1/*
2 * Audio and Music Data Transmission Protocol (IEC 61883-6) streams
3 * with Common Isochronous Packet (IEC 61883-1) headers
4 *
5 * Copyright (c) Clemens Ladisch <clemens@ladisch.de>
6 * Licensed under the terms of the GNU General Public License, version 2.
7 */
8
9#include <linux/device.h>
10#include <linux/err.h>
11#include <linux/firewire.h>
12#include <linux/module.h>
13#include <linux/slab.h>
14#include <linux/sched.h>
15#include <sound/pcm.h>
16#include <sound/pcm_params.h>
17#include <sound/rawmidi.h>
18#include "amdtp.h"
19
20#define TICKS_PER_CYCLE		3072
21#define CYCLES_PER_SECOND	8000
22#define TICKS_PER_SECOND	(TICKS_PER_CYCLE * CYCLES_PER_SECOND)
23
24/*
25 * Nominally 3125 bytes/second, but the MIDI port's clock might be
26 * 1% too slow, and the bus clock 100 ppm too fast.
27 */
28#define MIDI_BYTES_PER_SECOND	3093
29
30/*
31 * Several devices look only at the first eight data blocks.
32 * In any case, this is more than enough for the MIDI data rate.
33 */
34#define MAX_MIDI_RX_BLOCKS	8
35
36#define TRANSFER_DELAY_TICKS	0x2e00 /* 479.17 microseconds */
37
38/* isochronous header parameters */
39#define ISO_DATA_LENGTH_SHIFT	16
40#define TAG_CIP			1
41
42/* common isochronous packet header parameters */
43#define CIP_EOH			(1u << 31)
44#define CIP_EOH_MASK		0x80000000
45#define CIP_FMT_AM		(0x10 << 24)
46#define CIP_FMT_MASK		0x3f000000
47#define CIP_SYT_MASK		0x0000ffff
48#define CIP_SYT_NO_INFO		0xffff
49#define CIP_FDF_MASK		0x00ff0000
50#define CIP_FDF_SFC_SHIFT	16
51
52/*
53 * Audio and Music transfer protocol specific parameters
54 * only "Clock-based rate control mode" is supported
55 */
56#define AMDTP_FDF_AM824		(0 << (CIP_FDF_SFC_SHIFT + 3))
57#define AMDTP_FDF_NO_DATA	0xff
58#define AMDTP_DBS_MASK		0x00ff0000
59#define AMDTP_DBS_SHIFT		16
60#define AMDTP_DBC_MASK		0x000000ff
61
62/* TODO: make these configurable */
63#define INTERRUPT_INTERVAL	16
64#define QUEUE_LENGTH		48
65
66#define IN_PACKET_HEADER_SIZE	4
67#define OUT_PACKET_HEADER_SIZE	0
68
69static void pcm_period_tasklet(unsigned long data);
70
71/**
72 * amdtp_stream_init - initialize an AMDTP stream structure
73 * @s: the AMDTP stream to initialize
74 * @unit: the target of the stream
75 * @dir: the direction of stream
76 * @flags: the packet transmission method to use
77 */
78int amdtp_stream_init(struct amdtp_stream *s, struct fw_unit *unit,
79		      enum amdtp_stream_direction dir, enum cip_flags flags)
80{
81	s->unit = unit;
82	s->direction = dir;
83	s->flags = flags;
84	s->context = ERR_PTR(-1);
85	mutex_init(&s->mutex);
86	tasklet_init(&s->period_tasklet, pcm_period_tasklet, (unsigned long)s);
87	s->packet_index = 0;
88
89	init_waitqueue_head(&s->callback_wait);
90	s->callbacked = false;
91	s->sync_slave = NULL;
92
93	return 0;
94}
95EXPORT_SYMBOL(amdtp_stream_init);
96
97/**
98 * amdtp_stream_destroy - free stream resources
99 * @s: the AMDTP stream to destroy
100 */
101void amdtp_stream_destroy(struct amdtp_stream *s)
102{
103	WARN_ON(amdtp_stream_running(s));
104	mutex_destroy(&s->mutex);
105}
106EXPORT_SYMBOL(amdtp_stream_destroy);
107
108const unsigned int amdtp_syt_intervals[CIP_SFC_COUNT] = {
109	[CIP_SFC_32000]  =  8,
110	[CIP_SFC_44100]  =  8,
111	[CIP_SFC_48000]  =  8,
112	[CIP_SFC_88200]  = 16,
113	[CIP_SFC_96000]  = 16,
114	[CIP_SFC_176400] = 32,
115	[CIP_SFC_192000] = 32,
116};
117EXPORT_SYMBOL(amdtp_syt_intervals);
118
119const unsigned int amdtp_rate_table[CIP_SFC_COUNT] = {
120	[CIP_SFC_32000]  =  32000,
121	[CIP_SFC_44100]  =  44100,
122	[CIP_SFC_48000]  =  48000,
123	[CIP_SFC_88200]  =  88200,
124	[CIP_SFC_96000]  =  96000,
125	[CIP_SFC_176400] = 176400,
126	[CIP_SFC_192000] = 192000,
127};
128EXPORT_SYMBOL(amdtp_rate_table);
129
130/**
131 * amdtp_stream_add_pcm_hw_constraints - add hw constraints for PCM substream
132 * @s:		the AMDTP stream, which must be initialized.
133 * @runtime:	the PCM substream runtime
134 */
135int amdtp_stream_add_pcm_hw_constraints(struct amdtp_stream *s,
136					struct snd_pcm_runtime *runtime)
137{
138	int err;
139
140	/* AM824 in IEC 61883-6 can deliver 24bit data */
141	err = snd_pcm_hw_constraint_msbits(runtime, 0, 32, 24);
142	if (err < 0)
143		goto end;
144
145	/*
146	 * Currently firewire-lib processes 16 packets in one software
147	 * interrupt callback. This equals to 2msec but actually the
148	 * interval of the interrupts has a jitter.
149	 * Additionally, even if adding a constraint to fit period size to
150	 * 2msec, actual calculated frames per period doesn't equal to 2msec,
151	 * depending on sampling rate.
152	 * Anyway, the interval to call snd_pcm_period_elapsed() cannot 2msec.
153	 * Here let us use 5msec for safe period interrupt.
154	 */
155	err = snd_pcm_hw_constraint_minmax(runtime,
156					   SNDRV_PCM_HW_PARAM_PERIOD_TIME,
157					   5000, UINT_MAX);
158	if (err < 0)
159		goto end;
160
161	/* Non-Blocking stream has no more constraints */
162	if (!(s->flags & CIP_BLOCKING))
163		goto end;
164
165	/*
166	 * One AMDTP packet can include some frames. In blocking mode, the
167	 * number equals to SYT_INTERVAL. So the number is 8, 16 or 32,
168	 * depending on its sampling rate. For accurate period interrupt, it's
169	 * preferrable to align period/buffer sizes to current SYT_INTERVAL.
170	 *
171	 * TODO: These constraints can be improved with proper rules.
172	 * Currently apply LCM of SYT_INTERVALs.
173	 */
174	err = snd_pcm_hw_constraint_step(runtime, 0,
175					 SNDRV_PCM_HW_PARAM_PERIOD_SIZE, 32);
176	if (err < 0)
177		goto end;
178	err = snd_pcm_hw_constraint_step(runtime, 0,
179					 SNDRV_PCM_HW_PARAM_BUFFER_SIZE, 32);
180end:
181	return err;
182}
183EXPORT_SYMBOL(amdtp_stream_add_pcm_hw_constraints);
184
185/**
186 * amdtp_stream_set_parameters - set stream parameters
187 * @s: the AMDTP stream to configure
188 * @rate: the sample rate
189 * @pcm_channels: the number of PCM samples in each data block, to be encoded
190 *                as AM824 multi-bit linear audio
191 * @midi_ports: the number of MIDI ports (i.e., MPX-MIDI Data Channels)
192 *
193 * The parameters must be set before the stream is started, and must not be
194 * changed while the stream is running.
195 */
196void amdtp_stream_set_parameters(struct amdtp_stream *s,
197				 unsigned int rate,
198				 unsigned int pcm_channels,
199				 unsigned int midi_ports)
200{
201	unsigned int i, sfc, midi_channels;
202
203	midi_channels = DIV_ROUND_UP(midi_ports, 8);
204
205	if (WARN_ON(amdtp_stream_running(s)) |
206	    WARN_ON(pcm_channels > AMDTP_MAX_CHANNELS_FOR_PCM) |
207	    WARN_ON(midi_channels > AMDTP_MAX_CHANNELS_FOR_MIDI))
208		return;
209
210	for (sfc = 0; sfc < ARRAY_SIZE(amdtp_rate_table); ++sfc)
211		if (amdtp_rate_table[sfc] == rate)
212			goto sfc_found;
213	WARN_ON(1);
214	return;
215
216sfc_found:
217	s->pcm_channels = pcm_channels;
218	s->sfc = sfc;
219	s->data_block_quadlets = s->pcm_channels + midi_channels;
220	s->midi_ports = midi_ports;
221
222	s->syt_interval = amdtp_syt_intervals[sfc];
223
224	/* default buffering in the device */
225	s->transfer_delay = TRANSFER_DELAY_TICKS - TICKS_PER_CYCLE;
226	if (s->flags & CIP_BLOCKING)
227		/* additional buffering needed to adjust for no-data packets */
228		s->transfer_delay += TICKS_PER_SECOND * s->syt_interval / rate;
229
230	/* init the position map for PCM and MIDI channels */
231	for (i = 0; i < pcm_channels; i++)
232		s->pcm_positions[i] = i;
233	s->midi_position = s->pcm_channels;
234
235	/*
236	 * We do not know the actual MIDI FIFO size of most devices.  Just
237	 * assume two bytes, i.e., one byte can be received over the bus while
238	 * the previous one is transmitted over MIDI.
239	 * (The value here is adjusted for midi_ratelimit_per_packet().)
240	 */
241	s->midi_fifo_limit = rate - MIDI_BYTES_PER_SECOND * s->syt_interval + 1;
242}
243EXPORT_SYMBOL(amdtp_stream_set_parameters);
244
245/**
246 * amdtp_stream_get_max_payload - get the stream's packet size
247 * @s: the AMDTP stream
248 *
249 * This function must not be called before the stream has been configured
250 * with amdtp_stream_set_parameters().
251 */
252unsigned int amdtp_stream_get_max_payload(struct amdtp_stream *s)
253{
254	return 8 + s->syt_interval * s->data_block_quadlets * 4;
255}
256EXPORT_SYMBOL(amdtp_stream_get_max_payload);
257
258static void amdtp_write_s16(struct amdtp_stream *s,
259			    struct snd_pcm_substream *pcm,
260			    __be32 *buffer, unsigned int frames);
261static void amdtp_write_s32(struct amdtp_stream *s,
262			    struct snd_pcm_substream *pcm,
263			    __be32 *buffer, unsigned int frames);
264static void amdtp_read_s32(struct amdtp_stream *s,
265			   struct snd_pcm_substream *pcm,
266			   __be32 *buffer, unsigned int frames);
267
268/**
269 * amdtp_stream_set_pcm_format - set the PCM format
270 * @s: the AMDTP stream to configure
271 * @format: the format of the ALSA PCM device
272 *
273 * The sample format must be set after the other parameters (rate/PCM channels/
274 * MIDI) and before the stream is started, and must not be changed while the
275 * stream is running.
276 */
277void amdtp_stream_set_pcm_format(struct amdtp_stream *s,
278				 snd_pcm_format_t format)
279{
280	if (WARN_ON(amdtp_stream_pcm_running(s)))
281		return;
282
283	switch (format) {
284	default:
285		WARN_ON(1);
286		/* fall through */
287	case SNDRV_PCM_FORMAT_S16:
288		if (s->direction == AMDTP_OUT_STREAM) {
289			s->transfer_samples = amdtp_write_s16;
290			break;
291		}
292		WARN_ON(1);
293		/* fall through */
294	case SNDRV_PCM_FORMAT_S32:
295		if (s->direction == AMDTP_OUT_STREAM)
296			s->transfer_samples = amdtp_write_s32;
297		else
298			s->transfer_samples = amdtp_read_s32;
299		break;
300	}
301}
302EXPORT_SYMBOL(amdtp_stream_set_pcm_format);
303
304/**
305 * amdtp_stream_pcm_prepare - prepare PCM device for running
306 * @s: the AMDTP stream
307 *
308 * This function should be called from the PCM device's .prepare callback.
309 */
310void amdtp_stream_pcm_prepare(struct amdtp_stream *s)
311{
312	tasklet_kill(&s->period_tasklet);
313	s->pcm_buffer_pointer = 0;
314	s->pcm_period_pointer = 0;
315	s->pointer_flush = true;
316}
317EXPORT_SYMBOL(amdtp_stream_pcm_prepare);
318
319static unsigned int calculate_data_blocks(struct amdtp_stream *s)
320{
321	unsigned int phase, data_blocks;
322
323	if (s->flags & CIP_BLOCKING)
324		data_blocks = s->syt_interval;
325	else if (!cip_sfc_is_base_44100(s->sfc)) {
326		/* Sample_rate / 8000 is an integer, and precomputed. */
327		data_blocks = s->data_block_state;
328	} else {
329		phase = s->data_block_state;
330
331		/*
332		 * This calculates the number of data blocks per packet so that
333		 * 1) the overall rate is correct and exactly synchronized to
334		 *    the bus clock, and
335		 * 2) packets with a rounded-up number of blocks occur as early
336		 *    as possible in the sequence (to prevent underruns of the
337		 *    device's buffer).
338		 */
339		if (s->sfc == CIP_SFC_44100)
340			/* 6 6 5 6 5 6 5 ... */
341			data_blocks = 5 + ((phase & 1) ^
342					   (phase == 0 || phase >= 40));
343		else
344			/* 12 11 11 11 11 ... or 23 22 22 22 22 ... */
345			data_blocks = 11 * (s->sfc >> 1) + (phase == 0);
346		if (++phase >= (80 >> (s->sfc >> 1)))
347			phase = 0;
348		s->data_block_state = phase;
349	}
350
351	return data_blocks;
352}
353
354static unsigned int calculate_syt(struct amdtp_stream *s,
355				  unsigned int cycle)
356{
357	unsigned int syt_offset, phase, index, syt;
358
359	if (s->last_syt_offset < TICKS_PER_CYCLE) {
360		if (!cip_sfc_is_base_44100(s->sfc))
361			syt_offset = s->last_syt_offset + s->syt_offset_state;
362		else {
363		/*
364		 * The time, in ticks, of the n'th SYT_INTERVAL sample is:
365		 *   n * SYT_INTERVAL * 24576000 / sample_rate
366		 * Modulo TICKS_PER_CYCLE, the difference between successive
367		 * elements is about 1386.23.  Rounding the results of this
368		 * formula to the SYT precision results in a sequence of
369		 * differences that begins with:
370		 *   1386 1386 1387 1386 1386 1386 1387 1386 1386 1386 1387 ...
371		 * This code generates _exactly_ the same sequence.
372		 */
373			phase = s->syt_offset_state;
374			index = phase % 13;
375			syt_offset = s->last_syt_offset;
376			syt_offset += 1386 + ((index && !(index & 3)) ||
377					      phase == 146);
378			if (++phase >= 147)
379				phase = 0;
380			s->syt_offset_state = phase;
381		}
382	} else
383		syt_offset = s->last_syt_offset - TICKS_PER_CYCLE;
384	s->last_syt_offset = syt_offset;
385
386	if (syt_offset < TICKS_PER_CYCLE) {
387		syt_offset += s->transfer_delay;
388		syt = (cycle + syt_offset / TICKS_PER_CYCLE) << 12;
389		syt += syt_offset % TICKS_PER_CYCLE;
390
391		return syt & CIP_SYT_MASK;
392	} else {
393		return CIP_SYT_NO_INFO;
394	}
395}
396
397static void amdtp_write_s32(struct amdtp_stream *s,
398			    struct snd_pcm_substream *pcm,
399			    __be32 *buffer, unsigned int frames)
400{
401	struct snd_pcm_runtime *runtime = pcm->runtime;
402	unsigned int channels, remaining_frames, i, c;
403	const u32 *src;
404
405	channels = s->pcm_channels;
406	src = (void *)runtime->dma_area +
407			frames_to_bytes(runtime, s->pcm_buffer_pointer);
408	remaining_frames = runtime->buffer_size - s->pcm_buffer_pointer;
409
410	for (i = 0; i < frames; ++i) {
411		for (c = 0; c < channels; ++c) {
412			buffer[s->pcm_positions[c]] =
413					cpu_to_be32((*src >> 8) | 0x40000000);
414			src++;
415		}
416		buffer += s->data_block_quadlets;
417		if (--remaining_frames == 0)
418			src = (void *)runtime->dma_area;
419	}
420}
421
422static void amdtp_write_s16(struct amdtp_stream *s,
423			    struct snd_pcm_substream *pcm,
424			    __be32 *buffer, unsigned int frames)
425{
426	struct snd_pcm_runtime *runtime = pcm->runtime;
427	unsigned int channels, remaining_frames, i, c;
428	const u16 *src;
429
430	channels = s->pcm_channels;
431	src = (void *)runtime->dma_area +
432			frames_to_bytes(runtime, s->pcm_buffer_pointer);
433	remaining_frames = runtime->buffer_size - s->pcm_buffer_pointer;
434
435	for (i = 0; i < frames; ++i) {
436		for (c = 0; c < channels; ++c) {
437			buffer[s->pcm_positions[c]] =
438					cpu_to_be32((*src << 8) | 0x42000000);
439			src++;
440		}
441		buffer += s->data_block_quadlets;
442		if (--remaining_frames == 0)
443			src = (void *)runtime->dma_area;
444	}
445}
446
447static void amdtp_read_s32(struct amdtp_stream *s,
448			   struct snd_pcm_substream *pcm,
449			   __be32 *buffer, unsigned int frames)
450{
451	struct snd_pcm_runtime *runtime = pcm->runtime;
452	unsigned int channels, remaining_frames, i, c;
453	u32 *dst;
454
455	channels = s->pcm_channels;
456	dst  = (void *)runtime->dma_area +
457			frames_to_bytes(runtime, s->pcm_buffer_pointer);
458	remaining_frames = runtime->buffer_size - s->pcm_buffer_pointer;
459
460	for (i = 0; i < frames; ++i) {
461		for (c = 0; c < channels; ++c) {
462			*dst = be32_to_cpu(buffer[s->pcm_positions[c]]) << 8;
463			dst++;
464		}
465		buffer += s->data_block_quadlets;
466		if (--remaining_frames == 0)
467			dst = (void *)runtime->dma_area;
468	}
469}
470
471static void amdtp_fill_pcm_silence(struct amdtp_stream *s,
472				   __be32 *buffer, unsigned int frames)
473{
474	unsigned int i, c;
475
476	for (i = 0; i < frames; ++i) {
477		for (c = 0; c < s->pcm_channels; ++c)
478			buffer[s->pcm_positions[c]] = cpu_to_be32(0x40000000);
479		buffer += s->data_block_quadlets;
480	}
481}
482
483/*
484 * To avoid sending MIDI bytes at too high a rate, assume that the receiving
485 * device has a FIFO, and track how much it is filled.  This values increases
486 * by one whenever we send one byte in a packet, but the FIFO empties at
487 * a constant rate independent of our packet rate.  One packet has syt_interval
488 * samples, so the number of bytes that empty out of the FIFO, per packet(!),
489 * is MIDI_BYTES_PER_SECOND * syt_interval / sample_rate.  To avoid storing
490 * fractional values, the values in midi_fifo_used[] are measured in bytes
491 * multiplied by the sample rate.
492 */
493static bool midi_ratelimit_per_packet(struct amdtp_stream *s, unsigned int port)
494{
495	int used;
496
497	used = s->midi_fifo_used[port];
498	if (used == 0) /* common shortcut */
499		return true;
500
501	used -= MIDI_BYTES_PER_SECOND * s->syt_interval;
502	used = max(used, 0);
503	s->midi_fifo_used[port] = used;
504
505	return used < s->midi_fifo_limit;
506}
507
508static void midi_rate_use_one_byte(struct amdtp_stream *s, unsigned int port)
509{
510	s->midi_fifo_used[port] += amdtp_rate_table[s->sfc];
511}
512
513static void amdtp_fill_midi(struct amdtp_stream *s,
514			    __be32 *buffer, unsigned int frames)
515{
516	unsigned int f, port;
517	u8 *b;
518
519	for (f = 0; f < frames; f++) {
520		b = (u8 *)&buffer[s->midi_position];
521
522		port = (s->data_block_counter + f) % 8;
523		if (f < MAX_MIDI_RX_BLOCKS &&
524		    midi_ratelimit_per_packet(s, port) &&
525		    s->midi[port] != NULL &&
526		    snd_rawmidi_transmit(s->midi[port], &b[1], 1) == 1) {
527			midi_rate_use_one_byte(s, port);
528			b[0] = 0x81;
529		} else {
530			b[0] = 0x80;
531			b[1] = 0;
532		}
533		b[2] = 0;
534		b[3] = 0;
535
536		buffer += s->data_block_quadlets;
537	}
538}
539
540static void amdtp_pull_midi(struct amdtp_stream *s,
541			    __be32 *buffer, unsigned int frames)
542{
543	unsigned int f, port;
544	int len;
545	u8 *b;
546
547	for (f = 0; f < frames; f++) {
548		port = (s->data_block_counter + f) % 8;
549		b = (u8 *)&buffer[s->midi_position];
550
551		len = b[0] - 0x80;
552		if ((1 <= len) &&  (len <= 3) && (s->midi[port]))
553			snd_rawmidi_receive(s->midi[port], b + 1, len);
554
555		buffer += s->data_block_quadlets;
556	}
557}
558
559static void update_pcm_pointers(struct amdtp_stream *s,
560				struct snd_pcm_substream *pcm,
561				unsigned int frames)
562{
563	unsigned int ptr;
564
565	/*
566	 * In IEC 61883-6, one data block represents one event. In ALSA, one
567	 * event equals to one PCM frame. But Dice has a quirk to transfer
568	 * two PCM frames in one data block.
569	 */
570	if (s->double_pcm_frames)
571		frames *= 2;
572
573	ptr = s->pcm_buffer_pointer + frames;
574	if (ptr >= pcm->runtime->buffer_size)
575		ptr -= pcm->runtime->buffer_size;
576	ACCESS_ONCE(s->pcm_buffer_pointer) = ptr;
577
578	s->pcm_period_pointer += frames;
579	if (s->pcm_period_pointer >= pcm->runtime->period_size) {
580		s->pcm_period_pointer -= pcm->runtime->period_size;
581		s->pointer_flush = false;
582		tasklet_hi_schedule(&s->period_tasklet);
583	}
584}
585
586static void pcm_period_tasklet(unsigned long data)
587{
588	struct amdtp_stream *s = (void *)data;
589	struct snd_pcm_substream *pcm = ACCESS_ONCE(s->pcm);
590
591	if (pcm)
592		snd_pcm_period_elapsed(pcm);
593}
594
595static int queue_packet(struct amdtp_stream *s,
596			unsigned int header_length,
597			unsigned int payload_length, bool skip)
598{
599	struct fw_iso_packet p = {0};
600	int err = 0;
601
602	if (IS_ERR(s->context))
603		goto end;
604
605	p.interrupt = IS_ALIGNED(s->packet_index + 1, INTERRUPT_INTERVAL);
606	p.tag = TAG_CIP;
607	p.header_length = header_length;
608	p.payload_length = (!skip) ? payload_length : 0;
609	p.skip = skip;
610	err = fw_iso_context_queue(s->context, &p, &s->buffer.iso_buffer,
611				   s->buffer.packets[s->packet_index].offset);
612	if (err < 0) {
613		dev_err(&s->unit->device, "queueing error: %d\n", err);
614		goto end;
615	}
616
617	if (++s->packet_index >= QUEUE_LENGTH)
618		s->packet_index = 0;
619end:
620	return err;
621}
622
623static inline int queue_out_packet(struct amdtp_stream *s,
624				   unsigned int payload_length, bool skip)
625{
626	return queue_packet(s, OUT_PACKET_HEADER_SIZE,
627			    payload_length, skip);
628}
629
630static inline int queue_in_packet(struct amdtp_stream *s)
631{
632	return queue_packet(s, IN_PACKET_HEADER_SIZE,
633			    amdtp_stream_get_max_payload(s), false);
634}
635
636static void handle_out_packet(struct amdtp_stream *s, unsigned int syt)
637{
638	__be32 *buffer;
639	unsigned int data_blocks, payload_length;
640	struct snd_pcm_substream *pcm;
641
642	if (s->packet_index < 0)
643		return;
644
645	/* this module generate empty packet for 'no data' */
646	if (!(s->flags & CIP_BLOCKING) || (syt != CIP_SYT_NO_INFO))
647		data_blocks = calculate_data_blocks(s);
648	else
649		data_blocks = 0;
650
651	buffer = s->buffer.packets[s->packet_index].buffer;
652	buffer[0] = cpu_to_be32(ACCESS_ONCE(s->source_node_id_field) |
653				(s->data_block_quadlets << AMDTP_DBS_SHIFT) |
654				s->data_block_counter);
655	buffer[1] = cpu_to_be32(CIP_EOH | CIP_FMT_AM | AMDTP_FDF_AM824 |
656				(s->sfc << CIP_FDF_SFC_SHIFT) | syt);
657	buffer += 2;
658
659	pcm = ACCESS_ONCE(s->pcm);
660	if (pcm)
661		s->transfer_samples(s, pcm, buffer, data_blocks);
662	else
663		amdtp_fill_pcm_silence(s, buffer, data_blocks);
664	if (s->midi_ports)
665		amdtp_fill_midi(s, buffer, data_blocks);
666
667	s->data_block_counter = (s->data_block_counter + data_blocks) & 0xff;
668
669	payload_length = 8 + data_blocks * 4 * s->data_block_quadlets;
670	if (queue_out_packet(s, payload_length, false) < 0) {
671		s->packet_index = -1;
672		amdtp_stream_pcm_abort(s);
673		return;
674	}
675
676	if (pcm)
677		update_pcm_pointers(s, pcm, data_blocks);
678}
679
680static void handle_in_packet(struct amdtp_stream *s,
681			     unsigned int payload_quadlets,
682			     __be32 *buffer)
683{
684	u32 cip_header[2];
685	unsigned int data_blocks, data_block_quadlets, data_block_counter,
686		     dbc_interval;
687	struct snd_pcm_substream *pcm = NULL;
688	bool lost;
689
690	cip_header[0] = be32_to_cpu(buffer[0]);
691	cip_header[1] = be32_to_cpu(buffer[1]);
692
693	/*
694	 * This module supports 'Two-quadlet CIP header with SYT field'.
695	 * For convenience, also check FMT field is AM824 or not.
696	 */
697	if (((cip_header[0] & CIP_EOH_MASK) == CIP_EOH) ||
698	    ((cip_header[1] & CIP_EOH_MASK) != CIP_EOH) ||
699	    ((cip_header[1] & CIP_FMT_MASK) != CIP_FMT_AM)) {
700		dev_info_ratelimited(&s->unit->device,
701				"Invalid CIP header for AMDTP: %08X:%08X\n",
702				cip_header[0], cip_header[1]);
703		goto end;
704	}
705
706	/* Calculate data blocks */
707	if (payload_quadlets < 3 ||
708	    ((cip_header[1] & CIP_FDF_MASK) ==
709				(AMDTP_FDF_NO_DATA << CIP_FDF_SFC_SHIFT))) {
710		data_blocks = 0;
711	} else {
712		data_block_quadlets =
713			(cip_header[0] & AMDTP_DBS_MASK) >> AMDTP_DBS_SHIFT;
714		/* avoid division by zero */
715		if (data_block_quadlets == 0) {
716			dev_info_ratelimited(&s->unit->device,
717				"Detect invalid value in dbs field: %08X\n",
718				cip_header[0]);
719			goto err;
720		}
721		if (s->flags & CIP_WRONG_DBS)
722			data_block_quadlets = s->data_block_quadlets;
723
724		data_blocks = (payload_quadlets - 2) / data_block_quadlets;
725	}
726
727	/* Check data block counter continuity */
728	data_block_counter = cip_header[0] & AMDTP_DBC_MASK;
729	if (data_blocks == 0 && (s->flags & CIP_EMPTY_HAS_WRONG_DBC) &&
730	    s->data_block_counter != UINT_MAX)
731		data_block_counter = s->data_block_counter;
732
733	if (((s->flags & CIP_SKIP_DBC_ZERO_CHECK) &&
734	     data_block_counter == s->tx_first_dbc) ||
735	    s->data_block_counter == UINT_MAX) {
736		lost = false;
737	} else if (!(s->flags & CIP_DBC_IS_END_EVENT)) {
738		lost = data_block_counter != s->data_block_counter;
739	} else {
740		if ((data_blocks > 0) && (s->tx_dbc_interval > 0))
741			dbc_interval = s->tx_dbc_interval;
742		else
743			dbc_interval = data_blocks;
744
745		lost = data_block_counter !=
746		       ((s->data_block_counter + dbc_interval) & 0xff);
747	}
748
749	if (lost) {
750		dev_info(&s->unit->device,
751			 "Detect discontinuity of CIP: %02X %02X\n",
752			 s->data_block_counter, data_block_counter);
753		goto err;
754	}
755
756	if (data_blocks > 0) {
757		buffer += 2;
758
759		pcm = ACCESS_ONCE(s->pcm);
760		if (pcm)
761			s->transfer_samples(s, pcm, buffer, data_blocks);
762
763		if (s->midi_ports)
764			amdtp_pull_midi(s, buffer, data_blocks);
765	}
766
767	if (s->flags & CIP_DBC_IS_END_EVENT)
768		s->data_block_counter = data_block_counter;
769	else
770		s->data_block_counter =
771				(data_block_counter + data_blocks) & 0xff;
772end:
773	if (queue_in_packet(s) < 0)
774		goto err;
775
776	if (pcm)
777		update_pcm_pointers(s, pcm, data_blocks);
778
779	return;
780err:
781	s->packet_index = -1;
782	amdtp_stream_pcm_abort(s);
783}
784
785static void out_stream_callback(struct fw_iso_context *context, u32 cycle,
786				size_t header_length, void *header,
787				void *private_data)
788{
789	struct amdtp_stream *s = private_data;
790	unsigned int i, syt, packets = header_length / 4;
791
792	/*
793	 * Compute the cycle of the last queued packet.
794	 * (We need only the four lowest bits for the SYT, so we can ignore
795	 * that bits 0-11 must wrap around at 3072.)
796	 */
797	cycle += QUEUE_LENGTH - packets;
798
799	for (i = 0; i < packets; ++i) {
800		syt = calculate_syt(s, ++cycle);
801		handle_out_packet(s, syt);
802	}
803	fw_iso_context_queue_flush(s->context);
804}
805
806static void in_stream_callback(struct fw_iso_context *context, u32 cycle,
807			       size_t header_length, void *header,
808			       void *private_data)
809{
810	struct amdtp_stream *s = private_data;
811	unsigned int p, syt, packets, payload_quadlets;
812	__be32 *buffer, *headers = header;
813
814	/* The number of packets in buffer */
815	packets = header_length / IN_PACKET_HEADER_SIZE;
816
817	for (p = 0; p < packets; p++) {
818		if (s->packet_index < 0)
819			break;
820
821		buffer = s->buffer.packets[s->packet_index].buffer;
822
823		/* Process sync slave stream */
824		if (s->sync_slave && s->sync_slave->callbacked) {
825			syt = be32_to_cpu(buffer[1]) & CIP_SYT_MASK;
826			handle_out_packet(s->sync_slave, syt);
827		}
828
829		/* The number of quadlets in this packet */
830		payload_quadlets =
831			(be32_to_cpu(headers[p]) >> ISO_DATA_LENGTH_SHIFT) / 4;
832		handle_in_packet(s, payload_quadlets, buffer);
833	}
834
835	/* Queueing error or detecting discontinuity */
836	if (s->packet_index < 0) {
837		/* Abort sync slave. */
838		if (s->sync_slave) {
839			s->sync_slave->packet_index = -1;
840			amdtp_stream_pcm_abort(s->sync_slave);
841		}
842		return;
843	}
844
845	/* when sync to device, flush the packets for slave stream */
846	if (s->sync_slave && s->sync_slave->callbacked)
847		fw_iso_context_queue_flush(s->sync_slave->context);
848
849	fw_iso_context_queue_flush(s->context);
850}
851
852/* processing is done by master callback */
853static void slave_stream_callback(struct fw_iso_context *context, u32 cycle,
854				  size_t header_length, void *header,
855				  void *private_data)
856{
857	return;
858}
859
860/* this is executed one time */
861static void amdtp_stream_first_callback(struct fw_iso_context *context,
862					u32 cycle, size_t header_length,
863					void *header, void *private_data)
864{
865	struct amdtp_stream *s = private_data;
866
867	/*
868	 * For in-stream, first packet has come.
869	 * For out-stream, prepared to transmit first packet
870	 */
871	s->callbacked = true;
872	wake_up(&s->callback_wait);
873
874	if (s->direction == AMDTP_IN_STREAM)
875		context->callback.sc = in_stream_callback;
876	else if ((s->flags & CIP_BLOCKING) && (s->flags & CIP_SYNC_TO_DEVICE))
877		context->callback.sc = slave_stream_callback;
878	else
879		context->callback.sc = out_stream_callback;
880
881	context->callback.sc(context, cycle, header_length, header, s);
882}
883
884/**
885 * amdtp_stream_start - start transferring packets
886 * @s: the AMDTP stream to start
887 * @channel: the isochronous channel on the bus
888 * @speed: firewire speed code
889 *
890 * The stream cannot be started until it has been configured with
891 * amdtp_stream_set_parameters() and it must be started before any PCM or MIDI
892 * device can be started.
893 */
894int amdtp_stream_start(struct amdtp_stream *s, int channel, int speed)
895{
896	static const struct {
897		unsigned int data_block;
898		unsigned int syt_offset;
899	} initial_state[] = {
900		[CIP_SFC_32000]  = {  4, 3072 },
901		[CIP_SFC_48000]  = {  6, 1024 },
902		[CIP_SFC_96000]  = { 12, 1024 },
903		[CIP_SFC_192000] = { 24, 1024 },
904		[CIP_SFC_44100]  = {  0,   67 },
905		[CIP_SFC_88200]  = {  0,   67 },
906		[CIP_SFC_176400] = {  0,   67 },
907	};
908	unsigned int header_size;
909	enum dma_data_direction dir;
910	int type, tag, err;
911
912	mutex_lock(&s->mutex);
913
914	if (WARN_ON(amdtp_stream_running(s) ||
915		    (s->data_block_quadlets < 1))) {
916		err = -EBADFD;
917		goto err_unlock;
918	}
919
920	if (s->direction == AMDTP_IN_STREAM &&
921	    s->flags & CIP_SKIP_INIT_DBC_CHECK)
922		s->data_block_counter = UINT_MAX;
923	else
924		s->data_block_counter = 0;
925	s->data_block_state = initial_state[s->sfc].data_block;
926	s->syt_offset_state = initial_state[s->sfc].syt_offset;
927	s->last_syt_offset = TICKS_PER_CYCLE;
928
929	/* initialize packet buffer */
930	if (s->direction == AMDTP_IN_STREAM) {
931		dir = DMA_FROM_DEVICE;
932		type = FW_ISO_CONTEXT_RECEIVE;
933		header_size = IN_PACKET_HEADER_SIZE;
934	} else {
935		dir = DMA_TO_DEVICE;
936		type = FW_ISO_CONTEXT_TRANSMIT;
937		header_size = OUT_PACKET_HEADER_SIZE;
938	}
939	err = iso_packets_buffer_init(&s->buffer, s->unit, QUEUE_LENGTH,
940				      amdtp_stream_get_max_payload(s), dir);
941	if (err < 0)
942		goto err_unlock;
943
944	s->context = fw_iso_context_create(fw_parent_device(s->unit)->card,
945					   type, channel, speed, header_size,
946					   amdtp_stream_first_callback, s);
947	if (IS_ERR(s->context)) {
948		err = PTR_ERR(s->context);
949		if (err == -EBUSY)
950			dev_err(&s->unit->device,
951				"no free stream on this controller\n");
952		goto err_buffer;
953	}
954
955	amdtp_stream_update(s);
956
957	s->packet_index = 0;
958	do {
959		if (s->direction == AMDTP_IN_STREAM)
960			err = queue_in_packet(s);
961		else
962			err = queue_out_packet(s, 0, true);
963		if (err < 0)
964			goto err_context;
965	} while (s->packet_index > 0);
966
967	/* NOTE: TAG1 matches CIP. This just affects in stream. */
968	tag = FW_ISO_CONTEXT_MATCH_TAG1;
969	if (s->flags & CIP_EMPTY_WITH_TAG0)
970		tag |= FW_ISO_CONTEXT_MATCH_TAG0;
971
972	s->callbacked = false;
973	err = fw_iso_context_start(s->context, -1, 0, tag);
974	if (err < 0)
975		goto err_context;
976
977	mutex_unlock(&s->mutex);
978
979	return 0;
980
981err_context:
982	fw_iso_context_destroy(s->context);
983	s->context = ERR_PTR(-1);
984err_buffer:
985	iso_packets_buffer_destroy(&s->buffer, s->unit);
986err_unlock:
987	mutex_unlock(&s->mutex);
988
989	return err;
990}
991EXPORT_SYMBOL(amdtp_stream_start);
992
993/**
994 * amdtp_stream_pcm_pointer - get the PCM buffer position
995 * @s: the AMDTP stream that transports the PCM data
996 *
997 * Returns the current buffer position, in frames.
998 */
999unsigned long amdtp_stream_pcm_pointer(struct amdtp_stream *s)
1000{
1001	/* this optimization is allowed to be racy */
1002	if (s->pointer_flush && amdtp_stream_running(s))
1003		fw_iso_context_flush_completions(s->context);
1004	else
1005		s->pointer_flush = true;
1006
1007	return ACCESS_ONCE(s->pcm_buffer_pointer);
1008}
1009EXPORT_SYMBOL(amdtp_stream_pcm_pointer);
1010
1011/**
1012 * amdtp_stream_update - update the stream after a bus reset
1013 * @s: the AMDTP stream
1014 */
1015void amdtp_stream_update(struct amdtp_stream *s)
1016{
1017	ACCESS_ONCE(s->source_node_id_field) =
1018		(fw_parent_device(s->unit)->card->node_id & 0x3f) << 24;
1019}
1020EXPORT_SYMBOL(amdtp_stream_update);
1021
1022/**
1023 * amdtp_stream_stop - stop sending packets
1024 * @s: the AMDTP stream to stop
1025 *
1026 * All PCM and MIDI devices of the stream must be stopped before the stream
1027 * itself can be stopped.
1028 */
1029void amdtp_stream_stop(struct amdtp_stream *s)
1030{
1031	mutex_lock(&s->mutex);
1032
1033	if (!amdtp_stream_running(s)) {
1034		mutex_unlock(&s->mutex);
1035		return;
1036	}
1037
1038	tasklet_kill(&s->period_tasklet);
1039	fw_iso_context_stop(s->context);
1040	fw_iso_context_destroy(s->context);
1041	s->context = ERR_PTR(-1);
1042	iso_packets_buffer_destroy(&s->buffer, s->unit);
1043
1044	s->callbacked = false;
1045
1046	mutex_unlock(&s->mutex);
1047}
1048EXPORT_SYMBOL(amdtp_stream_stop);
1049
1050/**
1051 * amdtp_stream_pcm_abort - abort the running PCM device
1052 * @s: the AMDTP stream about to be stopped
1053 *
1054 * If the isochronous stream needs to be stopped asynchronously, call this
1055 * function first to stop the PCM device.
1056 */
1057void amdtp_stream_pcm_abort(struct amdtp_stream *s)
1058{
1059	struct snd_pcm_substream *pcm;
1060
1061	pcm = ACCESS_ONCE(s->pcm);
1062	if (pcm)
1063		snd_pcm_stop_xrun(pcm);
1064}
1065EXPORT_SYMBOL(amdtp_stream_pcm_abort);
1066