xref: /openbmc/linux/drivers/dma/bcm2835-dma.c (revision c67e8ec0)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * BCM2835 DMA engine support
4  *
5  * This driver only supports cyclic DMA transfers
6  * as needed for the I2S module.
7  *
8  * Author:      Florian Meier <florian.meier@koalo.de>
9  *              Copyright 2013
10  *
11  * Based on
12  *	OMAP DMAengine support by Russell King
13  *
14  *	BCM2708 DMA Driver
15  *	Copyright (C) 2010 Broadcom
16  *
17  *	Raspberry Pi PCM I2S ALSA Driver
18  *	Copyright (c) by Phil Poole 2013
19  *
20  *	MARVELL MMP Peripheral DMA Driver
21  *	Copyright 2012 Marvell International Ltd.
22  */
23 #include <linux/dmaengine.h>
24 #include <linux/dma-mapping.h>
25 #include <linux/dmapool.h>
26 #include <linux/err.h>
27 #include <linux/init.h>
28 #include <linux/interrupt.h>
29 #include <linux/list.h>
30 #include <linux/module.h>
31 #include <linux/platform_device.h>
32 #include <linux/slab.h>
33 #include <linux/io.h>
34 #include <linux/spinlock.h>
35 #include <linux/of.h>
36 #include <linux/of_dma.h>
37 
38 #include "virt-dma.h"
39 
40 #define BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED 14
41 #define BCM2835_DMA_CHAN_NAME_SIZE 8
42 
43 struct bcm2835_dmadev {
44 	struct dma_device ddev;
45 	spinlock_t lock;
46 	void __iomem *base;
47 	struct device_dma_parameters dma_parms;
48 };
49 
50 struct bcm2835_dma_cb {
51 	uint32_t info;
52 	uint32_t src;
53 	uint32_t dst;
54 	uint32_t length;
55 	uint32_t stride;
56 	uint32_t next;
57 	uint32_t pad[2];
58 };
59 
60 struct bcm2835_cb_entry {
61 	struct bcm2835_dma_cb *cb;
62 	dma_addr_t paddr;
63 };
64 
65 struct bcm2835_chan {
66 	struct virt_dma_chan vc;
67 	struct list_head node;
68 
69 	struct dma_slave_config	cfg;
70 	unsigned int dreq;
71 
72 	int ch;
73 	struct bcm2835_desc *desc;
74 	struct dma_pool *cb_pool;
75 
76 	void __iomem *chan_base;
77 	int irq_number;
78 	unsigned int irq_flags;
79 
80 	bool is_lite_channel;
81 };
82 
83 struct bcm2835_desc {
84 	struct bcm2835_chan *c;
85 	struct virt_dma_desc vd;
86 	enum dma_transfer_direction dir;
87 
88 	unsigned int frames;
89 	size_t size;
90 
91 	bool cyclic;
92 
93 	struct bcm2835_cb_entry cb_list[];
94 };
95 
96 #define BCM2835_DMA_CS		0x00
97 #define BCM2835_DMA_ADDR	0x04
98 #define BCM2835_DMA_TI		0x08
99 #define BCM2835_DMA_SOURCE_AD	0x0c
100 #define BCM2835_DMA_DEST_AD	0x10
101 #define BCM2835_DMA_LEN		0x14
102 #define BCM2835_DMA_STRIDE	0x18
103 #define BCM2835_DMA_NEXTCB	0x1c
104 #define BCM2835_DMA_DEBUG	0x20
105 
106 /* DMA CS Control and Status bits */
107 #define BCM2835_DMA_ACTIVE	BIT(0)  /* activate the DMA */
108 #define BCM2835_DMA_END		BIT(1)  /* current CB has ended */
109 #define BCM2835_DMA_INT		BIT(2)  /* interrupt status */
110 #define BCM2835_DMA_DREQ	BIT(3)  /* DREQ state */
111 #define BCM2835_DMA_ISPAUSED	BIT(4)  /* Pause requested or not active */
112 #define BCM2835_DMA_ISHELD	BIT(5)  /* Is held by DREQ flow control */
113 #define BCM2835_DMA_WAITING_FOR_WRITES BIT(6) /* waiting for last
114 					       * AXI-write to ack
115 					       */
116 #define BCM2835_DMA_ERR		BIT(8)
117 #define BCM2835_DMA_PRIORITY(x) ((x & 15) << 16) /* AXI priority */
118 #define BCM2835_DMA_PANIC_PRIORITY(x) ((x & 15) << 20) /* panic priority */
119 /* current value of TI.BCM2835_DMA_WAIT_RESP */
120 #define BCM2835_DMA_WAIT_FOR_WRITES BIT(28)
121 #define BCM2835_DMA_DIS_DEBUG	BIT(29) /* disable debug pause signal */
122 #define BCM2835_DMA_ABORT	BIT(30) /* Stop current CB, go to next, WO */
123 #define BCM2835_DMA_RESET	BIT(31) /* WO, self clearing */
124 
125 /* Transfer information bits - also bcm2835_cb.info field */
126 #define BCM2835_DMA_INT_EN	BIT(0)
127 #define BCM2835_DMA_TDMODE	BIT(1) /* 2D-Mode */
128 #define BCM2835_DMA_WAIT_RESP	BIT(3) /* wait for AXI-write to be acked */
129 #define BCM2835_DMA_D_INC	BIT(4)
130 #define BCM2835_DMA_D_WIDTH	BIT(5) /* 128bit writes if set */
131 #define BCM2835_DMA_D_DREQ	BIT(6) /* enable DREQ for destination */
132 #define BCM2835_DMA_D_IGNORE	BIT(7) /* ignore destination writes */
133 #define BCM2835_DMA_S_INC	BIT(8)
134 #define BCM2835_DMA_S_WIDTH	BIT(9) /* 128bit writes if set */
135 #define BCM2835_DMA_S_DREQ	BIT(10) /* enable SREQ for source */
136 #define BCM2835_DMA_S_IGNORE	BIT(11) /* ignore source reads - read 0 */
137 #define BCM2835_DMA_BURST_LENGTH(x) ((x & 15) << 12)
138 #define BCM2835_DMA_PER_MAP(x)	((x & 31) << 16) /* REQ source */
139 #define BCM2835_DMA_WAIT(x)	((x & 31) << 21) /* add DMA-wait cycles */
140 #define BCM2835_DMA_NO_WIDE_BURSTS BIT(26) /* no 2 beat write bursts */
141 
142 /* debug register bits */
143 #define BCM2835_DMA_DEBUG_LAST_NOT_SET_ERR	BIT(0)
144 #define BCM2835_DMA_DEBUG_FIFO_ERR		BIT(1)
145 #define BCM2835_DMA_DEBUG_READ_ERR		BIT(2)
146 #define BCM2835_DMA_DEBUG_OUTSTANDING_WRITES_SHIFT 4
147 #define BCM2835_DMA_DEBUG_OUTSTANDING_WRITES_BITS 4
148 #define BCM2835_DMA_DEBUG_ID_SHIFT		16
149 #define BCM2835_DMA_DEBUG_ID_BITS		9
150 #define BCM2835_DMA_DEBUG_STATE_SHIFT		16
151 #define BCM2835_DMA_DEBUG_STATE_BITS		9
152 #define BCM2835_DMA_DEBUG_VERSION_SHIFT		25
153 #define BCM2835_DMA_DEBUG_VERSION_BITS		3
154 #define BCM2835_DMA_DEBUG_LITE			BIT(28)
155 
156 /* shared registers for all dma channels */
157 #define BCM2835_DMA_INT_STATUS         0xfe0
158 #define BCM2835_DMA_ENABLE             0xff0
159 
160 #define BCM2835_DMA_DATA_TYPE_S8	1
161 #define BCM2835_DMA_DATA_TYPE_S16	2
162 #define BCM2835_DMA_DATA_TYPE_S32	4
163 #define BCM2835_DMA_DATA_TYPE_S128	16
164 
165 /* Valid only for channels 0 - 14, 15 has its own base address */
166 #define BCM2835_DMA_CHAN(n)	((n) << 8) /* Base address */
167 #define BCM2835_DMA_CHANIO(base, n) ((base) + BCM2835_DMA_CHAN(n))
168 
169 /* the max dma length for different channels */
170 #define MAX_DMA_LEN SZ_1G
171 #define MAX_LITE_DMA_LEN (SZ_64K - 4)
172 
173 static inline size_t bcm2835_dma_max_frame_length(struct bcm2835_chan *c)
174 {
175 	/* lite and normal channels have different max frame length */
176 	return c->is_lite_channel ? MAX_LITE_DMA_LEN : MAX_DMA_LEN;
177 }
178 
179 /* how many frames of max_len size do we need to transfer len bytes */
180 static inline size_t bcm2835_dma_frames_for_length(size_t len,
181 						   size_t max_len)
182 {
183 	return DIV_ROUND_UP(len, max_len);
184 }
185 
186 static inline struct bcm2835_dmadev *to_bcm2835_dma_dev(struct dma_device *d)
187 {
188 	return container_of(d, struct bcm2835_dmadev, ddev);
189 }
190 
191 static inline struct bcm2835_chan *to_bcm2835_dma_chan(struct dma_chan *c)
192 {
193 	return container_of(c, struct bcm2835_chan, vc.chan);
194 }
195 
196 static inline struct bcm2835_desc *to_bcm2835_dma_desc(
197 		struct dma_async_tx_descriptor *t)
198 {
199 	return container_of(t, struct bcm2835_desc, vd.tx);
200 }
201 
202 static void bcm2835_dma_free_cb_chain(struct bcm2835_desc *desc)
203 {
204 	size_t i;
205 
206 	for (i = 0; i < desc->frames; i++)
207 		dma_pool_free(desc->c->cb_pool, desc->cb_list[i].cb,
208 			      desc->cb_list[i].paddr);
209 
210 	kfree(desc);
211 }
212 
213 static void bcm2835_dma_desc_free(struct virt_dma_desc *vd)
214 {
215 	bcm2835_dma_free_cb_chain(
216 		container_of(vd, struct bcm2835_desc, vd));
217 }
218 
219 static void bcm2835_dma_create_cb_set_length(
220 	struct bcm2835_chan *chan,
221 	struct bcm2835_dma_cb *control_block,
222 	size_t len,
223 	size_t period_len,
224 	size_t *total_len,
225 	u32 finalextrainfo)
226 {
227 	size_t max_len = bcm2835_dma_max_frame_length(chan);
228 
229 	/* set the length taking lite-channel limitations into account */
230 	control_block->length = min_t(u32, len, max_len);
231 
232 	/* finished if we have no period_length */
233 	if (!period_len)
234 		return;
235 
236 	/*
237 	 * period_len means: that we need to generate
238 	 * transfers that are terminating at every
239 	 * multiple of period_len - this is typically
240 	 * used to set the interrupt flag in info
241 	 * which is required during cyclic transfers
242 	 */
243 
244 	/* have we filled in period_length yet? */
245 	if (*total_len + control_block->length < period_len) {
246 		/* update number of bytes in this period so far */
247 		*total_len += control_block->length;
248 		return;
249 	}
250 
251 	/* calculate the length that remains to reach period_length */
252 	control_block->length = period_len - *total_len;
253 
254 	/* reset total_length for next period */
255 	*total_len = 0;
256 
257 	/* add extrainfo bits in info */
258 	control_block->info |= finalextrainfo;
259 }
260 
261 static inline size_t bcm2835_dma_count_frames_for_sg(
262 	struct bcm2835_chan *c,
263 	struct scatterlist *sgl,
264 	unsigned int sg_len)
265 {
266 	size_t frames = 0;
267 	struct scatterlist *sgent;
268 	unsigned int i;
269 	size_t plength = bcm2835_dma_max_frame_length(c);
270 
271 	for_each_sg(sgl, sgent, sg_len, i)
272 		frames += bcm2835_dma_frames_for_length(
273 			sg_dma_len(sgent), plength);
274 
275 	return frames;
276 }
277 
278 /**
279  * bcm2835_dma_create_cb_chain - create a control block and fills data in
280  *
281  * @chan:           the @dma_chan for which we run this
282  * @direction:      the direction in which we transfer
283  * @cyclic:         it is a cyclic transfer
284  * @info:           the default info bits to apply per controlblock
285  * @frames:         number of controlblocks to allocate
286  * @src:            the src address to assign (if the S_INC bit is set
287  *                  in @info, then it gets incremented)
288  * @dst:            the dst address to assign (if the D_INC bit is set
289  *                  in @info, then it gets incremented)
290  * @buf_len:        the full buffer length (may also be 0)
291  * @period_len:     the period length when to apply @finalextrainfo
292  *                  in addition to the last transfer
293  *                  this will also break some control-blocks early
294  * @finalextrainfo: additional bits in last controlblock
295  *                  (or when period_len is reached in case of cyclic)
296  * @gfp:            the GFP flag to use for allocation
297  */
298 static struct bcm2835_desc *bcm2835_dma_create_cb_chain(
299 	struct dma_chan *chan, enum dma_transfer_direction direction,
300 	bool cyclic, u32 info, u32 finalextrainfo, size_t frames,
301 	dma_addr_t src, dma_addr_t dst, size_t buf_len,
302 	size_t period_len, gfp_t gfp)
303 {
304 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
305 	size_t len = buf_len, total_len;
306 	size_t frame;
307 	struct bcm2835_desc *d;
308 	struct bcm2835_cb_entry *cb_entry;
309 	struct bcm2835_dma_cb *control_block;
310 
311 	if (!frames)
312 		return NULL;
313 
314 	/* allocate and setup the descriptor. */
315 	d = kzalloc(sizeof(*d) + frames * sizeof(struct bcm2835_cb_entry),
316 		    gfp);
317 	if (!d)
318 		return NULL;
319 
320 	d->c = c;
321 	d->dir = direction;
322 	d->cyclic = cyclic;
323 
324 	/*
325 	 * Iterate over all frames, create a control block
326 	 * for each frame and link them together.
327 	 */
328 	for (frame = 0, total_len = 0; frame < frames; d->frames++, frame++) {
329 		cb_entry = &d->cb_list[frame];
330 		cb_entry->cb = dma_pool_alloc(c->cb_pool, gfp,
331 					      &cb_entry->paddr);
332 		if (!cb_entry->cb)
333 			goto error_cb;
334 
335 		/* fill in the control block */
336 		control_block = cb_entry->cb;
337 		control_block->info = info;
338 		control_block->src = src;
339 		control_block->dst = dst;
340 		control_block->stride = 0;
341 		control_block->next = 0;
342 		/* set up length in control_block if requested */
343 		if (buf_len) {
344 			/* calculate length honoring period_length */
345 			bcm2835_dma_create_cb_set_length(
346 				c, control_block,
347 				len, period_len, &total_len,
348 				cyclic ? finalextrainfo : 0);
349 
350 			/* calculate new remaining length */
351 			len -= control_block->length;
352 		}
353 
354 		/* link this the last controlblock */
355 		if (frame)
356 			d->cb_list[frame - 1].cb->next = cb_entry->paddr;
357 
358 		/* update src and dst and length */
359 		if (src && (info & BCM2835_DMA_S_INC))
360 			src += control_block->length;
361 		if (dst && (info & BCM2835_DMA_D_INC))
362 			dst += control_block->length;
363 
364 		/* Length of total transfer */
365 		d->size += control_block->length;
366 	}
367 
368 	/* the last frame requires extra flags */
369 	d->cb_list[d->frames - 1].cb->info |= finalextrainfo;
370 
371 	/* detect a size missmatch */
372 	if (buf_len && (d->size != buf_len))
373 		goto error_cb;
374 
375 	return d;
376 error_cb:
377 	bcm2835_dma_free_cb_chain(d);
378 
379 	return NULL;
380 }
381 
382 static void bcm2835_dma_fill_cb_chain_with_sg(
383 	struct dma_chan *chan,
384 	enum dma_transfer_direction direction,
385 	struct bcm2835_cb_entry *cb,
386 	struct scatterlist *sgl,
387 	unsigned int sg_len)
388 {
389 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
390 	size_t len, max_len;
391 	unsigned int i;
392 	dma_addr_t addr;
393 	struct scatterlist *sgent;
394 
395 	max_len = bcm2835_dma_max_frame_length(c);
396 	for_each_sg(sgl, sgent, sg_len, i) {
397 		for (addr = sg_dma_address(sgent), len = sg_dma_len(sgent);
398 		     len > 0;
399 		     addr += cb->cb->length, len -= cb->cb->length, cb++) {
400 			if (direction == DMA_DEV_TO_MEM)
401 				cb->cb->dst = addr;
402 			else
403 				cb->cb->src = addr;
404 			cb->cb->length = min(len, max_len);
405 		}
406 	}
407 }
408 
409 static int bcm2835_dma_abort(void __iomem *chan_base)
410 {
411 	unsigned long cs;
412 	long int timeout = 10000;
413 
414 	cs = readl(chan_base + BCM2835_DMA_CS);
415 	if (!(cs & BCM2835_DMA_ACTIVE))
416 		return 0;
417 
418 	/* Write 0 to the active bit - Pause the DMA */
419 	writel(0, chan_base + BCM2835_DMA_CS);
420 
421 	/* Wait for any current AXI transfer to complete */
422 	while ((cs & BCM2835_DMA_ISPAUSED) && --timeout) {
423 		cpu_relax();
424 		cs = readl(chan_base + BCM2835_DMA_CS);
425 	}
426 
427 	/* We'll un-pause when we set of our next DMA */
428 	if (!timeout)
429 		return -ETIMEDOUT;
430 
431 	if (!(cs & BCM2835_DMA_ACTIVE))
432 		return 0;
433 
434 	/* Terminate the control block chain */
435 	writel(0, chan_base + BCM2835_DMA_NEXTCB);
436 
437 	/* Abort the whole DMA */
438 	writel(BCM2835_DMA_ABORT | BCM2835_DMA_ACTIVE,
439 	       chan_base + BCM2835_DMA_CS);
440 
441 	return 0;
442 }
443 
444 static void bcm2835_dma_start_desc(struct bcm2835_chan *c)
445 {
446 	struct virt_dma_desc *vd = vchan_next_desc(&c->vc);
447 	struct bcm2835_desc *d;
448 
449 	if (!vd) {
450 		c->desc = NULL;
451 		return;
452 	}
453 
454 	list_del(&vd->node);
455 
456 	c->desc = d = to_bcm2835_dma_desc(&vd->tx);
457 
458 	writel(d->cb_list[0].paddr, c->chan_base + BCM2835_DMA_ADDR);
459 	writel(BCM2835_DMA_ACTIVE, c->chan_base + BCM2835_DMA_CS);
460 }
461 
462 static irqreturn_t bcm2835_dma_callback(int irq, void *data)
463 {
464 	struct bcm2835_chan *c = data;
465 	struct bcm2835_desc *d;
466 	unsigned long flags;
467 
468 	/* check the shared interrupt */
469 	if (c->irq_flags & IRQF_SHARED) {
470 		/* check if the interrupt is enabled */
471 		flags = readl(c->chan_base + BCM2835_DMA_CS);
472 		/* if not set then we are not the reason for the irq */
473 		if (!(flags & BCM2835_DMA_INT))
474 			return IRQ_NONE;
475 	}
476 
477 	spin_lock_irqsave(&c->vc.lock, flags);
478 
479 	/* Acknowledge interrupt */
480 	writel(BCM2835_DMA_INT, c->chan_base + BCM2835_DMA_CS);
481 
482 	d = c->desc;
483 
484 	if (d) {
485 		if (d->cyclic) {
486 			/* call the cyclic callback */
487 			vchan_cyclic_callback(&d->vd);
488 
489 			/* Keep the DMA engine running */
490 			writel(BCM2835_DMA_ACTIVE,
491 			       c->chan_base + BCM2835_DMA_CS);
492 		} else {
493 			vchan_cookie_complete(&c->desc->vd);
494 			bcm2835_dma_start_desc(c);
495 		}
496 	}
497 
498 	spin_unlock_irqrestore(&c->vc.lock, flags);
499 
500 	return IRQ_HANDLED;
501 }
502 
503 static int bcm2835_dma_alloc_chan_resources(struct dma_chan *chan)
504 {
505 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
506 	struct device *dev = c->vc.chan.device->dev;
507 
508 	dev_dbg(dev, "Allocating DMA channel %d\n", c->ch);
509 
510 	c->cb_pool = dma_pool_create(dev_name(dev), dev,
511 				     sizeof(struct bcm2835_dma_cb), 0, 0);
512 	if (!c->cb_pool) {
513 		dev_err(dev, "unable to allocate descriptor pool\n");
514 		return -ENOMEM;
515 	}
516 
517 	return request_irq(c->irq_number, bcm2835_dma_callback,
518 			   c->irq_flags, "DMA IRQ", c);
519 }
520 
521 static void bcm2835_dma_free_chan_resources(struct dma_chan *chan)
522 {
523 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
524 
525 	vchan_free_chan_resources(&c->vc);
526 	free_irq(c->irq_number, c);
527 	dma_pool_destroy(c->cb_pool);
528 
529 	dev_dbg(c->vc.chan.device->dev, "Freeing DMA channel %u\n", c->ch);
530 }
531 
532 static size_t bcm2835_dma_desc_size(struct bcm2835_desc *d)
533 {
534 	return d->size;
535 }
536 
537 static size_t bcm2835_dma_desc_size_pos(struct bcm2835_desc *d, dma_addr_t addr)
538 {
539 	unsigned int i;
540 	size_t size;
541 
542 	for (size = i = 0; i < d->frames; i++) {
543 		struct bcm2835_dma_cb *control_block = d->cb_list[i].cb;
544 		size_t this_size = control_block->length;
545 		dma_addr_t dma;
546 
547 		if (d->dir == DMA_DEV_TO_MEM)
548 			dma = control_block->dst;
549 		else
550 			dma = control_block->src;
551 
552 		if (size)
553 			size += this_size;
554 		else if (addr >= dma && addr < dma + this_size)
555 			size += dma + this_size - addr;
556 	}
557 
558 	return size;
559 }
560 
561 static enum dma_status bcm2835_dma_tx_status(struct dma_chan *chan,
562 	dma_cookie_t cookie, struct dma_tx_state *txstate)
563 {
564 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
565 	struct virt_dma_desc *vd;
566 	enum dma_status ret;
567 	unsigned long flags;
568 
569 	ret = dma_cookie_status(chan, cookie, txstate);
570 	if (ret == DMA_COMPLETE || !txstate)
571 		return ret;
572 
573 	spin_lock_irqsave(&c->vc.lock, flags);
574 	vd = vchan_find_desc(&c->vc, cookie);
575 	if (vd) {
576 		txstate->residue =
577 			bcm2835_dma_desc_size(to_bcm2835_dma_desc(&vd->tx));
578 	} else if (c->desc && c->desc->vd.tx.cookie == cookie) {
579 		struct bcm2835_desc *d = c->desc;
580 		dma_addr_t pos;
581 
582 		if (d->dir == DMA_MEM_TO_DEV)
583 			pos = readl(c->chan_base + BCM2835_DMA_SOURCE_AD);
584 		else if (d->dir == DMA_DEV_TO_MEM)
585 			pos = readl(c->chan_base + BCM2835_DMA_DEST_AD);
586 		else
587 			pos = 0;
588 
589 		txstate->residue = bcm2835_dma_desc_size_pos(d, pos);
590 	} else {
591 		txstate->residue = 0;
592 	}
593 
594 	spin_unlock_irqrestore(&c->vc.lock, flags);
595 
596 	return ret;
597 }
598 
599 static void bcm2835_dma_issue_pending(struct dma_chan *chan)
600 {
601 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
602 	unsigned long flags;
603 
604 	spin_lock_irqsave(&c->vc.lock, flags);
605 	if (vchan_issue_pending(&c->vc) && !c->desc)
606 		bcm2835_dma_start_desc(c);
607 
608 	spin_unlock_irqrestore(&c->vc.lock, flags);
609 }
610 
611 static struct dma_async_tx_descriptor *bcm2835_dma_prep_dma_memcpy(
612 	struct dma_chan *chan, dma_addr_t dst, dma_addr_t src,
613 	size_t len, unsigned long flags)
614 {
615 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
616 	struct bcm2835_desc *d;
617 	u32 info = BCM2835_DMA_D_INC | BCM2835_DMA_S_INC;
618 	u32 extra = BCM2835_DMA_INT_EN | BCM2835_DMA_WAIT_RESP;
619 	size_t max_len = bcm2835_dma_max_frame_length(c);
620 	size_t frames;
621 
622 	/* if src, dst or len is not given return with an error */
623 	if (!src || !dst || !len)
624 		return NULL;
625 
626 	/* calculate number of frames */
627 	frames = bcm2835_dma_frames_for_length(len, max_len);
628 
629 	/* allocate the CB chain - this also fills in the pointers */
630 	d = bcm2835_dma_create_cb_chain(chan, DMA_MEM_TO_MEM, false,
631 					info, extra, frames,
632 					src, dst, len, 0, GFP_KERNEL);
633 	if (!d)
634 		return NULL;
635 
636 	return vchan_tx_prep(&c->vc, &d->vd, flags);
637 }
638 
639 static struct dma_async_tx_descriptor *bcm2835_dma_prep_slave_sg(
640 	struct dma_chan *chan,
641 	struct scatterlist *sgl, unsigned int sg_len,
642 	enum dma_transfer_direction direction,
643 	unsigned long flags, void *context)
644 {
645 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
646 	struct bcm2835_desc *d;
647 	dma_addr_t src = 0, dst = 0;
648 	u32 info = BCM2835_DMA_WAIT_RESP;
649 	u32 extra = BCM2835_DMA_INT_EN;
650 	size_t frames;
651 
652 	if (!is_slave_direction(direction)) {
653 		dev_err(chan->device->dev,
654 			"%s: bad direction?\n", __func__);
655 		return NULL;
656 	}
657 
658 	if (c->dreq != 0)
659 		info |= BCM2835_DMA_PER_MAP(c->dreq);
660 
661 	if (direction == DMA_DEV_TO_MEM) {
662 		if (c->cfg.src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
663 			return NULL;
664 		src = c->cfg.src_addr;
665 		info |= BCM2835_DMA_S_DREQ | BCM2835_DMA_D_INC;
666 	} else {
667 		if (c->cfg.dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
668 			return NULL;
669 		dst = c->cfg.dst_addr;
670 		info |= BCM2835_DMA_D_DREQ | BCM2835_DMA_S_INC;
671 	}
672 
673 	/* count frames in sg list */
674 	frames = bcm2835_dma_count_frames_for_sg(c, sgl, sg_len);
675 
676 	/* allocate the CB chain */
677 	d = bcm2835_dma_create_cb_chain(chan, direction, false,
678 					info, extra,
679 					frames, src, dst, 0, 0,
680 					GFP_KERNEL);
681 	if (!d)
682 		return NULL;
683 
684 	/* fill in frames with scatterlist pointers */
685 	bcm2835_dma_fill_cb_chain_with_sg(chan, direction, d->cb_list,
686 					  sgl, sg_len);
687 
688 	return vchan_tx_prep(&c->vc, &d->vd, flags);
689 }
690 
691 static struct dma_async_tx_descriptor *bcm2835_dma_prep_dma_cyclic(
692 	struct dma_chan *chan, dma_addr_t buf_addr, size_t buf_len,
693 	size_t period_len, enum dma_transfer_direction direction,
694 	unsigned long flags)
695 {
696 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
697 	struct bcm2835_desc *d;
698 	dma_addr_t src, dst;
699 	u32 info = BCM2835_DMA_WAIT_RESP;
700 	u32 extra = BCM2835_DMA_INT_EN;
701 	size_t max_len = bcm2835_dma_max_frame_length(c);
702 	size_t frames;
703 
704 	/* Grab configuration */
705 	if (!is_slave_direction(direction)) {
706 		dev_err(chan->device->dev, "%s: bad direction?\n", __func__);
707 		return NULL;
708 	}
709 
710 	if (!buf_len) {
711 		dev_err(chan->device->dev,
712 			"%s: bad buffer length (= 0)\n", __func__);
713 		return NULL;
714 	}
715 
716 	/*
717 	 * warn if buf_len is not a multiple of period_len - this may leed
718 	 * to unexpected latencies for interrupts and thus audiable clicks
719 	 */
720 	if (buf_len % period_len)
721 		dev_warn_once(chan->device->dev,
722 			      "%s: buffer_length (%zd) is not a multiple of period_len (%zd)\n",
723 			      __func__, buf_len, period_len);
724 
725 	/* Setup DREQ channel */
726 	if (c->dreq != 0)
727 		info |= BCM2835_DMA_PER_MAP(c->dreq);
728 
729 	if (direction == DMA_DEV_TO_MEM) {
730 		if (c->cfg.src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
731 			return NULL;
732 		src = c->cfg.src_addr;
733 		dst = buf_addr;
734 		info |= BCM2835_DMA_S_DREQ | BCM2835_DMA_D_INC;
735 	} else {
736 		if (c->cfg.dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
737 			return NULL;
738 		dst = c->cfg.dst_addr;
739 		src = buf_addr;
740 		info |= BCM2835_DMA_D_DREQ | BCM2835_DMA_S_INC;
741 	}
742 
743 	/* calculate number of frames */
744 	frames = /* number of periods */
745 		 DIV_ROUND_UP(buf_len, period_len) *
746 		 /* number of frames per period */
747 		 bcm2835_dma_frames_for_length(period_len, max_len);
748 
749 	/*
750 	 * allocate the CB chain
751 	 * note that we need to use GFP_NOWAIT, as the ALSA i2s dmaengine
752 	 * implementation calls prep_dma_cyclic with interrupts disabled.
753 	 */
754 	d = bcm2835_dma_create_cb_chain(chan, direction, true,
755 					info, extra,
756 					frames, src, dst, buf_len,
757 					period_len, GFP_NOWAIT);
758 	if (!d)
759 		return NULL;
760 
761 	/* wrap around into a loop */
762 	d->cb_list[d->frames - 1].cb->next = d->cb_list[0].paddr;
763 
764 	return vchan_tx_prep(&c->vc, &d->vd, flags);
765 }
766 
767 static int bcm2835_dma_slave_config(struct dma_chan *chan,
768 				    struct dma_slave_config *cfg)
769 {
770 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
771 
772 	c->cfg = *cfg;
773 
774 	return 0;
775 }
776 
777 static int bcm2835_dma_terminate_all(struct dma_chan *chan)
778 {
779 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
780 	struct bcm2835_dmadev *d = to_bcm2835_dma_dev(c->vc.chan.device);
781 	unsigned long flags;
782 	int timeout = 10000;
783 	LIST_HEAD(head);
784 
785 	spin_lock_irqsave(&c->vc.lock, flags);
786 
787 	/* Prevent this channel being scheduled */
788 	spin_lock(&d->lock);
789 	list_del_init(&c->node);
790 	spin_unlock(&d->lock);
791 
792 	/*
793 	 * Stop DMA activity: we assume the callback will not be called
794 	 * after bcm_dma_abort() returns (even if it does, it will see
795 	 * c->desc is NULL and exit.)
796 	 */
797 	if (c->desc) {
798 		vchan_terminate_vdesc(&c->desc->vd);
799 		c->desc = NULL;
800 		bcm2835_dma_abort(c->chan_base);
801 
802 		/* Wait for stopping */
803 		while (--timeout) {
804 			if (!(readl(c->chan_base + BCM2835_DMA_CS) &
805 						BCM2835_DMA_ACTIVE))
806 				break;
807 
808 			cpu_relax();
809 		}
810 
811 		if (!timeout)
812 			dev_err(d->ddev.dev, "DMA transfer could not be terminated\n");
813 	}
814 
815 	vchan_get_all_descriptors(&c->vc, &head);
816 	spin_unlock_irqrestore(&c->vc.lock, flags);
817 	vchan_dma_desc_free_list(&c->vc, &head);
818 
819 	return 0;
820 }
821 
822 static void bcm2835_dma_synchronize(struct dma_chan *chan)
823 {
824 	struct bcm2835_chan *c = to_bcm2835_dma_chan(chan);
825 
826 	vchan_synchronize(&c->vc);
827 }
828 
829 static int bcm2835_dma_chan_init(struct bcm2835_dmadev *d, int chan_id,
830 				 int irq, unsigned int irq_flags)
831 {
832 	struct bcm2835_chan *c;
833 
834 	c = devm_kzalloc(d->ddev.dev, sizeof(*c), GFP_KERNEL);
835 	if (!c)
836 		return -ENOMEM;
837 
838 	c->vc.desc_free = bcm2835_dma_desc_free;
839 	vchan_init(&c->vc, &d->ddev);
840 	INIT_LIST_HEAD(&c->node);
841 
842 	c->chan_base = BCM2835_DMA_CHANIO(d->base, chan_id);
843 	c->ch = chan_id;
844 	c->irq_number = irq;
845 	c->irq_flags = irq_flags;
846 
847 	/* check in DEBUG register if this is a LITE channel */
848 	if (readl(c->chan_base + BCM2835_DMA_DEBUG) &
849 		BCM2835_DMA_DEBUG_LITE)
850 		c->is_lite_channel = true;
851 
852 	return 0;
853 }
854 
855 static void bcm2835_dma_free(struct bcm2835_dmadev *od)
856 {
857 	struct bcm2835_chan *c, *next;
858 
859 	list_for_each_entry_safe(c, next, &od->ddev.channels,
860 				 vc.chan.device_node) {
861 		list_del(&c->vc.chan.device_node);
862 		tasklet_kill(&c->vc.task);
863 	}
864 }
865 
866 static const struct of_device_id bcm2835_dma_of_match[] = {
867 	{ .compatible = "brcm,bcm2835-dma", },
868 	{},
869 };
870 MODULE_DEVICE_TABLE(of, bcm2835_dma_of_match);
871 
872 static struct dma_chan *bcm2835_dma_xlate(struct of_phandle_args *spec,
873 					   struct of_dma *ofdma)
874 {
875 	struct bcm2835_dmadev *d = ofdma->of_dma_data;
876 	struct dma_chan *chan;
877 
878 	chan = dma_get_any_slave_channel(&d->ddev);
879 	if (!chan)
880 		return NULL;
881 
882 	/* Set DREQ from param */
883 	to_bcm2835_dma_chan(chan)->dreq = spec->args[0];
884 
885 	return chan;
886 }
887 
888 static int bcm2835_dma_probe(struct platform_device *pdev)
889 {
890 	struct bcm2835_dmadev *od;
891 	struct resource *res;
892 	void __iomem *base;
893 	int rc;
894 	int i, j;
895 	int irq[BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED + 1];
896 	int irq_flags;
897 	uint32_t chans_available;
898 	char chan_name[BCM2835_DMA_CHAN_NAME_SIZE];
899 
900 	if (!pdev->dev.dma_mask)
901 		pdev->dev.dma_mask = &pdev->dev.coherent_dma_mask;
902 
903 	rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
904 	if (rc)
905 		return rc;
906 
907 	od = devm_kzalloc(&pdev->dev, sizeof(*od), GFP_KERNEL);
908 	if (!od)
909 		return -ENOMEM;
910 
911 	pdev->dev.dma_parms = &od->dma_parms;
912 	dma_set_max_seg_size(&pdev->dev, 0x3FFFFFFF);
913 
914 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
915 	base = devm_ioremap_resource(&pdev->dev, res);
916 	if (IS_ERR(base))
917 		return PTR_ERR(base);
918 
919 	od->base = base;
920 
921 	dma_cap_set(DMA_SLAVE, od->ddev.cap_mask);
922 	dma_cap_set(DMA_PRIVATE, od->ddev.cap_mask);
923 	dma_cap_set(DMA_CYCLIC, od->ddev.cap_mask);
924 	dma_cap_set(DMA_SLAVE, od->ddev.cap_mask);
925 	dma_cap_set(DMA_MEMCPY, od->ddev.cap_mask);
926 	od->ddev.device_alloc_chan_resources = bcm2835_dma_alloc_chan_resources;
927 	od->ddev.device_free_chan_resources = bcm2835_dma_free_chan_resources;
928 	od->ddev.device_tx_status = bcm2835_dma_tx_status;
929 	od->ddev.device_issue_pending = bcm2835_dma_issue_pending;
930 	od->ddev.device_prep_dma_cyclic = bcm2835_dma_prep_dma_cyclic;
931 	od->ddev.device_prep_slave_sg = bcm2835_dma_prep_slave_sg;
932 	od->ddev.device_prep_dma_memcpy = bcm2835_dma_prep_dma_memcpy;
933 	od->ddev.device_config = bcm2835_dma_slave_config;
934 	od->ddev.device_terminate_all = bcm2835_dma_terminate_all;
935 	od->ddev.device_synchronize = bcm2835_dma_synchronize;
936 	od->ddev.src_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_4_BYTES);
937 	od->ddev.dst_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_4_BYTES);
938 	od->ddev.directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV) |
939 			      BIT(DMA_MEM_TO_MEM);
940 	od->ddev.residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
941 	od->ddev.dev = &pdev->dev;
942 	INIT_LIST_HEAD(&od->ddev.channels);
943 	spin_lock_init(&od->lock);
944 
945 	platform_set_drvdata(pdev, od);
946 
947 	/* Request DMA channel mask from device tree */
948 	if (of_property_read_u32(pdev->dev.of_node,
949 			"brcm,dma-channel-mask",
950 			&chans_available)) {
951 		dev_err(&pdev->dev, "Failed to get channel mask\n");
952 		rc = -EINVAL;
953 		goto err_no_dma;
954 	}
955 
956 	/* get irqs for each channel that we support */
957 	for (i = 0; i <= BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED; i++) {
958 		/* skip masked out channels */
959 		if (!(chans_available & (1 << i))) {
960 			irq[i] = -1;
961 			continue;
962 		}
963 
964 		/* get the named irq */
965 		snprintf(chan_name, sizeof(chan_name), "dma%i", i);
966 		irq[i] = platform_get_irq_byname(pdev, chan_name);
967 		if (irq[i] >= 0)
968 			continue;
969 
970 		/* legacy device tree case handling */
971 		dev_warn_once(&pdev->dev,
972 			      "missing interrupt-names property in device tree - legacy interpretation is used\n");
973 		/*
974 		 * in case of channel >= 11
975 		 * use the 11th interrupt and that is shared
976 		 */
977 		irq[i] = platform_get_irq(pdev, i < 11 ? i : 11);
978 	}
979 
980 	/* get irqs for each channel */
981 	for (i = 0; i <= BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED; i++) {
982 		/* skip channels without irq */
983 		if (irq[i] < 0)
984 			continue;
985 
986 		/* check if there are other channels that also use this irq */
987 		irq_flags = 0;
988 		for (j = 0; j <= BCM2835_DMA_MAX_DMA_CHAN_SUPPORTED; j++)
989 			if ((i != j) && (irq[j] == irq[i])) {
990 				irq_flags = IRQF_SHARED;
991 				break;
992 			}
993 
994 		/* initialize the channel */
995 		rc = bcm2835_dma_chan_init(od, i, irq[i], irq_flags);
996 		if (rc)
997 			goto err_no_dma;
998 	}
999 
1000 	dev_dbg(&pdev->dev, "Initialized %i DMA channels\n", i);
1001 
1002 	/* Device-tree DMA controller registration */
1003 	rc = of_dma_controller_register(pdev->dev.of_node,
1004 			bcm2835_dma_xlate, od);
1005 	if (rc) {
1006 		dev_err(&pdev->dev, "Failed to register DMA controller\n");
1007 		goto err_no_dma;
1008 	}
1009 
1010 	rc = dma_async_device_register(&od->ddev);
1011 	if (rc) {
1012 		dev_err(&pdev->dev,
1013 			"Failed to register slave DMA engine device: %d\n", rc);
1014 		goto err_no_dma;
1015 	}
1016 
1017 	dev_dbg(&pdev->dev, "Load BCM2835 DMA engine driver\n");
1018 
1019 	return 0;
1020 
1021 err_no_dma:
1022 	bcm2835_dma_free(od);
1023 	return rc;
1024 }
1025 
1026 static int bcm2835_dma_remove(struct platform_device *pdev)
1027 {
1028 	struct bcm2835_dmadev *od = platform_get_drvdata(pdev);
1029 
1030 	dma_async_device_unregister(&od->ddev);
1031 	bcm2835_dma_free(od);
1032 
1033 	return 0;
1034 }
1035 
1036 static struct platform_driver bcm2835_dma_driver = {
1037 	.probe	= bcm2835_dma_probe,
1038 	.remove	= bcm2835_dma_remove,
1039 	.driver = {
1040 		.name = "bcm2835-dma",
1041 		.of_match_table = of_match_ptr(bcm2835_dma_of_match),
1042 	},
1043 };
1044 
1045 module_platform_driver(bcm2835_dma_driver);
1046 
1047 MODULE_ALIAS("platform:bcm2835-dma");
1048 MODULE_DESCRIPTION("BCM2835 DMA engine driver");
1049 MODULE_AUTHOR("Florian Meier <florian.meier@koalo.de>");
1050 MODULE_LICENSE("GPL");
1051