xref: /linux/drivers/crypto/cavium/cpt/cptvf_main.c (revision 95298d63c67673c654c08952672d016212b26054)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2016 Cavium, Inc.
4  */
5 
6 #include <linux/interrupt.h>
7 #include <linux/module.h>
8 
9 #include "cptvf.h"
10 
11 #define DRV_NAME	"thunder-cptvf"
12 #define DRV_VERSION	"1.0"
13 
14 struct cptvf_wqe {
15 	struct tasklet_struct twork;
16 	void *cptvf;
17 	u32 qno;
18 };
19 
20 struct cptvf_wqe_info {
21 	struct cptvf_wqe vq_wqe[CPT_NUM_QS_PER_VF];
22 };
23 
24 static void vq_work_handler(unsigned long data)
25 {
26 	struct cptvf_wqe_info *cwqe_info = (struct cptvf_wqe_info *)data;
27 	struct cptvf_wqe *cwqe = &cwqe_info->vq_wqe[0];
28 
29 	vq_post_process(cwqe->cptvf, cwqe->qno);
30 }
31 
32 static int init_worker_threads(struct cpt_vf *cptvf)
33 {
34 	struct pci_dev *pdev = cptvf->pdev;
35 	struct cptvf_wqe_info *cwqe_info;
36 	int i;
37 
38 	cwqe_info = kzalloc(sizeof(*cwqe_info), GFP_KERNEL);
39 	if (!cwqe_info)
40 		return -ENOMEM;
41 
42 	if (cptvf->nr_queues) {
43 		dev_info(&pdev->dev, "Creating VQ worker threads (%d)\n",
44 			 cptvf->nr_queues);
45 	}
46 
47 	for (i = 0; i < cptvf->nr_queues; i++) {
48 		tasklet_init(&cwqe_info->vq_wqe[i].twork, vq_work_handler,
49 			     (u64)cwqe_info);
50 		cwqe_info->vq_wqe[i].qno = i;
51 		cwqe_info->vq_wqe[i].cptvf = cptvf;
52 	}
53 
54 	cptvf->wqe_info = cwqe_info;
55 
56 	return 0;
57 }
58 
59 static void cleanup_worker_threads(struct cpt_vf *cptvf)
60 {
61 	struct cptvf_wqe_info *cwqe_info;
62 	struct pci_dev *pdev = cptvf->pdev;
63 	int i;
64 
65 	cwqe_info = (struct cptvf_wqe_info *)cptvf->wqe_info;
66 	if (!cwqe_info)
67 		return;
68 
69 	if (cptvf->nr_queues) {
70 		dev_info(&pdev->dev, "Cleaning VQ worker threads (%u)\n",
71 			 cptvf->nr_queues);
72 	}
73 
74 	for (i = 0; i < cptvf->nr_queues; i++)
75 		tasklet_kill(&cwqe_info->vq_wqe[i].twork);
76 
77 	kzfree(cwqe_info);
78 	cptvf->wqe_info = NULL;
79 }
80 
81 static void free_pending_queues(struct pending_qinfo *pqinfo)
82 {
83 	int i;
84 	struct pending_queue *queue;
85 
86 	for_each_pending_queue(pqinfo, queue, i) {
87 		if (!queue->head)
88 			continue;
89 
90 		/* free single queue */
91 		kzfree((queue->head));
92 
93 		queue->front = 0;
94 		queue->rear = 0;
95 
96 		return;
97 	}
98 
99 	pqinfo->qlen = 0;
100 	pqinfo->nr_queues = 0;
101 }
102 
103 static int alloc_pending_queues(struct pending_qinfo *pqinfo, u32 qlen,
104 				u32 nr_queues)
105 {
106 	u32 i;
107 	size_t size;
108 	int ret;
109 	struct pending_queue *queue = NULL;
110 
111 	pqinfo->nr_queues = nr_queues;
112 	pqinfo->qlen = qlen;
113 
114 	size = (qlen * sizeof(struct pending_entry));
115 
116 	for_each_pending_queue(pqinfo, queue, i) {
117 		queue->head = kzalloc((size), GFP_KERNEL);
118 		if (!queue->head) {
119 			ret = -ENOMEM;
120 			goto pending_qfail;
121 		}
122 
123 		queue->front = 0;
124 		queue->rear = 0;
125 		atomic64_set((&queue->pending_count), (0));
126 
127 		/* init queue spin lock */
128 		spin_lock_init(&queue->lock);
129 	}
130 
131 	return 0;
132 
133 pending_qfail:
134 	free_pending_queues(pqinfo);
135 
136 	return ret;
137 }
138 
139 static int init_pending_queues(struct cpt_vf *cptvf, u32 qlen, u32 nr_queues)
140 {
141 	struct pci_dev *pdev = cptvf->pdev;
142 	int ret;
143 
144 	if (!nr_queues)
145 		return 0;
146 
147 	ret = alloc_pending_queues(&cptvf->pqinfo, qlen, nr_queues);
148 	if (ret) {
149 		dev_err(&pdev->dev, "failed to setup pending queues (%u)\n",
150 			nr_queues);
151 		return ret;
152 	}
153 
154 	return 0;
155 }
156 
157 static void cleanup_pending_queues(struct cpt_vf *cptvf)
158 {
159 	struct pci_dev *pdev = cptvf->pdev;
160 
161 	if (!cptvf->nr_queues)
162 		return;
163 
164 	dev_info(&pdev->dev, "Cleaning VQ pending queue (%u)\n",
165 		 cptvf->nr_queues);
166 	free_pending_queues(&cptvf->pqinfo);
167 }
168 
169 static void free_command_queues(struct cpt_vf *cptvf,
170 				struct command_qinfo *cqinfo)
171 {
172 	int i;
173 	struct command_queue *queue = NULL;
174 	struct command_chunk *chunk = NULL;
175 	struct pci_dev *pdev = cptvf->pdev;
176 	struct hlist_node *node;
177 
178 	/* clean up for each queue */
179 	for (i = 0; i < cptvf->nr_queues; i++) {
180 		queue = &cqinfo->queue[i];
181 		if (hlist_empty(&cqinfo->queue[i].chead))
182 			continue;
183 
184 		hlist_for_each_entry_safe(chunk, node, &cqinfo->queue[i].chead,
185 					  nextchunk) {
186 			dma_free_coherent(&pdev->dev, chunk->size,
187 					  chunk->head,
188 					  chunk->dma_addr);
189 			chunk->head = NULL;
190 			chunk->dma_addr = 0;
191 			hlist_del(&chunk->nextchunk);
192 			kzfree(chunk);
193 		}
194 
195 		queue->nchunks = 0;
196 		queue->idx = 0;
197 	}
198 
199 	/* common cleanup */
200 	cqinfo->cmd_size = 0;
201 }
202 
203 static int alloc_command_queues(struct cpt_vf *cptvf,
204 				struct command_qinfo *cqinfo, size_t cmd_size,
205 				u32 qlen)
206 {
207 	int i;
208 	size_t q_size;
209 	struct command_queue *queue = NULL;
210 	struct pci_dev *pdev = cptvf->pdev;
211 
212 	/* common init */
213 	cqinfo->cmd_size = cmd_size;
214 	/* Qsize in dwords, needed for SADDR config, 1-next chunk pointer */
215 	cptvf->qsize = min(qlen, cqinfo->qchunksize) *
216 			CPT_NEXT_CHUNK_PTR_SIZE + 1;
217 	/* Qsize in bytes to create space for alignment */
218 	q_size = qlen * cqinfo->cmd_size;
219 
220 	/* per queue initialization */
221 	for (i = 0; i < cptvf->nr_queues; i++) {
222 		size_t c_size = 0;
223 		size_t rem_q_size = q_size;
224 		struct command_chunk *curr = NULL, *first = NULL, *last = NULL;
225 		u32 qcsize_bytes = cqinfo->qchunksize * cqinfo->cmd_size;
226 
227 		queue = &cqinfo->queue[i];
228 		INIT_HLIST_HEAD(&cqinfo->queue[i].chead);
229 		do {
230 			curr = kzalloc(sizeof(*curr), GFP_KERNEL);
231 			if (!curr)
232 				goto cmd_qfail;
233 
234 			c_size = (rem_q_size > qcsize_bytes) ? qcsize_bytes :
235 					rem_q_size;
236 			curr->head = (u8 *)dma_alloc_coherent(&pdev->dev,
237 							      c_size + CPT_NEXT_CHUNK_PTR_SIZE,
238 							      &curr->dma_addr,
239 							      GFP_KERNEL);
240 			if (!curr->head) {
241 				dev_err(&pdev->dev, "Command Q (%d) chunk (%d) allocation failed\n",
242 					i, queue->nchunks);
243 				kfree(curr);
244 				goto cmd_qfail;
245 			}
246 
247 			curr->size = c_size;
248 			if (queue->nchunks == 0) {
249 				hlist_add_head(&curr->nextchunk,
250 					       &cqinfo->queue[i].chead);
251 				first = curr;
252 			} else {
253 				hlist_add_behind(&curr->nextchunk,
254 						 &last->nextchunk);
255 			}
256 
257 			queue->nchunks++;
258 			rem_q_size -= c_size;
259 			if (last)
260 				*((u64 *)(&last->head[last->size])) = (u64)curr->dma_addr;
261 
262 			last = curr;
263 		} while (rem_q_size);
264 
265 		/* Make the queue circular */
266 		/* Tie back last chunk entry to head */
267 		curr = first;
268 		*((u64 *)(&last->head[last->size])) = (u64)curr->dma_addr;
269 		queue->qhead = curr;
270 		spin_lock_init(&queue->lock);
271 	}
272 	return 0;
273 
274 cmd_qfail:
275 	free_command_queues(cptvf, cqinfo);
276 	return -ENOMEM;
277 }
278 
279 static int init_command_queues(struct cpt_vf *cptvf, u32 qlen)
280 {
281 	struct pci_dev *pdev = cptvf->pdev;
282 	int ret;
283 
284 	/* setup AE command queues */
285 	ret = alloc_command_queues(cptvf, &cptvf->cqinfo, CPT_INST_SIZE,
286 				   qlen);
287 	if (ret) {
288 		dev_err(&pdev->dev, "failed to allocate AE command queues (%u)\n",
289 			cptvf->nr_queues);
290 		return ret;
291 	}
292 
293 	return ret;
294 }
295 
296 static void cleanup_command_queues(struct cpt_vf *cptvf)
297 {
298 	struct pci_dev *pdev = cptvf->pdev;
299 
300 	if (!cptvf->nr_queues)
301 		return;
302 
303 	dev_info(&pdev->dev, "Cleaning VQ command queue (%u)\n",
304 		 cptvf->nr_queues);
305 	free_command_queues(cptvf, &cptvf->cqinfo);
306 }
307 
308 static void cptvf_sw_cleanup(struct cpt_vf *cptvf)
309 {
310 	cleanup_worker_threads(cptvf);
311 	cleanup_pending_queues(cptvf);
312 	cleanup_command_queues(cptvf);
313 }
314 
315 static int cptvf_sw_init(struct cpt_vf *cptvf, u32 qlen, u32 nr_queues)
316 {
317 	struct pci_dev *pdev = cptvf->pdev;
318 	int ret = 0;
319 	u32 max_dev_queues = 0;
320 
321 	max_dev_queues = CPT_NUM_QS_PER_VF;
322 	/* possible cpus */
323 	nr_queues = min_t(u32, nr_queues, max_dev_queues);
324 	cptvf->nr_queues = nr_queues;
325 
326 	ret = init_command_queues(cptvf, qlen);
327 	if (ret) {
328 		dev_err(&pdev->dev, "Failed to setup command queues (%u)\n",
329 			nr_queues);
330 		return ret;
331 	}
332 
333 	ret = init_pending_queues(cptvf, qlen, nr_queues);
334 	if (ret) {
335 		dev_err(&pdev->dev, "Failed to setup pending queues (%u)\n",
336 			nr_queues);
337 		goto setup_pqfail;
338 	}
339 
340 	/* Create worker threads for BH processing */
341 	ret = init_worker_threads(cptvf);
342 	if (ret) {
343 		dev_err(&pdev->dev, "Failed to setup worker threads\n");
344 		goto init_work_fail;
345 	}
346 
347 	return 0;
348 
349 init_work_fail:
350 	cleanup_worker_threads(cptvf);
351 	cleanup_pending_queues(cptvf);
352 
353 setup_pqfail:
354 	cleanup_command_queues(cptvf);
355 
356 	return ret;
357 }
358 
359 static void cptvf_free_irq_affinity(struct cpt_vf *cptvf, int vec)
360 {
361 	irq_set_affinity_hint(pci_irq_vector(cptvf->pdev, vec), NULL);
362 	free_cpumask_var(cptvf->affinity_mask[vec]);
363 }
364 
365 static void cptvf_write_vq_ctl(struct cpt_vf *cptvf, bool val)
366 {
367 	union cptx_vqx_ctl vqx_ctl;
368 
369 	vqx_ctl.u = cpt_read_csr64(cptvf->reg_base, CPTX_VQX_CTL(0, 0));
370 	vqx_ctl.s.ena = val;
371 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_CTL(0, 0), vqx_ctl.u);
372 }
373 
374 void cptvf_write_vq_doorbell(struct cpt_vf *cptvf, u32 val)
375 {
376 	union cptx_vqx_doorbell vqx_dbell;
377 
378 	vqx_dbell.u = cpt_read_csr64(cptvf->reg_base,
379 				     CPTX_VQX_DOORBELL(0, 0));
380 	vqx_dbell.s.dbell_cnt = val * 8; /* Num of Instructions * 8 words */
381 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_DOORBELL(0, 0),
382 			vqx_dbell.u);
383 }
384 
385 static void cptvf_write_vq_inprog(struct cpt_vf *cptvf, u8 val)
386 {
387 	union cptx_vqx_inprog vqx_inprg;
388 
389 	vqx_inprg.u = cpt_read_csr64(cptvf->reg_base, CPTX_VQX_INPROG(0, 0));
390 	vqx_inprg.s.inflight = val;
391 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_INPROG(0, 0), vqx_inprg.u);
392 }
393 
394 static void cptvf_write_vq_done_numwait(struct cpt_vf *cptvf, u32 val)
395 {
396 	union cptx_vqx_done_wait vqx_dwait;
397 
398 	vqx_dwait.u = cpt_read_csr64(cptvf->reg_base,
399 				     CPTX_VQX_DONE_WAIT(0, 0));
400 	vqx_dwait.s.num_wait = val;
401 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_DONE_WAIT(0, 0),
402 			vqx_dwait.u);
403 }
404 
405 static void cptvf_write_vq_done_timewait(struct cpt_vf *cptvf, u16 time)
406 {
407 	union cptx_vqx_done_wait vqx_dwait;
408 
409 	vqx_dwait.u = cpt_read_csr64(cptvf->reg_base,
410 				     CPTX_VQX_DONE_WAIT(0, 0));
411 	vqx_dwait.s.time_wait = time;
412 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_DONE_WAIT(0, 0),
413 			vqx_dwait.u);
414 }
415 
416 static void cptvf_enable_swerr_interrupts(struct cpt_vf *cptvf)
417 {
418 	union cptx_vqx_misc_ena_w1s vqx_misc_ena;
419 
420 	vqx_misc_ena.u = cpt_read_csr64(cptvf->reg_base,
421 					CPTX_VQX_MISC_ENA_W1S(0, 0));
422 	/* Set mbox(0) interupts for the requested vf */
423 	vqx_misc_ena.s.swerr = 1;
424 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_MISC_ENA_W1S(0, 0),
425 			vqx_misc_ena.u);
426 }
427 
428 static void cptvf_enable_mbox_interrupts(struct cpt_vf *cptvf)
429 {
430 	union cptx_vqx_misc_ena_w1s vqx_misc_ena;
431 
432 	vqx_misc_ena.u = cpt_read_csr64(cptvf->reg_base,
433 					CPTX_VQX_MISC_ENA_W1S(0, 0));
434 	/* Set mbox(0) interupts for the requested vf */
435 	vqx_misc_ena.s.mbox = 1;
436 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_MISC_ENA_W1S(0, 0),
437 			vqx_misc_ena.u);
438 }
439 
440 static void cptvf_enable_done_interrupts(struct cpt_vf *cptvf)
441 {
442 	union cptx_vqx_done_ena_w1s vqx_done_ena;
443 
444 	vqx_done_ena.u = cpt_read_csr64(cptvf->reg_base,
445 					CPTX_VQX_DONE_ENA_W1S(0, 0));
446 	/* Set DONE interrupt for the requested vf */
447 	vqx_done_ena.s.done = 1;
448 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_DONE_ENA_W1S(0, 0),
449 			vqx_done_ena.u);
450 }
451 
452 static void cptvf_clear_dovf_intr(struct cpt_vf *cptvf)
453 {
454 	union cptx_vqx_misc_int vqx_misc_int;
455 
456 	vqx_misc_int.u = cpt_read_csr64(cptvf->reg_base,
457 					CPTX_VQX_MISC_INT(0, 0));
458 	/* W1C for the VF */
459 	vqx_misc_int.s.dovf = 1;
460 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_MISC_INT(0, 0),
461 			vqx_misc_int.u);
462 }
463 
464 static void cptvf_clear_irde_intr(struct cpt_vf *cptvf)
465 {
466 	union cptx_vqx_misc_int vqx_misc_int;
467 
468 	vqx_misc_int.u = cpt_read_csr64(cptvf->reg_base,
469 					CPTX_VQX_MISC_INT(0, 0));
470 	/* W1C for the VF */
471 	vqx_misc_int.s.irde = 1;
472 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_MISC_INT(0, 0),
473 			vqx_misc_int.u);
474 }
475 
476 static void cptvf_clear_nwrp_intr(struct cpt_vf *cptvf)
477 {
478 	union cptx_vqx_misc_int vqx_misc_int;
479 
480 	vqx_misc_int.u = cpt_read_csr64(cptvf->reg_base,
481 					CPTX_VQX_MISC_INT(0, 0));
482 	/* W1C for the VF */
483 	vqx_misc_int.s.nwrp = 1;
484 	cpt_write_csr64(cptvf->reg_base,
485 			CPTX_VQX_MISC_INT(0, 0), vqx_misc_int.u);
486 }
487 
488 static void cptvf_clear_mbox_intr(struct cpt_vf *cptvf)
489 {
490 	union cptx_vqx_misc_int vqx_misc_int;
491 
492 	vqx_misc_int.u = cpt_read_csr64(cptvf->reg_base,
493 					CPTX_VQX_MISC_INT(0, 0));
494 	/* W1C for the VF */
495 	vqx_misc_int.s.mbox = 1;
496 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_MISC_INT(0, 0),
497 			vqx_misc_int.u);
498 }
499 
500 static void cptvf_clear_swerr_intr(struct cpt_vf *cptvf)
501 {
502 	union cptx_vqx_misc_int vqx_misc_int;
503 
504 	vqx_misc_int.u = cpt_read_csr64(cptvf->reg_base,
505 					CPTX_VQX_MISC_INT(0, 0));
506 	/* W1C for the VF */
507 	vqx_misc_int.s.swerr = 1;
508 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_MISC_INT(0, 0),
509 			vqx_misc_int.u);
510 }
511 
512 static u64 cptvf_read_vf_misc_intr_status(struct cpt_vf *cptvf)
513 {
514 	return cpt_read_csr64(cptvf->reg_base, CPTX_VQX_MISC_INT(0, 0));
515 }
516 
517 static irqreturn_t cptvf_misc_intr_handler(int irq, void *cptvf_irq)
518 {
519 	struct cpt_vf *cptvf = (struct cpt_vf *)cptvf_irq;
520 	struct pci_dev *pdev = cptvf->pdev;
521 	u64 intr;
522 
523 	intr = cptvf_read_vf_misc_intr_status(cptvf);
524 	/*Check for MISC interrupt types*/
525 	if (likely(intr & CPT_VF_INTR_MBOX_MASK)) {
526 		dev_dbg(&pdev->dev, "Mailbox interrupt 0x%llx on CPT VF %d\n",
527 			intr, cptvf->vfid);
528 		cptvf_handle_mbox_intr(cptvf);
529 		cptvf_clear_mbox_intr(cptvf);
530 	} else if (unlikely(intr & CPT_VF_INTR_DOVF_MASK)) {
531 		cptvf_clear_dovf_intr(cptvf);
532 		/*Clear doorbell count*/
533 		cptvf_write_vq_doorbell(cptvf, 0);
534 		dev_err(&pdev->dev, "Doorbell overflow error interrupt 0x%llx on CPT VF %d\n",
535 			intr, cptvf->vfid);
536 	} else if (unlikely(intr & CPT_VF_INTR_IRDE_MASK)) {
537 		cptvf_clear_irde_intr(cptvf);
538 		dev_err(&pdev->dev, "Instruction NCB read error interrupt 0x%llx on CPT VF %d\n",
539 			intr, cptvf->vfid);
540 	} else if (unlikely(intr & CPT_VF_INTR_NWRP_MASK)) {
541 		cptvf_clear_nwrp_intr(cptvf);
542 		dev_err(&pdev->dev, "NCB response write error interrupt 0x%llx on CPT VF %d\n",
543 			intr, cptvf->vfid);
544 	} else if (unlikely(intr & CPT_VF_INTR_SERR_MASK)) {
545 		cptvf_clear_swerr_intr(cptvf);
546 		dev_err(&pdev->dev, "Software error interrupt 0x%llx on CPT VF %d\n",
547 			intr, cptvf->vfid);
548 	} else {
549 		dev_err(&pdev->dev, "Unhandled interrupt in CPT VF %d\n",
550 			cptvf->vfid);
551 	}
552 
553 	return IRQ_HANDLED;
554 }
555 
556 static inline struct cptvf_wqe *get_cptvf_vq_wqe(struct cpt_vf *cptvf,
557 						 int qno)
558 {
559 	struct cptvf_wqe_info *nwqe_info;
560 
561 	if (unlikely(qno >= cptvf->nr_queues))
562 		return NULL;
563 	nwqe_info = (struct cptvf_wqe_info *)cptvf->wqe_info;
564 
565 	return &nwqe_info->vq_wqe[qno];
566 }
567 
568 static inline u32 cptvf_read_vq_done_count(struct cpt_vf *cptvf)
569 {
570 	union cptx_vqx_done vqx_done;
571 
572 	vqx_done.u = cpt_read_csr64(cptvf->reg_base, CPTX_VQX_DONE(0, 0));
573 	return vqx_done.s.done;
574 }
575 
576 static inline void cptvf_write_vq_done_ack(struct cpt_vf *cptvf,
577 					   u32 ackcnt)
578 {
579 	union cptx_vqx_done_ack vqx_dack_cnt;
580 
581 	vqx_dack_cnt.u = cpt_read_csr64(cptvf->reg_base,
582 					CPTX_VQX_DONE_ACK(0, 0));
583 	vqx_dack_cnt.s.done_ack = ackcnt;
584 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_DONE_ACK(0, 0),
585 			vqx_dack_cnt.u);
586 }
587 
588 static irqreturn_t cptvf_done_intr_handler(int irq, void *cptvf_irq)
589 {
590 	struct cpt_vf *cptvf = (struct cpt_vf *)cptvf_irq;
591 	struct pci_dev *pdev = cptvf->pdev;
592 	/* Read the number of completions */
593 	u32 intr = cptvf_read_vq_done_count(cptvf);
594 
595 	if (intr) {
596 		struct cptvf_wqe *wqe;
597 
598 		/* Acknowledge the number of
599 		 * scheduled completions for processing
600 		 */
601 		cptvf_write_vq_done_ack(cptvf, intr);
602 		wqe = get_cptvf_vq_wqe(cptvf, 0);
603 		if (unlikely(!wqe)) {
604 			dev_err(&pdev->dev, "No work to schedule for VF (%d)",
605 				cptvf->vfid);
606 			return IRQ_NONE;
607 		}
608 		tasklet_hi_schedule(&wqe->twork);
609 	}
610 
611 	return IRQ_HANDLED;
612 }
613 
614 static void cptvf_set_irq_affinity(struct cpt_vf *cptvf, int vec)
615 {
616 	struct pci_dev *pdev = cptvf->pdev;
617 	int cpu;
618 
619 	if (!zalloc_cpumask_var(&cptvf->affinity_mask[vec],
620 				GFP_KERNEL)) {
621 		dev_err(&pdev->dev, "Allocation failed for affinity_mask for VF %d",
622 			cptvf->vfid);
623 		return;
624 	}
625 
626 	cpu = cptvf->vfid % num_online_cpus();
627 	cpumask_set_cpu(cpumask_local_spread(cpu, cptvf->node),
628 			cptvf->affinity_mask[vec]);
629 	irq_set_affinity_hint(pci_irq_vector(pdev, vec),
630 			cptvf->affinity_mask[vec]);
631 }
632 
633 static void cptvf_write_vq_saddr(struct cpt_vf *cptvf, u64 val)
634 {
635 	union cptx_vqx_saddr vqx_saddr;
636 
637 	vqx_saddr.u = val;
638 	cpt_write_csr64(cptvf->reg_base, CPTX_VQX_SADDR(0, 0), vqx_saddr.u);
639 }
640 
641 static void cptvf_device_init(struct cpt_vf *cptvf)
642 {
643 	u64 base_addr = 0;
644 
645 	/* Disable the VQ */
646 	cptvf_write_vq_ctl(cptvf, 0);
647 	/* Reset the doorbell */
648 	cptvf_write_vq_doorbell(cptvf, 0);
649 	/* Clear inflight */
650 	cptvf_write_vq_inprog(cptvf, 0);
651 	/* Write VQ SADDR */
652 	/* TODO: for now only one queue, so hard coded */
653 	base_addr = (u64)(cptvf->cqinfo.queue[0].qhead->dma_addr);
654 	cptvf_write_vq_saddr(cptvf, base_addr);
655 	/* Configure timerhold / coalescence */
656 	cptvf_write_vq_done_timewait(cptvf, CPT_TIMER_THOLD);
657 	cptvf_write_vq_done_numwait(cptvf, 1);
658 	/* Enable the VQ */
659 	cptvf_write_vq_ctl(cptvf, 1);
660 	/* Flag the VF ready */
661 	cptvf->flags |= CPT_FLAG_DEVICE_READY;
662 }
663 
664 static int cptvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
665 {
666 	struct device *dev = &pdev->dev;
667 	struct cpt_vf *cptvf;
668 	int    err;
669 
670 	cptvf = devm_kzalloc(dev, sizeof(*cptvf), GFP_KERNEL);
671 	if (!cptvf)
672 		return -ENOMEM;
673 
674 	pci_set_drvdata(pdev, cptvf);
675 	cptvf->pdev = pdev;
676 	err = pci_enable_device(pdev);
677 	if (err) {
678 		dev_err(dev, "Failed to enable PCI device\n");
679 		pci_set_drvdata(pdev, NULL);
680 		return err;
681 	}
682 
683 	err = pci_request_regions(pdev, DRV_NAME);
684 	if (err) {
685 		dev_err(dev, "PCI request regions failed 0x%x\n", err);
686 		goto cptvf_err_disable_device;
687 	}
688 	/* Mark as VF driver */
689 	cptvf->flags |= CPT_FLAG_VF_DRIVER;
690 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(48));
691 	if (err) {
692 		dev_err(dev, "Unable to get usable DMA configuration\n");
693 		goto cptvf_err_release_regions;
694 	}
695 
696 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(48));
697 	if (err) {
698 		dev_err(dev, "Unable to get 48-bit DMA for consistent allocations\n");
699 		goto cptvf_err_release_regions;
700 	}
701 
702 	/* MAP PF's configuration registers */
703 	cptvf->reg_base = pcim_iomap(pdev, 0, 0);
704 	if (!cptvf->reg_base) {
705 		dev_err(dev, "Cannot map config register space, aborting\n");
706 		err = -ENOMEM;
707 		goto cptvf_err_release_regions;
708 	}
709 
710 	cptvf->node = dev_to_node(&pdev->dev);
711 	err = pci_alloc_irq_vectors(pdev, CPT_VF_MSIX_VECTORS,
712 			CPT_VF_MSIX_VECTORS, PCI_IRQ_MSIX);
713 	if (err < 0) {
714 		dev_err(dev, "Request for #%d msix vectors failed\n",
715 			CPT_VF_MSIX_VECTORS);
716 		goto cptvf_err_release_regions;
717 	}
718 
719 	err = request_irq(pci_irq_vector(pdev, CPT_VF_INT_VEC_E_MISC),
720 			  cptvf_misc_intr_handler, 0, "CPT VF misc intr",
721 			  cptvf);
722 	if (err) {
723 		dev_err(dev, "Request misc irq failed");
724 		goto cptvf_free_vectors;
725 	}
726 
727 	/* Enable mailbox interrupt */
728 	cptvf_enable_mbox_interrupts(cptvf);
729 	cptvf_enable_swerr_interrupts(cptvf);
730 
731 	/* Check ready with PF */
732 	/* Gets chip ID / device Id from PF if ready */
733 	err = cptvf_check_pf_ready(cptvf);
734 	if (err) {
735 		dev_err(dev, "PF not responding to READY msg");
736 		goto cptvf_free_misc_irq;
737 	}
738 
739 	/* CPT VF software resources initialization */
740 	cptvf->cqinfo.qchunksize = CPT_CMD_QCHUNK_SIZE;
741 	err = cptvf_sw_init(cptvf, CPT_CMD_QLEN, CPT_NUM_QS_PER_VF);
742 	if (err) {
743 		dev_err(dev, "cptvf_sw_init() failed");
744 		goto cptvf_free_misc_irq;
745 	}
746 	/* Convey VQ LEN to PF */
747 	err = cptvf_send_vq_size_msg(cptvf);
748 	if (err) {
749 		dev_err(dev, "PF not responding to QLEN msg");
750 		goto cptvf_free_misc_irq;
751 	}
752 
753 	/* CPT VF device initialization */
754 	cptvf_device_init(cptvf);
755 	/* Send msg to PF to assign currnet Q to required group */
756 	cptvf->vfgrp = 1;
757 	err = cptvf_send_vf_to_grp_msg(cptvf);
758 	if (err) {
759 		dev_err(dev, "PF not responding to VF_GRP msg");
760 		goto cptvf_free_misc_irq;
761 	}
762 
763 	cptvf->priority = 1;
764 	err = cptvf_send_vf_priority_msg(cptvf);
765 	if (err) {
766 		dev_err(dev, "PF not responding to VF_PRIO msg");
767 		goto cptvf_free_misc_irq;
768 	}
769 
770 	err = request_irq(pci_irq_vector(pdev, CPT_VF_INT_VEC_E_DONE),
771 			  cptvf_done_intr_handler, 0, "CPT VF done intr",
772 			  cptvf);
773 	if (err) {
774 		dev_err(dev, "Request done irq failed\n");
775 		goto cptvf_free_misc_irq;
776 	}
777 
778 	/* Enable mailbox interrupt */
779 	cptvf_enable_done_interrupts(cptvf);
780 
781 	/* Set irq affinity masks */
782 	cptvf_set_irq_affinity(cptvf, CPT_VF_INT_VEC_E_MISC);
783 	cptvf_set_irq_affinity(cptvf, CPT_VF_INT_VEC_E_DONE);
784 
785 	err = cptvf_send_vf_up(cptvf);
786 	if (err) {
787 		dev_err(dev, "PF not responding to UP msg");
788 		goto cptvf_free_irq_affinity;
789 	}
790 	err = cvm_crypto_init(cptvf);
791 	if (err) {
792 		dev_err(dev, "Algorithm register failed\n");
793 		goto cptvf_free_irq_affinity;
794 	}
795 	return 0;
796 
797 cptvf_free_irq_affinity:
798 	cptvf_free_irq_affinity(cptvf, CPT_VF_INT_VEC_E_DONE);
799 	cptvf_free_irq_affinity(cptvf, CPT_VF_INT_VEC_E_MISC);
800 cptvf_free_misc_irq:
801 	free_irq(pci_irq_vector(pdev, CPT_VF_INT_VEC_E_MISC), cptvf);
802 cptvf_free_vectors:
803 	pci_free_irq_vectors(cptvf->pdev);
804 cptvf_err_release_regions:
805 	pci_release_regions(pdev);
806 cptvf_err_disable_device:
807 	pci_disable_device(pdev);
808 	pci_set_drvdata(pdev, NULL);
809 
810 	return err;
811 }
812 
813 static void cptvf_remove(struct pci_dev *pdev)
814 {
815 	struct cpt_vf *cptvf = pci_get_drvdata(pdev);
816 
817 	if (!cptvf) {
818 		dev_err(&pdev->dev, "Invalid CPT-VF device\n");
819 		return;
820 	}
821 
822 	/* Convey DOWN to PF */
823 	if (cptvf_send_vf_down(cptvf)) {
824 		dev_err(&pdev->dev, "PF not responding to DOWN msg");
825 	} else {
826 		cptvf_free_irq_affinity(cptvf, CPT_VF_INT_VEC_E_DONE);
827 		cptvf_free_irq_affinity(cptvf, CPT_VF_INT_VEC_E_MISC);
828 		free_irq(pci_irq_vector(pdev, CPT_VF_INT_VEC_E_DONE), cptvf);
829 		free_irq(pci_irq_vector(pdev, CPT_VF_INT_VEC_E_MISC), cptvf);
830 		pci_free_irq_vectors(cptvf->pdev);
831 		cptvf_sw_cleanup(cptvf);
832 		pci_set_drvdata(pdev, NULL);
833 		pci_release_regions(pdev);
834 		pci_disable_device(pdev);
835 		cvm_crypto_exit();
836 	}
837 }
838 
839 static void cptvf_shutdown(struct pci_dev *pdev)
840 {
841 	cptvf_remove(pdev);
842 }
843 
844 /* Supported devices */
845 static const struct pci_device_id cptvf_id_table[] = {
846 	{PCI_VDEVICE(CAVIUM, CPT_81XX_PCI_VF_DEVICE_ID), 0},
847 	{ 0, }  /* end of table */
848 };
849 
850 static struct pci_driver cptvf_pci_driver = {
851 	.name = DRV_NAME,
852 	.id_table = cptvf_id_table,
853 	.probe = cptvf_probe,
854 	.remove = cptvf_remove,
855 	.shutdown = cptvf_shutdown,
856 };
857 
858 module_pci_driver(cptvf_pci_driver);
859 
860 MODULE_AUTHOR("George Cherian <george.cherian@cavium.com>");
861 MODULE_DESCRIPTION("Cavium Thunder CPT Virtual Function Driver");
862 MODULE_LICENSE("GPL v2");
863 MODULE_VERSION(DRV_VERSION);
864 MODULE_DEVICE_TABLE(pci, cptvf_id_table);
865