[alsa-devel] [PATCH] ALSA: firewire-lib: permit to flush queued packets only in process context for better PCM period granularity
These three commits were merged to improve PCM pointer granularity. commit 76fb87894828 ("ALSA: firewire-lib: taskletize the snd_pcm_period_elapsed() call") commit e9148dddc3c7 ("ALSA: firewire-lib: flush completed packets when reading PCM position") commit 92b862c7d685 ("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 --- sound/firewire/amdtp-stream.c | 23 +++++++++++++++++------ sound/firewire/amdtp-stream.h | 1 - 2 files changed, 17 insertions(+), 7 deletions(-)
diff --git a/sound/firewire/amdtp-stream.c b/sound/firewire/amdtp-stream.c index 830a95c..024ab7f 100644 --- a/sound/firewire/amdtp-stream.c +++ b/sound/firewire/amdtp-stream.c @@ -251,7 +251,6 @@ void amdtp_stream_pcm_prepare(struct amdtp_stream *s) tasklet_kill(&s->period_tasklet); s->pcm_buffer_pointer = 0; s->pcm_period_pointer = 0; - s->pointer_flush = true; } EXPORT_SYMBOL(amdtp_stream_pcm_prepare);
@@ -356,7 +355,6 @@ static void update_pcm_pointers(struct amdtp_stream *s, s->pcm_period_pointer += frames; if (s->pcm_period_pointer >= pcm->runtime->period_size) { s->pcm_period_pointer -= pcm->runtime->period_size; - s->pointer_flush = false; tasklet_hi_schedule(&s->period_tasklet); } } @@ -803,11 +801,24 @@ EXPORT_SYMBOL(amdtp_stream_start); */ unsigned long amdtp_stream_pcm_pointer(struct amdtp_stream *s) { - /* this optimization is allowed to be racy */ - if (s->pointer_flush && amdtp_stream_running(s)) + /* + * This function is called in software IRQ context of period_tasklet or + * process context. + * + * When the software IRQ context was scheduled by software IRQ context + * of IR/IT contexts, queued packets were already handled. Therefore, + * no need to flush the queue in buffer anymore. + * + * When the process context reach here, some packets will be already + * queued in the buffer. These packets should be handled immediately + * to keep better granularity of PCM pointer. + * + * Later, the process context will sometimes schedules software IRQ + * context of the period_tasklet. Then, no need to flush the queue by + * the same reason as described for IR/IT contexts. + */ + if (!in_interrupt() && amdtp_stream_running(s)) fw_iso_context_flush_completions(s->context); - else - s->pointer_flush = true;
return ACCESS_ONCE(s->pcm_buffer_pointer); } diff --git a/sound/firewire/amdtp-stream.h b/sound/firewire/amdtp-stream.h index 349c405..c1bc7fa 100644 --- a/sound/firewire/amdtp-stream.h +++ b/sound/firewire/amdtp-stream.h @@ -126,7 +126,6 @@ struct amdtp_stream { struct tasklet_struct period_tasklet; unsigned int pcm_buffer_pointer; unsigned int pcm_period_pointer; - bool pointer_flush;
/* To wait for first packet. */ bool callbacked;
On Wed, 11 May 2016 19:17:39 +0200, Takashi Sakamoto wrote:
These three commits were merged to improve PCM pointer granularity. commit 76fb87894828 ("ALSA: firewire-lib: taskletize the snd_pcm_period_elapsed() call") commit e9148dddc3c7 ("ALSA: firewire-lib: flush completed packets when reading PCM position") commit 92b862c7d685 ("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
Applied, thanks.
Takashi
Hi,
Thanks for applying this. And this is my last patches for Linux 4.7.
In this development period, for ALSA firewire stack, I did: * Apply delayed registration of sound card instance for all of drivers. * When registering it during bus-reset state of IEEE 1394 bus and userspace applications immediately start PCM substreams, corresponding packet streams are easily stopped. Additionally, some of audio and music units on IEEE 1394 bus are easily hung-up. This change is to avoid this situation, by postpone sound card registration 2 seconds after recorded bus-reset time. * Add tracepoints to dump a part of isochronous packet data. * This is to detect quirks of incoming packets from peer units, or for future development. * Cancel reuse of incoming packet parameter for outgoing packet parameter. * In some behaviours of peer units, we can see that the units don't expect drivers to perform as audio function block in IEC 61883-6. The reuse is over-engineering, and should be dropped for better usage of system resources. * Add support for M-Audio profire series. * Fix minor issues.
Past 8 months, since a development period for Linux 4.4, 152 patches have been merged, and I got tired. In next development cycle for 4.8, I'd like to have a rest from ALSA firewire stack.
I plan to do for 4.8: * Add APIs of control element set for ALSA userspace library. * Handle EPIPE error in ALSA Rawmidi core to stop substream in driver side. * Add tracepoints for refine processing of hardware parameter in ALSA PCM core, and obsolete DEBUG macro.
Pending issues (not yet scheduled): * Driver module to communicate between two Linux systems via PCM/Rawmidi interfaces. * Mmap implementation for hwdep character devices of firewire-tascam to share status and control messages transferred from units. * Driver module for a part of units in RME FireFace series. * Driver module for a part of units in Mark of the unicorn firewire series. * Propose in systemd project about udev rules for supported units. * Implement unique timestamping for Dice quirks. * Dynamic interval for software IRQ of IR/IT context. * Support 'No period wakeup', and work to enable 'timer scheduling' of PulseAudio.
Regards
Takashi Sakamoto
On May 12 2016 03:38, Takashi Iwai wrote:
On Wed, 11 May 2016 19:17:39 +0200, Takashi Sakamoto wrote:
These three commits were merged to improve PCM pointer granularity. commit 76fb87894828 ("ALSA: firewire-lib: taskletize the snd_pcm_period_elapsed() call") commit e9148dddc3c7 ("ALSA: firewire-lib: flush completed packets when reading PCM position") commit 92b862c7d685 ("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
Applied, thanks.
Takashi
participants (2)
-
Takashi Iwai -
Takashi Sakamoto