xref: /linux/sound/firewire/amdtp-stream.c (revision 52338415)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Audio and Music Data Transmission Protocol (IEC 61883-6) streams
4  * with Common Isochronous Packet (IEC 61883-1) headers
5  *
6  * Copyright (c) Clemens Ladisch <clemens@ladisch.de>
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 <sound/pcm.h>
15 #include <sound/pcm_params.h>
16 #include "amdtp-stream.h"
17 
18 #define TICKS_PER_CYCLE		3072
19 #define CYCLES_PER_SECOND	8000
20 #define TICKS_PER_SECOND	(TICKS_PER_CYCLE * CYCLES_PER_SECOND)
21 
22 /* Always support Linux tracing subsystem. */
23 #define CREATE_TRACE_POINTS
24 #include "amdtp-stream-trace.h"
25 
26 #define TRANSFER_DELAY_TICKS	0x2e00 /* 479.17 microseconds */
27 
28 /* isochronous header parameters */
29 #define ISO_DATA_LENGTH_SHIFT	16
30 #define TAG_NO_CIP_HEADER	0
31 #define TAG_CIP			1
32 
33 /* common isochronous packet header parameters */
34 #define CIP_EOH_SHIFT		31
35 #define CIP_EOH			(1u << CIP_EOH_SHIFT)
36 #define CIP_EOH_MASK		0x80000000
37 #define CIP_SID_SHIFT		24
38 #define CIP_SID_MASK		0x3f000000
39 #define CIP_DBS_MASK		0x00ff0000
40 #define CIP_DBS_SHIFT		16
41 #define CIP_SPH_MASK		0x00000400
42 #define CIP_SPH_SHIFT		10
43 #define CIP_DBC_MASK		0x000000ff
44 #define CIP_FMT_SHIFT		24
45 #define CIP_FMT_MASK		0x3f000000
46 #define CIP_FDF_MASK		0x00ff0000
47 #define CIP_FDF_SHIFT		16
48 #define CIP_SYT_MASK		0x0000ffff
49 #define CIP_SYT_NO_INFO		0xffff
50 
51 /* Audio and Music transfer protocol specific parameters */
52 #define CIP_FMT_AM		0x10
53 #define AMDTP_FDF_NO_DATA	0xff
54 
55 /* TODO: make these configurable */
56 #define INTERRUPT_INTERVAL	16
57 #define QUEUE_LENGTH		48
58 
59 // For iso header, tstamp and 2 CIP header.
60 #define IR_CTX_HEADER_SIZE_CIP		16
61 // For iso header and tstamp.
62 #define IR_CTX_HEADER_SIZE_NO_CIP	8
63 #define HEADER_TSTAMP_MASK	0x0000ffff
64 
65 #define IT_PKT_HEADER_SIZE_CIP		8 // For 2 CIP header.
66 #define IT_PKT_HEADER_SIZE_NO_CIP	0 // Nothing.
67 
68 static void pcm_period_tasklet(unsigned long data);
69 
70 /**
71  * amdtp_stream_init - initialize an AMDTP stream structure
72  * @s: the AMDTP stream to initialize
73  * @unit: the target of the stream
74  * @dir: the direction of stream
75  * @flags: the packet transmission method to use
76  * @fmt: the value of fmt field in CIP header
77  * @process_ctx_payloads: callback handler to process payloads of isoc context
78  * @protocol_size: the size to allocate newly for protocol
79  */
80 int amdtp_stream_init(struct amdtp_stream *s, struct fw_unit *unit,
81 		      enum amdtp_stream_direction dir, enum cip_flags flags,
82 		      unsigned int fmt,
83 		      amdtp_stream_process_ctx_payloads_t process_ctx_payloads,
84 		      unsigned int protocol_size)
85 {
86 	if (process_ctx_payloads == NULL)
87 		return -EINVAL;
88 
89 	s->protocol = kzalloc(protocol_size, GFP_KERNEL);
90 	if (!s->protocol)
91 		return -ENOMEM;
92 
93 	s->unit = unit;
94 	s->direction = dir;
95 	s->flags = flags;
96 	s->context = ERR_PTR(-1);
97 	mutex_init(&s->mutex);
98 	tasklet_init(&s->period_tasklet, pcm_period_tasklet, (unsigned long)s);
99 	s->packet_index = 0;
100 
101 	init_waitqueue_head(&s->callback_wait);
102 	s->callbacked = false;
103 
104 	s->fmt = fmt;
105 	s->process_ctx_payloads = process_ctx_payloads;
106 
107 	if (dir == AMDTP_OUT_STREAM)
108 		s->ctx_data.rx.syt_override = -1;
109 
110 	return 0;
111 }
112 EXPORT_SYMBOL(amdtp_stream_init);
113 
114 /**
115  * amdtp_stream_destroy - free stream resources
116  * @s: the AMDTP stream to destroy
117  */
118 void amdtp_stream_destroy(struct amdtp_stream *s)
119 {
120 	/* Not initialized. */
121 	if (s->protocol == NULL)
122 		return;
123 
124 	WARN_ON(amdtp_stream_running(s));
125 	kfree(s->protocol);
126 	mutex_destroy(&s->mutex);
127 }
128 EXPORT_SYMBOL(amdtp_stream_destroy);
129 
130 const unsigned int amdtp_syt_intervals[CIP_SFC_COUNT] = {
131 	[CIP_SFC_32000]  =  8,
132 	[CIP_SFC_44100]  =  8,
133 	[CIP_SFC_48000]  =  8,
134 	[CIP_SFC_88200]  = 16,
135 	[CIP_SFC_96000]  = 16,
136 	[CIP_SFC_176400] = 32,
137 	[CIP_SFC_192000] = 32,
138 };
139 EXPORT_SYMBOL(amdtp_syt_intervals);
140 
141 const unsigned int amdtp_rate_table[CIP_SFC_COUNT] = {
142 	[CIP_SFC_32000]  =  32000,
143 	[CIP_SFC_44100]  =  44100,
144 	[CIP_SFC_48000]  =  48000,
145 	[CIP_SFC_88200]  =  88200,
146 	[CIP_SFC_96000]  =  96000,
147 	[CIP_SFC_176400] = 176400,
148 	[CIP_SFC_192000] = 192000,
149 };
150 EXPORT_SYMBOL(amdtp_rate_table);
151 
152 static int apply_constraint_to_size(struct snd_pcm_hw_params *params,
153 				    struct snd_pcm_hw_rule *rule)
154 {
155 	struct snd_interval *s = hw_param_interval(params, rule->var);
156 	const struct snd_interval *r =
157 		hw_param_interval_c(params, SNDRV_PCM_HW_PARAM_RATE);
158 	struct snd_interval t = {0};
159 	unsigned int step = 0;
160 	int i;
161 
162 	for (i = 0; i < CIP_SFC_COUNT; ++i) {
163 		if (snd_interval_test(r, amdtp_rate_table[i]))
164 			step = max(step, amdtp_syt_intervals[i]);
165 	}
166 
167 	t.min = roundup(s->min, step);
168 	t.max = rounddown(s->max, step);
169 	t.integer = 1;
170 
171 	return snd_interval_refine(s, &t);
172 }
173 
174 /**
175  * amdtp_stream_add_pcm_hw_constraints - add hw constraints for PCM substream
176  * @s:		the AMDTP stream, which must be initialized.
177  * @runtime:	the PCM substream runtime
178  */
179 int amdtp_stream_add_pcm_hw_constraints(struct amdtp_stream *s,
180 					struct snd_pcm_runtime *runtime)
181 {
182 	struct snd_pcm_hardware *hw = &runtime->hw;
183 	int err;
184 
185 	hw->info = SNDRV_PCM_INFO_BATCH |
186 		   SNDRV_PCM_INFO_BLOCK_TRANSFER |
187 		   SNDRV_PCM_INFO_INTERLEAVED |
188 		   SNDRV_PCM_INFO_JOINT_DUPLEX |
189 		   SNDRV_PCM_INFO_MMAP |
190 		   SNDRV_PCM_INFO_MMAP_VALID;
191 
192 	/* SNDRV_PCM_INFO_BATCH */
193 	hw->periods_min = 2;
194 	hw->periods_max = UINT_MAX;
195 
196 	/* bytes for a frame */
197 	hw->period_bytes_min = 4 * hw->channels_max;
198 
199 	/* Just to prevent from allocating much pages. */
200 	hw->period_bytes_max = hw->period_bytes_min * 2048;
201 	hw->buffer_bytes_max = hw->period_bytes_max * hw->periods_min;
202 
203 	/*
204 	 * Currently firewire-lib processes 16 packets in one software
205 	 * interrupt callback. This equals to 2msec but actually the
206 	 * interval of the interrupts has a jitter.
207 	 * Additionally, even if adding a constraint to fit period size to
208 	 * 2msec, actual calculated frames per period doesn't equal to 2msec,
209 	 * depending on sampling rate.
210 	 * Anyway, the interval to call snd_pcm_period_elapsed() cannot 2msec.
211 	 * Here let us use 5msec for safe period interrupt.
212 	 */
213 	err = snd_pcm_hw_constraint_minmax(runtime,
214 					   SNDRV_PCM_HW_PARAM_PERIOD_TIME,
215 					   5000, UINT_MAX);
216 	if (err < 0)
217 		goto end;
218 
219 	/* Non-Blocking stream has no more constraints */
220 	if (!(s->flags & CIP_BLOCKING))
221 		goto end;
222 
223 	/*
224 	 * One AMDTP packet can include some frames. In blocking mode, the
225 	 * number equals to SYT_INTERVAL. So the number is 8, 16 or 32,
226 	 * depending on its sampling rate. For accurate period interrupt, it's
227 	 * preferrable to align period/buffer sizes to current SYT_INTERVAL.
228 	 */
229 	err = snd_pcm_hw_rule_add(runtime, 0, SNDRV_PCM_HW_PARAM_PERIOD_SIZE,
230 				  apply_constraint_to_size, NULL,
231 				  SNDRV_PCM_HW_PARAM_PERIOD_SIZE,
232 				  SNDRV_PCM_HW_PARAM_RATE, -1);
233 	if (err < 0)
234 		goto end;
235 	err = snd_pcm_hw_rule_add(runtime, 0, SNDRV_PCM_HW_PARAM_BUFFER_SIZE,
236 				  apply_constraint_to_size, NULL,
237 				  SNDRV_PCM_HW_PARAM_BUFFER_SIZE,
238 				  SNDRV_PCM_HW_PARAM_RATE, -1);
239 	if (err < 0)
240 		goto end;
241 end:
242 	return err;
243 }
244 EXPORT_SYMBOL(amdtp_stream_add_pcm_hw_constraints);
245 
246 /**
247  * amdtp_stream_set_parameters - set stream parameters
248  * @s: the AMDTP stream to configure
249  * @rate: the sample rate
250  * @data_block_quadlets: the size of a data block in quadlet unit
251  *
252  * The parameters must be set before the stream is started, and must not be
253  * changed while the stream is running.
254  */
255 int amdtp_stream_set_parameters(struct amdtp_stream *s, unsigned int rate,
256 				unsigned int data_block_quadlets)
257 {
258 	unsigned int sfc;
259 
260 	for (sfc = 0; sfc < ARRAY_SIZE(amdtp_rate_table); ++sfc) {
261 		if (amdtp_rate_table[sfc] == rate)
262 			break;
263 	}
264 	if (sfc == ARRAY_SIZE(amdtp_rate_table))
265 		return -EINVAL;
266 
267 	s->sfc = sfc;
268 	s->data_block_quadlets = data_block_quadlets;
269 	s->syt_interval = amdtp_syt_intervals[sfc];
270 
271 	// default buffering in the device.
272 	if (s->direction == AMDTP_OUT_STREAM) {
273 		s->ctx_data.rx.transfer_delay =
274 					TRANSFER_DELAY_TICKS - TICKS_PER_CYCLE;
275 
276 		if (s->flags & CIP_BLOCKING) {
277 			// additional buffering needed to adjust for no-data
278 			// packets.
279 			s->ctx_data.rx.transfer_delay +=
280 				TICKS_PER_SECOND * s->syt_interval / rate;
281 		}
282 	}
283 
284 	return 0;
285 }
286 EXPORT_SYMBOL(amdtp_stream_set_parameters);
287 
288 /**
289  * amdtp_stream_get_max_payload - get the stream's packet size
290  * @s: the AMDTP stream
291  *
292  * This function must not be called before the stream has been configured
293  * with amdtp_stream_set_parameters().
294  */
295 unsigned int amdtp_stream_get_max_payload(struct amdtp_stream *s)
296 {
297 	unsigned int multiplier = 1;
298 	unsigned int cip_header_size = 0;
299 
300 	if (s->flags & CIP_JUMBO_PAYLOAD)
301 		multiplier = 5;
302 	if (!(s->flags & CIP_NO_HEADER))
303 		cip_header_size = sizeof(__be32) * 2;
304 
305 	return cip_header_size +
306 		s->syt_interval * s->data_block_quadlets * sizeof(__be32) * multiplier;
307 }
308 EXPORT_SYMBOL(amdtp_stream_get_max_payload);
309 
310 /**
311  * amdtp_stream_pcm_prepare - prepare PCM device for running
312  * @s: the AMDTP stream
313  *
314  * This function should be called from the PCM device's .prepare callback.
315  */
316 void amdtp_stream_pcm_prepare(struct amdtp_stream *s)
317 {
318 	tasklet_kill(&s->period_tasklet);
319 	s->pcm_buffer_pointer = 0;
320 	s->pcm_period_pointer = 0;
321 }
322 EXPORT_SYMBOL(amdtp_stream_pcm_prepare);
323 
324 static unsigned int calculate_data_blocks(struct amdtp_stream *s,
325 					  unsigned int syt)
326 {
327 	unsigned int phase, data_blocks;
328 
329 	/* Blocking mode. */
330 	if (s->flags & CIP_BLOCKING) {
331 		/* This module generate empty packet for 'no data'. */
332 		if (syt == CIP_SYT_NO_INFO)
333 			data_blocks = 0;
334 		else
335 			data_blocks = s->syt_interval;
336 	/* Non-blocking mode. */
337 	} else {
338 		if (!cip_sfc_is_base_44100(s->sfc)) {
339 			// Sample_rate / 8000 is an integer, and precomputed.
340 			data_blocks = s->ctx_data.rx.data_block_state;
341 		} else {
342 			phase = s->ctx_data.rx.data_block_state;
343 
344 		/*
345 		 * This calculates the number of data blocks per packet so that
346 		 * 1) the overall rate is correct and exactly synchronized to
347 		 *    the bus clock, and
348 		 * 2) packets with a rounded-up number of blocks occur as early
349 		 *    as possible in the sequence (to prevent underruns of the
350 		 *    device's buffer).
351 		 */
352 			if (s->sfc == CIP_SFC_44100)
353 				/* 6 6 5 6 5 6 5 ... */
354 				data_blocks = 5 + ((phase & 1) ^
355 						   (phase == 0 || phase >= 40));
356 			else
357 				/* 12 11 11 11 11 ... or 23 22 22 22 22 ... */
358 				data_blocks = 11 * (s->sfc >> 1) + (phase == 0);
359 			if (++phase >= (80 >> (s->sfc >> 1)))
360 				phase = 0;
361 			s->ctx_data.rx.data_block_state = phase;
362 		}
363 	}
364 
365 	return data_blocks;
366 }
367 
368 static unsigned int calculate_syt(struct amdtp_stream *s,
369 				  unsigned int cycle)
370 {
371 	unsigned int syt_offset, phase, index, syt;
372 
373 	if (s->ctx_data.rx.last_syt_offset < TICKS_PER_CYCLE) {
374 		if (!cip_sfc_is_base_44100(s->sfc))
375 			syt_offset = s->ctx_data.rx.last_syt_offset +
376 				     s->ctx_data.rx.syt_offset_state;
377 		else {
378 		/*
379 		 * The time, in ticks, of the n'th SYT_INTERVAL sample is:
380 		 *   n * SYT_INTERVAL * 24576000 / sample_rate
381 		 * Modulo TICKS_PER_CYCLE, the difference between successive
382 		 * elements is about 1386.23.  Rounding the results of this
383 		 * formula to the SYT precision results in a sequence of
384 		 * differences that begins with:
385 		 *   1386 1386 1387 1386 1386 1386 1387 1386 1386 1386 1387 ...
386 		 * This code generates _exactly_ the same sequence.
387 		 */
388 			phase = s->ctx_data.rx.syt_offset_state;
389 			index = phase % 13;
390 			syt_offset = s->ctx_data.rx.last_syt_offset;
391 			syt_offset += 1386 + ((index && !(index & 3)) ||
392 					      phase == 146);
393 			if (++phase >= 147)
394 				phase = 0;
395 			s->ctx_data.rx.syt_offset_state = phase;
396 		}
397 	} else
398 		syt_offset = s->ctx_data.rx.last_syt_offset - TICKS_PER_CYCLE;
399 	s->ctx_data.rx.last_syt_offset = syt_offset;
400 
401 	if (syt_offset < TICKS_PER_CYCLE) {
402 		syt_offset += s->ctx_data.rx.transfer_delay;
403 		syt = (cycle + syt_offset / TICKS_PER_CYCLE) << 12;
404 		syt += syt_offset % TICKS_PER_CYCLE;
405 
406 		return syt & CIP_SYT_MASK;
407 	} else {
408 		return CIP_SYT_NO_INFO;
409 	}
410 }
411 
412 static void update_pcm_pointers(struct amdtp_stream *s,
413 				struct snd_pcm_substream *pcm,
414 				unsigned int frames)
415 {
416 	unsigned int ptr;
417 
418 	ptr = s->pcm_buffer_pointer + frames;
419 	if (ptr >= pcm->runtime->buffer_size)
420 		ptr -= pcm->runtime->buffer_size;
421 	WRITE_ONCE(s->pcm_buffer_pointer, ptr);
422 
423 	s->pcm_period_pointer += frames;
424 	if (s->pcm_period_pointer >= pcm->runtime->period_size) {
425 		s->pcm_period_pointer -= pcm->runtime->period_size;
426 		tasklet_hi_schedule(&s->period_tasklet);
427 	}
428 }
429 
430 static void pcm_period_tasklet(unsigned long data)
431 {
432 	struct amdtp_stream *s = (void *)data;
433 	struct snd_pcm_substream *pcm = READ_ONCE(s->pcm);
434 
435 	if (pcm)
436 		snd_pcm_period_elapsed(pcm);
437 }
438 
439 static int queue_packet(struct amdtp_stream *s, struct fw_iso_packet *params)
440 {
441 	int err;
442 
443 	params->interrupt = IS_ALIGNED(s->packet_index + 1, INTERRUPT_INTERVAL);
444 	params->tag = s->tag;
445 	params->sy = 0;
446 
447 	err = fw_iso_context_queue(s->context, params, &s->buffer.iso_buffer,
448 				   s->buffer.packets[s->packet_index].offset);
449 	if (err < 0) {
450 		dev_err(&s->unit->device, "queueing error: %d\n", err);
451 		goto end;
452 	}
453 
454 	if (++s->packet_index >= QUEUE_LENGTH)
455 		s->packet_index = 0;
456 end:
457 	return err;
458 }
459 
460 static inline int queue_out_packet(struct amdtp_stream *s,
461 				   struct fw_iso_packet *params)
462 {
463 	params->skip =
464 		!!(params->header_length == 0 && params->payload_length == 0);
465 	return queue_packet(s, params);
466 }
467 
468 static inline int queue_in_packet(struct amdtp_stream *s,
469 				  struct fw_iso_packet *params)
470 {
471 	// Queue one packet for IR context.
472 	params->header_length = s->ctx_data.tx.ctx_header_size;
473 	params->payload_length = s->ctx_data.tx.max_ctx_payload_length;
474 	params->skip = false;
475 	return queue_packet(s, params);
476 }
477 
478 static void generate_cip_header(struct amdtp_stream *s, __be32 cip_header[2],
479 			unsigned int data_block_counter, unsigned int syt)
480 {
481 	cip_header[0] = cpu_to_be32(READ_ONCE(s->source_node_id_field) |
482 				(s->data_block_quadlets << CIP_DBS_SHIFT) |
483 				((s->sph << CIP_SPH_SHIFT) & CIP_SPH_MASK) |
484 				data_block_counter);
485 	cip_header[1] = cpu_to_be32(CIP_EOH |
486 			((s->fmt << CIP_FMT_SHIFT) & CIP_FMT_MASK) |
487 			((s->ctx_data.rx.fdf << CIP_FDF_SHIFT) & CIP_FDF_MASK) |
488 			(syt & CIP_SYT_MASK));
489 }
490 
491 static void build_it_pkt_header(struct amdtp_stream *s, unsigned int cycle,
492 				struct fw_iso_packet *params,
493 				unsigned int data_blocks,
494 				unsigned int data_block_counter,
495 				unsigned int syt, unsigned int index)
496 {
497 	unsigned int payload_length;
498 	__be32 *cip_header;
499 
500 	payload_length = data_blocks * sizeof(__be32) * s->data_block_quadlets;
501 	params->payload_length = payload_length;
502 
503 	if (!(s->flags & CIP_NO_HEADER)) {
504 		cip_header = (__be32 *)params->header;
505 		generate_cip_header(s, cip_header, data_block_counter, syt);
506 		params->header_length = 2 * sizeof(__be32);
507 		payload_length += params->header_length;
508 	} else {
509 		cip_header = NULL;
510 	}
511 
512 	trace_amdtp_packet(s, cycle, cip_header, payload_length, data_blocks,
513 			   data_block_counter, index);
514 }
515 
516 static int check_cip_header(struct amdtp_stream *s, const __be32 *buf,
517 			    unsigned int payload_length,
518 			    unsigned int *data_blocks,
519 			    unsigned int *data_block_counter, unsigned int *syt)
520 {
521 	u32 cip_header[2];
522 	unsigned int sph;
523 	unsigned int fmt;
524 	unsigned int fdf;
525 	unsigned int dbc;
526 	bool lost;
527 
528 	cip_header[0] = be32_to_cpu(buf[0]);
529 	cip_header[1] = be32_to_cpu(buf[1]);
530 
531 	/*
532 	 * This module supports 'Two-quadlet CIP header with SYT field'.
533 	 * For convenience, also check FMT field is AM824 or not.
534 	 */
535 	if ((((cip_header[0] & CIP_EOH_MASK) == CIP_EOH) ||
536 	     ((cip_header[1] & CIP_EOH_MASK) != CIP_EOH)) &&
537 	    (!(s->flags & CIP_HEADER_WITHOUT_EOH))) {
538 		dev_info_ratelimited(&s->unit->device,
539 				"Invalid CIP header for AMDTP: %08X:%08X\n",
540 				cip_header[0], cip_header[1]);
541 		return -EAGAIN;
542 	}
543 
544 	/* Check valid protocol or not. */
545 	sph = (cip_header[0] & CIP_SPH_MASK) >> CIP_SPH_SHIFT;
546 	fmt = (cip_header[1] & CIP_FMT_MASK) >> CIP_FMT_SHIFT;
547 	if (sph != s->sph || fmt != s->fmt) {
548 		dev_info_ratelimited(&s->unit->device,
549 				     "Detect unexpected protocol: %08x %08x\n",
550 				     cip_header[0], cip_header[1]);
551 		return -EAGAIN;
552 	}
553 
554 	/* Calculate data blocks */
555 	fdf = (cip_header[1] & CIP_FDF_MASK) >> CIP_FDF_SHIFT;
556 	if (payload_length < sizeof(__be32) * 2 ||
557 	    (fmt == CIP_FMT_AM && fdf == AMDTP_FDF_NO_DATA)) {
558 		*data_blocks = 0;
559 	} else {
560 		unsigned int data_block_quadlets =
561 				(cip_header[0] & CIP_DBS_MASK) >> CIP_DBS_SHIFT;
562 		/* avoid division by zero */
563 		if (data_block_quadlets == 0) {
564 			dev_err(&s->unit->device,
565 				"Detect invalid value in dbs field: %08X\n",
566 				cip_header[0]);
567 			return -EPROTO;
568 		}
569 		if (s->flags & CIP_WRONG_DBS)
570 			data_block_quadlets = s->data_block_quadlets;
571 
572 		*data_blocks = (payload_length / sizeof(__be32) - 2) /
573 							data_block_quadlets;
574 	}
575 
576 	/* Check data block counter continuity */
577 	dbc = cip_header[0] & CIP_DBC_MASK;
578 	if (*data_blocks == 0 && (s->flags & CIP_EMPTY_HAS_WRONG_DBC) &&
579 	    *data_block_counter != UINT_MAX)
580 		dbc = *data_block_counter;
581 
582 	if ((dbc == 0x00 && (s->flags & CIP_SKIP_DBC_ZERO_CHECK)) ||
583 	    *data_block_counter == UINT_MAX) {
584 		lost = false;
585 	} else if (!(s->flags & CIP_DBC_IS_END_EVENT)) {
586 		lost = dbc != *data_block_counter;
587 	} else {
588 		unsigned int dbc_interval;
589 
590 		if (*data_blocks > 0 && s->ctx_data.tx.dbc_interval > 0)
591 			dbc_interval = s->ctx_data.tx.dbc_interval;
592 		else
593 			dbc_interval = *data_blocks;
594 
595 		lost = dbc != ((*data_block_counter + dbc_interval) & 0xff);
596 	}
597 
598 	if (lost) {
599 		dev_err(&s->unit->device,
600 			"Detect discontinuity of CIP: %02X %02X\n",
601 			*data_block_counter, dbc);
602 		return -EIO;
603 	}
604 
605 	*data_block_counter = dbc;
606 
607 	*syt = cip_header[1] & CIP_SYT_MASK;
608 
609 	return 0;
610 }
611 
612 static int parse_ir_ctx_header(struct amdtp_stream *s, unsigned int cycle,
613 			       const __be32 *ctx_header,
614 			       unsigned int *payload_length,
615 			       unsigned int *data_blocks,
616 			       unsigned int *data_block_counter,
617 			       unsigned int *syt, unsigned int index)
618 {
619 	const __be32 *cip_header;
620 	int err;
621 
622 	*payload_length = be32_to_cpu(ctx_header[0]) >> ISO_DATA_LENGTH_SHIFT;
623 	if (*payload_length > s->ctx_data.tx.ctx_header_size +
624 					s->ctx_data.tx.max_ctx_payload_length) {
625 		dev_err(&s->unit->device,
626 			"Detect jumbo payload: %04x %04x\n",
627 			*payload_length, s->ctx_data.tx.max_ctx_payload_length);
628 		return -EIO;
629 	}
630 
631 	if (!(s->flags & CIP_NO_HEADER)) {
632 		cip_header = ctx_header + 2;
633 		err = check_cip_header(s, cip_header, *payload_length,
634 				       data_blocks, data_block_counter, syt);
635 		if (err < 0)
636 			return err;
637 	} else {
638 		cip_header = NULL;
639 		err = 0;
640 		*data_blocks = *payload_length / sizeof(__be32) /
641 			       s->data_block_quadlets;
642 		*syt = 0;
643 
644 		if (*data_block_counter == UINT_MAX)
645 			*data_block_counter = 0;
646 	}
647 
648 	trace_amdtp_packet(s, cycle, cip_header, *payload_length, *data_blocks,
649 			   *data_block_counter, index);
650 
651 	return err;
652 }
653 
654 // In CYCLE_TIMER register of IEEE 1394, 7 bits are used to represent second. On
655 // the other hand, in DMA descriptors of 1394 OHCI, 3 bits are used to represent
656 // it. Thus, via Linux firewire subsystem, we can get the 3 bits for second.
657 static inline u32 compute_cycle_count(__be32 ctx_header_tstamp)
658 {
659 	u32 tstamp = be32_to_cpu(ctx_header_tstamp) & HEADER_TSTAMP_MASK;
660 	return (((tstamp >> 13) & 0x07) * 8000) + (tstamp & 0x1fff);
661 }
662 
663 static inline u32 increment_cycle_count(u32 cycle, unsigned int addend)
664 {
665 	cycle += addend;
666 	if (cycle >= 8 * CYCLES_PER_SECOND)
667 		cycle -= 8 * CYCLES_PER_SECOND;
668 	return cycle;
669 }
670 
671 // Align to actual cycle count for the packet which is going to be scheduled.
672 // This module queued the same number of isochronous cycle as QUEUE_LENGTH to
673 // skip isochronous cycle, therefore it's OK to just increment the cycle by
674 // QUEUE_LENGTH for scheduled cycle.
675 static inline u32 compute_it_cycle(const __be32 ctx_header_tstamp)
676 {
677 	u32 cycle = compute_cycle_count(ctx_header_tstamp);
678 	return increment_cycle_count(cycle, QUEUE_LENGTH);
679 }
680 
681 static int generate_device_pkt_descs(struct amdtp_stream *s,
682 				     struct pkt_desc *descs,
683 				     const __be32 *ctx_header,
684 				     unsigned int packets)
685 {
686 	unsigned int dbc = s->data_block_counter;
687 	int i;
688 	int err;
689 
690 	for (i = 0; i < packets; ++i) {
691 		struct pkt_desc *desc = descs + i;
692 		unsigned int index = (s->packet_index + i) % QUEUE_LENGTH;
693 		unsigned int cycle;
694 		unsigned int payload_length;
695 		unsigned int data_blocks;
696 		unsigned int syt;
697 
698 		cycle = compute_cycle_count(ctx_header[1]);
699 
700 		err = parse_ir_ctx_header(s, cycle, ctx_header, &payload_length,
701 					  &data_blocks, &dbc, &syt, i);
702 		if (err < 0)
703 			return err;
704 
705 		desc->cycle = cycle;
706 		desc->syt = syt;
707 		desc->data_blocks = data_blocks;
708 		desc->data_block_counter = dbc;
709 		desc->ctx_payload = s->buffer.packets[index].buffer;
710 
711 		if (!(s->flags & CIP_DBC_IS_END_EVENT))
712 			dbc = (dbc + desc->data_blocks) & 0xff;
713 
714 		ctx_header +=
715 			s->ctx_data.tx.ctx_header_size / sizeof(*ctx_header);
716 	}
717 
718 	s->data_block_counter = dbc;
719 
720 	return 0;
721 }
722 
723 static void generate_ideal_pkt_descs(struct amdtp_stream *s,
724 				     struct pkt_desc *descs,
725 				     const __be32 *ctx_header,
726 				     unsigned int packets)
727 {
728 	unsigned int dbc = s->data_block_counter;
729 	int i;
730 
731 	for (i = 0; i < packets; ++i) {
732 		struct pkt_desc *desc = descs + i;
733 		unsigned int index = (s->packet_index + i) % QUEUE_LENGTH;
734 
735 		desc->cycle = compute_it_cycle(*ctx_header);
736 		desc->syt = calculate_syt(s, desc->cycle);
737 		desc->data_blocks = calculate_data_blocks(s, desc->syt);
738 
739 		if (s->flags & CIP_DBC_IS_END_EVENT)
740 			dbc = (dbc + desc->data_blocks) & 0xff;
741 
742 		desc->data_block_counter = dbc;
743 
744 		if (!(s->flags & CIP_DBC_IS_END_EVENT))
745 			dbc = (dbc + desc->data_blocks) & 0xff;
746 
747 		desc->ctx_payload = s->buffer.packets[index].buffer;
748 
749 		++ctx_header;
750 	}
751 
752 	s->data_block_counter = dbc;
753 }
754 
755 static inline void cancel_stream(struct amdtp_stream *s)
756 {
757 	s->packet_index = -1;
758 	if (in_interrupt())
759 		amdtp_stream_pcm_abort(s);
760 	WRITE_ONCE(s->pcm_buffer_pointer, SNDRV_PCM_POS_XRUN);
761 }
762 
763 static void process_ctx_payloads(struct amdtp_stream *s,
764 				 const struct pkt_desc *descs,
765 				 unsigned int packets)
766 {
767 	struct snd_pcm_substream *pcm;
768 	unsigned int pcm_frames;
769 
770 	pcm = READ_ONCE(s->pcm);
771 	pcm_frames = s->process_ctx_payloads(s, descs, packets, pcm);
772 	if (pcm)
773 		update_pcm_pointers(s, pcm, pcm_frames);
774 }
775 
776 static void out_stream_callback(struct fw_iso_context *context, u32 tstamp,
777 				size_t header_length, void *header,
778 				void *private_data)
779 {
780 	struct amdtp_stream *s = private_data;
781 	const __be32 *ctx_header = header;
782 	unsigned int packets = header_length / sizeof(*ctx_header);
783 	int i;
784 
785 	if (s->packet_index < 0)
786 		return;
787 
788 	generate_ideal_pkt_descs(s, s->pkt_descs, ctx_header, packets);
789 
790 	process_ctx_payloads(s, s->pkt_descs, packets);
791 
792 	for (i = 0; i < packets; ++i) {
793 		const struct pkt_desc *desc = s->pkt_descs + i;
794 		unsigned int syt;
795 		struct {
796 			struct fw_iso_packet params;
797 			__be32 header[IT_PKT_HEADER_SIZE_CIP / sizeof(__be32)];
798 		} template = { {0}, {0} };
799 
800 		if (s->ctx_data.rx.syt_override < 0)
801 			syt = desc->syt;
802 		else
803 			syt = s->ctx_data.rx.syt_override;
804 
805 		build_it_pkt_header(s, desc->cycle, &template.params,
806 				    desc->data_blocks, desc->data_block_counter,
807 				    syt, i);
808 
809 		if (queue_out_packet(s, &template.params) < 0) {
810 			cancel_stream(s);
811 			return;
812 		}
813 	}
814 
815 	fw_iso_context_queue_flush(s->context);
816 }
817 
818 static void in_stream_callback(struct fw_iso_context *context, u32 tstamp,
819 			       size_t header_length, void *header,
820 			       void *private_data)
821 {
822 	struct amdtp_stream *s = private_data;
823 	unsigned int packets;
824 	__be32 *ctx_header = header;
825 	int i;
826 	int err;
827 
828 	if (s->packet_index < 0)
829 		return;
830 
831 	// The number of packets in buffer.
832 	packets = header_length / s->ctx_data.tx.ctx_header_size;
833 
834 	err = generate_device_pkt_descs(s, s->pkt_descs, ctx_header, packets);
835 	if (err < 0) {
836 		if (err != -EAGAIN) {
837 			cancel_stream(s);
838 			return;
839 		}
840 	} else {
841 		process_ctx_payloads(s, s->pkt_descs, packets);
842 	}
843 
844 	for (i = 0; i < packets; ++i) {
845 		struct fw_iso_packet params = {0};
846 
847 		if (queue_in_packet(s, &params) < 0) {
848 			cancel_stream(s);
849 			return;
850 		}
851 	}
852 
853 	fw_iso_context_queue_flush(s->context);
854 }
855 
856 /* this is executed one time */
857 static void amdtp_stream_first_callback(struct fw_iso_context *context,
858 					u32 tstamp, size_t header_length,
859 					void *header, void *private_data)
860 {
861 	struct amdtp_stream *s = private_data;
862 	const __be32 *ctx_header = header;
863 	u32 cycle;
864 
865 	/*
866 	 * For in-stream, first packet has come.
867 	 * For out-stream, prepared to transmit first packet
868 	 */
869 	s->callbacked = true;
870 	wake_up(&s->callback_wait);
871 
872 	if (s->direction == AMDTP_IN_STREAM) {
873 		cycle = compute_cycle_count(ctx_header[1]);
874 
875 		context->callback.sc = in_stream_callback;
876 	} else {
877 		cycle = compute_it_cycle(*ctx_header);
878 
879 		context->callback.sc = out_stream_callback;
880 	}
881 
882 	s->start_cycle = cycle;
883 
884 	context->callback.sc(context, tstamp, header_length, header, s);
885 }
886 
887 /**
888  * amdtp_stream_start - start transferring packets
889  * @s: the AMDTP stream to start
890  * @channel: the isochronous channel on the bus
891  * @speed: firewire speed code
892  *
893  * The stream cannot be started until it has been configured with
894  * amdtp_stream_set_parameters() and it must be started before any PCM or MIDI
895  * device can be started.
896  */
897 static int amdtp_stream_start(struct amdtp_stream *s, int channel, int speed)
898 {
899 	static const struct {
900 		unsigned int data_block;
901 		unsigned int syt_offset;
902 	} *entry, initial_state[] = {
903 		[CIP_SFC_32000]  = {  4, 3072 },
904 		[CIP_SFC_48000]  = {  6, 1024 },
905 		[CIP_SFC_96000]  = { 12, 1024 },
906 		[CIP_SFC_192000] = { 24, 1024 },
907 		[CIP_SFC_44100]  = {  0,   67 },
908 		[CIP_SFC_88200]  = {  0,   67 },
909 		[CIP_SFC_176400] = {  0,   67 },
910 	};
911 	unsigned int ctx_header_size;
912 	unsigned int max_ctx_payload_size;
913 	enum dma_data_direction dir;
914 	int type, tag, err;
915 
916 	mutex_lock(&s->mutex);
917 
918 	if (WARN_ON(amdtp_stream_running(s) ||
919 		    (s->data_block_quadlets < 1))) {
920 		err = -EBADFD;
921 		goto err_unlock;
922 	}
923 
924 	if (s->direction == AMDTP_IN_STREAM) {
925 		s->data_block_counter = UINT_MAX;
926 	} else {
927 		entry = &initial_state[s->sfc];
928 
929 		s->data_block_counter = 0;
930 		s->ctx_data.rx.data_block_state = entry->data_block;
931 		s->ctx_data.rx.syt_offset_state = entry->syt_offset;
932 		s->ctx_data.rx.last_syt_offset = TICKS_PER_CYCLE;
933 	}
934 
935 	/* initialize packet buffer */
936 	if (s->direction == AMDTP_IN_STREAM) {
937 		dir = DMA_FROM_DEVICE;
938 		type = FW_ISO_CONTEXT_RECEIVE;
939 		if (!(s->flags & CIP_NO_HEADER))
940 			ctx_header_size = IR_CTX_HEADER_SIZE_CIP;
941 		else
942 			ctx_header_size = IR_CTX_HEADER_SIZE_NO_CIP;
943 
944 		max_ctx_payload_size = amdtp_stream_get_max_payload(s) -
945 				       ctx_header_size;
946 	} else {
947 		dir = DMA_TO_DEVICE;
948 		type = FW_ISO_CONTEXT_TRANSMIT;
949 		ctx_header_size = 0;	// No effect for IT context.
950 
951 		max_ctx_payload_size = amdtp_stream_get_max_payload(s);
952 		if (!(s->flags & CIP_NO_HEADER))
953 			max_ctx_payload_size -= IT_PKT_HEADER_SIZE_CIP;
954 	}
955 
956 	err = iso_packets_buffer_init(&s->buffer, s->unit, QUEUE_LENGTH,
957 				      max_ctx_payload_size, dir);
958 	if (err < 0)
959 		goto err_unlock;
960 
961 	s->context = fw_iso_context_create(fw_parent_device(s->unit)->card,
962 					  type, channel, speed, ctx_header_size,
963 					  amdtp_stream_first_callback, s);
964 	if (IS_ERR(s->context)) {
965 		err = PTR_ERR(s->context);
966 		if (err == -EBUSY)
967 			dev_err(&s->unit->device,
968 				"no free stream on this controller\n");
969 		goto err_buffer;
970 	}
971 
972 	amdtp_stream_update(s);
973 
974 	if (s->direction == AMDTP_IN_STREAM) {
975 		s->ctx_data.tx.max_ctx_payload_length = max_ctx_payload_size;
976 		s->ctx_data.tx.ctx_header_size = ctx_header_size;
977 	}
978 
979 	if (s->flags & CIP_NO_HEADER)
980 		s->tag = TAG_NO_CIP_HEADER;
981 	else
982 		s->tag = TAG_CIP;
983 
984 	s->pkt_descs = kcalloc(INTERRUPT_INTERVAL, sizeof(*s->pkt_descs),
985 			       GFP_KERNEL);
986 	if (!s->pkt_descs) {
987 		err = -ENOMEM;
988 		goto err_context;
989 	}
990 
991 	s->packet_index = 0;
992 	do {
993 		struct fw_iso_packet params;
994 		if (s->direction == AMDTP_IN_STREAM) {
995 			err = queue_in_packet(s, &params);
996 		} else {
997 			params.header_length = 0;
998 			params.payload_length = 0;
999 			err = queue_out_packet(s, &params);
1000 		}
1001 		if (err < 0)
1002 			goto err_pkt_descs;
1003 	} while (s->packet_index > 0);
1004 
1005 	/* NOTE: TAG1 matches CIP. This just affects in stream. */
1006 	tag = FW_ISO_CONTEXT_MATCH_TAG1;
1007 	if ((s->flags & CIP_EMPTY_WITH_TAG0) || (s->flags & CIP_NO_HEADER))
1008 		tag |= FW_ISO_CONTEXT_MATCH_TAG0;
1009 
1010 	s->callbacked = false;
1011 	err = fw_iso_context_start(s->context, -1, 0, tag);
1012 	if (err < 0)
1013 		goto err_pkt_descs;
1014 
1015 	mutex_unlock(&s->mutex);
1016 
1017 	return 0;
1018 err_pkt_descs:
1019 	kfree(s->pkt_descs);
1020 err_context:
1021 	fw_iso_context_destroy(s->context);
1022 	s->context = ERR_PTR(-1);
1023 err_buffer:
1024 	iso_packets_buffer_destroy(&s->buffer, s->unit);
1025 err_unlock:
1026 	mutex_unlock(&s->mutex);
1027 
1028 	return err;
1029 }
1030 
1031 /**
1032  * amdtp_stream_pcm_pointer - get the PCM buffer position
1033  * @s: the AMDTP stream that transports the PCM data
1034  *
1035  * Returns the current buffer position, in frames.
1036  */
1037 unsigned long amdtp_stream_pcm_pointer(struct amdtp_stream *s)
1038 {
1039 	/*
1040 	 * This function is called in software IRQ context of period_tasklet or
1041 	 * process context.
1042 	 *
1043 	 * When the software IRQ context was scheduled by software IRQ context
1044 	 * of IR/IT contexts, queued packets were already handled. Therefore,
1045 	 * no need to flush the queue in buffer anymore.
1046 	 *
1047 	 * When the process context reach here, some packets will be already
1048 	 * queued in the buffer. These packets should be handled immediately
1049 	 * to keep better granularity of PCM pointer.
1050 	 *
1051 	 * Later, the process context will sometimes schedules software IRQ
1052 	 * context of the period_tasklet. Then, no need to flush the queue by
1053 	 * the same reason as described for IR/IT contexts.
1054 	 */
1055 	if (!in_interrupt() && amdtp_stream_running(s))
1056 		fw_iso_context_flush_completions(s->context);
1057 
1058 	return READ_ONCE(s->pcm_buffer_pointer);
1059 }
1060 EXPORT_SYMBOL(amdtp_stream_pcm_pointer);
1061 
1062 /**
1063  * amdtp_stream_pcm_ack - acknowledge queued PCM frames
1064  * @s: the AMDTP stream that transfers the PCM frames
1065  *
1066  * Returns zero always.
1067  */
1068 int amdtp_stream_pcm_ack(struct amdtp_stream *s)
1069 {
1070 	/*
1071 	 * Process isochronous packets for recent isochronous cycle to handle
1072 	 * queued PCM frames.
1073 	 */
1074 	if (amdtp_stream_running(s))
1075 		fw_iso_context_flush_completions(s->context);
1076 
1077 	return 0;
1078 }
1079 EXPORT_SYMBOL(amdtp_stream_pcm_ack);
1080 
1081 /**
1082  * amdtp_stream_update - update the stream after a bus reset
1083  * @s: the AMDTP stream
1084  */
1085 void amdtp_stream_update(struct amdtp_stream *s)
1086 {
1087 	/* Precomputing. */
1088 	WRITE_ONCE(s->source_node_id_field,
1089                    (fw_parent_device(s->unit)->card->node_id << CIP_SID_SHIFT) & CIP_SID_MASK);
1090 }
1091 EXPORT_SYMBOL(amdtp_stream_update);
1092 
1093 /**
1094  * amdtp_stream_stop - stop sending packets
1095  * @s: the AMDTP stream to stop
1096  *
1097  * All PCM and MIDI devices of the stream must be stopped before the stream
1098  * itself can be stopped.
1099  */
1100 static void amdtp_stream_stop(struct amdtp_stream *s)
1101 {
1102 	mutex_lock(&s->mutex);
1103 
1104 	if (!amdtp_stream_running(s)) {
1105 		mutex_unlock(&s->mutex);
1106 		return;
1107 	}
1108 
1109 	tasklet_kill(&s->period_tasklet);
1110 	fw_iso_context_stop(s->context);
1111 	fw_iso_context_destroy(s->context);
1112 	s->context = ERR_PTR(-1);
1113 	iso_packets_buffer_destroy(&s->buffer, s->unit);
1114 	kfree(s->pkt_descs);
1115 
1116 	s->callbacked = false;
1117 
1118 	mutex_unlock(&s->mutex);
1119 }
1120 
1121 /**
1122  * amdtp_stream_pcm_abort - abort the running PCM device
1123  * @s: the AMDTP stream about to be stopped
1124  *
1125  * If the isochronous stream needs to be stopped asynchronously, call this
1126  * function first to stop the PCM device.
1127  */
1128 void amdtp_stream_pcm_abort(struct amdtp_stream *s)
1129 {
1130 	struct snd_pcm_substream *pcm;
1131 
1132 	pcm = READ_ONCE(s->pcm);
1133 	if (pcm)
1134 		snd_pcm_stop_xrun(pcm);
1135 }
1136 EXPORT_SYMBOL(amdtp_stream_pcm_abort);
1137 
1138 /**
1139  * amdtp_domain_init - initialize an AMDTP domain structure
1140  * @d: the AMDTP domain to initialize.
1141  */
1142 int amdtp_domain_init(struct amdtp_domain *d)
1143 {
1144 	INIT_LIST_HEAD(&d->streams);
1145 
1146 	return 0;
1147 }
1148 EXPORT_SYMBOL_GPL(amdtp_domain_init);
1149 
1150 /**
1151  * amdtp_domain_destroy - destroy an AMDTP domain structure
1152  * @d: the AMDTP domain to destroy.
1153  */
1154 void amdtp_domain_destroy(struct amdtp_domain *d)
1155 {
1156 	// At present nothing to do.
1157 	return;
1158 }
1159 EXPORT_SYMBOL_GPL(amdtp_domain_destroy);
1160 
1161 /**
1162  * amdtp_domain_add_stream - register isoc context into the domain.
1163  * @d: the AMDTP domain.
1164  * @s: the AMDTP stream.
1165  * @channel: the isochronous channel on the bus.
1166  * @speed: firewire speed code.
1167  */
1168 int amdtp_domain_add_stream(struct amdtp_domain *d, struct amdtp_stream *s,
1169 			    int channel, int speed)
1170 {
1171 	struct amdtp_stream *tmp;
1172 
1173 	list_for_each_entry(tmp, &d->streams, list) {
1174 		if (s == tmp)
1175 			return -EBUSY;
1176 	}
1177 
1178 	list_add(&s->list, &d->streams);
1179 
1180 	s->channel = channel;
1181 	s->speed = speed;
1182 
1183 	return 0;
1184 }
1185 EXPORT_SYMBOL_GPL(amdtp_domain_add_stream);
1186 
1187 /**
1188  * amdtp_domain_start - start sending packets for isoc context in the domain.
1189  * @d: the AMDTP domain.
1190  */
1191 int amdtp_domain_start(struct amdtp_domain *d)
1192 {
1193 	struct amdtp_stream *s;
1194 	int err = 0;
1195 
1196 	list_for_each_entry(s, &d->streams, list) {
1197 		err = amdtp_stream_start(s, s->channel, s->speed);
1198 		if (err < 0)
1199 			break;
1200 	}
1201 
1202 	if (err < 0) {
1203 		list_for_each_entry(s, &d->streams, list)
1204 			amdtp_stream_stop(s);
1205 	}
1206 
1207 	return err;
1208 }
1209 EXPORT_SYMBOL_GPL(amdtp_domain_start);
1210 
1211 /**
1212  * amdtp_domain_stop - stop sending packets for isoc context in the same domain.
1213  * @d: the AMDTP domain to which the isoc contexts belong.
1214  */
1215 void amdtp_domain_stop(struct amdtp_domain *d)
1216 {
1217 	struct amdtp_stream *s, *next;
1218 
1219 	list_for_each_entry_safe(s, next, &d->streams, list) {
1220 		list_del(&s->list);
1221 
1222 		amdtp_stream_stop(s);
1223 	}
1224 }
1225 EXPORT_SYMBOL_GPL(amdtp_domain_stop);
1226