xref: /linux/sound/firewire/amdtp-stream.c (revision 1d717123)
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  */
amdtp_stream_init(struct amdtp_stream * s,struct fw_unit * unit,enum amdtp_stream_direction dir,unsigned int flags,unsigned int fmt,amdtp_stream_process_ctx_payloads_t process_ctx_payloads,unsigned int protocol_size)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  */
amdtp_stream_destroy(struct amdtp_stream * s)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 
apply_constraint_to_size(struct snd_pcm_hw_params * params,struct snd_pcm_hw_rule * rule)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  */
amdtp_stream_add_pcm_hw_constraints(struct amdtp_stream * s,struct snd_pcm_runtime * runtime)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  */
amdtp_stream_set_parameters(struct amdtp_stream * s,unsigned int rate,unsigned int data_block_quadlets,unsigned int pcm_frame_multiplier)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.
amdtp_stream_get_max_ctx_payload_size(struct amdtp_stream * s)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  */
amdtp_stream_get_max_payload(struct amdtp_stream * s)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  */
amdtp_stream_pcm_prepare(struct amdtp_stream * s)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 
pool_blocking_data_blocks(struct amdtp_stream * s,struct seq_desc * descs,unsigned int size,unsigned int pos,unsigned int count)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 
pool_ideal_nonblocking_data_blocks(struct amdtp_stream * s,struct seq_desc * descs,unsigned int size,unsigned int pos,unsigned int count)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 
calculate_syt_offset(unsigned int * last_syt_offset,unsigned int * syt_offset_state,enum cip_sfc sfc)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 
pool_ideal_syt_offsets(struct amdtp_stream * s,struct seq_desc * descs,unsigned int size,unsigned int pos,unsigned int count)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 
compute_syt_offset(unsigned int syt,unsigned int cycle,unsigned int transfer_delay)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.
calculate_cached_cycle_count(struct amdtp_stream * s,unsigned int head)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 
cache_seq(struct amdtp_stream * s,const struct pkt_desc * src,unsigned int desc_count)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 
pool_ideal_seq_descs(struct amdtp_stream * s,struct seq_desc * descs,unsigned int size,unsigned int pos,unsigned int count)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 
pool_replayed_seq(struct amdtp_stream * s,struct seq_desc * descs,unsigned int size,unsigned int pos,unsigned int count)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 
pool_seq_descs(struct amdtp_stream * s,struct seq_desc * descs,unsigned int size,unsigned int pos,unsigned int count)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 
update_pcm_pointers(struct amdtp_stream * s,struct snd_pcm_substream * pcm,unsigned int frames)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 
queue_packet(struct amdtp_stream * s,struct fw_iso_packet * params,bool sched_irq)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 
queue_out_packet(struct amdtp_stream * s,struct fw_iso_packet * params,bool sched_irq)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 
queue_in_packet(struct amdtp_stream * s,struct fw_iso_packet * params)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 
generate_cip_header(struct amdtp_stream * s,__be32 cip_header[2],unsigned int data_block_counter,unsigned int syt)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 
build_it_pkt_header(struct amdtp_stream * s,unsigned int cycle,struct fw_iso_packet * params,unsigned int header_length,unsigned int data_blocks,unsigned int data_block_counter,unsigned int syt,unsigned int index,u32 curr_cycle_time)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 
check_cip_header(struct amdtp_stream * s,const __be32 * buf,unsigned int payload_length,unsigned int * data_blocks,unsigned int * data_block_counter,unsigned int * syt)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 (!(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 
parse_ir_ctx_header(struct amdtp_stream * s,unsigned int cycle,const __be32 * ctx_header,unsigned int * data_blocks,unsigned int * data_block_counter,unsigned int * syt,unsigned int packet_index,unsigned int index,u32 curr_cycle_time)803 static 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.
compute_ohci_iso_ctx_cycle_count(u32 tstamp)861 static inline u32 compute_ohci_iso_ctx_cycle_count(u32 tstamp)
862 {
863 	return (((tstamp >> 13) & 0x07) * CYCLES_PER_SECOND) + (tstamp & 0x1fff);
864 }
865 
compute_ohci_cycle_count(__be32 ctx_header_tstamp)866 static 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 
increment_ohci_cycle_count(u32 cycle,unsigned int addend)872 static 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 
decrement_ohci_cycle_count(u32 minuend,u32 subtrahend)880 static 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 
compare_ohci_cycle_count(u32 lval,u32 rval)888 static 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.
compute_ohci_it_cycle(const __be32 ctx_header_tstamp,unsigned int queue_size)902 static 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 
generate_tx_packet_descs(struct amdtp_stream * s,struct pkt_desc * desc,const __be32 * ctx_header,unsigned int packet_count,unsigned int * desc_count)909 static 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 
compute_syt(unsigned int syt_offset,unsigned int cycle,unsigned int transfer_delay)994 static 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 
generate_rx_packet_descs(struct amdtp_stream * s,struct pkt_desc * desc,const __be32 * ctx_header,unsigned int packet_count)1005 static 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 
cancel_stream(struct amdtp_stream * s)1050 static 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 
compute_pcm_extra_delay(struct amdtp_stream * s,const struct pkt_desc * desc,unsigned int count)1058 static 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 	}
1111 end:
1112 	return data_block_count * s->pcm_frame_multiplier;
1113 }
1114 
process_ctx_payloads(struct amdtp_stream * s,const struct pkt_desc * desc,unsigned int count)1115 static 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 
process_rx_packets(struct fw_iso_context * context,u32 tstamp,size_t header_length,void * header,void * private_data)1139 static 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 		DEFINE_FLEX(struct fw_iso_packet, template, header,
1184 			    header_length, CIP_HEADER_QUADLETS);
1185 		bool sched_irq = false;
1186 
1187 		build_it_pkt_header(s, desc->cycle, template, pkt_header_length,
1188 				    desc->data_blocks, desc->data_block_counter,
1189 				    desc->syt, i, curr_cycle_time);
1190 
1191 		if (s == s->domain->irq_target) {
1192 			event_count += desc->data_blocks;
1193 			if (event_count >= events_per_period) {
1194 				event_count -= events_per_period;
1195 				sched_irq = need_hw_irq;
1196 			}
1197 		}
1198 
1199 		if (queue_out_packet(s, template, sched_irq) < 0) {
1200 			cancel_stream(s);
1201 			return;
1202 		}
1203 
1204 		desc = amdtp_stream_next_packet_desc(s, desc);
1205 	}
1206 
1207 	s->ctx_data.rx.event_count = event_count;
1208 	s->packet_descs_cursor = desc;
1209 }
1210 
skip_rx_packets(struct fw_iso_context * context,u32 tstamp,size_t header_length,void * header,void * private_data)1211 static void skip_rx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1212 			    void *header, void *private_data)
1213 {
1214 	struct amdtp_stream *s = private_data;
1215 	struct amdtp_domain *d = s->domain;
1216 	const __be32 *ctx_header = header;
1217 	unsigned int packets;
1218 	unsigned int cycle;
1219 	int i;
1220 
1221 	if (s->packet_index < 0)
1222 		return;
1223 
1224 	packets = header_length / sizeof(*ctx_header);
1225 
1226 	cycle = compute_ohci_it_cycle(ctx_header[packets - 1], s->queue_size);
1227 	s->next_cycle = increment_ohci_cycle_count(cycle, 1);
1228 
1229 	for (i = 0; i < packets; ++i) {
1230 		struct fw_iso_packet params = {
1231 			.header_length = 0,
1232 			.payload_length = 0,
1233 		};
1234 		bool sched_irq = (s == d->irq_target && i == packets - 1);
1235 
1236 		if (queue_out_packet(s, &params, sched_irq) < 0) {
1237 			cancel_stream(s);
1238 			return;
1239 		}
1240 	}
1241 }
1242 
1243 static void irq_target_callback(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1244 				void *header, void *private_data);
1245 
process_rx_packets_intermediately(struct fw_iso_context * context,u32 tstamp,size_t header_length,void * header,void * private_data)1246 static void process_rx_packets_intermediately(struct fw_iso_context *context, u32 tstamp,
1247 					size_t header_length, void *header, void *private_data)
1248 {
1249 	struct amdtp_stream *s = private_data;
1250 	struct amdtp_domain *d = s->domain;
1251 	__be32 *ctx_header = header;
1252 	const unsigned int queue_size = s->queue_size;
1253 	unsigned int packets;
1254 	unsigned int offset;
1255 
1256 	if (s->packet_index < 0)
1257 		return;
1258 
1259 	packets = header_length / sizeof(*ctx_header);
1260 
1261 	offset = 0;
1262 	while (offset < packets) {
1263 		unsigned int cycle = compute_ohci_it_cycle(ctx_header[offset], queue_size);
1264 
1265 		if (compare_ohci_cycle_count(cycle, d->processing_cycle.rx_start) >= 0)
1266 			break;
1267 
1268 		++offset;
1269 	}
1270 
1271 	if (offset > 0) {
1272 		unsigned int length = sizeof(*ctx_header) * offset;
1273 
1274 		skip_rx_packets(context, tstamp, length, ctx_header, private_data);
1275 		if (amdtp_streaming_error(s))
1276 			return;
1277 
1278 		ctx_header += offset;
1279 		header_length -= length;
1280 	}
1281 
1282 	if (offset < packets) {
1283 		s->ready_processing = true;
1284 		wake_up(&s->ready_wait);
1285 
1286 		if (d->replay.enable)
1287 			s->ctx_data.rx.cache_pos = 0;
1288 
1289 		process_rx_packets(context, tstamp, header_length, ctx_header, private_data);
1290 		if (amdtp_streaming_error(s))
1291 			return;
1292 
1293 		if (s == d->irq_target)
1294 			s->context->callback.sc = irq_target_callback;
1295 		else
1296 			s->context->callback.sc = process_rx_packets;
1297 	}
1298 }
1299 
process_tx_packets(struct fw_iso_context * context,u32 tstamp,size_t header_length,void * header,void * private_data)1300 static void process_tx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1301 			       void *header, void *private_data)
1302 {
1303 	struct amdtp_stream *s = private_data;
1304 	__be32 *ctx_header = header;
1305 	struct pkt_desc *desc = s->packet_descs_cursor;
1306 	unsigned int packet_count;
1307 	unsigned int desc_count;
1308 	int i;
1309 	int err;
1310 
1311 	if (s->packet_index < 0)
1312 		return;
1313 
1314 	// Calculate the number of packets in buffer and check XRUN.
1315 	packet_count = header_length / s->ctx_data.tx.ctx_header_size;
1316 
1317 	desc_count = 0;
1318 	err = generate_tx_packet_descs(s, desc, ctx_header, packet_count, &desc_count);
1319 	if (err < 0) {
1320 		if (err != -EAGAIN) {
1321 			cancel_stream(s);
1322 			return;
1323 		}
1324 	} else {
1325 		struct amdtp_domain *d = s->domain;
1326 
1327 		process_ctx_payloads(s, desc, desc_count);
1328 
1329 		if (d->replay.enable)
1330 			cache_seq(s, desc, desc_count);
1331 
1332 		for (i = 0; i < desc_count; ++i)
1333 			desc = amdtp_stream_next_packet_desc(s, desc);
1334 		s->packet_descs_cursor = desc;
1335 	}
1336 
1337 	for (i = 0; i < packet_count; ++i) {
1338 		struct fw_iso_packet params = {0};
1339 
1340 		if (queue_in_packet(s, &params) < 0) {
1341 			cancel_stream(s);
1342 			return;
1343 		}
1344 	}
1345 }
1346 
drop_tx_packets(struct fw_iso_context * context,u32 tstamp,size_t header_length,void * header,void * private_data)1347 static void drop_tx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1348 			    void *header, void *private_data)
1349 {
1350 	struct amdtp_stream *s = private_data;
1351 	const __be32 *ctx_header = header;
1352 	unsigned int packets;
1353 	unsigned int cycle;
1354 	int i;
1355 
1356 	if (s->packet_index < 0)
1357 		return;
1358 
1359 	packets = header_length / s->ctx_data.tx.ctx_header_size;
1360 
1361 	ctx_header += (packets - 1) * s->ctx_data.tx.ctx_header_size / sizeof(*ctx_header);
1362 	cycle = compute_ohci_cycle_count(ctx_header[1]);
1363 	s->next_cycle = increment_ohci_cycle_count(cycle, 1);
1364 
1365 	for (i = 0; i < packets; ++i) {
1366 		struct fw_iso_packet params = {0};
1367 
1368 		if (queue_in_packet(s, &params) < 0) {
1369 			cancel_stream(s);
1370 			return;
1371 		}
1372 	}
1373 }
1374 
process_tx_packets_intermediately(struct fw_iso_context * context,u32 tstamp,size_t header_length,void * header,void * private_data)1375 static void process_tx_packets_intermediately(struct fw_iso_context *context, u32 tstamp,
1376 					size_t header_length, void *header, void *private_data)
1377 {
1378 	struct amdtp_stream *s = private_data;
1379 	struct amdtp_domain *d = s->domain;
1380 	__be32 *ctx_header;
1381 	unsigned int packets;
1382 	unsigned int offset;
1383 
1384 	if (s->packet_index < 0)
1385 		return;
1386 
1387 	packets = header_length / s->ctx_data.tx.ctx_header_size;
1388 
1389 	offset = 0;
1390 	ctx_header = header;
1391 	while (offset < packets) {
1392 		unsigned int cycle = compute_ohci_cycle_count(ctx_header[1]);
1393 
1394 		if (compare_ohci_cycle_count(cycle, d->processing_cycle.tx_start) >= 0)
1395 			break;
1396 
1397 		ctx_header += s->ctx_data.tx.ctx_header_size / sizeof(__be32);
1398 		++offset;
1399 	}
1400 
1401 	ctx_header = header;
1402 
1403 	if (offset > 0) {
1404 		size_t length = s->ctx_data.tx.ctx_header_size * offset;
1405 
1406 		drop_tx_packets(context, tstamp, length, ctx_header, s);
1407 		if (amdtp_streaming_error(s))
1408 			return;
1409 
1410 		ctx_header += length / sizeof(*ctx_header);
1411 		header_length -= length;
1412 	}
1413 
1414 	if (offset < packets) {
1415 		s->ready_processing = true;
1416 		wake_up(&s->ready_wait);
1417 
1418 		process_tx_packets(context, tstamp, header_length, ctx_header, s);
1419 		if (amdtp_streaming_error(s))
1420 			return;
1421 
1422 		context->callback.sc = process_tx_packets;
1423 	}
1424 }
1425 
drop_tx_packets_initially(struct fw_iso_context * context,u32 tstamp,size_t header_length,void * header,void * private_data)1426 static void drop_tx_packets_initially(struct fw_iso_context *context, u32 tstamp,
1427 				      size_t header_length, void *header, void *private_data)
1428 {
1429 	struct amdtp_stream *s = private_data;
1430 	struct amdtp_domain *d = s->domain;
1431 	__be32 *ctx_header;
1432 	unsigned int count;
1433 	unsigned int events;
1434 	int i;
1435 
1436 	if (s->packet_index < 0)
1437 		return;
1438 
1439 	count = header_length / s->ctx_data.tx.ctx_header_size;
1440 
1441 	// Attempt to detect any event in the batch of packets.
1442 	events = 0;
1443 	ctx_header = header;
1444 	for (i = 0; i < count; ++i) {
1445 		unsigned int payload_quads =
1446 			(be32_to_cpu(*ctx_header) >> ISO_DATA_LENGTH_SHIFT) / sizeof(__be32);
1447 		unsigned int data_blocks;
1448 
1449 		if (s->flags & CIP_NO_HEADER) {
1450 			data_blocks = payload_quads / s->data_block_quadlets;
1451 		} else {
1452 			__be32 *cip_headers = ctx_header + IR_CTX_HEADER_DEFAULT_QUADLETS;
1453 
1454 			if (payload_quads < CIP_HEADER_QUADLETS) {
1455 				data_blocks = 0;
1456 			} else {
1457 				payload_quads -= CIP_HEADER_QUADLETS;
1458 
1459 				if (s->flags & CIP_UNAWARE_SYT) {
1460 					data_blocks = payload_quads / s->data_block_quadlets;
1461 				} else {
1462 					u32 cip1 = be32_to_cpu(cip_headers[1]);
1463 
1464 					// NODATA packet can includes any data blocks but they are
1465 					// not available as event.
1466 					if ((cip1 & CIP_NO_DATA) == CIP_NO_DATA)
1467 						data_blocks = 0;
1468 					else
1469 						data_blocks = payload_quads / s->data_block_quadlets;
1470 				}
1471 			}
1472 		}
1473 
1474 		events += data_blocks;
1475 
1476 		ctx_header += s->ctx_data.tx.ctx_header_size / sizeof(__be32);
1477 	}
1478 
1479 	drop_tx_packets(context, tstamp, header_length, header, s);
1480 
1481 	if (events > 0)
1482 		s->ctx_data.tx.event_starts = true;
1483 
1484 	// Decide the cycle count to begin processing content of packet in IR contexts.
1485 	{
1486 		unsigned int stream_count = 0;
1487 		unsigned int event_starts_count = 0;
1488 		unsigned int cycle = UINT_MAX;
1489 
1490 		list_for_each_entry(s, &d->streams, list) {
1491 			if (s->direction == AMDTP_IN_STREAM) {
1492 				++stream_count;
1493 				if (s->ctx_data.tx.event_starts)
1494 					++event_starts_count;
1495 			}
1496 		}
1497 
1498 		if (stream_count == event_starts_count) {
1499 			unsigned int next_cycle;
1500 
1501 			list_for_each_entry(s, &d->streams, list) {
1502 				if (s->direction != AMDTP_IN_STREAM)
1503 					continue;
1504 
1505 				next_cycle = increment_ohci_cycle_count(s->next_cycle,
1506 								d->processing_cycle.tx_init_skip);
1507 				if (cycle == UINT_MAX ||
1508 				    compare_ohci_cycle_count(next_cycle, cycle) > 0)
1509 					cycle = next_cycle;
1510 
1511 				s->context->callback.sc = process_tx_packets_intermediately;
1512 			}
1513 
1514 			d->processing_cycle.tx_start = cycle;
1515 		}
1516 	}
1517 }
1518 
process_ctxs_in_domain(struct amdtp_domain * d)1519 static void process_ctxs_in_domain(struct amdtp_domain *d)
1520 {
1521 	struct amdtp_stream *s;
1522 
1523 	list_for_each_entry(s, &d->streams, list) {
1524 		if (s != d->irq_target && amdtp_stream_running(s))
1525 			fw_iso_context_flush_completions(s->context);
1526 
1527 		if (amdtp_streaming_error(s))
1528 			goto error;
1529 	}
1530 
1531 	return;
1532 error:
1533 	if (amdtp_stream_running(d->irq_target))
1534 		cancel_stream(d->irq_target);
1535 
1536 	list_for_each_entry(s, &d->streams, list) {
1537 		if (amdtp_stream_running(s))
1538 			cancel_stream(s);
1539 	}
1540 }
1541 
irq_target_callback(struct fw_iso_context * context,u32 tstamp,size_t header_length,void * header,void * private_data)1542 static void irq_target_callback(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1543 				void *header, void *private_data)
1544 {
1545 	struct amdtp_stream *s = private_data;
1546 	struct amdtp_domain *d = s->domain;
1547 
1548 	process_rx_packets(context, tstamp, header_length, header, private_data);
1549 	process_ctxs_in_domain(d);
1550 }
1551 
irq_target_callback_intermediately(struct fw_iso_context * context,u32 tstamp,size_t header_length,void * header,void * private_data)1552 static void irq_target_callback_intermediately(struct fw_iso_context *context, u32 tstamp,
1553 					size_t header_length, void *header, void *private_data)
1554 {
1555 	struct amdtp_stream *s = private_data;
1556 	struct amdtp_domain *d = s->domain;
1557 
1558 	process_rx_packets_intermediately(context, tstamp, header_length, header, private_data);
1559 	process_ctxs_in_domain(d);
1560 }
1561 
irq_target_callback_skip(struct fw_iso_context * context,u32 tstamp,size_t header_length,void * header,void * private_data)1562 static void irq_target_callback_skip(struct fw_iso_context *context, u32 tstamp,
1563 				     size_t header_length, void *header, void *private_data)
1564 {
1565 	struct amdtp_stream *s = private_data;
1566 	struct amdtp_domain *d = s->domain;
1567 	bool ready_to_start;
1568 
1569 	skip_rx_packets(context, tstamp, header_length, header, private_data);
1570 	process_ctxs_in_domain(d);
1571 
1572 	if (d->replay.enable && !d->replay.on_the_fly) {
1573 		unsigned int rx_count = 0;
1574 		unsigned int rx_ready_count = 0;
1575 		struct amdtp_stream *rx;
1576 
1577 		list_for_each_entry(rx, &d->streams, list) {
1578 			struct amdtp_stream *tx;
1579 			unsigned int cached_cycles;
1580 
1581 			if (rx->direction != AMDTP_OUT_STREAM)
1582 				continue;
1583 			++rx_count;
1584 
1585 			tx = rx->ctx_data.rx.replay_target;
1586 			cached_cycles = calculate_cached_cycle_count(tx, 0);
1587 			if (cached_cycles > tx->ctx_data.tx.cache.size / 2)
1588 				++rx_ready_count;
1589 		}
1590 
1591 		ready_to_start = (rx_count == rx_ready_count);
1592 	} else {
1593 		ready_to_start = true;
1594 	}
1595 
1596 	// Decide the cycle count to begin processing content of packet in IT contexts. All of IT
1597 	// contexts are expected to start and get callback when reaching here.
1598 	if (ready_to_start) {
1599 		unsigned int cycle = s->next_cycle;
1600 		list_for_each_entry(s, &d->streams, list) {
1601 			if (s->direction != AMDTP_OUT_STREAM)
1602 				continue;
1603 
1604 			if (compare_ohci_cycle_count(s->next_cycle, cycle) > 0)
1605 				cycle = s->next_cycle;
1606 
1607 			if (s == d->irq_target)
1608 				s->context->callback.sc = irq_target_callback_intermediately;
1609 			else
1610 				s->context->callback.sc = process_rx_packets_intermediately;
1611 		}
1612 
1613 		d->processing_cycle.rx_start = cycle;
1614 	}
1615 }
1616 
1617 // This is executed one time. For in-stream, first packet has come. For out-stream, prepared to
1618 // transmit first packet.
amdtp_stream_first_callback(struct fw_iso_context * context,u32 tstamp,size_t header_length,void * header,void * private_data)1619 static void amdtp_stream_first_callback(struct fw_iso_context *context,
1620 					u32 tstamp, size_t header_length,
1621 					void *header, void *private_data)
1622 {
1623 	struct amdtp_stream *s = private_data;
1624 	struct amdtp_domain *d = s->domain;
1625 
1626 	if (s->direction == AMDTP_IN_STREAM) {
1627 		context->callback.sc = drop_tx_packets_initially;
1628 	} else {
1629 		if (s == d->irq_target)
1630 			context->callback.sc = irq_target_callback_skip;
1631 		else
1632 			context->callback.sc = skip_rx_packets;
1633 	}
1634 
1635 	context->callback.sc(context, tstamp, header_length, header, s);
1636 }
1637 
1638 /**
1639  * amdtp_stream_start - start transferring packets
1640  * @s: the AMDTP stream to start
1641  * @channel: the isochronous channel on the bus
1642  * @speed: firewire speed code
1643  * @queue_size: The number of packets in the queue.
1644  * @idle_irq_interval: the interval to queue packet during initial state.
1645  *
1646  * The stream cannot be started until it has been configured with
1647  * amdtp_stream_set_parameters() and it must be started before any PCM or MIDI
1648  * device can be started.
1649  */
amdtp_stream_start(struct amdtp_stream * s,int channel,int speed,unsigned int queue_size,unsigned int idle_irq_interval)1650 static int amdtp_stream_start(struct amdtp_stream *s, int channel, int speed,
1651 			      unsigned int queue_size, unsigned int idle_irq_interval)
1652 {
1653 	bool is_irq_target = (s == s->domain->irq_target);
1654 	unsigned int ctx_header_size;
1655 	unsigned int max_ctx_payload_size;
1656 	enum dma_data_direction dir;
1657 	struct pkt_desc *descs;
1658 	int i, type, tag, err;
1659 
1660 	mutex_lock(&s->mutex);
1661 
1662 	if (WARN_ON(amdtp_stream_running(s) ||
1663 		    (s->data_block_quadlets < 1))) {
1664 		err = -EBADFD;
1665 		goto err_unlock;
1666 	}
1667 
1668 	if (s->direction == AMDTP_IN_STREAM) {
1669 		// NOTE: IT context should be used for constant IRQ.
1670 		if (is_irq_target) {
1671 			err = -EINVAL;
1672 			goto err_unlock;
1673 		}
1674 
1675 		s->data_block_counter = UINT_MAX;
1676 	} else {
1677 		s->data_block_counter = 0;
1678 	}
1679 
1680 	// initialize packet buffer.
1681 	if (s->direction == AMDTP_IN_STREAM) {
1682 		dir = DMA_FROM_DEVICE;
1683 		type = FW_ISO_CONTEXT_RECEIVE;
1684 		if (!(s->flags & CIP_NO_HEADER))
1685 			ctx_header_size = IR_CTX_HEADER_SIZE_CIP;
1686 		else
1687 			ctx_header_size = IR_CTX_HEADER_SIZE_NO_CIP;
1688 	} else {
1689 		dir = DMA_TO_DEVICE;
1690 		type = FW_ISO_CONTEXT_TRANSMIT;
1691 		ctx_header_size = 0;	// No effect for IT context.
1692 	}
1693 	max_ctx_payload_size = amdtp_stream_get_max_ctx_payload_size(s);
1694 
1695 	err = iso_packets_buffer_init(&s->buffer, s->unit, queue_size, max_ctx_payload_size, dir);
1696 	if (err < 0)
1697 		goto err_unlock;
1698 	s->queue_size = queue_size;
1699 
1700 	s->context = fw_iso_context_create(fw_parent_device(s->unit)->card,
1701 					  type, channel, speed, ctx_header_size,
1702 					  amdtp_stream_first_callback, s);
1703 	if (IS_ERR(s->context)) {
1704 		err = PTR_ERR(s->context);
1705 		if (err == -EBUSY)
1706 			dev_err(&s->unit->device,
1707 				"no free stream on this controller\n");
1708 		goto err_buffer;
1709 	}
1710 
1711 	amdtp_stream_update(s);
1712 
1713 	if (s->direction == AMDTP_IN_STREAM) {
1714 		s->ctx_data.tx.max_ctx_payload_length = max_ctx_payload_size;
1715 		s->ctx_data.tx.ctx_header_size = ctx_header_size;
1716 		s->ctx_data.tx.event_starts = false;
1717 
1718 		if (s->domain->replay.enable) {
1719 			// struct fw_iso_context.drop_overflow_headers is false therefore it's
1720 			// possible to cache much unexpectedly.
1721 			s->ctx_data.tx.cache.size = max_t(unsigned int, s->syt_interval * 2,
1722 							  queue_size * 3 / 2);
1723 			s->ctx_data.tx.cache.pos = 0;
1724 			s->ctx_data.tx.cache.descs = kcalloc(s->ctx_data.tx.cache.size,
1725 						sizeof(*s->ctx_data.tx.cache.descs), GFP_KERNEL);
1726 			if (!s->ctx_data.tx.cache.descs) {
1727 				err = -ENOMEM;
1728 				goto err_context;
1729 			}
1730 		}
1731 	} else {
1732 		static const struct {
1733 			unsigned int data_block;
1734 			unsigned int syt_offset;
1735 		} *entry, initial_state[] = {
1736 			[CIP_SFC_32000]  = {  4, 3072 },
1737 			[CIP_SFC_48000]  = {  6, 1024 },
1738 			[CIP_SFC_96000]  = { 12, 1024 },
1739 			[CIP_SFC_192000] = { 24, 1024 },
1740 			[CIP_SFC_44100]  = {  0,   67 },
1741 			[CIP_SFC_88200]  = {  0,   67 },
1742 			[CIP_SFC_176400] = {  0,   67 },
1743 		};
1744 
1745 		s->ctx_data.rx.seq.descs = kcalloc(queue_size, sizeof(*s->ctx_data.rx.seq.descs), GFP_KERNEL);
1746 		if (!s->ctx_data.rx.seq.descs) {
1747 			err = -ENOMEM;
1748 			goto err_context;
1749 		}
1750 		s->ctx_data.rx.seq.size = queue_size;
1751 		s->ctx_data.rx.seq.pos = 0;
1752 
1753 		entry = &initial_state[s->sfc];
1754 		s->ctx_data.rx.data_block_state = entry->data_block;
1755 		s->ctx_data.rx.syt_offset_state = entry->syt_offset;
1756 		s->ctx_data.rx.last_syt_offset = TICKS_PER_CYCLE;
1757 
1758 		s->ctx_data.rx.event_count = 0;
1759 	}
1760 
1761 	if (s->flags & CIP_NO_HEADER)
1762 		s->tag = TAG_NO_CIP_HEADER;
1763 	else
1764 		s->tag = TAG_CIP;
1765 
1766 	// NOTE: When operating without hardIRQ/softIRQ, applications tends to call ioctl request
1767 	// for runtime of PCM substream in the interval equivalent to the size of PCM buffer. It
1768 	// could take a round over queue of AMDTP packet descriptors and small loss of history. For
1769 	// safe, keep more 8 elements for the queue, equivalent to 1 ms.
1770 	descs = kcalloc(s->queue_size + 8, sizeof(*descs), GFP_KERNEL);
1771 	if (!descs) {
1772 		err = -ENOMEM;
1773 		goto err_context;
1774 	}
1775 	s->packet_descs = descs;
1776 
1777 	INIT_LIST_HEAD(&s->packet_descs_list);
1778 	for (i = 0; i < s->queue_size; ++i) {
1779 		INIT_LIST_HEAD(&descs->link);
1780 		list_add_tail(&descs->link, &s->packet_descs_list);
1781 		++descs;
1782 	}
1783 	s->packet_descs_cursor = list_first_entry(&s->packet_descs_list, struct pkt_desc, link);
1784 
1785 	s->packet_index = 0;
1786 	do {
1787 		struct fw_iso_packet params;
1788 
1789 		if (s->direction == AMDTP_IN_STREAM) {
1790 			err = queue_in_packet(s, &params);
1791 		} else {
1792 			bool sched_irq = false;
1793 
1794 			params.header_length = 0;
1795 			params.payload_length = 0;
1796 
1797 			if (is_irq_target) {
1798 				sched_irq = !((s->packet_index + 1) %
1799 					      idle_irq_interval);
1800 			}
1801 
1802 			err = queue_out_packet(s, &params, sched_irq);
1803 		}
1804 		if (err < 0)
1805 			goto err_pkt_descs;
1806 	} while (s->packet_index > 0);
1807 
1808 	/* NOTE: TAG1 matches CIP. This just affects in stream. */
1809 	tag = FW_ISO_CONTEXT_MATCH_TAG1;
1810 	if ((s->flags & CIP_EMPTY_WITH_TAG0) || (s->flags & CIP_NO_HEADER))
1811 		tag |= FW_ISO_CONTEXT_MATCH_TAG0;
1812 
1813 	s->ready_processing = false;
1814 	err = fw_iso_context_start(s->context, -1, 0, tag);
1815 	if (err < 0)
1816 		goto err_pkt_descs;
1817 
1818 	mutex_unlock(&s->mutex);
1819 
1820 	return 0;
1821 err_pkt_descs:
1822 	kfree(s->packet_descs);
1823 	s->packet_descs = NULL;
1824 err_context:
1825 	if (s->direction == AMDTP_OUT_STREAM) {
1826 		kfree(s->ctx_data.rx.seq.descs);
1827 	} else {
1828 		if (s->domain->replay.enable)
1829 			kfree(s->ctx_data.tx.cache.descs);
1830 	}
1831 	fw_iso_context_destroy(s->context);
1832 	s->context = ERR_PTR(-1);
1833 err_buffer:
1834 	iso_packets_buffer_destroy(&s->buffer, s->unit);
1835 err_unlock:
1836 	mutex_unlock(&s->mutex);
1837 
1838 	return err;
1839 }
1840 
1841 /**
1842  * amdtp_domain_stream_pcm_pointer - get the PCM buffer position
1843  * @d: the AMDTP domain.
1844  * @s: the AMDTP stream that transports the PCM data
1845  *
1846  * Returns the current buffer position, in frames.
1847  */
amdtp_domain_stream_pcm_pointer(struct amdtp_domain * d,struct amdtp_stream * s)1848 unsigned long amdtp_domain_stream_pcm_pointer(struct amdtp_domain *d,
1849 					      struct amdtp_stream *s)
1850 {
1851 	struct amdtp_stream *irq_target = d->irq_target;
1852 
1853 	// Process isochronous packets queued till recent isochronous cycle to handle PCM frames.
1854 	if (irq_target && amdtp_stream_running(irq_target)) {
1855 		// In software IRQ context, the call causes dead-lock to disable the tasklet
1856 		// synchronously.
1857 		if (!in_softirq())
1858 			fw_iso_context_flush_completions(irq_target->context);
1859 	}
1860 
1861 	return READ_ONCE(s->pcm_buffer_pointer);
1862 }
1863 EXPORT_SYMBOL_GPL(amdtp_domain_stream_pcm_pointer);
1864 
1865 /**
1866  * amdtp_domain_stream_pcm_ack - acknowledge queued PCM frames
1867  * @d: the AMDTP domain.
1868  * @s: the AMDTP stream that transfers the PCM frames
1869  *
1870  * Returns zero always.
1871  */
amdtp_domain_stream_pcm_ack(struct amdtp_domain * d,struct amdtp_stream * s)1872 int amdtp_domain_stream_pcm_ack(struct amdtp_domain *d, struct amdtp_stream *s)
1873 {
1874 	struct amdtp_stream *irq_target = d->irq_target;
1875 
1876 	// Process isochronous packets for recent isochronous cycle to handle
1877 	// queued PCM frames.
1878 	if (irq_target && amdtp_stream_running(irq_target))
1879 		fw_iso_context_flush_completions(irq_target->context);
1880 
1881 	return 0;
1882 }
1883 EXPORT_SYMBOL_GPL(amdtp_domain_stream_pcm_ack);
1884 
1885 /**
1886  * amdtp_stream_update - update the stream after a bus reset
1887  * @s: the AMDTP stream
1888  */
amdtp_stream_update(struct amdtp_stream * s)1889 void amdtp_stream_update(struct amdtp_stream *s)
1890 {
1891 	/* Precomputing. */
1892 	WRITE_ONCE(s->source_node_id_field,
1893                    (fw_parent_device(s->unit)->card->node_id << CIP_SID_SHIFT) & CIP_SID_MASK);
1894 }
1895 EXPORT_SYMBOL(amdtp_stream_update);
1896 
1897 /**
1898  * amdtp_stream_stop - stop sending packets
1899  * @s: the AMDTP stream to stop
1900  *
1901  * All PCM and MIDI devices of the stream must be stopped before the stream
1902  * itself can be stopped.
1903  */
amdtp_stream_stop(struct amdtp_stream * s)1904 static void amdtp_stream_stop(struct amdtp_stream *s)
1905 {
1906 	mutex_lock(&s->mutex);
1907 
1908 	if (!amdtp_stream_running(s)) {
1909 		mutex_unlock(&s->mutex);
1910 		return;
1911 	}
1912 
1913 	fw_iso_context_stop(s->context);
1914 	fw_iso_context_destroy(s->context);
1915 	s->context = ERR_PTR(-1);
1916 	iso_packets_buffer_destroy(&s->buffer, s->unit);
1917 	kfree(s->packet_descs);
1918 	s->packet_descs = NULL;
1919 
1920 	if (s->direction == AMDTP_OUT_STREAM) {
1921 		kfree(s->ctx_data.rx.seq.descs);
1922 	} else {
1923 		if (s->domain->replay.enable)
1924 			kfree(s->ctx_data.tx.cache.descs);
1925 	}
1926 
1927 	mutex_unlock(&s->mutex);
1928 }
1929 
1930 /**
1931  * amdtp_stream_pcm_abort - abort the running PCM device
1932  * @s: the AMDTP stream about to be stopped
1933  *
1934  * If the isochronous stream needs to be stopped asynchronously, call this
1935  * function first to stop the PCM device.
1936  */
amdtp_stream_pcm_abort(struct amdtp_stream * s)1937 void amdtp_stream_pcm_abort(struct amdtp_stream *s)
1938 {
1939 	struct snd_pcm_substream *pcm;
1940 
1941 	pcm = READ_ONCE(s->pcm);
1942 	if (pcm)
1943 		snd_pcm_stop_xrun(pcm);
1944 }
1945 EXPORT_SYMBOL(amdtp_stream_pcm_abort);
1946 
1947 /**
1948  * amdtp_domain_init - initialize an AMDTP domain structure
1949  * @d: the AMDTP domain to initialize.
1950  */
amdtp_domain_init(struct amdtp_domain * d)1951 int amdtp_domain_init(struct amdtp_domain *d)
1952 {
1953 	INIT_LIST_HEAD(&d->streams);
1954 
1955 	d->events_per_period = 0;
1956 
1957 	return 0;
1958 }
1959 EXPORT_SYMBOL_GPL(amdtp_domain_init);
1960 
1961 /**
1962  * amdtp_domain_destroy - destroy an AMDTP domain structure
1963  * @d: the AMDTP domain to destroy.
1964  */
amdtp_domain_destroy(struct amdtp_domain * d)1965 void amdtp_domain_destroy(struct amdtp_domain *d)
1966 {
1967 	// At present nothing to do.
1968 	return;
1969 }
1970 EXPORT_SYMBOL_GPL(amdtp_domain_destroy);
1971 
1972 /**
1973  * amdtp_domain_add_stream - register isoc context into the domain.
1974  * @d: the AMDTP domain.
1975  * @s: the AMDTP stream.
1976  * @channel: the isochronous channel on the bus.
1977  * @speed: firewire speed code.
1978  */
amdtp_domain_add_stream(struct amdtp_domain * d,struct amdtp_stream * s,int channel,int speed)1979 int amdtp_domain_add_stream(struct amdtp_domain *d, struct amdtp_stream *s,
1980 			    int channel, int speed)
1981 {
1982 	struct amdtp_stream *tmp;
1983 
1984 	list_for_each_entry(tmp, &d->streams, list) {
1985 		if (s == tmp)
1986 			return -EBUSY;
1987 	}
1988 
1989 	list_add(&s->list, &d->streams);
1990 
1991 	s->channel = channel;
1992 	s->speed = speed;
1993 	s->domain = d;
1994 
1995 	return 0;
1996 }
1997 EXPORT_SYMBOL_GPL(amdtp_domain_add_stream);
1998 
1999 // Make the reference from rx stream to tx stream for sequence replay. When the number of tx streams
2000 // is less than the number of rx streams, the first tx stream is selected.
make_association(struct amdtp_domain * d)2001 static int make_association(struct amdtp_domain *d)
2002 {
2003 	unsigned int dst_index = 0;
2004 	struct amdtp_stream *rx;
2005 
2006 	// Make association to replay target.
2007 	list_for_each_entry(rx, &d->streams, list) {
2008 		if (rx->direction == AMDTP_OUT_STREAM) {
2009 			unsigned int src_index = 0;
2010 			struct amdtp_stream *tx = NULL;
2011 			struct amdtp_stream *s;
2012 
2013 			list_for_each_entry(s, &d->streams, list) {
2014 				if (s->direction == AMDTP_IN_STREAM) {
2015 					if (dst_index == src_index) {
2016 						tx = s;
2017 						break;
2018 					}
2019 
2020 					++src_index;
2021 				}
2022 			}
2023 			if (!tx) {
2024 				// Select the first entry.
2025 				list_for_each_entry(s, &d->streams, list) {
2026 					if (s->direction == AMDTP_IN_STREAM) {
2027 						tx = s;
2028 						break;
2029 					}
2030 				}
2031 				// No target is available to replay sequence.
2032 				if (!tx)
2033 					return -EINVAL;
2034 			}
2035 
2036 			rx->ctx_data.rx.replay_target = tx;
2037 
2038 			++dst_index;
2039 		}
2040 	}
2041 
2042 	return 0;
2043 }
2044 
2045 /**
2046  * amdtp_domain_start - start sending packets for isoc context in the domain.
2047  * @d: the AMDTP domain.
2048  * @tx_init_skip_cycles: the number of cycles to skip processing packets at initial stage of IR
2049  *			 contexts.
2050  * @replay_seq: whether to replay the sequence of packet in IR context for the sequence of packet in
2051  *		IT context.
2052  * @replay_on_the_fly: transfer rx packets according to nominal frequency, then begin to replay
2053  *		       according to arrival of events in tx packets.
2054  */
amdtp_domain_start(struct amdtp_domain * d,unsigned int tx_init_skip_cycles,bool replay_seq,bool replay_on_the_fly)2055 int amdtp_domain_start(struct amdtp_domain *d, unsigned int tx_init_skip_cycles, bool replay_seq,
2056 		       bool replay_on_the_fly)
2057 {
2058 	unsigned int events_per_buffer = d->events_per_buffer;
2059 	unsigned int events_per_period = d->events_per_period;
2060 	unsigned int queue_size;
2061 	struct amdtp_stream *s;
2062 	bool found = false;
2063 	int err;
2064 
2065 	if (replay_seq) {
2066 		err = make_association(d);
2067 		if (err < 0)
2068 			return err;
2069 	}
2070 	d->replay.enable = replay_seq;
2071 	d->replay.on_the_fly = replay_on_the_fly;
2072 
2073 	// Select an IT context as IRQ target.
2074 	list_for_each_entry(s, &d->streams, list) {
2075 		if (s->direction == AMDTP_OUT_STREAM) {
2076 			found = true;
2077 			break;
2078 		}
2079 	}
2080 	if (!found)
2081 		return -ENXIO;
2082 	d->irq_target = s;
2083 
2084 	d->processing_cycle.tx_init_skip = tx_init_skip_cycles;
2085 
2086 	// This is a case that AMDTP streams in domain run just for MIDI
2087 	// substream. Use the number of events equivalent to 10 msec as
2088 	// interval of hardware IRQ.
2089 	if (events_per_period == 0)
2090 		events_per_period = amdtp_rate_table[d->irq_target->sfc] / 100;
2091 	if (events_per_buffer == 0)
2092 		events_per_buffer = events_per_period * 3;
2093 
2094 	queue_size = DIV_ROUND_UP(CYCLES_PER_SECOND * events_per_buffer,
2095 				  amdtp_rate_table[d->irq_target->sfc]);
2096 
2097 	list_for_each_entry(s, &d->streams, list) {
2098 		unsigned int idle_irq_interval = 0;
2099 
2100 		if (s->direction == AMDTP_OUT_STREAM && s == d->irq_target) {
2101 			idle_irq_interval = DIV_ROUND_UP(CYCLES_PER_SECOND * events_per_period,
2102 							 amdtp_rate_table[d->irq_target->sfc]);
2103 		}
2104 
2105 		// Starts immediately but actually DMA context starts several hundred cycles later.
2106 		err = amdtp_stream_start(s, s->channel, s->speed, queue_size, idle_irq_interval);
2107 		if (err < 0)
2108 			goto error;
2109 	}
2110 
2111 	return 0;
2112 error:
2113 	list_for_each_entry(s, &d->streams, list)
2114 		amdtp_stream_stop(s);
2115 	return err;
2116 }
2117 EXPORT_SYMBOL_GPL(amdtp_domain_start);
2118 
2119 /**
2120  * amdtp_domain_stop - stop sending packets for isoc context in the same domain.
2121  * @d: the AMDTP domain to which the isoc contexts belong.
2122  */
amdtp_domain_stop(struct amdtp_domain * d)2123 void amdtp_domain_stop(struct amdtp_domain *d)
2124 {
2125 	struct amdtp_stream *s, *next;
2126 
2127 	if (d->irq_target)
2128 		amdtp_stream_stop(d->irq_target);
2129 
2130 	list_for_each_entry_safe(s, next, &d->streams, list) {
2131 		list_del(&s->list);
2132 
2133 		if (s != d->irq_target)
2134 			amdtp_stream_stop(s);
2135 	}
2136 
2137 	d->events_per_period = 0;
2138 	d->irq_target = NULL;
2139 }
2140 EXPORT_SYMBOL_GPL(amdtp_domain_stop);
2141