xref: /openbmc/linux/drivers/dma/bcm2835-dma.c (revision b830f94f)
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 struct bcm2835_dmadev {
41 	struct dma_device ddev;
42 	void __iomem *base;
43 	struct device_dma_parameters dma_parms;
44 };
45 
46 struct bcm2835_dma_cb {
47 	uint32_t info;
48 	uint32_t src;
49 	uint32_t dst;
50 	uint32_t length;
51 	uint32_t stride;
52 	uint32_t next;
53 	uint32_t pad[2];
54 };
55 
56 struct bcm2835_cb_entry {
57 	struct bcm2835_dma_cb *cb;
58 	dma_addr_t paddr;
59 };
60 
61 struct bcm2835_chan {
62 	struct virt_dma_chan vc;
63 
64 	struct dma_slave_config	cfg;
65 	unsigned int dreq;
66 
67 	int ch;
68 	struct bcm2835_desc *desc;
69 	struct dma_pool *cb_pool;
70 
71 	void __iomem *chan_base;
72 	int irq_number;
73 	unsigned int irq_flags;
74 
75 	bool is_lite_channel;
76 };
77 
78 struct bcm2835_desc {
79 	struct bcm2835_chan *c;
80 	struct virt_dma_desc vd;
81 	enum dma_transfer_direction dir;
82 
83 	unsigned int frames;
84 	size_t size;
85 
86 	bool cyclic;
87 
88 	struct bcm2835_cb_entry cb_list[];
89 };
90 
91 #define BCM2835_DMA_CS		0x00
92 #define BCM2835_DMA_ADDR	0x04
93 #define BCM2835_DMA_TI		0x08
94 #define BCM2835_DMA_SOURCE_AD	0x0c
95 #define BCM2835_DMA_DEST_AD	0x10
96 #define BCM2835_DMA_LEN		0x14
97 #define BCM2835_DMA_STRIDE	0x18
98 #define BCM2835_DMA_NEXTCB	0x1c
99 #define BCM2835_DMA_DEBUG	0x20
100 
101 /* DMA CS Control and Status bits */
102 #define BCM2835_DMA_ACTIVE	BIT(0)  /* activate the DMA */
103 #define BCM2835_DMA_END		BIT(1)  /* current CB has ended */
104 #define BCM2835_DMA_INT		BIT(2)  /* interrupt status */
105 #define BCM2835_DMA_DREQ	BIT(3)  /* DREQ state */
106 #define BCM2835_DMA_ISPAUSED	BIT(4)  /* Pause requested or not active */
107 #define BCM2835_DMA_ISHELD	BIT(5)  /* Is held by DREQ flow control */
108 #define BCM2835_DMA_WAITING_FOR_WRITES BIT(6) /* waiting for last
109 					       * AXI-write to ack
110 					       */
111 #define BCM2835_DMA_ERR		BIT(8)
112 #define BCM2835_DMA_PRIORITY(x) ((x & 15) << 16) /* AXI priority */
113 #define BCM2835_DMA_PANIC_PRIORITY(x) ((x & 15) << 20) /* panic priority */
114 /* current value of TI.BCM2835_DMA_WAIT_RESP */
115 #define BCM2835_DMA_WAIT_FOR_WRITES BIT(28)
116 #define BCM2835_DMA_DIS_DEBUG	BIT(29) /* disable debug pause signal */
117 #define BCM2835_DMA_ABORT	BIT(30) /* Stop current CB, go to next, WO */
118 #define BCM2835_DMA_RESET	BIT(31) /* WO, self clearing */
119 
120 /* Transfer information bits - also bcm2835_cb.info field */
121 #define BCM2835_DMA_INT_EN	BIT(0)
122 #define BCM2835_DMA_TDMODE	BIT(1) /* 2D-Mode */
123 #define BCM2835_DMA_WAIT_RESP	BIT(3) /* wait for AXI-write to be acked */
124 #define BCM2835_DMA_D_INC	BIT(4)
125 #define BCM2835_DMA_D_WIDTH	BIT(5) /* 128bit writes if set */
126 #define BCM2835_DMA_D_DREQ	BIT(6) /* enable DREQ for destination */
127 #define BCM2835_DMA_D_IGNORE	BIT(7) /* ignore destination writes */
128 #define BCM2835_DMA_S_INC	BIT(8)
129 #define BCM2835_DMA_S_WIDTH	BIT(9) /* 128bit writes if set */
130 #define BCM2835_DMA_S_DREQ	BIT(10) /* enable SREQ for source */
131 #define BCM2835_DMA_S_IGNORE	BIT(11) /* ignore source reads - read 0 */
132 #define BCM2835_DMA_BURST_LENGTH(x) ((x & 15) << 12)
133 #define BCM2835_DMA_PER_MAP(x)	((x & 31) << 16) /* REQ source */
134 #define BCM2835_DMA_WAIT(x)	((x & 31) << 21) /* add DMA-wait cycles */
135 #define BCM2835_DMA_NO_WIDE_BURSTS BIT(26) /* no 2 beat write bursts */
136 
137 /* debug register bits */
138 #define BCM2835_DMA_DEBUG_LAST_NOT_SET_ERR	BIT(0)
139 #define BCM2835_DMA_DEBUG_FIFO_ERR		BIT(1)
140 #define BCM2835_DMA_DEBUG_READ_ERR		BIT(2)
141 #define BCM2835_DMA_DEBUG_OUTSTANDING_WRITES_SHIFT 4
142 #define BCM2835_DMA_DEBUG_OUTSTANDING_WRITES_BITS 4
143 #define BCM2835_DMA_DEBUG_ID_SHIFT		16
144 #define BCM2835_DMA_DEBUG_ID_BITS		9
145 #define BCM2835_DMA_DEBUG_STATE_SHIFT		16
146 #define BCM2835_DMA_DEBUG_STATE_BITS		9
147 #define BCM2835_DMA_DEBUG_VERSION_SHIFT		25
148 #define BCM2835_DMA_DEBUG_VERSION_BITS		3
149 #define BCM2835_DMA_DEBUG_LITE			BIT(28)
150 
151 /* shared registers for all dma channels */
152 #define BCM2835_DMA_INT_STATUS         0xfe0
153 #define BCM2835_DMA_ENABLE             0xff0
154 
155 #define BCM2835_DMA_DATA_TYPE_S8	1
156 #define BCM2835_DMA_DATA_TYPE_S16	2
157 #define BCM2835_DMA_DATA_TYPE_S32	4
158 #define BCM2835_DMA_DATA_TYPE_S128	16
159 
160 /* Valid only for channels 0 - 14, 15 has its own base address */
161 #define BCM2835_DMA_CHAN(n)	((n) << 8) /* Base address */
162 #define BCM2835_DMA_CHANIO(base, n) ((base) + BCM2835_DMA_CHAN(n))
163 
164 /* the max dma length for different channels */
165 #define MAX_DMA_LEN SZ_1G
166 #define MAX_LITE_DMA_LEN (SZ_64K - 4)
167 
168 static inline size_t bcm2835_dma_max_frame_length(struct bcm2835_chan *c)
169 {
170 	/* lite and normal channels have different max frame length */
171 	return c->is_lite_channel ? MAX_LITE_DMA_LEN : MAX_DMA_LEN;
172 }
173 
174 /* how many frames of max_len size do we need to transfer len bytes */
175 static inline size_t bcm2835_dma_frames_for_length(size_t len,
176 						   size_t max_len)
177 {
178 	return DIV_ROUND_UP(len, max_len);
179 }
180 
181 static inline struct bcm2835_dmadev *to_bcm2835_dma_dev(struct dma_device *d)
182 {
183 	return container_of(d, struct bcm2835_dmadev, ddev);
184 }
185 
186 static inline struct bcm2835_chan *to_bcm2835_dma_chan(struct dma_chan *c)
187 {
188 	return container_of(c, struct bcm2835_chan, vc.chan);
189 }
190 
191 static inline struct bcm2835_desc *to_bcm2835_dma_desc(
192 		struct dma_async_tx_descriptor *t)
193 {
194 	return container_of(t, struct bcm2835_desc, vd.tx);
195 }
196 
197 static void bcm2835_dma_free_cb_chain(struct bcm2835_desc *desc)
198 {
199 	size_t i;
200 
201 	for (i = 0; i < desc->frames; i++)
202 		dma_pool_free(desc->c->cb_pool, desc->cb_list[i].cb,
203 			      desc->cb_list[i].paddr);
204 
205 	kfree(desc);
206 }
207 
208 static void bcm2835_dma_desc_free(struct virt_dma_desc *vd)
209 {
210 	bcm2835_dma_free_cb_chain(
211 		container_of(vd, struct bcm2835_desc, vd));
212 }
213 
214 static void bcm2835_dma_create_cb_set_length(
215 	struct bcm2835_chan *chan,
216 	struct bcm2835_dma_cb *control_block,
217 	size_t len,
218 	size_t period_len,
219 	size_t *total_len,
220 	u32 finalextrainfo)
221 {
222 	size_t max_len = bcm2835_dma_max_frame_length(chan);
223 
224 	/* set the length taking lite-channel limitations into account */
225 	control_block->length = min_t(u32, len, max_len);
226 
227 	/* finished if we have no period_length */
228 	if (!period_len)
229 		return;
230 
231 	/*
232 	 * period_len means: that we need to generate
233 	 * transfers that are terminating at every
234 	 * multiple of period_len - this is typically
235 	 * used to set the interrupt flag in info
236 	 * which is required during cyclic transfers
237 	 */
238 
239 	/* have we filled in period_length yet? */
240 	if (*total_len + control_block->length < period_len) {
241 		/* update number of bytes in this period so far */
242 		*total_len += control_block->length;
243 		return;
244 	}
245 
246 	/* calculate the length that remains to reach period_length */
247 	control_block->length = period_len - *total_len;
248 
249 	/* reset total_length for next period */
250 	*total_len = 0;
251 
252 	/* add extrainfo bits in info */
253 	control_block->info |= finalextrainfo;
254 }
255 
256 static inline size_t bcm2835_dma_count_frames_for_sg(
257 	struct bcm2835_chan *c,
258 	struct scatterlist *sgl,
259 	unsigned int sg_len)
260 {
261 	size_t frames = 0;
262 	struct scatterlist *sgent;
263 	unsigned int i;
264 	size_t plength = bcm2835_dma_max_frame_length(c);
265 
266 	for_each_sg(sgl, sgent, sg_len, i)
267 		frames += bcm2835_dma_frames_for_length(
268 			sg_dma_len(sgent), plength);
269 
270 	return frames;
271 }
272 
273 /**
274  * bcm2835_dma_create_cb_chain - create a control block and fills data in
275  *
276  * @chan:           the @dma_chan for which we run this
277  * @direction:      the direction in which we transfer
278  * @cyclic:         it is a cyclic transfer
279  * @info:           the default info bits to apply per controlblock
280  * @frames:         number of controlblocks to allocate
281  * @src:            the src address to assign (if the S_INC bit is set
282  *                  in @info, then it gets incremented)
283  * @dst:            the dst address to assign (if the D_INC bit is set
284  *                  in @info, then it gets incremented)
285  * @buf_len:        the full buffer length (may also be 0)
286  * @period_len:     the period length when to apply @finalextrainfo
287  *                  in addition to the last transfer
288  *                  this will also break some control-blocks early
289  * @finalextrainfo: additional bits in last controlblock
290  *                  (or when period_len is reached in case of cyclic)
291  * @gfp:            the GFP flag to use for allocation
292  */
293 static struct bcm2835_desc *bcm2835_dma_create_cb_chain(
294 	struct dma_chan *chan, enum dma_transfer_direction direction,
295 	bool cyclic, u32 info, u32 finalextrainfo, size_t frames,
296 	dma_addr_t src, dma_addr_t dst, size_t buf_len,
297 	size_t period_len, gfp_t gfp)
298 {
299 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
300 	size_t len = buf_len, total_len;
301 	size_t frame;
302 	struct bcm2835_desc *d;
303 	struct bcm2835_cb_entry *cb_entry;
304 	struct bcm2835_dma_cb *control_block;
305 
306 	if (!frames)
307 		return NULL;
308 
309 	/* allocate and setup the descriptor. */
310 	d = kzalloc(struct_size(d, cb_list, frames), gfp);
311 	if (!d)
312 		return NULL;
313 
314 	d->c = c;
315 	d->dir = direction;
316 	d->cyclic = cyclic;
317 
318 	/*
319 	 * Iterate over all frames, create a control block
320 	 * for each frame and link them together.
321 	 */
322 	for (frame = 0, total_len = 0; frame < frames; d->frames++, frame++) {
323 		cb_entry = &d->cb_list[frame];
324 		cb_entry->cb = dma_pool_alloc(c->cb_pool, gfp,
325 					      &cb_entry->paddr);
326 		if (!cb_entry->cb)
327 			goto error_cb;
328 
329 		/* fill in the control block */
330 		control_block = cb_entry->cb;
331 		control_block->info = info;
332 		control_block->src = src;
333 		control_block->dst = dst;
334 		control_block->stride = 0;
335 		control_block->next = 0;
336 		/* set up length in control_block if requested */
337 		if (buf_len) {
338 			/* calculate length honoring period_length */
339 			bcm2835_dma_create_cb_set_length(
340 				c, control_block,
341 				len, period_len, &total_len,
342 				cyclic ? finalextrainfo : 0);
343 
344 			/* calculate new remaining length */
345 			len -= control_block->length;
346 		}
347 
348 		/* link this the last controlblock */
349 		if (frame)
350 			d->cb_list[frame - 1].cb->next = cb_entry->paddr;
351 
352 		/* update src and dst and length */
353 		if (src && (info & BCM2835_DMA_S_INC))
354 			src += control_block->length;
355 		if (dst && (info & BCM2835_DMA_D_INC))
356 			dst += control_block->length;
357 
358 		/* Length of total transfer */
359 		d->size += control_block->length;
360 	}
361 
362 	/* the last frame requires extra flags */
363 	d->cb_list[d->frames - 1].cb->info |= finalextrainfo;
364 
365 	/* detect a size missmatch */
366 	if (buf_len && (d->size != buf_len))
367 		goto error_cb;
368 
369 	return d;
370 error_cb:
371 	bcm2835_dma_free_cb_chain(d);
372 
373 	return NULL;
374 }
375 
376 static void bcm2835_dma_fill_cb_chain_with_sg(
377 	struct dma_chan *chan,
378 	enum dma_transfer_direction direction,
379 	struct bcm2835_cb_entry *cb,
380 	struct scatterlist *sgl,
381 	unsigned int sg_len)
382 {
383 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
384 	size_t len, max_len;
385 	unsigned int i;
386 	dma_addr_t addr;
387 	struct scatterlist *sgent;
388 
389 	max_len = bcm2835_dma_max_frame_length(c);
390 	for_each_sg(sgl, sgent, sg_len, i) {
391 		for (addr = sg_dma_address(sgent), len = sg_dma_len(sgent);
392 		     len > 0;
393 		     addr += cb->cb->length, len -= cb->cb->length, cb++) {
394 			if (direction == DMA_DEV_TO_MEM)
395 				cb->cb->dst = addr;
396 			else
397 				cb->cb->src = addr;
398 			cb->cb->length = min(len, max_len);
399 		}
400 	}
401 }
402 
403 static void bcm2835_dma_abort(struct bcm2835_chan *c)
404 {
405 	void __iomem *chan_base = c->chan_base;
406 	long int timeout = 10000;
407 
408 	/*
409 	 * A zero control block address means the channel is idle.
410 	 * (The ACTIVE flag in the CS register is not a reliable indicator.)
411 	 */
412 	if (!readl(chan_base + BCM2835_DMA_ADDR))
413 		return;
414 
415 	/* Write 0 to the active bit - Pause the DMA */
416 	writel(0, chan_base + BCM2835_DMA_CS);
417 
418 	/* Wait for any current AXI transfer to complete */
419 	while ((readl(chan_base + BCM2835_DMA_CS) &
420 		BCM2835_DMA_WAITING_FOR_WRITES) && --timeout)
421 		cpu_relax();
422 
423 	/* Peripheral might be stuck and fail to signal AXI write responses */
424 	if (!timeout)
425 		dev_err(c->vc.chan.device->dev,
426 			"failed to complete outstanding writes\n");
427 
428 	writel(BCM2835_DMA_RESET, chan_base + BCM2835_DMA_CS);
429 }
430 
431 static void bcm2835_dma_start_desc(struct bcm2835_chan *c)
432 {
433 	struct virt_dma_desc *vd = vchan_next_desc(&c->vc);
434 	struct bcm2835_desc *d;
435 
436 	if (!vd) {
437 		c->desc = NULL;
438 		return;
439 	}
440 
441 	list_del(&vd->node);
442 
443 	c->desc = d = to_bcm2835_dma_desc(&vd->tx);
444 
445 	writel(d->cb_list[0].paddr, c->chan_base + BCM2835_DMA_ADDR);
446 	writel(BCM2835_DMA_ACTIVE, c->chan_base + BCM2835_DMA_CS);
447 }
448 
449 static irqreturn_t bcm2835_dma_callback(int irq, void *data)
450 {
451 	struct bcm2835_chan *c = data;
452 	struct bcm2835_desc *d;
453 	unsigned long flags;
454 
455 	/* check the shared interrupt */
456 	if (c->irq_flags & IRQF_SHARED) {
457 		/* check if the interrupt is enabled */
458 		flags = readl(c->chan_base + BCM2835_DMA_CS);
459 		/* if not set then we are not the reason for the irq */
460 		if (!(flags & BCM2835_DMA_INT))
461 			return IRQ_NONE;
462 	}
463 
464 	spin_lock_irqsave(&c->vc.lock, flags);
465 
466 	/*
467 	 * Clear the INT flag to receive further interrupts. Keep the channel
468 	 * active in case the descriptor is cyclic or in case the client has
469 	 * already terminated the descriptor and issued a new one. (May happen
470 	 * if this IRQ handler is threaded.) If the channel is finished, it
471 	 * will remain idle despite the ACTIVE flag being set.
472 	 */
473 	writel(BCM2835_DMA_INT | BCM2835_DMA_ACTIVE,
474 	       c->chan_base + BCM2835_DMA_CS);
475 
476 	d = c->desc;
477 
478 	if (d) {
479 		if (d->cyclic) {
480 			/* call the cyclic callback */
481 			vchan_cyclic_callback(&d->vd);
482 		} else if (!readl(c->chan_base + BCM2835_DMA_ADDR)) {
483 			vchan_cookie_complete(&c->desc->vd);
484 			bcm2835_dma_start_desc(c);
485 		}
486 	}
487 
488 	spin_unlock_irqrestore(&c->vc.lock, flags);
489 
490 	return IRQ_HANDLED;
491 }
492 
493 static int bcm2835_dma_alloc_chan_resources(struct dma_chan *chan)
494 {
495 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
496 	struct device *dev = c->vc.chan.device->dev;
497 
498 	dev_dbg(dev, "Allocating DMA channel %d\n", c->ch);
499 
500 	/*
501 	 * Control blocks are 256 bit in length and must start at a 256 bit
502 	 * (32 byte) aligned address (BCM2835 ARM Peripherals, sec. 4.2.1.1).
503 	 */
504 	c->cb_pool = dma_pool_create(dev_name(dev), dev,
505 				     sizeof(struct bcm2835_dma_cb), 32, 0);
506 	if (!c->cb_pool) {
507 		dev_err(dev, "unable to allocate descriptor pool\n");
508 		return -ENOMEM;
509 	}
510 
511 	return request_irq(c->irq_number, bcm2835_dma_callback,
512 			   c->irq_flags, "DMA IRQ", c);
513 }
514 
515 static void bcm2835_dma_free_chan_resources(struct dma_chan *chan)
516 {
517 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
518 
519 	vchan_free_chan_resources(&c->vc);
520 	free_irq(c->irq_number, c);
521 	dma_pool_destroy(c->cb_pool);
522 
523 	dev_dbg(c->vc.chan.device->dev, "Freeing DMA channel %u\n", c->ch);
524 }
525 
526 static size_t bcm2835_dma_desc_size(struct bcm2835_desc *d)
527 {
528 	return d->size;
529 }
530 
531 static size_t bcm2835_dma_desc_size_pos(struct bcm2835_desc *d, dma_addr_t addr)
532 {
533 	unsigned int i;
534 	size_t size;
535 
536 	for (size = i = 0; i < d->frames; i++) {
537 		struct bcm2835_dma_cb *control_block = d->cb_list[i].cb;
538 		size_t this_size = control_block->length;
539 		dma_addr_t dma;
540 
541 		if (d->dir == DMA_DEV_TO_MEM)
542 			dma = control_block->dst;
543 		else
544 			dma = control_block->src;
545 
546 		if (size)
547 			size += this_size;
548 		else if (addr >= dma && addr < dma + this_size)
549 			size += dma + this_size - addr;
550 	}
551 
552 	return size;
553 }
554 
555 static enum dma_status bcm2835_dma_tx_status(struct dma_chan *chan,
556 	dma_cookie_t cookie, struct dma_tx_state *txstate)
557 {
558 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
559 	struct virt_dma_desc *vd;
560 	enum dma_status ret;
561 	unsigned long flags;
562 
563 	ret = dma_cookie_status(chan, cookie, txstate);
564 	if (ret == DMA_COMPLETE || !txstate)
565 		return ret;
566 
567 	spin_lock_irqsave(&c->vc.lock, flags);
568 	vd = vchan_find_desc(&c->vc, cookie);
569 	if (vd) {
570 		txstate->residue =
571 			bcm2835_dma_desc_size(to_bcm2835_dma_desc(&vd->tx));
572 	} else if (c->desc && c->desc->vd.tx.cookie == cookie) {
573 		struct bcm2835_desc *d = c->desc;
574 		dma_addr_t pos;
575 
576 		if (d->dir == DMA_MEM_TO_DEV)
577 			pos = readl(c->chan_base + BCM2835_DMA_SOURCE_AD);
578 		else if (d->dir == DMA_DEV_TO_MEM)
579 			pos = readl(c->chan_base + BCM2835_DMA_DEST_AD);
580 		else
581 			pos = 0;
582 
583 		txstate->residue = bcm2835_dma_desc_size_pos(d, pos);
584 	} else {
585 		txstate->residue = 0;
586 	}
587 
588 	spin_unlock_irqrestore(&c->vc.lock, flags);
589 
590 	return ret;
591 }
592 
593 static void bcm2835_dma_issue_pending(struct dma_chan *chan)
594 {
595 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
596 	unsigned long flags;
597 
598 	spin_lock_irqsave(&c->vc.lock, flags);
599 	if (vchan_issue_pending(&c->vc) && !c->desc)
600 		bcm2835_dma_start_desc(c);
601 
602 	spin_unlock_irqrestore(&c->vc.lock, flags);
603 }
604 
605 static struct dma_async_tx_descriptor *bcm2835_dma_prep_dma_memcpy(
606 	struct dma_chan *chan, dma_addr_t dst, dma_addr_t src,
607 	size_t len, unsigned long flags)
608 {
609 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
610 	struct bcm2835_desc *d;
611 	u32 info = BCM2835_DMA_D_INC | BCM2835_DMA_S_INC;
612 	u32 extra = BCM2835_DMA_INT_EN | BCM2835_DMA_WAIT_RESP;
613 	size_t max_len = bcm2835_dma_max_frame_length(c);
614 	size_t frames;
615 
616 	/* if src, dst or len is not given return with an error */
617 	if (!src || !dst || !len)
618 		return NULL;
619 
620 	/* calculate number of frames */
621 	frames = bcm2835_dma_frames_for_length(len, max_len);
622 
623 	/* allocate the CB chain - this also fills in the pointers */
624 	d = bcm2835_dma_create_cb_chain(chan, DMA_MEM_TO_MEM, false,
625 					info, extra, frames,
626 					src, dst, len, 0, GFP_KERNEL);
627 	if (!d)
628 		return NULL;
629 
630 	return vchan_tx_prep(&c->vc, &d->vd, flags);
631 }
632 
633 static struct dma_async_tx_descriptor *bcm2835_dma_prep_slave_sg(
634 	struct dma_chan *chan,
635 	struct scatterlist *sgl, unsigned int sg_len,
636 	enum dma_transfer_direction direction,
637 	unsigned long flags, void *context)
638 {
639 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
640 	struct bcm2835_desc *d;
641 	dma_addr_t src = 0, dst = 0;
642 	u32 info = BCM2835_DMA_WAIT_RESP;
643 	u32 extra = BCM2835_DMA_INT_EN;
644 	size_t frames;
645 
646 	if (!is_slave_direction(direction)) {
647 		dev_err(chan->device->dev,
648 			"%s: bad direction?\n", __func__);
649 		return NULL;
650 	}
651 
652 	if (c->dreq != 0)
653 		info |= BCM2835_DMA_PER_MAP(c->dreq);
654 
655 	if (direction == DMA_DEV_TO_MEM) {
656 		if (c->cfg.src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
657 			return NULL;
658 		src = c->cfg.src_addr;
659 		info |= BCM2835_DMA_S_DREQ | BCM2835_DMA_D_INC;
660 	} else {
661 		if (c->cfg.dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
662 			return NULL;
663 		dst = c->cfg.dst_addr;
664 		info |= BCM2835_DMA_D_DREQ | BCM2835_DMA_S_INC;
665 	}
666 
667 	/* count frames in sg list */
668 	frames = bcm2835_dma_count_frames_for_sg(c, sgl, sg_len);
669 
670 	/* allocate the CB chain */
671 	d = bcm2835_dma_create_cb_chain(chan, direction, false,
672 					info, extra,
673 					frames, src, dst, 0, 0,
674 					GFP_NOWAIT);
675 	if (!d)
676 		return NULL;
677 
678 	/* fill in frames with scatterlist pointers */
679 	bcm2835_dma_fill_cb_chain_with_sg(chan, direction, d->cb_list,
680 					  sgl, sg_len);
681 
682 	return vchan_tx_prep(&c->vc, &d->vd, flags);
683 }
684 
685 static struct dma_async_tx_descriptor *bcm2835_dma_prep_dma_cyclic(
686 	struct dma_chan *chan, dma_addr_t buf_addr, size_t buf_len,
687 	size_t period_len, enum dma_transfer_direction direction,
688 	unsigned long flags)
689 {
690 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
691 	struct bcm2835_desc *d;
692 	dma_addr_t src, dst;
693 	u32 info = BCM2835_DMA_WAIT_RESP;
694 	u32 extra = BCM2835_DMA_INT_EN;
695 	size_t max_len = bcm2835_dma_max_frame_length(c);
696 	size_t frames;
697 
698 	/* Grab configuration */
699 	if (!is_slave_direction(direction)) {
700 		dev_err(chan->device->dev, "%s: bad direction?\n", __func__);
701 		return NULL;
702 	}
703 
704 	if (!buf_len) {
705 		dev_err(chan->device->dev,
706 			"%s: bad buffer length (= 0)\n", __func__);
707 		return NULL;
708 	}
709 
710 	/*
711 	 * warn if buf_len is not a multiple of period_len - this may leed
712 	 * to unexpected latencies for interrupts and thus audiable clicks
713 	 */
714 	if (buf_len % period_len)
715 		dev_warn_once(chan->device->dev,
716 			      "%s: buffer_length (%zd) is not a multiple of period_len (%zd)\n",
717 			      __func__, buf_len, period_len);
718 
719 	/* Setup DREQ channel */
720 	if (c->dreq != 0)
721 		info |= BCM2835_DMA_PER_MAP(c->dreq);
722 
723 	if (direction == DMA_DEV_TO_MEM) {
724 		if (c->cfg.src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
725 			return NULL;
726 		src = c->cfg.src_addr;
727 		dst = buf_addr;
728 		info |= BCM2835_DMA_S_DREQ | BCM2835_DMA_D_INC;
729 	} else {
730 		if (c->cfg.dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
731 			return NULL;
732 		dst = c->cfg.dst_addr;
733 		src = buf_addr;
734 		info |= BCM2835_DMA_D_DREQ | BCM2835_DMA_S_INC;
735 	}
736 
737 	/* calculate number of frames */
738 	frames = /* number of periods */
739 		 DIV_ROUND_UP(buf_len, period_len) *
740 		 /* number of frames per period */
741 		 bcm2835_dma_frames_for_length(period_len, max_len);
742 
743 	/*
744 	 * allocate the CB chain
745 	 * note that we need to use GFP_NOWAIT, as the ALSA i2s dmaengine
746 	 * implementation calls prep_dma_cyclic with interrupts disabled.
747 	 */
748 	d = bcm2835_dma_create_cb_chain(chan, direction, true,
749 					info, extra,
750 					frames, src, dst, buf_len,
751 					period_len, GFP_NOWAIT);
752 	if (!d)
753 		return NULL;
754 
755 	/* wrap around into a loop */
756 	d->cb_list[d->frames - 1].cb->next = d->cb_list[0].paddr;
757 
758 	return vchan_tx_prep(&c->vc, &d->vd, flags);
759 }
760 
761 static int bcm2835_dma_slave_config(struct dma_chan *chan,
762 				    struct dma_slave_config *cfg)
763 {
764 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
765 
766 	c->cfg = *cfg;
767 
768 	return 0;
769 }
770 
771 static int bcm2835_dma_terminate_all(struct dma_chan *chan)
772 {
773 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
774 	unsigned long flags;
775 	LIST_HEAD(head);
776 
777 	spin_lock_irqsave(&c->vc.lock, flags);
778 
779 	/* stop DMA activity */
780 	if (c->desc) {
781 		vchan_terminate_vdesc(&c->desc->vd);
782 		c->desc = NULL;
783 		bcm2835_dma_abort(c);
784 	}
785 
786 	vchan_get_all_descriptors(&c->vc, &head);
787 	spin_unlock_irqrestore(&c->vc.lock, flags);
788 	vchan_dma_desc_free_list(&c->vc, &head);
789 
790 	return 0;
791 }
792 
793 static void bcm2835_dma_synchronize(struct dma_chan *chan)
794 {
795 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
796 
797 	vchan_synchronize(&c->vc);
798 }
799 
800 static int bcm2835_dma_chan_init(struct bcm2835_dmadev *d, int chan_id,
801 				 int irq, unsigned int irq_flags)
802 {
803 	struct bcm2835_chan *c;
804 
805 	c = devm_kzalloc(d->ddev.dev, sizeof(*c), GFP_KERNEL);
806 	if (!c)
807 		return -ENOMEM;
808 
809 	c->vc.desc_free = bcm2835_dma_desc_free;
810 	vchan_init(&c->vc, &d->ddev);
811 
812 	c->chan_base = BCM2835_DMA_CHANIO(d->base, chan_id);
813 	c->ch = chan_id;
814 	c->irq_number = irq;
815 	c->irq_flags = irq_flags;
816 
817 	/* check in DEBUG register if this is a LITE channel */
818 	if (readl(c->chan_base + BCM2835_DMA_DEBUG) &
819 		BCM2835_DMA_DEBUG_LITE)
820 		c->is_lite_channel = true;
821 
822 	return 0;
823 }
824 
825 static void bcm2835_dma_free(struct bcm2835_dmadev *od)
826 {
827 	struct bcm2835_chan *c, *next;
828 
829 	list_for_each_entry_safe(c, next, &od->ddev.channels,
830 				 vc.chan.device_node) {
831 		list_del(&c->vc.chan.device_node);
832 		tasklet_kill(&c->vc.task);
833 	}
834 }
835 
836 static const struct of_device_id bcm2835_dma_of_match[] = {
837 	{ .compatible = "brcm,bcm2835-dma", },
838 	{},
839 };
840 MODULE_DEVICE_TABLE(of, bcm2835_dma_of_match);
841 
842 static struct dma_chan *bcm2835_dma_xlate(struct of_phandle_args *spec,
843 					   struct of_dma *ofdma)
844 {
845 	struct bcm2835_dmadev *d = ofdma->of_dma_data;
846 	struct dma_chan *chan;
847 
848 	chan = dma_get_any_slave_channel(&d->ddev);
849 	if (!chan)
850 		return NULL;
851 
852 	/* Set DREQ from param */
853 	to_bcm2835_dma_chan(chan)->dreq = spec->args[0];
854 
855 	return chan;
856 }
857 
858 static int bcm2835_dma_probe(struct platform_device *pdev)
859 {
860 	struct bcm2835_dmadev *od;
861 	struct resource *res;
862 	void __iomem *base;
863 	int rc;
864 	int i, j;
865 	int irq[BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED + 1];
866 	int irq_flags;
867 	uint32_t chans_available;
868 	char chan_name[BCM2835_DMA_CHAN_NAME_SIZE];
869 
870 	if (!pdev->dev.dma_mask)
871 		pdev->dev.dma_mask = &pdev->dev.coherent_dma_mask;
872 
873 	rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
874 	if (rc)
875 		return rc;
876 
877 	od = devm_kzalloc(&pdev->dev, sizeof(*od), GFP_KERNEL);
878 	if (!od)
879 		return -ENOMEM;
880 
881 	pdev->dev.dma_parms = &od->dma_parms;
882 	dma_set_max_seg_size(&pdev->dev, 0x3FFFFFFF);
883 
884 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
885 	base = devm_ioremap_resource(&pdev->dev, res);
886 	if (IS_ERR(base))
887 		return PTR_ERR(base);
888 
889 	od->base = base;
890 
891 	dma_cap_set(DMA_SLAVE, od->ddev.cap_mask);
892 	dma_cap_set(DMA_PRIVATE, od->ddev.cap_mask);
893 	dma_cap_set(DMA_CYCLIC, od->ddev.cap_mask);
894 	dma_cap_set(DMA_MEMCPY, od->ddev.cap_mask);
895 	od->ddev.device_alloc_chan_resources = bcm2835_dma_alloc_chan_resources;
896 	od->ddev.device_free_chan_resources = bcm2835_dma_free_chan_resources;
897 	od->ddev.device_tx_status = bcm2835_dma_tx_status;
898 	od->ddev.device_issue_pending = bcm2835_dma_issue_pending;
899 	od->ddev.device_prep_dma_cyclic = bcm2835_dma_prep_dma_cyclic;
900 	od->ddev.device_prep_slave_sg = bcm2835_dma_prep_slave_sg;
901 	od->ddev.device_prep_dma_memcpy = bcm2835_dma_prep_dma_memcpy;
902 	od->ddev.device_config = bcm2835_dma_slave_config;
903 	od->ddev.device_terminate_all = bcm2835_dma_terminate_all;
904 	od->ddev.device_synchronize = bcm2835_dma_synchronize;
905 	od->ddev.src_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_4_BYTES);
906 	od->ddev.dst_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_4_BYTES);
907 	od->ddev.directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV) |
908 			      BIT(DMA_MEM_TO_MEM);
909 	od->ddev.residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
910 	od->ddev.dev = &pdev->dev;
911 	INIT_LIST_HEAD(&od->ddev.channels);
912 
913 	platform_set_drvdata(pdev, od);
914 
915 	/* Request DMA channel mask from device tree */
916 	if (of_property_read_u32(pdev->dev.of_node,
917 			"brcm,dma-channel-mask",
918 			&chans_available)) {
919 		dev_err(&pdev->dev, "Failed to get channel mask\n");
920 		rc = -EINVAL;
921 		goto err_no_dma;
922 	}
923 
924 	/* get irqs for each channel that we support */
925 	for (i = 0; i <= BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED; i++) {
926 		/* skip masked out channels */
927 		if (!(chans_available & (1 << i))) {
928 			irq[i] = -1;
929 			continue;
930 		}
931 
932 		/* get the named irq */
933 		snprintf(chan_name, sizeof(chan_name), "dma%i", i);
934 		irq[i] = platform_get_irq_byname(pdev, chan_name);
935 		if (irq[i] >= 0)
936 			continue;
937 
938 		/* legacy device tree case handling */
939 		dev_warn_once(&pdev->dev,
940 			      "missing interrupt-names property in device tree - legacy interpretation is used\n");
941 		/*
942 		 * in case of channel >= 11
943 		 * use the 11th interrupt and that is shared
944 		 */
945 		irq[i] = platform_get_irq(pdev, i < 11 ? i : 11);
946 	}
947 
948 	/* get irqs for each channel */
949 	for (i = 0; i <= BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED; i++) {
950 		/* skip channels without irq */
951 		if (irq[i] < 0)
952 			continue;
953 
954 		/* check if there are other channels that also use this irq */
955 		irq_flags = 0;
956 		for (j = 0; j <= BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED; j++)
957 			if ((i != j) && (irq[j] == irq[i])) {
958 				irq_flags = IRQF_SHARED;
959 				break;
960 			}
961 
962 		/* initialize the channel */
963 		rc = bcm2835_dma_chan_init(od, i, irq[i], irq_flags);
964 		if (rc)
965 			goto err_no_dma;
966 	}
967 
968 	dev_dbg(&pdev->dev, "Initialized %i DMA channels\n", i);
969 
970 	/* Device-tree DMA controller registration */
971 	rc = of_dma_controller_register(pdev->dev.of_node,
972 			bcm2835_dma_xlate, od);
973 	if (rc) {
974 		dev_err(&pdev->dev, "Failed to register DMA controller\n");
975 		goto err_no_dma;
976 	}
977 
978 	rc = dma_async_device_register(&od->ddev);
979 	if (rc) {
980 		dev_err(&pdev->dev,
981 			"Failed to register slave DMA engine device: %d\n", rc);
982 		goto err_no_dma;
983 	}
984 
985 	dev_dbg(&pdev->dev, "Load BCM2835 DMA engine driver\n");
986 
987 	return 0;
988 
989 err_no_dma:
990 	bcm2835_dma_free(od);
991 	return rc;
992 }
993 
994 static int bcm2835_dma_remove(struct platform_device *pdev)
995 {
996 	struct bcm2835_dmadev *od = platform_get_drvdata(pdev);
997 
998 	dma_async_device_unregister(&od->ddev);
999 	bcm2835_dma_free(od);
1000 
1001 	return 0;
1002 }
1003 
1004 static struct platform_driver bcm2835_dma_driver = {
1005 	.probe	= bcm2835_dma_probe,
1006 	.remove	= bcm2835_dma_remove,
1007 	.driver = {
1008 		.name = "bcm2835-dma",
1009 		.of_match_table = of_match_ptr(bcm2835_dma_of_match),
1010 	},
1011 };
1012 
1013 module_platform_driver(bcm2835_dma_driver);
1014 
1015 MODULE_ALIAS("platform:bcm2835-dma");
1016 MODULE_DESCRIPTION("BCM2835 DMA engine driver");
1017 MODULE_AUTHOR("Florian Meier <florian.meier@koalo.de>");
1018 MODULE_LICENSE("GPL");
1019