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
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
/*
 * This file is provided under a dual BSD/GPLv2 license.  When using or
 *   redistributing this file, you may do so under either license.
 *
 *   GPL LICENSE SUMMARY
 *
 *   Copyright(c) 2015 Intel Corporation. All rights reserved.
 *
 *   This program is free software; you can redistribute it and/or modify
 *   it under the terms of version 2 of the GNU General Public License as
 *   published by the Free Software Foundation.
 *
 *   BSD LICENSE
 *
 *   Copyright(c) 2015 Intel Corporation. All rights reserved.
 *
 *   Redistribution and use in source and binary forms, with or without
 *   modification, are permitted provided that the following conditions
 *   are met:
 *
 *     * Redistributions of source code must retain the above copyright
 *       notice, this list of conditions and the following disclaimer.
 *     * Redistributions in binary form must reproduce the above copy
 *       notice, this list of conditions and the following disclaimer in
 *       the documentation and/or other materials provided with the
 *       distribution.
 *     * Neither the name of Intel Corporation nor the names of its
 *       contributors may be used to endorse or promote products derived
 *       from this software without specific prior written permission.
 *
 *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 *   PCIe NTB Perf Linux driver
 */

#include <linux/init.h>
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/kthread.h>
#include <linux/time.h>
#include <linux/timer.h>
#include <linux/dma-mapping.h>
#include <linux/pci.h>
#include <linux/slab.h>
#include <linux/spinlock.h>
#include <linux/debugfs.h>
#include <linux/dmaengine.h>
#include <linux/delay.h>
#include <linux/sizes.h>
#include <linux/ntb.h>
#include <linux/mutex.h>

#define DRIVER_NAME		"ntb_perf"
#define DRIVER_DESCRIPTION	"PCIe NTB Performance Measurement Tool"

#define DRIVER_LICENSE		"Dual BSD/GPL"
#define DRIVER_VERSION		"1.0"
#define DRIVER_AUTHOR		"Dave Jiang <dave.jiang@intel.com>"

#define PERF_LINK_DOWN_TIMEOUT	10
#define PERF_VERSION		0xffff0001
#define MAX_THREADS		32
#define MAX_TEST_SIZE		SZ_1M
#define MAX_SRCS		32
#define DMA_OUT_RESOURCE_TO	msecs_to_jiffies(50)
#define DMA_RETRIES		20
#define SZ_4G			(1ULL << 32)
#define MAX_SEG_ORDER		20 /* no larger than 1M for kmalloc buffer */
#define PIDX			NTB_DEF_PEER_IDX

MODULE_LICENSE(DRIVER_LICENSE);
MODULE_VERSION(DRIVER_VERSION);
MODULE_AUTHOR(DRIVER_AUTHOR);
MODULE_DESCRIPTION(DRIVER_DESCRIPTION);

static struct dentry *perf_debugfs_dir;

static unsigned long max_mw_size;
module_param(max_mw_size, ulong, 0644);
MODULE_PARM_DESC(max_mw_size, "Limit size of large memory windows");

static unsigned int seg_order = 19; /* 512K */
module_param(seg_order, uint, 0644);
MODULE_PARM_DESC(seg_order, "size order [2^n] of buffer segment for testing");

static unsigned int run_order = 32; /* 4G */
module_param(run_order, uint, 0644);
MODULE_PARM_DESC(run_order, "size order [2^n] of total data to transfer");

static bool use_dma; /* default to 0 */
module_param(use_dma, bool, 0644);
MODULE_PARM_DESC(use_dma, "Using DMA engine to measure performance");

static bool on_node = true; /* default to 1 */
module_param(on_node, bool, 0644);
MODULE_PARM_DESC(on_node, "Run threads only on NTB device node (default: true)");

struct perf_mw {
	phys_addr_t	phys_addr;
	resource_size_t	phys_size;
	resource_size_t	xlat_align;
	resource_size_t	xlat_align_size;
	void __iomem	*vbase;
	size_t		xlat_size;
	size_t		buf_size;
	void		*virt_addr;
	dma_addr_t	dma_addr;
};

struct perf_ctx;

struct pthr_ctx {
	struct task_struct	*thread;
	struct perf_ctx		*perf;
	atomic_t		dma_sync;
	struct dma_chan		*dma_chan;
	int			dma_prep_err;
	int			src_idx;
	void			*srcs[MAX_SRCS];
	wait_queue_head_t       *wq;
	int			status;
	u64			copied;
	u64			diff_us;
};

struct perf_ctx {
	struct ntb_dev		*ntb;
	spinlock_t		db_lock;
	struct perf_mw		mw;
	bool			link_is_up;
	struct delayed_work	link_work;
	wait_queue_head_t	link_wq;
	u8			perf_threads;
	/* mutex ensures only one set of threads run at once */
	struct mutex		run_mutex;
	struct pthr_ctx		pthr_ctx[MAX_THREADS];
	atomic_t		tsync;
	atomic_t                tdone;
};

enum {
	VERSION = 0,
	MW_SZ_HIGH,
	MW_SZ_LOW,
	MAX_SPAD
};

static void perf_link_event(void *ctx)
{
	struct perf_ctx *perf = ctx;

	if (ntb_link_is_up(perf->ntb, NULL, NULL) == 1) {
		schedule_delayed_work(&perf->link_work, 2*HZ);
	} else {
		dev_dbg(&perf->ntb->pdev->dev, "link down\n");

		if (!perf->link_is_up)
			cancel_delayed_work_sync(&perf->link_work);

		perf->link_is_up = false;
	}
}

static void perf_db_event(void *ctx, int vec)
{
	struct perf_ctx *perf = ctx;
	u64 db_bits, db_mask;

	db_mask = ntb_db_vector_mask(perf->ntb, vec);
	db_bits = ntb_db_read(perf->ntb);

	dev_dbg(&perf->ntb->dev, "doorbell vec %d mask %#llx bits %#llx\n",
		vec, db_mask, db_bits);
}

static const struct ntb_ctx_ops perf_ops = {
	.link_event = perf_link_event,
	.db_event = perf_db_event,
};

static void perf_copy_callback(void *data)
{
	struct pthr_ctx *pctx = data;

	atomic_dec(&pctx->dma_sync);
}

static ssize_t perf_copy(struct pthr_ctx *pctx, char __iomem *dst,
			 char *src, size_t size)
{
	struct perf_ctx *perf = pctx->perf;
	struct dma_async_tx_descriptor *txd;
	struct dma_chan *chan = pctx->dma_chan;
	struct dma_device *device;
	struct dmaengine_unmap_data *unmap;
	dma_cookie_t cookie;
	size_t src_off, dst_off;
	struct perf_mw *mw = &perf->mw;
	void __iomem *vbase;
	void __iomem *dst_vaddr;
	dma_addr_t dst_phys;
	int retries = 0;

	if (!use_dma) {
		memcpy_toio(dst, src, size);
		return size;
	}

	if (!chan) {
		dev_err(&perf->ntb->dev, "DMA engine does not exist\n");
		return -EINVAL;
	}

	device = chan->device;
	src_off = (uintptr_t)src & ~PAGE_MASK;
	dst_off = (uintptr_t __force)dst & ~PAGE_MASK;

	if (!is_dma_copy_aligned(device, src_off, dst_off, size))
		return -ENODEV;

	vbase = mw->vbase;
	dst_vaddr = dst;
	dst_phys = mw->phys_addr + (dst_vaddr - vbase);

	unmap = dmaengine_get_unmap_data(device->dev, 1, GFP_NOWAIT);
	if (!unmap)
		return -ENOMEM;

	unmap->len = size;
	unmap->addr[0] = dma_map_page(device->dev, virt_to_page(src),
				      src_off, size, DMA_TO_DEVICE);
	if (dma_mapping_error(device->dev, unmap->addr[0]))
		goto err_get_unmap;

	unmap->to_cnt = 1;

	do {
		txd = device->device_prep_dma_memcpy(chan, dst_phys,
						     unmap->addr[0],
						     size, DMA_PREP_INTERRUPT);
		if (!txd) {
			set_current_state(TASK_INTERRUPTIBLE);
			schedule_timeout(DMA_OUT_RESOURCE_TO);
		}
	} while (!txd && (++retries < DMA_RETRIES));

	if (!txd) {
		pctx->dma_prep_err++;
		goto err_get_unmap;
	}

	txd->callback = perf_copy_callback;
	txd->callback_param = pctx;
	dma_set_unmap(txd, unmap);

	cookie = dmaengine_submit(txd);
	if (dma_submit_error(cookie))
		goto err_set_unmap;

	dmaengine_unmap_put(unmap);

	atomic_inc(&pctx->dma_sync);
	dma_async_issue_pending(chan);

	return size;

err_set_unmap:
	dmaengine_unmap_put(unmap);
err_get_unmap:
	dmaengine_unmap_put(unmap);
	return 0;
}

static int perf_move_data(struct pthr_ctx *pctx, char __iomem *dst, char *src,
			  u64 buf_size, u64 win_size, u64 total)
{
	int chunks, total_chunks, i;
	int copied_chunks = 0;
	u64 copied = 0, result;
	char __iomem *tmp = dst;
	u64 perf, diff_us;
	ktime_t kstart, kstop, kdiff;
	unsigned long last_sleep = jiffies;

	chunks = div64_u64(win_size, buf_size);
	total_chunks = div64_u64(total, buf_size);
	kstart = ktime_get();

	for (i = 0; i < total_chunks; i++) {
		result = perf_copy(pctx, tmp, src, buf_size);
		copied += result;
		copied_chunks++;
		if (copied_chunks == chunks) {
			tmp = dst;
			copied_chunks = 0;
		} else
			tmp += buf_size;

		/* Probably should schedule every 5s to prevent soft hang. */
		if (unlikely((jiffies - last_sleep) > 5 * HZ)) {
			last_sleep = jiffies;
			set_current_state(TASK_INTERRUPTIBLE);
			schedule_timeout(1);
		}

		if (unlikely(kthread_should_stop()))
			break;
	}

	if (use_dma) {
		pr_debug("%s: All DMA descriptors submitted\n", current->comm);
		while (atomic_read(&pctx->dma_sync) != 0) {
			if (kthread_should_stop())
				break;
			msleep(20);
		}
	}

	kstop = ktime_get();
	kdiff = ktime_sub(kstop, kstart);
	diff_us = ktime_to_us(kdiff);

	pr_debug("%s: copied %llu bytes\n", current->comm, copied);

	pr_debug("%s: lasted %llu usecs\n", current->comm, diff_us);

	perf = div64_u64(copied, diff_us);

	pr_debug("%s: MBytes/s: %llu\n", current->comm, perf);

	pctx->copied = copied;
	pctx->diff_us = diff_us;

	return 0;
}

static bool perf_dma_filter_fn(struct dma_chan *chan, void *node)
{
	/* Is the channel required to be on the same node as the device? */
	if (!on_node)
		return true;

	return dev_to_node(&chan->dev->device) == (int)(unsigned long)node;
}

static int ntb_perf_thread(void *data)
{
	struct pthr_ctx *pctx = data;
	struct perf_ctx *perf = pctx->perf;
	struct pci_dev *pdev = perf->ntb->pdev;
	struct perf_mw *mw = &perf->mw;
	char __iomem *dst;
	u64 win_size, buf_size, total;
	void *src;
	int rc, node, i;
	struct dma_chan *dma_chan = NULL;

	pr_debug("kthread %s starting...\n", current->comm);

	node = on_node ? dev_to_node(&pdev->dev) : NUMA_NO_NODE;

	if (use_dma && !pctx->dma_chan) {
		dma_cap_mask_t dma_mask;

		dma_cap_zero(dma_mask);
		dma_cap_set(DMA_MEMCPY, dma_mask);
		dma_chan = dma_request_channel(dma_mask, perf_dma_filter_fn,
					       (void *)(unsigned long)node);
		if (!dma_chan) {
			pr_warn("%s: cannot acquire DMA channel, quitting\n",
				current->comm);
			return -ENODEV;
		}
		pctx->dma_chan = dma_chan;
	}

	for (i = 0; i < MAX_SRCS; i++) {
		pctx->srcs[i] = kmalloc_node(MAX_TEST_SIZE, GFP_KERNEL, node);
		if (!pctx->srcs[i]) {
			rc = -ENOMEM;
			goto err;
		}
	}

	win_size = mw->phys_size;
	buf_size = 1ULL << seg_order;
	total = 1ULL << run_order;

	if (buf_size > MAX_TEST_SIZE)
		buf_size = MAX_TEST_SIZE;

	dst = (char __iomem *)mw->vbase;

	atomic_inc(&perf->tsync);
	while (atomic_read(&perf->tsync) != perf->perf_threads)
		schedule();

	src = pctx->srcs[pctx->src_idx];
	pctx->src_idx = (pctx->src_idx + 1) & (MAX_SRCS - 1);

	rc = perf_move_data(pctx, dst, src, buf_size, win_size, total);

	atomic_dec(&perf->tsync);

	if (rc < 0) {
		pr_err("%s: failed\n", current->comm);
		rc = -ENXIO;
		goto err;
	}

	for (i = 0; i < MAX_SRCS; i++) {
		kfree(pctx->srcs[i]);
		pctx->srcs[i] = NULL;
	}

	atomic_inc(&perf->tdone);
	wake_up(pctx->wq);
	rc = 0;
	goto done;

err:
	for (i = 0; i < MAX_SRCS; i++) {
		kfree(pctx->srcs[i]);
		pctx->srcs[i] = NULL;
	}

	if (dma_chan) {
		dma_release_channel(dma_chan);
		pctx->dma_chan = NULL;
	}

done:
	/* Wait until we are told to stop */
	for (;;) {
		set_current_state(TASK_INTERRUPTIBLE);
		if (kthread_should_stop())
			break;
		schedule();
	}
	__set_current_state(TASK_RUNNING);

	return rc;
}

static void perf_free_mw(struct perf_ctx *perf)
{
	struct perf_mw *mw = &perf->mw;
	struct pci_dev *pdev = perf->ntb->pdev;

	if (!mw->virt_addr)
		return;

	ntb_mw_clear_trans(perf->ntb, PIDX, 0);
	dma_free_coherent(&pdev->dev, mw->buf_size,
			  mw->virt_addr, mw->dma_addr);
	mw->xlat_size = 0;
	mw->buf_size = 0;
	mw->virt_addr = NULL;
}

static int perf_set_mw(struct perf_ctx *perf, resource_size_t size)
{
	struct perf_mw *mw = &perf->mw;
	size_t xlat_size, buf_size;
	int rc;

	if (!size)
		return -EINVAL;

	xlat_size = round_up(size, mw->xlat_align_size);
	buf_size = round_up(size, mw->xlat_align);

	if (mw->xlat_size == xlat_size)
		return 0;

	if (mw->buf_size)
		perf_free_mw(perf);

	mw->xlat_size = xlat_size;
	mw->buf_size = buf_size;

	mw->virt_addr = dma_alloc_coherent(&perf->ntb->pdev->dev, buf_size,
					   &mw->dma_addr, GFP_KERNEL);
	if (!mw->virt_addr) {
		mw->xlat_size = 0;
		mw->buf_size = 0;
	}

	rc = ntb_mw_set_trans(perf->ntb, PIDX, 0, mw->dma_addr, mw->xlat_size);
	if (rc) {
		dev_err(&perf->ntb->dev, "Unable to set mw0 translation\n");
		perf_free_mw(perf);
		return -EIO;
	}

	return 0;
}

static void perf_link_work(struct work_struct *work)
{
	struct perf_ctx *perf =
		container_of(work, struct perf_ctx, link_work.work);
	struct ntb_dev *ndev = perf->ntb;
	struct pci_dev *pdev = ndev->pdev;
	u32 val;
	u64 size;
	int rc;

	dev_dbg(&perf->ntb->pdev->dev, "%s called\n", __func__);

	size = perf->mw.phys_size;

	if (max_mw_size && size > max_mw_size)
		size = max_mw_size;

	ntb_peer_spad_write(ndev, PIDX, MW_SZ_HIGH, upper_32_bits(size));
	ntb_peer_spad_write(ndev, PIDX, MW_SZ_LOW, lower_32_bits(size));
	ntb_peer_spad_write(ndev, PIDX, VERSION, PERF_VERSION);

	/* now read what peer wrote */
	val = ntb_spad_read(ndev, VERSION);
	if (val != PERF_VERSION) {
		dev_dbg(&pdev->dev, "Remote version = %#x\n", val);
		goto out;
	}

	val = ntb_spad_read(ndev, MW_SZ_HIGH);
	size = (u64)val << 32;

	val = ntb_spad_read(ndev, MW_SZ_LOW);
	size |= val;

	dev_dbg(&pdev->dev, "Remote MW size = %#llx\n", size);

	rc = perf_set_mw(perf, size);
	if (rc)
		goto out1;

	perf->link_is_up = true;
	wake_up(&perf->link_wq);

	return;

out1:
	perf_free_mw(perf);

out:
	if (ntb_link_is_up(ndev, NULL, NULL) == 1)
		schedule_delayed_work(&perf->link_work,
				      msecs_to_jiffies(PERF_LINK_DOWN_TIMEOUT));
}

static int perf_setup_mw(struct ntb_dev *ntb, struct perf_ctx *perf)
{
	struct perf_mw *mw;
	int rc;

	mw = &perf->mw;

	rc = ntb_mw_get_align(ntb, PIDX, 0, &mw->xlat_align,
			      &mw->xlat_align_size, NULL);
	if (rc)
		return rc;

	rc = ntb_peer_mw_get_addr(ntb, 0, &mw->phys_addr, &mw->phys_size);
	if (rc)
		return rc;

	perf->mw.vbase = ioremap_wc(mw->phys_addr, mw->phys_size);
	if (!mw->vbase)
		return -ENOMEM;

	return 0;
}

static ssize_t debugfs_run_read(struct file *filp, char __user *ubuf,
				size_t count, loff_t *offp)
{
	struct perf_ctx *perf = filp->private_data;
	char *buf;
	ssize_t ret, out_off = 0;
	struct pthr_ctx *pctx;
	int i;
	u64 rate;

	if (!perf)
		return 0;

	buf = kmalloc(1024, GFP_KERNEL);
	if (!buf)
		return -ENOMEM;

	if (mutex_is_locked(&perf->run_mutex)) {
		out_off = scnprintf(buf, 64, "running\n");
		goto read_from_buf;
	}

	for (i = 0; i < MAX_THREADS; i++) {
		pctx = &perf->pthr_ctx[i];

		if (pctx->status == -ENODATA)
			break;

		if (pctx->status) {
			out_off += scnprintf(buf + out_off, 1024 - out_off,
					    "%d: error %d\n", i,
					    pctx->status);
			continue;
		}

		rate = div64_u64(pctx->copied, pctx->diff_us);
		out_off += scnprintf(buf + out_off, 1024 - out_off,
			"%d: copied %llu bytes in %llu usecs, %llu MBytes/s\n",
			i, pctx->copied, pctx->diff_us, rate);
	}

read_from_buf:
	ret = simple_read_from_buffer(ubuf, count, offp, buf, out_off);
	kfree(buf);

	return ret;
}

static void threads_cleanup(struct perf_ctx *perf)
{
	struct pthr_ctx *pctx;
	int i;

	for (i = 0; i < MAX_THREADS; i++) {
		pctx = &perf->pthr_ctx[i];
		if (pctx->thread) {
			pctx->status = kthread_stop(pctx->thread);
			pctx->thread = NULL;
		}
	}
}

static void perf_clear_thread_status(struct perf_ctx *perf)
{
	int i;

	for (i = 0; i < MAX_THREADS; i++)
		perf->pthr_ctx[i].status = -ENODATA;
}

static ssize_t debugfs_run_write(struct file *filp, const char __user *ubuf,
				 size_t count, loff_t *offp)
{
	struct perf_ctx *perf = filp->private_data;
	int node, i;
	DECLARE_WAIT_QUEUE_HEAD(wq);

	if (wait_event_interruptible(perf->link_wq, perf->link_is_up))
		return -ENOLINK;

	if (perf->perf_threads == 0)
		return -EINVAL;

	if (!mutex_trylock(&perf->run_mutex))
		return -EBUSY;

	perf_clear_thread_status(perf);

	if (perf->perf_threads > MAX_THREADS) {
		perf->perf_threads = MAX_THREADS;
		pr_info("Reset total threads to: %u\n", MAX_THREADS);
	}

	/* no greater than 1M */
	if (seg_order > MAX_SEG_ORDER) {
		seg_order = MAX_SEG_ORDER;
		pr_info("Fix seg_order to %u\n", seg_order);
	}

	if (run_order < seg_order) {
		run_order = seg_order;
		pr_info("Fix run_order to %u\n", run_order);
	}

	node = on_node ? dev_to_node(&perf->ntb->pdev->dev)
		       : NUMA_NO_NODE;
	atomic_set(&perf->tdone, 0);

	/* launch kernel thread */
	for (i = 0; i < perf->perf_threads; i++) {
		struct pthr_ctx *pctx;

		pctx = &perf->pthr_ctx[i];
		atomic_set(&pctx->dma_sync, 0);
		pctx->perf = perf;
		pctx->wq = &wq;
		pctx->thread =
			kthread_create_on_node(ntb_perf_thread,
					       (void *)pctx,
					       node, "ntb_perf %d", i);
		if (IS_ERR(pctx->thread)) {
			pctx->thread = NULL;
			goto err;
		} else {
			wake_up_process(pctx->thread);
		}
	}

	wait_event_interruptible(wq,
		atomic_read(&perf->tdone) == perf->perf_threads);

	threads_cleanup(perf);
	mutex_unlock(&perf->run_mutex);
	return count;

err:
	threads_cleanup(perf);
	mutex_unlock(&perf->run_mutex);
	return -ENXIO;
}

static const struct file_operations ntb_perf_debugfs_run = {
	.owner = THIS_MODULE,
	.open = simple_open,
	.read = debugfs_run_read,
	.write = debugfs_run_write,
};

static int perf_debugfs_setup(struct perf_ctx *perf)
{
	struct pci_dev *pdev = perf->ntb->pdev;
	struct dentry *debugfs_node_dir;
	struct dentry *debugfs_run;
	struct dentry *debugfs_threads;
	struct dentry *debugfs_seg_order;
	struct dentry *debugfs_run_order;
	struct dentry *debugfs_use_dma;
	struct dentry *debugfs_on_node;

	if (!debugfs_initialized())
		return -ENODEV;

	/* Assumpion: only one NTB device in the system */
	if (!perf_debugfs_dir) {
		perf_debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL);
		if (!perf_debugfs_dir)
			return -ENODEV;
	}

	debugfs_node_dir = debugfs_create_dir(pci_name(pdev),
					      perf_debugfs_dir);
	if (!debugfs_node_dir)
		goto err;

	debugfs_run = debugfs_create_file("run", S_IRUSR | S_IWUSR,
					  debugfs_node_dir, perf,
					  &ntb_perf_debugfs_run);
	if (!debugfs_run)
		goto err;

	debugfs_threads = debugfs_create_u8("threads", S_IRUSR | S_IWUSR,
					    debugfs_node_dir,
					    &perf->perf_threads);
	if (!debugfs_threads)
		goto err;

	debugfs_seg_order = debugfs_create_u32("seg_order", 0600,
					       debugfs_node_dir,
					       &seg_order);
	if (!debugfs_seg_order)
		goto err;

	debugfs_run_order = debugfs_create_u32("run_order", 0600,
					       debugfs_node_dir,
					       &run_order);
	if (!debugfs_run_order)
		goto err;

	debugfs_use_dma = debugfs_create_bool("use_dma", 0600,
					       debugfs_node_dir,
					       &use_dma);
	if (!debugfs_use_dma)
		goto err;

	debugfs_on_node = debugfs_create_bool("on_node", 0600,
					      debugfs_node_dir,
					      &on_node);
	if (!debugfs_on_node)
		goto err;

	return 0;

err:
	debugfs_remove_recursive(perf_debugfs_dir);
	perf_debugfs_dir = NULL;
	return -ENODEV;
}

static int perf_probe(struct ntb_client *client, struct ntb_dev *ntb)
{
	struct pci_dev *pdev = ntb->pdev;
	struct perf_ctx *perf;
	int node;
	int rc = 0;

	if (ntb_spad_count(ntb) < MAX_SPAD) {
		dev_err(&ntb->dev, "Not enough scratch pad registers for %s",
			DRIVER_NAME);
		return -EIO;
	}

	if (!ntb->ops->mw_set_trans) {
		dev_err(&ntb->dev, "Need inbound MW based NTB API\n");
		return -EINVAL;
	}

	if (ntb_peer_port_count(ntb) != NTB_DEF_PEER_CNT)
		dev_warn(&ntb->dev, "Multi-port NTB devices unsupported\n");

	node = on_node ? dev_to_node(&pdev->dev) : NUMA_NO_NODE;
	perf = kzalloc_node(sizeof(*perf), GFP_KERNEL, node);
	if (!perf) {
		rc = -ENOMEM;
		goto err_perf;
	}

	perf->ntb = ntb;
	perf->perf_threads = 1;
	atomic_set(&perf->tsync, 0);
	mutex_init(&perf->run_mutex);
	spin_lock_init(&perf->db_lock);
	perf_setup_mw(ntb, perf);
	init_waitqueue_head(&perf->link_wq);
	INIT_DELAYED_WORK(&perf->link_work, perf_link_work);

	rc = ntb_set_ctx(ntb, perf, &perf_ops);
	if (rc)
		goto err_ctx;

	perf->link_is_up = false;
	ntb_link_enable(ntb, NTB_SPEED_AUTO, NTB_WIDTH_AUTO);
	ntb_link_event(ntb);

	rc = perf_debugfs_setup(perf);
	if (rc)
		goto err_ctx;

	perf_clear_thread_status(perf);

	return 0;

err_ctx:
	cancel_delayed_work_sync(&perf->link_work);
	kfree(perf);
err_perf:
	return rc;
}

static void perf_remove(struct ntb_client *client, struct ntb_dev *ntb)
{
	struct perf_ctx *perf = ntb->ctx;
	int i;

	dev_dbg(&perf->ntb->dev, "%s called\n", __func__);

	mutex_lock(&perf->run_mutex);

	cancel_delayed_work_sync(&perf->link_work);

	ntb_clear_ctx(ntb);
	ntb_link_disable(ntb);

	debugfs_remove_recursive(perf_debugfs_dir);
	perf_debugfs_dir = NULL;

	if (use_dma) {
		for (i = 0; i < MAX_THREADS; i++) {
			struct pthr_ctx *pctx = &perf->pthr_ctx[i];

			if (pctx->dma_chan)
				dma_release_channel(pctx->dma_chan);
		}
	}

	kfree(perf);
}

static struct ntb_client perf_client = {
	.ops = {
		.probe = perf_probe,
		.remove = perf_remove,
	},
};
module_ntb_client(perf_client);