linux-stable/sound/firewire/amdtp-stream.c

2141 lines
60 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-only
/*
* Audio and Music Data Transmission Protocol (IEC 61883-6) streams
* with Common Isochronous Packet (IEC 61883-1) headers
*
* Copyright (c) Clemens Ladisch <clemens@ladisch.de>
*/
#include <linux/device.h>
#include <linux/err.h>
#include <linux/firewire.h>
#include <linux/firewire-constants.h>
#include <linux/module.h>
#include <linux/slab.h>
#include <sound/pcm.h>
#include <sound/pcm_params.h>
#include "amdtp-stream.h"
#define TICKS_PER_CYCLE 3072
#define CYCLES_PER_SECOND 8000
#define TICKS_PER_SECOND (TICKS_PER_CYCLE * CYCLES_PER_SECOND)
#define OHCI_SECOND_MODULUS 8
ALSA: firewire-lib: add tracepoints to dump a part of isochronous packet data When audio and music units have some quirks in their sequence of packet, it's really hard for non-owners to identify the quirks. Although developers need dumps for sequence of packets, it's difficult for users who have no knowledges and no equipments for this purpose. This commit adds tracepoints for this situation. When users encounter the issue, they can dump a part of packet data via Linux tracing framework as long as using drivers in ALSA firewire stack. Additionally, tracepoints for outgoing packets will be our help to check and debug packet processing of ALSA firewire stack. This commit newly adds 'snd_firewire_lib' subsystem with 'in_packet' and 'out_packet' events. In the events, some attributes of packets and the index of packet managed by this module are recorded per packet. This is an usage: $ trace-cmd record -e snd_firewire_lib:out_packet \ -e snd_firewire_lib:in_packet /sys/kernel/tracing/events/snd_firewire_lib/out_packet/filter /sys/kernel/tracing/events/snd_firewire_lib/in_packet/filter Hit Ctrl^C to stop recording ^C $ trace-cmd report trace.dat ... 23647.033934: in_packet: 01 4073 ffc0 ffc1 00 000f0040 9001b2d1 122 44 23647.033936: in_packet: 01 4074 ffc0 ffc1 00 000f0048 9001c83b 122 45 23647.033937: in_packet: 01 4075 ffc0 ffc1 00 000f0050 9001ffff 002 46 23647.033938: in_packet: 01 4076 ffc0 ffc1 00 000f0050 9001e1a6 122 47 23647.035426: out_packet: 01 4123 ffc1 ffc0 01 010f00d0 9001fb40 122 17 23647.035428: out_packet: 01 4124 ffc1 ffc0 01 010f00d8 9001ffff 002 18 23647.035429: out_packet: 01 4125 ffc1 ffc0 01 010f00d8 900114aa 122 19 23647.035430: out_packet: 01 4126 ffc1 ffc0 01 010f00e0 90012a15 122 20 (Here, some common fields are omitted so that a line to be within 80 characters.) ... One line represent one packet. The legend for the last nine fields is: - The second of cycle scheduled for the packet - The count of cycle scheduled for the packet - The ID of node as source (hex) - Some devices transfer packets with invalid source node ID in their CIP header. - The ID of node as destination (hex) - The value is not in CIP header of packets. - The value of isochronous channel - The first quadlet of CIP header (hex) - The second quadlet of CIP header (hex) - The number of included quadlets - The index of packet in a buffer maintained by this module This is an example to parse these lines from text file by Python3 script: \#!/usr/bin/env python3 import sys def parse_ts(second, cycle, syt): offset = syt & 0xfff syt >>= 12 if cycle & 0x0f > syt: cycle += 0x10 cycle &= 0x1ff0 cycle |= syt second += cycle // 8000 cycle %= 8000 # In CYCLE_TIMER of 1394 OHCI, second is represented in 8 bit. second %= 128 return (second, cycle, offset) def calc_ts(second, cycle, offset): ts = offset ts += cycle * 3072 # In DMA descriptor of 1394 OHCI, second is represented in 3 bit. ts += (second % 8) * 8000 * 3072 return ts def subtract_ts(minuend, subtrahend): # In DMA descriptor of 1394 OHCI, second is represented in 3 bit. if minuend < subtrahend: minuend += 8 * 8000 * 3072 return minuend - subtrahend if len(sys.argv) != 2: print('At least, one argument is required for packet dump.') sys.exit() filename = sys.argv[1] data = [] prev = 0 with open(filename, 'r') as f: for line in f: pos = line.find('packet:') if pos < 0: continue pos += len('packet:') line = line[pos:].strip() fields = line.split(' ') datum = [] datum.append(fields[8]) syt = int(fields[6][4:], 16) # Empty packet in IEC 61883-1, or NODATA in IEC 61883-6 if syt == 0xffff: data_blocks = 0 else: payload_size = int(fields[7], 10) data_block_size = int(fields[5][2:4], 16) data_blocks = (payload_size - 2) / data_block_size datum.append(data_blocks) second = int(fields[0], 10) cycle = int(fields[1], 10) start = (second << 25) | (cycle << 12) datum.append('0x{0:08x}'.format(start)) start = calc_ts(second, cycle, 0) datum.append("0x" + fields[5]) datum.append("0x" + fields[6]) if syt == 0xffff: second = 0 cycle = 0 tick = 0 else: second, cycle, tick = parse_ts(second, cycle, syt) ts = calc_ts(second, cycle, tick) datum.append(start) datum.append(ts) if ts == 0: datum.append(0) datum.append(0) else: # Usual case, or a case over 8 seconds. if ts > start or start > 7 * 8000 * 3072: datum.append(subtract_ts(ts, start)) if ts > prev or start > 7 * 8000 * 3072: gap = subtract_ts(ts, prev) datum.append(gap) else: datum.append('backward') else: datum.append('invalid') prev = ts data.append(datum) sys.exit() The data variable includes array with these elements: - The index of the packet - The number of data blocks in the packet - The value of cycle count (hex) - The value of CIP header 1 (hex) - The value of CIP header 2 (hex) - The value of cycle count (tick) - The value of calculated presentation timestamp (tick) - The offset between the cycle count and presentation timestamp - The elapsed ticks from the previous presentation timestamp Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Iwai <tiwai@suse.de>
2016-05-09 12:12:46 +00:00
/* Always support Linux tracing subsystem. */
#define CREATE_TRACE_POINTS
#include "amdtp-stream-trace.h"
#define TRANSFER_DELAY_TICKS 0x2e00 /* 479.17 microseconds */
/* isochronous header parameters */
#define ISO_DATA_LENGTH_SHIFT 16
2017-03-31 13:06:07 +00:00
#define TAG_NO_CIP_HEADER 0
#define TAG_CIP 1
// Common Isochronous Packet (CIP) header parameters. Use two quadlets CIP header when supported.
#define CIP_HEADER_QUADLETS 2
#define CIP_EOH_SHIFT 31
#define CIP_EOH (1u << CIP_EOH_SHIFT)
#define CIP_EOH_MASK 0x80000000
#define CIP_SID_SHIFT 24
#define CIP_SID_MASK 0x3f000000
#define CIP_DBS_MASK 0x00ff0000
#define CIP_DBS_SHIFT 16
#define CIP_SPH_MASK 0x00000400
#define CIP_SPH_SHIFT 10
#define CIP_DBC_MASK 0x000000ff
#define CIP_FMT_SHIFT 24
#define CIP_FMT_MASK 0x3f000000
#define CIP_FDF_MASK 0x00ff0000
#define CIP_FDF_SHIFT 16
#define CIP_FDF_NO_DATA 0xff
#define CIP_SYT_MASK 0x0000ffff
#define CIP_SYT_NO_INFO 0xffff
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
#define CIP_SYT_CYCLE_MODULUS 16
#define CIP_NO_DATA ((CIP_FDF_NO_DATA << CIP_FDF_SHIFT) | CIP_SYT_NO_INFO)
#define CIP_HEADER_SIZE (sizeof(__be32) * CIP_HEADER_QUADLETS)
/* Audio and Music transfer protocol specific parameters */
#define CIP_FMT_AM 0x10
#define AMDTP_FDF_NO_DATA 0xff
// For iso header and tstamp.
#define IR_CTX_HEADER_DEFAULT_QUADLETS 2
// Add nothing.
#define IR_CTX_HEADER_SIZE_NO_CIP (sizeof(__be32) * IR_CTX_HEADER_DEFAULT_QUADLETS)
// Add two quadlets CIP header.
#define IR_CTX_HEADER_SIZE_CIP (IR_CTX_HEADER_SIZE_NO_CIP + CIP_HEADER_SIZE)
#define HEADER_TSTAMP_MASK 0x0000ffff
#define IT_PKT_HEADER_SIZE_CIP CIP_HEADER_SIZE
#define IT_PKT_HEADER_SIZE_NO_CIP 0 // Nothing.
// The initial firmware of OXFW970 can postpone transmission of packet during finishing
// asynchronous transaction. This module accepts 5 cycles to skip as maximum to avoid buffer
// overrun. Actual device can skip more, then this module stops the packet streaming.
#define IR_JUMBO_PAYLOAD_MAX_SKIP_CYCLES 5
/**
* amdtp_stream_init - initialize an AMDTP stream structure
* @s: the AMDTP stream to initialize
* @unit: the target of the stream
* @dir: the direction of stream
* @flags: the details of the streaming protocol consist of cip_flags enumeration-constants.
* @fmt: the value of fmt field in CIP header
* @process_ctx_payloads: callback handler to process payloads of isoc context
* @protocol_size: the size to allocate newly for protocol
*/
int amdtp_stream_init(struct amdtp_stream *s, struct fw_unit *unit,
enum amdtp_stream_direction dir, unsigned int flags,
unsigned int fmt,
amdtp_stream_process_ctx_payloads_t process_ctx_payloads,
unsigned int protocol_size)
{
if (process_ctx_payloads == NULL)
return -EINVAL;
s->protocol = kzalloc(protocol_size, GFP_KERNEL);
if (!s->protocol)
return -ENOMEM;
s->unit = unit;
s->direction = dir;
s->flags = flags;
s->context = ERR_PTR(-1);
mutex_init(&s->mutex);
s->packet_index = 0;
init_waitqueue_head(&s->ready_wait);
s->fmt = fmt;
s->process_ctx_payloads = process_ctx_payloads;
return 0;
}
EXPORT_SYMBOL(amdtp_stream_init);
/**
* amdtp_stream_destroy - free stream resources
* @s: the AMDTP stream to destroy
*/
void amdtp_stream_destroy(struct amdtp_stream *s)
{
/* Not initialized. */
if (s->protocol == NULL)
return;
WARN_ON(amdtp_stream_running(s));
kfree(s->protocol);
mutex_destroy(&s->mutex);
}
EXPORT_SYMBOL(amdtp_stream_destroy);
const unsigned int amdtp_syt_intervals[CIP_SFC_COUNT] = {
[CIP_SFC_32000] = 8,
[CIP_SFC_44100] = 8,
[CIP_SFC_48000] = 8,
[CIP_SFC_88200] = 16,
[CIP_SFC_96000] = 16,
[CIP_SFC_176400] = 32,
[CIP_SFC_192000] = 32,
};
EXPORT_SYMBOL(amdtp_syt_intervals);
const unsigned int amdtp_rate_table[CIP_SFC_COUNT] = {
[CIP_SFC_32000] = 32000,
[CIP_SFC_44100] = 44100,
[CIP_SFC_48000] = 48000,
[CIP_SFC_88200] = 88200,
[CIP_SFC_96000] = 96000,
[CIP_SFC_176400] = 176400,
[CIP_SFC_192000] = 192000,
};
EXPORT_SYMBOL(amdtp_rate_table);
static int apply_constraint_to_size(struct snd_pcm_hw_params *params,
struct snd_pcm_hw_rule *rule)
{
struct snd_interval *s = hw_param_interval(params, rule->var);
const struct snd_interval *r =
hw_param_interval_c(params, SNDRV_PCM_HW_PARAM_RATE);
struct snd_interval t = {0};
unsigned int step = 0;
int i;
for (i = 0; i < CIP_SFC_COUNT; ++i) {
if (snd_interval_test(r, amdtp_rate_table[i]))
step = max(step, amdtp_syt_intervals[i]);
}
t.min = roundup(s->min, step);
t.max = rounddown(s->max, step);
t.integer = 1;
return snd_interval_refine(s, &t);
}
/**
* amdtp_stream_add_pcm_hw_constraints - add hw constraints for PCM substream
* @s: the AMDTP stream, which must be initialized.
* @runtime: the PCM substream runtime
*/
int amdtp_stream_add_pcm_hw_constraints(struct amdtp_stream *s,
struct snd_pcm_runtime *runtime)
{
struct snd_pcm_hardware *hw = &runtime->hw;
unsigned int ctx_header_size;
unsigned int maximum_usec_per_period;
int err;
ALSA: firewire-lib: support NO_PERIOD_WAKEUP in ALSA PCM runtime Drivers of ALSA firewire stack can process packets for IT/IR context in process context when the process operates ALSA PCM character device by calling ioctl(2) with some requests. The ioctl requests are: * SNDRV_PCM_IOCTL_HWSYNC * SNDRV_PCM_IOCTL_SYNC_PTR * SNDRV_PCM_IOCTL_REWIND * SNDRV_PCM_IOCTL_FORWARD * SNDRV_PCM_IOCTL_WRITEI_FRAMES * SNDRV_PCM_IOCTL_READI_FRAMES * SNDRV_PCM_IOCTL_WRITEN_FRAMES * SNDRV_PCM_IOCTL_READN_FRAMES This means that general application can process PCM frames apart from hardware IRQ invocation, even if they are programmed by either IRQ-based scheduling model or Timer-based scheduling model. This commit add support for Timer-based scheduling model by allowing PCM runtime to suppress both process wakeup per period and scheduling hardware IRQ. SNDRV_PCM_INFO_BATCH is obsoleted since ALSA IEC 61883-1/6 packet streaming engine can report the number of transferred PCM frames within PCM period boundary. The granularity equals to SYT_INTERVAL in blocking transmission. In non-blocking transmission, it doesn't equal to SYT_INTERVAL but doesn't exceed. This patch is tested with PulseAudio, and --sched-model option of axfer with fix against the issue reported at: * https://lore.kernel.org/alsa-devel/687f9871-7484-1370-04d1-9c968e86f72b@linux.intel.com/#r Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527123253.174315-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:32:53 +00:00
hw->info = SNDRV_PCM_INFO_BLOCK_TRANSFER |
SNDRV_PCM_INFO_INTERLEAVED |
SNDRV_PCM_INFO_JOINT_DUPLEX |
SNDRV_PCM_INFO_MMAP |
ALSA: firewire-lib: support NO_PERIOD_WAKEUP in ALSA PCM runtime Drivers of ALSA firewire stack can process packets for IT/IR context in process context when the process operates ALSA PCM character device by calling ioctl(2) with some requests. The ioctl requests are: * SNDRV_PCM_IOCTL_HWSYNC * SNDRV_PCM_IOCTL_SYNC_PTR * SNDRV_PCM_IOCTL_REWIND * SNDRV_PCM_IOCTL_FORWARD * SNDRV_PCM_IOCTL_WRITEI_FRAMES * SNDRV_PCM_IOCTL_READI_FRAMES * SNDRV_PCM_IOCTL_WRITEN_FRAMES * SNDRV_PCM_IOCTL_READN_FRAMES This means that general application can process PCM frames apart from hardware IRQ invocation, even if they are programmed by either IRQ-based scheduling model or Timer-based scheduling model. This commit add support for Timer-based scheduling model by allowing PCM runtime to suppress both process wakeup per period and scheduling hardware IRQ. SNDRV_PCM_INFO_BATCH is obsoleted since ALSA IEC 61883-1/6 packet streaming engine can report the number of transferred PCM frames within PCM period boundary. The granularity equals to SYT_INTERVAL in blocking transmission. In non-blocking transmission, it doesn't equal to SYT_INTERVAL but doesn't exceed. This patch is tested with PulseAudio, and --sched-model option of axfer with fix against the issue reported at: * https://lore.kernel.org/alsa-devel/687f9871-7484-1370-04d1-9c968e86f72b@linux.intel.com/#r Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527123253.174315-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:32:53 +00:00
SNDRV_PCM_INFO_MMAP_VALID |
SNDRV_PCM_INFO_NO_PERIOD_WAKEUP;
hw->periods_min = 2;
hw->periods_max = UINT_MAX;
/* bytes for a frame */
hw->period_bytes_min = 4 * hw->channels_max;
/* Just to prevent from allocating much pages. */
hw->period_bytes_max = hw->period_bytes_min * 2048;
hw->buffer_bytes_max = hw->period_bytes_max * hw->periods_min;
// Linux driver for 1394 OHCI controller voluntarily flushes isoc
// context when total size of accumulated context header reaches
// PAGE_SIZE. This kicks work for the isoc context and brings
// callback in the middle of scheduled interrupts.
// Although AMDTP streams in the same domain use the same events per
// IRQ, use the largest size of context header between IT/IR contexts.
// Here, use the value of context header in IR context is for both
// contexts.
if (!(s->flags & CIP_NO_HEADER))
ctx_header_size = IR_CTX_HEADER_SIZE_CIP;
else
ctx_header_size = IR_CTX_HEADER_SIZE_NO_CIP;
maximum_usec_per_period = USEC_PER_SEC * PAGE_SIZE /
CYCLES_PER_SECOND / ctx_header_size;
// In IEC 61883-6, one isoc packet can transfer events up to the value
// of syt interval. This comes from the interval of isoc cycle. As 1394
// OHCI controller can generate hardware IRQ per isoc packet, the
// interval is 125 usec.
// However, there are two ways of transmission in IEC 61883-6; blocking
// and non-blocking modes. In blocking mode, the sequence of isoc packet
// includes 'empty' or 'NODATA' packets which include no event. In
// non-blocking mode, the number of events per packet is variable up to
// the syt interval.
// Due to the above protocol design, the minimum PCM frames per
// interrupt should be double of the value of syt interval, thus it is
// 250 usec.
err = snd_pcm_hw_constraint_minmax(runtime,
SNDRV_PCM_HW_PARAM_PERIOD_TIME,
250, maximum_usec_per_period);
if (err < 0)
goto end;
/* Non-Blocking stream has no more constraints */
if (!(s->flags & CIP_BLOCKING))
goto end;
/*
* One AMDTP packet can include some frames. In blocking mode, the
* number equals to SYT_INTERVAL. So the number is 8, 16 or 32,
* depending on its sampling rate. For accurate period interrupt, it's
* preferrable to align period/buffer sizes to current SYT_INTERVAL.
*/
err = snd_pcm_hw_rule_add(runtime, 0, SNDRV_PCM_HW_PARAM_PERIOD_SIZE,
apply_constraint_to_size, NULL,
SNDRV_PCM_HW_PARAM_PERIOD_SIZE,
SNDRV_PCM_HW_PARAM_RATE, -1);
if (err < 0)
goto end;
err = snd_pcm_hw_rule_add(runtime, 0, SNDRV_PCM_HW_PARAM_BUFFER_SIZE,
apply_constraint_to_size, NULL,
SNDRV_PCM_HW_PARAM_BUFFER_SIZE,
SNDRV_PCM_HW_PARAM_RATE, -1);
if (err < 0)
goto end;
end:
return err;
}
EXPORT_SYMBOL(amdtp_stream_add_pcm_hw_constraints);
/**
* amdtp_stream_set_parameters - set stream parameters
* @s: the AMDTP stream to configure
* @rate: the sample rate
* @data_block_quadlets: the size of a data block in quadlet unit
* @pcm_frame_multiplier: the multiplier to compute the number of PCM frames by the number of AMDTP
* events.
*
* The parameters must be set before the stream is started, and must not be
* changed while the stream is running.
*/
int amdtp_stream_set_parameters(struct amdtp_stream *s, unsigned int rate,
unsigned int data_block_quadlets, unsigned int pcm_frame_multiplier)
{
unsigned int sfc;
for (sfc = 0; sfc < ARRAY_SIZE(amdtp_rate_table); ++sfc) {
if (amdtp_rate_table[sfc] == rate)
break;
}
if (sfc == ARRAY_SIZE(amdtp_rate_table))
return -EINVAL;
s->sfc = sfc;
s->data_block_quadlets = data_block_quadlets;
s->syt_interval = amdtp_syt_intervals[sfc];
// default buffering in the device.
s->transfer_delay = TRANSFER_DELAY_TICKS - TICKS_PER_CYCLE;
// additional buffering needed to adjust for no-data packets.
if (s->flags & CIP_BLOCKING)
s->transfer_delay += TICKS_PER_SECOND * s->syt_interval / rate;
s->pcm_frame_multiplier = pcm_frame_multiplier;
return 0;
}
EXPORT_SYMBOL(amdtp_stream_set_parameters);
// The CIP header is processed in context header apart from context payload.
static int amdtp_stream_get_max_ctx_payload_size(struct amdtp_stream *s)
{
unsigned int multiplier;
if (s->flags & CIP_JUMBO_PAYLOAD)
multiplier = IR_JUMBO_PAYLOAD_MAX_SKIP_CYCLES;
else
multiplier = 1;
return s->syt_interval * s->data_block_quadlets * sizeof(__be32) * multiplier;
}
/**
* amdtp_stream_get_max_payload - get the stream's packet size
* @s: the AMDTP stream
*
* This function must not be called before the stream has been configured
* with amdtp_stream_set_parameters().
*/
unsigned int amdtp_stream_get_max_payload(struct amdtp_stream *s)
{
unsigned int cip_header_size;
2017-03-31 13:06:07 +00:00
if (!(s->flags & CIP_NO_HEADER))
cip_header_size = CIP_HEADER_SIZE;
else
cip_header_size = 0;
return cip_header_size + amdtp_stream_get_max_ctx_payload_size(s);
}
EXPORT_SYMBOL(amdtp_stream_get_max_payload);
/**
* amdtp_stream_pcm_prepare - prepare PCM device for running
* @s: the AMDTP stream
*
* This function should be called from the PCM device's .prepare callback.
*/
void amdtp_stream_pcm_prepare(struct amdtp_stream *s)
{
s->pcm_buffer_pointer = 0;
s->pcm_period_pointer = 0;
}
EXPORT_SYMBOL(amdtp_stream_pcm_prepare);
#define prev_packet_desc(s, desc) \
list_prev_entry_circular(desc, &s->packet_descs_list, link)
static void pool_blocking_data_blocks(struct amdtp_stream *s, struct seq_desc *descs,
unsigned int size, unsigned int pos, unsigned int count)
{
const unsigned int syt_interval = s->syt_interval;
int i;
for (i = 0; i < count; ++i) {
struct seq_desc *desc = descs + pos;
if (desc->syt_offset != CIP_SYT_NO_INFO)
desc->data_blocks = syt_interval;
else
desc->data_blocks = 0;
pos = (pos + 1) % size;
}
}
static void pool_ideal_nonblocking_data_blocks(struct amdtp_stream *s, struct seq_desc *descs,
unsigned int size, unsigned int pos,
unsigned int count)
{
const enum cip_sfc sfc = s->sfc;
unsigned int state = s->ctx_data.rx.data_block_state;
int i;
for (i = 0; i < count; ++i) {
struct seq_desc *desc = descs + pos;
if (!cip_sfc_is_base_44100(sfc)) {
// Sample_rate / 8000 is an integer, and precomputed.
desc->data_blocks = state;
} else {
unsigned int phase = state;
/*
* This calculates the number of data blocks per packet so that
* 1) the overall rate is correct and exactly synchronized to
* the bus clock, and
* 2) packets with a rounded-up number of blocks occur as early
* as possible in the sequence (to prevent underruns of the
* device's buffer).
*/
if (sfc == CIP_SFC_44100)
/* 6 6 5 6 5 6 5 ... */
desc->data_blocks = 5 + ((phase & 1) ^ (phase == 0 || phase >= 40));
else
/* 12 11 11 11 11 ... or 23 22 22 22 22 ... */
desc->data_blocks = 11 * (sfc >> 1) + (phase == 0);
if (++phase >= (80 >> (sfc >> 1)))
phase = 0;
state = phase;
}
pos = (pos + 1) % size;
}
s->ctx_data.rx.data_block_state = state;
}
static unsigned int calculate_syt_offset(unsigned int *last_syt_offset,
unsigned int *syt_offset_state, enum cip_sfc sfc)
{
unsigned int syt_offset;
if (*last_syt_offset < TICKS_PER_CYCLE) {
if (!cip_sfc_is_base_44100(sfc))
syt_offset = *last_syt_offset + *syt_offset_state;
else {
/*
* The time, in ticks, of the n'th SYT_INTERVAL sample is:
* n * SYT_INTERVAL * 24576000 / sample_rate
* Modulo TICKS_PER_CYCLE, the difference between successive
* elements is about 1386.23. Rounding the results of this
* formula to the SYT precision results in a sequence of
* differences that begins with:
* 1386 1386 1387 1386 1386 1386 1387 1386 1386 1386 1387 ...
* This code generates _exactly_ the same sequence.
*/
unsigned int phase = *syt_offset_state;
unsigned int index = phase % 13;
syt_offset = *last_syt_offset;
syt_offset += 1386 + ((index && !(index & 3)) ||
phase == 146);
if (++phase >= 147)
phase = 0;
*syt_offset_state = phase;
}
} else
syt_offset = *last_syt_offset - TICKS_PER_CYCLE;
*last_syt_offset = syt_offset;
if (syt_offset >= TICKS_PER_CYCLE)
syt_offset = CIP_SYT_NO_INFO;
return syt_offset;
}
static void pool_ideal_syt_offsets(struct amdtp_stream *s, struct seq_desc *descs,
unsigned int size, unsigned int pos, unsigned int count)
{
const enum cip_sfc sfc = s->sfc;
unsigned int last = s->ctx_data.rx.last_syt_offset;
unsigned int state = s->ctx_data.rx.syt_offset_state;
int i;
for (i = 0; i < count; ++i) {
struct seq_desc *desc = descs + pos;
desc->syt_offset = calculate_syt_offset(&last, &state, sfc);
pos = (pos + 1) % size;
}
s->ctx_data.rx.last_syt_offset = last;
s->ctx_data.rx.syt_offset_state = state;
}
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
static unsigned int compute_syt_offset(unsigned int syt, unsigned int cycle,
unsigned int transfer_delay)
{
unsigned int cycle_lo = (cycle % CYCLES_PER_SECOND) & 0x0f;
unsigned int syt_cycle_lo = (syt & 0xf000) >> 12;
unsigned int syt_offset;
// Round up.
if (syt_cycle_lo < cycle_lo)
syt_cycle_lo += CIP_SYT_CYCLE_MODULUS;
syt_cycle_lo -= cycle_lo;
// Subtract transfer delay so that the synchronization offset is not so large
// at transmission.
syt_offset = syt_cycle_lo * TICKS_PER_CYCLE + (syt & 0x0fff);
if (syt_offset < transfer_delay)
syt_offset += CIP_SYT_CYCLE_MODULUS * TICKS_PER_CYCLE;
return syt_offset - transfer_delay;
}
ALSA: firewire-lib: replay sequence of incoming packets for outgoing packets ALSA IEC 61883-1/6 packet streaming engine uses pre-computed parameters ideal for nominal sampling transfer frequency (STF) to transfer packets to device since it was added 2011. As a result of user experience for a decade, it is clear that the sequence is not suitable to some actual devices. It takes the devices to generate noise, and causes any type of discontinuity in the series of packet transferred from the device. It's required for the engine to transfer packets according to effective STF. The effective STF is given by media clock recovered by the sequence of packet transferred from the target device. In the previous commit, the sequence is already cached. The media clock recovery can be achieved by analyzing the sequence. In technological world, many ideas are proposed for media clock recovery. However, the small part of them could be actually adopted in our case since floating point arithmetic is not mostly available in Linux kernel land. This commit adopts the simple way from them; sequence replay, which means that the sequence of parameters from incoming packet is used as is to transfer outgoing packets. The media clock is not computed internally, but the sequence of outgoing packet superficially looks to be generated by the media clock. The association between source and destination is decided when starting AMDTP domain. When the target device supports a pair of isochronous packet streams, the tx stream is source and the rx stream is destination. When it supports two pair of streams, each of tx stream is associated to corresponding rx stream in its order. When it supports less number of tx streams than rx streams, the fist tx stream is selected for all of rx streams. When it supports more tx streams than rx streams, the first tx packet is associated to the rx stream. As I noted in previous commit, the sequence of parameters from incoming packet is different between devices, time to time. It is worse idea to replay the sequence of parameters from a device for the sequence of packet to the other devices even if they are in the same category of device. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-3-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:10 +00:00
// Both of the producer and consumer of the queue runs in the same clock of IEEE 1394 bus.
// Additionally, the sequence of tx packets is severely checked against any discontinuity
// before filling entries in the queue. The calculation is safe even if it looks fragile by
// overrun.
static unsigned int calculate_cached_cycle_count(struct amdtp_stream *s, unsigned int head)
{
const unsigned int cache_size = s->ctx_data.tx.cache.size;
unsigned int cycles = s->ctx_data.tx.cache.pos;
ALSA: firewire-lib: replay sequence of incoming packets for outgoing packets ALSA IEC 61883-1/6 packet streaming engine uses pre-computed parameters ideal for nominal sampling transfer frequency (STF) to transfer packets to device since it was added 2011. As a result of user experience for a decade, it is clear that the sequence is not suitable to some actual devices. It takes the devices to generate noise, and causes any type of discontinuity in the series of packet transferred from the device. It's required for the engine to transfer packets according to effective STF. The effective STF is given by media clock recovered by the sequence of packet transferred from the target device. In the previous commit, the sequence is already cached. The media clock recovery can be achieved by analyzing the sequence. In technological world, many ideas are proposed for media clock recovery. However, the small part of them could be actually adopted in our case since floating point arithmetic is not mostly available in Linux kernel land. This commit adopts the simple way from them; sequence replay, which means that the sequence of parameters from incoming packet is used as is to transfer outgoing packets. The media clock is not computed internally, but the sequence of outgoing packet superficially looks to be generated by the media clock. The association between source and destination is decided when starting AMDTP domain. When the target device supports a pair of isochronous packet streams, the tx stream is source and the rx stream is destination. When it supports two pair of streams, each of tx stream is associated to corresponding rx stream in its order. When it supports less number of tx streams than rx streams, the fist tx stream is selected for all of rx streams. When it supports more tx streams than rx streams, the first tx packet is associated to the rx stream. As I noted in previous commit, the sequence of parameters from incoming packet is different between devices, time to time. It is worse idea to replay the sequence of parameters from a device for the sequence of packet to the other devices even if they are in the same category of device. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-3-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:10 +00:00
if (cycles < head)
cycles += cache_size;
cycles -= head;
return cycles;
}
static void cache_seq(struct amdtp_stream *s, const struct pkt_desc *src, unsigned int desc_count)
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
{
const unsigned int transfer_delay = s->transfer_delay;
const unsigned int cache_size = s->ctx_data.tx.cache.size;
struct seq_desc *cache = s->ctx_data.tx.cache.descs;
unsigned int cache_pos = s->ctx_data.tx.cache.pos;
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
bool aware_syt = !(s->flags & CIP_UNAWARE_SYT);
int i;
for (i = 0; i < desc_count; ++i) {
struct seq_desc *dst = cache + cache_pos;
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
if (aware_syt && src->syt != CIP_SYT_NO_INFO)
dst->syt_offset = compute_syt_offset(src->syt, src->cycle, transfer_delay);
else
dst->syt_offset = CIP_SYT_NO_INFO;
dst->data_blocks = src->data_blocks;
cache_pos = (cache_pos + 1) % cache_size;
src = amdtp_stream_next_packet_desc(s, src);
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
}
s->ctx_data.tx.cache.pos = cache_pos;
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
}
static void pool_ideal_seq_descs(struct amdtp_stream *s, struct seq_desc *descs, unsigned int size,
unsigned int pos, unsigned int count)
{
pool_ideal_syt_offsets(s, descs, size, pos, count);
if (s->flags & CIP_BLOCKING)
pool_blocking_data_blocks(s, descs, size, pos, count);
else
pool_ideal_nonblocking_data_blocks(s, descs, size, pos, count);
}
static void pool_replayed_seq(struct amdtp_stream *s, struct seq_desc *descs, unsigned int size,
unsigned int pos, unsigned int count)
ALSA: firewire-lib: replay sequence of incoming packets for outgoing packets ALSA IEC 61883-1/6 packet streaming engine uses pre-computed parameters ideal for nominal sampling transfer frequency (STF) to transfer packets to device since it was added 2011. As a result of user experience for a decade, it is clear that the sequence is not suitable to some actual devices. It takes the devices to generate noise, and causes any type of discontinuity in the series of packet transferred from the device. It's required for the engine to transfer packets according to effective STF. The effective STF is given by media clock recovered by the sequence of packet transferred from the target device. In the previous commit, the sequence is already cached. The media clock recovery can be achieved by analyzing the sequence. In technological world, many ideas are proposed for media clock recovery. However, the small part of them could be actually adopted in our case since floating point arithmetic is not mostly available in Linux kernel land. This commit adopts the simple way from them; sequence replay, which means that the sequence of parameters from incoming packet is used as is to transfer outgoing packets. The media clock is not computed internally, but the sequence of outgoing packet superficially looks to be generated by the media clock. The association between source and destination is decided when starting AMDTP domain. When the target device supports a pair of isochronous packet streams, the tx stream is source and the rx stream is destination. When it supports two pair of streams, each of tx stream is associated to corresponding rx stream in its order. When it supports less number of tx streams than rx streams, the fist tx stream is selected for all of rx streams. When it supports more tx streams than rx streams, the first tx packet is associated to the rx stream. As I noted in previous commit, the sequence of parameters from incoming packet is different between devices, time to time. It is worse idea to replay the sequence of parameters from a device for the sequence of packet to the other devices even if they are in the same category of device. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-3-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:10 +00:00
{
struct amdtp_stream *target = s->ctx_data.rx.replay_target;
const struct seq_desc *cache = target->ctx_data.tx.cache.descs;
const unsigned int cache_size = target->ctx_data.tx.cache.size;
unsigned int cache_pos = s->ctx_data.rx.cache_pos;
ALSA: firewire-lib: replay sequence of incoming packets for outgoing packets ALSA IEC 61883-1/6 packet streaming engine uses pre-computed parameters ideal for nominal sampling transfer frequency (STF) to transfer packets to device since it was added 2011. As a result of user experience for a decade, it is clear that the sequence is not suitable to some actual devices. It takes the devices to generate noise, and causes any type of discontinuity in the series of packet transferred from the device. It's required for the engine to transfer packets according to effective STF. The effective STF is given by media clock recovered by the sequence of packet transferred from the target device. In the previous commit, the sequence is already cached. The media clock recovery can be achieved by analyzing the sequence. In technological world, many ideas are proposed for media clock recovery. However, the small part of them could be actually adopted in our case since floating point arithmetic is not mostly available in Linux kernel land. This commit adopts the simple way from them; sequence replay, which means that the sequence of parameters from incoming packet is used as is to transfer outgoing packets. The media clock is not computed internally, but the sequence of outgoing packet superficially looks to be generated by the media clock. The association between source and destination is decided when starting AMDTP domain. When the target device supports a pair of isochronous packet streams, the tx stream is source and the rx stream is destination. When it supports two pair of streams, each of tx stream is associated to corresponding rx stream in its order. When it supports less number of tx streams than rx streams, the fist tx stream is selected for all of rx streams. When it supports more tx streams than rx streams, the first tx packet is associated to the rx stream. As I noted in previous commit, the sequence of parameters from incoming packet is different between devices, time to time. It is worse idea to replay the sequence of parameters from a device for the sequence of packet to the other devices even if they are in the same category of device. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-3-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:10 +00:00
int i;
for (i = 0; i < count; ++i) {
descs[pos] = cache[cache_pos];
cache_pos = (cache_pos + 1) % cache_size;
pos = (pos + 1) % size;
ALSA: firewire-lib: replay sequence of incoming packets for outgoing packets ALSA IEC 61883-1/6 packet streaming engine uses pre-computed parameters ideal for nominal sampling transfer frequency (STF) to transfer packets to device since it was added 2011. As a result of user experience for a decade, it is clear that the sequence is not suitable to some actual devices. It takes the devices to generate noise, and causes any type of discontinuity in the series of packet transferred from the device. It's required for the engine to transfer packets according to effective STF. The effective STF is given by media clock recovered by the sequence of packet transferred from the target device. In the previous commit, the sequence is already cached. The media clock recovery can be achieved by analyzing the sequence. In technological world, many ideas are proposed for media clock recovery. However, the small part of them could be actually adopted in our case since floating point arithmetic is not mostly available in Linux kernel land. This commit adopts the simple way from them; sequence replay, which means that the sequence of parameters from incoming packet is used as is to transfer outgoing packets. The media clock is not computed internally, but the sequence of outgoing packet superficially looks to be generated by the media clock. The association between source and destination is decided when starting AMDTP domain. When the target device supports a pair of isochronous packet streams, the tx stream is source and the rx stream is destination. When it supports two pair of streams, each of tx stream is associated to corresponding rx stream in its order. When it supports less number of tx streams than rx streams, the fist tx stream is selected for all of rx streams. When it supports more tx streams than rx streams, the first tx packet is associated to the rx stream. As I noted in previous commit, the sequence of parameters from incoming packet is different between devices, time to time. It is worse idea to replay the sequence of parameters from a device for the sequence of packet to the other devices even if they are in the same category of device. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-3-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:10 +00:00
}
s->ctx_data.rx.cache_pos = cache_pos;
ALSA: firewire-lib: replay sequence of incoming packets for outgoing packets ALSA IEC 61883-1/6 packet streaming engine uses pre-computed parameters ideal for nominal sampling transfer frequency (STF) to transfer packets to device since it was added 2011. As a result of user experience for a decade, it is clear that the sequence is not suitable to some actual devices. It takes the devices to generate noise, and causes any type of discontinuity in the series of packet transferred from the device. It's required for the engine to transfer packets according to effective STF. The effective STF is given by media clock recovered by the sequence of packet transferred from the target device. In the previous commit, the sequence is already cached. The media clock recovery can be achieved by analyzing the sequence. In technological world, many ideas are proposed for media clock recovery. However, the small part of them could be actually adopted in our case since floating point arithmetic is not mostly available in Linux kernel land. This commit adopts the simple way from them; sequence replay, which means that the sequence of parameters from incoming packet is used as is to transfer outgoing packets. The media clock is not computed internally, but the sequence of outgoing packet superficially looks to be generated by the media clock. The association between source and destination is decided when starting AMDTP domain. When the target device supports a pair of isochronous packet streams, the tx stream is source and the rx stream is destination. When it supports two pair of streams, each of tx stream is associated to corresponding rx stream in its order. When it supports less number of tx streams than rx streams, the fist tx stream is selected for all of rx streams. When it supports more tx streams than rx streams, the first tx packet is associated to the rx stream. As I noted in previous commit, the sequence of parameters from incoming packet is different between devices, time to time. It is worse idea to replay the sequence of parameters from a device for the sequence of packet to the other devices even if they are in the same category of device. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-3-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:10 +00:00
}
static void pool_seq_descs(struct amdtp_stream *s, struct seq_desc *descs, unsigned int size,
unsigned int pos, unsigned int count)
ALSA: firewire-lib: replay sequence of incoming packets for outgoing packets ALSA IEC 61883-1/6 packet streaming engine uses pre-computed parameters ideal for nominal sampling transfer frequency (STF) to transfer packets to device since it was added 2011. As a result of user experience for a decade, it is clear that the sequence is not suitable to some actual devices. It takes the devices to generate noise, and causes any type of discontinuity in the series of packet transferred from the device. It's required for the engine to transfer packets according to effective STF. The effective STF is given by media clock recovered by the sequence of packet transferred from the target device. In the previous commit, the sequence is already cached. The media clock recovery can be achieved by analyzing the sequence. In technological world, many ideas are proposed for media clock recovery. However, the small part of them could be actually adopted in our case since floating point arithmetic is not mostly available in Linux kernel land. This commit adopts the simple way from them; sequence replay, which means that the sequence of parameters from incoming packet is used as is to transfer outgoing packets. The media clock is not computed internally, but the sequence of outgoing packet superficially looks to be generated by the media clock. The association between source and destination is decided when starting AMDTP domain. When the target device supports a pair of isochronous packet streams, the tx stream is source and the rx stream is destination. When it supports two pair of streams, each of tx stream is associated to corresponding rx stream in its order. When it supports less number of tx streams than rx streams, the fist tx stream is selected for all of rx streams. When it supports more tx streams than rx streams, the first tx packet is associated to the rx stream. As I noted in previous commit, the sequence of parameters from incoming packet is different between devices, time to time. It is worse idea to replay the sequence of parameters from a device for the sequence of packet to the other devices even if they are in the same category of device. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-3-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:10 +00:00
{
struct amdtp_domain *d = s->domain;
void (*pool_seq_descs)(struct amdtp_stream *s, struct seq_desc *descs, unsigned int size,
unsigned int pos, unsigned int count);
ALSA: firewire-lib: replay sequence of incoming packets for outgoing packets ALSA IEC 61883-1/6 packet streaming engine uses pre-computed parameters ideal for nominal sampling transfer frequency (STF) to transfer packets to device since it was added 2011. As a result of user experience for a decade, it is clear that the sequence is not suitable to some actual devices. It takes the devices to generate noise, and causes any type of discontinuity in the series of packet transferred from the device. It's required for the engine to transfer packets according to effective STF. The effective STF is given by media clock recovered by the sequence of packet transferred from the target device. In the previous commit, the sequence is already cached. The media clock recovery can be achieved by analyzing the sequence. In technological world, many ideas are proposed for media clock recovery. However, the small part of them could be actually adopted in our case since floating point arithmetic is not mostly available in Linux kernel land. This commit adopts the simple way from them; sequence replay, which means that the sequence of parameters from incoming packet is used as is to transfer outgoing packets. The media clock is not computed internally, but the sequence of outgoing packet superficially looks to be generated by the media clock. The association between source and destination is decided when starting AMDTP domain. When the target device supports a pair of isochronous packet streams, the tx stream is source and the rx stream is destination. When it supports two pair of streams, each of tx stream is associated to corresponding rx stream in its order. When it supports less number of tx streams than rx streams, the fist tx stream is selected for all of rx streams. When it supports more tx streams than rx streams, the first tx packet is associated to the rx stream. As I noted in previous commit, the sequence of parameters from incoming packet is different between devices, time to time. It is worse idea to replay the sequence of parameters from a device for the sequence of packet to the other devices even if they are in the same category of device. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-3-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:10 +00:00
if (!d->replay.enable || !s->ctx_data.rx.replay_target) {
pool_seq_descs = pool_ideal_seq_descs;
} else {
if (!d->replay.on_the_fly) {
pool_seq_descs = pool_replayed_seq;
} else {
struct amdtp_stream *tx = s->ctx_data.rx.replay_target;
const unsigned int cache_size = tx->ctx_data.tx.cache.size;
const unsigned int cache_pos = s->ctx_data.rx.cache_pos;
unsigned int cached_cycles = calculate_cached_cycle_count(tx, cache_pos);
if (cached_cycles > count && cached_cycles > cache_size / 2)
pool_seq_descs = pool_replayed_seq;
else
pool_seq_descs = pool_ideal_seq_descs;
}
}
pool_seq_descs(s, descs, size, pos, count);
ALSA: firewire-lib: replay sequence of incoming packets for outgoing packets ALSA IEC 61883-1/6 packet streaming engine uses pre-computed parameters ideal for nominal sampling transfer frequency (STF) to transfer packets to device since it was added 2011. As a result of user experience for a decade, it is clear that the sequence is not suitable to some actual devices. It takes the devices to generate noise, and causes any type of discontinuity in the series of packet transferred from the device. It's required for the engine to transfer packets according to effective STF. The effective STF is given by media clock recovered by the sequence of packet transferred from the target device. In the previous commit, the sequence is already cached. The media clock recovery can be achieved by analyzing the sequence. In technological world, many ideas are proposed for media clock recovery. However, the small part of them could be actually adopted in our case since floating point arithmetic is not mostly available in Linux kernel land. This commit adopts the simple way from them; sequence replay, which means that the sequence of parameters from incoming packet is used as is to transfer outgoing packets. The media clock is not computed internally, but the sequence of outgoing packet superficially looks to be generated by the media clock. The association between source and destination is decided when starting AMDTP domain. When the target device supports a pair of isochronous packet streams, the tx stream is source and the rx stream is destination. When it supports two pair of streams, each of tx stream is associated to corresponding rx stream in its order. When it supports less number of tx streams than rx streams, the fist tx stream is selected for all of rx streams. When it supports more tx streams than rx streams, the first tx packet is associated to the rx stream. As I noted in previous commit, the sequence of parameters from incoming packet is different between devices, time to time. It is worse idea to replay the sequence of parameters from a device for the sequence of packet to the other devices even if they are in the same category of device. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-3-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:10 +00:00
}
static void update_pcm_pointers(struct amdtp_stream *s,
struct snd_pcm_substream *pcm,
unsigned int frames)
{
unsigned int ptr;
ptr = s->pcm_buffer_pointer + frames;
if (ptr >= pcm->runtime->buffer_size)
ptr -= pcm->runtime->buffer_size;
locking/atomics: COCCINELLE/treewide: Convert trivial ACCESS_ONCE() patterns to READ_ONCE()/WRITE_ONCE() Please do not apply this to mainline directly, instead please re-run the coccinelle script shown below and apply its output. For several reasons, it is desirable to use {READ,WRITE}_ONCE() in preference to ACCESS_ONCE(), and new code is expected to use one of the former. So far, there's been no reason to change most existing uses of ACCESS_ONCE(), as these aren't harmful, and changing them results in churn. However, for some features, the read/write distinction is critical to correct operation. To distinguish these cases, separate read/write accessors must be used. This patch migrates (most) remaining ACCESS_ONCE() instances to {READ,WRITE}_ONCE(), using the following coccinelle script: ---- // Convert trivial ACCESS_ONCE() uses to equivalent READ_ONCE() and // WRITE_ONCE() // $ make coccicheck COCCI=/home/mark/once.cocci SPFLAGS="--include-headers" MODE=patch virtual patch @ depends on patch @ expression E1, E2; @@ - ACCESS_ONCE(E1) = E2 + WRITE_ONCE(E1, E2) @ depends on patch @ expression E; @@ - ACCESS_ONCE(E) + READ_ONCE(E) ---- Signed-off-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: davem@davemloft.net Cc: linux-arch@vger.kernel.org Cc: mpe@ellerman.id.au Cc: shuah@kernel.org Cc: snitzer@redhat.com Cc: thor.thayer@linux.intel.com Cc: tj@kernel.org Cc: viro@zeniv.linux.org.uk Cc: will.deacon@arm.com Link: http://lkml.kernel.org/r/1508792849-3115-19-git-send-email-paulmck@linux.vnet.ibm.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-10-23 21:07:29 +00:00
WRITE_ONCE(s->pcm_buffer_pointer, ptr);
s->pcm_period_pointer += frames;
if (s->pcm_period_pointer >= pcm->runtime->period_size) {
s->pcm_period_pointer -= pcm->runtime->period_size;
ALSA: firewire-lib: support NO_PERIOD_WAKEUP in ALSA PCM runtime Drivers of ALSA firewire stack can process packets for IT/IR context in process context when the process operates ALSA PCM character device by calling ioctl(2) with some requests. The ioctl requests are: * SNDRV_PCM_IOCTL_HWSYNC * SNDRV_PCM_IOCTL_SYNC_PTR * SNDRV_PCM_IOCTL_REWIND * SNDRV_PCM_IOCTL_FORWARD * SNDRV_PCM_IOCTL_WRITEI_FRAMES * SNDRV_PCM_IOCTL_READI_FRAMES * SNDRV_PCM_IOCTL_WRITEN_FRAMES * SNDRV_PCM_IOCTL_READN_FRAMES This means that general application can process PCM frames apart from hardware IRQ invocation, even if they are programmed by either IRQ-based scheduling model or Timer-based scheduling model. This commit add support for Timer-based scheduling model by allowing PCM runtime to suppress both process wakeup per period and scheduling hardware IRQ. SNDRV_PCM_INFO_BATCH is obsoleted since ALSA IEC 61883-1/6 packet streaming engine can report the number of transferred PCM frames within PCM period boundary. The granularity equals to SYT_INTERVAL in blocking transmission. In non-blocking transmission, it doesn't equal to SYT_INTERVAL but doesn't exceed. This patch is tested with PulseAudio, and --sched-model option of axfer with fix against the issue reported at: * https://lore.kernel.org/alsa-devel/687f9871-7484-1370-04d1-9c968e86f72b@linux.intel.com/#r Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527123253.174315-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:32:53 +00:00
// The program in user process should periodically check the status of intermediate
// buffer associated to PCM substream to process PCM frames in the buffer, instead
// of receiving notification of period elapsed by poll wait.
if (!pcm->runtime->no_period_wakeup) {
if (in_softirq()) {
// In software IRQ context for 1394 OHCI.
snd_pcm_period_elapsed(pcm);
} else {
// In process context of ALSA PCM application under acquired lock of
// PCM substream.
snd_pcm_period_elapsed_under_stream_lock(pcm);
}
}
}
}
static int queue_packet(struct amdtp_stream *s, struct fw_iso_packet *params,
bool sched_irq)
{
int err;
params->interrupt = sched_irq;
params->tag = s->tag;
params->sy = 0;
err = fw_iso_context_queue(s->context, params, &s->buffer.iso_buffer,
s->buffer.packets[s->packet_index].offset);
if (err < 0) {
dev_err(&s->unit->device, "queueing error: %d\n", err);
goto end;
}
if (++s->packet_index >= s->queue_size)
s->packet_index = 0;
end:
return err;
}
static inline int queue_out_packet(struct amdtp_stream *s,
struct fw_iso_packet *params, bool sched_irq)
{
params->skip =
!!(params->header_length == 0 && params->payload_length == 0);
return queue_packet(s, params, sched_irq);
}
static inline int queue_in_packet(struct amdtp_stream *s,
struct fw_iso_packet *params)
{
// Queue one packet for IR context.
params->header_length = s->ctx_data.tx.ctx_header_size;
params->payload_length = s->ctx_data.tx.max_ctx_payload_length;
params->skip = false;
return queue_packet(s, params, false);
}
static void generate_cip_header(struct amdtp_stream *s, __be32 cip_header[2],
unsigned int data_block_counter, unsigned int syt)
{
cip_header[0] = cpu_to_be32(READ_ONCE(s->source_node_id_field) |
(s->data_block_quadlets << CIP_DBS_SHIFT) |
((s->sph << CIP_SPH_SHIFT) & CIP_SPH_MASK) |
data_block_counter);
cip_header[1] = cpu_to_be32(CIP_EOH |
((s->fmt << CIP_FMT_SHIFT) & CIP_FMT_MASK) |
((s->ctx_data.rx.fdf << CIP_FDF_SHIFT) & CIP_FDF_MASK) |
(syt & CIP_SYT_MASK));
}
static void build_it_pkt_header(struct amdtp_stream *s, unsigned int cycle,
struct fw_iso_packet *params, unsigned int header_length,
unsigned int data_blocks,
unsigned int data_block_counter,
unsigned int syt, unsigned int index, u32 curr_cycle_time)
{
unsigned int payload_length;
__be32 *cip_header;
ALSA: firewire-lib: add helper functions as interfaces between packet streaming layer and data block processing layer ALSA PCM framework uses PCM buffer with a concept of 'period' to synchronize userspace operations to hardware for nearly-realtime processing. Each driver implements snd_pcm_period_elapsed() to tell across of the period boundary to ALSA PCM middleware. To call the function, some drivers utilize hardware interrupt handlers, the others count handled PCM frames. Drivers for sound units on IEEE 1394 bus are the latter. They use two buffers; PCM buffer and DMA buffer for IEEE 1394 isochronous packet. PCM frames are copied between these two buffers and 'amdtp_stream' structure counts the handled PCM frames. Then, snd_pcm_period_elapsed() is called if required. Essentially, packet streaming layer should not be responsible for PCM frame processing. The PCM frame processing should be handled in each data block processing layer as a result of handling data blocks. Although, PCM frame counting is a common work for all of protocols which ALSA firewire stack is going to support. This commit adds two new helper functions as interfaces between packet streaming layer to data block processing layer. In future, each data block processing layer implements these functions. The packet streaming layer calls data block processing layer per packet by calling the functions. The data block processing layer processes data blocks and PCM frames, and returns the number of processed PCM frames. Then the packet streaming layer calculates handled PCM frames and calls snd_pcm_period_elapsed(). Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Iwai <tiwai@suse.de>
2015-09-19 02:21:52 +00:00
payload_length = data_blocks * sizeof(__be32) * s->data_block_quadlets;
params->payload_length = payload_length;
if (header_length > 0) {
cip_header = (__be32 *)params->header;
generate_cip_header(s, cip_header, data_block_counter, syt);
params->header_length = header_length;
} else {
cip_header = NULL;
}
trace_amdtp_packet(s, cycle, cip_header, payload_length + header_length, data_blocks,
data_block_counter, s->packet_index, index, curr_cycle_time);
2017-03-31 13:06:07 +00:00
}
static int check_cip_header(struct amdtp_stream *s, const __be32 *buf,
unsigned int payload_length,
unsigned int *data_blocks,
unsigned int *data_block_counter, unsigned int *syt)
{
u32 cip_header[2];
unsigned int sph;
unsigned int fmt;
unsigned int fdf;
unsigned int dbc;
bool lost;
cip_header[0] = be32_to_cpu(buf[0]);
cip_header[1] = be32_to_cpu(buf[1]);
/*
* This module supports 'Two-quadlet CIP header with SYT field'.
* For convenience, also check FMT field is AM824 or not.
*/
if ((((cip_header[0] & CIP_EOH_MASK) == CIP_EOH) ||
((cip_header[1] & CIP_EOH_MASK) != CIP_EOH)) &&
(!(s->flags & CIP_HEADER_WITHOUT_EOH))) {
dev_info_ratelimited(&s->unit->device,
"Invalid CIP header for AMDTP: %08X:%08X\n",
cip_header[0], cip_header[1]);
return -EAGAIN;
}
/* Check valid protocol or not. */
sph = (cip_header[0] & CIP_SPH_MASK) >> CIP_SPH_SHIFT;
fmt = (cip_header[1] & CIP_FMT_MASK) >> CIP_FMT_SHIFT;
if (sph != s->sph || fmt != s->fmt) {
dev_info_ratelimited(&s->unit->device,
"Detect unexpected protocol: %08x %08x\n",
cip_header[0], cip_header[1]);
return -EAGAIN;
}
/* Calculate data blocks */
fdf = (cip_header[1] & CIP_FDF_MASK) >> CIP_FDF_SHIFT;
if (payload_length == 0 || (fmt == CIP_FMT_AM && fdf == AMDTP_FDF_NO_DATA)) {
*data_blocks = 0;
} else {
unsigned int data_block_quadlets =
(cip_header[0] & CIP_DBS_MASK) >> CIP_DBS_SHIFT;
/* avoid division by zero */
if (data_block_quadlets == 0) {
dev_err(&s->unit->device,
"Detect invalid value in dbs field: %08X\n",
cip_header[0]);
return -EPROTO;
}
if (s->flags & CIP_WRONG_DBS)
data_block_quadlets = s->data_block_quadlets;
*data_blocks = payload_length / sizeof(__be32) / data_block_quadlets;
}
/* Check data block counter continuity */
dbc = cip_header[0] & CIP_DBC_MASK;
if (*data_blocks == 0 && (s->flags & CIP_EMPTY_HAS_WRONG_DBC) &&
*data_block_counter != UINT_MAX)
dbc = *data_block_counter;
if ((dbc == 0x00 && (s->flags & CIP_SKIP_DBC_ZERO_CHECK)) ||
*data_block_counter == UINT_MAX) {
lost = false;
} else if (!(s->flags & CIP_DBC_IS_END_EVENT)) {
lost = dbc != *data_block_counter;
} else {
unsigned int dbc_interval;
if (!(s->flags & CIP_DBC_IS_PAYLOAD_QUADLETS)) {
if (*data_blocks > 0 && s->ctx_data.tx.dbc_interval > 0)
dbc_interval = s->ctx_data.tx.dbc_interval;
else
dbc_interval = *data_blocks;
} else {
dbc_interval = payload_length / sizeof(__be32);
}
lost = dbc != ((*data_block_counter + dbc_interval) & 0xff);
}
if (lost) {
dev_err(&s->unit->device,
"Detect discontinuity of CIP: %02X %02X\n",
*data_block_counter, dbc);
return -EIO;
}
*data_block_counter = dbc;
if (!(s->flags & CIP_UNAWARE_SYT))
*syt = cip_header[1] & CIP_SYT_MASK;
return 0;
}
static int parse_ir_ctx_header(struct amdtp_stream *s, unsigned int cycle,
const __be32 *ctx_header,
unsigned int *data_blocks,
unsigned int *data_block_counter,
unsigned int *syt, unsigned int packet_index, unsigned int index,
u32 curr_cycle_time)
{
unsigned int payload_length;
const __be32 *cip_header;
unsigned int cip_header_size;
payload_length = be32_to_cpu(ctx_header[0]) >> ISO_DATA_LENGTH_SHIFT;
if (!(s->flags & CIP_NO_HEADER))
cip_header_size = CIP_HEADER_SIZE;
else
cip_header_size = 0;
if (payload_length > cip_header_size + s->ctx_data.tx.max_ctx_payload_length) {
dev_err(&s->unit->device,
"Detect jumbo payload: %04x %04x\n",
payload_length, cip_header_size + s->ctx_data.tx.max_ctx_payload_length);
return -EIO;
}
if (cip_header_size > 0) {
if (payload_length >= cip_header_size) {
int err;
cip_header = ctx_header + IR_CTX_HEADER_DEFAULT_QUADLETS;
err = check_cip_header(s, cip_header, payload_length - cip_header_size,
data_blocks, data_block_counter, syt);
if (err < 0)
return err;
} else {
// Handle the cycle so that empty packet arrives.
cip_header = NULL;
*data_blocks = 0;
*syt = 0;
}
} else {
cip_header = NULL;
*data_blocks = payload_length / sizeof(__be32) / s->data_block_quadlets;
*syt = 0;
if (*data_block_counter == UINT_MAX)
*data_block_counter = 0;
}
trace_amdtp_packet(s, cycle, cip_header, payload_length, *data_blocks,
*data_block_counter, packet_index, index, curr_cycle_time);
return 0;
}
// In CYCLE_TIMER register of IEEE 1394, 7 bits are used to represent second. On
// the other hand, in DMA descriptors of 1394 OHCI, 3 bits are used to represent
// it. Thus, via Linux firewire subsystem, we can get the 3 bits for second.
static inline u32 compute_ohci_iso_ctx_cycle_count(u32 tstamp)
{
return (((tstamp >> 13) & 0x07) * CYCLES_PER_SECOND) + (tstamp & 0x1fff);
}
static inline u32 compute_ohci_cycle_count(__be32 ctx_header_tstamp)
{
u32 tstamp = be32_to_cpu(ctx_header_tstamp) & HEADER_TSTAMP_MASK;
return compute_ohci_iso_ctx_cycle_count(tstamp);
}
static inline u32 increment_ohci_cycle_count(u32 cycle, unsigned int addend)
{
cycle += addend;
if (cycle >= OHCI_SECOND_MODULUS * CYCLES_PER_SECOND)
cycle -= OHCI_SECOND_MODULUS * CYCLES_PER_SECOND;
return cycle;
}
static inline u32 decrement_ohci_cycle_count(u32 minuend, u32 subtrahend)
{
if (minuend < subtrahend)
minuend += OHCI_SECOND_MODULUS * CYCLES_PER_SECOND;
return minuend - subtrahend;
}
static int compare_ohci_cycle_count(u32 lval, u32 rval)
{
if (lval == rval)
return 0;
else if (lval < rval && rval - lval < OHCI_SECOND_MODULUS * CYCLES_PER_SECOND / 2)
return -1;
else
return 1;
}
// Align to actual cycle count for the packet which is going to be scheduled.
// This module queued the same number of isochronous cycle as the size of queue
// to kip isochronous cycle, therefore it's OK to just increment the cycle by
// the size of queue for scheduled cycle.
static inline u32 compute_ohci_it_cycle(const __be32 ctx_header_tstamp,
unsigned int queue_size)
{
u32 cycle = compute_ohci_cycle_count(ctx_header_tstamp);
return increment_ohci_cycle_count(cycle, queue_size);
}
static int generate_tx_packet_descs(struct amdtp_stream *s, struct pkt_desc *desc,
const __be32 *ctx_header, unsigned int packet_count,
unsigned int *desc_count)
{
unsigned int next_cycle = s->next_cycle;
unsigned int dbc = s->data_block_counter;
unsigned int packet_index = s->packet_index;
unsigned int queue_size = s->queue_size;
u32 curr_cycle_time = 0;
int i;
int err;
if (trace_amdtp_packet_enabled())
(void)fw_card_read_cycle_time(fw_parent_device(s->unit)->card, &curr_cycle_time);
*desc_count = 0;
for (i = 0; i < packet_count; ++i) {
unsigned int cycle;
bool lost;
unsigned int data_blocks;
unsigned int syt;
cycle = compute_ohci_cycle_count(ctx_header[1]);
lost = (next_cycle != cycle);
if (lost) {
if (s->flags & CIP_NO_HEADER) {
// Fireface skips transmission just for an isoc cycle corresponding
// to empty packet.
unsigned int prev_cycle = next_cycle;
next_cycle = increment_ohci_cycle_count(next_cycle, 1);
lost = (next_cycle != cycle);
if (!lost) {
// Prepare a description for the skipped cycle for
// sequence replay.
desc->cycle = prev_cycle;
desc->syt = 0;
desc->data_blocks = 0;
desc->data_block_counter = dbc;
desc->ctx_payload = NULL;
desc = amdtp_stream_next_packet_desc(s, desc);
++(*desc_count);
}
} else if (s->flags & CIP_JUMBO_PAYLOAD) {
// OXFW970 skips transmission for several isoc cycles during
// asynchronous transaction. The sequence replay is impossible due
// to the reason.
unsigned int safe_cycle = increment_ohci_cycle_count(next_cycle,
IR_JUMBO_PAYLOAD_MAX_SKIP_CYCLES);
lost = (compare_ohci_cycle_count(safe_cycle, cycle) < 0);
}
if (lost) {
dev_err(&s->unit->device, "Detect discontinuity of cycle: %d %d\n",
next_cycle, cycle);
return -EIO;
}
}
err = parse_ir_ctx_header(s, cycle, ctx_header, &data_blocks, &dbc, &syt,
packet_index, i, curr_cycle_time);
if (err < 0)
return err;
desc->cycle = cycle;
desc->syt = syt;
desc->data_blocks = data_blocks;
desc->data_block_counter = dbc;
desc->ctx_payload = s->buffer.packets[packet_index].buffer;
if (!(s->flags & CIP_DBC_IS_END_EVENT))
dbc = (dbc + desc->data_blocks) & 0xff;
next_cycle = increment_ohci_cycle_count(next_cycle, 1);
desc = amdtp_stream_next_packet_desc(s, desc);
++(*desc_count);
ctx_header += s->ctx_data.tx.ctx_header_size / sizeof(*ctx_header);
packet_index = (packet_index + 1) % queue_size;
}
s->next_cycle = next_cycle;
s->data_block_counter = dbc;
return 0;
}
static unsigned int compute_syt(unsigned int syt_offset, unsigned int cycle,
unsigned int transfer_delay)
{
unsigned int syt;
syt_offset += transfer_delay;
syt = ((cycle + syt_offset / TICKS_PER_CYCLE) << 12) |
(syt_offset % TICKS_PER_CYCLE);
return syt & CIP_SYT_MASK;
}
static void generate_rx_packet_descs(struct amdtp_stream *s, struct pkt_desc *desc,
const __be32 *ctx_header, unsigned int packet_count)
{
struct seq_desc *seq_descs = s->ctx_data.rx.seq.descs;
unsigned int seq_size = s->ctx_data.rx.seq.size;
unsigned int seq_pos = s->ctx_data.rx.seq.pos;
unsigned int dbc = s->data_block_counter;
bool aware_syt = !(s->flags & CIP_UNAWARE_SYT);
int i;
pool_seq_descs(s, seq_descs, seq_size, seq_pos, packet_count);
for (i = 0; i < packet_count; ++i) {
unsigned int index = (s->packet_index + i) % s->queue_size;
const struct seq_desc *seq = seq_descs + seq_pos;
desc->cycle = compute_ohci_it_cycle(*ctx_header, s->queue_size);
if (aware_syt && seq->syt_offset != CIP_SYT_NO_INFO)
desc->syt = compute_syt(seq->syt_offset, desc->cycle, s->transfer_delay);
else
desc->syt = CIP_SYT_NO_INFO;
desc->data_blocks = seq->data_blocks;
if (s->flags & CIP_DBC_IS_END_EVENT)
dbc = (dbc + desc->data_blocks) & 0xff;
desc->data_block_counter = dbc;
if (!(s->flags & CIP_DBC_IS_END_EVENT))
dbc = (dbc + desc->data_blocks) & 0xff;
desc->ctx_payload = s->buffer.packets[index].buffer;
seq_pos = (seq_pos + 1) % seq_size;
desc = amdtp_stream_next_packet_desc(s, desc);
++ctx_header;
}
s->data_block_counter = dbc;
s->ctx_data.rx.seq.pos = seq_pos;
}
static inline void cancel_stream(struct amdtp_stream *s)
{
s->packet_index = -1;
if (in_softirq())
amdtp_stream_pcm_abort(s);
WRITE_ONCE(s->pcm_buffer_pointer, SNDRV_PCM_POS_XRUN);
}
static snd_pcm_sframes_t compute_pcm_extra_delay(struct amdtp_stream *s,
const struct pkt_desc *desc, unsigned int count)
{
unsigned int data_block_count = 0;
u32 latest_cycle;
u32 cycle_time;
u32 curr_cycle;
u32 cycle_gap;
int i, err;
if (count == 0)
goto end;
// Forward to the latest record.
for (i = 0; i < count - 1; ++i)
desc = amdtp_stream_next_packet_desc(s, desc);
latest_cycle = desc->cycle;
err = fw_card_read_cycle_time(fw_parent_device(s->unit)->card, &cycle_time);
if (err < 0)
goto end;
// Compute cycle count with lower 3 bits of second field and cycle field like timestamp
// format of 1394 OHCI isochronous context.
curr_cycle = compute_ohci_iso_ctx_cycle_count((cycle_time >> 12) & 0x0000ffff);
if (s->direction == AMDTP_IN_STREAM) {
// NOTE: The AMDTP packet descriptor should be for the past isochronous cycle since
// it corresponds to arrived isochronous packet.
if (compare_ohci_cycle_count(latest_cycle, curr_cycle) > 0)
goto end;
cycle_gap = decrement_ohci_cycle_count(curr_cycle, latest_cycle);
// NOTE: estimate delay by recent history of arrived AMDTP packets. The estimated
// value expectedly corresponds to a few packets (0-2) since the packet arrived at
// the most recent isochronous cycle has been already processed.
for (i = 0; i < cycle_gap; ++i) {
desc = amdtp_stream_next_packet_desc(s, desc);
data_block_count += desc->data_blocks;
}
} else {
// NOTE: The AMDTP packet descriptor should be for the future isochronous cycle
// since it was already scheduled.
if (compare_ohci_cycle_count(latest_cycle, curr_cycle) < 0)
goto end;
cycle_gap = decrement_ohci_cycle_count(latest_cycle, curr_cycle);
// NOTE: use history of scheduled packets.
for (i = 0; i < cycle_gap; ++i) {
data_block_count += desc->data_blocks;
desc = prev_packet_desc(s, desc);
}
}
end:
return data_block_count * s->pcm_frame_multiplier;
}
static void process_ctx_payloads(struct amdtp_stream *s,
const struct pkt_desc *desc,
unsigned int count)
{
struct snd_pcm_substream *pcm;
int i;
pcm = READ_ONCE(s->pcm);
s->process_ctx_payloads(s, desc, count, pcm);
if (pcm) {
unsigned int data_block_count = 0;
pcm->runtime->delay = compute_pcm_extra_delay(s, desc, count);
for (i = 0; i < count; ++i) {
data_block_count += desc->data_blocks;
desc = amdtp_stream_next_packet_desc(s, desc);
}
update_pcm_pointers(s, pcm, data_block_count * s->pcm_frame_multiplier);
}
}
static void process_rx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
void *header, void *private_data)
{
struct amdtp_stream *s = private_data;
const struct amdtp_domain *d = s->domain;
const __be32 *ctx_header = header;
const unsigned int events_per_period = d->events_per_period;
unsigned int event_count = s->ctx_data.rx.event_count;
struct pkt_desc *desc = s->packet_descs_cursor;
unsigned int pkt_header_length;
unsigned int packets;
u32 curr_cycle_time;
ALSA: firewire-lib: support NO_PERIOD_WAKEUP in ALSA PCM runtime Drivers of ALSA firewire stack can process packets for IT/IR context in process context when the process operates ALSA PCM character device by calling ioctl(2) with some requests. The ioctl requests are: * SNDRV_PCM_IOCTL_HWSYNC * SNDRV_PCM_IOCTL_SYNC_PTR * SNDRV_PCM_IOCTL_REWIND * SNDRV_PCM_IOCTL_FORWARD * SNDRV_PCM_IOCTL_WRITEI_FRAMES * SNDRV_PCM_IOCTL_READI_FRAMES * SNDRV_PCM_IOCTL_WRITEN_FRAMES * SNDRV_PCM_IOCTL_READN_FRAMES This means that general application can process PCM frames apart from hardware IRQ invocation, even if they are programmed by either IRQ-based scheduling model or Timer-based scheduling model. This commit add support for Timer-based scheduling model by allowing PCM runtime to suppress both process wakeup per period and scheduling hardware IRQ. SNDRV_PCM_INFO_BATCH is obsoleted since ALSA IEC 61883-1/6 packet streaming engine can report the number of transferred PCM frames within PCM period boundary. The granularity equals to SYT_INTERVAL in blocking transmission. In non-blocking transmission, it doesn't equal to SYT_INTERVAL but doesn't exceed. This patch is tested with PulseAudio, and --sched-model option of axfer with fix against the issue reported at: * https://lore.kernel.org/alsa-devel/687f9871-7484-1370-04d1-9c968e86f72b@linux.intel.com/#r Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527123253.174315-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:32:53 +00:00
bool need_hw_irq;
int i;
if (s->packet_index < 0)
return;
// Calculate the number of packets in buffer and check XRUN.
packets = header_length / sizeof(*ctx_header);
generate_rx_packet_descs(s, desc, ctx_header, packets);
process_ctx_payloads(s, desc, packets);
if (!(s->flags & CIP_NO_HEADER))
pkt_header_length = IT_PKT_HEADER_SIZE_CIP;
else
pkt_header_length = 0;
ALSA: firewire-lib: support NO_PERIOD_WAKEUP in ALSA PCM runtime Drivers of ALSA firewire stack can process packets for IT/IR context in process context when the process operates ALSA PCM character device by calling ioctl(2) with some requests. The ioctl requests are: * SNDRV_PCM_IOCTL_HWSYNC * SNDRV_PCM_IOCTL_SYNC_PTR * SNDRV_PCM_IOCTL_REWIND * SNDRV_PCM_IOCTL_FORWARD * SNDRV_PCM_IOCTL_WRITEI_FRAMES * SNDRV_PCM_IOCTL_READI_FRAMES * SNDRV_PCM_IOCTL_WRITEN_FRAMES * SNDRV_PCM_IOCTL_READN_FRAMES This means that general application can process PCM frames apart from hardware IRQ invocation, even if they are programmed by either IRQ-based scheduling model or Timer-based scheduling model. This commit add support for Timer-based scheduling model by allowing PCM runtime to suppress both process wakeup per period and scheduling hardware IRQ. SNDRV_PCM_INFO_BATCH is obsoleted since ALSA IEC 61883-1/6 packet streaming engine can report the number of transferred PCM frames within PCM period boundary. The granularity equals to SYT_INTERVAL in blocking transmission. In non-blocking transmission, it doesn't equal to SYT_INTERVAL but doesn't exceed. This patch is tested with PulseAudio, and --sched-model option of axfer with fix against the issue reported at: * https://lore.kernel.org/alsa-devel/687f9871-7484-1370-04d1-9c968e86f72b@linux.intel.com/#r Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527123253.174315-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:32:53 +00:00
if (s == d->irq_target) {
// At NO_PERIOD_WAKEUP mode, the packets for all IT/IR contexts are processed by
// the tasks of user process operating ALSA PCM character device by calling ioctl(2)
// with some requests, instead of scheduled hardware IRQ of an IT context.
struct snd_pcm_substream *pcm = READ_ONCE(s->pcm);
need_hw_irq = !pcm || !pcm->runtime->no_period_wakeup;
} else {
need_hw_irq = false;
}
if (trace_amdtp_packet_enabled())
(void)fw_card_read_cycle_time(fw_parent_device(s->unit)->card, &curr_cycle_time);
for (i = 0; i < packets; ++i) {
DEFINE_FLEX(struct fw_iso_packet, template, header,
header_length, CIP_HEADER_QUADLETS);
bool sched_irq = false;
build_it_pkt_header(s, desc->cycle, template, pkt_header_length,
desc->data_blocks, desc->data_block_counter,
desc->syt, i, curr_cycle_time);
if (s == s->domain->irq_target) {
event_count += desc->data_blocks;
if (event_count >= events_per_period) {
event_count -= events_per_period;
ALSA: firewire-lib: support NO_PERIOD_WAKEUP in ALSA PCM runtime Drivers of ALSA firewire stack can process packets for IT/IR context in process context when the process operates ALSA PCM character device by calling ioctl(2) with some requests. The ioctl requests are: * SNDRV_PCM_IOCTL_HWSYNC * SNDRV_PCM_IOCTL_SYNC_PTR * SNDRV_PCM_IOCTL_REWIND * SNDRV_PCM_IOCTL_FORWARD * SNDRV_PCM_IOCTL_WRITEI_FRAMES * SNDRV_PCM_IOCTL_READI_FRAMES * SNDRV_PCM_IOCTL_WRITEN_FRAMES * SNDRV_PCM_IOCTL_READN_FRAMES This means that general application can process PCM frames apart from hardware IRQ invocation, even if they are programmed by either IRQ-based scheduling model or Timer-based scheduling model. This commit add support for Timer-based scheduling model by allowing PCM runtime to suppress both process wakeup per period and scheduling hardware IRQ. SNDRV_PCM_INFO_BATCH is obsoleted since ALSA IEC 61883-1/6 packet streaming engine can report the number of transferred PCM frames within PCM period boundary. The granularity equals to SYT_INTERVAL in blocking transmission. In non-blocking transmission, it doesn't equal to SYT_INTERVAL but doesn't exceed. This patch is tested with PulseAudio, and --sched-model option of axfer with fix against the issue reported at: * https://lore.kernel.org/alsa-devel/687f9871-7484-1370-04d1-9c968e86f72b@linux.intel.com/#r Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527123253.174315-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:32:53 +00:00
sched_irq = need_hw_irq;
}
}
if (queue_out_packet(s, template, sched_irq) < 0) {
cancel_stream(s);
return;
}
desc = amdtp_stream_next_packet_desc(s, desc);
}
s->ctx_data.rx.event_count = event_count;
s->packet_descs_cursor = desc;
}
static void skip_rx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
void *header, void *private_data)
{
struct amdtp_stream *s = private_data;
struct amdtp_domain *d = s->domain;
const __be32 *ctx_header = header;
unsigned int packets;
unsigned int cycle;
int i;
if (s->packet_index < 0)
return;
packets = header_length / sizeof(*ctx_header);
cycle = compute_ohci_it_cycle(ctx_header[packets - 1], s->queue_size);
s->next_cycle = increment_ohci_cycle_count(cycle, 1);
for (i = 0; i < packets; ++i) {
struct fw_iso_packet params = {
.header_length = 0,
.payload_length = 0,
};
bool sched_irq = (s == d->irq_target && i == packets - 1);
if (queue_out_packet(s, &params, sched_irq) < 0) {
cancel_stream(s);
return;
}
}
}
static void irq_target_callback(struct fw_iso_context *context, u32 tstamp, size_t header_length,
void *header, void *private_data);
static void process_rx_packets_intermediately(struct fw_iso_context *context, u32 tstamp,
size_t header_length, void *header, void *private_data)
{
struct amdtp_stream *s = private_data;
struct amdtp_domain *d = s->domain;
__be32 *ctx_header = header;
const unsigned int queue_size = s->queue_size;
unsigned int packets;
unsigned int offset;
if (s->packet_index < 0)
return;
packets = header_length / sizeof(*ctx_header);
offset = 0;
while (offset < packets) {
unsigned int cycle = compute_ohci_it_cycle(ctx_header[offset], queue_size);
if (compare_ohci_cycle_count(cycle, d->processing_cycle.rx_start) >= 0)
break;
++offset;
}
if (offset > 0) {
unsigned int length = sizeof(*ctx_header) * offset;
skip_rx_packets(context, tstamp, length, ctx_header, private_data);
if (amdtp_streaming_error(s))
return;
ctx_header += offset;
header_length -= length;
}
if (offset < packets) {
s->ready_processing = true;
wake_up(&s->ready_wait);
if (d->replay.enable)
s->ctx_data.rx.cache_pos = 0;
process_rx_packets(context, tstamp, header_length, ctx_header, private_data);
if (amdtp_streaming_error(s))
return;
if (s == d->irq_target)
s->context->callback.sc = irq_target_callback;
else
s->context->callback.sc = process_rx_packets;
}
}
static void process_tx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
void *header, void *private_data)
{
struct amdtp_stream *s = private_data;
__be32 *ctx_header = header;
struct pkt_desc *desc = s->packet_descs_cursor;
unsigned int packet_count;
unsigned int desc_count;
int i;
int err;
if (s->packet_index < 0)
return;
// Calculate the number of packets in buffer and check XRUN.
packet_count = header_length / s->ctx_data.tx.ctx_header_size;
desc_count = 0;
err = generate_tx_packet_descs(s, desc, ctx_header, packet_count, &desc_count);
if (err < 0) {
if (err != -EAGAIN) {
cancel_stream(s);
return;
}
} else {
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
struct amdtp_domain *d = s->domain;
process_ctx_payloads(s, desc, desc_count);
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
if (d->replay.enable)
cache_seq(s, desc, desc_count);
for (i = 0; i < desc_count; ++i)
desc = amdtp_stream_next_packet_desc(s, desc);
s->packet_descs_cursor = desc;
}
for (i = 0; i < packet_count; ++i) {
struct fw_iso_packet params = {0};
if (queue_in_packet(s, &params) < 0) {
cancel_stream(s);
return;
}
}
}
static void drop_tx_packets(struct fw_iso_context *context, u32 tstamp, size_t header_length,
void *header, void *private_data)
{
struct amdtp_stream *s = private_data;
const __be32 *ctx_header = header;
unsigned int packets;
unsigned int cycle;
int i;
if (s->packet_index < 0)
return;
packets = header_length / s->ctx_data.tx.ctx_header_size;
ctx_header += (packets - 1) * s->ctx_data.tx.ctx_header_size / sizeof(*ctx_header);
cycle = compute_ohci_cycle_count(ctx_header[1]);
s->next_cycle = increment_ohci_cycle_count(cycle, 1);
for (i = 0; i < packets; ++i) {
struct fw_iso_packet params = {0};
if (queue_in_packet(s, &params) < 0) {
cancel_stream(s);
return;
}
}
}
static void process_tx_packets_intermediately(struct fw_iso_context *context, u32 tstamp,
size_t header_length, void *header, void *private_data)
{
struct amdtp_stream *s = private_data;
struct amdtp_domain *d = s->domain;
__be32 *ctx_header;
unsigned int packets;
unsigned int offset;
if (s->packet_index < 0)
return;
packets = header_length / s->ctx_data.tx.ctx_header_size;
offset = 0;
ctx_header = header;
while (offset < packets) {
unsigned int cycle = compute_ohci_cycle_count(ctx_header[1]);
if (compare_ohci_cycle_count(cycle, d->processing_cycle.tx_start) >= 0)
break;
ctx_header += s->ctx_data.tx.ctx_header_size / sizeof(__be32);
++offset;
}
ctx_header = header;
if (offset > 0) {
size_t length = s->ctx_data.tx.ctx_header_size * offset;
drop_tx_packets(context, tstamp, length, ctx_header, s);
if (amdtp_streaming_error(s))
return;
ctx_header += length / sizeof(*ctx_header);
header_length -= length;
}
if (offset < packets) {
s->ready_processing = true;
wake_up(&s->ready_wait);
process_tx_packets(context, tstamp, header_length, ctx_header, s);
if (amdtp_streaming_error(s))
return;
context->callback.sc = process_tx_packets;
}
}
static void drop_tx_packets_initially(struct fw_iso_context *context, u32 tstamp,
size_t header_length, void *header, void *private_data)
{
struct amdtp_stream *s = private_data;
struct amdtp_domain *d = s->domain;
__be32 *ctx_header;
unsigned int count;
unsigned int events;
int i;
if (s->packet_index < 0)
return;
count = header_length / s->ctx_data.tx.ctx_header_size;
// Attempt to detect any event in the batch of packets.
events = 0;
ctx_header = header;
for (i = 0; i < count; ++i) {
unsigned int payload_quads =
(be32_to_cpu(*ctx_header) >> ISO_DATA_LENGTH_SHIFT) / sizeof(__be32);
unsigned int data_blocks;
if (s->flags & CIP_NO_HEADER) {
data_blocks = payload_quads / s->data_block_quadlets;
} else {
__be32 *cip_headers = ctx_header + IR_CTX_HEADER_DEFAULT_QUADLETS;
if (payload_quads < CIP_HEADER_QUADLETS) {
data_blocks = 0;
} else {
payload_quads -= CIP_HEADER_QUADLETS;
if (s->flags & CIP_UNAWARE_SYT) {
data_blocks = payload_quads / s->data_block_quadlets;
} else {
u32 cip1 = be32_to_cpu(cip_headers[1]);
// NODATA packet can includes any data blocks but they are
// not available as event.
if ((cip1 & CIP_NO_DATA) == CIP_NO_DATA)
data_blocks = 0;
else
data_blocks = payload_quads / s->data_block_quadlets;
}
}
}
events += data_blocks;
ctx_header += s->ctx_data.tx.ctx_header_size / sizeof(__be32);
}
drop_tx_packets(context, tstamp, header_length, header, s);
if (events > 0)
s->ctx_data.tx.event_starts = true;
// Decide the cycle count to begin processing content of packet in IR contexts.
{
unsigned int stream_count = 0;
unsigned int event_starts_count = 0;
unsigned int cycle = UINT_MAX;
list_for_each_entry(s, &d->streams, list) {
if (s->direction == AMDTP_IN_STREAM) {
++stream_count;
if (s->ctx_data.tx.event_starts)
++event_starts_count;
}
}
if (stream_count == event_starts_count) {
unsigned int next_cycle;
list_for_each_entry(s, &d->streams, list) {
if (s->direction != AMDTP_IN_STREAM)
continue;
next_cycle = increment_ohci_cycle_count(s->next_cycle,
d->processing_cycle.tx_init_skip);
if (cycle == UINT_MAX ||
compare_ohci_cycle_count(next_cycle, cycle) > 0)
cycle = next_cycle;
s->context->callback.sc = process_tx_packets_intermediately;
}
d->processing_cycle.tx_start = cycle;
}
}
}
static void process_ctxs_in_domain(struct amdtp_domain *d)
{
struct amdtp_stream *s;
list_for_each_entry(s, &d->streams, list) {
if (s != d->irq_target && amdtp_stream_running(s))
fw_iso_context_flush_completions(s->context);
if (amdtp_streaming_error(s))
goto error;
}
return;
error:
if (amdtp_stream_running(d->irq_target))
cancel_stream(d->irq_target);
list_for_each_entry(s, &d->streams, list) {
if (amdtp_stream_running(s))
cancel_stream(s);
}
}
static void irq_target_callback(struct fw_iso_context *context, u32 tstamp, size_t header_length,
void *header, void *private_data)
{
struct amdtp_stream *s = private_data;
struct amdtp_domain *d = s->domain;
process_rx_packets(context, tstamp, header_length, header, private_data);
process_ctxs_in_domain(d);
}
static void irq_target_callback_intermediately(struct fw_iso_context *context, u32 tstamp,
size_t header_length, void *header, void *private_data)
{
struct amdtp_stream *s = private_data;
struct amdtp_domain *d = s->domain;
process_rx_packets_intermediately(context, tstamp, header_length, header, private_data);
process_ctxs_in_domain(d);
}
static void irq_target_callback_skip(struct fw_iso_context *context, u32 tstamp,
size_t header_length, void *header, void *private_data)
{
struct amdtp_stream *s = private_data;
struct amdtp_domain *d = s->domain;
ALSA: firewire-lib: replay sequence of incoming packets for outgoing packets ALSA IEC 61883-1/6 packet streaming engine uses pre-computed parameters ideal for nominal sampling transfer frequency (STF) to transfer packets to device since it was added 2011. As a result of user experience for a decade, it is clear that the sequence is not suitable to some actual devices. It takes the devices to generate noise, and causes any type of discontinuity in the series of packet transferred from the device. It's required for the engine to transfer packets according to effective STF. The effective STF is given by media clock recovered by the sequence of packet transferred from the target device. In the previous commit, the sequence is already cached. The media clock recovery can be achieved by analyzing the sequence. In technological world, many ideas are proposed for media clock recovery. However, the small part of them could be actually adopted in our case since floating point arithmetic is not mostly available in Linux kernel land. This commit adopts the simple way from them; sequence replay, which means that the sequence of parameters from incoming packet is used as is to transfer outgoing packets. The media clock is not computed internally, but the sequence of outgoing packet superficially looks to be generated by the media clock. The association between source and destination is decided when starting AMDTP domain. When the target device supports a pair of isochronous packet streams, the tx stream is source and the rx stream is destination. When it supports two pair of streams, each of tx stream is associated to corresponding rx stream in its order. When it supports less number of tx streams than rx streams, the fist tx stream is selected for all of rx streams. When it supports more tx streams than rx streams, the first tx packet is associated to the rx stream. As I noted in previous commit, the sequence of parameters from incoming packet is different between devices, time to time. It is worse idea to replay the sequence of parameters from a device for the sequence of packet to the other devices even if they are in the same category of device. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-3-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:10 +00:00
bool ready_to_start;
skip_rx_packets(context, tstamp, header_length, header, private_data);
process_ctxs_in_domain(d);
if (d->replay.enable && !d->replay.on_the_fly) {
ALSA: firewire-lib: replay sequence of incoming packets for outgoing packets ALSA IEC 61883-1/6 packet streaming engine uses pre-computed parameters ideal for nominal sampling transfer frequency (STF) to transfer packets to device since it was added 2011. As a result of user experience for a decade, it is clear that the sequence is not suitable to some actual devices. It takes the devices to generate noise, and causes any type of discontinuity in the series of packet transferred from the device. It's required for the engine to transfer packets according to effective STF. The effective STF is given by media clock recovered by the sequence of packet transferred from the target device. In the previous commit, the sequence is already cached. The media clock recovery can be achieved by analyzing the sequence. In technological world, many ideas are proposed for media clock recovery. However, the small part of them could be actually adopted in our case since floating point arithmetic is not mostly available in Linux kernel land. This commit adopts the simple way from them; sequence replay, which means that the sequence of parameters from incoming packet is used as is to transfer outgoing packets. The media clock is not computed internally, but the sequence of outgoing packet superficially looks to be generated by the media clock. The association between source and destination is decided when starting AMDTP domain. When the target device supports a pair of isochronous packet streams, the tx stream is source and the rx stream is destination. When it supports two pair of streams, each of tx stream is associated to corresponding rx stream in its order. When it supports less number of tx streams than rx streams, the fist tx stream is selected for all of rx streams. When it supports more tx streams than rx streams, the first tx packet is associated to the rx stream. As I noted in previous commit, the sequence of parameters from incoming packet is different between devices, time to time. It is worse idea to replay the sequence of parameters from a device for the sequence of packet to the other devices even if they are in the same category of device. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-3-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:10 +00:00
unsigned int rx_count = 0;
unsigned int rx_ready_count = 0;
struct amdtp_stream *rx;
list_for_each_entry(rx, &d->streams, list) {
struct amdtp_stream *tx;
unsigned int cached_cycles;
if (rx->direction != AMDTP_OUT_STREAM)
continue;
++rx_count;
tx = rx->ctx_data.rx.replay_target;
cached_cycles = calculate_cached_cycle_count(tx, 0);
if (cached_cycles > tx->ctx_data.tx.cache.size / 2)
++rx_ready_count;
}
ready_to_start = (rx_count == rx_ready_count);
} else {
ready_to_start = true;
}
// Decide the cycle count to begin processing content of packet in IT contexts. All of IT
// contexts are expected to start and get callback when reaching here.
ALSA: firewire-lib: replay sequence of incoming packets for outgoing packets ALSA IEC 61883-1/6 packet streaming engine uses pre-computed parameters ideal for nominal sampling transfer frequency (STF) to transfer packets to device since it was added 2011. As a result of user experience for a decade, it is clear that the sequence is not suitable to some actual devices. It takes the devices to generate noise, and causes any type of discontinuity in the series of packet transferred from the device. It's required for the engine to transfer packets according to effective STF. The effective STF is given by media clock recovered by the sequence of packet transferred from the target device. In the previous commit, the sequence is already cached. The media clock recovery can be achieved by analyzing the sequence. In technological world, many ideas are proposed for media clock recovery. However, the small part of them could be actually adopted in our case since floating point arithmetic is not mostly available in Linux kernel land. This commit adopts the simple way from them; sequence replay, which means that the sequence of parameters from incoming packet is used as is to transfer outgoing packets. The media clock is not computed internally, but the sequence of outgoing packet superficially looks to be generated by the media clock. The association between source and destination is decided when starting AMDTP domain. When the target device supports a pair of isochronous packet streams, the tx stream is source and the rx stream is destination. When it supports two pair of streams, each of tx stream is associated to corresponding rx stream in its order. When it supports less number of tx streams than rx streams, the fist tx stream is selected for all of rx streams. When it supports more tx streams than rx streams, the first tx packet is associated to the rx stream. As I noted in previous commit, the sequence of parameters from incoming packet is different between devices, time to time. It is worse idea to replay the sequence of parameters from a device for the sequence of packet to the other devices even if they are in the same category of device. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-3-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:10 +00:00
if (ready_to_start) {
unsigned int cycle = s->next_cycle;
list_for_each_entry(s, &d->streams, list) {
if (s->direction != AMDTP_OUT_STREAM)
continue;
ALSA: firewire-lib: replay sequence of incoming packets for outgoing packets ALSA IEC 61883-1/6 packet streaming engine uses pre-computed parameters ideal for nominal sampling transfer frequency (STF) to transfer packets to device since it was added 2011. As a result of user experience for a decade, it is clear that the sequence is not suitable to some actual devices. It takes the devices to generate noise, and causes any type of discontinuity in the series of packet transferred from the device. It's required for the engine to transfer packets according to effective STF. The effective STF is given by media clock recovered by the sequence of packet transferred from the target device. In the previous commit, the sequence is already cached. The media clock recovery can be achieved by analyzing the sequence. In technological world, many ideas are proposed for media clock recovery. However, the small part of them could be actually adopted in our case since floating point arithmetic is not mostly available in Linux kernel land. This commit adopts the simple way from them; sequence replay, which means that the sequence of parameters from incoming packet is used as is to transfer outgoing packets. The media clock is not computed internally, but the sequence of outgoing packet superficially looks to be generated by the media clock. The association between source and destination is decided when starting AMDTP domain. When the target device supports a pair of isochronous packet streams, the tx stream is source and the rx stream is destination. When it supports two pair of streams, each of tx stream is associated to corresponding rx stream in its order. When it supports less number of tx streams than rx streams, the fist tx stream is selected for all of rx streams. When it supports more tx streams than rx streams, the first tx packet is associated to the rx stream. As I noted in previous commit, the sequence of parameters from incoming packet is different between devices, time to time. It is worse idea to replay the sequence of parameters from a device for the sequence of packet to the other devices even if they are in the same category of device. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-3-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:10 +00:00
if (compare_ohci_cycle_count(s->next_cycle, cycle) > 0)
cycle = s->next_cycle;
ALSA: firewire-lib: replay sequence of incoming packets for outgoing packets ALSA IEC 61883-1/6 packet streaming engine uses pre-computed parameters ideal for nominal sampling transfer frequency (STF) to transfer packets to device since it was added 2011. As a result of user experience for a decade, it is clear that the sequence is not suitable to some actual devices. It takes the devices to generate noise, and causes any type of discontinuity in the series of packet transferred from the device. It's required for the engine to transfer packets according to effective STF. The effective STF is given by media clock recovered by the sequence of packet transferred from the target device. In the previous commit, the sequence is already cached. The media clock recovery can be achieved by analyzing the sequence. In technological world, many ideas are proposed for media clock recovery. However, the small part of them could be actually adopted in our case since floating point arithmetic is not mostly available in Linux kernel land. This commit adopts the simple way from them; sequence replay, which means that the sequence of parameters from incoming packet is used as is to transfer outgoing packets. The media clock is not computed internally, but the sequence of outgoing packet superficially looks to be generated by the media clock. The association between source and destination is decided when starting AMDTP domain. When the target device supports a pair of isochronous packet streams, the tx stream is source and the rx stream is destination. When it supports two pair of streams, each of tx stream is associated to corresponding rx stream in its order. When it supports less number of tx streams than rx streams, the fist tx stream is selected for all of rx streams. When it supports more tx streams than rx streams, the first tx packet is associated to the rx stream. As I noted in previous commit, the sequence of parameters from incoming packet is different between devices, time to time. It is worse idea to replay the sequence of parameters from a device for the sequence of packet to the other devices even if they are in the same category of device. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-3-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:10 +00:00
if (s == d->irq_target)
s->context->callback.sc = irq_target_callback_intermediately;
else
s->context->callback.sc = process_rx_packets_intermediately;
}
ALSA: firewire-lib: replay sequence of incoming packets for outgoing packets ALSA IEC 61883-1/6 packet streaming engine uses pre-computed parameters ideal for nominal sampling transfer frequency (STF) to transfer packets to device since it was added 2011. As a result of user experience for a decade, it is clear that the sequence is not suitable to some actual devices. It takes the devices to generate noise, and causes any type of discontinuity in the series of packet transferred from the device. It's required for the engine to transfer packets according to effective STF. The effective STF is given by media clock recovered by the sequence of packet transferred from the target device. In the previous commit, the sequence is already cached. The media clock recovery can be achieved by analyzing the sequence. In technological world, many ideas are proposed for media clock recovery. However, the small part of them could be actually adopted in our case since floating point arithmetic is not mostly available in Linux kernel land. This commit adopts the simple way from them; sequence replay, which means that the sequence of parameters from incoming packet is used as is to transfer outgoing packets. The media clock is not computed internally, but the sequence of outgoing packet superficially looks to be generated by the media clock. The association between source and destination is decided when starting AMDTP domain. When the target device supports a pair of isochronous packet streams, the tx stream is source and the rx stream is destination. When it supports two pair of streams, each of tx stream is associated to corresponding rx stream in its order. When it supports less number of tx streams than rx streams, the fist tx stream is selected for all of rx streams. When it supports more tx streams than rx streams, the first tx packet is associated to the rx stream. As I noted in previous commit, the sequence of parameters from incoming packet is different between devices, time to time. It is worse idea to replay the sequence of parameters from a device for the sequence of packet to the other devices even if they are in the same category of device. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-3-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:10 +00:00
d->processing_cycle.rx_start = cycle;
}
}
// This is executed one time. For in-stream, first packet has come. For out-stream, prepared to
// transmit first packet.
static void amdtp_stream_first_callback(struct fw_iso_context *context,
u32 tstamp, size_t header_length,
void *header, void *private_data)
{
struct amdtp_stream *s = private_data;
struct amdtp_domain *d = s->domain;
if (s->direction == AMDTP_IN_STREAM) {
context->callback.sc = drop_tx_packets_initially;
} else {
if (s == d->irq_target)
context->callback.sc = irq_target_callback_skip;
else
context->callback.sc = skip_rx_packets;
}
context->callback.sc(context, tstamp, header_length, header, s);
}
/**
* amdtp_stream_start - start transferring packets
* @s: the AMDTP stream to start
* @channel: the isochronous channel on the bus
* @speed: firewire speed code
* @queue_size: The number of packets in the queue.
* @idle_irq_interval: the interval to queue packet during initial state.
*
* The stream cannot be started until it has been configured with
* amdtp_stream_set_parameters() and it must be started before any PCM or MIDI
* device can be started.
*/
static int amdtp_stream_start(struct amdtp_stream *s, int channel, int speed,
unsigned int queue_size, unsigned int idle_irq_interval)
{
bool is_irq_target = (s == s->domain->irq_target);
unsigned int ctx_header_size;
unsigned int max_ctx_payload_size;
enum dma_data_direction dir;
struct pkt_desc *descs;
int i, type, tag, err;
mutex_lock(&s->mutex);
if (WARN_ON(amdtp_stream_running(s) ||
(s->data_block_quadlets < 1))) {
err = -EBADFD;
goto err_unlock;
}
if (s->direction == AMDTP_IN_STREAM) {
// NOTE: IT context should be used for constant IRQ.
if (is_irq_target) {
err = -EINVAL;
goto err_unlock;
}
ALSA: bebob/firewire-lib: Add a quirk for discontinuity at bus reset Normal BeBoB firmware has a quirk. When receiving bus reset, it transmits packets with discontinuous value in dbc field. This causes two situation, one is to abort streaming by firewire-lib as a result of detecting the discontinuity. Another is to call driver's .update() because of bus reset. These two is generated independently. (The former depends on isochronous stream and the latter depends on IEEE1394 bus driver.) When BeBoB driver works with XRUN-recoverable applications, this situation looks like stream_start_duplex() call followed by stream_update_duplex() call because applications will call snd_pcm_prepare() immediately at XRUN. To update connections and streams at first, this commit use completion. When queueing error occurs, stream_start_duplex() is forced to wait maximum 1000msec. During this, when .update() is called, the completion is waken and stream_start_duplex() is processed without breaking connections. At bus reset, stream_start_duplex() shouldn't break/establish connections and stream_update_duplex() should update connections because a caller of fw_iso_resources_allocate() is responsible for calling fw_iso_resources_update() on bus reset. This commit also adds a flag, which has an effect to skip checking continuity for first packet. This flag is useful for BeBoB quirk to start handling packets during streaming. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Iwai <tiwai@suse.de>
2014-04-25 13:45:16 +00:00
s->data_block_counter = UINT_MAX;
} else {
ALSA: bebob/firewire-lib: Add a quirk for discontinuity at bus reset Normal BeBoB firmware has a quirk. When receiving bus reset, it transmits packets with discontinuous value in dbc field. This causes two situation, one is to abort streaming by firewire-lib as a result of detecting the discontinuity. Another is to call driver's .update() because of bus reset. These two is generated independently. (The former depends on isochronous stream and the latter depends on IEEE1394 bus driver.) When BeBoB driver works with XRUN-recoverable applications, this situation looks like stream_start_duplex() call followed by stream_update_duplex() call because applications will call snd_pcm_prepare() immediately at XRUN. To update connections and streams at first, this commit use completion. When queueing error occurs, stream_start_duplex() is forced to wait maximum 1000msec. During this, when .update() is called, the completion is waken and stream_start_duplex() is processed without breaking connections. At bus reset, stream_start_duplex() shouldn't break/establish connections and stream_update_duplex() should update connections because a caller of fw_iso_resources_allocate() is responsible for calling fw_iso_resources_update() on bus reset. This commit also adds a flag, which has an effect to skip checking continuity for first packet. This flag is useful for BeBoB quirk to start handling packets during streaming. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Iwai <tiwai@suse.de>
2014-04-25 13:45:16 +00:00
s->data_block_counter = 0;
}
// initialize packet buffer.
if (s->direction == AMDTP_IN_STREAM) {
dir = DMA_FROM_DEVICE;
type = FW_ISO_CONTEXT_RECEIVE;
if (!(s->flags & CIP_NO_HEADER))
ctx_header_size = IR_CTX_HEADER_SIZE_CIP;
else
ctx_header_size = IR_CTX_HEADER_SIZE_NO_CIP;
} else {
dir = DMA_TO_DEVICE;
type = FW_ISO_CONTEXT_TRANSMIT;
ctx_header_size = 0; // No effect for IT context.
}
max_ctx_payload_size = amdtp_stream_get_max_ctx_payload_size(s);
err = iso_packets_buffer_init(&s->buffer, s->unit, queue_size, max_ctx_payload_size, dir);
if (err < 0)
goto err_unlock;
s->queue_size = queue_size;
s->context = fw_iso_context_create(fw_parent_device(s->unit)->card,
type, channel, speed, ctx_header_size,
amdtp_stream_first_callback, s);
if (IS_ERR(s->context)) {
err = PTR_ERR(s->context);
if (err == -EBUSY)
dev_err(&s->unit->device,
"no free stream on this controller\n");
goto err_buffer;
}
amdtp_stream_update(s);
if (s->direction == AMDTP_IN_STREAM) {
s->ctx_data.tx.max_ctx_payload_length = max_ctx_payload_size;
s->ctx_data.tx.ctx_header_size = ctx_header_size;
s->ctx_data.tx.event_starts = false;
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
if (s->domain->replay.enable) {
// struct fw_iso_context.drop_overflow_headers is false therefore it's
// possible to cache much unexpectedly.
s->ctx_data.tx.cache.size = max_t(unsigned int, s->syt_interval * 2,
queue_size * 3 / 2);
s->ctx_data.tx.cache.pos = 0;
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
s->ctx_data.tx.cache.descs = kcalloc(s->ctx_data.tx.cache.size,
sizeof(*s->ctx_data.tx.cache.descs), GFP_KERNEL);
if (!s->ctx_data.tx.cache.descs) {
err = -ENOMEM;
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
goto err_context;
}
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
}
} else {
static const struct {
unsigned int data_block;
unsigned int syt_offset;
} *entry, initial_state[] = {
[CIP_SFC_32000] = { 4, 3072 },
[CIP_SFC_48000] = { 6, 1024 },
[CIP_SFC_96000] = { 12, 1024 },
[CIP_SFC_192000] = { 24, 1024 },
[CIP_SFC_44100] = { 0, 67 },
[CIP_SFC_88200] = { 0, 67 },
[CIP_SFC_176400] = { 0, 67 },
};
s->ctx_data.rx.seq.descs = kcalloc(queue_size, sizeof(*s->ctx_data.rx.seq.descs), GFP_KERNEL);
if (!s->ctx_data.rx.seq.descs) {
err = -ENOMEM;
goto err_context;
}
s->ctx_data.rx.seq.size = queue_size;
s->ctx_data.rx.seq.pos = 0;
entry = &initial_state[s->sfc];
s->ctx_data.rx.data_block_state = entry->data_block;
s->ctx_data.rx.syt_offset_state = entry->syt_offset;
s->ctx_data.rx.last_syt_offset = TICKS_PER_CYCLE;
s->ctx_data.rx.event_count = 0;
}
ALSA: dice: fix kernel NULL pointer dereference due to invalid calculation for array index At a commit f91c9d7610a ('ALSA: firewire-lib: cache maximum length of payload to reduce function calls'), maximum size of payload for tx isochronous packet is cached to reduce the number of function calls. This cache was programmed to updated at a first callback of ohci1394 IR context. However, the maximum size is required to queueing packets before starting the isochronous context. As a result, the cached value is reused to queue packets in next time to starting the isochronous context. Then the cache is updated in a first callback of the isochronous context. This can cause kernel NULL pointer dereference in a below call graph: (sound/firewire/amdtp-stream.c) amdtp_stream_start() ->queue_in_packet() ->queue_packet() (drivers/firewire/core-iso.c) ->fw_iso_context_queue() ->struct fw_card_driver.queue_iso() (drivers/firewire/ohci.c) = ohci_queue_iso() ->queue_iso_packet_per_buffer() buffer->pages[page] The issued dereference occurs in a case that: - target unit supports different stream formats for sampling transmission frequency. - maximum length of payload for tx stream in a first trial is bigger than the length in a second trial. In this case, correct number of pages are allocated for DMA and the 'pages' array has enough elements, while index of the element is wrongly calculated according to the old value of length of payload in a call of 'queue_in_packet()'. Then it causes the issue. This commit fixes the critical bug. This affects all of drivers in ALSA firewire stack in Linux kernel v4.12 or later. [12665.302360] BUG: unable to handle kernel NULL pointer dereference at 0000000000000030 [12665.302415] IP: ohci_queue_iso+0x47c/0x800 [firewire_ohci] [12665.302439] PGD 0 [12665.302440] P4D 0 [12665.302450] [12665.302470] Oops: 0000 [#1] SMP PTI [12665.302487] Modules linked in: ... [12665.303096] CPU: 1 PID: 12760 Comm: jackd Tainted: P OE 4.13.0-38-generic #43-Ubuntu [12665.303154] Hardware name: /DH77DF, BIOS KCH7710H.86A.0069.2012.0224.1825 02/24/2012 [12665.303215] task: ffff9ce87da2ae80 task.stack: ffffb5b8823d0000 [12665.303258] RIP: 0010:ohci_queue_iso+0x47c/0x800 [firewire_ohci] [12665.303301] RSP: 0018:ffffb5b8823d3ab8 EFLAGS: 00010086 [12665.303337] RAX: ffff9ce4f4876930 RBX: 0000000000000008 RCX: ffff9ce88a3955e0 [12665.303384] RDX: 0000000000000000 RSI: 0000000034877f00 RDI: 0000000000000000 [12665.303427] RBP: ffffb5b8823d3b68 R08: ffff9ce8ccb390a0 R09: ffff9ce877639ab0 [12665.303475] R10: 0000000000000108 R11: 0000000000000000 R12: 0000000000000003 [12665.303513] R13: 0000000000000000 R14: ffff9ce4f4876950 R15: 0000000000000000 [12665.303554] FS: 00007f2ec467f8c0(0000) GS:ffff9ce8df280000(0000) knlGS:0000000000000000 [12665.303600] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [12665.303633] CR2: 0000000000000030 CR3: 00000002dcf90004 CR4: 00000000000606e0 [12665.303674] Call Trace: [12665.303698] fw_iso_context_queue+0x18/0x20 [firewire_core] [12665.303735] queue_packet+0x88/0xe0 [snd_firewire_lib] [12665.303770] amdtp_stream_start+0x19b/0x270 [snd_firewire_lib] [12665.303811] start_streams+0x276/0x3c0 [snd_dice] [12665.303840] snd_dice_stream_start_duplex+0x1bf/0x480 [snd_dice] [12665.303882] ? vma_gap_callbacks_rotate+0x1e/0x30 [12665.303914] ? __rb_insert_augmented+0xab/0x240 [12665.303936] capture_prepare+0x3c/0x70 [snd_dice] [12665.303961] snd_pcm_do_prepare+0x1d/0x30 [snd_pcm] [12665.303985] snd_pcm_action_single+0x3b/0x90 [snd_pcm] [12665.304009] snd_pcm_action_nonatomic+0x68/0x70 [snd_pcm] [12665.304035] snd_pcm_prepare+0x68/0x90 [snd_pcm] [12665.304058] snd_pcm_common_ioctl1+0x4c0/0x940 [snd_pcm] [12665.304083] snd_pcm_capture_ioctl1+0x19b/0x250 [snd_pcm] [12665.304108] snd_pcm_capture_ioctl+0x27/0x40 [snd_pcm] [12665.304131] do_vfs_ioctl+0xa8/0x630 [12665.304148] ? entry_SYSCALL_64_after_hwframe+0xe9/0x139 [12665.304172] ? entry_SYSCALL_64_after_hwframe+0xe2/0x139 [12665.304195] ? entry_SYSCALL_64_after_hwframe+0xdb/0x139 [12665.304218] ? entry_SYSCALL_64_after_hwframe+0xd4/0x139 [12665.304242] ? entry_SYSCALL_64_after_hwframe+0xcd/0x139 [12665.304265] ? entry_SYSCALL_64_after_hwframe+0xc6/0x139 [12665.304288] ? entry_SYSCALL_64_after_hwframe+0xbf/0x139 [12665.304312] ? entry_SYSCALL_64_after_hwframe+0xb8/0x139 [12665.304335] ? entry_SYSCALL_64_after_hwframe+0xb1/0x139 [12665.304358] SyS_ioctl+0x79/0x90 [12665.304374] ? entry_SYSCALL_64_after_hwframe+0x72/0x139 [12665.304397] entry_SYSCALL_64_fastpath+0x24/0xab [12665.304417] RIP: 0033:0x7f2ec3750ef7 [12665.304433] RSP: 002b:00007fff99e31388 EFLAGS: 00000246 ORIG_RAX: 0000000000000010 [12665.304465] RAX: ffffffffffffffda RBX: 00007fff99e312f0 RCX: 00007f2ec3750ef7 [12665.304494] RDX: 0000000000000000 RSI: 0000000000004140 RDI: 0000000000000007 [12665.304522] RBP: 0000556ebc63fd60 R08: 0000556ebc640560 R09: 0000000000000000 [12665.304553] R10: 0000000000000001 R11: 0000000000000246 R12: 0000556ebc63fcf0 [12665.304584] R13: 0000000000000000 R14: 0000000000000007 R15: 0000000000000000 [12665.304612] Code: 01 00 00 44 89 eb 45 31 ed 45 31 db 66 41 89 1e 66 41 89 5e 0c 66 45 89 5e 0e 49 8b 49 08 49 63 d4 4d 85 c0 49 63 ff 48 8b 14 d1 <48> 8b 72 30 41 8d 14 37 41 89 56 04 48 63 d3 0f 84 ce 00 00 00 [12665.304713] RIP: ohci_queue_iso+0x47c/0x800 [firewire_ohci] RSP: ffffb5b8823d3ab8 [12665.304743] CR2: 0000000000000030 [12665.317701] ---[ end trace 9d55b056dd52a19f ]--- Fixes: f91c9d7610a ('ALSA: firewire-lib: cache maximum length of payload to reduce function calls') Cc: <stable@vger.kernel.org> # v4.12+ Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Iwai <tiwai@suse.de>
2018-04-29 06:01:46 +00:00
2017-03-31 13:06:07 +00:00
if (s->flags & CIP_NO_HEADER)
s->tag = TAG_NO_CIP_HEADER;
else
s->tag = TAG_CIP;
// NOTE: When operating without hardIRQ/softIRQ, applications tends to call ioctl request
// for runtime of PCM substream in the interval equivalent to the size of PCM buffer. It
// could take a round over queue of AMDTP packet descriptors and small loss of history. For
// safe, keep more 8 elements for the queue, equivalent to 1 ms.
descs = kcalloc(s->queue_size + 8, sizeof(*descs), GFP_KERNEL);
if (!descs) {
err = -ENOMEM;
goto err_context;
}
s->packet_descs = descs;
INIT_LIST_HEAD(&s->packet_descs_list);
for (i = 0; i < s->queue_size; ++i) {
INIT_LIST_HEAD(&descs->link);
list_add_tail(&descs->link, &s->packet_descs_list);
++descs;
}
s->packet_descs_cursor = list_first_entry(&s->packet_descs_list, struct pkt_desc, link);
s->packet_index = 0;
do {
struct fw_iso_packet params;
if (s->direction == AMDTP_IN_STREAM) {
err = queue_in_packet(s, &params);
} else {
bool sched_irq = false;
params.header_length = 0;
params.payload_length = 0;
if (is_irq_target) {
sched_irq = !((s->packet_index + 1) %
idle_irq_interval);
}
err = queue_out_packet(s, &params, sched_irq);
}
if (err < 0)
goto err_pkt_descs;
} while (s->packet_index > 0);
/* NOTE: TAG1 matches CIP. This just affects in stream. */
tag = FW_ISO_CONTEXT_MATCH_TAG1;
2017-03-31 13:06:07 +00:00
if ((s->flags & CIP_EMPTY_WITH_TAG0) || (s->flags & CIP_NO_HEADER))
tag |= FW_ISO_CONTEXT_MATCH_TAG0;
s->ready_processing = false;
err = fw_iso_context_start(s->context, -1, 0, tag);
if (err < 0)
goto err_pkt_descs;
mutex_unlock(&s->mutex);
return 0;
err_pkt_descs:
kfree(s->packet_descs);
s->packet_descs = NULL;
err_context:
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
if (s->direction == AMDTP_OUT_STREAM) {
kfree(s->ctx_data.rx.seq.descs);
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
} else {
if (s->domain->replay.enable)
kfree(s->ctx_data.tx.cache.descs);
}
fw_iso_context_destroy(s->context);
s->context = ERR_PTR(-1);
err_buffer:
iso_packets_buffer_destroy(&s->buffer, s->unit);
err_unlock:
mutex_unlock(&s->mutex);
return err;
}
/**
* amdtp_domain_stream_pcm_pointer - get the PCM buffer position
* @d: the AMDTP domain.
* @s: the AMDTP stream that transports the PCM data
*
* Returns the current buffer position, in frames.
*/
unsigned long amdtp_domain_stream_pcm_pointer(struct amdtp_domain *d,
struct amdtp_stream *s)
{
struct amdtp_stream *irq_target = d->irq_target;
// Process isochronous packets queued till recent isochronous cycle to handle PCM frames.
if (irq_target && amdtp_stream_running(irq_target)) {
// In software IRQ context, the call causes dead-lock to disable the tasklet
// synchronously.
if (!in_softirq())
fw_iso_context_flush_completions(irq_target->context);
}
locking/atomics: COCCINELLE/treewide: Convert trivial ACCESS_ONCE() patterns to READ_ONCE()/WRITE_ONCE() Please do not apply this to mainline directly, instead please re-run the coccinelle script shown below and apply its output. For several reasons, it is desirable to use {READ,WRITE}_ONCE() in preference to ACCESS_ONCE(), and new code is expected to use one of the former. So far, there's been no reason to change most existing uses of ACCESS_ONCE(), as these aren't harmful, and changing them results in churn. However, for some features, the read/write distinction is critical to correct operation. To distinguish these cases, separate read/write accessors must be used. This patch migrates (most) remaining ACCESS_ONCE() instances to {READ,WRITE}_ONCE(), using the following coccinelle script: ---- // Convert trivial ACCESS_ONCE() uses to equivalent READ_ONCE() and // WRITE_ONCE() // $ make coccicheck COCCI=/home/mark/once.cocci SPFLAGS="--include-headers" MODE=patch virtual patch @ depends on patch @ expression E1, E2; @@ - ACCESS_ONCE(E1) = E2 + WRITE_ONCE(E1, E2) @ depends on patch @ expression E; @@ - ACCESS_ONCE(E) + READ_ONCE(E) ---- Signed-off-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: davem@davemloft.net Cc: linux-arch@vger.kernel.org Cc: mpe@ellerman.id.au Cc: shuah@kernel.org Cc: snitzer@redhat.com Cc: thor.thayer@linux.intel.com Cc: tj@kernel.org Cc: viro@zeniv.linux.org.uk Cc: will.deacon@arm.com Link: http://lkml.kernel.org/r/1508792849-3115-19-git-send-email-paulmck@linux.vnet.ibm.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-10-23 21:07:29 +00:00
return READ_ONCE(s->pcm_buffer_pointer);
}
EXPORT_SYMBOL_GPL(amdtp_domain_stream_pcm_pointer);
2017-06-07 00:38:05 +00:00
/**
* amdtp_domain_stream_pcm_ack - acknowledge queued PCM frames
* @d: the AMDTP domain.
2017-06-07 00:38:05 +00:00
* @s: the AMDTP stream that transfers the PCM frames
*
* Returns zero always.
*/
int amdtp_domain_stream_pcm_ack(struct amdtp_domain *d, struct amdtp_stream *s)
2017-06-07 00:38:05 +00:00
{
struct amdtp_stream *irq_target = d->irq_target;
// Process isochronous packets for recent isochronous cycle to handle
// queued PCM frames.
if (irq_target && amdtp_stream_running(irq_target))
fw_iso_context_flush_completions(irq_target->context);
2017-06-07 00:38:05 +00:00
return 0;
}
EXPORT_SYMBOL_GPL(amdtp_domain_stream_pcm_ack);
2017-06-07 00:38:05 +00:00
/**
* amdtp_stream_update - update the stream after a bus reset
* @s: the AMDTP stream
*/
void amdtp_stream_update(struct amdtp_stream *s)
{
/* Precomputing. */
locking/atomics: COCCINELLE/treewide: Convert trivial ACCESS_ONCE() patterns to READ_ONCE()/WRITE_ONCE() Please do not apply this to mainline directly, instead please re-run the coccinelle script shown below and apply its output. For several reasons, it is desirable to use {READ,WRITE}_ONCE() in preference to ACCESS_ONCE(), and new code is expected to use one of the former. So far, there's been no reason to change most existing uses of ACCESS_ONCE(), as these aren't harmful, and changing them results in churn. However, for some features, the read/write distinction is critical to correct operation. To distinguish these cases, separate read/write accessors must be used. This patch migrates (most) remaining ACCESS_ONCE() instances to {READ,WRITE}_ONCE(), using the following coccinelle script: ---- // Convert trivial ACCESS_ONCE() uses to equivalent READ_ONCE() and // WRITE_ONCE() // $ make coccicheck COCCI=/home/mark/once.cocci SPFLAGS="--include-headers" MODE=patch virtual patch @ depends on patch @ expression E1, E2; @@ - ACCESS_ONCE(E1) = E2 + WRITE_ONCE(E1, E2) @ depends on patch @ expression E; @@ - ACCESS_ONCE(E) + READ_ONCE(E) ---- Signed-off-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: davem@davemloft.net Cc: linux-arch@vger.kernel.org Cc: mpe@ellerman.id.au Cc: shuah@kernel.org Cc: snitzer@redhat.com Cc: thor.thayer@linux.intel.com Cc: tj@kernel.org Cc: viro@zeniv.linux.org.uk Cc: will.deacon@arm.com Link: http://lkml.kernel.org/r/1508792849-3115-19-git-send-email-paulmck@linux.vnet.ibm.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-10-23 21:07:29 +00:00
WRITE_ONCE(s->source_node_id_field,
(fw_parent_device(s->unit)->card->node_id << CIP_SID_SHIFT) & CIP_SID_MASK);
}
EXPORT_SYMBOL(amdtp_stream_update);
/**
* amdtp_stream_stop - stop sending packets
* @s: the AMDTP stream to stop
*
* All PCM and MIDI devices of the stream must be stopped before the stream
* itself can be stopped.
*/
static void amdtp_stream_stop(struct amdtp_stream *s)
{
mutex_lock(&s->mutex);
if (!amdtp_stream_running(s)) {
mutex_unlock(&s->mutex);
return;
}
fw_iso_context_stop(s->context);
fw_iso_context_destroy(s->context);
s->context = ERR_PTR(-1);
iso_packets_buffer_destroy(&s->buffer, s->unit);
kfree(s->packet_descs);
s->packet_descs = NULL;
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
if (s->direction == AMDTP_OUT_STREAM) {
kfree(s->ctx_data.rx.seq.descs);
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
} else {
if (s->domain->replay.enable)
kfree(s->ctx_data.tx.cache.descs);
}
mutex_unlock(&s->mutex);
}
/**
* amdtp_stream_pcm_abort - abort the running PCM device
* @s: the AMDTP stream about to be stopped
*
* If the isochronous stream needs to be stopped asynchronously, call this
* function first to stop the PCM device.
*/
void amdtp_stream_pcm_abort(struct amdtp_stream *s)
{
struct snd_pcm_substream *pcm;
locking/atomics: COCCINELLE/treewide: Convert trivial ACCESS_ONCE() patterns to READ_ONCE()/WRITE_ONCE() Please do not apply this to mainline directly, instead please re-run the coccinelle script shown below and apply its output. For several reasons, it is desirable to use {READ,WRITE}_ONCE() in preference to ACCESS_ONCE(), and new code is expected to use one of the former. So far, there's been no reason to change most existing uses of ACCESS_ONCE(), as these aren't harmful, and changing them results in churn. However, for some features, the read/write distinction is critical to correct operation. To distinguish these cases, separate read/write accessors must be used. This patch migrates (most) remaining ACCESS_ONCE() instances to {READ,WRITE}_ONCE(), using the following coccinelle script: ---- // Convert trivial ACCESS_ONCE() uses to equivalent READ_ONCE() and // WRITE_ONCE() // $ make coccicheck COCCI=/home/mark/once.cocci SPFLAGS="--include-headers" MODE=patch virtual patch @ depends on patch @ expression E1, E2; @@ - ACCESS_ONCE(E1) = E2 + WRITE_ONCE(E1, E2) @ depends on patch @ expression E; @@ - ACCESS_ONCE(E) + READ_ONCE(E) ---- Signed-off-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: davem@davemloft.net Cc: linux-arch@vger.kernel.org Cc: mpe@ellerman.id.au Cc: shuah@kernel.org Cc: snitzer@redhat.com Cc: thor.thayer@linux.intel.com Cc: tj@kernel.org Cc: viro@zeniv.linux.org.uk Cc: will.deacon@arm.com Link: http://lkml.kernel.org/r/1508792849-3115-19-git-send-email-paulmck@linux.vnet.ibm.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-10-23 21:07:29 +00:00
pcm = READ_ONCE(s->pcm);
if (pcm)
snd_pcm_stop_xrun(pcm);
}
EXPORT_SYMBOL(amdtp_stream_pcm_abort);
/**
* amdtp_domain_init - initialize an AMDTP domain structure
* @d: the AMDTP domain to initialize.
*/
int amdtp_domain_init(struct amdtp_domain *d)
{
INIT_LIST_HEAD(&d->streams);
d->events_per_period = 0;
return 0;
}
EXPORT_SYMBOL_GPL(amdtp_domain_init);
/**
* amdtp_domain_destroy - destroy an AMDTP domain structure
* @d: the AMDTP domain to destroy.
*/
void amdtp_domain_destroy(struct amdtp_domain *d)
{
// At present nothing to do.
return;
}
EXPORT_SYMBOL_GPL(amdtp_domain_destroy);
/**
* amdtp_domain_add_stream - register isoc context into the domain.
* @d: the AMDTP domain.
* @s: the AMDTP stream.
* @channel: the isochronous channel on the bus.
* @speed: firewire speed code.
*/
int amdtp_domain_add_stream(struct amdtp_domain *d, struct amdtp_stream *s,
int channel, int speed)
{
struct amdtp_stream *tmp;
list_for_each_entry(tmp, &d->streams, list) {
if (s == tmp)
return -EBUSY;
}
list_add(&s->list, &d->streams);
s->channel = channel;
s->speed = speed;
s->domain = d;
return 0;
}
EXPORT_SYMBOL_GPL(amdtp_domain_add_stream);
ALSA: firewire-lib: replay sequence of incoming packets for outgoing packets ALSA IEC 61883-1/6 packet streaming engine uses pre-computed parameters ideal for nominal sampling transfer frequency (STF) to transfer packets to device since it was added 2011. As a result of user experience for a decade, it is clear that the sequence is not suitable to some actual devices. It takes the devices to generate noise, and causes any type of discontinuity in the series of packet transferred from the device. It's required for the engine to transfer packets according to effective STF. The effective STF is given by media clock recovered by the sequence of packet transferred from the target device. In the previous commit, the sequence is already cached. The media clock recovery can be achieved by analyzing the sequence. In technological world, many ideas are proposed for media clock recovery. However, the small part of them could be actually adopted in our case since floating point arithmetic is not mostly available in Linux kernel land. This commit adopts the simple way from them; sequence replay, which means that the sequence of parameters from incoming packet is used as is to transfer outgoing packets. The media clock is not computed internally, but the sequence of outgoing packet superficially looks to be generated by the media clock. The association between source and destination is decided when starting AMDTP domain. When the target device supports a pair of isochronous packet streams, the tx stream is source and the rx stream is destination. When it supports two pair of streams, each of tx stream is associated to corresponding rx stream in its order. When it supports less number of tx streams than rx streams, the fist tx stream is selected for all of rx streams. When it supports more tx streams than rx streams, the first tx packet is associated to the rx stream. As I noted in previous commit, the sequence of parameters from incoming packet is different between devices, time to time. It is worse idea to replay the sequence of parameters from a device for the sequence of packet to the other devices even if they are in the same category of device. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-3-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:10 +00:00
// Make the reference from rx stream to tx stream for sequence replay. When the number of tx streams
// is less than the number of rx streams, the first tx stream is selected.
static int make_association(struct amdtp_domain *d)
{
unsigned int dst_index = 0;
struct amdtp_stream *rx;
// Make association to replay target.
list_for_each_entry(rx, &d->streams, list) {
if (rx->direction == AMDTP_OUT_STREAM) {
unsigned int src_index = 0;
struct amdtp_stream *tx = NULL;
struct amdtp_stream *s;
list_for_each_entry(s, &d->streams, list) {
if (s->direction == AMDTP_IN_STREAM) {
if (dst_index == src_index) {
tx = s;
break;
}
++src_index;
}
}
if (!tx) {
// Select the first entry.
list_for_each_entry(s, &d->streams, list) {
if (s->direction == AMDTP_IN_STREAM) {
tx = s;
break;
}
}
// No target is available to replay sequence.
if (!tx)
return -EINVAL;
}
rx->ctx_data.rx.replay_target = tx;
++dst_index;
}
}
return 0;
}
/**
* amdtp_domain_start - start sending packets for isoc context in the domain.
* @d: the AMDTP domain.
* @tx_init_skip_cycles: the number of cycles to skip processing packets at initial stage of IR
* contexts.
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
* @replay_seq: whether to replay the sequence of packet in IR context for the sequence of packet in
* IT context.
* @replay_on_the_fly: transfer rx packets according to nominal frequency, then begin to replay
* according to arrival of events in tx packets.
*/
int amdtp_domain_start(struct amdtp_domain *d, unsigned int tx_init_skip_cycles, bool replay_seq,
bool replay_on_the_fly)
{
unsigned int events_per_buffer = d->events_per_buffer;
unsigned int events_per_period = d->events_per_period;
unsigned int queue_size;
struct amdtp_stream *s;
bool found = false;
int err;
ALSA: firewire-lib: replay sequence of incoming packets for outgoing packets ALSA IEC 61883-1/6 packet streaming engine uses pre-computed parameters ideal for nominal sampling transfer frequency (STF) to transfer packets to device since it was added 2011. As a result of user experience for a decade, it is clear that the sequence is not suitable to some actual devices. It takes the devices to generate noise, and causes any type of discontinuity in the series of packet transferred from the device. It's required for the engine to transfer packets according to effective STF. The effective STF is given by media clock recovered by the sequence of packet transferred from the target device. In the previous commit, the sequence is already cached. The media clock recovery can be achieved by analyzing the sequence. In technological world, many ideas are proposed for media clock recovery. However, the small part of them could be actually adopted in our case since floating point arithmetic is not mostly available in Linux kernel land. This commit adopts the simple way from them; sequence replay, which means that the sequence of parameters from incoming packet is used as is to transfer outgoing packets. The media clock is not computed internally, but the sequence of outgoing packet superficially looks to be generated by the media clock. The association between source and destination is decided when starting AMDTP domain. When the target device supports a pair of isochronous packet streams, the tx stream is source and the rx stream is destination. When it supports two pair of streams, each of tx stream is associated to corresponding rx stream in its order. When it supports less number of tx streams than rx streams, the fist tx stream is selected for all of rx streams. When it supports more tx streams than rx streams, the first tx packet is associated to the rx stream. As I noted in previous commit, the sequence of parameters from incoming packet is different between devices, time to time. It is worse idea to replay the sequence of parameters from a device for the sequence of packet to the other devices even if they are in the same category of device. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-3-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:10 +00:00
if (replay_seq) {
err = make_association(d);
if (err < 0)
return err;
}
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
d->replay.enable = replay_seq;
d->replay.on_the_fly = replay_on_the_fly;
ALSA: firewire-lib: add replay target to cache sequence of packet In design of audio and music unit in IEEE 1394 bus, feedback of effective sampling transfer frequency (STF) is delivered by packets transferred from device. The devices supported by ALSA firewire stack are categorized to three groups regarding to it. * Group 1: * Echo Audio Fireworks board module * Oxford Semiconductor OXFW971 ASIC * Digidesign Digi00x family * Tascam FireWire series * RME Fireface series * Group 2: * BridgeCo. DM1000/DM1100/DM1500 ASICs for BeBoB solution * TC Applied Technologies DICE ASICs * Group 3: * Mark of the Unicord FireWire series In group 1, the effective STF is determined by the sequence of the number of events per packet. In group 2, the sequence of presentation timestamp expressed in syt field of CIP header is interpreted as well. In group 3, the presentation timestamp is expressed in source packet header (SPH) of each data block. I note that some models doesn't take care of effective STF with large internal buffer. It's reasonable to name it as group 0: * Group 0 * Oxford Semiconductor OXFW970 ASIC The effective STF is known to be slightly different from nominal STF for all of devices, and to be different between the devices. Furthermore, the effective STF is known to be shifted for long-period transmission. This makes it hard for software to satisfy the effective STF when processing packets to the device. The effective STF is deterministic as a result of analyzing the batch of packet transferred from the device. For the analysis, caching the sequence of parameter in the packet is required. This commit adds an option so that AMDTP domain structure takes AMDTP stream structure to cache the sequence of parameters in packet transferred from the device. The parameters are offset ticks of syt field against the cycle to receive the packet and the number of data blocks per packet. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210527122611.173711-2-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-27 12:26:09 +00:00
// Select an IT context as IRQ target.
list_for_each_entry(s, &d->streams, list) {
if (s->direction == AMDTP_OUT_STREAM) {
found = true;
break;
}
}
if (!found)
return -ENXIO;
d->irq_target = s;
d->processing_cycle.tx_init_skip = tx_init_skip_cycles;
// This is a case that AMDTP streams in domain run just for MIDI
// substream. Use the number of events equivalent to 10 msec as
// interval of hardware IRQ.
if (events_per_period == 0)
events_per_period = amdtp_rate_table[d->irq_target->sfc] / 100;
if (events_per_buffer == 0)
events_per_buffer = events_per_period * 3;
queue_size = DIV_ROUND_UP(CYCLES_PER_SECOND * events_per_buffer,
amdtp_rate_table[d->irq_target->sfc]);
list_for_each_entry(s, &d->streams, list) {
unsigned int idle_irq_interval = 0;
if (s->direction == AMDTP_OUT_STREAM && s == d->irq_target) {
idle_irq_interval = DIV_ROUND_UP(CYCLES_PER_SECOND * events_per_period,
amdtp_rate_table[d->irq_target->sfc]);
}
// Starts immediately but actually DMA context starts several hundred cycles later.
err = amdtp_stream_start(s, s->channel, s->speed, queue_size, idle_irq_interval);
if (err < 0)
goto error;
}
return 0;
error:
list_for_each_entry(s, &d->streams, list)
amdtp_stream_stop(s);
return err;
}
EXPORT_SYMBOL_GPL(amdtp_domain_start);
/**
* amdtp_domain_stop - stop sending packets for isoc context in the same domain.
* @d: the AMDTP domain to which the isoc contexts belong.
*/
void amdtp_domain_stop(struct amdtp_domain *d)
{
struct amdtp_stream *s, *next;
if (d->irq_target)
amdtp_stream_stop(d->irq_target);
list_for_each_entry_safe(s, next, &d->streams, list) {
list_del(&s->list);
if (s != d->irq_target)
amdtp_stream_stop(s);
}
d->events_per_period = 0;
d->irq_target = NULL;
}
EXPORT_SYMBOL_GPL(amdtp_domain_stop);