aboutsummaryrefslogtreecommitdiffstats
path: root/target/linux/brcm2708/patches-4.19/950-0698-Ported-pcie-brcmstb-bounce-buffer-implementation-to-.patch
blob: c69a8ca16ed15d3d10c0a9e20b67f3ccf7ae7f42 (plain)
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
From ccd23ce562e8223ba7c6acf7dcb7058ff89ff7ec Mon Sep 17 00:00:00 2001
From: yaroslavros <yaroslavros@gmail.com>
Date: Wed, 14 Aug 2019 15:22:55 +0100
Subject: [PATCH] Ported pcie-brcmstb bounce buffer implementation to
 ARM64. (#3144)

Ported pcie-brcmstb bounce buffer implementation to ARM64.
This enables full 4G RAM usage on Raspberry Pi in 64-bit mode.

Signed-off-by: Yaroslav Rosomakho <yaroslavros@gmail.com>
---
 arch/arm64/include/asm/dma-mapping.h          |  21 +
 arch/arm64/mm/dma-mapping.c                   |  50 ++
 drivers/pci/controller/Makefile               |   3 +
 drivers/pci/controller/pcie-brcmstb-bounce.h  |   2 +-
 .../pci/controller/pcie-brcmstb-bounce64.c    | 576 ++++++++++++++++++
 drivers/pci/controller/pcie-brcmstb.c         |  30 +-
 6 files changed, 658 insertions(+), 24 deletions(-)
 create mode 100644 drivers/pci/controller/pcie-brcmstb-bounce64.c

--- a/arch/arm64/include/asm/dma-mapping.h
+++ b/arch/arm64/include/asm/dma-mapping.h
@@ -24,6 +24,27 @@
 #include <xen/xen.h>
 #include <asm/xen/hypervisor.h>
 
+extern void *arm64_dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
+			   gfp_t gfp, unsigned long attrs);
+extern void arm64_dma_free(struct device *dev, size_t size, void *cpu_addr,
+			 dma_addr_t handle, unsigned long attrs);
+extern int arm64_dma_mmap(struct device *dev, struct vm_area_struct *vma,
+			void *cpu_addr, dma_addr_t dma_addr, size_t size,
+			unsigned long attrs);
+extern int arm64_dma_get_sgtable(struct device *dev, struct sg_table *sgt,
+		void *cpu_addr, dma_addr_t dma_addr, size_t size,
+		unsigned long attrs);
+extern int arm64_dma_map_sg(struct device *dev, struct scatterlist *sgl, int nelems,
+		enum dma_data_direction dir, unsigned long attrs);
+extern void arm64_dma_unmap_sg(struct device *dev, struct scatterlist *sgl, int,
+		enum dma_data_direction dir, unsigned long attrs);
+extern void arm64_dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *sgl, int nelems,
+		enum dma_data_direction dir);
+extern void arm64_dma_sync_sg_for_device(struct device *dev, struct scatterlist *sgl, int nelems,
+		enum dma_data_direction dir);
+
+
+
 extern const struct dma_map_ops dummy_dma_ops;
 
 static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
--- a/arch/arm64/mm/dma-mapping.c
+++ b/arch/arm64/mm/dma-mapping.c
@@ -138,6 +138,12 @@ no_mem:
 	return NULL;
 }
 
+void *arm64_dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
+                           gfp_t gfp, unsigned long attrs)
+{
+        return __dma_alloc(dev, size, handle, gfp, attrs);
+}
+
 static void __dma_free(struct device *dev, size_t size,
 		       void *vaddr, dma_addr_t dma_handle,
 		       unsigned long attrs)
@@ -154,6 +160,12 @@ static void __dma_free(struct device *de
 	swiotlb_free(dev, size, swiotlb_addr, dma_handle, attrs);
 }
 
+void arm64_dma_free(struct device *dev, size_t size, void *cpu_addr,
+                         dma_addr_t handle, unsigned long attrs)
+{
+        __dma_free(dev, size, cpu_addr, handle, attrs);
+}
+
 static dma_addr_t __swiotlb_map_page(struct device *dev, struct page *page,
 				     unsigned long offset, size_t size,
 				     enum dma_data_direction dir,
@@ -197,6 +209,12 @@ static int __swiotlb_map_sg_attrs(struct
 	return ret;
 }
 
+int arm64_dma_map_sg(struct device *dev, struct scatterlist *sgl, int nelems,
+                enum dma_data_direction dir, unsigned long attrs)
+{
+	return __swiotlb_map_sg_attrs(dev, sgl, nelems, dir, attrs);
+}
+
 static void __swiotlb_unmap_sg_attrs(struct device *dev,
 				     struct scatterlist *sgl, int nelems,
 				     enum dma_data_direction dir,
@@ -213,6 +231,12 @@ static void __swiotlb_unmap_sg_attrs(str
 	swiotlb_unmap_sg_attrs(dev, sgl, nelems, dir, attrs);
 }
 
+void arm64_dma_unmap_sg(struct device *dev, struct scatterlist *sgl, int nelems,
+                enum dma_data_direction dir, unsigned long attrs)
+{
+	__swiotlb_unmap_sg_attrs(dev, sgl, nelems, dir, attrs);
+}
+
 static void __swiotlb_sync_single_for_cpu(struct device *dev,
 					  dma_addr_t dev_addr, size_t size,
 					  enum dma_data_direction dir)
@@ -245,6 +269,12 @@ static void __swiotlb_sync_sg_for_cpu(st
 	swiotlb_sync_sg_for_cpu(dev, sgl, nelems, dir);
 }
 
+void arm64_dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *sgl, int nelems,
+                enum dma_data_direction dir)
+{
+	__swiotlb_sync_sg_for_cpu(dev, sgl, nelems, dir);
+}
+
 static void __swiotlb_sync_sg_for_device(struct device *dev,
 					 struct scatterlist *sgl, int nelems,
 					 enum dma_data_direction dir)
@@ -259,6 +289,12 @@ static void __swiotlb_sync_sg_for_device
 				       sg->length, dir);
 }
 
+void arm64_dma_sync_sg_for_device(struct device *dev, struct scatterlist *sgl, int nelems,
+                enum dma_data_direction dir)
+{
+	__swiotlb_sync_sg_for_device(dev, sgl, nelems, dir);
+}
+
 static int __swiotlb_mmap_pfn(struct vm_area_struct *vma,
 			      unsigned long pfn, size_t size)
 {
@@ -294,6 +330,13 @@ static int __swiotlb_mmap(struct device
 	return __swiotlb_mmap_pfn(vma, pfn, size);
 }
 
+int arm64_dma_mmap(struct device *dev, struct vm_area_struct *vma,
+                        void *cpu_addr, dma_addr_t dma_addr, size_t size,
+                        unsigned long attrs)
+{
+	return __swiotlb_mmap(dev, vma, cpu_addr, dma_addr, size, attrs);
+}
+
 static int __swiotlb_get_sgtable_page(struct sg_table *sgt,
 				      struct page *page, size_t size)
 {
@@ -314,6 +357,13 @@ static int __swiotlb_get_sgtable(struct
 	return __swiotlb_get_sgtable_page(sgt, page, size);
 }
 
+int arm64_dma_get_sgtable(struct device *dev, struct sg_table *sgt,
+                void *cpu_addr, dma_addr_t dma_addr, size_t size,
+                unsigned long attrs)
+{
+	return __swiotlb_get_sgtable(dev, sgt, cpu_addr, dma_addr, size, attrs);
+}
+
 static int __swiotlb_dma_supported(struct device *hwdev, u64 mask)
 {
 	if (swiotlb)
--- a/drivers/pci/controller/Makefile
+++ b/drivers/pci/controller/Makefile
@@ -32,6 +32,9 @@ obj-$(CONFIG_PCIE_BRCMSTB) += pcie-brcms
 ifdef CONFIG_ARM
 obj-$(CONFIG_PCIE_BRCMSTB) += pcie-brcmstb-bounce.o
 endif
+ifdef CONFIG_ARM64
+obj-$(CONFIG_PCIE_BRCMSTB) += pcie-brcmstb-bounce64.o
+endif
 
 obj-$(CONFIG_VMD) += vmd.o
 # pcie-hisi.o quirks are needed even without CONFIG_PCIE_DW
--- a/drivers/pci/controller/pcie-brcmstb-bounce.h
+++ b/drivers/pci/controller/pcie-brcmstb-bounce.h
@@ -6,7 +6,7 @@
 #ifndef _PCIE_BRCMSTB_BOUNCE_H
 #define _PCIE_BRCMSTB_BOUNCE_H
 
-#ifdef CONFIG_ARM
+#if defined(CONFIG_ARM) || defined(CONFIG_ARM64)
 
 int brcm_pcie_bounce_init(struct device *dev, unsigned long buffer_size,
 			  dma_addr_t threshold);
--- /dev/null
+++ b/drivers/pci/controller/pcie-brcmstb-bounce64.c
@@ -0,0 +1,576 @@
+/*
+ *  This code started out as a version of arch/arm/common/dmabounce.c,
+ *  modified to cope with highmem pages. Now it has been changed heavily -
+ *  it now preallocates a large block (currently 4MB) and carves it up
+ *  sequentially in ring fashion, and DMA is used to copy the data - to the
+ *  point where very little of the original remains.
+ *
+ *  Copyright (C) 2019 Raspberry Pi (Trading) Ltd.
+ *
+ *  Original version by Brad Parker (brad@heeltoe.com)
+ *  Re-written by Christopher Hoover <ch@murgatroid.com>
+ *  Made generic by Deepak Saxena <dsaxena@plexity.net>
+ *
+ *  Copyright (C) 2002 Hewlett Packard Company.
+ *  Copyright (C) 2004 MontaVista Software, Inc.
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  version 2 as published by the Free Software Foundation.
+ */
+
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/page-flags.h>
+#include <linux/device.h>
+#include <linux/dma-mapping.h>
+#include <linux/dma-direct.h>
+#include <linux/dmapool.h>
+#include <linux/list.h>
+#include <linux/scatterlist.h>
+#include <linux/bitmap.h>
+#include <linux/swiotlb.h>
+
+#include <asm/cacheflush.h>
+
+#define STATS
+
+#ifdef STATS
+#define DO_STATS(X) do { X ; } while (0)
+#else
+#define DO_STATS(X) do { } while (0)
+#endif
+
+/* ************************************************** */
+
+struct safe_buffer {
+	struct list_head node;
+
+	/* original request */
+	size_t		size;
+	int		direction;
+
+	struct dmabounce_pool *pool;
+	void		*safe;
+	dma_addr_t	unsafe_dma_addr;
+	dma_addr_t	safe_dma_addr;
+};
+
+struct dmabounce_pool {
+	unsigned long	pages;
+	void		*virt_addr;
+	dma_addr_t	dma_addr;
+	unsigned long	*alloc_map;
+	unsigned long	alloc_pos;
+	spinlock_t	lock;
+	struct device	*dev;
+	unsigned long   num_pages;
+#ifdef STATS
+	size_t		max_size;
+	unsigned long	num_bufs;
+	unsigned long   max_bufs;
+	unsigned long   max_pages;
+#endif
+};
+
+struct dmabounce_device_info {
+	struct device *dev;
+	dma_addr_t threshold;
+	struct list_head safe_buffers;
+	struct dmabounce_pool pool;
+	rwlock_t lock;
+#ifdef STATS
+	unsigned long map_count;
+	unsigned long unmap_count;
+	unsigned long sync_dev_count;
+	unsigned long sync_cpu_count;
+	unsigned long fail_count;
+	int attr_res;
+#endif
+};
+
+static struct dmabounce_device_info *g_dmabounce_device_info;
+
+extern int bcm2838_dma40_memcpy_init(void);
+extern void bcm2838_dma40_memcpy(dma_addr_t dst, dma_addr_t src, size_t size);
+
+#ifdef STATS
+static ssize_t
+bounce_show(struct device *dev, struct device_attribute *attr, char *buf)
+{
+	struct dmabounce_device_info *device_info = g_dmabounce_device_info;
+	return sprintf(buf, "m:%lu/%lu s:%lu/%lu f:%lu s:%zu b:%lu/%lu a:%lu/%lu\n",
+		device_info->map_count,
+		device_info->unmap_count,
+		device_info->sync_dev_count,
+		device_info->sync_cpu_count,
+		device_info->fail_count,
+		device_info->pool.max_size,
+		device_info->pool.num_bufs,
+		device_info->pool.max_bufs,
+		device_info->pool.num_pages * PAGE_SIZE,
+		device_info->pool.max_pages * PAGE_SIZE);
+}
+
+static DEVICE_ATTR(dmabounce_stats, 0444, bounce_show, NULL);
+#endif
+
+static int bounce_create(struct dmabounce_pool *pool, struct device *dev,
+			 unsigned long buffer_size)
+{
+	int ret = -ENOMEM;
+	pool->pages = (buffer_size + PAGE_SIZE - 1)/PAGE_SIZE;
+	pool->alloc_map = bitmap_zalloc(pool->pages, GFP_KERNEL);
+	if (!pool->alloc_map)
+		goto err_bitmap;
+	pool->virt_addr = dma_alloc_coherent(dev, pool->pages * PAGE_SIZE,
+					     &pool->dma_addr, GFP_KERNEL);
+	if (!pool->virt_addr)
+		goto err_dmabuf;
+
+	pool->alloc_pos = 0;
+	spin_lock_init(&pool->lock);
+	pool->dev = dev;
+	pool->num_pages = 0;
+
+	DO_STATS(pool->max_size = 0);
+	DO_STATS(pool->num_bufs = 0);
+	DO_STATS(pool->max_bufs = 0);
+	DO_STATS(pool->max_pages = 0);
+
+	return  0;
+
+err_dmabuf:
+	bitmap_free(pool->alloc_map);
+err_bitmap:
+	return ret;
+}
+
+static void bounce_destroy(struct dmabounce_pool *pool)
+{
+	dma_free_coherent(pool->dev, pool->pages * PAGE_SIZE, pool->virt_addr,
+			  pool->dma_addr);
+
+	bitmap_free(pool->alloc_map);
+}
+
+static void *bounce_alloc(struct dmabounce_pool *pool, size_t size,
+			  dma_addr_t *dmaaddrp)
+{
+	unsigned long pages;
+	unsigned long flags;
+	unsigned long pos;
+
+	pages = (size + PAGE_SIZE - 1)/PAGE_SIZE;
+
+	DO_STATS(pool->max_size = max(size, pool->max_size));
+
+	spin_lock_irqsave(&pool->lock, flags);
+	pos = bitmap_find_next_zero_area(pool->alloc_map, pool->pages,
+					 pool->alloc_pos, pages, 0);
+	/* If not found, try from the start */
+	if (pos >= pool->pages && pool->alloc_pos)
+		pos = bitmap_find_next_zero_area(pool->alloc_map, pool->pages,
+						 0, pages, 0);
+
+	if (pos >= pool->pages) {
+		spin_unlock_irqrestore(&pool->lock, flags);
+		return NULL;
+	}
+
+	bitmap_set(pool->alloc_map, pos, pages);
+	pool->alloc_pos = (pos + pages) % pool->pages;
+	pool->num_pages += pages;
+
+	DO_STATS(pool->num_bufs++);
+	DO_STATS(pool->max_bufs = max(pool->num_bufs, pool->max_bufs));
+	DO_STATS(pool->max_pages = max(pool->num_pages, pool->max_pages));
+
+	spin_unlock_irqrestore(&pool->lock, flags);
+
+	*dmaaddrp = pool->dma_addr + pos * PAGE_SIZE;
+
+	return pool->virt_addr + pos * PAGE_SIZE;
+}
+
+static void
+bounce_free(struct dmabounce_pool *pool, void *buf, size_t size)
+{
+	unsigned long pages;
+	unsigned long flags;
+	unsigned long pos;
+
+	pages = (size + PAGE_SIZE - 1)/PAGE_SIZE;
+	pos = (buf - pool->virt_addr)/PAGE_SIZE;
+
+	BUG_ON((buf - pool->virt_addr) & (PAGE_SIZE - 1));
+
+	spin_lock_irqsave(&pool->lock, flags);
+	bitmap_clear(pool->alloc_map, pos, pages);
+	pool->num_pages -= pages;
+	if (pool->num_pages == 0)
+		pool->alloc_pos = 0;
+	DO_STATS(pool->num_bufs--);
+	spin_unlock_irqrestore(&pool->lock, flags);
+}
+
+/* allocate a 'safe' buffer and keep track of it */
+static struct safe_buffer *
+alloc_safe_buffer(struct dmabounce_device_info *device_info,
+		  dma_addr_t dma_addr, size_t size, enum dma_data_direction dir)
+{
+	struct safe_buffer *buf;
+	struct dmabounce_pool *pool = &device_info->pool;
+	struct device *dev = device_info->dev;
+	unsigned long flags;
+
+	/*
+	 * Although one might expect this to be called in thread context,
+	 * using GFP_KERNEL here leads to hard-to-debug lockups. in_atomic()
+	 * was previously used to select the appropriate allocation mode,
+	 * but this is unsafe.
+	 */
+	buf = kmalloc(sizeof(struct safe_buffer), GFP_ATOMIC);
+	if (!buf) {
+		dev_warn(dev, "%s: kmalloc failed\n", __func__);
+		return NULL;
+	}
+
+	buf->unsafe_dma_addr = dma_addr;
+	buf->size = size;
+	buf->direction = dir;
+	buf->pool = pool;
+
+	buf->safe = bounce_alloc(pool, size, &buf->safe_dma_addr);
+
+	if (!buf->safe) {
+		dev_warn(dev,
+			 "%s: could not alloc dma memory (size=%d)\n",
+			 __func__, size);
+		kfree(buf);
+		return NULL;
+	}
+
+	write_lock_irqsave(&device_info->lock, flags);
+	list_add(&buf->node, &device_info->safe_buffers);
+	write_unlock_irqrestore(&device_info->lock, flags);
+
+	return buf;
+}
+
+/* determine if a buffer is from our "safe" pool */
+static struct safe_buffer *
+find_safe_buffer(struct dmabounce_device_info *device_info,
+		 dma_addr_t safe_dma_addr)
+{
+	struct safe_buffer *b, *rb = NULL;
+	unsigned long flags;
+
+	read_lock_irqsave(&device_info->lock, flags);
+
+	list_for_each_entry(b, &device_info->safe_buffers, node)
+		if (b->safe_dma_addr <= safe_dma_addr &&
+		    b->safe_dma_addr + b->size > safe_dma_addr) {
+			rb = b;
+			break;
+		}
+
+	read_unlock_irqrestore(&device_info->lock, flags);
+	return rb;
+}
+
+static void
+free_safe_buffer(struct dmabounce_device_info *device_info,
+		 struct safe_buffer *buf)
+{
+	unsigned long flags;
+
+	write_lock_irqsave(&device_info->lock, flags);
+	list_del(&buf->node);
+	write_unlock_irqrestore(&device_info->lock, flags);
+
+	bounce_free(buf->pool, buf->safe, buf->size);
+
+	kfree(buf);
+}
+
+/* ************************************************** */
+
+static struct safe_buffer *
+find_safe_buffer_dev(struct device *dev, dma_addr_t dma_addr, const char *where)
+{
+	if (!dev || !g_dmabounce_device_info)
+		return NULL;
+	if (dma_mapping_error(dev, dma_addr)) {
+		dev_err(dev, "Trying to %s invalid mapping\n", where);
+		return NULL;
+	}
+	return find_safe_buffer(g_dmabounce_device_info, dma_addr);
+}
+
+static dma_addr_t
+map_single(struct device *dev, struct safe_buffer *buf, size_t size,
+	   enum dma_data_direction dir, unsigned long attrs)
+{
+	BUG_ON(buf->size != size);
+	BUG_ON(buf->direction != dir);
+
+	dev_dbg(dev, "map: %llx->%llx\n", (u64)buf->unsafe_dma_addr,
+		(u64)buf->safe_dma_addr);
+
+	if ((dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL) &&
+	    !(attrs & DMA_ATTR_SKIP_CPU_SYNC))
+		bcm2838_dma40_memcpy(buf->safe_dma_addr, buf->unsafe_dma_addr,
+				     size);
+
+	return buf->safe_dma_addr;
+}
+
+static dma_addr_t
+unmap_single(struct device *dev, struct safe_buffer *buf, size_t size,
+	     enum dma_data_direction dir, unsigned long attrs)
+{
+	BUG_ON(buf->size != size);
+	BUG_ON(buf->direction != dir);
+
+	if ((dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL) &&
+	    !(attrs & DMA_ATTR_SKIP_CPU_SYNC)) {
+		dev_dbg(dev, "unmap: %llx->%llx\n", (u64)buf->safe_dma_addr,
+			(u64)buf->unsafe_dma_addr);
+
+		bcm2838_dma40_memcpy(buf->unsafe_dma_addr, buf->safe_dma_addr,
+				     size);
+	}
+	return buf->unsafe_dma_addr;
+}
+
+/* ************************************************** */
+
+/*
+ * see if a buffer address is in an 'unsafe' range.  if it is
+ * allocate a 'safe' buffer and copy the unsafe buffer into it.
+ * substitute the safe buffer for the unsafe one.
+ * (basically move the buffer from an unsafe area to a safe one)
+ */
+static dma_addr_t
+dmabounce_map_page(struct device *dev, struct page *page, unsigned long offset,
+		   size_t size, enum dma_data_direction dir,
+		   unsigned long attrs)
+{
+	struct dmabounce_device_info *device_info = g_dmabounce_device_info;
+	dma_addr_t dma_addr;
+
+	dma_addr = phys_to_dma(dev, page_to_phys(page)) + offset;
+
+	swiotlb_sync_single_for_device(dev, dma_addr, size, dir);
+        if (!is_device_dma_coherent(dev))
+		__dma_map_area(phys_to_virt(dma_to_phys(dev, dma_addr)), size, dir);
+
+	if (device_info && (dma_addr + size) > device_info->threshold) {
+		struct safe_buffer *buf;
+
+		buf = alloc_safe_buffer(device_info, dma_addr, size, dir);
+		if (!buf) {
+			DO_STATS(device_info->fail_count++);
+			return (~(dma_addr_t)0x0);
+		}
+
+		DO_STATS(device_info->map_count++);
+
+		dma_addr = map_single(dev, buf, size, dir, attrs);
+	}
+	return dma_addr;
+}
+
+/*
+ * see if a mapped address was really a "safe" buffer and if so, copy
+ * the data from the safe buffer back to the unsafe buffer and free up
+ * the safe buffer.  (basically return things back to the way they
+ * should be)
+ */
+static void
+dmabounce_unmap_page(struct device *dev, dma_addr_t dma_addr, size_t size,
+		     enum dma_data_direction dir, unsigned long attrs)
+{
+	struct safe_buffer *buf;
+
+	buf = find_safe_buffer_dev(dev, dma_addr, __func__);
+	if (buf) {
+		DO_STATS(g_dmabounce_device_info->unmap_count++);
+		dma_addr = unmap_single(dev, buf, size, dir, attrs);
+		free_safe_buffer(g_dmabounce_device_info, buf);
+	}
+
+        if (!is_device_dma_coherent(dev))
+		__dma_unmap_area(phys_to_virt(dma_to_phys(dev, dma_addr)), size, dir);
+	swiotlb_sync_single_for_cpu(dev, dma_addr, size, dir);
+}
+
+/*
+ * A version of dmabounce_map_page that assumes the mapping has already
+ * been created - intended for streaming operation.
+ */
+static void
+dmabounce_sync_for_device(struct device *dev, dma_addr_t dma_addr, size_t size,
+			  enum dma_data_direction dir)
+{
+	struct safe_buffer *buf;
+
+        swiotlb_sync_single_for_device(dev, dma_addr, size, dir);
+        if (!is_device_dma_coherent(dev))
+                __dma_map_area(phys_to_virt(dma_to_phys(dev, dma_addr)), size, dir);
+
+	buf = find_safe_buffer_dev(dev, dma_addr, __func__);
+	if (buf) {
+		DO_STATS(g_dmabounce_device_info->sync_dev_count++);
+		map_single(dev, buf, size, dir, 0);
+	}
+}
+
+/*
+ * A version of dmabounce_unmap_page that doesn't destroy the mapping -
+ * intended for streaming operation.
+ */
+static void
+dmabounce_sync_for_cpu(struct device *dev, dma_addr_t dma_addr,
+		       size_t size, enum dma_data_direction dir)
+{
+	struct safe_buffer *buf;
+
+	buf = find_safe_buffer_dev(dev, dma_addr, __func__);
+	if (buf) {
+		DO_STATS(g_dmabounce_device_info->sync_cpu_count++);
+		dma_addr = unmap_single(dev, buf, size, dir, 0);
+	}
+
+        if (!is_device_dma_coherent(dev))
+                __dma_unmap_area(phys_to_virt(dma_to_phys(dev, dma_addr)), size, dir);
+        swiotlb_sync_single_for_cpu(dev, dma_addr, size, dir);
+}
+
+static int dmabounce_dma_supported(struct device *dev, u64 dma_mask)
+{
+	if (g_dmabounce_device_info)
+		return 0;
+
+	return swiotlb_dma_supported(dev, dma_mask);
+}
+
+static int dmabounce_mapping_error(struct device *dev, dma_addr_t dma_addr)
+{
+	return swiotlb_dma_mapping_error(dev, dma_addr);
+}
+
+static const struct dma_map_ops dmabounce_ops = {
+	.alloc			= arm64_dma_alloc,
+	.free			= arm64_dma_free,
+	.mmap			= arm64_dma_mmap,
+	.get_sgtable		= arm64_dma_get_sgtable,
+	.map_page		= dmabounce_map_page,
+	.unmap_page		= dmabounce_unmap_page,
+	.sync_single_for_cpu	= dmabounce_sync_for_cpu,
+	.sync_single_for_device	= dmabounce_sync_for_device,
+	.map_sg			= arm64_dma_map_sg,
+	.unmap_sg		= arm64_dma_unmap_sg,
+	.sync_sg_for_cpu	= arm64_dma_sync_sg_for_cpu,
+	.sync_sg_for_device	= arm64_dma_sync_sg_for_device,
+	.dma_supported		= dmabounce_dma_supported,
+	.mapping_error		= dmabounce_mapping_error,
+};
+
+int brcm_pcie_bounce_init(struct device *dev,
+			  unsigned long buffer_size,
+			  dma_addr_t threshold)
+{
+	struct dmabounce_device_info *device_info;
+	int ret;
+
+	/* Only support a single client */
+	if (g_dmabounce_device_info)
+		return -EBUSY;
+
+	ret = bcm2838_dma40_memcpy_init();
+	if (ret)
+		return ret;
+
+	device_info = kmalloc(sizeof(struct dmabounce_device_info), GFP_ATOMIC);
+	if (!device_info) {
+		dev_err(dev,
+			"Could not allocated dmabounce_device_info\n");
+		return -ENOMEM;
+	}
+
+	ret = bounce_create(&device_info->pool, dev, buffer_size);
+	if (ret) {
+		dev_err(dev,
+			"dmabounce: could not allocate %ld byte DMA pool\n",
+			buffer_size);
+		goto err_bounce;
+	}
+
+	device_info->dev = dev;
+	device_info->threshold = threshold;
+	INIT_LIST_HEAD(&device_info->safe_buffers);
+	rwlock_init(&device_info->lock);
+
+	DO_STATS(device_info->map_count = 0);
+	DO_STATS(device_info->unmap_count = 0);
+	DO_STATS(device_info->sync_dev_count = 0);
+	DO_STATS(device_info->sync_cpu_count = 0);
+	DO_STATS(device_info->fail_count = 0);
+	DO_STATS(device_info->attr_res =
+		 device_create_file(dev, &dev_attr_dmabounce_stats));
+
+	g_dmabounce_device_info = device_info;
+
+	dev_err(dev, "dmabounce: initialised - %ld kB, threshold %pad\n",
+		 buffer_size / 1024, &threshold);
+
+	return 0;
+
+ err_bounce:
+	kfree(device_info);
+	return ret;
+}
+EXPORT_SYMBOL(brcm_pcie_bounce_init);
+
+void brcm_pcie_bounce_uninit(struct device *dev)
+{
+	struct dmabounce_device_info *device_info = g_dmabounce_device_info;
+
+	g_dmabounce_device_info = NULL;
+
+	if (!device_info) {
+		dev_warn(dev,
+			 "Never registered with dmabounce but attempting"
+			 "to unregister!\n");
+		return;
+	}
+
+	if (!list_empty(&device_info->safe_buffers)) {
+		dev_err(dev,
+			"Removing from dmabounce with pending buffers!\n");
+		BUG();
+	}
+
+	bounce_destroy(&device_info->pool);
+
+	DO_STATS(if (device_info->attr_res == 0)
+			 device_remove_file(dev, &dev_attr_dmabounce_stats));
+
+	kfree(device_info);
+}
+EXPORT_SYMBOL(brcm_pcie_bounce_uninit);
+
+int brcm_pcie_bounce_register_dev(struct device *dev)
+{
+	set_dma_ops(dev, &dmabounce_ops);
+
+	return 0;
+}
+EXPORT_SYMBOL(brcm_pcie_bounce_register_dev);
+
+MODULE_AUTHOR("Phil Elwell <phil@raspberrypi.org>");
+MODULE_DESCRIPTION("Dedicate DMA bounce support for pcie-brcmstb");
+MODULE_LICENSE("GPL");
--- a/drivers/pci/controller/pcie-brcmstb.c
+++ b/drivers/pci/controller/pcie-brcmstb.c
@@ -617,28 +617,6 @@ static const struct dma_map_ops brcm_dma
 
 static void brcm_set_dma_ops(struct device *dev)
 {
-	int ret;
-
-	if (IS_ENABLED(CONFIG_ARM64)) {
-		/*
-		 * We are going to invoke get_dma_ops().  That
-		 * function, at this point in time, invokes
-		 * get_arch_dma_ops(), and for ARM64 that function
-		 * returns a pointer to dummy_dma_ops.  So then we'd
-		 * like to call arch_setup_dma_ops(), but that isn't
-		 * exported.  Instead, we call of_dma_configure(),
-		 * which is exported, and this calls
-		 * arch_setup_dma_ops().  Once we do this the call to
-		 * get_dma_ops() will work properly because
-		 * dev->dma_ops will be set.
-		 */
-		ret = of_dma_configure(dev, dev->of_node, true);
-		if (ret) {
-			dev_err(dev, "of_dma_configure() failed: %d\n", ret);
-			return;
-		}
-	}
-
 	arch_dma_ops = get_dma_ops(dev);
 	if (!arch_dma_ops) {
 		dev_err(dev, "failed to get arch_dma_ops\n");
@@ -657,12 +635,12 @@ static int brcmstb_platform_notifier(str
 	extern unsigned long max_pfn;
 	struct device *dev = __dev;
 	const char *rc_name = "0000:00:00.0";
+	int ret;
 
 	switch (event) {
 	case BUS_NOTIFY_ADD_DEVICE:
 		if (max_pfn > (bounce_threshold/PAGE_SIZE) &&
 		    strcmp(dev->kobj.name, rc_name)) {
-			int ret;
 
 			ret = brcm_pcie_bounce_register_dev(dev);
 			if (ret) {
@@ -671,6 +649,12 @@ static int brcmstb_platform_notifier(str
 					ret);
 				return ret;
 			}
+		} else if (IS_ENABLED(CONFIG_ARM64)) {
+			ret = of_dma_configure(dev, dev->of_node, true);
+			if (ret) {
+				dev_err(dev, "of_dma_configure() failed: %d\n", ret);
+				return;
+			}
 		}
 		brcm_set_dma_ops(dev);
 		return NOTIFY_OK;