1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * BCM2835 DMA engine support
4 *
5 * Author:      Florian Meier <florian.meier@koalo.de>
6 *              Copyright 2013
7 *
8 * Based on
9 *	OMAP DMAengine support by Russell King
10 *
11 *	BCM2708 DMA Driver
12 *	Copyright (C) 2010 Broadcom
13 *
14 *	Raspberry Pi PCM I2S ALSA Driver
15 *	Copyright (c) by Phil Poole 2013
16 *
17 *	MARVELL MMP Peripheral DMA Driver
18 *	Copyright 2012 Marvell International Ltd.
19 */
20#include <linux/dmaengine.h>
21#include <linux/dma-mapping.h>
22#include <linux/dmapool.h>
23#include <linux/err.h>
24#include <linux/init.h>
25#include <linux/interrupt.h>
26#include <linux/list.h>
27#include <linux/module.h>
28#include <linux/platform_device.h>
29#include <linux/slab.h>
30#include <linux/io.h>
31#include <linux/spinlock.h>
32#include <linux/of.h>
33#include <linux/of_dma.h>
34
35#include "virt-dma.h"
36
37#define BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED 14
38#define BCM2835_DMA_CHAN_NAME_SIZE 8
39
40/**
41 * struct bcm2835_dmadev - BCM2835 DMA controller
42 * @ddev: DMA device
43 * @base: base address of register map
44 * @zero_page: bus address of zero page (to detect transactions copying from
45 *	zero page and avoid accessing memory if so)
46 */
47struct bcm2835_dmadev {
48	struct dma_device ddev;
49	void __iomem *base;
50	dma_addr_t zero_page;
51};
52
53struct bcm2835_dma_cb {
54	uint32_t info;
55	uint32_t src;
56	uint32_t dst;
57	uint32_t length;
58	uint32_t stride;
59	uint32_t next;
60	uint32_t pad[2];
61};
62
63struct bcm2835_cb_entry {
64	struct bcm2835_dma_cb *cb;
65	dma_addr_t paddr;
66};
67
68struct bcm2835_chan {
69	struct virt_dma_chan vc;
70
71	struct dma_slave_config	cfg;
72	unsigned int dreq;
73
74	int ch;
75	struct bcm2835_desc *desc;
76	struct dma_pool *cb_pool;
77
78	void __iomem *chan_base;
79	int irq_number;
80	unsigned int irq_flags;
81
82	bool is_lite_channel;
83};
84
85struct bcm2835_desc {
86	struct bcm2835_chan *c;
87	struct virt_dma_desc vd;
88	enum dma_transfer_direction dir;
89
90	unsigned int frames;
91	size_t size;
92
93	bool cyclic;
94
95	struct bcm2835_cb_entry cb_list[];
96};
97
98#define BCM2835_DMA_CS		0x00
99#define BCM2835_DMA_ADDR	0x04
100#define BCM2835_DMA_TI		0x08
101#define BCM2835_DMA_SOURCE_AD	0x0c
102#define BCM2835_DMA_DEST_AD	0x10
103#define BCM2835_DMA_LEN		0x14
104#define BCM2835_DMA_STRIDE	0x18
105#define BCM2835_DMA_NEXTCB	0x1c
106#define BCM2835_DMA_DEBUG	0x20
107
108/* DMA CS Control and Status bits */
109#define BCM2835_DMA_ACTIVE	BIT(0)  /* activate the DMA */
110#define BCM2835_DMA_END		BIT(1)  /* current CB has ended */
111#define BCM2835_DMA_INT		BIT(2)  /* interrupt status */
112#define BCM2835_DMA_DREQ	BIT(3)  /* DREQ state */
113#define BCM2835_DMA_ISPAUSED	BIT(4)  /* Pause requested or not active */
114#define BCM2835_DMA_ISHELD	BIT(5)  /* Is held by DREQ flow control */
115#define BCM2835_DMA_WAITING_FOR_WRITES BIT(6) /* waiting for last
116					       * AXI-write to ack
117					       */
118#define BCM2835_DMA_ERR		BIT(8)
119#define BCM2835_DMA_PRIORITY(x) ((x & 15) << 16) /* AXI priority */
120#define BCM2835_DMA_PANIC_PRIORITY(x) ((x & 15) << 20) /* panic priority */
121/* current value of TI.BCM2835_DMA_WAIT_RESP */
122#define BCM2835_DMA_WAIT_FOR_WRITES BIT(28)
123#define BCM2835_DMA_DIS_DEBUG	BIT(29) /* disable debug pause signal */
124#define BCM2835_DMA_ABORT	BIT(30) /* Stop current CB, go to next, WO */
125#define BCM2835_DMA_RESET	BIT(31) /* WO, self clearing */
126
127/* Transfer information bits - also bcm2835_cb.info field */
128#define BCM2835_DMA_INT_EN	BIT(0)
129#define BCM2835_DMA_TDMODE	BIT(1) /* 2D-Mode */
130#define BCM2835_DMA_WAIT_RESP	BIT(3) /* wait for AXI-write to be acked */
131#define BCM2835_DMA_D_INC	BIT(4)
132#define BCM2835_DMA_D_WIDTH	BIT(5) /* 128bit writes if set */
133#define BCM2835_DMA_D_DREQ	BIT(6) /* enable DREQ for destination */
134#define BCM2835_DMA_D_IGNORE	BIT(7) /* ignore destination writes */
135#define BCM2835_DMA_S_INC	BIT(8)
136#define BCM2835_DMA_S_WIDTH	BIT(9) /* 128bit writes if set */
137#define BCM2835_DMA_S_DREQ	BIT(10) /* enable SREQ for source */
138#define BCM2835_DMA_S_IGNORE	BIT(11) /* ignore source reads - read 0 */
139#define BCM2835_DMA_BURST_LENGTH(x) ((x & 15) << 12)
140#define BCM2835_DMA_PER_MAP(x)	((x & 31) << 16) /* REQ source */
141#define BCM2835_DMA_WAIT(x)	((x & 31) << 21) /* add DMA-wait cycles */
142#define BCM2835_DMA_NO_WIDE_BURSTS BIT(26) /* no 2 beat write bursts */
143
144/* debug register bits */
145#define BCM2835_DMA_DEBUG_LAST_NOT_SET_ERR	BIT(0)
146#define BCM2835_DMA_DEBUG_FIFO_ERR		BIT(1)
147#define BCM2835_DMA_DEBUG_READ_ERR		BIT(2)
148#define BCM2835_DMA_DEBUG_OUTSTANDING_WRITES_SHIFT 4
149#define BCM2835_DMA_DEBUG_OUTSTANDING_WRITES_BITS 4
150#define BCM2835_DMA_DEBUG_ID_SHIFT		16
151#define BCM2835_DMA_DEBUG_ID_BITS		9
152#define BCM2835_DMA_DEBUG_STATE_SHIFT		16
153#define BCM2835_DMA_DEBUG_STATE_BITS		9
154#define BCM2835_DMA_DEBUG_VERSION_SHIFT		25
155#define BCM2835_DMA_DEBUG_VERSION_BITS		3
156#define BCM2835_DMA_DEBUG_LITE			BIT(28)
157
158/* shared registers for all dma channels */
159#define BCM2835_DMA_INT_STATUS         0xfe0
160#define BCM2835_DMA_ENABLE             0xff0
161
162#define BCM2835_DMA_DATA_TYPE_S8	1
163#define BCM2835_DMA_DATA_TYPE_S16	2
164#define BCM2835_DMA_DATA_TYPE_S32	4
165#define BCM2835_DMA_DATA_TYPE_S128	16
166
167/* Valid only for channels 0 - 14, 15 has its own base address */
168#define BCM2835_DMA_CHAN(n)	((n) << 8) /* Base address */
169#define BCM2835_DMA_CHANIO(base, n) ((base) + BCM2835_DMA_CHAN(n))
170
171/* the max dma length for different channels */
172#define MAX_DMA_LEN SZ_1G
173#define MAX_LITE_DMA_LEN (SZ_64K - 4)
174
175static inline size_t bcm2835_dma_max_frame_length(struct bcm2835_chan *c)
176{
177	/* lite and normal channels have different max frame length */
178	return c->is_lite_channel ? MAX_LITE_DMA_LEN : MAX_DMA_LEN;
179}
180
181/* how many frames of max_len size do we need to transfer len bytes */
182static inline size_t bcm2835_dma_frames_for_length(size_t len,
183						   size_t max_len)
184{
185	return DIV_ROUND_UP(len, max_len);
186}
187
188static inline struct bcm2835_dmadev *to_bcm2835_dma_dev(struct dma_device *d)
189{
190	return container_of(d, struct bcm2835_dmadev, ddev);
191}
192
193static inline struct bcm2835_chan *to_bcm2835_dma_chan(struct dma_chan *c)
194{
195	return container_of(c, struct bcm2835_chan, vc.chan);
196}
197
198static inline struct bcm2835_desc *to_bcm2835_dma_desc(
199		struct dma_async_tx_descriptor *t)
200{
201	return container_of(t, struct bcm2835_desc, vd.tx);
202}
203
204static void bcm2835_dma_free_cb_chain(struct bcm2835_desc *desc)
205{
206	size_t i;
207
208	for (i = 0; i < desc->frames; i++)
209		dma_pool_free(desc->c->cb_pool, desc->cb_list[i].cb,
210			      desc->cb_list[i].paddr);
211
212	kfree(desc);
213}
214
215static void bcm2835_dma_desc_free(struct virt_dma_desc *vd)
216{
217	bcm2835_dma_free_cb_chain(
218		container_of(vd, struct bcm2835_desc, vd));
219}
220
221static void bcm2835_dma_create_cb_set_length(
222	struct bcm2835_chan *chan,
223	struct bcm2835_dma_cb *control_block,
224	size_t len,
225	size_t period_len,
226	size_t *total_len,
227	u32 finalextrainfo)
228{
229	size_t max_len = bcm2835_dma_max_frame_length(chan);
230
231	/* set the length taking lite-channel limitations into account */
232	control_block->length = min_t(u32, len, max_len);
233
234	/* finished if we have no period_length */
235	if (!period_len)
236		return;
237
238	/*
239	 * period_len means: that we need to generate
240	 * transfers that are terminating at every
241	 * multiple of period_len - this is typically
242	 * used to set the interrupt flag in info
243	 * which is required during cyclic transfers
244	 */
245
246	/* have we filled in period_length yet? */
247	if (*total_len + control_block->length < period_len) {
248		/* update number of bytes in this period so far */
249		*total_len += control_block->length;
250		return;
251	}
252
253	/* calculate the length that remains to reach period_length */
254	control_block->length = period_len - *total_len;
255
256	/* reset total_length for next period */
257	*total_len = 0;
258
259	/* add extrainfo bits in info */
260	control_block->info |= finalextrainfo;
261}
262
263static inline size_t bcm2835_dma_count_frames_for_sg(
264	struct bcm2835_chan *c,
265	struct scatterlist *sgl,
266	unsigned int sg_len)
267{
268	size_t frames = 0;
269	struct scatterlist *sgent;
270	unsigned int i;
271	size_t plength = bcm2835_dma_max_frame_length(c);
272
273	for_each_sg(sgl, sgent, sg_len, i)
274		frames += bcm2835_dma_frames_for_length(
275			sg_dma_len(sgent), plength);
276
277	return frames;
278}
279
280/**
281 * bcm2835_dma_create_cb_chain - create a control block and fills data in
282 *
283 * @chan:           the @dma_chan for which we run this
284 * @direction:      the direction in which we transfer
285 * @cyclic:         it is a cyclic transfer
286 * @info:           the default info bits to apply per controlblock
287 * @frames:         number of controlblocks to allocate
288 * @src:            the src address to assign (if the S_INC bit is set
289 *                  in @info, then it gets incremented)
290 * @dst:            the dst address to assign (if the D_INC bit is set
291 *                  in @info, then it gets incremented)
292 * @buf_len:        the full buffer length (may also be 0)
293 * @period_len:     the period length when to apply @finalextrainfo
294 *                  in addition to the last transfer
295 *                  this will also break some control-blocks early
296 * @finalextrainfo: additional bits in last controlblock
297 *                  (or when period_len is reached in case of cyclic)
298 * @gfp:            the GFP flag to use for allocation
299 */
300static struct bcm2835_desc *bcm2835_dma_create_cb_chain(
301	struct dma_chan *chan, enum dma_transfer_direction direction,
302	bool cyclic, u32 info, u32 finalextrainfo, size_t frames,
303	dma_addr_t src, dma_addr_t dst, size_t buf_len,
304	size_t period_len, gfp_t gfp)
305{
306	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
307	size_t len = buf_len, total_len;
308	size_t frame;
309	struct bcm2835_desc *d;
310	struct bcm2835_cb_entry *cb_entry;
311	struct bcm2835_dma_cb *control_block;
312
313	if (!frames)
314		return NULL;
315
316	/* allocate and setup the descriptor. */
317	d = kzalloc(struct_size(d, cb_list, frames), gfp);
318	if (!d)
319		return NULL;
320
321	d->c = c;
322	d->dir = direction;
323	d->cyclic = cyclic;
324
325	/*
326	 * Iterate over all frames, create a control block
327	 * for each frame and link them together.
328	 */
329	for (frame = 0, total_len = 0; frame < frames; d->frames++, frame++) {
330		cb_entry = &d->cb_list[frame];
331		cb_entry->cb = dma_pool_alloc(c->cb_pool, gfp,
332					      &cb_entry->paddr);
333		if (!cb_entry->cb)
334			goto error_cb;
335
336		/* fill in the control block */
337		control_block = cb_entry->cb;
338		control_block->info = info;
339		control_block->src = src;
340		control_block->dst = dst;
341		control_block->stride = 0;
342		control_block->next = 0;
343		/* set up length in control_block if requested */
344		if (buf_len) {
345			/* calculate length honoring period_length */
346			bcm2835_dma_create_cb_set_length(
347				c, control_block,
348				len, period_len, &total_len,
349				cyclic ? finalextrainfo : 0);
350
351			/* calculate new remaining length */
352			len -= control_block->length;
353		}
354
355		/* link this the last controlblock */
356		if (frame)
357			d->cb_list[frame - 1].cb->next = cb_entry->paddr;
358
359		/* update src and dst and length */
360		if (src && (info & BCM2835_DMA_S_INC))
361			src += control_block->length;
362		if (dst && (info & BCM2835_DMA_D_INC))
363			dst += control_block->length;
364
365		/* Length of total transfer */
366		d->size += control_block->length;
367	}
368
369	/* the last frame requires extra flags */
370	d->cb_list[d->frames - 1].cb->info |= finalextrainfo;
371
372	/* detect a size missmatch */
373	if (buf_len && (d->size != buf_len))
374		goto error_cb;
375
376	return d;
377error_cb:
378	bcm2835_dma_free_cb_chain(d);
379
380	return NULL;
381}
382
383static void bcm2835_dma_fill_cb_chain_with_sg(
384	struct dma_chan *chan,
385	enum dma_transfer_direction direction,
386	struct bcm2835_cb_entry *cb,
387	struct scatterlist *sgl,
388	unsigned int sg_len)
389{
390	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
391	size_t len, max_len;
392	unsigned int i;
393	dma_addr_t addr;
394	struct scatterlist *sgent;
395
396	max_len = bcm2835_dma_max_frame_length(c);
397	for_each_sg(sgl, sgent, sg_len, i) {
398		for (addr = sg_dma_address(sgent), len = sg_dma_len(sgent);
399		     len > 0;
400		     addr += cb->cb->length, len -= cb->cb->length, cb++) {
401			if (direction == DMA_DEV_TO_MEM)
402				cb->cb->dst = addr;
403			else
404				cb->cb->src = addr;
405			cb->cb->length = min(len, max_len);
406		}
407	}
408}
409
410static void bcm2835_dma_abort(struct bcm2835_chan *c)
411{
412	void __iomem *chan_base = c->chan_base;
413	long int timeout = 10000;
414
415	/*
416	 * A zero control block address means the channel is idle.
417	 * (The ACTIVE flag in the CS register is not a reliable indicator.)
418	 */
419	if (!readl(chan_base + BCM2835_DMA_ADDR))
420		return;
421
422	/* Write 0 to the active bit - Pause the DMA */
423	writel(0, chan_base + BCM2835_DMA_CS);
424
425	/* Wait for any current AXI transfer to complete */
426	while ((readl(chan_base + BCM2835_DMA_CS) &
427		BCM2835_DMA_WAITING_FOR_WRITES) && --timeout)
428		cpu_relax();
429
430	/* Peripheral might be stuck and fail to signal AXI write responses */
431	if (!timeout)
432		dev_err(c->vc.chan.device->dev,
433			"failed to complete outstanding writes\n");
434
435	writel(BCM2835_DMA_RESET, chan_base + BCM2835_DMA_CS);
436}
437
438static void bcm2835_dma_start_desc(struct bcm2835_chan *c)
439{
440	struct virt_dma_desc *vd = vchan_next_desc(&c->vc);
441	struct bcm2835_desc *d;
442
443	if (!vd) {
444		c->desc = NULL;
445		return;
446	}
447
448	list_del(&vd->node);
449
450	c->desc = d = to_bcm2835_dma_desc(&vd->tx);
451
452	writel(d->cb_list[0].paddr, c->chan_base + BCM2835_DMA_ADDR);
453	writel(BCM2835_DMA_ACTIVE, c->chan_base + BCM2835_DMA_CS);
454}
455
456static irqreturn_t bcm2835_dma_callback(int irq, void *data)
457{
458	struct bcm2835_chan *c = data;
459	struct bcm2835_desc *d;
460	unsigned long flags;
461
462	/* check the shared interrupt */
463	if (c->irq_flags & IRQF_SHARED) {
464		/* check if the interrupt is enabled */
465		flags = readl(c->chan_base + BCM2835_DMA_CS);
466		/* if not set then we are not the reason for the irq */
467		if (!(flags & BCM2835_DMA_INT))
468			return IRQ_NONE;
469	}
470
471	spin_lock_irqsave(&c->vc.lock, flags);
472
473	/*
474	 * Clear the INT flag to receive further interrupts. Keep the channel
475	 * active in case the descriptor is cyclic or in case the client has
476	 * already terminated the descriptor and issued a new one. (May happen
477	 * if this IRQ handler is threaded.) If the channel is finished, it
478	 * will remain idle despite the ACTIVE flag being set.
479	 */
480	writel(BCM2835_DMA_INT | BCM2835_DMA_ACTIVE,
481	       c->chan_base + BCM2835_DMA_CS);
482
483	d = c->desc;
484
485	if (d) {
486		if (d->cyclic) {
487			/* call the cyclic callback */
488			vchan_cyclic_callback(&d->vd);
489		} else if (!readl(c->chan_base + BCM2835_DMA_ADDR)) {
490			vchan_cookie_complete(&c->desc->vd);
491			bcm2835_dma_start_desc(c);
492		}
493	}
494
495	spin_unlock_irqrestore(&c->vc.lock, flags);
496
497	return IRQ_HANDLED;
498}
499
500static int bcm2835_dma_alloc_chan_resources(struct dma_chan *chan)
501{
502	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
503	struct device *dev = c->vc.chan.device->dev;
504
505	dev_dbg(dev, "Allocating DMA channel %d\n", c->ch);
506
507	/*
508	 * Control blocks are 256 bit in length and must start at a 256 bit
509	 * (32 byte) aligned address (BCM2835 ARM Peripherals, sec. 4.2.1.1).
510	 */
511	c->cb_pool = dma_pool_create(dev_name(dev), dev,
512				     sizeof(struct bcm2835_dma_cb), 32, 0);
513	if (!c->cb_pool) {
514		dev_err(dev, "unable to allocate descriptor pool\n");
515		return -ENOMEM;
516	}
517
518	return request_irq(c->irq_number, bcm2835_dma_callback,
519			   c->irq_flags, "DMA IRQ", c);
520}
521
522static void bcm2835_dma_free_chan_resources(struct dma_chan *chan)
523{
524	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
525
526	vchan_free_chan_resources(&c->vc);
527	free_irq(c->irq_number, c);
528	dma_pool_destroy(c->cb_pool);
529
530	dev_dbg(c->vc.chan.device->dev, "Freeing DMA channel %u\n", c->ch);
531}
532
533static size_t bcm2835_dma_desc_size(struct bcm2835_desc *d)
534{
535	return d->size;
536}
537
538static size_t bcm2835_dma_desc_size_pos(struct bcm2835_desc *d, dma_addr_t addr)
539{
540	unsigned int i;
541	size_t size;
542
543	for (size = i = 0; i < d->frames; i++) {
544		struct bcm2835_dma_cb *control_block = d->cb_list[i].cb;
545		size_t this_size = control_block->length;
546		dma_addr_t dma;
547
548		if (d->dir == DMA_DEV_TO_MEM)
549			dma = control_block->dst;
550		else
551			dma = control_block->src;
552
553		if (size)
554			size += this_size;
555		else if (addr >= dma && addr < dma + this_size)
556			size += dma + this_size - addr;
557	}
558
559	return size;
560}
561
562static enum dma_status bcm2835_dma_tx_status(struct dma_chan *chan,
563	dma_cookie_t cookie, struct dma_tx_state *txstate)
564{
565	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
566	struct virt_dma_desc *vd;
567	enum dma_status ret;
568	unsigned long flags;
569
570	ret = dma_cookie_status(chan, cookie, txstate);
571	if (ret == DMA_COMPLETE || !txstate)
572		return ret;
573
574	spin_lock_irqsave(&c->vc.lock, flags);
575	vd = vchan_find_desc(&c->vc, cookie);
576	if (vd) {
577		txstate->residue =
578			bcm2835_dma_desc_size(to_bcm2835_dma_desc(&vd->tx));
579	} else if (c->desc && c->desc->vd.tx.cookie == cookie) {
580		struct bcm2835_desc *d = c->desc;
581		dma_addr_t pos;
582
583		if (d->dir == DMA_MEM_TO_DEV)
584			pos = readl(c->chan_base + BCM2835_DMA_SOURCE_AD);
585		else if (d->dir == DMA_DEV_TO_MEM)
586			pos = readl(c->chan_base + BCM2835_DMA_DEST_AD);
587		else
588			pos = 0;
589
590		txstate->residue = bcm2835_dma_desc_size_pos(d, pos);
591	} else {
592		txstate->residue = 0;
593	}
594
595	spin_unlock_irqrestore(&c->vc.lock, flags);
596
597	return ret;
598}
599
600static void bcm2835_dma_issue_pending(struct dma_chan *chan)
601{
602	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
603	unsigned long flags;
604
605	spin_lock_irqsave(&c->vc.lock, flags);
606	if (vchan_issue_pending(&c->vc) && !c->desc)
607		bcm2835_dma_start_desc(c);
608
609	spin_unlock_irqrestore(&c->vc.lock, flags);
610}
611
612static struct dma_async_tx_descriptor *bcm2835_dma_prep_dma_memcpy(
613	struct dma_chan *chan, dma_addr_t dst, dma_addr_t src,
614	size_t len, unsigned long flags)
615{
616	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
617	struct bcm2835_desc *d;
618	u32 info = BCM2835_DMA_D_INC | BCM2835_DMA_S_INC;
619	u32 extra = BCM2835_DMA_INT_EN | BCM2835_DMA_WAIT_RESP;
620	size_t max_len = bcm2835_dma_max_frame_length(c);
621	size_t frames;
622
623	/* if src, dst or len is not given return with an error */
624	if (!src || !dst || !len)
625		return NULL;
626
627	/* calculate number of frames */
628	frames = bcm2835_dma_frames_for_length(len, max_len);
629
630	/* allocate the CB chain - this also fills in the pointers */
631	d = bcm2835_dma_create_cb_chain(chan, DMA_MEM_TO_MEM, false,
632					info, extra, frames,
633					src, dst, len, 0, GFP_KERNEL);
634	if (!d)
635		return NULL;
636
637	return vchan_tx_prep(&c->vc, &d->vd, flags);
638}
639
640static struct dma_async_tx_descriptor *bcm2835_dma_prep_slave_sg(
641	struct dma_chan *chan,
642	struct scatterlist *sgl, unsigned int sg_len,
643	enum dma_transfer_direction direction,
644	unsigned long flags, void *context)
645{
646	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
647	struct bcm2835_desc *d;
648	dma_addr_t src = 0, dst = 0;
649	u32 info = BCM2835_DMA_WAIT_RESP;
650	u32 extra = BCM2835_DMA_INT_EN;
651	size_t frames;
652
653	if (!is_slave_direction(direction)) {
654		dev_err(chan->device->dev,
655			"%s: bad direction?\n", __func__);
656		return NULL;
657	}
658
659	if (c->dreq != 0)
660		info |= BCM2835_DMA_PER_MAP(c->dreq);
661
662	if (direction == DMA_DEV_TO_MEM) {
663		if (c->cfg.src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
664			return NULL;
665		src = c->cfg.src_addr;
666		info |= BCM2835_DMA_S_DREQ | BCM2835_DMA_D_INC;
667	} else {
668		if (c->cfg.dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
669			return NULL;
670		dst = c->cfg.dst_addr;
671		info |= BCM2835_DMA_D_DREQ | BCM2835_DMA_S_INC;
672	}
673
674	/* count frames in sg list */
675	frames = bcm2835_dma_count_frames_for_sg(c, sgl, sg_len);
676
677	/* allocate the CB chain */
678	d = bcm2835_dma_create_cb_chain(chan, direction, false,
679					info, extra,
680					frames, src, dst, 0, 0,
681					GFP_NOWAIT);
682	if (!d)
683		return NULL;
684
685	/* fill in frames with scatterlist pointers */
686	bcm2835_dma_fill_cb_chain_with_sg(chan, direction, d->cb_list,
687					  sgl, sg_len);
688
689	return vchan_tx_prep(&c->vc, &d->vd, flags);
690}
691
692static struct dma_async_tx_descriptor *bcm2835_dma_prep_dma_cyclic(
693	struct dma_chan *chan, dma_addr_t buf_addr, size_t buf_len,
694	size_t period_len, enum dma_transfer_direction direction,
695	unsigned long flags)
696{
697	struct bcm2835_dmadev *od = to_bcm2835_dma_dev(chan->device);
698	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
699	struct bcm2835_desc *d;
700	dma_addr_t src, dst;
701	u32 info = BCM2835_DMA_WAIT_RESP;
702	u32 extra = 0;
703	size_t max_len = bcm2835_dma_max_frame_length(c);
704	size_t frames;
705
706	/* Grab configuration */
707	if (!is_slave_direction(direction)) {
708		dev_err(chan->device->dev, "%s: bad direction?\n", __func__);
709		return NULL;
710	}
711
712	if (!buf_len) {
713		dev_err(chan->device->dev,
714			"%s: bad buffer length (= 0)\n", __func__);
715		return NULL;
716	}
717
718	if (flags & DMA_PREP_INTERRUPT)
719		extra |= BCM2835_DMA_INT_EN;
720	else
721		period_len = buf_len;
722
723	/*
724	 * warn if buf_len is not a multiple of period_len - this may leed
725	 * to unexpected latencies for interrupts and thus audiable clicks
726	 */
727	if (buf_len % period_len)
728		dev_warn_once(chan->device->dev,
729			      "%s: buffer_length (%zd) is not a multiple of period_len (%zd)\n",
730			      __func__, buf_len, period_len);
731
732	/* Setup DREQ channel */
733	if (c->dreq != 0)
734		info |= BCM2835_DMA_PER_MAP(c->dreq);
735
736	if (direction == DMA_DEV_TO_MEM) {
737		if (c->cfg.src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
738			return NULL;
739		src = c->cfg.src_addr;
740		dst = buf_addr;
741		info |= BCM2835_DMA_S_DREQ | BCM2835_DMA_D_INC;
742	} else {
743		if (c->cfg.dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
744			return NULL;
745		dst = c->cfg.dst_addr;
746		src = buf_addr;
747		info |= BCM2835_DMA_D_DREQ | BCM2835_DMA_S_INC;
748
749		/* non-lite channels can write zeroes w/o accessing memory */
750		if (buf_addr == od->zero_page && !c->is_lite_channel)
751			info |= BCM2835_DMA_S_IGNORE;
752	}
753
754	/* calculate number of frames */
755	frames = /* number of periods */
756		 DIV_ROUND_UP(buf_len, period_len) *
757		 /* number of frames per period */
758		 bcm2835_dma_frames_for_length(period_len, max_len);
759
760	/*
761	 * allocate the CB chain
762	 * note that we need to use GFP_NOWAIT, as the ALSA i2s dmaengine
763	 * implementation calls prep_dma_cyclic with interrupts disabled.
764	 */
765	d = bcm2835_dma_create_cb_chain(chan, direction, true,
766					info, extra,
767					frames, src, dst, buf_len,
768					period_len, GFP_NOWAIT);
769	if (!d)
770		return NULL;
771
772	/* wrap around into a loop */
773	d->cb_list[d->frames - 1].cb->next = d->cb_list[0].paddr;
774
775	return vchan_tx_prep(&c->vc, &d->vd, flags);
776}
777
778static int bcm2835_dma_slave_config(struct dma_chan *chan,
779				    struct dma_slave_config *cfg)
780{
781	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
782
783	c->cfg = *cfg;
784
785	return 0;
786}
787
788static int bcm2835_dma_terminate_all(struct dma_chan *chan)
789{
790	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
791	unsigned long flags;
792	LIST_HEAD(head);
793
794	spin_lock_irqsave(&c->vc.lock, flags);
795
796	/* stop DMA activity */
797	if (c->desc) {
798		vchan_terminate_vdesc(&c->desc->vd);
799		c->desc = NULL;
800		bcm2835_dma_abort(c);
801	}
802
803	vchan_get_all_descriptors(&c->vc, &head);
804	spin_unlock_irqrestore(&c->vc.lock, flags);
805	vchan_dma_desc_free_list(&c->vc, &head);
806
807	return 0;
808}
809
810static void bcm2835_dma_synchronize(struct dma_chan *chan)
811{
812	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
813
814	vchan_synchronize(&c->vc);
815}
816
817static int bcm2835_dma_chan_init(struct bcm2835_dmadev *d, int chan_id,
818				 int irq, unsigned int irq_flags)
819{
820	struct bcm2835_chan *c;
821
822	c = devm_kzalloc(d->ddev.dev, sizeof(*c), GFP_KERNEL);
823	if (!c)
824		return -ENOMEM;
825
826	c->vc.desc_free = bcm2835_dma_desc_free;
827	vchan_init(&c->vc, &d->ddev);
828
829	c->chan_base = BCM2835_DMA_CHANIO(d->base, chan_id);
830	c->ch = chan_id;
831	c->irq_number = irq;
832	c->irq_flags = irq_flags;
833
834	/* check in DEBUG register if this is a LITE channel */
835	if (readl(c->chan_base + BCM2835_DMA_DEBUG) &
836		BCM2835_DMA_DEBUG_LITE)
837		c->is_lite_channel = true;
838
839	return 0;
840}
841
842static void bcm2835_dma_free(struct bcm2835_dmadev *od)
843{
844	struct bcm2835_chan *c, *next;
845
846	list_for_each_entry_safe(c, next, &od->ddev.channels,
847				 vc.chan.device_node) {
848		list_del(&c->vc.chan.device_node);
849		tasklet_kill(&c->vc.task);
850	}
851
852	dma_unmap_page_attrs(od->ddev.dev, od->zero_page, PAGE_SIZE,
853			     DMA_TO_DEVICE, DMA_ATTR_SKIP_CPU_SYNC);
854}
855
856static const struct of_device_id bcm2835_dma_of_match[] = {
857	{ .compatible = "brcm,bcm2835-dma", },
858	{},
859};
860MODULE_DEVICE_TABLE(of, bcm2835_dma_of_match);
861
862static struct dma_chan *bcm2835_dma_xlate(struct of_phandle_args *spec,
863					   struct of_dma *ofdma)
864{
865	struct bcm2835_dmadev *d = ofdma->of_dma_data;
866	struct dma_chan *chan;
867
868	chan = dma_get_any_slave_channel(&d->ddev);
869	if (!chan)
870		return NULL;
871
872	/* Set DREQ from param */
873	to_bcm2835_dma_chan(chan)->dreq = spec->args[0];
874
875	return chan;
876}
877
878static int bcm2835_dma_probe(struct platform_device *pdev)
879{
880	struct bcm2835_dmadev *od;
881	void __iomem *base;
882	int rc;
883	int i, j;
884	int irq[BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED + 1];
885	int irq_flags;
886	uint32_t chans_available;
887	char chan_name[BCM2835_DMA_CHAN_NAME_SIZE];
888
889	if (!pdev->dev.dma_mask)
890		pdev->dev.dma_mask = &pdev->dev.coherent_dma_mask;
891
892	rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
893	if (rc) {
894		dev_err(&pdev->dev, "Unable to set DMA mask\n");
895		return rc;
896	}
897
898	od = devm_kzalloc(&pdev->dev, sizeof(*od), GFP_KERNEL);
899	if (!od)
900		return -ENOMEM;
901
902	dma_set_max_seg_size(&pdev->dev, 0x3FFFFFFF);
903
904	base = devm_platform_ioremap_resource(pdev, 0);
905	if (IS_ERR(base))
906		return PTR_ERR(base);
907
908	od->base = base;
909
910	dma_cap_set(DMA_SLAVE, od->ddev.cap_mask);
911	dma_cap_set(DMA_PRIVATE, od->ddev.cap_mask);
912	dma_cap_set(DMA_CYCLIC, od->ddev.cap_mask);
913	dma_cap_set(DMA_MEMCPY, od->ddev.cap_mask);
914	od->ddev.device_alloc_chan_resources = bcm2835_dma_alloc_chan_resources;
915	od->ddev.device_free_chan_resources = bcm2835_dma_free_chan_resources;
916	od->ddev.device_tx_status = bcm2835_dma_tx_status;
917	od->ddev.device_issue_pending = bcm2835_dma_issue_pending;
918	od->ddev.device_prep_dma_cyclic = bcm2835_dma_prep_dma_cyclic;
919	od->ddev.device_prep_slave_sg = bcm2835_dma_prep_slave_sg;
920	od->ddev.device_prep_dma_memcpy = bcm2835_dma_prep_dma_memcpy;
921	od->ddev.device_config = bcm2835_dma_slave_config;
922	od->ddev.device_terminate_all = bcm2835_dma_terminate_all;
923	od->ddev.device_synchronize = bcm2835_dma_synchronize;
924	od->ddev.src_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_4_BYTES);
925	od->ddev.dst_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_4_BYTES);
926	od->ddev.directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV) |
927			      BIT(DMA_MEM_TO_MEM);
928	od->ddev.residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
929	od->ddev.descriptor_reuse = true;
930	od->ddev.dev = &pdev->dev;
931	INIT_LIST_HEAD(&od->ddev.channels);
932
933	platform_set_drvdata(pdev, od);
934
935	od->zero_page = dma_map_page_attrs(od->ddev.dev, ZERO_PAGE(0), 0,
936					   PAGE_SIZE, DMA_TO_DEVICE,
937					   DMA_ATTR_SKIP_CPU_SYNC);
938	if (dma_mapping_error(od->ddev.dev, od->zero_page)) {
939		dev_err(&pdev->dev, "Failed to map zero page\n");
940		return -ENOMEM;
941	}
942
943	/* Request DMA channel mask from device tree */
944	if (of_property_read_u32(pdev->dev.of_node,
945			"brcm,dma-channel-mask",
946			&chans_available)) {
947		dev_err(&pdev->dev, "Failed to get channel mask\n");
948		rc = -EINVAL;
949		goto err_no_dma;
950	}
951
952	/* get irqs for each channel that we support */
953	for (i = 0; i <= BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED; i++) {
954		/* skip masked out channels */
955		if (!(chans_available & (1 << i))) {
956			irq[i] = -1;
957			continue;
958		}
959
960		/* get the named irq */
961		snprintf(chan_name, sizeof(chan_name), "dma%i", i);
962		irq[i] = platform_get_irq_byname(pdev, chan_name);
963		if (irq[i] >= 0)
964			continue;
965
966		/* legacy device tree case handling */
967		dev_warn_once(&pdev->dev,
968			      "missing interrupt-names property in device tree - legacy interpretation is used\n");
969		/*
970		 * in case of channel >= 11
971		 * use the 11th interrupt and that is shared
972		 */
973		irq[i] = platform_get_irq(pdev, i < 11 ? i : 11);
974	}
975
976	/* get irqs for each channel */
977	for (i = 0; i <= BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED; i++) {
978		/* skip channels without irq */
979		if (irq[i] < 0)
980			continue;
981
982		/* check if there are other channels that also use this irq */
983		irq_flags = 0;
984		for (j = 0; j <= BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED; j++)
985			if ((i != j) && (irq[j] == irq[i])) {
986				irq_flags = IRQF_SHARED;
987				break;
988			}
989
990		/* initialize the channel */
991		rc = bcm2835_dma_chan_init(od, i, irq[i], irq_flags);
992		if (rc)
993			goto err_no_dma;
994	}
995
996	dev_dbg(&pdev->dev, "Initialized %i DMA channels\n", i);
997
998	/* Device-tree DMA controller registration */
999	rc = of_dma_controller_register(pdev->dev.of_node,
1000			bcm2835_dma_xlate, od);
1001	if (rc) {
1002		dev_err(&pdev->dev, "Failed to register DMA controller\n");
1003		goto err_no_dma;
1004	}
1005
1006	rc = dma_async_device_register(&od->ddev);
1007	if (rc) {
1008		dev_err(&pdev->dev,
1009			"Failed to register slave DMA engine device: %d\n", rc);
1010		goto err_no_dma;
1011	}
1012
1013	dev_dbg(&pdev->dev, "Load BCM2835 DMA engine driver\n");
1014
1015	return 0;
1016
1017err_no_dma:
1018	bcm2835_dma_free(od);
1019	return rc;
1020}
1021
1022static void bcm2835_dma_remove(struct platform_device *pdev)
1023{
1024	struct bcm2835_dmadev *od = platform_get_drvdata(pdev);
1025
1026	dma_async_device_unregister(&od->ddev);
1027	bcm2835_dma_free(od);
1028}
1029
1030static struct platform_driver bcm2835_dma_driver = {
1031	.probe	= bcm2835_dma_probe,
1032	.remove_new = bcm2835_dma_remove,
1033	.driver = {
1034		.name = "bcm2835-dma",
1035		.of_match_table = of_match_ptr(bcm2835_dma_of_match),
1036	},
1037};
1038
1039module_platform_driver(bcm2835_dma_driver);
1040
1041MODULE_ALIAS("platform:bcm2835-dma");
1042MODULE_DESCRIPTION("BCM2835 DMA engine driver");
1043MODULE_AUTHOR("Florian Meier <florian.meier@koalo.de>");
1044MODULE_LICENSE("GPL");
1045