Free Electrons

Embedded Linux Experts

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
/*
 * CAAM/SEC 4.x QI transport/backend driver
 * Queue Interface backend functionality
 *
 * Copyright 2013-2016 Freescale Semiconductor, Inc.
 * Copyright 2016-2017 NXP
 */

#include <linux/cpumask.h>
#include <linux/kthread.h>
#include <soc/fsl/qman.h>

#include "regs.h"
#include "qi.h"
#include "desc.h"
#include "intern.h"
#include "desc_constr.h"

#define PREHDR_RSLS_SHIFT	31

/*
 * Use a reasonable backlog of frames (per CPU) as congestion threshold,
 * so that resources used by the in-flight buffers do not become a memory hog.
 */
#define MAX_RSP_FQ_BACKLOG_PER_CPU	256

/* Length of a single buffer in the QI driver memory cache */
#define CAAM_QI_MEMCACHE_SIZE	512

#define CAAM_QI_ENQUEUE_RETRIES	10000

#define CAAM_NAPI_WEIGHT	63

/*
 * caam_napi - struct holding CAAM NAPI-related params
 * @irqtask: IRQ task for QI backend
 * @p: QMan portal
 */
struct caam_napi {
	struct napi_struct irqtask;
	struct qman_portal *p;
};

/*
 * caam_qi_pcpu_priv - percpu private data structure to main list of pending
 *                     responses expected on each cpu.
 * @caam_napi: CAAM NAPI params
 * @net_dev: netdev used by NAPI
 * @rsp_fq: response FQ from CAAM
 */
struct caam_qi_pcpu_priv {
	struct caam_napi caam_napi;
	struct net_device net_dev;
	struct qman_fq *rsp_fq;
} ____cacheline_aligned;

static DEFINE_PER_CPU(struct caam_qi_pcpu_priv, pcpu_qipriv);
static DEFINE_PER_CPU(int, last_cpu);

/*
 * caam_qi_priv - CAAM QI backend private params
 * @cgr: QMan congestion group
 * @qi_pdev: platform device for QI backend
 */
struct caam_qi_priv {
	struct qman_cgr cgr;
	struct platform_device *qi_pdev;
};

static struct caam_qi_priv qipriv ____cacheline_aligned;

/*
 * This is written by only one core - the one that initialized the CGR - and
 * read by multiple cores (all the others).
 */
bool caam_congested __read_mostly;
EXPORT_SYMBOL(caam_congested);

#ifdef CONFIG_DEBUG_FS
/*
 * This is a counter for the number of times the congestion group (where all
 * the request and response queueus are) reached congestion. Incremented
 * each time the congestion callback is called with congested == true.
 */
static u64 times_congested;
#endif

/*
 * CPU from where the module initialised. This is required because QMan driver
 * requires CGRs to be removed from same CPU from where they were originally
 * allocated.
 */
static int mod_init_cpu;

/*
 * This is a a cache of buffers, from which the users of CAAM QI driver
 * can allocate short (CAAM_QI_MEMCACHE_SIZE) buffers. It's faster than
 * doing malloc on the hotpath.
 * NOTE: A more elegant solution would be to have some headroom in the frames
 *       being processed. This could be added by the dpaa-ethernet driver.
 *       This would pose a problem for userspace application processing which
 *       cannot know of this limitation. So for now, this will work.
 * NOTE: The memcache is SMP-safe. No need to handle spinlocks in-here
 */
static struct kmem_cache *qi_cache;

int caam_qi_enqueue(struct device *qidev, struct caam_drv_req *req)
{
	struct qm_fd fd;
	dma_addr_t addr;
	int ret;
	int num_retries = 0;

	qm_fd_clear_fd(&fd);
	qm_fd_set_compound(&fd, qm_sg_entry_get_len(&req->fd_sgt[1]));

	addr = dma_map_single(qidev, req->fd_sgt, sizeof(req->fd_sgt),
			      DMA_BIDIRECTIONAL);
	if (dma_mapping_error(qidev, addr)) {
		dev_err(qidev, "DMA mapping error for QI enqueue request\n");
		return -EIO;
	}
	qm_fd_addr_set64(&fd, addr);

	do {
		ret = qman_enqueue(req->drv_ctx->req_fq, &fd);
		if (likely(!ret))
			return 0;

		if (ret != -EBUSY)
			break;
		num_retries++;
	} while (num_retries < CAAM_QI_ENQUEUE_RETRIES);

	dev_err(qidev, "qman_enqueue failed: %d\n", ret);

	return ret;
}
EXPORT_SYMBOL(caam_qi_enqueue);

static void caam_fq_ern_cb(struct qman_portal *qm, struct qman_fq *fq,
			   const union qm_mr_entry *msg)
{
	const struct qm_fd *fd;
	struct caam_drv_req *drv_req;
	struct device *qidev = &(raw_cpu_ptr(&pcpu_qipriv)->net_dev.dev);

	fd = &msg->ern.fd;

	if (qm_fd_get_format(fd) != qm_fd_compound) {
		dev_err(qidev, "Non-compound FD from CAAM\n");
		return;
	}

	drv_req = (struct caam_drv_req *)phys_to_virt(qm_fd_addr_get64(fd));
	if (!drv_req) {
		dev_err(qidev,
			"Can't find original request for CAAM response\n");
		return;
	}

	dma_unmap_single(drv_req->drv_ctx->qidev, qm_fd_addr(fd),
			 sizeof(drv_req->fd_sgt), DMA_BIDIRECTIONAL);

	drv_req->cbk(drv_req, -EIO);
}

static struct qman_fq *create_caam_req_fq(struct device *qidev,
					  struct qman_fq *rsp_fq,
					  dma_addr_t hwdesc,
					  int fq_sched_flag)
{
	int ret;
	struct qman_fq *req_fq;
	struct qm_mcc_initfq opts;

	req_fq = kzalloc(sizeof(*req_fq), GFP_ATOMIC);
	if (!req_fq)
		return ERR_PTR(-ENOMEM);

	req_fq->cb.ern = caam_fq_ern_cb;
	req_fq->cb.fqs = NULL;

	ret = qman_create_fq(0, QMAN_FQ_FLAG_DYNAMIC_FQID |
				QMAN_FQ_FLAG_TO_DCPORTAL, req_fq);
	if (ret) {
		dev_err(qidev, "Failed to create session req FQ\n");
		goto create_req_fq_fail;
	}

	memset(&opts, 0, sizeof(opts));
	opts.we_mask = cpu_to_be16(QM_INITFQ_WE_FQCTRL | QM_INITFQ_WE_DESTWQ |
				   QM_INITFQ_WE_CONTEXTB |
				   QM_INITFQ_WE_CONTEXTA | QM_INITFQ_WE_CGID);
	opts.fqd.fq_ctrl = cpu_to_be16(QM_FQCTRL_CPCSTASH | QM_FQCTRL_CGE);
	qm_fqd_set_destwq(&opts.fqd, qm_channel_caam, 2);
	opts.fqd.context_b = cpu_to_be32(qman_fq_fqid(rsp_fq));
	qm_fqd_context_a_set64(&opts.fqd, hwdesc);
	opts.fqd.cgid = qipriv.cgr.cgrid;

	ret = qman_init_fq(req_fq, fq_sched_flag, &opts);
	if (ret) {
		dev_err(qidev, "Failed to init session req FQ\n");
		goto init_req_fq_fail;
	}

	dev_info(qidev, "Allocated request FQ %u for CPU %u\n", req_fq->fqid,
		 smp_processor_id());
	return req_fq;

init_req_fq_fail:
	qman_destroy_fq(req_fq);
create_req_fq_fail:
	kfree(req_fq);
	return ERR_PTR(ret);
}

static int empty_retired_fq(struct device *qidev, struct qman_fq *fq)
{
	int ret;

	ret = qman_volatile_dequeue(fq, QMAN_VOLATILE_FLAG_WAIT_INT |
				    QMAN_VOLATILE_FLAG_FINISH,
				    QM_VDQCR_PRECEDENCE_VDQCR |
				    QM_VDQCR_NUMFRAMES_TILLEMPTY);
	if (ret) {
		dev_err(qidev, "Volatile dequeue fail for FQ: %u\n", fq->fqid);
		return ret;
	}

	do {
		struct qman_portal *p;

		p = qman_get_affine_portal(smp_processor_id());
		qman_p_poll_dqrr(p, 16);
	} while (fq->flags & QMAN_FQ_STATE_NE);

	return 0;
}

static int kill_fq(struct device *qidev, struct qman_fq *fq)
{
	u32 flags;
	int ret;

	ret = qman_retire_fq(fq, &flags);
	if (ret < 0) {
		dev_err(qidev, "qman_retire_fq failed: %d\n", ret);
		return ret;
	}

	if (!ret)
		goto empty_fq;

	/* Async FQ retirement condition */
	if (ret == 1) {
		/* Retry till FQ gets in retired state */
		do {
			msleep(20);
		} while (fq->state != qman_fq_state_retired);

		WARN_ON(fq->flags & QMAN_FQ_STATE_BLOCKOOS);
		WARN_ON(fq->flags & QMAN_FQ_STATE_ORL);
	}

empty_fq:
	if (fq->flags & QMAN_FQ_STATE_NE) {
		ret = empty_retired_fq(qidev, fq);
		if (ret) {
			dev_err(qidev, "empty_retired_fq fail for FQ: %u\n",
				fq->fqid);
			return ret;
		}
	}

	ret = qman_oos_fq(fq);
	if (ret)
		dev_err(qidev, "OOS of FQID: %u failed\n", fq->fqid);

	qman_destroy_fq(fq);

	return ret;
}

static int empty_caam_fq(struct qman_fq *fq)
{
	int ret;
	struct qm_mcr_queryfq_np np;

	/* Wait till the older CAAM FQ get empty */
	do {
		ret = qman_query_fq_np(fq, &np);
		if (ret)
			return ret;

		if (!qm_mcr_np_get(&np, frm_cnt))
			break;

		msleep(20);
	} while (1);

	/*
	 * Give extra time for pending jobs from this FQ in holding tanks
	 * to get processed
	 */
	msleep(20);
	return 0;
}

int caam_drv_ctx_update(struct caam_drv_ctx *drv_ctx, u32 *sh_desc)
{
	int ret;
	u32 num_words;
	struct qman_fq *new_fq, *old_fq;
	struct device *qidev = drv_ctx->qidev;

	num_words = desc_len(sh_desc);
	if (num_words > MAX_SDLEN) {
		dev_err(qidev, "Invalid descriptor len: %d words\n", num_words);
		return -EINVAL;
	}

	/* Note down older req FQ */
	old_fq = drv_ctx->req_fq;

	/* Create a new req FQ in parked state */
	new_fq = create_caam_req_fq(drv_ctx->qidev, drv_ctx->rsp_fq,
				    drv_ctx->context_a, 0);
	if (unlikely(IS_ERR_OR_NULL(new_fq))) {
		dev_err(qidev, "FQ allocation for shdesc update failed\n");
		return PTR_ERR(new_fq);
	}

	/* Hook up new FQ to context so that new requests keep queuing */
	drv_ctx->req_fq = new_fq;

	/* Empty and remove the older FQ */
	ret = empty_caam_fq(old_fq);
	if (ret) {
		dev_err(qidev, "Old CAAM FQ empty failed: %d\n", ret);

		/* We can revert to older FQ */
		drv_ctx->req_fq = old_fq;

		if (kill_fq(qidev, new_fq))
			dev_warn(qidev, "New CAAM FQ: %u kill failed\n",
				 new_fq->fqid);

		return ret;
	}

	/*
	 * Re-initialise pre-header. Set RSLS and SDLEN.
	 * Update the shared descriptor for driver context.
	 */
	drv_ctx->prehdr[0] = cpu_to_caam32((1 << PREHDR_RSLS_SHIFT) |
					   num_words);
	memcpy(drv_ctx->sh_desc, sh_desc, desc_bytes(sh_desc));
	dma_sync_single_for_device(qidev, drv_ctx->context_a,
				   sizeof(drv_ctx->sh_desc) +
				   sizeof(drv_ctx->prehdr),
				   DMA_BIDIRECTIONAL);

	/* Put the new FQ in scheduled state */
	ret = qman_schedule_fq(new_fq);
	if (ret) {
		dev_err(qidev, "Fail to sched new CAAM FQ, ecode = %d\n", ret);

		/*
		 * We can kill new FQ and revert to old FQ.
		 * Since the desc is already modified, it is success case
		 */

		drv_ctx->req_fq = old_fq;

		if (kill_fq(qidev, new_fq))
			dev_warn(qidev, "New CAAM FQ: %u kill failed\n",
				 new_fq->fqid);
	} else if (kill_fq(qidev, old_fq)) {
		dev_warn(qidev, "Old CAAM FQ: %u kill failed\n", old_fq->fqid);
	}

	return 0;
}
EXPORT_SYMBOL(caam_drv_ctx_update);

struct caam_drv_ctx *caam_drv_ctx_init(struct device *qidev,
				       int *cpu,
				       u32 *sh_desc)
{
	size_t size;
	u32 num_words;
	dma_addr_t hwdesc;
	struct caam_drv_ctx *drv_ctx;
	const cpumask_t *cpus = qman_affine_cpus();

	num_words = desc_len(sh_desc);
	if (num_words > MAX_SDLEN) {
		dev_err(qidev, "Invalid descriptor len: %d words\n",
			num_words);
		return ERR_PTR(-EINVAL);
	}

	drv_ctx = kzalloc(sizeof(*drv_ctx), GFP_ATOMIC);
	if (!drv_ctx)
		return ERR_PTR(-ENOMEM);

	/*
	 * Initialise pre-header - set RSLS and SDLEN - and shared descriptor
	 * and dma-map them.
	 */
	drv_ctx->prehdr[0] = cpu_to_caam32((1 << PREHDR_RSLS_SHIFT) |
					   num_words);
	memcpy(drv_ctx->sh_desc, sh_desc, desc_bytes(sh_desc));
	size = sizeof(drv_ctx->prehdr) + sizeof(drv_ctx->sh_desc);
	hwdesc = dma_map_single(qidev, drv_ctx->prehdr, size,
				DMA_BIDIRECTIONAL);
	if (dma_mapping_error(qidev, hwdesc)) {
		dev_err(qidev, "DMA map error for preheader + shdesc\n");
		kfree(drv_ctx);
		return ERR_PTR(-ENOMEM);
	}
	drv_ctx->context_a = hwdesc;

	/* If given CPU does not own the portal, choose another one that does */
	if (!cpumask_test_cpu(*cpu, cpus)) {
		int *pcpu = &get_cpu_var(last_cpu);

		*pcpu = cpumask_next(*pcpu, cpus);
		if (*pcpu >= nr_cpu_ids)
			*pcpu = cpumask_first(cpus);
		*cpu = *pcpu;

		put_cpu_var(last_cpu);
	}
	drv_ctx->cpu = *cpu;

	/* Find response FQ hooked with this CPU */
	drv_ctx->rsp_fq = per_cpu(pcpu_qipriv.rsp_fq, drv_ctx->cpu);

	/* Attach request FQ */
	drv_ctx->req_fq = create_caam_req_fq(qidev, drv_ctx->rsp_fq, hwdesc,
					     QMAN_INITFQ_FLAG_SCHED);
	if (unlikely(IS_ERR_OR_NULL(drv_ctx->req_fq))) {
		dev_err(qidev, "create_caam_req_fq failed\n");
		dma_unmap_single(qidev, hwdesc, size, DMA_BIDIRECTIONAL);
		kfree(drv_ctx);
		return ERR_PTR(-ENOMEM);
	}

	drv_ctx->qidev = qidev;
	return drv_ctx;
}
EXPORT_SYMBOL(caam_drv_ctx_init);

void *qi_cache_alloc(gfp_t flags)
{
	return kmem_cache_alloc(qi_cache, flags);
}
EXPORT_SYMBOL(qi_cache_alloc);

void qi_cache_free(void *obj)
{
	kmem_cache_free(qi_cache, obj);
}
EXPORT_SYMBOL(qi_cache_free);

static int caam_qi_poll(struct napi_struct *napi, int budget)
{
	struct caam_napi *np = container_of(napi, struct caam_napi, irqtask);

	int cleaned = qman_p_poll_dqrr(np->p, budget);

	if (cleaned < budget) {
		napi_complete(napi);
		qman_p_irqsource_add(np->p, QM_PIRQ_DQRI);
	}

	return cleaned;
}

void caam_drv_ctx_rel(struct caam_drv_ctx *drv_ctx)
{
	if (IS_ERR_OR_NULL(drv_ctx))
		return;

	/* Remove request FQ */
	if (kill_fq(drv_ctx->qidev, drv_ctx->req_fq))
		dev_err(drv_ctx->qidev, "Crypto session req FQ kill failed\n");

	dma_unmap_single(drv_ctx->qidev, drv_ctx->context_a,
			 sizeof(drv_ctx->sh_desc) + sizeof(drv_ctx->prehdr),
			 DMA_BIDIRECTIONAL);
	kfree(drv_ctx);
}
EXPORT_SYMBOL(caam_drv_ctx_rel);

int caam_qi_shutdown(struct device *qidev)
{
	int i, ret;
	struct caam_qi_priv *priv = dev_get_drvdata(qidev);
	const cpumask_t *cpus = qman_affine_cpus();
	struct cpumask old_cpumask = current->cpus_allowed;

	for_each_cpu(i, cpus) {
		struct napi_struct *irqtask;

		irqtask = &per_cpu_ptr(&pcpu_qipriv.caam_napi, i)->irqtask;
		napi_disable(irqtask);
		netif_napi_del(irqtask);

		if (kill_fq(qidev, per_cpu(pcpu_qipriv.rsp_fq, i)))
			dev_err(qidev, "Rsp FQ kill failed, cpu: %d\n", i);
		kfree(per_cpu(pcpu_qipriv.rsp_fq, i));
	}

	/*
	 * QMan driver requires CGRs to be deleted from same CPU from where they
	 * were instantiated. Hence we get the module removal execute from the
	 * same CPU from where it was originally inserted.
	 */
	set_cpus_allowed_ptr(current, get_cpu_mask(mod_init_cpu));

	ret = qman_delete_cgr(&priv->cgr);
	if (ret)
		dev_err(qidev, "Deletion of CGR failed: %d\n", ret);
	else
		qman_release_cgrid(priv->cgr.cgrid);

	kmem_cache_destroy(qi_cache);

	/* Now that we're done with the CGRs, restore the cpus allowed mask */
	set_cpus_allowed_ptr(current, &old_cpumask);

	platform_device_unregister(priv->qi_pdev);
	return ret;
}

static void cgr_cb(struct qman_portal *qm, struct qman_cgr *cgr, int congested)
{
	caam_congested = congested;

	if (congested) {
#ifdef CONFIG_DEBUG_FS
		times_congested++;
#endif
		pr_debug_ratelimited("CAAM entered congestion\n");

	} else {
		pr_debug_ratelimited("CAAM exited congestion\n");
	}
}

static int caam_qi_napi_schedule(struct qman_portal *p, struct caam_napi *np)
{
	/*
	 * In case of threaded ISR, for RT kernels in_irq() does not return
	 * appropriate value, so use in_serving_softirq to distinguish between
	 * softirq and irq contexts.
	 */
	if (unlikely(in_irq() || !in_serving_softirq())) {
		/* Disable QMan IRQ source and invoke NAPI */
		qman_p_irqsource_remove(p, QM_PIRQ_DQRI);
		np->p = p;
		napi_schedule(&np->irqtask);
		return 1;
	}
	return 0;
}

static enum qman_cb_dqrr_result caam_rsp_fq_dqrr_cb(struct qman_portal *p,
						    struct qman_fq *rsp_fq,
						    const struct qm_dqrr_entry *dqrr)
{
	struct caam_napi *caam_napi = raw_cpu_ptr(&pcpu_qipriv.caam_napi);
	struct caam_drv_req *drv_req;
	const struct qm_fd *fd;
	struct device *qidev = &(raw_cpu_ptr(&pcpu_qipriv)->net_dev.dev);
	u32 status;

	if (caam_qi_napi_schedule(p, caam_napi))
		return qman_cb_dqrr_stop;

	fd = &dqrr->fd;
	status = be32_to_cpu(fd->status);
	if (unlikely(status))
		dev_err(qidev, "Error: %#x in CAAM response FD\n", status);

	if (unlikely(qm_fd_get_format(fd) != qm_fd_compound)) {
		dev_err(qidev, "Non-compound FD from CAAM\n");
		return qman_cb_dqrr_consume;
	}

	drv_req = (struct caam_drv_req *)phys_to_virt(qm_fd_addr_get64(fd));
	if (unlikely(!drv_req)) {
		dev_err(qidev,
			"Can't find original request for caam response\n");
		return qman_cb_dqrr_consume;
	}

	dma_unmap_single(drv_req->drv_ctx->qidev, qm_fd_addr(fd),
			 sizeof(drv_req->fd_sgt), DMA_BIDIRECTIONAL);

	drv_req->cbk(drv_req, status);
	return qman_cb_dqrr_consume;
}

static int alloc_rsp_fq_cpu(struct device *qidev, unsigned int cpu)
{
	struct qm_mcc_initfq opts;
	struct qman_fq *fq;
	int ret;

	fq = kzalloc(sizeof(*fq), GFP_KERNEL | GFP_DMA);
	if (!fq)
		return -ENOMEM;

	fq->cb.dqrr = caam_rsp_fq_dqrr_cb;

	ret = qman_create_fq(0, QMAN_FQ_FLAG_NO_ENQUEUE |
			     QMAN_FQ_FLAG_DYNAMIC_FQID, fq);
	if (ret) {
		dev_err(qidev, "Rsp FQ create failed\n");
		kfree(fq);
		return -ENODEV;
	}

	memset(&opts, 0, sizeof(opts));
	opts.we_mask = cpu_to_be16(QM_INITFQ_WE_FQCTRL | QM_INITFQ_WE_DESTWQ |
				   QM_INITFQ_WE_CONTEXTB |
				   QM_INITFQ_WE_CONTEXTA | QM_INITFQ_WE_CGID);
	opts.fqd.fq_ctrl = cpu_to_be16(QM_FQCTRL_CTXASTASHING |
				       QM_FQCTRL_CPCSTASH | QM_FQCTRL_CGE);
	qm_fqd_set_destwq(&opts.fqd, qman_affine_channel(cpu), 3);
	opts.fqd.cgid = qipriv.cgr.cgrid;
	opts.fqd.context_a.stashing.exclusive =	QM_STASHING_EXCL_CTX |
						QM_STASHING_EXCL_DATA;
	qm_fqd_set_stashing(&opts.fqd, 0, 1, 1);

	ret = qman_init_fq(fq, QMAN_INITFQ_FLAG_SCHED, &opts);
	if (ret) {
		dev_err(qidev, "Rsp FQ init failed\n");
		kfree(fq);
		return -ENODEV;
	}

	per_cpu(pcpu_qipriv.rsp_fq, cpu) = fq;

	dev_info(qidev, "Allocated response FQ %u for CPU %u", fq->fqid, cpu);
	return 0;
}

static int init_cgr(struct device *qidev)
{
	int ret;
	struct qm_mcc_initcgr opts;
	const u64 cpus = *(u64 *)qman_affine_cpus();
	const int num_cpus = hweight64(cpus);
	const u64 val = num_cpus * MAX_RSP_FQ_BACKLOG_PER_CPU;

	ret = qman_alloc_cgrid(&qipriv.cgr.cgrid);
	if (ret) {
		dev_err(qidev, "CGR alloc failed for rsp FQs: %d\n", ret);
		return ret;
	}

	qipriv.cgr.cb = cgr_cb;
	memset(&opts, 0, sizeof(opts));
	opts.we_mask = cpu_to_be16(QM_CGR_WE_CSCN_EN | QM_CGR_WE_CS_THRES |
				   QM_CGR_WE_MODE);
	opts.cgr.cscn_en = QM_CGR_EN;
	opts.cgr.mode = QMAN_CGR_MODE_FRAME;
	qm_cgr_cs_thres_set64(&opts.cgr.cs_thres, val, 1);

	ret = qman_create_cgr(&qipriv.cgr, QMAN_CGR_FLAG_USE_INIT, &opts);
	if (ret) {
		dev_err(qidev, "Error %d creating CAAM CGRID: %u\n", ret,
			qipriv.cgr.cgrid);
		return ret;
	}

	dev_info(qidev, "Congestion threshold set to %llu\n", val);
	return 0;
}

static int alloc_rsp_fqs(struct device *qidev)
{
	int ret, i;
	const cpumask_t *cpus = qman_affine_cpus();

	/*Now create response FQs*/
	for_each_cpu(i, cpus) {
		ret = alloc_rsp_fq_cpu(qidev, i);
		if (ret) {
			dev_err(qidev, "CAAM rsp FQ alloc failed, cpu: %u", i);
			return ret;
		}
	}

	return 0;
}

static void free_rsp_fqs(void)
{
	int i;
	const cpumask_t *cpus = qman_affine_cpus();

	for_each_cpu(i, cpus)
		kfree(per_cpu(pcpu_qipriv.rsp_fq, i));
}

int caam_qi_init(struct platform_device *caam_pdev)
{
	int err, i;
	struct platform_device *qi_pdev;
	struct device *ctrldev = &caam_pdev->dev, *qidev;
	struct caam_drv_private *ctrlpriv;
	const cpumask_t *cpus = qman_affine_cpus();
	struct cpumask old_cpumask = current->cpus_allowed;
	static struct platform_device_info qi_pdev_info = {
		.name = "caam_qi",
		.id = PLATFORM_DEVID_NONE
	};

	/*
	 * QMAN requires CGRs to be removed from same CPU+portal from where it
	 * was originally allocated. Hence we need to note down the
	 * initialisation CPU and use the same CPU for module exit.
	 * We select the first CPU to from the list of portal owning CPUs.
	 * Then we pin module init to this CPU.
	 */
	mod_init_cpu = cpumask_first(cpus);
	set_cpus_allowed_ptr(current, get_cpu_mask(mod_init_cpu));

	qi_pdev_info.parent = ctrldev;
	qi_pdev_info.dma_mask = dma_get_mask(ctrldev);
	qi_pdev = platform_device_register_full(&qi_pdev_info);
	if (IS_ERR(qi_pdev))
		return PTR_ERR(qi_pdev);

	ctrlpriv = dev_get_drvdata(ctrldev);
	qidev = &qi_pdev->dev;

	qipriv.qi_pdev = qi_pdev;
	dev_set_drvdata(qidev, &qipriv);

	/* Initialize the congestion detection */
	err = init_cgr(qidev);
	if (err) {
		dev_err(qidev, "CGR initialization failed: %d\n", err);
		platform_device_unregister(qi_pdev);
		return err;
	}

	/* Initialise response FQs */
	err = alloc_rsp_fqs(qidev);
	if (err) {
		dev_err(qidev, "Can't allocate CAAM response FQs: %d\n", err);
		free_rsp_fqs();
		platform_device_unregister(qi_pdev);
		return err;
	}

	/*
	 * Enable the NAPI contexts on each of the core which has an affine
	 * portal.
	 */
	for_each_cpu(i, cpus) {
		struct caam_qi_pcpu_priv *priv = per_cpu_ptr(&pcpu_qipriv, i);
		struct caam_napi *caam_napi = &priv->caam_napi;
		struct napi_struct *irqtask = &caam_napi->irqtask;
		struct net_device *net_dev = &priv->net_dev;

		net_dev->dev = *qidev;
		INIT_LIST_HEAD(&net_dev->napi_list);

		netif_napi_add(net_dev, irqtask, caam_qi_poll,
			       CAAM_NAPI_WEIGHT);

		napi_enable(irqtask);
	}

	/* Hook up QI device to parent controlling caam device */
	ctrlpriv->qidev = qidev;

	qi_cache = kmem_cache_create("caamqicache", CAAM_QI_MEMCACHE_SIZE, 0,
				     SLAB_CACHE_DMA, NULL);
	if (!qi_cache) {
		dev_err(qidev, "Can't allocate CAAM cache\n");
		free_rsp_fqs();
		platform_device_unregister(qi_pdev);
		return -ENOMEM;
	}

	/* Done with the CGRs; restore the cpus allowed mask */
	set_cpus_allowed_ptr(current, &old_cpumask);
#ifdef CONFIG_DEBUG_FS
	ctrlpriv->qi_congested = debugfs_create_file("qi_congested", 0444,
						     ctrlpriv->ctl,
						     &times_congested,
						     &caam_fops_u64_ro);
#endif
	dev_info(qidev, "Linux CAAM Queue I/F driver initialised\n");
	return 0;
}