At Linux v3.5, packet processing can be done in process context of ALSA
PCM application as well as software IRQ context for OHCI 1394. Below is
an example of the callgraph (some calls are omitted).
ioctl(2) with e.g. HWSYNC
(sound/core/pcm_native.c)
->snd_pcm_common_ioctl1()
->snd_pcm_hwsync()
->snd_pcm_stream_lock_irq
(sound/core/pcm_lib.c)
->snd_pcm_update_hw_ptr()
->snd_pcm_udpate_hw_ptr0()
->struct snd_pcm_ops.pointer()
(sound/firewire/*)
= Each handler on drivers in ALSA firewire stack
(sound/firewire/amdtp-stream.c)
->amdtp_stream_pcm_pointer()
(drivers/firewire/core-iso.c)
->fw_iso_context_flush_completions()
->struct fw_card_driver.flush_iso_completion()
(drivers/firewire/ohci.c)
= flush_iso_completions()
->struct fw_iso_context.callback.sc
(sound/firewire/amdtp-stream.c)
= in_stream_callback() or out_stream_callback()
->...
->snd_pcm_stream_unlock_irq
When packet queueing error occurs or detecting invalid packets in
'in_stream_callback()' or 'out_stream_callback()', 'snd_pcm_stop_xrun()'
is called on local CPU with disabled IRQ.
(sound/firewire/amdtp-stream.c)
in_stream_callback() or out_stream_callback()
->amdtp_stream_pcm_abort()
->snd_pcm_stop_xrun()
->snd_pcm_stream_lock_irqsave()
->snd_pcm_stop()
->snd_pcm_stream_unlock_irqrestore()
The process is stalled on the CPU due to attempt to acquire recursive lock.
[ 562.630853] INFO: rcu_sched detected stalls on CPUs/tasks:
[ 562.630861] 2-...: (1 GPs behind) idle=37d/140000000000000/0 softirq=38323/38323 fqs=7140
[ 562.630862] (detected by 3, t=15002 jiffies, g=21036, c=21035, q=5933)
[ 562.630866] Task dump for CPU 2:
[ 562.630867] alsa-source-OXF R running task 0 6619 1 0x00000008
[ 562.630870] Call Trace:
[ 562.630876] ? vt_console_print+0x79/0x3e0
[ 562.630880] ? msg_print_text+0x9d/0x100
[ 562.630883] ? up+0x32/0x50
[ 562.630885] ? irq_work_queue+0x8d/0xa0
[ 562.630886] ? console_unlock+0x2b6/0x4b0
[ 562.630888] ? vprintk_emit+0x312/0x4a0
[ 562.630892] ? dev_vprintk_emit+0xbf/0x230
[ 562.630895] ? do_sys_poll+0x37a/0x550
[ 562.630897] ? dev_printk_emit+0x4e/0x70
[ 562.630900] ? __dev_printk+0x3c/0x80
[ 562.630903] ? _raw_spin_lock+0x20/0x30
[ 562.630909] ? snd_pcm_stream_lock+0x31/0x50 [snd_pcm]
[ 562.630914] ? _snd_pcm_stream_lock_irqsave+0x2e/0x40 [snd_pcm]
[ 562.630918] ? snd_pcm_stop_xrun+0x16/0x70 [snd_pcm]
[ 562.630922] ? in_stream_callback+0x3e6/0x450 [snd_firewire_lib]
[ 562.630925] ? handle_ir_packet_per_buffer+0x8e/0x1a0 [firewire_ohci]
[ 562.630928] ? ohci_flush_iso_completions+0xa3/0x130 [firewire_ohci]
[ 562.630932] ? fw_iso_context_flush_completions+0x15/0x20 [firewire_core]
[ 562.630935] ? amdtp_stream_pcm_pointer+0x2d/0x40 [snd_firewire_lib]
[ 562.630938] ? pcm_capture_pointer+0x19/0x20 [snd_oxfw]
[ 562.630943] ? snd_pcm_update_hw_ptr0+0x47/0x3d0 [snd_pcm]
[ 562.630945] ? poll_select_copy_remaining+0x150/0x150
[ 562.630947] ? poll_select_copy_remaining+0x150/0x150
[ 562.630952] ? snd_pcm_update_hw_ptr+0x10/0x20 [snd_pcm]
[ 562.630956] ? snd_pcm_hwsync+0x45/0xb0 [snd_pcm]
[ 562.630960] ? snd_pcm_common_ioctl1+0x1ff/0xc90 [snd_pcm]
[ 562.630962] ? futex_wake+0x90/0x170
[ 562.630966] ? snd_pcm_capture_ioctl1+0x136/0x260 [snd_pcm]
[ 562.630970] ? snd_pcm_capture_ioctl+0x27/0x40 [snd_pcm]
[ 562.630972] ? do_vfs_ioctl+0xa3/0x610
[ 562.630974] ? vfs_read+0x11b/0x130
[ 562.630976] ? SyS_ioctl+0x79/0x90
[ 562.630978] ? entry_SYSCALL_64_fastpath+0x1e/0xad
This commit fixes the above bug. This assumes two cases:
1. Any error is detected in software IRQ context of OHCI 1394 context.
In this case, PCM substream should be aborted in packet handler. On the
other hand, it should not be done in any process context. TO distinguish
these two context, use 'in_interrupt()' macro.
2. Any error is detect in process context of ALSA PCM application.
In this case, PCM substream should not be aborted in packet handler
because PCM substream lock is acquired. The task to abort PCM substream
should be done in ALSA PCM core. For this purpose, SNDRV_PCM_POS_XRUN is
returned at 'struct snd_pcm_ops.pointer()'.
Suggested-by: Clemens Ladisch <clemens@ladisch.de>
Fixes: e9148dddc3c7("ALSA: firewire-lib: flush completed packets when reading PCM position")
Cc: <stable@vger.kernel.org> # 4.9+
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
In recent commit for ALSA PCM core, some arrangement is done for
'struct snd_pcm_ops.ack' callback. This is called when appl_ptr is
explicitly moved in intermediate buffer for PCM frames, except for
some cases described later.
For drivers in ALSA firewire stack, usage of this callback has a merit to
reduce latency between time of PCM frame queueing and handling actual
packets in recent isochronous cycle, because no need to wait for software
IRQ context from isochronous context of OHCI 1394.
If this works well in a case that mapped page frame is used for the
intermediate buffer, user process should execute some commands for ioctl(2)
to tell the number of handled PCM frames in the intermediate buffer just
after handling them. Therefore, at present, with a combination of below
conditions, this doesn't work as expected and user process should wait for
the software IRQ context as usual:
- when ALSA PCM core judges page frame mapping is available for status
data (struct snd_pcm_mmap_status) and control data
(struct snd_pcm_mmap_control).
- user process handles PCM frames by loop just with 'snd_pcm_mmap_begin()'
and 'snd_pcm_mmap_commit()'.
- user process uses PCM hw plugin in alsa-lib to operate I/O without
'sync_ptr_ioctl' option.
Unfortunately, major use case include these three conditions.
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
During packet streaming, maximum length of payload for isochronous packet
is invariable, therefore no need to recalculate. Current ALSA IEC 61883-1/6
engine calls a function to calculate it 8,000 or more times per second
for incoming packet processing.
This commit adds a member to have maximum length of payload into 'struct
amdtp_stream', to reduces the function calls. At first callback from
isochronous context, the length is calculated and stored for later
processing.
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
As long as investigating Fireface 400, IEC 61883-1/6 is not applied to
its packet streaming protocol. Remarks of the specific protocol are:
* Each packet doesn't include CIP headers.
* 64,0 and 128,0 kHz are supported.
* The device doesn't necessarily transmit 8,000 packets per second.
* 0, 1, 2, 3 are used as tag for rx isochronous packets, however 0 is
used for tx isochronous packets.
On the other hand, there's a common feature. The number of data blocks
transferred in a second is the same as sampling transmission frequency.
Current ALSA IEC 61883-1/6 engine already has a method to calculate it and
this driver can utilize it for rx packets, as well as tx packets.
This commit adds support for the transferring protocol. CIP_NO_HEADERS
flag is newly added. When this flag is set:
* Both of 0 (without CIP header) and 1 (with CIP header) are used as tag
to handle incoming isochronous packet.
* 0 (without CIP header) is used as tag to transfer outgoing isochronous
packet.
* Skip CIP header evaluation.
* Use unique way to calculate the quadlets of isochronous packet payload.
In ALSA PCM interface, 128.0 kHz is not supported, and the ALSA
IEC 61883-1/6 engine doesn't support 64.0 kHz. These modes are dropped.
The sequence of rx packet has a remarkable quirk about tag. This will be
described in later commits.
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
In IEC 61883-1, when two quadlets CIP header is used, the most significant
bit in second CIP header stands. However, packets from units with MOTU
protocol version 3 have a quirk without this flag. Current packet streaming
layer handles this as protocol error.
This commit adds a new enumeration constant for this quirk, to handle MOTU
protocol version 3.
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Commit c8bdf49b9935("ALSA: fireworks/firewire-lib: Add a quirk for the
meaning of dbc") adds CIP_DBC_IS_END_EVENT flag just for tx packets.
However, MOTU FireWire series has this quirk for rx packets.
This commit allows both directions with the flag.
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
In IEC 61883-1, CIP headers can have a SPH field. When a packet has 1 in
SPH field of its CIP header, the packet has a source packet headers. A
source packet header consists of 32 bit field (= 1 quadlet) and it
transfers time stamp, which is the same value as the lower 25 bits of the
IEEE 1394 CYCLE_TIMER register and the rest is zero.
This commit just supports source packet header field because IEC 61883-1
includes ambiguity the position of this header and its count. Each
protocol layer is allowed to have actual implementation according its
requirements.
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Currently, packet streaming layer passes generated SYT value to data block
processing layer. However, this is not enough in a case that the data block
processing layer generates time stamps by its own ways.
For out-packet stream, the packet streaming layer guarantees 8,000 times
calls of data block processing layers per sec. Therefore, when cycle count
of the first packet is recorded, data block processing layers can calculate
own time stamps with the recorded value.
For the reason, this commit allows packet streaming layer to record the
first cycle count. Each data block processing layer can read the count by
accessing a member of structure for packet streaming layer.
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
These three commits were merged to improve PCM pointer granularity.
commit 76fb878948 ("ALSA: firewire-lib: taskletize the snd_pcm_period_elapsed() call")
commit e9148dddc3 ("ALSA: firewire-lib: flush completed packets when reading PCM position")
commit 92b862c7d6 ("ALSA: firewire-lib: optimize packet flushing")
The point of them is to handle queued packets not only in software IRQ
context of IR/IT contexts, but also in process context. As a result of
handling packets, period tasklet is scheduled when acrossing PCM period
boundary. This is to prevent recursive call of
'struct snd_pcm_ops.pointer()' in the same context.
When the pointer callback is executed in the process context, it's
better to avoid the second callback in the software IRQ context. The
software IRQ context runs immediately after scheduled in the process
context because few packets are queued yet.
For the aim, 'pointer_flush' is used, however it causes a race condition
between the process context and software IRQ context of IR/IT contexts.
Practically, this race is not so critical because it influences process
context to skip flushing queued packet and to get worse granularity of
PCM pointer. The race condition is quite rare but it should be improved
for stable service.
The similar effect can be achieved by using 'in_interrupt()' macro. This
commit obsoletes 'pointer_flush' with it.
Acked-by: Clemens Ladisch <clemens@ladisch.de>
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
In former commit, drivers in ALSA firewire stack always starts IT context
before IR context. If IR context starts after packets are transmitted by
peer unit, packet discontinuity may be detected because the context starts
in the middle of packet streaming. This situation is rare because IT
context usually starts immediately. However, it's better to solve this
issue. This is suppressed with CIP_SKIP_INIT_DBC_CHECK flag.
This commit enables the same feature as CIP_SKIP_INIT_DBC_CHECK.
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
In clause 6.3 of IEC 61883-6:2000, there's an explanation about processing
of presentation timestamp. In the clause, we can see "If a function block
receives a CIP, processes it and subsequently re-transmits it, then the
SYT of the outgoing CIP shall be the sum of the incoming SYT and the
processing delay." ALSA firewire stack has an implementation to partly
satisfy this specification. Developers assumed the stack to perform as an
Audio function block[1].
Following to the assumption, current implementation of ALSA firewire stack
use one software interrupt context to handle both of in/out packets. In
most case, this is processed in 1394 OHCI IR context independently of the
opposite context. Thus, this implementation uses longer CPU time in the
software interrupt context. This is not better for whole system.
Against the assumption, I confirmed that each ASIC for IEC 61883-1/6
doesn't necessarily expect it to the stack. Thus, current implementation
of ALSA firewire stack includes over-engineering.
This commit purges the implementation. As a result, packets of one
direction are handled in one software interrupt context and spends
minimum CPU time.
[1] [alsa-devel] [PATCH 0/8] [RFC] new driver for Echo Audio's Fireworks based devices
http://mailman.alsa-project.org/pipermail/alsa-devel/2013-June/062660.html
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This commit moves the codes related to data block processing from packet
streaming layer to AM824 layer.
Each driver initializes amdtp stream structure for AM824 data block by
calling amdtp_am824_init(). Then, a memory block is allocated for AM824
specific structure. This memory block is released by calling
amdtp_stream_destroy().
When setting streaming parameters, it calls amdtp_am824_set_parameters().
When starting packet streaming, it calls amdtp_stream_start(). When
stopping packet streaming, it calls amdtp_stream_stop().
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This commit renames some macros just related to AM824 format. In later
commit, they're moved to AM824 layer.
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Setting the format of PCM substream to AMDTP stream structure is important
to set a handler to copy actual PCM samples between buffers. The
processing should be in data block processing layer because essentially
it has no relationship to packet streaming.
This commit renames PCM format setting function to prepare for integrating
AM824 layer.
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
In IEC 61883-6, MIDI messages are transferred in MIDI conformant data
channel. Essentially, packet streaming layer is not responsible for MIDI
functionality.
This commit moves MIDI trigger helper function from the layer to AM824
layer. The rest of codes related to MIDI functionality will be moved in
later commits.
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The value of FDF field in CIP header is protocol-dependent. Thus, it's
better to allow data block processing layer to decide the value in any
timing.
In AM824 data format, the value of FDF field in CIP header indicates
N-flag and Nominal Sampling Frequency Code (sfc). The N-flag is for
switching 'Clock-based rate control mode' and 'Command-based rate control
mode'. In our implementation, 'Clock-based rate control mode' is just
supported. Therefore, When sampling transfer frequency is decided, then
the FDF can be set.
This commit replaces 'amdtp_stream_set_parameters' with
'amdtp_am824_set_parameters' to set the FDF. This is the same timing
to decide the ration between the number of data blocks and the number of
PCM frames.
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This commit adds data block processing layer for AM824 format. The new
layer initializes streaming layer with its value for fmt field.
Currently, most implementation of data block processing still remains
streaming layer. In later commits, these codes will be moved to the layer.
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
In later commit, data block processing layer will be newly added. This
layer will be named as 'amdtp-am824'.
This commit renames current amdtp file to amdtp-stream, to distinguish it
from the new layer.
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
