xref: /linux/sound/soc/fsl/fsl_dma.c (revision a8e7ef3cec99ba2487110e01d77a8a278593b3e9)
1 // SPDX-License-Identifier: GPL-2.0
2 //
3 // Freescale DMA ALSA SoC PCM driver
4 //
5 // Author: Timur Tabi <timur@freescale.com>
6 //
7 // Copyright 2007-2010 Freescale Semiconductor, Inc.
8 //
9 // This driver implements ASoC support for the Elo DMA controller, which is
10 // the DMA controller on Freescale 83xx, 85xx, and 86xx SOCs. In ALSA terms,
11 // the PCM driver is what handles the DMA buffer.
12 
13 #include <linux/module.h>
14 #include <linux/init.h>
15 #include <linux/platform_device.h>
16 #include <linux/dma-mapping.h>
17 #include <linux/interrupt.h>
18 #include <linux/delay.h>
19 #include <linux/gfp.h>
20 #include <linux/of_address.h>
21 #include <linux/of_irq.h>
22 #include <linux/of_platform.h>
23 #include <linux/list.h>
24 #include <linux/slab.h>
25 
26 #include <sound/core.h>
27 #include <sound/pcm.h>
28 #include <sound/pcm_params.h>
29 #include <sound/soc.h>
30 
31 #include <asm/io.h>
32 
33 #include "fsl_dma.h"
34 #include "fsl_ssi.h"	/* For the offset of stx0 and srx0 */
35 
36 #define DRV_NAME "fsl_dma"
37 
38 /*
39  * The formats that the DMA controller supports, which is anything
40  * that is 8, 16, or 32 bits.
41  */
42 #define FSLDMA_PCM_FORMATS (SNDRV_PCM_FMTBIT_S8 	| \
43 			    SNDRV_PCM_FMTBIT_U8 	| \
44 			    SNDRV_PCM_FMTBIT_S16_LE     | \
45 			    SNDRV_PCM_FMTBIT_S16_BE     | \
46 			    SNDRV_PCM_FMTBIT_U16_LE     | \
47 			    SNDRV_PCM_FMTBIT_U16_BE     | \
48 			    SNDRV_PCM_FMTBIT_S24_LE     | \
49 			    SNDRV_PCM_FMTBIT_S24_BE     | \
50 			    SNDRV_PCM_FMTBIT_U24_LE     | \
51 			    SNDRV_PCM_FMTBIT_U24_BE     | \
52 			    SNDRV_PCM_FMTBIT_S32_LE     | \
53 			    SNDRV_PCM_FMTBIT_S32_BE     | \
54 			    SNDRV_PCM_FMTBIT_U32_LE     | \
55 			    SNDRV_PCM_FMTBIT_U32_BE)
56 struct dma_object {
57 	struct snd_soc_component_driver dai;
58 	dma_addr_t ssi_stx_phys;
59 	dma_addr_t ssi_srx_phys;
60 	unsigned int ssi_fifo_depth;
61 	struct ccsr_dma_channel __iomem *channel;
62 	unsigned int irq;
63 	bool assigned;
64 };
65 
66 /*
67  * The number of DMA links to use.  Two is the bare minimum, but if you
68  * have really small links you might need more.
69  */
70 #define NUM_DMA_LINKS   2
71 
72 /** fsl_dma_private: p-substream DMA data
73  *
74  * Each substream has a 1-to-1 association with a DMA channel.
75  *
76  * The link[] array is first because it needs to be aligned on a 32-byte
77  * boundary, so putting it first will ensure alignment without padding the
78  * structure.
79  *
80  * @link[]: array of link descriptors
81  * @dma_channel: pointer to the DMA channel's registers
82  * @irq: IRQ for this DMA channel
83  * @substream: pointer to the substream object, needed by the ISR
84  * @ssi_sxx_phys: bus address of the STX or SRX register to use
85  * @ld_buf_phys: physical address of the LD buffer
86  * @current_link: index into link[] of the link currently being processed
87  * @dma_buf_phys: physical address of the DMA buffer
88  * @dma_buf_next: physical address of the next period to process
89  * @dma_buf_end: physical address of the byte after the end of the DMA
90  * @buffer period_size: the size of a single period
91  * @num_periods: the number of periods in the DMA buffer
92  */
93 struct fsl_dma_private {
94 	struct fsl_dma_link_descriptor link[NUM_DMA_LINKS];
95 	struct ccsr_dma_channel __iomem *dma_channel;
96 	unsigned int irq;
97 	struct snd_pcm_substream *substream;
98 	dma_addr_t ssi_sxx_phys;
99 	unsigned int ssi_fifo_depth;
100 	dma_addr_t ld_buf_phys;
101 	unsigned int current_link;
102 	dma_addr_t dma_buf_phys;
103 	dma_addr_t dma_buf_next;
104 	dma_addr_t dma_buf_end;
105 	size_t period_size;
106 	unsigned int num_periods;
107 };
108 
109 /**
110  * fsl_dma_hardare: define characteristics of the PCM hardware.
111  *
112  * The PCM hardware is the Freescale DMA controller.  This structure defines
113  * the capabilities of that hardware.
114  *
115  * Since the sampling rate and data format are not controlled by the DMA
116  * controller, we specify no limits for those values.  The only exception is
117  * period_bytes_min, which is set to a reasonably low value to prevent the
118  * DMA controller from generating too many interrupts per second.
119  *
120  * Since each link descriptor has a 32-bit byte count field, we set
121  * period_bytes_max to the largest 32-bit number.  We also have no maximum
122  * number of periods.
123  *
124  * Note that we specify SNDRV_PCM_INFO_JOINT_DUPLEX here, but only because a
125  * limitation in the SSI driver requires the sample rates for playback and
126  * capture to be the same.
127  */
128 static const struct snd_pcm_hardware fsl_dma_hardware = {
129 
130 	.info   		= SNDRV_PCM_INFO_INTERLEAVED |
131 				  SNDRV_PCM_INFO_MMAP |
132 				  SNDRV_PCM_INFO_MMAP_VALID |
133 				  SNDRV_PCM_INFO_JOINT_DUPLEX |
134 				  SNDRV_PCM_INFO_PAUSE,
135 	.formats		= FSLDMA_PCM_FORMATS,
136 	.period_bytes_min       = 512,  	/* A reasonable limit */
137 	.period_bytes_max       = (u32) -1,
138 	.periods_min    	= NUM_DMA_LINKS,
139 	.periods_max    	= (unsigned int) -1,
140 	.buffer_bytes_max       = 128 * 1024,   /* A reasonable limit */
141 };
142 
143 /**
144  * fsl_dma_abort_stream: tell ALSA that the DMA transfer has aborted
145  *
146  * This function should be called by the ISR whenever the DMA controller
147  * halts data transfer.
148  */
149 static void fsl_dma_abort_stream(struct snd_pcm_substream *substream)
150 {
151 	snd_pcm_stop_xrun(substream);
152 }
153 
154 /**
155  * fsl_dma_update_pointers - update LD pointers to point to the next period
156  *
157  * As each period is completed, this function changes the link
158  * descriptor pointers for that period to point to the next period.
159  */
160 static void fsl_dma_update_pointers(struct fsl_dma_private *dma_private)
161 {
162 	struct fsl_dma_link_descriptor *link =
163 		&dma_private->link[dma_private->current_link];
164 
165 	/* Update our link descriptors to point to the next period. On a 36-bit
166 	 * system, we also need to update the ESAD bits.  We also set (keep) the
167 	 * snoop bits.  See the comments in fsl_dma_hw_params() about snooping.
168 	 */
169 	if (dma_private->substream->stream == SNDRV_PCM_STREAM_PLAYBACK) {
170 		link->source_addr = cpu_to_be32(dma_private->dma_buf_next);
171 #ifdef CONFIG_PHYS_64BIT
172 		link->source_attr = cpu_to_be32(CCSR_DMA_ATR_SNOOP |
173 			upper_32_bits(dma_private->dma_buf_next));
174 #endif
175 	} else {
176 		link->dest_addr = cpu_to_be32(dma_private->dma_buf_next);
177 #ifdef CONFIG_PHYS_64BIT
178 		link->dest_attr = cpu_to_be32(CCSR_DMA_ATR_SNOOP |
179 			upper_32_bits(dma_private->dma_buf_next));
180 #endif
181 	}
182 
183 	/* Update our variables for next time */
184 	dma_private->dma_buf_next += dma_private->period_size;
185 
186 	if (dma_private->dma_buf_next >= dma_private->dma_buf_end)
187 		dma_private->dma_buf_next = dma_private->dma_buf_phys;
188 
189 	if (++dma_private->current_link >= NUM_DMA_LINKS)
190 		dma_private->current_link = 0;
191 }
192 
193 /**
194  * fsl_dma_isr: interrupt handler for the DMA controller
195  *
196  * @irq: IRQ of the DMA channel
197  * @dev_id: pointer to the dma_private structure for this DMA channel
198  */
199 static irqreturn_t fsl_dma_isr(int irq, void *dev_id)
200 {
201 	struct fsl_dma_private *dma_private = dev_id;
202 	struct snd_pcm_substream *substream = dma_private->substream;
203 	struct snd_soc_pcm_runtime *rtd = snd_soc_substream_to_rtd(substream);
204 	struct device *dev = rtd->dev;
205 	struct ccsr_dma_channel __iomem *dma_channel = dma_private->dma_channel;
206 	irqreturn_t ret = IRQ_NONE;
207 	u32 sr, sr2 = 0;
208 
209 	/* We got an interrupt, so read the status register to see what we
210 	   were interrupted for.
211 	 */
212 	sr = in_be32(&dma_channel->sr);
213 
214 	if (sr & CCSR_DMA_SR_TE) {
215 		dev_err(dev, "dma transmit error\n");
216 		fsl_dma_abort_stream(substream);
217 		sr2 |= CCSR_DMA_SR_TE;
218 		ret = IRQ_HANDLED;
219 	}
220 
221 	if (sr & CCSR_DMA_SR_CH)
222 		ret = IRQ_HANDLED;
223 
224 	if (sr & CCSR_DMA_SR_PE) {
225 		dev_err(dev, "dma programming error\n");
226 		fsl_dma_abort_stream(substream);
227 		sr2 |= CCSR_DMA_SR_PE;
228 		ret = IRQ_HANDLED;
229 	}
230 
231 	if (sr & CCSR_DMA_SR_EOLNI) {
232 		sr2 |= CCSR_DMA_SR_EOLNI;
233 		ret = IRQ_HANDLED;
234 	}
235 
236 	if (sr & CCSR_DMA_SR_CB)
237 		ret = IRQ_HANDLED;
238 
239 	if (sr & CCSR_DMA_SR_EOSI) {
240 		/* Tell ALSA we completed a period. */
241 		snd_pcm_period_elapsed(substream);
242 
243 		/*
244 		 * Update our link descriptors to point to the next period. We
245 		 * only need to do this if the number of periods is not equal to
246 		 * the number of links.
247 		 */
248 		if (dma_private->num_periods != NUM_DMA_LINKS)
249 			fsl_dma_update_pointers(dma_private);
250 
251 		sr2 |= CCSR_DMA_SR_EOSI;
252 		ret = IRQ_HANDLED;
253 	}
254 
255 	if (sr & CCSR_DMA_SR_EOLSI) {
256 		sr2 |= CCSR_DMA_SR_EOLSI;
257 		ret = IRQ_HANDLED;
258 	}
259 
260 	/* Clear the bits that we set */
261 	if (sr2)
262 		out_be32(&dma_channel->sr, sr2);
263 
264 	return ret;
265 }
266 
267 /**
268  * fsl_dma_new: initialize this PCM driver.
269  *
270  * This function is called by soc_new_pcm(), once for each DAI link
271  * in the machine driver's snd_soc_card structure.
272  *
273  * Regardless of where the memory is actually allocated, since the device can
274  * technically DMA to any 36-bit address, we do need to set the DMA mask to 36.
275  */
276 static int fsl_dma_new(struct snd_soc_component *component,
277 		       struct snd_soc_pcm_runtime *rtd)
278 {
279 	struct snd_card *card = rtd->card->snd_card;
280 	struct snd_pcm *pcm = rtd->pcm;
281 	int ret;
282 
283 	ret = dma_coerce_mask_and_coherent(card->dev, DMA_BIT_MASK(36));
284 	if (ret)
285 		return ret;
286 
287 	return snd_pcm_set_fixed_buffer_all(pcm, SNDRV_DMA_TYPE_DEV,
288 					    card->dev,
289 					    fsl_dma_hardware.buffer_bytes_max);
290 }
291 
292 /**
293  * fsl_dma_open: open a new substream.
294  *
295  * Each substream has its own DMA buffer.
296  *
297  * ALSA divides the DMA buffer into N periods.  We create NUM_DMA_LINKS link
298  * descriptors that ping-pong from one period to the next.  For example, if
299  * there are six periods and two link descriptors, this is how they look
300  * before playback starts:
301  *
302  *      	   The last link descriptor
303  *   ____________  points back to the first
304  *  |   	 |
305  *  V   	 |
306  *  ___    ___   |
307  * |   |->|   |->|
308  * |___|  |___|
309  *   |      |
310  *   |      |
311  *   V      V
312  *  _________________________________________
313  * |      |      |      |      |      |      |  The DMA buffer is
314  * |      |      |      |      |      |      |    divided into 6 parts
315  * |______|______|______|______|______|______|
316  *
317  * and here's how they look after the first period is finished playing:
318  *
319  *   ____________
320  *  |   	 |
321  *  V   	 |
322  *  ___    ___   |
323  * |   |->|   |->|
324  * |___|  |___|
325  *   |      |
326  *   |______________
327  *          |       |
328  *          V       V
329  *  _________________________________________
330  * |      |      |      |      |      |      |
331  * |      |      |      |      |      |      |
332  * |______|______|______|______|______|______|
333  *
334  * The first link descriptor now points to the third period.  The DMA
335  * controller is currently playing the second period.  When it finishes, it
336  * will jump back to the first descriptor and play the third period.
337  *
338  * There are four reasons we do this:
339  *
340  * 1. The only way to get the DMA controller to automatically restart the
341  *    transfer when it gets to the end of the buffer is to use chaining
342  *    mode.  Basic direct mode doesn't offer that feature.
343  * 2. We need to receive an interrupt at the end of every period.  The DMA
344  *    controller can generate an interrupt at the end of every link transfer
345  *    (aka segment).  Making each period into a DMA segment will give us the
346  *    interrupts we need.
347  * 3. By creating only two link descriptors, regardless of the number of
348  *    periods, we do not need to reallocate the link descriptors if the
349  *    number of periods changes.
350  * 4. All of the audio data is still stored in a single, contiguous DMA
351  *    buffer, which is what ALSA expects.  We're just dividing it into
352  *    contiguous parts, and creating a link descriptor for each one.
353  */
354 static int fsl_dma_open(struct snd_soc_component *component,
355 			struct snd_pcm_substream *substream)
356 {
357 	struct snd_pcm_runtime *runtime = substream->runtime;
358 	struct device *dev = component->dev;
359 	struct dma_object *dma =
360 		container_of(component->driver, struct dma_object, dai);
361 	struct fsl_dma_private *dma_private;
362 	struct ccsr_dma_channel __iomem *dma_channel;
363 	dma_addr_t ld_buf_phys;
364 	u64 temp_link;  	/* Pointer to next link descriptor */
365 	u32 mr;
366 	int ret = 0;
367 	unsigned int i;
368 
369 	/*
370 	 * Reject any DMA buffer whose size is not a multiple of the period
371 	 * size.  We need to make sure that the DMA buffer can be evenly divided
372 	 * into periods.
373 	 */
374 	ret = snd_pcm_hw_constraint_integer(runtime,
375 		SNDRV_PCM_HW_PARAM_PERIODS);
376 	if (ret < 0) {
377 		dev_err(dev, "invalid buffer size\n");
378 		return ret;
379 	}
380 
381 	if (dma->assigned) {
382 		dev_err(dev, "dma channel already assigned\n");
383 		return -EBUSY;
384 	}
385 
386 	dma_private = dma_alloc_coherent(dev, sizeof(struct fsl_dma_private),
387 					 &ld_buf_phys, GFP_KERNEL);
388 	if (!dma_private) {
389 		dev_err(dev, "can't allocate dma private data\n");
390 		return -ENOMEM;
391 	}
392 	if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK)
393 		dma_private->ssi_sxx_phys = dma->ssi_stx_phys;
394 	else
395 		dma_private->ssi_sxx_phys = dma->ssi_srx_phys;
396 
397 	dma_private->ssi_fifo_depth = dma->ssi_fifo_depth;
398 	dma_private->dma_channel = dma->channel;
399 	dma_private->irq = dma->irq;
400 	dma_private->substream = substream;
401 	dma_private->ld_buf_phys = ld_buf_phys;
402 	dma_private->dma_buf_phys = substream->dma_buffer.addr;
403 
404 	ret = request_irq(dma_private->irq, fsl_dma_isr, 0, "fsldma-audio",
405 			  dma_private);
406 	if (ret) {
407 		dev_err(dev, "can't register ISR for IRQ %u (ret=%i)\n",
408 			dma_private->irq, ret);
409 		dma_free_coherent(dev, sizeof(struct fsl_dma_private),
410 			dma_private, dma_private->ld_buf_phys);
411 		return ret;
412 	}
413 
414 	dma->assigned = true;
415 
416 	snd_soc_set_runtime_hwparams(substream, &fsl_dma_hardware);
417 	runtime->private_data = dma_private;
418 
419 	/* Program the fixed DMA controller parameters */
420 
421 	dma_channel = dma_private->dma_channel;
422 
423 	temp_link = dma_private->ld_buf_phys +
424 		sizeof(struct fsl_dma_link_descriptor);
425 
426 	for (i = 0; i < NUM_DMA_LINKS; i++) {
427 		dma_private->link[i].next = cpu_to_be64(temp_link);
428 
429 		temp_link += sizeof(struct fsl_dma_link_descriptor);
430 	}
431 	/* The last link descriptor points to the first */
432 	dma_private->link[i - 1].next = cpu_to_be64(dma_private->ld_buf_phys);
433 
434 	/* Tell the DMA controller where the first link descriptor is */
435 	out_be32(&dma_channel->clndar,
436 		CCSR_DMA_CLNDAR_ADDR(dma_private->ld_buf_phys));
437 	out_be32(&dma_channel->eclndar,
438 		CCSR_DMA_ECLNDAR_ADDR(dma_private->ld_buf_phys));
439 
440 	/* The manual says the BCR must be clear before enabling EMP */
441 	out_be32(&dma_channel->bcr, 0);
442 
443 	/*
444 	 * Program the mode register for interrupts, external master control,
445 	 * and source/destination hold.  Also clear the Channel Abort bit.
446 	 */
447 	mr = in_be32(&dma_channel->mr) &
448 		~(CCSR_DMA_MR_CA | CCSR_DMA_MR_DAHE | CCSR_DMA_MR_SAHE);
449 
450 	/*
451 	 * We want External Master Start and External Master Pause enabled,
452 	 * because the SSI is controlling the DMA controller.  We want the DMA
453 	 * controller to be set up in advance, and then we signal only the SSI
454 	 * to start transferring.
455 	 *
456 	 * We want End-Of-Segment Interrupts enabled, because this will generate
457 	 * an interrupt at the end of each segment (each link descriptor
458 	 * represents one segment).  Each DMA segment is the same thing as an
459 	 * ALSA period, so this is how we get an interrupt at the end of every
460 	 * period.
461 	 *
462 	 * We want Error Interrupt enabled, so that we can get an error if
463 	 * the DMA controller is mis-programmed somehow.
464 	 */
465 	mr |= CCSR_DMA_MR_EOSIE | CCSR_DMA_MR_EIE | CCSR_DMA_MR_EMP_EN |
466 		CCSR_DMA_MR_EMS_EN;
467 
468 	/* For playback, we want the destination address to be held.  For
469 	   capture, set the source address to be held. */
470 	mr |= (substream->stream == SNDRV_PCM_STREAM_PLAYBACK) ?
471 		CCSR_DMA_MR_DAHE : CCSR_DMA_MR_SAHE;
472 
473 	out_be32(&dma_channel->mr, mr);
474 
475 	return 0;
476 }
477 
478 /**
479  * fsl_dma_hw_params: continue initializing the DMA links
480  *
481  * This function obtains hardware parameters about the opened stream and
482  * programs the DMA controller accordingly.
483  *
484  * One drawback of big-endian is that when copying integers of different
485  * sizes to a fixed-sized register, the address to which the integer must be
486  * copied is dependent on the size of the integer.
487  *
488  * For example, if P is the address of a 32-bit register, and X is a 32-bit
489  * integer, then X should be copied to address P.  However, if X is a 16-bit
490  * integer, then it should be copied to P+2.  If X is an 8-bit register,
491  * then it should be copied to P+3.
492  *
493  * So for playback of 8-bit samples, the DMA controller must transfer single
494  * bytes from the DMA buffer to the last byte of the STX0 register, i.e.
495  * offset by 3 bytes. For 16-bit samples, the offset is two bytes.
496  *
497  * For 24-bit samples, the offset is 1 byte.  However, the DMA controller
498  * does not support 3-byte copies (the DAHTS register supports only 1, 2, 4,
499  * and 8 bytes at a time).  So we do not support packed 24-bit samples.
500  * 24-bit data must be padded to 32 bits.
501  */
502 static int fsl_dma_hw_params(struct snd_soc_component *component,
503 			     struct snd_pcm_substream *substream,
504 			     struct snd_pcm_hw_params *hw_params)
505 {
506 	struct snd_pcm_runtime *runtime = substream->runtime;
507 	struct fsl_dma_private *dma_private = runtime->private_data;
508 	struct device *dev = component->dev;
509 
510 	/* Number of bits per sample */
511 	unsigned int sample_bits =
512 		snd_pcm_format_physical_width(params_format(hw_params));
513 
514 	/* Number of bytes per frame */
515 	unsigned int sample_bytes = sample_bits / 8;
516 
517 	/* Bus address of SSI STX register */
518 	dma_addr_t ssi_sxx_phys = dma_private->ssi_sxx_phys;
519 
520 	/* Size of the DMA buffer, in bytes */
521 	size_t buffer_size = params_buffer_bytes(hw_params);
522 
523 	/* Number of bytes per period */
524 	size_t period_size = params_period_bytes(hw_params);
525 
526 	/* Pointer to next period */
527 	dma_addr_t temp_addr = substream->dma_buffer.addr;
528 
529 	/* Pointer to DMA controller */
530 	struct ccsr_dma_channel __iomem *dma_channel = dma_private->dma_channel;
531 
532 	u32 mr; /* DMA Mode Register */
533 
534 	unsigned int i;
535 
536 	/* Initialize our DMA tracking variables */
537 	dma_private->period_size = period_size;
538 	dma_private->num_periods = params_periods(hw_params);
539 	dma_private->dma_buf_end = dma_private->dma_buf_phys + buffer_size;
540 	dma_private->dma_buf_next = dma_private->dma_buf_phys +
541 		(NUM_DMA_LINKS * period_size);
542 
543 	if (dma_private->dma_buf_next >= dma_private->dma_buf_end)
544 		/* This happens if the number of periods == NUM_DMA_LINKS */
545 		dma_private->dma_buf_next = dma_private->dma_buf_phys;
546 
547 	mr = in_be32(&dma_channel->mr) & ~(CCSR_DMA_MR_BWC_MASK |
548 		  CCSR_DMA_MR_SAHTS_MASK | CCSR_DMA_MR_DAHTS_MASK);
549 
550 	/* Due to a quirk of the SSI's STX register, the target address
551 	 * for the DMA operations depends on the sample size.  So we calculate
552 	 * that offset here.  While we're at it, also tell the DMA controller
553 	 * how much data to transfer per sample.
554 	 */
555 	switch (sample_bits) {
556 	case 8:
557 		mr |= CCSR_DMA_MR_DAHTS_1 | CCSR_DMA_MR_SAHTS_1;
558 		ssi_sxx_phys += 3;
559 		break;
560 	case 16:
561 		mr |= CCSR_DMA_MR_DAHTS_2 | CCSR_DMA_MR_SAHTS_2;
562 		ssi_sxx_phys += 2;
563 		break;
564 	case 32:
565 		mr |= CCSR_DMA_MR_DAHTS_4 | CCSR_DMA_MR_SAHTS_4;
566 		break;
567 	default:
568 		/* We should never get here */
569 		dev_err(dev, "unsupported sample size %u\n", sample_bits);
570 		return -EINVAL;
571 	}
572 
573 	/*
574 	 * BWC determines how many bytes are sent/received before the DMA
575 	 * controller checks the SSI to see if it needs to stop. BWC should
576 	 * always be a multiple of the frame size, so that we always transmit
577 	 * whole frames.  Each frame occupies two slots in the FIFO.  The
578 	 * parameter for CCSR_DMA_MR_BWC() is rounded down the next power of two
579 	 * (MR[BWC] can only represent even powers of two).
580 	 *
581 	 * To simplify the process, we set BWC to the largest value that is
582 	 * less than or equal to the FIFO watermark.  For playback, this ensures
583 	 * that we transfer the maximum amount without overrunning the FIFO.
584 	 * For capture, this ensures that we transfer the maximum amount without
585 	 * underrunning the FIFO.
586 	 *
587 	 * f = SSI FIFO depth
588 	 * w = SSI watermark value (which equals f - 2)
589 	 * b = DMA bandwidth count (in bytes)
590 	 * s = sample size (in bytes, which equals frame_size * 2)
591 	 *
592 	 * For playback, we never transmit more than the transmit FIFO
593 	 * watermark, otherwise we might write more data than the FIFO can hold.
594 	 * The watermark is equal to the FIFO depth minus two.
595 	 *
596 	 * For capture, two equations must hold:
597 	 *	w > f - (b / s)
598 	 *	w >= b / s
599 	 *
600 	 * So, b > 2 * s, but b must also be <= s * w.  To simplify, we set
601 	 * b = s * w, which is equal to
602 	 *      (dma_private->ssi_fifo_depth - 2) * sample_bytes.
603 	 */
604 	mr |= CCSR_DMA_MR_BWC((dma_private->ssi_fifo_depth - 2) * sample_bytes);
605 
606 	out_be32(&dma_channel->mr, mr);
607 
608 	for (i = 0; i < NUM_DMA_LINKS; i++) {
609 		struct fsl_dma_link_descriptor *link = &dma_private->link[i];
610 
611 		link->count = cpu_to_be32(period_size);
612 
613 		/* The snoop bit tells the DMA controller whether it should tell
614 		 * the ECM to snoop during a read or write to an address. For
615 		 * audio, we use DMA to transfer data between memory and an I/O
616 		 * device (the SSI's STX0 or SRX0 register). Snooping is only
617 		 * needed if there is a cache, so we need to snoop memory
618 		 * addresses only.  For playback, that means we snoop the source
619 		 * but not the destination.  For capture, we snoop the
620 		 * destination but not the source.
621 		 *
622 		 * Note that failing to snoop properly is unlikely to cause
623 		 * cache incoherency if the period size is larger than the
624 		 * size of L1 cache.  This is because filling in one period will
625 		 * flush out the data for the previous period.  So if you
626 		 * increased period_bytes_min to a large enough size, you might
627 		 * get more performance by not snooping, and you'll still be
628 		 * okay.  You'll need to update fsl_dma_update_pointers() also.
629 		 */
630 		if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK) {
631 			link->source_addr = cpu_to_be32(temp_addr);
632 			link->source_attr = cpu_to_be32(CCSR_DMA_ATR_SNOOP |
633 				upper_32_bits(temp_addr));
634 
635 			link->dest_addr = cpu_to_be32(ssi_sxx_phys);
636 			link->dest_attr = cpu_to_be32(CCSR_DMA_ATR_NOSNOOP |
637 				upper_32_bits(ssi_sxx_phys));
638 		} else {
639 			link->source_addr = cpu_to_be32(ssi_sxx_phys);
640 			link->source_attr = cpu_to_be32(CCSR_DMA_ATR_NOSNOOP |
641 				upper_32_bits(ssi_sxx_phys));
642 
643 			link->dest_addr = cpu_to_be32(temp_addr);
644 			link->dest_attr = cpu_to_be32(CCSR_DMA_ATR_SNOOP |
645 				upper_32_bits(temp_addr));
646 		}
647 
648 		temp_addr += period_size;
649 	}
650 
651 	return 0;
652 }
653 
654 /**
655  * fsl_dma_pointer: determine the current position of the DMA transfer
656  *
657  * This function is called by ALSA when ALSA wants to know where in the
658  * stream buffer the hardware currently is.
659  *
660  * For playback, the SAR register contains the physical address of the most
661  * recent DMA transfer.  For capture, the value is in the DAR register.
662  *
663  * The base address of the buffer is stored in the source_addr field of the
664  * first link descriptor.
665  */
666 static snd_pcm_uframes_t fsl_dma_pointer(struct snd_soc_component *component,
667 					 struct snd_pcm_substream *substream)
668 {
669 	struct snd_pcm_runtime *runtime = substream->runtime;
670 	struct fsl_dma_private *dma_private = runtime->private_data;
671 	struct device *dev = component->dev;
672 	struct ccsr_dma_channel __iomem *dma_channel = dma_private->dma_channel;
673 	dma_addr_t position;
674 	snd_pcm_uframes_t frames;
675 
676 	/* Obtain the current DMA pointer, but don't read the ESAD bits if we
677 	 * only have 32-bit DMA addresses.  This function is typically called
678 	 * in interrupt context, so we need to optimize it.
679 	 */
680 	if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK) {
681 		position = in_be32(&dma_channel->sar);
682 #ifdef CONFIG_PHYS_64BIT
683 		position |= (u64)(in_be32(&dma_channel->satr) &
684 				  CCSR_DMA_ATR_ESAD_MASK) << 32;
685 #endif
686 	} else {
687 		position = in_be32(&dma_channel->dar);
688 #ifdef CONFIG_PHYS_64BIT
689 		position |= (u64)(in_be32(&dma_channel->datr) &
690 				  CCSR_DMA_ATR_ESAD_MASK) << 32;
691 #endif
692 	}
693 
694 	/*
695 	 * When capture is started, the SSI immediately starts to fill its FIFO.
696 	 * This means that the DMA controller is not started until the FIFO is
697 	 * full.  However, ALSA calls this function before that happens, when
698 	 * MR.DAR is still zero.  In this case, just return zero to indicate
699 	 * that nothing has been received yet.
700 	 */
701 	if (!position)
702 		return 0;
703 
704 	if ((position < dma_private->dma_buf_phys) ||
705 	    (position > dma_private->dma_buf_end)) {
706 		dev_err(dev, "dma pointer is out of range, halting stream\n");
707 		return SNDRV_PCM_POS_XRUN;
708 	}
709 
710 	frames = bytes_to_frames(runtime, position - dma_private->dma_buf_phys);
711 
712 	/*
713 	 * If the current address is just past the end of the buffer, wrap it
714 	 * around.
715 	 */
716 	if (frames == runtime->buffer_size)
717 		frames = 0;
718 
719 	return frames;
720 }
721 
722 /**
723  * fsl_dma_hw_free: release resources allocated in fsl_dma_hw_params()
724  *
725  * Release the resources allocated in fsl_dma_hw_params() and de-program the
726  * registers.
727  *
728  * This function can be called multiple times.
729  */
730 static int fsl_dma_hw_free(struct snd_soc_component *component,
731 			   struct snd_pcm_substream *substream)
732 {
733 	struct snd_pcm_runtime *runtime = substream->runtime;
734 	struct fsl_dma_private *dma_private = runtime->private_data;
735 
736 	if (dma_private) {
737 		struct ccsr_dma_channel __iomem *dma_channel;
738 
739 		dma_channel = dma_private->dma_channel;
740 
741 		/* Stop the DMA */
742 		out_be32(&dma_channel->mr, CCSR_DMA_MR_CA);
743 		out_be32(&dma_channel->mr, 0);
744 
745 		/* Reset all the other registers */
746 		out_be32(&dma_channel->sr, -1);
747 		out_be32(&dma_channel->clndar, 0);
748 		out_be32(&dma_channel->eclndar, 0);
749 		out_be32(&dma_channel->satr, 0);
750 		out_be32(&dma_channel->sar, 0);
751 		out_be32(&dma_channel->datr, 0);
752 		out_be32(&dma_channel->dar, 0);
753 		out_be32(&dma_channel->bcr, 0);
754 		out_be32(&dma_channel->nlndar, 0);
755 		out_be32(&dma_channel->enlndar, 0);
756 	}
757 
758 	return 0;
759 }
760 
761 /**
762  * fsl_dma_close: close the stream.
763  */
764 static int fsl_dma_close(struct snd_soc_component *component,
765 			 struct snd_pcm_substream *substream)
766 {
767 	struct snd_pcm_runtime *runtime = substream->runtime;
768 	struct fsl_dma_private *dma_private = runtime->private_data;
769 	struct device *dev = component->dev;
770 	struct dma_object *dma =
771 		container_of(component->driver, struct dma_object, dai);
772 
773 	if (dma_private) {
774 		if (dma_private->irq)
775 			free_irq(dma_private->irq, dma_private);
776 
777 		/* Deallocate the fsl_dma_private structure */
778 		dma_free_coherent(dev, sizeof(struct fsl_dma_private),
779 				  dma_private, dma_private->ld_buf_phys);
780 		substream->runtime->private_data = NULL;
781 	}
782 
783 	dma->assigned = false;
784 
785 	return 0;
786 }
787 
788 /**
789  * find_ssi_node -- returns the SSI node that points to its DMA channel node
790  *
791  * Although this DMA driver attempts to operate independently of the other
792  * devices, it still needs to determine some information about the SSI device
793  * that it's working with.  Unfortunately, the device tree does not contain
794  * a pointer from the DMA channel node to the SSI node -- the pointer goes the
795  * other way.  So we need to scan the device tree for SSI nodes until we find
796  * the one that points to the given DMA channel node.  It's ugly, but at least
797  * it's contained in this one function.
798  */
799 static struct device_node *find_ssi_node(struct device_node *dma_channel_np)
800 {
801 	struct device_node *ssi_np, *np;
802 
803 	for_each_compatible_node(ssi_np, NULL, "fsl,mpc8610-ssi") {
804 		/* Check each DMA phandle to see if it points to us.  We
805 		 * assume that device_node pointers are a valid comparison.
806 		 */
807 		np = of_parse_phandle(ssi_np, "fsl,playback-dma", 0);
808 		of_node_put(np);
809 		if (np == dma_channel_np)
810 			return ssi_np;
811 
812 		np = of_parse_phandle(ssi_np, "fsl,capture-dma", 0);
813 		of_node_put(np);
814 		if (np == dma_channel_np)
815 			return ssi_np;
816 	}
817 
818 	return NULL;
819 }
820 
821 static int fsl_soc_dma_probe(struct platform_device *pdev)
822 {
823 	struct dma_object *dma;
824 	struct device_node *np = pdev->dev.of_node;
825 	struct device_node *ssi_np;
826 	struct resource res;
827 	const uint32_t *iprop;
828 	int ret;
829 
830 	/* Find the SSI node that points to us. */
831 	ssi_np = find_ssi_node(np);
832 	if (!ssi_np) {
833 		dev_err(&pdev->dev, "cannot find parent SSI node\n");
834 		return -ENODEV;
835 	}
836 
837 	ret = of_address_to_resource(ssi_np, 0, &res);
838 	if (ret) {
839 		dev_err(&pdev->dev, "could not determine resources for %pOF\n",
840 			ssi_np);
841 		of_node_put(ssi_np);
842 		return ret;
843 	}
844 
845 	dma = kzalloc_obj(*dma);
846 	if (!dma) {
847 		of_node_put(ssi_np);
848 		return -ENOMEM;
849 	}
850 
851 	dma->dai.name = DRV_NAME;
852 	dma->dai.open = fsl_dma_open;
853 	dma->dai.close = fsl_dma_close;
854 	dma->dai.hw_params = fsl_dma_hw_params;
855 	dma->dai.hw_free = fsl_dma_hw_free;
856 	dma->dai.pointer = fsl_dma_pointer;
857 	dma->dai.pcm_new = fsl_dma_new;
858 
859 	/* Store the SSI-specific information that we need */
860 	dma->ssi_stx_phys = res.start + REG_SSI_STX0;
861 	dma->ssi_srx_phys = res.start + REG_SSI_SRX0;
862 
863 	iprop = of_get_property(ssi_np, "fsl,fifo-depth", NULL);
864 	if (iprop)
865 		dma->ssi_fifo_depth = be32_to_cpup(iprop);
866 	else
867                 /* Older 8610 DTs didn't have the fifo-depth property */
868 		dma->ssi_fifo_depth = 8;
869 
870 	of_node_put(ssi_np);
871 
872 	ret = devm_snd_soc_register_component(&pdev->dev, &dma->dai, NULL, 0);
873 	if (ret) {
874 		dev_err(&pdev->dev, "could not register platform\n");
875 		kfree(dma);
876 		return ret;
877 	}
878 
879 	dma->channel = of_iomap(np, 0);
880 	dma->irq = irq_of_parse_and_map(np, 0);
881 
882 	dev_set_drvdata(&pdev->dev, dma);
883 
884 	return 0;
885 }
886 
887 static void fsl_soc_dma_remove(struct platform_device *pdev)
888 {
889 	struct dma_object *dma = dev_get_drvdata(&pdev->dev);
890 
891 	iounmap(dma->channel);
892 	irq_dispose_mapping(dma->irq);
893 	kfree(dma);
894 }
895 
896 static const struct of_device_id fsl_soc_dma_ids[] = {
897 	{ .compatible = "fsl,ssi-dma-channel", },
898 	{}
899 };
900 MODULE_DEVICE_TABLE(of, fsl_soc_dma_ids);
901 
902 static struct platform_driver fsl_soc_dma_driver = {
903 	.driver = {
904 		.name = "fsl-pcm-audio",
905 		.of_match_table = fsl_soc_dma_ids,
906 	},
907 	.probe = fsl_soc_dma_probe,
908 	.remove = fsl_soc_dma_remove,
909 };
910 
911 module_platform_driver(fsl_soc_dma_driver);
912 
913 MODULE_AUTHOR("Timur Tabi <timur@freescale.com>");
914 MODULE_DESCRIPTION("Freescale Elo DMA ASoC PCM Driver");
915 MODULE_LICENSE("GPL v2");
916