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 */
90int 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}
117EXPORT_SYMBOL(amdtp_stream_init);
118
119/**
120 * amdtp_stream_destroy - free stream resources
121 * @s: the AMDTP stream to destroy
122 */
123void 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}
133EXPORT_SYMBOL(amdtp_stream_destroy);
134
135const 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};
144EXPORT_SYMBOL(amdtp_syt_intervals);
145
146const 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};
155EXPORT_SYMBOL(amdtp_rate_table);
156
157static 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 */
184int 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;
264end:
265	return err;
266}
267EXPORT_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 */
280int 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}
307EXPORT_SYMBOL(amdtp_stream_set_parameters);
308
309// The CIP header is processed in context header apart from context payload.
310static 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 */
329unsigned 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}
340EXPORT_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 */
348void amdtp_stream_pcm_prepare(struct amdtp_stream *s)
349{
350	s->pcm_buffer_pointer = 0;
351	s->pcm_period_pointer = 0;
352}
353EXPORT_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
358static 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
376static 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
418static 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
457static 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
477static 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.
502static 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
514static 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
539static 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
550static 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
568static 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
596static 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
627static 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;
645end:
646	return err;
647}
648
649static 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
657static 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
667static 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
680static 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
704static 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 (!(s->flags & CIP_DBC_IS_PAYLOAD_QUADLETS)) {
777			if (*data_blocks > 0 && s->ctx_data.tx.dbc_interval > 0)
778				dbc_interval = s->ctx_data.tx.dbc_interval;
779			else
780				dbc_interval = *data_blocks;
781		} else {
782			dbc_interval = payload_length / sizeof(__be32);
783		}
784
785		lost = dbc != ((*data_block_counter + dbc_interval) & 0xff);
786	}
787
788	if (lost) {
789		dev_err(&s->unit->device,
790			"Detect discontinuity of CIP: %02X %02X\n",
791			*data_block_counter, dbc);
792		return -EIO;
793	}
794
795	*data_block_counter = dbc;
796
797	if (!(s->flags & CIP_UNAWARE_SYT))
798		*syt = cip_header[1] & CIP_SYT_MASK;
799
800	return 0;
801}
802
803static int parse_ir_ctx_header(struct amdtp_stream *s, unsigned int cycle,
804			       const __be32 *ctx_header,
805			       unsigned int *data_blocks,
806			       unsigned int *data_block_counter,
807			       unsigned int *syt, unsigned int packet_index, unsigned int index,
808			       u32 curr_cycle_time)
809{
810	unsigned int payload_length;
811	const __be32 *cip_header;
812	unsigned int cip_header_size;
813
814	payload_length = be32_to_cpu(ctx_header[0]) >> ISO_DATA_LENGTH_SHIFT;
815
816	if (!(s->flags & CIP_NO_HEADER))
817		cip_header_size = CIP_HEADER_SIZE;
818	else
819		cip_header_size = 0;
820
821	if (payload_length > cip_header_size + s->ctx_data.tx.max_ctx_payload_length) {
822		dev_err(&s->unit->device,
823			"Detect jumbo payload: %04x %04x\n",
824			payload_length, cip_header_size + s->ctx_data.tx.max_ctx_payload_length);
825		return -EIO;
826	}
827
828	if (cip_header_size > 0) {
829		if (payload_length >= cip_header_size) {
830			int err;
831
832			cip_header = ctx_header + IR_CTX_HEADER_DEFAULT_QUADLETS;
833			err = check_cip_header(s, cip_header, payload_length - cip_header_size,
834					       data_blocks, data_block_counter, syt);
835			if (err < 0)
836				return err;
837		} else {
838			// Handle the cycle so that empty packet arrives.
839			cip_header = NULL;
840			*data_blocks = 0;
841			*syt = 0;
842		}
843	} else {
844		cip_header = NULL;
845		*data_blocks = payload_length / sizeof(__be32) / s->data_block_quadlets;
846		*syt = 0;
847
848		if (*data_block_counter == UINT_MAX)
849			*data_block_counter = 0;
850	}
851
852	trace_amdtp_packet(s, cycle, cip_header, payload_length, *data_blocks,
853			   *data_block_counter, packet_index, index, curr_cycle_time);
854
855	return 0;
856}
857
858// In CYCLE_TIMER register of IEEE 1394, 7 bits are used to represent second. On
859// the other hand, in DMA descriptors of 1394 OHCI, 3 bits are used to represent
860// it. Thus, via Linux firewire subsystem, we can get the 3 bits for second.
861static inline u32 compute_ohci_iso_ctx_cycle_count(u32 tstamp)
862{
863	return (((tstamp >> 13) & 0x07) * CYCLES_PER_SECOND) + (tstamp & 0x1fff);
864}
865
866static inline u32 compute_ohci_cycle_count(__be32 ctx_header_tstamp)
867{
868	u32 tstamp = be32_to_cpu(ctx_header_tstamp) & HEADER_TSTAMP_MASK;
869	return compute_ohci_iso_ctx_cycle_count(tstamp);
870}
871
872static inline u32 increment_ohci_cycle_count(u32 cycle, unsigned int addend)
873{
874	cycle += addend;
875	if (cycle >= OHCI_SECOND_MODULUS * CYCLES_PER_SECOND)
876		cycle -= OHCI_SECOND_MODULUS * CYCLES_PER_SECOND;
877	return cycle;
878}
879
880static inline u32 decrement_ohci_cycle_count(u32 minuend, u32 subtrahend)
881{
882	if (minuend < subtrahend)
883		minuend += OHCI_SECOND_MODULUS * CYCLES_PER_SECOND;
884
885	return minuend - subtrahend;
886}
887
888static int compare_ohci_cycle_count(u32 lval, u32 rval)
889{
890	if (lval == rval)
891		return 0;
892	else if (lval < rval && rval - lval < OHCI_SECOND_MODULUS * CYCLES_PER_SECOND / 2)
893		return -1;
894	else
895		return 1;
896}
897
898// Align to actual cycle count for the packet which is going to be scheduled.
899// This module queued the same number of isochronous cycle as the size of queue
900// to kip isochronous cycle, therefore it's OK to just increment the cycle by
901// the size of queue for scheduled cycle.
902static inline u32 compute_ohci_it_cycle(const __be32 ctx_header_tstamp,
903					unsigned int queue_size)
904{
905	u32 cycle = compute_ohci_cycle_count(ctx_header_tstamp);
906	return increment_ohci_cycle_count(cycle, queue_size);
907}
908
909static int generate_tx_packet_descs(struct amdtp_stream *s, struct pkt_desc *desc,
910				    const __be32 *ctx_header, unsigned int packet_count,
911				    unsigned int *desc_count)
912{
913	unsigned int next_cycle = s->next_cycle;
914	unsigned int dbc = s->data_block_counter;
915	unsigned int packet_index = s->packet_index;
916	unsigned int queue_size = s->queue_size;
917	u32 curr_cycle_time = 0;
918	int i;
919	int err;
920
921	if (trace_amdtp_packet_enabled())
922		(void)fw_card_read_cycle_time(fw_parent_device(s->unit)->card, &curr_cycle_time);
923
924	*desc_count = 0;
925	for (i = 0; i < packet_count; ++i) {
926		unsigned int cycle;
927		bool lost;
928		unsigned int data_blocks;
929		unsigned int syt;
930
931		cycle = compute_ohci_cycle_count(ctx_header[1]);
932		lost = (next_cycle != cycle);
933		if (lost) {
934			if (s->flags & CIP_NO_HEADER) {
935				// Fireface skips transmission just for an isoc cycle corresponding
936				// to empty packet.
937				unsigned int prev_cycle = next_cycle;
938
939				next_cycle = increment_ohci_cycle_count(next_cycle, 1);
940				lost = (next_cycle != cycle);
941				if (!lost) {
942					// Prepare a description for the skipped cycle for
943					// sequence replay.
944					desc->cycle = prev_cycle;
945					desc->syt = 0;
946					desc->data_blocks = 0;
947					desc->data_block_counter = dbc;
948					desc->ctx_payload = NULL;
949					desc = amdtp_stream_next_packet_desc(s, desc);
950					++(*desc_count);
951				}
952			} else if (s->flags & CIP_JUMBO_PAYLOAD) {
953				// OXFW970 skips transmission for several isoc cycles during
954				// asynchronous transaction. The sequence replay is impossible due
955				// to the reason.
956				unsigned int safe_cycle = increment_ohci_cycle_count(next_cycle,
957								IR_JUMBO_PAYLOAD_MAX_SKIP_CYCLES);
958				lost = (compare_ohci_cycle_count(safe_cycle, cycle) < 0);
959			}
960			if (lost) {
961				dev_err(&s->unit->device, "Detect discontinuity of cycle: %d %d\n",
962					next_cycle, cycle);
963				return -EIO;
964			}
965		}
966
967		err = parse_ir_ctx_header(s, cycle, ctx_header, &data_blocks, &dbc, &syt,
968					  packet_index, i, curr_cycle_time);
969		if (err < 0)
970			return err;
971
972		desc->cycle = cycle;
973		desc->syt = syt;
974		desc->data_blocks = data_blocks;
975		desc->data_block_counter = dbc;
976		desc->ctx_payload = s->buffer.packets[packet_index].buffer;
977
978		if (!(s->flags & CIP_DBC_IS_END_EVENT))
979			dbc = (dbc + desc->data_blocks) & 0xff;
980
981		next_cycle = increment_ohci_cycle_count(next_cycle, 1);
982		desc = amdtp_stream_next_packet_desc(s, desc);
983		++(*desc_count);
984		ctx_header += s->ctx_data.tx.ctx_header_size / sizeof(*ctx_header);
985		packet_index = (packet_index + 1) % queue_size;
986	}
987
988	s->next_cycle = next_cycle;
989	s->data_block_counter = dbc;
990
991	return 0;
992}
993
994static unsigned int compute_syt(unsigned int syt_offset, unsigned int cycle,
995				unsigned int transfer_delay)
996{
997	unsigned int syt;
998
999	syt_offset += transfer_delay;
1000	syt = ((cycle + syt_offset / TICKS_PER_CYCLE) << 12) |
1001	      (syt_offset % TICKS_PER_CYCLE);
1002	return syt & CIP_SYT_MASK;
1003}
1004
1005static void generate_rx_packet_descs(struct amdtp_stream *s, struct pkt_desc *desc,
1006				     const __be32 *ctx_header, unsigned int packet_count)
1007{
1008	struct seq_desc *seq_descs = s->ctx_data.rx.seq.descs;
1009	unsigned int seq_size = s->ctx_data.rx.seq.size;
1010	unsigned int seq_pos = s->ctx_data.rx.seq.pos;
1011	unsigned int dbc = s->data_block_counter;
1012	bool aware_syt = !(s->flags & CIP_UNAWARE_SYT);
1013	int i;
1014
1015	pool_seq_descs(s, seq_descs, seq_size, seq_pos, packet_count);
1016
1017	for (i = 0; i < packet_count; ++i) {
1018		unsigned int index = (s->packet_index + i) % s->queue_size;
1019		const struct seq_desc *seq = seq_descs + seq_pos;
1020
1021		desc->cycle = compute_ohci_it_cycle(*ctx_header, s->queue_size);
1022
1023		if (aware_syt && seq->syt_offset != CIP_SYT_NO_INFO)
1024			desc->syt = compute_syt(seq->syt_offset, desc->cycle, s->transfer_delay);
1025		else
1026			desc->syt = CIP_SYT_NO_INFO;
1027
1028		desc->data_blocks = seq->data_blocks;
1029
1030		if (s->flags & CIP_DBC_IS_END_EVENT)
1031			dbc = (dbc + desc->data_blocks) & 0xff;
1032
1033		desc->data_block_counter = dbc;
1034
1035		if (!(s->flags & CIP_DBC_IS_END_EVENT))
1036			dbc = (dbc + desc->data_blocks) & 0xff;
1037
1038		desc->ctx_payload = s->buffer.packets[index].buffer;
1039
1040		seq_pos = (seq_pos + 1) % seq_size;
1041		desc = amdtp_stream_next_packet_desc(s, desc);
1042
1043		++ctx_header;
1044	}
1045
1046	s->data_block_counter = dbc;
1047	s->ctx_data.rx.seq.pos = seq_pos;
1048}
1049
1050static inline void cancel_stream(struct amdtp_stream *s)
1051{
1052	s->packet_index = -1;
1053	if (in_softirq())
1054		amdtp_stream_pcm_abort(s);
1055	WRITE_ONCE(s->pcm_buffer_pointer, SNDRV_PCM_POS_XRUN);
1056}
1057
1058static snd_pcm_sframes_t compute_pcm_extra_delay(struct amdtp_stream *s,
1059						 const struct pkt_desc *desc, unsigned int count)
1060{
1061	unsigned int data_block_count = 0;
1062	u32 latest_cycle;
1063	u32 cycle_time;
1064	u32 curr_cycle;
1065	u32 cycle_gap;
1066	int i, err;
1067
1068	if (count == 0)
1069		goto end;
1070
1071	// Forward to the latest record.
1072	for (i = 0; i < count - 1; ++i)
1073		desc = amdtp_stream_next_packet_desc(s, desc);
1074	latest_cycle = desc->cycle;
1075
1076	err = fw_card_read_cycle_time(fw_parent_device(s->unit)->card, &cycle_time);
1077	if (err < 0)
1078		goto end;
1079
1080	// Compute cycle count with lower 3 bits of second field and cycle field like timestamp
1081	// format of 1394 OHCI isochronous context.
1082	curr_cycle = compute_ohci_iso_ctx_cycle_count((cycle_time >> 12) & 0x0000ffff);
1083
1084	if (s->direction == AMDTP_IN_STREAM) {
1085		// NOTE: The AMDTP packet descriptor should be for the past isochronous cycle since
1086		// it corresponds to arrived isochronous packet.
1087		if (compare_ohci_cycle_count(latest_cycle, curr_cycle) > 0)
1088			goto end;
1089		cycle_gap = decrement_ohci_cycle_count(curr_cycle, latest_cycle);
1090
1091		// NOTE: estimate delay by recent history of arrived AMDTP packets. The estimated
1092		// value expectedly corresponds to a few packets (0-2) since the packet arrived at
1093		// the most recent isochronous cycle has been already processed.
1094		for (i = 0; i < cycle_gap; ++i) {
1095			desc = amdtp_stream_next_packet_desc(s, desc);
1096			data_block_count += desc->data_blocks;
1097		}
1098	} else {
1099		// NOTE: The AMDTP packet descriptor should be for the future isochronous cycle
1100		// since it was already scheduled.
1101		if (compare_ohci_cycle_count(latest_cycle, curr_cycle) < 0)
1102			goto end;
1103		cycle_gap = decrement_ohci_cycle_count(latest_cycle, curr_cycle);
1104
1105		// NOTE: use history of scheduled packets.
1106		for (i = 0; i < cycle_gap; ++i) {
1107			data_block_count += desc->data_blocks;
1108			desc = prev_packet_desc(s, desc);
1109		}
1110	}
1111end:
1112	return data_block_count * s->pcm_frame_multiplier;
1113}
1114
1115static void process_ctx_payloads(struct amdtp_stream *s,
1116				 const struct pkt_desc *desc,
1117				 unsigned int count)
1118{
1119	struct snd_pcm_substream *pcm;
1120	int i;
1121
1122	pcm = READ_ONCE(s->pcm);
1123	s->process_ctx_payloads(s, desc, count, pcm);
1124
1125	if (pcm) {
1126		unsigned int data_block_count = 0;
1127
1128		pcm->runtime->delay = compute_pcm_extra_delay(s, desc, count);
1129
1130		for (i = 0; i < count; ++i) {
1131			data_block_count += desc->data_blocks;
1132			desc = amdtp_stream_next_packet_desc(s, desc);
1133		}
1134
1135		update_pcm_pointers(s, pcm, data_block_count * s->pcm_frame_multiplier);
1136	}
1137}
1138
1139static void process_rx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1140			       void *header, void *private_data)
1141{
1142	struct amdtp_stream *s = private_data;
1143	const struct amdtp_domain *d = s->domain;
1144	const __be32 *ctx_header = header;
1145	const unsigned int events_per_period = d->events_per_period;
1146	unsigned int event_count = s->ctx_data.rx.event_count;
1147	struct pkt_desc *desc = s->packet_descs_cursor;
1148	unsigned int pkt_header_length;
1149	unsigned int packets;
1150	u32 curr_cycle_time;
1151	bool need_hw_irq;
1152	int i;
1153
1154	if (s->packet_index < 0)
1155		return;
1156
1157	// Calculate the number of packets in buffer and check XRUN.
1158	packets = header_length / sizeof(*ctx_header);
1159
1160	generate_rx_packet_descs(s, desc, ctx_header, packets);
1161
1162	process_ctx_payloads(s, desc, packets);
1163
1164	if (!(s->flags & CIP_NO_HEADER))
1165		pkt_header_length = IT_PKT_HEADER_SIZE_CIP;
1166	else
1167		pkt_header_length = 0;
1168
1169	if (s == d->irq_target) {
1170		// At NO_PERIOD_WAKEUP mode, the packets for all IT/IR contexts are processed by
1171		// the tasks of user process operating ALSA PCM character device by calling ioctl(2)
1172		// with some requests, instead of scheduled hardware IRQ of an IT context.
1173		struct snd_pcm_substream *pcm = READ_ONCE(s->pcm);
1174		need_hw_irq = !pcm || !pcm->runtime->no_period_wakeup;
1175	} else {
1176		need_hw_irq = false;
1177	}
1178
1179	if (trace_amdtp_packet_enabled())
1180		(void)fw_card_read_cycle_time(fw_parent_device(s->unit)->card, &curr_cycle_time);
1181
1182	for (i = 0; i < packets; ++i) {
1183		struct {
1184			struct fw_iso_packet params;
1185			__be32 header[CIP_HEADER_QUADLETS];
1186		} template = { {0}, {0} };
1187		bool sched_irq = false;
1188
1189		build_it_pkt_header(s, desc->cycle, &template.params, pkt_header_length,
1190				    desc->data_blocks, desc->data_block_counter,
1191				    desc->syt, i, curr_cycle_time);
1192
1193		if (s == s->domain->irq_target) {
1194			event_count += desc->data_blocks;
1195			if (event_count >= events_per_period) {
1196				event_count -= events_per_period;
1197				sched_irq = need_hw_irq;
1198			}
1199		}
1200
1201		if (queue_out_packet(s, &template.params, sched_irq) < 0) {
1202			cancel_stream(s);
1203			return;
1204		}
1205
1206		desc = amdtp_stream_next_packet_desc(s, desc);
1207	}
1208
1209	s->ctx_data.rx.event_count = event_count;
1210	s->packet_descs_cursor = desc;
1211}
1212
1213static void skip_rx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1214			    void *header, void *private_data)
1215{
1216	struct amdtp_stream *s = private_data;
1217	struct amdtp_domain *d = s->domain;
1218	const __be32 *ctx_header = header;
1219	unsigned int packets;
1220	unsigned int cycle;
1221	int i;
1222
1223	if (s->packet_index < 0)
1224		return;
1225
1226	packets = header_length / sizeof(*ctx_header);
1227
1228	cycle = compute_ohci_it_cycle(ctx_header[packets - 1], s->queue_size);
1229	s->next_cycle = increment_ohci_cycle_count(cycle, 1);
1230
1231	for (i = 0; i < packets; ++i) {
1232		struct fw_iso_packet params = {
1233			.header_length = 0,
1234			.payload_length = 0,
1235		};
1236		bool sched_irq = (s == d->irq_target && i == packets - 1);
1237
1238		if (queue_out_packet(s, &params, sched_irq) < 0) {
1239			cancel_stream(s);
1240			return;
1241		}
1242	}
1243}
1244
1245static void irq_target_callback(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1246				void *header, void *private_data);
1247
1248static void process_rx_packets_intermediately(struct fw_iso_context *context, u32 tstamp,
1249					size_t header_length, void *header, void *private_data)
1250{
1251	struct amdtp_stream *s = private_data;
1252	struct amdtp_domain *d = s->domain;
1253	__be32 *ctx_header = header;
1254	const unsigned int queue_size = s->queue_size;
1255	unsigned int packets;
1256	unsigned int offset;
1257
1258	if (s->packet_index < 0)
1259		return;
1260
1261	packets = header_length / sizeof(*ctx_header);
1262
1263	offset = 0;
1264	while (offset < packets) {
1265		unsigned int cycle = compute_ohci_it_cycle(ctx_header[offset], queue_size);
1266
1267		if (compare_ohci_cycle_count(cycle, d->processing_cycle.rx_start) >= 0)
1268			break;
1269
1270		++offset;
1271	}
1272
1273	if (offset > 0) {
1274		unsigned int length = sizeof(*ctx_header) * offset;
1275
1276		skip_rx_packets(context, tstamp, length, ctx_header, private_data);
1277		if (amdtp_streaming_error(s))
1278			return;
1279
1280		ctx_header += offset;
1281		header_length -= length;
1282	}
1283
1284	if (offset < packets) {
1285		s->ready_processing = true;
1286		wake_up(&s->ready_wait);
1287
1288		if (d->replay.enable)
1289			s->ctx_data.rx.cache_pos = 0;
1290
1291		process_rx_packets(context, tstamp, header_length, ctx_header, private_data);
1292		if (amdtp_streaming_error(s))
1293			return;
1294
1295		if (s == d->irq_target)
1296			s->context->callback.sc = irq_target_callback;
1297		else
1298			s->context->callback.sc = process_rx_packets;
1299	}
1300}
1301
1302static void process_tx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1303			       void *header, void *private_data)
1304{
1305	struct amdtp_stream *s = private_data;
1306	__be32 *ctx_header = header;
1307	struct pkt_desc *desc = s->packet_descs_cursor;
1308	unsigned int packet_count;
1309	unsigned int desc_count;
1310	int i;
1311	int err;
1312
1313	if (s->packet_index < 0)
1314		return;
1315
1316	// Calculate the number of packets in buffer and check XRUN.
1317	packet_count = header_length / s->ctx_data.tx.ctx_header_size;
1318
1319	desc_count = 0;
1320	err = generate_tx_packet_descs(s, desc, ctx_header, packet_count, &desc_count);
1321	if (err < 0) {
1322		if (err != -EAGAIN) {
1323			cancel_stream(s);
1324			return;
1325		}
1326	} else {
1327		struct amdtp_domain *d = s->domain;
1328
1329		process_ctx_payloads(s, desc, desc_count);
1330
1331		if (d->replay.enable)
1332			cache_seq(s, desc, desc_count);
1333
1334		for (i = 0; i < desc_count; ++i)
1335			desc = amdtp_stream_next_packet_desc(s, desc);
1336		s->packet_descs_cursor = desc;
1337	}
1338
1339	for (i = 0; i < packet_count; ++i) {
1340		struct fw_iso_packet params = {0};
1341
1342		if (queue_in_packet(s, &params) < 0) {
1343			cancel_stream(s);
1344			return;
1345		}
1346	}
1347}
1348
1349static void drop_tx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1350			    void *header, void *private_data)
1351{
1352	struct amdtp_stream *s = private_data;
1353	const __be32 *ctx_header = header;
1354	unsigned int packets;
1355	unsigned int cycle;
1356	int i;
1357
1358	if (s->packet_index < 0)
1359		return;
1360
1361	packets = header_length / s->ctx_data.tx.ctx_header_size;
1362
1363	ctx_header += (packets - 1) * s->ctx_data.tx.ctx_header_size / sizeof(*ctx_header);
1364	cycle = compute_ohci_cycle_count(ctx_header[1]);
1365	s->next_cycle = increment_ohci_cycle_count(cycle, 1);
1366
1367	for (i = 0; i < packets; ++i) {
1368		struct fw_iso_packet params = {0};
1369
1370		if (queue_in_packet(s, &params) < 0) {
1371			cancel_stream(s);
1372			return;
1373		}
1374	}
1375}
1376
1377static void process_tx_packets_intermediately(struct fw_iso_context *context, u32 tstamp,
1378					size_t header_length, void *header, void *private_data)
1379{
1380	struct amdtp_stream *s = private_data;
1381	struct amdtp_domain *d = s->domain;
1382	__be32 *ctx_header;
1383	unsigned int packets;
1384	unsigned int offset;
1385
1386	if (s->packet_index < 0)
1387		return;
1388
1389	packets = header_length / s->ctx_data.tx.ctx_header_size;
1390
1391	offset = 0;
1392	ctx_header = header;
1393	while (offset < packets) {
1394		unsigned int cycle = compute_ohci_cycle_count(ctx_header[1]);
1395
1396		if (compare_ohci_cycle_count(cycle, d->processing_cycle.tx_start) >= 0)
1397			break;
1398
1399		ctx_header += s->ctx_data.tx.ctx_header_size / sizeof(__be32);
1400		++offset;
1401	}
1402
1403	ctx_header = header;
1404
1405	if (offset > 0) {
1406		size_t length = s->ctx_data.tx.ctx_header_size * offset;
1407
1408		drop_tx_packets(context, tstamp, length, ctx_header, s);
1409		if (amdtp_streaming_error(s))
1410			return;
1411
1412		ctx_header += length / sizeof(*ctx_header);
1413		header_length -= length;
1414	}
1415
1416	if (offset < packets) {
1417		s->ready_processing = true;
1418		wake_up(&s->ready_wait);
1419
1420		process_tx_packets(context, tstamp, header_length, ctx_header, s);
1421		if (amdtp_streaming_error(s))
1422			return;
1423
1424		context->callback.sc = process_tx_packets;
1425	}
1426}
1427
1428static void drop_tx_packets_initially(struct fw_iso_context *context, u32 tstamp,
1429				      size_t header_length, void *header, void *private_data)
1430{
1431	struct amdtp_stream *s = private_data;
1432	struct amdtp_domain *d = s->domain;
1433	__be32 *ctx_header;
1434	unsigned int count;
1435	unsigned int events;
1436	int i;
1437
1438	if (s->packet_index < 0)
1439		return;
1440
1441	count = header_length / s->ctx_data.tx.ctx_header_size;
1442
1443	// Attempt to detect any event in the batch of packets.
1444	events = 0;
1445	ctx_header = header;
1446	for (i = 0; i < count; ++i) {
1447		unsigned int payload_quads =
1448			(be32_to_cpu(*ctx_header) >> ISO_DATA_LENGTH_SHIFT) / sizeof(__be32);
1449		unsigned int data_blocks;
1450
1451		if (s->flags & CIP_NO_HEADER) {
1452			data_blocks = payload_quads / s->data_block_quadlets;
1453		} else {
1454			__be32 *cip_headers = ctx_header + IR_CTX_HEADER_DEFAULT_QUADLETS;
1455
1456			if (payload_quads < CIP_HEADER_QUADLETS) {
1457				data_blocks = 0;
1458			} else {
1459				payload_quads -= CIP_HEADER_QUADLETS;
1460
1461				if (s->flags & CIP_UNAWARE_SYT) {
1462					data_blocks = payload_quads / s->data_block_quadlets;
1463				} else {
1464					u32 cip1 = be32_to_cpu(cip_headers[1]);
1465
1466					// NODATA packet can includes any data blocks but they are
1467					// not available as event.
1468					if ((cip1 & CIP_NO_DATA) == CIP_NO_DATA)
1469						data_blocks = 0;
1470					else
1471						data_blocks = payload_quads / s->data_block_quadlets;
1472				}
1473			}
1474		}
1475
1476		events += data_blocks;
1477
1478		ctx_header += s->ctx_data.tx.ctx_header_size / sizeof(__be32);
1479	}
1480
1481	drop_tx_packets(context, tstamp, header_length, header, s);
1482
1483	if (events > 0)
1484		s->ctx_data.tx.event_starts = true;
1485
1486	// Decide the cycle count to begin processing content of packet in IR contexts.
1487	{
1488		unsigned int stream_count = 0;
1489		unsigned int event_starts_count = 0;
1490		unsigned int cycle = UINT_MAX;
1491
1492		list_for_each_entry(s, &d->streams, list) {
1493			if (s->direction == AMDTP_IN_STREAM) {
1494				++stream_count;
1495				if (s->ctx_data.tx.event_starts)
1496					++event_starts_count;
1497			}
1498		}
1499
1500		if (stream_count == event_starts_count) {
1501			unsigned int next_cycle;
1502
1503			list_for_each_entry(s, &d->streams, list) {
1504				if (s->direction != AMDTP_IN_STREAM)
1505					continue;
1506
1507				next_cycle = increment_ohci_cycle_count(s->next_cycle,
1508								d->processing_cycle.tx_init_skip);
1509				if (cycle == UINT_MAX ||
1510				    compare_ohci_cycle_count(next_cycle, cycle) > 0)
1511					cycle = next_cycle;
1512
1513				s->context->callback.sc = process_tx_packets_intermediately;
1514			}
1515
1516			d->processing_cycle.tx_start = cycle;
1517		}
1518	}
1519}
1520
1521static void process_ctxs_in_domain(struct amdtp_domain *d)
1522{
1523	struct amdtp_stream *s;
1524
1525	list_for_each_entry(s, &d->streams, list) {
1526		if (s != d->irq_target && amdtp_stream_running(s))
1527			fw_iso_context_flush_completions(s->context);
1528
1529		if (amdtp_streaming_error(s))
1530			goto error;
1531	}
1532
1533	return;
1534error:
1535	if (amdtp_stream_running(d->irq_target))
1536		cancel_stream(d->irq_target);
1537
1538	list_for_each_entry(s, &d->streams, list) {
1539		if (amdtp_stream_running(s))
1540			cancel_stream(s);
1541	}
1542}
1543
1544static void irq_target_callback(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1545				void *header, void *private_data)
1546{
1547	struct amdtp_stream *s = private_data;
1548	struct amdtp_domain *d = s->domain;
1549
1550	process_rx_packets(context, tstamp, header_length, header, private_data);
1551	process_ctxs_in_domain(d);
1552}
1553
1554static void irq_target_callback_intermediately(struct fw_iso_context *context, u32 tstamp,
1555					size_t header_length, void *header, void *private_data)
1556{
1557	struct amdtp_stream *s = private_data;
1558	struct amdtp_domain *d = s->domain;
1559
1560	process_rx_packets_intermediately(context, tstamp, header_length, header, private_data);
1561	process_ctxs_in_domain(d);
1562}
1563
1564static void irq_target_callback_skip(struct fw_iso_context *context, u32 tstamp,
1565				     size_t header_length, void *header, void *private_data)
1566{
1567	struct amdtp_stream *s = private_data;
1568	struct amdtp_domain *d = s->domain;
1569	bool ready_to_start;
1570
1571	skip_rx_packets(context, tstamp, header_length, header, private_data);
1572	process_ctxs_in_domain(d);
1573
1574	if (d->replay.enable && !d->replay.on_the_fly) {
1575		unsigned int rx_count = 0;
1576		unsigned int rx_ready_count = 0;
1577		struct amdtp_stream *rx;
1578
1579		list_for_each_entry(rx, &d->streams, list) {
1580			struct amdtp_stream *tx;
1581			unsigned int cached_cycles;
1582
1583			if (rx->direction != AMDTP_OUT_STREAM)
1584				continue;
1585			++rx_count;
1586
1587			tx = rx->ctx_data.rx.replay_target;
1588			cached_cycles = calculate_cached_cycle_count(tx, 0);
1589			if (cached_cycles > tx->ctx_data.tx.cache.size / 2)
1590				++rx_ready_count;
1591		}
1592
1593		ready_to_start = (rx_count == rx_ready_count);
1594	} else {
1595		ready_to_start = true;
1596	}
1597
1598	// Decide the cycle count to begin processing content of packet in IT contexts. All of IT
1599	// contexts are expected to start and get callback when reaching here.
1600	if (ready_to_start) {
1601		unsigned int cycle = s->next_cycle;
1602		list_for_each_entry(s, &d->streams, list) {
1603			if (s->direction != AMDTP_OUT_STREAM)
1604				continue;
1605
1606			if (compare_ohci_cycle_count(s->next_cycle, cycle) > 0)
1607				cycle = s->next_cycle;
1608
1609			if (s == d->irq_target)
1610				s->context->callback.sc = irq_target_callback_intermediately;
1611			else
1612				s->context->callback.sc = process_rx_packets_intermediately;
1613		}
1614
1615		d->processing_cycle.rx_start = cycle;
1616	}
1617}
1618
1619// This is executed one time. For in-stream, first packet has come. For out-stream, prepared to
1620// transmit first packet.
1621static void amdtp_stream_first_callback(struct fw_iso_context *context,
1622					u32 tstamp, size_t header_length,
1623					void *header, void *private_data)
1624{
1625	struct amdtp_stream *s = private_data;
1626	struct amdtp_domain *d = s->domain;
1627
1628	if (s->direction == AMDTP_IN_STREAM) {
1629		context->callback.sc = drop_tx_packets_initially;
1630	} else {
1631		if (s == d->irq_target)
1632			context->callback.sc = irq_target_callback_skip;
1633		else
1634			context->callback.sc = skip_rx_packets;
1635	}
1636
1637	context->callback.sc(context, tstamp, header_length, header, s);
1638}
1639
1640/**
1641 * amdtp_stream_start - start transferring packets
1642 * @s: the AMDTP stream to start
1643 * @channel: the isochronous channel on the bus
1644 * @speed: firewire speed code
1645 * @queue_size: The number of packets in the queue.
1646 * @idle_irq_interval: the interval to queue packet during initial state.
1647 *
1648 * The stream cannot be started until it has been configured with
1649 * amdtp_stream_set_parameters() and it must be started before any PCM or MIDI
1650 * device can be started.
1651 */
1652static int amdtp_stream_start(struct amdtp_stream *s, int channel, int speed,
1653			      unsigned int queue_size, unsigned int idle_irq_interval)
1654{
1655	bool is_irq_target = (s == s->domain->irq_target);
1656	unsigned int ctx_header_size;
1657	unsigned int max_ctx_payload_size;
1658	enum dma_data_direction dir;
1659	struct pkt_desc *descs;
1660	int i, type, tag, err;
1661
1662	mutex_lock(&s->mutex);
1663
1664	if (WARN_ON(amdtp_stream_running(s) ||
1665		    (s->data_block_quadlets < 1))) {
1666		err = -EBADFD;
1667		goto err_unlock;
1668	}
1669
1670	if (s->direction == AMDTP_IN_STREAM) {
1671		// NOTE: IT context should be used for constant IRQ.
1672		if (is_irq_target) {
1673			err = -EINVAL;
1674			goto err_unlock;
1675		}
1676
1677		s->data_block_counter = UINT_MAX;
1678	} else {
1679		s->data_block_counter = 0;
1680	}
1681
1682	// initialize packet buffer.
1683	if (s->direction == AMDTP_IN_STREAM) {
1684		dir = DMA_FROM_DEVICE;
1685		type = FW_ISO_CONTEXT_RECEIVE;
1686		if (!(s->flags & CIP_NO_HEADER))
1687			ctx_header_size = IR_CTX_HEADER_SIZE_CIP;
1688		else
1689			ctx_header_size = IR_CTX_HEADER_SIZE_NO_CIP;
1690	} else {
1691		dir = DMA_TO_DEVICE;
1692		type = FW_ISO_CONTEXT_TRANSMIT;
1693		ctx_header_size = 0;	// No effect for IT context.
1694	}
1695	max_ctx_payload_size = amdtp_stream_get_max_ctx_payload_size(s);
1696
1697	err = iso_packets_buffer_init(&s->buffer, s->unit, queue_size, max_ctx_payload_size, dir);
1698	if (err < 0)
1699		goto err_unlock;
1700	s->queue_size = queue_size;
1701
1702	s->context = fw_iso_context_create(fw_parent_device(s->unit)->card,
1703					  type, channel, speed, ctx_header_size,
1704					  amdtp_stream_first_callback, s);
1705	if (IS_ERR(s->context)) {
1706		err = PTR_ERR(s->context);
1707		if (err == -EBUSY)
1708			dev_err(&s->unit->device,
1709				"no free stream on this controller\n");
1710		goto err_buffer;
1711	}
1712
1713	amdtp_stream_update(s);
1714
1715	if (s->direction == AMDTP_IN_STREAM) {
1716		s->ctx_data.tx.max_ctx_payload_length = max_ctx_payload_size;
1717		s->ctx_data.tx.ctx_header_size = ctx_header_size;
1718		s->ctx_data.tx.event_starts = false;
1719
1720		if (s->domain->replay.enable) {
1721			// struct fw_iso_context.drop_overflow_headers is false therefore it's
1722			// possible to cache much unexpectedly.
1723			s->ctx_data.tx.cache.size = max_t(unsigned int, s->syt_interval * 2,
1724							  queue_size * 3 / 2);
1725			s->ctx_data.tx.cache.pos = 0;
1726			s->ctx_data.tx.cache.descs = kcalloc(s->ctx_data.tx.cache.size,
1727						sizeof(*s->ctx_data.tx.cache.descs), GFP_KERNEL);
1728			if (!s->ctx_data.tx.cache.descs) {
1729				err = -ENOMEM;
1730				goto err_context;
1731			}
1732		}
1733	} else {
1734		static const struct {
1735			unsigned int data_block;
1736			unsigned int syt_offset;
1737		} *entry, initial_state[] = {
1738			[CIP_SFC_32000]  = {  4, 3072 },
1739			[CIP_SFC_48000]  = {  6, 1024 },
1740			[CIP_SFC_96000]  = { 12, 1024 },
1741			[CIP_SFC_192000] = { 24, 1024 },
1742			[CIP_SFC_44100]  = {  0,   67 },
1743			[CIP_SFC_88200]  = {  0,   67 },
1744			[CIP_SFC_176400] = {  0,   67 },
1745		};
1746
1747		s->ctx_data.rx.seq.descs = kcalloc(queue_size, sizeof(*s->ctx_data.rx.seq.descs), GFP_KERNEL);
1748		if (!s->ctx_data.rx.seq.descs) {
1749			err = -ENOMEM;
1750			goto err_context;
1751		}
1752		s->ctx_data.rx.seq.size = queue_size;
1753		s->ctx_data.rx.seq.pos = 0;
1754
1755		entry = &initial_state[s->sfc];
1756		s->ctx_data.rx.data_block_state = entry->data_block;
1757		s->ctx_data.rx.syt_offset_state = entry->syt_offset;
1758		s->ctx_data.rx.last_syt_offset = TICKS_PER_CYCLE;
1759
1760		s->ctx_data.rx.event_count = 0;
1761	}
1762
1763	if (s->flags & CIP_NO_HEADER)
1764		s->tag = TAG_NO_CIP_HEADER;
1765	else
1766		s->tag = TAG_CIP;
1767
1768	// NOTE: When operating without hardIRQ/softIRQ, applications tends to call ioctl request
1769	// for runtime of PCM substream in the interval equivalent to the size of PCM buffer. It
1770	// could take a round over queue of AMDTP packet descriptors and small loss of history. For
1771	// safe, keep more 8 elements for the queue, equivalent to 1 ms.
1772	descs = kcalloc(s->queue_size + 8, sizeof(*descs), GFP_KERNEL);
1773	if (!descs) {
1774		err = -ENOMEM;
1775		goto err_context;
1776	}
1777	s->packet_descs = descs;
1778
1779	INIT_LIST_HEAD(&s->packet_descs_list);
1780	for (i = 0; i < s->queue_size; ++i) {
1781		INIT_LIST_HEAD(&descs->link);
1782		list_add_tail(&descs->link, &s->packet_descs_list);
1783		++descs;
1784	}
1785	s->packet_descs_cursor = list_first_entry(&s->packet_descs_list, struct pkt_desc, link);
1786
1787	s->packet_index = 0;
1788	do {
1789		struct fw_iso_packet params;
1790
1791		if (s->direction == AMDTP_IN_STREAM) {
1792			err = queue_in_packet(s, &params);
1793		} else {
1794			bool sched_irq = false;
1795
1796			params.header_length = 0;
1797			params.payload_length = 0;
1798
1799			if (is_irq_target) {
1800				sched_irq = !((s->packet_index + 1) %
1801					      idle_irq_interval);
1802			}
1803
1804			err = queue_out_packet(s, &params, sched_irq);
1805		}
1806		if (err < 0)
1807			goto err_pkt_descs;
1808	} while (s->packet_index > 0);
1809
1810	/* NOTE: TAG1 matches CIP. This just affects in stream. */
1811	tag = FW_ISO_CONTEXT_MATCH_TAG1;
1812	if ((s->flags & CIP_EMPTY_WITH_TAG0) || (s->flags & CIP_NO_HEADER))
1813		tag |= FW_ISO_CONTEXT_MATCH_TAG0;
1814
1815	s->ready_processing = false;
1816	err = fw_iso_context_start(s->context, -1, 0, tag);
1817	if (err < 0)
1818		goto err_pkt_descs;
1819
1820	mutex_unlock(&s->mutex);
1821
1822	return 0;
1823err_pkt_descs:
1824	kfree(s->packet_descs);
1825	s->packet_descs = NULL;
1826err_context:
1827	if (s->direction == AMDTP_OUT_STREAM) {
1828		kfree(s->ctx_data.rx.seq.descs);
1829	} else {
1830		if (s->domain->replay.enable)
1831			kfree(s->ctx_data.tx.cache.descs);
1832	}
1833	fw_iso_context_destroy(s->context);
1834	s->context = ERR_PTR(-1);
1835err_buffer:
1836	iso_packets_buffer_destroy(&s->buffer, s->unit);
1837err_unlock:
1838	mutex_unlock(&s->mutex);
1839
1840	return err;
1841}
1842
1843/**
1844 * amdtp_domain_stream_pcm_pointer - get the PCM buffer position
1845 * @d: the AMDTP domain.
1846 * @s: the AMDTP stream that transports the PCM data
1847 *
1848 * Returns the current buffer position, in frames.
1849 */
1850unsigned long amdtp_domain_stream_pcm_pointer(struct amdtp_domain *d,
1851					      struct amdtp_stream *s)
1852{
1853	struct amdtp_stream *irq_target = d->irq_target;
1854
1855	// Process isochronous packets queued till recent isochronous cycle to handle PCM frames.
1856	if (irq_target && amdtp_stream_running(irq_target)) {
1857		// In software IRQ context, the call causes dead-lock to disable the tasklet
1858		// synchronously.
1859		if (!in_softirq())
1860			fw_iso_context_flush_completions(irq_target->context);
1861	}
1862
1863	return READ_ONCE(s->pcm_buffer_pointer);
1864}
1865EXPORT_SYMBOL_GPL(amdtp_domain_stream_pcm_pointer);
1866
1867/**
1868 * amdtp_domain_stream_pcm_ack - acknowledge queued PCM frames
1869 * @d: the AMDTP domain.
1870 * @s: the AMDTP stream that transfers the PCM frames
1871 *
1872 * Returns zero always.
1873 */
1874int amdtp_domain_stream_pcm_ack(struct amdtp_domain *d, struct amdtp_stream *s)
1875{
1876	struct amdtp_stream *irq_target = d->irq_target;
1877
1878	// Process isochronous packets for recent isochronous cycle to handle
1879	// queued PCM frames.
1880	if (irq_target && amdtp_stream_running(irq_target))
1881		fw_iso_context_flush_completions(irq_target->context);
1882
1883	return 0;
1884}
1885EXPORT_SYMBOL_GPL(amdtp_domain_stream_pcm_ack);
1886
1887/**
1888 * amdtp_stream_update - update the stream after a bus reset
1889 * @s: the AMDTP stream
1890 */
1891void amdtp_stream_update(struct amdtp_stream *s)
1892{
1893	/* Precomputing. */
1894	WRITE_ONCE(s->source_node_id_field,
1895                   (fw_parent_device(s->unit)->card->node_id << CIP_SID_SHIFT) & CIP_SID_MASK);
1896}
1897EXPORT_SYMBOL(amdtp_stream_update);
1898
1899/**
1900 * amdtp_stream_stop - stop sending packets
1901 * @s: the AMDTP stream to stop
1902 *
1903 * All PCM and MIDI devices of the stream must be stopped before the stream
1904 * itself can be stopped.
1905 */
1906static void amdtp_stream_stop(struct amdtp_stream *s)
1907{
1908	mutex_lock(&s->mutex);
1909
1910	if (!amdtp_stream_running(s)) {
1911		mutex_unlock(&s->mutex);
1912		return;
1913	}
1914
1915	fw_iso_context_stop(s->context);
1916	fw_iso_context_destroy(s->context);
1917	s->context = ERR_PTR(-1);
1918	iso_packets_buffer_destroy(&s->buffer, s->unit);
1919	kfree(s->packet_descs);
1920	s->packet_descs = NULL;
1921
1922	if (s->direction == AMDTP_OUT_STREAM) {
1923		kfree(s->ctx_data.rx.seq.descs);
1924	} else {
1925		if (s->domain->replay.enable)
1926			kfree(s->ctx_data.tx.cache.descs);
1927	}
1928
1929	mutex_unlock(&s->mutex);
1930}
1931
1932/**
1933 * amdtp_stream_pcm_abort - abort the running PCM device
1934 * @s: the AMDTP stream about to be stopped
1935 *
1936 * If the isochronous stream needs to be stopped asynchronously, call this
1937 * function first to stop the PCM device.
1938 */
1939void amdtp_stream_pcm_abort(struct amdtp_stream *s)
1940{
1941	struct snd_pcm_substream *pcm;
1942
1943	pcm = READ_ONCE(s->pcm);
1944	if (pcm)
1945		snd_pcm_stop_xrun(pcm);
1946}
1947EXPORT_SYMBOL(amdtp_stream_pcm_abort);
1948
1949/**
1950 * amdtp_domain_init - initialize an AMDTP domain structure
1951 * @d: the AMDTP domain to initialize.
1952 */
1953int amdtp_domain_init(struct amdtp_domain *d)
1954{
1955	INIT_LIST_HEAD(&d->streams);
1956
1957	d->events_per_period = 0;
1958
1959	return 0;
1960}
1961EXPORT_SYMBOL_GPL(amdtp_domain_init);
1962
1963/**
1964 * amdtp_domain_destroy - destroy an AMDTP domain structure
1965 * @d: the AMDTP domain to destroy.
1966 */
1967void amdtp_domain_destroy(struct amdtp_domain *d)
1968{
1969	// At present nothing to do.
1970	return;
1971}
1972EXPORT_SYMBOL_GPL(amdtp_domain_destroy);
1973
1974/**
1975 * amdtp_domain_add_stream - register isoc context into the domain.
1976 * @d: the AMDTP domain.
1977 * @s: the AMDTP stream.
1978 * @channel: the isochronous channel on the bus.
1979 * @speed: firewire speed code.
1980 */
1981int amdtp_domain_add_stream(struct amdtp_domain *d, struct amdtp_stream *s,
1982			    int channel, int speed)
1983{
1984	struct amdtp_stream *tmp;
1985
1986	list_for_each_entry(tmp, &d->streams, list) {
1987		if (s == tmp)
1988			return -EBUSY;
1989	}
1990
1991	list_add(&s->list, &d->streams);
1992
1993	s->channel = channel;
1994	s->speed = speed;
1995	s->domain = d;
1996
1997	return 0;
1998}
1999EXPORT_SYMBOL_GPL(amdtp_domain_add_stream);
2000
2001// Make the reference from rx stream to tx stream for sequence replay. When the number of tx streams
2002// is less than the number of rx streams, the first tx stream is selected.
2003static int make_association(struct amdtp_domain *d)
2004{
2005	unsigned int dst_index = 0;
2006	struct amdtp_stream *rx;
2007
2008	// Make association to replay target.
2009	list_for_each_entry(rx, &d->streams, list) {
2010		if (rx->direction == AMDTP_OUT_STREAM) {
2011			unsigned int src_index = 0;
2012			struct amdtp_stream *tx = NULL;
2013			struct amdtp_stream *s;
2014
2015			list_for_each_entry(s, &d->streams, list) {
2016				if (s->direction == AMDTP_IN_STREAM) {
2017					if (dst_index == src_index) {
2018						tx = s;
2019						break;
2020					}
2021
2022					++src_index;
2023				}
2024			}
2025			if (!tx) {
2026				// Select the first entry.
2027				list_for_each_entry(s, &d->streams, list) {
2028					if (s->direction == AMDTP_IN_STREAM) {
2029						tx = s;
2030						break;
2031					}
2032				}
2033				// No target is available to replay sequence.
2034				if (!tx)
2035					return -EINVAL;
2036			}
2037
2038			rx->ctx_data.rx.replay_target = tx;
2039
2040			++dst_index;
2041		}
2042	}
2043
2044	return 0;
2045}
2046
2047/**
2048 * amdtp_domain_start - start sending packets for isoc context in the domain.
2049 * @d: the AMDTP domain.
2050 * @tx_init_skip_cycles: the number of cycles to skip processing packets at initial stage of IR
2051 *			 contexts.
2052 * @replay_seq: whether to replay the sequence of packet in IR context for the sequence of packet in
2053 *		IT context.
2054 * @replay_on_the_fly: transfer rx packets according to nominal frequency, then begin to replay
2055 *		       according to arrival of events in tx packets.
2056 */
2057int amdtp_domain_start(struct amdtp_domain *d, unsigned int tx_init_skip_cycles, bool replay_seq,
2058		       bool replay_on_the_fly)
2059{
2060	unsigned int events_per_buffer = d->events_per_buffer;
2061	unsigned int events_per_period = d->events_per_period;
2062	unsigned int queue_size;
2063	struct amdtp_stream *s;
2064	bool found = false;
2065	int err;
2066
2067	if (replay_seq) {
2068		err = make_association(d);
2069		if (err < 0)
2070			return err;
2071	}
2072	d->replay.enable = replay_seq;
2073	d->replay.on_the_fly = replay_on_the_fly;
2074
2075	// Select an IT context as IRQ target.
2076	list_for_each_entry(s, &d->streams, list) {
2077		if (s->direction == AMDTP_OUT_STREAM) {
2078			found = true;
2079			break;
2080		}
2081	}
2082	if (!found)
2083		return -ENXIO;
2084	d->irq_target = s;
2085
2086	d->processing_cycle.tx_init_skip = tx_init_skip_cycles;
2087
2088	// This is a case that AMDTP streams in domain run just for MIDI
2089	// substream. Use the number of events equivalent to 10 msec as
2090	// interval of hardware IRQ.
2091	if (events_per_period == 0)
2092		events_per_period = amdtp_rate_table[d->irq_target->sfc] / 100;
2093	if (events_per_buffer == 0)
2094		events_per_buffer = events_per_period * 3;
2095
2096	queue_size = DIV_ROUND_UP(CYCLES_PER_SECOND * events_per_buffer,
2097				  amdtp_rate_table[d->irq_target->sfc]);
2098
2099	list_for_each_entry(s, &d->streams, list) {
2100		unsigned int idle_irq_interval = 0;
2101
2102		if (s->direction == AMDTP_OUT_STREAM && s == d->irq_target) {
2103			idle_irq_interval = DIV_ROUND_UP(CYCLES_PER_SECOND * events_per_period,
2104							 amdtp_rate_table[d->irq_target->sfc]);
2105		}
2106
2107		// Starts immediately but actually DMA context starts several hundred cycles later.
2108		err = amdtp_stream_start(s, s->channel, s->speed, queue_size, idle_irq_interval);
2109		if (err < 0)
2110			goto error;
2111	}
2112
2113	return 0;
2114error:
2115	list_for_each_entry(s, &d->streams, list)
2116		amdtp_stream_stop(s);
2117	return err;
2118}
2119EXPORT_SYMBOL_GPL(amdtp_domain_start);
2120
2121/**
2122 * amdtp_domain_stop - stop sending packets for isoc context in the same domain.
2123 * @d: the AMDTP domain to which the isoc contexts belong.
2124 */
2125void amdtp_domain_stop(struct amdtp_domain *d)
2126{
2127	struct amdtp_stream *s, *next;
2128
2129	if (d->irq_target)
2130		amdtp_stream_stop(d->irq_target);
2131
2132	list_for_each_entry_safe(s, next, &d->streams, list) {
2133		list_del(&s->list);
2134
2135		if (s != d->irq_target)
2136			amdtp_stream_stop(s);
2137	}
2138
2139	d->events_per_period = 0;
2140	d->irq_target = NULL;
2141}
2142EXPORT_SYMBOL_GPL(amdtp_domain_stop);
2143