xref: /openbmc/linux/sound/firewire/amdtp-stream.c (revision 0bf49ffb)
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_work(struct work_struct *work);
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 	INIT_WORK(&s->period_work, pcm_period_work);
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 work 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 	cancel_work_sync(&s->period_work);
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 		queue_work(system_highpri_wq, &s->period_work);
441 	}
442 }
443 
444 static void pcm_period_work(struct work_struct *work)
445 {
446 	struct amdtp_stream *s = container_of(work, struct amdtp_stream,
447 					      period_work);
448 	struct snd_pcm_substream *pcm = READ_ONCE(s->pcm);
449 
450 	if (pcm)
451 		snd_pcm_period_elapsed(pcm);
452 }
453 
454 static int queue_packet(struct amdtp_stream *s, struct fw_iso_packet *params,
455 			bool sched_irq)
456 {
457 	int err;
458 
459 	params->interrupt = sched_irq;
460 	params->tag = s->tag;
461 	params->sy = 0;
462 
463 	err = fw_iso_context_queue(s->context, params, &s->buffer.iso_buffer,
464 				   s->buffer.packets[s->packet_index].offset);
465 	if (err < 0) {
466 		dev_err(&s->unit->device, "queueing error: %d\n", err);
467 		goto end;
468 	}
469 
470 	if (++s->packet_index >= s->queue_size)
471 		s->packet_index = 0;
472 end:
473 	return err;
474 }
475 
476 static inline int queue_out_packet(struct amdtp_stream *s,
477 				   struct fw_iso_packet *params, bool sched_irq)
478 {
479 	params->skip =
480 		!!(params->header_length == 0 && params->payload_length == 0);
481 	return queue_packet(s, params, sched_irq);
482 }
483 
484 static inline int queue_in_packet(struct amdtp_stream *s,
485 				  struct fw_iso_packet *params)
486 {
487 	// Queue one packet for IR context.
488 	params->header_length = s->ctx_data.tx.ctx_header_size;
489 	params->payload_length = s->ctx_data.tx.max_ctx_payload_length;
490 	params->skip = false;
491 	return queue_packet(s, params, false);
492 }
493 
494 static void generate_cip_header(struct amdtp_stream *s, __be32 cip_header[2],
495 			unsigned int data_block_counter, unsigned int syt)
496 {
497 	cip_header[0] = cpu_to_be32(READ_ONCE(s->source_node_id_field) |
498 				(s->data_block_quadlets << CIP_DBS_SHIFT) |
499 				((s->sph << CIP_SPH_SHIFT) & CIP_SPH_MASK) |
500 				data_block_counter);
501 	cip_header[1] = cpu_to_be32(CIP_EOH |
502 			((s->fmt << CIP_FMT_SHIFT) & CIP_FMT_MASK) |
503 			((s->ctx_data.rx.fdf << CIP_FDF_SHIFT) & CIP_FDF_MASK) |
504 			(syt & CIP_SYT_MASK));
505 }
506 
507 static void build_it_pkt_header(struct amdtp_stream *s, unsigned int cycle,
508 				struct fw_iso_packet *params,
509 				unsigned int data_blocks,
510 				unsigned int data_block_counter,
511 				unsigned int syt, unsigned int index)
512 {
513 	unsigned int payload_length;
514 	__be32 *cip_header;
515 
516 	payload_length = data_blocks * sizeof(__be32) * s->data_block_quadlets;
517 	params->payload_length = payload_length;
518 
519 	if (!(s->flags & CIP_NO_HEADER)) {
520 		cip_header = (__be32 *)params->header;
521 		generate_cip_header(s, cip_header, data_block_counter, syt);
522 		params->header_length = 2 * sizeof(__be32);
523 		payload_length += params->header_length;
524 	} else {
525 		cip_header = NULL;
526 	}
527 
528 	trace_amdtp_packet(s, cycle, cip_header, payload_length, data_blocks,
529 			   data_block_counter, index);
530 }
531 
532 static int check_cip_header(struct amdtp_stream *s, const __be32 *buf,
533 			    unsigned int payload_length,
534 			    unsigned int *data_blocks,
535 			    unsigned int *data_block_counter, unsigned int *syt)
536 {
537 	u32 cip_header[2];
538 	unsigned int sph;
539 	unsigned int fmt;
540 	unsigned int fdf;
541 	unsigned int dbc;
542 	bool lost;
543 
544 	cip_header[0] = be32_to_cpu(buf[0]);
545 	cip_header[1] = be32_to_cpu(buf[1]);
546 
547 	/*
548 	 * This module supports 'Two-quadlet CIP header with SYT field'.
549 	 * For convenience, also check FMT field is AM824 or not.
550 	 */
551 	if ((((cip_header[0] & CIP_EOH_MASK) == CIP_EOH) ||
552 	     ((cip_header[1] & CIP_EOH_MASK) != CIP_EOH)) &&
553 	    (!(s->flags & CIP_HEADER_WITHOUT_EOH))) {
554 		dev_info_ratelimited(&s->unit->device,
555 				"Invalid CIP header for AMDTP: %08X:%08X\n",
556 				cip_header[0], cip_header[1]);
557 		return -EAGAIN;
558 	}
559 
560 	/* Check valid protocol or not. */
561 	sph = (cip_header[0] & CIP_SPH_MASK) >> CIP_SPH_SHIFT;
562 	fmt = (cip_header[1] & CIP_FMT_MASK) >> CIP_FMT_SHIFT;
563 	if (sph != s->sph || fmt != s->fmt) {
564 		dev_info_ratelimited(&s->unit->device,
565 				     "Detect unexpected protocol: %08x %08x\n",
566 				     cip_header[0], cip_header[1]);
567 		return -EAGAIN;
568 	}
569 
570 	/* Calculate data blocks */
571 	fdf = (cip_header[1] & CIP_FDF_MASK) >> CIP_FDF_SHIFT;
572 	if (payload_length < sizeof(__be32) * 2 ||
573 	    (fmt == CIP_FMT_AM && fdf == AMDTP_FDF_NO_DATA)) {
574 		*data_blocks = 0;
575 	} else {
576 		unsigned int data_block_quadlets =
577 				(cip_header[0] & CIP_DBS_MASK) >> CIP_DBS_SHIFT;
578 		/* avoid division by zero */
579 		if (data_block_quadlets == 0) {
580 			dev_err(&s->unit->device,
581 				"Detect invalid value in dbs field: %08X\n",
582 				cip_header[0]);
583 			return -EPROTO;
584 		}
585 		if (s->flags & CIP_WRONG_DBS)
586 			data_block_quadlets = s->data_block_quadlets;
587 
588 		*data_blocks = (payload_length / sizeof(__be32) - 2) /
589 							data_block_quadlets;
590 	}
591 
592 	/* Check data block counter continuity */
593 	dbc = cip_header[0] & CIP_DBC_MASK;
594 	if (*data_blocks == 0 && (s->flags & CIP_EMPTY_HAS_WRONG_DBC) &&
595 	    *data_block_counter != UINT_MAX)
596 		dbc = *data_block_counter;
597 
598 	if ((dbc == 0x00 && (s->flags & CIP_SKIP_DBC_ZERO_CHECK)) ||
599 	    *data_block_counter == UINT_MAX) {
600 		lost = false;
601 	} else if (!(s->flags & CIP_DBC_IS_END_EVENT)) {
602 		lost = dbc != *data_block_counter;
603 	} else {
604 		unsigned int dbc_interval;
605 
606 		if (*data_blocks > 0 && s->ctx_data.tx.dbc_interval > 0)
607 			dbc_interval = s->ctx_data.tx.dbc_interval;
608 		else
609 			dbc_interval = *data_blocks;
610 
611 		lost = dbc != ((*data_block_counter + dbc_interval) & 0xff);
612 	}
613 
614 	if (lost) {
615 		dev_err(&s->unit->device,
616 			"Detect discontinuity of CIP: %02X %02X\n",
617 			*data_block_counter, dbc);
618 		return -EIO;
619 	}
620 
621 	*data_block_counter = dbc;
622 
623 	*syt = cip_header[1] & CIP_SYT_MASK;
624 
625 	return 0;
626 }
627 
628 static int parse_ir_ctx_header(struct amdtp_stream *s, unsigned int cycle,
629 			       const __be32 *ctx_header,
630 			       unsigned int *payload_length,
631 			       unsigned int *data_blocks,
632 			       unsigned int *data_block_counter,
633 			       unsigned int *syt, unsigned int index)
634 {
635 	const __be32 *cip_header;
636 	int err;
637 
638 	*payload_length = be32_to_cpu(ctx_header[0]) >> ISO_DATA_LENGTH_SHIFT;
639 	if (*payload_length > s->ctx_data.tx.ctx_header_size +
640 					s->ctx_data.tx.max_ctx_payload_length) {
641 		dev_err(&s->unit->device,
642 			"Detect jumbo payload: %04x %04x\n",
643 			*payload_length, s->ctx_data.tx.max_ctx_payload_length);
644 		return -EIO;
645 	}
646 
647 	if (!(s->flags & CIP_NO_HEADER)) {
648 		cip_header = ctx_header + 2;
649 		err = check_cip_header(s, cip_header, *payload_length,
650 				       data_blocks, data_block_counter, syt);
651 		if (err < 0)
652 			return err;
653 	} else {
654 		cip_header = NULL;
655 		err = 0;
656 		*data_blocks = *payload_length / sizeof(__be32) /
657 			       s->data_block_quadlets;
658 		*syt = 0;
659 
660 		if (*data_block_counter == UINT_MAX)
661 			*data_block_counter = 0;
662 	}
663 
664 	trace_amdtp_packet(s, cycle, cip_header, *payload_length, *data_blocks,
665 			   *data_block_counter, index);
666 
667 	return err;
668 }
669 
670 // In CYCLE_TIMER register of IEEE 1394, 7 bits are used to represent second. On
671 // the other hand, in DMA descriptors of 1394 OHCI, 3 bits are used to represent
672 // it. Thus, via Linux firewire subsystem, we can get the 3 bits for second.
673 static inline u32 compute_cycle_count(__be32 ctx_header_tstamp)
674 {
675 	u32 tstamp = be32_to_cpu(ctx_header_tstamp) & HEADER_TSTAMP_MASK;
676 	return (((tstamp >> 13) & 0x07) * 8000) + (tstamp & 0x1fff);
677 }
678 
679 static inline u32 increment_cycle_count(u32 cycle, unsigned int addend)
680 {
681 	cycle += addend;
682 	if (cycle >= OHCI_MAX_SECOND * CYCLES_PER_SECOND)
683 		cycle -= OHCI_MAX_SECOND * CYCLES_PER_SECOND;
684 	return cycle;
685 }
686 
687 // Align to actual cycle count for the packet which is going to be scheduled.
688 // This module queued the same number of isochronous cycle as the size of queue
689 // to kip isochronous cycle, therefore it's OK to just increment the cycle by
690 // the size of queue for scheduled cycle.
691 static inline u32 compute_it_cycle(const __be32 ctx_header_tstamp,
692 				   unsigned int queue_size)
693 {
694 	u32 cycle = compute_cycle_count(ctx_header_tstamp);
695 	return increment_cycle_count(cycle, queue_size);
696 }
697 
698 static int generate_device_pkt_descs(struct amdtp_stream *s,
699 				     struct pkt_desc *descs,
700 				     const __be32 *ctx_header,
701 				     unsigned int packets)
702 {
703 	unsigned int dbc = s->data_block_counter;
704 	int i;
705 	int err;
706 
707 	for (i = 0; i < packets; ++i) {
708 		struct pkt_desc *desc = descs + i;
709 		unsigned int index = (s->packet_index + i) % s->queue_size;
710 		unsigned int cycle;
711 		unsigned int payload_length;
712 		unsigned int data_blocks;
713 		unsigned int syt;
714 
715 		cycle = compute_cycle_count(ctx_header[1]);
716 
717 		err = parse_ir_ctx_header(s, cycle, ctx_header, &payload_length,
718 					  &data_blocks, &dbc, &syt, i);
719 		if (err < 0)
720 			return err;
721 
722 		desc->cycle = cycle;
723 		desc->syt = syt;
724 		desc->data_blocks = data_blocks;
725 		desc->data_block_counter = dbc;
726 		desc->ctx_payload = s->buffer.packets[index].buffer;
727 
728 		if (!(s->flags & CIP_DBC_IS_END_EVENT))
729 			dbc = (dbc + desc->data_blocks) & 0xff;
730 
731 		ctx_header +=
732 			s->ctx_data.tx.ctx_header_size / sizeof(*ctx_header);
733 	}
734 
735 	s->data_block_counter = dbc;
736 
737 	return 0;
738 }
739 
740 static unsigned int compute_syt(unsigned int syt_offset, unsigned int cycle,
741 				unsigned int transfer_delay)
742 {
743 	unsigned int syt;
744 
745 	syt_offset += transfer_delay;
746 	syt = ((cycle + syt_offset / TICKS_PER_CYCLE) << 12) |
747 	      (syt_offset % TICKS_PER_CYCLE);
748 	return syt & CIP_SYT_MASK;
749 }
750 
751 static void generate_pkt_descs(struct amdtp_stream *s, struct pkt_desc *descs,
752 			       const __be32 *ctx_header, unsigned int packets,
753 			       const struct seq_desc *seq_descs,
754 			       unsigned int seq_size)
755 {
756 	unsigned int dbc = s->data_block_counter;
757 	unsigned int seq_index = s->ctx_data.rx.seq_index;
758 	int i;
759 
760 	for (i = 0; i < packets; ++i) {
761 		struct pkt_desc *desc = descs + i;
762 		unsigned int index = (s->packet_index + i) % s->queue_size;
763 		const struct seq_desc *seq = seq_descs + seq_index;
764 		unsigned int syt;
765 
766 		desc->cycle = compute_it_cycle(*ctx_header, s->queue_size);
767 
768 		syt = seq->syt_offset;
769 		if (syt != CIP_SYT_NO_INFO) {
770 			syt = compute_syt(syt, desc->cycle,
771 					  s->ctx_data.rx.transfer_delay);
772 		}
773 		desc->syt = syt;
774 		desc->data_blocks = seq->data_blocks;
775 
776 		if (s->flags & CIP_DBC_IS_END_EVENT)
777 			dbc = (dbc + desc->data_blocks) & 0xff;
778 
779 		desc->data_block_counter = dbc;
780 
781 		if (!(s->flags & CIP_DBC_IS_END_EVENT))
782 			dbc = (dbc + desc->data_blocks) & 0xff;
783 
784 		desc->ctx_payload = s->buffer.packets[index].buffer;
785 
786 		seq_index = (seq_index + 1) % seq_size;
787 
788 		++ctx_header;
789 	}
790 
791 	s->data_block_counter = dbc;
792 	s->ctx_data.rx.seq_index = seq_index;
793 }
794 
795 static inline void cancel_stream(struct amdtp_stream *s)
796 {
797 	s->packet_index = -1;
798 	if (current_work() == &s->period_work)
799 		amdtp_stream_pcm_abort(s);
800 	WRITE_ONCE(s->pcm_buffer_pointer, SNDRV_PCM_POS_XRUN);
801 }
802 
803 static void process_ctx_payloads(struct amdtp_stream *s,
804 				 const struct pkt_desc *descs,
805 				 unsigned int packets)
806 {
807 	struct snd_pcm_substream *pcm;
808 	unsigned int pcm_frames;
809 
810 	pcm = READ_ONCE(s->pcm);
811 	pcm_frames = s->process_ctx_payloads(s, descs, packets, pcm);
812 	if (pcm)
813 		update_pcm_pointers(s, pcm, pcm_frames);
814 }
815 
816 static void out_stream_callback(struct fw_iso_context *context, u32 tstamp,
817 				size_t header_length, void *header,
818 				void *private_data)
819 {
820 	struct amdtp_stream *s = private_data;
821 	const struct amdtp_domain *d = s->domain;
822 	const __be32 *ctx_header = header;
823 	unsigned int events_per_period = s->ctx_data.rx.events_per_period;
824 	unsigned int event_count = s->ctx_data.rx.event_count;
825 	unsigned int packets;
826 	int i;
827 
828 	if (s->packet_index < 0)
829 		return;
830 
831 	// Calculate the number of packets in buffer and check XRUN.
832 	packets = header_length / sizeof(*ctx_header);
833 
834 	generate_pkt_descs(s, s->pkt_descs, ctx_header, packets, d->seq_descs,
835 			   d->seq_size);
836 
837 	process_ctx_payloads(s, s->pkt_descs, packets);
838 
839 	for (i = 0; i < packets; ++i) {
840 		const struct pkt_desc *desc = s->pkt_descs + i;
841 		unsigned int syt;
842 		struct {
843 			struct fw_iso_packet params;
844 			__be32 header[IT_PKT_HEADER_SIZE_CIP / sizeof(__be32)];
845 		} template = { {0}, {0} };
846 		bool sched_irq = false;
847 
848 		if (s->ctx_data.rx.syt_override < 0)
849 			syt = desc->syt;
850 		else
851 			syt = s->ctx_data.rx.syt_override;
852 
853 		build_it_pkt_header(s, desc->cycle, &template.params,
854 				    desc->data_blocks, desc->data_block_counter,
855 				    syt, i);
856 
857 		if (s == s->domain->irq_target) {
858 			event_count += desc->data_blocks;
859 			if (event_count >= events_per_period) {
860 				event_count -= events_per_period;
861 				sched_irq = true;
862 			}
863 		}
864 
865 		if (queue_out_packet(s, &template.params, sched_irq) < 0) {
866 			cancel_stream(s);
867 			return;
868 		}
869 	}
870 
871 	s->ctx_data.rx.event_count = event_count;
872 }
873 
874 static void in_stream_callback(struct fw_iso_context *context, u32 tstamp,
875 			       size_t header_length, void *header,
876 			       void *private_data)
877 {
878 	struct amdtp_stream *s = private_data;
879 	__be32 *ctx_header = header;
880 	unsigned int packets;
881 	int i;
882 	int err;
883 
884 	if (s->packet_index < 0)
885 		return;
886 
887 	// Calculate the number of packets in buffer and check XRUN.
888 	packets = header_length / s->ctx_data.tx.ctx_header_size;
889 
890 	err = generate_device_pkt_descs(s, s->pkt_descs, ctx_header, packets);
891 	if (err < 0) {
892 		if (err != -EAGAIN) {
893 			cancel_stream(s);
894 			return;
895 		}
896 	} else {
897 		process_ctx_payloads(s, s->pkt_descs, packets);
898 	}
899 
900 	for (i = 0; i < packets; ++i) {
901 		struct fw_iso_packet params = {0};
902 
903 		if (queue_in_packet(s, &params) < 0) {
904 			cancel_stream(s);
905 			return;
906 		}
907 	}
908 }
909 
910 static void pool_ideal_seq_descs(struct amdtp_domain *d, unsigned int packets)
911 {
912 	struct amdtp_stream *irq_target = d->irq_target;
913 	unsigned int seq_tail = d->seq_tail;
914 	unsigned int seq_size = d->seq_size;
915 	unsigned int min_avail;
916 	struct amdtp_stream *s;
917 
918 	min_avail = d->seq_size;
919 	list_for_each_entry(s, &d->streams, list) {
920 		unsigned int seq_index;
921 		unsigned int avail;
922 
923 		if (s->direction == AMDTP_IN_STREAM)
924 			continue;
925 
926 		seq_index = s->ctx_data.rx.seq_index;
927 		avail = d->seq_tail;
928 		if (seq_index > avail)
929 			avail += d->seq_size;
930 		avail -= seq_index;
931 
932 		if (avail < min_avail)
933 			min_avail = avail;
934 	}
935 
936 	while (min_avail < packets) {
937 		struct seq_desc *desc = d->seq_descs + seq_tail;
938 
939 		desc->syt_offset = calculate_syt_offset(&d->last_syt_offset,
940 					&d->syt_offset_state, irq_target->sfc);
941 		desc->data_blocks = calculate_data_blocks(&d->data_block_state,
942 				!!(irq_target->flags & CIP_BLOCKING),
943 				desc->syt_offset == CIP_SYT_NO_INFO,
944 				irq_target->syt_interval, irq_target->sfc);
945 
946 		++seq_tail;
947 		seq_tail %= seq_size;
948 
949 		++min_avail;
950 	}
951 
952 	d->seq_tail = seq_tail;
953 }
954 
955 static void irq_target_callback(struct fw_iso_context *context, u32 tstamp,
956 				size_t header_length, void *header,
957 				void *private_data)
958 {
959 	struct amdtp_stream *irq_target = private_data;
960 	struct amdtp_domain *d = irq_target->domain;
961 	unsigned int packets = header_length / sizeof(__be32);
962 	struct amdtp_stream *s;
963 
964 	// Record enough entries with extra 3 cycles at least.
965 	pool_ideal_seq_descs(d, packets + 3);
966 
967 	out_stream_callback(context, tstamp, header_length, header, irq_target);
968 	if (amdtp_streaming_error(irq_target))
969 		goto error;
970 
971 	list_for_each_entry(s, &d->streams, list) {
972 		if (s != irq_target && amdtp_stream_running(s)) {
973 			fw_iso_context_flush_completions(s->context);
974 			if (amdtp_streaming_error(s))
975 				goto error;
976 		}
977 	}
978 
979 	return;
980 error:
981 	if (amdtp_stream_running(irq_target))
982 		cancel_stream(irq_target);
983 
984 	list_for_each_entry(s, &d->streams, list) {
985 		if (amdtp_stream_running(s))
986 			cancel_stream(s);
987 	}
988 }
989 
990 // this is executed one time.
991 static void amdtp_stream_first_callback(struct fw_iso_context *context,
992 					u32 tstamp, size_t header_length,
993 					void *header, void *private_data)
994 {
995 	struct amdtp_stream *s = private_data;
996 	const __be32 *ctx_header = header;
997 	u32 cycle;
998 
999 	/*
1000 	 * For in-stream, first packet has come.
1001 	 * For out-stream, prepared to transmit first packet
1002 	 */
1003 	s->callbacked = true;
1004 	wake_up(&s->callback_wait);
1005 
1006 	if (s->direction == AMDTP_IN_STREAM) {
1007 		cycle = compute_cycle_count(ctx_header[1]);
1008 
1009 		context->callback.sc = in_stream_callback;
1010 	} else {
1011 		cycle = compute_it_cycle(*ctx_header, s->queue_size);
1012 
1013 		if (s == s->domain->irq_target)
1014 			context->callback.sc = irq_target_callback;
1015 		else
1016 			context->callback.sc = out_stream_callback;
1017 	}
1018 
1019 	s->start_cycle = cycle;
1020 
1021 	context->callback.sc(context, tstamp, header_length, header, s);
1022 }
1023 
1024 /**
1025  * amdtp_stream_start - start transferring packets
1026  * @s: the AMDTP stream to start
1027  * @channel: the isochronous channel on the bus
1028  * @speed: firewire speed code
1029  * @start_cycle: the isochronous cycle to start the context. Start immediately
1030  *		 if negative value is given.
1031  * @queue_size: The number of packets in the queue.
1032  * @idle_irq_interval: the interval to queue packet during initial state.
1033  *
1034  * The stream cannot be started until it has been configured with
1035  * amdtp_stream_set_parameters() and it must be started before any PCM or MIDI
1036  * device can be started.
1037  */
1038 static int amdtp_stream_start(struct amdtp_stream *s, int channel, int speed,
1039 			      int start_cycle, unsigned int queue_size,
1040 			      unsigned int idle_irq_interval)
1041 {
1042 	bool is_irq_target = (s == s->domain->irq_target);
1043 	unsigned int ctx_header_size;
1044 	unsigned int max_ctx_payload_size;
1045 	enum dma_data_direction dir;
1046 	int type, tag, err;
1047 
1048 	mutex_lock(&s->mutex);
1049 
1050 	if (WARN_ON(amdtp_stream_running(s) ||
1051 		    (s->data_block_quadlets < 1))) {
1052 		err = -EBADFD;
1053 		goto err_unlock;
1054 	}
1055 
1056 	if (s->direction == AMDTP_IN_STREAM) {
1057 		// NOTE: IT context should be used for constant IRQ.
1058 		if (is_irq_target) {
1059 			err = -EINVAL;
1060 			goto err_unlock;
1061 		}
1062 
1063 		s->data_block_counter = UINT_MAX;
1064 	} else {
1065 		s->data_block_counter = 0;
1066 	}
1067 
1068 	/* initialize packet buffer */
1069 	if (s->direction == AMDTP_IN_STREAM) {
1070 		dir = DMA_FROM_DEVICE;
1071 		type = FW_ISO_CONTEXT_RECEIVE;
1072 		if (!(s->flags & CIP_NO_HEADER))
1073 			ctx_header_size = IR_CTX_HEADER_SIZE_CIP;
1074 		else
1075 			ctx_header_size = IR_CTX_HEADER_SIZE_NO_CIP;
1076 
1077 		max_ctx_payload_size = amdtp_stream_get_max_payload(s) -
1078 				       ctx_header_size;
1079 	} else {
1080 		dir = DMA_TO_DEVICE;
1081 		type = FW_ISO_CONTEXT_TRANSMIT;
1082 		ctx_header_size = 0;	// No effect for IT context.
1083 
1084 		max_ctx_payload_size = amdtp_stream_get_max_payload(s);
1085 		if (!(s->flags & CIP_NO_HEADER))
1086 			max_ctx_payload_size -= IT_PKT_HEADER_SIZE_CIP;
1087 	}
1088 
1089 	err = iso_packets_buffer_init(&s->buffer, s->unit, queue_size,
1090 				      max_ctx_payload_size, dir);
1091 	if (err < 0)
1092 		goto err_unlock;
1093 	s->queue_size = queue_size;
1094 
1095 	s->context = fw_iso_context_create(fw_parent_device(s->unit)->card,
1096 					  type, channel, speed, ctx_header_size,
1097 					  amdtp_stream_first_callback, s);
1098 	if (IS_ERR(s->context)) {
1099 		err = PTR_ERR(s->context);
1100 		if (err == -EBUSY)
1101 			dev_err(&s->unit->device,
1102 				"no free stream on this controller\n");
1103 		goto err_buffer;
1104 	}
1105 
1106 	amdtp_stream_update(s);
1107 
1108 	if (s->direction == AMDTP_IN_STREAM) {
1109 		s->ctx_data.tx.max_ctx_payload_length = max_ctx_payload_size;
1110 		s->ctx_data.tx.ctx_header_size = ctx_header_size;
1111 	}
1112 
1113 	if (s->flags & CIP_NO_HEADER)
1114 		s->tag = TAG_NO_CIP_HEADER;
1115 	else
1116 		s->tag = TAG_CIP;
1117 
1118 	s->pkt_descs = kcalloc(s->queue_size, sizeof(*s->pkt_descs),
1119 			       GFP_KERNEL);
1120 	if (!s->pkt_descs) {
1121 		err = -ENOMEM;
1122 		goto err_context;
1123 	}
1124 
1125 	s->packet_index = 0;
1126 	do {
1127 		struct fw_iso_packet params;
1128 
1129 		if (s->direction == AMDTP_IN_STREAM) {
1130 			err = queue_in_packet(s, &params);
1131 		} else {
1132 			bool sched_irq = false;
1133 
1134 			params.header_length = 0;
1135 			params.payload_length = 0;
1136 
1137 			if (is_irq_target) {
1138 				sched_irq = !((s->packet_index + 1) %
1139 					      idle_irq_interval);
1140 			}
1141 
1142 			err = queue_out_packet(s, &params, sched_irq);
1143 		}
1144 		if (err < 0)
1145 			goto err_pkt_descs;
1146 	} while (s->packet_index > 0);
1147 
1148 	/* NOTE: TAG1 matches CIP. This just affects in stream. */
1149 	tag = FW_ISO_CONTEXT_MATCH_TAG1;
1150 	if ((s->flags & CIP_EMPTY_WITH_TAG0) || (s->flags & CIP_NO_HEADER))
1151 		tag |= FW_ISO_CONTEXT_MATCH_TAG0;
1152 
1153 	s->callbacked = false;
1154 	err = fw_iso_context_start(s->context, start_cycle, 0, tag);
1155 	if (err < 0)
1156 		goto err_pkt_descs;
1157 
1158 	mutex_unlock(&s->mutex);
1159 
1160 	return 0;
1161 err_pkt_descs:
1162 	kfree(s->pkt_descs);
1163 err_context:
1164 	fw_iso_context_destroy(s->context);
1165 	s->context = ERR_PTR(-1);
1166 err_buffer:
1167 	iso_packets_buffer_destroy(&s->buffer, s->unit);
1168 err_unlock:
1169 	mutex_unlock(&s->mutex);
1170 
1171 	return err;
1172 }
1173 
1174 /**
1175  * amdtp_domain_stream_pcm_pointer - get the PCM buffer position
1176  * @d: the AMDTP domain.
1177  * @s: the AMDTP stream that transports the PCM data
1178  *
1179  * Returns the current buffer position, in frames.
1180  */
1181 unsigned long amdtp_domain_stream_pcm_pointer(struct amdtp_domain *d,
1182 					      struct amdtp_stream *s)
1183 {
1184 	struct amdtp_stream *irq_target = d->irq_target;
1185 
1186 	if (irq_target && amdtp_stream_running(irq_target)) {
1187 		// This function is called in software IRQ context of
1188 		// period_work or process context.
1189 		//
1190 		// When the software IRQ context was scheduled by software IRQ
1191 		// context of IT contexts, queued packets were already handled.
1192 		// Therefore, no need to flush the queue in buffer furthermore.
1193 		//
1194 		// When the process context reach here, some packets will be
1195 		// already queued in the buffer. These packets should be handled
1196 		// immediately to keep better granularity of PCM pointer.
1197 		//
1198 		// Later, the process context will sometimes schedules software
1199 		// IRQ context of the period_work. Then, no need to flush the
1200 		// queue by the same reason as described in the above
1201 		if (current_work() != &s->period_work) {
1202 			// Queued packet should be processed without any kernel
1203 			// preemption to keep latency against bus cycle.
1204 			preempt_disable();
1205 			fw_iso_context_flush_completions(irq_target->context);
1206 			preempt_enable();
1207 		}
1208 	}
1209 
1210 	return READ_ONCE(s->pcm_buffer_pointer);
1211 }
1212 EXPORT_SYMBOL_GPL(amdtp_domain_stream_pcm_pointer);
1213 
1214 /**
1215  * amdtp_domain_stream_pcm_ack - acknowledge queued PCM frames
1216  * @d: the AMDTP domain.
1217  * @s: the AMDTP stream that transfers the PCM frames
1218  *
1219  * Returns zero always.
1220  */
1221 int amdtp_domain_stream_pcm_ack(struct amdtp_domain *d, struct amdtp_stream *s)
1222 {
1223 	struct amdtp_stream *irq_target = d->irq_target;
1224 
1225 	// Process isochronous packets for recent isochronous cycle to handle
1226 	// queued PCM frames.
1227 	if (irq_target && amdtp_stream_running(irq_target)) {
1228 		// Queued packet should be processed without any kernel
1229 		// preemption to keep latency against bus cycle.
1230 		preempt_disable();
1231 		fw_iso_context_flush_completions(irq_target->context);
1232 		preempt_enable();
1233 	}
1234 
1235 	return 0;
1236 }
1237 EXPORT_SYMBOL_GPL(amdtp_domain_stream_pcm_ack);
1238 
1239 /**
1240  * amdtp_stream_update - update the stream after a bus reset
1241  * @s: the AMDTP stream
1242  */
1243 void amdtp_stream_update(struct amdtp_stream *s)
1244 {
1245 	/* Precomputing. */
1246 	WRITE_ONCE(s->source_node_id_field,
1247                    (fw_parent_device(s->unit)->card->node_id << CIP_SID_SHIFT) & CIP_SID_MASK);
1248 }
1249 EXPORT_SYMBOL(amdtp_stream_update);
1250 
1251 /**
1252  * amdtp_stream_stop - stop sending packets
1253  * @s: the AMDTP stream to stop
1254  *
1255  * All PCM and MIDI devices of the stream must be stopped before the stream
1256  * itself can be stopped.
1257  */
1258 static void amdtp_stream_stop(struct amdtp_stream *s)
1259 {
1260 	mutex_lock(&s->mutex);
1261 
1262 	if (!amdtp_stream_running(s)) {
1263 		mutex_unlock(&s->mutex);
1264 		return;
1265 	}
1266 
1267 	cancel_work_sync(&s->period_work);
1268 	fw_iso_context_stop(s->context);
1269 	fw_iso_context_destroy(s->context);
1270 	s->context = ERR_PTR(-1);
1271 	iso_packets_buffer_destroy(&s->buffer, s->unit);
1272 	kfree(s->pkt_descs);
1273 
1274 	s->callbacked = false;
1275 
1276 	mutex_unlock(&s->mutex);
1277 }
1278 
1279 /**
1280  * amdtp_stream_pcm_abort - abort the running PCM device
1281  * @s: the AMDTP stream about to be stopped
1282  *
1283  * If the isochronous stream needs to be stopped asynchronously, call this
1284  * function first to stop the PCM device.
1285  */
1286 void amdtp_stream_pcm_abort(struct amdtp_stream *s)
1287 {
1288 	struct snd_pcm_substream *pcm;
1289 
1290 	pcm = READ_ONCE(s->pcm);
1291 	if (pcm)
1292 		snd_pcm_stop_xrun(pcm);
1293 }
1294 EXPORT_SYMBOL(amdtp_stream_pcm_abort);
1295 
1296 /**
1297  * amdtp_domain_init - initialize an AMDTP domain structure
1298  * @d: the AMDTP domain to initialize.
1299  */
1300 int amdtp_domain_init(struct amdtp_domain *d)
1301 {
1302 	INIT_LIST_HEAD(&d->streams);
1303 
1304 	d->events_per_period = 0;
1305 
1306 	d->seq_descs = NULL;
1307 
1308 	return 0;
1309 }
1310 EXPORT_SYMBOL_GPL(amdtp_domain_init);
1311 
1312 /**
1313  * amdtp_domain_destroy - destroy an AMDTP domain structure
1314  * @d: the AMDTP domain to destroy.
1315  */
1316 void amdtp_domain_destroy(struct amdtp_domain *d)
1317 {
1318 	// At present nothing to do.
1319 	return;
1320 }
1321 EXPORT_SYMBOL_GPL(amdtp_domain_destroy);
1322 
1323 /**
1324  * amdtp_domain_add_stream - register isoc context into the domain.
1325  * @d: the AMDTP domain.
1326  * @s: the AMDTP stream.
1327  * @channel: the isochronous channel on the bus.
1328  * @speed: firewire speed code.
1329  */
1330 int amdtp_domain_add_stream(struct amdtp_domain *d, struct amdtp_stream *s,
1331 			    int channel, int speed)
1332 {
1333 	struct amdtp_stream *tmp;
1334 
1335 	list_for_each_entry(tmp, &d->streams, list) {
1336 		if (s == tmp)
1337 			return -EBUSY;
1338 	}
1339 
1340 	list_add(&s->list, &d->streams);
1341 
1342 	s->channel = channel;
1343 	s->speed = speed;
1344 	s->domain = d;
1345 
1346 	return 0;
1347 }
1348 EXPORT_SYMBOL_GPL(amdtp_domain_add_stream);
1349 
1350 static int get_current_cycle_time(struct fw_card *fw_card, int *cur_cycle)
1351 {
1352 	int generation;
1353 	int rcode;
1354 	__be32 reg;
1355 	u32 data;
1356 
1357 	// This is a request to local 1394 OHCI controller and expected to
1358 	// complete without any event waiting.
1359 	generation = fw_card->generation;
1360 	smp_rmb();	// node_id vs. generation.
1361 	rcode = fw_run_transaction(fw_card, TCODE_READ_QUADLET_REQUEST,
1362 				   fw_card->node_id, generation, SCODE_100,
1363 				   CSR_REGISTER_BASE + CSR_CYCLE_TIME,
1364 				   &reg, sizeof(reg));
1365 	if (rcode != RCODE_COMPLETE)
1366 		return -EIO;
1367 
1368 	data = be32_to_cpu(reg);
1369 	*cur_cycle = data >> 12;
1370 
1371 	return 0;
1372 }
1373 
1374 /**
1375  * amdtp_domain_start - start sending packets for isoc context in the domain.
1376  * @d: the AMDTP domain.
1377  * @ir_delay_cycle: the cycle delay to start all IR contexts.
1378  */
1379 int amdtp_domain_start(struct amdtp_domain *d, unsigned int ir_delay_cycle)
1380 {
1381 	static const struct {
1382 		unsigned int data_block;
1383 		unsigned int syt_offset;
1384 	} *entry, initial_state[] = {
1385 		[CIP_SFC_32000]  = {  4, 3072 },
1386 		[CIP_SFC_48000]  = {  6, 1024 },
1387 		[CIP_SFC_96000]  = { 12, 1024 },
1388 		[CIP_SFC_192000] = { 24, 1024 },
1389 		[CIP_SFC_44100]  = {  0,   67 },
1390 		[CIP_SFC_88200]  = {  0,   67 },
1391 		[CIP_SFC_176400] = {  0,   67 },
1392 	};
1393 	unsigned int events_per_buffer = d->events_per_buffer;
1394 	unsigned int events_per_period = d->events_per_period;
1395 	unsigned int idle_irq_interval;
1396 	unsigned int queue_size;
1397 	struct amdtp_stream *s;
1398 	int cycle;
1399 	int err;
1400 
1401 	// Select an IT context as IRQ target.
1402 	list_for_each_entry(s, &d->streams, list) {
1403 		if (s->direction == AMDTP_OUT_STREAM)
1404 			break;
1405 	}
1406 	if (!s)
1407 		return -ENXIO;
1408 	d->irq_target = s;
1409 
1410 	// This is a case that AMDTP streams in domain run just for MIDI
1411 	// substream. Use the number of events equivalent to 10 msec as
1412 	// interval of hardware IRQ.
1413 	if (events_per_period == 0)
1414 		events_per_period = amdtp_rate_table[d->irq_target->sfc] / 100;
1415 	if (events_per_buffer == 0)
1416 		events_per_buffer = events_per_period * 3;
1417 
1418 	queue_size = DIV_ROUND_UP(CYCLES_PER_SECOND * events_per_buffer,
1419 				  amdtp_rate_table[d->irq_target->sfc]);
1420 
1421 	d->seq_descs = kcalloc(queue_size, sizeof(*d->seq_descs), GFP_KERNEL);
1422 	if (!d->seq_descs)
1423 		return -ENOMEM;
1424 	d->seq_size = queue_size;
1425 	d->seq_tail = 0;
1426 
1427 	entry = &initial_state[s->sfc];
1428 	d->data_block_state = entry->data_block;
1429 	d->syt_offset_state = entry->syt_offset;
1430 	d->last_syt_offset = TICKS_PER_CYCLE;
1431 
1432 	if (ir_delay_cycle > 0) {
1433 		struct fw_card *fw_card = fw_parent_device(s->unit)->card;
1434 
1435 		err = get_current_cycle_time(fw_card, &cycle);
1436 		if (err < 0)
1437 			goto error;
1438 
1439 		// No need to care overflow in cycle field because of enough
1440 		// width.
1441 		cycle += ir_delay_cycle;
1442 
1443 		// Round up to sec field.
1444 		if ((cycle & 0x00001fff) >= CYCLES_PER_SECOND) {
1445 			unsigned int sec;
1446 
1447 			// The sec field can overflow.
1448 			sec = (cycle & 0xffffe000) >> 13;
1449 			cycle = (++sec << 13) |
1450 				((cycle & 0x00001fff) / CYCLES_PER_SECOND);
1451 		}
1452 
1453 		// In OHCI 1394 specification, lower 2 bits are available for
1454 		// sec field.
1455 		cycle &= 0x00007fff;
1456 	} else {
1457 		cycle = -1;
1458 	}
1459 
1460 	list_for_each_entry(s, &d->streams, list) {
1461 		int cycle_match;
1462 
1463 		if (s->direction == AMDTP_IN_STREAM) {
1464 			cycle_match = cycle;
1465 		} else {
1466 			// IT context starts immediately.
1467 			cycle_match = -1;
1468 			s->ctx_data.rx.seq_index = 0;
1469 		}
1470 
1471 		if (s != d->irq_target) {
1472 			err = amdtp_stream_start(s, s->channel, s->speed,
1473 						 cycle_match, queue_size, 0);
1474 			if (err < 0)
1475 				goto error;
1476 		}
1477 	}
1478 
1479 	s = d->irq_target;
1480 	s->ctx_data.rx.events_per_period = events_per_period;
1481 	s->ctx_data.rx.event_count = 0;
1482 	s->ctx_data.rx.seq_index = 0;
1483 
1484 	idle_irq_interval = DIV_ROUND_UP(CYCLES_PER_SECOND * events_per_period,
1485 					 amdtp_rate_table[d->irq_target->sfc]);
1486 	err = amdtp_stream_start(s, s->channel, s->speed, -1, queue_size,
1487 				 idle_irq_interval);
1488 	if (err < 0)
1489 		goto error;
1490 
1491 	return 0;
1492 error:
1493 	list_for_each_entry(s, &d->streams, list)
1494 		amdtp_stream_stop(s);
1495 	kfree(d->seq_descs);
1496 	d->seq_descs = NULL;
1497 	return err;
1498 }
1499 EXPORT_SYMBOL_GPL(amdtp_domain_start);
1500 
1501 /**
1502  * amdtp_domain_stop - stop sending packets for isoc context in the same domain.
1503  * @d: the AMDTP domain to which the isoc contexts belong.
1504  */
1505 void amdtp_domain_stop(struct amdtp_domain *d)
1506 {
1507 	struct amdtp_stream *s, *next;
1508 
1509 	if (d->irq_target)
1510 		amdtp_stream_stop(d->irq_target);
1511 
1512 	list_for_each_entry_safe(s, next, &d->streams, list) {
1513 		list_del(&s->list);
1514 
1515 		if (s != d->irq_target)
1516 			amdtp_stream_stop(s);
1517 	}
1518 
1519 	d->events_per_period = 0;
1520 	d->irq_target = NULL;
1521 
1522 	kfree(d->seq_descs);
1523 	d->seq_descs = NULL;
1524 }
1525 EXPORT_SYMBOL_GPL(amdtp_domain_stop);
1526