xref: /openbmc/linux/sound/firewire/amdtp-stream.c (revision 82df5b73)
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_MAX_SECOND		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 header parameters */
37 #define CIP_EOH_SHIFT		31
38 #define CIP_EOH			(1u << CIP_EOH_SHIFT)
39 #define CIP_EOH_MASK		0x80000000
40 #define CIP_SID_SHIFT		24
41 #define CIP_SID_MASK		0x3f000000
42 #define CIP_DBS_MASK		0x00ff0000
43 #define CIP_DBS_SHIFT		16
44 #define CIP_SPH_MASK		0x00000400
45 #define CIP_SPH_SHIFT		10
46 #define CIP_DBC_MASK		0x000000ff
47 #define CIP_FMT_SHIFT		24
48 #define CIP_FMT_MASK		0x3f000000
49 #define CIP_FDF_MASK		0x00ff0000
50 #define CIP_FDF_SHIFT		16
51 #define CIP_SYT_MASK		0x0000ffff
52 #define CIP_SYT_NO_INFO		0xffff
53 
54 /* Audio and Music transfer protocol specific parameters */
55 #define CIP_FMT_AM		0x10
56 #define AMDTP_FDF_NO_DATA	0xff
57 
58 // For iso header, tstamp and 2 CIP header.
59 #define IR_CTX_HEADER_SIZE_CIP		16
60 // For iso header and tstamp.
61 #define IR_CTX_HEADER_SIZE_NO_CIP	8
62 #define HEADER_TSTAMP_MASK	0x0000ffff
63 
64 #define IT_PKT_HEADER_SIZE_CIP		8 // For 2 CIP header.
65 #define IT_PKT_HEADER_SIZE_NO_CIP	0 // Nothing.
66 
67 static void pcm_period_tasklet(unsigned long data);
68 
69 /**
70  * amdtp_stream_init - initialize an AMDTP stream structure
71  * @s: the AMDTP stream to initialize
72  * @unit: the target of the stream
73  * @dir: the direction of stream
74  * @flags: the packet transmission method to use
75  * @fmt: the value of fmt field in CIP header
76  * @process_ctx_payloads: callback handler to process payloads of isoc context
77  * @protocol_size: the size to allocate newly for protocol
78  */
79 int amdtp_stream_init(struct amdtp_stream *s, struct fw_unit *unit,
80 		      enum amdtp_stream_direction dir, enum cip_flags flags,
81 		      unsigned int fmt,
82 		      amdtp_stream_process_ctx_payloads_t process_ctx_payloads,
83 		      unsigned int protocol_size)
84 {
85 	if (process_ctx_payloads == NULL)
86 		return -EINVAL;
87 
88 	s->protocol = kzalloc(protocol_size, GFP_KERNEL);
89 	if (!s->protocol)
90 		return -ENOMEM;
91 
92 	s->unit = unit;
93 	s->direction = dir;
94 	s->flags = flags;
95 	s->context = ERR_PTR(-1);
96 	mutex_init(&s->mutex);
97 	tasklet_init(&s->period_tasklet, pcm_period_tasklet, (unsigned long)s);
98 	s->packet_index = 0;
99 
100 	init_waitqueue_head(&s->callback_wait);
101 	s->callbacked = false;
102 
103 	s->fmt = fmt;
104 	s->process_ctx_payloads = process_ctx_payloads;
105 
106 	if (dir == AMDTP_OUT_STREAM)
107 		s->ctx_data.rx.syt_override = -1;
108 
109 	return 0;
110 }
111 EXPORT_SYMBOL(amdtp_stream_init);
112 
113 /**
114  * amdtp_stream_destroy - free stream resources
115  * @s: the AMDTP stream to destroy
116  */
117 void amdtp_stream_destroy(struct amdtp_stream *s)
118 {
119 	/* Not initialized. */
120 	if (s->protocol == NULL)
121 		return;
122 
123 	WARN_ON(amdtp_stream_running(s));
124 	kfree(s->protocol);
125 	mutex_destroy(&s->mutex);
126 }
127 EXPORT_SYMBOL(amdtp_stream_destroy);
128 
129 const unsigned int amdtp_syt_intervals[CIP_SFC_COUNT] = {
130 	[CIP_SFC_32000]  =  8,
131 	[CIP_SFC_44100]  =  8,
132 	[CIP_SFC_48000]  =  8,
133 	[CIP_SFC_88200]  = 16,
134 	[CIP_SFC_96000]  = 16,
135 	[CIP_SFC_176400] = 32,
136 	[CIP_SFC_192000] = 32,
137 };
138 EXPORT_SYMBOL(amdtp_syt_intervals);
139 
140 const unsigned int amdtp_rate_table[CIP_SFC_COUNT] = {
141 	[CIP_SFC_32000]  =  32000,
142 	[CIP_SFC_44100]  =  44100,
143 	[CIP_SFC_48000]  =  48000,
144 	[CIP_SFC_88200]  =  88200,
145 	[CIP_SFC_96000]  =  96000,
146 	[CIP_SFC_176400] = 176400,
147 	[CIP_SFC_192000] = 192000,
148 };
149 EXPORT_SYMBOL(amdtp_rate_table);
150 
151 static int apply_constraint_to_size(struct snd_pcm_hw_params *params,
152 				    struct snd_pcm_hw_rule *rule)
153 {
154 	struct snd_interval *s = hw_param_interval(params, rule->var);
155 	const struct snd_interval *r =
156 		hw_param_interval_c(params, SNDRV_PCM_HW_PARAM_RATE);
157 	struct snd_interval t = {0};
158 	unsigned int step = 0;
159 	int i;
160 
161 	for (i = 0; i < CIP_SFC_COUNT; ++i) {
162 		if (snd_interval_test(r, amdtp_rate_table[i]))
163 			step = max(step, amdtp_syt_intervals[i]);
164 	}
165 
166 	t.min = roundup(s->min, step);
167 	t.max = rounddown(s->max, step);
168 	t.integer = 1;
169 
170 	return snd_interval_refine(s, &t);
171 }
172 
173 /**
174  * amdtp_stream_add_pcm_hw_constraints - add hw constraints for PCM substream
175  * @s:		the AMDTP stream, which must be initialized.
176  * @runtime:	the PCM substream runtime
177  */
178 int amdtp_stream_add_pcm_hw_constraints(struct amdtp_stream *s,
179 					struct snd_pcm_runtime *runtime)
180 {
181 	struct snd_pcm_hardware *hw = &runtime->hw;
182 	unsigned int ctx_header_size;
183 	unsigned int maximum_usec_per_period;
184 	int err;
185 
186 	hw->info = SNDRV_PCM_INFO_BATCH |
187 		   SNDRV_PCM_INFO_BLOCK_TRANSFER |
188 		   SNDRV_PCM_INFO_INTERLEAVED |
189 		   SNDRV_PCM_INFO_JOINT_DUPLEX |
190 		   SNDRV_PCM_INFO_MMAP |
191 		   SNDRV_PCM_INFO_MMAP_VALID;
192 
193 	/* SNDRV_PCM_INFO_BATCH */
194 	hw->periods_min = 2;
195 	hw->periods_max = UINT_MAX;
196 
197 	/* bytes for a frame */
198 	hw->period_bytes_min = 4 * hw->channels_max;
199 
200 	/* Just to prevent from allocating much pages. */
201 	hw->period_bytes_max = hw->period_bytes_min * 2048;
202 	hw->buffer_bytes_max = hw->period_bytes_max * hw->periods_min;
203 
204 	// Linux driver for 1394 OHCI controller voluntarily flushes isoc
205 	// context when total size of accumulated context header reaches
206 	// PAGE_SIZE. This kicks tasklet for the isoc context and brings
207 	// callback in the middle of scheduled interrupts.
208 	// Although AMDTP streams in the same domain use the same events per
209 	// IRQ, use the largest size of context header between IT/IR contexts.
210 	// Here, use the value of context header in IR context is for both
211 	// contexts.
212 	if (!(s->flags & CIP_NO_HEADER))
213 		ctx_header_size = IR_CTX_HEADER_SIZE_CIP;
214 	else
215 		ctx_header_size = IR_CTX_HEADER_SIZE_NO_CIP;
216 	maximum_usec_per_period = USEC_PER_SEC * PAGE_SIZE /
217 				  CYCLES_PER_SECOND / ctx_header_size;
218 
219 	// In IEC 61883-6, one isoc packet can transfer events up to the value
220 	// of syt interval. This comes from the interval of isoc cycle. As 1394
221 	// OHCI controller can generate hardware IRQ per isoc packet, the
222 	// interval is 125 usec.
223 	// However, there are two ways of transmission in IEC 61883-6; blocking
224 	// and non-blocking modes. In blocking mode, the sequence of isoc packet
225 	// includes 'empty' or 'NODATA' packets which include no event. In
226 	// non-blocking mode, the number of events per packet is variable up to
227 	// the syt interval.
228 	// Due to the above protocol design, the minimum PCM frames per
229 	// interrupt should be double of the value of syt interval, thus it is
230 	// 250 usec.
231 	err = snd_pcm_hw_constraint_minmax(runtime,
232 					   SNDRV_PCM_HW_PARAM_PERIOD_TIME,
233 					   250, maximum_usec_per_period);
234 	if (err < 0)
235 		goto end;
236 
237 	/* Non-Blocking stream has no more constraints */
238 	if (!(s->flags & CIP_BLOCKING))
239 		goto end;
240 
241 	/*
242 	 * One AMDTP packet can include some frames. In blocking mode, the
243 	 * number equals to SYT_INTERVAL. So the number is 8, 16 or 32,
244 	 * depending on its sampling rate. For accurate period interrupt, it's
245 	 * preferrable to align period/buffer sizes to current SYT_INTERVAL.
246 	 */
247 	err = snd_pcm_hw_rule_add(runtime, 0, SNDRV_PCM_HW_PARAM_PERIOD_SIZE,
248 				  apply_constraint_to_size, NULL,
249 				  SNDRV_PCM_HW_PARAM_PERIOD_SIZE,
250 				  SNDRV_PCM_HW_PARAM_RATE, -1);
251 	if (err < 0)
252 		goto end;
253 	err = snd_pcm_hw_rule_add(runtime, 0, SNDRV_PCM_HW_PARAM_BUFFER_SIZE,
254 				  apply_constraint_to_size, NULL,
255 				  SNDRV_PCM_HW_PARAM_BUFFER_SIZE,
256 				  SNDRV_PCM_HW_PARAM_RATE, -1);
257 	if (err < 0)
258 		goto end;
259 end:
260 	return err;
261 }
262 EXPORT_SYMBOL(amdtp_stream_add_pcm_hw_constraints);
263 
264 /**
265  * amdtp_stream_set_parameters - set stream parameters
266  * @s: the AMDTP stream to configure
267  * @rate: the sample rate
268  * @data_block_quadlets: the size of a data block in quadlet unit
269  *
270  * The parameters must be set before the stream is started, and must not be
271  * changed while the stream is running.
272  */
273 int amdtp_stream_set_parameters(struct amdtp_stream *s, unsigned int rate,
274 				unsigned int data_block_quadlets)
275 {
276 	unsigned int sfc;
277 
278 	for (sfc = 0; sfc < ARRAY_SIZE(amdtp_rate_table); ++sfc) {
279 		if (amdtp_rate_table[sfc] == rate)
280 			break;
281 	}
282 	if (sfc == ARRAY_SIZE(amdtp_rate_table))
283 		return -EINVAL;
284 
285 	s->sfc = sfc;
286 	s->data_block_quadlets = data_block_quadlets;
287 	s->syt_interval = amdtp_syt_intervals[sfc];
288 
289 	// default buffering in the device.
290 	if (s->direction == AMDTP_OUT_STREAM) {
291 		s->ctx_data.rx.transfer_delay =
292 					TRANSFER_DELAY_TICKS - TICKS_PER_CYCLE;
293 
294 		if (s->flags & CIP_BLOCKING) {
295 			// additional buffering needed to adjust for no-data
296 			// packets.
297 			s->ctx_data.rx.transfer_delay +=
298 				TICKS_PER_SECOND * s->syt_interval / rate;
299 		}
300 	}
301 
302 	return 0;
303 }
304 EXPORT_SYMBOL(amdtp_stream_set_parameters);
305 
306 /**
307  * amdtp_stream_get_max_payload - get the stream's packet size
308  * @s: the AMDTP stream
309  *
310  * This function must not be called before the stream has been configured
311  * with amdtp_stream_set_parameters().
312  */
313 unsigned int amdtp_stream_get_max_payload(struct amdtp_stream *s)
314 {
315 	unsigned int multiplier = 1;
316 	unsigned int cip_header_size = 0;
317 
318 	if (s->flags & CIP_JUMBO_PAYLOAD)
319 		multiplier = 5;
320 	if (!(s->flags & CIP_NO_HEADER))
321 		cip_header_size = sizeof(__be32) * 2;
322 
323 	return cip_header_size +
324 		s->syt_interval * s->data_block_quadlets * sizeof(__be32) * multiplier;
325 }
326 EXPORT_SYMBOL(amdtp_stream_get_max_payload);
327 
328 /**
329  * amdtp_stream_pcm_prepare - prepare PCM device for running
330  * @s: the AMDTP stream
331  *
332  * This function should be called from the PCM device's .prepare callback.
333  */
334 void amdtp_stream_pcm_prepare(struct amdtp_stream *s)
335 {
336 	tasklet_kill(&s->period_tasklet);
337 	s->pcm_buffer_pointer = 0;
338 	s->pcm_period_pointer = 0;
339 }
340 EXPORT_SYMBOL(amdtp_stream_pcm_prepare);
341 
342 static unsigned int calculate_data_blocks(unsigned int *data_block_state,
343 				bool is_blocking, bool is_no_info,
344 				unsigned int syt_interval, enum cip_sfc sfc)
345 {
346 	unsigned int data_blocks;
347 
348 	/* Blocking mode. */
349 	if (is_blocking) {
350 		/* This module generate empty packet for 'no data'. */
351 		if (is_no_info)
352 			data_blocks = 0;
353 		else
354 			data_blocks = syt_interval;
355 	/* Non-blocking mode. */
356 	} else {
357 		if (!cip_sfc_is_base_44100(sfc)) {
358 			// Sample_rate / 8000 is an integer, and precomputed.
359 			data_blocks = *data_block_state;
360 		} else {
361 			unsigned int phase = *data_block_state;
362 
363 		/*
364 		 * This calculates the number of data blocks per packet so that
365 		 * 1) the overall rate is correct and exactly synchronized to
366 		 *    the bus clock, and
367 		 * 2) packets with a rounded-up number of blocks occur as early
368 		 *    as possible in the sequence (to prevent underruns of the
369 		 *    device's buffer).
370 		 */
371 			if (sfc == CIP_SFC_44100)
372 				/* 6 6 5 6 5 6 5 ... */
373 				data_blocks = 5 + ((phase & 1) ^
374 						   (phase == 0 || phase >= 40));
375 			else
376 				/* 12 11 11 11 11 ... or 23 22 22 22 22 ... */
377 				data_blocks = 11 * (sfc >> 1) + (phase == 0);
378 			if (++phase >= (80 >> (sfc >> 1)))
379 				phase = 0;
380 			*data_block_state = phase;
381 		}
382 	}
383 
384 	return data_blocks;
385 }
386 
387 static unsigned int calculate_syt_offset(unsigned int *last_syt_offset,
388 			unsigned int *syt_offset_state, enum cip_sfc sfc)
389 {
390 	unsigned int syt_offset;
391 
392 	if (*last_syt_offset < TICKS_PER_CYCLE) {
393 		if (!cip_sfc_is_base_44100(sfc))
394 			syt_offset = *last_syt_offset + *syt_offset_state;
395 		else {
396 		/*
397 		 * The time, in ticks, of the n'th SYT_INTERVAL sample is:
398 		 *   n * SYT_INTERVAL * 24576000 / sample_rate
399 		 * Modulo TICKS_PER_CYCLE, the difference between successive
400 		 * elements is about 1386.23.  Rounding the results of this
401 		 * formula to the SYT precision results in a sequence of
402 		 * differences that begins with:
403 		 *   1386 1386 1387 1386 1386 1386 1387 1386 1386 1386 1387 ...
404 		 * This code generates _exactly_ the same sequence.
405 		 */
406 			unsigned int phase = *syt_offset_state;
407 			unsigned int index = phase % 13;
408 
409 			syt_offset = *last_syt_offset;
410 			syt_offset += 1386 + ((index && !(index & 3)) ||
411 					      phase == 146);
412 			if (++phase >= 147)
413 				phase = 0;
414 			*syt_offset_state = phase;
415 		}
416 	} else
417 		syt_offset = *last_syt_offset - TICKS_PER_CYCLE;
418 	*last_syt_offset = syt_offset;
419 
420 	if (syt_offset >= TICKS_PER_CYCLE)
421 		syt_offset = CIP_SYT_NO_INFO;
422 
423 	return syt_offset;
424 }
425 
426 static void update_pcm_pointers(struct amdtp_stream *s,
427 				struct snd_pcm_substream *pcm,
428 				unsigned int frames)
429 {
430 	unsigned int ptr;
431 
432 	ptr = s->pcm_buffer_pointer + frames;
433 	if (ptr >= pcm->runtime->buffer_size)
434 		ptr -= pcm->runtime->buffer_size;
435 	WRITE_ONCE(s->pcm_buffer_pointer, ptr);
436 
437 	s->pcm_period_pointer += frames;
438 	if (s->pcm_period_pointer >= pcm->runtime->period_size) {
439 		s->pcm_period_pointer -= pcm->runtime->period_size;
440 		tasklet_hi_schedule(&s->period_tasklet);
441 	}
442 }
443 
444 static void pcm_period_tasklet(unsigned long data)
445 {
446 	struct amdtp_stream *s = (void *)data;
447 	struct snd_pcm_substream *pcm = READ_ONCE(s->pcm);
448 
449 	if (pcm)
450 		snd_pcm_period_elapsed(pcm);
451 }
452 
453 static int queue_packet(struct amdtp_stream *s, struct fw_iso_packet *params,
454 			bool sched_irq)
455 {
456 	int err;
457 
458 	params->interrupt = sched_irq;
459 	params->tag = s->tag;
460 	params->sy = 0;
461 
462 	err = fw_iso_context_queue(s->context, params, &s->buffer.iso_buffer,
463 				   s->buffer.packets[s->packet_index].offset);
464 	if (err < 0) {
465 		dev_err(&s->unit->device, "queueing error: %d\n", err);
466 		goto end;
467 	}
468 
469 	if (++s->packet_index >= s->queue_size)
470 		s->packet_index = 0;
471 end:
472 	return err;
473 }
474 
475 static inline int queue_out_packet(struct amdtp_stream *s,
476 				   struct fw_iso_packet *params, bool sched_irq)
477 {
478 	params->skip =
479 		!!(params->header_length == 0 && params->payload_length == 0);
480 	return queue_packet(s, params, sched_irq);
481 }
482 
483 static inline int queue_in_packet(struct amdtp_stream *s,
484 				  struct fw_iso_packet *params)
485 {
486 	// Queue one packet for IR context.
487 	params->header_length = s->ctx_data.tx.ctx_header_size;
488 	params->payload_length = s->ctx_data.tx.max_ctx_payload_length;
489 	params->skip = false;
490 	return queue_packet(s, params, false);
491 }
492 
493 static void generate_cip_header(struct amdtp_stream *s, __be32 cip_header[2],
494 			unsigned int data_block_counter, unsigned int syt)
495 {
496 	cip_header[0] = cpu_to_be32(READ_ONCE(s->source_node_id_field) |
497 				(s->data_block_quadlets << CIP_DBS_SHIFT) |
498 				((s->sph << CIP_SPH_SHIFT) & CIP_SPH_MASK) |
499 				data_block_counter);
500 	cip_header[1] = cpu_to_be32(CIP_EOH |
501 			((s->fmt << CIP_FMT_SHIFT) & CIP_FMT_MASK) |
502 			((s->ctx_data.rx.fdf << CIP_FDF_SHIFT) & CIP_FDF_MASK) |
503 			(syt & CIP_SYT_MASK));
504 }
505 
506 static void build_it_pkt_header(struct amdtp_stream *s, unsigned int cycle,
507 				struct fw_iso_packet *params,
508 				unsigned int data_blocks,
509 				unsigned int data_block_counter,
510 				unsigned int syt, unsigned int index)
511 {
512 	unsigned int payload_length;
513 	__be32 *cip_header;
514 
515 	payload_length = data_blocks * sizeof(__be32) * s->data_block_quadlets;
516 	params->payload_length = payload_length;
517 
518 	if (!(s->flags & CIP_NO_HEADER)) {
519 		cip_header = (__be32 *)params->header;
520 		generate_cip_header(s, cip_header, data_block_counter, syt);
521 		params->header_length = 2 * sizeof(__be32);
522 		payload_length += params->header_length;
523 	} else {
524 		cip_header = NULL;
525 	}
526 
527 	trace_amdtp_packet(s, cycle, cip_header, payload_length, data_blocks,
528 			   data_block_counter, index);
529 }
530 
531 static int check_cip_header(struct amdtp_stream *s, const __be32 *buf,
532 			    unsigned int payload_length,
533 			    unsigned int *data_blocks,
534 			    unsigned int *data_block_counter, unsigned int *syt)
535 {
536 	u32 cip_header[2];
537 	unsigned int sph;
538 	unsigned int fmt;
539 	unsigned int fdf;
540 	unsigned int dbc;
541 	bool lost;
542 
543 	cip_header[0] = be32_to_cpu(buf[0]);
544 	cip_header[1] = be32_to_cpu(buf[1]);
545 
546 	/*
547 	 * This module supports 'Two-quadlet CIP header with SYT field'.
548 	 * For convenience, also check FMT field is AM824 or not.
549 	 */
550 	if ((((cip_header[0] & CIP_EOH_MASK) == CIP_EOH) ||
551 	     ((cip_header[1] & CIP_EOH_MASK) != CIP_EOH)) &&
552 	    (!(s->flags & CIP_HEADER_WITHOUT_EOH))) {
553 		dev_info_ratelimited(&s->unit->device,
554 				"Invalid CIP header for AMDTP: %08X:%08X\n",
555 				cip_header[0], cip_header[1]);
556 		return -EAGAIN;
557 	}
558 
559 	/* Check valid protocol or not. */
560 	sph = (cip_header[0] & CIP_SPH_MASK) >> CIP_SPH_SHIFT;
561 	fmt = (cip_header[1] & CIP_FMT_MASK) >> CIP_FMT_SHIFT;
562 	if (sph != s->sph || fmt != s->fmt) {
563 		dev_info_ratelimited(&s->unit->device,
564 				     "Detect unexpected protocol: %08x %08x\n",
565 				     cip_header[0], cip_header[1]);
566 		return -EAGAIN;
567 	}
568 
569 	/* Calculate data blocks */
570 	fdf = (cip_header[1] & CIP_FDF_MASK) >> CIP_FDF_SHIFT;
571 	if (payload_length < sizeof(__be32) * 2 ||
572 	    (fmt == CIP_FMT_AM && fdf == AMDTP_FDF_NO_DATA)) {
573 		*data_blocks = 0;
574 	} else {
575 		unsigned int data_block_quadlets =
576 				(cip_header[0] & CIP_DBS_MASK) >> CIP_DBS_SHIFT;
577 		/* avoid division by zero */
578 		if (data_block_quadlets == 0) {
579 			dev_err(&s->unit->device,
580 				"Detect invalid value in dbs field: %08X\n",
581 				cip_header[0]);
582 			return -EPROTO;
583 		}
584 		if (s->flags & CIP_WRONG_DBS)
585 			data_block_quadlets = s->data_block_quadlets;
586 
587 		*data_blocks = (payload_length / sizeof(__be32) - 2) /
588 							data_block_quadlets;
589 	}
590 
591 	/* Check data block counter continuity */
592 	dbc = cip_header[0] & CIP_DBC_MASK;
593 	if (*data_blocks == 0 && (s->flags & CIP_EMPTY_HAS_WRONG_DBC) &&
594 	    *data_block_counter != UINT_MAX)
595 		dbc = *data_block_counter;
596 
597 	if ((dbc == 0x00 && (s->flags & CIP_SKIP_DBC_ZERO_CHECK)) ||
598 	    *data_block_counter == UINT_MAX) {
599 		lost = false;
600 	} else if (!(s->flags & CIP_DBC_IS_END_EVENT)) {
601 		lost = dbc != *data_block_counter;
602 	} else {
603 		unsigned int dbc_interval;
604 
605 		if (*data_blocks > 0 && s->ctx_data.tx.dbc_interval > 0)
606 			dbc_interval = s->ctx_data.tx.dbc_interval;
607 		else
608 			dbc_interval = *data_blocks;
609 
610 		lost = dbc != ((*data_block_counter + dbc_interval) & 0xff);
611 	}
612 
613 	if (lost) {
614 		dev_err(&s->unit->device,
615 			"Detect discontinuity of CIP: %02X %02X\n",
616 			*data_block_counter, dbc);
617 		return -EIO;
618 	}
619 
620 	*data_block_counter = dbc;
621 
622 	*syt = cip_header[1] & CIP_SYT_MASK;
623 
624 	return 0;
625 }
626 
627 static int parse_ir_ctx_header(struct amdtp_stream *s, unsigned int cycle,
628 			       const __be32 *ctx_header,
629 			       unsigned int *payload_length,
630 			       unsigned int *data_blocks,
631 			       unsigned int *data_block_counter,
632 			       unsigned int *syt, unsigned int index)
633 {
634 	const __be32 *cip_header;
635 	int err;
636 
637 	*payload_length = be32_to_cpu(ctx_header[0]) >> ISO_DATA_LENGTH_SHIFT;
638 	if (*payload_length > s->ctx_data.tx.ctx_header_size +
639 					s->ctx_data.tx.max_ctx_payload_length) {
640 		dev_err(&s->unit->device,
641 			"Detect jumbo payload: %04x %04x\n",
642 			*payload_length, s->ctx_data.tx.max_ctx_payload_length);
643 		return -EIO;
644 	}
645 
646 	if (!(s->flags & CIP_NO_HEADER)) {
647 		cip_header = ctx_header + 2;
648 		err = check_cip_header(s, cip_header, *payload_length,
649 				       data_blocks, data_block_counter, syt);
650 		if (err < 0)
651 			return err;
652 	} else {
653 		cip_header = NULL;
654 		err = 0;
655 		*data_blocks = *payload_length / sizeof(__be32) /
656 			       s->data_block_quadlets;
657 		*syt = 0;
658 
659 		if (*data_block_counter == UINT_MAX)
660 			*data_block_counter = 0;
661 	}
662 
663 	trace_amdtp_packet(s, cycle, cip_header, *payload_length, *data_blocks,
664 			   *data_block_counter, index);
665 
666 	return err;
667 }
668 
669 // In CYCLE_TIMER register of IEEE 1394, 7 bits are used to represent second. On
670 // the other hand, in DMA descriptors of 1394 OHCI, 3 bits are used to represent
671 // it. Thus, via Linux firewire subsystem, we can get the 3 bits for second.
672 static inline u32 compute_cycle_count(__be32 ctx_header_tstamp)
673 {
674 	u32 tstamp = be32_to_cpu(ctx_header_tstamp) & HEADER_TSTAMP_MASK;
675 	return (((tstamp >> 13) & 0x07) * 8000) + (tstamp & 0x1fff);
676 }
677 
678 static inline u32 increment_cycle_count(u32 cycle, unsigned int addend)
679 {
680 	cycle += addend;
681 	if (cycle >= OHCI_MAX_SECOND * CYCLES_PER_SECOND)
682 		cycle -= OHCI_MAX_SECOND * CYCLES_PER_SECOND;
683 	return cycle;
684 }
685 
686 // Align to actual cycle count for the packet which is going to be scheduled.
687 // This module queued the same number of isochronous cycle as the size of queue
688 // to kip isochronous cycle, therefore it's OK to just increment the cycle by
689 // the size of queue for scheduled cycle.
690 static inline u32 compute_it_cycle(const __be32 ctx_header_tstamp,
691 				   unsigned int queue_size)
692 {
693 	u32 cycle = compute_cycle_count(ctx_header_tstamp);
694 	return increment_cycle_count(cycle, queue_size);
695 }
696 
697 static int generate_device_pkt_descs(struct amdtp_stream *s,
698 				     struct pkt_desc *descs,
699 				     const __be32 *ctx_header,
700 				     unsigned int packets)
701 {
702 	unsigned int dbc = s->data_block_counter;
703 	int i;
704 	int err;
705 
706 	for (i = 0; i < packets; ++i) {
707 		struct pkt_desc *desc = descs + i;
708 		unsigned int index = (s->packet_index + i) % s->queue_size;
709 		unsigned int cycle;
710 		unsigned int payload_length;
711 		unsigned int data_blocks;
712 		unsigned int syt;
713 
714 		cycle = compute_cycle_count(ctx_header[1]);
715 
716 		err = parse_ir_ctx_header(s, cycle, ctx_header, &payload_length,
717 					  &data_blocks, &dbc, &syt, i);
718 		if (err < 0)
719 			return err;
720 
721 		desc->cycle = cycle;
722 		desc->syt = syt;
723 		desc->data_blocks = data_blocks;
724 		desc->data_block_counter = dbc;
725 		desc->ctx_payload = s->buffer.packets[index].buffer;
726 
727 		if (!(s->flags & CIP_DBC_IS_END_EVENT))
728 			dbc = (dbc + desc->data_blocks) & 0xff;
729 
730 		ctx_header +=
731 			s->ctx_data.tx.ctx_header_size / sizeof(*ctx_header);
732 	}
733 
734 	s->data_block_counter = dbc;
735 
736 	return 0;
737 }
738 
739 static unsigned int compute_syt(unsigned int syt_offset, unsigned int cycle,
740 				unsigned int transfer_delay)
741 {
742 	unsigned int syt;
743 
744 	syt_offset += transfer_delay;
745 	syt = ((cycle + syt_offset / TICKS_PER_CYCLE) << 12) |
746 	      (syt_offset % TICKS_PER_CYCLE);
747 	return syt & CIP_SYT_MASK;
748 }
749 
750 static void generate_pkt_descs(struct amdtp_stream *s, struct pkt_desc *descs,
751 			       const __be32 *ctx_header, unsigned int packets,
752 			       const struct seq_desc *seq_descs,
753 			       unsigned int seq_size)
754 {
755 	unsigned int dbc = s->data_block_counter;
756 	unsigned int seq_index = s->ctx_data.rx.seq_index;
757 	int i;
758 
759 	for (i = 0; i < packets; ++i) {
760 		struct pkt_desc *desc = descs + i;
761 		unsigned int index = (s->packet_index + i) % s->queue_size;
762 		const struct seq_desc *seq = seq_descs + seq_index;
763 		unsigned int syt;
764 
765 		desc->cycle = compute_it_cycle(*ctx_header, s->queue_size);
766 
767 		syt = seq->syt_offset;
768 		if (syt != CIP_SYT_NO_INFO) {
769 			syt = compute_syt(syt, desc->cycle,
770 					  s->ctx_data.rx.transfer_delay);
771 		}
772 		desc->syt = syt;
773 		desc->data_blocks = seq->data_blocks;
774 
775 		if (s->flags & CIP_DBC_IS_END_EVENT)
776 			dbc = (dbc + desc->data_blocks) & 0xff;
777 
778 		desc->data_block_counter = dbc;
779 
780 		if (!(s->flags & CIP_DBC_IS_END_EVENT))
781 			dbc = (dbc + desc->data_blocks) & 0xff;
782 
783 		desc->ctx_payload = s->buffer.packets[index].buffer;
784 
785 		seq_index = (seq_index + 1) % seq_size;
786 
787 		++ctx_header;
788 	}
789 
790 	s->data_block_counter = dbc;
791 	s->ctx_data.rx.seq_index = seq_index;
792 }
793 
794 static inline void cancel_stream(struct amdtp_stream *s)
795 {
796 	s->packet_index = -1;
797 	if (in_interrupt())
798 		amdtp_stream_pcm_abort(s);
799 	WRITE_ONCE(s->pcm_buffer_pointer, SNDRV_PCM_POS_XRUN);
800 }
801 
802 static void process_ctx_payloads(struct amdtp_stream *s,
803 				 const struct pkt_desc *descs,
804 				 unsigned int packets)
805 {
806 	struct snd_pcm_substream *pcm;
807 	unsigned int pcm_frames;
808 
809 	pcm = READ_ONCE(s->pcm);
810 	pcm_frames = s->process_ctx_payloads(s, descs, packets, pcm);
811 	if (pcm)
812 		update_pcm_pointers(s, pcm, pcm_frames);
813 }
814 
815 static void out_stream_callback(struct fw_iso_context *context, u32 tstamp,
816 				size_t header_length, void *header,
817 				void *private_data)
818 {
819 	struct amdtp_stream *s = private_data;
820 	const struct amdtp_domain *d = s->domain;
821 	const __be32 *ctx_header = header;
822 	unsigned int events_per_period = s->ctx_data.rx.events_per_period;
823 	unsigned int event_count = s->ctx_data.rx.event_count;
824 	unsigned int packets;
825 	int i;
826 
827 	if (s->packet_index < 0)
828 		return;
829 
830 	// Calculate the number of packets in buffer and check XRUN.
831 	packets = header_length / sizeof(*ctx_header);
832 
833 	generate_pkt_descs(s, s->pkt_descs, ctx_header, packets, d->seq_descs,
834 			   d->seq_size);
835 
836 	process_ctx_payloads(s, s->pkt_descs, packets);
837 
838 	for (i = 0; i < packets; ++i) {
839 		const struct pkt_desc *desc = s->pkt_descs + i;
840 		unsigned int syt;
841 		struct {
842 			struct fw_iso_packet params;
843 			__be32 header[IT_PKT_HEADER_SIZE_CIP / sizeof(__be32)];
844 		} template = { {0}, {0} };
845 		bool sched_irq = false;
846 
847 		if (s->ctx_data.rx.syt_override < 0)
848 			syt = desc->syt;
849 		else
850 			syt = s->ctx_data.rx.syt_override;
851 
852 		build_it_pkt_header(s, desc->cycle, &template.params,
853 				    desc->data_blocks, desc->data_block_counter,
854 				    syt, i);
855 
856 		if (s == s->domain->irq_target) {
857 			event_count += desc->data_blocks;
858 			if (event_count >= events_per_period) {
859 				event_count -= events_per_period;
860 				sched_irq = true;
861 			}
862 		}
863 
864 		if (queue_out_packet(s, &template.params, sched_irq) < 0) {
865 			cancel_stream(s);
866 			return;
867 		}
868 	}
869 
870 	s->ctx_data.rx.event_count = event_count;
871 }
872 
873 static void in_stream_callback(struct fw_iso_context *context, u32 tstamp,
874 			       size_t header_length, void *header,
875 			       void *private_data)
876 {
877 	struct amdtp_stream *s = private_data;
878 	__be32 *ctx_header = header;
879 	unsigned int packets;
880 	int i;
881 	int err;
882 
883 	if (s->packet_index < 0)
884 		return;
885 
886 	// Calculate the number of packets in buffer and check XRUN.
887 	packets = header_length / s->ctx_data.tx.ctx_header_size;
888 
889 	err = generate_device_pkt_descs(s, s->pkt_descs, ctx_header, packets);
890 	if (err < 0) {
891 		if (err != -EAGAIN) {
892 			cancel_stream(s);
893 			return;
894 		}
895 	} else {
896 		process_ctx_payloads(s, s->pkt_descs, packets);
897 	}
898 
899 	for (i = 0; i < packets; ++i) {
900 		struct fw_iso_packet params = {0};
901 
902 		if (queue_in_packet(s, &params) < 0) {
903 			cancel_stream(s);
904 			return;
905 		}
906 	}
907 }
908 
909 static void pool_ideal_seq_descs(struct amdtp_domain *d, unsigned int packets)
910 {
911 	struct amdtp_stream *irq_target = d->irq_target;
912 	unsigned int seq_tail = d->seq_tail;
913 	unsigned int seq_size = d->seq_size;
914 	unsigned int min_avail;
915 	struct amdtp_stream *s;
916 
917 	min_avail = d->seq_size;
918 	list_for_each_entry(s, &d->streams, list) {
919 		unsigned int seq_index;
920 		unsigned int avail;
921 
922 		if (s->direction == AMDTP_IN_STREAM)
923 			continue;
924 
925 		seq_index = s->ctx_data.rx.seq_index;
926 		avail = d->seq_tail;
927 		if (seq_index > avail)
928 			avail += d->seq_size;
929 		avail -= seq_index;
930 
931 		if (avail < min_avail)
932 			min_avail = avail;
933 	}
934 
935 	while (min_avail < packets) {
936 		struct seq_desc *desc = d->seq_descs + seq_tail;
937 
938 		desc->syt_offset = calculate_syt_offset(&d->last_syt_offset,
939 					&d->syt_offset_state, irq_target->sfc);
940 		desc->data_blocks = calculate_data_blocks(&d->data_block_state,
941 				!!(irq_target->flags & CIP_BLOCKING),
942 				desc->syt_offset == CIP_SYT_NO_INFO,
943 				irq_target->syt_interval, irq_target->sfc);
944 
945 		++seq_tail;
946 		seq_tail %= seq_size;
947 
948 		++min_avail;
949 	}
950 
951 	d->seq_tail = seq_tail;
952 }
953 
954 static void irq_target_callback(struct fw_iso_context *context, u32 tstamp,
955 				size_t header_length, void *header,
956 				void *private_data)
957 {
958 	struct amdtp_stream *irq_target = private_data;
959 	struct amdtp_domain *d = irq_target->domain;
960 	unsigned int packets = header_length / sizeof(__be32);
961 	struct amdtp_stream *s;
962 
963 	// Record enough entries with extra 3 cycles at least.
964 	pool_ideal_seq_descs(d, packets + 3);
965 
966 	out_stream_callback(context, tstamp, header_length, header, irq_target);
967 	if (amdtp_streaming_error(irq_target))
968 		goto error;
969 
970 	list_for_each_entry(s, &d->streams, list) {
971 		if (s != irq_target && amdtp_stream_running(s)) {
972 			fw_iso_context_flush_completions(s->context);
973 			if (amdtp_streaming_error(s))
974 				goto error;
975 		}
976 	}
977 
978 	return;
979 error:
980 	if (amdtp_stream_running(irq_target))
981 		cancel_stream(irq_target);
982 
983 	list_for_each_entry(s, &d->streams, list) {
984 		if (amdtp_stream_running(s))
985 			cancel_stream(s);
986 	}
987 }
988 
989 // this is executed one time.
990 static void amdtp_stream_first_callback(struct fw_iso_context *context,
991 					u32 tstamp, size_t header_length,
992 					void *header, void *private_data)
993 {
994 	struct amdtp_stream *s = private_data;
995 	const __be32 *ctx_header = header;
996 	u32 cycle;
997 
998 	/*
999 	 * For in-stream, first packet has come.
1000 	 * For out-stream, prepared to transmit first packet
1001 	 */
1002 	s->callbacked = true;
1003 	wake_up(&s->callback_wait);
1004 
1005 	if (s->direction == AMDTP_IN_STREAM) {
1006 		cycle = compute_cycle_count(ctx_header[1]);
1007 
1008 		context->callback.sc = in_stream_callback;
1009 	} else {
1010 		cycle = compute_it_cycle(*ctx_header, s->queue_size);
1011 
1012 		if (s == s->domain->irq_target)
1013 			context->callback.sc = irq_target_callback;
1014 		else
1015 			context->callback.sc = out_stream_callback;
1016 	}
1017 
1018 	s->start_cycle = cycle;
1019 
1020 	context->callback.sc(context, tstamp, header_length, header, s);
1021 }
1022 
1023 /**
1024  * amdtp_stream_start - start transferring packets
1025  * @s: the AMDTP stream to start
1026  * @channel: the isochronous channel on the bus
1027  * @speed: firewire speed code
1028  * @start_cycle: the isochronous cycle to start the context. Start immediately
1029  *		 if negative value is given.
1030  * @queue_size: The number of packets in the queue.
1031  * @idle_irq_interval: the interval to queue packet during initial state.
1032  *
1033  * The stream cannot be started until it has been configured with
1034  * amdtp_stream_set_parameters() and it must be started before any PCM or MIDI
1035  * device can be started.
1036  */
1037 static int amdtp_stream_start(struct amdtp_stream *s, int channel, int speed,
1038 			      int start_cycle, unsigned int queue_size,
1039 			      unsigned int idle_irq_interval)
1040 {
1041 	bool is_irq_target = (s == s->domain->irq_target);
1042 	unsigned int ctx_header_size;
1043 	unsigned int max_ctx_payload_size;
1044 	enum dma_data_direction dir;
1045 	int type, tag, err;
1046 
1047 	mutex_lock(&s->mutex);
1048 
1049 	if (WARN_ON(amdtp_stream_running(s) ||
1050 		    (s->data_block_quadlets < 1))) {
1051 		err = -EBADFD;
1052 		goto err_unlock;
1053 	}
1054 
1055 	if (s->direction == AMDTP_IN_STREAM) {
1056 		// NOTE: IT context should be used for constant IRQ.
1057 		if (is_irq_target) {
1058 			err = -EINVAL;
1059 			goto err_unlock;
1060 		}
1061 
1062 		s->data_block_counter = UINT_MAX;
1063 	} else {
1064 		s->data_block_counter = 0;
1065 	}
1066 
1067 	/* initialize packet buffer */
1068 	if (s->direction == AMDTP_IN_STREAM) {
1069 		dir = DMA_FROM_DEVICE;
1070 		type = FW_ISO_CONTEXT_RECEIVE;
1071 		if (!(s->flags & CIP_NO_HEADER))
1072 			ctx_header_size = IR_CTX_HEADER_SIZE_CIP;
1073 		else
1074 			ctx_header_size = IR_CTX_HEADER_SIZE_NO_CIP;
1075 
1076 		max_ctx_payload_size = amdtp_stream_get_max_payload(s) -
1077 				       ctx_header_size;
1078 	} else {
1079 		dir = DMA_TO_DEVICE;
1080 		type = FW_ISO_CONTEXT_TRANSMIT;
1081 		ctx_header_size = 0;	// No effect for IT context.
1082 
1083 		max_ctx_payload_size = amdtp_stream_get_max_payload(s);
1084 		if (!(s->flags & CIP_NO_HEADER))
1085 			max_ctx_payload_size -= IT_PKT_HEADER_SIZE_CIP;
1086 	}
1087 
1088 	err = iso_packets_buffer_init(&s->buffer, s->unit, queue_size,
1089 				      max_ctx_payload_size, dir);
1090 	if (err < 0)
1091 		goto err_unlock;
1092 	s->queue_size = queue_size;
1093 
1094 	s->context = fw_iso_context_create(fw_parent_device(s->unit)->card,
1095 					  type, channel, speed, ctx_header_size,
1096 					  amdtp_stream_first_callback, s);
1097 	if (IS_ERR(s->context)) {
1098 		err = PTR_ERR(s->context);
1099 		if (err == -EBUSY)
1100 			dev_err(&s->unit->device,
1101 				"no free stream on this controller\n");
1102 		goto err_buffer;
1103 	}
1104 
1105 	amdtp_stream_update(s);
1106 
1107 	if (s->direction == AMDTP_IN_STREAM) {
1108 		s->ctx_data.tx.max_ctx_payload_length = max_ctx_payload_size;
1109 		s->ctx_data.tx.ctx_header_size = ctx_header_size;
1110 	}
1111 
1112 	if (s->flags & CIP_NO_HEADER)
1113 		s->tag = TAG_NO_CIP_HEADER;
1114 	else
1115 		s->tag = TAG_CIP;
1116 
1117 	s->pkt_descs = kcalloc(s->queue_size, sizeof(*s->pkt_descs),
1118 			       GFP_KERNEL);
1119 	if (!s->pkt_descs) {
1120 		err = -ENOMEM;
1121 		goto err_context;
1122 	}
1123 
1124 	s->packet_index = 0;
1125 	do {
1126 		struct fw_iso_packet params;
1127 
1128 		if (s->direction == AMDTP_IN_STREAM) {
1129 			err = queue_in_packet(s, &params);
1130 		} else {
1131 			bool sched_irq = false;
1132 
1133 			params.header_length = 0;
1134 			params.payload_length = 0;
1135 
1136 			if (is_irq_target) {
1137 				sched_irq = !((s->packet_index + 1) %
1138 					      idle_irq_interval);
1139 			}
1140 
1141 			err = queue_out_packet(s, &params, sched_irq);
1142 		}
1143 		if (err < 0)
1144 			goto err_pkt_descs;
1145 	} while (s->packet_index > 0);
1146 
1147 	/* NOTE: TAG1 matches CIP. This just affects in stream. */
1148 	tag = FW_ISO_CONTEXT_MATCH_TAG1;
1149 	if ((s->flags & CIP_EMPTY_WITH_TAG0) || (s->flags & CIP_NO_HEADER))
1150 		tag |= FW_ISO_CONTEXT_MATCH_TAG0;
1151 
1152 	s->callbacked = false;
1153 	err = fw_iso_context_start(s->context, start_cycle, 0, tag);
1154 	if (err < 0)
1155 		goto err_pkt_descs;
1156 
1157 	mutex_unlock(&s->mutex);
1158 
1159 	return 0;
1160 err_pkt_descs:
1161 	kfree(s->pkt_descs);
1162 err_context:
1163 	fw_iso_context_destroy(s->context);
1164 	s->context = ERR_PTR(-1);
1165 err_buffer:
1166 	iso_packets_buffer_destroy(&s->buffer, s->unit);
1167 err_unlock:
1168 	mutex_unlock(&s->mutex);
1169 
1170 	return err;
1171 }
1172 
1173 /**
1174  * amdtp_domain_stream_pcm_pointer - get the PCM buffer position
1175  * @d: the AMDTP domain.
1176  * @s: the AMDTP stream that transports the PCM data
1177  *
1178  * Returns the current buffer position, in frames.
1179  */
1180 unsigned long amdtp_domain_stream_pcm_pointer(struct amdtp_domain *d,
1181 					      struct amdtp_stream *s)
1182 {
1183 	struct amdtp_stream *irq_target = d->irq_target;
1184 
1185 	if (irq_target && amdtp_stream_running(irq_target)) {
1186 		// This function is called in software IRQ context of
1187 		// period_tasklet or process context.
1188 		//
1189 		// When the software IRQ context was scheduled by software IRQ
1190 		// context of IT contexts, queued packets were already handled.
1191 		// Therefore, no need to flush the queue in buffer furthermore.
1192 		//
1193 		// When the process context reach here, some packets will be
1194 		// already queued in the buffer. These packets should be handled
1195 		// immediately to keep better granularity of PCM pointer.
1196 		//
1197 		// Later, the process context will sometimes schedules software
1198 		// IRQ context of the period_tasklet. Then, no need to flush the
1199 		// queue by the same reason as described in the above
1200 		if (!in_interrupt()) {
1201 			// Queued packet should be processed without any kernel
1202 			// preemption to keep latency against bus cycle.
1203 			preempt_disable();
1204 			fw_iso_context_flush_completions(irq_target->context);
1205 			preempt_enable();
1206 		}
1207 	}
1208 
1209 	return READ_ONCE(s->pcm_buffer_pointer);
1210 }
1211 EXPORT_SYMBOL_GPL(amdtp_domain_stream_pcm_pointer);
1212 
1213 /**
1214  * amdtp_domain_stream_pcm_ack - acknowledge queued PCM frames
1215  * @d: the AMDTP domain.
1216  * @s: the AMDTP stream that transfers the PCM frames
1217  *
1218  * Returns zero always.
1219  */
1220 int amdtp_domain_stream_pcm_ack(struct amdtp_domain *d, struct amdtp_stream *s)
1221 {
1222 	struct amdtp_stream *irq_target = d->irq_target;
1223 
1224 	// Process isochronous packets for recent isochronous cycle to handle
1225 	// queued PCM frames.
1226 	if (irq_target && amdtp_stream_running(irq_target)) {
1227 		// Queued packet should be processed without any kernel
1228 		// preemption to keep latency against bus cycle.
1229 		preempt_disable();
1230 		fw_iso_context_flush_completions(irq_target->context);
1231 		preempt_enable();
1232 	}
1233 
1234 	return 0;
1235 }
1236 EXPORT_SYMBOL_GPL(amdtp_domain_stream_pcm_ack);
1237 
1238 /**
1239  * amdtp_stream_update - update the stream after a bus reset
1240  * @s: the AMDTP stream
1241  */
1242 void amdtp_stream_update(struct amdtp_stream *s)
1243 {
1244 	/* Precomputing. */
1245 	WRITE_ONCE(s->source_node_id_field,
1246                    (fw_parent_device(s->unit)->card->node_id << CIP_SID_SHIFT) & CIP_SID_MASK);
1247 }
1248 EXPORT_SYMBOL(amdtp_stream_update);
1249 
1250 /**
1251  * amdtp_stream_stop - stop sending packets
1252  * @s: the AMDTP stream to stop
1253  *
1254  * All PCM and MIDI devices of the stream must be stopped before the stream
1255  * itself can be stopped.
1256  */
1257 static void amdtp_stream_stop(struct amdtp_stream *s)
1258 {
1259 	mutex_lock(&s->mutex);
1260 
1261 	if (!amdtp_stream_running(s)) {
1262 		mutex_unlock(&s->mutex);
1263 		return;
1264 	}
1265 
1266 	tasklet_kill(&s->period_tasklet);
1267 	fw_iso_context_stop(s->context);
1268 	fw_iso_context_destroy(s->context);
1269 	s->context = ERR_PTR(-1);
1270 	iso_packets_buffer_destroy(&s->buffer, s->unit);
1271 	kfree(s->pkt_descs);
1272 
1273 	s->callbacked = false;
1274 
1275 	mutex_unlock(&s->mutex);
1276 }
1277 
1278 /**
1279  * amdtp_stream_pcm_abort - abort the running PCM device
1280  * @s: the AMDTP stream about to be stopped
1281  *
1282  * If the isochronous stream needs to be stopped asynchronously, call this
1283  * function first to stop the PCM device.
1284  */
1285 void amdtp_stream_pcm_abort(struct amdtp_stream *s)
1286 {
1287 	struct snd_pcm_substream *pcm;
1288 
1289 	pcm = READ_ONCE(s->pcm);
1290 	if (pcm)
1291 		snd_pcm_stop_xrun(pcm);
1292 }
1293 EXPORT_SYMBOL(amdtp_stream_pcm_abort);
1294 
1295 /**
1296  * amdtp_domain_init - initialize an AMDTP domain structure
1297  * @d: the AMDTP domain to initialize.
1298  */
1299 int amdtp_domain_init(struct amdtp_domain *d)
1300 {
1301 	INIT_LIST_HEAD(&d->streams);
1302 
1303 	d->events_per_period = 0;
1304 
1305 	d->seq_descs = NULL;
1306 
1307 	return 0;
1308 }
1309 EXPORT_SYMBOL_GPL(amdtp_domain_init);
1310 
1311 /**
1312  * amdtp_domain_destroy - destroy an AMDTP domain structure
1313  * @d: the AMDTP domain to destroy.
1314  */
1315 void amdtp_domain_destroy(struct amdtp_domain *d)
1316 {
1317 	// At present nothing to do.
1318 	return;
1319 }
1320 EXPORT_SYMBOL_GPL(amdtp_domain_destroy);
1321 
1322 /**
1323  * amdtp_domain_add_stream - register isoc context into the domain.
1324  * @d: the AMDTP domain.
1325  * @s: the AMDTP stream.
1326  * @channel: the isochronous channel on the bus.
1327  * @speed: firewire speed code.
1328  */
1329 int amdtp_domain_add_stream(struct amdtp_domain *d, struct amdtp_stream *s,
1330 			    int channel, int speed)
1331 {
1332 	struct amdtp_stream *tmp;
1333 
1334 	list_for_each_entry(tmp, &d->streams, list) {
1335 		if (s == tmp)
1336 			return -EBUSY;
1337 	}
1338 
1339 	list_add(&s->list, &d->streams);
1340 
1341 	s->channel = channel;
1342 	s->speed = speed;
1343 	s->domain = d;
1344 
1345 	return 0;
1346 }
1347 EXPORT_SYMBOL_GPL(amdtp_domain_add_stream);
1348 
1349 static int get_current_cycle_time(struct fw_card *fw_card, int *cur_cycle)
1350 {
1351 	int generation;
1352 	int rcode;
1353 	__be32 reg;
1354 	u32 data;
1355 
1356 	// This is a request to local 1394 OHCI controller and expected to
1357 	// complete without any event waiting.
1358 	generation = fw_card->generation;
1359 	smp_rmb();	// node_id vs. generation.
1360 	rcode = fw_run_transaction(fw_card, TCODE_READ_QUADLET_REQUEST,
1361 				   fw_card->node_id, generation, SCODE_100,
1362 				   CSR_REGISTER_BASE + CSR_CYCLE_TIME,
1363 				   &reg, sizeof(reg));
1364 	if (rcode != RCODE_COMPLETE)
1365 		return -EIO;
1366 
1367 	data = be32_to_cpu(reg);
1368 	*cur_cycle = data >> 12;
1369 
1370 	return 0;
1371 }
1372 
1373 /**
1374  * amdtp_domain_start - start sending packets for isoc context in the domain.
1375  * @d: the AMDTP domain.
1376  * @ir_delay_cycle: the cycle delay to start all IR contexts.
1377  */
1378 int amdtp_domain_start(struct amdtp_domain *d, unsigned int ir_delay_cycle)
1379 {
1380 	static const struct {
1381 		unsigned int data_block;
1382 		unsigned int syt_offset;
1383 	} *entry, initial_state[] = {
1384 		[CIP_SFC_32000]  = {  4, 3072 },
1385 		[CIP_SFC_48000]  = {  6, 1024 },
1386 		[CIP_SFC_96000]  = { 12, 1024 },
1387 		[CIP_SFC_192000] = { 24, 1024 },
1388 		[CIP_SFC_44100]  = {  0,   67 },
1389 		[CIP_SFC_88200]  = {  0,   67 },
1390 		[CIP_SFC_176400] = {  0,   67 },
1391 	};
1392 	unsigned int events_per_buffer = d->events_per_buffer;
1393 	unsigned int events_per_period = d->events_per_period;
1394 	unsigned int idle_irq_interval;
1395 	unsigned int queue_size;
1396 	struct amdtp_stream *s;
1397 	int cycle;
1398 	int err;
1399 
1400 	// Select an IT context as IRQ target.
1401 	list_for_each_entry(s, &d->streams, list) {
1402 		if (s->direction == AMDTP_OUT_STREAM)
1403 			break;
1404 	}
1405 	if (!s)
1406 		return -ENXIO;
1407 	d->irq_target = s;
1408 
1409 	// This is a case that AMDTP streams in domain run just for MIDI
1410 	// substream. Use the number of events equivalent to 10 msec as
1411 	// interval of hardware IRQ.
1412 	if (events_per_period == 0)
1413 		events_per_period = amdtp_rate_table[d->irq_target->sfc] / 100;
1414 	if (events_per_buffer == 0)
1415 		events_per_buffer = events_per_period * 3;
1416 
1417 	queue_size = DIV_ROUND_UP(CYCLES_PER_SECOND * events_per_buffer,
1418 				  amdtp_rate_table[d->irq_target->sfc]);
1419 
1420 	d->seq_descs = kcalloc(queue_size, sizeof(*d->seq_descs), GFP_KERNEL);
1421 	if (!d->seq_descs)
1422 		return -ENOMEM;
1423 	d->seq_size = queue_size;
1424 	d->seq_tail = 0;
1425 
1426 	entry = &initial_state[s->sfc];
1427 	d->data_block_state = entry->data_block;
1428 	d->syt_offset_state = entry->syt_offset;
1429 	d->last_syt_offset = TICKS_PER_CYCLE;
1430 
1431 	if (ir_delay_cycle > 0) {
1432 		struct fw_card *fw_card = fw_parent_device(s->unit)->card;
1433 
1434 		err = get_current_cycle_time(fw_card, &cycle);
1435 		if (err < 0)
1436 			goto error;
1437 
1438 		// No need to care overflow in cycle field because of enough
1439 		// width.
1440 		cycle += ir_delay_cycle;
1441 
1442 		// Round up to sec field.
1443 		if ((cycle & 0x00001fff) >= CYCLES_PER_SECOND) {
1444 			unsigned int sec;
1445 
1446 			// The sec field can overflow.
1447 			sec = (cycle & 0xffffe000) >> 13;
1448 			cycle = (++sec << 13) |
1449 				((cycle & 0x00001fff) / CYCLES_PER_SECOND);
1450 		}
1451 
1452 		// In OHCI 1394 specification, lower 2 bits are available for
1453 		// sec field.
1454 		cycle &= 0x00007fff;
1455 	} else {
1456 		cycle = -1;
1457 	}
1458 
1459 	list_for_each_entry(s, &d->streams, list) {
1460 		int cycle_match;
1461 
1462 		if (s->direction == AMDTP_IN_STREAM) {
1463 			cycle_match = cycle;
1464 		} else {
1465 			// IT context starts immediately.
1466 			cycle_match = -1;
1467 			s->ctx_data.rx.seq_index = 0;
1468 		}
1469 
1470 		if (s != d->irq_target) {
1471 			err = amdtp_stream_start(s, s->channel, s->speed,
1472 						 cycle_match, queue_size, 0);
1473 			if (err < 0)
1474 				goto error;
1475 		}
1476 	}
1477 
1478 	s = d->irq_target;
1479 	s->ctx_data.rx.events_per_period = events_per_period;
1480 	s->ctx_data.rx.event_count = 0;
1481 	s->ctx_data.rx.seq_index = 0;
1482 
1483 	idle_irq_interval = DIV_ROUND_UP(CYCLES_PER_SECOND * events_per_period,
1484 					 amdtp_rate_table[d->irq_target->sfc]);
1485 	err = amdtp_stream_start(s, s->channel, s->speed, -1, queue_size,
1486 				 idle_irq_interval);
1487 	if (err < 0)
1488 		goto error;
1489 
1490 	return 0;
1491 error:
1492 	list_for_each_entry(s, &d->streams, list)
1493 		amdtp_stream_stop(s);
1494 	kfree(d->seq_descs);
1495 	d->seq_descs = NULL;
1496 	return err;
1497 }
1498 EXPORT_SYMBOL_GPL(amdtp_domain_start);
1499 
1500 /**
1501  * amdtp_domain_stop - stop sending packets for isoc context in the same domain.
1502  * @d: the AMDTP domain to which the isoc contexts belong.
1503  */
1504 void amdtp_domain_stop(struct amdtp_domain *d)
1505 {
1506 	struct amdtp_stream *s, *next;
1507 
1508 	if (d->irq_target)
1509 		amdtp_stream_stop(d->irq_target);
1510 
1511 	list_for_each_entry_safe(s, next, &d->streams, list) {
1512 		list_del(&s->list);
1513 
1514 		if (s != d->irq_target)
1515 			amdtp_stream_stop(s);
1516 	}
1517 
1518 	d->events_per_period = 0;
1519 	d->irq_target = NULL;
1520 
1521 	kfree(d->seq_descs);
1522 	d->seq_descs = NULL;
1523 }
1524 EXPORT_SYMBOL_GPL(amdtp_domain_stop);
1525