xref: /openbmc/linux/sound/firewire/amdtp-stream.c (revision 9470114d)
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/firewire-constants.h>
13 #include <linux/module.h>
14 #include <linux/slab.h>
15 #include <sound/pcm.h>
16 #include <sound/pcm_params.h>
17 #include "amdtp-stream.h"
18 
19 #define TICKS_PER_CYCLE		3072
20 #define CYCLES_PER_SECOND	8000
21 #define TICKS_PER_SECOND	(TICKS_PER_CYCLE * CYCLES_PER_SECOND)
22 
23 #define OHCI_SECOND_MODULUS		8
24 
25 /* Always support Linux tracing subsystem. */
26 #define CREATE_TRACE_POINTS
27 #include "amdtp-stream-trace.h"
28 
29 #define TRANSFER_DELAY_TICKS	0x2e00 /* 479.17 microseconds */
30 
31 /* isochronous header parameters */
32 #define ISO_DATA_LENGTH_SHIFT	16
33 #define TAG_NO_CIP_HEADER	0
34 #define TAG_CIP			1
35 
36 // Common Isochronous Packet (CIP) header parameters. Use two quadlets CIP header when supported.
37 #define CIP_HEADER_QUADLETS	2
38 #define CIP_EOH_SHIFT		31
39 #define CIP_EOH			(1u << CIP_EOH_SHIFT)
40 #define CIP_EOH_MASK		0x80000000
41 #define CIP_SID_SHIFT		24
42 #define CIP_SID_MASK		0x3f000000
43 #define CIP_DBS_MASK		0x00ff0000
44 #define CIP_DBS_SHIFT		16
45 #define CIP_SPH_MASK		0x00000400
46 #define CIP_SPH_SHIFT		10
47 #define CIP_DBC_MASK		0x000000ff
48 #define CIP_FMT_SHIFT		24
49 #define CIP_FMT_MASK		0x3f000000
50 #define CIP_FDF_MASK		0x00ff0000
51 #define CIP_FDF_SHIFT		16
52 #define CIP_FDF_NO_DATA		0xff
53 #define CIP_SYT_MASK		0x0000ffff
54 #define CIP_SYT_NO_INFO		0xffff
55 #define CIP_SYT_CYCLE_MODULUS	16
56 #define CIP_NO_DATA		((CIP_FDF_NO_DATA << CIP_FDF_SHIFT) | CIP_SYT_NO_INFO)
57 
58 #define CIP_HEADER_SIZE		(sizeof(__be32) * CIP_HEADER_QUADLETS)
59 
60 /* Audio and Music transfer protocol specific parameters */
61 #define CIP_FMT_AM		0x10
62 #define AMDTP_FDF_NO_DATA	0xff
63 
64 // For iso header and tstamp.
65 #define IR_CTX_HEADER_DEFAULT_QUADLETS	2
66 // Add nothing.
67 #define IR_CTX_HEADER_SIZE_NO_CIP	(sizeof(__be32) * IR_CTX_HEADER_DEFAULT_QUADLETS)
68 // Add two quadlets CIP header.
69 #define IR_CTX_HEADER_SIZE_CIP		(IR_CTX_HEADER_SIZE_NO_CIP + CIP_HEADER_SIZE)
70 #define HEADER_TSTAMP_MASK	0x0000ffff
71 
72 #define IT_PKT_HEADER_SIZE_CIP		CIP_HEADER_SIZE
73 #define IT_PKT_HEADER_SIZE_NO_CIP	0 // Nothing.
74 
75 // The initial firmware of OXFW970 can postpone transmission of packet during finishing
76 // asynchronous transaction. This module accepts 5 cycles to skip as maximum to avoid buffer
77 // overrun. Actual device can skip more, then this module stops the packet streaming.
78 #define IR_JUMBO_PAYLOAD_MAX_SKIP_CYCLES	5
79 
80 /**
81  * amdtp_stream_init - initialize an AMDTP stream structure
82  * @s: the AMDTP stream to initialize
83  * @unit: the target of the stream
84  * @dir: the direction of stream
85  * @flags: the details of the streaming protocol consist of cip_flags enumeration-constants.
86  * @fmt: the value of fmt field in CIP header
87  * @process_ctx_payloads: callback handler to process payloads of isoc context
88  * @protocol_size: the size to allocate newly for protocol
89  */
90 int amdtp_stream_init(struct amdtp_stream *s, struct fw_unit *unit,
91 		      enum amdtp_stream_direction dir, unsigned int flags,
92 		      unsigned int fmt,
93 		      amdtp_stream_process_ctx_payloads_t process_ctx_payloads,
94 		      unsigned int protocol_size)
95 {
96 	if (process_ctx_payloads == NULL)
97 		return -EINVAL;
98 
99 	s->protocol = kzalloc(protocol_size, GFP_KERNEL);
100 	if (!s->protocol)
101 		return -ENOMEM;
102 
103 	s->unit = unit;
104 	s->direction = dir;
105 	s->flags = flags;
106 	s->context = ERR_PTR(-1);
107 	mutex_init(&s->mutex);
108 	s->packet_index = 0;
109 
110 	init_waitqueue_head(&s->ready_wait);
111 
112 	s->fmt = fmt;
113 	s->process_ctx_payloads = process_ctx_payloads;
114 
115 	return 0;
116 }
117 EXPORT_SYMBOL(amdtp_stream_init);
118 
119 /**
120  * amdtp_stream_destroy - free stream resources
121  * @s: the AMDTP stream to destroy
122  */
123 void amdtp_stream_destroy(struct amdtp_stream *s)
124 {
125 	/* Not initialized. */
126 	if (s->protocol == NULL)
127 		return;
128 
129 	WARN_ON(amdtp_stream_running(s));
130 	kfree(s->protocol);
131 	mutex_destroy(&s->mutex);
132 }
133 EXPORT_SYMBOL(amdtp_stream_destroy);
134 
135 const unsigned int amdtp_syt_intervals[CIP_SFC_COUNT] = {
136 	[CIP_SFC_32000]  =  8,
137 	[CIP_SFC_44100]  =  8,
138 	[CIP_SFC_48000]  =  8,
139 	[CIP_SFC_88200]  = 16,
140 	[CIP_SFC_96000]  = 16,
141 	[CIP_SFC_176400] = 32,
142 	[CIP_SFC_192000] = 32,
143 };
144 EXPORT_SYMBOL(amdtp_syt_intervals);
145 
146 const unsigned int amdtp_rate_table[CIP_SFC_COUNT] = {
147 	[CIP_SFC_32000]  =  32000,
148 	[CIP_SFC_44100]  =  44100,
149 	[CIP_SFC_48000]  =  48000,
150 	[CIP_SFC_88200]  =  88200,
151 	[CIP_SFC_96000]  =  96000,
152 	[CIP_SFC_176400] = 176400,
153 	[CIP_SFC_192000] = 192000,
154 };
155 EXPORT_SYMBOL(amdtp_rate_table);
156 
157 static int apply_constraint_to_size(struct snd_pcm_hw_params *params,
158 				    struct snd_pcm_hw_rule *rule)
159 {
160 	struct snd_interval *s = hw_param_interval(params, rule->var);
161 	const struct snd_interval *r =
162 		hw_param_interval_c(params, SNDRV_PCM_HW_PARAM_RATE);
163 	struct snd_interval t = {0};
164 	unsigned int step = 0;
165 	int i;
166 
167 	for (i = 0; i < CIP_SFC_COUNT; ++i) {
168 		if (snd_interval_test(r, amdtp_rate_table[i]))
169 			step = max(step, amdtp_syt_intervals[i]);
170 	}
171 
172 	t.min = roundup(s->min, step);
173 	t.max = rounddown(s->max, step);
174 	t.integer = 1;
175 
176 	return snd_interval_refine(s, &t);
177 }
178 
179 /**
180  * amdtp_stream_add_pcm_hw_constraints - add hw constraints for PCM substream
181  * @s:		the AMDTP stream, which must be initialized.
182  * @runtime:	the PCM substream runtime
183  */
184 int amdtp_stream_add_pcm_hw_constraints(struct amdtp_stream *s,
185 					struct snd_pcm_runtime *runtime)
186 {
187 	struct snd_pcm_hardware *hw = &runtime->hw;
188 	unsigned int ctx_header_size;
189 	unsigned int maximum_usec_per_period;
190 	int err;
191 
192 	hw->info = SNDRV_PCM_INFO_BLOCK_TRANSFER |
193 		   SNDRV_PCM_INFO_INTERLEAVED |
194 		   SNDRV_PCM_INFO_JOINT_DUPLEX |
195 		   SNDRV_PCM_INFO_MMAP |
196 		   SNDRV_PCM_INFO_MMAP_VALID |
197 		   SNDRV_PCM_INFO_NO_PERIOD_WAKEUP;
198 
199 	hw->periods_min = 2;
200 	hw->periods_max = UINT_MAX;
201 
202 	/* bytes for a frame */
203 	hw->period_bytes_min = 4 * hw->channels_max;
204 
205 	/* Just to prevent from allocating much pages. */
206 	hw->period_bytes_max = hw->period_bytes_min * 2048;
207 	hw->buffer_bytes_max = hw->period_bytes_max * hw->periods_min;
208 
209 	// Linux driver for 1394 OHCI controller voluntarily flushes isoc
210 	// context when total size of accumulated context header reaches
211 	// PAGE_SIZE. This kicks work for the isoc context and brings
212 	// callback in the middle of scheduled interrupts.
213 	// Although AMDTP streams in the same domain use the same events per
214 	// IRQ, use the largest size of context header between IT/IR contexts.
215 	// Here, use the value of context header in IR context is for both
216 	// contexts.
217 	if (!(s->flags & CIP_NO_HEADER))
218 		ctx_header_size = IR_CTX_HEADER_SIZE_CIP;
219 	else
220 		ctx_header_size = IR_CTX_HEADER_SIZE_NO_CIP;
221 	maximum_usec_per_period = USEC_PER_SEC * PAGE_SIZE /
222 				  CYCLES_PER_SECOND / ctx_header_size;
223 
224 	// In IEC 61883-6, one isoc packet can transfer events up to the value
225 	// of syt interval. This comes from the interval of isoc cycle. As 1394
226 	// OHCI controller can generate hardware IRQ per isoc packet, the
227 	// interval is 125 usec.
228 	// However, there are two ways of transmission in IEC 61883-6; blocking
229 	// and non-blocking modes. In blocking mode, the sequence of isoc packet
230 	// includes 'empty' or 'NODATA' packets which include no event. In
231 	// non-blocking mode, the number of events per packet is variable up to
232 	// the syt interval.
233 	// Due to the above protocol design, the minimum PCM frames per
234 	// interrupt should be double of the value of syt interval, thus it is
235 	// 250 usec.
236 	err = snd_pcm_hw_constraint_minmax(runtime,
237 					   SNDRV_PCM_HW_PARAM_PERIOD_TIME,
238 					   250, maximum_usec_per_period);
239 	if (err < 0)
240 		goto end;
241 
242 	/* Non-Blocking stream has no more constraints */
243 	if (!(s->flags & CIP_BLOCKING))
244 		goto end;
245 
246 	/*
247 	 * One AMDTP packet can include some frames. In blocking mode, the
248 	 * number equals to SYT_INTERVAL. So the number is 8, 16 or 32,
249 	 * depending on its sampling rate. For accurate period interrupt, it's
250 	 * preferrable to align period/buffer sizes to current SYT_INTERVAL.
251 	 */
252 	err = snd_pcm_hw_rule_add(runtime, 0, SNDRV_PCM_HW_PARAM_PERIOD_SIZE,
253 				  apply_constraint_to_size, NULL,
254 				  SNDRV_PCM_HW_PARAM_PERIOD_SIZE,
255 				  SNDRV_PCM_HW_PARAM_RATE, -1);
256 	if (err < 0)
257 		goto end;
258 	err = snd_pcm_hw_rule_add(runtime, 0, SNDRV_PCM_HW_PARAM_BUFFER_SIZE,
259 				  apply_constraint_to_size, NULL,
260 				  SNDRV_PCM_HW_PARAM_BUFFER_SIZE,
261 				  SNDRV_PCM_HW_PARAM_RATE, -1);
262 	if (err < 0)
263 		goto end;
264 end:
265 	return err;
266 }
267 EXPORT_SYMBOL(amdtp_stream_add_pcm_hw_constraints);
268 
269 /**
270  * amdtp_stream_set_parameters - set stream parameters
271  * @s: the AMDTP stream to configure
272  * @rate: the sample rate
273  * @data_block_quadlets: the size of a data block in quadlet unit
274  * @pcm_frame_multiplier: the multiplier to compute the number of PCM frames by the number of AMDTP
275  *			  events.
276  *
277  * The parameters must be set before the stream is started, and must not be
278  * changed while the stream is running.
279  */
280 int amdtp_stream_set_parameters(struct amdtp_stream *s, unsigned int rate,
281 				unsigned int data_block_quadlets, unsigned int pcm_frame_multiplier)
282 {
283 	unsigned int sfc;
284 
285 	for (sfc = 0; sfc < ARRAY_SIZE(amdtp_rate_table); ++sfc) {
286 		if (amdtp_rate_table[sfc] == rate)
287 			break;
288 	}
289 	if (sfc == ARRAY_SIZE(amdtp_rate_table))
290 		return -EINVAL;
291 
292 	s->sfc = sfc;
293 	s->data_block_quadlets = data_block_quadlets;
294 	s->syt_interval = amdtp_syt_intervals[sfc];
295 
296 	// default buffering in the device.
297 	s->transfer_delay = TRANSFER_DELAY_TICKS - TICKS_PER_CYCLE;
298 
299 	// additional buffering needed to adjust for no-data packets.
300 	if (s->flags & CIP_BLOCKING)
301 		s->transfer_delay += TICKS_PER_SECOND * s->syt_interval / rate;
302 
303 	s->pcm_frame_multiplier = pcm_frame_multiplier;
304 
305 	return 0;
306 }
307 EXPORT_SYMBOL(amdtp_stream_set_parameters);
308 
309 // The CIP header is processed in context header apart from context payload.
310 static int amdtp_stream_get_max_ctx_payload_size(struct amdtp_stream *s)
311 {
312 	unsigned int multiplier;
313 
314 	if (s->flags & CIP_JUMBO_PAYLOAD)
315 		multiplier = IR_JUMBO_PAYLOAD_MAX_SKIP_CYCLES;
316 	else
317 		multiplier = 1;
318 
319 	return s->syt_interval * s->data_block_quadlets * sizeof(__be32) * multiplier;
320 }
321 
322 /**
323  * amdtp_stream_get_max_payload - get the stream's packet size
324  * @s: the AMDTP stream
325  *
326  * This function must not be called before the stream has been configured
327  * with amdtp_stream_set_parameters().
328  */
329 unsigned int amdtp_stream_get_max_payload(struct amdtp_stream *s)
330 {
331 	unsigned int cip_header_size;
332 
333 	if (!(s->flags & CIP_NO_HEADER))
334 		cip_header_size = CIP_HEADER_SIZE;
335 	else
336 		cip_header_size = 0;
337 
338 	return cip_header_size + amdtp_stream_get_max_ctx_payload_size(s);
339 }
340 EXPORT_SYMBOL(amdtp_stream_get_max_payload);
341 
342 /**
343  * amdtp_stream_pcm_prepare - prepare PCM device for running
344  * @s: the AMDTP stream
345  *
346  * This function should be called from the PCM device's .prepare callback.
347  */
348 void amdtp_stream_pcm_prepare(struct amdtp_stream *s)
349 {
350 	s->pcm_buffer_pointer = 0;
351 	s->pcm_period_pointer = 0;
352 }
353 EXPORT_SYMBOL(amdtp_stream_pcm_prepare);
354 
355 #define prev_packet_desc(s, desc) \
356 	list_prev_entry_circular(desc, &s->packet_descs_list, link)
357 
358 static void pool_blocking_data_blocks(struct amdtp_stream *s, struct seq_desc *descs,
359 				      unsigned int size, unsigned int pos, unsigned int count)
360 {
361 	const unsigned int syt_interval = s->syt_interval;
362 	int i;
363 
364 	for (i = 0; i < count; ++i) {
365 		struct seq_desc *desc = descs + pos;
366 
367 		if (desc->syt_offset != CIP_SYT_NO_INFO)
368 			desc->data_blocks = syt_interval;
369 		else
370 			desc->data_blocks = 0;
371 
372 		pos = (pos + 1) % size;
373 	}
374 }
375 
376 static void pool_ideal_nonblocking_data_blocks(struct amdtp_stream *s, struct seq_desc *descs,
377 					       unsigned int size, unsigned int pos,
378 					       unsigned int count)
379 {
380 	const enum cip_sfc sfc = s->sfc;
381 	unsigned int state = s->ctx_data.rx.data_block_state;
382 	int i;
383 
384 	for (i = 0; i < count; ++i) {
385 		struct seq_desc *desc = descs + pos;
386 
387 		if (!cip_sfc_is_base_44100(sfc)) {
388 			// Sample_rate / 8000 is an integer, and precomputed.
389 			desc->data_blocks = state;
390 		} else {
391 			unsigned int phase = state;
392 
393 		/*
394 		 * This calculates the number of data blocks per packet so that
395 		 * 1) the overall rate is correct and exactly synchronized to
396 		 *    the bus clock, and
397 		 * 2) packets with a rounded-up number of blocks occur as early
398 		 *    as possible in the sequence (to prevent underruns of the
399 		 *    device's buffer).
400 		 */
401 			if (sfc == CIP_SFC_44100)
402 				/* 6 6 5 6 5 6 5 ... */
403 				desc->data_blocks = 5 + ((phase & 1) ^ (phase == 0 || phase >= 40));
404 			else
405 				/* 12 11 11 11 11 ... or 23 22 22 22 22 ... */
406 				desc->data_blocks = 11 * (sfc >> 1) + (phase == 0);
407 			if (++phase >= (80 >> (sfc >> 1)))
408 				phase = 0;
409 			state = phase;
410 		}
411 
412 		pos = (pos + 1) % size;
413 	}
414 
415 	s->ctx_data.rx.data_block_state = state;
416 }
417 
418 static unsigned int calculate_syt_offset(unsigned int *last_syt_offset,
419 			unsigned int *syt_offset_state, enum cip_sfc sfc)
420 {
421 	unsigned int syt_offset;
422 
423 	if (*last_syt_offset < TICKS_PER_CYCLE) {
424 		if (!cip_sfc_is_base_44100(sfc))
425 			syt_offset = *last_syt_offset + *syt_offset_state;
426 		else {
427 		/*
428 		 * The time, in ticks, of the n'th SYT_INTERVAL sample is:
429 		 *   n * SYT_INTERVAL * 24576000 / sample_rate
430 		 * Modulo TICKS_PER_CYCLE, the difference between successive
431 		 * elements is about 1386.23.  Rounding the results of this
432 		 * formula to the SYT precision results in a sequence of
433 		 * differences that begins with:
434 		 *   1386 1386 1387 1386 1386 1386 1387 1386 1386 1386 1387 ...
435 		 * This code generates _exactly_ the same sequence.
436 		 */
437 			unsigned int phase = *syt_offset_state;
438 			unsigned int index = phase % 13;
439 
440 			syt_offset = *last_syt_offset;
441 			syt_offset += 1386 + ((index && !(index & 3)) ||
442 					      phase == 146);
443 			if (++phase >= 147)
444 				phase = 0;
445 			*syt_offset_state = phase;
446 		}
447 	} else
448 		syt_offset = *last_syt_offset - TICKS_PER_CYCLE;
449 	*last_syt_offset = syt_offset;
450 
451 	if (syt_offset >= TICKS_PER_CYCLE)
452 		syt_offset = CIP_SYT_NO_INFO;
453 
454 	return syt_offset;
455 }
456 
457 static void pool_ideal_syt_offsets(struct amdtp_stream *s, struct seq_desc *descs,
458 				   unsigned int size, unsigned int pos, unsigned int count)
459 {
460 	const enum cip_sfc sfc = s->sfc;
461 	unsigned int last = s->ctx_data.rx.last_syt_offset;
462 	unsigned int state = s->ctx_data.rx.syt_offset_state;
463 	int i;
464 
465 	for (i = 0; i < count; ++i) {
466 		struct seq_desc *desc = descs + pos;
467 
468 		desc->syt_offset = calculate_syt_offset(&last, &state, sfc);
469 
470 		pos = (pos + 1) % size;
471 	}
472 
473 	s->ctx_data.rx.last_syt_offset = last;
474 	s->ctx_data.rx.syt_offset_state = state;
475 }
476 
477 static unsigned int compute_syt_offset(unsigned int syt, unsigned int cycle,
478 				       unsigned int transfer_delay)
479 {
480 	unsigned int cycle_lo = (cycle % CYCLES_PER_SECOND) & 0x0f;
481 	unsigned int syt_cycle_lo = (syt & 0xf000) >> 12;
482 	unsigned int syt_offset;
483 
484 	// Round up.
485 	if (syt_cycle_lo < cycle_lo)
486 		syt_cycle_lo += CIP_SYT_CYCLE_MODULUS;
487 	syt_cycle_lo -= cycle_lo;
488 
489 	// Subtract transfer delay so that the synchronization offset is not so large
490 	// at transmission.
491 	syt_offset = syt_cycle_lo * TICKS_PER_CYCLE + (syt & 0x0fff);
492 	if (syt_offset < transfer_delay)
493 		syt_offset += CIP_SYT_CYCLE_MODULUS * TICKS_PER_CYCLE;
494 
495 	return syt_offset - transfer_delay;
496 }
497 
498 // Both of the producer and consumer of the queue runs in the same clock of IEEE 1394 bus.
499 // Additionally, the sequence of tx packets is severely checked against any discontinuity
500 // before filling entries in the queue. The calculation is safe even if it looks fragile by
501 // overrun.
502 static unsigned int calculate_cached_cycle_count(struct amdtp_stream *s, unsigned int head)
503 {
504 	const unsigned int cache_size = s->ctx_data.tx.cache.size;
505 	unsigned int cycles = s->ctx_data.tx.cache.pos;
506 
507 	if (cycles < head)
508 		cycles += cache_size;
509 	cycles -= head;
510 
511 	return cycles;
512 }
513 
514 static void cache_seq(struct amdtp_stream *s, const struct pkt_desc *src, unsigned int desc_count)
515 {
516 	const unsigned int transfer_delay = s->transfer_delay;
517 	const unsigned int cache_size = s->ctx_data.tx.cache.size;
518 	struct seq_desc *cache = s->ctx_data.tx.cache.descs;
519 	unsigned int cache_pos = s->ctx_data.tx.cache.pos;
520 	bool aware_syt = !(s->flags & CIP_UNAWARE_SYT);
521 	int i;
522 
523 	for (i = 0; i < desc_count; ++i) {
524 		struct seq_desc *dst = cache + cache_pos;
525 
526 		if (aware_syt && src->syt != CIP_SYT_NO_INFO)
527 			dst->syt_offset = compute_syt_offset(src->syt, src->cycle, transfer_delay);
528 		else
529 			dst->syt_offset = CIP_SYT_NO_INFO;
530 		dst->data_blocks = src->data_blocks;
531 
532 		cache_pos = (cache_pos + 1) % cache_size;
533 		src = amdtp_stream_next_packet_desc(s, src);
534 	}
535 
536 	s->ctx_data.tx.cache.pos = cache_pos;
537 }
538 
539 static void pool_ideal_seq_descs(struct amdtp_stream *s, struct seq_desc *descs, unsigned int size,
540 				 unsigned int pos, unsigned int count)
541 {
542 	pool_ideal_syt_offsets(s, descs, size, pos, count);
543 
544 	if (s->flags & CIP_BLOCKING)
545 		pool_blocking_data_blocks(s, descs, size, pos, count);
546 	else
547 		pool_ideal_nonblocking_data_blocks(s, descs, size, pos, count);
548 }
549 
550 static void pool_replayed_seq(struct amdtp_stream *s, struct seq_desc *descs, unsigned int size,
551 			      unsigned int pos, unsigned int count)
552 {
553 	struct amdtp_stream *target = s->ctx_data.rx.replay_target;
554 	const struct seq_desc *cache = target->ctx_data.tx.cache.descs;
555 	const unsigned int cache_size = target->ctx_data.tx.cache.size;
556 	unsigned int cache_pos = s->ctx_data.rx.cache_pos;
557 	int i;
558 
559 	for (i = 0; i < count; ++i) {
560 		descs[pos] = cache[cache_pos];
561 		cache_pos = (cache_pos + 1) % cache_size;
562 		pos = (pos + 1) % size;
563 	}
564 
565 	s->ctx_data.rx.cache_pos = cache_pos;
566 }
567 
568 static void pool_seq_descs(struct amdtp_stream *s, struct seq_desc *descs, unsigned int size,
569 			   unsigned int pos, unsigned int count)
570 {
571 	struct amdtp_domain *d = s->domain;
572 	void (*pool_seq_descs)(struct amdtp_stream *s, struct seq_desc *descs, unsigned int size,
573 			       unsigned int pos, unsigned int count);
574 
575 	if (!d->replay.enable || !s->ctx_data.rx.replay_target) {
576 		pool_seq_descs = pool_ideal_seq_descs;
577 	} else {
578 		if (!d->replay.on_the_fly) {
579 			pool_seq_descs = pool_replayed_seq;
580 		} else {
581 			struct amdtp_stream *tx = s->ctx_data.rx.replay_target;
582 			const unsigned int cache_size = tx->ctx_data.tx.cache.size;
583 			const unsigned int cache_pos = s->ctx_data.rx.cache_pos;
584 			unsigned int cached_cycles = calculate_cached_cycle_count(tx, cache_pos);
585 
586 			if (cached_cycles > count && cached_cycles > cache_size / 2)
587 				pool_seq_descs = pool_replayed_seq;
588 			else
589 				pool_seq_descs = pool_ideal_seq_descs;
590 		}
591 	}
592 
593 	pool_seq_descs(s, descs, size, pos, count);
594 }
595 
596 static void update_pcm_pointers(struct amdtp_stream *s,
597 				struct snd_pcm_substream *pcm,
598 				unsigned int frames)
599 {
600 	unsigned int ptr;
601 
602 	ptr = s->pcm_buffer_pointer + frames;
603 	if (ptr >= pcm->runtime->buffer_size)
604 		ptr -= pcm->runtime->buffer_size;
605 	WRITE_ONCE(s->pcm_buffer_pointer, ptr);
606 
607 	s->pcm_period_pointer += frames;
608 	if (s->pcm_period_pointer >= pcm->runtime->period_size) {
609 		s->pcm_period_pointer -= pcm->runtime->period_size;
610 
611 		// The program in user process should periodically check the status of intermediate
612 		// buffer associated to PCM substream to process PCM frames in the buffer, instead
613 		// of receiving notification of period elapsed by poll wait.
614 		if (!pcm->runtime->no_period_wakeup) {
615 			if (in_softirq()) {
616 				// In software IRQ context for 1394 OHCI.
617 				snd_pcm_period_elapsed(pcm);
618 			} else {
619 				// In process context of ALSA PCM application under acquired lock of
620 				// PCM substream.
621 				snd_pcm_period_elapsed_under_stream_lock(pcm);
622 			}
623 		}
624 	}
625 }
626 
627 static int queue_packet(struct amdtp_stream *s, struct fw_iso_packet *params,
628 			bool sched_irq)
629 {
630 	int err;
631 
632 	params->interrupt = sched_irq;
633 	params->tag = s->tag;
634 	params->sy = 0;
635 
636 	err = fw_iso_context_queue(s->context, params, &s->buffer.iso_buffer,
637 				   s->buffer.packets[s->packet_index].offset);
638 	if (err < 0) {
639 		dev_err(&s->unit->device, "queueing error: %d\n", err);
640 		goto end;
641 	}
642 
643 	if (++s->packet_index >= s->queue_size)
644 		s->packet_index = 0;
645 end:
646 	return err;
647 }
648 
649 static inline int queue_out_packet(struct amdtp_stream *s,
650 				   struct fw_iso_packet *params, bool sched_irq)
651 {
652 	params->skip =
653 		!!(params->header_length == 0 && params->payload_length == 0);
654 	return queue_packet(s, params, sched_irq);
655 }
656 
657 static inline int queue_in_packet(struct amdtp_stream *s,
658 				  struct fw_iso_packet *params)
659 {
660 	// Queue one packet for IR context.
661 	params->header_length = s->ctx_data.tx.ctx_header_size;
662 	params->payload_length = s->ctx_data.tx.max_ctx_payload_length;
663 	params->skip = false;
664 	return queue_packet(s, params, false);
665 }
666 
667 static void generate_cip_header(struct amdtp_stream *s, __be32 cip_header[2],
668 			unsigned int data_block_counter, unsigned int syt)
669 {
670 	cip_header[0] = cpu_to_be32(READ_ONCE(s->source_node_id_field) |
671 				(s->data_block_quadlets << CIP_DBS_SHIFT) |
672 				((s->sph << CIP_SPH_SHIFT) & CIP_SPH_MASK) |
673 				data_block_counter);
674 	cip_header[1] = cpu_to_be32(CIP_EOH |
675 			((s->fmt << CIP_FMT_SHIFT) & CIP_FMT_MASK) |
676 			((s->ctx_data.rx.fdf << CIP_FDF_SHIFT) & CIP_FDF_MASK) |
677 			(syt & CIP_SYT_MASK));
678 }
679 
680 static void build_it_pkt_header(struct amdtp_stream *s, unsigned int cycle,
681 				struct fw_iso_packet *params, unsigned int header_length,
682 				unsigned int data_blocks,
683 				unsigned int data_block_counter,
684 				unsigned int syt, unsigned int index, u32 curr_cycle_time)
685 {
686 	unsigned int payload_length;
687 	__be32 *cip_header;
688 
689 	payload_length = data_blocks * sizeof(__be32) * s->data_block_quadlets;
690 	params->payload_length = payload_length;
691 
692 	if (header_length > 0) {
693 		cip_header = (__be32 *)params->header;
694 		generate_cip_header(s, cip_header, data_block_counter, syt);
695 		params->header_length = header_length;
696 	} else {
697 		cip_header = NULL;
698 	}
699 
700 	trace_amdtp_packet(s, cycle, cip_header, payload_length + header_length, data_blocks,
701 			   data_block_counter, s->packet_index, index, curr_cycle_time);
702 }
703 
704 static int check_cip_header(struct amdtp_stream *s, const __be32 *buf,
705 			    unsigned int payload_length,
706 			    unsigned int *data_blocks,
707 			    unsigned int *data_block_counter, unsigned int *syt)
708 {
709 	u32 cip_header[2];
710 	unsigned int sph;
711 	unsigned int fmt;
712 	unsigned int fdf;
713 	unsigned int dbc;
714 	bool lost;
715 
716 	cip_header[0] = be32_to_cpu(buf[0]);
717 	cip_header[1] = be32_to_cpu(buf[1]);
718 
719 	/*
720 	 * This module supports 'Two-quadlet CIP header with SYT field'.
721 	 * For convenience, also check FMT field is AM824 or not.
722 	 */
723 	if ((((cip_header[0] & CIP_EOH_MASK) == CIP_EOH) ||
724 	     ((cip_header[1] & CIP_EOH_MASK) != CIP_EOH)) &&
725 	    (!(s->flags & CIP_HEADER_WITHOUT_EOH))) {
726 		dev_info_ratelimited(&s->unit->device,
727 				"Invalid CIP header for AMDTP: %08X:%08X\n",
728 				cip_header[0], cip_header[1]);
729 		return -EAGAIN;
730 	}
731 
732 	/* Check valid protocol or not. */
733 	sph = (cip_header[0] & CIP_SPH_MASK) >> CIP_SPH_SHIFT;
734 	fmt = (cip_header[1] & CIP_FMT_MASK) >> CIP_FMT_SHIFT;
735 	if (sph != s->sph || fmt != s->fmt) {
736 		dev_info_ratelimited(&s->unit->device,
737 				     "Detect unexpected protocol: %08x %08x\n",
738 				     cip_header[0], cip_header[1]);
739 		return -EAGAIN;
740 	}
741 
742 	/* Calculate data blocks */
743 	fdf = (cip_header[1] & CIP_FDF_MASK) >> CIP_FDF_SHIFT;
744 	if (payload_length == 0 || (fmt == CIP_FMT_AM && fdf == AMDTP_FDF_NO_DATA)) {
745 		*data_blocks = 0;
746 	} else {
747 		unsigned int data_block_quadlets =
748 				(cip_header[0] & CIP_DBS_MASK) >> CIP_DBS_SHIFT;
749 		/* avoid division by zero */
750 		if (data_block_quadlets == 0) {
751 			dev_err(&s->unit->device,
752 				"Detect invalid value in dbs field: %08X\n",
753 				cip_header[0]);
754 			return -EPROTO;
755 		}
756 		if (s->flags & CIP_WRONG_DBS)
757 			data_block_quadlets = s->data_block_quadlets;
758 
759 		*data_blocks = payload_length / sizeof(__be32) / data_block_quadlets;
760 	}
761 
762 	/* Check data block counter continuity */
763 	dbc = cip_header[0] & CIP_DBC_MASK;
764 	if (*data_blocks == 0 && (s->flags & CIP_EMPTY_HAS_WRONG_DBC) &&
765 	    *data_block_counter != UINT_MAX)
766 		dbc = *data_block_counter;
767 
768 	if ((dbc == 0x00 && (s->flags & CIP_SKIP_DBC_ZERO_CHECK)) ||
769 	    *data_block_counter == UINT_MAX) {
770 		lost = false;
771 	} else if (!(s->flags & CIP_DBC_IS_END_EVENT)) {
772 		lost = dbc != *data_block_counter;
773 	} else {
774 		unsigned int dbc_interval;
775 
776 		if (*data_blocks > 0 && s->ctx_data.tx.dbc_interval > 0)
777 			dbc_interval = s->ctx_data.tx.dbc_interval;
778 		else
779 			dbc_interval = *data_blocks;
780 
781 		lost = dbc != ((*data_block_counter + dbc_interval) & 0xff);
782 	}
783 
784 	if (lost) {
785 		dev_err(&s->unit->device,
786 			"Detect discontinuity of CIP: %02X %02X\n",
787 			*data_block_counter, dbc);
788 		return -EIO;
789 	}
790 
791 	*data_block_counter = dbc;
792 
793 	if (!(s->flags & CIP_UNAWARE_SYT))
794 		*syt = cip_header[1] & CIP_SYT_MASK;
795 
796 	return 0;
797 }
798 
799 static int parse_ir_ctx_header(struct amdtp_stream *s, unsigned int cycle,
800 			       const __be32 *ctx_header,
801 			       unsigned int *data_blocks,
802 			       unsigned int *data_block_counter,
803 			       unsigned int *syt, unsigned int packet_index, unsigned int index,
804 			       u32 curr_cycle_time)
805 {
806 	unsigned int payload_length;
807 	const __be32 *cip_header;
808 	unsigned int cip_header_size;
809 
810 	payload_length = be32_to_cpu(ctx_header[0]) >> ISO_DATA_LENGTH_SHIFT;
811 
812 	if (!(s->flags & CIP_NO_HEADER))
813 		cip_header_size = CIP_HEADER_SIZE;
814 	else
815 		cip_header_size = 0;
816 
817 	if (payload_length > cip_header_size + s->ctx_data.tx.max_ctx_payload_length) {
818 		dev_err(&s->unit->device,
819 			"Detect jumbo payload: %04x %04x\n",
820 			payload_length, cip_header_size + s->ctx_data.tx.max_ctx_payload_length);
821 		return -EIO;
822 	}
823 
824 	if (cip_header_size > 0) {
825 		if (payload_length >= cip_header_size) {
826 			int err;
827 
828 			cip_header = ctx_header + IR_CTX_HEADER_DEFAULT_QUADLETS;
829 			err = check_cip_header(s, cip_header, payload_length - cip_header_size,
830 					       data_blocks, data_block_counter, syt);
831 			if (err < 0)
832 				return err;
833 		} else {
834 			// Handle the cycle so that empty packet arrives.
835 			cip_header = NULL;
836 			*data_blocks = 0;
837 			*syt = 0;
838 		}
839 	} else {
840 		cip_header = NULL;
841 		*data_blocks = payload_length / sizeof(__be32) / s->data_block_quadlets;
842 		*syt = 0;
843 
844 		if (*data_block_counter == UINT_MAX)
845 			*data_block_counter = 0;
846 	}
847 
848 	trace_amdtp_packet(s, cycle, cip_header, payload_length, *data_blocks,
849 			   *data_block_counter, packet_index, index, curr_cycle_time);
850 
851 	return 0;
852 }
853 
854 // In CYCLE_TIMER register of IEEE 1394, 7 bits are used to represent second. On
855 // the other hand, in DMA descriptors of 1394 OHCI, 3 bits are used to represent
856 // it. Thus, via Linux firewire subsystem, we can get the 3 bits for second.
857 static inline u32 compute_ohci_iso_ctx_cycle_count(u32 tstamp)
858 {
859 	return (((tstamp >> 13) & 0x07) * CYCLES_PER_SECOND) + (tstamp & 0x1fff);
860 }
861 
862 static inline u32 compute_ohci_cycle_count(__be32 ctx_header_tstamp)
863 {
864 	u32 tstamp = be32_to_cpu(ctx_header_tstamp) & HEADER_TSTAMP_MASK;
865 	return compute_ohci_iso_ctx_cycle_count(tstamp);
866 }
867 
868 static inline u32 increment_ohci_cycle_count(u32 cycle, unsigned int addend)
869 {
870 	cycle += addend;
871 	if (cycle >= OHCI_SECOND_MODULUS * CYCLES_PER_SECOND)
872 		cycle -= OHCI_SECOND_MODULUS * CYCLES_PER_SECOND;
873 	return cycle;
874 }
875 
876 static inline u32 decrement_ohci_cycle_count(u32 minuend, u32 subtrahend)
877 {
878 	if (minuend < subtrahend)
879 		minuend += OHCI_SECOND_MODULUS * CYCLES_PER_SECOND;
880 
881 	return minuend - subtrahend;
882 }
883 
884 static int compare_ohci_cycle_count(u32 lval, u32 rval)
885 {
886 	if (lval == rval)
887 		return 0;
888 	else if (lval < rval && rval - lval < OHCI_SECOND_MODULUS * CYCLES_PER_SECOND / 2)
889 		return -1;
890 	else
891 		return 1;
892 }
893 
894 // Align to actual cycle count for the packet which is going to be scheduled.
895 // This module queued the same number of isochronous cycle as the size of queue
896 // to kip isochronous cycle, therefore it's OK to just increment the cycle by
897 // the size of queue for scheduled cycle.
898 static inline u32 compute_ohci_it_cycle(const __be32 ctx_header_tstamp,
899 					unsigned int queue_size)
900 {
901 	u32 cycle = compute_ohci_cycle_count(ctx_header_tstamp);
902 	return increment_ohci_cycle_count(cycle, queue_size);
903 }
904 
905 static int generate_tx_packet_descs(struct amdtp_stream *s, struct pkt_desc *desc,
906 				    const __be32 *ctx_header, unsigned int packet_count,
907 				    unsigned int *desc_count)
908 {
909 	unsigned int next_cycle = s->next_cycle;
910 	unsigned int dbc = s->data_block_counter;
911 	unsigned int packet_index = s->packet_index;
912 	unsigned int queue_size = s->queue_size;
913 	u32 curr_cycle_time = 0;
914 	int i;
915 	int err;
916 
917 	if (trace_amdtp_packet_enabled())
918 		(void)fw_card_read_cycle_time(fw_parent_device(s->unit)->card, &curr_cycle_time);
919 
920 	*desc_count = 0;
921 	for (i = 0; i < packet_count; ++i) {
922 		unsigned int cycle;
923 		bool lost;
924 		unsigned int data_blocks;
925 		unsigned int syt;
926 
927 		cycle = compute_ohci_cycle_count(ctx_header[1]);
928 		lost = (next_cycle != cycle);
929 		if (lost) {
930 			if (s->flags & CIP_NO_HEADER) {
931 				// Fireface skips transmission just for an isoc cycle corresponding
932 				// to empty packet.
933 				unsigned int prev_cycle = next_cycle;
934 
935 				next_cycle = increment_ohci_cycle_count(next_cycle, 1);
936 				lost = (next_cycle != cycle);
937 				if (!lost) {
938 					// Prepare a description for the skipped cycle for
939 					// sequence replay.
940 					desc->cycle = prev_cycle;
941 					desc->syt = 0;
942 					desc->data_blocks = 0;
943 					desc->data_block_counter = dbc;
944 					desc->ctx_payload = NULL;
945 					desc = amdtp_stream_next_packet_desc(s, desc);
946 					++(*desc_count);
947 				}
948 			} else if (s->flags & CIP_JUMBO_PAYLOAD) {
949 				// OXFW970 skips transmission for several isoc cycles during
950 				// asynchronous transaction. The sequence replay is impossible due
951 				// to the reason.
952 				unsigned int safe_cycle = increment_ohci_cycle_count(next_cycle,
953 								IR_JUMBO_PAYLOAD_MAX_SKIP_CYCLES);
954 				lost = (compare_ohci_cycle_count(safe_cycle, cycle) < 0);
955 			}
956 			if (lost) {
957 				dev_err(&s->unit->device, "Detect discontinuity of cycle: %d %d\n",
958 					next_cycle, cycle);
959 				return -EIO;
960 			}
961 		}
962 
963 		err = parse_ir_ctx_header(s, cycle, ctx_header, &data_blocks, &dbc, &syt,
964 					  packet_index, i, curr_cycle_time);
965 		if (err < 0)
966 			return err;
967 
968 		desc->cycle = cycle;
969 		desc->syt = syt;
970 		desc->data_blocks = data_blocks;
971 		desc->data_block_counter = dbc;
972 		desc->ctx_payload = s->buffer.packets[packet_index].buffer;
973 
974 		if (!(s->flags & CIP_DBC_IS_END_EVENT))
975 			dbc = (dbc + desc->data_blocks) & 0xff;
976 
977 		next_cycle = increment_ohci_cycle_count(next_cycle, 1);
978 		desc = amdtp_stream_next_packet_desc(s, desc);
979 		++(*desc_count);
980 		ctx_header += s->ctx_data.tx.ctx_header_size / sizeof(*ctx_header);
981 		packet_index = (packet_index + 1) % queue_size;
982 	}
983 
984 	s->next_cycle = next_cycle;
985 	s->data_block_counter = dbc;
986 
987 	return 0;
988 }
989 
990 static unsigned int compute_syt(unsigned int syt_offset, unsigned int cycle,
991 				unsigned int transfer_delay)
992 {
993 	unsigned int syt;
994 
995 	syt_offset += transfer_delay;
996 	syt = ((cycle + syt_offset / TICKS_PER_CYCLE) << 12) |
997 	      (syt_offset % TICKS_PER_CYCLE);
998 	return syt & CIP_SYT_MASK;
999 }
1000 
1001 static void generate_rx_packet_descs(struct amdtp_stream *s, struct pkt_desc *desc,
1002 				     const __be32 *ctx_header, unsigned int packet_count)
1003 {
1004 	struct seq_desc *seq_descs = s->ctx_data.rx.seq.descs;
1005 	unsigned int seq_size = s->ctx_data.rx.seq.size;
1006 	unsigned int seq_pos = s->ctx_data.rx.seq.pos;
1007 	unsigned int dbc = s->data_block_counter;
1008 	bool aware_syt = !(s->flags & CIP_UNAWARE_SYT);
1009 	int i;
1010 
1011 	pool_seq_descs(s, seq_descs, seq_size, seq_pos, packet_count);
1012 
1013 	for (i = 0; i < packet_count; ++i) {
1014 		unsigned int index = (s->packet_index + i) % s->queue_size;
1015 		const struct seq_desc *seq = seq_descs + seq_pos;
1016 
1017 		desc->cycle = compute_ohci_it_cycle(*ctx_header, s->queue_size);
1018 
1019 		if (aware_syt && seq->syt_offset != CIP_SYT_NO_INFO)
1020 			desc->syt = compute_syt(seq->syt_offset, desc->cycle, s->transfer_delay);
1021 		else
1022 			desc->syt = CIP_SYT_NO_INFO;
1023 
1024 		desc->data_blocks = seq->data_blocks;
1025 
1026 		if (s->flags & CIP_DBC_IS_END_EVENT)
1027 			dbc = (dbc + desc->data_blocks) & 0xff;
1028 
1029 		desc->data_block_counter = dbc;
1030 
1031 		if (!(s->flags & CIP_DBC_IS_END_EVENT))
1032 			dbc = (dbc + desc->data_blocks) & 0xff;
1033 
1034 		desc->ctx_payload = s->buffer.packets[index].buffer;
1035 
1036 		seq_pos = (seq_pos + 1) % seq_size;
1037 		desc = amdtp_stream_next_packet_desc(s, desc);
1038 
1039 		++ctx_header;
1040 	}
1041 
1042 	s->data_block_counter = dbc;
1043 	s->ctx_data.rx.seq.pos = seq_pos;
1044 }
1045 
1046 static inline void cancel_stream(struct amdtp_stream *s)
1047 {
1048 	s->packet_index = -1;
1049 	if (in_softirq())
1050 		amdtp_stream_pcm_abort(s);
1051 	WRITE_ONCE(s->pcm_buffer_pointer, SNDRV_PCM_POS_XRUN);
1052 }
1053 
1054 static snd_pcm_sframes_t compute_pcm_extra_delay(struct amdtp_stream *s,
1055 						 const struct pkt_desc *desc, unsigned int count)
1056 {
1057 	unsigned int data_block_count = 0;
1058 	u32 latest_cycle;
1059 	u32 cycle_time;
1060 	u32 curr_cycle;
1061 	u32 cycle_gap;
1062 	int i, err;
1063 
1064 	if (count == 0)
1065 		goto end;
1066 
1067 	// Forward to the latest record.
1068 	for (i = 0; i < count - 1; ++i)
1069 		desc = amdtp_stream_next_packet_desc(s, desc);
1070 	latest_cycle = desc->cycle;
1071 
1072 	err = fw_card_read_cycle_time(fw_parent_device(s->unit)->card, &cycle_time);
1073 	if (err < 0)
1074 		goto end;
1075 
1076 	// Compute cycle count with lower 3 bits of second field and cycle field like timestamp
1077 	// format of 1394 OHCI isochronous context.
1078 	curr_cycle = compute_ohci_iso_ctx_cycle_count((cycle_time >> 12) & 0x0000ffff);
1079 
1080 	if (s->direction == AMDTP_IN_STREAM) {
1081 		// NOTE: The AMDTP packet descriptor should be for the past isochronous cycle since
1082 		// it corresponds to arrived isochronous packet.
1083 		if (compare_ohci_cycle_count(latest_cycle, curr_cycle) > 0)
1084 			goto end;
1085 		cycle_gap = decrement_ohci_cycle_count(curr_cycle, latest_cycle);
1086 
1087 		// NOTE: estimate delay by recent history of arrived AMDTP packets. The estimated
1088 		// value expectedly corresponds to a few packets (0-2) since the packet arrived at
1089 		// the most recent isochronous cycle has been already processed.
1090 		for (i = 0; i < cycle_gap; ++i) {
1091 			desc = amdtp_stream_next_packet_desc(s, desc);
1092 			data_block_count += desc->data_blocks;
1093 		}
1094 	} else {
1095 		// NOTE: The AMDTP packet descriptor should be for the future isochronous cycle
1096 		// since it was already scheduled.
1097 		if (compare_ohci_cycle_count(latest_cycle, curr_cycle) < 0)
1098 			goto end;
1099 		cycle_gap = decrement_ohci_cycle_count(latest_cycle, curr_cycle);
1100 
1101 		// NOTE: use history of scheduled packets.
1102 		for (i = 0; i < cycle_gap; ++i) {
1103 			data_block_count += desc->data_blocks;
1104 			desc = prev_packet_desc(s, desc);
1105 		}
1106 	}
1107 end:
1108 	return data_block_count * s->pcm_frame_multiplier;
1109 }
1110 
1111 static void process_ctx_payloads(struct amdtp_stream *s,
1112 				 const struct pkt_desc *desc,
1113 				 unsigned int count)
1114 {
1115 	struct snd_pcm_substream *pcm;
1116 	int i;
1117 
1118 	pcm = READ_ONCE(s->pcm);
1119 	s->process_ctx_payloads(s, desc, count, pcm);
1120 
1121 	if (pcm) {
1122 		unsigned int data_block_count = 0;
1123 
1124 		pcm->runtime->delay = compute_pcm_extra_delay(s, desc, count);
1125 
1126 		for (i = 0; i < count; ++i) {
1127 			data_block_count += desc->data_blocks;
1128 			desc = amdtp_stream_next_packet_desc(s, desc);
1129 		}
1130 
1131 		update_pcm_pointers(s, pcm, data_block_count * s->pcm_frame_multiplier);
1132 	}
1133 }
1134 
1135 static void process_rx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1136 			       void *header, void *private_data)
1137 {
1138 	struct amdtp_stream *s = private_data;
1139 	const struct amdtp_domain *d = s->domain;
1140 	const __be32 *ctx_header = header;
1141 	const unsigned int events_per_period = d->events_per_period;
1142 	unsigned int event_count = s->ctx_data.rx.event_count;
1143 	struct pkt_desc *desc = s->packet_descs_cursor;
1144 	unsigned int pkt_header_length;
1145 	unsigned int packets;
1146 	u32 curr_cycle_time;
1147 	bool need_hw_irq;
1148 	int i;
1149 
1150 	if (s->packet_index < 0)
1151 		return;
1152 
1153 	// Calculate the number of packets in buffer and check XRUN.
1154 	packets = header_length / sizeof(*ctx_header);
1155 
1156 	generate_rx_packet_descs(s, desc, ctx_header, packets);
1157 
1158 	process_ctx_payloads(s, desc, packets);
1159 
1160 	if (!(s->flags & CIP_NO_HEADER))
1161 		pkt_header_length = IT_PKT_HEADER_SIZE_CIP;
1162 	else
1163 		pkt_header_length = 0;
1164 
1165 	if (s == d->irq_target) {
1166 		// At NO_PERIOD_WAKEUP mode, the packets for all IT/IR contexts are processed by
1167 		// the tasks of user process operating ALSA PCM character device by calling ioctl(2)
1168 		// with some requests, instead of scheduled hardware IRQ of an IT context.
1169 		struct snd_pcm_substream *pcm = READ_ONCE(s->pcm);
1170 		need_hw_irq = !pcm || !pcm->runtime->no_period_wakeup;
1171 	} else {
1172 		need_hw_irq = false;
1173 	}
1174 
1175 	if (trace_amdtp_packet_enabled())
1176 		(void)fw_card_read_cycle_time(fw_parent_device(s->unit)->card, &curr_cycle_time);
1177 
1178 	for (i = 0; i < packets; ++i) {
1179 		struct {
1180 			struct fw_iso_packet params;
1181 			__be32 header[CIP_HEADER_QUADLETS];
1182 		} template = { {0}, {0} };
1183 		bool sched_irq = false;
1184 
1185 		build_it_pkt_header(s, desc->cycle, &template.params, pkt_header_length,
1186 				    desc->data_blocks, desc->data_block_counter,
1187 				    desc->syt, i, curr_cycle_time);
1188 
1189 		if (s == s->domain->irq_target) {
1190 			event_count += desc->data_blocks;
1191 			if (event_count >= events_per_period) {
1192 				event_count -= events_per_period;
1193 				sched_irq = need_hw_irq;
1194 			}
1195 		}
1196 
1197 		if (queue_out_packet(s, &template.params, sched_irq) < 0) {
1198 			cancel_stream(s);
1199 			return;
1200 		}
1201 
1202 		desc = amdtp_stream_next_packet_desc(s, desc);
1203 	}
1204 
1205 	s->ctx_data.rx.event_count = event_count;
1206 	s->packet_descs_cursor = desc;
1207 }
1208 
1209 static void skip_rx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1210 			    void *header, void *private_data)
1211 {
1212 	struct amdtp_stream *s = private_data;
1213 	struct amdtp_domain *d = s->domain;
1214 	const __be32 *ctx_header = header;
1215 	unsigned int packets;
1216 	unsigned int cycle;
1217 	int i;
1218 
1219 	if (s->packet_index < 0)
1220 		return;
1221 
1222 	packets = header_length / sizeof(*ctx_header);
1223 
1224 	cycle = compute_ohci_it_cycle(ctx_header[packets - 1], s->queue_size);
1225 	s->next_cycle = increment_ohci_cycle_count(cycle, 1);
1226 
1227 	for (i = 0; i < packets; ++i) {
1228 		struct fw_iso_packet params = {
1229 			.header_length = 0,
1230 			.payload_length = 0,
1231 		};
1232 		bool sched_irq = (s == d->irq_target && i == packets - 1);
1233 
1234 		if (queue_out_packet(s, &params, sched_irq) < 0) {
1235 			cancel_stream(s);
1236 			return;
1237 		}
1238 	}
1239 }
1240 
1241 static void irq_target_callback(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1242 				void *header, void *private_data);
1243 
1244 static void process_rx_packets_intermediately(struct fw_iso_context *context, u32 tstamp,
1245 					size_t header_length, void *header, void *private_data)
1246 {
1247 	struct amdtp_stream *s = private_data;
1248 	struct amdtp_domain *d = s->domain;
1249 	__be32 *ctx_header = header;
1250 	const unsigned int queue_size = s->queue_size;
1251 	unsigned int packets;
1252 	unsigned int offset;
1253 
1254 	if (s->packet_index < 0)
1255 		return;
1256 
1257 	packets = header_length / sizeof(*ctx_header);
1258 
1259 	offset = 0;
1260 	while (offset < packets) {
1261 		unsigned int cycle = compute_ohci_it_cycle(ctx_header[offset], queue_size);
1262 
1263 		if (compare_ohci_cycle_count(cycle, d->processing_cycle.rx_start) >= 0)
1264 			break;
1265 
1266 		++offset;
1267 	}
1268 
1269 	if (offset > 0) {
1270 		unsigned int length = sizeof(*ctx_header) * offset;
1271 
1272 		skip_rx_packets(context, tstamp, length, ctx_header, private_data);
1273 		if (amdtp_streaming_error(s))
1274 			return;
1275 
1276 		ctx_header += offset;
1277 		header_length -= length;
1278 	}
1279 
1280 	if (offset < packets) {
1281 		s->ready_processing = true;
1282 		wake_up(&s->ready_wait);
1283 
1284 		if (d->replay.enable)
1285 			s->ctx_data.rx.cache_pos = 0;
1286 
1287 		process_rx_packets(context, tstamp, header_length, ctx_header, private_data);
1288 		if (amdtp_streaming_error(s))
1289 			return;
1290 
1291 		if (s == d->irq_target)
1292 			s->context->callback.sc = irq_target_callback;
1293 		else
1294 			s->context->callback.sc = process_rx_packets;
1295 	}
1296 }
1297 
1298 static void process_tx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1299 			       void *header, void *private_data)
1300 {
1301 	struct amdtp_stream *s = private_data;
1302 	__be32 *ctx_header = header;
1303 	struct pkt_desc *desc = s->packet_descs_cursor;
1304 	unsigned int packet_count;
1305 	unsigned int desc_count;
1306 	int i;
1307 	int err;
1308 
1309 	if (s->packet_index < 0)
1310 		return;
1311 
1312 	// Calculate the number of packets in buffer and check XRUN.
1313 	packet_count = header_length / s->ctx_data.tx.ctx_header_size;
1314 
1315 	desc_count = 0;
1316 	err = generate_tx_packet_descs(s, desc, ctx_header, packet_count, &desc_count);
1317 	if (err < 0) {
1318 		if (err != -EAGAIN) {
1319 			cancel_stream(s);
1320 			return;
1321 		}
1322 	} else {
1323 		struct amdtp_domain *d = s->domain;
1324 
1325 		process_ctx_payloads(s, desc, desc_count);
1326 
1327 		if (d->replay.enable)
1328 			cache_seq(s, desc, desc_count);
1329 
1330 		for (i = 0; i < desc_count; ++i)
1331 			desc = amdtp_stream_next_packet_desc(s, desc);
1332 		s->packet_descs_cursor = desc;
1333 	}
1334 
1335 	for (i = 0; i < packet_count; ++i) {
1336 		struct fw_iso_packet params = {0};
1337 
1338 		if (queue_in_packet(s, &params) < 0) {
1339 			cancel_stream(s);
1340 			return;
1341 		}
1342 	}
1343 }
1344 
1345 static void drop_tx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1346 			    void *header, void *private_data)
1347 {
1348 	struct amdtp_stream *s = private_data;
1349 	const __be32 *ctx_header = header;
1350 	unsigned int packets;
1351 	unsigned int cycle;
1352 	int i;
1353 
1354 	if (s->packet_index < 0)
1355 		return;
1356 
1357 	packets = header_length / s->ctx_data.tx.ctx_header_size;
1358 
1359 	ctx_header += (packets - 1) * s->ctx_data.tx.ctx_header_size / sizeof(*ctx_header);
1360 	cycle = compute_ohci_cycle_count(ctx_header[1]);
1361 	s->next_cycle = increment_ohci_cycle_count(cycle, 1);
1362 
1363 	for (i = 0; i < packets; ++i) {
1364 		struct fw_iso_packet params = {0};
1365 
1366 		if (queue_in_packet(s, &params) < 0) {
1367 			cancel_stream(s);
1368 			return;
1369 		}
1370 	}
1371 }
1372 
1373 static void process_tx_packets_intermediately(struct fw_iso_context *context, u32 tstamp,
1374 					size_t header_length, void *header, void *private_data)
1375 {
1376 	struct amdtp_stream *s = private_data;
1377 	struct amdtp_domain *d = s->domain;
1378 	__be32 *ctx_header;
1379 	unsigned int packets;
1380 	unsigned int offset;
1381 
1382 	if (s->packet_index < 0)
1383 		return;
1384 
1385 	packets = header_length / s->ctx_data.tx.ctx_header_size;
1386 
1387 	offset = 0;
1388 	ctx_header = header;
1389 	while (offset < packets) {
1390 		unsigned int cycle = compute_ohci_cycle_count(ctx_header[1]);
1391 
1392 		if (compare_ohci_cycle_count(cycle, d->processing_cycle.tx_start) >= 0)
1393 			break;
1394 
1395 		ctx_header += s->ctx_data.tx.ctx_header_size / sizeof(__be32);
1396 		++offset;
1397 	}
1398 
1399 	ctx_header = header;
1400 
1401 	if (offset > 0) {
1402 		size_t length = s->ctx_data.tx.ctx_header_size * offset;
1403 
1404 		drop_tx_packets(context, tstamp, length, ctx_header, s);
1405 		if (amdtp_streaming_error(s))
1406 			return;
1407 
1408 		ctx_header += length / sizeof(*ctx_header);
1409 		header_length -= length;
1410 	}
1411 
1412 	if (offset < packets) {
1413 		s->ready_processing = true;
1414 		wake_up(&s->ready_wait);
1415 
1416 		process_tx_packets(context, tstamp, header_length, ctx_header, s);
1417 		if (amdtp_streaming_error(s))
1418 			return;
1419 
1420 		context->callback.sc = process_tx_packets;
1421 	}
1422 }
1423 
1424 static void drop_tx_packets_initially(struct fw_iso_context *context, u32 tstamp,
1425 				      size_t header_length, void *header, void *private_data)
1426 {
1427 	struct amdtp_stream *s = private_data;
1428 	struct amdtp_domain *d = s->domain;
1429 	__be32 *ctx_header;
1430 	unsigned int count;
1431 	unsigned int events;
1432 	int i;
1433 
1434 	if (s->packet_index < 0)
1435 		return;
1436 
1437 	count = header_length / s->ctx_data.tx.ctx_header_size;
1438 
1439 	// Attempt to detect any event in the batch of packets.
1440 	events = 0;
1441 	ctx_header = header;
1442 	for (i = 0; i < count; ++i) {
1443 		unsigned int payload_quads =
1444 			(be32_to_cpu(*ctx_header) >> ISO_DATA_LENGTH_SHIFT) / sizeof(__be32);
1445 		unsigned int data_blocks;
1446 
1447 		if (s->flags & CIP_NO_HEADER) {
1448 			data_blocks = payload_quads / s->data_block_quadlets;
1449 		} else {
1450 			__be32 *cip_headers = ctx_header + IR_CTX_HEADER_DEFAULT_QUADLETS;
1451 
1452 			if (payload_quads < CIP_HEADER_QUADLETS) {
1453 				data_blocks = 0;
1454 			} else {
1455 				payload_quads -= CIP_HEADER_QUADLETS;
1456 
1457 				if (s->flags & CIP_UNAWARE_SYT) {
1458 					data_blocks = payload_quads / s->data_block_quadlets;
1459 				} else {
1460 					u32 cip1 = be32_to_cpu(cip_headers[1]);
1461 
1462 					// NODATA packet can includes any data blocks but they are
1463 					// not available as event.
1464 					if ((cip1 & CIP_NO_DATA) == CIP_NO_DATA)
1465 						data_blocks = 0;
1466 					else
1467 						data_blocks = payload_quads / s->data_block_quadlets;
1468 				}
1469 			}
1470 		}
1471 
1472 		events += data_blocks;
1473 
1474 		ctx_header += s->ctx_data.tx.ctx_header_size / sizeof(__be32);
1475 	}
1476 
1477 	drop_tx_packets(context, tstamp, header_length, header, s);
1478 
1479 	if (events > 0)
1480 		s->ctx_data.tx.event_starts = true;
1481 
1482 	// Decide the cycle count to begin processing content of packet in IR contexts.
1483 	{
1484 		unsigned int stream_count = 0;
1485 		unsigned int event_starts_count = 0;
1486 		unsigned int cycle = UINT_MAX;
1487 
1488 		list_for_each_entry(s, &d->streams, list) {
1489 			if (s->direction == AMDTP_IN_STREAM) {
1490 				++stream_count;
1491 				if (s->ctx_data.tx.event_starts)
1492 					++event_starts_count;
1493 			}
1494 		}
1495 
1496 		if (stream_count == event_starts_count) {
1497 			unsigned int next_cycle;
1498 
1499 			list_for_each_entry(s, &d->streams, list) {
1500 				if (s->direction != AMDTP_IN_STREAM)
1501 					continue;
1502 
1503 				next_cycle = increment_ohci_cycle_count(s->next_cycle,
1504 								d->processing_cycle.tx_init_skip);
1505 				if (cycle == UINT_MAX ||
1506 				    compare_ohci_cycle_count(next_cycle, cycle) > 0)
1507 					cycle = next_cycle;
1508 
1509 				s->context->callback.sc = process_tx_packets_intermediately;
1510 			}
1511 
1512 			d->processing_cycle.tx_start = cycle;
1513 		}
1514 	}
1515 }
1516 
1517 static void process_ctxs_in_domain(struct amdtp_domain *d)
1518 {
1519 	struct amdtp_stream *s;
1520 
1521 	list_for_each_entry(s, &d->streams, list) {
1522 		if (s != d->irq_target && amdtp_stream_running(s))
1523 			fw_iso_context_flush_completions(s->context);
1524 
1525 		if (amdtp_streaming_error(s))
1526 			goto error;
1527 	}
1528 
1529 	return;
1530 error:
1531 	if (amdtp_stream_running(d->irq_target))
1532 		cancel_stream(d->irq_target);
1533 
1534 	list_for_each_entry(s, &d->streams, list) {
1535 		if (amdtp_stream_running(s))
1536 			cancel_stream(s);
1537 	}
1538 }
1539 
1540 static void irq_target_callback(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1541 				void *header, void *private_data)
1542 {
1543 	struct amdtp_stream *s = private_data;
1544 	struct amdtp_domain *d = s->domain;
1545 
1546 	process_rx_packets(context, tstamp, header_length, header, private_data);
1547 	process_ctxs_in_domain(d);
1548 }
1549 
1550 static void irq_target_callback_intermediately(struct fw_iso_context *context, u32 tstamp,
1551 					size_t header_length, void *header, void *private_data)
1552 {
1553 	struct amdtp_stream *s = private_data;
1554 	struct amdtp_domain *d = s->domain;
1555 
1556 	process_rx_packets_intermediately(context, tstamp, header_length, header, private_data);
1557 	process_ctxs_in_domain(d);
1558 }
1559 
1560 static void irq_target_callback_skip(struct fw_iso_context *context, u32 tstamp,
1561 				     size_t header_length, void *header, void *private_data)
1562 {
1563 	struct amdtp_stream *s = private_data;
1564 	struct amdtp_domain *d = s->domain;
1565 	bool ready_to_start;
1566 
1567 	skip_rx_packets(context, tstamp, header_length, header, private_data);
1568 	process_ctxs_in_domain(d);
1569 
1570 	if (d->replay.enable && !d->replay.on_the_fly) {
1571 		unsigned int rx_count = 0;
1572 		unsigned int rx_ready_count = 0;
1573 		struct amdtp_stream *rx;
1574 
1575 		list_for_each_entry(rx, &d->streams, list) {
1576 			struct amdtp_stream *tx;
1577 			unsigned int cached_cycles;
1578 
1579 			if (rx->direction != AMDTP_OUT_STREAM)
1580 				continue;
1581 			++rx_count;
1582 
1583 			tx = rx->ctx_data.rx.replay_target;
1584 			cached_cycles = calculate_cached_cycle_count(tx, 0);
1585 			if (cached_cycles > tx->ctx_data.tx.cache.size / 2)
1586 				++rx_ready_count;
1587 		}
1588 
1589 		ready_to_start = (rx_count == rx_ready_count);
1590 	} else {
1591 		ready_to_start = true;
1592 	}
1593 
1594 	// Decide the cycle count to begin processing content of packet in IT contexts. All of IT
1595 	// contexts are expected to start and get callback when reaching here.
1596 	if (ready_to_start) {
1597 		unsigned int cycle = s->next_cycle;
1598 		list_for_each_entry(s, &d->streams, list) {
1599 			if (s->direction != AMDTP_OUT_STREAM)
1600 				continue;
1601 
1602 			if (compare_ohci_cycle_count(s->next_cycle, cycle) > 0)
1603 				cycle = s->next_cycle;
1604 
1605 			if (s == d->irq_target)
1606 				s->context->callback.sc = irq_target_callback_intermediately;
1607 			else
1608 				s->context->callback.sc = process_rx_packets_intermediately;
1609 		}
1610 
1611 		d->processing_cycle.rx_start = cycle;
1612 	}
1613 }
1614 
1615 // This is executed one time. For in-stream, first packet has come. For out-stream, prepared to
1616 // transmit first packet.
1617 static void amdtp_stream_first_callback(struct fw_iso_context *context,
1618 					u32 tstamp, size_t header_length,
1619 					void *header, void *private_data)
1620 {
1621 	struct amdtp_stream *s = private_data;
1622 	struct amdtp_domain *d = s->domain;
1623 
1624 	if (s->direction == AMDTP_IN_STREAM) {
1625 		context->callback.sc = drop_tx_packets_initially;
1626 	} else {
1627 		if (s == d->irq_target)
1628 			context->callback.sc = irq_target_callback_skip;
1629 		else
1630 			context->callback.sc = skip_rx_packets;
1631 	}
1632 
1633 	context->callback.sc(context, tstamp, header_length, header, s);
1634 }
1635 
1636 /**
1637  * amdtp_stream_start - start transferring packets
1638  * @s: the AMDTP stream to start
1639  * @channel: the isochronous channel on the bus
1640  * @speed: firewire speed code
1641  * @queue_size: The number of packets in the queue.
1642  * @idle_irq_interval: the interval to queue packet during initial state.
1643  *
1644  * The stream cannot be started until it has been configured with
1645  * amdtp_stream_set_parameters() and it must be started before any PCM or MIDI
1646  * device can be started.
1647  */
1648 static int amdtp_stream_start(struct amdtp_stream *s, int channel, int speed,
1649 			      unsigned int queue_size, unsigned int idle_irq_interval)
1650 {
1651 	bool is_irq_target = (s == s->domain->irq_target);
1652 	unsigned int ctx_header_size;
1653 	unsigned int max_ctx_payload_size;
1654 	enum dma_data_direction dir;
1655 	struct pkt_desc *descs;
1656 	int i, type, tag, err;
1657 
1658 	mutex_lock(&s->mutex);
1659 
1660 	if (WARN_ON(amdtp_stream_running(s) ||
1661 		    (s->data_block_quadlets < 1))) {
1662 		err = -EBADFD;
1663 		goto err_unlock;
1664 	}
1665 
1666 	if (s->direction == AMDTP_IN_STREAM) {
1667 		// NOTE: IT context should be used for constant IRQ.
1668 		if (is_irq_target) {
1669 			err = -EINVAL;
1670 			goto err_unlock;
1671 		}
1672 
1673 		s->data_block_counter = UINT_MAX;
1674 	} else {
1675 		s->data_block_counter = 0;
1676 	}
1677 
1678 	// initialize packet buffer.
1679 	if (s->direction == AMDTP_IN_STREAM) {
1680 		dir = DMA_FROM_DEVICE;
1681 		type = FW_ISO_CONTEXT_RECEIVE;
1682 		if (!(s->flags & CIP_NO_HEADER))
1683 			ctx_header_size = IR_CTX_HEADER_SIZE_CIP;
1684 		else
1685 			ctx_header_size = IR_CTX_HEADER_SIZE_NO_CIP;
1686 	} else {
1687 		dir = DMA_TO_DEVICE;
1688 		type = FW_ISO_CONTEXT_TRANSMIT;
1689 		ctx_header_size = 0;	// No effect for IT context.
1690 	}
1691 	max_ctx_payload_size = amdtp_stream_get_max_ctx_payload_size(s);
1692 
1693 	err = iso_packets_buffer_init(&s->buffer, s->unit, queue_size, max_ctx_payload_size, dir);
1694 	if (err < 0)
1695 		goto err_unlock;
1696 	s->queue_size = queue_size;
1697 
1698 	s->context = fw_iso_context_create(fw_parent_device(s->unit)->card,
1699 					  type, channel, speed, ctx_header_size,
1700 					  amdtp_stream_first_callback, s);
1701 	if (IS_ERR(s->context)) {
1702 		err = PTR_ERR(s->context);
1703 		if (err == -EBUSY)
1704 			dev_err(&s->unit->device,
1705 				"no free stream on this controller\n");
1706 		goto err_buffer;
1707 	}
1708 
1709 	amdtp_stream_update(s);
1710 
1711 	if (s->direction == AMDTP_IN_STREAM) {
1712 		s->ctx_data.tx.max_ctx_payload_length = max_ctx_payload_size;
1713 		s->ctx_data.tx.ctx_header_size = ctx_header_size;
1714 		s->ctx_data.tx.event_starts = false;
1715 
1716 		if (s->domain->replay.enable) {
1717 			// struct fw_iso_context.drop_overflow_headers is false therefore it's
1718 			// possible to cache much unexpectedly.
1719 			s->ctx_data.tx.cache.size = max_t(unsigned int, s->syt_interval * 2,
1720 							  queue_size * 3 / 2);
1721 			s->ctx_data.tx.cache.pos = 0;
1722 			s->ctx_data.tx.cache.descs = kcalloc(s->ctx_data.tx.cache.size,
1723 						sizeof(*s->ctx_data.tx.cache.descs), GFP_KERNEL);
1724 			if (!s->ctx_data.tx.cache.descs) {
1725 				err = -ENOMEM;
1726 				goto err_context;
1727 			}
1728 		}
1729 	} else {
1730 		static const struct {
1731 			unsigned int data_block;
1732 			unsigned int syt_offset;
1733 		} *entry, initial_state[] = {
1734 			[CIP_SFC_32000]  = {  4, 3072 },
1735 			[CIP_SFC_48000]  = {  6, 1024 },
1736 			[CIP_SFC_96000]  = { 12, 1024 },
1737 			[CIP_SFC_192000] = { 24, 1024 },
1738 			[CIP_SFC_44100]  = {  0,   67 },
1739 			[CIP_SFC_88200]  = {  0,   67 },
1740 			[CIP_SFC_176400] = {  0,   67 },
1741 		};
1742 
1743 		s->ctx_data.rx.seq.descs = kcalloc(queue_size, sizeof(*s->ctx_data.rx.seq.descs), GFP_KERNEL);
1744 		if (!s->ctx_data.rx.seq.descs) {
1745 			err = -ENOMEM;
1746 			goto err_context;
1747 		}
1748 		s->ctx_data.rx.seq.size = queue_size;
1749 		s->ctx_data.rx.seq.pos = 0;
1750 
1751 		entry = &initial_state[s->sfc];
1752 		s->ctx_data.rx.data_block_state = entry->data_block;
1753 		s->ctx_data.rx.syt_offset_state = entry->syt_offset;
1754 		s->ctx_data.rx.last_syt_offset = TICKS_PER_CYCLE;
1755 
1756 		s->ctx_data.rx.event_count = 0;
1757 	}
1758 
1759 	if (s->flags & CIP_NO_HEADER)
1760 		s->tag = TAG_NO_CIP_HEADER;
1761 	else
1762 		s->tag = TAG_CIP;
1763 
1764 	// NOTE: When operating without hardIRQ/softIRQ, applications tends to call ioctl request
1765 	// for runtime of PCM substream in the interval equivalent to the size of PCM buffer. It
1766 	// could take a round over queue of AMDTP packet descriptors and small loss of history. For
1767 	// safe, keep more 8 elements for the queue, equivalent to 1 ms.
1768 	descs = kcalloc(s->queue_size + 8, sizeof(*descs), GFP_KERNEL);
1769 	if (!descs) {
1770 		err = -ENOMEM;
1771 		goto err_context;
1772 	}
1773 	s->packet_descs = descs;
1774 
1775 	INIT_LIST_HEAD(&s->packet_descs_list);
1776 	for (i = 0; i < s->queue_size; ++i) {
1777 		INIT_LIST_HEAD(&descs->link);
1778 		list_add_tail(&descs->link, &s->packet_descs_list);
1779 		++descs;
1780 	}
1781 	s->packet_descs_cursor = list_first_entry(&s->packet_descs_list, struct pkt_desc, link);
1782 
1783 	s->packet_index = 0;
1784 	do {
1785 		struct fw_iso_packet params;
1786 
1787 		if (s->direction == AMDTP_IN_STREAM) {
1788 			err = queue_in_packet(s, &params);
1789 		} else {
1790 			bool sched_irq = false;
1791 
1792 			params.header_length = 0;
1793 			params.payload_length = 0;
1794 
1795 			if (is_irq_target) {
1796 				sched_irq = !((s->packet_index + 1) %
1797 					      idle_irq_interval);
1798 			}
1799 
1800 			err = queue_out_packet(s, &params, sched_irq);
1801 		}
1802 		if (err < 0)
1803 			goto err_pkt_descs;
1804 	} while (s->packet_index > 0);
1805 
1806 	/* NOTE: TAG1 matches CIP. This just affects in stream. */
1807 	tag = FW_ISO_CONTEXT_MATCH_TAG1;
1808 	if ((s->flags & CIP_EMPTY_WITH_TAG0) || (s->flags & CIP_NO_HEADER))
1809 		tag |= FW_ISO_CONTEXT_MATCH_TAG0;
1810 
1811 	s->ready_processing = false;
1812 	err = fw_iso_context_start(s->context, -1, 0, tag);
1813 	if (err < 0)
1814 		goto err_pkt_descs;
1815 
1816 	mutex_unlock(&s->mutex);
1817 
1818 	return 0;
1819 err_pkt_descs:
1820 	kfree(s->packet_descs);
1821 	s->packet_descs = NULL;
1822 err_context:
1823 	if (s->direction == AMDTP_OUT_STREAM) {
1824 		kfree(s->ctx_data.rx.seq.descs);
1825 	} else {
1826 		if (s->domain->replay.enable)
1827 			kfree(s->ctx_data.tx.cache.descs);
1828 	}
1829 	fw_iso_context_destroy(s->context);
1830 	s->context = ERR_PTR(-1);
1831 err_buffer:
1832 	iso_packets_buffer_destroy(&s->buffer, s->unit);
1833 err_unlock:
1834 	mutex_unlock(&s->mutex);
1835 
1836 	return err;
1837 }
1838 
1839 /**
1840  * amdtp_domain_stream_pcm_pointer - get the PCM buffer position
1841  * @d: the AMDTP domain.
1842  * @s: the AMDTP stream that transports the PCM data
1843  *
1844  * Returns the current buffer position, in frames.
1845  */
1846 unsigned long amdtp_domain_stream_pcm_pointer(struct amdtp_domain *d,
1847 					      struct amdtp_stream *s)
1848 {
1849 	struct amdtp_stream *irq_target = d->irq_target;
1850 
1851 	// Process isochronous packets queued till recent isochronous cycle to handle PCM frames.
1852 	if (irq_target && amdtp_stream_running(irq_target)) {
1853 		// In software IRQ context, the call causes dead-lock to disable the tasklet
1854 		// synchronously.
1855 		if (!in_softirq())
1856 			fw_iso_context_flush_completions(irq_target->context);
1857 	}
1858 
1859 	return READ_ONCE(s->pcm_buffer_pointer);
1860 }
1861 EXPORT_SYMBOL_GPL(amdtp_domain_stream_pcm_pointer);
1862 
1863 /**
1864  * amdtp_domain_stream_pcm_ack - acknowledge queued PCM frames
1865  * @d: the AMDTP domain.
1866  * @s: the AMDTP stream that transfers the PCM frames
1867  *
1868  * Returns zero always.
1869  */
1870 int amdtp_domain_stream_pcm_ack(struct amdtp_domain *d, struct amdtp_stream *s)
1871 {
1872 	struct amdtp_stream *irq_target = d->irq_target;
1873 
1874 	// Process isochronous packets for recent isochronous cycle to handle
1875 	// queued PCM frames.
1876 	if (irq_target && amdtp_stream_running(irq_target))
1877 		fw_iso_context_flush_completions(irq_target->context);
1878 
1879 	return 0;
1880 }
1881 EXPORT_SYMBOL_GPL(amdtp_domain_stream_pcm_ack);
1882 
1883 /**
1884  * amdtp_stream_update - update the stream after a bus reset
1885  * @s: the AMDTP stream
1886  */
1887 void amdtp_stream_update(struct amdtp_stream *s)
1888 {
1889 	/* Precomputing. */
1890 	WRITE_ONCE(s->source_node_id_field,
1891                    (fw_parent_device(s->unit)->card->node_id << CIP_SID_SHIFT) & CIP_SID_MASK);
1892 }
1893 EXPORT_SYMBOL(amdtp_stream_update);
1894 
1895 /**
1896  * amdtp_stream_stop - stop sending packets
1897  * @s: the AMDTP stream to stop
1898  *
1899  * All PCM and MIDI devices of the stream must be stopped before the stream
1900  * itself can be stopped.
1901  */
1902 static void amdtp_stream_stop(struct amdtp_stream *s)
1903 {
1904 	mutex_lock(&s->mutex);
1905 
1906 	if (!amdtp_stream_running(s)) {
1907 		mutex_unlock(&s->mutex);
1908 		return;
1909 	}
1910 
1911 	fw_iso_context_stop(s->context);
1912 	fw_iso_context_destroy(s->context);
1913 	s->context = ERR_PTR(-1);
1914 	iso_packets_buffer_destroy(&s->buffer, s->unit);
1915 	kfree(s->packet_descs);
1916 	s->packet_descs = NULL;
1917 
1918 	if (s->direction == AMDTP_OUT_STREAM) {
1919 		kfree(s->ctx_data.rx.seq.descs);
1920 	} else {
1921 		if (s->domain->replay.enable)
1922 			kfree(s->ctx_data.tx.cache.descs);
1923 	}
1924 
1925 	mutex_unlock(&s->mutex);
1926 }
1927 
1928 /**
1929  * amdtp_stream_pcm_abort - abort the running PCM device
1930  * @s: the AMDTP stream about to be stopped
1931  *
1932  * If the isochronous stream needs to be stopped asynchronously, call this
1933  * function first to stop the PCM device.
1934  */
1935 void amdtp_stream_pcm_abort(struct amdtp_stream *s)
1936 {
1937 	struct snd_pcm_substream *pcm;
1938 
1939 	pcm = READ_ONCE(s->pcm);
1940 	if (pcm)
1941 		snd_pcm_stop_xrun(pcm);
1942 }
1943 EXPORT_SYMBOL(amdtp_stream_pcm_abort);
1944 
1945 /**
1946  * amdtp_domain_init - initialize an AMDTP domain structure
1947  * @d: the AMDTP domain to initialize.
1948  */
1949 int amdtp_domain_init(struct amdtp_domain *d)
1950 {
1951 	INIT_LIST_HEAD(&d->streams);
1952 
1953 	d->events_per_period = 0;
1954 
1955 	return 0;
1956 }
1957 EXPORT_SYMBOL_GPL(amdtp_domain_init);
1958 
1959 /**
1960  * amdtp_domain_destroy - destroy an AMDTP domain structure
1961  * @d: the AMDTP domain to destroy.
1962  */
1963 void amdtp_domain_destroy(struct amdtp_domain *d)
1964 {
1965 	// At present nothing to do.
1966 	return;
1967 }
1968 EXPORT_SYMBOL_GPL(amdtp_domain_destroy);
1969 
1970 /**
1971  * amdtp_domain_add_stream - register isoc context into the domain.
1972  * @d: the AMDTP domain.
1973  * @s: the AMDTP stream.
1974  * @channel: the isochronous channel on the bus.
1975  * @speed: firewire speed code.
1976  */
1977 int amdtp_domain_add_stream(struct amdtp_domain *d, struct amdtp_stream *s,
1978 			    int channel, int speed)
1979 {
1980 	struct amdtp_stream *tmp;
1981 
1982 	list_for_each_entry(tmp, &d->streams, list) {
1983 		if (s == tmp)
1984 			return -EBUSY;
1985 	}
1986 
1987 	list_add(&s->list, &d->streams);
1988 
1989 	s->channel = channel;
1990 	s->speed = speed;
1991 	s->domain = d;
1992 
1993 	return 0;
1994 }
1995 EXPORT_SYMBOL_GPL(amdtp_domain_add_stream);
1996 
1997 // Make the reference from rx stream to tx stream for sequence replay. When the number of tx streams
1998 // is less than the number of rx streams, the first tx stream is selected.
1999 static int make_association(struct amdtp_domain *d)
2000 {
2001 	unsigned int dst_index = 0;
2002 	struct amdtp_stream *rx;
2003 
2004 	// Make association to replay target.
2005 	list_for_each_entry(rx, &d->streams, list) {
2006 		if (rx->direction == AMDTP_OUT_STREAM) {
2007 			unsigned int src_index = 0;
2008 			struct amdtp_stream *tx = NULL;
2009 			struct amdtp_stream *s;
2010 
2011 			list_for_each_entry(s, &d->streams, list) {
2012 				if (s->direction == AMDTP_IN_STREAM) {
2013 					if (dst_index == src_index) {
2014 						tx = s;
2015 						break;
2016 					}
2017 
2018 					++src_index;
2019 				}
2020 			}
2021 			if (!tx) {
2022 				// Select the first entry.
2023 				list_for_each_entry(s, &d->streams, list) {
2024 					if (s->direction == AMDTP_IN_STREAM) {
2025 						tx = s;
2026 						break;
2027 					}
2028 				}
2029 				// No target is available to replay sequence.
2030 				if (!tx)
2031 					return -EINVAL;
2032 			}
2033 
2034 			rx->ctx_data.rx.replay_target = tx;
2035 
2036 			++dst_index;
2037 		}
2038 	}
2039 
2040 	return 0;
2041 }
2042 
2043 /**
2044  * amdtp_domain_start - start sending packets for isoc context in the domain.
2045  * @d: the AMDTP domain.
2046  * @tx_init_skip_cycles: the number of cycles to skip processing packets at initial stage of IR
2047  *			 contexts.
2048  * @replay_seq: whether to replay the sequence of packet in IR context for the sequence of packet in
2049  *		IT context.
2050  * @replay_on_the_fly: transfer rx packets according to nominal frequency, then begin to replay
2051  *		       according to arrival of events in tx packets.
2052  */
2053 int amdtp_domain_start(struct amdtp_domain *d, unsigned int tx_init_skip_cycles, bool replay_seq,
2054 		       bool replay_on_the_fly)
2055 {
2056 	unsigned int events_per_buffer = d->events_per_buffer;
2057 	unsigned int events_per_period = d->events_per_period;
2058 	unsigned int queue_size;
2059 	struct amdtp_stream *s;
2060 	bool found = false;
2061 	int err;
2062 
2063 	if (replay_seq) {
2064 		err = make_association(d);
2065 		if (err < 0)
2066 			return err;
2067 	}
2068 	d->replay.enable = replay_seq;
2069 	d->replay.on_the_fly = replay_on_the_fly;
2070 
2071 	// Select an IT context as IRQ target.
2072 	list_for_each_entry(s, &d->streams, list) {
2073 		if (s->direction == AMDTP_OUT_STREAM) {
2074 			found = true;
2075 			break;
2076 		}
2077 	}
2078 	if (!found)
2079 		return -ENXIO;
2080 	d->irq_target = s;
2081 
2082 	d->processing_cycle.tx_init_skip = tx_init_skip_cycles;
2083 
2084 	// This is a case that AMDTP streams in domain run just for MIDI
2085 	// substream. Use the number of events equivalent to 10 msec as
2086 	// interval of hardware IRQ.
2087 	if (events_per_period == 0)
2088 		events_per_period = amdtp_rate_table[d->irq_target->sfc] / 100;
2089 	if (events_per_buffer == 0)
2090 		events_per_buffer = events_per_period * 3;
2091 
2092 	queue_size = DIV_ROUND_UP(CYCLES_PER_SECOND * events_per_buffer,
2093 				  amdtp_rate_table[d->irq_target->sfc]);
2094 
2095 	list_for_each_entry(s, &d->streams, list) {
2096 		unsigned int idle_irq_interval = 0;
2097 
2098 		if (s->direction == AMDTP_OUT_STREAM && s == d->irq_target) {
2099 			idle_irq_interval = DIV_ROUND_UP(CYCLES_PER_SECOND * events_per_period,
2100 							 amdtp_rate_table[d->irq_target->sfc]);
2101 		}
2102 
2103 		// Starts immediately but actually DMA context starts several hundred cycles later.
2104 		err = amdtp_stream_start(s, s->channel, s->speed, queue_size, idle_irq_interval);
2105 		if (err < 0)
2106 			goto error;
2107 	}
2108 
2109 	return 0;
2110 error:
2111 	list_for_each_entry(s, &d->streams, list)
2112 		amdtp_stream_stop(s);
2113 	return err;
2114 }
2115 EXPORT_SYMBOL_GPL(amdtp_domain_start);
2116 
2117 /**
2118  * amdtp_domain_stop - stop sending packets for isoc context in the same domain.
2119  * @d: the AMDTP domain to which the isoc contexts belong.
2120  */
2121 void amdtp_domain_stop(struct amdtp_domain *d)
2122 {
2123 	struct amdtp_stream *s, *next;
2124 
2125 	if (d->irq_target)
2126 		amdtp_stream_stop(d->irq_target);
2127 
2128 	list_for_each_entry_safe(s, next, &d->streams, list) {
2129 		list_del(&s->list);
2130 
2131 		if (s != d->irq_target)
2132 			amdtp_stream_stop(s);
2133 	}
2134 
2135 	d->events_per_period = 0;
2136 	d->irq_target = NULL;
2137 }
2138 EXPORT_SYMBOL_GPL(amdtp_domain_stop);
2139