1 /*
2  * Copyright (C) 2016 Cavium, Inc.
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of version 2 of the GNU General Public License
6  * as published by the Free Software Foundation.
7  */
8 
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 
12 #include "cptvf.h"
13 
14 #define DRV_NAME	"thunder-cptvf"
15 #define DRV_VERSION	"1.0"
16 
17 struct cptvf_wqe {
18 	struct tasklet_struct twork;
19 	void *cptvf;
20 	u32 qno;
21 };
22 
23 struct cptvf_wqe_info {
24 	struct cptvf_wqe vq_wqe[CPT_NUM_QS_PER_VF];
25 };
26 
27 static void vq_work_handler(unsigned long data)
28 {
29 	struct cptvf_wqe_info *cwqe_info = (struct cptvf_wqe_info *)data;
30 	struct cptvf_wqe *cwqe = &cwqe_info->vq_wqe[0];
31 
32 	vq_post_process(cwqe->cptvf, cwqe->qno);
33 }
34 
35 static int init_worker_threads(struct cpt_vf *cptvf)
36 {
37 	struct pci_dev *pdev = cptvf->pdev;
38 	struct cptvf_wqe_info *cwqe_info;
39 	int i;
40 
41 	cwqe_info = kzalloc(sizeof(*cwqe_info), GFP_KERNEL);
42 	if (!cwqe_info)
43 		return -ENOMEM;
44 
45 	if (cptvf->nr_queues) {
46 		dev_info(&pdev->dev, "Creating VQ worker threads (%d)\n",
47 			 cptvf->nr_queues);
48 	}
49 
50 	for (i = 0; i < cptvf->nr_queues; i++) {
51 		tasklet_init(&cwqe_info->vq_wqe[i].twork, vq_work_handler,
52 			     (u64)cwqe_info);
53 		cwqe_info->vq_wqe[i].qno = i;
54 		cwqe_info->vq_wqe[i].cptvf = cptvf;
55 	}
56 
57 	cptvf->wqe_info = cwqe_info;
58 
59 	return 0;
60 }
61 
62 static void cleanup_worker_threads(struct cpt_vf *cptvf)
63 {
64 	struct cptvf_wqe_info *cwqe_info;
65 	struct pci_dev *pdev = cptvf->pdev;
66 	int i;
67 
68 	cwqe_info = (struct cptvf_wqe_info *)cptvf->wqe_info;
69 	if (!cwqe_info)
70 		return;
71 
72 	if (cptvf->nr_queues) {
73 		dev_info(&pdev->dev, "Cleaning VQ worker threads (%u)\n",
74 			 cptvf->nr_queues);
75 	}
76 
77 	for (i = 0; i < cptvf->nr_queues; i++)
78 		tasklet_kill(&cwqe_info->vq_wqe[i].twork);
79 
80 	kzfree(cwqe_info);
81 	cptvf->wqe_info = NULL;
82 }
83 
84 static void free_pending_queues(struct pending_qinfo *pqinfo)
85 {
86 	int i;
87 	struct pending_queue *queue;
88 
89 	for_each_pending_queue(pqinfo, queue, i) {
90 		if (!queue->head)
91 			continue;
92 
93 		/* free single queue */
94 		kzfree((queue->head));
95 
96 		queue->front = 0;
97 		queue->rear = 0;
98 
99 		return;
100 	}
101 
102 	pqinfo->qlen = 0;
103 	pqinfo->nr_queues = 0;
104 }
105 
106 static int alloc_pending_queues(struct pending_qinfo *pqinfo, u32 qlen,
107 				u32 nr_queues)
108 {
109 	u32 i;
110 	size_t size;
111 	int ret;
112 	struct pending_queue *queue = NULL;
113 
114 	pqinfo->nr_queues = nr_queues;
115 	pqinfo->qlen = qlen;
116 
117 	size = (qlen * sizeof(struct pending_entry));
118 
119 	for_each_pending_queue(pqinfo, queue, i) {
120 		queue->head = kzalloc((size), GFP_KERNEL);
121 		if (!queue->head) {
122 			ret = -ENOMEM;
123 			goto pending_qfail;
124 		}
125 
126 		queue->front = 0;
127 		queue->rear = 0;
128 		atomic64_set((&queue->pending_count), (0));
129 
130 		/* init queue spin lock */
131 		spin_lock_init(&queue->lock);
132 	}
133 
134 	return 0;
135 
136 pending_qfail:
137 	free_pending_queues(pqinfo);
138 
139 	return ret;
140 }
141 
142 static int init_pending_queues(struct cpt_vf *cptvf, u32 qlen, u32 nr_queues)
143 {
144 	struct pci_dev *pdev = cptvf->pdev;
145 	int ret;
146 
147 	if (!nr_queues)
148 		return 0;
149 
150 	ret = alloc_pending_queues(&cptvf->pqinfo, qlen, nr_queues);
151 	if (ret) {
152 		dev_err(&pdev->dev, "failed to setup pending queues (%u)\n",
153 			nr_queues);
154 		return ret;
155 	}
156 
157 	return 0;
158 }
159 
160 static void cleanup_pending_queues(struct cpt_vf *cptvf)
161 {
162 	struct pci_dev *pdev = cptvf->pdev;
163 
164 	if (!cptvf->nr_queues)
165 		return;
166 
167 	dev_info(&pdev->dev, "Cleaning VQ pending queue (%u)\n",
168 		 cptvf->nr_queues);
169 	free_pending_queues(&cptvf->pqinfo);
170 }
171 
172 static void free_command_queues(struct cpt_vf *cptvf,
173 				struct command_qinfo *cqinfo)
174 {
175 	int i;
176 	struct command_queue *queue = NULL;
177 	struct command_chunk *chunk = NULL;
178 	struct pci_dev *pdev = cptvf->pdev;
179 	struct hlist_node *node;
180 
181 	/* clean up for each queue */
182 	for (i = 0; i < cptvf->nr_queues; i++) {
183 		queue = &cqinfo->queue[i];
184 		if (hlist_empty(&cqinfo->queue[i].chead))
185 			continue;
186 
187 		hlist_for_each_entry_safe(chunk, node, &cqinfo->queue[i].chead,
188 					  nextchunk) {
189 			dma_free_coherent(&pdev->dev, chunk->size,
190 					  chunk->head,
191 					  chunk->dma_addr);
192 			chunk->head = NULL;
193 			chunk->dma_addr = 0;
194 			hlist_del(&chunk->nextchunk);
195 			kzfree(chunk);
196 		}
197 
198 		queue->nchunks = 0;
199 		queue->idx = 0;
200 	}
201 
202 	/* common cleanup */
203 	cqinfo->cmd_size = 0;
204 }
205 
206 static int alloc_command_queues(struct cpt_vf *cptvf,
207 				struct command_qinfo *cqinfo, size_t cmd_size,
208 				u32 qlen)
209 {
210 	int i;
211 	size_t q_size;
212 	struct command_queue *queue = NULL;
213 	struct pci_dev *pdev = cptvf->pdev;
214 
215 	/* common init */
216 	cqinfo->cmd_size = cmd_size;
217 	/* Qsize in dwords, needed for SADDR config, 1-next chunk pointer */
218 	cptvf->qsize = min(qlen, cqinfo->qchunksize) *
219 			CPT_NEXT_CHUNK_PTR_SIZE + 1;
220 	/* Qsize in bytes to create space for alignment */
221 	q_size = qlen * cqinfo->cmd_size;
222 
223 	/* per queue initialization */
224 	for (i = 0; i < cptvf->nr_queues; i++) {
225 		size_t c_size = 0;
226 		size_t rem_q_size = q_size;
227 		struct command_chunk *curr = NULL, *first = NULL, *last = NULL;
228 		u32 qcsize_bytes = cqinfo->qchunksize * cqinfo->cmd_size;
229 
230 		queue = &cqinfo->queue[i];
231 		INIT_HLIST_HEAD(&cqinfo->queue[i].chead);
232 		do {
233 			curr = kzalloc(sizeof(*curr), GFP_KERNEL);
234 			if (!curr)
235 				goto cmd_qfail;
236 
237 			c_size = (rem_q_size > qcsize_bytes) ? qcsize_bytes :
238 					rem_q_size;
239 			curr->head = (u8 *)dma_zalloc_coherent(&pdev->dev,
240 					  c_size + CPT_NEXT_CHUNK_PTR_SIZE,
241 					  &curr->dma_addr, GFP_KERNEL);
242 			if (!curr->head) {
243 				dev_err(&pdev->dev, "Command Q (%d) chunk (%d) allocation failed\n",
244 					i, queue->nchunks);
245 				kfree(curr);
246 				goto cmd_qfail;
247 			}
248 
249 			curr->size = c_size;
250 			if (queue->nchunks == 0) {
251 				hlist_add_head(&curr->nextchunk,
252 					       &cqinfo->queue[i].chead);
253 				first = curr;
254 			} else {
255 				hlist_add_behind(&curr->nextchunk,
256 						 &last->nextchunk);
257 			}
258 
259 			queue->nchunks++;
260 			rem_q_size -= c_size;
261 			if (last)
262 				*((u64 *)(&last->head[last->size])) = (u64)curr->dma_addr;
263 
264 			last = curr;
265 		} while (rem_q_size);
266 
267 		/* Make the queue circular */
268 		/* Tie back last chunk entry to head */
269 		curr = first;
270 		*((u64 *)(&last->head[last->size])) = (u64)curr->dma_addr;
271 		queue->qhead = curr;
272 		spin_lock_init(&queue->lock);
273 	}
274 	return 0;
275 
276 cmd_qfail:
277 	free_command_queues(cptvf, cqinfo);
278 	return -ENOMEM;
279 }
280 
281 static int init_command_queues(struct cpt_vf *cptvf, u32 qlen)
282 {
283 	struct pci_dev *pdev = cptvf->pdev;
284 	int ret;
285 
286 	/* setup AE command queues */
287 	ret = alloc_command_queues(cptvf, &cptvf->cqinfo, CPT_INST_SIZE,
288 				   qlen);
289 	if (ret) {
290 		dev_err(&pdev->dev, "failed to allocate AE command queues (%u)\n",
291 			cptvf->nr_queues);
292 		return ret;
293 	}
294 
295 	return ret;
296 }
297 
298 static void cleanup_command_queues(struct cpt_vf *cptvf)
299 {
300 	struct pci_dev *pdev = cptvf->pdev;
301 
302 	if (!cptvf->nr_queues)
303 		return;
304 
305 	dev_info(&pdev->dev, "Cleaning VQ command queue (%u)\n",
306 		 cptvf->nr_queues);
307 	free_command_queues(cptvf, &cptvf->cqinfo);
308 }
309 
310 static void cptvf_sw_cleanup(struct cpt_vf *cptvf)
311 {
312 	cleanup_worker_threads(cptvf);
313 	cleanup_pending_queues(cptvf);
314 	cleanup_command_queues(cptvf);
315 }
316 
317 static int cptvf_sw_init(struct cpt_vf *cptvf, u32 qlen, u32 nr_queues)
318 {
319 	struct pci_dev *pdev = cptvf->pdev;
320 	int ret = 0;
321 	u32 max_dev_queues = 0;
322 
323 	max_dev_queues = CPT_NUM_QS_PER_VF;
324 	/* possible cpus */
325 	nr_queues = min_t(u32, nr_queues, max_dev_queues);
326 	cptvf->nr_queues = nr_queues;
327 
328 	ret = init_command_queues(cptvf, qlen);
329 	if (ret) {
330 		dev_err(&pdev->dev, "Failed to setup command queues (%u)\n",
331 			nr_queues);
332 		return ret;
333 	}
334 
335 	ret = init_pending_queues(cptvf, qlen, nr_queues);
336 	if (ret) {
337 		dev_err(&pdev->dev, "Failed to setup pending queues (%u)\n",
338 			nr_queues);
339 		goto setup_pqfail;
340 	}
341 
342 	/* Create worker threads for BH processing */
343 	ret = init_worker_threads(cptvf);
344 	if (ret) {
345 		dev_err(&pdev->dev, "Failed to setup worker threads\n");
346 		goto init_work_fail;
347 	}
348 
349 	return 0;
350 
351 init_work_fail:
352 	cleanup_worker_threads(cptvf);
353 	cleanup_pending_queues(cptvf);
354 
355 setup_pqfail:
356 	cleanup_command_queues(cptvf);
357 
358 	return ret;
359 }
360 
361 static void cptvf_free_irq_affinity(struct cpt_vf *cptvf, int vec)
362 {
363 	irq_set_affinity_hint(pci_irq_vector(cptvf->pdev, vec), NULL);
364 	free_cpumask_var(cptvf->affinity_mask[vec]);
365 }
366 
367 static void cptvf_write_vq_ctl(struct cpt_vf *cptvf, bool val)
368 {
369 	union cptx_vqx_ctl vqx_ctl;
370 
371 	vqx_ctl.u = cpt_read_csr64(cptvf->reg_base, CPTX_VQX_CTL(0, 0));
372 	vqx_ctl.s.ena = val;
373 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_CTL(0, 0), vqx_ctl.u);
374 }
375 
376 void cptvf_write_vq_doorbell(struct cpt_vf *cptvf, u32 val)
377 {
378 	union cptx_vqx_doorbell vqx_dbell;
379 
380 	vqx_dbell.u = cpt_read_csr64(cptvf->reg_base,
381 				     CPTX_VQX_DOORBELL(0, 0));
382 	vqx_dbell.s.dbell_cnt = val * 8; /* Num of Instructions * 8 words */
383 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_DOORBELL(0, 0),
384 			vqx_dbell.u);
385 }
386 
387 static void cptvf_write_vq_inprog(struct cpt_vf *cptvf, u8 val)
388 {
389 	union cptx_vqx_inprog vqx_inprg;
390 
391 	vqx_inprg.u = cpt_read_csr64(cptvf->reg_base, CPTX_VQX_INPROG(0, 0));
392 	vqx_inprg.s.inflight = val;
393 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_INPROG(0, 0), vqx_inprg.u);
394 }
395 
396 static void cptvf_write_vq_done_numwait(struct cpt_vf *cptvf, u32 val)
397 {
398 	union cptx_vqx_done_wait vqx_dwait;
399 
400 	vqx_dwait.u = cpt_read_csr64(cptvf->reg_base,
401 				     CPTX_VQX_DONE_WAIT(0, 0));
402 	vqx_dwait.s.num_wait = val;
403 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_DONE_WAIT(0, 0),
404 			vqx_dwait.u);
405 }
406 
407 static void cptvf_write_vq_done_timewait(struct cpt_vf *cptvf, u16 time)
408 {
409 	union cptx_vqx_done_wait vqx_dwait;
410 
411 	vqx_dwait.u = cpt_read_csr64(cptvf->reg_base,
412 				     CPTX_VQX_DONE_WAIT(0, 0));
413 	vqx_dwait.s.time_wait = time;
414 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_DONE_WAIT(0, 0),
415 			vqx_dwait.u);
416 }
417 
418 static void cptvf_enable_swerr_interrupts(struct cpt_vf *cptvf)
419 {
420 	union cptx_vqx_misc_ena_w1s vqx_misc_ena;
421 
422 	vqx_misc_ena.u = cpt_read_csr64(cptvf->reg_base,
423 					CPTX_VQX_MISC_ENA_W1S(0, 0));
424 	/* Set mbox(0) interupts for the requested vf */
425 	vqx_misc_ena.s.swerr = 1;
426 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_MISC_ENA_W1S(0, 0),
427 			vqx_misc_ena.u);
428 }
429 
430 static void cptvf_enable_mbox_interrupts(struct cpt_vf *cptvf)
431 {
432 	union cptx_vqx_misc_ena_w1s vqx_misc_ena;
433 
434 	vqx_misc_ena.u = cpt_read_csr64(cptvf->reg_base,
435 					CPTX_VQX_MISC_ENA_W1S(0, 0));
436 	/* Set mbox(0) interupts for the requested vf */
437 	vqx_misc_ena.s.mbox = 1;
438 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_MISC_ENA_W1S(0, 0),
439 			vqx_misc_ena.u);
440 }
441 
442 static void cptvf_enable_done_interrupts(struct cpt_vf *cptvf)
443 {
444 	union cptx_vqx_done_ena_w1s vqx_done_ena;
445 
446 	vqx_done_ena.u = cpt_read_csr64(cptvf->reg_base,
447 					CPTX_VQX_DONE_ENA_W1S(0, 0));
448 	/* Set DONE interrupt for the requested vf */
449 	vqx_done_ena.s.done = 1;
450 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_DONE_ENA_W1S(0, 0),
451 			vqx_done_ena.u);
452 }
453 
454 static void cptvf_clear_dovf_intr(struct cpt_vf *cptvf)
455 {
456 	union cptx_vqx_misc_int vqx_misc_int;
457 
458 	vqx_misc_int.u = cpt_read_csr64(cptvf->reg_base,
459 					CPTX_VQX_MISC_INT(0, 0));
460 	/* W1C for the VF */
461 	vqx_misc_int.s.dovf = 1;
462 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_MISC_INT(0, 0),
463 			vqx_misc_int.u);
464 }
465 
466 static void cptvf_clear_irde_intr(struct cpt_vf *cptvf)
467 {
468 	union cptx_vqx_misc_int vqx_misc_int;
469 
470 	vqx_misc_int.u = cpt_read_csr64(cptvf->reg_base,
471 					CPTX_VQX_MISC_INT(0, 0));
472 	/* W1C for the VF */
473 	vqx_misc_int.s.irde = 1;
474 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_MISC_INT(0, 0),
475 			vqx_misc_int.u);
476 }
477 
478 static void cptvf_clear_nwrp_intr(struct cpt_vf *cptvf)
479 {
480 	union cptx_vqx_misc_int vqx_misc_int;
481 
482 	vqx_misc_int.u = cpt_read_csr64(cptvf->reg_base,
483 					CPTX_VQX_MISC_INT(0, 0));
484 	/* W1C for the VF */
485 	vqx_misc_int.s.nwrp = 1;
486 	cpt_write_csr64(cptvf->reg_base,
487 			CPTX_VQX_MISC_INT(0, 0), vqx_misc_int.u);
488 }
489 
490 static void cptvf_clear_mbox_intr(struct cpt_vf *cptvf)
491 {
492 	union cptx_vqx_misc_int vqx_misc_int;
493 
494 	vqx_misc_int.u = cpt_read_csr64(cptvf->reg_base,
495 					CPTX_VQX_MISC_INT(0, 0));
496 	/* W1C for the VF */
497 	vqx_misc_int.s.mbox = 1;
498 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_MISC_INT(0, 0),
499 			vqx_misc_int.u);
500 }
501 
502 static void cptvf_clear_swerr_intr(struct cpt_vf *cptvf)
503 {
504 	union cptx_vqx_misc_int vqx_misc_int;
505 
506 	vqx_misc_int.u = cpt_read_csr64(cptvf->reg_base,
507 					CPTX_VQX_MISC_INT(0, 0));
508 	/* W1C for the VF */
509 	vqx_misc_int.s.swerr = 1;
510 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_MISC_INT(0, 0),
511 			vqx_misc_int.u);
512 }
513 
514 static u64 cptvf_read_vf_misc_intr_status(struct cpt_vf *cptvf)
515 {
516 	return cpt_read_csr64(cptvf->reg_base, CPTX_VQX_MISC_INT(0, 0));
517 }
518 
519 static irqreturn_t cptvf_misc_intr_handler(int irq, void *cptvf_irq)
520 {
521 	struct cpt_vf *cptvf = (struct cpt_vf *)cptvf_irq;
522 	struct pci_dev *pdev = cptvf->pdev;
523 	u64 intr;
524 
525 	intr = cptvf_read_vf_misc_intr_status(cptvf);
526 	/*Check for MISC interrupt types*/
527 	if (likely(intr & CPT_VF_INTR_MBOX_MASK)) {
528 		dev_dbg(&pdev->dev, "Mailbox interrupt 0x%llx on CPT VF %d\n",
529 			intr, cptvf->vfid);
530 		cptvf_handle_mbox_intr(cptvf);
531 		cptvf_clear_mbox_intr(cptvf);
532 	} else if (unlikely(intr & CPT_VF_INTR_DOVF_MASK)) {
533 		cptvf_clear_dovf_intr(cptvf);
534 		/*Clear doorbell count*/
535 		cptvf_write_vq_doorbell(cptvf, 0);
536 		dev_err(&pdev->dev, "Doorbell overflow error interrupt 0x%llx on CPT VF %d\n",
537 			intr, cptvf->vfid);
538 	} else if (unlikely(intr & CPT_VF_INTR_IRDE_MASK)) {
539 		cptvf_clear_irde_intr(cptvf);
540 		dev_err(&pdev->dev, "Instruction NCB read error interrupt 0x%llx on CPT VF %d\n",
541 			intr, cptvf->vfid);
542 	} else if (unlikely(intr & CPT_VF_INTR_NWRP_MASK)) {
543 		cptvf_clear_nwrp_intr(cptvf);
544 		dev_err(&pdev->dev, "NCB response write error interrupt 0x%llx on CPT VF %d\n",
545 			intr, cptvf->vfid);
546 	} else if (unlikely(intr & CPT_VF_INTR_SERR_MASK)) {
547 		cptvf_clear_swerr_intr(cptvf);
548 		dev_err(&pdev->dev, "Software error interrupt 0x%llx on CPT VF %d\n",
549 			intr, cptvf->vfid);
550 	} else {
551 		dev_err(&pdev->dev, "Unhandled interrupt in CPT VF %d\n",
552 			cptvf->vfid);
553 	}
554 
555 	return IRQ_HANDLED;
556 }
557 
558 static inline struct cptvf_wqe *get_cptvf_vq_wqe(struct cpt_vf *cptvf,
559 						 int qno)
560 {
561 	struct cptvf_wqe_info *nwqe_info;
562 
563 	if (unlikely(qno >= cptvf->nr_queues))
564 		return NULL;
565 	nwqe_info = (struct cptvf_wqe_info *)cptvf->wqe_info;
566 
567 	return &nwqe_info->vq_wqe[qno];
568 }
569 
570 static inline u32 cptvf_read_vq_done_count(struct cpt_vf *cptvf)
571 {
572 	union cptx_vqx_done vqx_done;
573 
574 	vqx_done.u = cpt_read_csr64(cptvf->reg_base, CPTX_VQX_DONE(0, 0));
575 	return vqx_done.s.done;
576 }
577 
578 static inline void cptvf_write_vq_done_ack(struct cpt_vf *cptvf,
579 					   u32 ackcnt)
580 {
581 	union cptx_vqx_done_ack vqx_dack_cnt;
582 
583 	vqx_dack_cnt.u = cpt_read_csr64(cptvf->reg_base,
584 					CPTX_VQX_DONE_ACK(0, 0));
585 	vqx_dack_cnt.s.done_ack = ackcnt;
586 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_DONE_ACK(0, 0),
587 			vqx_dack_cnt.u);
588 }
589 
590 static irqreturn_t cptvf_done_intr_handler(int irq, void *cptvf_irq)
591 {
592 	struct cpt_vf *cptvf = (struct cpt_vf *)cptvf_irq;
593 	struct pci_dev *pdev = cptvf->pdev;
594 	/* Read the number of completions */
595 	u32 intr = cptvf_read_vq_done_count(cptvf);
596 
597 	if (intr) {
598 		struct cptvf_wqe *wqe;
599 
600 		/* Acknowledge the number of
601 		 * scheduled completions for processing
602 		 */
603 		cptvf_write_vq_done_ack(cptvf, intr);
604 		wqe = get_cptvf_vq_wqe(cptvf, 0);
605 		if (unlikely(!wqe)) {
606 			dev_err(&pdev->dev, "No work to schedule for VF (%d)",
607 				cptvf->vfid);
608 			return IRQ_NONE;
609 		}
610 		tasklet_hi_schedule(&wqe->twork);
611 	}
612 
613 	return IRQ_HANDLED;
614 }
615 
616 static void cptvf_set_irq_affinity(struct cpt_vf *cptvf, int vec)
617 {
618 	struct pci_dev *pdev = cptvf->pdev;
619 	int cpu;
620 
621 	if (!zalloc_cpumask_var(&cptvf->affinity_mask[vec],
622 				GFP_KERNEL)) {
623 		dev_err(&pdev->dev, "Allocation failed for affinity_mask for VF %d",
624 			cptvf->vfid);
625 		return;
626 	}
627 
628 	cpu = cptvf->vfid % num_online_cpus();
629 	cpumask_set_cpu(cpumask_local_spread(cpu, cptvf->node),
630 			cptvf->affinity_mask[vec]);
631 	irq_set_affinity_hint(pci_irq_vector(pdev, vec),
632 			cptvf->affinity_mask[vec]);
633 }
634 
635 static void cptvf_write_vq_saddr(struct cpt_vf *cptvf, u64 val)
636 {
637 	union cptx_vqx_saddr vqx_saddr;
638 
639 	vqx_saddr.u = val;
640 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_SADDR(0, 0), vqx_saddr.u);
641 }
642 
643 void cptvf_device_init(struct cpt_vf *cptvf)
644 {
645 	u64 base_addr = 0;
646 
647 	/* Disable the VQ */
648 	cptvf_write_vq_ctl(cptvf, 0);
649 	/* Reset the doorbell */
650 	cptvf_write_vq_doorbell(cptvf, 0);
651 	/* Clear inflight */
652 	cptvf_write_vq_inprog(cptvf, 0);
653 	/* Write VQ SADDR */
654 	/* TODO: for now only one queue, so hard coded */
655 	base_addr = (u64)(cptvf->cqinfo.queue[0].qhead->dma_addr);
656 	cptvf_write_vq_saddr(cptvf, base_addr);
657 	/* Configure timerhold / coalescence */
658 	cptvf_write_vq_done_timewait(cptvf, CPT_TIMER_THOLD);
659 	cptvf_write_vq_done_numwait(cptvf, 1);
660 	/* Enable the VQ */
661 	cptvf_write_vq_ctl(cptvf, 1);
662 	/* Flag the VF ready */
663 	cptvf->flags |= CPT_FLAG_DEVICE_READY;
664 }
665 
666 static int cptvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
667 {
668 	struct device *dev = &pdev->dev;
669 	struct cpt_vf *cptvf;
670 	int    err;
671 
672 	cptvf = devm_kzalloc(dev, sizeof(*cptvf), GFP_KERNEL);
673 	if (!cptvf)
674 		return -ENOMEM;
675 
676 	pci_set_drvdata(pdev, cptvf);
677 	cptvf->pdev = pdev;
678 	err = pci_enable_device(pdev);
679 	if (err) {
680 		dev_err(dev, "Failed to enable PCI device\n");
681 		pci_set_drvdata(pdev, NULL);
682 		return err;
683 	}
684 
685 	err = pci_request_regions(pdev, DRV_NAME);
686 	if (err) {
687 		dev_err(dev, "PCI request regions failed 0x%x\n", err);
688 		goto cptvf_err_disable_device;
689 	}
690 	/* Mark as VF driver */
691 	cptvf->flags |= CPT_FLAG_VF_DRIVER;
692 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(48));
693 	if (err) {
694 		dev_err(dev, "Unable to get usable DMA configuration\n");
695 		goto cptvf_err_release_regions;
696 	}
697 
698 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(48));
699 	if (err) {
700 		dev_err(dev, "Unable to get 48-bit DMA for consistent allocations\n");
701 		goto cptvf_err_release_regions;
702 	}
703 
704 	/* MAP PF's configuration registers */
705 	cptvf->reg_base = pcim_iomap(pdev, 0, 0);
706 	if (!cptvf->reg_base) {
707 		dev_err(dev, "Cannot map config register space, aborting\n");
708 		err = -ENOMEM;
709 		goto cptvf_err_release_regions;
710 	}
711 
712 	cptvf->node = dev_to_node(&pdev->dev);
713 	err = pci_alloc_irq_vectors(pdev, CPT_VF_MSIX_VECTORS,
714 			CPT_VF_MSIX_VECTORS, PCI_IRQ_MSIX);
715 	if (err < 0) {
716 		dev_err(dev, "Request for #%d msix vectors failed\n",
717 			CPT_VF_MSIX_VECTORS);
718 		goto cptvf_err_release_regions;
719 	}
720 
721 	err = request_irq(pci_irq_vector(pdev, CPT_VF_INT_VEC_E_MISC),
722 			  cptvf_misc_intr_handler, 0, "CPT VF misc intr",
723 			  cptvf);
724 	if (err) {
725 		dev_err(dev, "Request misc irq failed");
726 		goto cptvf_free_vectors;
727 	}
728 
729 	/* Enable mailbox interrupt */
730 	cptvf_enable_mbox_interrupts(cptvf);
731 	cptvf_enable_swerr_interrupts(cptvf);
732 
733 	/* Check ready with PF */
734 	/* Gets chip ID / device Id from PF if ready */
735 	err = cptvf_check_pf_ready(cptvf);
736 	if (err) {
737 		dev_err(dev, "PF not responding to READY msg");
738 		goto cptvf_free_misc_irq;
739 	}
740 
741 	/* CPT VF software resources initialization */
742 	cptvf->cqinfo.qchunksize = CPT_CMD_QCHUNK_SIZE;
743 	err = cptvf_sw_init(cptvf, CPT_CMD_QLEN, CPT_NUM_QS_PER_VF);
744 	if (err) {
745 		dev_err(dev, "cptvf_sw_init() failed");
746 		goto cptvf_free_misc_irq;
747 	}
748 	/* Convey VQ LEN to PF */
749 	err = cptvf_send_vq_size_msg(cptvf);
750 	if (err) {
751 		dev_err(dev, "PF not responding to QLEN msg");
752 		goto cptvf_free_misc_irq;
753 	}
754 
755 	/* CPT VF device initialization */
756 	cptvf_device_init(cptvf);
757 	/* Send msg to PF to assign currnet Q to required group */
758 	cptvf->vfgrp = 1;
759 	err = cptvf_send_vf_to_grp_msg(cptvf);
760 	if (err) {
761 		dev_err(dev, "PF not responding to VF_GRP msg");
762 		goto cptvf_free_misc_irq;
763 	}
764 
765 	cptvf->priority = 1;
766 	err = cptvf_send_vf_priority_msg(cptvf);
767 	if (err) {
768 		dev_err(dev, "PF not responding to VF_PRIO msg");
769 		goto cptvf_free_misc_irq;
770 	}
771 
772 	err = request_irq(pci_irq_vector(pdev, CPT_VF_INT_VEC_E_DONE),
773 			  cptvf_done_intr_handler, 0, "CPT VF done intr",
774 			  cptvf);
775 	if (err) {
776 		dev_err(dev, "Request done irq failed\n");
777 		goto cptvf_free_misc_irq;
778 	}
779 
780 	/* Enable mailbox interrupt */
781 	cptvf_enable_done_interrupts(cptvf);
782 
783 	/* Set irq affinity masks */
784 	cptvf_set_irq_affinity(cptvf, CPT_VF_INT_VEC_E_MISC);
785 	cptvf_set_irq_affinity(cptvf, CPT_VF_INT_VEC_E_DONE);
786 
787 	err = cptvf_send_vf_up(cptvf);
788 	if (err) {
789 		dev_err(dev, "PF not responding to UP msg");
790 		goto cptvf_free_irq_affinity;
791 	}
792 	err = cvm_crypto_init(cptvf);
793 	if (err) {
794 		dev_err(dev, "Algorithm register failed\n");
795 		goto cptvf_free_irq_affinity;
796 	}
797 	return 0;
798 
799 cptvf_free_irq_affinity:
800 	cptvf_free_irq_affinity(cptvf, CPT_VF_INT_VEC_E_DONE);
801 	cptvf_free_irq_affinity(cptvf, CPT_VF_INT_VEC_E_MISC);
802 cptvf_free_misc_irq:
803 	free_irq(pci_irq_vector(pdev, CPT_VF_INT_VEC_E_MISC), cptvf);
804 cptvf_free_vectors:
805 	pci_free_irq_vectors(cptvf->pdev);
806 cptvf_err_release_regions:
807 	pci_release_regions(pdev);
808 cptvf_err_disable_device:
809 	pci_disable_device(pdev);
810 	pci_set_drvdata(pdev, NULL);
811 
812 	return err;
813 }
814 
815 static void cptvf_remove(struct pci_dev *pdev)
816 {
817 	struct cpt_vf *cptvf = pci_get_drvdata(pdev);
818 
819 	if (!cptvf) {
820 		dev_err(&pdev->dev, "Invalid CPT-VF device\n");
821 		return;
822 	}
823 
824 	/* Convey DOWN to PF */
825 	if (cptvf_send_vf_down(cptvf)) {
826 		dev_err(&pdev->dev, "PF not responding to DOWN msg");
827 	} else {
828 		cptvf_free_irq_affinity(cptvf, CPT_VF_INT_VEC_E_DONE);
829 		cptvf_free_irq_affinity(cptvf, CPT_VF_INT_VEC_E_MISC);
830 		free_irq(pci_irq_vector(pdev, CPT_VF_INT_VEC_E_DONE), cptvf);
831 		free_irq(pci_irq_vector(pdev, CPT_VF_INT_VEC_E_MISC), cptvf);
832 		pci_free_irq_vectors(cptvf->pdev);
833 		cptvf_sw_cleanup(cptvf);
834 		pci_set_drvdata(pdev, NULL);
835 		pci_release_regions(pdev);
836 		pci_disable_device(pdev);
837 		cvm_crypto_exit();
838 	}
839 }
840 
841 static void cptvf_shutdown(struct pci_dev *pdev)
842 {
843 	cptvf_remove(pdev);
844 }
845 
846 /* Supported devices */
847 static const struct pci_device_id cptvf_id_table[] = {
848 	{PCI_VDEVICE(CAVIUM, CPT_81XX_PCI_VF_DEVICE_ID), 0},
849 	{ 0, }  /* end of table */
850 };
851 
852 static struct pci_driver cptvf_pci_driver = {
853 	.name = DRV_NAME,
854 	.id_table = cptvf_id_table,
855 	.probe = cptvf_probe,
856 	.remove = cptvf_remove,
857 	.shutdown = cptvf_shutdown,
858 };
859 
860 module_pci_driver(cptvf_pci_driver);
861 
862 MODULE_AUTHOR("George Cherian <george.cherian@cavium.com>");
863 MODULE_DESCRIPTION("Cavium Thunder CPT Virtual Function Driver");
864 MODULE_LICENSE("GPL v2");
865 MODULE_VERSION(DRV_VERSION);
866 MODULE_DEVICE_TABLE(pci, cptvf_id_table);
867