xref: /linux/sound/soc/fsl/fsl_dma.c (revision e5c91aac491def6ab3f90c4cc246e3fcb0f8f058)
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/list.h>
22 #include <linux/slab.h>
23 
24 #include <sound/core.h>
25 #include <sound/pcm.h>
26 #include <sound/pcm_params.h>
27 #include <sound/soc.h>
28 
29 #include <asm/io.h>
30 
31 #include "fsl_dma.h"
32 #include "fsl_ssi.h"	/* For the offset of stx0 and srx0 */
33 
34 #define DRV_NAME "fsl_dma"
35 
36 /*
37  * The formats that the DMA controller supports, which is anything
38  * that is 8, 16, or 32 bits.
39  */
40 #define FSLDMA_PCM_FORMATS (SNDRV_PCM_FMTBIT_S8 	| \
41 			    SNDRV_PCM_FMTBIT_U8 	| \
42 			    SNDRV_PCM_FMTBIT_S16_LE     | \
43 			    SNDRV_PCM_FMTBIT_S16_BE     | \
44 			    SNDRV_PCM_FMTBIT_U16_LE     | \
45 			    SNDRV_PCM_FMTBIT_U16_BE     | \
46 			    SNDRV_PCM_FMTBIT_S24_LE     | \
47 			    SNDRV_PCM_FMTBIT_S24_BE     | \
48 			    SNDRV_PCM_FMTBIT_U24_LE     | \
49 			    SNDRV_PCM_FMTBIT_U24_BE     | \
50 			    SNDRV_PCM_FMTBIT_S32_LE     | \
51 			    SNDRV_PCM_FMTBIT_S32_BE     | \
52 			    SNDRV_PCM_FMTBIT_U32_LE     | \
53 			    SNDRV_PCM_FMTBIT_U32_BE)
54 struct dma_object {
55 	struct snd_soc_component_driver dai;
56 	dma_addr_t ssi_stx_phys;
57 	dma_addr_t ssi_srx_phys;
58 	unsigned int ssi_fifo_depth;
59 	struct ccsr_dma_channel __iomem *channel;
60 	unsigned int irq;
61 	bool assigned;
62 };
63 
64 /*
65  * The number of DMA links to use.  Two is the bare minimum, but if you
66  * have really small links you might need more.
67  */
68 #define NUM_DMA_LINKS   2
69 
70 /** fsl_dma_private: p-substream DMA data
71  *
72  * Each substream has a 1-to-1 association with a DMA channel.
73  *
74  * The link[] array is first because it needs to be aligned on a 32-byte
75  * boundary, so putting it first will ensure alignment without padding the
76  * structure.
77  *
78  * @link[]: array of link descriptors
79  * @dma_channel: pointer to the DMA channel's registers
80  * @irq: IRQ for this DMA channel
81  * @substream: pointer to the substream object, needed by the ISR
82  * @ssi_sxx_phys: bus address of the STX or SRX register to use
83  * @ld_buf_phys: physical address of the LD buffer
84  * @current_link: index into link[] of the link currently being processed
85  * @dma_buf_phys: physical address of the DMA buffer
86  * @dma_buf_next: physical address of the next period to process
87  * @dma_buf_end: physical address of the byte after the end of the DMA
88  * @buffer period_size: the size of a single period
89  * @num_periods: the number of periods in the DMA buffer
90  */
91 struct fsl_dma_private {
92 	struct fsl_dma_link_descriptor link[NUM_DMA_LINKS];
93 	struct ccsr_dma_channel __iomem *dma_channel;
94 	unsigned int irq;
95 	struct snd_pcm_substream *substream;
96 	dma_addr_t ssi_sxx_phys;
97 	unsigned int ssi_fifo_depth;
98 	dma_addr_t ld_buf_phys;
99 	unsigned int current_link;
100 	dma_addr_t dma_buf_phys;
101 	dma_addr_t dma_buf_next;
102 	dma_addr_t dma_buf_end;
103 	size_t period_size;
104 	unsigned int num_periods;
105 };
106 
107 /**
108  * fsl_dma_hardare: define characteristics of the PCM hardware.
109  *
110  * The PCM hardware is the Freescale DMA controller.  This structure defines
111  * the capabilities of that hardware.
112  *
113  * Since the sampling rate and data format are not controlled by the DMA
114  * controller, we specify no limits for those values.  The only exception is
115  * period_bytes_min, which is set to a reasonably low value to prevent the
116  * DMA controller from generating too many interrupts per second.
117  *
118  * Since each link descriptor has a 32-bit byte count field, we set
119  * period_bytes_max to the largest 32-bit number.  We also have no maximum
120  * number of periods.
121  *
122  * Note that we specify SNDRV_PCM_INFO_JOINT_DUPLEX here, but only because a
123  * limitation in the SSI driver requires the sample rates for playback and
124  * capture to be the same.
125  */
126 static const struct snd_pcm_hardware fsl_dma_hardware = {
127 
128 	.info   		= SNDRV_PCM_INFO_INTERLEAVED |
129 				  SNDRV_PCM_INFO_MMAP |
130 				  SNDRV_PCM_INFO_MMAP_VALID |
131 				  SNDRV_PCM_INFO_JOINT_DUPLEX |
132 				  SNDRV_PCM_INFO_PAUSE,
133 	.formats		= FSLDMA_PCM_FORMATS,
134 	.period_bytes_min       = 512,  	/* A reasonable limit */
135 	.period_bytes_max       = (u32) -1,
136 	.periods_min    	= NUM_DMA_LINKS,
137 	.periods_max    	= (unsigned int) -1,
138 	.buffer_bytes_max       = 128 * 1024,   /* A reasonable limit */
139 };
140 
141 /**
142  * fsl_dma_abort_stream: tell ALSA that the DMA transfer has aborted
143  *
144  * This function should be called by the ISR whenever the DMA controller
145  * halts data transfer.
146  */
147 static void fsl_dma_abort_stream(struct snd_pcm_substream *substream)
148 {
149 	snd_pcm_stop_xrun(substream);
150 }
151 
152 /**
153  * fsl_dma_update_pointers - update LD pointers to point to the next period
154  *
155  * As each period is completed, this function changes the link
156  * descriptor pointers for that period to point to the next period.
157  */
158 static void fsl_dma_update_pointers(struct fsl_dma_private *dma_private)
159 {
160 	struct fsl_dma_link_descriptor *link =
161 		&dma_private->link[dma_private->current_link];
162 
163 	/* Update our link descriptors to point to the next period. On a 36-bit
164 	 * system, we also need to update the ESAD bits.  We also set (keep) the
165 	 * snoop bits.  See the comments in fsl_dma_hw_params() about snooping.
166 	 */
167 	if (dma_private->substream->stream == SNDRV_PCM_STREAM_PLAYBACK) {
168 		link->source_addr = cpu_to_be32(dma_private->dma_buf_next);
169 #ifdef CONFIG_PHYS_64BIT
170 		link->source_attr = cpu_to_be32(CCSR_DMA_ATR_SNOOP |
171 			upper_32_bits(dma_private->dma_buf_next));
172 #endif
173 	} else {
174 		link->dest_addr = cpu_to_be32(dma_private->dma_buf_next);
175 #ifdef CONFIG_PHYS_64BIT
176 		link->dest_attr = cpu_to_be32(CCSR_DMA_ATR_SNOOP |
177 			upper_32_bits(dma_private->dma_buf_next));
178 #endif
179 	}
180 
181 	/* Update our variables for next time */
182 	dma_private->dma_buf_next += dma_private->period_size;
183 
184 	if (dma_private->dma_buf_next >= dma_private->dma_buf_end)
185 		dma_private->dma_buf_next = dma_private->dma_buf_phys;
186 
187 	if (++dma_private->current_link >= NUM_DMA_LINKS)
188 		dma_private->current_link = 0;
189 }
190 
191 /**
192  * fsl_dma_isr: interrupt handler for the DMA controller
193  *
194  * @irq: IRQ of the DMA channel
195  * @dev_id: pointer to the dma_private structure for this DMA channel
196  */
197 static irqreturn_t fsl_dma_isr(int irq, void *dev_id)
198 {
199 	struct fsl_dma_private *dma_private = dev_id;
200 	struct snd_pcm_substream *substream = dma_private->substream;
201 	struct snd_soc_pcm_runtime *rtd = snd_soc_substream_to_rtd(substream);
202 	struct device *dev = rtd->dev;
203 	struct ccsr_dma_channel __iomem *dma_channel = dma_private->dma_channel;
204 	irqreturn_t ret = IRQ_NONE;
205 	u32 sr, sr2 = 0;
206 
207 	/* We got an interrupt, so read the status register to see what we
208 	   were interrupted for.
209 	 */
210 	sr = in_be32(&dma_channel->sr);
211 
212 	if (sr & CCSR_DMA_SR_TE) {
213 		dev_err(dev, "dma transmit error\n");
214 		fsl_dma_abort_stream(substream);
215 		sr2 |= CCSR_DMA_SR_TE;
216 		ret = IRQ_HANDLED;
217 	}
218 
219 	if (sr & CCSR_DMA_SR_CH)
220 		ret = IRQ_HANDLED;
221 
222 	if (sr & CCSR_DMA_SR_PE) {
223 		dev_err(dev, "dma programming error\n");
224 		fsl_dma_abort_stream(substream);
225 		sr2 |= CCSR_DMA_SR_PE;
226 		ret = IRQ_HANDLED;
227 	}
228 
229 	if (sr & CCSR_DMA_SR_EOLNI) {
230 		sr2 |= CCSR_DMA_SR_EOLNI;
231 		ret = IRQ_HANDLED;
232 	}
233 
234 	if (sr & CCSR_DMA_SR_CB)
235 		ret = IRQ_HANDLED;
236 
237 	if (sr & CCSR_DMA_SR_EOSI) {
238 		/* Tell ALSA we completed a period. */
239 		snd_pcm_period_elapsed(substream);
240 
241 		/*
242 		 * Update our link descriptors to point to the next period. We
243 		 * only need to do this if the number of periods is not equal to
244 		 * the number of links.
245 		 */
246 		if (dma_private->num_periods != NUM_DMA_LINKS)
247 			fsl_dma_update_pointers(dma_private);
248 
249 		sr2 |= CCSR_DMA_SR_EOSI;
250 		ret = IRQ_HANDLED;
251 	}
252 
253 	if (sr & CCSR_DMA_SR_EOLSI) {
254 		sr2 |= CCSR_DMA_SR_EOLSI;
255 		ret = IRQ_HANDLED;
256 	}
257 
258 	/* Clear the bits that we set */
259 	if (sr2)
260 		out_be32(&dma_channel->sr, sr2);
261 
262 	return ret;
263 }
264 
265 /**
266  * fsl_dma_new: initialize this PCM driver.
267  *
268  * This function is called by soc_new_pcm(), once for each DAI link
269  * in the machine driver's snd_soc_card structure.
270  *
271  * Regardless of where the memory is actually allocated, since the device can
272  * technically DMA to any 36-bit address, we do need to set the DMA mask to 36.
273  */
274 static int fsl_dma_new(struct snd_soc_component *component,
275 		       struct snd_soc_pcm_runtime *rtd)
276 {
277 	struct snd_card *card = rtd->card->snd_card;
278 	struct snd_pcm *pcm = rtd->pcm;
279 	int ret;
280 
281 	ret = dma_coerce_mask_and_coherent(card->dev, DMA_BIT_MASK(36));
282 	if (ret)
283 		return ret;
284 
285 	return snd_pcm_set_fixed_buffer_all(pcm, SNDRV_DMA_TYPE_DEV,
286 					    card->dev,
287 					    fsl_dma_hardware.buffer_bytes_max);
288 }
289 
290 /**
291  * fsl_dma_open: open a new substream.
292  *
293  * Each substream has its own DMA buffer.
294  *
295  * ALSA divides the DMA buffer into N periods.  We create NUM_DMA_LINKS link
296  * descriptors that ping-pong from one period to the next.  For example, if
297  * there are six periods and two link descriptors, this is how they look
298  * before playback starts:
299  *
300  *      	   The last link descriptor
301  *   ____________  points back to the first
302  *  |   	 |
303  *  V   	 |
304  *  ___    ___   |
305  * |   |->|   |->|
306  * |___|  |___|
307  *   |      |
308  *   |      |
309  *   V      V
310  *  _________________________________________
311  * |      |      |      |      |      |      |  The DMA buffer is
312  * |      |      |      |      |      |      |    divided into 6 parts
313  * |______|______|______|______|______|______|
314  *
315  * and here's how they look after the first period is finished playing:
316  *
317  *   ____________
318  *  |   	 |
319  *  V   	 |
320  *  ___    ___   |
321  * |   |->|   |->|
322  * |___|  |___|
323  *   |      |
324  *   |______________
325  *          |       |
326  *          V       V
327  *  _________________________________________
328  * |      |      |      |      |      |      |
329  * |      |      |      |      |      |      |
330  * |______|______|______|______|______|______|
331  *
332  * The first link descriptor now points to the third period.  The DMA
333  * controller is currently playing the second period.  When it finishes, it
334  * will jump back to the first descriptor and play the third period.
335  *
336  * There are four reasons we do this:
337  *
338  * 1. The only way to get the DMA controller to automatically restart the
339  *    transfer when it gets to the end of the buffer is to use chaining
340  *    mode.  Basic direct mode doesn't offer that feature.
341  * 2. We need to receive an interrupt at the end of every period.  The DMA
342  *    controller can generate an interrupt at the end of every link transfer
343  *    (aka segment).  Making each period into a DMA segment will give us the
344  *    interrupts we need.
345  * 3. By creating only two link descriptors, regardless of the number of
346  *    periods, we do not need to reallocate the link descriptors if the
347  *    number of periods changes.
348  * 4. All of the audio data is still stored in a single, contiguous DMA
349  *    buffer, which is what ALSA expects.  We're just dividing it into
350  *    contiguous parts, and creating a link descriptor for each one.
351  */
352 static int fsl_dma_open(struct snd_soc_component *component,
353 			struct snd_pcm_substream *substream)
354 {
355 	struct snd_pcm_runtime *runtime = substream->runtime;
356 	struct device *dev = component->dev;
357 	struct dma_object *dma =
358 		container_of(component->driver, struct dma_object, dai);
359 	struct fsl_dma_private *dma_private;
360 	struct ccsr_dma_channel __iomem *dma_channel;
361 	dma_addr_t ld_buf_phys;
362 	u64 temp_link;  	/* Pointer to next link descriptor */
363 	u32 mr;
364 	int ret = 0;
365 	unsigned int i;
366 
367 	/*
368 	 * Reject any DMA buffer whose size is not a multiple of the period
369 	 * size.  We need to make sure that the DMA buffer can be evenly divided
370 	 * into periods.
371 	 */
372 	ret = snd_pcm_hw_constraint_integer(runtime,
373 		SNDRV_PCM_HW_PARAM_PERIODS);
374 	if (ret < 0) {
375 		dev_err(dev, "invalid buffer size\n");
376 		return ret;
377 	}
378 
379 	if (dma->assigned) {
380 		dev_err(dev, "dma channel already assigned\n");
381 		return -EBUSY;
382 	}
383 
384 	dma_private = dma_alloc_coherent(dev, sizeof(struct fsl_dma_private),
385 					 &ld_buf_phys, GFP_KERNEL);
386 	if (!dma_private) {
387 		dev_err(dev, "can't allocate dma private data\n");
388 		return -ENOMEM;
389 	}
390 	if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK)
391 		dma_private->ssi_sxx_phys = dma->ssi_stx_phys;
392 	else
393 		dma_private->ssi_sxx_phys = dma->ssi_srx_phys;
394 
395 	dma_private->ssi_fifo_depth = dma->ssi_fifo_depth;
396 	dma_private->dma_channel = dma->channel;
397 	dma_private->irq = dma->irq;
398 	dma_private->substream = substream;
399 	dma_private->ld_buf_phys = ld_buf_phys;
400 	dma_private->dma_buf_phys = substream->dma_buffer.addr;
401 
402 	ret = request_irq(dma_private->irq, fsl_dma_isr, 0, "fsldma-audio",
403 			  dma_private);
404 	if (ret) {
405 		dev_err(dev, "can't register ISR for IRQ %u (ret=%i)\n",
406 			dma_private->irq, ret);
407 		dma_free_coherent(dev, sizeof(struct fsl_dma_private),
408 			dma_private, dma_private->ld_buf_phys);
409 		return ret;
410 	}
411 
412 	dma->assigned = true;
413 
414 	snd_soc_set_runtime_hwparams(substream, &fsl_dma_hardware);
415 	runtime->private_data = dma_private;
416 
417 	/* Program the fixed DMA controller parameters */
418 
419 	dma_channel = dma_private->dma_channel;
420 
421 	temp_link = dma_private->ld_buf_phys +
422 		sizeof(struct fsl_dma_link_descriptor);
423 
424 	for (i = 0; i < NUM_DMA_LINKS; i++) {
425 		dma_private->link[i].next = cpu_to_be64(temp_link);
426 
427 		temp_link += sizeof(struct fsl_dma_link_descriptor);
428 	}
429 	/* The last link descriptor points to the first */
430 	dma_private->link[i - 1].next = cpu_to_be64(dma_private->ld_buf_phys);
431 
432 	/* Tell the DMA controller where the first link descriptor is */
433 	out_be32(&dma_channel->clndar,
434 		CCSR_DMA_CLNDAR_ADDR(dma_private->ld_buf_phys));
435 	out_be32(&dma_channel->eclndar,
436 		CCSR_DMA_ECLNDAR_ADDR(dma_private->ld_buf_phys));
437 
438 	/* The manual says the BCR must be clear before enabling EMP */
439 	out_be32(&dma_channel->bcr, 0);
440 
441 	/*
442 	 * Program the mode register for interrupts, external master control,
443 	 * and source/destination hold.  Also clear the Channel Abort bit.
444 	 */
445 	mr = in_be32(&dma_channel->mr) &
446 		~(CCSR_DMA_MR_CA | CCSR_DMA_MR_DAHE | CCSR_DMA_MR_SAHE);
447 
448 	/*
449 	 * We want External Master Start and External Master Pause enabled,
450 	 * because the SSI is controlling the DMA controller.  We want the DMA
451 	 * controller to be set up in advance, and then we signal only the SSI
452 	 * to start transferring.
453 	 *
454 	 * We want End-Of-Segment Interrupts enabled, because this will generate
455 	 * an interrupt at the end of each segment (each link descriptor
456 	 * represents one segment).  Each DMA segment is the same thing as an
457 	 * ALSA period, so this is how we get an interrupt at the end of every
458 	 * period.
459 	 *
460 	 * We want Error Interrupt enabled, so that we can get an error if
461 	 * the DMA controller is mis-programmed somehow.
462 	 */
463 	mr |= CCSR_DMA_MR_EOSIE | CCSR_DMA_MR_EIE | CCSR_DMA_MR_EMP_EN |
464 		CCSR_DMA_MR_EMS_EN;
465 
466 	/* For playback, we want the destination address to be held.  For
467 	   capture, set the source address to be held. */
468 	mr |= (substream->stream == SNDRV_PCM_STREAM_PLAYBACK) ?
469 		CCSR_DMA_MR_DAHE : CCSR_DMA_MR_SAHE;
470 
471 	out_be32(&dma_channel->mr, mr);
472 
473 	return 0;
474 }
475 
476 /**
477  * fsl_dma_hw_params: continue initializing the DMA links
478  *
479  * This function obtains hardware parameters about the opened stream and
480  * programs the DMA controller accordingly.
481  *
482  * One drawback of big-endian is that when copying integers of different
483  * sizes to a fixed-sized register, the address to which the integer must be
484  * copied is dependent on the size of the integer.
485  *
486  * For example, if P is the address of a 32-bit register, and X is a 32-bit
487  * integer, then X should be copied to address P.  However, if X is a 16-bit
488  * integer, then it should be copied to P+2.  If X is an 8-bit register,
489  * then it should be copied to P+3.
490  *
491  * So for playback of 8-bit samples, the DMA controller must transfer single
492  * bytes from the DMA buffer to the last byte of the STX0 register, i.e.
493  * offset by 3 bytes. For 16-bit samples, the offset is two bytes.
494  *
495  * For 24-bit samples, the offset is 1 byte.  However, the DMA controller
496  * does not support 3-byte copies (the DAHTS register supports only 1, 2, 4,
497  * and 8 bytes at a time).  So we do not support packed 24-bit samples.
498  * 24-bit data must be padded to 32 bits.
499  */
500 static int fsl_dma_hw_params(struct snd_soc_component *component,
501 			     struct snd_pcm_substream *substream,
502 			     struct snd_pcm_hw_params *hw_params)
503 {
504 	struct snd_pcm_runtime *runtime = substream->runtime;
505 	struct fsl_dma_private *dma_private = runtime->private_data;
506 	struct device *dev = component->dev;
507 
508 	/* Number of bits per sample */
509 	unsigned int sample_bits =
510 		snd_pcm_format_physical_width(params_format(hw_params));
511 
512 	/* Number of bytes per frame */
513 	unsigned int sample_bytes = sample_bits / 8;
514 
515 	/* Bus address of SSI STX register */
516 	dma_addr_t ssi_sxx_phys = dma_private->ssi_sxx_phys;
517 
518 	/* Size of the DMA buffer, in bytes */
519 	size_t buffer_size = params_buffer_bytes(hw_params);
520 
521 	/* Number of bytes per period */
522 	size_t period_size = params_period_bytes(hw_params);
523 
524 	/* Pointer to next period */
525 	dma_addr_t temp_addr = substream->dma_buffer.addr;
526 
527 	/* Pointer to DMA controller */
528 	struct ccsr_dma_channel __iomem *dma_channel = dma_private->dma_channel;
529 
530 	u32 mr; /* DMA Mode Register */
531 
532 	unsigned int i;
533 
534 	/* Initialize our DMA tracking variables */
535 	dma_private->period_size = period_size;
536 	dma_private->num_periods = params_periods(hw_params);
537 	dma_private->dma_buf_end = dma_private->dma_buf_phys + buffer_size;
538 	dma_private->dma_buf_next = dma_private->dma_buf_phys +
539 		(NUM_DMA_LINKS * period_size);
540 
541 	if (dma_private->dma_buf_next >= dma_private->dma_buf_end)
542 		/* This happens if the number of periods == NUM_DMA_LINKS */
543 		dma_private->dma_buf_next = dma_private->dma_buf_phys;
544 
545 	mr = in_be32(&dma_channel->mr) & ~(CCSR_DMA_MR_BWC_MASK |
546 		  CCSR_DMA_MR_SAHTS_MASK | CCSR_DMA_MR_DAHTS_MASK);
547 
548 	/* Due to a quirk of the SSI's STX register, the target address
549 	 * for the DMA operations depends on the sample size.  So we calculate
550 	 * that offset here.  While we're at it, also tell the DMA controller
551 	 * how much data to transfer per sample.
552 	 */
553 	switch (sample_bits) {
554 	case 8:
555 		mr |= CCSR_DMA_MR_DAHTS_1 | CCSR_DMA_MR_SAHTS_1;
556 		ssi_sxx_phys += 3;
557 		break;
558 	case 16:
559 		mr |= CCSR_DMA_MR_DAHTS_2 | CCSR_DMA_MR_SAHTS_2;
560 		ssi_sxx_phys += 2;
561 		break;
562 	case 32:
563 		mr |= CCSR_DMA_MR_DAHTS_4 | CCSR_DMA_MR_SAHTS_4;
564 		break;
565 	default:
566 		/* We should never get here */
567 		dev_err(dev, "unsupported sample size %u\n", sample_bits);
568 		return -EINVAL;
569 	}
570 
571 	/*
572 	 * BWC determines how many bytes are sent/received before the DMA
573 	 * controller checks the SSI to see if it needs to stop. BWC should
574 	 * always be a multiple of the frame size, so that we always transmit
575 	 * whole frames.  Each frame occupies two slots in the FIFO.  The
576 	 * parameter for CCSR_DMA_MR_BWC() is rounded down the next power of two
577 	 * (MR[BWC] can only represent even powers of two).
578 	 *
579 	 * To simplify the process, we set BWC to the largest value that is
580 	 * less than or equal to the FIFO watermark.  For playback, this ensures
581 	 * that we transfer the maximum amount without overrunning the FIFO.
582 	 * For capture, this ensures that we transfer the maximum amount without
583 	 * underrunning the FIFO.
584 	 *
585 	 * f = SSI FIFO depth
586 	 * w = SSI watermark value (which equals f - 2)
587 	 * b = DMA bandwidth count (in bytes)
588 	 * s = sample size (in bytes, which equals frame_size * 2)
589 	 *
590 	 * For playback, we never transmit more than the transmit FIFO
591 	 * watermark, otherwise we might write more data than the FIFO can hold.
592 	 * The watermark is equal to the FIFO depth minus two.
593 	 *
594 	 * For capture, two equations must hold:
595 	 *	w > f - (b / s)
596 	 *	w >= b / s
597 	 *
598 	 * So, b > 2 * s, but b must also be <= s * w.  To simplify, we set
599 	 * b = s * w, which is equal to
600 	 *      (dma_private->ssi_fifo_depth - 2) * sample_bytes.
601 	 */
602 	mr |= CCSR_DMA_MR_BWC((dma_private->ssi_fifo_depth - 2) * sample_bytes);
603 
604 	out_be32(&dma_channel->mr, mr);
605 
606 	for (i = 0; i < NUM_DMA_LINKS; i++) {
607 		struct fsl_dma_link_descriptor *link = &dma_private->link[i];
608 
609 		link->count = cpu_to_be32(period_size);
610 
611 		/* The snoop bit tells the DMA controller whether it should tell
612 		 * the ECM to snoop during a read or write to an address. For
613 		 * audio, we use DMA to transfer data between memory and an I/O
614 		 * device (the SSI's STX0 or SRX0 register). Snooping is only
615 		 * needed if there is a cache, so we need to snoop memory
616 		 * addresses only.  For playback, that means we snoop the source
617 		 * but not the destination.  For capture, we snoop the
618 		 * destination but not the source.
619 		 *
620 		 * Note that failing to snoop properly is unlikely to cause
621 		 * cache incoherency if the period size is larger than the
622 		 * size of L1 cache.  This is because filling in one period will
623 		 * flush out the data for the previous period.  So if you
624 		 * increased period_bytes_min to a large enough size, you might
625 		 * get more performance by not snooping, and you'll still be
626 		 * okay.  You'll need to update fsl_dma_update_pointers() also.
627 		 */
628 		if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK) {
629 			link->source_addr = cpu_to_be32(temp_addr);
630 			link->source_attr = cpu_to_be32(CCSR_DMA_ATR_SNOOP |
631 				upper_32_bits(temp_addr));
632 
633 			link->dest_addr = cpu_to_be32(ssi_sxx_phys);
634 			link->dest_attr = cpu_to_be32(CCSR_DMA_ATR_NOSNOOP |
635 				upper_32_bits(ssi_sxx_phys));
636 		} else {
637 			link->source_addr = cpu_to_be32(ssi_sxx_phys);
638 			link->source_attr = cpu_to_be32(CCSR_DMA_ATR_NOSNOOP |
639 				upper_32_bits(ssi_sxx_phys));
640 
641 			link->dest_addr = cpu_to_be32(temp_addr);
642 			link->dest_attr = cpu_to_be32(CCSR_DMA_ATR_SNOOP |
643 				upper_32_bits(temp_addr));
644 		}
645 
646 		temp_addr += period_size;
647 	}
648 
649 	return 0;
650 }
651 
652 /**
653  * fsl_dma_pointer: determine the current position of the DMA transfer
654  *
655  * This function is called by ALSA when ALSA wants to know where in the
656  * stream buffer the hardware currently is.
657  *
658  * For playback, the SAR register contains the physical address of the most
659  * recent DMA transfer.  For capture, the value is in the DAR register.
660  *
661  * The base address of the buffer is stored in the source_addr field of the
662  * first link descriptor.
663  */
664 static snd_pcm_uframes_t fsl_dma_pointer(struct snd_soc_component *component,
665 					 struct snd_pcm_substream *substream)
666 {
667 	struct snd_pcm_runtime *runtime = substream->runtime;
668 	struct fsl_dma_private *dma_private = runtime->private_data;
669 	struct device *dev = component->dev;
670 	struct ccsr_dma_channel __iomem *dma_channel = dma_private->dma_channel;
671 	dma_addr_t position;
672 	snd_pcm_uframes_t frames;
673 
674 	/* Obtain the current DMA pointer, but don't read the ESAD bits if we
675 	 * only have 32-bit DMA addresses.  This function is typically called
676 	 * in interrupt context, so we need to optimize it.
677 	 */
678 	if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK) {
679 		position = in_be32(&dma_channel->sar);
680 #ifdef CONFIG_PHYS_64BIT
681 		position |= (u64)(in_be32(&dma_channel->satr) &
682 				  CCSR_DMA_ATR_ESAD_MASK) << 32;
683 #endif
684 	} else {
685 		position = in_be32(&dma_channel->dar);
686 #ifdef CONFIG_PHYS_64BIT
687 		position |= (u64)(in_be32(&dma_channel->datr) &
688 				  CCSR_DMA_ATR_ESAD_MASK) << 32;
689 #endif
690 	}
691 
692 	/*
693 	 * When capture is started, the SSI immediately starts to fill its FIFO.
694 	 * This means that the DMA controller is not started until the FIFO is
695 	 * full.  However, ALSA calls this function before that happens, when
696 	 * MR.DAR is still zero.  In this case, just return zero to indicate
697 	 * that nothing has been received yet.
698 	 */
699 	if (!position)
700 		return 0;
701 
702 	if ((position < dma_private->dma_buf_phys) ||
703 	    (position > dma_private->dma_buf_end)) {
704 		dev_err(dev, "dma pointer is out of range, halting stream\n");
705 		return SNDRV_PCM_POS_XRUN;
706 	}
707 
708 	frames = bytes_to_frames(runtime, position - dma_private->dma_buf_phys);
709 
710 	/*
711 	 * If the current address is just past the end of the buffer, wrap it
712 	 * around.
713 	 */
714 	if (frames == runtime->buffer_size)
715 		frames = 0;
716 
717 	return frames;
718 }
719 
720 /**
721  * fsl_dma_hw_free: release resources allocated in fsl_dma_hw_params()
722  *
723  * Release the resources allocated in fsl_dma_hw_params() and de-program the
724  * registers.
725  *
726  * This function can be called multiple times.
727  */
728 static int fsl_dma_hw_free(struct snd_soc_component *component,
729 			   struct snd_pcm_substream *substream)
730 {
731 	struct snd_pcm_runtime *runtime = substream->runtime;
732 	struct fsl_dma_private *dma_private = runtime->private_data;
733 
734 	if (dma_private) {
735 		struct ccsr_dma_channel __iomem *dma_channel;
736 
737 		dma_channel = dma_private->dma_channel;
738 
739 		/* Stop the DMA */
740 		out_be32(&dma_channel->mr, CCSR_DMA_MR_CA);
741 		out_be32(&dma_channel->mr, 0);
742 
743 		/* Reset all the other registers */
744 		out_be32(&dma_channel->sr, -1);
745 		out_be32(&dma_channel->clndar, 0);
746 		out_be32(&dma_channel->eclndar, 0);
747 		out_be32(&dma_channel->satr, 0);
748 		out_be32(&dma_channel->sar, 0);
749 		out_be32(&dma_channel->datr, 0);
750 		out_be32(&dma_channel->dar, 0);
751 		out_be32(&dma_channel->bcr, 0);
752 		out_be32(&dma_channel->nlndar, 0);
753 		out_be32(&dma_channel->enlndar, 0);
754 	}
755 
756 	return 0;
757 }
758 
759 /**
760  * fsl_dma_close: close the stream.
761  */
762 static int fsl_dma_close(struct snd_soc_component *component,
763 			 struct snd_pcm_substream *substream)
764 {
765 	struct snd_pcm_runtime *runtime = substream->runtime;
766 	struct fsl_dma_private *dma_private = runtime->private_data;
767 	struct device *dev = component->dev;
768 	struct dma_object *dma =
769 		container_of(component->driver, struct dma_object, dai);
770 
771 	if (dma_private) {
772 		if (dma_private->irq)
773 			free_irq(dma_private->irq, dma_private);
774 
775 		/* Deallocate the fsl_dma_private structure */
776 		dma_free_coherent(dev, sizeof(struct fsl_dma_private),
777 				  dma_private, dma_private->ld_buf_phys);
778 		substream->runtime->private_data = NULL;
779 	}
780 
781 	dma->assigned = false;
782 
783 	return 0;
784 }
785 
786 /**
787  * find_ssi_node -- returns the SSI node that points to its DMA channel node
788  *
789  * Although this DMA driver attempts to operate independently of the other
790  * devices, it still needs to determine some information about the SSI device
791  * that it's working with.  Unfortunately, the device tree does not contain
792  * a pointer from the DMA channel node to the SSI node -- the pointer goes the
793  * other way.  So we need to scan the device tree for SSI nodes until we find
794  * the one that points to the given DMA channel node.  It's ugly, but at least
795  * it's contained in this one function.
796  */
797 static struct device_node *find_ssi_node(struct device_node *dma_channel_np)
798 {
799 	struct device_node *ssi_np, *np;
800 
801 	for_each_compatible_node(ssi_np, NULL, "fsl,mpc8610-ssi") {
802 		/* Check each DMA phandle to see if it points to us.  We
803 		 * assume that device_node pointers are a valid comparison.
804 		 */
805 		np = of_parse_phandle(ssi_np, "fsl,playback-dma", 0);
806 		of_node_put(np);
807 		if (np == dma_channel_np)
808 			return ssi_np;
809 
810 		np = of_parse_phandle(ssi_np, "fsl,capture-dma", 0);
811 		of_node_put(np);
812 		if (np == dma_channel_np)
813 			return ssi_np;
814 	}
815 
816 	return NULL;
817 }
818 
819 static int fsl_soc_dma_probe(struct platform_device *pdev)
820 {
821 	struct dma_object *dma;
822 	struct device_node *np = pdev->dev.of_node;
823 	struct device_node *ssi_np;
824 	struct resource res;
825 	void __iomem *channel;
826 	const uint32_t *iprop;
827 	int irq;
828 	int ret;
829 
830 	channel = devm_platform_ioremap_resource(pdev, 0);
831 	if (IS_ERR(channel))
832 		return PTR_ERR(channel);
833 
834 	irq = platform_get_irq(pdev, 0);
835 	if (irq < 0)
836 		return irq;
837 
838 	dma = devm_kzalloc(&pdev->dev, sizeof(*dma), GFP_KERNEL);
839 	if (!dma)
840 		return -ENOMEM;
841 
842 	dma->dai.name = DRV_NAME;
843 	dma->dai.open = fsl_dma_open;
844 	dma->dai.close = fsl_dma_close;
845 	dma->dai.hw_params = fsl_dma_hw_params;
846 	dma->dai.hw_free = fsl_dma_hw_free;
847 	dma->dai.pointer = fsl_dma_pointer;
848 	dma->dai.pcm_new = fsl_dma_new;
849 
850 	dma->channel = channel;
851 	dma->irq = irq;
852 
853 	/* Find the SSI node that points to us. */
854 	ssi_np = find_ssi_node(np);
855 	if (!ssi_np) {
856 		dev_err(&pdev->dev, "cannot find parent SSI node\n");
857 		return -ENODEV;
858 	}
859 
860 	ret = of_address_to_resource(ssi_np, 0, &res);
861 	if (ret) {
862 		dev_err(&pdev->dev, "could not determine resources for %pOF\n",
863 			ssi_np);
864 		of_node_put(ssi_np);
865 		return ret;
866 	}
867 
868 	/* Store the SSI-specific information that we need */
869 	dma->ssi_stx_phys = res.start + REG_SSI_STX0;
870 	dma->ssi_srx_phys = res.start + REG_SSI_SRX0;
871 
872 	iprop = of_get_property(ssi_np, "fsl,fifo-depth", NULL);
873 	of_node_put(ssi_np);
874 	if (iprop)
875 		dma->ssi_fifo_depth = be32_to_cpup(iprop);
876 	else
877                 /* Older 8610 DTs didn't have the fifo-depth property */
878 		dma->ssi_fifo_depth = 8;
879 
880 	return devm_snd_soc_register_component(&pdev->dev, &dma->dai, NULL, 0);
881 }
882 
883 static const struct of_device_id fsl_soc_dma_ids[] = {
884 	{ .compatible = "fsl,ssi-dma-channel", },
885 	{}
886 };
887 MODULE_DEVICE_TABLE(of, fsl_soc_dma_ids);
888 
889 static struct platform_driver fsl_soc_dma_driver = {
890 	.driver = {
891 		.name = "fsl-pcm-audio",
892 		.of_match_table = fsl_soc_dma_ids,
893 	},
894 	.probe = fsl_soc_dma_probe,
895 };
896 
897 module_platform_driver(fsl_soc_dma_driver);
898 
899 MODULE_AUTHOR("Timur Tabi <timur@freescale.com>");
900 MODULE_DESCRIPTION("Freescale Elo DMA ASoC PCM Driver");
901 MODULE_LICENSE("GPL v2");
902