Merge tag 'vfio-v3.9-rc7' of git://github.com/awilliam/linux-vfio
[sfrench/cifs-2.6.git] / sound / firewire / amdtp.c
1 /*
2  * Audio and Music Data Transmission Protocol (IEC 61883-6) streams
3  * with Common Isochronous Packet (IEC 61883-1) headers
4  *
5  * Copyright (c) Clemens Ladisch <clemens@ladisch.de>
6  * Licensed under the terms of the GNU General Public License, version 2.
7  */
8
9 #include <linux/device.h>
10 #include <linux/err.h>
11 #include <linux/firewire.h>
12 #include <linux/module.h>
13 #include <linux/slab.h>
14 #include <sound/pcm.h>
15 #include "amdtp.h"
16
17 #define TICKS_PER_CYCLE         3072
18 #define CYCLES_PER_SECOND       8000
19 #define TICKS_PER_SECOND        (TICKS_PER_CYCLE * CYCLES_PER_SECOND)
20
21 #define TRANSFER_DELAY_TICKS    0x2e00 /* 479.17 µs */
22
23 #define TAG_CIP                 1
24
25 #define CIP_EOH                 (1u << 31)
26 #define CIP_FMT_AM              (0x10 << 24)
27 #define AMDTP_FDF_AM824         (0 << 19)
28 #define AMDTP_FDF_SFC_SHIFT     16
29
30 /* TODO: make these configurable */
31 #define INTERRUPT_INTERVAL      16
32 #define QUEUE_LENGTH            48
33
34 static void pcm_period_tasklet(unsigned long data);
35
36 /**
37  * amdtp_out_stream_init - initialize an AMDTP output stream structure
38  * @s: the AMDTP output stream to initialize
39  * @unit: the target of the stream
40  * @flags: the packet transmission method to use
41  */
42 int amdtp_out_stream_init(struct amdtp_out_stream *s, struct fw_unit *unit,
43                           enum cip_out_flags flags)
44 {
45         if (flags != CIP_NONBLOCKING)
46                 return -EINVAL;
47
48         s->unit = fw_unit_get(unit);
49         s->flags = flags;
50         s->context = ERR_PTR(-1);
51         mutex_init(&s->mutex);
52         tasklet_init(&s->period_tasklet, pcm_period_tasklet, (unsigned long)s);
53         s->packet_index = 0;
54
55         return 0;
56 }
57 EXPORT_SYMBOL(amdtp_out_stream_init);
58
59 /**
60  * amdtp_out_stream_destroy - free stream resources
61  * @s: the AMDTP output stream to destroy
62  */
63 void amdtp_out_stream_destroy(struct amdtp_out_stream *s)
64 {
65         WARN_ON(!IS_ERR(s->context));
66         mutex_destroy(&s->mutex);
67         fw_unit_put(s->unit);
68 }
69 EXPORT_SYMBOL(amdtp_out_stream_destroy);
70
71 /**
72  * amdtp_out_stream_set_rate - set the sample rate
73  * @s: the AMDTP output stream to configure
74  * @rate: the sample rate
75  *
76  * The sample rate must be set before the stream is started, and must not be
77  * changed while the stream is running.
78  */
79 void amdtp_out_stream_set_rate(struct amdtp_out_stream *s, unsigned int rate)
80 {
81         static const struct {
82                 unsigned int rate;
83                 unsigned int syt_interval;
84         } rate_info[] = {
85                 [CIP_SFC_32000]  = {  32000,  8, },
86                 [CIP_SFC_44100]  = {  44100,  8, },
87                 [CIP_SFC_48000]  = {  48000,  8, },
88                 [CIP_SFC_88200]  = {  88200, 16, },
89                 [CIP_SFC_96000]  = {  96000, 16, },
90                 [CIP_SFC_176400] = { 176400, 32, },
91                 [CIP_SFC_192000] = { 192000, 32, },
92         };
93         unsigned int sfc;
94
95         if (WARN_ON(!IS_ERR(s->context)))
96                 return;
97
98         for (sfc = 0; sfc < ARRAY_SIZE(rate_info); ++sfc)
99                 if (rate_info[sfc].rate == rate) {
100                         s->sfc = sfc;
101                         s->syt_interval = rate_info[sfc].syt_interval;
102                         return;
103                 }
104         WARN_ON(1);
105 }
106 EXPORT_SYMBOL(amdtp_out_stream_set_rate);
107
108 /**
109  * amdtp_out_stream_get_max_payload - get the stream's packet size
110  * @s: the AMDTP output stream
111  *
112  * This function must not be called before the stream has been configured
113  * with amdtp_out_stream_set_hw_params(), amdtp_out_stream_set_pcm(), and
114  * amdtp_out_stream_set_midi().
115  */
116 unsigned int amdtp_out_stream_get_max_payload(struct amdtp_out_stream *s)
117 {
118         static const unsigned int max_data_blocks[] = {
119                 [CIP_SFC_32000]  =  4,
120                 [CIP_SFC_44100]  =  6,
121                 [CIP_SFC_48000]  =  6,
122                 [CIP_SFC_88200]  = 12,
123                 [CIP_SFC_96000]  = 12,
124                 [CIP_SFC_176400] = 23,
125                 [CIP_SFC_192000] = 24,
126         };
127
128         s->data_block_quadlets = s->pcm_channels;
129         s->data_block_quadlets += DIV_ROUND_UP(s->midi_ports, 8);
130
131         return 8 + max_data_blocks[s->sfc] * 4 * s->data_block_quadlets;
132 }
133 EXPORT_SYMBOL(amdtp_out_stream_get_max_payload);
134
135 static void amdtp_write_s16(struct amdtp_out_stream *s,
136                             struct snd_pcm_substream *pcm,
137                             __be32 *buffer, unsigned int frames);
138 static void amdtp_write_s32(struct amdtp_out_stream *s,
139                             struct snd_pcm_substream *pcm,
140                             __be32 *buffer, unsigned int frames);
141
142 /**
143  * amdtp_out_stream_set_pcm_format - set the PCM format
144  * @s: the AMDTP output stream to configure
145  * @format: the format of the ALSA PCM device
146  *
147  * The sample format must be set before the stream is started, and must not be
148  * changed while the stream is running.
149  */
150 void amdtp_out_stream_set_pcm_format(struct amdtp_out_stream *s,
151                                      snd_pcm_format_t format)
152 {
153         if (WARN_ON(!IS_ERR(s->context)))
154                 return;
155
156         switch (format) {
157         default:
158                 WARN_ON(1);
159                 /* fall through */
160         case SNDRV_PCM_FORMAT_S16:
161                 s->transfer_samples = amdtp_write_s16;
162                 break;
163         case SNDRV_PCM_FORMAT_S32:
164                 s->transfer_samples = amdtp_write_s32;
165                 break;
166         }
167 }
168 EXPORT_SYMBOL(amdtp_out_stream_set_pcm_format);
169
170 /**
171  * amdtp_out_stream_pcm_prepare - prepare PCM device for running
172  * @s: the AMDTP output stream
173  *
174  * This function should be called from the PCM device's .prepare callback.
175  */
176 void amdtp_out_stream_pcm_prepare(struct amdtp_out_stream *s)
177 {
178         tasklet_kill(&s->period_tasklet);
179         s->pcm_buffer_pointer = 0;
180         s->pcm_period_pointer = 0;
181         s->pointer_flush = true;
182 }
183 EXPORT_SYMBOL(amdtp_out_stream_pcm_prepare);
184
185 static unsigned int calculate_data_blocks(struct amdtp_out_stream *s)
186 {
187         unsigned int phase, data_blocks;
188
189         if (!cip_sfc_is_base_44100(s->sfc)) {
190                 /* Sample_rate / 8000 is an integer, and precomputed. */
191                 data_blocks = s->data_block_state;
192         } else {
193                 phase = s->data_block_state;
194
195                 /*
196                  * This calculates the number of data blocks per packet so that
197                  * 1) the overall rate is correct and exactly synchronized to
198                  *    the bus clock, and
199                  * 2) packets with a rounded-up number of blocks occur as early
200                  *    as possible in the sequence (to prevent underruns of the
201                  *    device's buffer).
202                  */
203                 if (s->sfc == CIP_SFC_44100)
204                         /* 6 6 5 6 5 6 5 ... */
205                         data_blocks = 5 + ((phase & 1) ^
206                                            (phase == 0 || phase >= 40));
207                 else
208                         /* 12 11 11 11 11 ... or 23 22 22 22 22 ... */
209                         data_blocks = 11 * (s->sfc >> 1) + (phase == 0);
210                 if (++phase >= (80 >> (s->sfc >> 1)))
211                         phase = 0;
212                 s->data_block_state = phase;
213         }
214
215         return data_blocks;
216 }
217
218 static unsigned int calculate_syt(struct amdtp_out_stream *s,
219                                   unsigned int cycle)
220 {
221         unsigned int syt_offset, phase, index, syt;
222
223         if (s->last_syt_offset < TICKS_PER_CYCLE) {
224                 if (!cip_sfc_is_base_44100(s->sfc))
225                         syt_offset = s->last_syt_offset + s->syt_offset_state;
226                 else {
227                 /*
228                  * The time, in ticks, of the n'th SYT_INTERVAL sample is:
229                  *   n * SYT_INTERVAL * 24576000 / sample_rate
230                  * Modulo TICKS_PER_CYCLE, the difference between successive
231                  * elements is about 1386.23.  Rounding the results of this
232                  * formula to the SYT precision results in a sequence of
233                  * differences that begins with:
234                  *   1386 1386 1387 1386 1386 1386 1387 1386 1386 1386 1387 ...
235                  * This code generates _exactly_ the same sequence.
236                  */
237                         phase = s->syt_offset_state;
238                         index = phase % 13;
239                         syt_offset = s->last_syt_offset;
240                         syt_offset += 1386 + ((index && !(index & 3)) ||
241                                               phase == 146);
242                         if (++phase >= 147)
243                                 phase = 0;
244                         s->syt_offset_state = phase;
245                 }
246         } else
247                 syt_offset = s->last_syt_offset - TICKS_PER_CYCLE;
248         s->last_syt_offset = syt_offset;
249
250         if (syt_offset < TICKS_PER_CYCLE) {
251                 syt_offset += TRANSFER_DELAY_TICKS - TICKS_PER_CYCLE;
252                 syt = (cycle + syt_offset / TICKS_PER_CYCLE) << 12;
253                 syt += syt_offset % TICKS_PER_CYCLE;
254
255                 return syt & 0xffff;
256         } else {
257                 return 0xffff; /* no info */
258         }
259 }
260
261 static void amdtp_write_s32(struct amdtp_out_stream *s,
262                             struct snd_pcm_substream *pcm,
263                             __be32 *buffer, unsigned int frames)
264 {
265         struct snd_pcm_runtime *runtime = pcm->runtime;
266         unsigned int channels, remaining_frames, frame_step, i, c;
267         const u32 *src;
268
269         channels = s->pcm_channels;
270         src = (void *)runtime->dma_area +
271                         s->pcm_buffer_pointer * (runtime->frame_bits / 8);
272         remaining_frames = runtime->buffer_size - s->pcm_buffer_pointer;
273         frame_step = s->data_block_quadlets - channels;
274
275         for (i = 0; i < frames; ++i) {
276                 for (c = 0; c < channels; ++c) {
277                         *buffer = cpu_to_be32((*src >> 8) | 0x40000000);
278                         src++;
279                         buffer++;
280                 }
281                 buffer += frame_step;
282                 if (--remaining_frames == 0)
283                         src = (void *)runtime->dma_area;
284         }
285 }
286
287 static void amdtp_write_s16(struct amdtp_out_stream *s,
288                             struct snd_pcm_substream *pcm,
289                             __be32 *buffer, unsigned int frames)
290 {
291         struct snd_pcm_runtime *runtime = pcm->runtime;
292         unsigned int channels, remaining_frames, frame_step, i, c;
293         const u16 *src;
294
295         channels = s->pcm_channels;
296         src = (void *)runtime->dma_area +
297                         s->pcm_buffer_pointer * (runtime->frame_bits / 8);
298         remaining_frames = runtime->buffer_size - s->pcm_buffer_pointer;
299         frame_step = s->data_block_quadlets - channels;
300
301         for (i = 0; i < frames; ++i) {
302                 for (c = 0; c < channels; ++c) {
303                         *buffer = cpu_to_be32((*src << 8) | 0x40000000);
304                         src++;
305                         buffer++;
306                 }
307                 buffer += frame_step;
308                 if (--remaining_frames == 0)
309                         src = (void *)runtime->dma_area;
310         }
311 }
312
313 static void amdtp_fill_pcm_silence(struct amdtp_out_stream *s,
314                                    __be32 *buffer, unsigned int frames)
315 {
316         unsigned int i, c;
317
318         for (i = 0; i < frames; ++i) {
319                 for (c = 0; c < s->pcm_channels; ++c)
320                         buffer[c] = cpu_to_be32(0x40000000);
321                 buffer += s->data_block_quadlets;
322         }
323 }
324
325 static void amdtp_fill_midi(struct amdtp_out_stream *s,
326                             __be32 *buffer, unsigned int frames)
327 {
328         unsigned int i;
329
330         for (i = 0; i < frames; ++i)
331                 buffer[s->pcm_channels + i * s->data_block_quadlets] =
332                                                 cpu_to_be32(0x80000000);
333 }
334
335 static void queue_out_packet(struct amdtp_out_stream *s, unsigned int cycle)
336 {
337         __be32 *buffer;
338         unsigned int index, data_blocks, syt, ptr;
339         struct snd_pcm_substream *pcm;
340         struct fw_iso_packet packet;
341         int err;
342
343         if (s->packet_index < 0)
344                 return;
345         index = s->packet_index;
346
347         data_blocks = calculate_data_blocks(s);
348         syt = calculate_syt(s, cycle);
349
350         buffer = s->buffer.packets[index].buffer;
351         buffer[0] = cpu_to_be32(ACCESS_ONCE(s->source_node_id_field) |
352                                 (s->data_block_quadlets << 16) |
353                                 s->data_block_counter);
354         buffer[1] = cpu_to_be32(CIP_EOH | CIP_FMT_AM | AMDTP_FDF_AM824 |
355                                 (s->sfc << AMDTP_FDF_SFC_SHIFT) | syt);
356         buffer += 2;
357
358         pcm = ACCESS_ONCE(s->pcm);
359         if (pcm)
360                 s->transfer_samples(s, pcm, buffer, data_blocks);
361         else
362                 amdtp_fill_pcm_silence(s, buffer, data_blocks);
363         if (s->midi_ports)
364                 amdtp_fill_midi(s, buffer, data_blocks);
365
366         s->data_block_counter = (s->data_block_counter + data_blocks) & 0xff;
367
368         packet.payload_length = 8 + data_blocks * 4 * s->data_block_quadlets;
369         packet.interrupt = IS_ALIGNED(index + 1, INTERRUPT_INTERVAL);
370         packet.skip = 0;
371         packet.tag = TAG_CIP;
372         packet.sy = 0;
373         packet.header_length = 0;
374
375         err = fw_iso_context_queue(s->context, &packet, &s->buffer.iso_buffer,
376                                    s->buffer.packets[index].offset);
377         if (err < 0) {
378                 dev_err(&s->unit->device, "queueing error: %d\n", err);
379                 s->packet_index = -1;
380                 amdtp_out_stream_pcm_abort(s);
381                 return;
382         }
383
384         if (++index >= QUEUE_LENGTH)
385                 index = 0;
386         s->packet_index = index;
387
388         if (pcm) {
389                 ptr = s->pcm_buffer_pointer + data_blocks;
390                 if (ptr >= pcm->runtime->buffer_size)
391                         ptr -= pcm->runtime->buffer_size;
392                 ACCESS_ONCE(s->pcm_buffer_pointer) = ptr;
393
394                 s->pcm_period_pointer += data_blocks;
395                 if (s->pcm_period_pointer >= pcm->runtime->period_size) {
396                         s->pcm_period_pointer -= pcm->runtime->period_size;
397                         s->pointer_flush = false;
398                         tasklet_hi_schedule(&s->period_tasklet);
399                 }
400         }
401 }
402
403 static void pcm_period_tasklet(unsigned long data)
404 {
405         struct amdtp_out_stream *s = (void *)data;
406         struct snd_pcm_substream *pcm = ACCESS_ONCE(s->pcm);
407
408         if (pcm)
409                 snd_pcm_period_elapsed(pcm);
410 }
411
412 static void out_packet_callback(struct fw_iso_context *context, u32 cycle,
413                                 size_t header_length, void *header, void *data)
414 {
415         struct amdtp_out_stream *s = data;
416         unsigned int i, packets = header_length / 4;
417
418         /*
419          * Compute the cycle of the last queued packet.
420          * (We need only the four lowest bits for the SYT, so we can ignore
421          * that bits 0-11 must wrap around at 3072.)
422          */
423         cycle += QUEUE_LENGTH - packets;
424
425         for (i = 0; i < packets; ++i)
426                 queue_out_packet(s, ++cycle);
427         fw_iso_context_queue_flush(s->context);
428 }
429
430 static int queue_initial_skip_packets(struct amdtp_out_stream *s)
431 {
432         struct fw_iso_packet skip_packet = {
433                 .skip = 1,
434         };
435         unsigned int i;
436         int err;
437
438         for (i = 0; i < QUEUE_LENGTH; ++i) {
439                 skip_packet.interrupt = IS_ALIGNED(s->packet_index + 1,
440                                                    INTERRUPT_INTERVAL);
441                 err = fw_iso_context_queue(s->context, &skip_packet, NULL, 0);
442                 if (err < 0)
443                         return err;
444                 if (++s->packet_index >= QUEUE_LENGTH)
445                         s->packet_index = 0;
446         }
447
448         return 0;
449 }
450
451 /**
452  * amdtp_out_stream_start - start sending packets
453  * @s: the AMDTP output stream to start
454  * @channel: the isochronous channel on the bus
455  * @speed: firewire speed code
456  *
457  * The stream cannot be started until it has been configured with
458  * amdtp_out_stream_set_hw_params(), amdtp_out_stream_set_pcm(), and
459  * amdtp_out_stream_set_midi(); and it must be started before any
460  * PCM or MIDI device can be started.
461  */
462 int amdtp_out_stream_start(struct amdtp_out_stream *s, int channel, int speed)
463 {
464         static const struct {
465                 unsigned int data_block;
466                 unsigned int syt_offset;
467         } initial_state[] = {
468                 [CIP_SFC_32000]  = {  4, 3072 },
469                 [CIP_SFC_48000]  = {  6, 1024 },
470                 [CIP_SFC_96000]  = { 12, 1024 },
471                 [CIP_SFC_192000] = { 24, 1024 },
472                 [CIP_SFC_44100]  = {  0,   67 },
473                 [CIP_SFC_88200]  = {  0,   67 },
474                 [CIP_SFC_176400] = {  0,   67 },
475         };
476         int err;
477
478         mutex_lock(&s->mutex);
479
480         if (WARN_ON(!IS_ERR(s->context) ||
481                     (!s->pcm_channels && !s->midi_ports))) {
482                 err = -EBADFD;
483                 goto err_unlock;
484         }
485
486         s->data_block_state = initial_state[s->sfc].data_block;
487         s->syt_offset_state = initial_state[s->sfc].syt_offset;
488         s->last_syt_offset = TICKS_PER_CYCLE;
489
490         err = iso_packets_buffer_init(&s->buffer, s->unit, QUEUE_LENGTH,
491                                       amdtp_out_stream_get_max_payload(s),
492                                       DMA_TO_DEVICE);
493         if (err < 0)
494                 goto err_unlock;
495
496         s->context = fw_iso_context_create(fw_parent_device(s->unit)->card,
497                                            FW_ISO_CONTEXT_TRANSMIT,
498                                            channel, speed, 0,
499                                            out_packet_callback, s);
500         if (IS_ERR(s->context)) {
501                 err = PTR_ERR(s->context);
502                 if (err == -EBUSY)
503                         dev_err(&s->unit->device,
504                                 "no free output stream on this controller\n");
505                 goto err_buffer;
506         }
507
508         amdtp_out_stream_update(s);
509
510         s->packet_index = 0;
511         s->data_block_counter = 0;
512         err = queue_initial_skip_packets(s);
513         if (err < 0)
514                 goto err_context;
515
516         err = fw_iso_context_start(s->context, -1, 0, 0);
517         if (err < 0)
518                 goto err_context;
519
520         mutex_unlock(&s->mutex);
521
522         return 0;
523
524 err_context:
525         fw_iso_context_destroy(s->context);
526         s->context = ERR_PTR(-1);
527 err_buffer:
528         iso_packets_buffer_destroy(&s->buffer, s->unit);
529 err_unlock:
530         mutex_unlock(&s->mutex);
531
532         return err;
533 }
534 EXPORT_SYMBOL(amdtp_out_stream_start);
535
536 /**
537  * amdtp_out_stream_pcm_pointer - get the PCM buffer position
538  * @s: the AMDTP output stream that transports the PCM data
539  *
540  * Returns the current buffer position, in frames.
541  */
542 unsigned long amdtp_out_stream_pcm_pointer(struct amdtp_out_stream *s)
543 {
544         /* this optimization is allowed to be racy */
545         if (s->pointer_flush)
546                 fw_iso_context_flush_completions(s->context);
547         else
548                 s->pointer_flush = true;
549
550         return ACCESS_ONCE(s->pcm_buffer_pointer);
551 }
552 EXPORT_SYMBOL(amdtp_out_stream_pcm_pointer);
553
554 /**
555  * amdtp_out_stream_update - update the stream after a bus reset
556  * @s: the AMDTP output stream
557  */
558 void amdtp_out_stream_update(struct amdtp_out_stream *s)
559 {
560         ACCESS_ONCE(s->source_node_id_field) =
561                 (fw_parent_device(s->unit)->card->node_id & 0x3f) << 24;
562 }
563 EXPORT_SYMBOL(amdtp_out_stream_update);
564
565 /**
566  * amdtp_out_stream_stop - stop sending packets
567  * @s: the AMDTP output stream to stop
568  *
569  * All PCM and MIDI devices of the stream must be stopped before the stream
570  * itself can be stopped.
571  */
572 void amdtp_out_stream_stop(struct amdtp_out_stream *s)
573 {
574         mutex_lock(&s->mutex);
575
576         if (IS_ERR(s->context)) {
577                 mutex_unlock(&s->mutex);
578                 return;
579         }
580
581         tasklet_kill(&s->period_tasklet);
582         fw_iso_context_stop(s->context);
583         fw_iso_context_destroy(s->context);
584         s->context = ERR_PTR(-1);
585         iso_packets_buffer_destroy(&s->buffer, s->unit);
586
587         mutex_unlock(&s->mutex);
588 }
589 EXPORT_SYMBOL(amdtp_out_stream_stop);
590
591 /**
592  * amdtp_out_stream_pcm_abort - abort the running PCM device
593  * @s: the AMDTP stream about to be stopped
594  *
595  * If the isochronous stream needs to be stopped asynchronously, call this
596  * function first to stop the PCM device.
597  */
598 void amdtp_out_stream_pcm_abort(struct amdtp_out_stream *s)
599 {
600         struct snd_pcm_substream *pcm;
601
602         pcm = ACCESS_ONCE(s->pcm);
603         if (pcm) {
604                 snd_pcm_stream_lock_irq(pcm);
605                 if (snd_pcm_running(pcm))
606                         snd_pcm_stop(pcm, SNDRV_PCM_STATE_XRUN);
607                 snd_pcm_stream_unlock_irq(pcm);
608         }
609 }
610 EXPORT_SYMBOL(amdtp_out_stream_pcm_abort);