);
1474 			}
1475 		}
1476 
1477 		/* fallthrough */
1478 	case OCT_DEV_RESP_LIST_INIT_DONE:
1479 		octeon_delete_response_list(oct);
1480 
1481 		/* fallthrough */
1482 	case OCT_DEV_INSTR_QUEUE_INIT_DONE:
1483 		for (i = 0; i < MAX_OCTEON_INSTR_QUEUES(oct); i++) {
1484 			if (!(oct->io_qmask.iq & BIT_ULL(i)))
1485 				continue;
1486 			octeon_delete_instr_queue(oct, i);
1487 		}
1488 #ifdef CONFIG_PCI_IOV
1489 		if (oct->sriov_info.sriov_enabled)
1490 			pci_disable_sriov(oct->pci_dev);
1491 #endif
1492 		/* fallthrough */
1493 	case OCT_DEV_SC_BUFF_POOL_INIT_DONE:
1494 		octeon_free_sc_buffer_pool(oct);
1495 
1496 		/* fallthrough */
1497 	case OCT_DEV_DISPATCH_INIT_DONE:
1498 		octeon_delete_dispatch_list(oct);
1499 		cancel_delayed_work_sync(&oct->nic_poll_work.work);
1500 
1501 		/* fallthrough */
1502 	case OCT_DEV_PCI_MAP_DONE:
1503 		/* Soft reset the octeon device before exiting */
1504 		if ((!OCTEON_CN23XX_PF(oct)) || !oct->octeon_id)
1505 			oct->fn_list.soft_reset(oct);
1506 
1507 		octeon_unmap_pci_barx(oct, 0);
1508 		octeon_unmap_pci_barx(oct, 1);
1509 
1510 		/* fallthrough */
1511 	case OCT_DEV_PCI_ENABLE_DONE:
1512 		pci_clear_master(oct->pci_dev);
1513 		/* Disable the device, releasing the PCI INT */
1514 		pci_disable_device(oct->pci_dev);
1515 
1516 		/* fallthrough */
1517 	case OCT_DEV_BEGIN_STATE:
1518 		/* Nothing to be done here either */
1519 		break;
1520 	}                       /* end switch (oct->status) */
1521 
1522 	tasklet_kill(&oct_priv->droq_tasklet);
1523 }
1524 
1525 /**
1526  * \brief Callback for rx ctrl
1527  * @param status status of request
1528  * @param buf pointer to resp structure
1529  */
1530 static void rx_ctl_callback(struct octeon_device *oct,
1531 			    u32 status,
1532 			    void *buf)
1533 {
1534 	struct octeon_soft_command *sc = (struct octeon_soft_command *)buf;
1535 	struct liquidio_rx_ctl_context *ctx;
1536 
1537 	ctx  = (struct liquidio_rx_ctl_context *)sc->ctxptr;
1538 
1539 	oct = lio_get_device(ctx->octeon_id);
1540 	if (status)
1541 		dev_err(&oct->pci_dev->dev, "rx ctl instruction failed. Status: %llx\n",
1542 			CVM_CAST64(status));
1543 	WRITE_ONCE(ctx->cond, 1);
1544 
1545 	/* This barrier is required to be sure that the response has been
1546 	 * written fully before waking up the handler
1547 	 */
1548 	wmb();
1549 
1550 	wake_up_interruptible(&ctx->wc);
1551 }
1552 
1553 /**
1554  * \brief Send Rx control command
1555  * @param lio per-network private data
1556  * @param start_stop whether to start or stop
1557  */
1558 static void send_rx_ctrl_cmd(struct lio *lio, int start_stop)
1559 {
1560 	struct octeon_soft_command *sc;
1561 	struct liquidio_rx_ctl_context *ctx;
1562 	union octnet_cmd *ncmd;
1563 	int ctx_size = sizeof(struct liquidio_rx_ctl_context);
1564 	struct octeon_device *oct = (struct octeon_device *)lio->oct_dev;
1565 	int retval;
1566 
1567 	if (oct->props[lio->ifidx].rx_on == start_stop)
1568 		return;
1569 
1570 	sc = (struct octeon_soft_command *)
1571 		octeon_alloc_soft_command(oct, OCTNET_CMD_SIZE,
1572 					  16, ctx_size);
1573 
1574 	ncmd = (union octnet_cmd *)sc->virtdptr;
1575 	ctx  = (struct liquidio_rx_ctl_context *)sc->ctxptr;
1576 
1577 	WRITE_ONCE(ctx->cond, 0);
1578 	ctx->octeon_id = lio_get_device_id(oct);
1579 	init_waitqueue_head(&ctx->wc);
1580 
1581 	ncmd->u64 = 0;
1582 	ncmd->s.cmd = OCTNET_CMD_RX_CTL;
1583 	ncmd->s.param1 = start_stop;
1584 
1585 	octeon_swap_8B_data((u64 *)ncmd, (OCTNET_CMD_SIZE >> 3));
1586 
1587 	sc->iq_no = lio->linfo.txpciq[0].s.q_no;
1588 
1589 	octeon_prepare_soft_command(oct, sc, OPCODE_NIC,
1590 				    OPCODE_NIC_CMD, 0, 0, 0);
1591 
1592 	sc->callback = rx_ctl_callback;
1593 	sc->callback_arg = sc;
1594 	sc->wait_time = 5000;
1595 
1596 	retval = octeon_send_soft_command(oct, sc);
1597 	if (retval == IQ_SEND_FAILED) {
1598 		netif_info(lio, rx_err, lio->netdev, "Failed to send RX Control message\n");
1599 	} else {
1600 		/* Sleep on a wait queue till the cond flag indicates that the
1601 		 * response arrived or timed-out.
1602 		 */
1603 		if (sleep_cond(&ctx->wc, &ctx->cond) == -EINTR)
1604 			return;
1605 		oct->props[lio->ifidx].rx_on = start_stop;
1606 	}
1607 
1608 	octeon_free_soft_command(oct, sc);
1609 }
1610 
1611 /**
1612  * \brief Destroy NIC device interface
1613  * @param oct octeon device
1614  * @param ifidx which interface to destroy
1615  *
1616  * Cleanup associated with each interface for an Octeon device  when NIC
1617  * module is being unloaded or if initialization fails during load.
1618  */
1619 static void liquidio_destroy_nic_device(struct octeon_device *oct, int ifidx)
1620 {
1621 	struct net_device *netdev = oct->props[ifidx].netdev;
1622 	struct lio *lio;
1623 	struct napi_struct *napi, *n;
1624 
1625 	if (!netdev) {
1626 		dev_err(&oct->pci_dev->dev, "%s No netdevice ptr for index %d\n",
1627 			__func__, ifidx);
1628 		return;
1629 	}
1630 
1631 	lio = GET_LIO(netdev);
1632 
1633 	dev_dbg(&oct->pci_dev->dev, "NIC device cleanup\n");
1634 
1635 	if (atomic_read(&lio->ifstate) & LIO_IFSTATE_RUNNING)
1636 		liquidio_stop(netdev);
1637 
1638 	if (oct->props[lio->ifidx].napi_enabled == 1) {
1639 		list_for_each_entry_safe(napi, n, &netdev->napi_list, dev_list)
1640 			napi_disable(napi);
1641 
1642 		oct->props[lio->ifidx].napi_enabled = 0;
1643 
1644 		if (OCTEON_CN23XX_PF(oct))
1645 			oct->droq[0]->ops.poll_mode = 0;
1646 	}
1647 
1648 	if (atomic_read(&lio->ifstate) & LIO_IFSTATE_REGISTERED)
1649 		unregister_netdev(netdev);
1650 
1651 	cleanup_link_status_change_wq(netdev);
1652 
1653 	delete_glists(lio);
1654 
1655 	free_netdev(netdev);
1656 
1657 	oct->props[ifidx].gmxport = -1;
1658 
1659 	oct->props[ifidx].netdev = NULL;
1660 }
1661 
1662 /**
1663  * \brief Stop complete NIC functionality
1664  * @param oct octeon device
1665  */
1666 static int liquidio_stop_nic_module(struct octeon_device *oct)
1667 {
1668 	int i, j;
1669 	struct lio *lio;
1670 
1671 	dev_dbg(&oct->pci_dev->dev, "Stopping network interfaces\n");
1672 	if (!oct->ifcount) {
1673 		dev_err(&oct->pci_dev->dev, "Init for Octeon was not completed\n");
1674 		return 1;
1675 	}
1676 
1677 	spin_lock_bh(&oct->cmd_resp_wqlock);
1678 	oct->cmd_resp_state = OCT_DRV_OFFLINE;
1679 	spin_unlock_bh(&oct->cmd_resp_wqlock);
1680 
1681 	for (i = 0; i < oct->ifcount; i++) {
1682 		lio = GET_LIO(oct->props[i].netdev);
1683 		for (j = 0; j < lio->linfo.num_rxpciq; j++)
1684 			octeon_unregister_droq_ops(oct,
1685 						   lio->linfo.rxpciq[j].s.q_no);
1686 	}
1687 
1688 	for (i = 0; i < oct->ifcount; i++)
1689 		liquidio_destroy_nic_device(oct, i);
1690 
1691 	dev_dbg(&oct->pci_dev->dev, "Network interfaces stopped\n");
1692 	return 0;
1693 }
1694 
1695 /**
1696  * \brief Cleans up resources at unload time
1697  * @param pdev PCI device structure
1698  */
1699 static void liquidio_remove(struct pci_dev *pdev)
1700 {
1701 	struct octeon_device *oct_dev = pci_get_drvdata(pdev);
1702 
1703 	dev_dbg(&oct_dev->pci_dev->dev, "Stopping device\n");
1704 
1705 	if (oct_dev->watchdog_task)
1706 		kthread_stop(oct_dev->watchdog_task);
1707 
1708 	if (oct_dev->app_mode && (oct_dev->app_mode == CVM_DRV_NIC_APP))
1709 		liquidio_stop_nic_module(oct_dev);
1710 
1711 	/* Reset the octeon device and cleanup all memory allocated for
1712 	 * the octeon device by driver.
1713 	 */
1714 	octeon_destroy_resources(oct_dev);
1715 
1716 	dev_info(&oct_dev->pci_dev->dev, "Device removed\n");
1717 
1718 	/* This octeon device has been removed. Update the global
1719 	 * data structure to reflect this. Free the device structure.
1720 	 */
1721 	octeon_free_device_mem(oct_dev);
1722 }
1723 
1724 /**
1725  * \brief Identify the Octeon device and to map the BAR address space
1726  * @param oct octeon device
1727  */
1728 static int octeon_chip_specific_setup(struct octeon_device *oct)
1729 {
1730 	u32 dev_id, rev_id;
1731 	int ret = 1;
1732 	char *s;
1733 
1734 	pci_read_config_dword(oct->pci_dev, 0, &dev_id);
1735 	pci_read_config_dword(oct->pci_dev, 8, &rev_id);
1736 	oct->rev_id = rev_id & 0xff;
1737 
1738 	switch (dev_id) {
1739 	case OCTEON_CN68XX_PCIID:
1740 		oct->chip_id = OCTEON_CN68XX;
1741 		ret = lio_setup_cn68xx_octeon_device(oct);
1742 		s = "CN68XX";
1743 		break;
1744 
1745 	case OCTEON_CN66XX_PCIID:
1746 		oct->chip_id = OCTEON_CN66XX;
1747 		ret = lio_setup_cn66xx_octeon_device(oct);
1748 		s = "CN66XX";
1749 		break;
1750 
1751 	case OCTEON_CN23XX_PCIID_PF:
1752 		oct->chip_id = OCTEON_CN23XX_PF_VID;
1753 		ret = setup_cn23xx_octeon_pf_device(oct);
1754 		s = "CN23XX";
1755 		break;
1756 
1757 	default:
1758 		s = "?";
1759 		dev_err(&oct->pci_dev->dev, "Unknown device found (dev_id: %x)\n",
1760 			dev_id);
1761 	}
1762 
1763 	if (!ret)
1764 		dev_info(&oct->pci_dev->dev, "%s PASS%d.%d %s Version: %s\n", s,
1765 			 OCTEON_MAJOR_REV(oct),
1766 			 OCTEON_MINOR_REV(oct),
1767 			 octeon_get_conf(oct)->card_name,
1768 			 LIQUIDIO_VERSION);
1769 
1770 	return ret;
1771 }
1772 
1773 /**
1774  * \brief PCI initialization for each Octeon device.
1775  * @param oct octeon device
1776  */
1777 static int octeon_pci_os_setup(struct octeon_device *oct)
1778 {
1779 	/* setup PCI stuff first */
1780 	if (pci_enable_device(oct->pci_dev)) {
1781 		dev_err(&oct->pci_dev->dev, "pci_enable_device failed\n");
1782 		return 1;
1783 	}
1784 
1785 	if (dma_set_mask_and_coherent(&oct->pci_dev->dev, DMA_BIT_MASK(64))) {
1786 		dev_err(&oct->pci_dev->dev, "Unexpected DMA device capability\n");
1787 		pci_disable_device(oct->pci_dev);
1788 		return 1;
1789 	}
1790 
1791 	/* Enable PCI DMA Master. */
1792 	pci_set_master(oct->pci_dev);
1793 
1794 	return 0;
1795 }
1796 
1797 static inline int skb_iq(struct lio *lio, struct sk_buff *skb)
1798 {
1799 	int q = 0;
1800 
1801 	if (netif_is_multiqueue(lio->netdev))
1802 		q = skb->queue_mapping % lio->linfo.num_txpciq;
1803 
1804 	return q;
1805 }
1806 
1807 /**
1808  * \brief Check Tx queue state for a given network buffer
1809  * @param lio per-network private data
1810  * @param skb network buffer
1811  */
1812 static inline int check_txq_state(struct lio *lio, struct sk_buff *skb)
1813 {
1814 	int q = 0, iq = 0;
1815 
1816 	if (netif_is_multiqueue(lio->netdev)) {
1817 		q = skb->queue_mapping;
1818 		iq = lio->linfo.txpciq[(q % (lio->linfo.num_txpciq))].s.q_no;
1819 	} else {
1820 		iq = lio->txq;
1821 		q = iq;
1822 	}
1823 
1824 	if (octnet_iq_is_full(lio->oct_dev, iq))
1825 		return 0;
1826 
1827 	if (__netif_subqueue_stopped(lio->netdev, q)) {
1828 		INCR_INSTRQUEUE_PKT_COUNT(lio->oct_dev, iq, tx_restart, 1);
1829 		wake_q(lio->netdev, q);
1830 	}
1831 	return 1;
1832 }
1833 
1834 /**
1835  * \brief Unmap and free network buffer
1836  * @param buf buffer
1837  */
1838 static void free_netbuf(void *buf)
1839 {
1840 	struct sk_buff *skb;
1841 	struct octnet_buf_free_info *finfo;
1842 	struct lio *lio;
1843 
1844 	finfo = (struct octnet_buf_free_info *)buf;
1845 	skb = finfo->skb;
1846 	lio = finfo->lio;
1847 
1848 	dma_unmap_single(&lio->oct_dev->pci_dev->dev, finfo->dptr, skb->len,
1849 			 DMA_TO_DEVICE);
1850 
1851 	check_txq_state(lio, skb);
1852 
1853 	tx_buffer_free(skb);
1854 }
1855 
1856 /**
1857  * \brief Unmap and free gather buffer
1858  * @param buf buffer
1859  */
1860 static void free_netsgbuf(void *buf)
1861 {
1862 	struct octnet_buf_free_info *finfo;
1863 	struct sk_buff *skb;
1864 	struct lio *lio;
1865 	struct octnic_gather *g;
1866 	int i, frags, iq;
1867 
1868 	finfo = (struct octnet_buf_free_info *)buf;
1869 	skb = finfo->skb;
1870 	lio = finfo->lio;
1871 	g = finfo->g;
1872 	frags = skb_shinfo(skb)->nr_frags;
1873 
1874 	dma_unmap_single(&lio->oct_dev->pci_dev->dev,
1875 			 g->sg[0].ptr[0], (skb->len - skb->data_len),
1876 			 DMA_TO_DEVICE);
1877 
1878 	i = 1;
1879 	while (frags--) {
1880 		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
1881 
1882 		pci_unmap_page((lio->oct_dev)->pci_dev,
1883 			       g->sg[(i >> 2)].ptr[(i & 3)],
1884 			       frag->size, DMA_TO_DEVICE);
1885 		i++;
1886 	}
1887 
1888 	dma_sync_single_for_cpu(&lio->oct_dev->pci_dev->dev,
1889 				g->sg_dma_ptr, g->sg_size, DMA_TO_DEVICE);
1890 
1891 	iq = skb_iq(lio, skb);
1892 	spin_lock(&lio->glist_lock[iq]);
1893 	list_add_tail(&g->list, &lio->glist[iq]);
1894 	spin_unlock(&lio->glist_lock[iq]);
1895 
1896 	check_txq_state(lio, skb);     /* mq support: sub-queue state check */
1897 
1898 	tx_buffer_free(skb);
1899 }
1900 
1901 /**
1902  * \brief Unmap and free gather buffer with response
1903  * @param buf buffer
1904  */
1905 static void free_netsgbuf_with_resp(void *buf)
1906 {
1907 	struct octeon_soft_command *sc;
1908 	struct octnet_buf_free_info *finfo;
1909 	struct sk_buff *skb;
1910 	struct lio *lio;
1911 	struct octnic_gather *g;
1912 	int i, frags, iq;
1913 
1914 	sc = (struct octeon_soft_command *)buf;
1915 	skb = (struct sk_buff *)sc->callback_arg;
1916 	finfo = (struct octnet_buf_free_info *)&skb->cb;
1917 
1918 	lio = finfo->lio;
1919 	g = finfo->g;
1920 	frags = skb_shinfo(skb)->nr_frags;
1921 
1922 	dma_unmap_single(&lio->oct_dev->pci_dev->dev,
1923 			 g->sg[0].ptr[0], (skb->len - skb->data_len),
1924 			 DMA_TO_DEVICE);
1925 
1926 	i = 1;
1927 	while (frags--) {
1928 		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
1929 
1930 		pci_unmap_page((lio->oct_dev)->pci_dev,
1931 			       g->sg[(i >> 2)].ptr[(i & 3)],
1932 			       frag->size, DMA_TO_DEVICE);
1933 		i++;
1934 	}
1935 
1936 	dma_sync_single_for_cpu(&lio->oct_dev->pci_dev->dev,
1937 				g->sg_dma_ptr, g->sg_size, DMA_TO_DEVICE);
1938 
1939 	iq = skb_iq(lio, skb);
1940 
1941 	spin_lock(&lio->glist_lock[iq]);
1942 	list_add_tail(&g->list, &lio->glist[iq]);
1943 	spin_unlock(&lio->glist_lock[iq]);
1944 
1945 	/* Don't free the skb yet */
1946 
1947 	check_txq_state(lio, skb);
1948 }
1949 
1950 /**
1951  * \brief Adjust ptp frequency
1952  * @param ptp PTP clock info
1953  * @param ppb how much to adjust by, in parts-per-billion
1954  */
1955 static int liquidio_ptp_adjfreq(struct ptp_clock_info *ptp, s32 ppb)
1956 {
1957 	struct lio *lio = container_of(ptp, struct lio, ptp_info);
1958 	struct octeon_device *oct = (struct octeon_device *)lio->oct_dev;
1959 	u64 comp, delta;
1960 	unsigned long flags;
1961 	bool neg_adj = false;
1962 
1963 	if (ppb < 0) {
1964 		neg_adj = true;
1965 		ppb = -ppb;
1966 	}
1967 
1968 	/* The hardware adds the clock compensation value to the
1969 	 * PTP clock on every coprocessor clock cycle, so we
1970 	 * compute the delta in terms of coprocessor clocks.
1971 	 */
1972 	delta = (u64)ppb << 32;
1973 	do_div(delta, oct->coproc_clock_rate);
1974 
1975 	spin_lock_irqsave(&lio->ptp_lock, flags);
1976 	comp = lio_pci_readq(oct, CN6XXX_MIO_PTP_CLOCK_COMP);
1977 	if (neg_adj)
1978 		comp -= delta;
1979 	else
1980 		comp += delta;
1981 	lio_pci_writeq(oct, comp, CN6XXX_MIO_PTP_CLOCK_COMP);
1982 	spin_unlock_irqrestore(&lio->ptp_lock, flags);
1983 
1984 	return 0;
1985 }
1986 
1987 /**
1988  * \brief Adjust ptp time
1989  * @param ptp PTP clock info
1990  * @param delta how much to adjust by, in nanosecs
1991  */
1992 static int liquidio_ptp_adjtime(struct ptp_clock_info *ptp, s64 delta)
1993 {
1994 	unsigned long flags;
1995 	struct lio *lio = container_of(ptp, struct lio, ptp_info);
1996 
1997 	spin_lock_irqsave(&lio->ptp_lock, flags);
1998 	lio->ptp_adjust += delta;
1999 	spin_unlock_irqrestore(&lio->ptp_lock, flags);
2000 
2001 	return 0;
2002 }
2003 
2004 /**
2005  * \brief Get hardware clock time, including any adjustment
2006  * @param ptp PTP clock info
2007  * @param ts timespec
2008  */
2009 static int liquidio_ptp_gettime(struct ptp_clock_info *ptp,
2010 				struct timespec64 *ts)
2011 {
2012 	u64 ns;
2013 	unsigned long flags;
2014 	struct lio *lio = container_of(ptp, struct lio, ptp_info);
2015 	struct octeon_device *oct = (struct octeon_device *)lio->oct_dev;
2016 
2017 	spin_lock_irqsave(&lio->ptp_lock, flags);
2018 	ns = lio_pci_readq(oct, CN6XXX_MIO_PTP_CLOCK_HI);
2019 	ns += lio->ptp_adjust;
2020 	spin_unlock_irqrestore(&lio->ptp_lock, flags);
2021 
2022 	*ts = ns_to_timespec64(ns);
2023 
2024 	return 0;
2025 }
2026 
2027 /**
2028  * \brief Set hardware clock time. Reset adjustment
2029  * @param ptp PTP clock info
2030  * @param ts timespec
2031  */
2032 static int liquidio_ptp_settime(struct ptp_clock_info *ptp,
2033 				const struct timespec64 *ts)
2034 {
2035 	u64 ns;
2036 	unsigned long flags;
2037 	struct lio *lio = container_of(ptp, struct lio, ptp_info);
2038 	struct octeon_device *oct = (struct octeon_device *)lio->oct_dev;
2039 
2040 	ns = timespec_to_ns(ts);
2041 
2042 	spin_lock_irqsave(&lio->ptp_lock, flags);
2043 	lio_pci_writeq(oct, ns, CN6XXX_MIO_PTP_CLOCK_HI);
2044 	lio->ptp_adjust = 0;
2045 	spin_unlock_irqrestore(&lio->ptp_lock, flags);
2046 
2047 	return 0;
2048 }
2049 
2050 /**
2051  * \brief Check if PTP is enabled
2052  * @param ptp PTP clock info
2053  * @param rq request
2054  * @param on is it on
2055  */
2056 static int
2057 liquidio_ptp_enable(struct ptp_clock_info *ptp __attribute__((unused)),
2058 		    struct ptp_clock_request *rq __attribute__((unused)),
2059 		    int on __attribute__((unused)))
2060 {
2061 	return -EOPNOTSUPP;
2062 }
2063 
2064 /**
2065  * \brief Open PTP clock source
2066  * @param netdev network device
2067  */
2068 static void oct_ptp_open(struct net_device *netdev)
2069 {
2070 	struct lio *lio = GET_LIO(netdev);
2071 	struct octeon_device *oct = (struct octeon_device *)lio->oct_dev;
2072 
2073 	spin_lock_init(&lio->ptp_lock);
2074 
2075 	snprintf(lio->ptp_info.name, 16, "%s", netdev->name);
2076 	lio->ptp_info.owner = THIS_MODULE;
2077 	lio->ptp_info.max_adj = 250000000;
2078 	lio->ptp_info.n_alarm = 0;
2079 	lio->ptp_info.n_ext_ts = 0;
2080 	lio->ptp_info.n_per_out = 0;
2081 	lio->ptp_info.pps = 0;
2082 	lio->ptp_info.adjfreq = liquidio_ptp_adjfreq;
2083 	lio->ptp_info.adjtime = liquidio_ptp_adjtime;
2084 	lio->ptp_info.gettime64 = liquidio_ptp_gettime;
2085 	lio->ptp_info.settime64 = liquidio_ptp_settime;
2086 	lio->ptp_info.enable = liquidio_ptp_enable;
2087 
2088 	lio->ptp_adjust = 0;
2089 
2090 	lio->ptp_clock = ptp_clock_register(&lio->ptp_info,
2091 					     &oct->pci_dev->dev);
2092 
2093 	if (IS_ERR(lio->ptp_clock))
2094 		lio->ptp_clock = NULL;
2095 }
2096 
2097 /**
2098  * \brief Init PTP clock
2099  * @param oct octeon device
2100  */
2101 static void liquidio_ptp_init(struct octeon_device *oct)
2102 {
2103 	u64 clock_comp, cfg;
2104 
2105 	clock_comp = (u64)NSEC_PER_SEC << 32;
2106 	do_div(clock_comp, oct->coproc_clock_rate);
2107 	lio_pci_writeq(oct, clock_comp, CN6XXX_MIO_PTP_CLOCK_COMP);
2108 
2109 	/* Enable */
2110 	cfg = lio_pci_readq(oct, CN6XXX_MIO_PTP_CLOCK_CFG);
2111 	lio_pci_writeq(oct, cfg | 0x01, CN6XXX_MIO_PTP_CLOCK_CFG);
2112 }
2113 
2114 /**
2115  * \brief Load firmware to device
2116  * @param oct octeon device
2117  *
2118  * Maps device to firmware filename, requests firmware, and downloads it
2119  */
2120 static int load_firmware(struct octeon_device *oct)
2121 {
2122 	int ret = 0;
2123 	const struct firmware *fw;
2124 	char fw_name[LIO_MAX_FW_FILENAME_LEN];
2125 	char *tmp_fw_type;
2126 
2127 	if (strncmp(fw_type, LIO_FW_NAME_TYPE_NONE,
2128 		    sizeof(LIO_FW_NAME_TYPE_NONE)) == 0) {
2129 		dev_info(&oct->pci_dev->dev, "Skipping firmware load\n");
2130 		return ret;
2131 	}
2132 
2133 	if (fw_type[0] == '\0')
2134 		tmp_fw_type = LIO_FW_NAME_TYPE_NIC;
2135 	else
2136 		tmp_fw_type = fw_type;
2137 
2138 	sprintf(fw_name, "%s%s%s_%s%s", LIO_FW_DIR, LIO_FW_BASE_NAME,
2139 		octeon_get_conf(oct)->card_name, tmp_fw_type,
2140 		LIO_FW_NAME_SUFFIX);
2141 
2142 	ret = request_firmware(&fw, fw_name, &oct->pci_dev->dev);
2143 	if (ret) {
2144 		dev_err(&oct->pci_dev->dev, "Request firmware failed. Could not find file %s.\n.",
2145 			fw_name);
2146 		release_firmware(fw);
2147 		return ret;
2148 	}
2149 
2150 	ret = octeon_download_firmware(oct, fw->data, fw->size);
2151 
2152 	release_firmware(fw);
2153 
2154 	return ret;
2155 }
2156 
2157 /**
2158  * \brief Setup output queue
2159  * @param oct octeon device
2160  * @param q_no which queue
2161  * @param num_descs how many descriptors
2162  * @param desc_size size of each descriptor
2163  * @param app_ctx application context
2164  */
2165 static int octeon_setup_droq(struct octeon_device *oct, int q_no, int num_descs,
2166 			     int desc_size, void *app_ctx)
2167 {
2168 	int ret_val = 0;
2169 
2170 	dev_dbg(&oct->pci_dev->dev, "Creating Droq: %d\n", q_no);
2171 	/* droq creation and local register settings. */
2172 	ret_val = octeon_create_droq(oct, q_no, num_descs, desc_size, app_ctx);
2173 	if (ret_val < 0)
2174 		return ret_val;
2175 
2176 	if (ret_val == 1) {
2177 		dev_dbg(&oct->pci_dev->dev, "Using default droq %d\n", q_no);
2178 		return 0;
2179 	}
2180 	/* tasklet creation for the droq */
2181 
2182 	/* Enable the droq queues */
2183 	octeon_set_droq_pkt_op(oct, q_no, 1);
2184 
2185 	/* Send Credit for Octeon Output queues. Credits are always
2186 	 * sent after the output queue is enabled.
2187 	 */
2188 	writel(oct->droq[q_no]->max_count,
2189 	       oct->droq[q_no]->pkts_credit_reg);
2190 
2191 	return ret_val;
2192 }
2193 
2194 /**
2195  * \brief Callback for getting interface configuration
2196  * @param status status of request
2197  * @param buf pointer to resp structure
2198  */
2199 static void if_cfg_callback(struct octeon_device *oct,
2200 			    u32 status __attribute__((unused)),
2201 			    void *buf)
2202 {
2203 	struct octeon_soft_command *sc = (struct octeon_soft_command *)buf;
2204 	struct liquidio_if_cfg_resp *resp;
2205 	struct liquidio_if_cfg_context *ctx;
2206 
2207 	resp = (struct liquidio_if_cfg_resp *)sc->virtrptr;
2208 	ctx = (struct liquidio_if_cfg_context *)sc->ctxptr;
2209 
2210 	oct = lio_get_device(ctx->octeon_id);
2211 	if (resp->status)
2212 		dev_err(&oct->pci_dev->dev, "nic if cfg instruction failed. Status: %llx\n",
2213 			CVM_CAST64(resp->status));
2214 	WRITE_ONCE(ctx->cond, 1);
2215 
2216 	snprintf(oct->fw_info.liquidio_firmware_version, 32, "%s",
2217 		 resp->cfg_info.liquidio_firmware_version);
2218 
2219 	/* This barrier is required to be sure that the response has been
2220 	 * written fully before waking up the handler
2221 	 */
2222 	wmb();
2223 
2224 	wake_up_interruptible(&ctx->wc);
2225 }
2226 
2227 /** Routine to push packets arriving on Octeon interface upto network layer.
2228  * @param oct_id   - octeon device id.
2229  * @param skbuff   - skbuff struct to be passed to network layer.
2230  * @param len      - size of total data received.
2231  * @param rh       - Control header associated with the packet
2232  * @param param    - additional control data with the packet
2233  * @param arg	   - farg registered in droq_ops
2234  */
2235 static void
2236 liquidio_push_packet(u32 octeon_id __attribute__((unused)),
2237 		     void *skbuff,
2238 		     u32 len,
2239 		     union octeon_rh *rh,
2240 		     void *param,
2241 		     void *arg)
2242 {
2243 	struct napi_struct *napi = param;
2244 	struct sk_buff *skb = (struct sk_buff *)skbuff;
2245 	struct skb_shared_hwtstamps *shhwtstamps;
2246 	u64 ns;
2247 	u16 vtag = 0;
2248 	u32 r_dh_off;
2249 	struct net_device *netdev = (struct net_device *)arg;
2250 	struct octeon_droq *droq = container_of(param, struct octeon_droq,
2251 						napi);
2252 	if (netdev) {
2253 		int packet_was_received;
2254 		struct lio *lio = GET_LIO(netdev);
2255 		struct octeon_device *oct = lio->oct_dev;
2256 
2257 		/* Do not proceed if the interface is not in RUNNING state. */
2258 		if (!ifstate_check(lio, LIO_IFSTATE_RUNNING)) {
2259 			recv_buffer_free(skb);
2260 			droq->stats.rx_dropped++;
2261 			return;
2262 		}
2263 
2264 		skb->dev = netdev;
2265 
2266 		skb_record_rx_queue(skb, droq->q_no);
2267 		if (likely(len > MIN_SKB_SIZE)) {
2268 			struct octeon_skb_page_info *pg_info;
2269 			unsigned char *va;
2270 
2271 			pg_info = ((struct octeon_skb_page_info *)(skb->cb));
2272 			if (pg_info->page) {
2273 				/* For Paged allocation use the frags */
2274 				va = page_address(pg_info->page) +
2275 					pg_info->page_offset;
2276 				memcpy(skb->data, va, MIN_SKB_SIZE);
2277 				skb_put(skb, MIN_SKB_SIZE);
2278 				skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
2279 						pg_info->page,
2280 						pg_info->page_offset +
2281 						MIN_SKB_SIZE,
2282 						len - MIN_SKB_SIZE,
2283 						LIO_RXBUFFER_SZ);
2284 			}
2285 		} else {
2286 			struct octeon_skb_page_info *pg_info =
2287 				((struct octeon_skb_page_info *)(skb->cb));
2288 			skb_copy_to_linear_data(skb, page_address(pg_info->page)
2289 						+ pg_info->page_offset, len);
2290 			skb_put(skb, len);
2291 			put_page(pg_info->page);
2292 		}
2293 
2294 		r_dh_off = (rh->r_dh.len - 1) * BYTES_PER_DHLEN_UNIT;
2295 
2296 		if (((oct->chip_id == OCTEON_CN66XX) ||
2297 		     (oct->chip_id == OCTEON_CN68XX)) &&
2298 		    ptp_enable) {
2299 			if (rh->r_dh.has_hwtstamp) {
2300 				/* timestamp is included from the hardware at
2301 				 * the beginning of the packet.
2302 				 */
2303 				if (ifstate_check
2304 				    (lio, LIO_IFSTATE_RX_TIMESTAMP_ENABLED)) {
2305 					/* Nanoseconds are in the first 64-bits
2306 					 * of the packet.
2307 					 */
2308 					memcpy(&ns, (skb->data + r_dh_off),
2309 					       sizeof(ns));
2310 					r_dh_off -= BYTES_PER_DHLEN_UNIT;
2311 					shhwtstamps = skb_hwtstamps(skb);
2312 					shhwtstamps->hwtstamp =
2313 						ns_to_ktime(ns +
2314 							    lio->ptp_adjust);
2315 				}
2316 			}
2317 		}
2318 
2319 		if (rh->r_dh.has_hash) {
2320 			__be32 *hash_be = (__be32 *)(skb->data + r_dh_off);
2321 			u32 hash = be32_to_cpu(*hash_be);
2322 
2323 			skb_set_hash(skb, hash, PKT_HASH_TYPE_L4);
2324 			r_dh_off -= BYTES_PER_DHLEN_UNIT;
2325 		}
2326 
2327 		skb_pull(skb, rh->r_dh.len * BYTES_PER_DHLEN_UNIT);
2328 
2329 		skb->protocol = eth_type_trans(skb, skb->dev);
2330 		if ((netdev->features & NETIF_F_RXCSUM) &&
2331 		    (((rh->r_dh.encap_on) &&
2332 		      (rh->r_dh.csum_verified & CNNIC_TUN_CSUM_VERIFIED)) ||
2333 		     (!(rh->r_dh.encap_on) &&
2334 		      (rh->r_dh.csum_verified & CNNIC_CSUM_VERIFIED))))
2335 			/* checksum has already been verified */
2336 			skb->ip_summed = CHECKSUM_UNNECESSARY;
2337 		else
2338 			skb->ip_summed = CHECKSUM_NONE;
2339 
2340 		/* Setting Encapsulation field on basis of status received
2341 		 * from the firmware
2342 		 */
2343 		if (rh->r_dh.encap_on) {
2344 			skb->encapsulation = 1;
2345 			skb->csum_level = 1;
2346 			droq->stats.rx_vxlan++;
2347 		}
2348 
2349 		/* inbound VLAN tag */
2350 		if ((netdev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
2351 		    (rh->r_dh.vlan != 0)) {
2352 			u16 vid = rh->r_dh.vlan;
2353 			u16 priority = rh->r_dh.priority;
2354 
2355 			vtag = priority << 13 | vid;
2356 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vtag);
2357 		}
2358 
2359 		packet_was_received = napi_gro_receive(napi, skb) != GRO_DROP;
2360 
2361 		if (packet_was_received) {
2362 			droq->stats.rx_bytes_received += len;
2363 			droq->stats.rx_pkts_received++;
2364 		} else {
2365 			droq->stats.rx_dropped++;
2366 			netif_info(lio, rx_err, lio->netdev,
2367 				   "droq:%d  error rx_dropped:%llu\n",
2368 				   droq->q_no, droq->stats.rx_dropped);
2369 		}
2370 
2371 	} else {
2372 		recv_buffer_free(skb);
2373 	}
2374 }
2375 
2376 /**
2377  * \brief wrapper for calling napi_schedule
2378  * @param param parameters to pass to napi_schedule
2379  *
2380  * Used when scheduling on different CPUs
2381  */
2382 static void napi_schedule_wrapper(void *param)
2383 {
2384 	struct napi_struct *napi = param;
2385 
2386 	napi_schedule(napi);
2387 }
2388 
2389 /**
2390  * \brief callback when receive interrupt occurs and we are in NAPI mode
2391  * @param arg pointer to octeon output queue
2392  */
2393 static void liquidio_napi_drv_callback(void *arg)
2394 {
2395 	struct octeon_device *oct;
2396 	struct octeon_droq *droq = arg;
2397 	int this_cpu = smp_processor_id();
2398 
2399 	oct = droq->oct_dev;
2400 
2401 	if (OCTEON_CN23XX_PF(oct) || droq->cpu_id == this_cpu) {
2402 		napi_schedule_irqoff(&droq->napi);
2403 	} else {
2404 		struct call_single_data *csd = &droq->csd;
2405 
2406 		csd->func = napi_schedule_wrapper;
2407 		csd->info = &droq->napi;
2408 		csd->flags = 0;
2409 
2410 		smp_call_function_single_async(droq->cpu_id, csd);
2411 	}
2412 }
2413 
2414 /**
2415  * \brief Entry point for NAPI polling
2416  * @param napi NAPI structure
2417  * @param budget maximum number of items to process
2418  */
2419 static int liquidio_napi_poll(struct napi_struct *napi, int budget)
2420 {
2421 	struct octeon_droq *droq;
2422 	int work_done;
2423 	int tx_done = 0, iq_no;
2424 	struct octeon_instr_queue *iq;
2425 	struct octeon_device *oct;
2426 
2427 	droq = container_of(napi, struct octeon_droq, napi);
2428 	oct = droq->oct_dev;
2429 	iq_no = droq->q_no;
2430 	/* Handle Droq descriptors */
2431 	work_done = octeon_process_droq_poll_cmd(oct, droq->q_no,
2432 						 POLL_EVENT_PROCESS_PKTS,
2433 						 budget);
2434 
2435 	/* Flush the instruction queue */
2436 	iq = oct->instr_queue[iq_no];
2437 	if (iq) {
2438 		/* Process iq buffers with in the budget limits */
2439 		tx_done = octeon_flush_iq(oct, iq, budget);
2440 		/* Update iq read-index rather than waiting for next interrupt.
2441 		 * Return back if tx_done is false.
2442 		 */
2443 		update_txq_status(oct, iq_no);
2444 	} else {
2445 		dev_err(&oct->pci_dev->dev, "%s:  iq (%d) num invalid\n",
2446 			__func__, iq_no);
2447 	}
2448 
2449 	/* force enable interrupt if reg cnts are high to avoid wraparound */
2450 	if ((work_done < budget && tx_done) ||
2451 	    (iq && iq->pkt_in_done >= MAX_REG_CNT) ||
2452 	    (droq->pkt_count >= MAX_REG_CNT)) {
2453 		tx_done = 1;
2454 		napi_complete_done(napi, work_done);
2455 		octeon_process_droq_poll_cmd(droq->oct_dev, droq->q_no,
2456 					     POLL_EVENT_ENABLE_INTR, 0);
2457 		return 0;
2458 	}
2459 
2460 	return (!tx_done) ? (budget) : (work_done);
2461 }
2462 
2463 /**
2464  * \brief Setup input and output queues
2465  * @param octeon_dev octeon device
2466  * @param ifidx  Interface Index
2467  *
2468  * Note: Queues are with respect to the octeon device. Thus
2469  * an input queue is for egress packets, and output queues
2470  * are for ingress packets.
2471  */
2472 static inline int setup_io_queues(struct octeon_device *octeon_dev,
2473 				  int ifidx)
2474 {
2475 	struct octeon_droq_ops droq_ops;
2476 	struct net_device *netdev;
2477 	static int cpu_id;
2478 	static int cpu_id_modulus;
2479 	struct octeon_droq *droq;
2480 	struct napi_struct *napi;
2481 	int q, q_no, retval = 0;
2482 	struct lio *lio;
2483 	int num_tx_descs;
2484 
2485 	netdev = octeon_dev->props[ifidx].netdev;
2486 
2487 	lio = GET_LIO(netdev);
2488 
2489 	memset(&droq_ops, 0, sizeof(struct octeon_droq_ops));
2490 
2491 	droq_ops.fptr = liquidio_push_packet;
2492 	droq_ops.farg = (void *)netdev;
2493 
2494 	droq_ops.poll_mode = 1;
2495 	droq_ops.napi_fn = liquidio_napi_drv_callback;
2496 	cpu_id = 0;
2497 	cpu_id_modulus = num_present_cpus();
2498 
2499 	/* set up DROQs. */
2500 	for (q = 0; q < lio->linfo.num_rxpciq; q++) {
2501 		q_no = lio->linfo.rxpciq[q].s.q_no;
2502 		dev_dbg(&octeon_dev->pci_dev->dev,
2503 			"setup_io_queues index:%d linfo.rxpciq.s.q_no:%d\n",
2504 			q, q_no);
2505 		retval = octeon_setup_droq(octeon_dev, q_no,
2506 					   CFG_GET_NUM_RX_DESCS_NIC_IF
2507 						   (octeon_get_conf(octeon_dev),
2508 						   lio->ifidx),
2509 					   CFG_GET_NUM_RX_BUF_SIZE_NIC_IF
2510 						   (octeon_get_conf(octeon_dev),
2511 						   lio->ifidx), NULL);
2512 		if (retval) {
2513 			dev_err(&octeon_dev->pci_dev->dev,
2514 				"%s : Runtime DROQ(RxQ) creation failed.\n",
2515 				__func__);
2516 			return 1;
2517 		}
2518 
2519 		droq = octeon_dev->droq[q_no];
2520 		napi = &droq->napi;
2521 		dev_dbg(&octeon_dev->pci_dev->dev, "netif_napi_add netdev:%llx oct:%llx pf_num:%d\n",
2522 			(u64)netdev, (u64)octeon_dev, octeon_dev->pf_num);
2523 		netif_napi_add(netdev, napi, liquidio_napi_poll, 64);
2524 
2525 		/* designate a CPU for this droq */
2526 		droq->cpu_id = cpu_id;
2527 		cpu_id++;
2528 		if (cpu_id >= cpu_id_modulus)
2529 			cpu_id = 0;
2530 
2531 		octeon_register_droq_ops(octeon_dev, q_no, &droq_ops);
2532 	}
2533 
2534 	if (OCTEON_CN23XX_PF(octeon_dev)) {
2535 		/* 23XX PF can receive control messages (via the first PF-owned
2536 		 * droq) from the firmware even if the ethX interface is down,
2537 		 * so that's why poll_mode must be off for the first droq.
2538 		 */
2539 		octeon_dev->droq[0]->ops.poll_mode = 0;
2540 	}
2541 
2542 	/* set up IQs. */
2543 	for (q = 0; q < lio->linfo.num_txpciq; q++) {
2544 		num_tx_descs = CFG_GET_NUM_TX_DESCS_NIC_IF(octeon_get_conf
2545 							   (octeon_dev),
2546 							   lio->ifidx);
2547 		retval = octeon_setup_iq(octeon_dev, ifidx, q,
2548 					 lio->linfo.txpciq[q], num_tx_descs,
2549 					 netdev_get_tx_queue(netdev, q));
2550 		if (retval) {
2551 			dev_err(&octeon_dev->pci_dev->dev,
2552 				" %s : Runtime IQ(TxQ) creation failed.\n",
2553 				__func__);
2554 			return 1;
2555 		}
2556 	}
2557 
2558 	return 0;
2559 }
2560 
2561 /**
2562  * \brief Poll routine for checking transmit queue status
2563  * @param work work_struct data structure
2564  */
2565 static void octnet_poll_check_txq_status(struct work_struct *work)
2566 {
2567 	struct cavium_wk *wk = (struct cavium_wk *)work;
2568 	struct lio *lio = (struct lio *)wk->ctxptr;
2569 
2570 	if (!ifstate_check(lio, LIO_IFSTATE_RUNNING))
2571 		return;
2572 
2573 	check_txq_status(lio);
2574 	queue_delayed_work(lio->txq_status_wq.wq,
2575 			   &lio->txq_status_wq.wk.work, msecs_to_jiffies(1));
2576 }
2577 
2578 /**
2579  * \brief Sets up the txq poll check
2580  * @param netdev network device
2581  */
2582 static inline int setup_tx_poll_fn(struct net_device *netdev)
2583 {
2584 	struct lio *lio = GET_LIO(netdev);
2585 	struct octeon_device *oct = lio->oct_dev;
2586 
2587 	lio->txq_status_wq.wq = alloc_workqueue("txq-status",
2588 						WQ_MEM_RECLAIM, 0);
2589 	if (!lio->txq_status_wq.wq) {
2590 		dev_err(&oct->pci_dev->dev, "unable to create cavium txq status wq\n");
2591 		return -1;
2592 	}
2593 	INIT_DELAYED_WORK(&lio->txq_status_wq.wk.work,
2594 			  octnet_poll_check_txq_status);
2595 	lio->txq_status_wq.wk.ctxptr = lio;
2596 	queue_delayed_work(lio->txq_status_wq.wq,
2597 			   &lio->txq_status_wq.wk.work, msecs_to_jiffies(1));
2598 	return 0;
2599 }
2600 
2601 static inline void cleanup_tx_poll_fn(struct net_device *netdev)
2602 {
2603 	struct lio *lio = GET_LIO(netdev);
2604 
2605 	if (lio->txq_status_wq.wq) {
2606 		cancel_delayed_work_sync(&lio->txq_status_wq.wk.work);
2607 		destroy_workqueue(lio->txq_status_wq.wq);
2608 	}
2609 }
2610 
2611 /**
2612  * \brief Net device open for LiquidIO
2613  * @param netdev network device
2614  */
2615 static int liquidio_open(struct net_device *netdev)
2616 {
2617 	struct lio *lio = GET_LIO(netdev);
2618 	struct octeon_device *oct = lio->oct_dev;
2619 	struct napi_struct *napi, *n;
2620 
2621 	if (oct->props[lio->ifidx].napi_enabled == 0) {
2622 		list_for_each_entry_safe(napi, n, &netdev->napi_list, dev_list)
2623 			napi_enable(napi);
2624 
2625 		oct->props[lio->ifidx].napi_enabled = 1;
2626 
2627 		if (OCTEON_CN23XX_PF(oct))
2628 			oct->droq[0]->ops.poll_mode = 1;
2629 	}
2630 
2631 	if ((oct->chip_id == OCTEON_CN66XX || oct->chip_id == OCTEON_CN68XX) &&
2632 	    ptp_enable)
2633 		oct_ptp_open(netdev);
2634 
2635 	ifstate_set(lio, LIO_IFSTATE_RUNNING);
2636 
2637 	/* Ready for link status updates */
2638 	lio->intf_open = 1;
2639 
2640 	netif_info(lio, ifup, lio->netdev, "Interface Open, ready for traffic\n");
2641 
2642 	if (OCTEON_CN23XX_PF(oct)) {
2643 		if (!oct->msix_on)
2644 			if (setup_tx_poll_fn(netdev))
2645 				return -1;
2646 	} else {
2647 		if (setup_tx_poll_fn(netdev))
2648 			return -1;
2649 	}
2650 
2651 	start_txq(netdev);
2652 
2653 	/* tell Octeon to start forwarding packets to host */
2654 	send_rx_ctrl_cmd(lio, 1);
2655 
2656 	dev_info(&oct->pci_dev->dev, "%s interface is opened\n",
2657 		 netdev->name);
2658 
2659 	return 0;
2660 }
2661 
2662 /**
2663  * \brief Net device stop for LiquidIO
2664  * @param netdev network device
2665  */
2666 static int liquidio_stop(struct net_device *netdev)
2667 {
2668 	struct lio *lio = GET_LIO(netdev);
2669 	struct octeon_device *oct = lio->oct_dev;
2670 
2671 	ifstate_reset(lio, LIO_IFSTATE_RUNNING);
2672 
2673 	netif_tx_disable(netdev);
2674 
2675 	/* Inform that netif carrier is down */
2676 	netif_carrier_off(netdev);
2677 	lio->intf_open = 0;
2678 	lio->linfo.link.s.link_up = 0;
2679 	lio->link_changes++;
2680 
2681 	/* Tell Octeon that nic interface is down. */
2682 	send_rx_ctrl_cmd(lio, 0);
2683 
2684 	if (OCTEON_CN23XX_PF(oct)) {
2685 		if (!oct->msix_on)
2686 			cleanup_tx_poll_fn(netdev);
2687 	} else {
2688 		cleanup_tx_poll_fn(netdev);
2689 	}
2690 
2691 	if (lio->ptp_clock) {
2692 		ptp_clock_unregister(lio->ptp_clock);
2693 		lio->ptp_clock = NULL;
2694 	}
2695 
2696 	dev_info(&oct->pci_dev->dev, "%s interface is stopped\n", netdev->name);
2697 
2698 	return 0;
2699 }
2700 
2701 /**
2702  * \brief Converts a mask based on net device flags
2703  * @param netdev network device
2704  *
2705  * This routine generates a octnet_ifflags mask from the net device flags
2706  * received from the OS.
2707  */
2708 static inline enum octnet_ifflags get_new_flags(struct net_device *netdev)
2709 {
2710 	enum octnet_ifflags f = OCTNET_IFFLAG_UNICAST;
2711 
2712 	if (netdev->flags & IFF_PROMISC)
2713 		f |= OCTNET_IFFLAG_PROMISC;
2714 
2715 	if (netdev->flags & IFF_ALLMULTI)
2716 		f |= OCTNET_IFFLAG_ALLMULTI;
2717 
2718 	if (netdev->flags & IFF_MULTICAST) {
2719 		f |= OCTNET_IFFLAG_MULTICAST;
2720 
2721 		/* Accept all multicast addresses if there are more than we
2722 		 * can handle
2723 		 */
2724 		if (netdev_mc_count(netdev) > MAX_OCTEON_MULTICAST_ADDR)
2725 			f |= OCTNET_IFFLAG_ALLMULTI;
2726 	}
2727 
2728 	if (netdev->flags & IFF_BROADCAST)
2729 		f |= OCTNET_IFFLAG_BROADCAST;
2730 
2731 	return f;
2732 }
2733 
2734 /**
2735  * \brief Net device set_multicast_list
2736  * @param netdev network device
2737  */
2738 static void liquidio_set_mcast_list(struct net_device *netdev)
2739 {
2740 	struct lio *lio = GET_LIO(netdev);
2741 	struct octeon_device *oct = lio->oct_dev;
2742 	struct octnic_ctrl_pkt nctrl;
2743 	struct netdev_hw_addr *ha;
2744 	u64 *mc;
2745 	int ret;
2746 	int mc_count = min(netdev_mc_count(netdev), MAX_OCTEON_MULTICAST_ADDR);
2747 
2748 	memset(&nctrl, 0, sizeof(struct octnic_ctrl_pkt));
2749 
2750 	/* Create a ctrl pkt command to be sent to core app. */
2751 	nctrl.ncmd.u64 = 0;
2752 	nctrl.ncmd.s.cmd = OCTNET_CMD_SET_MULTI_LIST;
2753 	nctrl.ncmd.s.param1 = get_new_flags(netdev);
2754 	nctrl.ncmd.s.param2 = mc_count;
2755 	nctrl.ncmd.s.more = mc_count;
2756 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
2757 	nctrl.netpndev = (u64)netdev;
2758 	nctrl.cb_fn = liquidio_link_ctrl_cmd_completion;
2759 
2760 	/* copy all the addresses into the udd */
2761 	mc = &nctrl.udd[0];
2762 	netdev_for_each_mc_addr(ha, netdev) {
2763 		*mc = 0;
2764 		memcpy(((u8 *)mc) + 2, ha->addr, ETH_ALEN);
2765 		/* no need to swap bytes */
2766 
2767 		if (++mc > &nctrl.udd[mc_count])
2768 			break;
2769 	}
2770 
2771 	/* Apparently, any activity in this call from the kernel has to
2772 	 * be atomic. So we won't wait for response.
2773 	 */
2774 	nctrl.wait_time = 0;
2775 
2776 	ret = octnet_send_nic_ctrl_pkt(lio->oct_dev, &nctrl);
2777 	if (ret < 0) {
2778 		dev_err(&oct->pci_dev->dev, "DEVFLAGS change failed in core (ret: 0x%x)\n",
2779 			ret);
2780 	}
2781 }
2782 
2783 /**
2784  * \brief Net device set_mac_address
2785  * @param netdev network device
2786  */
2787 static int liquidio_set_mac(struct net_device *netdev, void *p)
2788 {
2789 	int ret = 0;
2790 	struct lio *lio = GET_LIO(netdev);
2791 	struct octeon_device *oct = lio->oct_dev;
2792 	struct sockaddr *addr = (struct sockaddr *)p;
2793 	struct octnic_ctrl_pkt nctrl;
2794 
2795 	if (!is_valid_ether_addr(addr->sa_data))
2796 		return -EADDRNOTAVAIL;
2797 
2798 	memset(&nctrl, 0, sizeof(struct octnic_ctrl_pkt));
2799 
2800 	nctrl.ncmd.u64 = 0;
2801 	nctrl.ncmd.s.cmd = OCTNET_CMD_CHANGE_MACADDR;
2802 	nctrl.ncmd.s.param1 = 0;
2803 	nctrl.ncmd.s.more = 1;
2804 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
2805 	nctrl.netpndev = (u64)netdev;
2806 	nctrl.cb_fn = liquidio_link_ctrl_cmd_completion;
2807 	nctrl.wait_time = 100;
2808 
2809 	nctrl.udd[0] = 0;
2810 	/* The MAC Address is presented in network byte order. */
2811 	memcpy((u8 *)&nctrl.udd[0] + 2, addr->sa_data, ETH_ALEN);
2812 
2813 	ret = octnet_send_nic_ctrl_pkt(lio->oct_dev, &nctrl);
2814 	if (ret < 0) {
2815 		dev_err(&oct->pci_dev->dev, "MAC Address change failed\n");
2816 		return -ENOMEM;
2817 	}
2818 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2819 	memcpy(((u8 *)&lio->linfo.hw_addr) + 2, addr->sa_data, ETH_ALEN);
2820 
2821 	return 0;
2822 }
2823 
2824 /**
2825  * \brief Net device get_stats
2826  * @param netdev network device
2827  */
2828 static struct net_device_stats *liquidio_get_stats(struct net_device *netdev)
2829 {
2830 	struct lio *lio = GET_LIO(netdev);
2831 	struct net_device_stats *stats = &netdev->stats;
2832 	struct octeon_device *oct;
2833 	u64 pkts = 0, drop = 0, bytes = 0;
2834 	struct oct_droq_stats *oq_stats;
2835 	struct oct_iq_stats *iq_stats;
2836 	int i, iq_no, oq_no;
2837 
2838 	oct = lio->oct_dev;
2839 
2840 	for (i = 0; i < lio->linfo.num_txpciq; i++) {
2841 		iq_no = lio->linfo.txpciq[i].s.q_no;
2842 		iq_stats = &oct->instr_queue[iq_no]->stats;
2843 		pkts += iq_stats->tx_done;
2844 		drop += iq_stats->tx_dropped;
2845 		bytes += iq_stats->tx_tot_bytes;
2846 	}
2847 
2848 	stats->tx_packets = pkts;
2849 	stats->tx_bytes = bytes;
2850 	stats->tx_dropped = drop;
2851 
2852 	pkts = 0;
2853 	drop = 0;
2854 	bytes = 0;
2855 
2856 	for (i = 0; i < lio->linfo.num_rxpciq; i++) {
2857 		oq_no = lio->linfo.rxpciq[i].s.q_no;
2858 		oq_stats = &oct->droq[oq_no]->stats;
2859 		pkts += oq_stats->rx_pkts_received;
2860 		drop += (oq_stats->rx_dropped +
2861 			 oq_stats->dropped_nodispatch +
2862 			 oq_stats->dropped_toomany +
2863 			 oq_stats->dropped_nomem);
2864 		bytes += oq_stats->rx_bytes_received;
2865 	}
2866 
2867 	stats->rx_bytes = bytes;
2868 	stats->rx_packets = pkts;
2869 	stats->rx_dropped = drop;
2870 
2871 	return stats;
2872 }
2873 
2874 /**
2875  * \brief Net device change_mtu
2876  * @param netdev network device
2877  */
2878 static int liquidio_change_mtu(struct net_device *netdev, int new_mtu)
2879 {
2880 	struct lio *lio = GET_LIO(netdev);
2881 	struct octeon_device *oct = lio->oct_dev;
2882 	struct octnic_ctrl_pkt nctrl;
2883 	int ret = 0;
2884 
2885 	memset(&nctrl, 0, sizeof(struct octnic_ctrl_pkt));
2886 
2887 	nctrl.ncmd.u64 = 0;
2888 	nctrl.ncmd.s.cmd = OCTNET_CMD_CHANGE_MTU;
2889 	nctrl.ncmd.s.param1 = new_mtu;
2890 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
2891 	nctrl.wait_time = 100;
2892 	nctrl.netpndev = (u64)netdev;
2893 	nctrl.cb_fn = liquidio_link_ctrl_cmd_completion;
2894 
2895 	ret = octnet_send_nic_ctrl_pkt(lio->oct_dev, &nctrl);
2896 	if (ret < 0) {
2897 		dev_err(&oct->pci_dev->dev, "Failed to set MTU\n");
2898 		return -1;
2899 	}
2900 
2901 	lio->mtu = new_mtu;
2902 
2903 	return 0;
2904 }
2905 
2906 /**
2907  * \brief Handler for SIOCSHWTSTAMP ioctl
2908  * @param netdev network device
2909  * @param ifr interface request
2910  * @param cmd command
2911  */
2912 static int hwtstamp_ioctl(struct net_device *netdev, struct ifreq *ifr)
2913 {
2914 	struct hwtstamp_config conf;
2915 	struct lio *lio = GET_LIO(netdev);
2916 
2917 	if (copy_from_user(&conf, ifr->ifr_data, sizeof(conf)))
2918 		return -EFAULT;
2919 
2920 	if (conf.flags)
2921 		return -EINVAL;
2922 
2923 	switch (conf.tx_type) {
2924 	case HWTSTAMP_TX_ON:
2925 	case HWTSTAMP_TX_OFF:
2926 		break;
2927 	default:
2928 		return -ERANGE;
2929 	}
2930 
2931 	switch (conf.rx_filter) {
2932 	case HWTSTAMP_FILTER_NONE:
2933 		break;
2934 	case HWTSTAMP_FILTER_ALL:
2935 	case HWTSTAMP_FILTER_SOME:
2936 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2937 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2938 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2939 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2940 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2941 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2942 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2943 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2944 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2945 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
2946 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
2947 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2948 		conf.rx_filter = HWTSTAMP_FILTER_ALL;
2949 		break;
2950 	default:
2951 		return -ERANGE;
2952 	}
2953 
2954 	if (conf.rx_filter == HWTSTAMP_FILTER_ALL)
2955 		ifstate_set(lio, LIO_IFSTATE_RX_TIMESTAMP_ENABLED);
2956 
2957 	else
2958 		ifstate_reset(lio, LIO_IFSTATE_RX_TIMESTAMP_ENABLED);
2959 
2960 	return copy_to_user(ifr->ifr_data, &conf, sizeof(conf)) ? -EFAULT : 0;
2961 }
2962 
2963 /**
2964  * \brief ioctl handler
2965  * @param netdev network device
2966  * @param ifr interface request
2967  * @param cmd command
2968  */
2969 static int liquidio_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2970 {
2971 	struct lio *lio = GET_LIO(netdev);
2972 
2973 	switch (cmd) {
2974 	case SIOCSHWTSTAMP:
2975 		if ((lio->oct_dev->chip_id == OCTEON_CN66XX ||
2976 		     lio->oct_dev->chip_id == OCTEON_CN68XX) && ptp_enable)
2977 			return hwtstamp_ioctl(netdev, ifr);
2978 	default:
2979 		return -EOPNOTSUPP;
2980 	}
2981 }
2982 
2983 /**
2984  * \brief handle a Tx timestamp response
2985  * @param status response status
2986  * @param buf pointer to skb
2987  */
2988 static void handle_timestamp(struct octeon_device *oct,
2989 			     u32 status,
2990 			     void *buf)
2991 {
2992 	struct octnet_buf_free_info *finfo;
2993 	struct octeon_soft_command *sc;
2994 	struct oct_timestamp_resp *resp;
2995 	struct lio *lio;
2996 	struct sk_buff *skb = (struct sk_buff *)buf;
2997 
2998 	finfo = (struct octnet_buf_free_info *)skb->cb;
2999 	lio = finfo->lio;
3000 	sc = finfo->sc;
3001 	oct = lio->oct_dev;
3002 	resp = (struct oct_timestamp_resp *)sc->virtrptr;
3003 
3004 	if (status != OCTEON_REQUEST_DONE) {
3005 		dev_err(&oct->pci_dev->dev, "Tx timestamp instruction failed. Status: %llx\n",
3006 			CVM_CAST64(status));
3007 		resp->timestamp = 0;
3008 	}
3009 
3010 	octeon_swap_8B_data(&resp->timestamp, 1);
3011 
3012 	if (unlikely((skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS) != 0)) {
3013 		struct skb_shared_hwtstamps ts;
3014 		u64 ns = resp->timestamp;
3015 
3016 		netif_info(lio, tx_done, lio->netdev,
3017 			   "Got resulting SKBTX_HW_TSTAMP skb=%p ns=%016llu\n",
3018 			   skb, (unsigned long long)ns);
3019 		ts.hwtstamp = ns_to_ktime(ns + lio->ptp_adjust);
3020 		skb_tstamp_tx(skb, &ts);
3021 	}
3022 
3023 	octeon_free_soft_command(oct, sc);
3024 	tx_buffer_free(skb);
3025 }
3026 
3027 /* \brief Send a data packet that will be timestamped
3028  * @param oct octeon device
3029  * @param ndata pointer to network data
3030  * @param finfo pointer to private network data
3031  */
3032 static inline int send_nic_timestamp_pkt(struct octeon_device *oct,
3033 					 struct octnic_data_pkt *ndata,
3034 					 struct octnet_buf_free_info *finfo)
3035 {
3036 	int retval;
3037 	struct octeon_soft_command *sc;
3038 	struct lio *lio;
3039 	int ring_doorbell;
3040 	u32 len;
3041 
3042 	lio = finfo->lio;
3043 
3044 	sc = octeon_alloc_soft_command_resp(oct, &ndata->cmd,
3045 					    sizeof(struct oct_timestamp_resp));
3046 	finfo->sc = sc;
3047 
3048 	if (!sc) {
3049 		dev_err(&oct->pci_dev->dev, "No memory for timestamped data packet\n");
3050 		return IQ_SEND_FAILED;
3051 	}
3052 
3053 	if (ndata->reqtype == REQTYPE_NORESP_NET)
3054 		ndata->reqtype = REQTYPE_RESP_NET;
3055 	else if (ndata->reqtype == REQTYPE_NORESP_NET_SG)
3056 		ndata->reqtype = REQTYPE_RESP_NET_SG;
3057 
3058 	sc->callback = handle_timestamp;
3059 	sc->callback_arg = finfo->skb;
3060 	sc->iq_no = ndata->q_no;
3061 
3062 	if (OCTEON_CN23XX_PF(oct))
3063 		len = (u32)((struct octeon_instr_ih3 *)
3064 			    (&sc->cmd.cmd3.ih3))->dlengsz;
3065 	else
3066 		len = (u32)((struct octeon_instr_ih2 *)
3067 			    (&sc->cmd.cmd2.ih2))->dlengsz;
3068 
3069 	ring_doorbell = 1;
3070 
3071 	retval = octeon_send_command(oct, sc->iq_no, ring_doorbell, &sc->cmd,
3072 				     sc, len, ndata->reqtype);
3073 
3074 	if (retval == IQ_SEND_FAILED) {
3075 		dev_err(&oct->pci_dev->dev, "timestamp data packet failed status: %x\n",
3076 			retval);
3077 		octeon_free_soft_command(oct, sc);
3078 	} else {
3079 		netif_info(lio, tx_queued, lio->netdev, "Queued timestamp packet\n");
3080 	}
3081 
3082 	return retval;
3083 }
3084 
3085 /** \brief Transmit networks packets to the Octeon interface
3086  * @param skbuff   skbuff struct to be passed to network layer.
3087  * @param netdev    pointer to network device
3088  * @returns whether the packet was transmitted to the device okay or not
3089  *             (NETDEV_TX_OK or NETDEV_TX_BUSY)
3090  */
3091 static int liquidio_xmit(struct sk_buff *skb, struct net_device *netdev)
3092 {
3093 	struct lio *lio;
3094 	struct octnet_buf_free_info *finfo;
3095 	union octnic_cmd_setup cmdsetup;
3096 	struct octnic_data_pkt ndata;
3097 	struct octeon_device *oct;
3098 	struct oct_iq_stats *stats;
3099 	struct octeon_instr_irh *irh;
3100 	union tx_info *tx_info;
3101 	int status = 0;
3102 	int q_idx = 0, iq_no = 0;
3103 	int j;
3104 	u64 dptr = 0;
3105 	u32 tag = 0;
3106 
3107 	lio = GET_LIO(netdev);
3108 	oct = lio->oct_dev;
3109 
3110 	if (netif_is_multiqueue(netdev)) {
3111 		q_idx = skb->queue_mapping;
3112 		q_idx = (q_idx % (lio->linfo.num_txpciq));
3113 		tag = q_idx;
3114 		iq_no = lio->linfo.txpciq[q_idx].s.q_no;
3115 	} else {
3116 		iq_no = lio->txq;
3117 	}
3118 
3119 	stats = &oct->instr_queue[iq_no]->stats;
3120 
3121 	/* Check for all conditions in which the current packet cannot be
3122 	 * transmitted.
3123 	 */
3124 	if (!(atomic_read(&lio->ifstate) & LIO_IFSTATE_RUNNING) ||
3125 	    (!lio->linfo.link.s.link_up) ||
3126 	    (skb->len <= 0)) {
3127 		netif_info(lio, tx_err, lio->netdev,
3128 			   "Transmit failed link_status : %d\n",
3129 			   lio->linfo.link.s.link_up);
3130 		goto lio_xmit_failed;
3131 	}
3132 
3133 	/* Use space in skb->cb to store info used to unmap and
3134 	 * free the buffers.
3135 	 */
3136 	finfo = (struct octnet_buf_free_info *)skb->cb;
3137 	finfo->lio = lio;
3138 	finfo->skb = skb;
3139 	finfo->sc = NULL;
3140 
3141 	/* Prepare the attributes for the data to be passed to OSI. */
3142 	memset(&ndata, 0, sizeof(struct octnic_data_pkt));
3143 
3144 	ndata.buf = (void *)finfo;
3145 
3146 	ndata.q_no = iq_no;
3147 
3148 	if (netif_is_multiqueue(netdev)) {
3149 		if (octnet_iq_is_full(oct, ndata.q_no)) {
3150 			/* defer sending if queue is full */
3151 			netif_info(lio, tx_err, lio->netdev, "Transmit failed iq:%d full\n",
3152 				   ndata.q_no);
3153 			stats->tx_iq_busy++;
3154 			return NETDEV_TX_BUSY;
3155 		}
3156 	} else {
3157 		if (octnet_iq_is_full(oct, lio->txq)) {
3158 			/* defer sending if queue is full */
3159 			stats->tx_iq_busy++;
3160 			netif_info(lio, tx_err, lio->netdev, "Transmit failed iq:%d full\n",
3161 				   lio->txq);
3162 			return NETDEV_TX_BUSY;
3163 		}
3164 	}
3165 	/* pr_info(" XMIT - valid Qs: %d, 1st Q no: %d, cpu:  %d, q_no:%d\n",
3166 	 *	lio->linfo.num_txpciq, lio->txq, cpu, ndata.q_no);
3167 	 */
3168 
3169 	ndata.datasize = skb->len;
3170 
3171 	cmdsetup.u64 = 0;
3172 	cmdsetup.s.iq_no = iq_no;
3173 
3174 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
3175 		if (skb->encapsulation) {
3176 			cmdsetup.s.tnl_csum = 1;
3177 			stats->tx_vxlan++;
3178 		} else {
3179 			cmdsetup.s.transport_csum = 1;
3180 		}
3181 	}
3182 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
3183 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
3184 		cmdsetup.s.timestamp = 1;
3185 	}
3186 
3187 	if (skb_shinfo(skb)->nr_frags == 0) {
3188 		cmdsetup.s.u.datasize = skb->len;
3189 		octnet_prepare_pci_cmd(oct, &ndata.cmd, &cmdsetup, tag);
3190 
3191 		/* Offload checksum calculation for TCP/UDP packets */
3192 		dptr = dma_map_single(&oct->pci_dev->dev,
3193 				      skb->data,
3194 				      skb->len,
3195 				      DMA_TO_DEVICE);
3196 		if (dma_mapping_error(&oct->pci_dev->dev, dptr)) {
3197 			dev_err(&oct->pci_dev->dev, "%s DMA mapping error 1\n",
3198 				__func__);
3199 			return NETDEV_TX_BUSY;
3200 		}
3201 
3202 		if (OCTEON_CN23XX_PF(oct))
3203 			ndata.cmd.cmd3.dptr = dptr;
3204 		else
3205 			ndata.cmd.cmd2.dptr = dptr;
3206 		finfo->dptr = dptr;
3207 		ndata.reqtype = REQTYPE_NORESP_NET;
3208 
3209 	} else {
3210 		int i, frags;
3211 		struct skb_frag_struct *frag;
3212 		struct octnic_gather *g;
3213 
3214 		spin_lock(&lio->glist_lock[q_idx]);
3215 		g = (struct octnic_gather *)
3216 			list_delete_head(&lio->glist[q_idx]);
3217 		spin_unlock(&lio->glist_lock[q_idx]);
3218 
3219 		if (!g) {
3220 			netif_info(lio, tx_err, lio->netdev,
3221 				   "Transmit scatter gather: glist null!\n");
3222 			goto lio_xmit_failed;
3223 		}
3224 
3225 		cmdsetup.s.gather = 1;
3226 		cmdsetup.s.u.gatherptrs = (skb_shinfo(skb)->nr_frags + 1);
3227 		octnet_prepare_pci_cmd(oct, &ndata.cmd, &cmdsetup, tag);
3228 
3229 		memset(g->sg, 0, g->sg_size);
3230 
3231 		g->sg[0].ptr[0] = dma_map_single(&oct->pci_dev->dev,
3232 						 skb->data,
3233 						 (skb->len - skb->data_len),
3234 						 DMA_TO_DEVICE);
3235 		if (dma_mapping_error(&oct->pci_dev->dev, g->sg[0].ptr[0])) {
3236 			dev_err(&oct->pci_dev->dev, "%s DMA mapping error 2\n",
3237 				__func__);
3238 			return NETDEV_TX_BUSY;
3239 		}
3240 		add_sg_size(&g->sg[0], (skb->len - skb->data_len), 0);
3241 
3242 		frags = skb_shinfo(skb)->nr_frags;
3243 		i = 1;
3244 		while (frags--) {
3245 			frag = &skb_shinfo(skb)->frags[i - 1];
3246 
3247 			g->sg[(i >> 2)].ptr[(i & 3)] =
3248 				dma_map_page(&oct->pci_dev->dev,
3249 					     frag->page.p,
3250 					     frag->page_offset,
3251 					     frag->size,
3252 					     DMA_TO_DEVICE);
3253 
3254 			if (dma_mapping_error(&oct->pci_dev->dev,
3255 					      g->sg[i >> 2].ptr[i & 3])) {
3256 				dma_unmap_single(&oct->pci_dev->dev,
3257 						 g->sg[0].ptr[0],
3258 						 skb->len - skb->data_len,
3259 						 DMA_TO_DEVICE);
3260 				for (j = 1; j < i; j++) {
3261 					frag = &skb_shinfo(skb)->frags[j - 1];
3262 					dma_unmap_page(&oct->pci_dev->dev,
3263 						       g->sg[j >> 2].ptr[j & 3],
3264 						       frag->size,
3265 						       DMA_TO_DEVICE);
3266 				}
3267 				dev_err(&oct->pci_dev->dev, "%s DMA mapping error 3\n",
3268 					__func__);
3269 				return NETDEV_TX_BUSY;
3270 			}
3271 
3272 			add_sg_size(&g->sg[(i >> 2)], frag->size, (i & 3));
3273 			i++;
3274 		}
3275 
3276 		dma_sync_single_for_device(&oct->pci_dev->dev, g->sg_dma_ptr,
3277 					   g->sg_size, DMA_TO_DEVICE);
3278 		dptr = g->sg_dma_ptr;
3279 
3280 		if (OCTEON_CN23XX_PF(oct))
3281 			ndata.cmd.cmd3.dptr = dptr;
3282 		else
3283 			ndata.cmd.cmd2.dptr = dptr;
3284 		finfo->dptr = dptr;
3285 		finfo->g = g;
3286 
3287 		ndata.reqtype = REQTYPE_NORESP_NET_SG;
3288 	}
3289 
3290 	if (OCTEON_CN23XX_PF(oct)) {
3291 		irh = (struct octeon_instr_irh *)&ndata.cmd.cmd3.irh;
3292 		tx_info = (union tx_info *)&ndata.cmd.cmd3.ossp[0];
3293 	} else {
3294 		irh = (struct octeon_instr_irh *)&ndata.cmd.cmd2.irh;
3295 		tx_info = (union tx_info *)&ndata.cmd.cmd2.ossp[0];
3296 	}
3297 
3298 	if (skb_shinfo(skb)->gso_size) {
3299 		tx_info->s.gso_size = skb_shinfo(skb)->gso_size;
3300 		tx_info->s.gso_segs = skb_shinfo(skb)->gso_segs;
3301 		stats->tx_gso++;
3302 	}
3303 
3304 	/* HW insert VLAN tag */
3305 	if (skb_vlan_tag_present(skb)) {
3306 		irh->priority = skb_vlan_tag_get(skb) >> 13;
3307 		irh->vlan = skb_vlan_tag_get(skb) & 0xfff;
3308 	}
3309 
3310 	if (unlikely(cmdsetup.s.timestamp))
3311 		status = send_nic_timestamp_pkt(oct, &ndata, finfo);
3312 	else
3313 		status = octnet_send_nic_data_pkt(oct, &ndata);
3314 	if (status == IQ_SEND_FAILED)
3315 		goto lio_xmit_failed;
3316 
3317 	netif_info(lio, tx_queued, lio->netdev, "Transmit queued successfully\n");
3318 
3319 	if (status == IQ_SEND_STOP)
3320 		stop_q(lio->netdev, q_idx);
3321 
3322 	netif_trans_update(netdev);
3323 
3324 	if (tx_info->s.gso_segs)
3325 		stats->tx_done += tx_info->s.gso_segs;
3326 	else
3327 		stats->tx_done++;
3328 	stats->tx_tot_bytes += ndata.datasize;
3329 
3330 	return NETDEV_TX_OK;
3331 
3332 lio_xmit_failed:
3333 	stats->tx_dropped++;
3334 	netif_info(lio, tx_err, lio->netdev, "IQ%d Transmit dropped:%llu\n",
3335 		   iq_no, stats->tx_dropped);
3336 	if (dptr)
3337 		dma_unmap_single(&oct->pci_dev->dev, dptr,
3338 				 ndata.datasize, DMA_TO_DEVICE);
3339 	tx_buffer_free(skb);
3340 	return NETDEV_TX_OK;
3341 }
3342 
3343 /** \brief Network device Tx timeout
3344  * @param netdev    pointer to network device
3345  */
3346 static void liquidio_tx_timeout(struct net_device *netdev)
3347 {
3348 	struct lio *lio;
3349 
3350 	lio = GET_LIO(netdev);
3351 
3352 	netif_info(lio, tx_err, lio->netdev,
3353 		   "Transmit timeout tx_dropped:%ld, waking up queues now!!\n",
3354 		   netdev->stats.tx_dropped);
3355 	netif_trans_update(netdev);
3356 	txqs_wake(netdev);
3357 }
3358 
3359 static int liquidio_vlan_rx_add_vid(struct net_device *netdev,
3360 				    __be16 proto __attribute__((unused)),
3361 				    u16 vid)
3362 {
3363 	struct lio *lio = GET_LIO(netdev);
3364 	struct octeon_device *oct = lio->oct_dev;
3365 	struct octnic_ctrl_pkt nctrl;
3366 	int ret = 0;
3367 
3368 	memset(&nctrl, 0, sizeof(struct octnic_ctrl_pkt));
3369 
3370 	nctrl.ncmd.u64 = 0;
3371 	nctrl.ncmd.s.cmd = OCTNET_CMD_ADD_VLAN_FILTER;
3372 	nctrl.ncmd.s.param1 = vid;
3373 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
3374 	nctrl.wait_time = 100;
3375 	nctrl.netpndev = (u64)netdev;
3376 	nctrl.cb_fn = liquidio_link_ctrl_cmd_completion;
3377 
3378 	ret = octnet_send_nic_ctrl_pkt(lio->oct_dev, &nctrl);
3379 	if (ret < 0) {
3380 		dev_err(&oct->pci_dev->dev, "Add VLAN filter failed in core (ret: 0x%x)\n",
3381 			ret);
3382 	}
3383 
3384 	return ret;
3385 }
3386 
3387 static int liquidio_vlan_rx_kill_vid(struct net_device *netdev,
3388 				     __be16 proto __attribute__((unused)),
3389 				     u16 vid)
3390 {
3391 	struct lio *lio = GET_LIO(netdev);
3392 	struct octeon_device *oct = lio->oct_dev;
3393 	struct octnic_ctrl_pkt nctrl;
3394 	int ret = 0;
3395 
3396 	memset(&nctrl, 0, sizeof(struct octnic_ctrl_pkt));
3397 
3398 	nctrl.ncmd.u64 = 0;
3399 	nctrl.ncmd.s.cmd = OCTNET_CMD_DEL_VLAN_FILTER;
3400 	nctrl.ncmd.s.param1 = vid;
3401 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
3402 	nctrl.wait_time = 100;
3403 	nctrl.netpndev = (u64)netdev;
3404 	nctrl.cb_fn = liquidio_link_ctrl_cmd_completion;
3405 
3406 	ret = octnet_send_nic_ctrl_pkt(lio->oct_dev, &nctrl);
3407 	if (ret < 0) {
3408 		dev_err(&oct->pci_dev->dev, "Add VLAN filter failed in core (ret: 0x%x)\n",
3409 			ret);
3410 	}
3411 	return ret;
3412 }
3413 
3414 /** Sending command to enable/disable RX checksum offload
3415  * @param netdev                pointer to network device
3416  * @param command               OCTNET_CMD_TNL_RX_CSUM_CTL
3417  * @param rx_cmd_bit            OCTNET_CMD_RXCSUM_ENABLE/
3418  *                              OCTNET_CMD_RXCSUM_DISABLE
3419  * @returns                     SUCCESS or FAILURE
3420  */
3421 static int liquidio_set_rxcsum_command(struct net_device *netdev, int command,
3422 				       u8 rx_cmd)
3423 {
3424 	struct lio *lio = GET_LIO(netdev);
3425 	struct octeon_device *oct = lio->oct_dev;
3426 	struct octnic_ctrl_pkt nctrl;
3427 	int ret = 0;
3428 
3429 	nctrl.ncmd.u64 = 0;
3430 	nctrl.ncmd.s.cmd = command;
3431 	nctrl.ncmd.s.param1 = rx_cmd;
3432 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
3433 	nctrl.wait_time = 100;
3434 	nctrl.netpndev = (u64)netdev;
3435 	nctrl.cb_fn = liquidio_link_ctrl_cmd_completion;
3436 
3437 	ret = octnet_send_nic_ctrl_pkt(lio->oct_dev, &nctrl);
3438 	if (ret < 0) {
3439 		dev_err(&oct->pci_dev->dev,
3440 			"DEVFLAGS RXCSUM change failed in core(ret:0x%x)\n",
3441 			ret);
3442 	}
3443 	return ret;
3444 }
3445 
3446 /** Sending command to add/delete VxLAN UDP port to firmware
3447  * @param netdev                pointer to network device
3448  * @param command               OCTNET_CMD_VXLAN_PORT_CONFIG
3449  * @param vxlan_port            VxLAN port to be added or deleted
3450  * @param vxlan_cmd_bit         OCTNET_CMD_VXLAN_PORT_ADD,
3451  *                              OCTNET_CMD_VXLAN_PORT_DEL
3452  * @returns                     SUCCESS or FAILURE
3453  */
3454 static int liquidio_vxlan_port_command(struct net_device *netdev, int command,
3455 				       u16 vxlan_port, u8 vxlan_cmd_bit)
3456 {
3457 	struct lio *lio = GET_LIO(netdev);
3458 	struct octeon_device *oct = lio->oct_dev;
3459 	struct octnic_ctrl_pkt nctrl;
3460 	int ret = 0;
3461 
3462 	nctrl.ncmd.u64 = 0;
3463 	nctrl.ncmd.s.cmd = command;
3464 	nctrl.ncmd.s.more = vxlan_cmd_bit;
3465 	nctrl.ncmd.s.param1 = vxlan_port;
3466 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
3467 	nctrl.wait_time = 100;
3468 	nctrl.netpndev = (u64)netdev;
3469 	nctrl.cb_fn = liquidio_link_ctrl_cmd_completion;
3470 
3471 	ret = octnet_send_nic_ctrl_pkt(lio->oct_dev, &nctrl);
3472 	if (ret < 0) {
3473 		dev_err(&oct->pci_dev->dev,
3474 			"VxLAN port add/delete failed in core (ret:0x%x)\n",
3475 			ret);
3476 	}
3477 	return ret;
3478 }
3479 
3480 /** \brief Net device fix features
3481  * @param netdev  pointer to network device
3482  * @param request features requested
3483  * @returns updated features list
3484  */
3485 static netdev_features_t liquidio_fix_features(struct net_device *netdev,
3486 					       netdev_features_t request)
3487 {
3488 	struct lio *lio = netdev_priv(netdev);
3489 
3490 	if ((request & NETIF_F_RXCSUM) &&
3491 	    !(lio->dev_capability & NETIF_F_RXCSUM))
3492 		request &= ~NETIF_F_RXCSUM;
3493 
3494 	if ((request & NETIF_F_HW_CSUM) &&
3495 	    !(lio->dev_capability & NETIF_F_HW_CSUM))
3496 		request &= ~NETIF_F_HW_CSUM;
3497 
3498 	if ((request & NETIF_F_TSO) && !(lio->dev_capability & NETIF_F_TSO))
3499 		request &= ~NETIF_F_TSO;
3500 
3501 	if ((request & NETIF_F_TSO6) && !(lio->dev_capability & NETIF_F_TSO6))
3502 		request &= ~NETIF_F_TSO6;
3503 
3504 	if ((request & NETIF_F_LRO) && !(lio->dev_capability & NETIF_F_LRO))
3505 		request &= ~NETIF_F_LRO;
3506 
3507 	/*Disable LRO if RXCSUM is off */
3508 	if (!(request & NETIF_F_RXCSUM) && (netdev->features & NETIF_F_LRO) &&
3509 	    (lio->dev_capability & NETIF_F_LRO))
3510 		request &= ~NETIF_F_LRO;
3511 
3512 	return request;
3513 }
3514 
3515 /** \brief Net device set features
3516  * @param netdev  pointer to network device
3517  * @param features features to enable/disable
3518  */
3519 static int liquidio_set_features(struct net_device *netdev,
3520 				 netdev_features_t features)
3521 {
3522 	struct lio *lio = netdev_priv(netdev);
3523 
3524 	if (!((netdev->features ^ features) & NETIF_F_LRO))
3525 		return 0;
3526 
3527 	if ((features & NETIF_F_LRO) && (lio->dev_capability & NETIF_F_LRO))
3528 		liquidio_set_feature(netdev, OCTNET_CMD_LRO_ENABLE,
3529 				     OCTNIC_LROIPV4 | OCTNIC_LROIPV6);
3530 	else if (!(features & NETIF_F_LRO) &&
3531 		 (lio->dev_capability & NETIF_F_LRO))
3532 		liquidio_set_feature(netdev, OCTNET_CMD_LRO_DISABLE,
3533 				     OCTNIC_LROIPV4 | OCTNIC_LROIPV6);
3534 
3535 	/* Sending command to firmware to enable/disable RX checksum
3536 	 * offload settings using ethtool
3537 	 */
3538 	if (!(netdev->features & NETIF_F_RXCSUM) &&
3539 	    (lio->enc_dev_capability & NETIF_F_RXCSUM) &&
3540 	    (features & NETIF_F_RXCSUM))
3541 		liquidio_set_rxcsum_command(netdev,
3542 					    OCTNET_CMD_TNL_RX_CSUM_CTL,
3543 					    OCTNET_CMD_RXCSUM_ENABLE);
3544 	else if ((netdev->features & NETIF_F_RXCSUM) &&
3545 		 (lio->enc_dev_capability & NETIF_F_RXCSUM) &&
3546 		 !(features & NETIF_F_RXCSUM))
3547 		liquidio_set_rxcsum_command(netdev, OCTNET_CMD_TNL_RX_CSUM_CTL,
3548 					    OCTNET_CMD_RXCSUM_DISABLE);
3549 
3550 	return 0;
3551 }
3552 
3553 static void liquidio_add_vxlan_port(struct net_device *netdev,
3554 				    struct udp_tunnel_info *ti)
3555 {
3556 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
3557 		return;
3558 
3559 	liquidio_vxlan_port_command(netdev,
3560 				    OCTNET_CMD_VXLAN_PORT_CONFIG,
3561 				    htons(ti->port),
3562 				    OCTNET_CMD_VXLAN_PORT_ADD);
3563 }
3564 
3565 static void liquidio_del_vxlan_port(struct net_device *netdev,
3566 				    struct udp_tunnel_info *ti)
3567 {
3568 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
3569 		return;
3570 
3571 	liquidio_vxlan_port_command(netdev,
3572 				    OCTNET_CMD_VXLAN_PORT_CONFIG,
3573 				    htons(ti->port),
3574 				    OCTNET_CMD_VXLAN_PORT_DEL);
3575 }
3576 
3577 static int __liquidio_set_vf_mac(struct net_device *netdev, int vfidx,
3578 				 u8 *mac, bool is_admin_assigned)
3579 {
3580 	struct lio *lio = GET_LIO(netdev);
3581 	struct octeon_device *oct = lio->oct_dev;
3582 	struct octnic_ctrl_pkt nctrl;
3583 
3584 	if (!is_valid_ether_addr(mac))
3585 		return -EINVAL;
3586 
3587 	if (vfidx < 0 || vfidx >= oct->sriov_info.max_vfs)
3588 		return -EINVAL;
3589 
3590 	memset(&nctrl, 0, sizeof(struct octnic_ctrl_pkt));
3591 
3592 	nctrl.ncmd.u64 = 0;
3593 	nctrl.ncmd.s.cmd = OCTNET_CMD_CHANGE_MACADDR;
3594 	/* vfidx is 0 based, but vf_num (param1) is 1 based */
3595 	nctrl.ncmd.s.param1 = vfidx + 1;
3596 	nctrl.ncmd.s.param2 = (is_admin_assigned ? 1 : 0);
3597 	nctrl.ncmd.s.more = 1;
3598 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
3599 	nctrl.cb_fn = 0;
3600 	nctrl.wait_time = LIO_CMD_WAIT_TM;
3601 
3602 	nctrl.udd[0] = 0;
3603 	/* The MAC Address is presented in network byte order. */
3604 	ether_addr_copy((u8 *)&nctrl.udd[0] + 2, mac);
3605 
3606 	oct->sriov_info.vf_macaddr[vfidx] = nctrl.udd[0];
3607 
3608 	octnet_send_nic_ctrl_pkt(oct, &nctrl);
3609 
3610 	return 0;
3611 }
3612 
3613 static int liquidio_set_vf_mac(struct net_device *netdev, int vfidx, u8 *mac)
3614 {
3615 	struct lio *lio = GET_LIO(netdev);
3616 	struct octeon_device *oct = lio->oct_dev;
3617 	int retval;
3618 
3619 	retval = __liquidio_set_vf_mac(netdev, vfidx, mac, true);
3620 	if (!retval)
3621 		cn23xx_tell_vf_its_macaddr_changed(oct, vfidx, mac);
3622 
3623 	return retval;
3624 }
3625 
3626 static int liquidio_set_vf_vlan(struct net_device *netdev, int vfidx,
3627 				u16 vlan, u8 qos, __be16 vlan_proto)
3628 {
3629 	struct lio *lio = GET_LIO(netdev);
3630 	struct octeon_device *oct = lio->oct_dev;
3631 	struct octnic_ctrl_pkt nctrl;
3632 	u16 vlantci;
3633 
3634 	if (vfidx < 0 || vfidx >= oct->sriov_info.num_vfs_alloced)
3635 		return -EINVAL;
3636 
3637 	if (vlan_proto != htons(ETH_P_8021Q))
3638 		return -EPROTONOSUPPORT;
3639 
3640 	if (vlan >= VLAN_N_VID || qos > 7)
3641 		return -EINVAL;
3642 
3643 	if (vlan)
3644 		vlantci = vlan | (u16)qos << VLAN_PRIO_SHIFT;
3645 	else
3646 		vlantci = 0;
3647 
3648 	if (oct->sriov_info.vf_vlantci[vfidx] == vlantci)
3649 		return 0;
3650 
3651 	memset(&nctrl, 0, sizeof(struct octnic_ctrl_pkt));
3652 
3653 	if (vlan)
3654 		nctrl.ncmd.s.cmd = OCTNET_CMD_ADD_VLAN_FILTER;
3655 	else
3656 		nctrl.ncmd.s.cmd = OCTNET_CMD_DEL_VLAN_FILTER;
3657 
3658 	nctrl.ncmd.s.param1 = vlantci;
3659 	nctrl.ncmd.s.param2 =
3660 	    vfidx + 1; /* vfidx is 0 based, but vf_num (param2) is 1 based */
3661 	nctrl.ncmd.s.more = 0;
3662 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
3663 	nctrl.cb_fn = 0;
3664 	nctrl.wait_time = LIO_CMD_WAIT_TM;
3665 
3666 	octnet_send_nic_ctrl_pkt(oct, &nctrl);
3667 
3668 	oct->sriov_info.vf_vlantci[vfidx] = vlantci;
3669 
3670 	return 0;
3671 }
3672 
3673 static int liquidio_get_vf_config(struct net_device *netdev, int vfidx,
3674 				  struct ifla_vf_info *ivi)
3675 {
3676 	struct lio *lio = GET_LIO(netdev);
3677 	struct octeon_device *oct = lio->oct_dev;
3678 	u8 *macaddr;
3679 
3680 	if (vfidx < 0 || vfidx >= oct->sriov_info.num_vfs_alloced)
3681 		return -EINVAL;
3682 
3683 	ivi->vf = vfidx;
3684 	macaddr = 2 + (u8 *)&oct->sriov_info.vf_macaddr[vfidx];
3685 	ether_addr_copy(&ivi->mac[0], macaddr);
3686 	ivi->vlan = oct->sriov_info.vf_vlantci[vfidx] & VLAN_VID_MASK;
3687 	ivi->qos = oct->sriov_info.vf_vlantci[vfidx] >> VLAN_PRIO_SHIFT;
3688 	ivi->linkstate = oct->sriov_info.vf_linkstate[vfidx];
3689 	return 0;
3690 }
3691 
3692 static int liquidio_set_vf_link_state(struct net_device *netdev, int vfidx,
3693 				      int linkstate)
3694 {
3695 	struct lio *lio = GET_LIO(netdev);
3696 	struct octeon_device *oct = lio->oct_dev;
3697 	struct octnic_ctrl_pkt nctrl;
3698 
3699 	if (vfidx < 0 || vfidx >= oct->sriov_info.num_vfs_alloced)
3700 		return -EINVAL;
3701 
3702 	if (oct->sriov_info.vf_linkstate[vfidx] == linkstate)
3703 		return 0;
3704 
3705 	memset(&nctrl, 0, sizeof(struct octnic_ctrl_pkt));
3706 	nctrl.ncmd.s.cmd = OCTNET_CMD_SET_VF_LINKSTATE;
3707 	nctrl.ncmd.s.param1 =
3708 	    vfidx + 1; /* vfidx is 0 based, but vf_num (param1) is 1 based */
3709 	nctrl.ncmd.s.param2 = linkstate;
3710 	nctrl.ncmd.s.more = 0;
3711 	nctrl.iq_no = lio->linfo.txpciq[0].s.q_no;
3712 	nctrl.cb_fn = 0;
3713 	nctrl.wait_time = LIO_CMD_WAIT_TM;
3714 
3715 	octnet_send_nic_ctrl_pkt(oct, &nctrl);
3716 
3717 	oct->sriov_info.vf_linkstate[vfidx] = linkstate;
3718 
3719 	return 0;
3720 }
3721 
3722 static const struct net_device_ops lionetdevops = {
3723 	.ndo_open		= liquidio_open,
3724 	.ndo_stop		= liquidio_stop,
3725 	.ndo_start_xmit		= liquidio_xmit,
3726 	.ndo_get_stats		= liquidio_get_stats,
3727 	.ndo_set_mac_address	= liquidio_set_mac,
3728 	.ndo_set_rx_mode	= liquidio_set_mcast_list,
3729 	.ndo_tx_timeout		= liquidio_tx_timeout,
3730 
3731 	.ndo_vlan_rx_add_vid    = liquidio_vlan_rx_add_vid,
3732 	.ndo_vlan_rx_kill_vid   = liquidio_vlan_rx_kill_vid,
3733 	.ndo_change_mtu		= liquidio_change_mtu,
3734 	.ndo_do_ioctl		= liquidio_ioctl,
3735 	.ndo_fix_features	= liquidio_fix_features,
3736 	.ndo_set_features	= liquidio_set_features,
3737 	.ndo_udp_tunnel_add	= liquidio_add_vxlan_port,
3738 	.ndo_udp_tunnel_del	= liquidio_del_vxlan_port,
3739 	.ndo_set_vf_mac		= liquidio_set_vf_mac,
3740 	.ndo_set_vf_vlan	= liquidio_set_vf_vlan,
3741 	.ndo_get_vf_config	= liquidio_get_vf_config,
3742 	.ndo_set_vf_link_state  = liquidio_set_vf_link_state,
3743 };
3744 
3745 /** \brief Entry point for the liquidio module
3746  */
3747 static int __init liquidio_init(void)
3748 {
3749 	int i;
3750 	struct handshake *hs;
3751 
3752 	init_completion(&first_stage);
3753 
3754 	octeon_init_device_list(OCTEON_CONFIG_TYPE_DEFAULT);
3755 
3756 	if (liquidio_init_pci())
3757 		return -EINVAL;
3758 
3759 	wait_for_completion_timeout(&first_stage, msecs_to_jiffies(1000));
3760 
3761 	for (i = 0; i < MAX_OCTEON_DEVICES; i++) {
3762 		hs = &handshake[i];
3763 		if (hs->pci_dev) {
3764 			wait_for_completion(&hs->init);
3765 			if (!hs->init_ok) {
3766 				/* init handshake failed */
3767 				dev_err(&hs->pci_dev->dev,
3768 					"Failed to init device\n");
3769 				liquidio_deinit_pci();
3770 				return -EIO;
3771 			}
3772 		}
3773 	}
3774 
3775 	for (i = 0; i < MAX_OCTEON_DEVICES; i++) {
3776 		hs = &handshake[i];
3777 		if (hs->pci_dev) {
3778 			wait_for_completion_timeout(&hs->started,
3779 						    msecs_to_jiffies(30000));
3780 			if (!hs->started_ok) {
3781 				/* starter handshake failed */
3782 				dev_err(&hs->pci_dev->dev,
3783 					"Firmware failed to start\n");
3784 				liquidio_deinit_pci();
3785 				return -EIO;
3786 			}
3787 		}
3788 	}
3789 
3790 	return 0;
3791 }
3792 
3793 static int lio_nic_info(struct octeon_recv_info *recv_info, void *buf)
3794 {
3795 	struct octeon_device *oct = (struct octeon_device *)buf;
3796 	struct octeon_recv_pkt *recv_pkt = recv_info->recv_pkt;
3797 	int gmxport = 0;
3798 	union oct_link_status *ls;
3799 	int i;
3800 
3801 	if (recv_pkt->buffer_size[0] != sizeof(*ls)) {
3802 		dev_err(&oct->pci_dev->dev, "Malformed NIC_INFO, len=%d, ifidx=%d\n",
3803 			recv_pkt->buffer_size[0],
3804 			recv_pkt->rh.r_nic_info.gmxport);
3805 		goto nic_info_err;
3806 	}
3807 
3808 	gmxport = recv_pkt->rh.r_nic_info.gmxport;
3809 	ls = (union oct_link_status *)get_rbd(recv_pkt->buffer_ptr[0]);
3810 
3811 	octeon_swap_8B_data((u64 *)ls, (sizeof(union oct_link_status)) >> 3);
3812 	for (i = 0; i < oct->ifcount; i++) {
3813 		if (oct->props[i].gmxport == gmxport) {
3814 			update_link_status(oct->props[i].netdev, ls);
3815 			break;
3816 		}
3817 	}
3818 
3819 nic_info_err:
3820 	for (i = 0; i < recv_pkt->buffer_count; i++)
3821 		recv_buffer_free(recv_pkt->buffer_ptr[i]);
3822 	octeon_free_recv_info(recv_info);
3823 	return 0;
3824 }
3825 
3826 /**
3827  * \brief Setup network interfaces
3828  * @param octeon_dev  octeon device
3829  *
3830  * Called during init time for each device. It assumes the NIC
3831  * is already up and running.  The link information for each
3832  * interface is passed in link_info.
3833  */
3834 static int setup_nic_devices(struct octeon_device *octeon_dev)
3835 {
3836 	struct lio *lio = NULL;
3837 	struct net_device *netdev;
3838 	u8 mac[6], i, j;
3839 	struct octeon_soft_command *sc;
3840 	struct liquidio_if_cfg_context *ctx;
3841 	struct liquidio_if_cfg_resp *resp;
3842 	struct octdev_props *props;
3843 	int retval, num_iqueues, num_oqueues;
3844 	union oct_nic_if_cfg if_cfg;
3845 	unsigned int base_queue;
3846 	unsigned int gmx_port_id;
3847 	u32 resp_size, ctx_size, data_size;
3848 	u32 ifidx_or_pfnum;
3849 	struct lio_version *vdata;
3850 
3851 	/* This is to handle link status changes */
3852 	octeon_register_dispatch_fn(octeon_dev, OPCODE_NIC,
3853 				    OPCODE_NIC_INFO,
3854 				    lio_nic_info, octeon_dev);
3855 
3856 	/* REQTYPE_RESP_NET and REQTYPE_SOFT_COMMAND do not have free functions.
3857 	 * They are handled directly.
3858 	 */
3859 	octeon_register_reqtype_free_fn(octeon_dev, REQTYPE_NORESP_NET,
3860 					free_netbuf);
3861 
3862 	octeon_register_reqtype_free_fn(octeon_dev, REQTYPE_NORESP_NET_SG,
3863 					free_netsgbuf);
3864 
3865 	octeon_register_reqtype_free_fn(octeon_dev, REQTYPE_RESP_NET_SG,
3866 					free_netsgbuf_with_resp);
3867 
3868 	for (i = 0; i < octeon_dev->ifcount; i++) {
3869 		resp_size = sizeof(struct liquidio_if_cfg_resp);
3870 		ctx_size = sizeof(struct liquidio_if_cfg_context);
3871 		data_size = sizeof(struct lio_version);
3872 		sc = (struct octeon_soft_command *)
3873 			octeon_alloc_soft_command(octeon_dev, data_size,
3874 						  resp_size, ctx_size);
3875 		resp = (struct liquidio_if_cfg_resp *)sc->virtrptr;
3876 		ctx  = (struct liquidio_if_cfg_context *)sc->ctxptr;
3877 		vdata = (struct lio_version *)sc->virtdptr;
3878 
3879 		*((u64 *)vdata) = 0;
3880 		vdata->major = cpu_to_be16(LIQUIDIO_BASE_MAJOR_VERSION);
3881 		vdata->minor = cpu_to_be16(LIQUIDIO_BASE_MINOR_VERSION);
3882 		vdata->micro = cpu_to_be16(LIQUIDIO_BASE_MICRO_VERSION);
3883 
3884 		if (OCTEON_CN23XX_PF(octeon_dev)) {
3885 			num_iqueues = octeon_dev->sriov_info.num_pf_rings;
3886 			num_oqueues = octeon_dev->sriov_info.num_pf_rings;
3887 			base_queue = octeon_dev->sriov_info.pf_srn;
3888 
3889 			gmx_port_id = octeon_dev->pf_num;
3890 			ifidx_or_pfnum = octeon_dev->pf_num;
3891 		} else {
3892 			num_iqueues = CFG_GET_NUM_TXQS_NIC_IF(
3893 						octeon_get_conf(octeon_dev), i);
3894 			num_oqueues = CFG_GET_NUM_RXQS_NIC_IF(
3895 						octeon_get_conf(octeon_dev), i);
3896 			base_queue = CFG_GET_BASE_QUE_NIC_IF(
3897 						octeon_get_conf(octeon_dev), i);
3898 			gmx_port_id = CFG_GET_GMXID_NIC_IF(
3899 						octeon_get_conf(octeon_dev), i);
3900 			ifidx_or_pfnum = i;
3901 		}
3902 
3903 		dev_dbg(&octeon_dev->pci_dev->dev,
3904 			"requesting config for interface %d, iqs %d, oqs %d\n",
3905 			ifidx_or_pfnum, num_iqueues, num_oqueues);
3906 		WRITE_ONCE(ctx->cond, 0);
3907 		ctx->octeon_id = lio_get_device_id(octeon_dev);
3908 		init_waitqueue_head(&ctx->wc);
3909 
3910 		if_cfg.u64 = 0;
3911 		if_cfg.s.num_iqueues = num_iqueues;
3912 		if_cfg.s.num_oqueues = num_oqueues;
3913 		if_cfg.s.base_queue = base_queue;
3914 		if_cfg.s.gmx_port_id = gmx_port_id;
3915 
3916 		sc->iq_no = 0;
3917 
3918 		octeon_prepare_soft_command(octeon_dev, sc, OPCODE_NIC,
3919 					    OPCODE_NIC_IF_CFG, 0,
3920 					    if_cfg.u64, 0);
3921 
3922 		sc->callback = if_cfg_callback;
3923 		sc->callback_arg = sc;
3924 		sc->wait_time = 3000;
3925 
3926 		retval = octeon_send_soft_command(octeon_dev, sc);
3927 		if (retval == IQ_SEND_FAILED) {
3928 			dev_err(&octeon_dev->pci_dev->dev,
3929 				"iq/oq config failed status: %x\n",
3930 				retval);
3931 			/* Soft instr is freed by driver in case of failure. */
3932 			goto setup_nic_dev_fail;
3933 		}
3934 
3935 		/* Sleep on a wait queue till the cond flag indicates that the
3936 		 * response arrived or timed-out.
3937 		 */
3938 		if (sleep_cond(&ctx->wc, &ctx->cond) == -EINTR) {
3939 			dev_err(&octeon_dev->pci_dev->dev, "Wait interrupted\n");
3940 			goto setup_nic_wait_intr;
3941 		}
3942 
3943 		retval = resp->status;
3944 		if (retval) {
3945 			dev_err(&octeon_dev->pci_dev->dev, "iq/oq config failed\n");
3946 			goto setup_nic_dev_fail;
3947 		}
3948 
3949 		octeon_swap_8B_data((u64 *)(&resp->cfg_info),
3950 				    (sizeof(struct liquidio_if_cfg_info)) >> 3);
3951 
3952 		num_iqueues = hweight64(resp->cfg_info.iqmask);
3953 		num_oqueues = hweight64(resp->cfg_info.oqmask);
3954 
3955 		if (!(num_iqueues) || !(num_oqueues)) {
3956 			dev_err(&octeon_dev->pci_dev->dev,
3957 				"Got bad iqueues (%016llx) or oqueues (%016llx) from firmware.\n",
3958 				resp->cfg_info.iqmask,
3959 				resp->cfg_info.oqmask);
3960 			goto setup_nic_dev_fail;
3961 		}
3962 		dev_dbg(&octeon_dev->pci_dev->dev,
3963 			"interface %d, iqmask %016llx, oqmask %016llx, numiqueues %d, numoqueues %d\n",
3964 			i, resp->cfg_info.iqmask, resp->cfg_info.oqmask,
3965 			num_iqueues, num_oqueues);
3966 		netdev = alloc_etherdev_mq(LIO_SIZE, num_iqueues);
3967 
3968 		if (!netdev) {
3969 			dev_err(&octeon_dev->pci_dev->dev, "Device allocation failed\n");
3970 			goto setup_nic_dev_fail;
3971 		}
3972 
3973 		SET_NETDEV_DEV(netdev, &octeon_dev->pci_dev->dev);
3974 
3975 		/* Associate the routines that will handle different
3976 		 * netdev tasks.
3977 		 */
3978 		netdev->netdev_ops = &lionetdevops;
3979 
3980 		lio = GET_LIO(netdev);
3981 
3982 		memset(lio, 0, sizeof(struct lio));
3983 
3984 		lio->ifidx = ifidx_or_pfnum;
3985 
3986 		props = &octeon_dev->props[i];
3987 		props->gmxport = resp->cfg_info.linfo.gmxport;
3988 		props->netdev = netdev;
3989 
3990 		lio->linfo.num_rxpciq = num_oqueues;
3991 		lio->linfo.num_txpciq = num_iqueues;
3992 		for (j = 0; j < num_oqueues; j++) {
3993 			lio->linfo.rxpciq[j].u64 =
3994 				resp->cfg_info.linfo.rxpciq[j].u64;
3995 		}
3996 		for (j = 0; j < num_iqueues; j++) {
3997 			lio->linfo.txpciq[j].u64 =
3998 				resp->cfg_info.linfo.txpciq[j].u64;
3999 		}
4000 		lio->linfo.hw_addr = resp->cfg_info.linfo.hw_addr;
4001 		lio->linfo.gmxport = resp->cfg_info.linfo.gmxport;
4002 		lio->linfo.link.u64 = resp->cfg_info.linfo.link.u64;
4003 
4004 		lio->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4005 
4006 		if (OCTEON_CN23XX_PF(octeon_dev) ||
4007 		    OCTEON_CN6XXX(octeon_dev)) {
4008 			lio->dev_capability = NETIF_F_HIGHDMA
4009 					      | NETIF_F_IP_CSUM
4010 					      | NETIF_F_IPV6_CSUM
4011 					      | NETIF_F_SG | NETIF_F_RXCSUM
4012 					      | NETIF_F_GRO
4013 					      | NETIF_F_TSO | NETIF_F_TSO6
4014 					      | NETIF_F_LRO;
4015 		}
4016 		netif_set_gso_max_size(netdev, OCTNIC_GSO_MAX_SIZE);
4017 
4018 		/*  Copy of transmit encapsulation capabilities:
4019 		 *  TSO, TSO6, Checksums for this device
4020 		 */
4021 		lio->enc_dev_capability = NETIF_F_IP_CSUM
4022 					  | NETIF_F_IPV6_CSUM
4023 					  | NETIF_F_GSO_UDP_TUNNEL
4024 					  | NETIF_F_HW_CSUM | NETIF_F_SG
4025 					  | NETIF_F_RXCSUM
4026 					  | NETIF_F_TSO | NETIF_F_TSO6
4027 					  | NETIF_F_LRO;
4028 
4029 		netdev->hw_enc_features = (lio->enc_dev_capability &
4030 					   ~NETIF_F_LRO);
4031 
4032 		lio->dev_capability |= NETIF_F_GSO_UDP_TUNNEL;
4033 
4034 		netdev->vlan_features = lio->dev_capability;
4035 		/* Add any unchangeable hw features */
4036 		lio->dev_capability |=  NETIF_F_HW_VLAN_CTAG_FILTER |
4037 					NETIF_F_HW_VLAN_CTAG_RX |
4038 					NETIF_F_HW_VLAN_CTAG_TX;
4039 
4040 		netdev->features = (lio->dev_capability & ~NETIF_F_LRO);
4041 
4042 		netdev->hw_features = lio->dev_capability;
4043 		/*HW_VLAN_RX and HW_VLAN_FILTER is always on*/
4044 		netdev->hw_features = netdev->hw_features &
4045 			~NETIF_F_HW_VLAN_CTAG_RX;
4046 
4047 		/* MTU range: 68 - 16000 */
4048 		netdev->min_mtu = LIO_MIN_MTU_SIZE;
4049 		netdev->max_mtu = LIO_MAX_MTU_SIZE;
4050 
4051 		/* Point to the  properties for octeon device to which this
4052 		 * interface belongs.
4053 		 */
4054 		lio->oct_dev = octeon_dev;
4055 		lio->octprops = props;
4056 		lio->netdev = netdev;
4057 
4058 		dev_dbg(&octeon_dev->pci_dev->dev,
4059 			"if%d gmx: %d hw_addr: 0x%llx\n", i,
4060 			lio->linfo.gmxport, CVM_CAST64(lio->linfo.hw_addr));
4061 
4062 		for (j = 0; j < octeon_dev->sriov_info.max_vfs; j++) {
4063 			u8 vfmac[ETH_ALEN];
4064 
4065 			random_ether_addr(&vfmac[0]);
4066 			if (__liquidio_set_vf_mac(netdev, j,
4067 						  &vfmac[0], false)) {
4068 				dev_err(&octeon_dev->pci_dev->dev,
4069 					"Error setting VF%d MAC address\n",
4070 					j);
4071 				goto setup_nic_dev_fail;
4072 			}
4073 		}
4074 
4075 		/* 64-bit swap required on LE machines */
4076 		octeon_swap_8B_data(&lio->linfo.hw_addr, 1);
4077 		for (j = 0; j < 6; j++)
4078 			mac[j] = *((u8 *)(((u8 *)&lio->linfo.hw_addr) + 2 + j));
4079 
4080 		/* Copy MAC Address to OS network device structure */
4081 
4082 		ether_addr_copy(netdev->dev_addr, mac);
4083 
4084 		/* By default all interfaces on a single Octeon uses the same
4085 		 * tx and rx queues
4086 		 */
4087 		lio->txq = lio->linfo.txpciq[0].s.q_no;
4088 		lio->rxq = lio->linfo.rxpciq[0].s.q_no;
4089 		if (setup_io_queues(octeon_dev, i)) {
4090 			dev_err(&octeon_dev->pci_dev->dev, "I/O queues creation failed\n");
4091 			goto setup_nic_dev_fail;
4092 		}
4093 
4094 		ifstate_set(lio, LIO_IFSTATE_DROQ_OPS);
4095 
4096 		lio->tx_qsize = octeon_get_tx_qsize(octeon_dev, lio->txq);
4097 		lio->rx_qsize = octeon_get_rx_qsize(octeon_dev, lio->rxq);
4098 
4099 		if (setup_glists(octeon_dev, lio, num_iqueues)) {
4100 			dev_err(&octeon_dev->pci_dev->dev,
4101 				"Gather list allocation failed\n");
4102 			goto setup_nic_dev_fail;
4103 		}
4104 
4105 		/* Register ethtool support */
4106 		liquidio_set_ethtool_ops(netdev);
4107 		if (lio->oct_dev->chip_id == OCTEON_CN23XX_PF_VID)
4108 			octeon_dev->priv_flags = OCT_PRIV_FLAG_DEFAULT;
4109 		else
4110 			octeon_dev->priv_flags = 0x0;
4111 
4112 		if (netdev->features & NETIF_F_LRO)
4113 			liquidio_set_feature(netdev, OCTNET_CMD_LRO_ENABLE,
4114 					     OCTNIC_LROIPV4 | OCTNIC_LROIPV6);
4115 
4116 		liquidio_set_feature(netdev, OCTNET_CMD_ENABLE_VLAN_FILTER, 0);
4117 
4118 		if ((debug != -1) && (debug & NETIF_MSG_HW))
4119 			liquidio_set_feature(netdev,
4120 					     OCTNET_CMD_VERBOSE_ENABLE, 0);
4121 
4122 		if (setup_link_status_change_wq(netdev))
4123 			goto setup_nic_dev_fail;
4124 
4125 		/* Register the network device with the OS */
4126 		if (register_netdev(netdev)) {
4127 			dev_err(&octeon_dev->pci_dev->dev, "Device registration failed\n");
4128 			goto setup_nic_dev_fail;
4129 		}
4130 
4131 		dev_dbg(&octeon_dev->pci_dev->dev,
4132 			"Setup NIC ifidx:%d mac:%02x%02x%02x%02x%02x%02x\n",
4133 			i, mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
4134 		netif_carrier_off(netdev);
4135 		lio->link_changes++;
4136 
4137 		ifstate_set(lio, LIO_IFSTATE_REGISTERED);
4138 
4139 		/* Sending command to firmware to enable Rx checksum offload
4140 		 * by default at the time of setup of Liquidio driver for
4141 		 * this device
4142 		 */
4143 		liquidio_set_rxcsum_command(netdev, OCTNET_CMD_TNL_RX_CSUM_CTL,
4144 					    OCTNET_CMD_RXCSUM_ENABLE);
4145 		liquidio_set_feature(netdev, OCTNET_CMD_TNL_TX_CSUM_CTL,
4146 				     OCTNET_CMD_TXCSUM_ENABLE);
4147 
4148 		dev_dbg(&octeon_dev->pci_dev->dev,
4149 			"NIC ifidx:%d Setup successful\n", i);
4150 
4151 		octeon_free_soft_command(octeon_dev, sc);
4152 	}
4153 
4154 	return 0;
4155 
4156 setup_nic_dev_fail:
4157 
4158 	octeon_free_soft_command(octeon_dev, sc);
4159 
4160 setup_nic_wait_intr:
4161 
4162 	while (i--) {
4163 		dev_err(&octeon_dev->pci_dev->dev,
4164 			"NIC ifidx:%d Setup failed\n", i);
4165 		liquidio_destroy_nic_device(octeon_dev, i);
4166 	}
4167 	return -ENODEV;
4168 }
4169 
4170 #ifdef CONFIG_PCI_IOV
4171 static int octeon_enable_sriov(struct octeon_device *oct)
4172 {
4173 	unsigned int num_vfs_alloced = oct->sriov_info.num_vfs_alloced;
4174 	struct pci_dev *vfdev;
4175 	int err;
4176 	u32 u;
4177 
4178 	if (OCTEON_CN23XX_PF(oct) && num_vfs_alloced) {
4179 		err = pci_enable_sriov(oct->pci_dev,
4180 				       oct->sriov_info.num_vfs_alloced);
4181 		if (err) {
4182 			dev_err(&oct->pci_dev->dev,
4183 				"OCTEON: Failed to enable PCI sriov: %d\n",
4184 				err);
4185 			oct->sriov_info.num_vfs_alloced = 0;
4186 			return err;
4187 		}
4188 		oct->sriov_info.sriov_enabled = 1;
4189 
4190 		/* init lookup table that maps DPI ring number to VF pci_dev
4191 		 * struct pointer
4192 		 */
4193 		u = 0;
4194 		vfdev = pci_get_device(PCI_VENDOR_ID_CAVIUM,
4195 				       OCTEON_CN23XX_VF_VID, NULL);
4196 		while (vfdev) {
4197 			if (vfdev->is_virtfn &&
4198 			    (vfdev->physfn == oct->pci_dev)) {
4199 				oct->sriov_info.dpiring_to_vfpcidev_lut[u] =
4200 					vfdev;
4201 				u += oct->sriov_info.rings_per_vf;
4202 			}
4203 			vfdev = pci_get_device(PCI_VENDOR_ID_CAVIUM,
4204 					       OCTEON_CN23XX_VF_VID, vfdev);
4205 		}
4206 	}
4207 
4208 	return num_vfs_alloced;
4209 }
4210 
4211 static int lio_pci_sriov_disable(struct octeon_device *oct)
4212 {
4213 	int u;
4214 
4215 	if (pci_vfs_assigned(oct->pci_dev)) {
4216 		dev_err(&oct->pci_dev->dev, "VFs are still assigned to VMs.\n");
4217 		return -EPERM;
4218 	}
4219 
4220 	pci_disable_sriov(oct->pci_dev);
4221 
4222 	u = 0;
4223 	while (u < MAX_POSSIBLE_VFS) {
4224 		oct->sriov_info.dpiring_to_vfpcidev_lut[u] = NULL;
4225 		u += oct->sriov_info.rings_per_vf;
4226 	}
4227 
4228 	oct->sriov_info.num_vfs_alloced = 0;
4229 	dev_info(&oct->pci_dev->dev, "oct->pf_num:%d disabled VFs\n",
4230 		 oct->pf_num);
4231 
4232 	return 0;
4233 }
4234 
4235 static int liquidio_enable_sriov(struct pci_dev *dev, int num_vfs)
4236 {
4237 	struct octeon_device *oct = pci_get_drvdata(dev);
4238 	int ret = 0;
4239 
4240 	if ((num_vfs == oct->sriov_info.num_vfs_alloced) &&
4241 	    (oct->sriov_info.sriov_enabled)) {
4242 		dev_info(&oct->pci_dev->dev, "oct->pf_num:%d already enabled num_vfs:%d\n",
4243 			 oct->pf_num, num_vfs);
4244 		return 0;
4245 	}
4246 
4247 	if (!num_vfs) {
4248 		ret = lio_pci_sriov_disable(oct);
4249 	} else if (num_vfs > oct->sriov_info.max_vfs) {
4250 		dev_err(&oct->pci_dev->dev,
4251 			"OCTEON: Max allowed VFs:%d user requested:%d",
4252 			oct->sriov_info.max_vfs, num_vfs);
4253 		ret = -EPERM;
4254 	} else {
4255 		oct->sriov_info.num_vfs_alloced = num_vfs;
4256 		ret = octeon_enable_sriov(oct);
4257 		dev_info(&oct->pci_dev->dev, "oct->pf_num:%d num_vfs:%d\n",
4258 			 oct->pf_num, num_vfs);
4259 	}
4260 
4261 	return ret;
4262 }
4263 #endif
4264 
4265 /**
4266  * \brief initialize the NIC
4267  * @param oct octeon device
4268  *
4269  * This initialization routine is called once the Octeon device application is
4270  * up and running
4271  */
4272 static int liquidio_init_nic_module(struct octeon_device *oct)
4273 {
4274 	struct oct_intrmod_cfg *intrmod_cfg;
4275 	int i, retval = 0;
4276 	int num_nic_ports = CFG_GET_NUM_NIC_PORTS(octeon_get_conf(oct));
4277 
4278 	dev_dbg(&oct->pci_dev->dev, "Initializing network interfaces\n");
4279 
4280 	/* only default iq and oq were initialized
4281 	 * initialize the rest as well
4282 	 */
4283 	/* run port_config command for each port */
4284 	oct->ifcount = num_nic_ports;
4285 
4286 	memset(oct->props, 0, sizeof(struct octdev_props) * num_nic_ports);
4287 
4288 	for (i = 0; i < MAX_OCTEON_LINKS; i++)
4289 		oct->props[i].gmxport = -1;
4290 
4291 	retval = setup_nic_devices(oct);
4292 	if (retval) {
4293 		dev_err(&oct->pci_dev->dev, "Setup NIC devices failed\n");
4294 		goto octnet_init_failure;
4295 	}
4296 
4297 	liquidio_ptp_init(oct);
4298 
4299 	/* Initialize interrupt moderation params */
4300 	intrmod_cfg = &((struct octeon_device *)oct)->intrmod;
4301 	intrmod_cfg->rx_enable = 1;
4302 	intrmod_cfg->check_intrvl = LIO_INTRMOD_CHECK_INTERVAL;
4303 	intrmod_cfg->maxpkt_ratethr = LIO_INTRMOD_MAXPKT_RATETHR;
4304 	intrmod_cfg->minpkt_ratethr = LIO_INTRMOD_MINPKT_RATETHR;
4305 	intrmod_cfg->rx_maxcnt_trigger = LIO_INTRMOD_RXMAXCNT_TRIGGER;
4306 	intrmod_cfg->rx_maxtmr_trigger = LIO_INTRMOD_RXMAXTMR_TRIGGER;
4307 	intrmod_cfg->rx_mintmr_trigger = LIO_INTRMOD_RXMINTMR_TRIGGER;
4308 	intrmod_cfg->rx_mincnt_trigger = LIO_INTRMOD_RXMINCNT_TRIGGER;
4309 	intrmod_cfg->tx_enable = 1;
4310 	intrmod_cfg->tx_maxcnt_trigger = LIO_INTRMOD_TXMAXCNT_TRIGGER;
4311 	intrmod_cfg->tx_mincnt_trigger = LIO_INTRMOD_TXMINCNT_TRIGGER;
4312 	intrmod_cfg->rx_frames = CFG_GET_OQ_INTR_PKT(octeon_get_conf(oct));
4313 	intrmod_cfg->rx_usecs = CFG_GET_OQ_INTR_TIME(octeon_get_conf(oct));
4314 	intrmod_cfg->tx_frames = CFG_GET_IQ_INTR_PKT(octeon_get_conf(oct));
4315 	dev_dbg(&oct->pci_dev->dev, "Network interfaces ready\n");
4316 
4317 	return retval;
4318 
4319 octnet_init_failure:
4320 
4321 	oct->ifcount = 0;
4322 
4323 	return retval;
4324 }
4325 
4326 /**
4327  * \brief starter callback that invokes the remaining initialization work after
4328  * the NIC is up and running.
4329  * @param octptr  work struct work_struct
4330  */
4331 static void nic_starter(struct work_struct *work)
4332 {
4333 	struct octeon_device *oct;
4334 	struct cavium_wk *wk = (struct cavium_wk *)work;
4335 
4336 	oct = (struct octeon_device *)wk->ctxptr;
4337 
4338 	if (atomic_read(&oct->status) == OCT_DEV_RUNNING)
4339 		return;
4340 
4341 	/* If the status of the device is CORE_OK, the core
4342 	 * application has reported its application type. Call
4343 	 * any registered handlers now and move to the RUNNING
4344 	 * state.
4345 	 */
4346 	if (atomic_read(&oct->status) != OCT_DEV_CORE_OK) {
4347 		schedule_delayed_work(&oct->nic_poll_work.work,
4348 				      LIQUIDIO_STARTER_POLL_INTERVAL_MS);
4349 		return;
4350 	}
4351 
4352 	atomic_set(&oct->status, OCT_DEV_RUNNING);
4353 
4354 	if (oct->app_mode && oct->app_mode == CVM_DRV_NIC_APP) {
4355 		dev_dbg(&oct->pci_dev->dev, "Starting NIC module\n");
4356 
4357 		if (liquidio_init_nic_module(oct))
4358 			dev_err(&oct->pci_dev->dev, "NIC initialization failed\n");
4359 		else
4360 			handshake[oct->octeon_id].started_ok = 1;
4361 	} else {
4362 		dev_err(&oct->pci_dev->dev,
4363 			"Unexpected application running on NIC (%d). Check firmware.\n",
4364 			oct->app_mode);
4365 	}
4366 
4367 	complete(&handshake[oct->octeon_id].started);
4368 }
4369 
4370 static int
4371 octeon_recv_vf_drv_notice(struct octeon_recv_info *recv_info, void *buf)
4372 {
4373 	struct octeon_device *oct = (struct octeon_device *)buf;
4374 	struct octeon_recv_pkt *recv_pkt = recv_info->recv_pkt;
4375 	int i, notice, vf_idx;
4376 	u64 *data, vf_num;
4377 
4378 	notice = recv_pkt->rh.r.ossp;
4379 	data = (u64 *)get_rbd(recv_pkt->buffer_ptr[0]);
4380 
4381 	/* the first 64-bit word of data is the vf_num */
4382 	vf_num = data[0];
4383 	octeon_swap_8B_data(&vf_num, 1);
4384 	vf_idx = (int)vf_num - 1;
4385 
4386 	if (notice == VF_DRV_LOADED) {
4387 		if (!(oct->sriov_info.vf_drv_loaded_mask & BIT_ULL(vf_idx))) {
4388 			oct->sriov_info.vf_drv_loaded_mask |= BIT_ULL(vf_idx);
4389 			dev_info(&oct->pci_dev->dev,
4390 				 "driver for VF%d was loaded\n", vf_idx);
4391 			try_module_get(THIS_MODULE);
4392 		}
4393 	} else if (notice == VF_DRV_REMOVED) {
4394 		if (oct->sriov_info.vf_drv_loaded_mask & BIT_ULL(vf_idx)) {
4395 			oct->sriov_info.vf_drv_loaded_mask &= ~BIT_ULL(vf_idx);
4396 			dev_info(&oct->pci_dev->dev,
4397 				 "driver for VF%d was removed\n", vf_idx);
4398 			module_put(THIS_MODULE);
4399 		}
4400 	} else if (notice == VF_DRV_MACADDR_CHANGED) {
4401 		u8 *b = (u8 *)&data[1];
4402 
4403 		oct->sriov_info.vf_macaddr[vf_idx] = data[1];
4404 		dev_info(&oct->pci_dev->dev,
4405 			 "VF driver changed VF%d's MAC address to %pM\n",
4406 			 vf_idx, b + 2);
4407 	}
4408 
4409 	for (i = 0; i < recv_pkt->buffer_count; i++)
4410 		recv_buffer_free(recv_pkt->buffer_ptr[i]);
4411 	octeon_free_recv_info(recv_info);
4412 
4413 	return 0;
4414 }
4415 
4416 /**
4417  * \brief Device initialization for each Octeon device that is probed
4418  * @param octeon_dev  octeon device
4419  */
4420 static int octeon_device_init(struct octeon_device *octeon_dev)
4421 {
4422 	int j, ret;
4423 	int fw_loaded = 0;
4424 	char bootcmd[] = "\n";
4425 	struct octeon_device_priv *oct_priv =
4426 		(struct octeon_device_priv *)octeon_dev->priv;
4427 	atomic_set(&octeon_dev->status, OCT_DEV_BEGIN_STATE);
4428 
4429 	/* Enable access to the octeon device and make its DMA capability
4430 	 * known to the OS.
4431 	 */
4432 	if (octeon_pci_os_setup(octeon_dev))
4433 		return 1;
4434 
4435 	atomic_set(&octeon_dev->status, OCT_DEV_PCI_ENABLE_DONE);
4436 
4437 	/* Identify the Octeon type and map the BAR address space. */
4438 	if (octeon_chip_specific_setup(octeon_dev)) {
4439 		dev_err(&octeon_dev->pci_dev->dev, "Chip specific setup failed\n");
4440 		return 1;
4441 	}
4442 
4443 	atomic_set(&octeon_dev->status, OCT_DEV_PCI_MAP_DONE);
4444 
4445 	octeon_dev->app_mode = CVM_DRV_INVALID_APP;
4446 
4447 	if (OCTEON_CN23XX_PF(octeon_dev)) {
4448 		if (!cn23xx_fw_loaded(octeon_dev)) {
4449 			fw_loaded = 0;
4450 			/* Do a soft reset of the Octeon device. */
4451 			if (octeon_dev->fn_list.soft_reset(octeon_dev))
4452 				return 1;
4453 			/* things might have changed */
4454 			if (!cn23xx_fw_loaded(octeon_dev))
4455 				fw_loaded = 0;
4456 			else
4457 				fw_loaded = 1;
4458 		} else {
4459 			fw_loaded = 1;
4460 		}
4461 	} else if (octeon_dev->fn_list.soft_reset(octeon_dev)) {
4462 		return 1;
4463 	}
4464 
4465 	/* Initialize the dispatch mechanism used to push packets arriving on
4466 	 * Octeon Output queues.
4467 	 */
4468 	if (octeon_init_dispatch_list(octeon_dev))
4469 		return 1;
4470 
4471 	octeon_register_dispatch_fn(octeon_dev, OPCODE_NIC,
4472 				    OPCODE_NIC_CORE_DRV_ACTIVE,
4473 				    octeon_core_drv_init,
4474 				    octeon_dev);
4475 
4476 	octeon_register_dispatch_fn(octeon_dev, OPCODE_NIC,
4477 				    OPCODE_NIC_VF_DRV_NOTICE,
4478 				    octeon_recv_vf_drv_notice, octeon_dev);
4479 	INIT_DELAYED_WORK(&octeon_dev->nic_poll_work.work, nic_starter);
4480 	octeon_dev->nic_poll_work.ctxptr = (void *)octeon_dev;
4481 	schedule_delayed_work(&octeon_dev->nic_poll_work.work,
4482 			      LIQUIDIO_STARTER_POLL_INTERVAL_MS);
4483 
4484 	atomic_set(&octeon_dev->status, OCT_DEV_DISPATCH_INIT_DONE);
4485 
4486 	if (octeon_set_io_queues_off(octeon_dev)) {
4487 		dev_err(&octeon_dev->pci_dev->dev, "setting io queues off failed\n");
4488 		return 1;
4489 	}
4490 
4491 	if (OCTEON_CN23XX_PF(octeon_dev)) {
4492 		ret = octeon_dev->fn_list.setup_device_regs(octeon_dev);
4493 		if (ret) {
4494 			dev_err(&octeon_dev->pci_dev->dev, "OCTEON: Failed to configure device registers\n");
4495 			return ret;
4496 		}
4497 	}
4498 
4499 	/* Initialize soft command buffer pool
4500 	 */
4501 	if (octeon_setup_sc_buffer_pool(octeon_dev)) {
4502 		dev_err(&octeon_dev->pci_dev->dev, "sc buffer pool allocation failed\n");
4503 		return 1;
4504 	}
4505 	atomic_set(&octeon_dev->status, OCT_DEV_SC_BUFF_POOL_INIT_DONE);
4506 
4507 	/*  Setup the data structures that manage this Octeon's Input queues. */
4508 	if (octeon_setup_instr_queues(octeon_dev)) {
4509 		dev_err(&octeon_dev->pci_dev->dev,
4510 			"instruction queue initialization failed\n");
4511 		return 1;
4512 	}
4513 	atomic_set(&octeon_dev->status, OCT_DEV_INSTR_QUEUE_INIT_DONE);
4514 
4515 	/* Initialize lists to manage the requests of different types that
4516 	 * arrive from user & kernel applications for this octeon device.
4517 	 */
4518 	if (octeon_setup_response_list(octeon_dev)) {
4519 		dev_err(&octeon_dev->pci_dev->dev, "Response list allocation failed\n");
4520 		return 1;
4521 	}
4522 	atomic_set(&octeon_dev->status, OCT_DEV_RESP_LIST_INIT_DONE);
4523 
4524 	if (octeon_setup_output_queues(octeon_dev)) {
4525 		dev_err(&octeon_dev->pci_dev->dev, "Output queue initialization failed\n");
4526 		return 1;
4527 	}
4528 
4529 	atomic_set(&octeon_dev->status, OCT_DEV_DROQ_INIT_DONE);
4530 
4531 	if (OCTEON_CN23XX_PF(octeon_dev)) {
4532 		if (octeon_dev->fn_list.setup_mbox(octeon_dev)) {
4533 			dev_err(&octeon_dev->pci_dev->dev, "OCTEON: Mailbox setup failed\n");
4534 			return 1;
4535 		}
4536 		atomic_set(&octeon_dev->status, OCT_DEV_MBOX_SETUP_DONE);
4537 
4538 		if (octeon_allocate_ioq_vector(octeon_dev)) {
4539 			dev_err(&octeon_dev->pci_dev->dev, "OCTEON: ioq vector allocation failed\n");
4540 			return 1;
4541 		}
4542 		atomic_set(&octeon_dev->status, OCT_DEV_MSIX_ALLOC_VECTOR_DONE);
4543 
4544 	} else {
4545 		/* The input and output queue registers were setup earlier (the
4546 		 * queues were not enabled). Any additional registers
4547 		 * that need to be programmed should be done now.
4548 		 */
4549 		ret = octeon_dev->fn_list.setup_device_regs(octeon_dev);
4550 		if (ret) {
4551 			dev_err(&octeon_dev->pci_dev->dev,
4552 				"Failed to configure device registers\n");
4553 			return ret;
4554 		}
4555 	}
4556 
4557 	/* Initialize the tasklet that handles output queue packet processing.*/
4558 	dev_dbg(&octeon_dev->pci_dev->dev, "Initializing droq tasklet\n");
4559 	tasklet_init(&oct_priv->droq_tasklet, octeon_droq_bh,
4560 		     (unsigned long)octeon_dev);
4561 
4562 	/* Setup the interrupt handler and record the INT SUM register address
4563 	 */
4564 	if (octeon_setup_interrupt(octeon_dev))
4565 		return 1;
4566 
4567 	/* Enable Octeon device interrupts */
4568 	octeon_dev->fn_list.enable_interrupt(octeon_dev, OCTEON_ALL_INTR);
4569 
4570 	atomic_set(&octeon_dev->status, OCT_DEV_INTR_SET_DONE);
4571 
4572 	/* Enable the input and output queues for this Octeon device */
4573 	ret = octeon_dev->fn_list.enable_io_queues(octeon_dev);
4574 	if (ret) {
4575 		dev_err(&octeon_dev->pci_dev->dev, "Failed to enable input/output queues");
4576 		return ret;
4577 	}
4578 
4579 	atomic_set(&octeon_dev->status, OCT_DEV_IO_QUEUES_DONE);
4580 
4581 	if ((!OCTEON_CN23XX_PF(octeon_dev)) || !fw_loaded) {
4582 		dev_dbg(&octeon_dev->pci_dev->dev, "Waiting for DDR initialization...\n");
4583 		if (!ddr_timeout) {
4584 			dev_info(&octeon_dev->pci_dev->dev,
4585 				 "WAITING. Set ddr_timeout to non-zero value to proceed with initialization.\n");
4586 		}
4587 
4588 		schedule_timeout_uninterruptible(HZ * LIO_RESET_SECS);
4589 
4590 		/* Wait for the octeon to initialize DDR after the soft-reset.*/
4591 		while (!ddr_timeout) {
4592 			set_current_state(TASK_INTERRUPTIBLE);
4593 			if (schedule_timeout(HZ / 10)) {
4594 				/* user probably pressed Control-C */
4595 				return 1;
4596 			}
4597 		}
4598 		ret = octeon_wait_for_ddr_init(octeon_dev, &ddr_timeout);
4599 		if (ret) {
4600 			dev_err(&octeon_dev->pci_dev->dev,
4601 				"DDR not initialized. Please confirm that board is configured to boot from Flash, ret: %d\n",
4602 				ret);
4603 			return 1;
4604 		}
4605 
4606 		if (octeon_wait_for_bootloader(octeon_dev, 1000)) {
4607 			dev_err(&octeon_dev->pci_dev->dev, "Board not responding\n");
4608 			return 1;
4609 		}
4610 
4611 		/* Divert uboot to take commands from host instead. */
4612 		ret = octeon_console_send_cmd(octeon_dev, bootcmd, 50);
4613 
4614 		dev_dbg(&octeon_dev->pci_dev->dev, "Initializing consoles\n");
4615 		ret = octeon_init_consoles(octeon_dev);
4616 		if (ret) {
4617 			dev_err(&octeon_dev->pci_dev->dev, "Could not access board consoles\n");
4618 			return 1;
4619 		}
4620 		ret = octeon_add_console(octeon_dev, 0);
4621 		if (ret) {
4622 			dev_err(&octeon_dev->pci_dev->dev, "Could not access board console\n");
4623 			return 1;
4624 		}
4625 
4626 		atomic_set(&octeon_dev->status, OCT_DEV_CONSOLE_INIT_DONE);
4627 
4628 		dev_dbg(&octeon_dev->pci_dev->dev, "Loading firmware\n");
4629 		ret = load_firmware(octeon_dev);
4630 		if (ret) {
4631 			dev_err(&octeon_dev->pci_dev->dev, "Could not load firmware to board\n");
4632 			return 1;
4633 		}
4634 		/* set bit 1 of SLI_SCRATCH_1 to indicate that firmware is
4635 		 * loaded
4636 		 */
4637 		if (OCTEON_CN23XX_PF(octeon_dev))
4638 			octeon_write_csr64(octeon_dev, CN23XX_SLI_SCRATCH1,
4639 					   2ULL);
4640 	}
4641 
4642 	handshake[octeon_dev->octeon_id].init_ok = 1;
4643 	complete(&handshake[octeon_dev->octeon_id].init);
4644 
4645 	atomic_set(&octeon_dev->status, OCT_DEV_HOST_OK);
4646 
4647 	/* Send Credit for Octeon Output queues. Credits are always sent after
4648 	 * the output queue is enabled.
4649 	 */
4650 	for (j = 0; j < octeon_dev->num_oqs; j++)
4651 		writel(octeon_dev->droq[j]->max_count,
4652 		       octeon_dev->droq[j]->pkts_credit_reg);
4653 
4654 	/* Packets can start arriving on the output queues from this point. */
4655 	return 0;
4656 }
4657 
4658 /**
4659  * \brief Exits the module
4660  */
4661 static void __exit liquidio_exit(void)
4662 {
4663 	liquidio_deinit_pci();
4664 
4665 	pr_info("LiquidIO network module is now unloaded\n");
4666 }
4667 
4668 module_init(liquidio_init);
4669 module_exit(liquidio_exit);
4670