1 // SPDX-License-Identifier: GPL-2.0-only
3 * Audio and Music Data Transmission Protocol (IEC 61883-6) streams
4 * with Common Isochronous Packet (IEC 61883-1) headers
6 * Copyright (c) Clemens Ladisch <clemens@ladisch.de>
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"
19 #define TICKS_PER_CYCLE 3072
20 #define CYCLES_PER_SECOND 8000
21 #define TICKS_PER_SECOND (TICKS_PER_CYCLE * CYCLES_PER_SECOND)
23 #define OHCI_SECOND_MODULUS 8
25 /* Always support Linux tracing subsystem. */
26 #define CREATE_TRACE_POINTS
27 #include "amdtp-stream-trace.h"
29 #define TRANSFER_DELAY_TICKS 0x2e00 /* 479.17 microseconds */
31 /* isochronous header parameters */
32 #define ISO_DATA_LENGTH_SHIFT 16
33 #define TAG_NO_CIP_HEADER 0
36 // Common Isochronous Packet (CIP) header parameters. Use two quadlets CIP header when supported.
37 #define CIP_HEADER_QUADLETS 2
38 #define CIP_EOH_SHIFT 31
39 #define CIP_EOH (1u << CIP_EOH_SHIFT)
40 #define CIP_EOH_MASK 0x80000000
41 #define CIP_SID_SHIFT 24
42 #define CIP_SID_MASK 0x3f000000
43 #define CIP_DBS_MASK 0x00ff0000
44 #define CIP_DBS_SHIFT 16
45 #define CIP_SPH_MASK 0x00000400
46 #define CIP_SPH_SHIFT 10
47 #define CIP_DBC_MASK 0x000000ff
48 #define CIP_FMT_SHIFT 24
49 #define CIP_FMT_MASK 0x3f000000
50 #define CIP_FDF_MASK 0x00ff0000
51 #define CIP_FDF_SHIFT 16
52 #define CIP_FDF_NO_DATA 0xff
53 #define CIP_SYT_MASK 0x0000ffff
54 #define CIP_SYT_NO_INFO 0xffff
55 #define CIP_SYT_CYCLE_MODULUS 16
56 #define CIP_NO_DATA ((CIP_FDF_NO_DATA << CIP_FDF_SHIFT) | CIP_SYT_NO_INFO)
58 #define CIP_HEADER_SIZE (sizeof(__be32) * CIP_HEADER_QUADLETS)
60 /* Audio and Music transfer protocol specific parameters */
61 #define CIP_FMT_AM 0x10
62 #define AMDTP_FDF_NO_DATA 0xff
64 // For iso header and tstamp.
65 #define IR_CTX_HEADER_DEFAULT_QUADLETS 2
67 #define IR_CTX_HEADER_SIZE_NO_CIP (sizeof(__be32) * IR_CTX_HEADER_DEFAULT_QUADLETS)
68 // Add two quadlets CIP header.
69 #define IR_CTX_HEADER_SIZE_CIP (IR_CTX_HEADER_SIZE_NO_CIP + CIP_HEADER_SIZE)
70 #define HEADER_TSTAMP_MASK 0x0000ffff
72 #define IT_PKT_HEADER_SIZE_CIP CIP_HEADER_SIZE
73 #define IT_PKT_HEADER_SIZE_NO_CIP 0 // Nothing.
75 // The initial firmware of OXFW970 can postpone transmission of packet during finishing
76 // asynchronous transaction. This module accepts 5 cycles to skip as maximum to avoid buffer
77 // overrun. Actual device can skip more, then this module stops the packet streaming.
78 #define IR_JUMBO_PAYLOAD_MAX_SKIP_CYCLES 5
81 * amdtp_stream_init - initialize an AMDTP stream structure
82 * @s: the AMDTP stream to initialize
83 * @unit: the target of the stream
84 * @dir: the direction of stream
85 * @flags: the details of the streaming protocol consist of cip_flags enumeration-constants.
86 * @fmt: the value of fmt field in CIP header
87 * @process_ctx_payloads: callback handler to process payloads of isoc context
88 * @protocol_size: the size to allocate newly for protocol
90 int amdtp_stream_init(struct amdtp_stream *s, struct fw_unit *unit,
91 enum amdtp_stream_direction dir, unsigned int flags,
93 amdtp_stream_process_ctx_payloads_t process_ctx_payloads,
94 unsigned int protocol_size)
96 if (process_ctx_payloads == NULL)
99 s->protocol = kzalloc(protocol_size, GFP_KERNEL);
106 s->context = ERR_PTR(-1);
107 mutex_init(&s->mutex);
110 init_waitqueue_head(&s->ready_wait);
113 s->process_ctx_payloads = process_ctx_payloads;
117 EXPORT_SYMBOL(amdtp_stream_init);
120 * amdtp_stream_destroy - free stream resources
121 * @s: the AMDTP stream to destroy
123 void amdtp_stream_destroy(struct amdtp_stream *s)
125 /* Not initialized. */
126 if (s->protocol == NULL)
129 WARN_ON(amdtp_stream_running(s));
131 mutex_destroy(&s->mutex);
133 EXPORT_SYMBOL(amdtp_stream_destroy);
135 const unsigned int amdtp_syt_intervals[CIP_SFC_COUNT] = {
139 [CIP_SFC_88200] = 16,
140 [CIP_SFC_96000] = 16,
141 [CIP_SFC_176400] = 32,
142 [CIP_SFC_192000] = 32,
144 EXPORT_SYMBOL(amdtp_syt_intervals);
146 const unsigned int amdtp_rate_table[CIP_SFC_COUNT] = {
147 [CIP_SFC_32000] = 32000,
148 [CIP_SFC_44100] = 44100,
149 [CIP_SFC_48000] = 48000,
150 [CIP_SFC_88200] = 88200,
151 [CIP_SFC_96000] = 96000,
152 [CIP_SFC_176400] = 176400,
153 [CIP_SFC_192000] = 192000,
155 EXPORT_SYMBOL(amdtp_rate_table);
157 static int apply_constraint_to_size(struct snd_pcm_hw_params *params,
158 struct snd_pcm_hw_rule *rule)
160 struct snd_interval *s = hw_param_interval(params, rule->var);
161 const struct snd_interval *r =
162 hw_param_interval_c(params, SNDRV_PCM_HW_PARAM_RATE);
163 struct snd_interval t = {0};
164 unsigned int step = 0;
167 for (i = 0; i < CIP_SFC_COUNT; ++i) {
168 if (snd_interval_test(r, amdtp_rate_table[i]))
169 step = max(step, amdtp_syt_intervals[i]);
172 t.min = roundup(s->min, step);
173 t.max = rounddown(s->max, step);
176 return snd_interval_refine(s, &t);
180 * amdtp_stream_add_pcm_hw_constraints - add hw constraints for PCM substream
181 * @s: the AMDTP stream, which must be initialized.
182 * @runtime: the PCM substream runtime
184 int amdtp_stream_add_pcm_hw_constraints(struct amdtp_stream *s,
185 struct snd_pcm_runtime *runtime)
187 struct snd_pcm_hardware *hw = &runtime->hw;
188 unsigned int ctx_header_size;
189 unsigned int maximum_usec_per_period;
192 hw->info = SNDRV_PCM_INFO_BLOCK_TRANSFER |
193 SNDRV_PCM_INFO_INTERLEAVED |
194 SNDRV_PCM_INFO_JOINT_DUPLEX |
195 SNDRV_PCM_INFO_MMAP |
196 SNDRV_PCM_INFO_MMAP_VALID |
197 SNDRV_PCM_INFO_NO_PERIOD_WAKEUP;
200 hw->periods_max = UINT_MAX;
202 /* bytes for a frame */
203 hw->period_bytes_min = 4 * hw->channels_max;
205 /* Just to prevent from allocating much pages. */
206 hw->period_bytes_max = hw->period_bytes_min * 2048;
207 hw->buffer_bytes_max = hw->period_bytes_max * hw->periods_min;
209 // Linux driver for 1394 OHCI controller voluntarily flushes isoc
210 // context when total size of accumulated context header reaches
211 // PAGE_SIZE. This kicks work for the isoc context and brings
212 // callback in the middle of scheduled interrupts.
213 // Although AMDTP streams in the same domain use the same events per
214 // IRQ, use the largest size of context header between IT/IR contexts.
215 // Here, use the value of context header in IR context is for both
217 if (!(s->flags & CIP_NO_HEADER))
218 ctx_header_size = IR_CTX_HEADER_SIZE_CIP;
220 ctx_header_size = IR_CTX_HEADER_SIZE_NO_CIP;
221 maximum_usec_per_period = USEC_PER_SEC * PAGE_SIZE /
222 CYCLES_PER_SECOND / ctx_header_size;
224 // In IEC 61883-6, one isoc packet can transfer events up to the value
225 // of syt interval. This comes from the interval of isoc cycle. As 1394
226 // OHCI controller can generate hardware IRQ per isoc packet, the
227 // interval is 125 usec.
228 // However, there are two ways of transmission in IEC 61883-6; blocking
229 // and non-blocking modes. In blocking mode, the sequence of isoc packet
230 // includes 'empty' or 'NODATA' packets which include no event. In
231 // non-blocking mode, the number of events per packet is variable up to
233 // Due to the above protocol design, the minimum PCM frames per
234 // interrupt should be double of the value of syt interval, thus it is
236 err = snd_pcm_hw_constraint_minmax(runtime,
237 SNDRV_PCM_HW_PARAM_PERIOD_TIME,
238 250, maximum_usec_per_period);
242 /* Non-Blocking stream has no more constraints */
243 if (!(s->flags & CIP_BLOCKING))
247 * One AMDTP packet can include some frames. In blocking mode, the
248 * number equals to SYT_INTERVAL. So the number is 8, 16 or 32,
249 * depending on its sampling rate. For accurate period interrupt, it's
250 * preferrable to align period/buffer sizes to current SYT_INTERVAL.
252 err = snd_pcm_hw_rule_add(runtime, 0, SNDRV_PCM_HW_PARAM_PERIOD_SIZE,
253 apply_constraint_to_size, NULL,
254 SNDRV_PCM_HW_PARAM_PERIOD_SIZE,
255 SNDRV_PCM_HW_PARAM_RATE, -1);
258 err = snd_pcm_hw_rule_add(runtime, 0, SNDRV_PCM_HW_PARAM_BUFFER_SIZE,
259 apply_constraint_to_size, NULL,
260 SNDRV_PCM_HW_PARAM_BUFFER_SIZE,
261 SNDRV_PCM_HW_PARAM_RATE, -1);
267 EXPORT_SYMBOL(amdtp_stream_add_pcm_hw_constraints);
270 * amdtp_stream_set_parameters - set stream parameters
271 * @s: the AMDTP stream to configure
272 * @rate: the sample rate
273 * @data_block_quadlets: the size of a data block in quadlet unit
274 * @pcm_frame_multiplier: the multiplier to compute the number of PCM frames by the number of AMDTP
277 * The parameters must be set before the stream is started, and must not be
278 * changed while the stream is running.
280 int amdtp_stream_set_parameters(struct amdtp_stream *s, unsigned int rate,
281 unsigned int data_block_quadlets, unsigned int pcm_frame_multiplier)
285 for (sfc = 0; sfc < ARRAY_SIZE(amdtp_rate_table); ++sfc) {
286 if (amdtp_rate_table[sfc] == rate)
289 if (sfc == ARRAY_SIZE(amdtp_rate_table))
293 s->data_block_quadlets = data_block_quadlets;
294 s->syt_interval = amdtp_syt_intervals[sfc];
296 // default buffering in the device.
297 s->transfer_delay = TRANSFER_DELAY_TICKS - TICKS_PER_CYCLE;
299 // additional buffering needed to adjust for no-data packets.
300 if (s->flags & CIP_BLOCKING)
301 s->transfer_delay += TICKS_PER_SECOND * s->syt_interval / rate;
303 s->pcm_frame_multiplier = pcm_frame_multiplier;
307 EXPORT_SYMBOL(amdtp_stream_set_parameters);
309 // The CIP header is processed in context header apart from context payload.
310 static int amdtp_stream_get_max_ctx_payload_size(struct amdtp_stream *s)
312 unsigned int multiplier;
314 if (s->flags & CIP_JUMBO_PAYLOAD)
315 multiplier = IR_JUMBO_PAYLOAD_MAX_SKIP_CYCLES;
319 return s->syt_interval * s->data_block_quadlets * sizeof(__be32) * multiplier;
323 * amdtp_stream_get_max_payload - get the stream's packet size
324 * @s: the AMDTP stream
326 * This function must not be called before the stream has been configured
327 * with amdtp_stream_set_parameters().
329 unsigned int amdtp_stream_get_max_payload(struct amdtp_stream *s)
331 unsigned int cip_header_size;
333 if (!(s->flags & CIP_NO_HEADER))
334 cip_header_size = CIP_HEADER_SIZE;
338 return cip_header_size + amdtp_stream_get_max_ctx_payload_size(s);
340 EXPORT_SYMBOL(amdtp_stream_get_max_payload);
343 * amdtp_stream_pcm_prepare - prepare PCM device for running
344 * @s: the AMDTP stream
346 * This function should be called from the PCM device's .prepare callback.
348 void amdtp_stream_pcm_prepare(struct amdtp_stream *s)
350 s->pcm_buffer_pointer = 0;
351 s->pcm_period_pointer = 0;
353 EXPORT_SYMBOL(amdtp_stream_pcm_prepare);
355 #define prev_packet_desc(s, desc) \
356 list_prev_entry_circular(desc, &s->packet_descs_list, link)
358 static void pool_blocking_data_blocks(struct amdtp_stream *s, struct seq_desc *descs,
359 unsigned int size, unsigned int pos, unsigned int count)
361 const unsigned int syt_interval = s->syt_interval;
364 for (i = 0; i < count; ++i) {
365 struct seq_desc *desc = descs + pos;
367 if (desc->syt_offset != CIP_SYT_NO_INFO)
368 desc->data_blocks = syt_interval;
370 desc->data_blocks = 0;
372 pos = (pos + 1) % size;
376 static void pool_ideal_nonblocking_data_blocks(struct amdtp_stream *s, struct seq_desc *descs,
377 unsigned int size, unsigned int pos,
380 const enum cip_sfc sfc = s->sfc;
381 unsigned int state = s->ctx_data.rx.data_block_state;
384 for (i = 0; i < count; ++i) {
385 struct seq_desc *desc = descs + pos;
387 if (!cip_sfc_is_base_44100(sfc)) {
388 // Sample_rate / 8000 is an integer, and precomputed.
389 desc->data_blocks = state;
391 unsigned int phase = state;
394 * This calculates the number of data blocks per packet so that
395 * 1) the overall rate is correct and exactly synchronized to
397 * 2) packets with a rounded-up number of blocks occur as early
398 * as possible in the sequence (to prevent underruns of the
401 if (sfc == CIP_SFC_44100)
402 /* 6 6 5 6 5 6 5 ... */
403 desc->data_blocks = 5 + ((phase & 1) ^ (phase == 0 || phase >= 40));
405 /* 12 11 11 11 11 ... or 23 22 22 22 22 ... */
406 desc->data_blocks = 11 * (sfc >> 1) + (phase == 0);
407 if (++phase >= (80 >> (sfc >> 1)))
412 pos = (pos + 1) % size;
415 s->ctx_data.rx.data_block_state = state;
418 static unsigned int calculate_syt_offset(unsigned int *last_syt_offset,
419 unsigned int *syt_offset_state, enum cip_sfc sfc)
421 unsigned int syt_offset;
423 if (*last_syt_offset < TICKS_PER_CYCLE) {
424 if (!cip_sfc_is_base_44100(sfc))
425 syt_offset = *last_syt_offset + *syt_offset_state;
428 * The time, in ticks, of the n'th SYT_INTERVAL sample is:
429 * n * SYT_INTERVAL * 24576000 / sample_rate
430 * Modulo TICKS_PER_CYCLE, the difference between successive
431 * elements is about 1386.23. Rounding the results of this
432 * formula to the SYT precision results in a sequence of
433 * differences that begins with:
434 * 1386 1386 1387 1386 1386 1386 1387 1386 1386 1386 1387 ...
435 * This code generates _exactly_ the same sequence.
437 unsigned int phase = *syt_offset_state;
438 unsigned int index = phase % 13;
440 syt_offset = *last_syt_offset;
441 syt_offset += 1386 + ((index && !(index & 3)) ||
445 *syt_offset_state = phase;
448 syt_offset = *last_syt_offset - TICKS_PER_CYCLE;
449 *last_syt_offset = syt_offset;
451 if (syt_offset >= TICKS_PER_CYCLE)
452 syt_offset = CIP_SYT_NO_INFO;
457 static void pool_ideal_syt_offsets(struct amdtp_stream *s, struct seq_desc *descs,
458 unsigned int size, unsigned int pos, unsigned int count)
460 const enum cip_sfc sfc = s->sfc;
461 unsigned int last = s->ctx_data.rx.last_syt_offset;
462 unsigned int state = s->ctx_data.rx.syt_offset_state;
465 for (i = 0; i < count; ++i) {
466 struct seq_desc *desc = descs + pos;
468 desc->syt_offset = calculate_syt_offset(&last, &state, sfc);
470 pos = (pos + 1) % size;
473 s->ctx_data.rx.last_syt_offset = last;
474 s->ctx_data.rx.syt_offset_state = state;
477 static unsigned int compute_syt_offset(unsigned int syt, unsigned int cycle,
478 unsigned int transfer_delay)
480 unsigned int cycle_lo = (cycle % CYCLES_PER_SECOND) & 0x0f;
481 unsigned int syt_cycle_lo = (syt & 0xf000) >> 12;
482 unsigned int syt_offset;
485 if (syt_cycle_lo < cycle_lo)
486 syt_cycle_lo += CIP_SYT_CYCLE_MODULUS;
487 syt_cycle_lo -= cycle_lo;
489 // Subtract transfer delay so that the synchronization offset is not so large
491 syt_offset = syt_cycle_lo * TICKS_PER_CYCLE + (syt & 0x0fff);
492 if (syt_offset < transfer_delay)
493 syt_offset += CIP_SYT_CYCLE_MODULUS * TICKS_PER_CYCLE;
495 return syt_offset - transfer_delay;
498 // Both of the producer and consumer of the queue runs in the same clock of IEEE 1394 bus.
499 // Additionally, the sequence of tx packets is severely checked against any discontinuity
500 // before filling entries in the queue. The calculation is safe even if it looks fragile by
502 static unsigned int calculate_cached_cycle_count(struct amdtp_stream *s, unsigned int head)
504 const unsigned int cache_size = s->ctx_data.tx.cache.size;
505 unsigned int cycles = s->ctx_data.tx.cache.pos;
508 cycles += cache_size;
514 static void cache_seq(struct amdtp_stream *s, const struct pkt_desc *src, unsigned int desc_count)
516 const unsigned int transfer_delay = s->transfer_delay;
517 const unsigned int cache_size = s->ctx_data.tx.cache.size;
518 struct seq_desc *cache = s->ctx_data.tx.cache.descs;
519 unsigned int cache_pos = s->ctx_data.tx.cache.pos;
520 bool aware_syt = !(s->flags & CIP_UNAWARE_SYT);
523 for (i = 0; i < desc_count; ++i) {
524 struct seq_desc *dst = cache + cache_pos;
526 if (aware_syt && src->syt != CIP_SYT_NO_INFO)
527 dst->syt_offset = compute_syt_offset(src->syt, src->cycle, transfer_delay);
529 dst->syt_offset = CIP_SYT_NO_INFO;
530 dst->data_blocks = src->data_blocks;
532 cache_pos = (cache_pos + 1) % cache_size;
533 src = amdtp_stream_next_packet_desc(s, src);
536 s->ctx_data.tx.cache.pos = cache_pos;
539 static void pool_ideal_seq_descs(struct amdtp_stream *s, struct seq_desc *descs, unsigned int size,
540 unsigned int pos, unsigned int count)
542 pool_ideal_syt_offsets(s, descs, size, pos, count);
544 if (s->flags & CIP_BLOCKING)
545 pool_blocking_data_blocks(s, descs, size, pos, count);
547 pool_ideal_nonblocking_data_blocks(s, descs, size, pos, count);
550 static void pool_replayed_seq(struct amdtp_stream *s, struct seq_desc *descs, unsigned int size,
551 unsigned int pos, unsigned int count)
553 struct amdtp_stream *target = s->ctx_data.rx.replay_target;
554 const struct seq_desc *cache = target->ctx_data.tx.cache.descs;
555 const unsigned int cache_size = target->ctx_data.tx.cache.size;
556 unsigned int cache_pos = s->ctx_data.rx.cache_pos;
559 for (i = 0; i < count; ++i) {
560 descs[pos] = cache[cache_pos];
561 cache_pos = (cache_pos + 1) % cache_size;
562 pos = (pos + 1) % size;
565 s->ctx_data.rx.cache_pos = cache_pos;
568 static void pool_seq_descs(struct amdtp_stream *s, struct seq_desc *descs, unsigned int size,
569 unsigned int pos, unsigned int count)
571 struct amdtp_domain *d = s->domain;
572 void (*pool_seq_descs)(struct amdtp_stream *s, struct seq_desc *descs, unsigned int size,
573 unsigned int pos, unsigned int count);
575 if (!d->replay.enable || !s->ctx_data.rx.replay_target) {
576 pool_seq_descs = pool_ideal_seq_descs;
578 if (!d->replay.on_the_fly) {
579 pool_seq_descs = pool_replayed_seq;
581 struct amdtp_stream *tx = s->ctx_data.rx.replay_target;
582 const unsigned int cache_size = tx->ctx_data.tx.cache.size;
583 const unsigned int cache_pos = s->ctx_data.rx.cache_pos;
584 unsigned int cached_cycles = calculate_cached_cycle_count(tx, cache_pos);
586 if (cached_cycles > count && cached_cycles > cache_size / 2)
587 pool_seq_descs = pool_replayed_seq;
589 pool_seq_descs = pool_ideal_seq_descs;
593 pool_seq_descs(s, descs, size, pos, count);
596 static void update_pcm_pointers(struct amdtp_stream *s,
597 struct snd_pcm_substream *pcm,
602 ptr = s->pcm_buffer_pointer + frames;
603 if (ptr >= pcm->runtime->buffer_size)
604 ptr -= pcm->runtime->buffer_size;
605 WRITE_ONCE(s->pcm_buffer_pointer, ptr);
607 s->pcm_period_pointer += frames;
608 if (s->pcm_period_pointer >= pcm->runtime->period_size) {
609 s->pcm_period_pointer -= pcm->runtime->period_size;
611 // The program in user process should periodically check the status of intermediate
612 // buffer associated to PCM substream to process PCM frames in the buffer, instead
613 // of receiving notification of period elapsed by poll wait.
614 if (!pcm->runtime->no_period_wakeup) {
616 // In software IRQ context for 1394 OHCI.
617 snd_pcm_period_elapsed(pcm);
619 // In process context of ALSA PCM application under acquired lock of
621 snd_pcm_period_elapsed_under_stream_lock(pcm);
627 static int queue_packet(struct amdtp_stream *s, struct fw_iso_packet *params,
632 params->interrupt = sched_irq;
633 params->tag = s->tag;
636 err = fw_iso_context_queue(s->context, params, &s->buffer.iso_buffer,
637 s->buffer.packets[s->packet_index].offset);
639 dev_err(&s->unit->device, "queueing error: %d\n", err);
643 if (++s->packet_index >= s->queue_size)
649 static inline int queue_out_packet(struct amdtp_stream *s,
650 struct fw_iso_packet *params, bool sched_irq)
653 !!(params->header_length == 0 && params->payload_length == 0);
654 return queue_packet(s, params, sched_irq);
657 static inline int queue_in_packet(struct amdtp_stream *s,
658 struct fw_iso_packet *params)
660 // Queue one packet for IR context.
661 params->header_length = s->ctx_data.tx.ctx_header_size;
662 params->payload_length = s->ctx_data.tx.max_ctx_payload_length;
663 params->skip = false;
664 return queue_packet(s, params, false);
667 static void generate_cip_header(struct amdtp_stream *s, __be32 cip_header[2],
668 unsigned int data_block_counter, unsigned int syt)
670 cip_header[0] = cpu_to_be32(READ_ONCE(s->source_node_id_field) |
671 (s->data_block_quadlets << CIP_DBS_SHIFT) |
672 ((s->sph << CIP_SPH_SHIFT) & CIP_SPH_MASK) |
674 cip_header[1] = cpu_to_be32(CIP_EOH |
675 ((s->fmt << CIP_FMT_SHIFT) & CIP_FMT_MASK) |
676 ((s->ctx_data.rx.fdf << CIP_FDF_SHIFT) & CIP_FDF_MASK) |
677 (syt & CIP_SYT_MASK));
680 static void build_it_pkt_header(struct amdtp_stream *s, unsigned int cycle,
681 struct fw_iso_packet *params, unsigned int header_length,
682 unsigned int data_blocks,
683 unsigned int data_block_counter,
684 unsigned int syt, unsigned int index, u32 curr_cycle_time)
686 unsigned int payload_length;
689 payload_length = data_blocks * sizeof(__be32) * s->data_block_quadlets;
690 params->payload_length = payload_length;
692 if (header_length > 0) {
693 cip_header = (__be32 *)params->header;
694 generate_cip_header(s, cip_header, data_block_counter, syt);
695 params->header_length = header_length;
700 trace_amdtp_packet(s, cycle, cip_header, payload_length + header_length, data_blocks,
701 data_block_counter, s->packet_index, index, curr_cycle_time);
704 static int check_cip_header(struct amdtp_stream *s, const __be32 *buf,
705 unsigned int payload_length,
706 unsigned int *data_blocks,
707 unsigned int *data_block_counter, unsigned int *syt)
716 cip_header[0] = be32_to_cpu(buf[0]);
717 cip_header[1] = be32_to_cpu(buf[1]);
720 * This module supports 'Two-quadlet CIP header with SYT field'.
721 * For convenience, also check FMT field is AM824 or not.
723 if ((((cip_header[0] & CIP_EOH_MASK) == CIP_EOH) ||
724 ((cip_header[1] & CIP_EOH_MASK) != CIP_EOH)) &&
725 (!(s->flags & CIP_HEADER_WITHOUT_EOH))) {
726 dev_info_ratelimited(&s->unit->device,
727 "Invalid CIP header for AMDTP: %08X:%08X\n",
728 cip_header[0], cip_header[1]);
732 /* Check valid protocol or not. */
733 sph = (cip_header[0] & CIP_SPH_MASK) >> CIP_SPH_SHIFT;
734 fmt = (cip_header[1] & CIP_FMT_MASK) >> CIP_FMT_SHIFT;
735 if (sph != s->sph || fmt != s->fmt) {
736 dev_info_ratelimited(&s->unit->device,
737 "Detect unexpected protocol: %08x %08x\n",
738 cip_header[0], cip_header[1]);
742 /* Calculate data blocks */
743 fdf = (cip_header[1] & CIP_FDF_MASK) >> CIP_FDF_SHIFT;
744 if (payload_length == 0 || (fmt == CIP_FMT_AM && fdf == AMDTP_FDF_NO_DATA)) {
747 unsigned int data_block_quadlets =
748 (cip_header[0] & CIP_DBS_MASK) >> CIP_DBS_SHIFT;
749 /* avoid division by zero */
750 if (data_block_quadlets == 0) {
751 dev_err(&s->unit->device,
752 "Detect invalid value in dbs field: %08X\n",
756 if (s->flags & CIP_WRONG_DBS)
757 data_block_quadlets = s->data_block_quadlets;
759 *data_blocks = payload_length / sizeof(__be32) / data_block_quadlets;
762 /* Check data block counter continuity */
763 dbc = cip_header[0] & CIP_DBC_MASK;
764 if (*data_blocks == 0 && (s->flags & CIP_EMPTY_HAS_WRONG_DBC) &&
765 *data_block_counter != UINT_MAX)
766 dbc = *data_block_counter;
768 if ((dbc == 0x00 && (s->flags & CIP_SKIP_DBC_ZERO_CHECK)) ||
769 *data_block_counter == UINT_MAX) {
771 } else if (!(s->flags & CIP_DBC_IS_END_EVENT)) {
772 lost = dbc != *data_block_counter;
774 unsigned int dbc_interval;
776 if (*data_blocks > 0 && s->ctx_data.tx.dbc_interval > 0)
777 dbc_interval = s->ctx_data.tx.dbc_interval;
779 dbc_interval = *data_blocks;
781 lost = dbc != ((*data_block_counter + dbc_interval) & 0xff);
785 dev_err(&s->unit->device,
786 "Detect discontinuity of CIP: %02X %02X\n",
787 *data_block_counter, dbc);
791 *data_block_counter = dbc;
793 if (!(s->flags & CIP_UNAWARE_SYT))
794 *syt = cip_header[1] & CIP_SYT_MASK;
799 static int parse_ir_ctx_header(struct amdtp_stream *s, unsigned int cycle,
800 const __be32 *ctx_header,
801 unsigned int *data_blocks,
802 unsigned int *data_block_counter,
803 unsigned int *syt, unsigned int packet_index, unsigned int index,
806 unsigned int payload_length;
807 const __be32 *cip_header;
808 unsigned int cip_header_size;
810 payload_length = be32_to_cpu(ctx_header[0]) >> ISO_DATA_LENGTH_SHIFT;
812 if (!(s->flags & CIP_NO_HEADER))
813 cip_header_size = CIP_HEADER_SIZE;
817 if (payload_length > cip_header_size + s->ctx_data.tx.max_ctx_payload_length) {
818 dev_err(&s->unit->device,
819 "Detect jumbo payload: %04x %04x\n",
820 payload_length, cip_header_size + s->ctx_data.tx.max_ctx_payload_length);
824 if (cip_header_size > 0) {
825 if (payload_length >= cip_header_size) {
828 cip_header = ctx_header + IR_CTX_HEADER_DEFAULT_QUADLETS;
829 err = check_cip_header(s, cip_header, payload_length - cip_header_size,
830 data_blocks, data_block_counter, syt);
834 // Handle the cycle so that empty packet arrives.
841 *data_blocks = payload_length / sizeof(__be32) / s->data_block_quadlets;
844 if (*data_block_counter == UINT_MAX)
845 *data_block_counter = 0;
848 trace_amdtp_packet(s, cycle, cip_header, payload_length, *data_blocks,
849 *data_block_counter, packet_index, index, curr_cycle_time);
854 // In CYCLE_TIMER register of IEEE 1394, 7 bits are used to represent second. On
855 // the other hand, in DMA descriptors of 1394 OHCI, 3 bits are used to represent
856 // it. Thus, via Linux firewire subsystem, we can get the 3 bits for second.
857 static inline u32 compute_ohci_iso_ctx_cycle_count(u32 tstamp)
859 return (((tstamp >> 13) & 0x07) * CYCLES_PER_SECOND) + (tstamp & 0x1fff);
862 static inline u32 compute_ohci_cycle_count(__be32 ctx_header_tstamp)
864 u32 tstamp = be32_to_cpu(ctx_header_tstamp) & HEADER_TSTAMP_MASK;
865 return compute_ohci_iso_ctx_cycle_count(tstamp);
868 static inline u32 increment_ohci_cycle_count(u32 cycle, unsigned int addend)
871 if (cycle >= OHCI_SECOND_MODULUS * CYCLES_PER_SECOND)
872 cycle -= OHCI_SECOND_MODULUS * CYCLES_PER_SECOND;
876 static inline u32 decrement_ohci_cycle_count(u32 minuend, u32 subtrahend)
878 if (minuend < subtrahend)
879 minuend += OHCI_SECOND_MODULUS * CYCLES_PER_SECOND;
881 return minuend - subtrahend;
884 static int compare_ohci_cycle_count(u32 lval, u32 rval)
888 else if (lval < rval && rval - lval < OHCI_SECOND_MODULUS * CYCLES_PER_SECOND / 2)
894 // Align to actual cycle count for the packet which is going to be scheduled.
895 // This module queued the same number of isochronous cycle as the size of queue
896 // to kip isochronous cycle, therefore it's OK to just increment the cycle by
897 // the size of queue for scheduled cycle.
898 static inline u32 compute_ohci_it_cycle(const __be32 ctx_header_tstamp,
899 unsigned int queue_size)
901 u32 cycle = compute_ohci_cycle_count(ctx_header_tstamp);
902 return increment_ohci_cycle_count(cycle, queue_size);
905 static int generate_tx_packet_descs(struct amdtp_stream *s, struct pkt_desc *desc,
906 const __be32 *ctx_header, unsigned int packet_count,
907 unsigned int *desc_count)
909 unsigned int next_cycle = s->next_cycle;
910 unsigned int dbc = s->data_block_counter;
911 unsigned int packet_index = s->packet_index;
912 unsigned int queue_size = s->queue_size;
913 u32 curr_cycle_time = 0;
917 if (trace_amdtp_packet_enabled())
918 (void)fw_card_read_cycle_time(fw_parent_device(s->unit)->card, &curr_cycle_time);
921 for (i = 0; i < packet_count; ++i) {
924 unsigned int data_blocks;
927 cycle = compute_ohci_cycle_count(ctx_header[1]);
928 lost = (next_cycle != cycle);
930 if (s->flags & CIP_NO_HEADER) {
931 // Fireface skips transmission just for an isoc cycle corresponding
933 unsigned int prev_cycle = next_cycle;
935 next_cycle = increment_ohci_cycle_count(next_cycle, 1);
936 lost = (next_cycle != cycle);
938 // Prepare a description for the skipped cycle for
940 desc->cycle = prev_cycle;
942 desc->data_blocks = 0;
943 desc->data_block_counter = dbc;
944 desc->ctx_payload = NULL;
945 desc = amdtp_stream_next_packet_desc(s, desc);
948 } else if (s->flags & CIP_JUMBO_PAYLOAD) {
949 // OXFW970 skips transmission for several isoc cycles during
950 // asynchronous transaction. The sequence replay is impossible due
952 unsigned int safe_cycle = increment_ohci_cycle_count(next_cycle,
953 IR_JUMBO_PAYLOAD_MAX_SKIP_CYCLES);
954 lost = (compare_ohci_cycle_count(safe_cycle, cycle) < 0);
957 dev_err(&s->unit->device, "Detect discontinuity of cycle: %d %d\n",
963 err = parse_ir_ctx_header(s, cycle, ctx_header, &data_blocks, &dbc, &syt,
964 packet_index, i, curr_cycle_time);
970 desc->data_blocks = data_blocks;
971 desc->data_block_counter = dbc;
972 desc->ctx_payload = s->buffer.packets[packet_index].buffer;
974 if (!(s->flags & CIP_DBC_IS_END_EVENT))
975 dbc = (dbc + desc->data_blocks) & 0xff;
977 next_cycle = increment_ohci_cycle_count(next_cycle, 1);
978 desc = amdtp_stream_next_packet_desc(s, desc);
980 ctx_header += s->ctx_data.tx.ctx_header_size / sizeof(*ctx_header);
981 packet_index = (packet_index + 1) % queue_size;
984 s->next_cycle = next_cycle;
985 s->data_block_counter = dbc;
990 static unsigned int compute_syt(unsigned int syt_offset, unsigned int cycle,
991 unsigned int transfer_delay)
995 syt_offset += transfer_delay;
996 syt = ((cycle + syt_offset / TICKS_PER_CYCLE) << 12) |
997 (syt_offset % TICKS_PER_CYCLE);
998 return syt & CIP_SYT_MASK;
1001 static void generate_rx_packet_descs(struct amdtp_stream *s, struct pkt_desc *desc,
1002 const __be32 *ctx_header, unsigned int packet_count)
1004 struct seq_desc *seq_descs = s->ctx_data.rx.seq.descs;
1005 unsigned int seq_size = s->ctx_data.rx.seq.size;
1006 unsigned int seq_pos = s->ctx_data.rx.seq.pos;
1007 unsigned int dbc = s->data_block_counter;
1008 bool aware_syt = !(s->flags & CIP_UNAWARE_SYT);
1011 pool_seq_descs(s, seq_descs, seq_size, seq_pos, packet_count);
1013 for (i = 0; i < packet_count; ++i) {
1014 unsigned int index = (s->packet_index + i) % s->queue_size;
1015 const struct seq_desc *seq = seq_descs + seq_pos;
1017 desc->cycle = compute_ohci_it_cycle(*ctx_header, s->queue_size);
1019 if (aware_syt && seq->syt_offset != CIP_SYT_NO_INFO)
1020 desc->syt = compute_syt(seq->syt_offset, desc->cycle, s->transfer_delay);
1022 desc->syt = CIP_SYT_NO_INFO;
1024 desc->data_blocks = seq->data_blocks;
1026 if (s->flags & CIP_DBC_IS_END_EVENT)
1027 dbc = (dbc + desc->data_blocks) & 0xff;
1029 desc->data_block_counter = dbc;
1031 if (!(s->flags & CIP_DBC_IS_END_EVENT))
1032 dbc = (dbc + desc->data_blocks) & 0xff;
1034 desc->ctx_payload = s->buffer.packets[index].buffer;
1036 seq_pos = (seq_pos + 1) % seq_size;
1037 desc = amdtp_stream_next_packet_desc(s, desc);
1042 s->data_block_counter = dbc;
1043 s->ctx_data.rx.seq.pos = seq_pos;
1046 static inline void cancel_stream(struct amdtp_stream *s)
1048 s->packet_index = -1;
1050 amdtp_stream_pcm_abort(s);
1051 WRITE_ONCE(s->pcm_buffer_pointer, SNDRV_PCM_POS_XRUN);
1054 static snd_pcm_sframes_t compute_pcm_extra_delay(struct amdtp_stream *s,
1055 const struct pkt_desc *desc, unsigned int count)
1057 unsigned int data_block_count = 0;
1067 // Forward to the latest record.
1068 for (i = 0; i < count - 1; ++i)
1069 desc = amdtp_stream_next_packet_desc(s, desc);
1070 latest_cycle = desc->cycle;
1072 err = fw_card_read_cycle_time(fw_parent_device(s->unit)->card, &cycle_time);
1076 // Compute cycle count with lower 3 bits of second field and cycle field like timestamp
1077 // format of 1394 OHCI isochronous context.
1078 curr_cycle = compute_ohci_iso_ctx_cycle_count((cycle_time >> 12) & 0x0000ffff);
1080 if (s->direction == AMDTP_IN_STREAM) {
1081 // NOTE: The AMDTP packet descriptor should be for the past isochronous cycle since
1082 // it corresponds to arrived isochronous packet.
1083 if (compare_ohci_cycle_count(latest_cycle, curr_cycle) > 0)
1085 cycle_gap = decrement_ohci_cycle_count(curr_cycle, latest_cycle);
1087 // NOTE: estimate delay by recent history of arrived AMDTP packets. The estimated
1088 // value expectedly corresponds to a few packets (0-2) since the packet arrived at
1089 // the most recent isochronous cycle has been already processed.
1090 for (i = 0; i < cycle_gap; ++i) {
1091 desc = amdtp_stream_next_packet_desc(s, desc);
1092 data_block_count += desc->data_blocks;
1095 // NOTE: The AMDTP packet descriptor should be for the future isochronous cycle
1096 // since it was already scheduled.
1097 if (compare_ohci_cycle_count(latest_cycle, curr_cycle) < 0)
1099 cycle_gap = decrement_ohci_cycle_count(latest_cycle, curr_cycle);
1101 // NOTE: use history of scheduled packets.
1102 for (i = 0; i < cycle_gap; ++i) {
1103 data_block_count += desc->data_blocks;
1104 desc = prev_packet_desc(s, desc);
1108 return data_block_count * s->pcm_frame_multiplier;
1111 static void process_ctx_payloads(struct amdtp_stream *s,
1112 const struct pkt_desc *desc,
1115 struct snd_pcm_substream *pcm;
1118 pcm = READ_ONCE(s->pcm);
1119 s->process_ctx_payloads(s, desc, count, pcm);
1122 unsigned int data_block_count = 0;
1124 pcm->runtime->delay = compute_pcm_extra_delay(s, desc, count);
1126 for (i = 0; i < count; ++i) {
1127 data_block_count += desc->data_blocks;
1128 desc = amdtp_stream_next_packet_desc(s, desc);
1131 update_pcm_pointers(s, pcm, data_block_count * s->pcm_frame_multiplier);
1135 static void process_rx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1136 void *header, void *private_data)
1138 struct amdtp_stream *s = private_data;
1139 const struct amdtp_domain *d = s->domain;
1140 const __be32 *ctx_header = header;
1141 const unsigned int events_per_period = d->events_per_period;
1142 unsigned int event_count = s->ctx_data.rx.event_count;
1143 struct pkt_desc *desc = s->packet_descs_cursor;
1144 unsigned int pkt_header_length;
1145 unsigned int packets;
1146 u32 curr_cycle_time;
1150 if (s->packet_index < 0)
1153 // Calculate the number of packets in buffer and check XRUN.
1154 packets = header_length / sizeof(*ctx_header);
1156 generate_rx_packet_descs(s, desc, ctx_header, packets);
1158 process_ctx_payloads(s, desc, packets);
1160 if (!(s->flags & CIP_NO_HEADER))
1161 pkt_header_length = IT_PKT_HEADER_SIZE_CIP;
1163 pkt_header_length = 0;
1165 if (s == d->irq_target) {
1166 // At NO_PERIOD_WAKEUP mode, the packets for all IT/IR contexts are processed by
1167 // the tasks of user process operating ALSA PCM character device by calling ioctl(2)
1168 // with some requests, instead of scheduled hardware IRQ of an IT context.
1169 struct snd_pcm_substream *pcm = READ_ONCE(s->pcm);
1170 need_hw_irq = !pcm || !pcm->runtime->no_period_wakeup;
1172 need_hw_irq = false;
1175 if (trace_amdtp_packet_enabled())
1176 (void)fw_card_read_cycle_time(fw_parent_device(s->unit)->card, &curr_cycle_time);
1178 for (i = 0; i < packets; ++i) {
1180 struct fw_iso_packet params;
1181 __be32 header[CIP_HEADER_QUADLETS];
1182 } template = { {0}, {0} };
1183 bool sched_irq = false;
1185 build_it_pkt_header(s, desc->cycle, &template.params, pkt_header_length,
1186 desc->data_blocks, desc->data_block_counter,
1187 desc->syt, i, curr_cycle_time);
1189 if (s == s->domain->irq_target) {
1190 event_count += desc->data_blocks;
1191 if (event_count >= events_per_period) {
1192 event_count -= events_per_period;
1193 sched_irq = need_hw_irq;
1197 if (queue_out_packet(s, &template.params, sched_irq) < 0) {
1202 desc = amdtp_stream_next_packet_desc(s, desc);
1205 s->ctx_data.rx.event_count = event_count;
1206 s->packet_descs_cursor = desc;
1209 static void skip_rx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1210 void *header, void *private_data)
1212 struct amdtp_stream *s = private_data;
1213 struct amdtp_domain *d = s->domain;
1214 const __be32 *ctx_header = header;
1215 unsigned int packets;
1219 if (s->packet_index < 0)
1222 packets = header_length / sizeof(*ctx_header);
1224 cycle = compute_ohci_it_cycle(ctx_header[packets - 1], s->queue_size);
1225 s->next_cycle = increment_ohci_cycle_count(cycle, 1);
1227 for (i = 0; i < packets; ++i) {
1228 struct fw_iso_packet params = {
1230 .payload_length = 0,
1232 bool sched_irq = (s == d->irq_target && i == packets - 1);
1234 if (queue_out_packet(s, ¶ms, sched_irq) < 0) {
1241 static void irq_target_callback(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1242 void *header, void *private_data);
1244 static void process_rx_packets_intermediately(struct fw_iso_context *context, u32 tstamp,
1245 size_t header_length, void *header, void *private_data)
1247 struct amdtp_stream *s = private_data;
1248 struct amdtp_domain *d = s->domain;
1249 __be32 *ctx_header = header;
1250 const unsigned int queue_size = s->queue_size;
1251 unsigned int packets;
1252 unsigned int offset;
1254 if (s->packet_index < 0)
1257 packets = header_length / sizeof(*ctx_header);
1260 while (offset < packets) {
1261 unsigned int cycle = compute_ohci_it_cycle(ctx_header[offset], queue_size);
1263 if (compare_ohci_cycle_count(cycle, d->processing_cycle.rx_start) >= 0)
1270 unsigned int length = sizeof(*ctx_header) * offset;
1272 skip_rx_packets(context, tstamp, length, ctx_header, private_data);
1273 if (amdtp_streaming_error(s))
1276 ctx_header += offset;
1277 header_length -= length;
1280 if (offset < packets) {
1281 s->ready_processing = true;
1282 wake_up(&s->ready_wait);
1284 if (d->replay.enable)
1285 s->ctx_data.rx.cache_pos = 0;
1287 process_rx_packets(context, tstamp, header_length, ctx_header, private_data);
1288 if (amdtp_streaming_error(s))
1291 if (s == d->irq_target)
1292 s->context->callback.sc = irq_target_callback;
1294 s->context->callback.sc = process_rx_packets;
1298 static void process_tx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1299 void *header, void *private_data)
1301 struct amdtp_stream *s = private_data;
1302 __be32 *ctx_header = header;
1303 struct pkt_desc *desc = s->packet_descs_cursor;
1304 unsigned int packet_count;
1305 unsigned int desc_count;
1309 if (s->packet_index < 0)
1312 // Calculate the number of packets in buffer and check XRUN.
1313 packet_count = header_length / s->ctx_data.tx.ctx_header_size;
1316 err = generate_tx_packet_descs(s, desc, ctx_header, packet_count, &desc_count);
1318 if (err != -EAGAIN) {
1323 struct amdtp_domain *d = s->domain;
1325 process_ctx_payloads(s, desc, desc_count);
1327 if (d->replay.enable)
1328 cache_seq(s, desc, desc_count);
1330 for (i = 0; i < desc_count; ++i)
1331 desc = amdtp_stream_next_packet_desc(s, desc);
1332 s->packet_descs_cursor = desc;
1335 for (i = 0; i < packet_count; ++i) {
1336 struct fw_iso_packet params = {0};
1338 if (queue_in_packet(s, ¶ms) < 0) {
1345 static void drop_tx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1346 void *header, void *private_data)
1348 struct amdtp_stream *s = private_data;
1349 const __be32 *ctx_header = header;
1350 unsigned int packets;
1354 if (s->packet_index < 0)
1357 packets = header_length / s->ctx_data.tx.ctx_header_size;
1359 ctx_header += (packets - 1) * s->ctx_data.tx.ctx_header_size / sizeof(*ctx_header);
1360 cycle = compute_ohci_cycle_count(ctx_header[1]);
1361 s->next_cycle = increment_ohci_cycle_count(cycle, 1);
1363 for (i = 0; i < packets; ++i) {
1364 struct fw_iso_packet params = {0};
1366 if (queue_in_packet(s, ¶ms) < 0) {
1373 static void process_tx_packets_intermediately(struct fw_iso_context *context, u32 tstamp,
1374 size_t header_length, void *header, void *private_data)
1376 struct amdtp_stream *s = private_data;
1377 struct amdtp_domain *d = s->domain;
1379 unsigned int packets;
1380 unsigned int offset;
1382 if (s->packet_index < 0)
1385 packets = header_length / s->ctx_data.tx.ctx_header_size;
1388 ctx_header = header;
1389 while (offset < packets) {
1390 unsigned int cycle = compute_ohci_cycle_count(ctx_header[1]);
1392 if (compare_ohci_cycle_count(cycle, d->processing_cycle.tx_start) >= 0)
1395 ctx_header += s->ctx_data.tx.ctx_header_size / sizeof(__be32);
1399 ctx_header = header;
1402 size_t length = s->ctx_data.tx.ctx_header_size * offset;
1404 drop_tx_packets(context, tstamp, length, ctx_header, s);
1405 if (amdtp_streaming_error(s))
1408 ctx_header += length / sizeof(*ctx_header);
1409 header_length -= length;
1412 if (offset < packets) {
1413 s->ready_processing = true;
1414 wake_up(&s->ready_wait);
1416 process_tx_packets(context, tstamp, header_length, ctx_header, s);
1417 if (amdtp_streaming_error(s))
1420 context->callback.sc = process_tx_packets;
1424 static void drop_tx_packets_initially(struct fw_iso_context *context, u32 tstamp,
1425 size_t header_length, void *header, void *private_data)
1427 struct amdtp_stream *s = private_data;
1428 struct amdtp_domain *d = s->domain;
1431 unsigned int events;
1434 if (s->packet_index < 0)
1437 count = header_length / s->ctx_data.tx.ctx_header_size;
1439 // Attempt to detect any event in the batch of packets.
1441 ctx_header = header;
1442 for (i = 0; i < count; ++i) {
1443 unsigned int payload_quads =
1444 (be32_to_cpu(*ctx_header) >> ISO_DATA_LENGTH_SHIFT) / sizeof(__be32);
1445 unsigned int data_blocks;
1447 if (s->flags & CIP_NO_HEADER) {
1448 data_blocks = payload_quads / s->data_block_quadlets;
1450 __be32 *cip_headers = ctx_header + IR_CTX_HEADER_DEFAULT_QUADLETS;
1452 if (payload_quads < CIP_HEADER_QUADLETS) {
1455 payload_quads -= CIP_HEADER_QUADLETS;
1457 if (s->flags & CIP_UNAWARE_SYT) {
1458 data_blocks = payload_quads / s->data_block_quadlets;
1460 u32 cip1 = be32_to_cpu(cip_headers[1]);
1462 // NODATA packet can includes any data blocks but they are
1463 // not available as event.
1464 if ((cip1 & CIP_NO_DATA) == CIP_NO_DATA)
1467 data_blocks = payload_quads / s->data_block_quadlets;
1472 events += data_blocks;
1474 ctx_header += s->ctx_data.tx.ctx_header_size / sizeof(__be32);
1477 drop_tx_packets(context, tstamp, header_length, header, s);
1480 s->ctx_data.tx.event_starts = true;
1482 // Decide the cycle count to begin processing content of packet in IR contexts.
1484 unsigned int stream_count = 0;
1485 unsigned int event_starts_count = 0;
1486 unsigned int cycle = UINT_MAX;
1488 list_for_each_entry(s, &d->streams, list) {
1489 if (s->direction == AMDTP_IN_STREAM) {
1491 if (s->ctx_data.tx.event_starts)
1492 ++event_starts_count;
1496 if (stream_count == event_starts_count) {
1497 unsigned int next_cycle;
1499 list_for_each_entry(s, &d->streams, list) {
1500 if (s->direction != AMDTP_IN_STREAM)
1503 next_cycle = increment_ohci_cycle_count(s->next_cycle,
1504 d->processing_cycle.tx_init_skip);
1505 if (cycle == UINT_MAX ||
1506 compare_ohci_cycle_count(next_cycle, cycle) > 0)
1509 s->context->callback.sc = process_tx_packets_intermediately;
1512 d->processing_cycle.tx_start = cycle;
1517 static void process_ctxs_in_domain(struct amdtp_domain *d)
1519 struct amdtp_stream *s;
1521 list_for_each_entry(s, &d->streams, list) {
1522 if (s != d->irq_target && amdtp_stream_running(s))
1523 fw_iso_context_flush_completions(s->context);
1525 if (amdtp_streaming_error(s))
1531 if (amdtp_stream_running(d->irq_target))
1532 cancel_stream(d->irq_target);
1534 list_for_each_entry(s, &d->streams, list) {
1535 if (amdtp_stream_running(s))
1540 static void irq_target_callback(struct fw_iso_context *context, u32 tstamp, size_t header_length,
1541 void *header, void *private_data)
1543 struct amdtp_stream *s = private_data;
1544 struct amdtp_domain *d = s->domain;
1546 process_rx_packets(context, tstamp, header_length, header, private_data);
1547 process_ctxs_in_domain(d);
1550 static void irq_target_callback_intermediately(struct fw_iso_context *context, u32 tstamp,
1551 size_t header_length, void *header, void *private_data)
1553 struct amdtp_stream *s = private_data;
1554 struct amdtp_domain *d = s->domain;
1556 process_rx_packets_intermediately(context, tstamp, header_length, header, private_data);
1557 process_ctxs_in_domain(d);
1560 static void irq_target_callback_skip(struct fw_iso_context *context, u32 tstamp,
1561 size_t header_length, void *header, void *private_data)
1563 struct amdtp_stream *s = private_data;
1564 struct amdtp_domain *d = s->domain;
1565 bool ready_to_start;
1567 skip_rx_packets(context, tstamp, header_length, header, private_data);
1568 process_ctxs_in_domain(d);
1570 if (d->replay.enable && !d->replay.on_the_fly) {
1571 unsigned int rx_count = 0;
1572 unsigned int rx_ready_count = 0;
1573 struct amdtp_stream *rx;
1575 list_for_each_entry(rx, &d->streams, list) {
1576 struct amdtp_stream *tx;
1577 unsigned int cached_cycles;
1579 if (rx->direction != AMDTP_OUT_STREAM)
1583 tx = rx->ctx_data.rx.replay_target;
1584 cached_cycles = calculate_cached_cycle_count(tx, 0);
1585 if (cached_cycles > tx->ctx_data.tx.cache.size / 2)
1589 ready_to_start = (rx_count == rx_ready_count);
1591 ready_to_start = true;
1594 // Decide the cycle count to begin processing content of packet in IT contexts. All of IT
1595 // contexts are expected to start and get callback when reaching here.
1596 if (ready_to_start) {
1597 unsigned int cycle = s->next_cycle;
1598 list_for_each_entry(s, &d->streams, list) {
1599 if (s->direction != AMDTP_OUT_STREAM)
1602 if (compare_ohci_cycle_count(s->next_cycle, cycle) > 0)
1603 cycle = s->next_cycle;
1605 if (s == d->irq_target)
1606 s->context->callback.sc = irq_target_callback_intermediately;
1608 s->context->callback.sc = process_rx_packets_intermediately;
1611 d->processing_cycle.rx_start = cycle;
1615 // This is executed one time. For in-stream, first packet has come. For out-stream, prepared to
1616 // transmit first packet.
1617 static void amdtp_stream_first_callback(struct fw_iso_context *context,
1618 u32 tstamp, size_t header_length,
1619 void *header, void *private_data)
1621 struct amdtp_stream *s = private_data;
1622 struct amdtp_domain *d = s->domain;
1624 if (s->direction == AMDTP_IN_STREAM) {
1625 context->callback.sc = drop_tx_packets_initially;
1627 if (s == d->irq_target)
1628 context->callback.sc = irq_target_callback_skip;
1630 context->callback.sc = skip_rx_packets;
1633 context->callback.sc(context, tstamp, header_length, header, s);
1637 * amdtp_stream_start - start transferring packets
1638 * @s: the AMDTP stream to start
1639 * @channel: the isochronous channel on the bus
1640 * @speed: firewire speed code
1641 * @queue_size: The number of packets in the queue.
1642 * @idle_irq_interval: the interval to queue packet during initial state.
1644 * The stream cannot be started until it has been configured with
1645 * amdtp_stream_set_parameters() and it must be started before any PCM or MIDI
1646 * device can be started.
1648 static int amdtp_stream_start(struct amdtp_stream *s, int channel, int speed,
1649 unsigned int queue_size, unsigned int idle_irq_interval)
1651 bool is_irq_target = (s == s->domain->irq_target);
1652 unsigned int ctx_header_size;
1653 unsigned int max_ctx_payload_size;
1654 enum dma_data_direction dir;
1655 struct pkt_desc *descs;
1656 int i, type, tag, err;
1658 mutex_lock(&s->mutex);
1660 if (WARN_ON(amdtp_stream_running(s) ||
1661 (s->data_block_quadlets < 1))) {
1666 if (s->direction == AMDTP_IN_STREAM) {
1667 // NOTE: IT context should be used for constant IRQ.
1668 if (is_irq_target) {
1673 s->data_block_counter = UINT_MAX;
1675 s->data_block_counter = 0;
1678 // initialize packet buffer.
1679 if (s->direction == AMDTP_IN_STREAM) {
1680 dir = DMA_FROM_DEVICE;
1681 type = FW_ISO_CONTEXT_RECEIVE;
1682 if (!(s->flags & CIP_NO_HEADER))
1683 ctx_header_size = IR_CTX_HEADER_SIZE_CIP;
1685 ctx_header_size = IR_CTX_HEADER_SIZE_NO_CIP;
1687 dir = DMA_TO_DEVICE;
1688 type = FW_ISO_CONTEXT_TRANSMIT;
1689 ctx_header_size = 0; // No effect for IT context.
1691 max_ctx_payload_size = amdtp_stream_get_max_ctx_payload_size(s);
1693 err = iso_packets_buffer_init(&s->buffer, s->unit, queue_size, max_ctx_payload_size, dir);
1696 s->queue_size = queue_size;
1698 s->context = fw_iso_context_create(fw_parent_device(s->unit)->card,
1699 type, channel, speed, ctx_header_size,
1700 amdtp_stream_first_callback, s);
1701 if (IS_ERR(s->context)) {
1702 err = PTR_ERR(s->context);
1704 dev_err(&s->unit->device,
1705 "no free stream on this controller\n");
1709 amdtp_stream_update(s);
1711 if (s->direction == AMDTP_IN_STREAM) {
1712 s->ctx_data.tx.max_ctx_payload_length = max_ctx_payload_size;
1713 s->ctx_data.tx.ctx_header_size = ctx_header_size;
1714 s->ctx_data.tx.event_starts = false;
1716 if (s->domain->replay.enable) {
1717 // struct fw_iso_context.drop_overflow_headers is false therefore it's
1718 // possible to cache much unexpectedly.
1719 s->ctx_data.tx.cache.size = max_t(unsigned int, s->syt_interval * 2,
1720 queue_size * 3 / 2);
1721 s->ctx_data.tx.cache.pos = 0;
1722 s->ctx_data.tx.cache.descs = kcalloc(s->ctx_data.tx.cache.size,
1723 sizeof(*s->ctx_data.tx.cache.descs), GFP_KERNEL);
1724 if (!s->ctx_data.tx.cache.descs) {
1730 static const struct {
1731 unsigned int data_block;
1732 unsigned int syt_offset;
1733 } *entry, initial_state[] = {
1734 [CIP_SFC_32000] = { 4, 3072 },
1735 [CIP_SFC_48000] = { 6, 1024 },
1736 [CIP_SFC_96000] = { 12, 1024 },
1737 [CIP_SFC_192000] = { 24, 1024 },
1738 [CIP_SFC_44100] = { 0, 67 },
1739 [CIP_SFC_88200] = { 0, 67 },
1740 [CIP_SFC_176400] = { 0, 67 },
1743 s->ctx_data.rx.seq.descs = kcalloc(queue_size, sizeof(*s->ctx_data.rx.seq.descs), GFP_KERNEL);
1744 if (!s->ctx_data.rx.seq.descs) {
1748 s->ctx_data.rx.seq.size = queue_size;
1749 s->ctx_data.rx.seq.pos = 0;
1751 entry = &initial_state[s->sfc];
1752 s->ctx_data.rx.data_block_state = entry->data_block;
1753 s->ctx_data.rx.syt_offset_state = entry->syt_offset;
1754 s->ctx_data.rx.last_syt_offset = TICKS_PER_CYCLE;
1756 s->ctx_data.rx.event_count = 0;
1759 if (s->flags & CIP_NO_HEADER)
1760 s->tag = TAG_NO_CIP_HEADER;
1764 // NOTE: When operating without hardIRQ/softIRQ, applications tends to call ioctl request
1765 // for runtime of PCM substream in the interval equivalent to the size of PCM buffer. It
1766 // could take a round over queue of AMDTP packet descriptors and small loss of history. For
1767 // safe, keep more 8 elements for the queue, equivalent to 1 ms.
1768 descs = kcalloc(s->queue_size + 8, sizeof(*descs), GFP_KERNEL);
1773 s->packet_descs = descs;
1775 INIT_LIST_HEAD(&s->packet_descs_list);
1776 for (i = 0; i < s->queue_size; ++i) {
1777 INIT_LIST_HEAD(&descs->link);
1778 list_add_tail(&descs->link, &s->packet_descs_list);
1781 s->packet_descs_cursor = list_first_entry(&s->packet_descs_list, struct pkt_desc, link);
1783 s->packet_index = 0;
1785 struct fw_iso_packet params;
1787 if (s->direction == AMDTP_IN_STREAM) {
1788 err = queue_in_packet(s, ¶ms);
1790 bool sched_irq = false;
1792 params.header_length = 0;
1793 params.payload_length = 0;
1795 if (is_irq_target) {
1796 sched_irq = !((s->packet_index + 1) %
1800 err = queue_out_packet(s, ¶ms, sched_irq);
1804 } while (s->packet_index > 0);
1806 /* NOTE: TAG1 matches CIP. This just affects in stream. */
1807 tag = FW_ISO_CONTEXT_MATCH_TAG1;
1808 if ((s->flags & CIP_EMPTY_WITH_TAG0) || (s->flags & CIP_NO_HEADER))
1809 tag |= FW_ISO_CONTEXT_MATCH_TAG0;
1811 s->ready_processing = false;
1812 err = fw_iso_context_start(s->context, -1, 0, tag);
1816 mutex_unlock(&s->mutex);
1820 kfree(s->packet_descs);
1821 s->packet_descs = NULL;
1823 if (s->direction == AMDTP_OUT_STREAM) {
1824 kfree(s->ctx_data.rx.seq.descs);
1826 if (s->domain->replay.enable)
1827 kfree(s->ctx_data.tx.cache.descs);
1829 fw_iso_context_destroy(s->context);
1830 s->context = ERR_PTR(-1);
1832 iso_packets_buffer_destroy(&s->buffer, s->unit);
1834 mutex_unlock(&s->mutex);
1840 * amdtp_domain_stream_pcm_pointer - get the PCM buffer position
1841 * @d: the AMDTP domain.
1842 * @s: the AMDTP stream that transports the PCM data
1844 * Returns the current buffer position, in frames.
1846 unsigned long amdtp_domain_stream_pcm_pointer(struct amdtp_domain *d,
1847 struct amdtp_stream *s)
1849 struct amdtp_stream *irq_target = d->irq_target;
1851 // Process isochronous packets queued till recent isochronous cycle to handle PCM frames.
1852 if (irq_target && amdtp_stream_running(irq_target)) {
1853 // In software IRQ context, the call causes dead-lock to disable the tasklet
1856 fw_iso_context_flush_completions(irq_target->context);
1859 return READ_ONCE(s->pcm_buffer_pointer);
1861 EXPORT_SYMBOL_GPL(amdtp_domain_stream_pcm_pointer);
1864 * amdtp_domain_stream_pcm_ack - acknowledge queued PCM frames
1865 * @d: the AMDTP domain.
1866 * @s: the AMDTP stream that transfers the PCM frames
1868 * Returns zero always.
1870 int amdtp_domain_stream_pcm_ack(struct amdtp_domain *d, struct amdtp_stream *s)
1872 struct amdtp_stream *irq_target = d->irq_target;
1874 // Process isochronous packets for recent isochronous cycle to handle
1875 // queued PCM frames.
1876 if (irq_target && amdtp_stream_running(irq_target))
1877 fw_iso_context_flush_completions(irq_target->context);
1881 EXPORT_SYMBOL_GPL(amdtp_domain_stream_pcm_ack);
1884 * amdtp_stream_update - update the stream after a bus reset
1885 * @s: the AMDTP stream
1887 void amdtp_stream_update(struct amdtp_stream *s)
1890 WRITE_ONCE(s->source_node_id_field,
1891 (fw_parent_device(s->unit)->card->node_id << CIP_SID_SHIFT) & CIP_SID_MASK);
1893 EXPORT_SYMBOL(amdtp_stream_update);
1896 * amdtp_stream_stop - stop sending packets
1897 * @s: the AMDTP stream to stop
1899 * All PCM and MIDI devices of the stream must be stopped before the stream
1900 * itself can be stopped.
1902 static void amdtp_stream_stop(struct amdtp_stream *s)
1904 mutex_lock(&s->mutex);
1906 if (!amdtp_stream_running(s)) {
1907 mutex_unlock(&s->mutex);
1911 fw_iso_context_stop(s->context);
1912 fw_iso_context_destroy(s->context);
1913 s->context = ERR_PTR(-1);
1914 iso_packets_buffer_destroy(&s->buffer, s->unit);
1915 kfree(s->packet_descs);
1916 s->packet_descs = NULL;
1918 if (s->direction == AMDTP_OUT_STREAM) {
1919 kfree(s->ctx_data.rx.seq.descs);
1921 if (s->domain->replay.enable)
1922 kfree(s->ctx_data.tx.cache.descs);
1925 mutex_unlock(&s->mutex);
1929 * amdtp_stream_pcm_abort - abort the running PCM device
1930 * @s: the AMDTP stream about to be stopped
1932 * If the isochronous stream needs to be stopped asynchronously, call this
1933 * function first to stop the PCM device.
1935 void amdtp_stream_pcm_abort(struct amdtp_stream *s)
1937 struct snd_pcm_substream *pcm;
1939 pcm = READ_ONCE(s->pcm);
1941 snd_pcm_stop_xrun(pcm);
1943 EXPORT_SYMBOL(amdtp_stream_pcm_abort);
1946 * amdtp_domain_init - initialize an AMDTP domain structure
1947 * @d: the AMDTP domain to initialize.
1949 int amdtp_domain_init(struct amdtp_domain *d)
1951 INIT_LIST_HEAD(&d->streams);
1953 d->events_per_period = 0;
1957 EXPORT_SYMBOL_GPL(amdtp_domain_init);
1960 * amdtp_domain_destroy - destroy an AMDTP domain structure
1961 * @d: the AMDTP domain to destroy.
1963 void amdtp_domain_destroy(struct amdtp_domain *d)
1965 // At present nothing to do.
1968 EXPORT_SYMBOL_GPL(amdtp_domain_destroy);
1971 * amdtp_domain_add_stream - register isoc context into the domain.
1972 * @d: the AMDTP domain.
1973 * @s: the AMDTP stream.
1974 * @channel: the isochronous channel on the bus.
1975 * @speed: firewire speed code.
1977 int amdtp_domain_add_stream(struct amdtp_domain *d, struct amdtp_stream *s,
1978 int channel, int speed)
1980 struct amdtp_stream *tmp;
1982 list_for_each_entry(tmp, &d->streams, list) {
1987 list_add(&s->list, &d->streams);
1989 s->channel = channel;
1995 EXPORT_SYMBOL_GPL(amdtp_domain_add_stream);
1997 // Make the reference from rx stream to tx stream for sequence replay. When the number of tx streams
1998 // is less than the number of rx streams, the first tx stream is selected.
1999 static int make_association(struct amdtp_domain *d)
2001 unsigned int dst_index = 0;
2002 struct amdtp_stream *rx;
2004 // Make association to replay target.
2005 list_for_each_entry(rx, &d->streams, list) {
2006 if (rx->direction == AMDTP_OUT_STREAM) {
2007 unsigned int src_index = 0;
2008 struct amdtp_stream *tx = NULL;
2009 struct amdtp_stream *s;
2011 list_for_each_entry(s, &d->streams, list) {
2012 if (s->direction == AMDTP_IN_STREAM) {
2013 if (dst_index == src_index) {
2022 // Select the first entry.
2023 list_for_each_entry(s, &d->streams, list) {
2024 if (s->direction == AMDTP_IN_STREAM) {
2029 // No target is available to replay sequence.
2034 rx->ctx_data.rx.replay_target = tx;
2044 * amdtp_domain_start - start sending packets for isoc context in the domain.
2045 * @d: the AMDTP domain.
2046 * @tx_init_skip_cycles: the number of cycles to skip processing packets at initial stage of IR
2048 * @replay_seq: whether to replay the sequence of packet in IR context for the sequence of packet in
2050 * @replay_on_the_fly: transfer rx packets according to nominal frequency, then begin to replay
2051 * according to arrival of events in tx packets.
2053 int amdtp_domain_start(struct amdtp_domain *d, unsigned int tx_init_skip_cycles, bool replay_seq,
2054 bool replay_on_the_fly)
2056 unsigned int events_per_buffer = d->events_per_buffer;
2057 unsigned int events_per_period = d->events_per_period;
2058 unsigned int queue_size;
2059 struct amdtp_stream *s;
2064 err = make_association(d);
2068 d->replay.enable = replay_seq;
2069 d->replay.on_the_fly = replay_on_the_fly;
2071 // Select an IT context as IRQ target.
2072 list_for_each_entry(s, &d->streams, list) {
2073 if (s->direction == AMDTP_OUT_STREAM) {
2082 d->processing_cycle.tx_init_skip = tx_init_skip_cycles;
2084 // This is a case that AMDTP streams in domain run just for MIDI
2085 // substream. Use the number of events equivalent to 10 msec as
2086 // interval of hardware IRQ.
2087 if (events_per_period == 0)
2088 events_per_period = amdtp_rate_table[d->irq_target->sfc] / 100;
2089 if (events_per_buffer == 0)
2090 events_per_buffer = events_per_period * 3;
2092 queue_size = DIV_ROUND_UP(CYCLES_PER_SECOND * events_per_buffer,
2093 amdtp_rate_table[d->irq_target->sfc]);
2095 list_for_each_entry(s, &d->streams, list) {
2096 unsigned int idle_irq_interval = 0;
2098 if (s->direction == AMDTP_OUT_STREAM && s == d->irq_target) {
2099 idle_irq_interval = DIV_ROUND_UP(CYCLES_PER_SECOND * events_per_period,
2100 amdtp_rate_table[d->irq_target->sfc]);
2103 // Starts immediately but actually DMA context starts several hundred cycles later.
2104 err = amdtp_stream_start(s, s->channel, s->speed, queue_size, idle_irq_interval);
2111 list_for_each_entry(s, &d->streams, list)
2112 amdtp_stream_stop(s);
2115 EXPORT_SYMBOL_GPL(amdtp_domain_start);
2118 * amdtp_domain_stop - stop sending packets for isoc context in the same domain.
2119 * @d: the AMDTP domain to which the isoc contexts belong.
2121 void amdtp_domain_stop(struct amdtp_domain *d)
2123 struct amdtp_stream *s, *next;
2126 amdtp_stream_stop(d->irq_target);
2128 list_for_each_entry_safe(s, next, &d->streams, list) {
2131 if (s != d->irq_target)
2132 amdtp_stream_stop(s);
2135 d->events_per_period = 0;
2136 d->irq_target = NULL;
2138 EXPORT_SYMBOL_GPL(amdtp_domain_stop);