1// SPDX-License-Identifier: GPL-2.0
2/* Marvell Octeon EP (EndPoint) Ethernet Driver
3 *
4 * Copyright (C) 2020 Marvell.
5 *
6 */
7
8#include <linux/types.h>
9#include <linux/module.h>
10#include <linux/pci.h>
11#include <linux/netdevice.h>
12#include <linux/etherdevice.h>
13#include <linux/rtnetlink.h>
14#include <linux/vmalloc.h>
15
16#include "octep_config.h"
17#include "octep_main.h"
18#include "octep_ctrl_net.h"
19#include "octep_pfvf_mbox.h"
20
21#define OCTEP_INTR_POLL_TIME_MSECS    100
22struct workqueue_struct *octep_wq;
23
24/* Supported Devices */
25static const struct pci_device_id octep_pci_id_tbl[] = {
26	{PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, OCTEP_PCI_DEVICE_ID_CN98_PF)},
27	{PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, OCTEP_PCI_DEVICE_ID_CN93_PF)},
28	{PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, OCTEP_PCI_DEVICE_ID_CNF95N_PF)},
29	{PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, OCTEP_PCI_DEVICE_ID_CN10KA_PF)},
30	{PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, OCTEP_PCI_DEVICE_ID_CNF10KA_PF)},
31	{PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, OCTEP_PCI_DEVICE_ID_CNF10KB_PF)},
32	{PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, OCTEP_PCI_DEVICE_ID_CN10KB_PF)},
33	{0, },
34};
35MODULE_DEVICE_TABLE(pci, octep_pci_id_tbl);
36
37MODULE_AUTHOR("Veerasenareddy Burru <vburru@marvell.com>");
38MODULE_DESCRIPTION(OCTEP_DRV_STRING);
39MODULE_LICENSE("GPL");
40
41/**
42 * octep_alloc_ioq_vectors() - Allocate Tx/Rx Queue interrupt info.
43 *
44 * @oct: Octeon device private data structure.
45 *
46 * Allocate resources to hold per Tx/Rx queue interrupt info.
47 * This is the information passed to interrupt handler, from which napi poll
48 * is scheduled and includes quick access to private data of Tx/Rx queue
49 * corresponding to the interrupt being handled.
50 *
51 * Return: 0, on successful allocation of resources for all queue interrupts.
52 *         -1, if failed to allocate any resource.
53 */
54static int octep_alloc_ioq_vectors(struct octep_device *oct)
55{
56	int i;
57	struct octep_ioq_vector *ioq_vector;
58
59	for (i = 0; i < oct->num_oqs; i++) {
60		oct->ioq_vector[i] = vzalloc(sizeof(*oct->ioq_vector[i]));
61		if (!oct->ioq_vector[i])
62			goto free_ioq_vector;
63
64		ioq_vector = oct->ioq_vector[i];
65		ioq_vector->iq = oct->iq[i];
66		ioq_vector->oq = oct->oq[i];
67		ioq_vector->octep_dev = oct;
68	}
69
70	dev_info(&oct->pdev->dev, "Allocated %d IOQ vectors\n", oct->num_oqs);
71	return 0;
72
73free_ioq_vector:
74	while (i) {
75		i--;
76		vfree(oct->ioq_vector[i]);
77		oct->ioq_vector[i] = NULL;
78	}
79	return -1;
80}
81
82/**
83 * octep_free_ioq_vectors() - Free Tx/Rx Queue interrupt vector info.
84 *
85 * @oct: Octeon device private data structure.
86 */
87static void octep_free_ioq_vectors(struct octep_device *oct)
88{
89	int i;
90
91	for (i = 0; i < oct->num_oqs; i++) {
92		if (oct->ioq_vector[i]) {
93			vfree(oct->ioq_vector[i]);
94			oct->ioq_vector[i] = NULL;
95		}
96	}
97	netdev_info(oct->netdev, "Freed IOQ Vectors\n");
98}
99
100/**
101 * octep_enable_msix_range() - enable MSI-x interrupts.
102 *
103 * @oct: Octeon device private data structure.
104 *
105 * Allocate and enable all MSI-x interrupts (queue and non-queue interrupts)
106 * for the Octeon device.
107 *
108 * Return: 0, on successfully enabling all MSI-x interrupts.
109 *         -1, if failed to enable any MSI-x interrupt.
110 */
111static int octep_enable_msix_range(struct octep_device *oct)
112{
113	int num_msix, msix_allocated;
114	int i;
115
116	/* Generic interrupts apart from input/output queues */
117	num_msix = oct->num_oqs + CFG_GET_NON_IOQ_MSIX(oct->conf);
118	oct->msix_entries = kcalloc(num_msix,
119				    sizeof(struct msix_entry), GFP_KERNEL);
120	if (!oct->msix_entries)
121		goto msix_alloc_err;
122
123	for (i = 0; i < num_msix; i++)
124		oct->msix_entries[i].entry = i;
125
126	msix_allocated = pci_enable_msix_range(oct->pdev, oct->msix_entries,
127					       num_msix, num_msix);
128	if (msix_allocated != num_msix) {
129		dev_err(&oct->pdev->dev,
130			"Failed to enable %d msix irqs; got only %d\n",
131			num_msix, msix_allocated);
132		goto enable_msix_err;
133	}
134	oct->num_irqs = msix_allocated;
135	dev_info(&oct->pdev->dev, "MSI-X enabled successfully\n");
136
137	return 0;
138
139enable_msix_err:
140	if (msix_allocated > 0)
141		pci_disable_msix(oct->pdev);
142	kfree(oct->msix_entries);
143	oct->msix_entries = NULL;
144msix_alloc_err:
145	return -1;
146}
147
148/**
149 * octep_disable_msix() - disable MSI-x interrupts.
150 *
151 * @oct: Octeon device private data structure.
152 *
153 * Disable MSI-x on the Octeon device.
154 */
155static void octep_disable_msix(struct octep_device *oct)
156{
157	pci_disable_msix(oct->pdev);
158	kfree(oct->msix_entries);
159	oct->msix_entries = NULL;
160	dev_info(&oct->pdev->dev, "Disabled MSI-X\n");
161}
162
163/**
164 * octep_mbox_intr_handler() - common handler for pfvf mbox interrupts.
165 *
166 * @irq: Interrupt number.
167 * @data: interrupt data.
168 *
169 * this is common handler for pfvf mbox interrupts.
170 */
171static irqreturn_t octep_mbox_intr_handler(int irq, void *data)
172{
173	struct octep_device *oct = data;
174
175	return oct->hw_ops.mbox_intr_handler(oct);
176}
177
178/**
179 * octep_oei_intr_handler() - common handler for output endpoint interrupts.
180 *
181 * @irq: Interrupt number.
182 * @data: interrupt data.
183 *
184 * this is common handler for all output endpoint interrupts.
185 */
186static irqreturn_t octep_oei_intr_handler(int irq, void *data)
187{
188	struct octep_device *oct = data;
189
190	return oct->hw_ops.oei_intr_handler(oct);
191}
192
193/**
194 * octep_ire_intr_handler() - common handler for input ring error interrupts.
195 *
196 * @irq: Interrupt number.
197 * @data: interrupt data.
198 *
199 * this is common handler for input ring error interrupts.
200 */
201static irqreturn_t octep_ire_intr_handler(int irq, void *data)
202{
203	struct octep_device *oct = data;
204
205	return oct->hw_ops.ire_intr_handler(oct);
206}
207
208/**
209 * octep_ore_intr_handler() - common handler for output ring error interrupts.
210 *
211 * @irq: Interrupt number.
212 * @data: interrupt data.
213 *
214 * this is common handler for output ring error interrupts.
215 */
216static irqreturn_t octep_ore_intr_handler(int irq, void *data)
217{
218	struct octep_device *oct = data;
219
220	return oct->hw_ops.ore_intr_handler(oct);
221}
222
223/**
224 * octep_vfire_intr_handler() - common handler for vf input ring error interrupts.
225 *
226 * @irq: Interrupt number.
227 * @data: interrupt data.
228 *
229 * this is common handler for vf input ring error interrupts.
230 */
231static irqreturn_t octep_vfire_intr_handler(int irq, void *data)
232{
233	struct octep_device *oct = data;
234
235	return oct->hw_ops.vfire_intr_handler(oct);
236}
237
238/**
239 * octep_vfore_intr_handler() - common handler for vf output ring error interrupts.
240 *
241 * @irq: Interrupt number.
242 * @data: interrupt data.
243 *
244 * this is common handler for vf output ring error interrupts.
245 */
246static irqreturn_t octep_vfore_intr_handler(int irq, void *data)
247{
248	struct octep_device *oct = data;
249
250	return oct->hw_ops.vfore_intr_handler(oct);
251}
252
253/**
254 * octep_dma_intr_handler() - common handler for dpi dma related interrupts.
255 *
256 * @irq: Interrupt number.
257 * @data: interrupt data.
258 *
259 * this is common handler for dpi dma related interrupts.
260 */
261static irqreturn_t octep_dma_intr_handler(int irq, void *data)
262{
263	struct octep_device *oct = data;
264
265	return oct->hw_ops.dma_intr_handler(oct);
266}
267
268/**
269 * octep_dma_vf_intr_handler() - common handler for dpi dma transaction error interrupts for VFs.
270 *
271 * @irq: Interrupt number.
272 * @data: interrupt data.
273 *
274 * this is common handler for dpi dma transaction error interrupts for VFs.
275 */
276static irqreturn_t octep_dma_vf_intr_handler(int irq, void *data)
277{
278	struct octep_device *oct = data;
279
280	return oct->hw_ops.dma_vf_intr_handler(oct);
281}
282
283/**
284 * octep_pp_vf_intr_handler() - common handler for pp transaction error interrupts for VFs.
285 *
286 * @irq: Interrupt number.
287 * @data: interrupt data.
288 *
289 * this is common handler for pp transaction error interrupts for VFs.
290 */
291static irqreturn_t octep_pp_vf_intr_handler(int irq, void *data)
292{
293	struct octep_device *oct = data;
294
295	return oct->hw_ops.pp_vf_intr_handler(oct);
296}
297
298/**
299 * octep_misc_intr_handler() - common handler for mac related interrupts.
300 *
301 * @irq: Interrupt number.
302 * @data: interrupt data.
303 *
304 * this is common handler for mac related interrupts.
305 */
306static irqreturn_t octep_misc_intr_handler(int irq, void *data)
307{
308	struct octep_device *oct = data;
309
310	return oct->hw_ops.misc_intr_handler(oct);
311}
312
313/**
314 * octep_rsvd_intr_handler() - common handler for reserved interrupts (future use).
315 *
316 * @irq: Interrupt number.
317 * @data: interrupt data.
318 *
319 * this is common handler for all reserved interrupts.
320 */
321static irqreturn_t octep_rsvd_intr_handler(int irq, void *data)
322{
323	struct octep_device *oct = data;
324
325	return oct->hw_ops.rsvd_intr_handler(oct);
326}
327
328/**
329 * octep_ioq_intr_handler() - handler for all Tx/Rx queue interrupts.
330 *
331 * @irq: Interrupt number.
332 * @data: interrupt data contains pointers to Tx/Rx queue private data
333 *         and correspong NAPI context.
334 *
335 * this is common handler for all non-queue (generic) interrupts.
336 */
337static irqreturn_t octep_ioq_intr_handler(int irq, void *data)
338{
339	struct octep_ioq_vector *ioq_vector = data;
340	struct octep_device *oct = ioq_vector->octep_dev;
341
342	return oct->hw_ops.ioq_intr_handler(ioq_vector);
343}
344
345/**
346 * octep_request_irqs() - Register interrupt handlers.
347 *
348 * @oct: Octeon device private data structure.
349 *
350 * Register handlers for all queue and non-queue interrupts.
351 *
352 * Return: 0, on successful registration of all interrupt handlers.
353 *         -1, on any error.
354 */
355static int octep_request_irqs(struct octep_device *oct)
356{
357	struct net_device *netdev = oct->netdev;
358	struct octep_ioq_vector *ioq_vector;
359	struct msix_entry *msix_entry;
360	char **non_ioq_msix_names;
361	int num_non_ioq_msix;
362	int ret, i, j;
363
364	num_non_ioq_msix = CFG_GET_NON_IOQ_MSIX(oct->conf);
365	non_ioq_msix_names = CFG_GET_NON_IOQ_MSIX_NAMES(oct->conf);
366
367	oct->non_ioq_irq_names = kcalloc(num_non_ioq_msix,
368					 OCTEP_MSIX_NAME_SIZE, GFP_KERNEL);
369	if (!oct->non_ioq_irq_names)
370		goto alloc_err;
371
372	/* First few MSI-X interrupts are non-queue interrupts */
373	for (i = 0; i < num_non_ioq_msix; i++) {
374		char *irq_name;
375
376		irq_name = &oct->non_ioq_irq_names[i * OCTEP_MSIX_NAME_SIZE];
377		msix_entry = &oct->msix_entries[i];
378
379		snprintf(irq_name, OCTEP_MSIX_NAME_SIZE,
380			 "%s-%s", netdev->name, non_ioq_msix_names[i]);
381		if (!strncmp(non_ioq_msix_names[i], "epf_mbox_rint", strlen("epf_mbox_rint"))) {
382			ret = request_irq(msix_entry->vector,
383					  octep_mbox_intr_handler, 0,
384					  irq_name, oct);
385		} else if (!strncmp(non_ioq_msix_names[i], "epf_oei_rint",
386			   strlen("epf_oei_rint"))) {
387			ret = request_irq(msix_entry->vector,
388					  octep_oei_intr_handler, 0,
389					  irq_name, oct);
390		} else if (!strncmp(non_ioq_msix_names[i], "epf_ire_rint",
391			   strlen("epf_ire_rint"))) {
392			ret = request_irq(msix_entry->vector,
393					  octep_ire_intr_handler, 0,
394					  irq_name, oct);
395		} else if (!strncmp(non_ioq_msix_names[i], "epf_ore_rint",
396			   strlen("epf_ore_rint"))) {
397			ret = request_irq(msix_entry->vector,
398					  octep_ore_intr_handler, 0,
399					  irq_name, oct);
400		} else if (!strncmp(non_ioq_msix_names[i], "epf_vfire_rint",
401			   strlen("epf_vfire_rint"))) {
402			ret = request_irq(msix_entry->vector,
403					  octep_vfire_intr_handler, 0,
404					  irq_name, oct);
405		} else if (!strncmp(non_ioq_msix_names[i], "epf_vfore_rint",
406			   strlen("epf_vfore_rint"))) {
407			ret = request_irq(msix_entry->vector,
408					  octep_vfore_intr_handler, 0,
409					  irq_name, oct);
410		} else if (!strncmp(non_ioq_msix_names[i], "epf_dma_rint",
411			   strlen("epf_dma_rint"))) {
412			ret = request_irq(msix_entry->vector,
413					  octep_dma_intr_handler, 0,
414					  irq_name, oct);
415		} else if (!strncmp(non_ioq_msix_names[i], "epf_dma_vf_rint",
416			   strlen("epf_dma_vf_rint"))) {
417			ret = request_irq(msix_entry->vector,
418					  octep_dma_vf_intr_handler, 0,
419					  irq_name, oct);
420		} else if (!strncmp(non_ioq_msix_names[i], "epf_pp_vf_rint",
421			   strlen("epf_pp_vf_rint"))) {
422			ret = request_irq(msix_entry->vector,
423					  octep_pp_vf_intr_handler, 0,
424					  irq_name, oct);
425		} else if (!strncmp(non_ioq_msix_names[i], "epf_misc_rint",
426			   strlen("epf_misc_rint"))) {
427			ret = request_irq(msix_entry->vector,
428					  octep_misc_intr_handler, 0,
429					  irq_name, oct);
430		} else {
431			ret = request_irq(msix_entry->vector,
432					  octep_rsvd_intr_handler, 0,
433					  irq_name, oct);
434		}
435
436		if (ret) {
437			netdev_err(netdev,
438				   "request_irq failed for %s; err=%d",
439				   irq_name, ret);
440			goto non_ioq_irq_err;
441		}
442	}
443
444	/* Request IRQs for Tx/Rx queues */
445	for (j = 0; j < oct->num_oqs; j++) {
446		ioq_vector = oct->ioq_vector[j];
447		msix_entry = &oct->msix_entries[j + num_non_ioq_msix];
448
449		snprintf(ioq_vector->name, sizeof(ioq_vector->name),
450			 "%s-q%d", netdev->name, j);
451		ret = request_irq(msix_entry->vector,
452				  octep_ioq_intr_handler, 0,
453				  ioq_vector->name, ioq_vector);
454		if (ret) {
455			netdev_err(netdev,
456				   "request_irq failed for Q-%d; err=%d",
457				   j, ret);
458			goto ioq_irq_err;
459		}
460
461		cpumask_set_cpu(j % num_online_cpus(),
462				&ioq_vector->affinity_mask);
463		irq_set_affinity_hint(msix_entry->vector,
464				      &ioq_vector->affinity_mask);
465	}
466
467	return 0;
468ioq_irq_err:
469	while (j) {
470		--j;
471		ioq_vector = oct->ioq_vector[j];
472		msix_entry = &oct->msix_entries[j + num_non_ioq_msix];
473
474		irq_set_affinity_hint(msix_entry->vector, NULL);
475		free_irq(msix_entry->vector, ioq_vector);
476	}
477non_ioq_irq_err:
478	while (i) {
479		--i;
480		free_irq(oct->msix_entries[i].vector, oct);
481	}
482	kfree(oct->non_ioq_irq_names);
483	oct->non_ioq_irq_names = NULL;
484alloc_err:
485	return -1;
486}
487
488/**
489 * octep_free_irqs() - free all registered interrupts.
490 *
491 * @oct: Octeon device private data structure.
492 *
493 * Free all queue and non-queue interrupts of the Octeon device.
494 */
495static void octep_free_irqs(struct octep_device *oct)
496{
497	int i;
498
499	/* First few MSI-X interrupts are non queue interrupts; free them */
500	for (i = 0; i < CFG_GET_NON_IOQ_MSIX(oct->conf); i++)
501		free_irq(oct->msix_entries[i].vector, oct);
502	kfree(oct->non_ioq_irq_names);
503
504	/* Free IRQs for Input/Output (Tx/Rx) queues */
505	for (i = CFG_GET_NON_IOQ_MSIX(oct->conf); i < oct->num_irqs; i++) {
506		irq_set_affinity_hint(oct->msix_entries[i].vector, NULL);
507		free_irq(oct->msix_entries[i].vector,
508			 oct->ioq_vector[i - CFG_GET_NON_IOQ_MSIX(oct->conf)]);
509	}
510	netdev_info(oct->netdev, "IRQs freed\n");
511}
512
513/**
514 * octep_setup_irqs() - setup interrupts for the Octeon device.
515 *
516 * @oct: Octeon device private data structure.
517 *
518 * Allocate data structures to hold per interrupt information, allocate/enable
519 * MSI-x interrupt and register interrupt handlers.
520 *
521 * Return: 0, on successful allocation and registration of all interrupts.
522 *         -1, on any error.
523 */
524static int octep_setup_irqs(struct octep_device *oct)
525{
526	if (octep_alloc_ioq_vectors(oct))
527		goto ioq_vector_err;
528
529	if (octep_enable_msix_range(oct))
530		goto enable_msix_err;
531
532	if (octep_request_irqs(oct))
533		goto request_irq_err;
534
535	return 0;
536
537request_irq_err:
538	octep_disable_msix(oct);
539enable_msix_err:
540	octep_free_ioq_vectors(oct);
541ioq_vector_err:
542	return -1;
543}
544
545/**
546 * octep_clean_irqs() - free all interrupts and its resources.
547 *
548 * @oct: Octeon device private data structure.
549 */
550static void octep_clean_irqs(struct octep_device *oct)
551{
552	octep_free_irqs(oct);
553	octep_disable_msix(oct);
554	octep_free_ioq_vectors(oct);
555}
556
557/**
558 * octep_enable_ioq_irq() - Enable MSI-x interrupt of a Tx/Rx queue.
559 *
560 * @iq: Octeon Tx queue data structure.
561 * @oq: Octeon Rx queue data structure.
562 */
563static void octep_enable_ioq_irq(struct octep_iq *iq, struct octep_oq *oq)
564{
565	u32 pkts_pend = oq->pkts_pending;
566
567	netdev_dbg(iq->netdev, "enabling intr for Q-%u\n", iq->q_no);
568	if (iq->pkts_processed) {
569		writel(iq->pkts_processed, iq->inst_cnt_reg);
570		iq->pkt_in_done -= iq->pkts_processed;
571		iq->pkts_processed = 0;
572	}
573	if (oq->last_pkt_count - pkts_pend) {
574		writel(oq->last_pkt_count - pkts_pend, oq->pkts_sent_reg);
575		oq->last_pkt_count = pkts_pend;
576	}
577
578	/* Flush the previous wrties before writing to RESEND bit */
579	wmb();
580	writeq(1UL << OCTEP_OQ_INTR_RESEND_BIT, oq->pkts_sent_reg);
581	writeq(1UL << OCTEP_IQ_INTR_RESEND_BIT, iq->inst_cnt_reg);
582}
583
584/**
585 * octep_napi_poll() - NAPI poll function for Tx/Rx.
586 *
587 * @napi: pointer to napi context.
588 * @budget: max number of packets to be processed in single invocation.
589 */
590static int octep_napi_poll(struct napi_struct *napi, int budget)
591{
592	struct octep_ioq_vector *ioq_vector =
593		container_of(napi, struct octep_ioq_vector, napi);
594	u32 tx_pending, rx_done;
595
596	tx_pending = octep_iq_process_completions(ioq_vector->iq, budget);
597	rx_done = octep_oq_process_rx(ioq_vector->oq, budget);
598
599	/* need more polling if tx completion processing is still pending or
600	 * processed at least 'budget' number of rx packets.
601	 */
602	if (tx_pending || rx_done >= budget)
603		return budget;
604
605	napi_complete(napi);
606	octep_enable_ioq_irq(ioq_vector->iq, ioq_vector->oq);
607	return rx_done;
608}
609
610/**
611 * octep_napi_add() - Add NAPI poll for all Tx/Rx queues.
612 *
613 * @oct: Octeon device private data structure.
614 */
615static void octep_napi_add(struct octep_device *oct)
616{
617	int i;
618
619	for (i = 0; i < oct->num_oqs; i++) {
620		netdev_dbg(oct->netdev, "Adding NAPI on Q-%d\n", i);
621		netif_napi_add(oct->netdev, &oct->ioq_vector[i]->napi,
622			       octep_napi_poll);
623		oct->oq[i]->napi = &oct->ioq_vector[i]->napi;
624	}
625}
626
627/**
628 * octep_napi_delete() - delete NAPI poll callback for all Tx/Rx queues.
629 *
630 * @oct: Octeon device private data structure.
631 */
632static void octep_napi_delete(struct octep_device *oct)
633{
634	int i;
635
636	for (i = 0; i < oct->num_oqs; i++) {
637		netdev_dbg(oct->netdev, "Deleting NAPI on Q-%d\n", i);
638		netif_napi_del(&oct->ioq_vector[i]->napi);
639		oct->oq[i]->napi = NULL;
640	}
641}
642
643/**
644 * octep_napi_enable() - enable NAPI for all Tx/Rx queues.
645 *
646 * @oct: Octeon device private data structure.
647 */
648static void octep_napi_enable(struct octep_device *oct)
649{
650	int i;
651
652	for (i = 0; i < oct->num_oqs; i++) {
653		netdev_dbg(oct->netdev, "Enabling NAPI on Q-%d\n", i);
654		napi_enable(&oct->ioq_vector[i]->napi);
655	}
656}
657
658/**
659 * octep_napi_disable() - disable NAPI for all Tx/Rx queues.
660 *
661 * @oct: Octeon device private data structure.
662 */
663static void octep_napi_disable(struct octep_device *oct)
664{
665	int i;
666
667	for (i = 0; i < oct->num_oqs; i++) {
668		netdev_dbg(oct->netdev, "Disabling NAPI on Q-%d\n", i);
669		napi_disable(&oct->ioq_vector[i]->napi);
670	}
671}
672
673static void octep_link_up(struct net_device *netdev)
674{
675	netif_carrier_on(netdev);
676	netif_tx_start_all_queues(netdev);
677}
678
679/**
680 * octep_open() - start the octeon network device.
681 *
682 * @netdev: pointer to kernel network device.
683 *
684 * setup Tx/Rx queues, interrupts and enable hardware operation of Tx/Rx queues
685 * and interrupts..
686 *
687 * Return: 0, on successfully setting up device and bring it up.
688 *         -1, on any error.
689 */
690static int octep_open(struct net_device *netdev)
691{
692	struct octep_device *oct = netdev_priv(netdev);
693	int err, ret;
694
695	netdev_info(netdev, "Starting netdev ...\n");
696	netif_carrier_off(netdev);
697
698	oct->hw_ops.reset_io_queues(oct);
699
700	if (octep_setup_iqs(oct))
701		goto setup_iq_err;
702	if (octep_setup_oqs(oct))
703		goto setup_oq_err;
704	if (octep_setup_irqs(oct))
705		goto setup_irq_err;
706
707	err = netif_set_real_num_tx_queues(netdev, oct->num_oqs);
708	if (err)
709		goto set_queues_err;
710	err = netif_set_real_num_rx_queues(netdev, oct->num_iqs);
711	if (err)
712		goto set_queues_err;
713
714	octep_napi_add(oct);
715	octep_napi_enable(oct);
716
717	oct->link_info.admin_up = 1;
718	octep_ctrl_net_set_rx_state(oct, OCTEP_CTRL_NET_INVALID_VFID, true,
719				    false);
720	octep_ctrl_net_set_link_status(oct, OCTEP_CTRL_NET_INVALID_VFID, true,
721				       false);
722	oct->poll_non_ioq_intr = false;
723
724	/* Enable the input and output queues for this Octeon device */
725	oct->hw_ops.enable_io_queues(oct);
726
727	/* Enable Octeon device interrupts */
728	oct->hw_ops.enable_interrupts(oct);
729
730	octep_oq_dbell_init(oct);
731
732	ret = octep_ctrl_net_get_link_status(oct, OCTEP_CTRL_NET_INVALID_VFID);
733	if (ret > 0)
734		octep_link_up(netdev);
735
736	return 0;
737
738set_queues_err:
739	octep_clean_irqs(oct);
740setup_irq_err:
741	octep_free_oqs(oct);
742setup_oq_err:
743	octep_free_iqs(oct);
744setup_iq_err:
745	return -1;
746}
747
748/**
749 * octep_stop() - stop the octeon network device.
750 *
751 * @netdev: pointer to kernel network device.
752 *
753 * stop the device Tx/Rx operations, bring down the link and
754 * free up all resources allocated for Tx/Rx queues and interrupts.
755 */
756static int octep_stop(struct net_device *netdev)
757{
758	struct octep_device *oct = netdev_priv(netdev);
759
760	netdev_info(netdev, "Stopping the device ...\n");
761
762	octep_ctrl_net_set_link_status(oct, OCTEP_CTRL_NET_INVALID_VFID, false,
763				       false);
764	octep_ctrl_net_set_rx_state(oct, OCTEP_CTRL_NET_INVALID_VFID, false,
765				    false);
766
767	/* Stop Tx from stack */
768	netif_tx_stop_all_queues(netdev);
769	netif_carrier_off(netdev);
770	netif_tx_disable(netdev);
771
772	oct->link_info.admin_up = 0;
773	oct->link_info.oper_up = 0;
774
775	oct->hw_ops.disable_interrupts(oct);
776	octep_napi_disable(oct);
777	octep_napi_delete(oct);
778
779	octep_clean_irqs(oct);
780	octep_clean_iqs(oct);
781
782	oct->hw_ops.disable_io_queues(oct);
783	oct->hw_ops.reset_io_queues(oct);
784	octep_free_oqs(oct);
785	octep_free_iqs(oct);
786
787	oct->poll_non_ioq_intr = true;
788	queue_delayed_work(octep_wq, &oct->intr_poll_task,
789			   msecs_to_jiffies(OCTEP_INTR_POLL_TIME_MSECS));
790
791	netdev_info(netdev, "Device stopped !!\n");
792	return 0;
793}
794
795/**
796 * octep_iq_full_check() - check if a Tx queue is full.
797 *
798 * @iq: Octeon Tx queue data structure.
799 *
800 * Return: 0, if the Tx queue is not full.
801 *         1, if the Tx queue is full.
802 */
803static inline int octep_iq_full_check(struct octep_iq *iq)
804{
805	if (likely((IQ_INSTR_SPACE(iq)) >
806		   OCTEP_WAKE_QUEUE_THRESHOLD))
807		return 0;
808
809	/* Stop the queue if unable to send */
810	netif_stop_subqueue(iq->netdev, iq->q_no);
811
812	/* Allow for pending updates in write index
813	 * from iq_process_completion in other cpus
814	 * to reflect, in case queue gets free
815	 * entries.
816	 */
817	smp_mb();
818
819	/* check again and restart the queue, in case NAPI has just freed
820	 * enough Tx ring entries.
821	 */
822	if (unlikely(IQ_INSTR_SPACE(iq) >
823		     OCTEP_WAKE_QUEUE_THRESHOLD)) {
824		netif_start_subqueue(iq->netdev, iq->q_no);
825		iq->stats.restart_cnt++;
826		return 0;
827	}
828
829	return 1;
830}
831
832/**
833 * octep_start_xmit() - Enqueue packet to Octoen hardware Tx Queue.
834 *
835 * @skb: packet skbuff pointer.
836 * @netdev: kernel network device.
837 *
838 * Return: NETDEV_TX_BUSY, if Tx Queue is full.
839 *         NETDEV_TX_OK, if successfully enqueued to hardware Tx queue.
840 */
841static netdev_tx_t octep_start_xmit(struct sk_buff *skb,
842				    struct net_device *netdev)
843{
844	struct octep_device *oct = netdev_priv(netdev);
845	netdev_features_t feat  = netdev->features;
846	struct octep_tx_sglist_desc *sglist;
847	struct octep_tx_buffer *tx_buffer;
848	struct octep_tx_desc_hw *hw_desc;
849	struct skb_shared_info *shinfo;
850	struct octep_instr_hdr *ih;
851	struct octep_iq *iq;
852	skb_frag_t *frag;
853	u16 nr_frags, si;
854	int xmit_more;
855	u16 q_no, wi;
856
857	if (skb_put_padto(skb, ETH_ZLEN))
858		return NETDEV_TX_OK;
859
860	q_no = skb_get_queue_mapping(skb);
861	if (q_no >= oct->num_iqs) {
862		netdev_err(netdev, "Invalid Tx skb->queue_mapping=%d\n", q_no);
863		q_no = q_no % oct->num_iqs;
864	}
865
866	iq = oct->iq[q_no];
867
868	shinfo = skb_shinfo(skb);
869	nr_frags = shinfo->nr_frags;
870
871	wi = iq->host_write_index;
872	hw_desc = &iq->desc_ring[wi];
873	hw_desc->ih64 = 0;
874
875	tx_buffer = iq->buff_info + wi;
876	tx_buffer->skb = skb;
877
878	ih = &hw_desc->ih;
879	ih->pkind = oct->conf->fw_info.pkind;
880	ih->fsz = oct->conf->fw_info.fsz;
881	ih->tlen = skb->len + ih->fsz;
882
883	if (!nr_frags) {
884		tx_buffer->gather = 0;
885		tx_buffer->dma = dma_map_single(iq->dev, skb->data,
886						skb->len, DMA_TO_DEVICE);
887		if (dma_mapping_error(iq->dev, tx_buffer->dma))
888			goto dma_map_err;
889		hw_desc->dptr = tx_buffer->dma;
890	} else {
891		/* Scatter/Gather */
892		dma_addr_t dma;
893		u16 len;
894
895		sglist = tx_buffer->sglist;
896
897		ih->gsz = nr_frags + 1;
898		ih->gather = 1;
899		tx_buffer->gather = 1;
900
901		len = skb_headlen(skb);
902		dma = dma_map_single(iq->dev, skb->data, len, DMA_TO_DEVICE);
903		if (dma_mapping_error(iq->dev, dma))
904			goto dma_map_err;
905
906		memset(sglist, 0, OCTEP_SGLIST_SIZE_PER_PKT);
907		sglist[0].len[3] = len;
908		sglist[0].dma_ptr[0] = dma;
909
910		si = 1; /* entry 0 is main skb, mapped above */
911		frag = &shinfo->frags[0];
912		while (nr_frags--) {
913			len = skb_frag_size(frag);
914			dma = skb_frag_dma_map(iq->dev, frag, 0,
915					       len, DMA_TO_DEVICE);
916			if (dma_mapping_error(iq->dev, dma))
917				goto dma_map_sg_err;
918
919			sglist[si >> 2].len[3 - (si & 3)] = len;
920			sglist[si >> 2].dma_ptr[si & 3] = dma;
921
922			frag++;
923			si++;
924		}
925		hw_desc->dptr = tx_buffer->sglist_dma;
926	}
927
928	if (oct->conf->fw_info.tx_ol_flags) {
929		if ((feat & (NETIF_F_TSO)) && (skb_is_gso(skb))) {
930			hw_desc->txm.ol_flags = OCTEP_TX_OFFLOAD_CKSUM;
931			hw_desc->txm.ol_flags |= OCTEP_TX_OFFLOAD_TSO;
932			hw_desc->txm.gso_size =  skb_shinfo(skb)->gso_size;
933			hw_desc->txm.gso_segs =  skb_shinfo(skb)->gso_segs;
934		} else if (feat & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
935			hw_desc->txm.ol_flags = OCTEP_TX_OFFLOAD_CKSUM;
936		}
937		/* due to ESR txm will be swapped by hw */
938		hw_desc->txm64[0] = (__force u64)cpu_to_be64(hw_desc->txm64[0]);
939	}
940
941	xmit_more = netdev_xmit_more();
942
943	__netdev_tx_sent_queue(iq->netdev_q, skb->len, xmit_more);
944
945	skb_tx_timestamp(skb);
946	iq->fill_cnt++;
947	wi++;
948	iq->host_write_index = wi & iq->ring_size_mask;
949
950	/* octep_iq_full_check stops the queue and returns
951	 * true if so, in case the queue has become full
952	 * by inserting current packet. If so, we can
953	 * go ahead and ring doorbell.
954	 */
955	if (!octep_iq_full_check(iq) && xmit_more &&
956	    iq->fill_cnt < iq->fill_threshold)
957		return NETDEV_TX_OK;
958
959	/* Flush the hw descriptor before writing to doorbell */
960	wmb();
961	/* Ring Doorbell to notify the NIC of new packets */
962	writel(iq->fill_cnt, iq->doorbell_reg);
963	iq->stats.instr_posted += iq->fill_cnt;
964	iq->fill_cnt = 0;
965	return NETDEV_TX_OK;
966
967dma_map_sg_err:
968	if (si > 0) {
969		dma_unmap_single(iq->dev, sglist[0].dma_ptr[0],
970				 sglist[0].len[3], DMA_TO_DEVICE);
971		sglist[0].len[3] = 0;
972	}
973	while (si > 1) {
974		dma_unmap_page(iq->dev, sglist[si >> 2].dma_ptr[si & 3],
975			       sglist[si >> 2].len[3 - (si & 3)], DMA_TO_DEVICE);
976		sglist[si >> 2].len[3 - (si & 3)] = 0;
977		si--;
978	}
979	tx_buffer->gather = 0;
980dma_map_err:
981	dev_kfree_skb_any(skb);
982	return NETDEV_TX_OK;
983}
984
985/**
986 * octep_get_stats64() - Get Octeon network device statistics.
987 *
988 * @netdev: kernel network device.
989 * @stats: pointer to stats structure to be filled in.
990 */
991static void octep_get_stats64(struct net_device *netdev,
992			      struct rtnl_link_stats64 *stats)
993{
994	u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
995	struct octep_device *oct = netdev_priv(netdev);
996	int q;
997
998	if (netif_running(netdev))
999		octep_ctrl_net_get_if_stats(oct,
1000					    OCTEP_CTRL_NET_INVALID_VFID,
1001					    &oct->iface_rx_stats,
1002					    &oct->iface_tx_stats);
1003
1004	tx_packets = 0;
1005	tx_bytes = 0;
1006	rx_packets = 0;
1007	rx_bytes = 0;
1008	for (q = 0; q < oct->num_oqs; q++) {
1009		struct octep_iq *iq = oct->iq[q];
1010		struct octep_oq *oq = oct->oq[q];
1011
1012		tx_packets += iq->stats.instr_completed;
1013		tx_bytes += iq->stats.bytes_sent;
1014		rx_packets += oq->stats.packets;
1015		rx_bytes += oq->stats.bytes;
1016	}
1017	stats->tx_packets = tx_packets;
1018	stats->tx_bytes = tx_bytes;
1019	stats->rx_packets = rx_packets;
1020	stats->rx_bytes = rx_bytes;
1021	stats->multicast = oct->iface_rx_stats.mcast_pkts;
1022	stats->rx_errors = oct->iface_rx_stats.err_pkts;
1023	stats->collisions = oct->iface_tx_stats.xscol;
1024	stats->tx_fifo_errors = oct->iface_tx_stats.undflw;
1025}
1026
1027/**
1028 * octep_tx_timeout_task - work queue task to Handle Tx queue timeout.
1029 *
1030 * @work: pointer to Tx queue timeout work_struct
1031 *
1032 * Stop and start the device so that it frees up all queue resources
1033 * and restarts the queues, that potentially clears a Tx queue timeout
1034 * condition.
1035 **/
1036static void octep_tx_timeout_task(struct work_struct *work)
1037{
1038	struct octep_device *oct = container_of(work, struct octep_device,
1039						tx_timeout_task);
1040	struct net_device *netdev = oct->netdev;
1041
1042	rtnl_lock();
1043	if (netif_running(netdev)) {
1044		octep_stop(netdev);
1045		octep_open(netdev);
1046	}
1047	rtnl_unlock();
1048}
1049
1050/**
1051 * octep_tx_timeout() - Handle Tx Queue timeout.
1052 *
1053 * @netdev: pointer to kernel network device.
1054 * @txqueue: Timed out Tx queue number.
1055 *
1056 * Schedule a work to handle Tx queue timeout.
1057 */
1058static void octep_tx_timeout(struct net_device *netdev, unsigned int txqueue)
1059{
1060	struct octep_device *oct = netdev_priv(netdev);
1061
1062	queue_work(octep_wq, &oct->tx_timeout_task);
1063}
1064
1065static int octep_set_mac(struct net_device *netdev, void *p)
1066{
1067	struct octep_device *oct = netdev_priv(netdev);
1068	struct sockaddr *addr = (struct sockaddr *)p;
1069	int err;
1070
1071	if (!is_valid_ether_addr(addr->sa_data))
1072		return -EADDRNOTAVAIL;
1073
1074	err = octep_ctrl_net_set_mac_addr(oct, OCTEP_CTRL_NET_INVALID_VFID,
1075					  addr->sa_data, true);
1076	if (err)
1077		return err;
1078
1079	memcpy(oct->mac_addr, addr->sa_data, ETH_ALEN);
1080	eth_hw_addr_set(netdev, addr->sa_data);
1081
1082	return 0;
1083}
1084
1085static int octep_change_mtu(struct net_device *netdev, int new_mtu)
1086{
1087	struct octep_device *oct = netdev_priv(netdev);
1088	struct octep_iface_link_info *link_info;
1089	int err = 0;
1090
1091	link_info = &oct->link_info;
1092	if (link_info->mtu == new_mtu)
1093		return 0;
1094
1095	err = octep_ctrl_net_set_mtu(oct, OCTEP_CTRL_NET_INVALID_VFID, new_mtu,
1096				     true);
1097	if (!err) {
1098		oct->link_info.mtu = new_mtu;
1099		netdev->mtu = new_mtu;
1100	}
1101
1102	return err;
1103}
1104
1105static int octep_set_features(struct net_device *dev, netdev_features_t features)
1106{
1107	struct octep_ctrl_net_offloads offloads = { 0 };
1108	struct octep_device *oct = netdev_priv(dev);
1109	int err;
1110
1111	/* We only support features received from firmware */
1112	if ((features & dev->hw_features) != features)
1113		return -EINVAL;
1114
1115	if (features & NETIF_F_TSO)
1116		offloads.tx_offloads |= OCTEP_TX_OFFLOAD_TSO;
1117
1118	if (features & NETIF_F_TSO6)
1119		offloads.tx_offloads |= OCTEP_TX_OFFLOAD_TSO;
1120
1121	if (features & NETIF_F_IP_CSUM)
1122		offloads.tx_offloads |= OCTEP_TX_OFFLOAD_CKSUM;
1123
1124	if (features & NETIF_F_IPV6_CSUM)
1125		offloads.tx_offloads |= OCTEP_TX_OFFLOAD_CKSUM;
1126
1127	if (features & NETIF_F_RXCSUM)
1128		offloads.rx_offloads |= OCTEP_RX_OFFLOAD_CKSUM;
1129
1130	err = octep_ctrl_net_set_offloads(oct,
1131					  OCTEP_CTRL_NET_INVALID_VFID,
1132					  &offloads,
1133					  true);
1134	if (!err)
1135		dev->features = features;
1136
1137	return err;
1138}
1139
1140static const struct net_device_ops octep_netdev_ops = {
1141	.ndo_open                = octep_open,
1142	.ndo_stop                = octep_stop,
1143	.ndo_start_xmit          = octep_start_xmit,
1144	.ndo_get_stats64         = octep_get_stats64,
1145	.ndo_tx_timeout          = octep_tx_timeout,
1146	.ndo_set_mac_address     = octep_set_mac,
1147	.ndo_change_mtu          = octep_change_mtu,
1148	.ndo_set_features        = octep_set_features,
1149};
1150
1151/**
1152 * octep_intr_poll_task - work queue task to process non-ioq interrupts.
1153 *
1154 * @work: pointer to mbox work_struct
1155 *
1156 * Process non-ioq interrupts to handle control mailbox, pfvf mailbox.
1157 **/
1158static void octep_intr_poll_task(struct work_struct *work)
1159{
1160	struct octep_device *oct = container_of(work, struct octep_device,
1161						intr_poll_task.work);
1162
1163	if (!oct->poll_non_ioq_intr) {
1164		dev_info(&oct->pdev->dev, "Interrupt poll task stopped.\n");
1165		return;
1166	}
1167
1168	oct->hw_ops.poll_non_ioq_interrupts(oct);
1169	queue_delayed_work(octep_wq, &oct->intr_poll_task,
1170			   msecs_to_jiffies(OCTEP_INTR_POLL_TIME_MSECS));
1171}
1172
1173/**
1174 * octep_hb_timeout_task - work queue task to check firmware heartbeat.
1175 *
1176 * @work: pointer to hb work_struct
1177 *
1178 * Check for heartbeat miss count. Uninitialize oct device if miss count
1179 * exceeds configured max heartbeat miss count.
1180 *
1181 **/
1182static void octep_hb_timeout_task(struct work_struct *work)
1183{
1184	struct octep_device *oct = container_of(work, struct octep_device,
1185						hb_task.work);
1186
1187	int miss_cnt;
1188
1189	miss_cnt = atomic_inc_return(&oct->hb_miss_cnt);
1190	if (miss_cnt < oct->conf->fw_info.hb_miss_count) {
1191		queue_delayed_work(octep_wq, &oct->hb_task,
1192				   msecs_to_jiffies(oct->conf->fw_info.hb_interval));
1193		return;
1194	}
1195
1196	dev_err(&oct->pdev->dev, "Missed %u heartbeats. Uninitializing\n",
1197		miss_cnt);
1198	rtnl_lock();
1199	if (netif_running(oct->netdev))
1200		octep_stop(oct->netdev);
1201	rtnl_unlock();
1202}
1203
1204/**
1205 * octep_ctrl_mbox_task - work queue task to handle ctrl mbox messages.
1206 *
1207 * @work: pointer to ctrl mbox work_struct
1208 *
1209 * Poll ctrl mbox message queue and handle control messages from firmware.
1210 **/
1211static void octep_ctrl_mbox_task(struct work_struct *work)
1212{
1213	struct octep_device *oct = container_of(work, struct octep_device,
1214						ctrl_mbox_task);
1215
1216	octep_ctrl_net_recv_fw_messages(oct);
1217}
1218
1219static const char *octep_devid_to_str(struct octep_device *oct)
1220{
1221	switch (oct->chip_id) {
1222	case OCTEP_PCI_DEVICE_ID_CN98_PF:
1223		return "CN98XX";
1224	case OCTEP_PCI_DEVICE_ID_CN93_PF:
1225		return "CN93XX";
1226	case OCTEP_PCI_DEVICE_ID_CNF95N_PF:
1227		return "CNF95N";
1228	case OCTEP_PCI_DEVICE_ID_CN10KA_PF:
1229		return "CN10KA";
1230	case OCTEP_PCI_DEVICE_ID_CNF10KA_PF:
1231		return "CNF10KA";
1232	case OCTEP_PCI_DEVICE_ID_CNF10KB_PF:
1233		return "CNF10KB";
1234	case OCTEP_PCI_DEVICE_ID_CN10KB_PF:
1235		return "CN10KB";
1236	default:
1237		return "Unsupported";
1238	}
1239}
1240
1241/**
1242 * octep_device_setup() - Setup Octeon Device.
1243 *
1244 * @oct: Octeon device private data structure.
1245 *
1246 * Setup Octeon device hardware operations, configuration, etc ...
1247 */
1248int octep_device_setup(struct octep_device *oct)
1249{
1250	struct pci_dev *pdev = oct->pdev;
1251	int i, ret;
1252
1253	/* allocate memory for oct->conf */
1254	oct->conf = kzalloc(sizeof(*oct->conf), GFP_KERNEL);
1255	if (!oct->conf)
1256		return -ENOMEM;
1257
1258	/* Map BAR regions */
1259	for (i = 0; i < OCTEP_MMIO_REGIONS; i++) {
1260		oct->mmio[i].hw_addr =
1261			ioremap(pci_resource_start(oct->pdev, i * 2),
1262				pci_resource_len(oct->pdev, i * 2));
1263		if (!oct->mmio[i].hw_addr)
1264			goto unmap_prev;
1265
1266		oct->mmio[i].mapped = 1;
1267	}
1268
1269	oct->chip_id = pdev->device;
1270	oct->rev_id = pdev->revision;
1271	dev_info(&pdev->dev, "chip_id = 0x%x\n", pdev->device);
1272
1273	switch (oct->chip_id) {
1274	case OCTEP_PCI_DEVICE_ID_CN98_PF:
1275	case OCTEP_PCI_DEVICE_ID_CN93_PF:
1276	case OCTEP_PCI_DEVICE_ID_CNF95N_PF:
1277		dev_info(&pdev->dev, "Setting up OCTEON %s PF PASS%d.%d\n",
1278			 octep_devid_to_str(oct), OCTEP_MAJOR_REV(oct),
1279			 OCTEP_MINOR_REV(oct));
1280		octep_device_setup_cn93_pf(oct);
1281		break;
1282	case OCTEP_PCI_DEVICE_ID_CNF10KA_PF:
1283	case OCTEP_PCI_DEVICE_ID_CN10KA_PF:
1284	case OCTEP_PCI_DEVICE_ID_CNF10KB_PF:
1285	case OCTEP_PCI_DEVICE_ID_CN10KB_PF:
1286		dev_info(&pdev->dev, "Setting up OCTEON %s PF PASS%d.%d\n",
1287			 octep_devid_to_str(oct), OCTEP_MAJOR_REV(oct), OCTEP_MINOR_REV(oct));
1288		octep_device_setup_cnxk_pf(oct);
1289		break;
1290	default:
1291		dev_err(&pdev->dev,
1292			"%s: unsupported device\n", __func__);
1293		goto unsupported_dev;
1294	}
1295
1296
1297	ret = octep_ctrl_net_init(oct);
1298	if (ret)
1299		return ret;
1300
1301	INIT_WORK(&oct->tx_timeout_task, octep_tx_timeout_task);
1302	INIT_WORK(&oct->ctrl_mbox_task, octep_ctrl_mbox_task);
1303	INIT_DELAYED_WORK(&oct->intr_poll_task, octep_intr_poll_task);
1304	oct->poll_non_ioq_intr = true;
1305	queue_delayed_work(octep_wq, &oct->intr_poll_task,
1306			   msecs_to_jiffies(OCTEP_INTR_POLL_TIME_MSECS));
1307
1308	atomic_set(&oct->hb_miss_cnt, 0);
1309	INIT_DELAYED_WORK(&oct->hb_task, octep_hb_timeout_task);
1310
1311	return 0;
1312
1313unsupported_dev:
1314	i = OCTEP_MMIO_REGIONS;
1315unmap_prev:
1316	while (i--)
1317		iounmap(oct->mmio[i].hw_addr);
1318
1319	kfree(oct->conf);
1320	return -1;
1321}
1322
1323/**
1324 * octep_device_cleanup() - Cleanup Octeon Device.
1325 *
1326 * @oct: Octeon device private data structure.
1327 *
1328 * Cleanup Octeon device allocated resources.
1329 */
1330static void octep_device_cleanup(struct octep_device *oct)
1331{
1332	int i;
1333
1334	oct->poll_non_ioq_intr = false;
1335	cancel_delayed_work_sync(&oct->intr_poll_task);
1336	cancel_work_sync(&oct->ctrl_mbox_task);
1337
1338	dev_info(&oct->pdev->dev, "Cleaning up Octeon Device ...\n");
1339
1340	for (i = 0; i < OCTEP_MAX_VF; i++) {
1341		vfree(oct->mbox[i]);
1342		oct->mbox[i] = NULL;
1343	}
1344
1345	octep_delete_pfvf_mbox(oct);
1346	octep_ctrl_net_uninit(oct);
1347	cancel_delayed_work_sync(&oct->hb_task);
1348
1349	oct->hw_ops.soft_reset(oct);
1350	for (i = 0; i < OCTEP_MMIO_REGIONS; i++) {
1351		if (oct->mmio[i].mapped)
1352			iounmap(oct->mmio[i].hw_addr);
1353	}
1354
1355	kfree(oct->conf);
1356	oct->conf = NULL;
1357}
1358
1359static bool get_fw_ready_status(struct pci_dev *pdev)
1360{
1361	u32 pos = 0;
1362	u16 vsec_id;
1363	u8 status;
1364
1365	while ((pos = pci_find_next_ext_capability(pdev, pos,
1366						   PCI_EXT_CAP_ID_VNDR))) {
1367		pci_read_config_word(pdev, pos + 4, &vsec_id);
1368#define FW_STATUS_VSEC_ID  0xA3
1369		if (vsec_id != FW_STATUS_VSEC_ID)
1370			continue;
1371
1372		pci_read_config_byte(pdev, (pos + 8), &status);
1373		dev_info(&pdev->dev, "Firmware ready status = %u\n", status);
1374#define FW_STATUS_READY 1ULL
1375		return status == FW_STATUS_READY;
1376	}
1377	return false;
1378}
1379
1380/**
1381 * octep_probe() - Octeon PCI device probe handler.
1382 *
1383 * @pdev: PCI device structure.
1384 * @ent: entry in Octeon PCI device ID table.
1385 *
1386 * Initializes and enables the Octeon PCI device for network operations.
1387 * Initializes Octeon private data structure and registers a network device.
1388 */
1389static int octep_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1390{
1391	struct octep_device *octep_dev = NULL;
1392	struct net_device *netdev;
1393	int max_rx_pktlen;
1394	int err;
1395
1396	err = pci_enable_device(pdev);
1397	if (err) {
1398		dev_err(&pdev->dev, "Failed to enable PCI device\n");
1399		return  err;
1400	}
1401
1402	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
1403	if (err) {
1404		dev_err(&pdev->dev, "Failed to set DMA mask !!\n");
1405		goto err_dma_mask;
1406	}
1407
1408	err = pci_request_mem_regions(pdev, OCTEP_DRV_NAME);
1409	if (err) {
1410		dev_err(&pdev->dev, "Failed to map PCI memory regions\n");
1411		goto err_pci_regions;
1412	}
1413
1414	pci_set_master(pdev);
1415
1416	if (!get_fw_ready_status(pdev)) {
1417		dev_notice(&pdev->dev, "Firmware not ready; defer probe.\n");
1418		err = -EPROBE_DEFER;
1419		goto err_alloc_netdev;
1420	}
1421
1422	netdev = alloc_etherdev_mq(sizeof(struct octep_device),
1423				   OCTEP_MAX_QUEUES);
1424	if (!netdev) {
1425		dev_err(&pdev->dev, "Failed to allocate netdev\n");
1426		err = -ENOMEM;
1427		goto err_alloc_netdev;
1428	}
1429	SET_NETDEV_DEV(netdev, &pdev->dev);
1430
1431	octep_dev = netdev_priv(netdev);
1432	octep_dev->netdev = netdev;
1433	octep_dev->pdev = pdev;
1434	octep_dev->dev = &pdev->dev;
1435	pci_set_drvdata(pdev, octep_dev);
1436
1437	err = octep_device_setup(octep_dev);
1438	if (err) {
1439		dev_err(&pdev->dev, "Device setup failed\n");
1440		goto err_octep_config;
1441	}
1442
1443	err = octep_setup_pfvf_mbox(octep_dev);
1444	if (err) {
1445		dev_err(&pdev->dev, "PF-VF mailbox setup failed\n");
1446		goto register_dev_err;
1447	}
1448
1449	err = octep_ctrl_net_get_info(octep_dev, OCTEP_CTRL_NET_INVALID_VFID,
1450				      &octep_dev->conf->fw_info);
1451	if (err) {
1452		dev_err(&pdev->dev, "Failed to get firmware info\n");
1453		goto register_dev_err;
1454	}
1455	dev_info(&octep_dev->pdev->dev, "Heartbeat interval %u msecs Heartbeat miss count %u\n",
1456		 octep_dev->conf->fw_info.hb_interval,
1457		 octep_dev->conf->fw_info.hb_miss_count);
1458	queue_delayed_work(octep_wq, &octep_dev->hb_task,
1459			   msecs_to_jiffies(octep_dev->conf->fw_info.hb_interval));
1460
1461	netdev->netdev_ops = &octep_netdev_ops;
1462	octep_set_ethtool_ops(netdev);
1463	netif_carrier_off(netdev);
1464
1465	netdev->hw_features = NETIF_F_SG;
1466	if (OCTEP_TX_IP_CSUM(octep_dev->conf->fw_info.tx_ol_flags))
1467		netdev->hw_features |= (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
1468
1469	if (OCTEP_RX_IP_CSUM(octep_dev->conf->fw_info.rx_ol_flags))
1470		netdev->hw_features |= NETIF_F_RXCSUM;
1471
1472	max_rx_pktlen = octep_ctrl_net_get_mtu(octep_dev, OCTEP_CTRL_NET_INVALID_VFID);
1473	if (max_rx_pktlen < 0) {
1474		dev_err(&octep_dev->pdev->dev,
1475			"Failed to get max receive packet size; err = %d\n", max_rx_pktlen);
1476		err = max_rx_pktlen;
1477		goto register_dev_err;
1478	}
1479	netdev->min_mtu = OCTEP_MIN_MTU;
1480	netdev->max_mtu = max_rx_pktlen - (ETH_HLEN + ETH_FCS_LEN);
1481	netdev->mtu = OCTEP_DEFAULT_MTU;
1482
1483	if (OCTEP_TX_TSO(octep_dev->conf->fw_info.tx_ol_flags)) {
1484		netdev->hw_features |= NETIF_F_TSO;
1485		netif_set_tso_max_size(netdev, netdev->max_mtu);
1486	}
1487
1488	netdev->features |= netdev->hw_features;
1489	err = octep_ctrl_net_get_mac_addr(octep_dev, OCTEP_CTRL_NET_INVALID_VFID,
1490					  octep_dev->mac_addr);
1491	if (err) {
1492		dev_err(&pdev->dev, "Failed to get mac address\n");
1493		goto register_dev_err;
1494	}
1495	eth_hw_addr_set(netdev, octep_dev->mac_addr);
1496
1497	err = register_netdev(netdev);
1498	if (err) {
1499		dev_err(&pdev->dev, "Failed to register netdev\n");
1500		goto register_dev_err;
1501	}
1502	dev_info(&pdev->dev, "Device probe successful\n");
1503	return 0;
1504
1505register_dev_err:
1506	octep_device_cleanup(octep_dev);
1507err_octep_config:
1508	free_netdev(netdev);
1509err_alloc_netdev:
1510	pci_release_mem_regions(pdev);
1511err_pci_regions:
1512err_dma_mask:
1513	pci_disable_device(pdev);
1514	return err;
1515}
1516
1517static int octep_sriov_disable(struct octep_device *oct)
1518{
1519	struct pci_dev *pdev = oct->pdev;
1520
1521	if (pci_vfs_assigned(oct->pdev)) {
1522		dev_warn(&pdev->dev, "Can't disable SRIOV while VFs are assigned\n");
1523		return -EPERM;
1524	}
1525
1526	pci_disable_sriov(pdev);
1527	CFG_GET_ACTIVE_VFS(oct->conf) = 0;
1528
1529	return 0;
1530}
1531
1532/**
1533 * octep_remove() - Remove Octeon PCI device from driver control.
1534 *
1535 * @pdev: PCI device structure of the Octeon device.
1536 *
1537 * Cleanup all resources allocated for the Octeon device.
1538 * Unregister from network device and disable the PCI device.
1539 */
1540static void octep_remove(struct pci_dev *pdev)
1541{
1542	struct octep_device *oct = pci_get_drvdata(pdev);
1543	struct net_device *netdev;
1544
1545	if (!oct)
1546		return;
1547
1548	netdev = oct->netdev;
1549	octep_sriov_disable(oct);
1550	if (netdev->reg_state == NETREG_REGISTERED)
1551		unregister_netdev(netdev);
1552
1553	cancel_work_sync(&oct->tx_timeout_task);
1554	octep_device_cleanup(oct);
1555	pci_release_mem_regions(pdev);
1556	free_netdev(netdev);
1557	pci_disable_device(pdev);
1558}
1559
1560static int octep_sriov_enable(struct octep_device *oct, int num_vfs)
1561{
1562	struct pci_dev *pdev = oct->pdev;
1563	int err;
1564
1565	CFG_GET_ACTIVE_VFS(oct->conf) = num_vfs;
1566	err = pci_enable_sriov(pdev, num_vfs);
1567	if (err) {
1568		dev_warn(&pdev->dev, "Failed to enable SRIOV err=%d\n", err);
1569		CFG_GET_ACTIVE_VFS(oct->conf) = 0;
1570		return err;
1571	}
1572
1573	return num_vfs;
1574}
1575
1576static int octep_sriov_configure(struct pci_dev *pdev, int num_vfs)
1577{
1578	struct octep_device *oct = pci_get_drvdata(pdev);
1579	int max_nvfs;
1580
1581	if (num_vfs == 0)
1582		return octep_sriov_disable(oct);
1583
1584	max_nvfs = CFG_GET_MAX_VFS(oct->conf);
1585
1586	if (num_vfs > max_nvfs) {
1587		dev_err(&pdev->dev, "Invalid VF count Max supported VFs = %d\n",
1588			max_nvfs);
1589		return -EINVAL;
1590	}
1591
1592	return octep_sriov_enable(oct, num_vfs);
1593}
1594
1595static struct pci_driver octep_driver = {
1596	.name = OCTEP_DRV_NAME,
1597	.id_table = octep_pci_id_tbl,
1598	.probe = octep_probe,
1599	.remove = octep_remove,
1600	.sriov_configure = octep_sriov_configure,
1601};
1602
1603/**
1604 * octep_init_module() - Module initialiation.
1605 *
1606 * create common resource for the driver and register PCI driver.
1607 */
1608static int __init octep_init_module(void)
1609{
1610	int ret;
1611
1612	pr_info("%s: Loading %s ...\n", OCTEP_DRV_NAME, OCTEP_DRV_STRING);
1613
1614	/* work queue for all deferred tasks */
1615	octep_wq = create_singlethread_workqueue(OCTEP_DRV_NAME);
1616	if (!octep_wq) {
1617		pr_err("%s: Failed to create common workqueue\n",
1618		       OCTEP_DRV_NAME);
1619		return -ENOMEM;
1620	}
1621
1622	ret = pci_register_driver(&octep_driver);
1623	if (ret < 0) {
1624		pr_err("%s: Failed to register PCI driver; err=%d\n",
1625		       OCTEP_DRV_NAME, ret);
1626		destroy_workqueue(octep_wq);
1627		return ret;
1628	}
1629
1630	pr_info("%s: Loaded successfully !\n", OCTEP_DRV_NAME);
1631
1632	return ret;
1633}
1634
1635/**
1636 * octep_exit_module() - Module exit routine.
1637 *
1638 * unregister the driver with PCI subsystem and cleanup common resources.
1639 */
1640static void __exit octep_exit_module(void)
1641{
1642	pr_info("%s: Unloading ...\n", OCTEP_DRV_NAME);
1643
1644	pci_unregister_driver(&octep_driver);
1645	destroy_workqueue(octep_wq);
1646
1647	pr_info("%s: Unloading complete\n", OCTEP_DRV_NAME);
1648}
1649
1650module_init(octep_init_module);
1651module_exit(octep_exit_module);
1652